key: cord-0285261-07hgml3z
authors: Ding, Baoqing; Xia, Rui; Lin, Qiaoshan; Gurung, Vandana; Sagawa, Janelle M.; Stanley, Lauren E.; Strobel, Matthew; Diggle, Pamela K.; Meyers, Blake C.; Yuan, Yao-Wu
title: Developmental genetics of corolla tube formation: role of the tasiRNA-ARF pathway and a conceptual model
date: 2018-06-19
journal: bioRxiv
DOI: 10.1101/253112
sha: 21d8c15be6b663d9ad88121663c7a64c6efc9a25
doc_id: 285261
cord_uid: 07hgml3z

More than 80,000 angiosperm species produce flowers with petals fused into a corolla tube. As an important element of the tremendous diversity of flower morphology, the corolla tube plays a critical role in many specialized interactions between plants and animal pollinators (e.g., beeflies, hawkmoths, hummingbirds, nectar bats), which in turn drives rapid plant speciation. Despite its clear significance in plant reproduction and evolution, the corolla tube remains one of the least understood plant structures from a developmental genetics perspective. Through mutant analyses and transgenic experiments, here we show that the tasiRNA-ARF pathway is required for corolla tube formation in the monkeyflower species Mimulus lewisii. Loss-of-function mutations in the M. lewisii orthologs of ARGONAUTE7 and SUPPRESSOR OF GENE SILENCING 3 cause a dramatic decrease in abundance of TAS3-derived small RNAs and a moderate up-regulation of AUXIN RESPONSE FACTOR 3 (ARF3) and ARF4, which lead to inhibition of lateral expansion of the bases of petal primordia and complete arrest of the upward growth of the inter-primordial regions, resulting in unfused corollas. By using an auxin reporter construct, we discovered that auxin distribution is continuous along the petal primordium base and the inter-primordial region during the critical stage of corolla tube formation in the wild-type, and that this auxin distribution is much weaker and more restricted in the mutant. Together, these results suggest a new conceptual model highlighting the central role of auxin directed synchronized growth of the petal primordium base and the inter-primordial region in corolla tube formation.

About one third of the ~275,000 angiosperm species produce flowers with petals fused into a corolla tube (i.e., sympetalous), forming a protective enclosure of nectaries and reproductive organs. Corolla tubes have evolved multiple times independently across the angiosperm tree of life (Endress, 2011) , most notably in the common ancestor of the Asterids, a clade containing more than 80,000 species (Schonenberger and Von Balthazar, 2013) . Subsequent elaboration in length, width, and curvature has led to a great variety of corolla tube shapes that enabled asterid species to exploit many specialized pollinator groups (e.g., beeflies, hawkmoths, hummingbirds, nectar bats), which in turn drives rapid plant speciation (Muchhala, 2006; Hermann and Kuhlemeier, 2011; Paudel et al., 2015; Lagomarsino et al., 2016) . As such, the corolla tube has long been considered a key morphological innovation that contributed to the radiation of the Asterids (Endress, 2011) . Despite its critical importance in the reproductive success and evolution of such a large number of species, the corolla tube remains one of the least understood plant structures from a developmental genetics perspective (Specht and Howarth, 2015; Zhong and Preston, 2015) .

Historically, corolla tube formation has been the subject of extensive morphological and anatomical studies (Boke, 1948; Kaplan, 1968; Govil, 1972; Nishino, 1976 Nishino, , 1978 Nishino, , 1983a Nishino, , 1983b Erbar, 1991; Erbar and Leins, 1996; Kajita and Nishino, 2009; El Ottra et al., 2013) . In particular, numerous studies have described the detailed ontogenetic process of corolla tube development in one subgroup of the asterid clade, the Lamiids, which contains some classical plant genetic model systems such as snapdragon (Antirrhinum), petunia (Petunia), and morning glory (Ipomoea) (Govil, 1972; Nishino, 1976 Nishino, , 1978 Nishino, , 1983a Nishino, , 1983b Singh and Jain, 1979; Erbar, 1991; Vincent and Coen, 2004; Kajita and Nishino, 2009; Erbar and Leins, 2011) . A common theme emerging from these studies is that during the early stage of petal development, petal primordia are initiated separately, followed by rapid extension of the petal bases toward the inter-primordial regions, which also grow coordinately, causing congenital "fusion" of the petal primordia and formation of the corolla tube. Little is known, however, about the genetic control of this early-phase lateral extension of the petal base or the nature of the coordinated interprimordial growth.

To date only a few genes have been implicated in corolla tube formation. Loss-offunction alleles of the FEATHERED gene in Japanese morning glory (Ipomoea nil) and the MAEWEST gene in petunia (Petunia x hybrida), both generated by transposon insertions, result in unfused corollas (Iwasaki and Nitasaka, 2006; Vandenbussche et al., 2009) . FEATHERED and MAEWEST encode KANADI and WOX transcription factors, and their Arabidopsis orthologs are KANADI1 and WOX1, respectively. In addition, ectopic expression of the Arabidopsis TCP5 protein fused with a repressor motif in Ipomoea also disrupted corolla tube formation (Ono et al., 2012) . However, whether the endogenous TCP5 ortholog in Ipomoea is involved in corolla tube development is unclear. More recently, it was reported that transient knock-down of the Petunia NAC-transcription factors NAM and NH16 via virus-induced gene silencing (VIGS) also caused decreased petal fusion (Zhong et al., 2016) , but the interpretation of this result was confounded by the observation that occasional flowers produced on the "escape shoots" of the loss-offunction nam mutants have normal corolla tubes (Souer et al., 1996) . The fact that these genes were characterized from different plant systems and through different methods (transposon insertion alleles, heterologous expression of chimeric repressor, and VIGS) makes it challenging to interpret their genetic relationships and their precise functional roles in corolla tube formation.

