Direct conversion of root primordium into shoot meristem relies on timing of stem cell niche development

Since the 1950s, in vitro culture methods have contributed to the study of plant cell totipotency and enabled the propagation of plant materials for research or commercial purposes. A large body of work illustrates the complexity of the mechanisms at play during organogenesis. Although plants can be regenerated from a wide variety of tissues, their in vitro response largely depends on cell types, species and genotypes (Cary et al., 2002; Ikeuchi et al., 2016; Motte et al., 2014; Pulianmackal et al., 2014; Skoog and Miller, 1957).

In Arabidopsis, organogenesis can be induced in explants prepared from root or hypocotyl (Atta et al., 2009; Che et al., 2006; Gordon et al., 2007; Valvekens et al., 1988). In two-step protocols, cell proliferation is first enhanced in explants placed on a callus-inducing medium (CIM) characterized by a high auxin/cytokinin ratio. After a few days, the resulting calli can be transferred on a root-inducing medium (RIM) that only contains auxin or on a shoot-inducing medium (SIM) with a high cytokinin/auxin ratio, with founder cells then initiating the formation of one or the other type of organ (Che et al., 2006). The balance between exogenously applied auxin and cytokinin thus directs the development of new organs.

The genetic control of in vitro plant regeneration occurs at multiple levels of regulation (Motte et al., 2014; Xu and Huang, 2014). Among the transcription factors involved, WUSCHEL (WUS) plays a key role in the initiation of shoot organogenesis (Chatfield et al., 2013; Gallois et al., 2004; Gordon et al., 2007; Su et al., 2015). SHOOT MERISTEMLESS (STM) is another transcription factor required for shoot regeneration in vitro (Barton and Poethig, 1993; Endrizzi et al., 1996; Hibara et al., 2003) and linked to the establishment of meristems in response to cytokinin (Brand et al., 2002; Scofield et al., 2013). Both WUS and STM transcripts over-accumulate in explants incubated on SIM (Chatfield et al., 2013; Gordon et al., 2007). In addition, specific histone methylation marks have been linked to shoot regeneration (He et al., 2012) and WUS induction is controlled in part through DNA demethylation in Arabidopsis calli on SIM (Li et al., 2011), indicating that epigenome reprogramming takes place during cell fate transitions.

Shoot-producing calli derived from various Arabidopsis explants seem to originate from perivascular cells, similar to the xylem-pole pericycle cells that are the founders of lateral roots (LR). Converging evidence suggests that explants compatible with de novo organogenesis do not undergo dedifferentiation and resemble LR primordia (LRP) because regenerative calli express genes associated with LR initiation and have transcript profiles reminiscent of root apical meristem (RAM), rather than shoot apical meristem (SAM) or globular embryo (Atta et al., 2009; Che et al., 2007; Sugimoto et al., 2010). Furthermore, LRP could eventually develop into roots or shoots, whether the root segments that produce them are placed on auxin- or cytokinin-rich medium, respectively (Atta et al., 2009; Cary et al., 2002; Che et al., 2007; Sugiyama, 1999), and the ability to regenerate shoot progenitor cells is controlled through a hormone-dependent pathway by root stem cell regulators (Kareem et al., 2015).

The stem cells at the core of plant apical meristems drive development and produce all post-embryonic organs in response to diverse cues (Heidstra and Sabatini, 2014; Scheres, 2007). The production of organs from in vitro-cultured tissues implies cell fate switches for the initiation of novel meristematic structures. Although such transitions are often assumed to involve a dedifferentiation phase, the de novo organogenesis observed in CIM/SIM systems is thought to be a transdifferentiation process, whereby a cell is directly transformed into another cell type without dedifferentiation (Sugimoto et al., 2011). The conversion of root meristems into shoot meristems (SMs), or vice versa, is a very rare phenomenon. But it has been observed in the dominant-negative tpl-1 mutant or through overexpression of a root master regulator in the embryo shoot pole (Smith and Long, 2010).

The successive phases of tissue regeneration are difficult to comprehend because they rely on the coordinated action of multiple factors. In particular, the early events leading to shoot formation in CIM/SIM protocols are difficult to capture because regenerating explants produce massive amounts of proliferating tissues that do not participate in de novo organogenesis. Hence, the molecular mechanisms as well as the cells involved in the very first steps of organ initiation remain poorly characterized.

To understand how the identity of an organ can be switched, we studied the direct conversion of a LRP into a SM combining cytological, molecular and genomic approaches. Our results show that the development of a root primordium can be diverted to yield a growing shoot in as few as four days following cytokinin treatment, without any proliferating callus being formed. Detailed analysis of the LRP/SM conversion proved that it is a transdifferentiation process concomitant with the differential expression of regulators of root and shoot development, dynamic hormonal crosstalks, and coordinated mitotic activity. The conversion is only possible within a narrow window during the development of the initial LRP and requires the presence of apical stem cells. However, within that window, the root and shoot organogenetic programs are remarkably plastic and the status of the stem cell niche can be reversed repeatedly, as previously observed in animal developmental processes.

