Cell-fate plasticity, adhesion and cell sorting complementarily establish a sharp midbrain-hindbrain boundary

The concept of boundaries between gene expression domains is central and crucial to our current understanding of organ development because some of these boundaries also act as organizers or local signaling centers (Kiecker and Lumsden, 2005; Dahmann et al., 2011). Organizers are groups of cells that instruct other cells in their vicinity to acquire specific developmental fates and this fundamental process is essential for proper embryonic development (Arias and Steventon, 2018). The midbrain-hindbrain boundary (MHB), also known as the isthmic organizer (midbrain-hindbrain organizer), is an example of such an organizer that forms at the interface of the midbrain (mesencephalon, mes) and the hindbrain (cerebellum, metencephalon, met), and is vital for the formation and function of both the midbrain and the cerebellum (Rhinn et al., 2006; Gibbs et al., 2017). Cells at these signaling centers secrete various morphogens such as Wnt and fibroblast growth factors (FGFs), which provide ‘positional information’ and help establish proper tissue patterning and cell fate commitment (Gibbs et al., 2017).

In vertebrates, the interface between the expression domains of two transcription factors, Otx and Gbx, constitutes the MHB; in zebrafish, the MHB develops from the initially overlapping expression domains of Otx2 and Gbx1 that subsequently sort out and form a sharp boundary (Raible and Brand, 2004; Rhinn et al., 2003). During this process, morphogens such as Wnt and Fgf, and transcription factors such as engrailed 1, engrailed 2, Pax2, Pax5 and Pax8, sequentially induce MHB formation, and their subsequent interplay is crucial for the maintenance of the MHB (Rhinn and Brand, 2001; Wurst and Bally-Cuif, 2001; Raible and Brand, 2004; Rhinn et al., 2006; Dworkin and Jane, 2013). The above-mentioned factors (Otx, Gbx, Wnt1, Fgf8, Pax and Eng) make up the core of the MHB signaling machinery and any functional disruption of these factors interferes with patterning at the MHB (Dworkin and Jane, 2013; Gibbs et al., 2017).

Several studies using various vertebrate models have addressed whether the MHB also acts as a lineage restriction boundary, apart from its role as an organizer, and most studies argue in favor of the MHB also being a lineage restriction boundary (Zervas et al., 2004; Sunmonu et al., 2011). We have previously demonstrated the existence of a lineage restriction boundary at the zebrafish MHB using a combination of high-resolution time-lapse imaging, single-cell labeling and transplantation experiments (Langenberg and Brand, 2005; Langenberg et al., 2006). Our observations imply a cell-cell communication mechanism that restricts the migration of cells across the presumptive MHB but still permits cell movement within the group of cells on either side of the boundary. Thus, although it is known that Otx2 and Gbx1 establish their expression domains on either side of the MHB, the mechanisms by which this Otx-Gbx interface acts as a lineage restriction boundary, and the cell biological processes that prevent the intermingling of cells destined to dissimilar developmental fates, remain poorly understood.

Therefore, we have addressed these issues using the developing zebrafish as a model. Specially, we visualized the establishment of the MHB in real time by generating various novel fluorescent reporter and Cre driver lines for otx2b, wnt1, gbx1, gbx2 and fgf8a using CRISPR/Cas9-mediated knock-in strategies (Kesavan et al., 2017, 2018). Using a combination of time-lapse imaging and Cre/lox-mediated lineage tracing, we show that lineage restriction at the MHB occurs by the end of gastrulation. Next, we demonstrate that cells indeed sort at the MHB to establish a sharp boundary and that the underlying molecular mechanisms include differential adhesion between prospective midbrain and hindbrain progenitors, with N-cadherin involvement and signaling through the Eph-ephrin pathway.

The gene expression boundary abutting the otx2b-gbx1 expression domains demarcates the MHB primordium. In order to visualize these developmental events in real time, we generated CRISPR/Cas9-based knock-in zebrafish lines. The coding sequences for the reporters were knocked-in upstream of the corresponding ATG and were, thus, under the control of the endogenous promoter/enhancer elements (Fig. 1A). Importantly, as shown previously, reporter expression faithfully reproduces endogenous gene expression patterns (Kesavan et al., 2017, 2018).

