Sequential changes in cellular properties accompanying amniote somite formation

Abstract Somites are transient structures derived from the pre‐somitic mesoderm (PSM), involving mesenchyme‐to‐epithelial transition (MET) where the cells change their shape and polarize. Using Scanning electron microscopy (SEM), immunocytochemistry and confocal microscopy, we study the progression of these events along the tail‐to‐head axis of the embryo, which mirrors the progression of somitogenesis (younger cells located more caudally). SEM revealed that PSM epithelialization is a gradual process, which begins much earlier than previously thought, starting with the dorsalmost cells, then the medial ones, and then, simultaneously, the ventral and lateral cells, before a somite fully separates from the PSM. The core (internal) cells of the PSM and somites never epithelialize, which suggests that the core cells could be ‘trapped’ within the somitocoele after cells at the surfaces of the PSM undergo MET. Three‐dimensional imaging of the distribution of the cell polarity markers PKCζ, PAR3, ZO1, the Golgi marker GM130 and the apical marker N‐cadherin reveal that the pattern of polarization is distinctive for each marker and for each surface of the PSM, but the order of these events is not the same as the progression of cell elongation. These observations challenge some assumptions underlying existing models of somite formation.


| INTRODUC TI ON
Somites, first described by Marcello Malphigi (Malpighi, 1686), are transient structures, forming sequentially in head-to-tail order at regular time intervals. The pattern of somites is fundamental for the organization of the adult segmental body plan as it guides the associated pattern of peripheral nervous system elements (nerves, neural crest cells and ganglia) and generates the skeletal musculature as well as the vertebral column. In the last few decades, several models have been proposed to explain the mechanisms responsible for controlling the size, number and timing of somite formation in space and time . These include pre-patterning (Christ et al., 1974;Meier, 1984;Menkes & Miclea, 1962;Menkes & Sandor, 1969;Packard Jr & Meier, 1984), the 'clock and wavefront' (Cooke & Zeeman, 1976) and the 'cell cycle' (Collier et al., 2000;Primmett et al., 1988Primmett et al., , 1989Roy et al., 1999) models, as well as 'reaction diffusion', 'clock and trail' (Kerszberg & Wolpert, 2000;Meinhardt, 1982), the 'wave and cell polarization' model (Beloussov & Naumidi, 1983;Polezhaev, 1992) and the 'progressive oscillatory reaction-diffusion' (PORD) model (Cotterell et al., 2015). Some molecular evidence apparently consistent with the 'clock and wavefront' model (Dubrulle et al., 2001;Dubrulle & Pourquie, 2004;Morelli et al., 2009;Oates et al., 2012;Palmeirim et al., 1997) has resulted in this model becoming dominant in the literature. However, it has also been argued that the 'clock-and-wavefront' model has been interpreted in various ways in the literature, resulting in several 'sub-models' that are not identical to each other (Glazier et al., 2008;Hester et al., 2011). Among the proposals is the idea that a combination of cell-repulsion and cell-adhesion in adjacent cells can be coupled to the events of somite formation to explain the progression of boundary formation (Glazier et al., 2008, Hester et al., 2011. Surprisingly, very few experimental studies have focused on cellular events within the PSM, and these have mainly examined aspects of the mesenchymal-to-epithelial transition (MET), performed cell tracking or studied extracellular matrix (ECM) assembly (Bellairs, 1979;Duband et al., 1987;Kulesa & Fraser, 2002;Martins et al., 2007Martins et al., , 2009. Some of the results of these studies are apparently contradictory, such as whether MET is initiated in the anteromedial or the antero-dorsal PSM (Beloussov & Naumidi, 1983;Kulesa & Fraser, 2002;Martins et al., 2009;Polezhaev, 1992). Most models of somitogenesis that have been proposed (see above) largely ignore the cellular dynamics within the PSM and generally assume that somite formation involves a sudden or 'catastrophic' MET causing cells to aggregate as a sphere. The main exceptions are the 'wave and cell polarization' model (Beloussov & Naumidi, 1983;Polezhaev, 1992) and three multi-scale models (Dias et al., 2014;Glazier et al., 2008;Hester et al., 2011), all of which are based on various cellular properties. Here, we address the dynamics of cell shape changes, the progression of how cells aggregate into future somites, and the dynamics of cell polarization, using scanning electron microscopy (SEM), immunostaining and confocal microscopy.
We find that the events of MET begin very posteriorly, therefore a long time before the formation of the respective somite. This implies that, rather than a sudden and catastrophic event just prior to the formation of a somite, epithelialization and cell polarization are gradual events. Moreover, these events are not exactly contemporary with each other because each appears at a distinct axial level within the PSM. A particularly surprising finding was that the different surfaces of the PSM undergo MET with different dynamics, rather than corresponding to the timing of formation of each somite. Since expression patterns of the 'segmentation clock' genes do seem to have boundaries covering the cross section of the PSM, these observations raise the question of how these gene expression patterns relate to the cellular events of somitogenesis.

