The emergence of a complex pore-canal system in the dermal skeleton of Tremataspis (Osteostraci)

Thyestiids are a group of osteostracans (sister-group to jawed vertebrates) ranging in time from the early Silurian to Middle Devonian. Tremataspis is unique among thyestiids in having a continuous mesodentine and enameloid cover on its dermal elements, and an embedded pore-canal system divided into lower and upper parts by a perforated septum. The origin of this upper mesh canal system and its potential homology to similar canal systems of other osteostracans has remained a matter of debate. To investigate this, we use synchrotron radiation microtomography data of four species of Tremataspis and three other thyestiid genera. Procephalaspis oeselensis lacks an upper mesh canal system entirely, but Aestiaspis viitaensis has partially enclosed upper canals formed between slightly modified tubercles that generally only cover separate pore fields. Further modification of tubercles in Dartmuthia gemmifera forms a more extensive, semi-enclosed upper mesh canal system that overlies an extensive perforated septum, similar to that found in Tremataspis . Lower mesh canals in P. oeselensis are radially arranged and buried tubercles indicate a continuous growth and addition of dermal hard tissues. These features are lacking to varying degrees in the other investigated thyestiids, and Tremataspis probably had a determinate growth accompanied by a single mineralization phase of its dermal hard tissues. The previously proposed homology between the semi-enclosed upper canal system in Dartmuthia to the pore-canal system in Tremataspis is supported in this study, but the suggested homologies between these canals and other parts of the thyestiid vasculature to those in non-thyestiid osteostracans remain unclear. This study shows that three-dimensional modeling of high-resolution data can provide histological and structural details that can help clarify homology issues and elucidate the evolution of dermal hard tissues in osteostracans. In extension, this can give insights into how these tissues relate to those found among jawed vertebrates.

ever, the correlation between the architecture of these superficial canals to the underlying dermal skeleton was not always evident according to Wängsjö (1952). The pore-canal system in Tremataspis is not only superficially similar to the convergently acquired, hard tissue-embedded pore-canal system of cephalaspid osteostracans (e.g., Stensiö, 1927), but also to the cosmine of extinct osteichthyans (e.g., MondéJar-Fern andez, 2018;Qu et al., 2017;Schultze, 2016). Similarly, the true functional role of the pore-canal system in Tremataspis has remained unclear, with suggestions ranging from mucous canals (Stensiö, 1932) or extensions of the lateral lines (Bölau, 1951;Denison, 1947Denison, , 1966 to a system for electroreceptors (Thomson, 1977). King et al. (2018) reviewed and discussed these previous suggestions and refuted an electroreceptive role for the porecanal system, leaning more toward a mechanoreceptive function similar to lateral lines. This view, however, was not shared by O'Shea et al. (2019), who thoroughly investigated the three-dimensional (3D) histology of a single specimen of Tremataspis mammillata and discussed its potential development and function.
In this study, we use propagation phase contrast synchrotron X-ray micro computed tomography (PPC-SRμCT) to analyze the 3D histology of the dermal skeleton of four different species of Tremataspis and the closely related D. gemmifera, as well as the other thyestiids Procephalaspis oeselensis and Aestiaspis viitaensis. Interpreting this in the phylogenetic context established for thyestiids (Sansom, 2008(Sansom, , 2009) enables us to infer the emergence of the complex pore-canal system in Tremataspis through the modification of the structures already present in other taxa.

| MATERIALS AND METHODS
The material for this study comes from the mid-Silurian of Saaremaa Island, Estonia, from the collection sampled by Tiiu Märss between 1970 and 2010. The P. oeselensis head shield fragment (GIT 769-17) comes from Silma pank locality, the A. viitaensis shield fragment  comes from Elda cliff (Bed 5), the D. gemmifera head shield fragment  and scale , as well as the T. mammillata  head shield fragment come from the Himmiste Quarry, the Tremataspis milleri (GIT 769-1) shield fragment comes from Vesiku creek, and the Tremataspis schmidti (GIT 769-11) shield fragment and the Tremataspis perforata scales (GIT 769-15 and 769-16) come from the Viita trench locality. The rock samples were subjected to acetic acid preparation and numerous microfossils were sorted from the residues. The head shield fragments and scales were identified following the taxonomic framework established by Märss et al. (2014). The specimens are housed at the Department of Geology, Tallinn University of Technology, Estonia.
All specimens were imaged using PPC-SRμCT at the ID19 beamline of the European Synchrotron Radiation Facility (Grenoble, France). The experimental setup consisted of: filtered white beam from a wiggler (W150B gap = 60 mm; filters: Al 2.8 mm, W 0.14 mm) which generated a spectrum with a total integrated detected energy of 69 keV; an indirect detector comprising a LuAG 0.25 μm scintillator, mirror reflecting the image upward, a ×10 Mitutoyo lens (Mitutoyo corporation, Kawasaki, Japan) and a PCO. In some cases, to increase the horizontal field of view (hFOV), the center of rotation was shifted by 0.65 mm, effectively doubling the hFOV. The sample detector distance was set to 100 mm, resulting in a measured isotropic pixel size on radiograph of 0.716 ± 0.01 mm. Additionally, two white field images (with no sample in the field of view), representing the intensity of the beam at the beginning and end of the acquisition, were generated from the median of 41 projections; a dark field image (no incoming X-ray, measuring only the noise of the camera) was generated from the average of 40 projections.
