Systematic analyses of the MIR172 family members of Arabidopsis define their distinct roles in regulation of APETALA2 during floral transition

MicroRNAs (miRNAs) play important roles in regulating flowering and reproduction of angiosperms. Mature miRNAs are encoded by multiple MIRNA genes that can differ in their spatiotemporal activities and their contributions to gene regulatory networks, but the functions of individual MIRNA genes are poorly defined. We functionally analyzed the activity of all 5 Arabidopsis thaliana MIR172 genes, which encode miR172 and promote the floral transition by inhibiting the accumulation of APETALA2 (AP2) and APETALA2-LIKE (AP2-LIKE) transcription factors (TFs). Through genome editing and detailed confocal microscopy, we show that the activity of miR172 at the shoot apex is encoded by 3 MIR172 genes, is critical for floral transition of the shoot meristem under noninductive photoperiods, and reduces accumulation of AP2 and TARGET OF EAT2 (TOE2), an AP2-LIKE TF, at the shoot meristem. Utilizing the genetic resources generated here, we show that the promotion of flowering by miR172 is enhanced by the MADS-domain TF FRUITFULL, which may facilitate long-term silencing of AP2-LIKE transcription, and that their activities are partially coordinated by the TF SQUAMOSA PROMOTER-BINDING-LIKE PROTEIN 15. Thus, we present a genetic framework for the depletion of AP2 and AP2-LIKE TFs at the shoot apex during floral transition and demonstrate that this plays a central role in floral induction.


Characterization of clustered regularly interspaced short palindromic repeats
(CRISPR)-induced mutant alleles and fluorescent protein reporter constructs of entire MIR172 family.

Introduction
Transcription factors (TFs) and microRNAs (miRNAs) are well-studied regulators of development in animals and plants. TFs regulate the transcription of their targets by interacting with cis-regulatory elements in the genome, whereas miRNAs bind to mRNAs to induce transcript degradation or to inhibit their translation. The mRNAs of TFs are prominent targets of miR-NAs, and in animal systems feedback loops between miRNAs and their target, TFs have been shown to confer robustness to gene expression patterns during development [1]. Plant and animal miRNAs are similar in length (20 to 24 nucleotides); however, sequence complementarity between plant miRNAs and their target mRNAs is more extensive than in animals, which is consistent with the higher proportion of miRNA-targeted genes in animals than in plants [2]. The increased specificity of plant miRNAs is offset by the expansion of gene families during plant evolution, particularly TFs, with most plant miRNAs controlling the activity of multigene TF families [3,4]. Plant miRNAs are encoded by families of MICRORNA (MIR) genes. The roles of individual MIRs are relatively poorly understood because loss-of-function mutant alleles are not available for most of them, and functional redundancy is assumed to occur among the members of a single MIR family. To understand the interactions between miRNAs and their targets, most research has been limited to the use of miRNA mimicry strategies or the expression of miRNA-resistant versions of the target genes [5][6][7][8][9][10]. Both of these approaches have been useful but are subject to artefacts caused by off-target effects or ectopic expression, and they do not offer insights into the contribution of individual MIR family members to developmental regulatory networks. The advent of clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR associated protein-9 (Cas-9) systems to induce mutations at specific genomic loci provides a new opportunity to study miRNA functions in plants [11][12][13][14]. MIRs can be inactivated by different types of mutation that could be induced by CRISPR, such as deletion of the coding sequences of the miRNA/miRNA � duplex or mutagenesis of the sites required for DICER-LIKE1 processing of the pri-miRNA [15,16]. Here, we follow both strategies to inactivate all 5 members of the MIR172 gene family in Arabidopsis thaliana.
The targets of miR172 are the mRNAs of the APETALA2-LIKE (AP2-LIKE) TF family, and this interaction is conserved across spermatophytes [17,18]. In A. thaliana, there are 6 members of the AP2-LIKE TF family APETALA2 (AP2), TARGET OF EAT1 (TOE1), TOE2, TOE3, SCHNARCHZAPFEN (SNZ), and SCHLAFMÜ TZE (SMZ). The founding member, AP2, was initially characterized because ap2 mutations exhibit floral organ defects that led to it being assigned A-class function in the ABC model of flower development [19,20]. Subsequently, AP2 and the other AP2-like genes were shown to act as repressors of the floral transition [7,8,[21][22][23]. Ectopic expression of MIR172 phenocopies strong ap2 mutants, whereas ectopic expression of a miR172-resistant version of AP2 mRNA (rAP2) or TOE3 (rTOE3) from the viral 35S promoter produced strongly indeterminate flowers [8,23,24], which is partially due to the negative regulation of the C class gene AGAMOUS by these AP2-LIKE TFs 2. microRNA172 activity at the shoot apical meristem (SAM) is crucially important for floral transition under short-day (SD) conditions. 3. Coordinated transcriptional and posttranscriptional mechanisms repress the activity of APETALA2-LIKE (AP2-LIKE) transcription factors (TFs) during flowering.
other members of the SPL family [9,46]. FUL encodes a MADS-domain containing TF that promotes the floral transition, fruit development, and floral meristem identity and was implicated in the control of inflorescence meristem determinacy, a process termed "global proliferative arrest" [47][48][49][50][51]. Recently, FUL was shown to directly negatively regulate transcription of several AP2-LIKE genes such as SNZ, TOE1, and AP2 during inflorescence development [47], although it is unclear if the regulation of these genes by FUL is important during the floral transition.
We present a comprehensive analysis of the role of MIR172 genes in floral transition by generating novel gene expression reporters and CRISPR-Cas-9-induced mutations for each of the 5 MIR172 members. Detailed phenotypic analysis of single and higher-order mutants as well as confocal microscopy determine the contribution of individual MIR172 genes to floral transition and demonstrate the role of miR172 is most critical under SDs. Further genetic and imaging experiments show that one function of miR172 at the shoot apex is to reduce AP2 protein levels during floral transition. By generating combinations of miR172 mutations, ful and spl15, we show that miR172 and FUL act in parallel to promote floral transition, and their activities are important downstream of SPL15. We propose that AP2 and FUL act at the core of a bistable switch mechanism regulating early stages of the floral transition at the SAM.

Expression analysis of the MIR172 family in long days
The expression patterns of the MIR172 genes were previously assessed by semiquantitative reverse transcription PCR (RT-PCR) in whole seedlings [22]; however, the expression of these genes in the SAM was not determined. Therefore, we harvested RNA from apical tissue of LDgrown plants through a developmental time course and performed quantitative reverse transcription PCR (RT-qPCR; Fig 1A-1F). We analyzed the accumulation of the MIR172 precursor RNAs, because the mature forms encoded by each precursor are too similar to distinguish by this method [8]. Under these conditions, the floral transition occurs between 14 and 17 days after germination, as defined by the level of AP1 transcript ( Fig 1A) and microscopic examination (see later). MIR172A was more highly expressed at 7 days, was at trough levels at 10 days, and then increased again through 12, 14, and 17 days (Fig 1B). In contrast, MIR172B and MIR172D RNAs were at trough levels during vegetative development, but increased during floral transition (Fig 1C-1E). Thus, the abundance of these 3 MIR172 RNAs increased most sharply at the apex during the floral transition and was closely correlated with AP1 expression, although significant increases in MIR172A RNA were observed prior to AP1 expression ( Fig 1B). MIR172C and MIR172E were very lowly expressed in samples derived from vegetative plants or plants undergoing the floral transition; however, MIR172C was expressed at 21 days (Fig 1D-1F).
To improve the resolution of the spatial expression patterns, we cloned the 5 intergenic regions containing the MIR172A-E genes and replaced the miR172 hairpin regions with the coding region for a nuclear localized fusion between the fluorescent protein Venus and β-glucuronidase (NLS-Venus-GUS, NVG) (S1 Fig). Notably, the size and genomic structure of each of the MIR172 genes varied greatly (S1 Fig), which might contribute to their diverse expression patterns. We analyzed several independent transformants of each construct and selected at least 2 for more detailed analysis (Fig 1, S2 and S3 Figs). Apical tissue of these lines was harvested at the same time points described for the RT-qPCR experiments and analyzed by confocal microscopy (Fig 1, S2 and S3 Figs). Only MIR172A-NVG, MIR172B-NVG, and MIR172D-NVG were reproducibly expressed at the shoot apex with each of their temporal expression patterns largely in agreement with the RNA profiling study (Fig 1G-1I ; however, we were unable to detect expression of MIR172E-NVG in the shoot apex. Taken together, these data demonstrate that the MIR172 genes exhibit diverse temporal and spatial expression patterns and that MIR172A, MIR172B, and MIR172D increase in expression in or near the shoot apex before and during the floral transition, whereas MIR172C is mainly expressed in the inflorescence stem, and MIR172E activity could not be detected in the apex of LD-grown plants.