| Embryos and scanning electron microscopy (SEM)
Fertilized domestic hens' eggs (Gallus gallus, Brown Bovan Gold, Henry Stewart & Co.) were incubated at 38°C to stages HH 11 (11-15 somites) (Hamburger & Hamilton, 1992). A newly formed somite 1 (s1) is defined as such when it is fully separated from the PSM, whereas somite 0 (s0) (at the tip of the PSM) is not separated by a completely formed cleft (Christ & Ordahl, 1995). In this study, we did not consider the first five somites (future occipital somites) because they form differently and have different structure to those of the trunk (Dias et al., 2014;Hamilton & Hinsch, 1956;Lim et al., 1987).
Embryos were harvested in PBS and processed for SEM as described (Bellairs, 1979), with the modification that after fixation in sodium cacodylate each embryo was cut only once, either transversely or sagittally, with a tissue chopper (Mickle Laboratory Engineering).
PSM lengths were measured (see Supplementary Methods) and excess tissue was removed. Next, embryos were dried in CO 2 (Leica CPD critical point dryer), mounted and sputter-coated with gold/ palladium. Images were taken on a JEOL JSM-740IF Field Emission Scanning Electron Microscope (SEM) with 2000x magnification at 2KV and pressure of 5.25 10 −4 Pa.

| SEM image processing and aspect ratio analysis
Montages of the images were made using Photoshop CS6 and analysed with (FIJI) (Schindelin et al., 2012). A touch screen (SmartPodium 624) and touch pen were used to draw the outlines of each cell manually using the 'freehand selection tool' in FIJI. Only cells that were in focus and not covered by neighbouring cells were considered. Next, each cell outline was added to the Region of Interest (ROI) manager tool in FIJI. The aspect ratio (AR) for each cell was automatically calculated in FIJI by dividing the longest by the shortest dimension of the outline of a given cell as seen in the (two-dimensional) SEM images. The AR values were colour-encoded using a pseudo-colour lookup

| 3D imaging, processing and analysis
3D Images were acquired with an Olympus Fluoview FV1000 confocal microscope and processed in FIJI (see Supplementary Methods).
For each sample, the image was optically re-sliced in the transverse plane and the PSM was divided into medial, lateral, dorsal and ventral domains, with each domain further divided into apical and basal halves (for details see Supplementary Methods). Next, the pixel intensity differences between apical and basal compartments were calculated for each domain and plotted against PSM distance represented as 0%-100% (see above and Supplementary Methods). To identify when the observed changes occur, two embryos with DiI-

| Whole-mount in situ hybridization
Embryos were fixed in 4% PFA and stored in methanol at −20°C.

| Sectioning and imaging
Images of whole-fixed embryos were acquired with an Olympus SZH10 Stereo microscope equipped with a Q-Imaging Retiga 2000 K camera, controlled using QCapturePro software. The embryos were positioned as desired in depression slides in PBS. Next, embryos were embedded in paraffin wax and sectioned sagittally at

| Hybridization chain reaction
Embryos were fixed in 4% PFA and stored overnight at −20°C in methanol. Hybridization Chain Reaction (HCR) (Dirks & Pierce, 2004) was performed according to the HCR v3.0 protocol from the manufacturers (Molecular Instruments) for whole-mounted chicken embryos. Embryos were mounted on glass slides using SecureSeal 13 mm × 0.12 mm imaging spacers (Grace Bio-Labs). A drop of SlowFade Gold (Invitrogen) was placed on the embryo, which was then covered with a No. 1 coverslip. Embryos were imaged using a Leica SP8 laser scanning confocal microscope with an Apochromat 40X 1.4 NA oil objective. 3D maximum projections and sagittal sections were generated using Imaris 8.2 (Bitplane).