One way to overcome this problem is to systematically analyze corolla tube mutants in a single model system. To this end, we have employed a new genetic model system, the monkeyflower species Mimulus lewisii, mainly for its ease in chemical mutagenesis and Agrobacterium-mediated in planta transformation (Owen and Bradshaw, 2011; Yuan et al., 2013a) . M. lewisii is a typical bumblebee-pollinated species with a conspicuous corolla tube ( Figure 1A ). Through ethyl methanesulfonate (EMS) mutagenesis, we have generated a dozen recessive mutants (named flayed) with split corolla tubes. Here we report the characterization of one group of mutants, caused by loss-of-function mutations in two genes that are required for the biogenesis of trans-acting short interfering RNAs (tasiRNAs).

Among the tasiRNA loci characterized to date, TAS3 is the most widely conserved, found in virtually all land plants (Xia et al., 2017) . TAS3 transcript bears two binding sites for miR390, which triggers the production of phased tasiRNAs, including the highly conserved "tasiARF" that targets AUXIN RESPONSE FACTOR 3 (ARF3) and ARF4 (Allen et al., 2005; Axtell et al., 2006) . This tasiRNA-ARF regulatory module has been shown to play a critical role in leaf adaxial/abaxial polarity and blade expansion (i.e., lamina growth) in both eudicots (Fahlgren et al., 2006; Garcia et al., 2006; Hunter et al., 2006; Yan et al., 2010; Yifhar et al., 2012; Zhou et al., 2013) and monocots (Nagasaki et al., 2007; Nogueira et al., 2007; Douglas et al., 2010) .

Consistent with previous studies, here we demonstrate that in the M. lewisii mutants, TAS3derived tasiRNAs decrease dramatically in abundance and MlARF3 and MlARF4 expression are upregulated. Importantly, we show that malfunction of the tasiRNA-ARF pathway in the M. lewisii mutants impedes the early lateral expansion of the petal primordium bases and the coordinated inter-primordial growth, most likely through change of auxin homeostasis, leading to unfused petal primordia. Integrating our molecular and phenotypic analyses of the tasiRNA-ARF pathway and auxin accumulation patterns in M. lewisii with historical insights from morphological and anatomical studies of various sympetalous species, we propose a new conceptual model for the genetic control of corolla tube formation, which offers logical connections among the sporadic previous reports of corolla tube mutants and makes clear predictions that can be readily tested using the Mimulus system.

Three of the recessive mutants recovered from EMS mutagenesis using the inbred line LF10, flayed1-flayed3, are morphologically indistinguishable. Pair-wise crosses suggested that they belong to two complementation groups, flayed1 and flayed2 (flayed3 is allelic to flayed2) ( Figure   1B and C). In addition to having unfused petals, these mutants display carpel fusion defects, with phenotypes varying from flower to flower within the same plant. Most mutant flowers have two fused carpels, as in the wild-type, but have partially split stigmas with more than two lobes ( Figure 1E ). Less frequently there are flowers with two almost completely separate styles. The length of mutant pistils is also reduced compared to the wild-type ( Figure 1E ). No obvious phenotypes were observed in the stamens of these mutants.

Another notable feature of the flayed1/2 mutants is the reduced width of lateral organs.

The dorsal and lateral petals show ~30% decrease in width compared to the wild-type, and the ventral petal shows ~37% decrease ( Figure 1F ; Table S1 ). Leaf width is also substantially reduced (by ~40%) in the mutants, but length is unaffected ( Figure 1G and H; Table S1 ). To determine whether the reduction in petal width is due to change in cell number, cell size, or both, we measured the width of abaxial epidermal cells of the dorsal petal lobe for both the wild-type and the flayed2 mutant. Because the petal lobe abaxial epidermal cells are irregularly shaped ( Figure 1I ), the width measurements were done on five contiguous cells to account for the variation among individual cells within the same sample. No significant difference in cell width was found between the wild-type and flayed2 ( Figure 1J ), which suggests that the difference in petal width between the mutant and the wild-type is primarily due to difference in cell number (i.e., number of cell divisions).

Unlike the morning glory mutant feathered (Iwasaki and Nitasaka, 2006) or the petunia mutant maewest (Vandenbussche et al., 2009) , flayed1/2 do not show any defects in tissue adaxial/abaxial polarity. Instead, the flayed1/2 mutants closely resemble the petunia mutant choripetala suzaane (chsu), which also have split corolla tubes, variable carpel fusion defects, and narrower leaf with normal adaxial/abaxial polarity. Unfortunately, the molecular identity of CHSU is still unknown.

To identify the causal genes of flayed1 and flayed2, we analyzed each mutant using a genomic approach that combines the advantages of bulk segregant analysis and comparison of single nucleotide polymorphism (SNP) profiles between multiple EMS mutants (Methods), as demonstrated in a previous study (LaFountain et al., 2017) . We narrowed the causal mutation of flayed1 and flayed2 down to 38 and 19 candidate SNPs, respectively (Table S2 and S3). The vast majority of these SNPs locate in non-coding, repetitive sequences, with only two or three mutations resulting in amino acid changes in each mutant (Table S2 and Table S2 and S3). AGO7 and SGS3 are part of the same tasiRNA biogenesis pathway (Peragine et al., 2004; Yoshikawa et al., 2005; Chen, 2010) , which would explain the indistinguishable mutant phenotypes of flayed1 and flayed2. Furthermore, sequencing the coding DNA (CDS) of MlSGS3 in flayed3, which is allelic to flayed2, revealed an independent mutation that also leads to a premature stop codon ( Figure 2B ). Together, these results suggested that MlAGO7 and MlSGS3 were the most promising candidate genes for showed a fully rescued phenotype that is indistinguishable from the wild-type; four lines showed a partially rescued phenotype, with petal and leaf width indistinguishable from wild-type but the petals remained unfused ( Figure 2C ). Similarly, six of the 18 MlSGS3 over-expression lines in the flayed2 background displayed a fully rescued phenotype and two displayed a partially rescued phenotype ( Figure 2D ). qRT-PCR assessment of MlAGO7 and MlSGS3 expression in 5mm floral buds showed that, in the fully rescued lines, expression levels of the transgenes are 4~64-fold higher than those of the corresponding endogenous genes ( Figure S1 ). These results confirmed that MlAGO7 and MlSGS3 are indeed the causal genes underlying flayed1 and flayed2, respectively.

