Development of a straight vertebrate body axis

The vertebral column or spine is the defining feature of vertebrate animals. This structure provides axial stability, leverages movement through the environment and protects the spinal cord. Typical patterning of the spine consists of a segmented alternating pattern of intercalated vertebral bodies, or centra, and intervertebral discs (Fig. 1A,B). In amniotes (e.g. mammals and birds), centra are assembled around the notochord and the neural arches enclose the spinal cord during embryonic development, whereas in many teleosts (bony fish) and amphibians, the formation of the centra and neural arches occurs in freely swimming larvae. Owing to its axial structure and central position in the body, morphogenesis of the adult spine requires the integration of several distinct musculoskeletal tissues including bone, cartilage and connective tissue, muscles, vasculature and the nervous system. Thus, the spine physically connects axial structures, providing a scaffold for the vertebrate body plan.

In both amniotes and bony fish, the spine shares the same metameric plan (Fleming et al., 2015) (Fig. 1A,B). Although the mechanisms controlling vertebral patterning are not fully understood, recent studies have shown clear differences between amniotes and teleost species. In amniotes, the segmented pattern of the spine is largely thought to be patterned during, and as a consequence of, somitogenesis: the process in which the presomitic mesoderm becomes compartmentalized into organized somites that flank the notochord along the midline. The molecular details of this segmentation process have been investigated intensively in a variety of vertebrate model systems (Sheeba et al., 2016; Venzin and Oates, 2020). Later in development, the myotome portion of the somites gives rise to the axial muscles, while the ventral most portion (the sclerotome) migrates from each side towards the midline and fuses, forming a cartilaginous progenitor tissue (anlage), which encircles the notochord and spinal cord (Eckalbar et al., 2012) (Fig. 1C). Sclerotome-derived tissue gives rise to the major structural elements of the spine including the osteoblast lineages forming the vertebral bones, the outer annulus fibrosus portion of the intervertebral discs, ligaments and tendons. In bony fishes, vertebral bodies are also formed from sclerotome-derived osteoblasts, but the patterning of centra/vertebral bodies is instructed via notochord-derived signals that override the somitic pattern (Peskin et al., 2020).

The dominant theory of metameric patterning of the spine in amniotes, first proposed by Remak (1855), proceeds as ‘resegmentation’ of the anterior and posterior halves of two adjacent somites, which collectively contribute to a single vertebral body (Huang et al., 2000). This theory suggests that the fidelity of somite boundaries are instructive, and directly pattern the segmentation of the spine. However, this model is controversial; for example, both zebrafish and chicken display ‘leaky’ resegmentation, where one somite can contribute to several adjacent vertebrate (Morin-Kensicki et al., 2002; Stern and Keynes, 1987). The disruption of somite boundaries in the T-box transcription factor 24 (fused somites; tbx6) mutant zebrafish does not alter the segmentation pattern of the centra, although the morphology of the neural arches is affected (Fleming et al., 2004; Wopat et al., 2018). This data suggested that, although neural arch formation follows somitic boundary cues, patterning of centra/vertebral bodies relies on cues from the notochord. Indeed, recent studies show that Notch-dependent patterning of the notochord dictates the segmentation of the vertebral column in zebrafish (Wopat et al., 2018) and when notochord segmentation cues are lost, zebrafish revert to a mechanism of vertebral patterning that relies on somite boundaries and is found in basal fishes, which gave rise to the tetrapod lineage (Peskin et al., 2020). Notochord grafting experiments in chicken provided evidence that cross talk between the notochord and the sclerotome is required for the fidelity of axial segmentation (Ward et al., 2018). Altogether, these results suggest a more complex model in which interactions between the notochord and sclerotome are necessary to refine axial patterning.

Given the tissue architecture and developmental origins of the structural elements of the spine, it is not surprising that embryonic defects of the somites and notochord are linked to the pathogenesis of spine disorders in humans and in animal models. The most common spine disorders in humans are atypical curvatures of the spine, collectively called scoliosis, a term which is derived from the ancient Greek word meaning ‘a bending’, first described generally by Hippocrates in his treatise On Articulations in 400 B.C.E. (Vasiliadis et al., 2009). In modern times, scoliosis in humans falls into three broad phenotypic groups: (1) congenital scoliosis (CS), which are observed at birth, involving vertebral malformations that directly contribute to regions of spine curvature; (2) idiopathic scoliosis (IS), which involves spine curvature without obvious vertebral malformations and in otherwise healthy children (Box 1); and (3) neuromuscular or syndromic/acquired scoliosis, which is a common co-occurring condition for a wide range of neurological or motor neuron or connective tissue diseases, Rett syndrome, amyotrophic lateral sclerosis, Ehlers-Danlos syndrome and spinal muscular atrophy (Halawi et al., 2015; Mary et al., 2018; Murphy and Mooney, 2019) or in cases of pediatric spinal cord injury (Mulcahey et al., 2013).

