miR167 limits anther growth to potentiate anther dehiscence

Release of pollen from anthers is essential for sexual reproduction of flowering plants. The diurnal and developmental timing of anther dehiscence affects the mode of pollination and the degree of selfing versus outcrossing. Timing of anther dehiscence is also important for pollination and seed production in breeding programs and agriculture. Anther dehiscence often occurs when flowers open at anthesis, at flower stage 13 in Arabidopsis thaliana (Smyth et al., 1990). Before anther dehiscence, cells in the septum between locules die and break down, lignin is deposited in cell walls surrounding the endothecium cell layer just beneath the epidermis, pectin in cell walls of the stomium breaks down and the anther desiccates. These processes lead to rupture of the stomium, anther opening and pollen release (Wilson et al., 2011; Nelson et al., 2012).

In Arabidopsis, a network of hormone signaling pathways regulates the timing of flower maturation and anther dehiscence. Gibberellins and the redundantly acting auxin response transcription factors ARF6 and ARF8 are each required for production of jasmonate hormones in flowers at stages 11-12, shortly before flowers open (Nagpal et al., 2005; Cheng et al., 2009; Tabata et al., 2010; Reeves et al., 2012). Jasmonates, in turn, activate genes encoding the transcription factors MYB21 and MYB24, which then stimulate anther dehiscence (Mandaokar et al., 2006; Cheng et al., 2009; Reeves et al., 2012). In addition, MYB21 and MYB24 proteins can interact with MYC bHLH transcription factors, which mediate the jasmonate response, and with JAZ repressors of jasmonate signaling, suggesting that jasmonate may also regulate the transcriptional activation activity of MYB21 and MYB24 (Song et al., 2011; Qi et al., 2015; Huang et al., 2017). Anthers of mutants deficient in these components arrest at flower stage 12 just prior to anther dehiscence, with absent or delayed stomium rupture (Feys et al., 1994; Sanders et al., 2000; Ishiguro et al., 2001; Park et al., 2002; von Malek et al., 2002; Nagpal et al., 2005; Mandaokar et al., 2006; Reeves et al., 2012). In Arabidopsis, ARF6 and ARF8 are also required for female fertility, whereas jasmonate, MYB21 and MYB24 only regulate stamen and petal development. In tomato, ARF6 and ARF8 are similarly important for gynoecium, stamen and petal development (Liu et al., 2014), whereas jasmonates and SlMYB21 regulate gynoecium development but not stamen development (Li et al., 2004; Schubert et al., 2019). In developing Arabidopsis flowers, ARF6 and ARF8 are expressed in stamen filaments, where jasmonate is predominantly produced (Ishiguro et al., 2001; Wu et al., 2006). Jasmonates can move to other cells, and jasmonate response in the anther epidermis is sufficient for anther dehiscence (Jewell and Browse, 2016). After flowers open, jasmonate and MYB21 levels decrease through a negative feedback (Nagpal et al., 2005; Reeves et al., 2012).

Although ARF6 and ARF8 are required to initiate this pathway for jasmonate production and anther dehiscence, they also need to be inhibited for anthers to dehisce. Loss of negative regulation by the microRNA miR167 by mutation of the miR167 target site in either ARF6 or ARF8 causes both male and female sterility, due to ectopic expression of ARF6 or ARF8 in anthers and ovules (Wu et al., 2006). Stomium rupture does not occur in anthers of strong mARF6 plants. Moreover, anthers of mARF6 and mARF8 plants are larger than anthers of wild-type, jasmonate-deficient or myb21 myb24 double mutants. mARF6 anthers have ectopic expression of ARF6 in cells surrounding the anther vasculature, where miR167 is present in wild-type anthers, and this expanded expression domain likely underlies the anther phenotype (Válóczi et al., 2006; Wu et al., 2006; Rubio-Somoza and Weigel, 2013).

These studies have shown that anther dehiscence occurring at flower stage 13 requires correct spatial regulation of ARF6 and ARF8. In addition, anthers of auxin receptor mutants with decreased auxin response, or of mutants with altered auxin transport, dehisced precociously, suggesting that auxin response normally inhibits anther dehiscence at earlier developmental stages (Cecchetti et al., 2008, 2013, 2015). Excess auxin production in rice anthers also delays dehiscence (Song et al., 2018). The early dehiscence of Arabidopsis auxin pathway mutants was attributed to a combination of precocious lignification and jasmonate production in auxin-resistant mutant anthers (Cecchetti et al., 2013). The precocious anther dehiscence of these mutants contrasts with the delayed dehiscence of arf6 arf8 double mutants, and suggests that distinct auxin responses may regulate anther dehiscence differently. In fact, auxin receptor mutants had decreased auxin response in the middle layer of the anther wall, whereas ARF6 and ARF8 are expressed in stamen filaments but not in cells that surround anther locules (Wu et al., 2006; Cecchetti et al., 2016). Thus, auxin response in different cells of the stamen may cause contrasting overall effects on anther dehiscence.

Here, we have engineered plants with mutations in the MIR167a gene encoding pre-miR167, and compared anther growth and morphology among wild-type, mir167a and mARF6/8 plants. Analyses of gene expression in wild-type, mARF6 and mARF8 stamens suggest that miR167-mediated clearance of ARF6 and ARF8 from anthers limits expression of growth-promoting genes at flower stages 11-12, and thereby enables anther dehiscence at stage 13. Manipulations of hydration status suggest that changes in miR167 function do not affect competence to dehisce. Rather, anther size affects the duration of anther desiccation, which then determines the timing of dehiscence.

mARF6 and mARF8 constructs have genomic ARF6 or ARF8 genes with mutations at the target site of miR167. Different transgenic plant lines carrying either of these constructs had a range of phenotypes, and were both male- and female-sterile (Wu et al., 2006). We screened additional transformants, and identified one mARF6 and two mARF8 lines that had delayed anther dehiscence (Fig. 1) and reduced fecundity, but that produced a small number of seeds after self-pollination of hemizygotes, or after outcrossing pollen from hemizygotes to a wild-type recipient. Homozygotes, obtained from self-pollinated hemizygotes, released a small amount of pollen only in quite old flowers, and were also entirely female-sterile due to defective ovule integument growth (Fig. 2). These lines segregated as single-locus lines in progeny of crosses. Stamens from mARF6 and mARF8 stage 11 flower buds had higher levels of ARF6 or ARF8 mRNA than did wild-type stamens (Fig. 1A). As these lines could be propagated, they allowed more extensive analyses of phenotypes than was achieved previously using T1 plants (Wu et al., 2006).

