The fully developed zebrafish vertebral column displays a mode of 33 vertebral bodies (Figure 1) including 4 centra in the Weberian apparatus, 10 abdominal vertebrae, 15 caudal vertebrae, and 3 to 4 caudal fin vertebrae (depending on the presence of a fourth preural) 15 .

The first centra to mineralize (Figure 2a) are 3 and 4 (4.0 mm TL) in an anterior-to-posterior direction, followed by the mineralization of centrum 5, and later centra 1 and 2, in specimens of 4.4 mm TL. The mineralization of the remaining abdominal and caudal centra follows an anterior-to-posterior direction (Figure 2b). The mineralization of the caudal fin centra starts with the U1 from the compound centrum [PU1⁺+U1], in specimens of 5.5 mm TL (Figure 2b). PU2 is the last centrum to be formed, in specimens of 6.7 mm TL, preceded by PU3 formation (Figure 2c). By then, the complex cartilaginous structure that forms the Weberian apparatus is already present in the anterior part of the vertebral column.

All Weberian, abdominal and caudal centra mineralize before development of the arches (Figure 3a). In the Weberian region, the haemal and neural arches develop as cartilaginous anlagen. These are Collagen type II positive, except for the associated spines that have no cartilaginous precursor. In the abdominal and caudal regions, the arches develop as intramembranous bones, with no cartilaginous anlage.

In contrast, in the modified caudal fin vertebrae, centrum mineralization occurs after arch formation. The (modified) arches associated to these centra have a cartilaginous anlage, also positive for Collagen type II. Staining of these arches was particularly strong at the bases (attachment areas to the notochord sheath) and distal parts, where no perichondral mineralization is present throughout the individual’s life span. Collagen type II immunostaining also showed protein accumulation in the notochord sheath, although less evident than in the cartilage (Figure 3a’).

In addition to the different timing of centrum mineralization versus arch formation, the present results also show a clear pattern concerning mineral expansion within the centra (Figure 3b, b’, c, c’).

With the exception of C1 and C2, vertebral bodies in the Weberian, abdominal and caudal regions start to mineralize in the form of a ring-shaped mineralized structure that expands in both anterior and posterior directions (Figure 3b and b’). These centra will be referred to as “ring centra”. In none of these vertebral bodies, have arches ever been observed prior to centrum mineralization. Nevertheless, the origin of mineralization matches the position of the myosepta, i.e. the site where neural and haemal arches will develop at a later stage. From this point, centrum mineralization expands one quarter anteriorly and three quarters posteriorly, showing clear incremental mineralization fronts (Figure 3b’).

Different from this pattern, in the caudal fin (PU4/3 to U2) region, mineralization starts as a broad ventral mineral deposition (basiventral origin) (Figure 3c and c’), which expands dorsally. Here, mineralization starts and expands from the attachment sites of the already developed modified cartilaginous haemal arches to the notochord. The developing centra show no apparent growth fronts as seen in the “ring centra” (Figure 3c’).

Irrespective of the shape of the mineralized centrum, (“ring” and caudal fin centra) (Figure 4a, b), the origin of mineralization was shown both with von Kossa staining (Figure 4c, c’, d, d’) and TEM (Figure 4e-f), to occur always within the notochord sheath, thus establishing the chordacentrum. Once the chordacentrum is fully formed, perichordal bone is deposited mostly at the anterior and posterior edges, in this way establishing the vertebral endplates of the centra, thus forming the autocentrum (Figure 5a-d). Autocentrum formation occurs irrespective of the initial mineralization pattern within the notochord sheath (“ring” vs. basiventral origin). Yet, in the caudal fin centra, decreased bone formation is observed in the areas where cartilaginous arches attach to the notochord (Figure 5c-d).

