HoxB genes regulate neuronal delamination in the trunk neural tube by controlling the expression of Lzts1 Free

Hox genes encode highly conserved homeodomain (HD) transcription factors essential to promote morphological diversification of the bilaterian body (Rezsohazy et al., 2015). In higher vertebrates, including human, mouse and chicken, 39 Hox genes specify the regional identity of body structures, including the axial skeleton, nervous system, limbs, genitalia, and the intestinal and reproductive tracts (Crawford, 2003). Hox genes are organized in four clusters located on different chromosomes named HoxA to HoxD, and in 13 paralog groups. Members of each paralog group, further classified in anterior, central and posterior classes, are deployed in ordered spatial and temporal patterns along the antero-posterior (AP) axis (Duboule, 2007; Duboule and Dollé, 1989; Iimura and Pourquié, 2006; Pearson et al., 2005). Genes located 3′ in a cluster are expressed earlier and more rostral than genes located more 5′. The correlation between the genomic organization and the spatio-temporal characteristics of Hox gene expression along the AP body axis is referred to as ‘collinearity’. Importantly, genes of the same paralog group, including in distant species, display higher sequence conservation and regulatory properties than different paralogs within the same species. Together with the collinear expression, this results in the deployment of distinct regulatory activities in distinct spatial territories, allowing for morphological diversification.

The expression and function of Hox genes in the developing spinal cord of vertebrates (the trunk neural tube) illustrate how Hox collinear expression generates morphological diversification. From a functional point of view, experimental evidence showed that Hox genes define the identity and synaptic pattern of neurons, setting distinctive features necessary for the building of locally distinct motor circuits ultimately controlling diverse functions such as locomotion or respiration (Dasen et al., 2003, 2008; Lacombe et al., 2013; Sweeney et al., 2018). Hox gene functions in the trunk neural tube include the segregation of motor neuron columns: the LMC (lateral motor column) in a ventrolateral position at limb levels (brachial and lumbar levels), and the PGC (preganglionic motor column) and HMCs (hypaxial motor neurons) at the thoracic level. This neuron segregation according to their final functions is essential for subsequent functional organization of the spinal cord.

Previous studies provided an extensive view of Hox gene expression in the chicken and mouse trunk neural tube (Dasen et al., 2005; Jung et al., 2010; Lacombe et al., 2013). The expression of HoxA, HoxC and HoxD genes displays a pronounced axial collinearity, with in most cases a preferential accumulation in motor neuron territories after the onset of neuronal differentiation, which suits the described pattern of activity of Hox genes in promoting motor neuronal diversification.

Available data on HoxB genes suggest expression with distinct levels of axial collinearity. HoxB genes start to be expressed very early in the embryo (in the epiblast/tailbud), and present a temporal collinear onset of expression: HoxB1 and HoxB2 at HH4; HoxB3 to HoxB6 at HH5; HoxB7 at HH6; HoxB8 and HoxB9 at HH7; and HoxB13, which is only expressed in the tail bud, at HH20 (Denans et al., 2015). In the trunk neural tube of chicken embryo from HH4 to HH17, HoxB genes can be split into two groups: HoxB1 to HoxB5 are expressed up to the otic vesicle but not in the caudal part, whereas HoxB6 to HoxB9 are expressed in the caudal part of the neural tube (Bel-Vialar et al., 2002). The exhaustive view of the expression of Hox genes and proteins much later (at E6) in the trunk neural tube shows, in contrast, that at later stages the expression of all HoxB genes is highly overlapping along the antero-posterior axis: HoxB3 to HoxB9 genes are all expressed at the brachial, thoracic and lumbar levels (Dasen et al., 2005). In addition, at E6, Hox from the B cluster display an almost complete absence of expression in motor neurons, where Hox genes from the other clusters (HoxA, HoxC and HoxD) display a strong collinear expression to specify columnar and pool subtypes (Dasen et al., 2005). Although less comprehensive, data in mouse are also consistent with B cluster Hox genes displaying characteristics of expression distinct from non-B cluster Hox genes (Graham et al., 1991; Jung et al., 2010; Lacombe et al., 2013). These observations indicate a different spatial deployment of B cluster Hox genes. Interestingly, HoxB8 protein is present in chicken neural tube progenitors (Asli and Kessel, 2010), i.e. long before the expression of non-B cluster Hox proteins. Early transcription at the progenitor stage of non-B cluster Hox genes has been described, but their proteins are either weakly expressed or undetectable, with proteins observed only in postmitotic neurons (Dasen et al., 2003). Therefore, HoxB genes may have earlier functions than non-B Hox genes in the trunk neural tube development.

The trunk neural tube is a pseudostratified epithelium that will sequentially give rise to a large variety of neurons and glial cells of the spinal cord. After an initial phase of proliferation resulting in the expansion of progenitors (P) by symmetrical divisions (P-P), neurogenesis and then gliogenesis are achieved via a succession of steps that follow a stereotypic temporal order. Concomitantly, progenitors become committed to differentiate into a specific neuronal (and later on glial) subtype according to their dorso-ventral position (Le Dréau and Martí, 2012). Postmitotic neurons (N) are produced by asymmetric (P-N) or symmetric (N-N) terminal divisions of progenitor cells (Götz and Huttner, 2005). As neural tube cells progress through the cell cycle, they undergo interkinetic nuclear migration with nuclei undergoing mitosis at the ventricular surface of the neural tube (at the apical part of the cells, close to the lumen of the neural tube) while their daughter cells reach the G1/S checkpoint as nuclei reach the basal limit of the progenitor zone, where they either re-enter or exit the cell cycle (Lee and Norden, 2013). Post-mitotic cells remain at the lateral face of the neural tube where they contribute to the mantle zone (MZ) and acquire further differentiated features. Therefore, as neurogenesis progresses, the MZ thickens. The intermediate layer between the progenitor area [or ventricular zone (VZ)] and the MZ, called intermediate zone (IZ), contains the newly born neurons on their way to their final position (del Corral and Storey, 2001).

