LEAFY maintains apical stem cell activity during shoot development in the fern Ceratopteris richardii

During land plant evolution, determinate spore-bearing axes (retained in extant bryophytes such as mosses) were progressively transformed into indeterminate branching shoots with specialized reproductive axes that form flowers. The LEAFY transcription factor, which is required for the first zygotic cell division in mosses and primarily for floral meristem identity in flowering plants, may have facilitated developmental innovations during these transitions. Mapping the LEAFY evolutionary trajectory has been challenging, however, because there is no functional overlap between mosses and flowering plants, and no functional data from intervening lineages. Here, we report a transgenic analysis in the fern Ceratopteris richardii that reveals a role for LEAFY in maintaining cell divisions in the apical stem cells of both haploid and diploid phases of the lifecycle. These results support an evolutionary trajectory in which an ancestral LEAFY module that promotes cell proliferation was progressively co-opted, adapted and specialized as novel shoot developmental contexts emerged.


Introduction
Land plants are characterized by the alternation of haploid (gametophyte) and diploid (sporophyte) phases within their lifecycle, both of which are multicellular (Niklas and Kutschera, 2010;Bowman et al., 2016). In the earliest diverging bryophyte lineages (liverworts, mosses and hornworts) the free-living indeterminate gametophyte predominates the lifecycle, producing gametes that fuse to form the sporophyte. The sporophyte embryo develops on the surface of the gametophyte, ultimately forming a simple determinate spore-producing axis (Kato and Akiyama, 2005;Ligrone et al., 2012). By contrast, angiosperm (flowering plant) sporophytes range from small herbaceous to large arborescent forms, all developing from an indeterminate vegetative shoot apex that ultimately transitions to flowering, and gametophytes are few-celled determinate structures produced within flowers . A series of developmental innovations during the course of land plant evolution thus simplified gametophyte form whilst increasing sporophyte complexity, with a prolonged and plastic phase of vegetative development arising in the sporophyte of all vascular plants (lycophytes, ferns, gymnosperms and angiosperms).
Studies aimed at understanding how gene function evolved to facilitate developmental innovations during land plant evolution have thus far largely relied on comparative analyses between bryophytes and angiosperms, lineages that diverged over 450 million years ago. Such comparisons have revealed examples of both sub-and neo-functionalization following gene duplication, and of cooption of existing gene regulatory networks into new developmental contexts. For example, a single bHLH transcription factor in the moss Physcomitrella patens regulates stomatal differentiation, whereas gene duplications have resulted in three homologs with sub-divided stomatal patterning eLife digest The first plants colonized land around 500 million years ago. These plants had simple shoots with no branches, similar to the mosses that live today. Later on, some plants evolved more complex structures including branched shoots and flowers (collectively known as the "flowering plants"). Ferns are a group of plants that evolved midway between the mosses and flowering plants and have branched shoots but no flowers.
The gradual transition from simple to more complex plant structures required changes to the way in which cells divide and grow within plant shoots. Whereas animals produce new cells throughout their body, most plant cells divide in areas known as meristems. All plants grow from embryos, which contain meristems that will form the roots and shoots of the mature plant. A gene called LEAFY is required for cells in moss embryos to divide. However, in flowering plants LEAFY does not carry out this role, instead it is only required to make the meristems that produce flowers.
How did LEAFY transition from a general role in embryos to a more specialized role in making flowers? To address this question, Plackett, Conway et al. studied the two LEAFY genes in a fern called Ceratopteris richardii. The experiments showed that at least one of these LEAFY genes was active in the meristems of fern shoots throughout the lifespan of the plant. The shoots of ferns with less active LEAFY genes could not form the leaves seen in normal C. richardii plants. This suggests that as land plants evolved, the role of LEAFY changed from forming embryos to forming complex shoot structures.
Most of our major crops are flowering plants. By understanding how the role of LEAFY has changed over the evolution of land plants, it might be possible to manipulate LEAFY genes in crop plants to alter shoot structures to better suit specific environments. context in non-flowering vascular plants, a context that cannot be predicted from existing bryophyte data.
The role of LFY in non-flowering vascular plant lineages has thus far been hypothesized on the basis of expression patterns in the lycophyte Isoetes sinensis (Yang et al., 2017), several gymnosperm species (Mellerowicz et al., 1998;Mouradov et al., 1998;Shindo et al., 2001;Carlsbecker et al., 2004;Vázquez-Lobo et al., 2007;Carlsbecker et al., 2013) and the fern Ceratopteris richardii (hereafter 'Ceratopteris') (Himi et al., 2001), which has been used as a model of fern development for a number of years (Hickok et al., 1995). These studies reported broad expression in vegetative and reproductive sporophyte tissues of I. sinensis and gymnosperms, and in both gametophytes and sporophytes of Ceratopteris. Although gene expression can be indicative of potential roles in each case, the possible evolutionary trajectories and differing ancestral functions proposed for LFY within the vascular plants (Theissen and Melzer, 2007;Moyroud et al., 2010) cannot be resolved without functional validation. Here we present a functional analysis in Ceratopteris that reveals a stem cell maintenance role for at least one of the two LFY homologs in both gametophyte and sporophyte shoots and discuss how that role informs our mechanistic understanding of developmental innovations during land plant evolution.