To address Otx-Gbx boundary formation in real time, we used time-lapse imaging of otx2b:HRAS-mKate2; gbx1:Venus double transgenic embryos wherein cell membranes of the midbrain cells and hindbrain cells themselves were labeled, respectively. However, this process could only be visualized from tail bud stage onwards (10 h post fertilization, hpf) because of the time required for the reporter (mKate2 or Venus) to become functional, despite being fast-folding proteins. Nonetheless, an overlap in the expression of otx2b (midbrain, mKate2) and gbx1 (hindbrain, Venus) can be observed at the neural plate during the one- to two-somite stage (10.5-11 hpf), as evidenced by the presence of cells that are double positive for both mKate2 and Venus (Fig. 1B). Importantly, these cells could only be observed close to the prospective MHB and were not present in the deeper cell layers of the expression domains of either otx2b or gbx1 (Fig. 1B). Subsequently, around the 10-14 somite stage (15-16 hpf), this intermingling of double-positive midbrain-hindbrain cells resolved into a sharper boundary (Fig. 1B,C; Movie 1). Overlapping expression domains were completely absent by around 30 hpf. Additionally, at this time point, the double-positive cells, along with gaps that represent otx2b-derived cells within hindbrain domain, were also not visible (Fig. 1B). Together, these observations imply a sequence of events during MHB formation wherein an initial overlap in the expression domains of Otx and Gbx resolves into a sharp, non-overlapping boundary over time. The process by which such initially overlapping gene expression domains segregate to form sharp boundaries can involve multiple mechanisms, such as loss of cell identity (cell-fate plasticity) or cell sorting (Batlle and Wilkinson, 2012). Hence, we next investigated whether these mechanisms play a role in MHB formation.

To understand changes in cell identity, i.e. the establishment of lineage restriction at the midbrain-hindbrain boundary (MHB), we attempted two types of lineage tracing: short-term or transient labeling; and long-term or permanent tracing. We used both approaches in order to derive additional information on potential identity change or fate plasticity in the cells abutting the MHB. First, for short-term tracing, CRISPR/Cas9-mediated knock-ins of fluorescent reporters for multiple genes expressed at the MHB, i.e. otx2b, wnt1, gbx1 and gbx2, were generated (Fig. 2A). Next, to follow the fate of Otx2- and Wnt1-expressing cells at the MHB, these fluorescent reporter lines were tagged with a nuclear localization signal (NLS) to generate otx2b:Venus-NLS and wnt1:Venus-NLS fish. Venus was used as the reporter because it is fast folding, which allows visualization of the early events during MHB formation. Moreover, it has a known lifetime of about 24 h (Li et al., 1998; Snapp, 2009), such that the ‘perdurance’ of the Venus protein, even when the promoter is turned off, can be used as a lineage-tracing marker.

In situ hybridization at 24 hpf showed that otx2b and wnt1 mRNA expression was restricted to the midbrain region with a sharp boundary abutting the MHB (Fig. 2B,C). In contrast, at 24 hpf, in the otx2b:Venus and the wnt1:Venus reporter embryos, the presence of Venus fluorescent protein (perdurance) was observed in the hindbrain domain, albeit with a weaker intensity than in the midbrain domain (Fig. 2D-E′), suggesting that these cells had indeed expressed otx2b and wnt1 at an earlier time point during MHB development (Fig. S1). Furthermore, time-lapse imaging of otx2b:Venus-NLS and wnt1:Venus-NLS embryos not only showed the presence of Venus-positive cells in the hindbrain domain, but also that these cells remained within this domain over time (Movies 2 and 3). Taken together, these observations imply that cells abutting the MHB are indeed capable of changing their gene expression patterns and that they may display cell-fate plasticity.

Next, to verify whether a similar mechanism also operates in hindbrain cells, hindbrain boundary markers like gbx1 and gbx2 were lineage traced using gbx1:Venus and gbx2:Venus-NLS reporter lines. In situ hybridization at 24 hpf for hindbrain markers, i.e. gbx1 and gbx2, showed expression boundaries posterior to rhombomere 1 for gbx1 and at the MHB for gbx2 (Fig. 2F,G). In contrast, at 24 hpf in the reporter fish, the presence of Venus fluorescent protein was observed in the midbrain domain, albeit with relatively weaker fluorescence intensity than that seen in the hindbrain domain (Fig. 2H-I′). Furthermore, as shown in Movie 1, these gbx1:Venus-positive cells were also otx2b positive and remained in the midbrain domain. Thus, these observations point to a pattern of perdurance of hindbrain markers (gbx1) in the midbrain domain and of midbrain markers (otx2b and wnt1) in the hindbrain domain.