Inactivation of all MIR172 genes using genome editing
Null or strong loss-of-function mutants for the MIR172 genes are not available in A. thaliana, although alleles that reduce the activity of MIR172A and MIR172D have been described [6,34]. Therefore, we generated mutations in each of the 5 MIR172 genes using previously described CRISPR-Cas-9 systems [52,53]. We inactivated the MIR172 genes by deleting or severely disrupting the sequences that comprise the miRNA/miRNA � duplex or by disrupting the sequences that are necessary for DICER-LIKE 1 (DCL1)-mediated processing [15,16,54]. The guide RNAs used, their activity, and the sequences of the stably inherited mutations are shown in S4 Fig. We identified 3 classes of mutation: (i) modifications of the stem region beneath the miRNA/miRNA � duplex (mir172a-2, mir172a-3, mir172b-1, and mir172b-3); (ii) modifications of the miRNA/miRNA � duplex (mir172d-2 and mir172d-3); and (iii) deletions of the entire miRNA region or the entire miRNA/miRNA � duplex (mir172c-1 and mir172e-1) (Fig  2A, S4M Fig). Notably, the predicted secondary structures of the first class of mir172 mutant precursors were altered relative to their WT counterparts (S5 Fig), suggesting that their processing would be impaired [15,16]. To test the effect of the mutations, we ectopically expressed the endogenous and mutant genes in planta to assay their effects on flowering time. While the mean total leaf number (TLN) produced by plants ectopically expressing the WT  MIR172 genes was lower than the empty vector control, the mean TLN produced by plants  expressing mutated versions was similar to the empty vector control (Fig 2B, Table 1, S1  Table). These data indicated that the mir172 mutant alleles were strongly impaired in function and may be null alleles, and this was further supported by testing mir172 levels in higher-order mutants (see below).

Analysis of flowering time behavior of mir172 mutants
We removed the T-DNA harboring the CRISPR-Cas-9 cassette by genetic crossing and generated quintuple mutants using a selection of these novel mir172 alleles. We performed whole genome sequencing of the resulting quintuple mutant and found no off-target mutations induced by the CRISPR-Cas-9 cassettes (Materials and methods). We re-isolated all individual single mutants from plants that were segregating for each selected mir172 mutant and performed flowering time assays in long and short photoperiods. Of the single mutants grown under LD conditions, the average TLN of mir172a-2 and mir172b-3 were reproducibly higher relative to the reference WT genotype, Col-0 ( Fig 2C, Table 1, S1 Table). In days to bolting (DTB), mir172a-2 was slightly higher than Col-0 ( Fig 2C, Table 1, S1 Table). By contrast, these metrics were mostly comparable between the other mir172 mutant alleles and Col-0 (Fig 2C,  Table 1, S1 Table). Therefore, MIR172A and MIR172B are the most important in floral promotion under LDs.
Under SD conditions, mir172a-2 also flowered later than Col-0, whereas mir172b-3 and mir172c-1 flowered at a time similar to Col-0 ( Fig 2D, Table 1, S1 Table). In contrast to LD conditions, mir172d-3 flowered much later than Col-0 when grown under SD conditions ( Fig  2D, Table 1, S1 Table). These results suggested that miR172 activity is more important in the absence of inductive photoperiods and that MIR172D contributes to a large degree of this activity. Therefore, we tested the expression of our MIR172 reporters under these conditions (Fig 2E-2G We then analyzed the flowering time behavior of mir172 mutant combinations and observed progressively later flowering under both LD and SD conditions as more MIR172 genes were inactivated (Fig 3A and 3B). The mir172abcd quadruple mutant flowered later than the mir172abc triple mutant in both conditions, but it was substantially later under SDs, consistent with an important role for MIR172D under these conditions ( Fig 3B, Table 1, S1 Table). The leaf number and DTB were comparable between quintuple mir172abcde mutant plants and quadruple mir172abcd mutant plants that retained a functional MIR172E allele, consistent with a weak role for this gene in controlling flowering time ( Fig 3B). Furthermore, both quadruple and quintuple mutants grown under short photoperiods initially flowered from axillary shoots rather than the primary shoot (70/77 plants), indicating that miR172 is essential for the earlier timing of the floral transition at the SAM compared to axillary meristems under these conditions ( Fig 3C). Siliques of quadruple and quintuple mutant plants exhibited a swollen shape, indicative of defective floral meristem termination (S7A Fig). The severity of this phenotype decreased acropetally, which mirrored previous reports of the severity of ap2 single mutant phenotypes increasing acropetally [55]. Plants lacking MIR172A, MIR172B, and MIR172D activity also bore severely serrated cauline leaves (S7B Fig), which were less serrated in mir172abc or mir172ab mutants and very similar to Col-0 in the single mutant combinations. Supported by the imaging and RT-qPCR results presented above, these results suggest that MIR172E plays a very minor role during the floral transition and floral organ development.
With the exception of the comparison between quadruple and quintuple mutants, the smallest effect size in terms of leaf number and DTB was observed when comparing mir172abc and mir172ab mutants ( Fig 3A and 3B, Table 1, S1 Table). The TLN and DTB of mir172c-1 e-1 double mutants were similar to Col-0 under LDs (S7C Fig, Table 1, S1 Table). Furthermore, all mir172abc plants grown under SD conditions flowered from the main shoot and bore siliques of WT appearance. This suggests that although MIR172C may play some role in controlling flowering time, its contribution is small when compared to MIR172A, MIR172B, and MIR172D.
The TLN and DTB of mir172ab were clearly higher than Col-0 in both photoperiods ( Fig  3A and 3B, Table 1, S1 Table). Although the differences between mir172a single mutants and Col-0 were similar to our previous results (Fig 2C and 2D, Table 1, S1 Table), the variation was increased in these experiments ( Fig 3A and 3B, Table 1, S1 Table). We noted that the number of leaves produced by mir172a mutants often appeared greater relative to Col-0 than the differences in DTB. Therefore, we tested the leaf initiation rate of mir172 mutants and found that the plastochron was longer than Col-0, with robust differences arising after 40 d of growth under short photoperiods (S7D Fig). Therefore, counting the leaf number of mir172 mutants alone is not sufficient to assess their flowering times.
Taken together, these genetic data supported the idea that MIR172A, MIR172B, and MIR172D are the main regulators of flowering time in the MIR172 gene family. This is in agreement with the expression patterns of the reporter constructs, with only MIR172A-NVG, MIR172B-NVG, and MIR172D-NVG being detectably expressed in the shoot apex prior to and around the time of the floral transition. To test this further, we produced mir172a-3 b-3 d-2 triple mutants and compared their flowering time behaviors under long and SDs with the mir172abcde quintuple mutants generated above and the previously described 35Spro: MIM172 transgenic line, which ectopically expresses a target mimicry that depletes miR172 activity [5] (Fig 3D-3F). As observed previously, mir172abcde plants flowered much later than Col-0 ( Fig 3D-3F). The leaf numbers and DTB of mir172abd were very similar to mir172abcde under both conditions (Fig 3D and 3E, Table 1, S1 Table), with most mir172abd and mir172abcde plants initially flowering from axillary shoots when grown under short photoperiods. The leaf numbers and DTB of 35Spro:MIM172 transgenic lines were lower than both the triple and quintuple mutants under LDs and SDs (Fig 3D and 3E, Table 1, S1 Table). Furthermore, the morphology of the 35Spro:MIM172 rosette leaves was markedly different to any of the mir172 mutant plants, and the rosette diameter was substantially smaller, which was particularly noticeable in SDs (S7F Fig). We conclude that MIR172A, MIR172B, and MIR172D are the primary regulators of flowering time in the MIR172 family. We also conclude that miR172 action is essential for flowering from the SAM under noninductive photoperiods and strongly delays flowering, while under inductive photoperiods, it plays a less important role in flowering, and this can be eventually bypassed.