Computed tomographic reconstruction was performed using the single distance phase retrieval approach (Paganin et al., 2002) of PyHST2 (Mirone et al., 2014). Parameters for phase retrieval were intentionally set with low values (δ/β = 50), effectively using the approach as a denoising low-pass filter. From the 32-bit data generated by PyHST2, postprocessing included: data conversion as 16-bit jpeg2000 stack with a compression ratio of 10, using the full 3D histogram and excluding 0.001% on both extremities; ring correction (Lyckegaard et al., 2011).
The reconstructed data were imported as jpeg2000 stacks into the software VGStudio Max 3.4 (Volume Graphics GmbH) installed on a Dell Computer Station Precision T7600 running Windows 7. The 3D-models presented in this work were created from regions of interest using selection tools in VGStudio Max 3.4. Virtual scan slices and images of the 3Dmodels were rendered in VGStudio Max 3.4 and exported using the export tool. These were then assembled in Adobe Photoshop CC 2015 and Adobe Illustrator CC 2015 (Adobe Inc. San Jose, CA) to create the figures.

| RESULTS
The general structure of the dermal skeleton is clearly visible in virtual scan slices of the elements (Figure 1), and its three layers are labeled in The basal layer is usually distinct because of its difference in architecture, but the boundary to the middle layer can sometimes be uneven (Figure 1 (a)). The middle layer varies in thickness between taxa and has dispersed cell spaces and lines around its canals (e.g., Figure 1(c)), that have often been identified as osteon-like structures. The transition between the middle and superficial layers is often hard to distinguish given the similar structure of mesodentine and cellular bone. However, the more dentinelike tissue, and especially the hypermineralized cap, is generally only present in the apex of large tubercles (e.g., Figure 1(c)), but covers the whole surface in Tremataspis (Figure 1(d)).
The terminology in this study for the canal systems within the three layers described above will follow the one adapted by Qu et al. (2015) from earlier works. These are basal cavities and canals in the laminated basal layer (Figure 1: yellow), mesh canals (Figure 1: pink) in the middle layer (divided into lower and upper [ Figure 1: teal] parts in Tremataspis), and the subepidermal vascular plexus (also colored pink in the figures) that arise from the mesh canals and supply the dentine of the superficial layer. Note that the partially enclosed canals above the pore fields in A. viitaensis and D. gemmifera (called partially or semi-enclosed upper canals in this study) are also highlighted in teal (Figure 1(b),(c)). The differently sized polygons of upper mesh canals in the Tremataspis taxa have historically been referred to as inter-and intra-areal canals after being homologized with the canals found in cephalaspids (Denison, 1951;Stensiö, 1927Stensiö, , 1932. However, they will here be referred to as polygonal and intra-polygonal canals to avoid assumptions of homology.

| Procephalaspis oeselensis
One of the most striking features of the head shield fragment of P. oeselensis are the buried surfaces (Figure 1(a), arrowhead). Threedimensional models of the fragment (Figure 2) show that these buried F I G U R E 1 Virtual scan slices and inserts of their respective positions on the dermal elements. The different canal systems are highlighted in yellow (basal cavities and canals), pink (lower mesh canals), and teal (upper canals/upper mesh canals). (a), Procephalaspis oeselensis head shield fragment (GIT 769-17); (b), Aestiaspis viitaensis head shield fragment (GIT 769-29); (c), Dartmuthia gemmifera scale (GIT 769-24); (d), Tremataspis milleri head shield fragment (GIT 769-1). The basal, middle, and superficial layers are indicated in (c). Arrowheads in (a) and (b) indicate overgrowth of tubercles, and arrows in (b)-(d) point to the pore field and perforated septum surfaces form tubercles (Figure 2(a),(d)) that are either partly or completely covered by younger generations of tubercles. Therefore, some of the smaller tubercles visible at the surface are actually the exposed tips of older generations (Figure 2(a)).
The cavities of the basal layer in the P. oeselensis head shield fragment are relatively small and scattered throughout the basal layer ( Figure 2(b)), and narrower canals connect the cavities occasionally (arrow in Figure 2(b)). Similar to all of the other investigated taxa, they only open to the canals of the middle layer in a few places.
The middle layer has a deep set of mesh canals just above the basal layer, which also includes the cavities of the oldest and often buried tubercles (Figures 1(a) and 2(c)). The deepest canals acquire a radial appearance where they connect three separate areas composing the head shield fragment (Figure 2(c)). The separate cavities of each tubercle in P. oeselensis also have canals that radiate away from them (Figure 2(c)), which was also pointed out by Denison (1951). The cavities of the younger generations of tubercles in P. oeselensis and their associated horizontal canal network lie above the deep canal network, but connect to it extensively. It is noteworthy that there are no pore fields present where the mesh canals open to the surface, or at the former openings that surround buried tubercles in P. oeselensis (Figure 1(a)).