Results
The CrLFY1 and CrLFY2 genes duplicated recently within the fern lineage The LFY gene family is present as a single gene copy in most land plant genomes (Sayou et al., 2014). In this regard, the presence of two LFY genes in Ceratopteris (Himi et al., 2001) is atypical. To determine whether this gene duplication is more broadly represented within the ferns and related species (hereafter 'ferns'), a previous amino acid alignment of LFY orthologs (Sayou et al., 2014) was pruned and supplemented with newly-available fern homologs (see Materials and methods) to create a dataset of 120 sequences,~50% of which were from the fern lineage (Supplementary file 1-3). The phylogenetic topology inferred within the vascular plants using the entire dataset (Figure 1-figure supplement 1) was consistent with previous analyses (Qiu et al., 2006;Wickett et al., 2014). Within the ferns (64 in total), phylogenetic relationships between LFY sequences indicated that the two gene copies identified in Equisetum arvense, Azolla caroliniana and Ceratopteris each resulted from recent independent duplication events ( Figure 1). Gel blot analysis confirmed the presence of no more than two LFY genes in the Ceratopteris genome ( Figure 1-figure supplement 2). Given that the topology of the tree excludes the possibility of a gene duplication prior to diversification of the ferns, CrLFY1 and CrLFY2 are equally orthologous to the single copy LFY representatives in other fern species.
CrLFY1 and CrLFY2 transcripts accumulate differentially during the Ceratopteris lifecycle The presence of two LFY genes in the Ceratopteris genome raises the possibility that gene activity was neo-or sub-functionalized following duplication. To test this hypothesis, transcript accumulation patterns of CrLFY1 and CrLFY2 were investigated throughout the Ceratopteris lifecycle (shown as a schematic in Figure 2 for reference). The developmental stages sampled spanned from imbibed spores prior to germination of the haploid gametophyte ( Figure 3A), to differentiated male and hermaphrodite gametophytes ( Figure 3B-D), through fertilization and formation of the diploid sporophyte embryo ( Figure 3E), to development of the increasingly complex sporophyte body plan ( Figure 3F-K). Quantitative realtime PCR (qRT-PCR) analysis detected transcripts of both CrLFY1 and CrLFY2 at all stages after spore germination, but only CrLFY2 transcripts were detected in spores prior to germination ( Figure 3L). A two-way ANOVA yielded a highly significant interaction (F(10,22) = 14.21; p<0.0001) between gene copy and developmental stage that had not been reported in earlier studies (Himi et al., 2001), and is indicative of differential gene expression between CrLFY1 and CrLFY2 that is dependent on developmental stage. Of particular note were significant differences between Figure 2. The lifecycle of Ceratopteris richardii. Ceratopteris propagates in the haploid gametophyte phase of its lifecycle (n) through single-celled spores (A) On spore germination (B) a two-dimensional photosynthetic thallus develops into one of two sexes, a default hermaphrodite (C) which produces eggs and sperm (D) or a hormoneinduced male that produces sperm only (E). Eggs are retained on the hermaphrodite thallus, and fertilization results in the development of a diploid (2n) embryo on the gametophyte (F), initiating the sporophyte phase of the lifecycle. The sporophyte establishes a vegetative shoot that initiates leaflike lateral organs (fronds) and roots from its apex (G). The first fronds produced are simple but later fronds become increasingly lobed and dissected (H, I).
The sporophyte undergoes a reproductive phase-change and subsequent fronds generate haploid spores by meiosis on their undersides (J), enclosed in a morphologically-distinct curled lamina. Mature spores are dispersed to restart the lifecycle. DOI: https://doi.org/10.7554/eLife.39625.006 . Imbibed spores (A); populations of developing gametophytes harvested at 5 (B, C) and 8 (D) days after spore-sowing (DPS), comprising only males (B) or a mixture of hermaphrodites (h) and males (m) (C, D); fertilized gametophyte subtending a developing sporophyte embryo (em) (E); whole sporophyte shoots comprising the shoot apex with 3 (F) or five expanded entire fronds attached (G); individual vegetative fronds demonstrating a heteroblastic progression in which frond complexity increases through successive iterations of lateral outgrowths (pinnae) (H-J); complex fertile frond with Figure 3 continued on next page CrLFY1 and CrLFY2 transcript levels during sporophyte development (Supplementary file 4). Whereas CrLFY2 transcript levels were similar across sporophyte samples, CrLFY1 transcript levels were much higher in samples that contained the shoot apex ( Figure 3F,G) than in those that contained just fronds ( Figure 3H-K). These data suggest that CrLFY1 and CrLFY2 genes may play divergent roles during sporophyte development, with CrLFY1 acting primarily in the shoot apex and CrLFY2 acting more generally.