Four MIR167 precursor genes encode miR167 in Arabidopsis. Analyses of promoter:GFP:GUS fusion plants have revealed that stamens express both MIR167a and MIR167c (Wu et al., 2006; Rubio-Somoza and Weigel, 2013). P_MIR167a:GUS anthers had X-Gluc staining in the vicinity of the anther vasculature at flower stages 10-12, and then throughout the anther starting at late stage 12, approximately one flower before the first open flower (Fig. 3A-E, Fig. S1A). In addition to the previously reported expression in stage 13 stamen filaments, P_MIR167c:GUS anthers had X-Gluc staining throughout young anthers until early stage 11, approximately six flowers before the first fully open flower, after which anther staining disappeared (Fig. S1G).

Strong P_MIR167a:GUS expression in stage 11 and 12 anthers suggested that the MIR167a precursor gene might produce the miR167 that represses ARF6 and ARF8 in anthers. We generated plants with mutations in MIR167a using CRISPR/Cas9-based genome engineering to target a site within the 21 nt miR167 sequence (Fig. 1K). Among the T2 plants derived from self-fertilization of T1 transformants that had carried the genome engineering construct, several had flower phenotypes resembling those of mARF6 or mARF8 plants (Fig. 1G). DNA sequencing of the MIR167a gene from five of these revealed that four plants were transheterozygous for two different mir167a alleles, and one plant was homozygous for a single allele (Fig. 1K). We recovered seed from two of these plants and identified progeny that were homozygous for either mir167a-1 (29 bp deletion including 14 bases of the 5′ end of mature miR167), mir167a-2 (1 bp insertion of a G after position 15 of miR167) or mir167a-8 (14 bp deletion including 7 bases at the 3′ end of mature miR167). qRT-PCR assays indicated that both the mir167a-1 and mir167a-2 mutant stamens had very low levels of mature miR167, although ARF6 and ARF8 transcript levels appeared normal, possibly reflecting high baseline expression in wild-type filaments (Fig. 1B).

mir167a mutants had delayed anther dehiscence (Fig. 1G) and produced no or very few seeds, either after manual self-pollination or in crosses with wild-type plants as either the male or female parent. Similarly to mARF6 and mARF8, many of the seeds that mir167a plants produced were inviable. mARF6-A/− hemizygotes, mARF8-3/− hemizygotes and mir167a mutant plants had a mixture of ovules with short integuments similar to those of mARF6 and mARF8 homozygotes, and ovules with normal integuments (Fig. 2B-F), which may have given rise to the viable seed. F1 heterozygous mir167a-1/+ and mir167a-2/+ plants were fully fertile, and produced F2 progeny that segregated into roughly 25% presumed mir167a homozygotes with largely sterile flowers, consistent with a single locus being responsible for the phenotype (wt:mutant numbers for mir167a-1, 174:45 and for mir167a-2, 118:41; data pooled from several crosses).

The very similar phenotypes of mir167a, mARF6 and mARF8 plants indicate that MIR167a is the major miR167 precursor gene regulating ARF6 and ARF8 in flowers. Although mARF6 and mARF8 plants were difficult to propagate because of poor seed viability, the stronger phenotypes of mARF6 and mARF8 homozygotes compared with mir167a mutants gave more definitive results in some experiments. Moreover, partially dominant mARF6 or mARF8 lines were convenient for experiments with reporter genes using F1 plants.

Wild-type anthers had a projected area of ∼0.15 mm² viewed from the adaxial side at late flower stage 11 or early stage 12 (gynoecium length 1.3-1.5 mm, approximately 4 flowers before opening), and grew slightly up through stage 12 just before flowers opened, when gynoecia were ∼2 mm long (Fig. 4A,B). During stage 12, stamen filaments typically grow much faster, roughly doubling in length (Smyth et al., 1990; Reeves et al., 2012). At stage 13 when flowers opened, wild-type anthers began to shrink, and they dehisced to release pollen (Fig. 4A-C).

In mARF6 and mARF8 flowers, petal and stamen filament growth proceeded normally through flower maturation and opening (Fig. S2A,B), but anthers grew too large at flower stages 11-12 (Smyth et al., 1990). At stage 11, approximately 5-6 flowers before the first open flower, mARF6 and mARF8 anthers developed a broader cap of cells at the apical end than did wild-type anthers, and they became measurably larger than wild-type anthers starting at approximately 4 flowers before the first open flower, when wild-type anther growth slowed (Fig. 4A,B, Fig. S3A,B,C). Hemizygous mARF6/− and mARF8/− anthers grew to be up to 25% larger in projected area than did wild-type anthers, shrank somewhat starting at the time of flower opening and dehisced 2 to 3 flowers later than did wild-type anthers, although they generally opened less widely and released less pollen than did wild-type anthers (Fig. 4A,C). Homozygous mARF6 and mARF8 anthers also grew larger than did wild-type anthers, shrank less after flower opening and usually failed to dehisce (Fig. 4A-C, Fig. S3A-C). mir167a mutant plants had enlarged and indehiscent anthers similar to those of mARF6 and mARF8 plants (Fig. 4A,G, Fig. S3A). Anther dehiscence in mir167a mutants occurred later than it did in mARF6/− or mARF8/− hemizygotes, but earlier than in mARF6 or mARF8 homozygotes (Fig. 4C, Fig. S2C). As a result of the excess growth of locules and of cells at the tip, the anthers of mir167a, mARF6 and mARF8 plants developed a spade-shaped appearance.

Epidermal cells of mir167a, mARF6, and mARF8 anthers were visibly larger than those in wild-type anthers, especially in the area close to the junction with the filament (Fig. 4D-F). In transverse cross-sections, mARF8 and mir167a anthers from the first open flower (stage 13) had similar cellular organization to wild-type anthers, but had larger connective, epidermis and endothecium cells, and were larger than wild-type anthers across the lateral and adaxial-abaxial axes, similarly to mARF6 anthers characterized previously (Fig. 5A-D) (Wu et al., 2006). In some mir167a-1 and mARF8-3 anther locules, the septum had degraded and the stomium was broken, although the anther locules had not opened to release pollen as they did in wild-type anthers (Fig. 5B,D). In other mutant anthers, the septum was incompletely degraded and/or the stomium was still connected (Fig. 5C). Often, different sections from the same anther revealed different degrees of septum or stomium breakage, implying that only part of the septum and stomium had broken. Similarly, in images taken by environmental scanning electron microscopy (eSEM) (Fig. 5E-L), stage 13 and 14 mARF8 anthers sometimes appeared partially open, with a slit along part of the dehiscence zone (Fig. 5G,K). In addition, some mARF8 anthers had pollen grains that had germinated within the anther locule (Fig. 5C,D, arrows).