| Aestiaspis viitaensis
The head shield of A. viitaensis (Figure 3(a)) shows evidence of superpositional growth, as the tubercles sometimes overlie older generations, essentially fusing with them on the surface and effectively forming a single large tubercle (arrowhead in Figure 1(b)).
The basal cavities of A. viitaensis are irregular and only occupy a narrow band of the head shield fragment anteriorly (Figure 3 (Afanassieva & Märss, 1997), but these are only sometimes preserved in the material investigated here (arrow in Figure 1(b)). Most of the openings are overarched by modified tubercles with a T-like cross-section (left in Figure 1(b)), that effectively produce partially enclosed upper canals. These partially enclosed canals form paired streaks between tubercles (teal in Figure 3(d)) that appear to be relatively independent from the mesh canal system of the middle layer, as are the tubercles at large, and connect to the mesh canal system in different places.

| Dartmuthia gemmifera
The head shield fragment of D. gemmifera investigated here ( Figure 4) probably comes from a more posterior and lateral part of the head F I G U R E 2 Three-dimensional models of the Procephalaspis oeselensis head shield fragment . Hard tissues (brown), basal cavities and canals (yellow), and lower mesh canals (pink). Contacts between basal canals and lower mesh canals are indicated with green, and openings of mesh canals to the surface are colored white. Buried tubercles have been colored gray in (a) and (d). Arrow in (b) points to a narrow canal connecting basal cavities. The element is oriented with its anterior pointing toward lower left shield judging by the outstretched and slightly curved tubercle (compare to Märss et al., 2014, fig. 11I). No natural borders are preserved and part of it was cropped in the scanning procedure (Figure 4(a)).
Unfortunately, the areas creating the semi-enclosed upper canals are poorly preserved in this fragment ( Figure S1A), so a scale of D. gemmifera ( Figure 5) is also included in this study in order to investigate this part more closely. The basal layer of scales generally differed from the head shield in the thyestiid T. mammillata (O'Shea et al., 2019). There were also some other topological differences between them, but the general histology of the middle and upper F I G U R E 3 Three-dimensional models of the Aestiaspis viitaensis head shield fragment .
The colors indicate the same as in previous figures with the addition of partially/semi-enclosed upper canals (teal) and highlighted pore fields/perforated septa (white). Arrow in (b) points to a narrow canal connecting basal cavities. The element is oriented with its anterior pointing toward lower left F I G U R E 4 Three-dimensional models of the Dartmuthia gemmifera head shield fragment . Colors indicate the same as in previous figures. The element is oriented with its anterior pointing toward lower left layer was similar as they produced similar structures, which is evident in the material studied here as well.
The basal layer of the D. gemmifera head shield fragment is relatively thin and consists of large fiber bundles ( Figure S1A), and only a few scattered and pillar-like canals are present (Figure 4(b)). The largest, vertical canal running through the basal layer and opening to the basal surface has, unlike all other taxa, conspicuously smooth walls (Figure S1A and Figure 4(b)). It also has a thin, inner lining of unclear nature, which is most likely not a primary structure, but the walls of the basal layer itself are smooth as well. The scale of D. gemmifera has a few isolated, pillar-like canals in its basal layer, as well as some larger cavities, that are interconnected by relatively large and almost horizontal canals (Figures 1(c) and 5(b)).
The head shield fragment has two main and large mesh canals that run deeper than the remaining mesh canal system (Figure 4(c) and Figure  and Figure S2A). The main canals intersect underneath the largest tubercle, where they also connect to the conspicuously smooth basal canal described above (Figure 4(c) and Figure S1A), but there is no large cavity formed in the middle layer under the largest tubercles.
There is another antero-posteriorly running mesh canal that connects to the main perpendicular canal (Figure 4(c)), but it is not as large and positioned vertically higher. The remainder of the mesh canal system is irregularly branching and sits higher up in the middle layer. However, slightly larger canals appear to converge at the center of each section created between the two main mesh canals, where they also connect to basal canals (Figure 4(b),(c) and Figure S2A). There are a few connections between the mesh canal system and the basal canals in other parts of the head shield fragment as well. Numerous thinner canals rise from the mesh canal system to supply the subepidermal vascular plexus of both the larger tubercles, as well as the tessera-like tubercles in-between that reach over the perforated septa (Figure 4(d)). Other mesh canals ascend from the deeper mesh canal network and reach toward the contact with the upper canal system (Figure 4(c)). The areas underneath the tessera-like tubercles are poorly preserved in the head shield fragment ( Figure S1A), so it is difficult to see the extent of the perforated septa and the canals underlying it. The deepest mesh canals do not have a radial layout, but some of the ascending mesh canals appear radial as they reach away from the large tubercles (on the right in Figure 4(d)).
The largest mesh canal of the D. gemmifera scale is deep and irregularly branching ( Figure 5(c)). Besides this, the mesh canal system of the middle layer branch seemingly independently from the surface structures, and even the largest tubercles lack a large cavity or any trace of one ( Figure 5(a),(c)). However, mesh canals ascend into both the large tubercles and the tessera-like plates that cover the bony septum. Other ascending mesh canals reach toward the perforated septum and expand horizontally to underlie the entirety of it, sometimes forming continuous meandering canals underneath ( Figure 5(c)).