During chordacentrum formation, notochord inner cells or chordoblasts do not accumulate in the prospective intervertebral space (Figure 4a, b), unlike what has been observed in other teleost species. However, once the centrum is mineralized, a higher number of chordoblasts can be seen in the non-mineralized intervertebral space (Figure 5a, c, e). By then, the chordoblasts are oblong, larger in size and are oriented in a dorsal-ventral direction (Figure 5e), as opposed to the squamous morphology displayed prior to and during chordacentrum development. In addition, chordoblasts proliferate in the intervertebral space, as shown by PCNA staining. Here, notochord sheath thickening occurs (Figure 5f). This pattern of chordoblast distribution is observed throughout the notochord, regardless of the type of centrum and its mineralization pattern (Figure 5a-d).

During chordacentrum formation, contrary to the inner notochord cells, the assumed sclerotome-derived notochord outer cells 35 in the “ring” vertebrae are visible as few scattered nuclei located around the notochord (Figure 4e). These cells are squamous and without apparent morphological features of active osteoblasts. In the caudal fin region, the cartilaginous haemal arches of the PUs (parhypural and hypurals 1, 2 and 4) directly attach to the notochord sheath with no space for other cells in-between the notochord sheath and the chondrocytes of the arches (Figure 4f).

During autocentrum formation, osteoblasts can be clearly identified in association with bone deposited around the “ring” chordacentra (Figure 5a). Major areas of the caudal fin chordacentra are covered by the cartilaginous arches attached to the notochord sheath surface (Figure 5c). Yet some osteoblasts can be identified wherever the arches are absent, lining the edges of the prospective vertebral endplates.

In order to clarify possible factors in chordacentrum mineralization, both in “ring” and caudal fin centrum, we have localized Alkaline phosphatase (ALP) and Osteocalcin (Figure 6).

In “ring” centra, ALP-positive signal is co-localized with the site of chordacentrum mineralization prior to any bone deposition outside the notochord sheath. The flattened shape of both outer and inner notochord cells makes it difficult to clearly distinguish the origin of the ALP signal, due to its membrane-bound characteristic (Figure 6a). In caudal fin centra, the arch chondrocytes, particularly those adjacent to the mineralizing notochord sheath, co-localize with a strong ALP signal (Figure 6b).

For Osteocalcin immunostaining, we used an antibody raised against Argyrosomus regius Oc1, which was previously validated 36 . It was considered to detect a single protein in zebrafish, since a western blot analysis gave a single band of approximately 5kDa. Nevertheless we cannot guarantee that it distinguishes zebrafish Oc1 from Oc2. However, since Oc2 appears only after day 7 post-fertilization in zebrafish as shown by qPCR analysis (Figure 6l, m), immunostaining in younger fish, such as during formation of chordacentra C1 to C5 (Figure 2a) could only represent sites of Oc1 protein accumulation.

In “ring” (Figure 6c) and caudal fin centra (Figure 6d), Oc1 protein accumulation is observed in the mineralizing matrix of the notochord sheath, even at very early stages of chordacentrum formation. Oc1 accumulation co-localizes with chordacentrum mineralization, reflecting the exact same pattern as observed with alizarin red staining (Figure 3b, b’, c, c’). Moreover, in later stages of centrum formation, we can detect progression fronts in “ring” centra (Figure 6e) while in caudal fin centra we observe homogenous protein accumulation (Figure 6f). In the caudal fin region, Oc1 is also detected in the vertebral arches and caudal fin rays (Figure 6d, f), which are examples of perichondral and intramembranous mineralization, respectively.

Analysis of semi-thin sections of the immunostained samples confirms accumulation of Osteocalcin in the notochord sheath and not around the notochord, during early (Figure 6g, g’, h, h’) and late (Figure 6i, j) stages of chordacentrum formation. In caudal fin centra (Figure 6h, j), the arch chondrocytes adjacent to the mineralizing notochord sheath also co-localize with Oc1 protein (Figure 6h’), as observed for ALP. No perichondral ossification occurred in the arch at sites positive for ALP and Oc1, suggesting that these signals are not associated to arch ossification.