Knowing the causal genes and mutations allowed direct genotyping of a "flayed1 x flayed2" F 2 population to identify flayed1 flayed2 double mutants, which are phenotypically indistinguishable from the single mutants ( Figure 1D , F, G, H). This further indicates that MlAGO7 and MlSGS3 function in the same genetic pathway in Mimulus, as expected.

The flayed1/2 Phenotypes Are Primarily Mediated Through the tasiRNA-ARF Pathway

Because AGO7 and SGS3 are necessary components of the miR390-TAS3-ARF pathway ( Figure   3A ), and the highly conserved, TAS3-derived tasiARFs are known to play a critical role in leaf polarity and lamina growth by repressing ARF3/4 expression, we hypothesized that the flayed1/2 phenotypes (e.g., reduced width of lateral organs) are primarily mediated through the tasiRNA- To test the first prediction, we sequenced the total small RNA pool from young floral buds (5-mm) of the wild-type, flayed1, and flayed 2. Like most other angiosperms, M. lewisii has two kinds of TAS3 genes (each represented by only a single copy in the M. lewisii genome):

TAS3S contains a single, centrally located tasiARF, whereas TAS3L contains two tandem tasiARFs (Xia et al., 2017 ) ( Figure S2 ). No TAS3S-derived small RNAs were detected in any of the sequenced samples, suggesting that the TAS3S gene is not expressed. TAS3L-derived small RNAs were detected at the level of ~600 per million reads in the wild-type, but decreased >50fold in both flayed1 and flayed2 ( Figure 3B ). In particular, the tasiARFs were almost entirely absent from the mutant samples ( Figure 3B ). These results confirmed the first prediction.

To test the second prediction, we first searched the M. lewisii genome for ARF3/4 homologs and found a single ortholog for each of the two genes. Similar to ARF3/4 in other species, both MlARF3 and MlARF4 have two binding sites with sequences complementary to tasiARF ( Figure 4A and B). qRT-PCR measurements in 5-mm floral buds showed that in the single and double mutants, MlARF3 was up-regulated by 1.7~2.5-fold and MlARF4 was upregulated by 2.7~3.7-fold ( Figure 4C ). This moderate up-regulation of ARF3/ARF4 in the ago7 and sgs3 mutant backgrounds is very similar to previous reports in Arabidopsis (Garcia et al., 2006; Hunter et al., 2006) , supporting the role of tasiARF in fine-tuning ARF3/4 expression level.

To test the third prediction, we transformed the wild-type with a tasiARF-insensitive version of MlARF3 (mMlARF3) and MlARF4 (mMlARF4) with several synonymous substitutions at the tasiARF binding sites ( Figure 4A and B), driven by the 35S promoter. We obtained seven independent 35S:mMlARF3 and 14 35S:mMlARF4 lines. In each case, only two transgenic lines showed obvious phenotypes: their leaves are very similar to the flayed1/2 mutants (i.e., narrower than the wild-type) and corollas are partially split (indicated by the red arrow heads in Figure 4D and E). qRT-PCR experiments on 5-mm floral buds of the transgenic lines with even the strongest phenotypes showed only moderate overexpression of MlARF3/4 relative to the wildtype (2~4-fold, Figure S3A and B). Examination of two random 35S:mMlARF4 lines without obvious phenotypes showed no increase in expression level of MlARF4 ( Figure S3C ). The lack of 35S:mMlARF3/4 lines with strong transgene expression is in contrast to ectopic expression of MlAGO7 and MlSGS3 ( Figure S1 ) as well as pigment-related transcription factors in M. lewisii (Yuan et al., 2014; Sagawa et al., 2016) , where the same 35S promoter could readily drive transgene expression level >10-fold higher than that of the endogenous genes. One possible explanation for this observation is that transgenic lines with very strong ARF3/4 expression in M.

lewisii are seedling lethal, as implicated by similar experiments in tomato (Yifhar et al., 2012) . 

To understand how malfunction of the tasiRNA-ARF pathway affects corolla tube formation in M. lewisii, we have studied floral organogenesis in the wild-type and the flayed2 mutant using scanning electron microscopy. Like other species in the lamiid clade (e.g., snapdragon, petunia, morning glory), M. lewisii petals are initiated as five separate primordia ( Figure 5A ). Petal development lags behind stamens in the early stages ( Figure 5B and C), but by the time the corolla reaches 0.5 mm in diameter ( Figure 5D ), petal development progresses rapidly and soon the stamens are found enclosed in the corolla ( Figure 5E -H). The developmental stage from 0.3 to 0.4 mm (corolla diameter) is critical for corolla tube formation: during this stage, the bases of the petal primordia quickly expand laterally (to a conspicuously greater extent than the upper portion of the petal primordia; Figure 5M ), and the inter-primordial regions also grow coordinately, connecting the initially separate petal primordia. Floral organogenesis of flayed2 is very similar to that of the wild-type at the early stages (before the corolla reaches 0.3 mm in diameter; Figure 5I ). However, during the critical period (0.3~0.4 mm), there is no preferential lateral expansion at the bases of the petal primordia, manifested as the truncate shape of the petal primordium base ( Figure 5N ), in contrast to the semi-circle shape of the wild-type ( Figure 5M ).