Next, we quantified this perdurance and show that, on average, 19 otx2b:Venus-NLS-derived cells and 9 wnt1-derived cells per embryo were found in the hindbrain domain (Fig. 2J), i.e. greater numbers of otx2b-derived cells showed perdurance than did wnt1-derived cells. Similarly, although there were 15 gbx-1-derived cells in the midbrain, there were only four gbx2-derived cells (Fig. 2K). To understand why there may be more otx2b than wnt1 cells, or more gbx1 than gbx2 cells, we evaluated gene expression onset by whole-mount in situ hybridization. The results revealed that otx2b expression occurs earlier than that of wnt1 (6 hpf versus 10 hpf); likewise, gbx1 expression occurs earlier (early onset) than that of gbx2 (6 hpf versus 10 hpf) (Fig. S2), indicating that genes with earlier onset of expression may show greater plasticity compared with those showing later onset. Furthermore, these observations, when combined with the pattern of perdurance seen above, imply that lineage restriction occurs at the end of gastrulation in both midbrain and hindbrain cells.

Next, to substantiate that lineage restriction occurs at the end of gastrulation, we permanently labeled cells in the MHB region by generating CRISPR/Cas9-based CreER^T2 knock-in lines and crossing them with a zebrabow responder line (Pan et al., 2013). Zebrabow-based lineage tracing allows stochastic recombination, and the resultant fluorescent protein combinations provide an opportunity to understand lineage decisions and clonal origins with high temporal resolution. CreER^T2-mediated recombination was induced with 4-hydroxy tamoxifen, administered at 6 hpf and/or 24 hpf, and all embryos were imaged at 48 hpf (Fig. 3A, schematic representation). In the otx2b:CreER^T2 line combined with the zebrabow line, several recombined cells were observed in the hindbrain domain when 4-hydroxy tamoxifen was used at 6 hpf. In contrast, no recombined cells were seen upon 4-hydroxy tamoxifen induction at 24 hpf (Fig. 3B-C″). Similarly, the fgf8a:CreER^T2 crossed with the zebrabow line revealed the presence of fgf8a-derived cells in the midbrain with 4-hydroxy tamoxifen induction at 6 hpf (Fig. 3D-D″). 4-Hydroxy tamoxifen induction at 24 hpf in this line did not yield a visible readout, probably because of low recombination efficiency and low Fgf8 expression (data not shown). These permanent lineage tracing results not only concur with the perdurance data described above, but also suggest that (1) lineage restriction indeed occurs at the end of gastrulation; and (2) there must be other mechanisms, such as cell sorting, that serve to subsequently establish and maintain sharp expression boundaries, i.e. from the tail bud stage onwards.

We looked at cell sorting as a possible mechanism of MHB formation and maintenance after lineage restriction was established at the tail bud stage. To track boundary cells with greater sensitivity in real time, we live imaged wnt1:Venus-NLS fluorescent reporter fish from the 1- to 2-somite stage until the 6- to 7-somite stage (10.5-12 hpf). We used Wnt1 as the marker because it accurately identifies midbrain boundary cells and, as mentioned earlier, imaging before 10 hpf was not possible due to the time required for reporter maturation. Using z-stacks of confocal images acquired over a narrow time interval (every 150 s), we were able to track individual cells (using labeled nuclei) with high spatial and temporal resolution (schematic representation of the cells tracked by time-lapse live imaging, Fig. 4A). All cells migrated only anteriorly and individual cell tracking of the wnt1 cells showed zig-zag movements and crossovers among these cells (Fig. 4B-F and Movie 5). Furthermore, from around 14 hpf, when the neural rod has formed, cell movement was restricted and only inter-kinetic nuclear migration is observed (Movie 3), rather than active whole-cell movement. Interestingly, we could identify individual cells that were initially distant from the group of wnt1:Venus-NLS-positive boundary cells migrating anteriorly (Fig. 4B-E, arrowheads). We tracked the movement of one such straggling wnt1:Venus-NLS-positive cell and found that this cell actively migrated towards the group of other boundary cells (Fig. 4B-E, Movie 4). Such active migration of straggling wnt1 cells was observed in multiple embryos (in five out of eight movies imaged; an additional time-lapse movie depicting such behavior is shown in Movie 6). We have also quantified the proportion of wnt-1-positive cells at a distance that sorted back to the midbrain (MB) domain. Enumerating such cells from time lapse movies (n=5) showed that about 50% of cells sorted back to the MB domain (51.6±11.9, mean±s.e.m.) (Table 1). Thus, it appears that cell sorting during MHB formation involves active cell mixing and migration of cells both within and between brain compartments, both of which may contribute to establishing sharp gene expression boundaries during neural rod formation.