Because these meandering canals often are connected to the canal system underlying the large tubercles, the canals that run away from the tubercles obtain a radiating appearance in the scale as well (bottom right in Figure 5(c)). The circular canals that were described in D. gemmifera by Afanassieva (2004) have not been identified in the head shield fragment or the scale.

| Tremataspis species
The Tremataspis material investigated here resembles that described  Figure 7(b)). The head shield of T. schmidti lacks a basal layer entirely ( Figure S1C), which is possibly related to its growth stage, as demonstrated in T. mammillata by Denison (1947, fig. 11). Tremataspis perforata (Figure 9 and Figure S1D) is only represented by scales in this study, which explains some of the observed histological (mainly in the basal layer) and topological differences. This Following the discovery of a perforated septum separating the mesh canals in three species of Tremataspis (Bölau, 1951;Denison, 1947), subsequent works have divided this canal system into lower and upper parts (e.g., Qu et al., 2015), with the septum itself being composed of bone (O'Shea et al., 2019). Because the upper mesh canal system is viewed as separate from the lower mesh canal system, it will be described separately below. The lower mesh canal system of Tremataspis has a polygon-like organization where it underlies the bony septum below the upper mesh canal system. This is best seen in T. mammillata, where lower mesh canals underlie the entirety F I G U R E 6 Three-dimensional models of the Tremataspis mammillata head shield fragment . Colors indicate the same as in previous figures, but teal represents the upper mesh canal system of the upper mesh canal system (Figure 6(c)). Within the polygons of T. mammillata, the lower mesh canals form loops that connect to the surrounding lower mesh canals (Figure 6(c) and Figure S2B). In some places, these connections continue between adjacent loops and form an irregularly branching network ( Figure S2B). The loops also send narrower canals toward the surface that form the subepidermal vascular plexus that supply the dentine of the continuous superficial layer (center of Figure S1B). The narrow canals that arise from the lower mesh canal system reach above the upper mesh canals toward the tips of similar canals from neighboring polygons, but they do not appear to connect (Figure 6(c)). Furthermore, each loop has a connection to the basal cavities, although they may sometimes be small, or only open into minuscule and blind basal canals (center of Figure 6(b)).
The polygonal architecture of the lower mesh canal system is interrupted under the large tubercle-like surface structure (top right in Figure 6(c)). Instead, three loop-like sections with separate connections to basal cavities appear to have merged under this surface structure (lower right in Figure S2B).
The majority of the lower mesh canal system in T. milleri also forms a polygonal network that reflects the architecture of the upper mesh canal system with a bony septum between them (Figure 7(c)), but there are also independent and tree-like lower mesh canals (- Figure S2C) similar to T. mammillata. Unlike T. mammillata, however, there are additional, intra-polygonal upper mesh canals within the larger polygons in T. milleri (Figure 7(c)) at a higher level compared to the largest upper mesh canals (Figure 1(d)). Only a few of these intrapolygonal upper mesh canals have an equivalent in the lower mesh canal system and a bony septum separating them (arrowhead in Figure 7(c)), but most of them do not (see below). Another big difference compared to T. mammillata is that the canals of the lower mesh canal system within the polygons, which rise to supply the subepidermal vascular plexus, actually connect to mesh canals of adjacent polygons above the upper mesh canals (Figures 1(d) and 7(c),(d)). This was also reported by Denison (1947) in both T. milleri and T. schmidti.
As mentioned before, the T. schmidti head shield fragment studied here ( Figure 8) has not been fully ossified because it is in a younger growth stage (compare Figure S1C to Denison, 1947, fig. 11B). Therefore, much of the lower mesh canal architecture remains obscure ( Figure 8(b)). However, the majority of it must have attained a polygonal pattern because it underlies most of the upper mesh canal system ( Figure S3), as it does in T. mammillata, and which was evident in the scale of T. schmidti investigated by Qu et al. (2015, fig . 2F).
The T. perforata scale has, similar to the D. gemmifera scale, a deeper and independently branching system of lower mesh canals (Figure 9(c) and Figures S1D and S2D), although they are more numerous. Above this, the lower mesh canals underlying the perforated septum form an independent and more polygonally arranged canal system (Figure 9(d)) similar to that found in the head shields of the F I G U R E 7 Three-dimensional models of the Tremataspis milleri head shield fragment (GIT 769-1). Colors indicate the same as in previous figures. The arrow in (c) points to a perforated septum within a polygon and the arrowhead in (d) points to a forked ascending canal of the upper mesh canal system other Tremataspis taxa. The polygonal and the deeper lower mesh canal systems are connected in some places through vertical canals, while other lower mesh canals ascend from the deeper lower mesh canal system and reach up into the polygons to supply the subepidermal vascular plexus (Figure 9(e)). In some places there are also horizontal canals connecting these mesh canals to the polygonally arranged lower mesh canals surrounding them.
In Tremataspis species, the upper mesh canal system is entirely enclosed within hard tissues, except for more or less regularly spaced pores that open to the surface. Subsequently, the entire surface in many Tremataspis taxa is covered by a continuous mesodentine layer with a hypermineralized cap, which is only pierced by these pores (Figures 6(a), 7(a), and 8(a)). In T. mammillata, the polygonal upper mesh canal network consists of large polygons with a continuous perforated septum connecting it to the lower mesh canal system ( Figure 6(d)). The relatively large pores on the surface connect to vertical canals of the upper mesh canal system that arise at the intersections of the polygonal canals, or halfway in-between intersections ( Figure 10(a)).