| RE SULTS
3.1 | Cell shape changes during somitogenesis 3.1.1 | Cell shape changes in sagittally fractured PSM A total of 8318 cells from 12 mid-sagittally fractured embryos were classified as belonging to the dorsal or ventral PSM domains or to the core (Supplementary Table S1) and AR frequency distributions per domain were plotted as histograms (see Supplementary Figure S1).
An example of a mid-sagittally fractured embryo, with cells colourcoded for their AR, is presented in Figure 1a Thus, for the sagittally fractured PSM, the order of epithelialization as indicated by AR is as follows: first dorsal at 40% of PSM distance, then ventral at about 70%; core cells never epithelialize.

| Cell shape changes in transversely fractured PSM
For the transverse PSM fractured embryos, 2267 cells from 20 embryos were assigned to dorsal, ventral, medial, lateral or core do- Table S2). Examples of transversely fractured embryos at different rostrocaudal levels of the PSM are shown in Cells of non-epithelial morphology with long protrusions are found in the core at all PSM levels, suggesting that these cells may be motile, consistent with previous reports of cell movement both within and between somites (Kulesa et al., 2007;Kulesa & Fraser, 2002;Martins et al., 2009), and in the PSM (Benazeraf et al., 2017).  Supplementary Table S2) because it was defined as a single layer of cells that is neither dorsal nor ventral nor core, and therefore comprises relatively few cells.

AR values for cells in each domain
Despite this, clear trends are apparent from the plots. The AR of cells in the dorsal domain starts to increase very early, at 40% of the PSM length, consistent with the results from the sagittally fractured PSM.
For medial cells, the AR starts to increase at 55% PSM. In both the ventral and lateral domains, AR starts to increase at 70%, also consistent with the sagittal sections. Cells in all four domains (excluding the core) increase their AR with PSM position (t-test p = 2 × 10 −16 ). The cells of the core domain do not increase in AR at any PSM position (t-test p = 0.016). No significant differences were found between the left and the right PSM at each position (t-test p = 0.608). Statistical In summary, cells located at the edges of the PSM (dorsal, lateral, ventral, and medial) begin to undergo elongation (increase in AR) at different rostro-caudal levels of the PSM: dorsal cells start elongating first (40%), followed by medial cells (55%), and finally the ventral and lateral domains, which undergo this transition at the same time (70%). In contrast, the core cells appear to be trapped between epithelializing cells, and never elongate except for occasional cells with long protrusions.

| Cell shape changes in sagittally fractured somites
Among the 12 embryos that were fractured sagittally through the segmental mesoderm, several fracture planes included the newly formed somites (s0-s5). In total, 2180 somite cells were analysed and assigned to dorsal, ventral, anterior and posterior domains and the somitic core (somitocoele) (Supplementary Table S4 Table S6). This result suggests that the cells of the epithelial walls of the somite continue to elongate, while the core cells remain mesenchymal. Wilcoxon pairwise comparison of consecutive somites revealed significant differences between the anterior domains of s1 and s2, and also between the posterior domains of these somites (Supplementary Table S7). This is because somites 0 and 1 do not have cells of high AR in their anterior and posterior domains, but the next somite, s2, has more elongated cells in its anterior and posterior domains, and more rostral (older) somites have highly elongated cells in those domains (Figure 3a,b). As no significant difference was found between the anterior and posterior domains within each somite, the anterior and posterior domains of consecutive somites were compared with a two-sided t-test Collectively, the results from sagittally and transversely fractured PSM suggest that MET, as indicated by increase in AR, starts at the caudal end of the PSM at about 40% of the length of the PSM. Therefore, these cellular changes occur gradually long before a somite buds off from the PSM, starting from the most dorsal cells, followed by those placed medially, and finally those in the ventral and lateral aspects of the PSM, like a long rectangular box gradually being constructed by laying down the roof, then one side, then the floor, and finally the other side before cutting it into cubes. The anterior and posterior cells of a somite elongate only after the border is formed. The border between a newly formed somite and the anterior PSM resembles a ball and socket, as previously described (Kulesa & Fraser, 2002;Martins et al., 2007). The posterior domain of the preceding somite contains more elongated cells than the anterior domain of the subsequent somite, which may indicate how the border is maintained.