Notably, growth of the inter-primordial regions is also arrested in flayed2, leading to a gap between two adjacent petal primordia (Figure 5J-L and indicated by the asterisk in Figure 5N ).

Given that disruption of tasiRNA biogenesis and the consequent up-regulation of ARF3/4 have been shown to cause reduced lamina growth of lateral organs in multiple plant species (Peragine et al., 2004; Douglas et al., 2010; Yan et al., 2010; Yifhar et al., 2012; Zhou et al., 2013) , it is not surprising to observe reduced lateral expansion at the bases of the petal primordia in flayed2 compared to the wild-type ( Figure 5M and N). But how does this relate to the arrest of upward growth of the inter-primordial regions?

In a series of careful anatomical studies of various taxa in the Asterids, Nishino recognized that the "co-operation" between the marginal meristem activities of the base of the petal primordia and the upward growth of the inter-primordial regions plays a pivotal role in corolla tube formation (Nishino, 1976 (Nishino, , 1978 (Nishino, , 1983a (Nishino, , 1983b , although the nature of this "cooperation" was unclear. Considering this earlier insight together with our own results ( Figure 5M and N), we speculated that in the wild-type, there is a molecular signal with continuous distribution in the marginal meristematic cells along the petal primordium base and the interprimordial region, stimulating synchronized growth between the two regions. In the flayed2 mutant, this growth signal is perhaps reduced or disrupted in spatial distribution, leading to arrested growth in the zone encompassing the margins of primordium base and the interprimordial region. An obvious candidate for this putative signal is the phytohormone auxin, which is known to promote localized tissue outgrowth and meanwhile suppress organ boundary genes such as CUP-SHAPED COTYLEDON 1 (CUC1) and CUC2 in Arabidopsis (Bilsborough et al., 2011) .

To examine auxin distribution in developing corolla buds, we introduced an auxin reporter construct DR5rev:mRFPer (Gallavotti et al., 2008) into the wild-type LF10. In the very early stage where petal primordia just initiate (corresponding to a stage between Figure 5A and B), auxin is accumulated in both petal primordia and inter-primordial regions, with a clear gap between the two ( Figure 6A ). As the corolla bud reaches 0.4 mm in diameter, auxin becomes concentrated on the apex of the petal primordium and the "synchronized growth zone" encompassing the margins along the primordium base and the inter-primordial region (between the arrow heads in Figure 6B ), with relatively weak signal along the upper part of the petal primordium margin. It is worth noting that the auxin localization pattern at this stage corresponds almost perfectly to the petal growth pattern that leads to corolla tube formation (i.e., preferential lateral expansion of the petal primordium base and the coordinated upward growth of the interprimordial region; Figure 5M ). As the corolla bud reaches 0.5 mm in diameter, auxin distribution remains continuous at the base of the petal primordium and the inter-primordial region (demarcated by the arrow heads in Figure 6C ), but the gap devoid of auxin between the petal apex and the synchronized growth zone becomes more conspicuous than the 0.4-mm stage.

When the corolla bud reaches 0.6 mm in diameter, auxin signal starts to spread more evenly along the entire petal margin ( Figure 6D ), likely corresponding to later growth of the entire corolla. While all confocal images shown in Figure 6 was taken using DR5rev:mRFPer line 2, examination of several additional independent transgenic lines showed very similar results.

To test whether auxin homeostasis is altered in the flayed2 mutant, we crossed DR5rev:mRFPer line 2 with flayed2, and analyzed F 2 individuals that are homozygous for the flayed2 mutation and with the DR5rev:mRFPer transgene. When imaged under the same conditions, the flayed2 corolla buds showed much weaker auxin signal overall than the wild-type ( Figure 6E and F). In particular, the junction between the petal primordium base and the interprimordial region (arrows in Figure 6E and F) showed no signal at all, consistent with the lack of synchronized growth between these two regions in the mutant ( Figure 5N ). The decreased auxin signals in flayed2 is likely caused by up-regulation of MlARF3/4, as ARF3 has been implicated in auxin homeostasis in Arabidopsis by directly repressing auxin biosynthesis and transport (Simonini et al., 2017) .

One potential consequence of the reduced and discontinuous auxin distribution in the "synchronized growth zone" of flayed2 corolla buds is up-regulation of organ boundary genes such as NAM/CUC1/CUC2 homologs (Bilsborough et al., 2011) . To test this possibility, we mined an unpublished transcriptome assembled from various developmental stages of M. lewisii floral buds, to search for genes that are closely related to NAM/CUC1/CUC2, which can be identified using the amino acid motif "HVSCFS[N/T/S]" downstream of the NAC domain (Ooka et al., 2003) in addition to overall sequence identity. The "HVSCFS[N/T/S]" motif also corresponds to the recognition site for miR164 (Laufs et al., 2004; Mallory et al., 2004) . We found three genes named MlNAC1/2/3 with highest sequence identity to NAM/CUC1/CUC2 and with the putative miR164 recognition site ( Figure S4A and B). qRT-PCR experiments on 2-mm floral buds showed no expression difference between the wild-type and flayed2 for MlNAC1 or MlNAC2; however, MlNAC3 transcript level is ~8-fold higher in flayed2 than in the wild-type ( Figure S4C ). The functional relevance of MlNAC3 up-regulation in corolla tube formation will need be tested by future transgenic experiments.

This study represents the first step of a systematic effort towards understanding the developmental genetics of corolla tube formation using the new model system Mimulus lewisii.

We show that the tasiRNA-ARF pathway is required for the synchronized growth between the petal primordium base and the inter-primordial region at early stages of corolla development.

