The neural tube (NT), the precursor of the vertebrate brain and spinal chord, is formed by a process known as neurulation. There are two major modes of neurulation in amniotes, primary and secondary neurulation, that occur in anterior and posterior regions of the embryo, respectively (Colas and Schoenwolf, 2001). Primary neurulation is achieved by the bending and folding of the epithelial neural plate to form a neural tube. Secondary neurulation involves the condensation of densely packed mesenchymal cells to shape a cord-like structure known as the medullary cord. These cells subsequently undergo a mesenchymal-to-epithelial transition and rearrange, by a process known as cavitation, to create a hollow tube (Schoenwolf, 1979; Schoenwolf and Delongo, 1980; Griffith et al., 1992). Neural tube defects (NTDs) resulting from impaired neurulation are the most common severely disabling birth defects in the United States, with a frequency of approximately 1/1,000 births (Detrait et al., 2005). Multiple neural tube defects in humans are associated with impaired cavitation in the posterior region of the neural tube specifically (Seller, 1990; Park et al., 1992; Saitsu et al., 2004; Donovan and Pedersen, 2005); however, a molecular understanding of the underlying causes is lacking. Studies of neurulation in the zebrafish have primarily focused on the morphogenetic movements in the head and anterior trunk region of the embryo. During the initial stages of anterior neurulation, the neuroepithelium narrows along the mediolateral (M–L) axis, to form a keel-shaped structure, a process known as neural convergence. The neural keel coalesces into a neural rod, which eventually cavitates and becomes the neural tube proper (Schmitz et al., 1993; Papan and Campos-Ortega, 1994; Kimmel et al., 1995; Concha and Adams, 1998; Geldmacher-Voss et al., 2003; Handrigan, 2003; Lowery and Sive, 2004).
Cellular analysis of anterior neurulation in zebrafish has revealed that the neural plate is composed of two cell layers, referred to as deep and superficial (Hong and Brewster, 2006). Deep cells occupy the lower layer of the neural plate. They have a columnar shape and maintain their relative M–L position and contact with the basal surface of the neuroepithelium throughout neurulation (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006). Despite these epithelial characteristics, deep cells do not exhibit a clearly defined apicobasal axis during the initial stages of neurulation, as they lack apical junctional complexes (Geldmacher-Voss et al., 2003; Hong and Brewster, 2006) and generate medially directed membrane protrusions, transiently interdigitating with cells from the opposite side of the midline (Hong and Brewster, 2006). As the neural keel converges, deep cells shift from a vertical to an oblique and finally a horizontal orientation, which is indicative of an epithelial “infolding” mechanism (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006). In contrast to the apparent organized infolding of deep cells, superficial cells in the zebrafish neural plate appear to migrate individually toward the midline and intercalate between deep cells to establish contact with the basal surface of the neuroepithelium (Hong and Brewster, 2006). Thus, the cellular behaviors underlying anterior neurulation involve an infolding process of deep cells, the active migration of superficial cells and the intercalation of superficial cells between deep cells to form a single cell layered neuroepithelium (Hong and Brewster, 2006). Cavitation begins at the neural rod stage and is mediated by apical membrane biogenesis (Munson et al., 2008), which establishes an epithelial seam and divides the left and right halves of the neural rod, lumen inflation, and localized cell proliferation (Lowery and Sive, 2005; Lowery et al., 2009).
The cellular dynamics discussed above focused exclusively on interphase cells. Dividing cells in anterior regions of the zebrafish neural keel/rod exhibit unique behaviors, as one daughter cell crosses the midline, while the other daughter cell remains on the same side of the neuroepithelium as the mother cell (Papan and Campos-Ortega, 1994; Concha and Adams, 1998; Geldmacher-Voss et al., 2003). Both daughter cells re-integrate into the neuroepithelium by extending membrane protrusions directed toward the basal surface of the neural keel/rod (Hong and Brewster, 2006). Midline crossing is a cell behavior apparently unique to the zebrafish neural tube and involves a 90° rotation of the mitotic spindle (Geldmacher-Voss et al., 2003).
The morphogenetic movements that shape the PNT of the zebrafish embryo are currently unknown. Given the distinct modes of primary and secondary neurulation in anterior and posterior regions of amniotes, respectively, it is possible that PNT formation in zebrafish may be quite distinct from the mechanisms described above. Fate mapping studies in the zebrafish embryo have shown that the neural tube in the tail region, with the exception of the floor plate (Shih and Fraser, 1995, 1996; Amacher et al., 2002; Kudoh et al., 2004), derives from ventral–vegetal ectoderm at the 70–80% epiboly stage, whereas more anterior neural tube originates from progressively more dorsal ectoderm (Kudoh et al., 2004). By 3 somites, presumptive posterior neural cells localize to the dorsomedial region of the anterior tailbud (Kanki and Ho, 1997). Fate mapping analysis demonstrated that this cell population moves caudally, over the posterior tailbud cells, during axis elongation (Kanki and Ho, 1997). These observations indicate that early stages of PNT formation are likely to involve morphogenetic movements that position ventrally derived cells in the tailbud. However, neither these cell behaviors nor those contributing to the continued shaping and elongation of the PNT have been described.
Here, we provide the first extensive analysis of the cellular mechanisms shaping the zebrafish PNT, using a variety of tools for imaging the neural tissue and tracking the behavior of individual cells in real time. We arbitrarily divided the events leading to PNT formation into three stages. During “Stage 1” neural precursor populations, originating in lateral and ventral domains of the pregastrula embryo, merge with dorsally derived neural cells in the anterior tailbud region, to form a continuous neuroepithelium. At this stage, these cells are organized as a single-cell layered tissue, that lacks distinct apicobasal polarity. During “Stage 2” the posterior neural domain thickens, narrows and elongates, to shape a rod-like structure. This is achieved by delamination of individual cells from the neuroepithelium and ipsilateral intercalation (intercalation of cells amongst neighbors derived from the same side of the neural plate). In addition, angular re-orientation of cells toward the midline during “Stage 2” further contributes to the shaping of the PNT. During “Stage 3,” the neural tissue continues to elongate in absence of further convergence movements. Using drugs that block mitosis, we demonstrate that elongation of the PNT during stages 2 and 3 is only partially dependent on cell proliferation, highlighting the importance of cellular rearrangements during these stages. Moreover, we provide evidence that apicobasal polarity is not established until the neural rod is formed, indicating that neural cells undergo progressive epithelialization as neurulation proceeds. The morphological changes and cell behaviors observed during PNT formation are for the most part similar to those previously described for neurulation in anterior regions. The main differences involve the type of cellular rearrangements driving neural convergence and the orientation of dividing cells, which appears random in posterior regions. Overall, these findings suggest that neurulation in the zebrafish is a fairly uniform process that most closely resembles secondary neurulation.