Pollen from mir167a, mARF6 or mARF8 plants had apparently normal exterior morphology by eSEM (Fig. S4A-D). As assessed by fluorescein diacetate staining, pollen from the first open flower of mir167a, mARF6 or mARF8 plants (obtained by cutting open the anthers) had similar viability to pollen from wild-type flowers, whereas pollen from the fifth open flower was somewhat less viable than pollen from the fifth open wild-type flower (Fig. S4E). Moreover, mir167a and mARF8 pollen germinated in vitro at a ∼20% lower rate than did wild-type pollen (Fig. S4F). The decreased pollen performance of these genotypes may contribute to their decreased pollination efficiency.

Consistent with their increased and persistent growth, mARF6/− and mARF8/− anthers had increased signal of a phloem unloading tracer, P_SUC2:YPET, which is expressed in phloem companion cells of source tissues and delivered to sink tissues such as anthers (Truernit and Sauer, 1995; Imlau et al., 1999; Stadler et al., 2005a,b). In wild-type anthers, P_SUC2:YPET fluorescence appeared in cells surrounding the vasculature in the center of the anther up until the third open flower, but then decreased starting after flower +3. (Fig. 6A,B,F). In mARF6/− and mARF8/− anthers, P_SUC2:YPET signal in cells surrounding the anther vasculature continued up to approximately flower +6 (Fig. 6C-E). In addition, the phloem domain was longer in mARF6/− and mARF8/− anthers than in wild-type anthers starting at flower −4 and continuing up until flower +6 (Fig. 6F). Thus, it appears that phloem content was delivered into mARF6/− and mARF8/− anthers over a larger area and until a later stage than occurs in the wild type. This phenotype may arise simply because the mARF anthers grow larger, but could also contribute to their increased growth.

In wild-type, mARF6 and mARF8 anthers, phloroglucinol stained a stripe of lignin near the future dehiscence zone on each half of the anther, similar in extent in each genotype (Fig. 6I-K) (Mizuno et al., 2007; Cecchetti et al., 2016). Appearance of the lignification zone in mARF6 and mARF8 anthers suggests that their delayed dehiscence does not arise from a failure of lignin deposition as in myb26 mutants (Yang et al., 2007). In addition, differentiated stomata were visible on the abaxial side of mARF6, mARF8 and mir167a anthers, as in wild-type anthers (Fig. 4D-G, Fig. 6G,H). Thus, mARF6 and mARF8 anthers differ from wild-type anthers mainly in size rather than in other anatomical features that can affect dehiscence.

We assayed global gene expression patterns by RNA-Seq in wild-type and mARF8 stamens from flower stage 11, when the excess growth begins. The mARF8 samples came from both stable mARF8 lines characterized here, and from a pooled sample from T1 mARF8 plants that were not recovered as stable lines (Table S1). In addition, we included samples from stamens of P_RPS5a>MIR167a plants overexpressing MIR167a. These plants resemble arf6 arf8 double mutant plants (Reed et al., 2018), and reveal effects of loss of the redundantly acting ARF6 and ARF8.

Using DESeq2 software, we identified genes whose expression was statistically significantly different between wild-type and P_RPS5a>MIR167a stamens, or between wild-type and mARF8 stamens, and that also had at least two-fold difference in expression levels (Fig. 7A, Tables S2-S4). In P_RPS5a>MIR167a stamens, more genes were underexpressed (969) than were overexpressed (292) (Table S2), consistent with previous microarray results of arf6 arf8 whole flowers (Nagpal et al., 2005; Reeves et al., 2012), and with the finding that ARF6 and ARF8 activate gene expression (Ulmasov et al., 1999). Of 624 genes previously found to be underexpressed in microarray analyses of arf6-2 arf8-3 stage 12 flower buds (Reeves et al., 2012), 177 (28%) were also underexpressed in the stage 11 P_RPS5a>MIR167a stamens studied here (Table S2). Considering that the previous experiment was based on whole flowers and at a slightly later stage than stamens of the current experiment, the overlap suggests good concordance of the datasets. Genes overexpressed in the ARF-deficient flowers or stamens in both datasets had less overlap (29 in common of 312 genes from the Reeves et al., 2012 dataset), suggesting that more of the overexpressed genes may be affected indirectly or in other organs.

mARF8 stamens overexpressed 1105 genes, 102 of which were underexpressed in P_RPS5a>MIR167a stamens (Tables S3 and S4). mARF8 stamens underexpressed 1686 genes, suggesting either that ectopic ARF8 also represses genes directly, or that many of the underexpressed genes are affected secondarily. qRT-PCR assays confirmed that LOX2/At3g45140, SAUR42/At2g28085, At5g50335, XTH19/At4g30290 and AXR2/IAA7/At3g23050 were each overexpressed in mir167a, hemizygous or homozygous mARF6 and/or mARF8 stamens at flower stage 11 (Fig. 7B,C, Fig. S5). In addition, MYB26/At3g13890 was overexpressed in mARF6 and mARF8 stamens but not in mir167a stamens, although the expression level of MYB26 in our experiment was generally very low and the results may therefore be sensitive to small differences in sample staging. In contrast, EXPA8/At2g40610 was not overexpressed in the tested genotypes in most experiments. However, as described below, a GUS fusion to the promoter of EXPA8/At2g40610 did have an expanded expression domain in mARF6/− anthers.