Proposed mechanisms of cell sorting include the differential tension hypothesis, the differential adhesion hypothesis and the repulsion hypothesis (Batlle and Wilkinson, 2012); here, we investigated whether differences in adhesion contribute to cell sorting at the MHB. Specifically, we used atomic force microscope-based single cell force spectroscopy (AFM-SCFS) (Fig. 5A, schematic representation of SCFS; Krieg et al., 2008) to determine potential differences in cell-cell adhesion properties of prospective midbrain and hindbrain cells.

Individual progenitor populations were isolated from various knock-in reporters such as the otx2b:Venus (for midbrain cells), gbx1:Venus (for hindbrain cells) and dusp6:d2eGFP transgenic line (for hindbrain cells) (Fig. 5B,C). Embryos were dissociated at the tail bud stage (10 hpf) and a single-cell suspension for AFM-SCFS was made by mechanical trituration (Fig. 5D,E). Measurements of the adhesive strength of progenitors of the same kind (homotypic adhesion or cohesion) showed that the midbrain cells have greater cohesion than hindbrain cells at both contact times tested (5 and 10 s; Fig. 5F). In contrast, adhesive forces between different cell types (heterotypic adhesion) were significantly lower than the cohesive force between two midbrain cells, but comparable with homotypic cohesion in hindbrain cells (Fig. 5G). These data imply the presence of differential adhesion between midbrain and hindbrain cells.

Calcium-dependent adhesion molecules (cadherins) are known to be essential for cell-cell adhesion during embryonic development. Therefore, to determine whether calcium is required for cohesion, we used EGTA to chelate calcium in the media and measured cohesion using AFM-SCFS. Calcium chelation dramatically reduced cohesive strength between midbrain cells and hindbrain cells, suggesting that cohesion is calcium dependent (Fig. 5H).

Next, as N-cadherin (N-cad, cdh2) is the major mediator of calcium-dependent adhesion, we used morpholinos to knockdown N-cad and demonstrate that the absence of N-cad led to a significant reduction in cohesive strength (Fig. 5I). Taken together, these findings suggest that the observed differential adhesion between midbrain and hindbrain cells requires calcium, and that it is primarily mediated by N-cad in the developing neural plate.

To substantiate the above-described role of N-cad in cell-cell adhesion at the MHB, N-cadherin (cdh2) expression was perturbed by two CRISPR/Cas9-mediated knockout approaches. In the first approach, global N-cad mutants were generated by injecting 2 sgRNAs targeting exon 1 and 2, along with Cas9 mRNA, in one-cell embryos. PCR genotyping showed efficient deletion and ‘crispants’ displayed phenotypes consistent with previously described cdh2 mutants at 24 hpf (Jiang et al., 1996; Lele et al., 2002). Importantly, in situ hybridization showed disorganized gene expression boundaries for otx2b and egr2b in these mutants. In contrast, control fish showed clearly demarcated boundaries for these markers at 24 hpf. Additionally, in the N-cad mutants, individual otx2b-positive cells were visible outside their expression domain; again, such cells were absent in the control fish (Fig. 6A,B). In the second approach, N-cad was conditionally ablated in the otx2b domain using Cre/lox-controlled Cas9 combined with the otx2b:CreER^T2 driver. This conditional perturbation of N-cad in the midbrain also resulted in irregular gene expression boundaries for otx2b, along with the presence of otx2b-positive cells outside the endogenous otx2b expression domain at 10 hpf (Fig. 6C,D). Thus, both targeted and global N-cad deficiency yielded very similar phenotypes. Taken together, these results imply that N-cad is necessary for establishing a sharp expression boundary for Otx2.

The Eph/ephrin signaling pathway is a known regulator of cell and tissue segregation during various stages of embryonic development (Xu et al., 1999). Importantly, it has been shown that the Eph receptor EphB4a is expressed in the midbrain (Cooke et al., 1997), while the ligand Efnb2a is expressed in a complementary manner in the hindbrain (rhombomere 1) (Cooke et al., 2005); we have observed the same expression pattern from late gastrulation stages (8-10 hpf). Therefore, to test whether Eph-ephrin signaling plays a role in cell sorting at the MHB, we expressed a truncated soluble form of Efnb2a (sol-efnb2a), which competes with various endogenous ephrin ligands that bind to Eph receptors, thereby disturbing both forward and reverse Eph/ephrin signaling (Cavodeassi et al., 2013; Cooke et al., 2001). Therefore, sol-efnb2a mRNA was injected into one-cell embryos of the otx2b:HRAS-mKate2, gbx1:Venus double transgenic line to monitor the Otx-Gbx gene expression boundary. Perturbed Eph-ephrin signaling resulted in mis-sorting of cells across the MHB and the presence of single otx2b-positive cells in the hindbrain. Additionally, otx2b-positive and gbx1-positive cells were seen distributed further from the Otx-Gbx overlapping domain compared with control embryos (Fig. 6E, Movies 7 and 8). These observations were confirmed by in situ hybridization for otx2b and egr2b, which revealed the presence of otx2b-positive cells outside their expression domain in the sol-efnb2a-injected embryos (Fig. 6F).