Stages of PNT Morphogenesis
In zebrafish, cells contributing to the PNT arise from lateral and ventral regions of the gastrulating embryo (Kimmel et al., 1990; Kudoh et al., 2004). Upon completion of epiboly, neural cells populate the anterior tailbud region, where they undergo NT morphogenesis (Kanki and Ho, 1997). To gain a better understanding of the timing and events that shape and elongate the PNT we analyzed the expression of molecular markers for the neural ectoderm (α-Sox3C), the non-neural ectoderm or epidermis (α-p63), and the cell surface (α-β-catenin), between the tailbud (TB) and 20 somite (som) stages. Embryos were triple-labeled with α-Sox3C, α-p63, and α-β-catenin and sectioned transversely or longitudinally through the tailbud, at the level of Kupffer's vesicle (KV, a morphological landmark in the tailbud; Kanki and Ho, 1997). Transverse sections enabled us to assess the extent of neural convergence at different stages of neurulation, whereas longitudinal sections revealed the elongation of neural tissue along the anterior–posterior (A-P) axis.
Analysis of transverse sections using these markers revealed that, at the TB stage, cells expressing high levels of Sox3C are found in two domains adjacent to the tailbud (Fig. 1A), that exhibits low levels of this marker (asterisk in Fig. 1A). As discussed below, the high levels of Sox3C identify populations of neural precursors, whereas low levels of Sox3C in the tailbud may mark mesodermal cells. By 4 som, neural cells populate the tailbud and form a continuous dorsal sheet that spans the midline and is flanked by epidermis (Fig. 1B). Concomitant with merging of the two lateral neural domains, the neuroectoderm thickens, transitioning from a single cell layer at the TB stage (Fig. 1A), 1–2 cell layers at the 2 somite stage (data not shown) and into a multi-layered tissue by 4 som (Fig. 1B). Narrowing and thickening of the neural tissue continues after the merge of the two lateral domains (Fig. 1B–D) and by 14 som, a rod-like structure is formed (Fig. 1D). This stage demarcates the completion of medial convergence, as the width of the PNT does not decrease further after 14 som (double arrows in Fig. 1D,E). Thus convergence, here defined as the narrowing of the neural tissue that takes place after merging of the neural domains, occurs between 4 and 14 som.
Imaging of longitudinal sections revealed that by 4 som, neural cells occupy the dorsal epiblast layer of the anterior tailbud (Fig. 2A), consistent with previous fate mapping studies (Kanki and Ho, 1997). The caudal limit of the neural domain at 4 som and continuing into 10 and 14 som stages, is at the level of KV (arrows Fig. 2A–C). However, by 20 som, posterior neural cells eventually spread over the entire surface of the tailbud, extending into the tail tip (Fig. 2D). To assess the amount of PNT elongation between 10 and 20 som, we measured the distance between the posterior-most domain of Sox3C expression and the tip of the tailbud as a ratio of total body length. We observed a significant decrease in this ratio as development proceeds, because the distance between the posterior end on the neural tube and the tip of the tailbud shortened faster than the total body length increased (10 som: 0.109 ± 0.011, n = 6 embryos; 14 som: 0.055 ± 0.002, n = 9 embryos; 20 som: 0.032 ± 0.005, n = 10 embryos; P value < 0.0001 between 10 and 14 som and < 0.0001 between 14 and 20 som. These values may be explained by the progressive lengthening of the neural tube, however it is important to also keep in mind that the anterior movement of posterior mesodermal tailbud cells is likely to contribute to the shortening of this distance (Kanki and Ho, 1997). Together these observations suggest that PNT elongation takes place at all stages analyzed.
Based on these observations, we arbitrarily subdivided PNT morphogenesis into three stages: Stage 1 (50% epiboly to 4 som) involves merging of ventrolateral neural domains into a continuous neural sheet that spans the midline; during Stage 2 (4 som to 14 som) the posterior neural tissue convergences medially, elongates and is shaped into a “rod-like” structure; and Stage 3 (beyond 14 som) is distinguished by elongation of the PNT in absence of further convergence. The rest of this study focuses mainly on events taking place during Stage 2 of PNT morphogenesis. We use several terms below to describe the cellular behaviors underlying PNT morphogenesis. The terms “merging” and “convergence” are used as described above, “shaping” specifically refers to the infolding event that orients the apical pole of neural cells toward the midline and “elongation” is the posterior extension of the PNT.
The Tailbud Mesenchyme Is Organized Into Tissue-Restricted Domains
Although Sox3C is generally considered a neural marker (Uwanogho et al., 1995; Rex et al., 1997; Streit et al., 1997; Mizuseki et al., 1998), we observed that it is expressed throughout the tailbud during Stage 1 (Figs. 1A, 3A). Cells expressing high levels of Sox3C are restricted to the dorsal–anterior epiblast and are likely to be neural precursors, based on previous fate mapping studies (Kanki and Ho, 1997). However, expression of Sox3C at low levels in ventromedial layers of the anterior tailbud (asterisk in Figs. 1A, 2A) was unexpected, as these cells are fated to become axial mesoderm (Kanki and Ho, 1997). Weak staining for Sox3C is unlikely due to nonspecific labeling, as this marker becomes restricted to neural tissue at later stages (Figs. 1C–E, 2B–D, 3B).
To verify the identity of cells expressing low levels of Sox3C, we immunolabeled adjacent sections through the tailbud with α-Sox3C and α-No Tail (Schulte-Merker et al., 1992). Ntl expression at the TB stage is low and is distributed throughout the medial and deep layers of the anterior tailbud. Interestingly, this pattern of expression appears to spatially overlapped with that of Sox3C (arrows in Fig. 3A,C), suggesting that tailbud cells expressing low levels of this marker may be mesodermal. By the 10 som stage, Sox3C and Ntl no longer overlap, as these markers become restricted to neural and axial mesodermal lineages, respectively (Fig. 3B,D). The levels of Ntl in the notochord are also increased at this stage (Fig. 3D).