The progenitor/neuron ratio is controlled by the proliferation properties of progenitors (lengths and rounds of cell cycles), by the survival of progenitors and differentiating neurons, and by progenitor cell fate decisions (to remain a progenitor or to differentiate). Progenitor cell fate decisions are based on the activation of a cascade of transcription factors triggered by proneural genes (Bertrand et al., 2002; Lacomme et al., 2012; Ma et al., 1996), the expression of which is largely controlled by the Notch signaling pathway (Formosa-Jordan et al., 2013; Hatakeyama, 2004; Hatakeyama et al., 2006). The mediolateral spatial organization of the differentiating neural tube into the three layers (VZ, IZ and MZ) is important for ensuring a proper differentiation rate because, in the nascent IZ neurons, proneural genes induce the expression of Notch ligands such as Delta1 and Jagged which in turn activate Notch1, which down-regulates proneural gene expression and inhibits neurogenesis in neighboring precursors. Correct spatial organization of the neural tube along the medio-lateral axis requires timely detachment of newborn neurons from the apical surface in order to exit this proliferative zone and begin the morphological reorganization that underlies neuronal differentiation. This apical detachment process is known as delamination (Kasioulis and Storey, 2018). Although it has been recently shown that the synchronization of the delamination is controlled by the Notch pathway (Baek et al., 2018), molecular mechanisms controlling the timing of the delamination are not fully understood. In this study, we aimed to investigate the function of B cluster Hox genes at early steps of trunk neural tube development using the chicken embryo as a model from E2 onwards, when non-B Hox genes are not yet expressed and before to the well-documented role of Hox genes in motor neuron differentiation.

Numerous studies have described Hox gene expression in the trunk neural tube of chicken embryo. The lack of marked axial specificity of the HoxB genes within the brachial, lumbar and sacral territories of the trunk neural tube at E6, with an expression profile very different from the Hox genes of other clusters (Dasen et al., 2005), prompted us to re-investigate the expression and function of B cluster Hox genes during early spinal cord development.

We started by exploring HoxB expression patterns between E3 and E5 in the trunk neural tube using whole-mount and transverse section in situ hybridization for anterior (HoxB2 and HoxB4), central (HoxB5, HoxB7 and HoxB8) and posterior (HoxB9) classes of Hox genes. As early as E3, HoxB genes are expressed in largely overlapping territories, from the neck (in which there is still a weak spatial collinearity) to the tail (Fig. 1A,B). In addition to highlighting the large overlap in the spatial expression domains of HoxB genes (except HoxB13, which is expressed only in the tail bud after stage HH20; Denans et al., 2015), our results confirm that HoxB transcripts are present in the trunk neural tube before the onset of neurogenesis (Fig. 1A,B). To assess the presence of HoxB proteins in the trunk neural tube, including at early stages, we raised an antibody specific to the posterior HoxB9 protein (Fig. S1 displays the specificity of the HoxB9 antibody). Immunostaining with this antibody on transverse sections show that the HoxB9 protein is present in the trunk neural tube as early as E3 (Fig. 1C) and is broadly expressed from the neck to the tail (Fig. 1D) (coincident with HoxB9 transcripts; Fig. 1A). This contrasts with its paralog HoxC9 protein, not yet expressed at E3 in the neural tube (Fig. 1C), expressed only later and only at the thoracic level (Fig. 1D). In situ hybridization with the HoxB8 probe (a similar pattern was described for the HoxB8 protein; Asli and Kessel, 2010) and immunostaining with the HoxB9 antibody on the same transverse sections at E4.5 highlight the strong overlap in the expression of these ‘central’ and ‘posterior’ HoxB members, along the antero-posterior axis of the chicken embryo, from cervical to sacral level (Fig. 1E).

Altogether, our expression data define a temporal and spatial time window that differs from non-B cluster Hox genes. The lack of clear axial specificity does not favor a role for HoxB genes in an antero-posterior axial diversification of the neural tube, but early and broad expression of HoxB rather suggests a generic function during neurogenesis of the developing trunk neural tube.

The comparison of the expression pattern of HoxB genes with markers of the three layers of the trunk neural tube (Sox2 for the VZ, NeuroD4 for the IZ and Tuj1 for the MZ) at E4 (Fig. 2A) shows that in addition to disappearing from the differentiating motor neuron domain where non-B cluster Hox genes are expressed, HoxB gene and protein expression at that stage is mainly restricted in the IZ, although weak expression is observed in the VZ (Figs 1C-E, 2A, S2). These expression dynamics suggest that HoxB genes, although not exclusively, may control neurogenesis progression by controlling the expression of genes expressed in the IZ. As HoxB genes are expressed in the IZ all along the dorso-ventral axis, and from the neck to the tail (Fig. 1C-E), this function would apply to all neuronal subtypes, i.e. motor neurons and all interneurons, and would be irrespective of the antero-posterior axial position (except in the neck) and of the paralog identity of the HoxB gene.