We surveyed groups of genes with different expression characteristics for enrichment of gene ontology (GO) categories, protein domains or Kyoto Encyclpedia of Genes and Genomes (KEGG) pathways (Tables S5-S10). The 102 genes that were underexpressed in P_RPS5a>MIR167a but overexpressed in mARF8 stamens were enriched for the GO categories response to jasmonate (including jasmonate biosynthesis genes) and cell wall loosening (Table S6). Some of these genes may be direct ARF6 and ARF8 targets. In fact, 20 of these genes were designated as ARF6 targets in seedlings based on expression and ChIP-Seq data (Oh et al., 2014) (Table S6). arf6 arf8 flowers underproduce jasmonates (Nagpal et al., 2005; Tabata et al., 2010; Reeves et al., 2012), and these results suggest that conversely, mARF8 stamens may produce excess jasmonate. Cell wall loosening functions that may contribute to excess growth in mARF8 anthers were represented by expansin genes (Cosgrove, 2016). Enriched protein domains included both expansins and a family of DUF642 proteins that are secreted to the apoplast and may be involved in pectin metabolism (Zúñiga-Sánchez et al., 2014; Salazar-Iribe et al., 2016).

The full set of 1105 genes that were overexpressed in mARF8 stamens were similarly enriched for jasmonate pathway and cell wall loosening functions, and for fatty acid biosynthesis proteins and pathway, and the KEGG pathway term phenylpropanoid biosynthesis (Table S5). Overexpression of MYB26/At3g13890 in mARF8 stamens may contribute to the increased phenylpropanoid pathway gene expression (Table S5, Fig. 7B, Fig. S5). The transcription factor MYB26 activates the lignin biosynthesis pathway required to make the lignification zone in anthers (Yang et al., 2007). The jasmonate pathway is normally expressed at stages 11-12, but is then switched off in mature open flowers (Nagpal et al., 2005; Reeves et al., 2012). Overexpression of many of these genes in mARF8 anthers may reflect persistent growth or delayed desiccation.

Enriched GO terms among genes that were underexpressed in both P_RPS5a>MIR167a and mARF8 stamens included pollen tube growth and cation transport (Table S9). Enriched protein domains included pectin esterase inhibitors, RALF (rapid alkalinization factor) peptides, H⁺/K⁺ exchange transporters, and universal stress protein A domain (UspA) kinases (Kerk et al., 2003; Sze et al., 2004; Murphy and De Smet, 2014). Many of these are expressed in mature pollen or growing pollen tubes (Honys and Twell, 2004; Verelst et al., 2007a; Wang et al., 2008; Ge et al., 2017; Mecchia et al., 2017). In fact, 200 of the 411 genes that were underexpressed in both P_RPS5a>MIR167a and mARF8 stamens were induced in tricellular and mature pollen (Honys and Twell, 2004; Verelst et al., 2007a) (Table S4). As that study used a microarray that has spots for approximately two-thirds of Arabidopsis genes, it seems likely that a majority of the genes underexpressed in both P_RPS5a>MIR167a and mARF8 are normally expressed in mature pollen. mARF8 stamens underexpressed AGL65, AGL66 and AGL104, which encode MADS-domain transcription factors that are required for late stages of pollen maturation (Verelst et al., 2007b). Their underexpression in P_RPS5a>MIR167a and mARF8 stage 11 stamens suggests that pollen maturation or desiccation might be delayed in these stamens, and may have contributed to decreased mARF8 pollination efficiency.

Both EXPA8/At2g40610, encoding an expansin protein, and XTH19/At4g30290, encoding a xyloglucan transglycosidase/hydrolase, were overexpressed in mARF8 stamens based on our RNA-Seq data, and their products may contribute to cell wall loosening and growth. EXPA8/At2g40610 is auxin regulated and considered to be a direct target of ARF6 in seedlings (Oh et al., 2014). We assessed expression of available P_EXPA8:GUS and P_XTH19:GUS fusions to the promoters of these two genes (Bassel et al., 2014) in F1 plants. Both had expanded expression domains in hemizygous mARF6/− or mARF8/− anthers. Wild-type P_EXPA8:GUS plants had X-Gluc staining in stamen filaments starting at stage 13, but did not have appreciable staining in anthers (Fig. 7D,E). In mARF6/− plants, staining appeared in anthers at stage 12 and persisted at stage 13 (Fig. 7F,G). The staining in mARF6/− anthers was confined to the central zone surrounding the vasculature and extended to the swollen tips of mARF6/− anthers. In wild-type P_XTH19:GUS stamens, X-Gluc staining was first visible in filaments starting at stage 12, and became stronger at stage 13; but was not detectable in anthers (Fig. 7H,I). In both mARF6/− and mARF8/− hemizygotes, staining was present in stage 12 and stage 13 anthers, with the strongest signal in cells surrounding the vasculature at the base of the anther (Fig. 7J-M).

Expanded expression domains of these reporters in mARF6/− or mARF8/− anthers could arise from excess ARF6 or ARF8 activity in the cells surrounding the anther vasculature. Visible X-Gluc stain in each case appeared at stage 12, which is slightly later than the action of miR167 at stage 11.

The auxin response reporter gene DR5:GUS is normally expressed in wild-type anthers primarily in the middle layer interior to the endothecium at flower stage 10, and is then turned off at approximately stage 11 (Aloni et al., 2006; Cecchetti et al., 2016) (Fig. S6A,H-J). In contrast, in anthers of mARF6 and mARF8 plants DR5:GUS expression persisted until well after flowers opened (Fig. S6B-E,K-M). Expression was generally stronger and persisted for longer in homozygotes than in hemizygotes. This expression was not confined to perivascular cells where ectopic ARF6 and ARF8 are likely to be present, but appeared to be in cells around the anther locules, suggesting indirect regulation.

To test whether ARF6 and ARF8 are necessary for DR5:GUS expression at earlier stages, we transformed the DR5:GUS line with a P_35S:MIR167a transgene that causes overproduction of miR167 and therefore reduces ARF6 and ARF8 mRNA levels (Wu et al., 2006). Inflorescences of several transformants with phenotypes resembling those of strong arf6 arf8 plants were stained with X-Gluc. In each case, significant staining was present in anthers (Fig. S6F,G). These results indicate that ARF6 and ARF8 are not required for DR5:GUS expression in anthers.