Spatial expression patterns of CrLFY1 are consistent with a retained ancestral role to facilitate cell divisions during embryogenesis Functional characterization in P. patens previously demonstrated that PpLFY promotes cell divisions during early sporophyte development . To determine whether the spatial domains of CrLFY1 expression are consistent with a similar role in Ceratopteris embryo (early sporophyte) development, transgenic lines were generated that expressed the reporter gene B-glucuronidase (GUS) driven by a 3.9 kb fragment of the CrLFY1 promoter (CrLFY1 pro ::GUS). This promoter fragment comprised genomic sequence encoding the entire published 5'UTR (Himi et al., 2001) plus a further 1910 bp upstream of the predicted transcription start site (Figure 1-figure supplement 2A). In the absence of a genome sequence, repeated attempts to isolate an analogous fragment of CrLFY2 sequence were unsuccessful (see Materials and methods for details). Construct maps plus DNA blot and PCR validation of transgenic lines are shown in Figure 4-figure supplements 1-4. GUS activity was monitored in individuals from three independent transgenic lines, sampling both before and up to six days after fertilization ( Figure 4A-O), using wild-type individuals as negative controls ( Figure 4P-T) and individuals from a transgenic line expressing GUS driven by the constitutive 35S promoter (35S pro ) as positive controls ( Figure 4U-Y). Notably, no GUS activity was detected in unfertilized archegonia of CrLFY1 pro ::GUS gametophytes ( Figure 4A,F,K) but by two days after fertilization (DAF) GUS activity was detected in most cells of the early sporophyte embryo ( Figure 4B,G,L). At 4 DAF, activity was similarly detected in all visible embryo cells, including the embryonic frond, but not in the surrounding gametophytic tissue (the calyptra) ( Figure 4C,H,M). This embryo-wide pattern of GUS activity became restricted in the final stages of development such that by the end of embryogenesis (6 DAF) GUS activity was predominantly localized in the newly-initiated shoot apex ( Figure 4D,E,I,J,N,O). Collectively, the GUS activity profiles indicate that CrLFY1 expression is induced following formation of the zygote, sustained in cells of the embryo that are actively dividing, and then restricted to the shoot apex at embryo maturity. This profile is consistent with the suggestion that CrLFY1 has retained the LFY role first identified in P. patens Relative expression levels of CrLFY1 and CrLFY2 (normalized against the housekeeping genes CrACTIN1 and CrTBP) at different stages of development. n = 3; Error bars = standard error of the mean (SEM). Pairwise statistical comparisons (ANOVA followed by Tukey's multiple comparisons test-Supplementary file 4) found no significant difference in CrLFY2 transcript levels between any gametophyte or sporophyte tissues sampled after spore germination (p>0.05) and no significant difference between CrLFY1 and CrLFY2 transcript levels during early gametophyte development (p>0.05) (B, C). Differences between CrLFY1 and CrLFY2 transcript levels were significant in gametophytes at 8 DPS (p<0.05) (D). CrLFY1 transcript levels were significantly higher in whole young sporophytes (F) and vegetative shoots (G) compared to isolated fronds (H-K) (p<0.05). CrLFY1 transcript levels in whole sporophytes and shoots were greater than CrLFY2, whereas in isolated fronds CrLFY1 transcript levels were consistently lower than CrLFY2 (p<0.05). Asterisks denote significant difference (*, p<0.05; **, p<0.01, ***, p<0.001; ****, p<0.0001) between CrLFY1 and CrLFY2 transcript levels (Sidak's multiple comparisons test) within a developmental stage. Letters denote significant difference (p<0.05) between developmental stages for CrLFY1 or CrLFY2 (Tukey's test). Groups marked with the same letter are not significantly different from each other (p>0.05). Statistical comparisons between developmental stages were considered separately for CrLFY1 and CrLFY2. The use of different letters between CrLFY1 and CrLFY2 does not indicate a significant difference. DOI: https://doi.org/10.7554/eLife.39625.007 The following source data is available for figure 3:  , namely to promote the development of a multicellular sporophyte, in part by facilitating the first cell division of the zygote.
CrLFY1 is expressed in dividing tissues throughout shoot development Both mosses and ferns form embryos, but moss sporophyte development is determinate postembryogenesis (Kato and Akiyama, 2005;Kofuji and Hasebe, 2014) whereas fern sporophytes are elaborated post-embryonically from indeterminate shoot apices (Bierhorst, 1977;White and Turner, 1995). The Ceratopteris shoot apex comprises a single apical cell that generates daughter cells through asymmetric divisions, and individual lateral organs (fronds and root) arise from their own apical cells specified within the grouped descendants of these daughter cells (Hou and Hill, 2002;Hou and Hill, 2004). CrLFY1 pro ::GUS expression in the shoot apex at the end of embryogenesis ( Figure 4E,J,O) and elevated transcript levels in shoot apex-containing sporophyte tissues ( Figure 3L) suggested an additional role for CrLFY1 relative to that seen in mosses, namely to promote proliferation in the indeterminate shoot apex. To monitor CrLFY1 expression patterns in postembryonic sporophytes, GUS activity was assessed in CrLFY1 pro ::GUS lines at two stages of vegetative development ( Figure  In young sporophytes (20 DAF), GUS activity was primarily localized in shoot apical tissues and newly-emerging frond primordia ( Figure 5A,F,K), with very little activity detected in the expanded simple fronds produced at this age ( Figure 5B,G,L). In older vegetative sporophytes (60 DAF), which develop complex dissected fronds ( Figure 5C,H,M), GUS activity was similarly localized in the shoot apex and young frond primordia in two out of the three fully characterized lines ( Figure 5D,I,N) and in a total of 8 out of 11 lines screened (from seven independent rounds of plant transformation). GUS activity was also detected in developing fronds in regions where the lamina was dividing to generate pinnae and pinnules ( Figure 5E,J,O). In some individuals GUS activity could be detected in frond tissues almost until maturity ( Figure 5C). Notably, patterns of CrLFY1 pro ::GUS expression were the same in the apex and complex fronds of shoots before (60 DAF) ( Figure 5C . Consistent with a general role for CrLFY1 in promoting cell proliferation in the shoot, GUS activity was also detected in shoot apices that initiate de novo at the lamina margin between pinnae ( Figure 5Z-AD). Together these data support the hypothesis that LFY function was recruited to regulate cell division processes in the shoot when sporophytes evolved from determinate to indeterminate structures.

Figure 4 continued
entry) are fertilized. After fertilization, the venter forms a jacket of haploid cells known as the calyptra (c) that surrounds the diploid embryo (em). Cell fates in the embryo (embryo frond (fr), embryo foot (ft), root apex (ra) and shoot apex (sa)) are established at the eight-celled stage (Johnson and Renzaglia, 2008), which is around 2 DAF under our growth conditions. Embryogenesis is complete at 6 DAF, after which fronds arise from the shoot apex. Scale bars = 50 mm. DOI: https://doi.org/10.7554/eLife.39625.009 The following figure supplements are available for figure 4:  To test the functional significance of CrLFY expression patterns, transgenic RNAi lines were generated in which one of four RNAi constructs targeted to CrLFY1, CrLFY2 or both were expressed from the maize ubiquitin promoter (ZmUbi pro ). Construct maps plus DNA blot and PCR validation of transgenic lines are shown in Figure 6-figure supplements 1-5. Genotypic screening identified 10 lines  Table 1. Summary of CrLFY RNAi transgenic lines and their phenotypic characterization. Transgenic lines exhibited gametophytic developmental arrest and/or sporophyte shoot termination at varying stages of development. '+' indicates that a particular line was phenotypically normal at the developmental stage indicated, 'À' indicates that development had arrested at or prior to this stage. In lines marked '+/-' the stage at which developmental defects occurred varied between individuals within the line, and at least some arrested individuals were identified at the stage indicated. The five ZmUbi pro ::CrLFY1/2-i1 lines shown were generated from three rounds of transformation, the pairs of lines B16 and B19 and D2 and D4 potentially arising from the same transformation event. The no hairpin control lines NHC-2 (F3) and NHC-3 (F4) may similarly have arisen from a single transformation event. In all other cases, each transgenic line arose from a separate round of transformation and so must represent independent T-DNA insertions. ZmUBI pro ::

RNAi transgene Line
ZmUBI pro :: ZmUBI pro :: ZmUBI pro ::  which contained the complete transgene cassette and three lines that contained a fragment of the transgene which included the antibiotic resistance marker but not the RNAi hairpin (Table 1).