Furthermore, we discovered that auxin distribution is continuous in this synchronized growth zone in the wild-type, but becomes much reduced in the flayed2 mutant with split corolla tubes, explaining the inhibition of lateral expansion at the base of the petal primordia and complete arrest of the upward growth of the inter-primordial regions observed in the mutant. Together, these results suggest a new conceptual model for the developmental genetic control of corolla tube formation.

Although this is the first detailed study investigating the role of the tasiRNA-ARF pathway in corolla tube formation, a previous study on leaf development in the family Solanaceae has mentioned in passing that malfunction of the tasiRNA-ARF pathway led to unfused corolla in tomato and tobacco flowers (Yifhar et al., 2012) . Like Mimulus, the family Solanaceae belongs to the asterid clade. This suggests that the role of the tasiRNA-ARF pathway in corolla tube formation is likely conserved across asterid plants. More surprisingly, two other studies on leaf development in Medicago truncatula and Lotus japonicus, both belonging to the legume family (Fabaceae), also implicated an indispensable role of the tasiRNA-ARF pathway in petal fusion (Yan et al., 2010; Zhou et al., 2013) . Typical legume flowers have separate dorsal and lateral petals; but the two ventral petals often fuse into a prow-shaped structure (i.e., the "keel"). In the ago7 mutants of Medicago truncatula and Lotus japonicus, the two ventral petals become separated. The family Fabaceae belongs to the clade Rosids. Given that the vast majority of rosid species (e.g., Arabidopsis thaliana) produce flowers with completely separate petals (i.e., polypetalous), with only a few exceptions in derived lineages (Zhong and Preston, 2015) , it is clear that the tasiRNA-ARF pathway was independently recruited to enable petal fusion in the legume species.

There are two significant insights emerging from our molecular and phenotypic analyses of the tasiRNA-ARF pathway in Mimulus that were not known from the aforementioned studies that focused on leaf development: (i) The tasiRNA-ARF pathway is required not only for the lamina growth of the petal primordia, but also for the upward growth of the inter-primordial regions ( Figure 5M and N) . In fact, we think that the synchronized growth between the petal primordium base and the inter-primordial region is the key to corolla tube formation; (ii) There is a tight correlation between auxin distribution and the growth pattern in developing corolla buds ( Figure 5M ; Figure 6B and C). The drastic reduction of auxin concentration in the flayed2 mutant compared to the wild-type is likely due to up-regulation of MlARF3/4, as there are evidence from Arabidopsis that ARF3 can directly repress auxin biosynthesis and transportation (Simonini et al., 2017) .

Given the well-known function of the tasiRNA-ARF pathway in lamina growth of lateral organs (Fahlgren et al., 2006; Garcia et al., 2006; Hunter et al., 2006; Nagasaki et al., 2007; Nogueira et al., 2007; Douglas et al., 2010Yan et al., 2010 Yifhar et al., 2012; Zhou et al., 2013) and the reduced petal width of the flayed1/2 mutants, it is easy to over-interpret the significance of lamina growth of the petal primordia in the "fusion" of adjacent petals. However, the following observations suggest that lamina growth of the petal primordia per se is not the key to corolla tube formation: (i) In some of the 35S:MlAGO7 and 35S:MlSGS3 transgenic lines (in the corresponding mutant background), petal width was restored to wild-type level, but the petals remained unfused (Figure 2C and D) . In addition, we have several other yet-to-be-characterized flayed mutants with split corolla but normal petal width. These observations suggest that petal lamina growth at the wild-type level is not sufficient for tube formation; (ii) Petal width of a previously characterized M. lewisii mutant, act1-D, is very similar to that of flayed1/2 , but the corolla tube of act1-D is intact. This suggests that reduced petal lamina growth does not necessarily affect petal fusion. Instead, we think it is the synchronization between the lateral expansion of the petal primordium base and the upward growth of the inter-primordia region, most likely directed by auxin signaling, that holds the key to corolla tube formation.

The difference in auxin accumulation patterns between flayed2 and the wild-type ( Figure 6) suggests that the role of the tasiRNA-ARF pathway in corolla tube formation lies in the regulation of auxin homeostasis. But how exactly does the tasiRNA-ARF pathway regulates the spatial pattern and dynamics of auxin distribution in the developing corolla buds of Mimulus remains to be elucidated.

Formation Two recent attempts of building a conceptual framework for floral organ fusion in general (Specht and Howarth, 2015) or petal fusion in particular (Zhong and Preston, 2015) both emphasized the genetic regulatory network underlying organ boundary formation and maintenance. The rationale for such emphasis was explicitly stated by Specht and Howarth (Specht and Howarth, 2015) : "fusion as a process may more accurately be defined as a lack of organ separation". While these attempts represent an important step towards a mechanistic understanding of the developmental process of corolla tube formation, to some degree they have neglected the insight provided by earlier morphological and anatomical studies (i.e., the importance of the "co-operation" between the rapid lateral expansion of the petal primordium base and the upward growth of the inter-primordial region), and have not provided a logical explanation to the previously reported corolla tube mutants (e.g., morning glory feathered, petunia maewest) (Iwasaki and Nitasaka, 2006; Vandenbussche et al., 2009) in terms of organ boundary formation or maintenance. Overemphasis on the "lack of organ separation" may represent an underestimation of the complexity of the corolla tube formation process, with the implication that simple loss-of-function mutations in some organ boundary genes in polypetalous species (e.g., Arabidopsis thaliana) could produce a functional corolla tube. As far as we are aware of, despite extensive mutagenesis of Arabidopsis in the past 40 years, such a mutant has not been reported.