Observation of the transgenic fish Tg(oc2:gfp; osx:mcherry) showed that no oc2 expression is present during early stages of centrum formation. In fact, oc2 is completely absent during chordacentrum formation, appearing first in intramembranous formation of abdominal and caudal arches, a process previously described as mediated by mature osteoblasts (Figure 6k, k’, k”). Analysis of relative oc1 and oc2 expression levels during zebrafish development shows that, while oc1 maternal transcripts are detected in early developmental stages, when no mature osteoblasts can yet be detected (Figure 6l), oc2 is only expressed after 7dpf (Figure 6m). Chordacentrum formation of the first vertebrae starts at 4.0 mm TL, as previously indicated, which, despite the natural variation in growth, corresponds to fish younger than 7 days.

The first mineralization of the vertebral bodies, independent of the vertebral type (Weberian, abdominal, caudal and caudal fin), occurs within the notochord sheath, establishing the chordacentrum. These are the first indications of segmented units in the zebrafish notochord. Other authors have already described that early centrum mineralization in teleosts occurs within the notochord sheath 2 12 13 19 38 39 . In addition, the notochord may actively contribute to segmentation and centrum formation 2 12 20 .

Thus, while the chordacentrum is always established through mineralization of the notochord sheath, the pattern of mineralization differs between caudal fin and remaining vertebral regions: (1) a caudal fin centrum mineralizes after arch formation, while a “ring” centrum is always formed prior to haemal or neural arches, (2) the modified arches from caudal fin vertebrae are composed of cartilage while “ring” centra outside the Weberian apparatus are associated to arches formed by intramembranous ossification; and (3) caudal fin centra display a basiventral origin of mineralization, while all other vertebrae start with a “ring”-shaped mineralization.

Arch histogenesis appears to be a clear mark that distinguishes caudal fin vertebrae from the remaining regions. A similar, regional distinct pattern of mineralization has been observed in medaka, with cartilaginous arches in the caudal fin region and intramembranous arches in the anterior part of the vertebral column 19 40 . Although notochord mineralization of anterior centra in medaka starts dorsal, subsequently chordacentrum mineralization will proceed as “ring” centra 19 . Yet, “ring” notochord-based centrum formation, is not an unique feature of zebrafish and medaka, occurring also in other taxa 41 . Considering caudal fin centra of zebrafish and medaka, both display similar patterns of mineralization 40 , linked to the presence of modified cartilaginous arches. Many basal ray-finned fishes have cartilaginous arches directly attached to an unconstricted notochord 18 suggesting that cartilaginous arches are an ancestral feature. Therefore, the caudal fin endoskeleton of zebrafish, and also medaka, appear to display the evolutionarily conserved condition, with cartilaginous arch formation.

Mineralization in the caudal fin centra shows a clear association with attachment sites of the cartilaginous haemal arches to the notochord (basiventral mineralization pattern). The association between basiventral mineralization and presence of cartilaginous haemal arches has also been described in other teleost species, such as the Goldeye (Hiodon alosoides) 42 , Atlantic salmon 2 and White seabream (Diplodus sargus) 43 . In these species too, centrum mineralization originates exactly where the cartilaginous haemal arches are attached to the notochord. Mineralization then progresses from a basiventral origin through bilateral wedge-shaped areas that meet dorsally.

Moreover, we have identified mineralization related proteins in the chondrocytes of the cartilaginous arches, in the immediate vicinity of the area of the incipient mineralization of the notochord. This supports the possible role of the cartilaginous arches in early centrum establishment, a hypothesis that will be discussed in more detail in the next section.

The cartilaginous arches appear to be strictly associated not only with chordacentrum formation but also with the start of autocentrum formation, given that the arches are directly attached to the notochord, affecting bone deposition around the notochord.

The type of arches and associated pattern of centrum mineralization appears to be correlated not only to the formation of the vertebral bodies but also to their ability to fuse or stay as individual units (Figure 7). In the medaka bis (biaxial symmetries) mutant, centra mineralize at sites where ectopic hypurals attach to the notochord, leading to the development of fused centra 40 . Thus, it is tempting to suggest from the zebrafish and medaka data, that the cartilaginous arches have a role in centrum mineralization. Perhaps the predisposition to develop vertebral fusions is related to a specific mineralization pattern with basiventral origin (cartilaginous arch associated).