In the no hairpin control (NHC) plants, post-embryonic shoot development initiated with the production of simple, spade-like fronds from the shoot apex ( Figure 6A) as in wild type. In eight transgenic lines, sub-populations of sporophytes developed in which this early stage of sporophyte development was perturbed, one line (E8) failing to initiate recognizable embryos ( Figure 6B) and the remainder exhibiting premature shoot apex termination, typically after producing several distorted fronds ( Figure 6C-H). Sub-populations of phenotypically normal transgenic sporophytes were also identified in some of these lines ( Figure 6I-L). The two remaining lines exhibited less severe shoot phenotypes, one undergoing shoot termination after the production of simple (B19a) or lobed (B19b) fronds at the stage when control sporophytes produced complex dissected fronds ( Figure 6M-O), and the other (D4b) completing sporophyte development but reduced in size to approximately 50% of controls ( Figure 6P,Q). Despite the predicted sequence specificity of CrLFY1-i3 and CrLFY2-i4 (Supplementary file 5), qRT-PCR analysis found that all four RNAi constructs led to suppressed transcript levels of both CrLFY genes ( Figure 6R,S). The severity of the shoot phenotype was correlated with the level of endogenous CrLFY transcripts detected across all lines ( Figure 6R, S), with relative levels of both CrLFY1 and CrLFY2 significantly reduced compared to controls in all early-terminating sporophytes (E8, G13, C3, D2, D4a, D13, F9) (p<0.01 or less). In phenotypically normal transgenic siblings CrLFY2 transcript levels were not significantly lower than controls (indeed in line D2, levels were higher p<0.01) whereas CrLFY1 levels were significantly reduced (p<0.0001), as in arrested siblings. Together, these data indicate that CrLFY2 can compensate for some loss of CrLFY1, but at least 22% of CrLFY1 activity is required for normal development (line D4a pn, Figure 6J,S). It can thus be concluded that CrLFY1 and CrLFY2 act partially redundantly to maintain indeterminacy of the shoot apex in Ceratopteris, a role not found in the early divergent bryophyte P. patens, nor known to be retained in the majority of later diverging flowering plants. . Transcript levels are depicted relative to no hairpin controls (NHC-1or À3), n = 3, error bars = standard error of the mean (SEM). CrLFY1 and CrLFY2 expression levels were significantly reduced compared to controls (p<0.01 or less) in all transgenic lines where sporophyte shoots undergo early termination (A-H), but in phenotypically normal (pn) sporophytes segregating in the same lines (I-L), only CrLFY1 transcript levels were reduced (p<0.0001). CrLFY2 transcript levels in pn sporophytes were not significantly lower than in controls. Asterisks denote level of significant difference from controls (**p<0.01, ***p<0.001; ****p<0.0001). DOI: https://doi.org/10.7554/eLife.39625.017 The following source data and figure supplements are available for figure 6: Source data 1. CrLFY RNAi lines qRT-PCR expression data DOI: https://doi.org/10.7554/eLife.39625.023  (A-C) In no hairpin control lines, the gametophyte established a triangular apical cell (ac) shortly after spore (sp) germination (A). Divisions of the apical cell established a photosynthetic thallus in both hermaphrodite and male gametophytes. At 10 days post spore sowing (DPS) both gametophyte sexes were approaching maturity, with the hermaphrodite (B) having formed a chordate shape from divisions at a lateral notch meristem (n) and having produced egg-containing archegonia (ar), sperm-containing antheridia (an), and rhizoids (rh). The male (C) had a more uniform shape with antheridia across the surface. These phenotypes were identical to wild-type. (D-J) When screened at 10-17 DPS, gametophytes from multiple RNAi lines (as indicated) exhibited developmental arrest, mostly associated with a failure of apical cell activity. Arrest occurred at various stages of development from failure to specify an apical cell, resulting in only a rhizoid being produced and no thallus (D) through subsequent thallus proliferation (E-I). Gametophyte development in one line progressed to initiation of the notch meristem but overall thallus size was severely reduced compared to wild-type (J). (K-R) In situ hybridization with antisense probes detected CrLFY transcripts in the apical cell and immediate daughter cells of wild-type gametophytes at 4 Figure 7 continued on next page CrLFY promotes apical cell divisions in the gametophyte In six of the RNAi lines that exhibited sporophyte developmental defects, it was notable that 50-99% of gametophytes arrested development prior to the sporophyte phase of the lifecycle ( Table 1). This observation suggested that LFY plays a role in Ceratopteris gametophyte development, a function not previously demonstrated in either bryophytes or angiosperms. During wild-type development, the Ceratopteris gametophyte germinates from a single-celled haploid spore, establishing a single apical cell (AC) within the first few cell divisions ( Figure 7A). Divisions of the AC go on to form a two-dimensional photosynthetic thallus in both the hermaphrodite, where a notch meristem takes on growth ( Figure 7B), and male sexes ( Figure 7C) (Banks, 1999). In contrast, the gametophytes from six RNAi lines (carrying either ZmUbi pro ::CrLFY1-i3, ZmUbi pro ::CrLFY1/2-i1 or ZmUbi pro :: CrLFY1/2-i2) exhibited developmental arrest ( Figure 7D-J), which in five lines clearly related to a failure of AC activity. The point at which AC arrest occurred varied, in the most severe line occurring prior to or during AC specification ( Figure 7D) and in others during AC-driven thallus proliferation ( Figure 7E-I). Failure of AC activity was observed in both hermaphrodites ( Figure 7E) and males ( Figure 7H,I). The phenotypically least-severe line exhibited hermaphrodite developmental arrest only after AC activity had been replaced by the notch meristem ( Figure 7J). A role for CrLFY in maintenance of gametophyte AC activity was supported by the detection of CrLFY transcripts in the AC and immediate daughter cells of wild-type gametophytes by in situ hybridization ( Figure 7K-N). By contrast CrLFY transcripts were not detected in arrested ZmUbi pro ::CrLFY1/2-i1 lines ( Figure 7O-R) in which the presence of the transgene was confirmed by genotyping of individual arrested gametophytes (Figure 7-figure supplement 1). CrLFY1 and CrLFY2 transcripts could not be clearly distinguished in situ due to sequence similarity (see Supplementary file 6), and hence the observed phenotypes could not be ascribed to a specific gene copy. However, these data support a role for at least one CrLFY homolog in AC maintenance during gametophyte development, and thus invoke a role for LFY in the regulation of apical activity in both the sporophyte and gametophyte phases of vascular plant development.