| Dynamics of cell polarity changes during somitogenesis
To obtain a 3D view of how cell polarity changes within the PSM as cells progress towards forming a somite, whole-mount embryos were stained for cell polarity markers GM130 (a Golgi marker), PAR3, ZO1 and PKCζ, and N-cadherin (the latter as an  Figure S11D-F), a pattern referred to as 'the basket' in a previous report (Martins et al., 2009) The dorsal domain then starts to polarize more rapidly at about 80% of the PSM; the last domain to polarize is the lateral aspect, just prior to somite formation at about 95% of the PSM.
Ncad staining (Figure 4e) reveals that the dorsal domain starts to polarize at 60% of the PSM and gradually increases anteriorly.
The medial, ventral and lateral domains epithelialize almost simultaneously at about 90% of PSM; thereafter epithelialization occurs rapidly, as suggested by the steep slope, consistent with a previous report (Duband et al., 1987).
In summary, cells in the dorsal, medial, lateral and ventral domains of the segmental mesoderm polarize very gradually. For a given polarity marker, each domain starts the process at a different level of the PSM and undergoes polarization at a different rate (and with characteristic patterns for each marker). Our findings are summarized in Figure 4f, which combines our results on MET (as indicated by change of AR) and cell polarization (indicated by GM130, PKCζ, PAR3, ZO1 and Ncad staining). The earliest events of polarization (PKCζ) and epithelialization (AR) are observed in the dorsal domain at around 40% of the rostro-caudal length of the PSM. Then polarization begins ventrally at 50% of the PSM, prior to epithelialization at about 65% of the PSM, followed by lateral polarization, and finally followed by the medial domain.

| Relationship between PSM position and the timing of segmentation
The relative rostro-caudal position of a given cell within the PSM is also a function of time elapsed since the cells entered the PSM from the primitive streak, and consequently of the time remaining for those cells before segmentation occurs. Because the PSM is a spati-  (Figures 5 and 6a,g black), slowing down as somitogenesis begins to take place, as previously described (Gomez et al., 2008).
The distance from the most caudal (most recently formed) somite border to the front of the labelled cells was plotted against time in reverse order to follow how soon the labelled group of cells contributed to somites, and whether this rate was constant ( Figures 5   and 6h). Despite the difference in the rate and overall number of somites formed for the two embryos, both embryos displayed the same trend: initially cells progressed slowly along the PSM before starting to accelerate, from about 10 h for embryo 1 and 11 h for embryo 2. After the initial 10-11 h, the rostral progression of cells in the PSM sped up at a constant rate (Figures 5 and 6h). This is consistent with previous studies (Benazeraf et al., 2017;Selleck & Stern, 1991;Stern et al., 1988) which suggest that cells intermingle extensively in the caudal PSM and then gradually stop doing so as they become located more rostrally. This new analysis therefore provides an approximate translation of PSM position (distance) at which a given event occurs into the time at which that event happens. The PSM distance [%] scale on the y axis represents the distance along the PSM to the level of the forming somite [μm] (Figures 5 and 6h), and the X coordinates from the inflection and 10% distance points were used to extrapolate PSM position to the timing of events within the PSM. Figures 5 and 6h summarizes these findings.

| Do the rostral and caudal subdivisions of the somite extend to the core?
Because core cells never epithelialize, we asked whether they behave differently from other somite mesoderm cells in other aspects of their cellular dynamics, for example whether they have rostrocaudal somite identity. The rostral-somite marker EphA4 (Baker & Antin, 2003;Gilardi-Hebenstreit et al., 1992;Schmidt et al., 2001) and caudal-somite markers Meso2 (Buchberger et al., 2002), Uncx4.1 (Schrägle et al., 2004), LFrng (McGrew et al., 1998 and Hairy-1 (Palmeirim et al., 1997)  Meso2 is expressed in the anterior PSM as 1-3 narrow bands (Buchberger et al., 2002). Embryos with two bands were selected for further analysis (Figure 7a verse orientation. Therefore, in newly formed somites, there is indeed a boundary that extends to the core, but this is not precisely orthogonal to the axis of the embryo. This observation suggests that von Ebner's fissure, which does appear to separate exact somite halves , may not exactly coincide with the boundary of expression of these markers. This suggests that cells adjust their expression a little later, or that cell mixing occurs at this stage, or that some cells may be eliminated. It has been reported that some core cells can incorporate into the epithelial wall of the somites (Huang et al., 1994(Huang et al., , 1996Kulesa et al., 2007;Kulesa & Fraser, 2002;Martins et al., 2009), which may contribute to tidying up this boundary.
In contrast to the pattern in formed somites, the domain of expression of EphA4 in the PSM appears much larger than the prospective rostral half (Figure 8a,d,e). A few cells apparently co-express both markers. Again, the change between the domains of expression before and just after segmentation could suggest cell sorting or changes in expression. In conclusion, our results suggest that the rostral-caudal subdivision of the somite extends into the core, although this boundary may not correspond precisely to that defined by von Ebner's fissure.