The onset of IS can occur over a range of developmental stages, yet it is most common during adolescence (AIS). For the most part, CS has been attributed to defects in somite segmentation, largely attributed to disruption of NOTCH signaling in both humans and animal models (reviewed by Giampietro et al., 2009; Venzin and Oates, 2020), or after disruption of notochord biogenesis and physiology, as outlined below. In contrast, the molecular genetics and pathogenesis of IS in humans is less well understood. Recent efforts integrating human genetics and functional analysis in both mouse and fish models have begun to highlight intriguing mechanisms for this historically unresolved disorder, which are also the focus of this Review. Models of IS in amniotes have been historically driven using surgically-induced rodent or large animal models (Liu and Gray, 2018). A recent genetic model of AIS has been reported in the G-coupled receptor 126 (Gpr126; Adgrg6) conditional knockout mouse (Karner et al., 2015), which was informed by the strong association of the GPR126 locus in human AIS cohorts (Kou et al., 2013) (Box 1). Additional genetic models that disrupt the proprioceptive system in mouse also display scoliosis without vertebral malformations (Assaraf et al., 2020; Blecher et al., 2017), which may provide a rationale mechanism underlying neuromuscular acquired scoliosis or AIS. Genetic studies in zebrafish have also discovered that disruption of motile cilia in the brain and spinal cord can generate adult-viable scoliosis phenotypes, which display spine disorders akin to those observed in AIS or neuromuscular acquired scoliosis. There is also evidence that seemingly distinct spine disorders such as CS and IS may have common genetic origins (Box 2). It remains to be determined whether these zebrafish models will provide a transition of knowledge to human spine disorders. The development of more sophisticated animal models modeling human alleles associated with spine disorders will be instructive for diagnosis and potential therapeutic interventions.

Here, we focus on the structural roles of the notochord that are linked to spine patterning defects such as CS. We identify unique functions of connective tissues implicated in the etiology of AIS. Finally, we discuss recent findings in zebrafish that are beginning to explain how the neural tube, cerebrospinal fluid (CSF) and the Reissner fiber contribute to axis straightening and spine morphogenesis.

In all vertebrates, the notochord is derived from the axial mesendoderm, which transitions into a rod-like tissue comprised of vacuolated cells, surrounded by a peri-notochordal basement membrane sheath (Balmer et al., 2016; Choi et al., 2008; Choi and Harfe, 2011; Jurand, 1974) (Fig. 1D). The notochord in swimming larval animals (e.g. teleosts and amphibians) is noticeably larger in diameter in comparison with the neural tube and displays a thicker extracellular matrix (ECM) sheath with a medial layer of circumferentially aligned matrix and outer layer which is orthogonally-aligned with the medial layer (Gotz et al., 1995; Gray et al., 2014; Stemple, 2005) (Fig. 1E). In zebrafish, direct imaging of transgenic-labeled vacuolated cells demonstrate that these cells expand in volume via the pressure created by the inflation of notochord vacuoles, utilizing both lysosomal and caveolar based mechanisms (Bagwell et al., 2020; Ellis et al., 2013). Both mouse and chicken also express caveolin in the notochord (Nixon et al., 2007), suggesting that similar mechanisms of notochord vacuolation are involved in amniotes as well. The expansion or vacuolation of the chordamesoderm occurs in an anteroposterior direction concurrent with the onset of somite segmentation (Dale and Topczewski, 2011). Another unique feature of the notochord in teleost species is the presence of a peri-notochordal sheath cell epithelial (also known as chordoblast) layer surrounding the vacuolated cells (Dale and Topczewski, 2011; Grotmol et al., 2003). (Fig. 1E). This sheath cell epithelium displays abundant rough endoplasmic reticulum, common in highly secretory cell types, which likely contributes to the secretion of the thick peri-notochordal basement membrane and lamellar sheath found in fish species (Coutinho et al., 2004; Trapani et al., 2017). Recent work in zebrafish showed that the molecular signal driving the segmented pattern of the spine does not in fact originate from somite boundaries and is instead generated by Notch-dependent signals within the notochord sheath cell epithelium (Lleras Forero et al., 2018; Pogoda et al., 2018; Wopat et al., 2018).

In this model, the sheath cell epithelium serves as a blueprint, directing sclerotome-derived osteoblast migration to specific regions of vertebral body formation and eventually gives rise to the mature vertebral bone (Wopat et al., 2018). Sheath cell epithelial regions are marked by expression of the enzyme Entpd5 (Ectonucleoside triphosphate diphosphohydrolase 5) in alternating domains, promoting mineralization of the ECM sheath (Huitema et al., 2012; Pogoda et al., 2018; Wopat et al., 2018). The mineralized rings that form around the notochord sheath are called chordacentra and are fish-specific structures that may function as a primitive spine in free-swimming larvae (Fleming et al., 2015). Chordacentra mineralization is controlled by retinoic acid via a complex mechanism in which production and degradation are fine tuned to define the shape and size of these calcified rings (Pogoda et al., 2018).

As mentioned earlier, notochord sheath formation and segmentation are key innovations in teleosts driving a complete developmental shift in spine patterning away from the somite-based mechanism found in amniotes (Fleming et al., 2015). The origin of this shift is unclear, but it is linked to emergence of the ECM gene calymmin (cmn) in teleosts. Recent work showed that a dominant mutation in zebrafish cmn called spondo abrogates notochord segmentation and shifts the vertebral patterning to an ancestral somite boundary-based mechanism. In spondo mutants, osteoblasts first form arches following somitic cues, as in amniotes, and then spill over the notochord to form hemi-vertebrae in a pattern that closely resembles the spine of basal fishes from the fossil record (Peskin et al., 2020).

In all vertebrates, the notochord is essential in the patterning of midline tissues including the neural tube, gut endoderm and its derivative tissues, such as the lungs, liver, pancreas and intestine (Cleaver and Krieg, 2001; Lumsden and Graham, 1995). In addition, studies in zebrafish discussed above and recent evidence from amniotes support a model whereby the notochord provides attractive signals for the recruitment of sclerotome-derived osteoblasts that form the vertebral bones (Ward et al., 2018), a process that requires hedgehog secretion by the notochord in mouse (Choi and Harfe, 2011).