In the mouse and chick embryo Sox3 is expressed throughout the epiblast and only later becomes restricted to the neurectoderm (Collignon et al., 1996; Penzel et al., 1997; Rex et al., 1997; Wood and Episkopou, 1999). We also observe that Sox3C is expressed throughout the zebrafish epiblast at 70% epiboly (Supporting Information Fig. S1A,B, which is available online), raising the possibility that the low levels of Sox3C protein we observe in the deep layers of the anterior tailbud correspond to residual protein and the high levels in the neural ectoderm may be protein produced de novo (Supporting Information Fig. S1). Consistent with this hypothesis, we observe that sox3c mRNA is only present in the epiblast layer of the anterior tailbud (Supporting Information Fig. S1C), in contrast to ntl mRNA that is found in deeper layers of the tailbud (Supporting Information Fig. S1D).
In addition to its expression in the mesoderm, Sox3C labeling is also seen in a row of three to four p63-positive cells, at the margin between the neural non neural ectoderm in the anterior tailbud (arrows in insets of Fig. 1B). Based on their location, these cells are likely to be neural crest. However, as with Ntl, the overlap between Sox3C and p63 is no longer observed at later stages (Fig. 1D). This observation suggests that Sox3C is transiently expressed in neural crest cells.
Thus, while Sox3C is not restricted to neural lineages during early stages of posterior development, its expression levels are tissue-specific, with high levels only observed in neural tissue and lower levels found in mesodermal or neural crest precursors. These observations are consistent with cell fate mapping studies in the zebrafish, demonstrating that the tailbud is organized into tissue-restricted domains as early as the TB stage (Kanki and Ho, 1997).
Convergence of Neural Tissue
During Stage 2, the PNT narrows dramatically along the M–L axis (Fig. 1B–D) and elongates along the perpendicular A–P axis. We hypothesized that this process could be brought about by several previously described cell behaviors that occur during gastrulation and neurulation in anterior regions, namely M–L intercalation (the rearrangement of cells along the M–L axis), delamination (a “dropping” of cells out of an epithelial sheet) or/and medial migration followed by migration along the A–P axis once cells have reached the midline (Keller et al., 1992, 2000; Wallingford et al., 2002; Solnica-Krezel, 2005). To image the cellular dynamics that narrow the PNT, we took advantage of the optical transparency of the zebrafish embryo to track the behavior of individual cells labeled with membrane-targeted green fluorescent protein (mGFP) in real time. Labeling with mGFP was performed by injecting DNA encoding this marker into newly fertilized embryos. Because DNA is mosaically inherited after injection in zebrafish, this allowed us to visualize cluster of cells in the neuroepithelium, as previously described (Hong and Brewster, 2006).
Time-lapse video microscopy of the tailbud, between 4 and 8 som (early mid Stage 2; 3 embryos), revealed that cells do not migrate individually, as do superficial cells in anterior regions of the zebrafish neuroepithelium (Hong and Brewster, 2006). Rather, cells in the posterior neural tissue appear to move en masse (Movie 1). Of interest, as cells approach the midline they “dive down” and disappear beneath the focal plane (Movie 1; red rectangular area in Fig. 4), suggesting that the neural tissue folds inward at the midline, as is known to occur in anterior regions of the zebrafish embryo (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006). Based on this observation, we ruled out intercalation between cells from opposite sides of the midline (contra-lateral intercalation) and medial followed by anterior or posterior migration as mechanisms contributing to the narrowing and elongation of the PNT. To investigate the cell movements occurring during the last part of Stage 2, we performed time lapse imaging on embryos between 10 and 12 som (2 embryos; Fig. 5). At these later stages, neural cells continue to move en masse and begin to elongate. They appear to be in close contact with the basal lamina and extend medially oriented membrane protrusions, reminiscent of cell behaviors previously reported in the hindbrain region.
To investigate other mechanisms that could contribute to the narrowing and elongation of the PNT during Stage 2, we traced the behavior of small clusters of cells in still frames of time-lapse videos. Only clusters of 3 or more cells, in which most cells remained in focus throughout the duration of the movie, were scored. This analysis revealed several intercalation events amongst cells on the same side of the midline, otherwise known as ipsilateral intercalations (n = 8 intercalation events observed of 8 cell clusters analyzed; three examples of which are shown in Fig. 6A–C). These intercalation events involved the repositioning of lateral cells into more medial regions.
As previously mentioned, the PNT undergoes a progressive thickening, that begins around the 2 som stage (Fig. 1A–D). Thickening is suggestive of cell delamination, a mechanism that could also account for the narrowing of the PNT. To address this possibility, both movies and still frames were analyzed for delamination cell behaviors. If delamination occurs, cells should “drop” out of the plane of view, allowing neighbors in more lateral regions to reposition themselves medially. The difference between this cell behavior and the afore-mentioned “infolding” mechanism, is that delamination may not be spatially restricted to the midline but is rather expected to occur throughout the neuroepithelium. We analyzed cell behaviors at both the 4 and 10 som stages and observed delamination at various positions along the M–L axis in movies (data not shown) and still frames (Fig. 6A, Cell 2). While the number of delamination events we observed is low, (n = 3 delamination events of 8 clusters analyzed), these cell behaviors were scored beginning at 4 som, by which stage the neuroectoderm is already considerably thickened (Fig. 1B).
Together, these data demonstrate that both ipsilateral intercalation and delamination are mechanisms used to narrow the PNT.