Finally, we used Cre/lox-based permanent labeling to understand the effects of disrupted Eph/ephrin signaling in lineage restriction and cell sorting at the MHB. Specifically, fgf8a:CreER^T2 fish were crossed with the Tg(hsp70l:loxP-DsRed-loxP-EGFPNLS) to not only visualize sorting defects, but also enumerate the number of cells that have mis-sorted. Embryos exposed to 4-hydroxy-tamoxifen at 6 hpf and imaged at 36 hpf revealed the presence of fgf8a-derived cells in the midbrain. Notably, when s-efnb2a mRNA was injected in these embryos at the one-cell stage, the number of such fgf8a-derived cells in the midbrain was higher compared with control fish (Fig. 6G,H). These results indicate that Eph-ephrin signaling is involved in cell sorting at the MHB.

Recently, actomyosin and Yap-mediated mechanisms have also been shown to be essential for maintaining rhombomere boundaries in zebrafish (Calzolari et al., 2014; Voltes et al., 2019). However, our analysis with F-actin reporters and active Yap signaling reporter did not reveal any involvement during MHB formation (Fig. S3), probably because both actomyosin-mediated tension and Yap appear much later at the hindbrain boundaries (Cayuso et al., 2019; Voltes et al., 2019). Therefore, these results suggest that, at the MHB, N-cad mediates cell-cell adhesion and that Eph/ephrin signaling is involved in cell sorting, both of which serve to establish and maintain sharp gene expression boundaries at the MHB.

Using multiple transgenic reporter fish for lineage tracing and live imaging, we show that lineage restriction and the formation of sharp gene expression boundaries in the developing MHB occur through multiple complementary mechanisms, namely cell-fate plasticity and cell sorting, and that these processes involve differential adhesion, N-cad and Eph-ephrin signaling (Fig. 7).

Previously, in zebrafish, using RNA in situ hybridization, we found that, at 60% epiboly (6 hpf), the anterior border of the gbx1 expression domain directly abuts the otx2b domain, with an overlap of the two domains covering three or four cell layers; this structure subsequently resolves into sharply defined non-overlapping adjacent segments (Rhinn et al., 2003). Owing to technical limitations (fixed samples) and the absence of reporter lines, these overlapping and segregation events could not be visualized by live imaging or be lineage traced to follow their fate in these earlier studies. Here, using various knock-in fluorescent reporters and Cre driver lines to visualize and follow cell fate, we show that cell-fate plasticity does occur across the gene expression domains abutting the developing MHB. This phenomenon predominantly occurs during early gastrulation stages (6-8 hpf) and reduces by the end of gastrulation (10 hpf). A possible explanation for this phenomenon is as follows. Morphogens such as Wnt (in the midbrain domain) and Fgf (in the hindbrain domain), transcription factors such as engrailed 1, engrailed 2, Pax2, Pax5 and Pax8 (across the MHB), and cell-adhesion molecules, including Eph/ephrin (present in adjacent domains), are expressed only between 8 and 10 hpf. Thus, the presence of such a multitude of factors can increase cellular complexity and lead to reinforced fate commitment and lineage restriction.

Comparable results from other studies support this explanation. For example, a recent study on zebrafish hindbrain development demonstrated cell identity switching through a mechanism involving segment identity and retinoic acid signaling, with intermingling between segments and consequent cell-identity changes occurring during early stages of rhombomere segment formation, i.e. before the establishment of robust Eph/ephrin signaling that causes cell segregation across rhombomere boundaries (Addison et al., 2018). Likewise, using a combination of cell transplantation, iontophoretic cell labeling and live-cell imaging, we found that lineage restriction occurs at the MHB during the late gastrulation stages in zebrafish (Langenberg and Brand, 2005; Langenberg et al., 2006), with similar observations being reported in chick and mice (Zervas et al., 2004; Sunmonu et al., 2011; Tossell et al., 2011). Similarly, cell lineage analysis during hindbrain development in chick has shown that the clonal progeny of cells labeled before the establishment of morphological boundaries display plasticity in cell-fate specification; in contrast, clones labeled after boundary formation are confined to their respective segments (Fraser et al., 1990).