In swimming larvae such as Xenopus, zebrafish and ascidian species, the notochord also acts as pressurized rod that serves as a hydrostatic scaffold providing rigidity to the main body axis. In non-vertebrate chordates, such as ascidians, an epithelial layer surrounds a continuous lumen and internal pressure is generated by ion and water transport (Deng et al., 2013; Dong et al., 2009). In contrast, in vertebrates, the notochord interior is made up of large fluid filled vacuolated cells, which essentially compartmentalize the notochord into discrete units (Ellis et al., 2013; Norman et al., 2018) (Fig. 2A). This compartmentalization allows the teleost notochord to be built progressively and is central for the formation of a straight spine (Bagwell et al., 2020). Remarkably, although the structures of the notochord differ considerably between ascidians and zebrafish, both buffer stresses acting on the notochord cells by using caveolae as a mechanoprotective mechanism (Bhattachan et al., 2020; Garcia et al., 2017; Lim et al., 2017; Nixon et al., 2007) (Fig. 2A). Caveolae are plasma membrane invaginations that serve as membrane reservoirs which can unfold when cells are stretched, thus preventing cell rupture (Sinha et al., 2011).

Much of what we know about the structural function of the notochord during vertebrate embryogenesis has come from genetic and cellular studies in zebrafish (Ellis et al., 2013; Nixon et al., 2007; Stemple et al., 1996), ex vivo experiments in Xenopus (Adams et al., 1990) and anatomical studies in salmon (Grotmol et al., 2003). In a seminal study, Adams and colleagues used dissected Xenopus notochords to probe the response of the tissue to osmotic challenge and mechanical deformation (Adams et al., 1990). This, and subsequent ex vivo and physical modeling studies (Koehl et al., 2000), led to the development of a basic model in which osmotic swelling of vacuolated cells inside a semi-rigid collagenous casing provides a hydraulic force that helps straighten and extend the embryonic axis and gives flexural stiffness to the body axis in free-swimming larvae (Adams et al., 1990). Experimental work in zebrafish also suggests that the mechanical properties of the vacuolated notochord act as a counter force against the accumulation of cortical bone, influencing the mineralization of centra and vice versa (Bagwell et al., 2020; Haga et al., 2009).

The mechanical-structural necessity of the vacuolated notochord tissue in amniotes is not known, and key developmental differences of the notochord comparing amniotes and swimming larval species exist. For example, in amniotes, the notochord transitions into segmented regions within the newly forming intervertebral disc (IVD) while the vertebral body regions are being replaced by trabecular bone in utero (Choi et al., 2008) (Fig. 1F), whereas in swimming larvae the morphogenesis of the IVD and formation of centra occurs in concert with active locomotion. Protein arginine methyltransferase 5 conditional knockout mice, display reduced trabecular bone formation in the vertebrae and scoliosis before weaning (Liu et al., 2019), which may underscore that the formation of trabecular bone in the vertebrate acts in an analogous manner to the vacuolated cells during cortical bone deposition of vertebrae in zebrafish (Bagwell et al., 2020) (Fig. 1G).

The function of the notochord in zebrafish axis elongation is dependent on the balanced differentiation of the outer sheath epithelial and inner vacuolated cells (Fig. 1E), which occurs at the end of gastrulation and is controlled by Notch signaling. Activation of Notch by expression of the Notch intracellular domain (NICD) leads to the generation of more sheath cells and fewer vacuolated cells, whereas Notch inhibition produces more vacuolated cells (Yamamoto et al., 2010). Live imaging has revealed that these two cell types segregate rapidly along with the inflation of vacuoles within vacuolated cells (Dale and Topczewski, 2011). As vacuole inflation proceeds, vacuolated cells arrange in a stereotypical staircase pattern via a self-organization process that is governed by packing principles that are solely dependent on the number of vacuolated cells that are generated, which in turn determines the size each cell achieves, and the geometry of the notochord rod (Norman et al., 2018).

Further work in zebrafish has shown that the function of vacuolated cells in embryonic axis elongation depends on the generation and maintenance of a single vacuole per cell (Ellis et al., 2013). Notochord vacuoles are lysosome-related organelles formed by biosynthetic trafficking regulated by the Rab32 GTPase (Ellis et al., 2013). Loss or fragmentation of notochord vacuoles leads to shortening of the embryonic axis (Bagwell et al., 2020; Ellis et al., 2013; Norman et al., 2018; Yamamoto et al., 2010) (Fig. 2B). Vacuoles are formed by biosynthetic transport emanating from the Golgi complex and subsequent trafficking and fusion of membranes regulated by the homotypic fusion and vacuole protein sorting (HOPS) complex and Rab32a, whereas vacuole inflation requires the activity of the V-ATPase, which supplies the driving force for fluid accumulation (Ellis et al., 2013). More recently, it has been shown in zebrafish that fusion of large pre-vacuolar compartments with the main vacuole is a rate limiting step in vacuole formation and maintenance, which is regulated by the activity of the Dual serine/threonine and tyrosine protein kinase (Dstyk) protein (Bagwell et al., 2020). dstyk mutants show arrested fusion of Rab32a+ vesicles within the vacuole and present a shorter embryonic axis (Bagwell et al., 2020). This phenotype results from a reduction in vacuolated cell volume and tighter packing of the cells (Bagwell et al., 2020). A separate zebrafish study also implicates Dstyk in the biogenesis of notochord vacuoles and axis formation, presenting in vitro data linking Dstyk to the regulation of Transcription factor EB (Tfeb) (Sun et al., 2020). However, as zebrafish tfeb mutants do not show notochord or axial defects (Meireles et al., 2018), it is unlikely that Dstyk regulates Tfeb in vivo.