In Atlantic salmon, notochord sheath mineralization is preceded by an apparent segmentation of the chordoblast layer 2 . The same holds for medaka (PEW personal observations). In contrast, differential chordoblast distribution in zebrafish is only clearly observed after chordacentrum mineralization, with few scattered squamous cells in the mineralized area and large, dorso-ventrally oriented, highly proliferative cells in the intervertebral region.

Regarding notochord outer cells, it appears that the segmental pattern of centra in the caudal fin region is associated with the anatomical pattern of arches. In contrast, no association can be found in “ring” centra between outer cell distribution and early chordacentrum formation. The latter is in agreement with the data from zebrafish fused somite mutants (fss), where paraxial mesoderm lacks a proper segmentation 44 . This mutant shows fusion of vertebral arches, while vertebral centra are individualized, suggesting that centrum segmentation does not rely on sclerotome patterning only. The resegmentation process in zebrafish has been suggested to follow non-lineage restricted compartments, where one somite contributes to more than one vertebral body, independently of the antero-posterior somitic domains (leaky-resegmentation) 1 .

Although mineralization is not preceded by a clear pattern of inner or outer cell distribution (exceptions are arch-chondrocytes in the caudal fin), these cells may nevertheless contribute to centrum mineralization. In Atlantic salmon, initial centrum mineralization has been related to chordoblasts, followed by sclerotome-derived bone formation 2 10 . Different from Atlantic salmon, in medaka sclerotome-derived osteoblasts, with no input of chordoblasts, have been described to form vertebral centra 19 35 . Yet, recent data show that conditional ablation of osteoblasts in medaka maintains notochord mineralization and even leads to vertebral fusion 20 . This suggested that notochord cells have the ability to induce mineralization of the notochord sheath. This is consistent with studies in zebrafish 12 45 that show that notochord cell ablation prevents early centrum mineralization.

Osteocalcin is the most abundant noncollagenous protein in bone of vertebrates, including all teleosts analyzed so far 46 . Although Osteocalcin was already described in the vertebral bodies of teleosts including zebrafish 47 48 49 50 , no previous reports demonstrated Oc in early notochordal mineralization events. Thus, the presence of Oc in stages of chordacentrum mineralization has to be considered in a perspective different from that of regular bone formation.

Within teleosts, Osteocalcin is represented by two distinct isoforms (Oc1 and Oc2) 25 51 , encoded by two different genes, which were proposed to originate from a duplication event 25 . Analysis of relative gene expression levels by qPCR for oc1 and oc2 showed that oc2 expression is first detected after 7dpf, while oc1 is maternally transcribed, suggesting a different role of the two isoforms. In fact, oc1 is detected prior to any osteoblast differentiation or mineralization event. This leads to the hypothesis that oc1 must be involved in processes prior to bone formation. Detection of Oc1 by immunostaining shows that this protein accumulates exactly where mineralization of the chordacentrum occurs. In the “ring” centra, the origin of the secreted protein is difficult to assess, due to the squamous morphology of the cells of the notochord epithelium and the inconspicuous presence of cells outside the notochord. In the caudal fin region, ALP positive chondrocytes, directly adjoining the mineralizing notochord sheath, are positive for Oc immunostaining, as well as notochord inner cells. In the Tg(oc2:gfp; osx:mcherry) fish, oc2 expression was located at the bone of arches of the “ring” centra, that form via intramembranous bone formation. This time point corresponds to the phase of autocentrum mineralization, as previously described in medaka 52 . Therefore, our results suggest that Oc1, but not Oc2, is connected to chordacentrum mineralization. Likewise, in larval and juvenile Atlantic cod 53 , oc1 and oc2 are expressed in different cells associated with mineralizing structures. In Atlantic salmon, some mineralized areas with reduced expression of osteocalcin (oc1) were considered to be associated with expression of a second isoform that the authors did not identify 49 . Laizé and co-workers 25 described Oc2 propeptide as serine-rich, potentially phosphorylated and containing acid residues. These characteristics allow this protein to bind numerous calcium ions. If Osteocalcin has an active role in the process of mineralization or if it is a bone-related hormone that passively binds to calcium 54 55 , is a current debate and must be further explored.