The largely overlapping HoxB expression patterns suggest HoxB gene functional redundancy, which compromises loss-of-function approaches. We thus probed the function of HoxB genes in neural tube development by gain-of-function experiments. HoxB4, HoxB8 and HoxB9 were chosen as representative of anterior, central and posterior HoxB genes, respectively. Neural tubes of E2 embryos were unilaterally electroporated with a control plasmid encoding GFP or with each HoxB expression vector co-expressing GFP (to report transfected cells). Immunostainings were performed with antibodies against the progenitor marker Sox2 and the pan-neuronal marker Tuj1. Results show the presence of ectopic Sox2-positive cells in the MZ on the electroporated side for all three HoxB gene gain-of-function experiments at either 2 (Fig. S3) or 3 days (Fig. S4) after electroporation. We observed the ectopic Sox2 phenotype at all dorso-ventral positions in the spinal cord (Fig. S4). However, the phenotype obtained is modest, with only a few ectopic Sox2-positive cells in the MZ. The low penetrance of the phenotype could result from elimination of ectopic Sox2 cells by apoptosis, a hypothesis consistent with increased apoptosis following HoxB8 electroporation (Fig. S5). The hypothesis was probed by analyzing embryos co-transfected with HoxB (HoxB4, HoxB8 or HoxB9) and P35 [an inhibitor of apoptosis (Sahdev et al., 2010)] expression vectors. Quantification of Sox2 ectopic cells in the MZ 72 h after HoxB4, HoxB8 or HoxB9 gain of function in the context of P35 expression shows a strong phenotype [around 35% of transfected cells (GFP+) in the MZ are Sox2 positive after the overexpression of any of the three HoxB genes]. Ectopic Sox2⁺ cells in the MZ are rarely seen under control conditions (control vector+P35) (Fig. 2C). These results indicate that cell elimination through apoptosis contributes to the modest phenotype observed in HoxB gain-of-function experiments, and that the full range of HoxB-induced phenotypes can only be observed when suppressing apoptosis.

The phenotypes triggered by overexpression of each of the three HoxB genes are similar, with no marked differences in their potential to induce ectopic Sox2-positive cells (Fig. 2C). The phenotype of HoxB gain of function is not strictly cell-autonomous as Sox2-positive GFP-negative cells can be found in the MZ (Fig. 2D). In addition, Sox2/pH3 and Sox2/EdU double staining following HoxB8 gain of function shows that Sox2 ectopic cells in the MZ are still mitotic (Fig. 2D and Fig. S6), a characteristic of progenitor cells.

We conclude from this set of experiments that HoxB genes, irrespective of their paralog identity, induce (when overexpressed) a similar phenotype consisting of the appearance of ectopic progenitors (Sox2-positive) cells in the MZ. Taken together with expression pattern data, this suggests a generic role for HoxB genes in the control of early neural tube differentiation (neurogenesis and/or neuronal delamination). We next questioned whether non-B Hox genes, although not expressed at these early stages, also have the capacity to induce ectopic Sox2 cells in the MZ. The hypothesis was probed by forcing the premature expression of HoxA7, HoxC8 and HoxD8 (one representative of each non-HoxB cluster Hox) from E2 (these experiments were carried out in a P35 context). Results showed that such an expression leads to a phenotype similar (ectopic Sox2 cells in the MZ) to those exhibited by B cluster Hox genes (Fig. S7), supporting that the induction of ectopic Sox2 cells in the MZ is a regulatory property also embedded in non-B Hox proteins.

To obtain molecular insights into HoxB gene function in the early neural tube, we aimed to identify downstream target genes of the HoxB transcription factors, focusing on the central class HoxB8 protein. E2 neural tubes were bilaterally electroporated with either a control vector encoding nuclear GFP (pCIG) or a HoxB8 expression vector co-expressing nuclear GFP (pCIG-HoxB8) (Fig. 3A). The regions of the neural tube expressing the GFP were dissected 18 h after electroporation and dissociated. GFP-expressing cells were isolated by FACS with the use of a dead cell exclusion (DCE)/discrimination dye (DAPI) to eliminate dying cells (Fig. S8). Two independent RNAs samples were extracted and reverse transcribed, and cDNAs were amplified using a linear amplification system and used for sequencing library building. Qualitative analysis of RNA-seq data from the two biological replicates shows a high Pearson correlation score (>0.98) indicative of the experimental reproducibility (Fig. S9). RNA-seq data from alignment to the Galgal4 genome assembly identified 1913 genes with significantly changed expression [Fig. 3B FDR5 (False Discovery Rate 5), Tables S1 and S2; see also Materials and Methods section], of which 1097 were upregulated (57%) (Table S1) and 816 were downregulated (43%) (Table S2) (Fig. 3C, left panel). This tendency of HoxB8 to act as activator rather than repressor is amplified when selecting genes differentially expressed by more than twofold, with 251 being upregulated (90%) and only 25 downregulated (10%) (Fig. 3C, right panel). Gene ontology enrichment analysis (GOEA) of the biological processes suggests pleiotropic functions of HoxB8 during spinal cord development (Fig. 3D and Table S3), including neuron differentiation, apoptotic process, cell cycle and cell migration (Fig. 3D). In particular, the Notch signaling pathway, a key regulator of neurogenesis (Formosa-Jordan et al., 2013; Hatakeyama, 2004; Hatakeyama et al., 2006), stands out from the GOEA (Fig. 3D, downregulated genes), suggesting that HoxB8 controls neurogenesis. This is illustrated by Hes5.1, a Notch pathway effector expressed in the VZ and known to keep neural tube cells in a progenitor state (Fior and Henrique, 2005), for which transcripts in situ hybridization shows strong transcriptional downregulation (Fig. S10, Table S2).