Arabidopsis primary roots can produce shoots that develop from LRP when treated with an exogenous cytokinin (Atta et al., 2009; Chatfield et al., 2013; Kareem et al., 2015). We noticed that LRs follow distinct organogenetic paths depending on their developmental stage when treated with the cytokinin 2-isopentenyladenine (2-iP): young root primordia rapidly terminate development, intermediate LRP convert into shoots and emerged LRs remain as such (Fig. 1). To investigate the determinants of root-to-shoot conversion, LR formation was primed with the application of the auxin 1-naphthaleneacetic acid (NAA) on primary roots segments, after confirming that induced LRP had the same structure as spontaneous ones (Malamy and Benfey, 1997). The response to cytokinin according to LRP developmental stages was measured based on 4453 LRP (in 210 root segments) of which 1072 switched into SMs (see supplementary Materials and Methods for details). When transferred on 2-iP medium at stage V or younger, almost all LRP terminate growth as green bulges (Fig. 2A,E,I,J), with very few converted shoots originating from such early stages (0.7% of all SMs). These arrested organs grow in size mainly through the enlargement of outer cell layers, with some aberrant cell divisions taking place shortly after transfer. Most shoots develop from LRP at stage VI or VII upon transfer (90.2%) (Fig. 2B,C,F,G). These new shoots have an actively dividing central zone and form leaf primordia as early as four days after 2-iP application (Fig. 2K). LRP at stage VIII (Fig. 2D,H) or older when transferred either sustain growth with no change in organ identity, or stop developing and turn green, in similar numbers (Fig. 2L,M), but few of the converted shoots originate from these later stages (9.1%).

In conclusion, exogenous cytokinin can induce the conversion of a LRP into a functional SM, but only within a very narrow developmental window, corresponding to stages VI and VII in Col-0 primary root segments. LRs at either of these stages are referred to as competent LRs (CLRs) hereafter.

Next, we studied the evolving structure of CLRs across the conversion sequence to understand the cytological features enabling the transformation of a root primordium into a functional shoot meristem. Based on the analysis of cell boundaries in fixed tissues, very few morphological changes occur after the first 24 h of 2-iP treatment, indicating that mitotic activity rapidly decreases upon transfer from NAA to 2-iP medium (Fig. 3A,B). This mitotic pause was confirmed with two mitotic markers in lines that express either the KNOLLE-GFP fusion protein, preferentially labeling the position of the newly formed cell wall during cytokinesis (Fig. 3G-K) (Boutté et al., 2010), or the Destruction Box-GUS fusion protein under the CYCB1;1 promoter marking the G2-M transition (Fig. 3L-P) (González-García et al., 2011). Both markers showed that cell divisions are frequent throughout the CLR before the transfer from NAA to 2-iP, but rare shortly after (paused CLR, pCLR; Fig. 3H,M,Q).

The organ resumes growth gradually during the second and third day of 2-iP treatment with the production of smaller cells indicative of mitotic activity in the provasculature and in the upper half of the dome (converting organ, CO; Fig. 3C,I,N,Q). In the stele, the procambial cells differentiate into vascular tissues. At the top of the CO, divisions result in a round shape after three days of 2-iP treatment (early shoot promeristem, eSP; Fig. 3D,J,O) and in a flattened structure distinct from a root after four days (late shoot promeristem, lSP; Fig. 3E) and that portends the leaf primordia bulging out after four to five days (shoot meristem, SM; Fig. 3F,K,P). Whereas root meristems produce files of cells, cells in all layers of the converting organ divide according to multiple planes as shown by the orientation of the newly formed cell walls (Fig. 3C-E,I,J). Only the epidermis shows a simple pattern with almost exclusively anticlinal divisions. The leaf primordia always form on opposite flanks of the promeristem, suggesting that the root-to-shoot conversion induced in our system does not involve the disorganized proliferation of cells into callus-like structures.

Thus, the developmental shift towards the shoot meristem seems to be engaged as soon as active cell divisions resume because their position and orientation differ in the converting organ compared with the emerging LR (Fig. 3R).