Animal handling and experiments were legally accredited by the Portuguese Direcção Geral de Veterinária (DGV) and all the experimental procedures involving animals followed the EU (Directive 86/609/CEE) and National (Directives 1005/92 from October 23, 466/95 from May 17 and 1 1131/97 from November 7) legislation for animal experimentation and welfare.

To establish a timeline of events for early vertebrae development, an ontogenic series of Danio rerio was studied. Zebrafish eggs were obtained from natural spawning of wild-type fish and larvae were maintained and reared at standard conditions 56 . Fish were collected at size intervals of 0.1 mm, between 4.0 and 8.0 mm of total length (TL). Similar-sized specimens were always selected among age-matched fish in order to avoid additional interspecific variation.

Larvae were euthanized with a lethal dose of MS222 (Sigma-Aldrich, St. Louis, MO) and fixed for 24 h in 4% buffered paraformaldehyde at 4ºC.

Early mineralization stages of the vertebral bodies were observed following 10 minutes staining with 0.01% alizarin red S in 70% ethanol solution. Specimens were observed under a Zeiss Axio Imager microscope, through fluorescence detection 57 .

Zebrafish specimens of 5.0 and 11.0 mm TL with no pre-staining were fixed in a mixture of 1.5% glutaraldehyde and 1.5% paraformaldehyde in 0.1 M cacodylate buffer and processed for embedding in Epon, according to procedures previously described 58 . Parasagittal 1 μm semi-thin sections of the sites of interest were stained with toluidine blue for 1–2 min (0.2% toluidine blue, 2% Na₂CO₃), rinsed with water, air-dried and mounted with DPX (Fluka, Buchs, Switzerland). Ultrathin sections were contrasted with uranyl acetate and lead citrate and observed with a Jeol JEM 1010 (Jeol, Tokyo, Japan), operating at 60 kV. Images were digitized using a DITABIS system (Pforzheim, Germany).

To determine sites of mineral deposition and of Alkaline phosphatase (ALP) activity during centrum formation, specimens ranging from 5.0 to 6.5 mm TL were dehydrated and embedded in glycol methacrylate, as previously described 59 , and sectioned at 5 μm.

For detection of minerals, a silver nitrate coupling method, according to von Kossa 60 , was used. Sections were counterstained with neutral red. Slides were examined for the appearance of optimal signal, rinsed in distilled water and mounted with DPX (Sigma-Aldrich, St. Louis, MO).

Alkaline phosphatase activity was detected using a previously described method 61 with some modification. Briefly, slides were incubated in staining buffer (100 mM Tris/HCl, pH 9.5; 50mM MgCl₂; 100 mM NaCl) for 10 min. ALP was demonstrated by adding 4-nitro blue tetrazolium chloride (NBT; 0.38 mg ml^-1; Roche) and 5-bromo-4-chloro-3-indolyl phosphate (BCIP; 0.175 mg ml^-1; Roche) to the buffer solution, for a maximum of 30 min at 37ºC, in the dark. Negative control slides were incubated for 10 min at 85ºC in staining buffer prior to addition of NBT and BCIP. Slides were examined for the appearance of optimal signal-noise ratio, counterstained with Nuclear Fast Red Solution (Sigma-Aldrich, St. Louis, MO) and mounted in DPX (Sigma-Aldrich, St. Louis, MO).

For detection of cell proliferation through Proliferating cell nuclear antigen (PCNA) immunolocalization, 6.2 mm TL fish were paraffin-embedded according to routine procedures and serially sectioned at 7 μm. A mouse monoclonal primary antibody produced against human antigen (Sigma-Aldrich, St. Louis, MO, dilution 1:1000) was used 62 together with a goat polyclonal anti-mouse immunoglobulin biotin-coupled secondary antibody (Dako, dilution 1:500).

Immunohistochemical detection of Osteocalcin and Collagen type II was performed on fish of 4 to 6.5 mm TL as wholemount staining, according to a procedure previously described 63 , using an anti-meagre (Argyrosomus regius) Osteocalcin rabbit polyclonal primary antibody, previously validated for zebrafish 36 . For Collagen type II an anti-chicken mouse antibody (II-II6B3, Hybridoma Bank) previously validated for zebrafish 64 , was used. Control fish were treated with secondary antibody alone.