The upper mesh canal system in the head shield of T. milleri (Figure 7(d)) has large, main canals that are continuously connected to the lower mesh canal system through a perforated septum (Figure 7 (c)). However, as mentioned before, T. milleri has additional upper mesh canals within the larger polygons, (Figure 10(b)), here termed intrapolygonal upper mesh canals. These further subdivide the polygons into smaller areas and lie vertically higher than the largest upper mesh canals ( Figure 1(d)). As described before, some of these intra-polygonal canals host at least a partial perforated septum with an equivalent canal of the lower mesh canal system (arrowhead in Figure 7(c)). Additional upper mesh canals that are much thinner and positioned even higher compared to all the other upper mesh canals further subdivide the polygons (Figures 7(d) and 10(b)), but these do not have any perforated septa or equivalent lower mesh canals. There are many more canals rising to the surface from the upper mesh canal system in T. milleri compared to T. mammillata, and subsequently the pores are more tightly spaced on the surface (Figure 7(a)). Furthermore, the majority of the ascending canals split into two toward the surface and give rise to two pore openings (e.g., arrowhead in Figure 7(d)).
The upper mesh canal system of T. schmidti (Figure 8 Figure 8(d)). The pore openings on the surface are denser in T. schmidti when compared to T. mammillata, but F I G U R E 8 Three-dimensional models of the Tremataspis schmidti head shield fragment . Colors indicate the same as in previous figures. Arrowheads in (d) point to paired or forked ascending canals of the upper mesh canal system. Arrows in (d) point to a complete (top) and incomplete (bottom) intra-areal canal not as dense as in T. milleri. Furthermore, the majority of the ascending canals leading to pores are singular as in T. mammillata, but they can sometimes be paired, or fork into two toward the surface (arrowheads in Figure 8 Figure 9(f)). This intra-areal canal appears to have formed within a single modified tubercle (arrow in Figure 11(a)). The second scale of T. perforata (GIT 769-16) investigated here is much more chaotic in its inner structure (Figure 11(b)) and has not been fully modeled out for this reason. However, in this specimen some of the upper canals clearly formed between separate, individual tubercle-like structures that sometimes fuse or overgrow each other (Figure 11(b),(c)). In fact, the tips of buried tubercles are sometimes visible inside the surface pores of the upper mesh canals (arrowheads in Figure 11(d)).

| Comparisons to previous histological studies
Buried tubercles in P. oeselensis were shown in a histological sketch by Janvier (1996, fig. 4.17A), but the original publication identifies the investigated specimen as a Cephalaspis sp. from Spitsbergen (Ørvig, 1951, fig. 11B). We agree that the specimen is similar to  Denison (1952) used size data to conclude that T. mammillata, unlike some other osteostracans (Hawthorn et al., 2008;Keating et al., 2012), probably had a determinate growth and that its hard tissues only formed when its shield was fully grown. Denison (1947) showed different stages of development in T. mammillata, where the youngest specimen only showed ossification of a continuous, thin outer dentine and enameloid cover, and the only ossification of the middle layer was found directly surrounding the polygonally arranged lower and upper mesh canals (Denison, 1947, fig. 11A). At later growth stages, the middle layer would become increasingly ossified (Denison, 1947, fig. 11B, C) and the last layer to form was the basal layer that thickened toward the base (Denison, 1947, fig. 11C, D). A single mineralization phase of an encapsulating head shield further supports a determinate growth in T. mammillata and Tremataspis in general. Denison (1951) suspected The D. gemmifera head shield fragment investigated here is unlike any head shield material previously investigated by both Wängsjö (1946: pl. 5, fig. 1) and Denison (1951, fig. 32C), where each tesserae with a large tubercle on the surface hosted a large central basal cavity. However, these features may depend on the life-position of the head shield fragments. The head shield fragment investigated here also has two main and large mesh canals that run deeper than the remaining mesh canal system, which is similar to what is observed in the scale of this taxon.
The presence of pore fields has been used as a synapomorphy for thyestiid osteostracans, (e.g., Afanassieva & Märss, 1997;Sansom, 2008), but it should be mentioned that the P. oeselensis material used for this character by Sansom (2008) was first described as Thyestes sp. in Gross (1968b, fig. 12B, D, 13A-E). Fredholm (1990) later referred these to Procephalaspis oeselensis?, but this was chal- Furthermore, the canal openings often have a second layer forming a rim around the opening, which could indicate the former presence of a pore field (e.g., Märss et al., 2014, fig. 34I).