Among all deregulated genes, the leucine zipper tumor suppressor 1 (Lzts1) gene (also known as FEZ1 and PSD-Zip70) (Baffa et al., 2008; Ishii et al., 2001; Vecchione et al., 2007) caught our attention for two reasons. First, Lzts1, which is upregulated by HoxB8 (Fig. 3B,E and Table S1) is, as HoxB, preferentially expressed in the IZ in the trunk neural tube of chicken and mouse embryos (Kropp and Wilson, 2012). Second, Lzts1 has been recently shown to control neuronal delamination during mammalian cerebral development (Kawaue et al., 2019), a function that, if conserved in the neural tube, could account for the HoxB-induced ectopic Sox2-positive cells found in the MZ.

We studied the dynamics of Lzts1 expression by in situ hybridization with an Lzts1 probe at E2, E3 and E4 stages (Fig. 4A,B and Fig. S11). At E3, Lzts1 transcripts are already found in the IZ, which, owing to the lack of differentiated neurons that will form the MZ at that stage, is in the most-lateral region of the neural tube (Fig. 4A). At E4, as previously described (Kropp and Wilson, 2012), Lzts1 transcripts are still associated with the IZ, located between progenitors of the VZ and differentiated neurons of the MZ (Fig. S11B), with an expression pattern very similar to HoxB genes (Fig. 4B and Fig. S12). Indeed, although not completely overlapping, as HoxB genes are still expressed in the VZ at low levels, Lzts1 transcripts are enriched where the HoxB9 protein levels are the highest (Fig. S12). The expression of the Lzts1 gene is not restricted to a specific antero-posterior region of the neural tube (Figs S11A and S12). The Lzts1 expression pattern is thus compatible with regulation by HoxB proteins in the IZ. Consistent with its identification as a HoxB8 target in the transcriptomic approach, in situ hybridization with a Lzts1 probe following HoxB8 gain of function shows ectopic Lzts1 expression in the trunk neural tube (Fig. 4C). If Lzts1 regulation illustrates at the level of a single target the generic control of early neurogenesis documented in Fig. 2, HoxB4 and HoxB9 should also induce ectopic Lzts1 expression, which is indeed observed (Fig. 4C). Lzts1 transcriptional activation is faint, consistent with the 2.7-fold transcript enrichment seen in the RNA-seq data, and is mainly observed in the ventricular zone (Fig. 4). Co-expressing HoxB4, HoxB8 or HoxB9 with the P35 apoptotic inhibitor does not allow for stronger and more frequent Lzts1 induction, in particular in the MZ, where under such conditions the frequency of Sox2-positive cells in the MZ is high (Fig. S13). This indicates that the lack of Lzts1 induction in cells of the MZ is not due to cell elimination by apoptosis, suggesting that HoxB proteins can only transcriptionally control Lzts1 expression within a precise time window: when cells are still in the VZ or IZ.

We also found that HoxA7, HoxC8 and HoxD8 gain of function induce Lzts1 expression (Fig. S14), showing that, as in the case of Sox2-positive ectopic cell induction in the MZ, Lzts1 transcriptional activation by Hox proteins relies on regulatory properties embedded in B and non-B Hox proteins. These results show that Lzts1, identified as a HoxB8 target, is a generic Hox target. However, only B cluster Hox genes are expressed at the proper time and space for assuming that function.

Premature delamination of neural progenitors may explain the presence of ectopic Sox2-positive cells in the MZ after the HoxB gain of function. The function of Lzts1 in neuronal development within the trunk neural tube is not known, but it has been described to positively control neuronal delamination in brain development in mammals (Kawaue et al., 2019). Owing to its expression in the IZ, where progenitors switch to neurons and lose their apical attachment, Lzts1 may also control delamination during spinal cord neurogenesis.

To examine this, we analyzed the consequences of Lzts1 gain of function, obtained through unilateral electroporation of an Lzts1 expression vector in the neural tube at E2 (Fig. S15). The tracking of cytoplasmic GFP demonstrated massive cell delamination, with nearly all electroporated cells losing their attachment to the lumen and found in the MZ (Fig. 5A). This phenotype is seen both at 2 and 3 days after electroporation (Fig. 5A). Under normal conditions, only newborn neurons lose their apical attachment (Kasioulis and Storey, 2018), suggesting that Lzts1 in the IZ is involved in the control of newborn neuron delamination. Immunostaining with Sox2 and Tuj1 antibodies (progenitor and neuronal markers, respectively) (Fig. 5 and Fig. S16) showed that nearly half (47.7%) of the Lzts1 gain-of-function cells in the MZ ectopically express Sox2 (Fig. 5C). This suggests that Lzts1 gain of function forces the delamination but not neural differentiation, as cells that prematurely delaminate stay in a progenitor state (Fig. 5B,C). This is distinct from neurogenin2 gain of function where electroporated cells massively delaminate but also prematurely differentiate (Garcia-Gutierrez et al., 2014). The Lzts1 delamination ‘only’ phenotype is confirmed by the finding that the MZ ectopic Sox2-positive cells retain progenitor characteristics: they express CCND1/cyclinD1 (Fig. 5D) and the pH3 mitotic marker (Fig. 5E), and Hes5.1 and NeuroD4 genes, which are (respectively) markers of the VZ and IZ (Fig. 5F,G). The phenotype induced by Lzts1 gain of function is independent of the dorso-ventral and antero-posterior position within the trunk neural tube (Fig. 5B and Fig. S16) and is not strictly cell-autonomous (ectopic Sox2-positive GFP-negative cells can be found following Lzts1 gain of function; Fig. S16).

