Ever since Rathke [1], who found a discontinuity between the rostral and caudal part of the cranium, a part of the gnathostome neurocranium known as the trabecula cranii has drawn the attention of vertebrate morphologists (Fig. 1). The difference between the anterior and posterior portions of the neurocranium resides primarily in their morphological relationships; specifically, whether or not the cranial base is medially associated with the notochord, in a manner similar to the vertebral column. The morphological interpretation of this division is profoundly related to the segmental theory of the vertebrate head, developed from Goethe’s and Oken’s ideas about the skull being composed of serial homologues of vertebrae [2–4]. Gegenbaur and Froriep thought that only the posterior part of the gnathostome neurocranium, which lies alongside the notochord, could be compared with the vertebral column; the neurocranium could thus be divided into prespinal and spinal parts [5, 6].

As recognized by Rathke, the above noted two regions initially arise as two pairs of rod-like cartilages, trabeculae and parachordals (Fig. 1). It was Thomas Huxley who first put forth the idea that the trabecula might represent a highly modified element originally belonging to a visceral arch situated rostral to the mandibular arch (premandibular arch) [7]. In the early twentieth century, the original idea of Huxley was elaborated by many morphologists, including Goodrich and Sewertzoff [8–10]. De Beer also illustrated a schematic diagram to explain how an ancestral amphioxus-like animal evolved into the gnathostome morphotype (Fig. 2a) [11]. This ancestor is assumed to have possessed a series of visceral arches along the anteroposterior axis, and these arches showed no clear differentiation. Of these, the rostral three arches represent prospective premandibular, mandibular and hyoid arches. On the other hand, the shark pharyngula-like gnathostome embryo has the trabecula, functioning as a floor for the expanded forebrain, derived from the premandibular arch. The ammocoete larva-like creature is placed in the middle as an intermediate stage, where the premandibular arch is positioned rostral to the mandibular arch.

Toward the end of the 20th century, the concepts developed by Gegenbaur and Froriep gained support from experimental embryology, which showed that the trabecular part of the neurocranium is of cephalic neural crest origin [12, 13], and only the notochord-associated posterior part of the neurocranium is of mesodermal origin, like the vertebral column (Fig. 3). Thus the neural crest/mesoderm distinction does not coincide with the functional distinction of the neurocranium versus viscerocranium. It was further suggested that the paraxial mesoderm requires a signal emanating from the notochord to differentiate into the skeletal tissues, whereas the cephalic crest cells can differentiate without this signal. On this basis, Couly and his colleagues designated the anterior and posterior parts of the neurocranium as the prechordal and chordal cranium, respectively [12]. The two types of mesenchymal cell lineages correspond perfectly to the presence/absence of the notochord, distinct molecular basis of developmental signaling, as well as two types of the neurocranial anlagen. Moreover, the neural crest generally contributes to the visceral part of the cranium (visceral arch skeletons), making the premandibular arch-hypothesis of Huxley [7] appear somewhat plausible, although the basic idea of ‘arches rostral to the mandibular arch’ is currently not supported by many morphologists, mainly due to the apparent lack of pharyngeal pouch-like structures rostral to the mandibular arch.

Although this theoretical framework for the interpretation of the gnathostome neurocranium has a prominent evolutionary dimension, made explicit in De Beer’s schematic diagram (Fig. 2a), it was developed almost entirely without reference to the fossil record. During the late 19th and early twentieth century, fossils of primitive armoured jawed vertebrates (placoderms) and jawless vertebrates (ostracoderms) had already been discovered in rocks from the Devonian period (419–359 million years ago), but they were known only from their external anatomy and were thus essentially uninformative about neurocranial evolution. However, starting with Erik Stensiö’s groundbreaking investigations of ostracoderm and placoderm cranial anatomy by serial grinding and the construction of wax models [14–16], a series of publications have illuminated the neurocranial anatomy of early vertebrates using the techniques of serial grinding, acid preparation, and more recently synchrotron microtomography [17–22].