Discussion
The results reported here reveal a role for LFY in the maintenance of apical cell activity throughout gametophyte and sporophyte shoot development in Ceratopteris. During sporophyte development, qRT-PCR and transgenic reporter lines demonstrated that CrLFY1 is preferentially expressed in the shoot apex (whether formed during embryogenesis or de novo on fronds, and both before and after the reproductive transition); in emerging lateral organ (frond) primordia; and in pinnae and pinnules as they form on dissected fronds (Figures 3-5). Notably, active cell division is the main feature in all of these contexts. CrLFY2 transcript levels were more uniform throughout sporophyte shoot development, in both dividing tissues and expanded fronds (Figure 3), and expression has previously been reported in roots (Himi et al., 2001). Simultaneous suppression of CrLFY1 and CrLFY2 activity by RNAi resulted in developmental arrest of both gametophyte and sporophyte shoot apices, with any fronds produced before termination of the sporophyte apex exhibiting abnormal morphologies (Figures 6 and 7). The severity of phenotypic perturbations in sporophytes of transgenic lines correlated with combined CrLFY1 and CrLFY2 transcript levels, with wild-type levels of CrLFY2 able to fully compensate for up to a 70% reduction in CrLFY1 levels ( Figure 6). The duplicate CrLFY genes therefore act at least partially redundantly during shoot development in Ceratopteris. A function for LFY in gametophyte development has not previously been reported in any land plant species. In the moss P. patens, PpLFY1 and PpLFY2 are expressed in both the main and lateral apices of gametophytic leafy shoots but double loss-offunction mutants develop normally, indicating that LFY is not necessary for maintenance of apical cell activity in the gametophyte . By contrast, loss of CrLFY expression from the gametophyte shoot apex results in loss of apical cell activity during thallus formation in Ceratopteris (Figure 7). The different DNA binding site preferences (and hence downstream target sequences) of PpLFY and CrLFY (Sayou et al., 2014) may be sufficient to explain the functional distinction in moss and fern gametophytes, but the conserved expression pattern is intriguing given that there should be no pressure to retain that pattern in P. patens in the absence of functional necessity. The thalloid gametophytes of the two other extant bryophyte lineages (liverworts and hornworts) resemble the fern gametophyte more closely than mosses (Ligrone et al., 2012), but LFY function in these contexts is not yet known. Overall the data are consistent with the hypothesis that in the last common ancestor of ferns and angiosperms, LFY functioned to promote cell proliferation in the thalloid gametophyte, a role that has been lost in angiosperms where gametophytes have no apical cell and are instead just few-celled determinate structures.
The range of reported roles for LFY in sporophyte development can be rationalized by hypothesizing three sequential changes in gene function during land plant evolution ( Figure 8). First, the ancestral LFY function to promote early cell divisions in the embryo was retained in vascular plants after they diverged from the bryophytes, leading to conserved roles in P. patens  and Ceratopteris (Figure 4). Second, within the vascular plants (preceding divergence of the ferns) this proliferative role expanded to maintain post-embryonic apical cell activity, and hence to enable indeterminate shoot growth. This is evidenced by CrLFY activity at the tips of shoots, fronds and pinnae (Figures 4-6), all of which develop from one or more apical cells (Hill, 2001;Hou and Hill, 2004). Whether fern fronds are homologous to shoots or to leaves in angiosperms is an area of debate (Tomescu, 2009;Vasco et al., 2013;Harrison and Morris, 2018), but there are angiosperm examples of LFY function in the vegetative SAM (Ahearn et al., 2001;Zhao et al., 2018), axillary meristems (Kanrar et al., 2008;Rao et al., 2008;Chahtane et al., 2013) and in actively dividing regions of compound leaves (Hofer et al., 1997;Molinero-Rosales et al., 1999;Champagne et al., 2007;Wang et al., 2008;Monniaux et al., 2017) indicating that a proliferative role in vegetative tissues has been retained in at least some angiosperm species. Consistent with the suggestion that the angiosperm floral meristem represents a modified vegetative meristem (Theiben et al., 2016), the third stage of LFY evolution could have been co-option and adaptation of this proliferation-promoting network into floral meristems, with subsequent restriction to just the flowering role in many species. This is consistent with multiple observations of LFY expression in both vegetative and reproductive shoots (developing cones) in gymnosperms (Mellerowicz et al., 1998;Mouradov et al., 1998;Shindo et al., 2001;Carlsbecker et al., 2004;Vázquez-Lobo et al., 2007;Carlsbecker et al., 2013;Moyroud et al., 2017) and suggests that pre-existing LFY-dependent vegetative gene networks might have been co-opted during the origin of specialized sporophyte reproductive axes in ancestral seed plants, prior to the divergence of angiosperms.