As cytokinin and auxin provide cues to position and maintain stem cell niches (Schaller et al., 2015), we analyzed these hormonal responses in the converting CLR. Auxin fluctuations were investigated through the activity of the auxin-responsive DR5 promoter (Ulmasov et al., 1997), the DII-VENUS fluorescent protein that is degraded in the presence of auxin (Brunoud et al., 2012), and the PIN-FORMED (PIN1) auxin transporter (Benková et al., 2003) (Fig. 4A-O). As expected, cells in NAA-primed CLR have a strong DR5 signal, no detectable DII-VENUS fluorescence, and a preferential localization of PIN1 in the basal plasma membrane, indicating that an auxin maximum is building up at the tip of the primordium where the new RAM should form (Fig. 4A,F,K) (Benková et al., 2003; Kleine-Vehn et al., 2009). The transfer to 2-iP medium induces a sharp drop in auxin level and the clearing of PIN1 from the plasma membranes during the mitotic pause (Fig. 4B,G,L). Auxin levels rise back in the first cells dividing at the tip of the converting organ (Fig. 4C,H), whereas PIN1 strongly marks the L2 and L3 cell layers, then L4 (Fig. 4M). This zone overlaps with the central cell cluster that forms the quiescent center (QC) in the CLR.

The transient auxin maximum together with the resurgence of a PIN1 domain (Fig. 4C,M) suggests that de novo auxin synthesis occurs in the apical portion of the organ. In the late shoot promeristem, the central auxin maximum disappears and a maximum forms in the subepidermal zone of each incipient leaf primordium (Fig. 4D,I), whereas the PIN1 domain extends in all dividing cell layers in the top half of the organ (Fig. 4N). In the fully structured shoot meristem, auxin maxima are restricted to the leaf primordia (Fig. 4E,J) and the PIN1 protein is predominantly located in the L1 layer and in the provascular strands of the growing leaves (Fig. 4O) (Heisler et al., 2005; Reinhardt et al., 2003). No PIN1 protein or auxin signal is detected in the bottom half of the converting organ where PIN1 is normally expressed during LR development (Benková et al., 2003).

Changes in cytokinin signaling were tracked with the TCSn reporter gene (Zürcher et al., 2013). No signal is detected in NAA-primed CLR (Fig. 4P), in agreement with previous reports (Bielach et al., 2012; Chang et al., 2013). Transfer onto 2-iP medium strongly induces the TCSn signal in the stele (Fig. 4Q) and this signal decreases rapidly once cell divisions resume, without expanding into the transient auxin maximum of the CO (Fig. 4R). Finally, the TCSn signal is detected in the inner part of the late shoot promeristem and the shoot meristem (Fig. 4S,T), consistent with the pattern observed in the SAM (Zürcher et al., 2013).

Thus, the auxin and cytokinin response domains rapidly evolve along the conversion process, from the patterns typical of a LRP to those of a SAM. Our data suggest that auxin and cytokinin shape partially complementary domains and act antagonistically to position the new shoot stem cell niche arising from the lateral root meristem.

The expression of genes involved in the initiation and maintenance of root or shoot meristems was further analyzed across the conversion sequence with previously characterized reporter lines, in situ hybridization (ISH) and RT-qPCR. The selected genes code for transcription factors and a secreted peptide and are involved in the development of the stem cell niche, specifically in the root [PLETHORA (PLT), SHORTROOT (SHR), WUSCHEL-RELATED HOMEOBOX 5 (WOX5), SCARECROW (SCR)] or in the shoot [WUS, CLAVATA3 (CLV3), STM] (Heidstra and Sabatini, 2014).

The PLT1, SHR and SCR genes are transcriptionally active in CLR tissues in domains similar to those observed in the primary root meristem (Aida et al., 2004; Helariutta et al., 2000; Terpstra and Heidstra, 2009). The expression of PLT1 and SHR is rapidly turned off after transfer to 2-iP medium and is not detected in the converting organ when active divisions resume (Fig. S1).

At the beginning of the conversion sequence, CLV3, WUS and STM are not expressed in the CLR (Fig. 5A-C,E,F). Within 24 h following transfer onto 2-iP medium, thus during the mitotic pause (pCLR), all three genes are induced in cells located at the center of the organ, corresponding to the root QC, endodermis, pericycle and a part of the stele (Fig. 5A,B,E,F). Once cell divisions resume (CO), the CLV3, STM and WUS domains are restricted to the apex (Fig. 5A,B,E,F) where the STM protein is initially detected (Fig. 5C). At this stage, CLV3, WUS and STM are expressed in overlapping domains. As the promeristem is formed, the CLV3 and WUS domains gradually narrow down to their typical organization in the SAM, with CLV3 marking the central zone containing the stem cells (Fig. 5A) and WUS the organizing center (OC) (Fig. 5B,E). Similar to the STM fusion protein, STM transcription is first restricted to the apical L2 cells layer, when divisions resume, then expands into the promeristem, and is eventually detected across the entire shoot meristem, at high level in the two peripheral zones where the first two leaf primordia bulge out (Fig. 5C,F) (Heisler et al., 2005). The RT-qPCR profiles followed the patterns seen in the reporter lines (Fig. 5D).