Vacuolated cells are unusually large (up to 50 µm in diameter) and have a rigid fluid filled vacuole that occupies most of their cell volume, which presents a structural challenge as the notochord is subject to mechanical stresses. Moreover, when caveolae are lost, vacuolated cells collapse during larval locomotion, and this also impairs axis elongation (Garcia et al., 2017; Lim et al., 2017). However, the effect is not as dramatic as that of vacuole fragmentation because vacuolated cell collapse or injury triggers a regenerative response in which sheath cells invade the affected areas of the notochord and differentiate into new vacuolated cells, thereby restoring notochord function (Garcia et al., 2017; Lopez-Baez et al., 2018). A similar response has been reported for the medaka notochord (Seleit et al., 2020), which is essentially identical structurally to that of zebrafish despite the long distance separating them in the phylogenetic tree.

Notochord vacuoles also play an essential role in straight spine formation in zebrafish (Ellis et al., 2013). Recent work using forward genetics and longitudinal live imaging of notochord and spine development has revealed that vacuole fragmentation causes kinking of the axis during vertebral bone growth (Bagwell et al., 2020). As vertebral bone is assembled around the notochord, the vacuoles initially resist compression thereby facilitating symmetrical centra formation (Fig. 2B). When vacuoles are fragmented prematurely or are lost, the notochord loses internal pressure and becomes more deformable, leading to asymmetrical vertebral bone growth and spine curvature (Bagwell et al., 2020). Interestingly, during late stages of normal centra formation, vacuolated cells re-arrange and stack under the central portion of each centrum, while vacuoles fragment selectively in the intervertebral areas (Bagwell et al., 2020). The stacked vacuolated cells may help to initially resist bone growth, thus facilitating centra formation, while the fragmentation later on may help absorb the compression generated by concentric mineralizing centra (Fig. 2B). The need to accommodate compressive bone growth may also explain the dramatic transition in notochord structure from the hollow lumen present in ascidians to the compartmentalization into vacuoles that is found in vertebrates.

The physical properties of the notochord and its ability to function in axis elongation also depend on the composition and crosslinking of the peri-notochordal ECM sheath (Fig. 1E), recently reviewed by Trapani et al. (2017). This was predicted by ex vivo studies of the Xenopus notochord (Adams et al., 1990) and was demonstrated by the identification of several mutations affecting the notochord sheath in zebrafish (Stemple et al., 1996). Subsequent studies have identified developmental roles for specific matrix components (Christiansen et al., 2009; Corallo et al., 2013; Gansner and Gitlin, 2008; Gray et al., 2014; Parsons et al., 2002), its secretory machinery (Coutinho et al., 2004) and collagen crosslinking (Anderson et al., 2007; Gansner et al., 2007; Mendelsohn et al., 2006), imperative for notochord morphogenesis. These genetic studies in zebrafish have revealed that when components of the ECM sheath surrounding the notochord are missing or mutated, the notochord develops an undulatory pattern (Fig. 3), owing to the inability of the peri-notochordal sheath to contain the internal pressure exerted by the highly vacuolated notochord tissue and by the deposition of cortical bone (Bagwell et al., 2020). Dorsal-ward hyperflexion of the notochord is also observed after disruption of polycystin genes pkd1a/b and pkd2 in zebrafish (Le Corre et al., 2014; Mangos et al., 2010). Loss of Pkd1/2 function leads to upregulation of collagen type-II protein deposition around the notochord, which is normally downregulated in the larval notochord, and increased expression of the endoplasmic reticulum (ER)/Golgi transport coat protein complex proteins (e.g. Sec proteins). Reduction of col2a1a expression or disruption of ER to Golgi transport, via knockdown of sec24d expression or treatment with Brefeldin A rescues the severe dorsal flexion of the notochord and body axis after loss of pkd2 function in zebrafish (Le Corre et al., 2014). Interestingly, loss of Pkd1/2 function does not alter the ultrastructure of the ECM sheath, suggesting that these receptors are aiding in the sensation of the ECM sheath in order to control the important developmental transitions of the notochord (Mangos et al., 2010).

Work in zebrafish has also demonstrated a crucial role for the notochord in the formation of a straight spine. As the notochord acts as the template and directs the segmentation of the spinal column (Wopat et al., 2018), structural defects of the notochord during embryonic development can cause the formation of vertebral malformations and scoliosis in the adult (Ellis et al., 2013; Gray et al., 2014) (Fig. 3). The loss of Collagen, type VIII, alpha 1 (Col8a1) function leads to thinning of the outer ECM sheath, longitudinal live imaging has revealed that this leads to undulation and bending of the notochord, which later on promotes the spatial joining of normally segmented osteoblast populations, resulting in focal regions of vertebral malformations and fusions (Gray et al., 2014) (Fig. 3). Importantly, the structural role of the notochord in spine morphogenesis is distinct from the fish-specific role in vertebral patterning controlled by segmentation of the notochord sheath cells (Lleras Forero et al., 2018; Pogoda et al., 2018; Wopat et al., 2018). However, it remains unclear whether spine defects arising from impaired collagen crosslinking or deposition of the ECM sheath (Anderson et al., 2007) result from alterations of notochord segmentation, surface defects, or both.