According to Afanassieva and Märss (1997), the pore fields in A. viitaensis may fuse in places to form a small, perforated septum, but this was not identified in the material studied here. Material from the dorsal side of the head shield of D. gemmifera presented by Wängsjö (1946) and Denison (1951) has extensive perforated septa overarched by modified tubercles that attain tessera-like morphologies. These sit in-between larger tubercles and create a semienclosed canal network (see also Gross, 1961). The same pattern was identified in both the head shield and scale fragments of D. gemmifera investigated in this study. The ventral side of the head shield in D. gemmifera on the other hand, lacks the large tubercles and is instead entirely covered by modified and tessera-like tubercles, which effectively forms a continuous, polygonal network of semi-enclosed upper canals (see Denison, 1951). Previous studies have pointed out the similarity of both the perforated septum and the polygonal network of these semi-enclosed canals in D. gemmifera to the upper mesh canal system found in Tremataspis.  (Figure 6(c)) or T. perforata (Figure 9(f)). This was also reported for T. mammillata by Denison (1947) fig. 2C). Furthermore, T. milleri has intrapolygonal upper mesh canals, that were homologized by both Stensiö (1932) and Denison (1951) to the intra-areal canals described in cephalaspids (see below).

| Potential homologies of canal systems
The deeper mesh canals that acquire a radial appearance in the P. oeselensis head shield fragment are reminiscent to the radial canals described in cephalaspids by Stensiö (1927Stensiö ( , 1932. Radial canals in cephalaspids constituted the deepest set of vascular canals that ran parallel to the external surface of the dermal elements and radiated out from the center of each tessera to connect to the radial canals of neighboring tesserae (Stensiö, 1927(Stensiö, , 1932. According to Afanassieva (1999), radiating canals are also present in the earlybranching osteostracan Ateleaspis, which suggests that it is plesiomorphic for osteostracans (O'Shea et al., 2019). Wängsjö (1952) drew comparisons between the deepest, but often irregular, vascular canals in several thyestiid taxa to the radial canals of other osteostracans, and Afanassieva and Märss (2014) homologized the network of deeper mesh canals in A. viitaensis to the radial canals of other osteostracans. The radially arranged canals surrounding large tubercles in D. gemmifera have also been homologized to the deep radial canals of other osteostracans (Afanassieva, 1995;Wängsjö, 1946). However, Gross (1961) stated that this could not be the case because they occur much higher in the middle layer compared to the deeply underlying radial canals proper, which is supported by the data presented here.
The potential homology between the semi-enclosed upper canal system of D. gemmifera to the pore-canal system of Tremataspis has been suggested before (e.g., Afanassieva, 1995;Denison, 1951). The similarity is perhaps best seen between the ventral shield of D. gemmifera (Denison, 1951, fig. 33A) and T. mammillata (Figure 6(d)).
Considering the phylogenetic framework presented by Sansom (2008Sansom ( , 2009), a version of this system can be seen in A. viitaensis (Figure 3(d)), where modified tubercles partially overhang pore fields.
Further modification of the morphology of these tubercles into tessera-like plates could form the more continuous, semi-enclosed upper canal system seen in D. gemmifera (Figure 5(d)). The dorsal head shield of D. gemmifera often has a large tubercle at the center of what appears to be proper tesserae, which in turn is surrounded by smaller and tessera-like, modified tubercles (Wängsjö, 1946: pl. 5 fig. 1).
These tessera-like tubercles overhang a perforated septum and essentially form a semi-enclosed upper canals system. According to Wängsjö (1946), however, they are all of similar dimensions, so there was no way of distinguishing between upper canals occurring between and within the tesserae proper. The vasculature of the tessera-like tubercles is positioned higher up in the middle layer and they essentially overgrow the flanks of the larger tubercles, which is also the case in the scale studied here (Figure 1(c): tessera-like tubercle). On the ventral side of the head shield, however, it appears that each main tubercle of the tesserae proper has been modified to overhang the perforated septa that border them (Denison, 1951, fig.   32B). The superficial similarity between the semi-enclosed upper canal system of the ventral head shield in D. gemmifera to the polygonal network of upper mesh canals seen in T. mammillata has led to suggestions that the large polygonal canals in the latter represent borders between tesserae (Stensiö, 1927). Stensiö (1927Stensiö ( , 1932 described so-called mucus canals in cephalaspids that were divided into inter-areal canals (at the borders between tesserae) and intra-areal canals (between tubercles within tesserae). In cephalaspids that had a continuous superficial layer, both the inter-and intra-areal canals were entirely enclosed in hard tissues and only opened to the surface through pores, but they did not arise from and were not connected to the subepidermal vascular plexus (Stensiö, 1932contra Stensiö, 1927. Therefore, the inter-and intraareal canals were fully enclosed in more superficial tissues and formed a canal system separate from the vasculature of the middle layer, which is similar to what is seen in Tremataspis (Stensiö, 1932). Following the comparison between D. gemmifera and T. mammillata, previous workers further homologized the different parts of the upper mesh canals in Tremataspis to the inter-and intra-areal canals of cephalaspids. However, unlike Tremataspis, the canals in cephalaspids only connected to the mesh canals of the middle layer in a few places (Stensiö, 1932, fig. 6) and not through a continuous perforated septum. further suggested that the upper mesh canals represent partial boundaries between regions of odontogenesis, because the vasculature (the lower mesh canals) of each polygon had no clear connections to each other above the upper mesh canals (e.g., Figure 6(d)). However, as pointed out by Denison (1947), and which is visible in the material presented here, the lower mesh canals of each polygon connect extensively to each other above the upper mesh canals in both T. milleri and T. schmidti, essentially fusing the vasculature of each region (Figure 7(d)). The largest upper mesh canals in these two taxa are of similar dimensions as the upper mesh canals in T. mammillata ( Figure 10(a),(b)), but there are also additional upper mesh canals within the polygons (i.e., intra-polygonal canals) that are positioned at a higher vertical level, some of which has a short perforated septum underlying it. Some of these intra-polygonal upper mesh canals also has lower mesh canals reaching over and connecting above them (Figure 7(d)). Furthermore, some of the smaller upper mesh canals in T. perforata appears to have formed within a modified tubercle itself (Figures 9(f) and 11(a)).