The opposite root versus shoot trends faithfully reported the dynamic remodeling that the converting organ operated through defined cell divisions. The transient expression of root- and shoot-specific genes in converting organs at the same developmental stage suggests that they might retain a dual identity during a short time window (Fig. 5; Fig. S1). Alternatively, certain root and shoot markers might be mutually exclusive. But either hypothesis is difficult to prove because the comparative analysis of patterns is limited by the quality of the synchrony between separate objects and by the sensitivity of signal detection.

To gain further insight into the root-to-shoot conversion molecular process, we analyzed the transcriptome signature of the early events involved in the transition phases between the two organogenetic developmental programs. To maximize specificity, the lateral organs were laser microdissected from the root segments and pooled according to four conversion stages: T₀, 42 h NAA priming, CLR; T₆, 6 h 2-iP treatment, paused CLR; T₃₄, 34 h 2-iP, CO resuming cell divisions; T₅₈, 58 h 2-iP, early shoot promeristem (Fig. 5G, see supplementary Materials and Methods for details). Hereafter, a transcript profile change at a given time point refers to a pairwise comparison with the previous time point. Accordingly, 748 genes were differentially expressed (DEGs) at T₆, 1806 at T₃₄ and 527 at T₅₈ (Fig. S2). Table S1 provides the list of all DEGs together with their functional classification and the P-value indicative of statistically significant differential expression. Table S2 presents a summary of the intersection between DEGs identified in this study (1) with genes associated with cell cycle (Vandepoele et al., 2002; Chatfield et al., 2013) or auxin and cytokinin metabolism or signaling (Nemhauser et al., 2006; Brenner and Schmülling, 2015), or (2) with DEGs identified in other studies.

The transcriptome profiles were congruent with cytological and molecular observations. However, the transcript level of some key regulators described above and expressed in specific cells or developmental stages (e.g. CLV3, PLT1, PLT2, WUS) could not be distinguished from background, despite microdissection tissue enrichment. Cell cycle and nucleosome component DEGs were downregulated at the T₆ paused CLR stage (CYCB1;4, CYCB2;4, CYCD3;3, 17 histone-encoding genes) whereas negative mitotic regulators were up (KRP2). Genes typically responsive to cytokinin were highly regulated in the conversion sequence. In comparison with the core set of 77 cytokinin-regulated genes deduced from multiple independent studies (Brenner and Schmülling, 2015), all 25 matching DEGs were up at T₆ (P=8.64×10⁻¹⁸) and most were downregulated at T₅₈ (Tables S1, S2). Inversely, genes associated with auxin responses were mostly down at T₆ but up at T₃₄ (Tables S1, S2), suggesting that auxin production is rapidly tuned after 6 h of 2-iP treatment and contributes to the formation of the shoot promeristem. The downregulation of genes controlling root meristem functions [e.g. GLV5 (also known as RGF2), LBD16, SHR, WOX5] and the induction of genes specific to shoot meristem development [e.g. BLH8 (also known as PNF), CUC1, MYB37 (also known as RAX1), PHV] were also confirmed, either during the mitotic pause (T₆) or after divisions resumed (T₃₄). Furthermore, the genes induced at T₆ indicated that the 2-iP treatment rapidly triggered anabolic processes and intense cellular activity, as they are significantly enriched in functions involved in amino acid, nucleotide, nitrogen and secondary metabolism, gluconeogenesis, and protein and transport pathways (Table S3).

We then compared our data with publicly available transcriptome datasets to examine the nature of the converting tissues. Che et al. (2006) profiled Arabidopsis explants after short or long auxin treatments, and producing shoots or roots depending on the medium they were transferred to. Explants undergoing direct conversion (this study) most closely match shoot-forming calli, but have some DEGs in common with root-forming ones, possibly indicating their dual identity and highlighting that mechanisms linked to shoot commitment are already upregulated 6 h after 2-iP treatment (Table S2). Additionally, we compared the most deregulated genes in the conversion sequence with genes expressed in different anatomical parts and classified as root, shoot or cell cultures (https://genevestigator.com/gv/; see supplementary Materials and Methods). Conversion DEGs are most similar to root and shoot gene sets, but are poorly related to genes associated with callus or cell culture or primary cell classes (Table S4, Fig. S3).

Collectively, these findings indicate that conversion is a transdifferentiation process because the transition from root to shoot operates without an intermediary step of tissue dedifferentiation.

Mutants resulting in the functional loss of DNA methylation showed earlier and more efficient de novo shoot regeneration that correlated with hypomethylated regions in the WUS locus, and with earlier and higher WUS transcription than in wild type (Li et al., 2011; Shemer et al., 2015). We thus tested whether the large-scale changes in gene expression observed through the root-to-shoot conversion are linked with variations in DNA methylation across the nuclear genome.