Activity of miR172 in the leaves and meristems under inductive photoperiods
Under inductive LD photoperiods, gene activity in leaves and shoot apices make major contributions to floral induction; therefore, to understand the roles of the MIR172 genes under these conditions, we examined the expression of the MIR172-NVG reporters in the leaves of plants grown under LDs. For this analysis, we used histological staining with the β-glucuronidase enzyme (GUS), which comprises part of our NVG reporters (S1, S8 and S9 Figs). Strong GUS enzyme activity was detected in the cotyledons and throughout the leaves of MIR172B-NVG plants, which grew weaker as the leaf aged. In contrast, MIR172A-NVG activity was weakly but reproducibly detected in the vasculature of young late-arising rosette and/or cauline leaves (approximately 21 d), and MIR172C-NVG was sometimes detected in a similar pattern (  Together, these data indicate that MIR172B makes a major contribution to the pool of miR172 present in the leaves. To quantify the amount of miR172 present, we performed RT-qPCR on mature miR172 in Col-0, mir172b single mutants, and mir172abcde quintuple mutants. MIR172B contributed 50% to 73% of total miR172 levels in leaves 1 and 2 of 12-dayold LD-grown plants (2-tailed paired t test p = 0.021). We did not test miR172 levels in the other single mutants, but in leaves of mir172abcde, quintuple mutants miR172 levels were at approximately 1% of those in Col-0 (2-tailed paired t test p = 0.009) (Fig 4D). This quantitative result is consistent with the absence of miR172 signal in the shoot apices of quintuple mutant plants, as determined by RNA in situ hybridization both under LDs and SDs (Fig 4E, S7E Fig). Some of the AP2-LIKE TFs whose mRNAs are miR172 targets are thought to directly suppress the transcription of the florigen-encoding FT gene in leaves [7,22,27]. We therefore tested FT mRNA levels in the same leaf material of these genetic backgrounds and found that FT mRNA levels were between 49% and 79% of Col-0 levels in mir172b mutants (2-tailed paired t test p = 0.008) (Fig 4D). In mir172abcde quintuple mutants, FT mRNA levels were between 27% and 31% of Col-0 levels (2-tailed paired t test p = 0.015), which suggests that reduced FT partially contributes to the late-flowering phenotype of LD-grown mir172 quintuple mutants ( Fig  4D). However, sufficient FT and TSF expression may be retained to facilitate some response to LDs, as suggested by the dramatic difference between miR172abcde mutant flowering in LDand SD-grown plants (Fig 2A and 2B).

Interplay of AP2 and MIR172 during the floral transition
MiR172 inhibits the activity of AP2-like proteins either by inducing degradation of their mRNAs or inhibiting their translation [8,22,45]. Unfortunately, the activities and the mRNA or protein expression profiles of most of these AP2-LIKE genes have not been precisely determined. Furthermore, two of the best characterized AP2-LIKE genes, TOE1 and SMZ, are genetically linked to MIR172A and MIR172D, respectively, complicating genetic analysis (S10 Fig). Moreover, little is known about TOE2 and SNZ activity, and the main function of TOE3 is probably during floral organ formation [24]. Therefore, we initially focused on testing the relationship between AP2 and miR172 during floral transition. Previous work that included genome-wide assessments of AP2 activity suggested it exerts the majority of its influence on flowering time at the shoot apex [21], although it appears to be expressed throughout the plant [57]. Furthermore, RNA in situ hybridization showed that AP2 mRNA is highly abundant in the vegetative SAM but reduced in the inflorescence meristem [31], which is consistent with its role as a negative regulator of flowering time. To understand how the expression pattern of AP2 progresses during the floral transition, we generated a fluorescently tagged version of AP2 driven by its endogenous regulatory sequences (S11 Fig). We then assessed the expression pattern of AP2-Venus in Col-0 by confocal laser microscopy under LDs (Fig 5A). AP2-Venus was present in the shoot meristem at 14 days after germination but at later time points, as the floral transition proceeded, AP2-VENUS was depleted from the apex. This pattern therefore was anticorrelated with the expression of the MIR172-NVG reporters (Fig 1). We then introgressed the AP2-VENUS transgene into the mir172abd triple mutant background and monitored the expression of AP2-VENUS in LDs (Fig 5A). AP2-VENUS was not detected in the mir172abd plants at 14 days after germination, but the signal appeared at 17 LDs and became strongly expressed in a broader domain at 19 LDs (Fig 5A). At 21 LDs, AP2-VENUS was no longer detected at the shoot meristem. This experiment indicated that in mir172abd plants the expression of AP2-VENUS protein at the shoot apex is delayed and that it then persists later in development than in Col plants. Interestingly, similar observations were made through in situ hybridizations with AP2 mRNA, as it was absent from the apices of WT plants by 16 LDs but  (Fig 5B). To provide a broader view of the effect of mir172abd mutations on the abundance of the mRNAs of AP2 and the AP2-L genes, we performed an RNA-seq analysis of apices of Col-0 and mir172abd plants grown under LDs and SDs (Fig 5C and 5D, S11 Fig, S3 Table). At this level of resolution, the abundance of the mRNAs of AP2 and most AP2-L genes in mir172abd plants was similar to Col-0 throughout the time course. However, the mRNA of TOE2 was significantly reduced in Col-0 compared to mir172abd at later time points under LDs and SDs. This result is consistent with a previous analysis that found that in plants overexpressing miR172, TOE2 mRNA was strongly reduced, while the mRNA levels of AP2 and other AP2-L genes were not significantly altered, perhaps due to impaired feedback regulation [45].
The extended duration of expression of AP2-VENUS in mir172abd plants indicated that the persistence of AP2 in the SAMs of mir172abd plants might contribute to their later flowering. To test this directly, we introduced the null ap2-12 allele into mir172abd triple mutants and scored their flowering time under long and short photoperiods (Fig 5E and 5F). As previously reported, ap2-12 produced fewer leaves and bolted earlier than Col-0 under both LDs and SDs (Fig 5E-5G, Table 1, S1 Table) [21]. Consistent with this, the number of leaves and bolting time of ap2 miR172abd quadruple mutants were much reduced compared to miR172abd triple mutants under both LDs and SDs (Fig 5E-5G, Table 1, S1 Table). Although inactivation of AP2 in the mir172abd mutant produced earlier flowering plants, the quadruple mutant bore more leaves and bolted later than Col-0 ( Fig 5E and 5F, Table 1, S1 Table). These results demonstrate that ectopic and/or prolonged AP2 activity contributes to the late-flowering phenotype of mir172abd mutants and that ectopic expression of other AP2-LIKE genes, including TOE2 (Fig 5C and 5D), likely participates in repressing the floral transition of mir172abd mutants under LDs and SDs.