In all vertebrates, the notochord is a transient structure, which eventually becomes completely segmented into the nucleus pulposus at the center of the IVD and typically persists throughout life (Choi et al., 2008; Choi and Harfe, 2011; Haga et al., 2009). In amniotes, the IVD is a fibrocartilaginous joint that connects two adjacent vertebrae and provides structural stability, flexibility and cushions axial loading of the spinal column (Cortes and Elliott, 2014), but the mechanical role in teleosts is less clear.

In amniotes, the disc is composed of the nucleus pulposus, a mass of notochord vacuolated cells embedded in a sulfated proteoglycan-rich matrix, which is surrounded by a multi-lamellar annulus fibrosus (Fig. 1F), derived from the sclerotome. The nucleus pulposus forms during embryonic development as vertebral bodies condense and grow into the notochord, eventually obliterating the notochord in those areas and leaving the remaining notochord vacuolated cells fated to become part of the IVD (Choi and Harfe, 2011; Jurand, 1974; Peck et al., 2017; Smith et al., 2011). The IVD is situated adjacent to the cartilaginous endplates of the vertebral bodies, which provide nutritional flux to the IVD (Urban et al., 2004; Wong et al., 2019).

Conversely, in bony fish the IVD is almost exclusively derived from the notochord, which remodels to resemble a nucleus pulposus-like tissue (Fig. 1G). Unlike in mammals, remnants of the notochord remain embedded within the vertebral bodies as large vacuolated cells in the adult spine (Bagwell et al., 2020; Haga et al., 2009) (Fig. 1G). There is no thick multi-lamellar annulus fibrosus in fish; however, there is a thin ligament that wraps circumferentially around and connects adjacent vertebrae (Inohaya et al., 2007), yet may provide a homologous function for containment of the nucleus pulposus.

Recent studies in the conditional Gpr126 knockout mouse model of AIS (Karner et al., 2015) (Box 1) have demonstrated that increased pro-inflammatory signaling and dysregulation of several anabolic and catabolic factors of the growth plate and IVD tissues, coupled with increased mechanical stiffness of the IVD unit, may underlie the pathology of scoliosis in this model (Liu et al., 2019). In the medaka fish, an undefined spontaneous scoliosis mutant has increased hypertrophy of the nucleus pulposus (Irie et al., 2016). These studies provide evidence that changes in the cellular and mechanical properties of the IVDs may drive the onset of scoliosis, particularly during the period of rapid growth, which is a common theme in AIS.

Additional intrinsic elements of the spine are important for spine stability. For example, low bone mineral density (osteopenia) of cortical bone is proposed as causative mechanism in a subset of AIS patients (Cheng et al., 2015). Abnormalities in bone metabolic factors, including increased adiponectin levels in circulation, which may increase bone-resorbing osteoclast activity (Luo et al., 2006), are associated with osteopenia-dependent adolescent IS (Zhang et al., 2019). Defects in the flux of terminal hypertrophic chondrocyte differentiation in the vertebral growth plate, leading to reduced trabecular bone formation in vertebra of the spine, is associated with early-onset scoliosis in mice (Liu et al., 2019). Paradoxically, the complete absence of mineralized bone in entpd5a mutant zebrafish does not affect the typical morphology of the spine (Huitema et al., 2012), suggesting that a vacuolated notochord exerts a stronger role in maintenance of a straight spine during skeletogenesis in teleosts.

A diverse array of ligaments and tendons make multiple insertions and connections with the musculature and bony spinal column. Thus, it is not surprising that connective tissue disorders have been also proposed for causing AIS in humans. For example, rare variants in the in ECM genes FIBRILLIN 1 (FBN1), FBN2 (Buchan et al., 2014), perlecan (HSPG2) (Baschal et al., 2014) and an overall increased polygenic burden of rare variants across ECM gene families have been associated with AIS (Haller et al., 2016). These genetic findings support a model whereby the accumulation of variants in ECM genes, which on their own may display low effect, can collectively synergize to compromise the typical properties or development of axial connective tissues (e.g. ligaments, tendons or annulus fibrosis) and lead to mechanical instability of the spine. Loss of fbn2 function in zebrafish causes defects in the formation of the peri-notochordal ECM sheath and undulations of the notochord (Gansner and Gitlin, 2008). Although these mutants did not survive long enough to assess spine morphology, similar defects of the notochord sheath are well-established to contribute to adult spine defects in zebrafish (Christiansen et al., 2009; Gansner and Gitlin, 2008; Gray et al., 2014) (Fig. 3). A reduction in the basal lamina sheath surrounding the notochord was observed in a global Sox6 (encoding the SRY-box transcription factor 6 protein) knockout mice, which led to defects in the development of the nucleus pulposus and kinks in the tail vertebrae (Smits and Lefebvre, 2003). However, because Sox6 is also expressed in the sclerotome-derived cartilaginous anlage, which give rise to the vertebral bodies and annulus fibrosus, a conditional Sox6 knockout approach specific to the notochord is needed to more rigorously assess the role of a reduced notochord sheath for axial stability in mouse. We refer the reader to an additional review of the role of the cartilage matrisome for scoliosis (Wise et al., 2020).