When mARF6, mARF8 and mir167a inflorescences were cut off the plant and allowed to desiccate, anthers of stage 13 flowers released pollen (Fig. 8A,B), in contrast to the indehiscent anthers of well-watered intact plants. This observation suggests that delayed or failed dehiscence of mARF6, mARF8 and mir167a anthers might be a consequence of excess water content arising from persistent growth. To test this model, we excised indehiscent stamens from flowers at defined developmental stages, and incubated them in air without a water source. Wild-type anthers from stage 12 unopened flowers usually dehisced to release pollen, whereas anthers from stage 11 flowers generally did not dehisce (Fig. 8D,E). mARF6, mARF8 and mir167a anthers also dehisced in these experiments if they came from flower stage 12 or later (Fig. 8D,E, Fig. S7A,B). Dehiscence after stamen excision usually occurred within 2 h. Pollen obtained from desiccated homozygous mARF6-A anthers was used successfully to fertilize wild-type gynoecia, indicating that at least a portion of it was viable. These results indicate that miR167 action accelerates dehiscence but is not required to establish competence to dehisce in anthers of stage 12 flowers.

To create conditions of high humidity, we placed cut inflorescences in agar on vertically oriented Petri plates sealed against water loss. Many water droplets condensed on the plate lid, indicating that humidity in the air space was close to saturation. Inflorescences grew for at least 2 days under these conditions, and flower buds opened. However, even wild-type anthers did not dehisce while kept in high humidity (Fig. 8C). Once inflorescences or excised anthers were removed to a dry environment, the anthers shrank and dehisced (Fig. 8F-J). Similarly, mir167a stamens subjected to the vacuum used for eSEM imaging without mounting on a cooled (−20°C) stage dehisced within minutes (Fig. 8K,L). We compared the speed of dehiscence of anthers of matched developmental stage upon a shift from high to low humidity, after excising stamens from stage 13 flowers that had opened under high humidity without anther dehiscence. Wild-type anthers generally dehisced within 10 min of the shift, whereas anthers from mir167a, mARF6 or mARF8 plants usually dehisced after more than 30 min (Fig. 8M, Fig. S8A). Anthers of the miR167 pathway mutants lost significant volume from their maximal size before desiccation, but were still larger than wild-type anthers at the time they dehisced (Fig. S8B,C). They also generally opened less widely and released less pollen than did wild-type anthers (Fig. 8H,J).

In an attempt to accelerate desiccation without excising stamens, we removed the perianth from stage 12 flower buds on otherwise intact wild-type plants. These anthers dehisced at the normal flower stage when the perianth would have opened (Fig. S7C,D), suggesting that under this condition anthers desiccate less than if excised, and that flower opening per se is not sufficient to trigger dehiscence.

ARF6 and ARF8 promote jasmonate production at flower stages 11-12, and this burst of jasmonate acts through MYB21 and MYB24 transcription factors to promote filament growth and anther dehiscence. To determine whether this pathway regulates miR167 production, we assayed P_MIR167a:GUS expression in stamens of aos-2 (jasmonate-deficient), coi1-1 (jasmonate-insensitive), myb21 myb24 and arf6-2 arf8-8 mutant plants (Xie et al., 1998; Reeves et al., 2012; Mlotshwa et al., 2016). Although these flowers lacked jasmonate production or response, and had anthers that arrested at flower stage 11, they each had significant X-Gluc staining in anthers (Fig. 3F,G, Fig. S1B-F). Moreover, qRT-PCR assays showed that miR167 was present in aos-2 stamens (Fig. S1H). These results indicate that the jasmonate-MYB21 pathway is not required for P_MIR167a:GUS expression and miR167 production in anthers of flower stages 11 and 12.

Anthers of the aos-2 jasmonate biosynthesis mutant did not normally dehisce (Park et al., 2002; von Malek et al., 2002; Reeves et al., 2012) (Fig. 5H,L, Fig. S3A). In some experiments, aos-2 anthers were smaller than wild-type indehiscent anthers, perhaps because of shrinkage after stage 13 without dehiscence (Fig. S3A, Fig. 5H,L, Figs S7E and S8D). aos-2 anthers became competent to dehisce after stamen excision at a later flower stage than did anthers of the other genotypes tested (Fig. 8D,E, Fig. S7F) and they took longer to dehisce than did wild-type anthers after excision from flowers that had opened under high humidity (Fig. 8M, Fig. S8C). aos-2 mir167a, aos-2 mARF6 and aos-2 mARF8 flowers had short stamen filaments similar to those of aos-2 plants; and enlarged anthers similar to mir167a, mARF6 or mARF8 anthers (Figs S3A, S8D,E and S9). aos-2 mARF8-3/− anthers usually failed to dehisce even after desiccation (Fig. S7E,F). Lastly, exogenous methyl jasmonate treatment caused aos-2 and aos-2 mir167a-1 stamen filaments to elongate and elicited dehiscence of aos-2 anthers, but did not cause aos-2 mir167a-1 anthers to dehisce (Fig. S9E-G). Together these results suggest that the anther dehiscence defect caused by loss of mir167 regulation is independent of jasmonate production and response in filaments, although it remains possible that jasmonate response is perturbed in the epidermis of mir167a mutant anthers.

The close similarity among mir167a, mARF6 and mARF8 phenotypes indicates that MIR167a is the main miR167 precursor gene targeting ARF6 and ARF8 transcripts in both anthers and ovules. Consistent with this view, P_MIR167a:GUS is expressed surrounding the anther vasculature at flower stages 11-12 and in developing ovules. Based on expression patterns, three other MIR167 genes may have functions in other tissues, such as sepals or roots (Wu et al., 2006; Gifford et al., 2008; Medina et al., 2017).

The mir167a mutant anther and fertility defects appeared slightly stronger than those of the hemizygous mARF6/- or mARF8/- plants characterized here, but weaker than those of homozygous mARF6 or mARF8 plants. In fact, we detected increased ARF6 or ARF8 transcript levels in mARF6 or mARF8 stamens, but not in mir167 mutant stamens. These observations suggest that mARF transgene copy number or position effects caused higher levels of ARF6 or ARF8 expression than did mutating MIR167a. The variable strength of different mARF6 and mARF8 lines (Wu et al., 2006) also shows that the dosage of mARF6 or mARF8 can affect the timing of anther dehiscence.

In mARF6, mARF8 and mir167a anthers, ectopic ARF6 and ARF8 promote excess cell expansion growth starting at flower stage 11, ∼2 days before flowers open at stage 13 and wild-type anthers normally dehisce. Ectopic ARF6 and ARF8 in cells surrounding the vasculature, where miR167 is normally present (Válóczi et al., 2006; Wu et al., 2006; Rubio-Somoza and Weigel, 2013), likely activate EXPA8, XTH19 and other genes whose products loosen cell walls and thereby increase expansion of perivascular cells.