We first compared the DEGs identified in this study with genes regulated by DNA methylation and possibly involved in shoot regeneration according to Li et al. (2011). These authors showed that over 300 genes differentially expressed in met1-1 mutant calli incubated on CIM medium, compared with wild type, were also differentially expressed in wild-type calli transferred from CIM to SIM, suggesting regulation through MET1-dependent DNA methylation. DEGs induced at T₃₄ and T₅₈ in our system are over-represented in this 300-gene subset (see supplementary Materials and Methods for details), pointing to the putative involvement of MET1-dependent transcriptional regulation in direct conversion (Table S2).

To explore this potential connection further, methylated regions were tracked via MeDIP-chip analysis in genomic DNA extracted from lateral organs whose development was synchronized by growing plantlets in the presence of NPA prior to NAA priming (Fig. 1, Fig. 5G; Materials and Methods). At the level of resolution afforded by MeDIP-chip, the methylome seems largely constant across the conversion sequence, with major root (WOX5, PLT1) and shoot (WUS, STM) development transcription factor genes remaining unmethylated across the series (data not shown). Nevertheless, 400 differentially methylated regions (DMRs) underwent significant albeit low-amplitude changes in methylation level (Fig. S4), of which only 10 coincided with nearby DEGs (Table S5). Whether or not these changes affect the three possible sequence contexts for DNA methylation, namely CG, CHG and CHH, remains to be investigated.

Most LRP at stage V or earlier stall when transferred onto 2-iP medium whereas LRP at stages VI and VII can be converted into shoot meristems. To understand what determines the ability to convert, we examined the primordium morphology at stage VI and VII, compared with V. A central cluster of cells characterizes the later stages (red cells in Fig. 2F-H) (Malamy and Benfey, 1997). It is positioned between the incipient root cap and the provasculature, and flanked on both sides by the forming cortex and endodermis. The precise location and developmental fate of these central LRP cells were confirmed with the expression profile of WOX5, encoding a transcription factor that positively controls cell pluripotency in the QC of the primary root (Lavenus et al., 2013; Pi et al., 2015). Whole-mount ISH revealed that WOX5 transcripts are first detected in these cells at the transition between stages V and VI. The initial WOX5 domain is wider than at stages VII and VIII, when only the third and fourth central cell layers remain labeled (Fig. S1A). Immediately following 2-iP exposure, the transcription of WOX5 collapses (Fig. S1B). As soon as divisions resume in the CO, the same apical stem cells that formed the QC express the shoot meristem genes WUS, CLV3 and STM (Fig. 5A-C,E,F).

In conclusion, LRP at competent stages already contain a stem cell niche, the development of which seems to be required for conversion. These results agree with the earlier observation that shoot primordia only poorly regenerate in root segments of a plt1/plt2 double mutant in which QC specification and maintenance are altered (Kareem et al., 2015). Upon 2-iP treatment, the same niche is reprogrammed to form a shoot, before cell divisions resume.

Last, we investigated whether the identity of the newly formed shoots is fixed or can also be reversed. As illustrated in Fig. 3, round-shaped eSP are visible on root explants placed for 3 days on 2-iP medium, following an initial NPA/NAA synchronization (Fig. 6A). When treated further with NAA, most of these converting organs revert back to lateral root development (Fig. 6B).

The reversion rate was measured by comparing explants kept either on 2-iP medium for 6 days, or incubated for 3 days on 2-iP medium then transferred back on NAA medium. We estimate that 9/10 of the organs at the early shoot primordium stage revert into emerged roots after the final NAA exposure in this experimental setup (see supplementary Materials and Methods for details).

Following NPA/NAA synchronization and 4 days on 2-iP medium, the late shoot promeristems flatten with the bulging of paired leaf primordia (Fig. 6C). When transferred to NAA medium, these more mature meristems do not revert into roots and develop anthocyanin-colored leaflets (Fig. 6D).

WOX5 and WUS expression patterns corroborated the observed reversion constraints. As described above, 2-iP rapidly represses WOX5 and gradually induces WUS in the converting organs (Fig. 6E). But the response to the subsequent 3-day NAA treatment depends on the duration of the prior 2-iP exposure. In all cases, WOX5 transcript level is increased by NAA, but to a lesser extent after 4-day compared with 3-day 2-iP treatments. Concomitantly, WUS is decreased by NAA, but is higher after 4-day than 3-day 2-iP exposure (Fig. 6F).

Whole-mount ISH shows that, in CLRs, WOX5 transcription is restricted to the QC (Fig. 6G) and WUS is not expressed (Fig. 6K). In early shoot promeristem, WOX5 is turned off (Fig. 6H) and WUS transcripts define a large zone (Fig. 6L), in a pattern that prefigures the locked shoot meristem in which WUS expression is reduced to the OC (Fig. 6I,M). However, in the early shoot promeristem exposed for 3 days to NAA, WOX5 is expressed again (Fig. 6J), whereas WUS is off (Fig. 6N), thereby matching the morphological switch back to root development. The WOX5 domain in reverted roots only includes a few cells, as observed in undisturbed LRP from late stage VI onward (Fig. S1A), but it is positioned at least four layers away from the outer boundary of the enlarged organ formed following 2-iP treatment. Presumably, the WOX5-labeled cells mark the new QC in the reverted roots that have other typical hallmarks, including two to three outer root cap layers, made of large cells containing amyloplasts, and radially organized endodermis- and pericycle-like layers surrounding the vascular cylinder (Fig. 6J,N).