The neural arches of each vertebral body collectively form a cavity called the spinal canal, which encases and protects the spinal cord (Fig. 1F,G). The nerve roots of the peripheral nervous system extend out from the spinal cord at distinct points along the spine and, together with the spinal cord, serve to integrate sensory information and motor control with the brain. The stability of the spine is vitally dependent on the integration and function of the neuromuscular system. For example, muscle disorders (such as Duchenne Muscular Dystrophy and spinal muscular atrophy), neurological disorders (such as Rett syndrome, Parkinson's disease) and pediatric spinal cord injuries (Mary et al., 2018; Mulcahey et al., 2013) are all associated with scoliosis in humans (Murphy and Mooney, 2019). Recent studies in mouse have shown an essential role for the proprioceptive system for spine stability (Blecher et al., 2017).

The brain and spinal cord are chemically connected via the ventricular and central canal system, which allows for the exchange of CSF secreted from the choroid plexus in the brain with the spinal cord. In all vertebrates, multiple physiological processes control the flow of CSF including the rates of CSF production, drainage out to the circulatory system, cardiopulmonary function and body movements (Fame and Lehtinen, 2020; Olstad et al., 2019). The bulk flow of CSF provides nutrients and helps to remove waste products from the central nervous system, but also provides key signaling molecules during development of these tissues, including maintenance of adult neural stem cells in the brain (Petrik et al., 2018). Disruption of normal CSF circulation and its turnover is believed to contribute to the development of many neurodegenerative conditions, such as Alzheimer's disease, and neuroinflammatory conditions, such as multiple sclerosis (Simon and Iliff, 2016).

In zebrafish, compartmentalized flows of CSF within the brain ventricles and central canal of the spinal cord require the polarized beating of motile cilia extending from the apical surface of ependymal cells which line the ventricular and spinal canal (Olstad et al., 2019; Thouvenin et al., 2020) (Fig. 4A). Several genetic models in both mouse and zebrafish that disrupt motile cilia or ependymal cell development display defects of CSF flow and hydrocephalus (Lee, 2013; Ringers et al., 2020), showing that, in addition to bulk CSF flows, motile-cilia-driven near-wall or laminar CSF flows (Spassky and Meunier, 2017) are important for homeostasis of the nervous system. Primary ciliary dyskinesia (PCD) patients commonly display pleiotropic defects in tissues, which require motile cilia function. However, hydrocephalus is rarely observed in human PCD patients (Vieira et al., 2012), suggesting that the role of motile ependymal cell cilia is not imperative for function of the human nervous system. However, recent human genetic studies suggest that the disruption of master transcriptional regulators of motile cilia development (Wallmeier et al., 2019) or of specific structural components (Morimoto et al., 2019) of the motile cilia are associated with hydrocephalus in humans. This data reinvigorates the need for understanding the role of motile cilia for the development and homeostasis of the ventricular system in humans.

Recent studies in zebrafish have uncovered a curious association between loss of ependymal motile cilia function, reduced CSF flow and scoliosis (Grimes et al., 2016; Konjikusic et al., 2018). An apparent consequence of disruption of ependymal motile cilia function in the protein tyrosine kinase 7a (ptk7a) mutant zebrafish is increased neuroinflammation in the brain and spinal cord. Moreover, stimulation of proinflammatory cues in the ependyma was sufficient to stimulate scoliosis in zebrafish, and scoliosis in zebrafish can be ameliorated after treatment with nonsteroidal anti-inflammatory drugs (Rose et al., 2020; Van Gennip et al., 2018). The coincidence of increased proinflammatory signals and scoliosis is also observed in stat3 mutant zebrafish (Liu et al., 2019), and scoliosis is observed in a rare primary immunodeficiency disease called Hyper IgE syndrome, which is attributed to STAT3 mutations in humans (Sowerwine et al., 2012). Obstruction of CSF flow in humans with syringomyelia, observed as cysts or cavity formation within the spinal cord, is strongly correlated with syndromic scoliosis (Strahle et al., 2015). Altogether these studies underscore the role of CSF physiology and neuron inflammation for the maintenance of spine stability.

Recently, a CSF resident structure called the Reissner fiber has become the focus for understanding the regulation of the straight body axis and spine morphogenesis (Driever, 2018). Since its discovery over a century ago (Reissner, 1860), the Reissner fiber has been shown to be an ancient conserved structure, which has been identified in ascidian tadpoles (Olsson, 1972) and is present in numerous chordate lineages, from cephalochordates (Gobron et al., 1999) to lamprey (Reissner, 1860), teleosts (Cantaut-Belarif et al., 2018), amniotes (Vio et al., 2008) and several primates (Castenholz and Zoltzer, 1980).

In zebrafish, the Reissner fiber is an extracellular glycoprotein thread, which extends from the brain through the central canal and terminates at the base of the spinal cord (Troutwine et al., 2020) (Fig. 4B). In zebrafish, the assembly of the Reissner fiber is dependent on the secretion of the SCO-spondin protein from the subcomissural organ (SCO) in the brain, as Reissner fiber formation is completely absent in scospondin mutant zebrafish (Cantaut-Belarif et al., 2018). The Reissner fiber has long been postulated to continually move from head to tail, which was experimentally demonstrated using pulse-chase labeling of the fiber with labeled monoamines in rat (Caprile et al., 2003). Live imaging of an endogenous knock-in scospondin-GFP^ut24 gene fusion zebrafish line has shown that the assembly of the Reissner fiber first extends out from the subcomissural organ in the brain at early tailbud stage (∼20 h post fertilization) and travels towards the base of the spinal cord where it terminates at the terminal ampulla (Fig. 4B′) and that the fiber continually treadmills in a rostral-caudal direction throughout larval development and remains assembled in adult zebrafish (Troutwine et al., 2020).