Shaping the PNT
Cell intercalation and delamination explain how the posterior neural ectoderm narrows, but it remains unclear how this tissue is shaped into a neural tube in which cells are polarized along the apicobasal axis and oriented in a stereotypic manner. Do cells coalesce to form the PNT in a random, disorganized manner or is there an infolding mechanism in place akin to what has been described in anterior regions (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006)? Infolding is suggested by the cell behaviors observed at the midline (Fig. 4, Movie 1). However, to further address this possibility, we examined the angular orientation of individual mGFP labeled cells in embryos at developmental time points between 4 and 20 som that were colabeled with α-Sox3C (Fig. 7A–E). mGFP-expressing cells were subdivided into two categories, medial cells (located immediately dorsal to KV) and lateral cells (located in regions lateral to KV), and were scored for the angular orientation of their long axis relative to the dorsoventral (D-V) axis of the embryo. Cells with angles between 0° and 30° were considered “vertical”; cells measuring between 31° and 60° were labeled as “oblique” and cells whose angles ranged between 61° and 90° were categorized as “horizontal” (Table 1). In lateral regions, cells exhibited a variety of angular orientations at the 4 and 6 som stage, as lateral-most cells were vertical (Fig. 7A) and mediolateral cells were either oblique (Fig. 7B) or horizontal at these stages. At later stages of neurulation, we observed that the majority of lateral cells adopted an oblique rather than a horizontal orientation (86.7% of oblique cells vs. 13.3% of horizontal cells at the 10 som stage, Fig. 7C). This observation was unexpected, as cells in anterior regions adopt a fully horizontal orientation by the neural rod stage (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006). In medial regions, cells appear to transition from a horizontal orientation at stages 4 and 6 som (Fig. 7A,B) to an oblique orientation by 14 som (100% and 96% of cells are horizontal at the 4 and 6 som stages, respectively, vs. only 58.8% of horizontal cells at the 8 som stage; Fig. 7C), which also differs from what was observed in anterior regions (Hong and Brewster, 2006). These findings are consistent with an infolding of the neuroepithelium, proceeding in a medial to lateral direction. Because medial cells infold first, they adopt a horizontal orientation before lateral cells. The final oblique orientation of both medial and lateral cells can potentially be explained by a dorsal elevation of the neuroepithelium above the mesoderm, after cells have began to orient medially. Indeed, in contrast to the anterior neural tube, the PNT lies on top of (rather than adjacent to) the paraxial mesoderm at 14 som and 20 som (Fig. 7D,E).
Table 1. Changes in Angular Orientation and Cell Shape During PNT Morphogenesisa
No. of Embryos
No. of Cells
Average cell angle (°)
Measurements and total cell counts are displayed as average number ± standard deviation. PNT, posterior neural tube; LWR, length-to-width ratios; som, somite.
2.2 ± 1.2
46.7 ± 27.9
2.4 ± 1.2
47.1 ± 25.2
3.2 ± 1.0
33.4 ± 16.9
5.2 ± 2.9
49.9 ± 9.0
3.5 ± 1.6
81.3 ± 10.7
3.2 ± 1.1
79.4 ± 17.7
4.4 ± 0.5
70.1 ± 20.4
3.4 ± 1.6
67.5 ± 14.7
To quantify changes in cell shape during NT morphogenesis, we also determined the length-to-width ratios (LWRs) of posterior neural cells (Table 1). The LWRs of lateral cells increase between 4 and 10 som (2.2 ± 1.2 μm at 4 som vs. 5.2 ± 2.9 μm at 10 som) while the LWRs of medial cells remain relatively constant (3.5 ± 1.6 μm at 4 som vs. 3.4 ± 1.6 μm at 10 som), suggesting that most of the changes in cell shape occur in lateral regions. Stretching of lateral cells, also observed in anterior regions (Hong and Brewster, 2006), may be required for cells to establish contact with the midline and span the width of the neuroepithelium.
Together these observations indicate that formation of the PNT during Stage 2 is brought about by three cell behaviors: ipsilateral intercalation, delamination and infolding of neural cells at the midline.
PNT Cells Undergo Progressive Epithelialization
Infolding of the neuroepithelium suggests that posterior cells have epithelial properties. To further determine the extent of epithelialization of posterior neural cells, we examined the expression of ZO-1 and aPKC, as apical localization of these markers is indicative of apicobasal polarity, a hallmark of epithelial cells. Embryos at the 8 som, 14 som, and 24 hours postfertilization (hpf) stages were immunolabeled with α-ZO-1 and α-aPKC and imaged in cross-sections through the tailbud region (Fig. 8). We observed that both markers are not expressed at 8 som (Fig. 8A). However, by 14 som apical ZO-1 and aPKC labeling is apparent in ventral regions of the neural rod (Fig. 8B) and extends to dorsal regions by 24 hpf (Fig. 8C). The dynamics of ZO-1 and a PKC labeling suggest that posterior neural tissue undergoes progressive epithelialization, that is completed by 24 hpf.
Cell Proliferation Occurs Uniformly in the PNT
The convergence movements described above not only narrow posterior neural tissue but also cause its elongation. However, given that convergence is for the most part complete by 14 som (compare Fig. 1D,E) while elongation is not (Fig. 2C,D), we sought to address whether cell proliferation might also contribute to PNT elongation during Stages 2 and 3. In principle, cell proliferation could either occur evenly along the neural tube or be concentrated in a particular region, such as the tailbud. Moreover, oriented cell division along the A-P axis specifically could be a driving force for PNT elongation.
To determine the distribution of dividing cells in posterior neural tissue, we immunolabeled embryos between TB and 20 som with the neural marker Sox3C and the mitotic marker α-phospho-Histone H3 (α-PH3) and imaged sagittal sections (Fig. 9A–C). The amount of cell division was determined by calculating the mitotic indices for three adjacent tail regions (regions 1–3, each ∼100 μm in length; Table 2). This analysis revealed that, at all stages examined, mitosis occurred fairly evenly throughout the posterior axis (even though mitotic indices appear higher in regions 1 and 2 than in region 3, these differences are not statistically significant) suggesting that the NT elongates uniformly during Stages 2 and 3. Very few cell divisions were observed in axial mesoderm (Fig. 9A–C).
Table 2. Mitotic Indices for Posterior Neural Cells in Control Embryos and Embryos in Which Cell Division Is Blockeda
(A) Mitotic indices for 3 different regions of the PNT at 4 som, 14 som, and 20 som in control and hydroxyurea/aphidicolin-treated embryos. (B) Illustration of the tail regions in which the α-PH3-positive neural cells were counted. Som, somite; PNT, posterior neural tube.