Synergistic functions of MIR172 and FUL during floral transition
In the mature inflorescence, the MADS-box TF FUL binds to the regulatory regions of AP2, SNZ, and TOE1 to repress their transcription [47], whereas miR172 represses these genes at the posttranscriptional level. Whether this dual mechanism of repression of AP2-LIKE gene expression occurs during floral transition is unclear. Therefore, we tested the effect of combining ful-2 and mir172 mutations on flowering time (Fig 6A, S12A and S12B Fig).
We constructed the ful-2 mir172a-2 b-3 c-1 d-3 quintuple mutant and compared its flowering time under LDs with Col-0 and parental controls ( Fig 6A, Table 1, S1 Table). As previously observed, ful-2 had a higher cauline leaf number (CLN) and the days to the first flower opening (DTF) was higher than Col-0 under these conditions, although the TLN of ful-2 and Col-0 was similar ( Fig 6A, Table 1, S1 Table). The quintuple ful-2 mir172a-2 b-3 c-1 d-3 mutants bore more leaves and opened the first flower later than mir172a-2 b-3 c-1 d-3 ( Fig 6A, Table 1, S1 Table). Under SD conditions, ful-2 was later flowering than Col-0 with slightly fewer rosette leaves, but more cauline leaves (S12B Fig, Table 1, S1 Table), in agreement with previous reports [58]. Under these conditions, the ful-2 mir172a-2 and ful-2 mir172b-3 double mutants were scored due to the strong late-flowering phenotype of the ful-2 mir172a-2 b-3 c-1 d-3 quintuple mutant in LDs (Fig 6A). The ful-2 mir172b-3 plants flowered substantially later than ful-2 mir172a-2 or the single mutant progenitors under SDs (S12B Fig, Table 1, S1 Table), which supports a synergistic interaction between FUL and MIR172B under these conditions. Notably, under LDs, the relative severity of the double mutant flowering time was reversed compared to SDs, so that the ful-2 mir172a-2 double mutant flowered later than ful-2 mir172b-3 with similar leaf numbers (S12A Fig, Table 1, S1 Table). This result suggests a photoperiod-dependent input into different MIR172 genes. Taken together, these data support a redundant role for FUL and miR172 in controlling flowering time. This genetic redundancy probably involves the suppression of AP2-LIKE activity through transcriptional and posttranscriptional means, as was observed in mature inflorescences [47].

MIR172 and FUL contribute to SPL15 function during floral transition
SPL15 is an important regulator of flowering time under SDs, although it also plays a minor role under LD conditions that is circumvented by the photoperiod pathway [35,41]. Notably, miR172 and the mRNA of FUL are undetectable by in situ hybridization in the shoot apices of spl15 mutants grown under SDs for 9 weeks, whereas in WT plants, both FUL mRNA and miR172 are detectable from 5 weeks [35]. Furthermore, SPL15 binds to the regulatory regions of MIR172B and FUL, indicating that these genes are direct targets of SPL15 [35]. To test further the idea that SPL15 activates transcription of MIR172 genes and FUL, we tested the levels of MIR172A-E RNAs and FUL mRNA in a background containing an SPL15 transgene controlled by its endogenous regulatory regions and engineered to be resistant to miR156 (Venus-rSPL15) (Fig 6E-6J) [35]. FUL and all MIR172 genes, except MIR172E, were expressed at higher levels than in Col-0. We then combined Venus-rSPL15, mir172a-2 b-3, and ful-2 and tested their flowering behaviors under both long and SDs (Fig 6K and 6L). Control experiments using Venus-rSPL15 and Col-0 plants showed that Venus-rSPL15 bolted earlier after germination and produced fewer leaves than Col-0 (Fig 6K and 6L, Table 1, S1 Table). However, the time between germination and the DTF was similar for both genotypes (S12C Fig). Thus, the time between DTB and DTF was extended in the Venus-rSPL15 genotype relative to Col-0, which was also reflected in the longer stem length of Venus-rSPL15 compared to Col-0 at the time DTF (S12C Fig). The introgression of either ful-2 or mir172a-2 b-3 into Venus-rSPL15 delayed flowering time. This delay was enhanced further in a Venus-rSPL15 ful-2 mir172a-2 b-3 background (Fig 6K and 6L, Table 1, S1 Table). However, Venus-rSPL15 ful-2 mir172a-2 b-3 still flowered earlier than the ful-2 mir172a-2 b-3 triple mutant. These data provide genetic support for MIR172AB and FUL acting downstream of SPL15, although SPL15 must also activate independent factors that can promote flowering in the absence of MIR172AB and FUL activity.

Late flowering of spl15 mutant is strongly enhanced by mir172 or ful mutations
To determine whether MIR172 and FUL only act downstream of SPL15 or whether they also have independent functions during floral transition, we constructed genotypes carrying the loss-of-function spl15-1 mutation as well as ful-2 or mir172a-2 and mir172b-3, and measured their flowering times (Fig 7A, S12D and S12E Fig).
Under LD conditions, the flowering times of spl15-1 mir172a-2 b-3 and spl15-1 ful-2 were delayed relative to Col-0 and the parental genotypes (S12D and S12E Fig). The interaction between spl15-1 and mir172a-2 b-3 appeared to be largely additive when compared with the expected additive interaction [Δ3.16 TLN (p = 0.13), ΔDTF (p = 0.94)] (S12D Fig). This result is consistent with MIR172A and MIR172B retaining activity in LD-grown spl15-1 mutants, and therefore that AP2-LIKE genes are down-regulated at least partially independently of SPL15 under these conditions [35,41]. The interaction between spl15-1 and ful-2 appeared to be synergistic when compared to the expected additive interaction [Δ7.73 TLN (p = 0.0001), Δ3.05 DTF (p = 0.0001)] (S12E Fig). Given that FUL is required for complete FT activity [49,59], the loss of FUL function in LDs may enhance the importance of SPL15 when the photoperiod pathway is partially disrupted, which indicates that FUL and SPL15 do not act in a simple linear pathway under LDs.
Under SD conditions, many of the genotypes produced extremely high numbers of leaves, so we recorded only the number of days from sowing to bolting for many genotypes in a largescale greenhouse experiment (Fig 7A), although we also performed smaller-scale experiments with combinations of ful-2, mir172a-2, and mir172b-3 in growth cabinets, as described above (S12B Fig). Each of the parental genotypes behaved as previously observed as spl15-1 bolted much later than Col-0, mir172a-2 bolted slightly later than Col-0, and mir172b-3 bolted at a similar time to Col-0 ( Fig 7A, Table 1, S1 Table). The spl15-1 mir172a-2 double mutant bolted much later than spl15-1 and spl15-1 mir172b-3 bolted even later, with both interactions appearing to be synergistic compared to the expected additive interaction [Δ22.73 DTB; p = 0.0002] and [Δ35.48 DTB; p < 10-5], respectively ( Fig 7A, Table 1, S1 Table). This result indicated that MIR172A and MIR172B activities are retained in SD-grown spl15-1 mutants and promote flowering, although miR172 abundance is reduced in spl15-1 [35].
Under SDs, ful-2 also bolted later than Col-0, as expected ( Fig 7A, Table 1, S1 Table). The spl15-1 ful-2 double mutant bolted extremely late compared to spl15-1 in a clearly synergistic interaction compared to the expected additive interaction (Δ33.78 DTB; p < 10−5) (Fig 7A,  Table 1, S1 Table). This experiment also indicated that FUL activity is retained in spl15-1, although its expression is reduced and delayed [35]. Taken together, these data indicate that although MIR172 and FUL are transcriptionally activated by SPL15, they retain activity in spl15-1 mutants, and mutations in MIR172 genes or FUL strongly enhance the spl15 mutant phenotype.
These results indicate that transcriptional activation of FUL and MIR172 by SPL15 may be essential for flowering of the SAM in some plants and that these genes ensure the temporal progression of flowering from the shoot apex to axillary branches during shoot growth.