These fossils provide detailed data on the shape of the cranial cavity, including the position and approximate morphology of the nasal sacs, inner ears, hypophysis, pineal and parapineal organs, cranial nerves and cranial blood vessels, as well as external features of the neurocranium, such as the articulation points for the visceral arch skeleton. Together these structures create a network of landmarks that can be used for drawing inferences about neurocranial composition, by comparison with developmental data from extant vertebrates. This approach has been greatly strengthened by the discovery that major parts of early cranial development, including the patterns of neural crest cell migration and distribution, are conserved between cyclostomes and gnathostomes [23, 24]. The fossil ostracoderms and placoderms all belong to the gnathostome stem group (Fig. 4), and thus fall within the ‘phylogenetic bracket’ of extant cyclostomes and gnathostomes [25], meaning that the inferences are methodologically robust.

The present review is intended to revisit this developmental morphological issue from an evolutionary standpoint, incorporating fossil data from stem gnathostomes and developmental data from cyclostomes, and to put forth a hypothesis that the ancestral vertebrate cranium may have shown a simpler pattern of development, in which the neurocranium (sensory capsules excluded) was purely mesodermal in origin, and the cephalic neural crest cells contributed only to the oro (naso)-visceral part of the head.

The heads of cyclostomes (lampreys and hagfishes) differ from those of gnathostomes in many ways other than the obvious absence of jaws. Notably, the forebrain of cyclostomes is anteroposteriorly shorter, the distance between the hypophysis and nasal sacs is smaller, and these organs are located in a common median nasohypophyseal duct [23, 24, 26, 27]. It has already been pointed out that the cephalic crest-derived premandibular ectomesenchyme should be found in the dorsal part of the lamprey oral apparatus, rather than ventral to the forebrain as in a gnathostome [27–29]. However, De Beer [11, 30] was correct in that the lamprey has a less expanded forebrain, and thus a rostral notochordal tip that reaches closer to the rostral end of the head, than a gnathostome. In an ammocoete-like embryo, or a hagfish embryo, it is therefore expected that the neurocranium will be predominantly chordal (of paraxial mesodermal origin) and that the trabecular homolog will not take part in the formation of the neurocranium (see below).

In the lamprey cranium also, traditional comparative morphology has described an inverted U-shaped cartilage called the trabecula (Fig. 2b) [23, 24, 27, 31–34]. It extends rostrally beyond the notochordal tip to the forebrain level and encircles the hypophysis. This structure, however, is not identical to the ‘premandibular arch’ in the ammocoete-like ancestor illustrated by De Beer as in Fig. 2a, nor the premandibular ectomesenchyme that forms the upper lip of ammocoetes. Indeed, it has been shown that the lamprey trabecula represents a rostrally extended parachordal, the paraxial mesodermal cartilage: the earliest anlage of the trabecula is found at the level of the mandibular arch, which grows secondarily rostrally during development [32, 34].

The developmental origin of the transverse commissure is not known; however, it is conceivable that it represents the rostralmost component of the mesodermal neurocranium; the rostralmost paraxial mesoderm in gnathostome embryos is also connected in the midline with its counterpart, just rostral to the notochord reviewed by [35]. The similarity of the morphological configuration between the premandibular mesoderm (cavity) in gnathostomes and transverse commissure in the lamprey supports the above inference, which, however, needs to be tested experimentally. Rostral to the lamprey trabecula is the nasal capsule that is thought to be of neural crest origin (ectomesenchyme in the “anterior process”) [23, 27].

Much less is known about hagfish cranial development [33, 36–38]. However, it has been suggested that, similar to the lamprey, longitudinal rod-like cartilages forming a functional neurocranium are derived from parachordal head mesoderm, and rostrally connect to the oral cartilages derived from the neural crest [23]. The apparent homolog of the ammocoete transverse commissure would be found in a cartilaginous bar, called the hypophyseal commissure, located slightly rostral to the hypophysis, showing a topographical relationship similar to that seen in the lamprey larva. Therefore, the position of this commissure may indicate the rostral limit of the mesodermal neurocranium in the hagfish, as has already been represented in a diagram in a previous paper (Fig. 10 in Reference [23]).

Thus, apart from the crest-derived nasal capsule, the cyclostome neurocranium is mostly formed of mesodermal mesenchyme, and there appears to be no distinct anterior crest-derived neurocranial moiety as seen in the gnathostome chondrocranium (Fig. 3). The question now is, whether the vertebrate neurocranium was composed of crest-derived and mesodermal moieties from the very beginning of their history, or whether the crest-derived part is a gnathostome innovation.