ARF6 and ARF8 similarly promote cell expansion in hypocotyls, inflorescence stems, petals, stamen filaments and stigma (Nagpal et al., 2005; Ru et al., 2006; Reeves et al., 2012; Reed et al., 2018), and many of the differentially expressed genes in arf6 arf8 and mARF8 stamens are also affected by arf6 mutations in seedlings or are auxin regulated in hypocotyls (Chapman et al., 2012; Oh et al., 2014). Despite this overlap, whereas SAUR (small auxin up RNA) genes are the dominant class of genes induced in hypocotyls by auxin or by light conditions that increase auxin production (Chapman et al., 2012; Kohnen et al., 2016; Sun et al., 2016), only one SAUR gene (SAUR42/At2g28085) was underexpressed in arf6 arf8 stamens and overexpressed in mARF8 stamens; and GO and protein domain analyses did not highlight SAUR genes in our dataset. This difference might arise if auxin response factors interact with tissue-specific factors to regulate different targets in different tissues.

Although the primary effects of ectopic ARF6 and ARF8 likely occur in cells near the central vasculature, cells throughout the anther grew excessively, and expression of the auxin-responsive reporter gene DR5:GUS persisted after flower stage 11 in mARF6 and mARF8 anthers. DR5:GUS is normally expressed in the middle cell layer surrounding anther locules, several cells distant from the location of miR167 action in cells surrounding the vasculature (Válóczi et al., 2006; Wu et al., 2006; Cecchetti et al., 2016). These phenotypes suggest that MIR167a has non-cell-autonomous effects on growth and auxin response in anthers. Ectopic mARF6 transcript was seen only in perivascular cells in anthers (Wu et al., 2006), suggesting that miR167 acts cell-autonomously. While ARF6 and ARF8 proteins might move between cells, DR5:GUS is still expressed in P_35S:MIR167a anthers deficient in both ARF6 and ARF8, indicating that additional transcription factors in anthers can activate this reporter and possibly control growth. Such factors could include other ARF proteins, or targets of brassinolide or gibberellin signaling pathways (Oh et al., 2014). Cells in sepals have variable growth rates that can compensate for each other to result in smooth overall organ growth (Hong et al., 2016), and similar compensation may underlie cell expansion throughout anthers driven by growth of cells in the center. Such effects might be mediated by tension in the tissue deriving from growth of the perivascular cells, increased water availability or other mobile signals.

mir167a and mARF8 pollen had slightly decreased viability and in vitro germination rate, similarly to pollen from inadequately desiccated ice1 mutant anthers (Wei et al., 2018). Delayed desiccation of the anther and its pollen might also delay pollen maturation, or alternatively allow pollen to bypass desiccation and dormancy, as suggested by the occasional pollen germination we observed in mARF8 anthers. Consistent with either model, many genes characteristic of late stages of pollen maturation were underexpressed in mARF8 stamens in our RNA-Seq data. It would be interesting to test directly whether anther desiccation induces late-stage pollen gene expression.

We propose that excess anther growth during flower stages 11-12 causes the subsequent dehiscence defect in miR167 pathway mutant anthers because mutant anthers become too large to desiccate efficiently. Some mutant stage 13 anthers had partially open gaps along the stomium, indicating that the cell wall breakdown pathway was likely active, and that even if partially intact at the time of anthesis, the stomium was weakened. In practice, it is difficult to pinpoint the timing of stomium breakage because protocols for microscopy specimen preparation include desiccation steps or vacuum (for eSEM) and chemical fixatives or freezing may not entirely prevent stomium breakage during processing. Several anatomical features appeared normal in the mutants, and, although likely causing faster water loss than normally occurs, stamen excision triggered miR167 pathway mutant anthers to shrink and then dehisce. These results suggest that dehydration limits anther dehiscence, and that these mutants fail to release pollen because they retain too much water.

The delay in desiccation in miR167 pathway mutants might limit production of mechanical forces to crack open the stomium and/or to evert the locule walls to release pollen (Wilson et al., 2011; Nelson et al., 2012; Wei et al., 2018). In this scenario, enlarged mutant anthers could preclude or delay generation of sufficient mechanical force, either because of delayed shrinkage to an appropriate size, or if greater forces are needed to evert the enlarged mutant locule walls than are needed in wild-type anthers. Prolonged hydration might also delay cell wall breakdown in the septum and/or stomium through regulatory pathways regulated by cytoplasmic or apoplastic water potential. Other factors needed for dehiscence such as stomatal differentiation or lignification (Yang et al., 2007; Wei et al., 2018) were apparently unaffected in miR167 pathway mutants. Changes in growth rate are also important for developmental events such as sperm release from pollen, where peptide signals from the female gametophyte induce growth imbalance leading to pollen tube rupture and sperm release (Ge et al., 2017).

Although excess prior growth is likely to be the primary cause of anther indehiscence in plants with defects in miR167 regulation, ectopic ARF6 and ARF8 might continue to act even after flowers open to promote growth and/or impede desiccation, as suggested by the persistent phloem unloading zone in mutant anthers. Increased expression of jasmonate biosynthesis genes and MYB26 should promote dehiscence, so may be symptoms of delayed desiccation.

Why do wild-type anthers normally dehisce just as flowers open? Perianth removal did not accelerate anther dehiscence in wild-type stage 12 flower buds, suggesting that flower opening per se would not cause sufficient desiccation to induce immediate anther dehiscence. Anther stomata develop at flower stage 12, and ice1 mutants lacking mature stomata have indehiscent anthers at high relative humidity (Wei et al., 2018), indicating that timing of stomatal development contributes to competence to desiccate and dehisce. Similarly, in the moss Physcomitrella patens, stomata contribute to sporophyte capsule rupture (Chater et al., 2016). However, ice1 mutant anthers did dehisce at low humidity (Wei et al., 2018), suggesting that anthers also lose water by a route that is independent of anther stomata. Water flux through vasculature may slow or even shift from import into anthers to export to other organs, or water may escape through ruptured stomia. ice1 anthers likely dehisce more readily than do mir167a and mARF6/8 anthers because they do not grow too large and so need to lose less water to open.