While the distribution of dividing cells appears uniform, it is possible that the orientation of the mitotic spindle is skewed toward one axis, resulting in enhanced growth along that axis (Gong et al., 2004). To determine the orientation of cell divisions during PNT morphogenesis, we first analyzed time lapse movies of mGFP-labeled embryos imaged from a dorsal view. These movies revealed that cell divisions occur along the M-L and A-P axes (Movie 1: M-L1 and A-P1 designate cells dividing along the M-L and A-P axes, respectively; D-V oriented cell divisions were not observed in these movies). In anterior regions, cells that divide along the M-L axis (perpendicular to the plane of the neuroepithelium) at the late neural keel and neural stage, have been shown to cross the midline (Papan and Campos-Ortega, 1994; Concha and Adams, 1998; Geldmacher-Voss et al., 2003). Midline-crossing divisions also occur in posterior regions, as evidence in Figure 5A1′–A2′.
In addition, we scored the orientation of mitoses in the tailbud using cross and longitudinal sections of embryos at 10, 14, and 20 som, labeled with α-Sox3C and the nuclear marker DAPI (4′,6-diamidine-2-phenylidole-dihydrochloride; two examples of cells dividing along the M-L axis are shown in Fig. 8B). Percentages of dividing cells calculated for cross-sections reflect cell divisions occurring along the M-L and D-V axis exclusively and percentages for sagittal sections correspond to A-P and D-V divisions only. We observed that at 10 som, the majority of divisions occur along the A-P axis (Table 3). At 14 som, there were more M-L and A-P divisions than D-V divisions and the percentages of M-L and A-P divisions were similar (Table 3). By 20 som, M-L divisions were the most abundant and there was roughly an equal ratio of A-P and D-V divisions (Table 3). Together these observations indicate that cell divisions occur along all three axes and that M-L, rather than A-P, divisions predominate in the PNT at stages following the completion of neural convergence.
Table 3. Orientation of Cell Divisions in the PNT
No. of embryos
No. of cells
(A) Analysis of cell divisions in cross sections. Percentages are calculated as the proportion of dividing cells that are oriented along the dorsoventral (D-V) and mediolateral (M-L) axes. Cells dividing along the anterior–posterior (A-P) axis were not scored in cross sections. (B) Analysis of cell divisions in sagittal sections. Percentages are calculated as the proportion of dividing cells in regions 1, 2 and 3 or in region 2 only (R2, numbers in parenthesis) that are oriented along the D-V and A-P axes. Cells dividing along the M-L axis were not scored in sagittal sections. PNT, posterior neural tube; som, somite.
AB 10 som
AB 14 som
AB 20 som
B. Sagittal sections
No. of embryos
No. of cells (R2)
D-V (R2) orientation
A-P (R2) orientation
AB 10 som
AB 14 som
AB 20 som
While these findings indicate that there is no bias toward A-P–oriented cells divisions in posterior regions between 10 and 20 som, it remains possible that such divisions occur in anterior regions, where neurulation is more advanced. Indeed, once the neural rod is formed, dividing cells in the hindbrain and trunk region cease to cross the midline and rather divide parallel to the forming ventricular surface (Geldmacher-Voss et al., 2003). We have observed that these divisions appear to orient primarily along the A-P axis (n = 4 out of 4 divisions observed in the hindbrain and trunk region [3 embryos]; see for example Fig. 5A3′–A4′). Thus, it is possible that A-P -biased divisions in more anterior regions, that have completed neural rod formation, provide a driving force for neural tube elongation.
Cell Proliferation Contributes Minimally to PNT Elongation
Despite the lack of bias for A-P–oriented mitoses, the high frequency of cell division in the PNT relative to the axial mesoderm suggests that cell proliferation may play an important role in PNT elongation during Stages 2 and 3. If this assumption is correct, then blocking cell division should result in a shorter NT. To inhibit cell division, wild-type (WT) embryos were treated with hydroxyurea and aphidicolin beginning at TB and ending at 20 som. Treated embryos appeared normal and showed no apparent signs of cell death, as compared to nontreated WT siblings (data not shown), consistent with previous observations (Tawk et al., 2002; Ciruna et al., 2006). Treated embryos and controls were labeled with α-Sox3C and α-PH3, sectioned sagittally and scored for mitotic indices and overall shape and size of the PNT.
A dramatic decrease in the number of α-PH3 positive cells was observed in treated embryos as compared to controls, demonstrating that the drug treatments were effective (Table 2; Figs. 8A′–C′, 9B). Surprisingly, blockage of cell division did not significantly affect the height of the PNT (Table 4; control: 30.8 ± 1.8 μm, treated: 28.3 ± 4.6 μm; P values for height > 0.2; Figs. 8A′–C′, 9B) and the width of the PNT was in fact slightly larger in treated embryos compared with controls (control: 46.5 ± 0.9 μm, treated 48.9 ± 3.2 μm; Fig. 10B). Further examination of cross-sectioned embryos revealed a decrease in the number of neural progenitor cells in treated embryos compared with controls (Table 4; control: 73.6 ± 15.4 cells; treated: 44.8 ± 13.5, Fig. 10B), as expected. However the reduction in cell number contrasted with an increase in cell size (estimated based on the size of Sox3C-positive nuclei; Table 4; height-control: 6.6 ± 1.1 μm, treated: 8.5 ± 1.9 μm; width-control: 3.5 ± 0.6 μm, treated: 3.8 ± 0.8 μm; Fig. 10B).
Table 4. Effect of Blocking Cell Division on the Size of the PNT and PNT Cells
A. Size of the PNT
No. of embryos
Height of PNT (average)
Width of PNT (average)
Length of PNT (average)
(A) Effect of hydroxyurea/aphidicolin treatment on the size of the PNT at 20 som. (B) Effect of hydroxyurea/aphidicolin on the size of PNT cells at 20 som. The size of cells was determined by measuring the dimensions of Sox3C-positive nuclei. Measurements and total cell counts are displayed as average number ± standard deviation. PNT, posterior neural tube; som, somite.