To support conclusion from Lzts1 gain-of-function experiments, we analyzed the effects of Lzts1 loss-of-function. Knockdown was obtained through unilateral electroporation of a shRNA-expressing plasmid at E2. The efficiency of the shRNA was assessed by following Lzts1 transcripts (Fig. 6A), showing a strong effect that was illustrated by the absence of the typical Lzts1 expression in the IZ. The effects of Lzts1 knockdown were studied using the Sox2 progenitor (Fig. 6B) and Tuj1 or HuC/D (Fig. 6C-F, Fig. S17) neuronal markers. Results show that Lzts1 knockdown does not lead to ectopic Sox2 cells, as induced by Lzts1 gain of function, but instead to an ectopic expression of Tuj1 or HuC/D neuronal markers in the VZ, with neurons retaining their apical attachment (Fig. 6C-F). The loss of function of Lzts1 thus results in neuronal delamination inhibition, a phenotype that mirrors the promotion of neuronal delamination seen in Lzts1 gain-of-function experiments (Fig. 5 and Fig. S16).

Collectively, Lzts1 gain- and loss-of-function experiments demonstrate a role for Lzts1 in controlling neural delamination in the trunk neural tube. As impaired delamination is a plausible explanation for the generic HoxB-induced MZ ectopic Sox2 cells, and as Lzts1 transcripts are induced by HoxB proteins, Lzts1 is likely to be a key HoxB effector leading to the MZ ectopic Sox2 phenotype. To probe this hypothesis, we performed epistatic experiments by co-expressing HoxB8 and the shRNA-Lzts1, in the P35 context in order to start with a stronger HoxB-induced phenotype. Results show that Lzts1 gene inactivation lowers significantly the occurrence of Sox2 ectopic cells in the MZ (Fig. 6G-I), supporting the hypothesis that Lzts1 is a key effector in the HoxB-induced delamination phenotype.

Our work extends the functional contribution of Hox genes to spinal cord development. Although largely shown to act as ‘choreographers’ of neural development in specifying motor neurons subtypes (Philippidou and Dasen, 2013), this study highlights an unexpected early and general role in controlling early neurogenesis and neuronal delamination. Previous expression data delineated that B cluster Hox gene expression at E6, a stage when motor neuron subtypes are defined, does not follow axial collinearity, as non-B cluster Hox genes do, and is generally excluded from differentiating motor neurons (Dasen et al., 2005; Jung et al., 2010; Lacombe et al., 2013). Based on the expression analysis of representatives of anterior, central and posterior Hox paralogs, we propose that B cluster Hox genes (excepted HoxB13) are expressed in the chicken neural tube earlier than non-B cluster Hox genes, in a largely ubiquitous pattern that later resolves in preferential expression in the IZ, the region of the trunk neural tube where neuronal progenitors exit the cell cycle and delaminate to transit toward the mantle zone. Consistent with the lack of axial collinearity already observed at E6, B cluster Hox gene expression displays little axial specificity, with most HoxB genes expressed in largely overlapping expression patterns in the trunk neural tube (Fig. 7). These expression patterns suggest a function distinct from endowing the neural tube with axial positional information required for proper setting of neuronal subtype along the AP axis, which is well documented for HoxA, HoxC and HoxD genes. It rather suggests that B cluster Hox genes act without paralog specificity all along the trunk neural tube, in a ‘generic’ manner, providing temporal instead of positional information. Although long underseen, a recent literature survey indicates that such generic functions are constitutive of Hox protein function (Saurin et al., 2018), and may be an intrinsic deeply rooted property of Hox proteins that reflects their phylogenetic common origin. An illustration of such a function is the generic control of autophagy by Hox proteins in the Drosophila fat body (Banreti et al., 2014), where, as seen here in the chicken neural tube, Hox genes are broadly expressed in the tissue. The difficulty in studying such generic function, which has contributed to its late recognition, is that revealing them can often not be achieved by conventional loss-of-function approaches, as mutating one or even a few Hox genes does not alter the shared generic function, due to inter-paralog functional compensation (Banreti et al., 2014).

To obtain insights into early B cluster Hox gene function in the chicken neural tube, we used a gain-of-function approach. Results obtained indicate that Hox gain of function results in the appearance of progenitor Sox2-positive cells in the MZ, a region of the neural tube that normally hosts differentiated post-mitotic neurons. Consistent with a shared ‘generic’ function suggested by the expression patterns, we found that anterior, central and posterior HoxB genes induce similar defects, all resulting in ectopic Sox2-positive cells in the MZ. This phenotype is observed all along the trunk neural tube and occurs at all dorso-ventral positions within the tube, indicating that this generic Hox function may be relevant to neurogenesis progression in general, irrespective of the final antero-posterior or dorso-ventral driven final neuronal identity. We also found that gain-of-function experiments conducted with the non-B proteins HoxA7, HoxC8 and HoxD8, not expressed at early stage in the neural tube, also result in ectopic Sox2-positive cells in the MZ. This suggests that the control of the process leading to the Sox2-positive cells in the MZ is a regulatory property likely embedded into Hox proteins in general, and may rely on the similar biochemical characteristics of Hox proteins, with most Hox proteins displaying similar DNA-binding properties (Hayashi and Scott, 1990; Mann and Chan, 1996; Mann et al., 2009; Merabet and Mann, 2016; Zandvakili and Gebelein, 2016). The B cluster specificity would thus arise strictly from the temporal and spatial deployment of B cluster proteins, and not from intrinsic properties specific to B cluster Hox proteins. In agreement with this hypothesis, in silico survey of sequence conservation in Hox proteins, including short linear motifs (SLiMs), does not reveal any characteristics specific to the B cluster Hox proteins (Rinaldi et al., 2018).