Because there are generally no preserved traces of odontodegenerations or centers of tesserae in Tremataspis, and because the shape of an odontode dictates the underlying vasculature (see Donoghue, 2002), it is not possible to determine if the upper mesh canals in Tremataspis represent borders between tesserae or tubercles. Furthermore, some of the smaller canals in T. perforata seem to have formed within modified tubercles themselves. A distinction between inter-and intra-areal canals seems difficult to make in D. gemmifera as well (Wängsjö, 1946). For these reasons, it is not possible to establish a homology between the polygonal and intrapolygonal upper mesh canals in Tremataspis to the inter-and intraareal canals of cephalaspids, as suggested by Stensiö (1927Stensiö ( , 1932. However, the results of this study strongly support the homology between the upper mesh canal system in Tremataspis to the partially and semi-enclosed upper canals of other thyestiids.  Reif (1982). Essentially, the mutual inhibition of dermal papillae would regulate the growth of odontodes in a reactiondiffusion system (see Cooper et al., 2018;Donoghue, 2002;Maisey & Denton, 2016). This also reconciled the single phase of mineralization with the "growth lines" that had been identified in some Tremataspis specimens (Denison, 1947;Janvier, 1985), by identifying them as the breakdown of Turing patterning (O'Shea et al., 2019). Indeed, the occurrence of tessera-like plates that sometimes attained more tubercle-like morphologies on the dorsal head shield of D. gemmifera (Gross, 1961;Wängsjö, 1946) Figure 12) that subdivide the polygon into 2-4 areas, and two levels of smaller intra-polygonal upper mesh canals (Numbers 3 and 4 in Figure 12) that further subdivide the polygon. As stated before, the lower mesh canal system connects regularly above the largest upper mesh canals (Number 1 in Figure 12(c)), and often over the intra-polygonal upper mesh canals as well (Number 2 in Figure 12(c)). The intra-polygonal upper mesh canals also appear to be influenced by the deeper generations of upper mesh canals that they connect with (Figure 12(b)). In theory, the same reaction- Considering the growth stages presented by Denison (1947) for Tremataspis, and the lack of primary odontodes or extensive remodeling in the canal system of T. mammillata (O'Shea et al., 2019), its dermal skeleton was most likely embedded in a single phase of mineralization. The scans of T. milleri and T. schmidti also lack buried tubercles or any signs of major remodeling (Figure 1(d sensu Gross, 1961) suggests the presence of an upper mesh canal system in the dermis between a first generation of odontodes, which is eventually covered by a second generation of hard tissue formation.
The initial formation of a continuous surface of enameloid and dentine that encapsulated both the surface of the dermal element and the upper mesh canal system below (Denison, 1947;O'Shea et al., 2019) points to the prior existence of a similar network in soft tissues in Tremataspis as well. Qu et al. (2015) suggested that the upper mesh canals in T. schmidti formed by invaginations of the epithelium. The different levels of upper mesh canals in T. milleri could therefore have formed by subsequent invaginations of epithelium in conjunction with a thickening of the dermis. In such a system, the outer surface together with the upper mesh canals would represent a continuous mesenchyme-epithelium contact. However, no direct trace of this is preserved in the hard tissues.
The phylogenetic framework presented by Sansom (2008Sansom ( , 2009 makes it possible to present a tentative model for the emergence of the upper mesh canal system in Tremataspis (Figure 13). The overgrowth of tubercles observed in P. oeselensis (Figure 13(a)) can be seen to a certain degree in A. viitaensis as well (Figure 13(b)). Aestiaspis viitaensis also has modified tubercles overarching pore fields, and tubercles are further modified into tessera-like plates in D. gemmifera F I G U R E 1 2 (a) Virtual scan slice of the Tremataspis milleri head shield fragment (GIT 769-1) with numbers 1-4 indicating different levels of upper mesh canals (teal); (b)-(c) threedimensional models of a single polygon from the same dermal element with the lower mesh canals (pink) and four levels of upper mesh canals in different shades of blue (labeled with numbers 1-4); (c) the relation between the upper mesh canals to the lower mesh canal system and the subepidermal vascular plexus. Note that the basal cavities and canals have not been included or highlighted ( Figure 13(c)). The dorsal head shield in D. gemmifera has a younger network of vasculature supplying the secondary, modified tubercles developing around each main tubercle (Figure 13(c1)), while the ventral head shield (Figure 13(c2)) could represent the previously mentioned reaction-diffusion system with a single generation of tubercles. The overgrowth of hard tissues in P. oeselensis and A. viitaensis potentially gave way to a single mineralization phase in most Tremataspis. The upper mesh canals in T. perforata appear to have formed both between tubercles and within modified tubercles themselves, but it is not possible to deduce how the upper mesh canals formed in other Tremataspis because no trace of this is preserved in their hard tissues. In another thyestiid, namely Thyestes verrucosus, a growth model was presented by Afanassieva (2014) where the earliest mineralized parts of tesserae had no contact with the mineralized parts of neighboring tesserae during growth, which suggests a delayed onset of mineralization. In such a scenario, much of the framework in Tremataspis could have formed in soft tissues first, with subsequent invaginations of epithelium ( Figure 13(d1), (d2)) before the initiation of mineralization in the adult-sized animal ( Figure 13(d3)), as postulated by Denison (1947).