The most detailed fossil evidence about the neurocranium of early vertebrates comes from the jawless osteostracans [14, 16, 18, 39] and galeaspids [20], and the jawed placoderms [15, 17, 21, 22]. These groups all have perichondrally ossified braincases, which preserve the morphology of the cranial cavity, nerve canals and vascular canals. Another fossil jawless group, the heterostracans, lack perichondral ossification, but show faint impressions of the brain, semicircular canals, and branchial pouches on the inner surface of the dermal bones of the head shield [39]. As mentioned above, all these fossil groups belong to the gnathostome stem group. In the older literature the jawless stem gnathostomes are often referred to, collectively, as ‘ostracoderms’ (see above). The uppermost part of the stem is composed of the placoderms, which are themselves paraphyletic (i.e. some placoderms are more closely related to crown gnathostomes than others) [21, 22, 40]. Below them sit the osteostracans, followed by the galeaspids; both groups are monophyletic (Fig. 4). Together, these fossil groups straddle the transition from jawless to jawed vertebrates.

Osteostracans and galeaspids have broad ‘head shields’ that in fact incorporate the entire craniopharyngeal region as far back as the pectoral fins (if present). The neurocranium, which is not subdivided by sutures, is very broad and covers the entire dorsal surface of the gill region. Both groups have a nasohypophyseal duct with a dorsally positioned external opening. The hyoidean branchial pouch appears to have been developed into a normal full-size gill, similar to those on the branchial arches (Fig. 5).

Past analyses of the fossil jawless vertebrates have been greatly influenced by the perception that osteostracans are essentially ‘lampreys in armour’. This idea originated from the observation that the cranial cavity ends anteriorly in a small, ventrally closed nasohypophyseal duct, which opens between the eyes in a lamprey-like manner [14, 16, 18, 39]. However, the positional relationship of the pharynx to the cranial cavity and nasohypophyseal duct is radically different in osteostracans compared to lampreys and other vertebrates (Fig. 5).

The branchial apparatus of osteostracans is displaced so far anteriorly that not only the hyoid gill pouch but also the first two branchial gill pouches lie anterior to the eyes [14, 16, 39]. This uniquely specialized morphology means that the nasohypophyseal duct cannot have a ventral opening onto the palate, because the position that this opening would occupy is already filled by the dorsal aorta of the branchial apparatus (see Fig. 349a in Reference [39]). The apparent similarity with the ventrally closed nasohypophyseal duct of a lamprey, which does not have this positional relationship to the gills, is thus probably convergent.

The specialized pharyngeal morphology of osteostracans limits their utility for investigating the evolution of the cranial cell population map. If we assume that the trigeminal crest cells descended in their customary place, between the optic and otic vesicles, we are forced to conclude that the resulting premandibular and mandibular ectomesenchyme underwent a dramatic forward migration to reach its final destination. The picture is made still more complicated by fan-shaped arrays of canals, emanating from the saccular regions of the inner ears, which terminate near the lateral margins of the head shield and are believed to have had a sensory function (Fig. 5c, d). As the inner ears of living vertebrates are always enclosed within the mesodermal otic capsule, these canal arrays suggest that otic capsule mesoderm extended laterally, above the branchial ectomesenchyme of the gill region, almost to the edges of the shield. However, it is impossible to map the precise boundaries between the premandibular ectomesenchyme (prechordal part of the neurocranium) and mesoderm (chordal cranium), let alone between different components of the ectomesenchyme.

By contrast, galeaspids, heterostracans and placoderms all have hyoid- and branchial arches posterior to the eye. There is some evidence that heterostracans had a hagfish-like nasohypophyseal duct that opened anteriorly [39], but as their neurocranial anatomy cannot be reconstructed in detail they will not be considered further. In galeaspids the nasohypophyseal duct, which is large and either oval or slit-shaped, opens dorsally on top of the head in front of the eyes, and ventrally on the palate (Fig. 5a, b). The nasal sacs and hypophysis open into the duct [20]. Interestingly, the anteriorly directed hypophysis and the olfactory tracts are separated by neurocranial tissue as in crown gnathostomes, which forms a short spike projecting into the nasohypophyseal duct. This spike has been interpreted as a rudimentary trabecula [20]. Its existence suggests that galeaspids had separate nasal and hypophyseal placodes, rather than a single nasohypophyseal placode as in cyclostomes, allowing premandibular ectomesenchyme to differentiate into primitive prechordal cranium that grew forward between the placodes.