In summary, cells previously involved in the formation of the early shoot promeristem are recruited back for the development of a new root meristem. However, the NAA-induced reversion is only possible during early shoot development, after a short exposure to cytokinin.

We have shown that three successive phases can be distinguished during the direct conversion of a lateral root into a shoot meristem. The LR initiates the formation of an apical stem cell niche necessary for sustained organ development. Exogenous cytokinin transiently pauses cell division and induces the organ switch. The cells forming the incipient QC and the surrounding cell layers are recruited to form the novel shoot meristem through active and coordinated cell divisions. The molecular factors involved in the transitions are discussed here below.

Genes involved in shoot development, including STM, CLV3 and WUS, are turned on during the mitotic pause that immediately follows cytokinin treatment. This early induction is most likely a key step triggering the conversion process (Gordon et al., 2007). It might be a direct cytokinin transcriptional response as the initial expression domain for these shoot genes is rather wide, encompassing the stele of the converting organ.

Concomitantly, genes involved in root development, including SCR and WOX5, are repressed, marking the loss of the QC (Zhang et al., 2013). Analysis of molecular reporters of auxin response and PIN1 protein localization showed that exogenous cytokinin also dramatically affects auxin fluxes and gradients necessary for the patterning of the root stem cell niche in the LRP, reflecting the known antagonism between auxin and cytokinin in the course of root development (Dello Ioio et al., 2008; Marhavý et al., 2011; Marhavý et al., 2014). These perturbations alter the maintenance of the lateral root meristem and result in a reduction of mitotic activity.

At the onset of LRP/SM conversion, the apical zone of the organ where mitotic activity resumes strictly overlaps with the transient auxin maximum in cells that accumulate the PIN1 protein in their plasma membrane. Therefore, auxin plays a key role in direct conversion, even though the switch of organ identity is induced by the replacement of an auxin by a cytokinin in the culture medium. This observation agrees with the demonstration that auxin action is required for the establishment of the stem cell organizing center and the SAM formation during somatic embryogenesis (Su et al., 2009, 2015). Furthermore, the presence of shoot progenitor cells is correlated with auxin action (Gordon et al., 2007, 2009; Kareem et al., 2015), and defects in auxin synthesis or transport dramatically disrupt shoot regeneration (Cheng et al., 2013; Kakani et al., 2009).

In the late phase of the shoot promeristem development, auxin maxima only mark the leaf primordia, in accordance with the observation that auxin response must be repressed for the proper homeostatic maintenance of the shoot meristem (de Reuille et al., 2006; Heisler et al., 2005; Roodbarkelari et al., 2015; Yadav et al., 2010). At these late stages, the central apical zone of the meristem is instead characterized by the cytokinin transcriptional response, as expected in an established shoot meristem (Gordon et al., 2009; Zürcher et al., 2013) and previously reported in shoots induced in CIM/SIM protocols (Che et al., 2002; Gordon et al., 2007).

Auxin and cytokinin define separate but complementary domains as the lateral root meristem is reprogrammed. Their spatial arrangement might position the new shoot stem cell niche and guide patterning, as observed during apical meristem initiation and maintenance or when an apical root meristem is regenerated after root tip excision (Efroni et al., 2016; Schaller et al., 2015).

The root-to-shoot conversion induced in our setup did not involve the disorganized proliferation of cells resulting in the formation of callus-like structures for several reasons. Detailed morphological analysis of the converting organ demonstrates that the root meristem cells were directly recruited for the formation of the shoot meristem whose shape rapidly conformed with that of a typical shoot apex, with no tearing or squashing of tissues. The SM central zone (CZ) was positioned at the apex of the lateral organ, its main axis was always perpendicular to that of the primary root segment, and the leaf primordia appeared on opposite sides of the CZ. Such robust patterns are not observed when SMs originate from calli.

Direct conversion is a developmental process that involves major transcriptional reprogramming. Approximately 3000 genes were differentially expressed during the four early time points analyzed. Most genes were significantly up- or downregulated at only one time point, suggesting that the switch of organ identity occurs through a succession of rapidly evolving states. The transition relies on the regulatory action of transcription factors, a class that is significantly overrepresented among the deregulated genes. For example, MYB37 (AT5G23000), among the most highly induced transcription factor genes during the mitotic pause, regulates the formation of the axillary shoot meristem and is its earliest spatial marker, in agreement with a plausible role in root-to-shoot conversion (Keller et al., 2006).