| CONCLUSIONS
The tessellate condition of the dermal skeleton is generally considered plesiomorphic for osteostracans (Janvier, 1996). In many groups, the deepest vasculature is composed of radial canals that emanate from the center of each tessera and eventually connect to the radial canals of neighboring tesserae (Afanassieva, 1999;Stensiö, 1927Stensiö, , 1932Wängsjö, 1952). Furthermore, the central part of each tessera generally has a group of basal cavities and canals underlying it (Stensiö, 1927;Wängsjö, 1952). The pattern of both the basal and deepest mesh canals seen in the P. oeselensis head shield fragment presented here (Figure 2(b),(c)) could reflect the condition seen outside of Thyestiida. This pattern appears to be lost in other thyestiids, where the deepest mesh canals form a much more irregular network (see also O'Shea et al., 2019;Qu et al., 2015;Wängsjö, 1952), but the basal canals become increasingly more regular in their layout. The regularly placed, large basal cavities seen in both D. gemmifera and Tremataspis have often been viewed as indicating separate tesserae (Denison, 1947;Wängsjö, 1946), although their relation to the vasculature of the middle layer may differ in detail (Wängsjö, 1946). For example, the descending canals at tessera-borders described in cephalaspids appear to be missing in D. gemmifera (Wängsjö, 1946).
Because the dermal elements of T. mammillata is superficially similar to the ventral head shield of D. gemmifera, its upper mesh canal system was considered homologous to the semi-enclosed upper canal system of D. gemmifera. Because of this, and because each polygon is generally underlain by a basal cavity, Denison (1951) also viewed them as homologous to the inter-areal canals of other osteostracans.
A polygonal network of canals is often present in osteostracans, although they may range from open grooves to fully enclosed canals F I G U R E 1 3 Phylogenetic relations and head shield outlines of the thyestiid taxa included in this study. The schematics to the right illustrate the suggested growth of hard tissues in each taxon. (a) Procephalaspis oeselensis; (b) Aestiaspis viitaensis; (c) Dartmuthia gemmifera with dorsal (C1) and ventral (C2) head shield structure inferred from Denison (1951, fig. 32B, C); (d) Tremataspis milleri with the possible development (indicated by black arrows) of soft tissues ((d1) and (d2)) and hard tissues after a single mineralization stage (d3). The lower mesh canals are expanded in (d1) and (d2) to represent presumed soft tissues (Stensiö, 1932;Wängsjö, 1952). This study confirms the primary homology between the semi-enclosed upper canals in D. gemmifera to the upper mesh canals in Tremataspis as a whole. However, because there are no traces of tubercle generations preserved in the hard tissues of most Tremataspis, it is less clear whether the inter-polygonal canals reflect borders between tesserae or tubercles. For the same reason, it is not possible to say whether the upper mesh canals in Tremataspis could equate to the inter-and intra-areal canals of other osteostracans. However, it is possible that both Tremataspis and other osteostracans embedded a canal system that existed ancestrally in soft tissues, as suggested by Stensiö (1932), although they differ in detail (e.g., the perforated septum dividing the canal system is only present in thyestiids).
It is clear that the complex pore-canal system in Tremataspis evolved through the modification of the semi-enclosed canal system present in closely related thyestiids. In turn, the semi-enclosed system in D. gemmifera was produced by modifying the growth and shape of superficial tubercles. However, it is not possible to homologize the different parts of the upper mesh canal system between different thyestiids, or to similar canal systems of other osteostracans. Similar investigations on a wider array of taxa may produce phylogenetically informative traits that can help test potential homologies and help us acquire a better understanding of the dermal hard tissues in osteostracans in general. In extension, this may help elucidate the origin and evolution of the dermal skeleton in jawed vertebrates.

ACKNOWLEDGMENTS
The authors would like to thank the reviewers for their input, which greatly improved this work. We also thank Ursula Toom, Tallinn, for help connected with the loan of material. The experiments were performed on beamline ID19 at the European Synchrotron Radiation Facility (ESRF, proposal ES 581) in Grenoble, France. We acknowledge the European Synchrotron Radiation Facility for provision of synchrotron radiation facilities and for assistance in using beamline ID19.
T.M. acknowledges the Department of Geology at TTU for allowing her to use their facilities for research.

CONFLICT OF INTEREST
The authors have no conflict of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available in the ESRF database (http://paleo.esrf.eu/).