| Order of cellular events leading to somitogenesis
A previous study reported that the posterior two-thirds of the PSM is mesenchymal and that the first signs of epithelialization are observed in the anterior PSM (Duband et al., 1987). The dorsal PSM appears to epithelialize more posteriorly (earlier) than the ventral PSM (Bellairs, 1979) and the medial PSM before the lateral PSM (Kulesa & Fraser, 2002;Martins et al., 2009). The present SEM study suggests that epithelialization begins at a position 40% of the length of the PSM (from the caudal end) for dorsal PSM cells, then medially at the 55% position, and then at the same level for ventral and lateral domains at 70% PSM. Thus, the first signs of epithelialization (40% PSM length) appear much more posteriorly (earlier) than previously reported (Bellairs, 1963;Bellairs, 1979;Duband et al., 1987;Kulesa & Fraser, 2002;Martins et al., 2009). Sagittal and transverse SEM sections also clarify the order of events between the dorsal and medial PSM. In contrast to previous reports of somitomeres spanning at least half the length of the PSM (7-9 prospective somites) (Meier, 1984), we could only observe 'pre-somite'-like cell arrangements for, at most, 1-3 presumptive somites in the most anterior-PSM.
Paraxis (TCF-15) is considered a good marker for the onset of epithelialization as it is first expressed in the anterior PSM (and retained in formed somites). Paraxis mutants generate somite-sized arrangements of cells but these do not have epithelial structure (Burgess et al., 1995(Burgess et al., , 1996Kulesa et al., 2007). However, the present results reveal that the first signs of epithelialization are seen at 40% position in the dorsal PSM, well before paraxis is expressed.
Epithelialization seems to generate discrete epithelialized clusters in the dorsal and then medial PSM (Figures 1h and 2d,f). This raises the question of what regulates the appearance and spacing of these clusters. Previous studies showed that ectoderm is required for epithelialization and, together with the neural tube, is required for paraxis expression, which is consistent with the appearance of the clusters in the dorsal and medial aspects of the PSM (Correia & Conlon, 2000;Lash & Yamada, 1986;Packard Jr, 1976;Sosic et al., 1997). However, the PSM was also shown to epithelialize autonomously, provided that the fibronectin matrix is intact (Lash & Yamada, 1986;Rifes et al., 2007).
3D-confocal analysis of polarity markers PKCζ , PAR3, ZO1, Golgi marker GM130 and apical marker N-cadherin revealed that each marker starts to be localized at a characteristic position and domain of the PSM: this pattern is different for each marker as well as from the pattern of cell elongation described above (Figure 4 and Supplementary Table S8). This observation suggests that epithelialization and polarization may be regulated independently of each other.