30.8 ± 1.8
46.5 ± 0.9
226.0 ± 16.9
28.3 ± 4.6
48.9 ± 3.2
199.5 ± 16.3
B. Size of PNT cells
No. of embryos
Length of nuclei (average)
Width of nuclei (average)
No. of Sox3C -positive cells
6.6 ± 1.1
3.5 ± 0.6
73.6 ± 15.4
8.5 ± 1.9
3.8 ± 0.8
44.8 ± 13.5
The relative length of the PNT was determined by measuring the distance between the posterior-most somite to the caudal-most Sox3C-positive domain (as KV is no longer visible at late somitic stages). We observed that the PNT was shorter in treated embryos than in controls (Table 4; control: 226 ± 16.9 μm, treated: 199.5 ± 16.3 μm; P < 0.05), but the visual effect of the treatment was very subtle (Fig. 9). These data indicate that blocking cell division from TB to 20 som stages has overall little effect on the width and height of the PNT and impacts PNT elongation only marginally. However, we cannot rule out that the increase in cell size observed in drug-treated embryos compensates for the absence of cell division, thereby explaining the subtle effect on PNT size. Given that cell size may have a significant impact on PNT elongation, we measured the surface area of nuclei in the tailbud region of untreated embryos at 4, 14 and 20 som, as an indirect measure of whether neural cells enlarge during late somitogenesis. These data revealed that the size of nuclei appears to in fact be decreased at 20 som relative to 4 som, arguing against cell growth as a contributing factor to elongation (data not shown).
These observations suggest an essential role for other mechanisms, such as oriented divisions in anterior regions and neural convergence, in elongating the PNT after gastrulation.
While neurulation in anterior regions has been studied extensively in several vertebrate model organisms, the cellular mechanisms that drive PNT morphogenesis are poorly understood and in the case of zebrafish are mostly unknown. In this study, we use a variety of tools for imaging neural tissue and tracking the behavior of individual cells in real time to provide an extensive analysis of neurulation in the tail region of the zebrafish embryo.
Stages on PNT Morphogenesis
In zebrafish, neural cells at all DV positions of the ectoderm give rise to neural tissue, with ventrally located ectoderm contributing specifically to caudal spinal chord (Kudoh et al., 2004). Upon completion of epiboly, ventral ectoderm cells populate the anterior tailbud region (Kanki and Ho, 1997), where they undergo PNT morphogenesis. We document here the different stages of PNT morphogenesis using KV as a reference point. Even though KV travels along the A-P axis during axis elongation, its position appears to remain constant relative to the posterior-most region of the neural tube. Indeed, time-lapse imaging revealed that neural cells do not migrate away from KV as they converge toward the midline. Furthermore, we observe that cell behaviors are homogenous throughout the region analyzed. Based on these observations, we infer that imaging at the level of this landmark, at different stages of development, provides snapshots of the same neural progenitor population over time and that the cell behaviors analyzed are likely to be representative of those occurring throughout the tailbud.
We demonstrate that PNT morphogenesis involves three distinct developmental stages. During Stage 1 (50% epiboly - 4 som) neuroectodermal domains derived from different regions of the embryo merge to form one continuous epithelial sheet across the dorsal midline of the tailbud. Stage 2 (4 som - 14 som) is marked by the dramatic narrowing and elongation of posterior neural tissue by convergence movements and shaping of this tissue into a neural rod. Neural convergence is brought about by intercalation of cells on the same side of the neuroectoderm (ipsilateral intercalation) and delamination of cells below the plane of the neuroectoderm. Shaping of the neural rod involves infolding of the neuroectoderm by angular re-orientation of cells such that their apical poles establish contact with the midline. Following 14 som, Stage 3, the PNT continues to elongate in absence of further convergence.
Before PNT Formation, the Tailbud Is Subdivided Into Distinct Domains
Cell fate mapping studies in zebrafish previously demonstrated that the tailbud is not a homogenous blastema but is rather subdivided into distinct regions (Kanki and Ho, 1997), similar to the tailbud organization in chick, mouse, and Xenopus embryos (Catala et al., 1995; Wilson and Beddington, 1996; Le Douarin et al., 1998; Davis and Kirschner, 2000). In zebrafish, the posterior tailbud domain gives rise to paraxial mesodermal fates exclusively. Cells in the anterior–dorsal region of the tailbud become neurons while notochord is derived by the anterior–ventral domain (Kanki and Ho, 1997). Notochord and spinal cord precursors in the anterior tailbud were found to be contiguous with these respective tissues, established during gastrulation, in the trunk (Kanki and Ho, 1997). Our observation that cells expressing high levels of Sox3C are restricted to the anterior epiblast layer of the tailbud while deeper anterior cells, appear to express lower levels of Sox3C and Ntl, is consistent with the known regionalization of the tailbud. The low levels of Sox3C in deep anterior cells, fated to become mesoderm, may be residual protein from the earlier expression of Sox3C throughout the epiblast.
Regionalization of the tailbud raises the interesting question of how different cell populations avoid mixing during the cellular rearrangements that takes in this region after gastrulation (Kanki and Ho, 1997). In addition, how do neural domains derived from opposite sides of the embryo accurately align and merge at the dorsal midline of the tailbud? Differential expression of members of the classical cadherin family of cell–cell adhesion molecules may provide an answer to the first question, as these molecules have been implicated in cell sorting (Steinberg, 2007). N-cadherin (N-cad), a prominent member of this family, is unlikely to provide cell sorting information as it is expressed at even levels throughout the tailbud (Harrington et al., 2007), suggesting that other members of the cadherin family may be involved.
Mechanisms of Neural Convergence
We provide evidence that neural convergence in posterior regions is mediated by infolding, ipsilateral intercalation and delamination. These cellular events are spatially distinct, as infolding occurs at the midline whereas intercalation and delamination take place in more lateral regions. Infolding appears to involve a zippering mechanism similar to what has been previously described in anterior regions (Papan and Campos-Ortega, 1994; Hong and Brewster, 2006), whereby cells move en masse toward the midline and medial (future ventral) cells infold first followed by lateral cells (that populate the dorsal neural tube). Delamination takes place between the tailbud and 2 som stage and is brought about by the exclusion of cells from the outer layer of neurectoderm. The cells that delaminate appear to rapidly re-establish contact with both the apical and basal surfaces of the neuroepithelium once internalized, as they increase their LWR and eventually span the entire width of the neuroepithelium. The end result of delamination is the formation of a pseudostratified epithelium and a net thickening and narrowing of the neuroectoderm. Delamination also contributes to neural convergence by creating room for lateral cells to populate more medial regions.