The proposed evolutionary trajectory for LFY function bears some resemblance to that seen for KNOX protein function. Class I KNOX genes are key regulators of indeterminacy in the vegetative shoot apical meristem of angiosperms (Gaillochet et al., 2015), and are required for compound leaf formation in both tomato and Cardamine hirsuta (Bar and Ori, 2015). In ferns, KNOX gene expression is observed both in the shoot apex and developing fronds (Sano et al., 2005;Ambrose and Vasco, 2016), and in P. patens the genes regulate cell division patterns in the determinate sporophyte (Sakakibara et al., 2008). It can thus be speculated that LFY and KNOX had overlapping functions in the sporophyte of the last common ancestor of land plants, but by the divergence of ancestral angiosperms from gymnosperms, KNOX genes had come to dominate in vegetative meristems whereas LFY became increasingly specialized for floral meristem function. Unlike LFY, however, there is not yet any evidence for KNOX function in the gametophyte of any land plant lineage, and thus if a pathway for regulating stem cell activity was co-opted from the gametophyte into the sporophyte, the LFY pathway is the more likely one.         Continued on next page better resolution within and between angiosperm clades, sequences from Spirodela polyrhiza (32G0007500), Zostera marina (27g00160.1), Aquilegia coerulea (5G327800.1) and Solanum tuberosum (PGSC0003DMT400036749) were added from Phytozome v12.1 (https://phytozome.jgi.doe. gov/pz/portal.html). Genome sequence from the early-diverging Eudicot Thalictrum thalictroides was searched by TBLASTX (Altschul et al., 1990) (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PRO-GRAM=tblastx&PAGE_TYPE=BlastSearch&BLAST_SPEC=&LINK_LOC=blasttab) with nucleotide sequence from the Arabidopsis LFY gene. A gene model was derived from sequence in two contigs (108877 and 116935) using Genewise (Birney et al., 2004) (https://www.ebi.ac.uk/Tools/psa/genewise/). Gymnosperm sequences were retained from Ginkgo biloba and from a subset of conifers included in Sayou et al. (2014), whilst sequences from conifers where in situ hybridization patterns have been reported were added from NCBI -Pinus radiata PRFLL AAB51587.1 and NLY AAB68601.1 (Mellerowicz et al., 1998;Mouradov et al., 1998) and Picea abies PaLFY AAV49504.1 and PaNLY AAV49503.1 (Carlsbecker et al., 2004). Fern sequences were retained except Angiopteris spp sequences which consistently disrupted the topology of the tree by grouping with gymnosperms. To better resolve relationships within the ferns, additional sequences were identified in both NCBI and 1KP (Matasci et al., 2014) databases. The protein sequence from Matteuccia struthiopteris AAF77608.1 MatstFLO (Himi et al., 2001) was retrieved from NCBI. Further sequences from horsetails (2), plus eusporangiate (1) and leptosporangiate (53) ferns were retrieved from the 1KP database (https://db.cngb.org/blast/) using BLASTP and the MatstFLO sequence as a query. Lycophyte and bryophyte sequences were all retained, but the liverwort Marchantia polymorpha predicted ORF sequence was updated from Phytozome v12.1 (Mpo0113s0034.1.p), the hornwort Nothoceros genome scaffold was replaced with a translated full length cDNA sequence (AHJ90704.1) from NCBI and two additional lycophyte sequences were added from the 1KP dataset (Isoetes tegetiformans scaffold 2013584 and Selaginella kraussiana scaffold 2008343). All of the charophyte scaffold sequences were substituted with Coleochaete scutata (AHJ90705.1) and Klebsormidium subtile (AHJ90707.1) translated full-length cDNAs from NCBI. The new/replacement sequences were trimmed and amino acids aligned to the existing alignment from Sayou et al. (2014) using CLUSTALW (Li et al., 2015) (Supplementary file 2 and 3). The best-fitting model parameters (JTT + I + G4) were estimated and consensus phylogenetic trees were run using Maximum Likelihood from 1000 bootstrap replicates, using IQTREE (Nguyen et al., 2015). Two trees were inferred. The first contained only a subset of fern and allied sequences to achieve a more balanced distribution across the major plant groups (81 sequences in total) (Figure 8), whereas the second used the entire dataset (120 sequences~50% of which are fern and allied sequences - Figure 1-figure supplement 1). The data were imported into ITOL (Letunic and Bork, 2016) to generate the pictorial representations. All branches with less than 50% bootstrap support were collapsed. Relationships within the ferns ( Figure 1) were represented by pruning the lycophyte and fern sequences (68 in total) from the tree containing all available fern sequences (Figure 1-figure supplement 1).