Jasmonate production starting at stage 11 is required to produce dehiscent anthers, and coincides with increased expression of MIR167a. However, MIR167a is expressed independently of jasmonate production or response, aos mir167a double mutants had an additive phenotype, and jasmonate could not rescue mir167a anther dehiscence. Finally, jasmonate is thought to be made in the filament and to act in the anther epidermis, whereas miR167 acts in cells surrounding the anther vasculature. These results suggest that miR167 and jasmonate act independently, although it remains possible that miR167 deficiency affects responsiveness to jasmonate specifically in anthers but not in stamen filaments. Signals and factors that initiate contemporaneous jasmonate production in filaments and MIR167a expression in anthers might include TCP transcription factors (Schommer et al., 2008; Danisman et al., 2012; Rubio-Somoza and Weigel, 2013).

All plant lines used in this work were in the Columbia (Col-0) ecotype. Most plant seeds were surface sterilized with 95% ethanol followed by bleach solution (2:1 H₂O:bleach with two or three drops of Tween-20), plated on Murashige and Skoog (MS) salts (Murashige and Skoog, 1962) containing 1% (w/v) sucrose and 0.6% (w/v) Phyto-agar (pH 5.7), cold stratified for 1 to 3 days at 4°C, then grown at 22°C with 16-h-light/8-h-dark photoperiod. mARF6, mARF8, and mir167a seeds have defective seed coats and were surface sterilized using 70% (v/v) ethanol, 0.5% (v/v) Triton X-100 (Yadav et al., 2014).

mARF6-A, mARF8-3 and mARF8-4 lines were derived using transgene constructs described previously (Wu et al., 2006). mir167a mutants were made by CRISPR/Cas9-targeted mutagenesis using the 19-mer sgRNA guide sequence AAAGCTAATTAGATCATGC (Fig. 1K). The guide RNA was cloned into pDONR207 vector (Invitrogen) using primers listed in Table S11, subcloned into binary vector pCUT6 (Peterson et al., 2016) by LR clonase (Invitrogen) and introduced into Arabidopsis plants using Agrobacterium tumefaciens strain GV3101 by floral dip (Clough, 2005). Basta-resistant T1 transformed plants were selected on soil by spraying with 0.01% Finale (Bayer Environmental Science, RTP, NC) and phenotypic plants identified in the T2 generation. Mutations were identified as having bands of altered mobility on 15% polyacrylamide gels of PCR products using primers listed in Table S11, and by sequencing PCR products spanning the targeted site. The P_MIR167a:GUS, P_MIR167c:GUS, P_EXPA8:GUS, P_XTH19:GUS, P_DR5:GUS and P_35S:MIR167a constructs and plants were described previously (Ulmasov et al., 1997; Wu et al., 2006; Bassel et al., 2014). To construct P_SUC2:YPET, a YPET open reading frame sequence with an A206K mutation to decrease protein dimerization was amplified by PCR and cloned into pENTR (Invitrogen). This was then recombined by Gateway cloning into a derivative of pB7WG2 destination vector (Karimi et al., 2002) in which the P_35S promoter sequences had been replaced with a 2.1 kb upstream promoter sequence from AtSUC2 (Truernit and Sauer, 1995). The resulting construct was transformed into wild-type Columbia plants, and a line with 3:1 segregation was crossed with mARF6/− and mARF8/− plants.

Flower X-gluc staining and sectioning of stained anthers were performed as described previously (Ploense et al., 2009). Phloroglucinol staining was based on methods described by Mizuno et al. (2007). Wild-type, mARF6 and mARF8 mutant flowers at stage 13 were fixed in 3.7% (v/v) formaldehyde, 5% (v/v) acetic acid, 50% (v/v) ethanol at 4°C for 24 h; and dehydrated with an ethanol series [50% (v/v) to 92% (v/v)]. Flowers were stained with 1% (w/v) phloroglucinol (Sigma-Aldrich) in 92% (v/v) ethanol for 5 min and moved to a glass dish, then 25% (v/v) HCl was dropped onto the flowers and stained anthers were photographed.

For microscopy of P_SUC2:YPET anthers, flowers of phloem loading tracer P_SUC2:YPET (514 nm excitation; emission collection at 535-600 nm) were washed [1× PBS, 2% (v/v) Tween-20 for 10 min, then 1× PBS] then stained with 0.0005% (w/v) ethidium bromide (560 nm excitation; emission collection at 592-643 nm) (Yang et al., 2007). A weak vacuum was pulled for 30 min, and samples were then left at room temperature for 30 min. Fluorescence was visualized on a Zeiss 710 DUO confocal microscope (Zeiss, Thornwood, NY, USA). Environmental scanning electron microscopy images of anthers were taken on a Hirox SH-3000 mini-SEM microscope (Fig. 4) or a FEI Quanta 200 eSEM (Figs 5 and 8KL, Fig. S4) at 10 kV. Samples for images shown in Figs 4, 5 and Fig. S4 were cooled to −20°C on a cooled specimen stage.

A P_35S:PM-mCherry transgene with a fusion between AtPIP2A and mCherry and a Basta-resistance selectable marker gene (Nelson et al., 2007) was transformed into Columbia plants and generously provided by Tim Ross-Elliott (University of North Carolina at Chapel Hill, USA). This line was crossed with mARF8-3, and individual F2 plants homozygous for mARF8-3 and carrying P_35S:PM-mCherry used for microscopy. For imaging, stamens were immersed in water between glass slide and coverslip separated by 20 µm spacers to prevent tissue crushing. Tiled 1024×1024 pixel images with 2 µm z-stack steps were acquired on a Leica TCS sp8 confocal microscope using excitation with a 552 nm solid-state laser and emission detection at 599-641 nm. Tiles were merged and z-stack projections generated using LAS X software (Leica Microsystems).

For anther microstructure, first open flowers were fixed in 1.6% paraformaldehyde, 2.5% glutaraldehyde, 50 mM potassium phosphate (pH 6.9), and washed and dehydrated in an ethanol series (Yeung and Chan, 2015). Tissue was embedded in Technovit 7100 resin (Electron Microscopy Sciences, Hatfield, PA) in 0.2 ml thin-walled PCR tubes, cut to 2 µm sections on a Leica JUNG RM2065 rotary microtome with glass knives, stained with 0.5% Toluidine Blue, and observed by light microscopy (Matsushima, 2014; Matsushima et al., 2014; Wei et al., 2018). Ovules were fixed, cleared and observed by differential interference contrast (DIC) microscopy, as described previously (Wu et al., 2006).