Galeaspids, osteostracans and the majority of placoderms all have very short forebrains, with the result that the cranial cavity hardly extends in front of the eyes [14–18, 21, 22, 39]. This resembles the condition in extant cyclostomes and most probably represents retention of the primitive vertebrate condition. Nevertheless, the broad and slab-like neurocrania of galeaspids and osteostracans extend not only laterally to cover the whole branchial region, but also anterior to the eyes and nasohypophyseal duct, right to the tip of the snout (Fig. 5). This preorbital region effectively corresponds to the lamprey upper lip [27]. In galeaspids the entire preorbital region lies anterior to the first pharyngeal pouch, and is presumably composed of some combination of mandibular and premandibular ectomesenchyme; in osteostracans it may also incorporate hyoid and even branchial ectomesenchyme, due to the anterior displacement of the pharynx. In other words, the neurocranium of these jawless stem gnathostomes incorporates an anterior component derived from premandibular ectomesenchyme, but unlike in crown gnathostomes it does not underlie the brain as the trabecula.

Placoderms have separate left and right nasal sacs opening onto the face, and an adenohypophysis opening onto the palate, in the typical gnathostome manner; there is no nasohypophyseal duct. As mentioned above, the forebrain cavity is usually very short. It is floored by a part of the neurocranium that lies in front of the buccohypophyseal foramen, medial to the palatoquadrates, and can thus be identified as a trabecular region (Fig. 5). Compared to crown gnathostomes this region is very broad, reflecting the generally broad and flat character of the neurocranium (probably a primitive character retained from jawless ancestors; see above).

Interestingly, its length is highly variable. In arthrodire placoderms (Fig. 5h-j), and in the so-called ‘maxillate placoderms’ which may be close to the origin of osteichthyans, the nasal capsules are positioned terminally on the snout and the trabecular region is short, reflecting the short forebrain cavity [15, 19, 39, 40]. However, in a few placoderms including the acanthothoracid Romundina [21, 22], the trabecular region is developed into a projecting ‘upper lip’ that extends forward below and in front of the nasal capsules (Fig. 5e-g). Its ventral position, and the fact that it is flanked throughout its length by the palatoquadrates, indicates that it is composed of premandibular infraoptic ectomesenchyme, the anlage that gives rise to the upper lip in the lamprey. In these placoderms the crest-derived part of the neurocranium is thus largely, but not entirely, anterior to the cranial cavity. The similarity with jawless vertebrates, especially galeaspids, is obvious and probably indicates that this is a retained primitive character [21].

It seems that the origin of the crown gnathostome trabeculae involved several steps and provides an example of how a seemingly trivial anatomical modification can open the door to major morphological change. In a cyclostome, the forebrain cavity is floored only by the soft dorsal wall of the nasohypophyseal duct. The first step in the origin of the trabeculae, starting perhaps already in galeaspids but brought to completion in early placoderms, was the re-routing of at least part of the premandibular ectomesenchyme into the space between the hypophyseal and nasal placodes. This obliterated the nasohypophyseal duct (although the buccohypophyseal foramen can be considered a remnant of its ventral opening on the palate), and gave the forebrain cavity a skeletal floor [41].

Initially the forebrain remained very short, as did the trabecular region sensu stricto, i.e. the part of the prechordal neurocranium that floors the forebrain cavity. The loss of the projecting premandibular “upper lip” in arthrodires and maxillate placoderms was not immediately accompanied by lengthening of the forebrain, but simply led to a shortening of the preorbital face. However, the fact that the forebrain was now enclosed in a protective skeletal box meant that it could be lengthened without deleterious effects, in a way that the unprotected forebrain of a cyclostome cannot. Within a relatively short time period, an elongated forebrain evolved at least three times in parallel in early jawed vertebrates; in tapinosteid arthrodires, in macropetalichthyid placoderms, and in crown gnathostomes [15, 21].