Mechanisms Contributing to PNT Elongation
The zebrafish embryo elongates dramatically during postgastrula stages. M-L intercalation is known to play an essential role in this process (Solnica–Krezel and Cooper, 2002; Glickman et al., 2003). However, the mechanisms that mediate extension of the neural tube are less well understood. One mechanism that may contribute to PNT extension is passive elongation due to stretching forces generated in the notochord during late somitogenesis stages. Indeed, previous studies have demonstrated that the notochord is a “driving force” for growth of the posterior body (Kitchin, 1949; Jacobson and Gordon, 1976). However, ventral marginal cell transplantation resulting in ectopic “tails” in the zebrafish has revealed that the dorsal spinal cord present in this tissue appears to undergo extension in absence of an underlying notochord (Agathon et al., 2003). This observation argues against the importance of axial mesoderm in PNT elongation in the zebrafish.
Kanki and Ho observed that zebrafish posterior neural precursors are restricted to the anterior tailbud at the TB stage, but eventually become distributed along the entire length of the tail, suggesting that either cell proliferation in dorsomedial regions or intercalation of cells associated with convergence drive elongation of the posterior neural tube (Kanki and Ho, 1997). Consistent with the former hypothesis, there appears to be a higher number of cell divisions in the PNT than in the underlying notochord. Thus, several events linked to cell proliferation could conceptually contribute to PNT elongation, namely: cell division per se in posterior regions, growth in neural cell size after cell division, preferential orientation of the mitotic spindle along the A-P axis and addition of new progenitors to the PNT, supplied by a stem cell population in the tailbud. These events need not be mutually exclusive.
To investigate the importance of cell proliferation on PNT elongation, we inhibited cell division and observed that this treatment had surprisingly little effect on the length of the PNT, as the latter appeared normal in sagittal sections and measurements revealed that it was only incrementally reduced. However, we cannot rule out that the increase in cell size observed in drug-treated embryos compensated for a decrease in cell number in drug-treated embryos, thereby enabling proper elongation. To address the possibility that cell growth after cell division could drive PNT elongation in untreated embryos, we measured the surface area of nuclei in the PNT and found that nuclei decrease rather than increase in size between stages 2 and 3, arguing against cell growth as a contributing factor.
During gastrulation in zebrafish, orientation of the mitotic spindle along the A-P axis has been shown to be a driving force for axis elongation (Gong et al., 2004), raising the possibility that A-P divisions may also promote PNT extension. While we do not observe an A-P bias between 10 and 20 som in the PNT, time lapse imaging of the hindbrain and trunk region during post neural convergence stages has revealed that cell divisions occur parallel to the ventricular surface (Geldmacher-Voss et al., 2003). Many of these divisions appear to orient along the A-P axis. Thus, it seems likely that divisions of neural precursors oriented along the A-P axis in the trunk and head region may provide one explanation for how the neural tube elongates.
In chick and mouse embryos a stem-cell population located in a region of the tailbud called the chordoneural hinge contributes cells to the growing end of the PNT (Cambray and Wilson, 2002, 2007; McGrew et al., 2008). This type of elongation involves cell proliferation but is distinct from cell division occurring in the neural tube proper. Whether a neural stem cell population exists in the zebrafish tailbud has not been investigated.
Together, these observations suggest that early stages of PNT elongation may be driven primarily by neural convergence during Stage 2 and by other mechanisms possibly involving A-P–oriented cell divisions in anterior regions during Stage 3.
How Similar Are Neurulation in Anterior and Posterior Regions in the Zebrafish?
In amniotes, neurulation occurs by means of two mechanisms in anterior and posterior regions, known as primary and secondary neurulation respectively (Colas and Schoenwolf, 2001). A distinguishing factor between these modes of neurulation is the level of epithelialization of neural cells at the onset of neurulation, as anterior cells are organized as an epithelial neural plate, while posterior cells are assemble in a condensed mesenchymal mass known as the medullary chord (Colas and Schoenwolf, 2001). This raises the question of whether PNT morphogenesis in zebrafish is different from neural tube formation previously analyzed in the hindbrain of this organism (Hong and Brewster, 2006).
Findings from this study provide evidence for homogeneity in the mechanisms of neurulation along the zebrafish A-P axis, although there are some subtle differences. At a morphological level, the neural tissue in both anterior and posterior regions appears to transition from a neural plate-like structure (observed upon completion of Stage 1 in the tailbud) to a neural keel and eventually a neural rod. At a cellular level, anterior and posterior cells re-orient their apical surface toward the midline and express the apical marker Z01 and aPKC only once the neural rod is formed. These observations suggest that cells along the entire length of the A-P axis have epithelial-like characteristic during neurulation, as they undergo an organized infolding process. However, the late onset of localization of these markers suggests that apicobasal polarity, a hallmark of epithelial cells, is only established once the neural rod is formed. At a molecular level, the calcium-dependent cell adhesion molecule N-cad appears to be required for neural convergence in both anterior and posterior regions (Hong and Brewster, 2006; Harrington et al., 2007), consistent with the homogeneity of the aforementioned cell behaviors along the A-P axis.
While, the mode of neurulation appears similar in anterior and posterior regions, there are some regional nuances. In posterior regions, cells were observed to move medially “en masse.” In contrast, superficial cells in the neuroepithelium appear to migrate individually toward the midline. Cellular rearrangements contributing to neural convergence are also different, as ipsilateral intercalation and delamination were not observed in anterior regions. Overall, these findings indicate that, unlike amniotes, zebrafish do not appear to undergo two distinct modes of neurulation.
Do Zebrafish Undergo Primary or Secondary Neurulation?
Whether or not neural tube formation in zebrafish more closely resembles primary or secondary neurulation has remained a point of controversy (Lowery and Sive, 2004). Certainly, the presence of a neural plate-like structure that undergoes epithelial infolding is reminiscent of primary neurulation (Lowery and Sive, 2004). However, arguing in favor of a mode of secondary neurulation is the neural keel-to-neural tube transition which resembles cavitation in amniotes at a morphological level (Schoenwolf, 1979; Schoenwolf and Delongo, 1980; Griffith et al., 1992). In addition, progressive epithelialization of zebrafish neural cells is akin to the mesenchymal-epithelial transition observed in the medullary cord. Given that the cellular basis for cavitation is not well understood, it is difficult to draw further analogies, but we speculate that the cell intercalation behaviors observed in zebrafish may also take place during secondary neurulation. Thus, we propose that neurulation in zebrafish most closely resembles secondary neurulation (Harrington et al., 2009).
Zebrafish Maintenance, Embryo Generation, Staging
Zebrafish embryos (Danio rerio) were collected from mated adult fish within 30 min postfertilization and maintained at 28.5°C until the desired developmental stage was reached (Kimmel et al., 1995). Offspring from AB or TL strains were used.