Interestingly, we have noted a consistent pattern in our lineage-tracing results, i.e. whereas gbx1-postive cells in the midbrain domain are always also otx2b positive (double positive) and are relatively fewer in number, otx2b-positive cells in the hindbrain domain are rarely gbx1 positive and are relatively greater in number. Although there are a few potential explanations, currently, there are no known molecular mechanisms that can adequately explain this observation.

Between 10 and 12 hpf, when the neural plate transforms into the neural keel, convergent extension occurs, involving extensive cell intercalation, cell division and intermingling of cells along the A-P axis (Kimmel et al., 1994), suggesting that mechanisms other than cell-fate switching may exist to maintain sharp expression boundaries. Consistently, using live imaging, we show that cells actively sort at the MHB between 10 and 12 hpf, and that this sorting is influenced by N-cad-mediated adhesion between the midbrain and hindbrain cells. These findings are consistent with previous data on the role of N-cad in cell convergence and maintenance of neuronal positioning during vertebrate neural tube development (Lele et al., 2002).

Over the years, three major classes of cell segregation mechanisms during boundary formation have been uncovered. The first is based on differential cell-cell adhesive property (differential adhesion hypothesis) that establishes interfacial tension across boundaries (Steinberg, 2007). The second is actomyosin-based establishment of cortical tension (differential tension hypothesis) at the boundaries (Harris, 1976), and the third involves cell-cell repulsion mediated by Eph/ephrin-like signaling molecules. These mechanisms, either individually or in combination, have been shown to establish and maintain boundaries in different tissues during development (Batlle and Wilkinson, 2012). Furthermore, in the developing spinal cord of zebrafish, it has been recently shown that a heterogeneous population of neuronal progenitors induced by sonic hedgehog signaling sort to rearrange and form sharply bordered domains, and that this sorting mechanism is mediated via N-cadherin (Xiong et al., 2013). Here, using an AFM-based assay to measure cell-adhesion properties, we show that the prospective midbrain and hindbrain cells indeed have differential adhesive strengths and that adhesion between a midbrain and a hindbrain cell is lower than that of a midbrain-midbrain cell combination. Time lapse movies of MHB formation show extensive cell movement and intermingling during which cells constantly change their partners. Intuitively, such cell behavior would require differential adhesion because this intermingling and subsequent sorting leads to compartmentalization of similar cells that ultimately form a pattern.

Additionally, our data also reveal that the cell-adhesion molecule N-cadherin is involved in contributing to this adhesion. Thus, in a scenario where adhesion is disrupted due to N-cad mutations, it is expected that cells will mis-sort, and we show that conditional N-cad mutants indeed display sorting defects at the MHB. These observations highlight the importance of differential adhesion during boundary formation at the MHB.

Interactions between Eph receptor tyrosine kinase and its ligand ephrin are known to control multiple cell biological processes such as polymerization of actin cytoskeleton, cadherin function and integrin-mediated adhesion (Batlle and Wilkinson, 2012). Several Eph receptors and their ligands are expressed in complementary domains in multiple tissues during development (e.g. in the developing rhombomere boundaries), and any perturbation in the signaling leads to cell intermingling between adjacent segments (Xu et al., 1999). However, loss-of-function studies using Eph/ephrin mutants are complicated due to redundancy among the many receptors and ligands expressed in the same tissue/cell type in vivo (Bush and Soriano, 2012). Nonetheless, using a soluble version of the ligand Efnb2a to block a wide range of Eph/ephrin bidirectional signaling, we find that Eph/ephrin signaling plays a significant role in establishing sharp gene expression boundaries. Further studies are required to identify the specific molecular combination of Eph receptors and ligands that mediate cell sorting at the MHB. Notably, individual perturbations of N-cad and Eph/ephrin signaling did not yield a comprehensive sorting phenotype in any of the embryos (i.e. it is never 100%), suggesting that both differential adhesion and Eph/ephrin pathway are required for sharpening the boundary at the MHB. Interestingly, previous studies and subsequent concept reviews have posited that Eph-ephrin signaling may drive cell segregation through heterotypic tension/repulsion by interplay with Ca²⁺-dependent cell-adhesion molecules like cadherins (Rohani et al., 2014; Winklbauer, 2015), and that differential expression of N-cad may not be necessary as changes in activity mediated by Eph/ephrin signaling or heterotypic tension alone are sufficient to facilitate cell segregation (Canty et al., 2017; Taylor et al., 2017). Thus, further investigations into the mechanical aspects of how Eph/ephrin signaling can modulate cadherin function and repulsion/tension at the cell-cell interface will provide greater clarity on this extremely complex process.