To circumvent the difficulty of gaining functional insights into HoxB gene function in the early chicken neural tube from loss-of-function approaches, we reasoned that identifying and studying HoxB downstream targets, including through loss-of-function approaches, would allow better assessment of how HoxB genes influence early spinal cord development. Transcriptomic data obtained 1 day after HoxB8 overexpression highlights genes and pathways well known to control multiple aspects of neurogenesis, including Notch and IGF pathway effectors (Fior and Henrique, 2005; Fishwick et al., 2010; Vilas-Boas and Henrique, 2010), suggesting that the influence of HoxB genes on early neurogenesis is diverse. For this study, we focused on Lzts1, which shares with HoxB genes a preferential expression in the IZ – the region of the trunk neural tube containing the newly born neurons on their way to their final position (the MZ) – and thus may account for the main phenotype (Sox2 ectopic cells in the MZ) seen in HoxB overexpression experiments.

The study of Lzts1 gain- and loss-of-function experiments showed that Lzts1 controls the delamination of newborn neurons: gain of function induces massive cell delamination with nearly all electroporated cells losing their attachment to the lumen and found in the MZ, whereas loss of function leads to differentiated neurons keeping their apical attachment. The promotion of delamination by Lzts1 further suggests that the appearance of Sox2-positive cells in the MZ, seen in Lzts1 and Hox gain-of-function experiments, results from loss of apical attachment and subsequent migration of progenitor Sox2-positive cells in the MZ. Consistent with the view that Lzts1 mediates HoxB generic function in the chicken early neural tube, we observed that Lzts1 expression is influenced not only by HoxB8, but also by all other HoxB cluster genes probed (the anterior HoxB4 and posterior HoxB9 class genes), and that Lzts1 gene knockdown in a HoxB8 gain-of-function experiment significantly lowers the HoxB8-induced ectopic Sox2 phenotype. Although both HoxB and Lzts1 overexpression result in ectopic Sox2-positive cells in the MZ, the HoxB overexpression phenotypes are less pronounced than Lzts1 gain of function (including in a P35 context which inhibits cell death), with fewer ectopic Sox2-positive cells seen in the MZ. This weaker phenotype is in line with the limited capacity of HoxB genes to induce Lzts1 expression in gain-of-function experiments, which likely reflects a precise time window within which Lzts1 transcriptional activation by HoxB proteins is possible.

HoxB-mediated control of neural delamination via the regulation of Lzts1 in the IZ uncovered by this study might be shared by higher vertebrates. Expression patterns of the HoxB and Lzts1 genes in the trunk neural tube of mouse embryo are highly reminiscent of those of the chicken embryo: Lzts1 is also expressed in the IZ (Kropp and Wilson, 2012); HoxB genes, also expressed earlier than non-B cluster Hox genes, also have broadly overlapping expression in the neural tube and are also excluded from the motor neuron area, as in chicken (Graham et al., 1991; Jung et al., 2010; Lacombe et al., 2013). However, HoxB gene expression does not seem to resolve in the IZ as sharply as in the chicken. HoxB activity might be restrained to the IZ through the expression and action in the VZ of geminin, a pleiotropic cell-cycle regulator also known to inhibit Hox proteins (Luo et al., 2004; Patterson et al., 2014). In this situation, only cells that express HoxB genes laterally to the limit of geminin expression, which corresponds to the IZ, would be free of the geminin inhibitor and allow HoxB-mediated Lzts1 transcriptional activation.

Lzts1 function in the control of neural delamination has already been described in mammals, in the context of the brain (cephalic neural tube) of mouse and ferret (Kawaue et al., 2019). Kawaue and colleagues demonstrated that Lzts1, which associates with microtubule components and is involved in microtubule assembly (Ishii et al., 2001), controls apical delamination of neuronally committed cells of the brain by altering apical junctional organization (Kawaue et al., 2019). Indeed, in neuronally differentiating cells of the brain, Lzts1 modulates the microtubule-actin-AJ system at the apical endfeet to evoke apical contraction and reduce N-cadherin expression (Kawaue et al., 2019). Molecular mechanisms by which Lzts1 controls delamination in the trunk neural tube might be the same as in the mammalian brain. However, Lzts1 upstream regulation has to be distinct, as the brain is known as a Hox-free territory.

In humans, expression of the LZTS1 gene (also named FEZ1) is altered in multiple tumors (Ishii et al., 1999). LZTS1 tumor suppressor function has been attributed at least in part to its role in the control of mitosis progression (Vecchione et al., 2007) and in regulating the Pi3k/AKT pathway (He and Liu, 2015; Zhou et al., 2015). As the Pi3k/AKT pathway is required for neuron production in the trunk neural tube in both mouse and chicken embryos (Fishwick et al., 2010), Lzts1 might also regulate neuronal production in the trunk neural tube by regulating the Pi3k/AKT pathway in addition to controlling delamination.