In Situ Hybridization and Immunolabeling
In situ hybridization was performed as described in Thisse et al. (1993). To synthesize antisense digoxigenin RNA probes, no tail (Schulte-Merker et al., 1994) was linearized with XbaI and transcribed with T7 and sox3 (Kudoh et al., 2001) was linearized with SalI and transcribed with T7 polymerase.
Whole-mount immunolabeling was carried out as described in Westerfield (2000). Embryos were fixed in 4% paraformaldehyde for 4 hr at 4°C. The following primary antibodies were used: α-Sox3C (Zhang et al., 2003) @1:2,000; α-Ntl (Schulte-Merker et al., 1992) @1:1,000, α-p63 (Lane and Benchimol, 1990), @1:200; α-β-catenin (BD Transduction Laboratories), @1:200; α-Phospho-Histone H3 (Molecular Probes) @1:200; α-ZO-1 (Molecular Probes) @1:200; α-aPKCζ (C-20; Santa Cruz Biotechnology) @1:200. Detection of primary antibodies was carried out using α-rabbit-Cy3 (Biomedia) @1:200 or/and α-mouse Alexa 488 @1:200. DAPI was used according to the manufacturer's instructions (Molecular Probes).
Injection of DNA
Plasmid encoding membrane-targeted enhanced GFP (mGFP, 40 ng/μl; obtained from R. Harland, University of California, Berkeley) was injected in embryos at the one- to two-cell stage and embryos were developed in the dark until the desired developmental stage. mGFP-injected embryos were either imaged using time lapse microscopy or processed for immunolabeling with α-Sox3C and stained with DAPI. Injection of mGFP was carried out using a Pico-injector (Harvard Apparatus, PLI-100).
Inhibition of Cell Proliferation
Cell proliferation was inhibited as described previously in Tawk et al. (2007). Embryos were cultured in embryo medium containing 100 M aphidicolin (Sigma) and 20 mM hydroxyurea (Sigma) dissolved in 4% dimethylsulfoxide, from approximately 90% epiboly until 24 hpf. At the desired stage, both untreated and drug treated embryos were immunolabeled with α-Sox3C and α-PH3 to confirm successful inhibition of cell proliferation and either imaged with a Zeiss Axioskop to measure body length (whole-mounts) or sectioned and imaged using a LSM510Meta Confocal Microscope.
Imaging and Sectioning
Time-lapse imaging: Analysis of single cell behaviors in live embryos was carried out at the 4, 6, 8, 10, and 12 som stage (1 movie per stage) for a duration of 2-hr time periods at 28.5°C. Embryos were dechorionated and imbedded into 1% agarose and covered with embryo media. Embryos were oriented so that the tailbud could be imaged from a dorsal view using a Leica SP5 Confocal Microscope. Z-stacks were taken from the dorsal-most to ventral-most regions of the embryo at a speed of 0.5 μm/sec.
Sectioning and imaging of fixed preparations: Whole-mount fluorescently labeled embryos were sectioned either transversely or longitudinally at a 60 μm thickness through the tailbud using a Vibratome (Vibratome, Inc.), as described in Harrington et al. (2007). To detect ZO-1 and aPKC expression, embryos were sectioned before immunolabeled. Fluorescently labeled specimens, (whole-mounts or sections) were placed in Aqua Poly/Mount (Polysciences, Inc.) and imaged using a Zeiss LSM 510-Meta Confocal Microscope or/and a Leica SP5 Confocal Microscope. A Leica M205 FA stereomicroscope was used for whole-mount imaging of samples for measurements of PNT length.
LWR (Length-to-width ratio) of cell and nuclei: were measured along the long and short axis of mGFP-labeled cells/Sox3c-labeled nuclei using the ImageJ64 software (NIH) in transverse sections. The LWR was calculated by dividing the length of the cell by its width. LWR is recorded as the average ± standard deviation.
Angular orientation of cells: The orientation of mGFP labeled cells was determined by measuring the angular distance between the long axis of each cell in relation to the embryonic D-V axis using the ImageJ64 software (NIH) in transverse sections. Cells were categorized into three groups: vertical, cells with an angular orientation between 0° and 30°; oblique, cells with an angular orientation between 31° and 60°; and horizontal, cells with an angular orientation between 61° and 90°. Angular orientation was recorded as the average angle ± standard deviation
Mitotic index: The mitotic indices in the neural tissue of control and drug-treated (aphidicolin and hydroxyurea) embryos were calculated by dividing the total number of proliferating cells in the neural tissue by the total number of Sox3C-positive neural cells in sagittal sections.
Orientation of cell divisions: The orientation of the mitotic spindle was analyzed in transverse and sagittal sections labeled with α-PH3 and DAPI. Dividing cells were categorized as oriented either along the A-P, D-V, or M-L axis. In sagittal sections, dividing cells were scored from the tail tip to a region located 300 μm anterior. Sagittal and transverse sections analyzed were ∼ 60 μm thick and at the level of KV.
Measurements of the PNT
Length and height of the PNT: Measurements for PNT length and height were calculated using longitudinal sections by measuring: (1) the distance between the posterior-most domain of Sox3C and the end of the tailbud as a ratio of body length or/and the distance between the posterior-most somite and the posterior end of the neural tube; and (2) height- the distance between the roof plate and floor plate of the PNT). Measurements were made using either LSM510Meta Software, Leica Application Suite Software and the Metamorph software. Measurements are recorded as the average ± standard deviation. Statistical analysis was performed using the Student t-test.
Width of the PNT: Transverse sections were used to measure the width of the PNT (the distance between the lateral edges of the PNT at its widest point). Measurements were made using either LSM510Meta Software or/and Leica Application Suite Software. Measurements are recorded as the average ± standard deviation.
We thank Marnie Halpern and Daphne Blumberg for their helpful comments on the manuscript, Chere Petty for assistance with imaging, and Michael Summers and Justine Johnson for their support. We also thank Michael Klymkowsky for his generous gift of the α-Sox3C antibody and Jeff Leips for helping us with the quantitative analysis of the data. R.B. was funded by a National Science Foundation grant, M.H. received a NIGMS initiative for minority student development grant, and the Leica SP5 confocal microscope was purchased with funds from the National Science Foundation.