Discussion
We used a combination of gene expression profiling and genome editing to demonstrate that miR172 is critical for flowering under noninductive SD photoperiods and that MIR172ABD contribute almost all of this activity. Under inductive LD photoperiods, MIR172ABD also promote flowering, but their contribution is hugely reduced compared to SDs, presumably because activation of the photoperiodic flowering pathway reduces the requirement for miR172. Delayed flowering of mir172 mutants was partially conferred by extended expression of the miR172 target AP2 at the shoot meristem. Previously, several AP2-LIKE genes were shown to be directly negatively regulated at the transcriptional level by the FUL TF in the mature inflorescence meristem, and this was proposed to enhance the effect of miR172 on inflorescence meristem determinacy [47]. Our genetic data extend this interaction between miR172 and FUL to floral transition, as we find a synergistic interaction between miR172 and ful mutations in delaying floral transition. We also use these genetic materials to extend the previous molecular analyses of Hyun and colleagues and show that the early flowering of a miR156-resistant version of SPL15 is partially dependent on FUL and MIR172 activities. Our data support a model whereby the promotion of both FUL and MIR172 activities by SPL15 leads to the transcriptional and posttranscriptional suppression of AP2-LIKE activity, which is essential for floral transition at the SAM under SDs.

Essential activity of miR172 at the shoot apex
Flowering time analysis of mir172 quintuple mutants combined with gene expression profiling showed that MIR172A, MIR172B, and MIR172D are the primary regulators of flowering time at the shoot apex (Figs 1 and 2). The product of MIR172B appears to also account for at least half of the total miR172 pool in the leaves (Fig 4D), but other MIR172 genes are also expressed in leaves (S8 Fig). Although miR172 increases FT expression indirectly, this regulation does not control the photoperiod response in a binary manner. Rather, it appears that miR172 modulates the response to photoperiod, perhaps as a means to provide developmental robustness, as has been observed for miRNAs in metazoans [60]. Such a role would explain the observation that MIR172 expression appears to overlap with AP2-LIKE expression in the leaves and cotyledons rather than being mutually exclusive [7,22,57], although understanding their functions in leaves will require more thorough investigation at the tissue and cellular levels. Responses to other environmental stimuli, such as ambient temperature and winter cold [61], may be modulated by miR172 in a similar manner.
At the shoot apex, expression of MIR172 and AP2, one of the 6 members of the AP2-LIKE family, appears to be mutually exclusive (Fig 5A). AP2 mRNA and protein is rapidly cleared from the shoot apex during floral transition, which is probably important for a robust flowering response. This mechanism is particularly important under noninductive photoperiods as the photoperiod pathway appears to largely bypass the requirement for miR172 activity under inductive photoperiods. Genome-wide transcriptome profiling also showed that the mRNA of TOE2 is highly increased under SDs in the mir172abd mutant background compared to Col-0, suggesting that TOE2 also contributes to late flowering of the mutant under these conditions. TOE2 mRNA was previously shown to be reduced by overexpression of miR172, whereas the mRNA of AP2 and other AP2-L genes was not [45]. It was proposed that these TFs feedback to repress their own transcription [21,45], so that reductions in their protein levels by overexpression of miR172 may lead to higher levels of their mRNAs. This feedback regulation together with cross-regulation among different members of the family makes it difficult to predict the temporal expression levels of AP2-L proteins in the miR172 mutants. Indeed, we detected an unexpected delay in AP2-Venus accumulation in the apices of mir172abd mutants. Expression and genetic analysis of the whole AP2-L family will be required to fully decipher the interactions among these proteins.
Loss of MIR172A-E or MIR172A, MIR172B, and MIR172D delayed flowering under noninductive photoperiods by approximately 3 months compared to Col-0 controls. Furthermore, once these mir172 mutants did flower, they mostly did not do so from the primary SAM but from axillary meristems, although usually the primary shoot did eventually flower. This loss of apical dominance suggests that MIR172 activity is required to confer a coherent spatial response to inductive flowering cues and not only for timing the transition. Inactivation of AP2 partially suppressed the late flowering of mir172abd plants, although the ap2 mir172abd plants still flowered much later than Col-0 (Fig 5B and 5C). The effects of introducing mutations in other AP2-L genes, particularly TOE2, into mir172abd mutants remain to be tested. Our RNA-seq analysis suggests that several of these are expressed at the shoot apex and may also be important in controlling the floral transition at the shoot apex. Notably, Zhang and colleagues showed that toe1 and toe1 toe2 plants are extremely early flowering under both SD and LD conditions.

Transcriptional and posttranscriptional regulation of AP2-LIKE activity by FUL and miR172
Previous work showed that FUL binds to regulatory regions in the proximity of AP2-LIKE genes (AP2, TOE1, SNZ, and TOE3) and that these genes are expressed at higher levels in ful mutant inflorescences [47]. Whether these interactions were relevant during floral transition was previously unclear. We tested genetic interactions between ful and mir172 mutants under both LD and SD conditions (Figs 6 and 7). Under LDs, ful mir172abcd mutant plants flowered much later than either mir172abcd or ful plants. Furthermore, ful mir172abcd flowers exhibited a severe loss of determinacy, which did not occur in either parental line. Similarly, the longevity of ful mir172abcd plants was dramatically extended, with LD-grown plants continuing to produce flowers 4 months after germinating. This longevity is consistent with reports that the AP2-miR172 module is involved in global proliferative arrest [47]; however, it should be noted that the fertility of ful mir172abcd plants was extremely low. These phenotypes are closely related to those of plants in which AP2-LIKE genes are ectopically expressed [7,8,24,27,47]. Taken together, this genetic analysis indicates that FUL and miR172 control AP2-LIKE gene activity at both the transcriptional and posttranscriptional levels to regulate the floral transition, although this remains to be confirmed by examining expression of AP2 and AP2-L genes in the ful mir172abd background. Furthermore, as described above, analyzing transcript levels of these genes in different backgrounds is complicated by feedback regulation.
Expression analyses, including in situ hybridizations, indicate that miR172 acts prior to FUL at the shoot apex [35]. Therefore, we propose that miR172 is initially required to reduce AP2-LIKE activity and destabilize the vegetative state at the shoot meristem ( Fig 7C). Furthermore, as AP2 directly represses FUL transcription [21], which is further supported by our analysis (S12F Fig), the depletion of AP2-LIKE TFs by miR172 is proposed to allow FUL transcription to increase. FUL would then bind to the promoters of AP2-LIKE genes initiating their transcriptional repression and thus enhancing the reduction of AP2-LIKE activity mediated by miR172 (Fig 7C). Affinity purification of FUL from inflorescences followed by protein mass spectrometry identified several chromatin remodeling proteins as interacting with FUL [62]. Therefore, binding of FUL to AP2-LIKE genes could mediate long-term transcriptional repression by compaction of the chromatin. We propose that these parallel mechanisms of posttranscriptional repression of AP2-LIKE gene expression by miR172 and transcriptional repression of the same genes by FUL provide a rapid and robust means to deplete these potent negative regulators of floral transition from the induced shoot apex as flowering proceeds ( Fig  7C).
Expression of FUL and MIR172 appears to be partially coordinated by SPL15 under both LDs and SDs, but its contribution is more important under SDs. When grown in SDs, FUL and miR172 are undetectable by RNA in situ hybridization in spl15 mutants and SPL15 binds to the promoter regions of FUL and MIR172B. However, given that mir172abd mutants flower much later than spl15-1 with FUL expression persisting in their SAMs (S12F Fig) and inactivation of FUL or MIR172A/B can strongly enhance the phenotype of spl15-1 (Fig 7A), miR172 and FUL activity must still occur in spl15-1. However, analysis of transgenic plants ectopically expressing a miR156-resistant version of SPL15 demonstrated that SPL15 can promote the transcription of MIR172A, MIR172B, MIR172C, and MIR172D (Fig 6E-6J). Furthermore, genetic analysis demonstrated that FUL and miR172 activity are partially required for rSPL15 activity, while combinations of ful and mir172 mutants can partially phenocopy the spl15 mutant (Figs 6K and 6L and 7A and 7B). These data support the notion that SPL15 acts through both FUL and MIR172 to promote flowering, but that SPL15 also acts independently of these factors (Fig 7C). Similarly, other unidentified regulators must promote FUL and MIR172 expression in parallel with SPL15. Thus, SPL15 mediates the transition from the stable vegetative state to the reproductive state by activating at least 2 repressors of AP2-LIKE TFs that act partially redundantly at different levels of regulation (Fig 7C). This is a form of feedforward circuit that has been described in other systems to mediate robust and stable repression of miRNA targets [60,63] and here likely plays a role in stabilizing the floral state.