Experiments performed with non-hatched avian embryos in the first two-thirds of embryonic developmental time are not considered animal experiments according to the directive 2010/63/EU.

Fertilized chicken eggs were obtained from EARL les Bruyeres (Dangers, France) and incubated horizontally at 38°C in a humidified incubator. Embryos were staged according to the developmental table of Hamburger and Hamilton (HH) (Hamburger and Hamilton, 1992) or according to days of incubation (E).

Neural tube in ovo electroporations were performed around HH12. Eggs were windowed, and the DNA solution was injected in neural tube lumen. Needle L-shape platinum electrodes (CUY613P5) were placed on both sides of the embryo at trunk level (5 mm apart), with the cathode always at its right. Five 50 ms pulses of 25 volts were given unilaterally (or bilaterally for RNAseq experiments) at 50 ms intervals with an electroporator NEPA21 (Nepagene).

The plasmids used for the gain-of-function experiments co-express a cytoplasmic or nuclear GFP (pCAGGS and pCIG, respectively, used alone as controls) and the coding sequence (CDS) of the gene of interest. Vectors used were: pCIZ-HoxB4, pCIG-HoxB8, pCIG-HoxB9 and pCIG-HoxC8 (gifts from Dr Olivier Pourquié, Harvard Medical School, Boston, MA, USA), pCAGGS-P35 (a gift from Dr Xavier Morin, IBENS, Paris, France), and pCAGGS-Lzts1, pCAGGS-HoxA7, pCAGGS-HoxD8, pCAGGS-HoxB4 pCAGGS-HoxB8 and pCAGGS-HoxB9 (this study). The CDS of HoxA7 and HoxD8 (second isoform, 567 bp) were PCR amplified from chicken neural tube cDNA; the CDS of HoxB4, HoxB8 and HoxB9 were PCR amplified from the plasmids described above and the CDS of Lzts1 was amplified from pGEMT-Lzts1 (a gift from Dr Sara Wilson, Umeå University, Sweden). All sequences were subcloned in the pCAGGS plasmid using In-Fusion HD Cloning Kit (Takara). RNA-interference technology was used to inhibit Lzts1, with the pRFPRNAiC vector, which contains an RFP reporter gene (Das et al., 2006), and insertion sites for two siRNAs in tandem. The two 22-nucleotide target sequences for the shRNA-Lzts1 plasmid were chosen using the design tool ‘siRNA Target Finder’ (AAGGTCAACCTGTTAGAGCAGG and AACATCATGCAGTGTGCCATCA). A shRNA-scrambled-Lzts1 plasmid was designed as control (AGAAGAGTGTACGGTCGCAGTC and GCATGTTGAACCGCAATACACT).

All the plasmids used for electroporation were purified using the Nucleobond Xtra Midi kit (Macherey-Nagel). Final concentrations of DNA delivered per embryo for electroporation were between 1 and 2 µg/µl, except for the epistatic experiment performed with DNA solution at 2.5 µg/µl, due to technical constraints (Table S4).

Embryos were fixed in 4% buffered formaldehyde in PBS then treated with a sucrose gradient (15% and 30% in PBS), embedded in OCT medium and stored at −80°C. Embryos were sectioned into 16 µm sections with a Leica cryostat and the slides were conserved at −80°C or directly used for fluorescent in situ hybridization and/or immunofluorescence.

Slides were rehydrated in PBS then blocked with 10% goat serum, 3% BSA and 0.4% Triton X-100 in PBS for 1 h. Primary antibodies were incubated overnight diluted in the same solution at 4°C. The following primary antibodies were used in this study: chicken anti-GFP 1:1000 (1020 AVES), rabbit anti-SOX2 1:500 (AB5603 Merck Millipore), mouse anti-Tuj1 1:500 (801202 Ozyme), mouse anti-HuC/D 1: 200 (ThermoFisher 16A11), rat anti-pH3 1: 250 (S28, Abcam ab10543), rabbit anti-caspase 3 1:500 (Asp175, CST 9661), guinea pig anti-HoxC9 1:1000 (NY1638, a gift from Dr Jeremy Dasen, NYU School of Medicine, USA) and rabbit anti-LZTS1 1:250 (Sigma HPA006294). Polyclonal HoxB9 antibodies were raised in guinea pig using the peptide ‘143-158 GIVSNQRPSFEDNKVC’ and used at 1:500. The secondary antibodies used were: anti-chicken, anti-rabbit, anti-mouse, anti-rat or anti-guinea pig conjugated with fluorochromes (488, 568 or 647) at 1:500 (Life Technologies). They were incubated for 1 h in the blocking solution containing Hoechst (1:1000). Slides were washed, mounted (ThermoFisher Scientific Shandon Immu-Mount) and imaged with a Zeiss microscope Z1 equipped with Apotome or a confocal LSM 780.

Proliferative cells were labeled with EdU using the Click-iT EdU Alexa Fluor 647 kit (ThermoFisher). 400 µl of a 0.5 mM EdU solution (in PBS) was applied on top of the embryo and incubated at 38°C for 30 min. Cryostat sections were stained for EdU using the manufacturer's instructions.