Taken together, our data suggest that the conversion process does not involve cell dedifferentiation but, instead, a rapid switch in the fate of the cells involved, referred to as transdifferentiation.

This report demonstrates that the networks at play in the meristem of a root can morph within a couple of division cycles into those of a shoot. Comparative analysis of root and shoot meristem functions points to common features (Heidstra and Sabatini, 2014), including molecular regulatory modules that might be interchangeable, thus enabling a seamless transition between LRP and SM, and vice versa (Hobe et al., 2003; Perilli et al., 2012; Sarkar et al., 2007; Sebastian et al., 2015; Zhou et al., 2015). For example, WOX5 is replaced by WUS within hours at the onset of conversion, and conversely upon reversion, although they control pluripotency, in the root and shoot apex, respectively.

When divisions resume, cells within the incipient QC and adjacent layers are directly recruited to form the shoot meristem, through a coordinated process circumscribed to the central cell cluster. Such de novo specification implies that positional signals exchanged between stem cells and differentiating daughter cells – proteins, RNAs, phytohormones, secreted peptides – can be rapidly reconfigured (Reddy and Meyerowitz, 2005; Stuurman et al., 2002; van den Berg et al., 1995; Yadav et al., 2010).

In all developmental transitions we studied, WOX5 and WUS are first expressed in a relatively wide domain when a new meristematic structure is initiated, then reach their narrower known positions in the established meristem. This was observed for WOX5 in the first LRP stages. During conversion, the large initial WUS domain also narrows down to the organizing center of the new SM. A similar early confinement of WUS was reported during zygotic embryogenesis (Mayer et al., 1998) and after the laser ablation of the SAM or RAM organizing center (Haecker et al., 2004; Reinhardt et al., 2003; Xu et al., 2006). The STM transcript profile at first overlaps with WUS, but eventually marks the entire shoot promeristem and meristem. We reasoned that the dynamic process leading to the stereotypical organization of an apical meristem is driven by complex regulatory loops, some of which are only triggered after specific signals, thus explaining the sequential confinement observed during conversion.

However, the induction of WUS is not sufficient to operate the root-to-shoot conversion because it is also triggered by cytokinin in stage V LRP that do not develop into shoots. In contrast to the pattern observed in converting organs, the WUS domain does not focus after 2-iP treatment in unconverted LRP (Fig. S5). In this LRP/SM system, as in others, the expression of shoot development genes is therefore not strictly correlated with shoot regeneration (Cary et al., 2002; Motte et al., 2011).

Finally, it remains unclear whether organ identity markers are mutually exclusive. Do cells expressing either WOX5 or WUS coexist in the same converting organ? Can both genes be expressed simultaneously in the same cell? If such instances occur in the LRP/SM system, they must be short-lived. At the tissue level, the superposition of cell identity markers has already been observed. For example, when a RAM regenerates after root tip excision, cells recruited from multiple tissues in the remaining stump first converge to a common mixed identity, before dividing to form the columella and QC in the new apex (Efroni et al., 2016).

In the direct conversion system described herein, the lateral root primordium can be switched into a shoot only after the onset of WOX5 expression and only for a few hours. Furthermore, the misspecification of the QC strongly affects conversion (Kareem et al., 2015). A simple interpretation of these results is that the developing organ must first acquire a functional stem cell niche before the application of exogenous cytokinin. Accordingly, the niche is able to withstand the hormonal shock and continues to provide cells to sustain organ growth, root or shoot, whereas a precocious cytokinin treatment results in the terminal differentiation of cells in the bulging primordium in the absence of a stem cell niche.

However, the presence of a stem cell niche is not always required for organogenesis, suggesting that different regeneration mechanisms might occur in plants. Indeed, a new RAM can be formed after the excision of the root tip removing the apical stem cells, even in mutants (plt1, plt2 and scr) in which the niche cannot be maintained in the adult root (Sena et al., 2009; Sena, 2014). In such excision experiments, the regeneration of the root apical meristem starts with an embryonic-like sequence and terminates with the formation of an active stem cell niche, with stem cell-like divisions observed before the expression of cell identity markers (Efroni et al., 2016).

Models of RAM and SAM function describe how organ growth is shaped through complex interconnections between the constraints imposed by the encasing cell walls, the genetic programs driving cell expansion and cell division, and the signals exchanged between neighboring cells (Drisch and Stahl, 2015; Gaillochet et al., 2015). Logically, the RAM and SAM structural features might lock their identity. But our results highlight that organ identity remains undefined for several days after the initiation of shoot development or root reversion. Thus, the lock on organ identity probably resides in the final pattern of the meristematic tissues. It might be encoded in the tensile stress networks that characterize different cell arrangements as mechanical forces regulate key meristem factors including STM (Landrein et al., 2015). Alternatively, the size and shape of the lateral organ might alter the distribution of morphogens or the penetration of exogenous cytokinin, as well as other compounds, which control meristem functions.