Cloning MIR172-NVG reporter constructs
Polymerase Incomplete Primer Extension (PIPE) [64] was used to generate the NVG-stb205 and MIR172(A-E)-NVG-stb205 plasmids. The NVG-stb205 plasmid, which is a variant of pDONR201, was generated with primers oDM-462-465 (S2 Table). To generate the MIR172 (A-E)-NVG-stb205 plasmids, initially, the entire intergenic region containing the MIR172 coding region of each gene was placed in stb205 with primers oDM-11-20 (S2 Table). The NVG coding region was used to replace the miRNA hairpin coding sequence in each MIR172 gene using primers oDM-31-50 (S2 Table). Either Phusion (New England Biolabs, United States of America) or PrimeSTAR GXL (TaKaRa Bio, Japan) high-fidelity polymerases were used for DNA amplification. The PCR products derived from plasmid templates were treated with DpnI (New England Biolabs) overnight at 37˚C to remove the template DNA. After purification of each PCR product with the NucleoSpin PCR clean-up kit (Macherey Nagel, Germany), a molar ratio of 1:10 plasmid:insert PCR products were mixed and 100-μL chemically competent Dh5α Escherichia coli were transformed using the heat shock method. Transformed bacteria were incubated with 500-μL LB for 1 h at 37˚C in a shaker. The entire liquid culture was then centrifuged for 5 min at 5 krpm, and 500 μL of the supernatant was removed. The other 100 μL of supernatant was used to resuspend the bacterial pellet, which was plated on LB-agar plates supplemented with 50 μg/mL kanamycin and incubated at 37˚C overnight. Colonies were tested by PCR using primers in S2 Table, and positive colonies were used to inoculate 5-mL liquid LB culture, which were supplemented with 50 μg/mL kanamycin and were grown overnight at 37˚C. Plasmids were extracted using the NucleoSpin Plasmid extraction kit (Macherey Nagel, Germany), and the success of the cloning was initially determined by DNA restriction enzyme analysis. Plasmids with the correct restriction patterns were subjected to Sanger sequencing (Eurofins Genomics, Germany) to verify that the correct sequences were present in the plasmids. MIR172(A-E)-NVG sequences were then transferred to the p � GW binary vector [65] using the Gateway LR clonase II enzyme mix (Invitrogen, USA), according to the manufacturer's protocol. After transformation protocols as described above, a 100-μL culture was plated on LB-agar plates supplemented with 100 μg/mL spectinomycin. Colonies were tested by PCR using primers in S2 Table. The success of the cloning was determined as described above although Sanger sequencing was not repeated. Plasmids were transformed into Agrobacterium tumefaciens strain GV3101 by electroporation, and 500 μL of LB was added and incubated at 28˚C for 3 to 4 h. Moreover, 100 μL of the culture was plated on LBagar plates supplemented with 100 μg/mL spectinomycin, 100 μg/mL rifampicin, and 25 μg/ mL gentamicin and incubated at 28˚C. Positive colonies were determined by colony PCR using primers in S2 Table, and cultures were grown in LB medium supplemented with 100 μg/ mL spectinomycin, 25 μg/mL gentamicin, and 100 μg/mL carbenicillin.

Cloning overexpression of wild-type and mutant MIR172 alleles
The A. thaliana UBIQUITIN10 promoter and the octopine synthase terminator sequences (from pBJ36 [66]) were introduced to the 5 0 and 3 0 sides of the multiple cloning site of pGREEN-0229 [67], respectively, using restriction enzyme based cloning to generate pGREEN-0229-UBQ10pro-3 0 OCS (S2 Table). The WT and mutant alleles for MIR172 were amplified with the Phusion (New England Biolabs) high-fidelity polymerase and primers that contained XhoI and NotI sites (S2 Table). PCR products and pGREEN-0229-UBQ10-3 0 OCS were incubated with XhoI and NotI (New England Biolabs) overnight at 37˚C. After gel extraction and purification of the digested products with the NucleoSpin PCR clean-up kit (Macherey Nagel, Germany), they were incubated at 4˚C overnight with T4 DNA ligase (New England Biolabs). The ligation reaction was then transformed into Dh5α using the heat shock method. Transformed bacteria were incubated with 500 μL of LB for 1 h at 37˚C in a shaker. Moreover, 100 μL of the supernatant was plated on LB-agar plates supplemented with 50 μg/ mL kanamycin and incubated at 37˚C overnight. Success of the cloning, transformation of A. tumefaciens strain GV3101 that contained the pSOUP helper plasmid, and culturing of positive colonies was performed as described above with the exception that 10 μg/mL of tetracycline was also added to the LB/LB-agar media.

Single guide RNA cloning, T7E1 assays, and CRISPR genotyping
Single guide RNA (SgRNA) cloning was performed according to the methods in [52,53], and the appropriate primers are listed in S2 Table. Pure genomic DNA was extracted from the cauline leaves of T1 plants [68]. The region in which the Cas-9 protein was targeted was flanked by 2 sets of nested primers (S2 Table). Amplified products were reannealed by denaturing them at high temperature and slowly decreasing the temperature [53]. Moreover, 1 μL of the T7E1 enzyme (New England Biolabs) was then introduced to 10 μL of the PCR reaction supplemented with sterile distilled H 2 O and the appropriate buffer (NEB), and samples were incubated at 37˚C for 20 min in a thermocycler. Incubated products were then run on an EtBr-supplemented 3% agarose gel and visualized under an ultraviolet lamp (S4 Fig). Seeds from plants that were positive in the T7E1 assay were then planted, and plants with stable mutations were identified through a combination of T7E1 assays, PCR amplification, and Sanger sequencing. Plants lacking the Cas-9 insertion were identified using a leaf-based phosphinothricin (PPT) resistance assay (S4N Fig) and genotyping (S2 Table).

Plant transformation and selection
The floral dip method [69] was used to transform A. thaliana. A solution containing the appropriate A. tumefaciens strain, 10% w/v sucrose, and 0.025% v/v Silwett L-77 was used to dip the flowers of A. thaliana. Plants were then incubated in the dark at room temperature overnight before being placed in LD-grown conditions. For selection of plants transformed with a bar selection cassette, seeds were germinated on soil, and seedlings were sprayed with 200 μg/ml ammonium-glufosinate. For selection of plants transformed with a hygromycin-resistance selection cassette, sterilized seeds were plated on MS agar supplemented with 12 μg/ml hygromycin, covered, and placed in a growth chamber for 3 d. After 3 d, the cover was removed, and seedlings were grown until a noticeable difference in hypocotyl length was observed. Seedlings with longer hypocotyls were presumed to be resistant to hygromycin. Putative transformants were genotyped using appropriate primers for confirmation.

Phosphinothricin (PPT) resistance assay
Personal communications with several researchers from independent groups highlighted the problem with segregating the Cas-9 containing T-DNAs from the mutation of interest, as they often appeared to be genetically linked. In order to identify plants lacking the Cas-9 containing T-DNAs without needing to genotype hundreds of plants or propagate seeds from hundreds of plants, we used a plate based PPT-resistance leaf assay. MS-agar plates (without sucrose) supplemented with 25 μg/mL PPT were poured. A leaf (preferably a young leaf but part of an old leaf is also suitable) was placed on the MS-agar and pressed down until it stuck to the agar. Plates were then sealed with medical tape and placed in a growth cabinet for 2 to 3 days. Leaves derived from PPT-resistant plants were dark green, whereas leaves from PPT-susceptible plants were light green (S4N Fig). Follow-up genotyping by PCR confirmed the absence or presence of the Cas-9 T-DNA (S2 Table). This assay was not performed in sterile conditions, and fungal growth was kept to a minimum by using sucrose-free MS-agar plates and limiting the time the MS-agar plates were open.