Integrating our results on the tasiRNA-ARF pathway and auxin distributions in Mimulus lewisii with previous reports of corolla tube mutants (Iwasaki and Nitasaka, 2006; Vandenbussche et al., 2009 ) as well as historical insights from anatomical studies (Nishino, 1978 (Nishino, , 1983a (Nishino, , 1983b , we propose a new conceptual model for the developmental genetic control of corolla tube formation (Figure 7) . At the heart of the model is auxin-induced synchronized growth of the marginal meristematic cells at the base of the petal primordia and the interprimordial cells, providing a molecular explanation for the "co-operation" between the petal primordium base and the inter-primordial region observed in anatomical studies. Upstream of this core module is the genetic regulatory network controlling adaxial/abaxial polarity and lamina growth of lateral organs (Nakata and Okada, 2013; Tsukaya, 2013; Kuhlemeier and Timmermans, 2016) , which is conserved in a wide range of angiosperms and is somehow coopted in sympetalous species to regulate auxin homeostasis in the synchronized growth zone.

Downstream of this core module lie the organ boundary genes that would suppress localized tissue growth if not repressed by auxin signaling.

This model can readily explain the phenotypes of loss-of-function mutations in the morning glory FEATHERED gene and the petunia MAEWEST gene, which encode KANADI and WOX transcription factors, respectively (Iwasaki and Nitasaka, 2006; Vandenbussche et al., 2009 ). Together with the tasiRNA-ARF pathway, these transcription factors are part of the genetic network regulating leaf adaxial-abaxial polarity and lamina growth (Nakata and Okada, 2013; Tsukaya, 2013; Kuhlemeier and Timmermans, 2016) . Recent studies in other model systems such as Arabidopsis, Medicago, and Nicotiana, have demonstrated that these polarity regulators largely function by modulating auxin homeostasis (Tadege et al., 2011; Huang et al., 2014; Simonini et al., 2017) . According to our model, interfering with this genetic regulatory network is expected to reduce or disrupt the continuous auxin distribution in the synchronized growth zone, as shown in the flayed2 mutant ( Figure 5N ; Figure 6E and F), and consequently resulting in unfused corolla.

Our model also provides a plausible explanation for the petal fusion defects observed when a chimeric repressor of AtTCP5 was over-expressed in the Japanese morning glory (Ono et al., 2012) . Chimeric repressors of CIN-like TCP transcription factors, including TCP5 in Arabidopsis, are known to activate organ boundary genes such as CUC1/2 (Koyama et al., 2007) , which are otherwise suppressed by auxin (Bilsborough et al., 2011;  Figure S4C ). Ectopic activation of CUC1/2 is expected to prevent the upward growth of the inter-primordial regions (i.e., boundary between adjacent petal primordia). Also consistent with the "organ boundary" module ( Figure 7C ) is a recent observation in snapdragon --the expression of the CUC ortholog, CUPULIFORMIS, is cleared from the inter-primordial regions shortly after petal initiation but later is reactivated in the sinuses between adjacent corolla lobes (Rebocho et al., 2017) . Through computational modeling, Rebocho et al. (2017) showed that this "gap" of CUPULIFORMIS expression (between the base of the corolla and the sinuses) is necessary for corolla tube formation.

In addition to explaining these previous observations, our model predicts that misregulation of auxin biosynthesis, polar transport, and signaling within and between petal primordia, in transgenic or mutant plants of sympetalous species should result in defective corolla tubes. It also predicts that transgenic manipulation of other components of the leaf polarity/lamina growth network (e.g., AS1, AS2, HD-ZIPIII) or ectopic expression of organ boundary genes (e.g., CUC, ORGAN BOUNDARY1, JAGGED LATERAL ORGAN) (Aida et al., 1997; Borghi et al., 2007; Cho and Zambryski, 2011) in sympetalous species may also result in unfused corollas. The availability of multiple corolla tube mutants, the ease of bulk segregant analysis to identify mutant genes, and the amenability of Agrobacterium-mediated in planta transformation make Mimulus a favorable system to test these predictions and to dissect the detailed molecular mechanisms and developmental process of corolla tube formation.

EMS mutagenesis was performed using the Mimulus lewisii Pursh inbred line LF10 as described (Owen and Bradshaw, 2011) . Another inbred line SL9 was used to generate the "flayed x SL9" F 2 populations. Plants were grown in the University of Connecticut research greenhouses under natural light supplemented with sodium vapor lamps, ensuring a 16-hr day length.

To quantify phenotypic differences between the mutants and wild-type, we measured the widths of the dorsal, ventral, and lateral petals using a digital caliper. We also measured the lengths and widths of the fourth leaf (the largest leaf) of mature plants. To further evaluate whether the width difference is due to change in cell number, cell size or both, width of the abaxial epidermal cells of the dorsal petal lobe was measured following a previously described procedure .

To identify causal genes underlying flayed1 and flayed2, we employed a hybrid strategy that combines the advantages of bulk segregant analysis and genome comparisons between multiple EMS mutants, as described previously (LaFountain et al., 2017) . Briefly, for each mutant an F 2 population was produced by crossing the homozygous mutant (generated in the LF10 background) and the mapping line SL9. DNA samples from 96 F 2 segregants displaying the mutant phenotype (i.e., homozygous for the causal mutation) were pooled with equal representation. A small-insert library was then prepared for the pooled sample and was sequenced using an Illumina HiSeq 2000 platform at the University of North Carolina High Throughput Sequencing Facility. About 213 and 448 million 100-bp paired-end reads were generated for flayed1 and flayed2, respectively.

The short reads were mapped to the LF10 genome assembly version 1.8 (http://monkeyflower.uconn.edu/resources/) using CLC Genomics Workbench 7.0 (Qiagen, Valencia, CA). The causal mutation should be: (i) homozygous for the pooled sample (i.e., 100% SNP frequency in the "F 2 reads -LF10 genome" alignment); and (ii) unique to each mutant (i.e., any shared 100% SNPs between mutants are most likely due to assembly error in the reference genome or non-specific mapping of repetitive sequences). After comparisons to the SNP profiles of previously published mutants, guideless (Yuan et al., 2013b) , rcp1 , act1-D , and rcp2 (Stanley et al., 2017) , we narrowed the causal mutation to 39 and 19 candidate SNPs for flayed1 and flayed2, respectively ( Figure S5 ; Table S2 and S3).