| Are the observed PSM cellular dynamics compatible with existing models of somite formation?
To date, none of the models can quite explain all the experimental observations. For example, the 'clock and wavefront' model is challenged by the finding that a single heat-shock can generate repeated (periodic) defects in segmentation, whereas the model predicts only one anomaly should occur (Cooke, 1978;Elsdale et al., 1976;Primmett et al., 1988Primmett et al., , 1989Roy et al., 1999). In addition, somites can be generated from posterior primitive streak, which normally does not form paraxial mesoderm; these somites form almost simultaneously, are not arranged along a line, and lack oscillatory expression of the 'clock' genes of the Notch pathway, yet have approximately normal size and shape. Therefore, they appear to form without participation of either the clock or a wavefront (Dias et al., 2014;Streit & Stern, 1999). In silico simulation predicts that somites can be generated spontaneously if cells are allowed to have neighbours, can 'see' both the apical and the basal sides of those neighbours, and can maximize their adhesion and epithelialize once they come into close contact with each other (Dias et al., 2014).
Hence, studying cellular PSM dynamics in vivo is of interest.
The 'clock and wavefront' model proposes that a group of PSM cells undergo 'rapid cell change' at the same time, as a sudden 'catastrophic' change (Cooke & Zeeman, 1976). Thus, the model predicts that a group of PSM cells cluster together in synchrony, rapidly, and at the same time as they commit to form a somite together. In this study, analysis of the sequence of cell shape and polarization changes revealed that the PSM is subdivided into domains that undergo the changes separately, and for each domain, the change is gradual (except for lateral PKCζ, Figure 4), begins at a different PSM level, and occurs at a different time and rate-in other words, the timing of events does not seem to be characteristic for a particular somite. These observations seem to contradict the idea that the clock-and-wavefront functions primarily to determine the size of somites (by regulating the number of cells that will later segment together) and the timing of segmentation. However, they do not exclude the possibility that the 'rapid cell change' may be manifested in other cellular behaviours not studied here, such as cell adhesion, or simply the 'commitment' to undergo these behaviours. The finding that the core domain never epithelializes is also not obviously consistent with the 'clock and wavefront' model, but may be accommodated by invoking that core cells may lack 'competence' to respond to the clock and wavefront information.
The 'cell cycle' model proposes that the cells within the PSM are organized according to their age order and their cell cycle phase, and that some cell-cycle-coupled event taking place prior to segmentation is responsible for gating cells into cohorts that will segment together (Collier et al., 2000;Primmett et al., 1989;Stern et al., 1988).
Although the original proposal was that this gating event might take place one cell cycle (about 10 h) before overt segmentation, it could also take place earlier. This would be more consistent with the finding that groups of cells are seen to start epithelialization as early as the 40% position in the PSM (which is a little longer than 10 h prior to overt somite formation). However, since epithelialization events start at different times/positions for each domain of the PSM, this would suggest that the interpretation of the gating information may differ for each property and position. As with the clock-andwavefront model, the failure of core cells to participate in epithelialization could be explained by lack of competence of these cells to interpret the information.
The 'reaction diffusion' model proposes a posterior-to-anterior morphogen gradient to which cells respond while oscillating between two cellular states (Meinhardt, 1982). In the context of epithelialization dynamics, the two cellular states could be epithelial and mesenchymal, one of which becomes fixed after a cell experiences a certain level of the morphogen. Because neighbouring cells promote the opposing state, this scenario is compatible with epithelialized clusters observed at the dorsal-posterior-PSM, but not in the dorsal-anterior-PSM where all cells are already epithelialized.
As with the cell cycle model, because the epithelialization pattern is domain specific, the model would also require separate, and different, interpretations of this information at each position/domain, including the core.
The PORD model (Cotterell et al., 2015) proposes that cellular interactions at local level induce the neighbouring cells to change their state, and that this change propagates posteriorly. The epithelialization pattern within the PSM fits this concept as the epithelialization is more advanced in the anterior than posterior PSM for most of the domains. As for other models, the PORD model does not predict that core cells never epithelialize, that each PSM domain epithelializes at a different PSM level, or that epithelialized clusters appear at posterior-dorsal-and posterior-medial-PSM levels, which may require differences in interpretation of the information.
The 'wave and polarization' model stemmed from observations of epithelialization dynamics within the PSM (Beloussov & Naumidi, 1983;Polezhaev, 1992  Our initial observations that dorsal-PSM forms an epithelial monolayer long before somite formation led to a model being proposed in which an interplay between apical constriction and a posteriorly progressing 'activation front' alone could segment the dorsal-PSM (Adhyapok et al., 2021). The model predicts that segment size increases as the speed of an 'activation front' increases, and that there is an inverse rate of increase for apical contractility.
However, the behaviour of other PSM domains was not integrated into this model-perhaps, separate activation fronts operate in each domain, but it is then unclear what is responsible for the formation of each somite, involving all the surfaces of the PSM (and not the core).
In conclusion, none of the existing models can easily explain our present findings. At the very least, differences in the interpretation of the segmentation information are required between different groups of cells to account for differences in the timing of segmentation in the different domains of the PSM, and the failure of the core to participate in segmentation.
The work presented here uncovered no significant difference in AR between the anterior and the posterior domains within forming and formed somites, suggesting that both domains may epithelialize together ( Figure 3 and Supplementary Table S5). Epithelialization within the newly formed somite was reported to coincide with separation from the PSM, with the separated somite having a rosette structure (Kulesa & Fraser, 2002). However, the present results show differ-  Table S5). This may be a consequence of epithelialization order within the PSM.
The core cells of the somite never epithelialize (Figure 3 Supplementary Tables S5 and S6). However, Beloussov and Naumidi proposed that the 'cell fan' of the posterior-somite domain is generated when the epithelialized dorsal domain recruits core cells, which then elongate. The 'fan' was said to progress in a dorsal-to-ventral direction (Beloussov & Naumidi, 1983), but in Figure 1e Table S4, Figure 3a,b). This suggests that the border between neighbouring somites may form due to different degrees of epithelialization between neighbouring somites, a mechanism previously proposed for the separation of somites from the PSM (Beloussov & Naumidi, 1983). Structures similar to the 'cell fans' are observed between somite 1 or 0 and the anterior front of the PSM (Figure 3a,b).
Cells within this region were not allocated to anterior-core-PSM, and were not analysed. However, Beloussov and Naumidi's proposed that the somitic border forms as a consequence of epithelialization differences are challenged by the paraxis mouse null mutant which undergoes segmentation but does not epithelialize, suggesting that epithelialization is dispensable for segmentation. Moreover, their model does not provide a mechanism for the regulation of somite size or the timing of segmentation.