ClearSee and confocal imaging
The ClearSee reagent (10% (w/v) xylitol, 15% (w/v) sodium deoxycholate, and 25% (w/v) urea) was prepared as previously described [70]. Tissue was dissected at approximately zeitgeber time (ZT) 6 and placed in 4˚C 4% paraformaldehyde prepared in PBS (pH 7), which was resting on ice. Tissue was then subjected to a vacuum of approximately 700 mbar at room temperature for 1 h. The vacuum was then released, and samples were incubated for a further 1 h. Samples were washed with 1x PBS 3 times and then incubated with ClearSee in the dark for up to 14 d. The shoot meristems were visualized with a LSM780 (Zeiss, Germany) laser scanning confocal microscope, and the settings were optimized for the visualization of the Venus fluorescent protein (laser wavelength, 514 nm; detection wavelength, 517 to 569 nm) and the SCRI Renaissance 2200 dye (laser wavelength, 405 nm; detection wavelength, 410 to 503 nm), which stains plant cell walls [71]. The same settings were used for each sample; however, the gain for visualizing the plant cell walls by Renaissance staining was modified for presentation purposes. This does not change the interpretation of the results, and original images are available upon request. The Renaissance dye was added to the samples 12 to 24 h before imaging. For Fig 5A, the samples in a Col-0 and mir172abd background are unpaired. The Col-0 background samples were imaged with a lower 514-nm gain. A paired Col-0 background sample was performed, but the signal was dimmer probably because of the quenching effects of ClearSee; however, the data were similar.

RNA extraction and RT-qPCR
Plant tissue was harvested at the indicated times on liquid nitrogen and immediately frozen at −80˚C. Samples were processed with the RNeasy Plant Mini Kit (Qiagen, the Netherlands) according to the manufacturer's protocol. RNA was quantified using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, USA), and up to 5 μg of RNA was incubated with DNase I (TURBO DNA-free kit, Invitrogen) to remove genomic DNA according to the manufacturer's protocol. After quantification, 1 μg of DNA-free RNA was used in a reverse transcription reaction (Superscript IV first-strand synthesis system, Invitrogen) and diluted 1:5 with pure and sterile H 2 O. Moreover, 1 μL of the diluted reverse transcription reaction was used in a PCR reaction with iQ™ SYBR green supermix (Bio-Rad, USA). Relative transcript abundance was determined with a Roche LightCycler 480 system. Measurements were taken for 3 independent biological triplicates, and the average transcript abundance for each biological sample was determined by performing 2 to 3 technical repeats. LightCycler melting curves were obtained for the reactions, revealing single peak melting curves for most amplification products. The amplification data were analyzed using the second derivative maximum method, and resulting Cp values were converted into relative expression values using the comparative cycle threshold method [72]. One reference gene (REF1 (PROTEIN PHOSPHATASE 2A (PP2A)) was used to normalize the data.

DNA extraction and genotyping
Pure genomic DNA was extracted for T7E1 assays according to [68], for next generation sequencing with the DNeasy plant mini kit (Qiagen) according to manufacturer's instructions, and for genotyping according to [73]. Platinum Taq DNA polymerase (Invitrogen) was used for all PCR assays except for cloning DNA fragments.

Statistical analysis
We have implemented some recommendations from The American Statistician [74]. Specifically, we avoid the term "statistically significant," and we report p-values as continuous where possible. We have supplied a Supporting information table in which all p-values are reported in addition to a description of the dataset (S1 Table). Although grouping data based on statistical thresholds is not recommended [74], we have provided a grouping threshold in Table 1 and S1 Table as we believe that it can be a helpful guide, but we do not interpret our results on the basis of these groupings.
Data were analyzed using the R studio software [75]. The distribution and variance of the data were analyzed initially by visualization and the base functions shapiro.test, and bartlett.test and/or fligner.test, respectively. Data were analyzed using 1-way analysis of variance (ANOVA) if data were deemed to be normally distributed (according to a combination of statistical analysis and data visualization) and if the variance of the data were equal (p > 0.01 for at least one of bartlett.test or fligner.test). If these assumptions were violated, data were transformed to meet these assumptions. The Kruskal-Wallis test was used (kruskal.test) if the data were not normally distributed; however, p > 0.01 was required for at least one of bartlett.test or fligner.test. If p < 0.01 for bartlett.test and fligner.test, and data transformation did not off-set the problem, the Brown-Forsythe (bf.test from the package onewaytests) test was implemented [76].
Post hoc tests were employed if a significant result was obtained from the above tests (p < 0.01). The base function pairwise.t.test was used following an ANOVA test, with analysis being paired or unpaired depending on the experimental design. Dunn test (dunn.test [77]) was used following the Kruskal-Wallis test. Games Howell test was used following the Brown-Forsythe test from the package userfriendlyscience [78]. Adjustments for multiple corrections were performed in each case using the Benjamini-Hochberg method [79].
Tests for genetic synergism were done by subtracting observed values from the mean of the reference Col-0. Then means and variances were added to simulate the expected additive effect of the 2 alleles. This was compared with the observed differences between the higher-order mutant and Col-0 using an unpaired 2-tailed Welch t test.
For the box plots, the box encapsulates the 25th to 75th percentile, and the error bars encapsulate the 10th and 90th percentiles. The horizontal line running through the box indicates the median, and each point represents an individual plant. Bars represent the mean value, and the error bars represent the standard error of the mean (SEM). Graphs were modified for style using Adobe Illustrator.

GUS staining
GUS staining was performed according to [80] with some modifications. Briefly, samples were harvested at approximately ZT6 in ice-cold 90% acetone and vacuum infiltrated at approximately 700 mbar for 10 min at room temperature. The vacuum was then released, and samples were incubated for a further 20 min. The acetone was removed and replaced with cold staining buffer (50 mM PO 4 , 0.2% Triton X-100, 2 mM ferro-ferricyanide) without X-glucuronidase (X-gluc) and vacuum infiltrated at approximately 700 mbar on ice for 10 min. Cold staining solution without X-gluc was removed, and staining solution with 2 mM X-gluc was added, and samples were vacuum infiltrated on ice for up to 20 min with the vacuum being released and reapplied 2 to 3 times. Samples were then incubated at 37˚C for 2 h or overnight. After this incubation, samples were submerged in 70% ethanol, which was refreshed until the chlorophyll was removed. Images were taken with a Zeiss SMZ-25 (Zeiss) stereomicroscope or a digital SLR camera.

RNA in situ hybridization
Plants were grown in LD conditions for the specified time. In situ hybridization was performed as previously described [59]. The osa-miR172a miRCURY LNA miRNA detection probe (Qiagen, Cat No. YD00618724) was used to detect miR172 RNA. A probe for AP2 was synthesized as described in [31].

Whole genome sequencing and off-target analysis
Library preparations of genomic DNA derived from quintuple mir172abcde mutant plants were performed using the NEBNext Ultra II FS DNA library preparation kit for Illumina. Whole genome sequencing (20X coverage) was performed with an Illumina HiSeq3000 with single-end 150 bp reads. Potential off-targets were identified by comparing the sgRNA sequences with the A. thaliana genome, allowing for up to 5 mismatches using the dedicated off-target search program CAS-OFFinder [81]. In order to detect mutations, the raw singleend reads were processed by removing the adapter sequences using cutadapt [82] and then trimming low-quality ends using Trimmomatic [83]. The cleaned reads were mapped to the A. thaliana TAIR10 genome using BWA [84] with default settings. BWA outputs both primary, and if found, secondary alignments. In this study, we only retained primary alignments with a mapping quality of at least 30. Samtools mpileup [85] (default settings) was used in conjunction with bcftools [86] for initial SNP calling. SNPs were filtered by minimum quality of 25, minimum depth of 5. Additionally, all regions of the genome that were devoid of any mapped read were labeled as potential larger deletions. Finally, we searched for overlap between the predicted off-targets, the SNPs, and larger deletions.