For small RNA sequencing, total RNA was first extracted using the Spectrum Plant Total RNA Kit (Sigma-Aldrich) from 5-mm floral buds of LF10, flayed1, and flayed2 (two biological replicates for each genotype). Small RNA libraries were then constructed using the Illumina TruSeq Small RNA Sample Preparation Kits, with the total RNA as starting material. The libraries were sequenced on an Illumina HiSeq 2500 at the Delaware Biotechnology Institute (Newark, DE) . Small RNA reads were quality-controlled and adaptor-trimmed before calculating tasiRNA abundance, as described in Xia et al. (2017) .

RNA extraction and cDNA synthesis were as previously described (Yuan et al., 2013a) . cDNA samples were diluted 10-fold before qRT-PCR. All qRT-PCRs were performed using iQ SYBR Green Supermix (Bio-Rad) in a CFX96 Touch Real-Time PCR Detection System (Bio-Rad).

Samples were amplified for 40 cycles of 95 o C for 15 s and 60 o C for 30 s. Amplification efficiencies for each primer pair were determined using critical threshold values obtained from a dilution series (1:4, 1:8, 1:16, 1:32) of pooled cDNA. MlUBC was used as a reference gene as described (Yuan et al., 2013a) . Primers used for qRT-PCR are listed in Table S4 .

To generate the 35S:MlAGO7 and 35S:MlSGS3 constructs for the rescue experiments, we first amplified the full-length CDS of MlAGO7 and MlSGS3 from the wild-type LF10 cDNA using the Phusion enzyme (NEB, Ipswich, MA). For each gene, the amplified fragment was cloned into the pENTR/D-TOPO vector (Invitrogen) and then a linear fragment containing the CDS flanked by the attL1 and attL2 sites was amplified using M13 primers. This linear fragment was subsequently recombined into the Gateway vector pEarleyGate 100 (Earley et al., 2006) , which drives transgene expression by the CaMV 35S promoter. To generate the 35S:mMlARF3/4 constructs, CDS of sRNA insensitive forms of MlARF3 (mARF3) and MlARF4 (mARF4) that carries synonymous substitutions in the two tasiRNA recognition sites were synthesized by GenScript (NJ, USA) and then cloned into the pEarleyGate 100 destination vector as described for the 35S:MlAGO7 and 35S:MlSGS3 constructs. All plasmids were verified by sequencing before being transformed into Agrobacterum tumefaciens strain GV3101 for subsequent plant transformation, as described in Yuan et al. (2013a) . Primers used for plasmid construction and sequencing are listed in Table S5 .

Scanning Electron Microscopy. Flower buds were fixed overnight in Formalin-Acetic-Alcohol (FAA) at 4°C, dehydrated for 30 min through a 50%, 60%, 70%, 95%, and 100% alcohol series.

Samples were then critical-point dried, mounted, and sputter coated before being observed using a NOVA NanoSEM with Oxford EDX at 35 kV at UConn's Bioscience Electron Microscopy Laboratory.

Early developing floral buds (with sepals removed if necessary) carrying the DR5rev:mRFPer construct were laser scanned in red channel with a Z-stack. All fluorescence images were acquired using a Nikon A1R confocal laser scanning microscope equipped with a 20X objective (Plan APO VC 20X/1.20) at UConn's Advanced Light Microscopy Facility.

Short read data have been deposited in NCBI SRA (BioProject PRJNA423263); small RNA data have been deposited in NCBI GEO (GSE108530); annotated gene sequences have been deposited in GenBank (MG669632-MG669634 and MF084285). Table S1 . Relative transcript level of MlAGO7 (A) and MlSGS3 (B) in 5-mm floral buds of four representative, fully rescued over-expression lines compared to the corresponding mutant backgrounds, as determined by qRT-PCR. R1 and R2 represent two biological replicates. MlUBC was used as the reference gene. Error bars represent 1 SD from three technical replicates. Table S1 . Measurements of width (mm) of the dorsal, lateral and ventral petal, and the length (mm) and width (mm) of the fourth leaf in Mimulus lewisii wild-type LF10 (n = 18), flayed1 (n = 10), flayed2 (n = 12) and the double mutant (n = 12) (mean±SD).

Wild - Table S4 . qRT-PCR Primers used in this study.

CGAGAATGAGGTCGCAAACTCA AGCTTCAGCTTCGGAGGCTGAA

GGACGACATTGATGACACTGA AGGCTCGTTTATCTGCTCAACA

CGCAGCTCAGATATGCATGGAA TGTCTCTTACAGCATGCCTGTC

GCGTGTTGTACACTGATAGCGA CTTTGTGTCGTCGCTATTCATCC

GCTCAAAGACGTGCTCATTGG TGACCTCATGGTTGGTGTTCAT

ACGAACAGAGCCACCGCAGAA GGAGAAGAATCGATCCTCATCA

GACAAACAGAGCTACATCTGCA TCAATCCTGAAATGTGGACCTT

GGCTTGGACTCTGCAGTCTGT TCTTCGGCATGGCAGCAAGTC 

Genes involved in organ separation in Arabidopsis: an analysis of the cup-shaped cotyledon mutant

microRNA-directed phasing during trans-acting siRNA biogenesis in plants

A two-hit trigger for siRNA biogenesis in plants

Model for the regulation of Arabidopsis thaliana leaf margin development

Development of the perianth in Vinca rosea L

Arabidopsis JAGGED LATERAL ORGANS is expressed in boundaries and coordinates KNOX and PIN activity