For pollen viability assays, a stock solution of 1% Fluorescein diacetate (FDA, Fluka BioChemika 31545) in acetone was diluted 1000× in 20% sucrose just before use. Pollen was released into a drop of diluted FDA by dabbing the anther on the drop on a glass slide, and a coverslip was overlaid. Pollen fluorescence indicating viable pollen was observed under a fluorescence microscope using GFP filters. Pollen germination was assayed as described (Dickinson et al., 2018) with the germination medium buffered at pH 7.5 and using cellophane membrane 95P (Futamura USA, Atlanta, GA).

Flowers and floral organs were photographed using a Leica DFC420 camera on a Wild dissecting microscope, and measured from images using ImageJ software or using a camera lucida attachment on the dissecting microscope (Abramoff et al., 2004; Reeves et al., 2012). Some panels in Figs 5B, 6E and 8K, Figs S1 and S6 are composite images of partly overlapping fields stitched together either manually or using the Automate>photomerge function in Photoshop software. Flower and anther images in desiccation experiments were obtained using a Keyence VHX-5000 microscope. Images presented in Fig. 8 are recalculated composites made by the Keyence microscope software from serial images in several consecutive focal planes.

Plants were grown in 14:10 h day:night cycles for RNA-Seq experiments or 16:8 h cycles for qRT-PCR. Samples for RNA analyses were collected between 1 and 3 h after dawn and immediately frozen in liquid nitrogen. Stage 11 flower buds were 1.5-1.8 mm long, generally corresponding to flowers −3 to −5. Stamens at this stage still appeared green before turning yellow at flower stage 12. For RNA-Seq, RNA was isolated using the Rneasy Plant Mini kit (Qiagen) according to the manufacturer's instructions. Libraries were constructed with the Truseq mRNA library prep kit (Illumina) according to the manufacturer's instructions. Samples were bar-coded and multiplexed, and sequenced using an Illumina HiSeq 2000 machine with single-end 50 base reads. Library construction and sequencing were performed at the University of North Carolina high-throughput sequencing facility.

For qRT-PCR, total RNA was extracted using Trizol (Invitrogen) as described by the manufacturer. cDNA was synthesized using the A3500 Reverse Transcription System (Promega) according to the manufacturer's instructions. For the mRNA, reverse transcription was conducted using Oligo(dT)15 Primer; for miRNA, reverse transcription used the Stem-loop RT primer (see Table S11). Total RNA of 1 µg was used in a 20 µl volume reaction. The cDNA was diluted 5-fold, and 1 µl was used as a template in a 10 µl PCR reaction. The real-time quantitative RT-PCR was conducted using a DNA Engine Opticon 2 System (Bio-Rad, USA), based on Power SYBR^® Green Master Mix (Applied Biosystems). Each 10 μl reaction comprised 1 μl template, 5 μl 2× SYBR Premix, 0.4 μl (10 µM) of each primer and 3.2 μl RNase-free ddH₂O. The primers for transcript analysis were designed using the Primer premier 5.0 software and are listed in Table S11. The reactions were incubated at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. A melting curve analysis was performed at the end of the PCR run over the range 60-95°C, increasing the temperature stepwise by 0.2°C every 2 s, normalized to the reference transcript UBQ10. Each PCR reaction was carried out in triplicate. C_q values were converted into relative quantities via the delta-C_q method. The amplification efficiencies (E) and correlation coefficient R² values were calculated using LinRegPCR software (Karlen et al., 2007; Ruijter et al., 2013).

Bioinformatics analyses were performed using a server with two 8-core processors per compute node, 128 gigabytes of RAM, and an Ubuntu Linux-Distribution operating system at the Bioinformatics Research Center (BRC) at North Carolina State University.

Low-quality cDNA sequences were filtered out with Ea-Utils software (Lindgreen, 2012). Reads with Q-score (phred-like quality score) greater than 20 and read length greater than 20-bp were kept and aligned against the TAIR 10 reference genome (Ensembl annotation) downloaded from the iGenome database (http://support.illumina.com/sequencing/sequencing_software/igenome.html).

TopHat2 (Trapnell et al., 2009), a splice junction mapper, was used to align cleaned reads against the entire Arabidopsis genome using the TopHat2 default parameters, except for maximal intron length (−I 2000). The large majority of known introns are smaller than the selected −I 2000 threshold (Li et al., 2013). To align reads solely and exclusively against TAIR 10 annotated gene models, the arguments ‘−T’ (transcriptome only) and ‘−no-novel-juncs’ (no novel junction) also were included. Only uniquely mapped reads to the entire genome were used for downstream analysis. Uniquely mapped reads were extracted from the alignment output binary (BAM) files using Samtools (Li et al., 2009).

The HTSeq: analyzing high-throughput sequencing data with Python software (Anders et al., 2015) was used with default parameters except for the ‘stranded=no’ mode to generate tables-counts. Tables-counts were analyzed with R packages edgeR (Robinson et al., 2010) and DESeq2 (Seyednasrollah et al., 2015).

Differential gene expression analyses were carried out using the R packages edgeR, and DESeq2 and Cufflinks program (Love et al., 2014) for DEGs and transcripts. Filters were applied to determine if a gene was detected as a means to enrich for true DEGs, to reduce type I error, and to improve P-value adjustment (Bourgon et al., 2010; Pimentel et al., 2014; Seyednasrollah et al., 2015). The edgeR function ‘counts per million’ (cpm) (Robinson et al., 2010) was used to discard those genes whose cpm was lower than a threshold of two reads per gene in at least two biological replicates. For Cufflinks, a minimum RPKM of 5 was set for a gene to be expressed, following the criteria of Suzuki et al. (2015).

We thank Bob Franks for supporting G.V.; Stéphane Verger, Lucie Riglet, Carrie Donley, Tony Perdue, Alex Davis, and Bill Kier for help with microscopy; Yoan Coudert for use of the Keyence microscope; Olivier Hamant for use of the sp8 confocal microscope; Tim Ross-Elliott for providing the P_35S:PM-mCherry line; and Marc Nishimura and Zack Nimchuk for vector pCUT6. Some eSEM images were taken at the Chapel Hill Analytical and Nanofabrication Laboratory (CHANL), a member of the North Carolina Research Triangle Nanotechnology Network (RTNN), which is supported by the National Science Foundation (ECCS-1542015) as part of the National Nanotechnology Coordinated Infrastructure (NNCI).