CrLFY locus characterization and DNA gel blot analysis
Because no reference genome has yet been established for Ceratopteris (or any fern), CrLFY copy number was quantified by DNA gel blot analysis. Ceratopteris genomic DNA was hybridized using both the highly conserved LFY DNA-binding domain diagnostic of the LFY gene family (Maizel et al., 2005) and also gene copy-specific sequences (Figure 1-figure supplement 2). CrLFY1 and CrLFY2 share 85% amino acid similarity, compared to 65% and 44% similarity of each to AtLFY. DNA gel blotting and hybridization was performed as described previously (Plackett et al., 2014). The results supported the presence of only two copies of LFY within the Ceratopteris genome. All primers used in probe preparation are supplied in the Key Resources Table. Genomic sequences for CrLFY1 and CrLFY2 open reading frames (ORFs) were amplified by PCR from wild-type genomic DNA using primers designed against published transcript sequences (Himi et al., 2001). ORFs of 1551 bp and 2108 bp were obtained, respectively (Figure 1-figure  supplement 2). Exon structure was determined by comparison between genomic and transcript sequences. The native promoter region of CrLFY1 was amplified from genomic template by sequential rounds of inverse PCR with initial primer pairs designed against published CrLFY1 5'UTR sequence and additional primers subsequently designed against additional contiguous sequence that was retrieved. A 3.9 kb contiguous promoter fragment was isolated for CrLFY1 containing the entire published 5'UTR and 1.9 kb of additional upstream sequence (Figure 1-figure supplement  2). Repeated attempts were made to obtain a CrLFY2 promoter fragment but this proved impossible in the absence of a reference genome. Some sequence contiguous with the CrLFY2 ORF was obtained by inverse PCR using primers designed against the previously published 5'UTR sequence of the CrLFY2 transcript (Himi et al., 2001). This sequence was extended to 1016 bp in length using additional primers against the isolated genomic sequence but this fragment did not contain the entire published 5'UTR. Numerous rounds of inverse PCR generated a second 3619 bp genomic fragment containing sequence identical to the remaining 5'UTR (see Figure 1-figure supplement 2, Supplementary file 7) but the presumed connecting sequence between these two fragments could not be amplified despite many attempts. It was eventually concluded that either the intervening promoter fragment was too long to amplify or that it was too GC rich for amplification. All primers used in ORF amplification and inverse PCR are supplied in the Key Resources table. The contiguous sequences obtained for the CrLFY1 and CrLFY2 genomic loci have been submitted to Genbank (accessions MH841970 and MH841971, respectively).
qRT-PCR analysis of gene expression RNA was extracted from Ceratopteris tissues using the Spectrum Total Plant RNA kit (Sigma-Aldrich, St. Louis, MO) and 480 ng were used as template in iScript cDNA synthesis (Bio-Rad). CrLFY1 and CrLFY2 locus-specific qRT-PCR primers were designed spanning intron 1. Amplification specificity of primers was validated via PCR followed by sequencing. qRT-PCR of three biological replicates and three technical replicates each was performed in a Bio-Rad CFX Connect with iTaq Universal SYBR Green Supermix (Bio-Rad, Hercules, CA). Primer amplification efficiency was checked with a cDNA serial dilution. Efficiency was determined using the slope of the linear regression line as calculated by Bio-Rad CFX Connect software. Primer specificity was tested via melting curve analysis, resulting in a single peak per primer set. CrLFY expression was calculated using the 2 -DDCt method (Livak and Schmittgen, 2001) and normalized against the geometric mean of the expression of two endogenous reference genes (Hellemans et al., 2007), CrACTIN1 and CrTATA-BINDING PROTEIN (TBP) (Ganger et al., 2015). The standard deviation of the Ct values of each reference gene was calculated to ensure minimal variation (<3%) in gene expression. Error bars represent ± the standard error of the mean of the 2 DD Ct values. All primers used in qRT-PCR are supplied in the Key Resources table.
Relative expression values of CrLFY from qRT-PCR were compared by one or two-way analysis of variance (ANOVA) for developmental stages followed by Tukey's or Sidak's multiple comparisons, respectively. To test whether genes were downregulated in transgenic RNAi lines, two-way ANOVA was perfomed with gene (CrLFY1 or CrLFY2) and transgenic line as factors, with 'gene' as a repeated factor when all transgenic lines had the same number of replicates. Where appropriate, expression of each gene in each line was compared to the expression of the respective control by Dunnet comparisons. Control plants had been transformed and were hygromycin-resistant, but did not contain the RNAi hairpin that triggers gene silencing (non-hairpin controls, NHC). For all experiments, NHCs were grown alongside transgenic lines. qRT-PCR of transgenic lines was necessarily conducted across several plates, each including a representative NHC, and statistical comparisons were performed within each plate relative to its respective control. The significance threshold (p) was set at 0.05. All statistical analyses were performed in Prism v. 6.0 (GraphPad Software, Inc., La Jolla, CA).

Generation of GUS reporter constructs
The CrLFY1 pro ::GUS reporter construct (Figure 4-figure supplement 1) was created by cloning the CrLFY1 promoter into pART7 as a NotI-XbaI restriction fragment, replacing the existing 35S promoter. A b-Glucuronidase (GUS) coding sequence (Ulmasov, 1997) was cloned downstream of pCrLFY1 as an XbaI-XbaI fragment. The same GUS XbaI-XbaI fragment was also cloned into pART7 to create a 35S pro ::GUS positive control (Figure 4-figure supplement 4). The resulting CrLFY1 pro :: GUS::ocs and 35S pro ::GUS::ocs cassettes were each cloned as NotI-NotI fragments into the pART27based binary transformation vector pBOMBER carrying a hygromycin resistance marker previously optimized for Ceratopteris transformation (Plackett et al., 2015). All primers used in GUS reporter component amplification are supplied in the Key Resources table.

Generation of RNAi constructs
RNAi constructs were designed and constructed using the pANDA RNAi expression system (Miki and Shimamoto, 2004). Four RNAi fragments were designed, two targeting a conserved region of the CrLFY1 and CrLFY2 coding sequence (77% nucleotide identity) using sequences from either CrLFY1 (CrLFY1/2-i1) or CrLFY2 (CrLFY1/2-i2), and two targeting gene-specific sequence within the 3'UTR of CrLFY1 (CrLFY1-i3) or CrLFY2 (CrLFY2-i4) ( Figure 6-figure supplement 1). Target fragments were amplified from cDNA and cloned into Gateway-compatible entry vector pDONR207 (Invitrogen, Carlsbad, CA). Each sequence was then recombined into the pANDA expression vector via Gateway LR cloning (Invitrogen, Carlsbad, CA). All primers used in RNAi target fragment amplification are supplied in the Key Resources table.