Small RNAs -secrets and surprises of the genome

ORGAN BOUNDARY1 defines a gene expressed at the junction between the shoot apical meristem and lateral organs

A dominant-negative actin mutation alters corolla tube width and pollinator visitation in Mimulus lewisii

ragged seedling2 encodes an ARGONAUTE7-like protein required for mediolateral expansion, but not dorsiventrality, of maize leaves

Gateway-compatible vectors for plant functional genomics and proteomics

Fusion within and between whorls of floral organs in Galipeinae (Rutaceae): structural features and evolutionary implications

Evolutionary diversification of the flowers in angiosperms

Sympetaly--a systematic character

Distribution of the character states "early sympetaly" and "late sympetaly" within the "Sympetalae Tetracyclicae" and presumably allied groups

Synopsis of some important, non-DNA character states in the asterids with special reference to sympetaly

Regulation of AUXIN RESPONSE FACTOR3 by TAS3 ta-siRNA affects developmental timing and patterning in Arabidopsis

The Relationship between auxin transport and maize branching

Specification of leaf polarity in Arabidopsis via the trans-acting siRNA pathway

Morphological studies in the family convolvulaceae

The genetic architecture of natural variation in flower morphology

Arabidopsis KANADI1 acts as a transcriptional repressor by interacting with a specific cis-element and regulates auxin biosynthesis, transport, and signaling in opposition to HD-ZIPIII Factors

Trans-acting siRNA-mediated repression of ETTIN and ARF4 regulates heteroblasty in Arabidopsis

The FEATHERED gene is required for polarity establishment in lateral organs especially flowers of the Japanese morning glory (Ipomoea nil)

Development of leaves and flowers in the wild-type and pleiotropic maple-willow mutant of Japanese morning glory (Ipomoea nil)

Structure and development of the perianth in Downingia bacigalupii

TCP transcription factors control the morphology of shoot lateral organs via negative regulation of the expression of boundary-specific genes in Arabidopsis

The Sussex signal: insights into leaf dorsiventrality

Molecular basis of overdominance at a flower color locus

The abiotic and biotic drivers of rapid diversification in Andean bellflowers (Campanulaceae)

MicroRNA regulation of the CUC genes is required for boundary size control in Arabidopsis meristems

MicroRNA regulation of NAC-domain targets is required for proper formation and separation of adjacent embryonic, vegetative, and floral organs

Nectar bat stows huge tongue in its rib cage

The small interfering RNA production pathway is required for shoot meristem initiation in rice

The leaf adaxial-abaxial boundary and lamina growth

Developmental anatomy of foliage leaves, bracts, calyx and corolla in Pharbitis nil

Corolla tube formation in four species of Solanaceae

Corolla tube formation in the Primulaceae and Ericales

Corolla tube formation in the Tubiflorae and Gentianales

Two small regulatory RNAs establish opposing fates of a developmental axis

Morphological changes in Ipomoea nil using chimeric repressors of Arabidopsis TCP3 and TCP5

Comprehensive analysis of NAC family genes in Oryza sativa and Arabidopsis thaliana

Induced mutations affecting pollinator choice in Mimulus lewisii (Phrymaceae)

Out of Africa: evidence of the obligate mutualism between long corolla tubed plant and long-tongued fly in the Himalayas

SGS3 and SGS2/SDE1/RDR6 are required for juvenile development and the production of transacting siRNAs in Arabidopsis

Formation and shaping of the Antirrhinum flower through modulation of the CUP boundary gene

An R2R3-MYB transcription factor regulates carotenoid pigmentation in Mimulus lewisii flowers

Auxin-induced modulation of ETTIN activity orchestrates gene expression in Arabidopsis

Floral organogenesis in Antirrhinum majus

The no apical meristem gene of Petunia is required for pattern formation in embryos and flowers and is expressed at meristem and primordia boundaries

Adaptation in flower form: a comparative evodevo approach

A tetratricopeptide repeat protein regulates carotenoid biosynthesis and chromoplast development in monkeyflowers (Mimulus)

STENOFOLIA regulates blade outgrowth and leaf vascular patterning in Medicago truncatula and Nicotiana sylvestris

Leaf development. The Arabidopsis Book

Differential recruitment of WOX transcription factors for lateral development and organ fusion in Petunia and Arabidopsis

Fusion events during floral morphogenesis

A temporal and morphological framework for flower development in Antirrhinum majus

The emergence, evolution, and diversification of the miR390-TAS3-ARF pathway in land plants

The REDUCED LEAFLET genes encode key components of the trans-acting small interfering RNA pathway and regulate compound leaf and flower development in Lotus japonicus

Failure of the tomato trans-acting short interfering RNA program to regulate AUXIN RESPONSE FACTOR3 and ARF4 underlies the wiry leaf syndrome

A pathway for the biogenesis of trans-acting siRNAs in Arabidopsis

Genetic dissection of a major anthocyanin QTL contributing to pollinator-mediated reproductive isolation between sister species of Mimulus

Bulk segregant analysis of an induced floral mutant identifies a MIXTA-like R2R3 MYB controlling nectar guide formation in Mimulus lewisii

Transcriptional control of floral anthocyanin pigmentation in monkeyflowers (Mimulus)

Competition between anthocyanin and flavonol biosynthesis produces spatial pattern variation of floral pigments between Mimulus species

Organ boundary NAC-domain transcription factors are implicated in the evolution of petal fusion

Bridging the gaps: evolution and development of perianth fusion

We are grateful to Dr. Toby Bradshaw (University of Washington) for encouragement and initial support for generating the bulk segregant data in his laboratory. We thank Clinton Morse, Matt Opel, and Adam Histen for plant care in the UConn EEB Research Greenhouses. The