The slides were treated with proteinase K 10 µg/ml (3 min at 37°C) in a solution of 50 mM TrisHCl (pH 7.5), then in 0.1 M triethanolamine and 0.25% acetic anhydride. They were pre-incubated with hybridization buffer (50% formamide, 5×SSC 5X, 5×Denhardt's, 250 µg/ml yeast tRNA and 500 µg/ml herring sperm DNA) for 3 h at room temperature, and incubated in the same buffer with DIG-labeled RNA probes overnight at 55°C in a wet chamber. The slides were then washed twice with 0.2× SSC for 30 min at 65°C. After 5 min in TNT buffer [100 mM Tris (pH 7.5), 150 mM NaCl and 0.1% Tween-20], they were then blocked for 1 h in buffer containing 1× TNT, 1% blocking reagent (Roche) and 10% goat serum, then incubated in the same buffer for 3 h with anti-DIG-POD antibodies (1:500, Roche) and revealed using the kit TSA-Plus Cyanin-3 (Perkin Elmer). RNA probes used for in situ hybridization were: Lzts1, Hes5.1, Ccnd1, NeuroD4, HoxB4, HoxB5, HoxB7, HoxB8 and HoxB9. The plasmids used to generate the Hox RNA probes were gifts from Jeremy Dasen and Olivier Pourquié (except HoxB8 PCR primer forward: CCAGCTCCCCTTACCAACAG and T7 reverse: TAATACGACTCACTATAGGGCCTCGGGGGCTCTTCTACCC, transcription from neural tube cDNA). The vector for the Lzts1 probe (pGEMT-Lzts1) was a kind gift from Dr Sara Wilson and the vector for the Hes5.1 probe was a gift from Dr Xavier Morin.

Embryos were fixed for 2 h at room temperature in 4% formaldehyde in PBS. Embryos were dehydrated with sequential washes in 50% ethanol/PBS+ 0.1% Tween20 and 100% ethanol and conserved at −20°C. Embryos were bleached for 45min in 80% ethanol+20% H₂O₂ at 30% and then rehydrated. They were treated with proteinase K (10 µg/ml) at room temperature and refixed with 4% formaldehyde and 0.2% glutaraldehyde. After 1 h of blocking in the hybridization buffer (50% formamide, 5× SSC, 50 μg/ml heparin, 50 μg/ml yeast tRNA, 1% SDS), hybridization with DIG-labelled RNA probes (HoxB2, HoxB4, HoxB7, HoxB8, HoxB9 and Lzts1) was performed at 68°C overnight. The next day, embryos were washed (three times for 30 min) in hybridization buffer and once in TBS [25 mM Tris, 150 mM NaCl, 2 mM KCl (pH 7.4)] +0.1% Tween 20. They were incubated for 1 h at room temperature in a blocking buffer (20% blocking reagent+20% goat serum) and then overnight with an anti-DIG-AP antibody (1:2000, Roche) in the blocking buffer. After three washes (1 h) in TBS+0.1% Tween 20, embryos were equilibrated (twice for 10 min) in NTMT buffer [NaCl 100 mM, TrisHCl 100 mM (pH 9.5), MgCl₂ 50 mM, 2% Tween20] and incubated in NBT/BCIP (Promega) at room temperature in the dark until color development. Pictures of whole embryos were made using a BinoFluo MZFLIII and a color camera.

Electroporations were carried out as described in a previous section but with five bilateral pulses. Plasmid DNA concentrations were: for the control mix, pCIG 2 µg/µl; for the HoxB8 mix, HoxB8-pCIG 1 µg/µl+pCIG 1 µg/µl. Parts of the neural tube expressing GFP were dissected 18 h after electroporation and dissociated (Trypsin-EDTA 0.25%). GFP and CDS take around 3 h to be expressed after electroporation. As a consequence, 18 h post-electroporation means that HoxB8 is overexpressed in neural tube cells for about 15 h. We have chosen this timing as it is the earliest at which the size of the neural tube allows for rapid dissection, a condition required for collecting sufficient starting material for FACS in a minimum timeframe. A highly enriched population of GFP-expressing cells was isolated by FACS with the use of a dead cell exclusion (DCE)/discrimination dye (DAPI) to eliminate dying cells (Fig. S8). RNA was extracted (RNeasy Mini Kit) and reverse transcribed, and cDNA was amplified using a linear amplification system and used for sequencing library building (GATC): random primed cDNA library, purification of poly-A containing mRNA molecules, mRNA fragmentation, random primed cDNA synthesis, adapter ligation and adapter specific PCR amplification, Illumina technology, 50,000,000 reads paired end with 2×50 bp read length. Bioinformatics analyses were carried out using the galgal4.0 chicken genome. Qualitative analysis of RNA-seq data from the two biological replicates shows a high Pearson Correlation score (>0.98) indicative of the experimental reproducibility (Fig. S9). RNA-seq data have been deposited in GEO and under accession number GSE162665.

The number of embryos and number sections analyzed are indicated in the figure legends. A minimum of three embryos and six sections per embryo were used for quantification. All quantifications were carried out using the cell counter tool of Fiji software. The results were analyzed and plotted using Prism 8 software (GraphPad software). Statistical analyses were performed using a two-tailed Mann–Whitney test and considered significant when P<0.05. All P-values are indicated on the graphs. The error bars represent the standard deviation (s.d.).

We thank Dr Olivier Pourquié, Dr Jeremy Dasen, Dr Xavier Morin, Dr Heather Etchevers and Dr Sara Wilson for their generous gifts of antibodies, RNA probes and/or expression vectors. We thank France-BioImaging/PICsL infrastructure (ANR-10-INSB-04-01). We sincerely thank Dr Samuel Tozer, Dr Heather Etchevers and Dr Xavier Morin for critical reading of the manuscript. FACS experiments were carried out at the CRCM (Marseille, France).

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.195404.reviewer-comments.pdf