A variety of hypotheses have proposed functions for the Reissner fiber, including regulation of CSF production and composition, circulation or detoxification. Early studies demonstrate that the Reissner fiber transports key signaling molecules, which may be important for the regulation of neurodevelopment and contribute to straightening of the body axis, from the brain to the spinal cord. For example, binding of the Reissner fiber to monoamines and catecholamines (e.g. L-DOPA, noradrenaline, adrenaline, serotonin) may provide a source-sink mechanism for transport and removal of these important neuronal signaling molecules from the brain and spinal canal (Caprile et al., 2003). Indeed, soluble SCO-spondin can also act as regulator of neurite outgrowth, neurogenesis and neurodifferentiation in chicken neural explant cultures (Gobron et al., 2000; Vera et al., 2015; Vera et al., 2013).

A class of sensory neurons commonly called cerebrospinal contacting neurons (CSF-cNs) are a specialized GABAergic sensory neuron found in most vertebrates that, in zebrafish, line the dorsal and ventral sides of the central canal (Djenoune et al., 2017) (Fig. 4). CSF-cNs extend out into the central canal and respond to lateral and longitudinal bending of the spine through their modulation of distinct premotor circuits (Djenoune et al., 2017; Djenoune et al., 2014; Orts-Del'Immagine et al., 2020). This lead Kolmer to hypothesize more than a century ago that the SCO, the Reissner fiber and its interaction with central canal components, acts as a ‘sagittal organ’, which functions as an integrated mechanosensory system of the body (Kolmer, 1921). In support of Kolmer's hypothesis, recent studies in zebrafish have shown that CSF-cN monocilia are in direct contact with the Reissner fiber and that the loss of the Reissner fiber results in a severe reduction of CSF-cN ability to respond to mechanical stimulus, such as axial bending (Orts-Del'Immagine et al., 2020). These data suggest an intriguing role for the Reissner fiber acting as a physical cue of the axial midline that can help signals to the CSF-cNs and help refine sensory-motor control of the body axis.

At a molecular level, CSF-cNs express mechanosensitive channel Polycystic kidney disease1-like 1 (Pkd2l1) localized within the apical microvilli extensions that interact with CSF (Bohm et al., 2016). pdk2l1 mutant CSF-cNs display a marked reduction of spontaneous calcium activity in vivo or in response to mechanical stimulus in vitro, and adult pkd2l1 mutant fish display an exacerbated kyphosis of the spine (Sternberg et al., 2018). Genetic loss of motile cilia activity or the Reissner fiber also reduces the spontaneous activity (assayed by calcium imaging) of CSF-cNs without affecting Pkd2l1 channel activity (Orts-Del'Immagine et al., 2020; Sternberg et al., 2018), suggesting that physical interaction of the Reissner fiber and the CSF-cNs is required to stimulate Pkd2l1 channels.

CSF-cNs also express the urotensin II-related neuropeptides Urp1/2 (Quan et al., 2015). Motile-cilia mutant zebrafish with severe ventral curvatures show diminished expression of urp1/2 and the ectopic expression of Urp1 is sufficient to straighten the body axis in these animals or lead to dorsal ward flexion in wild-type animals (Zhang et al., 2018). A variety of monoamines and catecholamines, such as dopamine, epinephrine or L-DOPA, which have been shown to bind the Reissner fiber in rat (Caprile et al., 2003), are sufficient to stimulate urp1/2 expression in the CSF-cNs of motile cilia mutant zebrafish and rescue body axis morphogenesis defects in zebrafish (Lu et al., 2020; Zhang et al., 2018). The putative Urp1 receptor Uts2ra (Uts2r3) is expressed in the dorsal slow-twitch muscle fibers in zebrafish and uts2ra mutant zebrafish display larval-onset scoliosis, without vertebral malformations (Zhang et al., 2018). As mentioned above, severe dorsal flexion in Pkd1/2 loss-of-function zebrafish models is also observed as a consequence of increased ectopic secretion of collagen type-II protein in the larval notochord (Le Corre et al., 2014; Mangos et al., 2010). However, the genetic ablation of the CSF-cN mechanosensitive receptor Pdk2l1 gene only results in mild kyphoscoliosis in adult viable zebrafish (Sternberg et al., 2018). It remains to be determined how severe dorsal flexion phenotypes related to ectopic secretion of embryonic collagens in the notochord and alterations in Urp regulation of dorsal musculature are related.