In summary, using novel fluorescent knock-in reporters, live imaging, Cre driver-based lineage tracing, and cutting-edge cell adhesion assays and mutant analysis, we describe the process of lineage restriction at the MHB. It occurs through multiple complementary mechanisms, which are, in sequence, cell-fate specification, lineage restriction and cell sorting; the last involving differential adhesion, N-cad and Eph-ephrin signaling.

Zebrafish (Danio rerio) embryos and adults were obtained, maintained and raised as described previously (Brand et al., 2002; Westerfield, 2000). Embryos were staged as hours post fertilization (hpf) (Kimmel et al., 1995). The wild-type strain AB was used to obtain knock-in lines, and transgenic fish lines were maintained as outcrosses. None of the larvae or adult fish showed any physiological or behavioral abnormalities.

All animal experiments were carried out in accordance with animal welfare laws of the Federal Republic of Germany (Tierschutzgesetz) that were enforced and approved by the competent local authority (Landesdirektion Sachsen; protocol numbers TVV21/2018; DD24-5131/346/11 and DD24-5131/346/12; TVT1/2019), by the institutional animal welfare committee (Tierschutzkommission der Technische Universität Dresden) and in accordance with EU directives (Directive 2010/63/EU).

The knock-in lines for various genes expressed at MHB were generated and maintained as previously described (Kesavan et al., 2017, 2018). The target site sequence and primers for generating baits are provided in Table S1. Transgenic animals used in this study are listed in Table S2.

mRNA for HRAS:mKate2 or soluble-efnb2a were prepared using mMessage mMachine Kit (ThermoFisher). 100 pg of mRNA in a 1 nl volume was microinjected into one-cell stage embryos (wildtype AB). For N-cadherin morpholino experiments, 0.5 pmol/embryo was injected in a 1 nl volume (wildtype AB). Plasmids and morpholinos used in this study are listed in Table S3.

For generating a global N-cadherin mutant, a 1 nl of solution containing 25 ng/µl of the two sgRNAs (Target sequence in Table S1) and 150 ng/µl of Cas9 mRNA was injected into one-cell stage embryos (wildtype AB). PCR genotyping of embryos absolutely correlated with the mutant phenotype and the two sgRNA showed high efficiency in cutting the target site.

For generating the N-cadherin conditional mutants, 1 nl of solution containing 25 ng/µl of the two sgRNAs and 25 ng/µl of circular plasmid DNA (Hsp70l:loxp_nonFP_loxp_Cas9eGFP) were injected into one-cell stage embryos obtained from otx2:CreET^T2. 4-OH tamoxifen (5 µM) induction was carried out at 4 hpf and embryos were heat shocked at 8 hpf for 30 min at 37°C. Embryos were sorted for eGFP fluorescence at 10.5 hpf and fixed for further analysis using in situ hybridization.

Embryos at specific developmental stages were derived by crossing wild-type fish (Ab strain). They were fixed in 4% PFA, stored in 100% methanol at −20°C and whole-mount in situ hybridization was performed as described elsewhere (Kesavan et al., 2017; Reifers et al., 1998). Briefly, using a RNA labeling and detection kit (Roche), digoxigenin (DIG)-labeled probes were synthesized from linear DNA and the hybridized probes were detected using anti-digoxigenin antibody. Antibody staining was visualized using BM purple (digoxigenin). In situ probe staining for otx2, gbx1 and gbx2 (Rhinn et al., 2003), and wnt1 (Lekven et al., 2003) matched expression patterns described previously.

For live imaging, early somite stage embryos (10 to 12 hpf) were embedded in 0.7% low melting agarose. Late somite stage embryos (15 hpf onwards) were treated with 1-phenyl-2-thiourea (PTU) to block pigmentation, and with MS-222 for anesthesia then were mounted on a glass-bottomed dish (MatTek) in 1% low melting agarose. All embryos were imaged on a Zeiss LSM 780. Images were analyzed using FIJI (open source software) or Imaris (v7, Bitplane), respective TIFF files were generated and figures assembled in Adobe Photoshop (version CS5 or CS6). For time-lapse imaging, tissue sections spanning 30-40 µm, with a z interval of 1 µm, were imaged every 2 min and 30 s at 28°C on the LSM780 (Zeiss) microscope. The duration of time-lapse imaging was different for each imaging experiment and is mentioned in the legend of the respective figures. Maximum intensity projections of fluorescence and transmitted light images were generated using Imaris (v7, Bitplane), FIJI (Schindelin et al., 2012 ) or Arivis 4D. Cell tracking was carried out using FIJI (Trackmate).