In addition to morphological constraints, epigenetic factors might prevent the organ switch as the expression of plant genes involved in cell differentiation or totipotency depends in part on changes in the chromatin landscape (Berdasco et al., 2008; Grafi, 2004; Grafi et al., 2007; He et al., 2012; Ikeuchi et al., 2015; Koukalova et al., 2005; Lafos et al., 2011; Zhang et al., 2011). Our data confirmed that, together with additional regulators of shoot development (e.g. STM), WUS is not methylated in the competent lateral root and remains in that state throughout the early stages of conversion. Since methylation is generally associated with gene repression, an unmethylated WUS locus might reflect that the developing LRP is open to a change in identity, in this case induced by cytokinin (Li et al., 2011; Shemer et al., 2015). The hypothesis that switchable organs have a chromatin state favorable to conversion is further supported by the identification of genes similarly regulated during direct conversion (this report) and in hypomethylated mutants prone to shoot organogenesis (Li et al., 2011). Thus, key regulators of shoot initiation that are actively repressed in the fully developed root would remain inducible until repression is established. However, further analysis will be required to determine whether changes in chromatin marks, including DNA methylation but also histone modifications, occur earlier in LR development or later when organ identity is locked.

This study stresses that subtle differences in development timing can result in dramatically contrasted in vitro responses, explaining why regeneration protocols must always be optimized locally. In our view, the precise and microscopic analysis of the cultured plant tissues is the best guide when test-proofing regeneration methods.

In conclusion, the mechanisms at play in direct organ conversion are complex and draw on functional modules involved in other developmental pathways. Their further study in the original and – now – well-controlled conversion program might help us distinguish the factors that determine cell totipotency in a generic plant apical meristem and that are involved in establishing and locking organ identity.

All plants were grown in climate-controlled growth chambers at 24°C, under a 16 h day/8 h night photoperiod (Philips 36 W 840 neon, 110 μE m⁻² s⁻¹). Sterilized Arabidopsis thaliana seeds were sown, stratified for one night at 4°C, and grown for 6 days on solid medium. In explants prepared for morphological and marker line analysis, LR initiation was induced with NAA priming for 42 h. In explants sampled for RT-qPCR, transcriptomic and methylome studies, LR initiation was further synchronized by germinating and growing plantlets in the presence of NPA prior to NAA priming (see supplementary Materials and Methods for additional details, including composition of media and Arabidopsis reporter lines).

Propidium iodide staining and confocal microscopy analysis were adapted from Truernit et al. (2008). Developing lateral organs were imaged as stacks of confocal optical sections and their organization was analyzed in the sagittal plane reconstructed for each object. Stacks were reoriented according to the primary root main axis to define transverse and sagittal planes across the LRP center by 3D multi-planar reconstruction with OsiriX software (Pixmeo). LRP developmental stages were identified based on epidermal cell numbers and stele cell organization, analyzed in reconstructed sagittal planes. To measure conversion and reversion rates according to LRP stages, samples were photographed across developmental series (Zeiss Axio zoom stereo-microscope) and each LRP of each explant was tracked with ImageJ (NIH). Whole-mount in situ hybridization was performed according to H. Morin and A. Bendahmane (Institute of Plant Sciences Paris-Saclay, France). Additional details and GUS staining protocols are provided in the supplementary Materials and Methods.

For transcriptome analysis, lateral organs at each time point (T₀, T₆, T₃₄, T₅₈; Fig. 5G) were microdissected from fresh explants with the Zeiss AxioObserver PALM. Total RNA was extracted from pooled dissected organs with the RNeasy Plus Micro kit (Qiagen). RNA integrity was tested with the Agilent Bioanalyzer. Transcript profile analysis was performed on two independent biological experiments with the 25 K CATMA_v2.1 microarray bearing 24,576 gene-specific tags. Details about quantitative RT-qPCR analysis and transcriptome data interpretation are available in the supplementary Materials and Methods.

Genome methylation marks were analyzed through MeDIP-chip analysis in two independent biological experiments at T₀, T₆, T₃₄ and T₅₈, as described by Seifert et al. (2012). In these experiments, root segments were NPA-synchronized but not microdissected. Methylome patterns were compared with transcriptome datasets generated from the same explants.

The authors thank the reviewers for their insightful comments, P. Laufs (Institut Jean-Pierre Bourgin, Versailles, France) and B. Müller [University of Zürich (UZH), Switzerland] for their generous gift of reporter lines seeds, J. C. Palauqui and K. Belcram for their help with microscope imaging and analyses, J. D. Faure, P. Laufs and J. C. Palauqui for their critical reading of the manuscript.