Generation of transgenic lines
Transformation of all transgenes into wild-type Hn-n Ceratopteris callus was performed as previously described (Plackett et al., 2015). T 0 sporophyte shoots were regenerated from transformed callus tissue, with each round of transformation using multiple separate pieces of callus as starting material. Transgenic T 1 spores were harvested from these T 0 shoots, germinated to form T 1 gametophytes and then self-fertilized to produce T 1 sporophytes. T 1 sporophytes were assessed for T-DNA copy number by DNA gel blot analysis (

GUS staining
GUS activity analysis in CrLFY1 pro ::GUS transgenic lines was conducted in the T 1 generation. GUS staining was conducted as described previously (Plackett et al., 2014). Optimum staining conditions (1 mg/ml X-GlcA, 5 mM potassium ferricyanide) were determined empirically. Tissue was cleared with sequential incubations in 70% ethanol until no further decolorization occurred. GUS-stained gametophytes were imaged with a Zeiss Axioplan microscope and GUS-stained sporophytes imaged with a dissecting microscope, both mounted with Q-imaging Micro-published 3.3 RTV cameras. Images were minimally processed for brightness and contrast in Photoshop (CS4).

Phenotypic characterization
Phenotypic characterization of RNAi transgenic lines was conducted in the T 2 or T 3 generation. Isogenic lines were obtained by isolating hermaphrodite gametophytes in individual wells at approximately 7 DPS (or when the notch became visible, whichever came first) and flooding them once they had developed mature gametangia (at approximately 9 DPS). All transgenic lines were grown alongside both wild-type and no hairpin controls, and phenotypes observed and recorded daily. Gametophytes exhibiting altered phenotypes were imaged at approximately 10 DPS with a Nikon Microphot-FX microscope. Sporophytes with abnormal phenotypes were imaged with a dissecting microscope.

In situ hybridization
Antisense and sense RNA probes for CrLFY1 and CrLFY2 were amplified and cloned into pCR 4-TOPO (Invitrogen) and DIG-labelled according to the manufacturer's instructions (Roche, Indianapolis, IN). Probes were designed to include the 5'UTR and ORF (CrLFY1 521 bp 5'UTR and 1113 bp ORF; CrLFY2 301 bp 5'UTR and 1185 bp ORF) (Supplementary file 6). All primers used in in situ probe amplification are supplied in the Key Resources table. We were unable to identify fragments that distinguished the two genes in whole mount in situ hybridizations. Tissue was fixed in FAA (3.7% formaldehyde, 5% acetic acid; 50% ethanol) for 1-4 hr and then stored in 70% ethanol. Whole mount in situ hybridization was carried out based on Hejátko et al. (2006), with the following modifications: hybridization and wash steps were carried out in 24-well plates with custom-made transfer baskets (0.5 mL microcentrifuge tubes and 30 mm nylon mesh, Small Parts Inc., Logansport, IN). Permeabilization and post-fixation steps were omitted depending on tissue type to avoid damaging fragile gametophytes, Acetic Anhydride (Sigma-Aldrich) and 0.5% Blocking Reagent (Roche) washing steps were added to decrease background staining, and tissue was hybridized at 45˚C. Photos were taken under bright-field with a Q-imaging Micro-publisher 3.3 RTV camera mounted on a Nikon Microphot-FX microscope. Images were minimally processed for brightness and contrast in Photoshop (CS4). . Supplementary file 5. Specificity of CrLFY RNAi target sequences. Alignment (prepared using Clustal Omega) of the full length transcript sequences for CrLFY1 and CrLFY2, with nucleotide identity between the two gene copies denoted by a subtending asterisk. The CDS for each gene is highlighted in bold. The target sequences for each RNAi construct are highlighted in light blue (CrLFY1/2-i1), light green (CrLFY1/2-i2), dark blue (CrLFY1-i3) or dark green (CrLFY2-i4). The CrLFY1/2-i1 and CrLFY1/2-i2 target sequences each have 77% similarity to the opposing gene transcript (BLAST2n, discontiguous megablast for highly similar sequences). The full-length CrLFY1-i3 and CrLFY2-i4 target sequences do not demonstrate significant sequence similarity to the opposing gene transcript or 3'UTR alone (BLAST2n, blastn for somewhat similar sequences) but short regions of similarity within the target sequence might explain the cross-reactivity observed. DOI: https://doi.org/10.7554/eLife.39625.031 . Supplementary file 6. Predicted hybridization and specificity of CrLFY in situ hybridization probes. Alignment (prepared using Clustal Omega) of full length CrLFY1 and CrLFY2 transcript sequences, with nucleotide identity between the two paralogs denoted by a subtending asterisk. The coding sequence (CDS) for each gene copy is highlighted in bold. Predicted sites of hybridization for the two probes are highlighted in blue (CrLFY1) and yellow (CrLFY2) respectively, with PCR primer sites underlined. The in situ probes span the complete CDS and 5'UTR of each gene copy. The CrLFY1 probe sequence shows 79% nucleotide identity to the CrLFY2 transcript (BLAST2n, discontiguous megablast for highly similar sequences). The CrLFY2 probe shows 79% nucleotide identity to the CrLFY1 transcript (BLAST2n, discontiguous megablast for highly similar sequences). . Supplementary file 8. Summary of published reports of LFY function in a range of angiosperm species. All citations included in reference list of main article (Bradley et al., 1996;Blázquez et al., 1997;Bradley et al., 1997;Kyozuka et al., 1998;Pnueli et al., 1998;Ratcliffe et al., 1999;Gourlay et al., 2000;Bomblies et al., 2003;Becker et al., 2005;Meng et al., 2007;Souer et al., 2008;Wreath et al., 2013).