The Cre drivers otx2b:CreER ^T2 and fgf8a:CreER ^T2 fish were crossed with the zebrabow responder line Tg(ubb:lox2272-loxp-RFP-lox2272-CFP-loxp-YFP) (Pan et al., 2013) or with Tg(hsp70l:loxP-DsRed-loxP-EGFPNLS) (Knopf et al., 2011). Embryos were treated with 1 μM 4-OH-tamoxifen (4-OHT) at 6 hpf for 12 h or with 10 μM 4-OH-tamoxifen at 24 hpf for 12 h. Control embryos were left untreated. All embryos were live imaged at 48 hpf as described above. In the zebrabow responder, a cell with multiple copies of RFP, CFP and YFP, CreER^T2-mediated stochastic recombination events lead to cell clones being labeled by different colors. The combinatorial nature of fluorescent protein expression gives a unique barcode to each cell and marks their progenies with the same color. Untreated embryos show only default RFP expression, i.e. they represent non-recombined cells.

The AFM-SCFS was performed as previously described (Krieg et al., 2008). AFM-SCFS is a high-precision force measuring tool that measures the difference in adhesion strength between two cells by initially bringing two isolated cells into contact. Next, after a given contact time, the cells are separated and the force required to separate them is quantified. All experiments and data analyses used the Nanowizard I AFM setup (JPK Instruments). Briefly, cantilevers (Nano world TR-TL-Au-20, nominal spring constant k=32 mN m^–1) that were coated with concanavalin A (ConA, 2.5 mg/µl, Sigma) were used. Two regions were marked in a glass-bottomed dish; one was coated with bovine serum albumin (1%) to obtain a non-adhesive substrate, while the other was coated with ConA (2.5 mg/µl) to obtain an adhesive substrate. Both substrates were gently rinsed with the 1×HBSS (Life Technologies) before the experiment.

Transgenic embryos staged between 10 and 10.5 hpf were carefully dissected to isolate midbrain or hindbrain progenitors, and a single-cell suspension was obtained by trituration that was diluted in 1×HBSS and seeded onto the substrate. For both homotypic and heterotypic adhesion experiments, cells were identified using fluorescence microscopy, whereas for heterotypic adhesion experiments, fluorescent dextran, labeled with Cy5, was injected into only one set of embryos to distinguish between the two cell types. A given ‘probe’-cell was selected from the non-adhesive side of the substrate with a ConA-coated cantilever by gently pressing on it with a controlled force of 1 nN, typically for 1 s. The cell was removed from the surface for 2-10 min to allow it to firmly attach to the cantilever. The probe cell was then moved above a ‘target’ cell that was firmly attached to the adhesive (ConA-coated) part of the substrate. Adhesion experiments (‘force-distance cycles’, see Fig. 1A) were performed at contact force of 1 nN, 10 μm s^–1 approach and retract velocities, and contact times of either 5 or 10 s. Each condition (i.e. same probe-target couple at the same contact time) was repeated up to three times, with a resting time of 30 s between successive contacts. Each probe cell was used to test several target cells and no more than 12 curves were obtained with any given probe cell. Cells were observed continuously during and between the force-distance cycles to judge whether they were intact and were firmly attached to the cantilever or substrate. Force-distance curves were derived and pooled data were used for statistical analysis (Graph pad prism, v8). Cadherin dependence of cell adhesion was tested after either depleting calcium by adding EGTA (5 mM, Sigma) to the medium, or by injecting embryos with morpholino oligonucleotides for N-cadherin (2 ng, Genetools).

Data are presented as mean±s.e.m., unless specified otherwise. A two-tailed, unpaired t-test was used to calculate statistical significance at P<0.05 (GraphPad prism, v8.0).

We are thankful to the Chen and Wente labs for providing plasmids to generate Cas9 and sgRNA mRNA (via Addgene); to Jens Ehrig (Molecular Imaging and Manipulation Facility, CMCB, Dresden), Jens Friedrichs, the Heisenberg lab and the Guck lab for help with various AFM experiments; to past and present members of the Brand lab for discussions; to Pablo Sanchez Quintana for help with imaging; and to Dr Vasuprada Iyengar for language and content editing. We thank Marika Fischer, Jitka Michling and Daniela Mögel for dedicated zebrafish care. The Light Microscopy Facility at CMCB also supported this work.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.186882.reviewer-comments.pdf