| THE CORE CELL S
The core cells, or somitocoele, are characteristic for amniotes like mouse and chick but are not a feature of anamniotes like zebrafish or frog, as the cellular organization of the somite is different in these taxa. In chick, core cells occupy the centre of a somite, but can relocate to the epithelial rosette, as established by time-lapse observation of living embryos (Kulesa et al., 2007;Kulesa & Fraser, 2002;Martins et al., 2009) as well as fate mapping of core cells (Huang et al., 1994(Huang et al., , 1996. Core cells have been reported to arise by proliferation and ingression from the epithelialized rosette (Martins et al., 2009;Wong et al., 1993). Here, the core cells seem to be recruited from the centre/middle cells of the unsegmented PSM, which are engulfed by the elongating cells of the surrounding domains. This observation raises the question as to whether core cells are merely an architectural leftover, or whether there is a mechanism regulating formation and number of the core cells, as well as a specific molecular identity and polarity.
The core cells of the PSM behave differently from the cells of other domains. Our results suggest that the core population never epithelializes ( Figure 1m). Also, although polarization was not directly studied within the core domain, the core cells do not seem to polarize, as no increased immunostaining for polarity markers is observed within this domain ( Supplementary Figures S3-S12). Since our observations are based on fixed embryos, they do not allow us to establish whether there is shuttling between the core cells and other domains at the PSM level. Indeed, this possibility cannot be excluded as other studies have suggested extensive cell mixing at posterior-PSM levels (Benazeraf et al., 2010(Benazeraf et al., , 2017Selleck & Stern, 1991;Stern et al., 1988). Our findings with HCR to reveal expression of a caudal (Uncx4.1) and a rostral (EphA4) marker in the same embryo show that the boundary between these markers changes over time. Prior to segmentation, the domains of expression of EphA4 are larger than expected for a somite half; as the somite forms, the domain of expression of Uncx4.1 becomes larger than that of EphA4. Moreover, the boundary between the two domains in the somite is not precisely orthogonal to the embryo's long axis. These observations are consistent either with cell sorting or with dynamic changes in gene expression.
Shuttling and the suggestion that core cell fate is not irreversibly determined (Martins et al., 2009;Senthinathan et al., 2012) could explain the versatile fate of the core cells, which contribute to myotome, sclerotome and its derivatives, ribs, and the annulus fibrosus of the intervertebral discs, as well as intervertebral joints (Huang et al., 1994(Huang et al., , 1996Mestres & Hinrichsen, 1976;Mittapalli et al., 2005;Williams, 1910). Shuttling could also explain disagreements about the origin of the annulus fibrosus which has been reported to be derived from rostral and caudal sclerotomes, as well as from the core cells (Bruggeman et al., 2012;Huang et al., 1994;Takahashi et al., 2013).
A related question is whether core cells have medio-lateral identity. cSim1 (lateral) and cSwiP (medial) markers are expressed in the core cells of newly formed somites (Pourquie et al., 1996;Vasiliauskas et al., 1999), suggesting that they do have medio-lateral identity. This medial-lateral boundary appears to be much sharper than the rostral-caudal boundary within the core. in the LMCB at UCL, and to Hyung-Chul Lee for advice on using FIJI for measuring cells.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest for this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
All original data are available from the authors on request.