In addition to its role in straightening the body axis during embryonic development, recent data show the Reissner fiber is continuously required for maintenance of a straight body axis and spine morphogenesis in larval zebrafish. In contrast to the complete loss-of-function of scospondin, recent studies show that hypomorphic recessive (Troutwine et al., 2020) or a dominant heterozygous (Rose et al., 2020) mutations in scospondin are not associated with defects of Reissner fiber assembly or body axis straightening during embryonic development. Instead, these mutants display progressive disassembly of the Reissner fiber during larval development, concurrent with the onset of axial curvatures in zebrafish. The onset of axial curvatures in these larval fish leads to scoliosis without vertebral malformations in the adult mutant fish (Troutwine et al., 2020) (Fig. 3). How might this maintenance be achieved? One hypothesis is that CSF-cNs synapse onto glutamatergic pre-motor interneurons to modulate spinal pre-motor circuits (Fidelin et al., 2015), which may facilitate neuromuscular stability of the body axis and subsequently maintain straightness of the body axis, and which indirectly controls the straightness of the notochord, which is the direct template for embryonic spine development (Wopat et al., 2018) (Fig. 1C). Interestingly, deficiencies in postural and balance control on the basis of proprioceptive cues are associated with a subset of adolescent IS patients (Cheng et al., 2015; Guo et al., 2006). Another, potentially non-exclusive, model supported by recent studies in zebrafish suggests that the loss of Reissner fiber leads to dysregulation of Urp signaling, leading to overexpression (Vesque et al., 2019 preprint) or underexpression (Lu et al., 2020; Zhang et al., 2018) of urotensin peptide from the CSF-cNs. This imbalance results in body axis instability from defects of the dorsal musculature during spine morphogenesis (Lu et al., 2020; Zhang et al., 2018). Finally, the loss of Reissner fiber has been associated with increased neuroinflammation of the brain and spinal cord (Rose et al., 2020); however, whether the loss of Reissner fiber assembly is stimulating or is a consequence of increased neuroinflammatory signaling remains undetermined. Together, these data suggest several complementary modes of action downstream of the Reissner fiber and CSF-cNs for spine morphogenesis: (1) as an instructive signaling modality involving direct stimulation of the CSF-cN activity by physical interactions with the Reissner fiber for downstream sensory motor modulation of neuromuscular muscle tone; and (2) to promote dorsal muscle contraction to facilitate straightening of the body axis, which has been recently reviewed (Grimes, 2019; Ringers and Jurisch-Yaksi, 2020).

The Reissner fiber and CSF-cNs are found in hundreds of chordate lineages (Djenoune and Wyart, 2017), including primates (Castenholz and Zoltzer, 1980; Djenoune et al., 2014). The extent to which the Reissner fiber instructs human development remains unclear. The SCO, which secretes the Reissner fiber in the brain, is regressive during postnatal development in human (Rodriguez et al., 2001). Although the Reissner fiber was historically observed in humans (Agduhr, 1922), neither the SCO nor the Reissner fiber are observed to immuno-react with any of the numerous polyclonal and monoclonal antibodies raised against Reissner fiber glycoproteins of animal origin in human or in several anthropoid apes (Rodriguez et al., 1990; Rodríguez et al., 1993; Rodriguez et al., 2001). The current Human Genome Organization Gene Nomenclature Committee now classifies the human SCO-spondin (HGNC:21998) as a pseudogene, a consequence of deleterious mutations common in the population. In addition, we can find no evidence of CSF-cNs being reported in human spinal cord. Altogether, this may suggest that anthropoid apes and humans represent a unique primate lineage that lost the ability to form or sustain the formation of Reissner fibers and utilize other mechanisms to maintain postural control. It is interesting to note that primates (Castenholz and Zoltzer, 1980) and other animals which have a Reissner fiber all display a relatively straight spinal column (Fig. 1B). In contrast, typical human spine morphology follows a natural cervical kyphosis and lumbar lordosis (Fig. 1A). Perhaps the loss of Reissner fiber underscores the establishment of natural spine morphology in humans. It also remains an open question as to the function of the Reissner fiber and CSF-cNs in experimental mammalian models that are known to display both components: it will be interesting to define how physiological control of posture has been innovated or coopted in humans.

Recent progress in our understanding of the multiple processes involved in embryonic axis elongation and straightening, spine development and scoliosis in a variety of model systems may be summarized in the following three emerging concepts. First, the notochord plays a key role as organizing center and template for the spine. Early defects affecting the structural integrity of the notochord can translate to severe notochord curves, leading to vertebral malformations with scoliosis (Fig. 3). Perhaps if the notochord is only slightly curved, as a function of neuromuscular defects of the body axis during larval development, then this may give rise to scoliosis without vertebral malformations akin to pathology commonly observed in IS or neuromuscular/syndromic scoliosis and in zebrafish models outlined in this review (Fig. 3). Although these principles have mostly been defined in zebrafish and other swimming larvae species, it is less clear whether this principle has a strong role in mammalian development because notochord and early spine development are difficult to study in these models. Second, notochord vacuoles play a key structural role in the vertebrate axis. As vacuoles fill with fluid and vacuolated cells expand within the casing of the sheath, the notochord becomes a pressurized rod that can provide a hydrostatic scaffold for embryonic axis elongation. Then, as vertebral bone grows it facilitates symmetrical centra formation by first resisting and then absorbing compression. Loss of vacuole integrity can lead to the formation of asymmetrical vertebrae that can cause CS-phenotypes in zebrafish and perhaps also in humans, although trabecular bone formation in the vertebrae may help to overcome mechanical defects of the notochord in these animals. Finally, the Reissner fiber has a crucial role in straightening the zebrafish body axis during embryogenesis and is required for the maintenance of a straight body axis during larval development to facilitate normal spine morphology. The role of the Reissner fiber for spine morphogenesis or nervous system physiology in mammals is unknown and warrants further study. Although it is unclear how the Reissner fiber might be related to scoliosis observed in IS or neuromuscular/syndromic scoliosis phenotypes, it is clearly linked to CSF sensing and influences neuromuscular postural control, which in turn can cause bending of the axis if it is inappropriately regulated. An open question in the field remains to understand how loss of motile cilia function in the central canal aids in the assembly of the Reissner fiber and how this translates to other amniote models.

We thank Jennifer Bagwell for comments on the manuscript and for the illustrations of Fig. 2.