RNA secondary structure analysis
Sequences from 24 nucleotides downstream and upstream of the miR172/miR172 � duplex coding sequences of WT and mutant MIR172 alleles were processed by mFold version 2.3 [87] with standard settings except that the temperature was set at 22˚C. The predicted structure with the lowest minimum free energy is presented.

RNA extraction and RNA sequencing transcript profiling
The shoot apex of Col-0 and mir172abd were dissected under a stereo microscope at 10, 12, 14, and 17 days in LD conditions and 3, 4, 5, and 6 weeks in SD conditions in 3 independent biological replicates. Total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen, USA) and subjected to DNase treatment using the TURBO DNase (Invitrogen). Poly(A) RNA enrichment, library preparation, and sequencing were carried out at the MPIPZ Genome Center, Cologne, Germany using the following conditions: The RNAs were processed by poly-A enrichment followed by application of basic components of "NEBNext Ultra II Directional RNA Library Prep Kit for Illumina" with a homebrew barcoding regime. Sequencing was performed on a HiSeq3000 sequencer by sequencing-by-synthesis with 1 × 150 bp single-read length. Sequence reads were preprocessed to remove any residual adaptors with CutAdapt, and the low-quality bases (Q<15) were trimmed from the ends with Trimmomatic [82,83]. Only reads with a minimum length of 50 nucleotides were kept. Salmon was used to quantify the abundance of transcripts from the Arabidopsis reference genome Reference Transcript Dataset for Arabidopsis (including guanine/cytosine bias, unstranded samples) [88,89]. Fragments Per Kilobase of transcript per Million (FPKM) values and corrected p-values were obtained using DESeq2 by comparing Col-0 to mir172abd in each time point using standard settings. Data from this RNA-seq transcript profiling are deposited in the BioProject database (https://www.ncbi.nlm.nih.gov/bioproject) under the accession number PRJNA669254.
Supporting information S1 Fig. Schematics of the reporter constructs generated for MIR172A-E. The NVG coding sequence is depicted by a short bright red box (NLS), a yellow box (Venus), and a blue box (GUS). This coding sequence was used to replace the hairpin region (dark red and orange lines) in each MIR172 gene. The beige boxes represent exons, gray lines represent introns, and black lines represent intergenic sequences including untranslated regions. Brown boxes indicate adjacent genes. At3g55513 is annotated as encoding a hypothetical protein, whereas At3g08495 and At2g08215 are annotated as expressing long noncoding RNAs, respectively, and were therefore included in the reporter constructs for MIR172A and MIR172D as they may contain regulatory elements relevant to their expression. The complete genomic structure has only been determined for MIR172A and MIR172B. Therefore, the most upstream transcription start site for MIR172C and MIR172E described in [90] were used. RNA-seq data available from Araport [91] were used to estimate which regions surrounding the hairpin coding regions of MIR172A-C are transcribed. Black arrows beneath the schematics indicate ATGs upstream of the NVG reporter, whereas gray arrows indicate ATG that were present before the MIR172 hairpin coding region that were omitted from the reporter constructs. � and miR172 coding sequences are highlighted in red and orange, respectively. The PAM sequence required for functionality of the Cas-9-sgRNA complex is highlighted in purple. (F-H, J-L) Agarose gels of the indicated T7E1 assays of functional sgRNAs. DNA was derived from T 1 plants transformed with the CRISPR-Cas-9 system from [53]. The expected patterns of restriction digest are indicated below the photograph, and the associated sgRNA is indicated above. (I) An agarose gel of a PCR of MIR172C using DNA derived from T 1 plants harboring the CRISPR-Cas-9 system from [52], indicating the presence of a deletion (asterisk). Data underlying panels F to L are provided in S3 Data. (M) The sequences of WT MIR172A-E genes and the mutants identified by CRISPR-Cas-9. The miR172 � and miR172 coding sequences are highlighted in red and orange, respectively, while the sequences required for the first processing cleavages by DCL1 are highlighted in green. Note that mir172d-3 contains several SNPs (highlighted in red) rather than a deletion. (N) An example of a PPT-resistance assay to identify plants that lack the Cas-9-containing T-DNA. Young leaves were places on an MS-agar (-sucrose) plate supplemented with 25 μg/mL PPT and incubated in a growth chamber for 3 days. The leaves derived from plants lacking the Cas-9-containing T-DNA are light green (red circles), whereas the leaves derived from plants harboring the Cas-9-containing T-DNA are dark green. Cas-9, CRISPR associated protein-9; CLN, cauline leaf number; CRISPR, clustered regularly interspaced short palindromic repeats; PAM, protospacer adjacent motif; PPT, phosphinothricin; sgRNA, single guide RNA; WT, wild-type. The Venus coding sequence is depicted by a yellow box. This coding sequence was fused in frame to the final exon of AP2 by removing the stop codon of AP2. The beige boxes represent exons, gray lines represent introns, and black lines represent intergenic sequences including untranslated regions. Brown boxes indicate adjacent genes. At4g09445/At4g09415 are annotated as long noncoding RNAs, and At4g09415 is annotated as "other RNA" and were therefore included in the reporter construct as they may contain regulatory elements important for AP2 expression. (B, C) Graphs depicting RNA-seq-derived data of the mRNA levels in apices of SMZ, SNZ, TOE1, TOE3, and AP1 in Col-0 and miR172abd at the indicated time points after germination in (B) LDs and (C) SDs. Error bars represent the standard deviation of 3 biological replicates. Data underlying panels (B) and (C) are provided in S1 Data. AP2, APETALA2; AP2-LIKE, APETALA2-LIKE; LD, long-day; RNA-seq, RNA-sequencing; SD, short-day. (TIF) S12 Fig. Interplay between Venus-rSPL15, ful-2, mir172, and spl15 in long and short days. (A, B) Graphs depicting DTF and TLN between the indicated combinations of ful-2 and mir172 mutants grown in (A) long and (B) SD conditions. (C) A bar chart depicting the DTB, DTF, the duration between DTB and DTF (ΔDTF-DTB), and the stem length at flowering. (D, E) Graphs depicting the days to open flower and TLN of combinations between spl15-1, ful-2, and mir172-a2 b-3 in LD conditions. Data underlying panels A to E are provided in S2 Data. (F) Micrographs of RNA in situ hybridizations using a probe designed to recognize FUL of Col-0 (upper panels) and mir172a-2 b-3 c-1 d-3 e-1 (lower panels) apices grown under SD conditions for the indicated times. DTB, days to bolting; DTF, days to the first flower opening; FUL, FRUITFULL; SD, short-day; TLN, total leaf number. (TIF) S13 Fig. Flowering architecture of combinations between spl15-1, ful-2, and mir172. Flowering architecture of indicated combinations between spl15-1, ful-2, and mir172 in (A-C) SDs and (D-H) LDs. LD, long-day; SD, short-day. (TIF) S1  Table. Number of mapped reads for RNA-seq of Col-0 and mir172abd apices in LDs and SDs. LD, long-day; RNA-seq, RNA-sequencing; SD, short-day. (XLSX) S1 Data. Data underlying Fig 1A-1F, Fig 4D, Fig 5C and 5D, S9O Fig, and S11B and S11C  Fig 2B and 2D, Fig 3A and 3B, Fig 3D and 3E, Fig 5E and 5F, Fig  6A, Fig 6K and 6L, Fig 7A and 7B, S7D Fig, and S12A