Osteocalcin stained specimens were subsequently dehydrated, embedded in epon, serially sectioned at 4 μm and mounted with DPX (Sigma-Aldrich, St. Louis, MO), to achieve a more detailed analysis. Wholemount fish stained for Collagen type II were observed using a Leica MZ Apo stereomicroscope. Sections were observed using a Zeiss Axio Imager Microscope and photographed using an Axiocam MRC videocamera.

The transgenic zebrafish line Tg(oc2:gfp; osx:mcherry), previously described by 65 , was provided by Stefan Schulte-Merker from Hubrecht Institute-KNAW and University Medical Centre (Uppsalalaan 8, 3584 CT Utrecht, The Netherlands).

Total RNA was extracted from a pool of up to twenty specimens following the Chomczynski and Sacchi method 66 . Samples were collected at stages 8-32 or 128 cells, at shield stage, at 12, 24, 48, 72 and 96 hours post-fertilization (hpf) and at 5, 7, 15, 30, 45 and 60 days post-fertilization (dpf). To prevent genomic DNA contamination, following RNA extraction, samples were DNase treated with RQ1 DNase (Promega, Madison, WI) according to the manufacturer’s protocol, followed by a phenol-chlorophorm purification step. RNA quantity and quality was determined by spectrophotometry (NanoDrop ND-1000, Thermo Scientific) and electrophoresis.

cDNA was prepared from 1μg of total RNA from each sample using Moloney-murine leukemia virus (M-MLV) reverse transcriptase (Invitrogen), RNase Out (Invitrogen), and an oligo(dT) adapter (5'-ACGCGTCGACCTCGAGATCGATG(T)13-3') in a total volume of 20μl for 1h at 37ºC. For quantitative real-time PCR a StepOnePlus 96 (Applied Biosystems) was used together with 1× SsoFastTM EvaGreen Supermix (BioRad, Richmond, USA), 0.2 μM of forward (Oc1: 5^′-GAAGCGAACATGAAGAGTCTGACAGTCC-3^′; Oc2: 5^′-CCAACTCCGCATCAGACTCCGCATCA-3^′) and reverse (Oc1- 5^′-TTTATAGGCGGCGATGATTCC-3^′; Oc2: 5^′-AGCAACACTCCGCTTCAGCAGCACAT-3^′) primers and 200 ng of reverse-transcribed RNA. PCR conditions were 15 min at 95°C followed by 40–50 amplification cycles (each cycle is 30 s at 95°C, 15 s at 68°C). Expression values were normalized using the Elongation factor 1α (Ef1α) housekeeping gene 67 . Gel electrophoresis and melt curve analysis were used to confirm specific product formation.

All data were statistically analyzed using a non-parametric Kruskal-Wallis test, followed by Wilcoxon multiple comparison with the GraphPad Prism software.

The subdivision of the vertebral column in different regions is based on the nomenclatures of Arratia et al. 16 , Bird and Mabee 31 and Nybelin 68 (Figure 1). From anterior to posterior, vertebrae were identified as (1) Weberian vertebrae, (2) abdominal vertebrae (rib-bearing with open haemal arches and without haemal spines), (3) caudal vertebrae (with haemal arches closed), and (4) caudal fin vertebrae (with modified haemal and neural arches and respective spines). The nomenclature of the Weberian apparatus follows 31 . The caudal fin vertebrae nomenclature follows Patterson 69 , adapted by Arratia and Schultze 70 and Bensimon-Brito et al. 32 . The caudal fin vertebral region contains preural and ural vertebral bodies. Preural vertebral bodies (abbreviated as PU) have haemal and neural arches that support the caudal fin. Ural vertebral bodies (briefly urals, abbreviated as U) support the caudal fin with modified haemal arches that do not enclose the caudal artery. A plus sign is added in superscript when a vertebral centrum is derived from a fusion event (compound centrum), as described previously 32 .