ABSTRACT
Sensory neurones display organizational features that are common to most populations of neurones in the vertebrate nervous system. First, their cell bodies are arranged in discrete groups (the sensory ganglia) each of which has characteristic receptive and projection fields (the peripheral and central target fields, respectively, to which the peripheral and central processes of the neurones grow in development). Second, nerve fibres make specific terminations within each target field; an important feature is that different kinds of sensory receptors in the periphery are connected to the appropriate kinds of second-order neurones in the central nervous system (CNS). Third, nerve fibre terminations in the peripheral and central target fields have a similar topographic order.
Introduction
To understand how these three facets of sensory neurone connections are established in development it is necessary to ascertain the following. First, the means by which sensory nerve fibres are guided to their specific target fields. Second, the factors in the peripheral and central target fields which selectively maintain the appropriate numbers of nerve fibre terminals belonging to each of the various functionally distinct classes of sensory neurones (nociceptive, proprioceptive, etc) and ensure that these terminate in relation to the correct cell types and neurones. Third, the arrangement of nerve fibres in developing sensory nerves to determine how somatotopic sensory projections are established.
Several characteristics of sensory neurones have facilitated investigation of these fundamental aspects of neural development.
(1) Sensory ganglia can be dissected from embryos from the earliest stages of their formation. This permits in vitro study of the factors that influence the growth and guidance of early sensory axons and the factors that regulate neuronal survival and maturation at later stages. Furthermore, the segregation of functionally distinct kinds of sensory neurones within certain ganglia permits investigation of the specific regulatory influences on the development of these neurones in relation to the kinds of sensory receptors and second-order neurones they innervate.
(2) The target fields of certain sensory ganglia are clearly circumscribed and can be dissected from embryos at stages prior to and after innervation. This permits investigation of the influence of virgin target territory on the growth and guidance of early sensory axons in culture. It is also possible to measure changes in the levels of regulatory molecules such as neurotrophic factors in the target field at stages throughout the phase of innervation and to study how the synthesis of these factors is controlled.
(3) The trigeminal system of rodents is one of the few cases in which developing topographic projections can be easily studied since the characteristic pattern of whisker follicles on the snout is replicated in the CNS by corresponding arrays of multineuronal units. Because the receptive field of this neural projection is situated in the skin, it is particularly accessible to experimental manipulation.
Guidance of sensory nerve fibres to their target fields
Mechanisms of axonal guidance in the vertebrate nervous system
The axons of almost all populations of neurones in the developing vertebrate nervous system grow with considerable precision to reach their target fields (Lance-Jones & Landmesser, 1981a; Scott, 1982; Jacobson & Huang, 1985). These axons may be guided in two ways: by the tissues lying on route to their targets or by target-derived chemotropic factors.
It is possible that the tissues lying between neurones and their targets may either provide preformed pathways which nonspecifically channel growing axons through a particular region or contain specific guidance cues to which certain axons selectively respond. Evidence consistent with the existence of preformed pathways is the finding that the growth of axons along stereotyped routes in certain parts of the embryo is preceded by the development of either mechanical guides such as aligned intercellular spaces (Singer, Nordlander & Egar, 1979) or tracts of molecules which provide a favourable substratum for axon growth such as the neural cell adhesion molecule (Silver & Rutishauser, 1984) and laminin (Cohen, Burne, Winter & Bartlett, 1986; Riggott & Moody, 1987). Evidence for the selective response of axons to specific guidance cues present in the tissues through which they grow comes from several neural systems. When the motoneurones that innervate the chick hindlimb bud are experimentally displaced by several spinal segments prior to axon outgrowth, their axons pursue a variety of aberrant routes to reach and innervate the correct muscles (Lance-Jones & Landmesser, 1981b). The finding that the axons of displaced motoneurones sort out predominantly within the nerve plexus region adjoining the spinal cord and that ablation of tissues distal to the plexus does not affect sorting (Tosney & Landmesser, 1984) suggests that motoneurones are specified with respect to their targets and that they respond to specific guidance cues which reside within the vicinity of the plexus. Although the nature of the guidance cues for motor axons in the limb bud is an enigma, in certain other situations the molecular basis of specific guidance cues lying on route to the target is clearer. There is some evidence that pre-existing axon bundles may influence the growth of later-generated axons, for example, á group of axons in the embryonic fish spinal cord selectively fasciculates with a specific subset of early axons (Kuwada, 1986). More extensive work on this mechanism of guidance in invertebrates has shown that some axons make a series of precise choices in moving from one subset of axons to another (Goodman et al. 1984) and recent work suggests that this phenomenon of selective fasciculation may be mediated by specific cell-surface glycoproteins (Bastiani, Harrelson, Snow & Goodman, 1987).
The most convincing in vivo evidence for chemotropic guidance of developing axons comes from a study of axon outgrowth from optic primordia transplanted to a variety of ectopic locations in the head of Xenopus embryos (Harris, 1986). Irrespective of the location of the transplant site, axons grew directly towards their target field (the optic tectum). Although the most parsimonious explanation for these findings is that the axons are responding to a concentration gradient of a specific chemotropic factor diffusing from the tectum, it has also been suggested that these findings are consistent with navigation with respect to ‘positional information’. That is, growth cones are able to ascertain their location in a coordinate system in the embryo by detecting ‘positional information’ (possibly in the form of stable gradients of substances) and navigate to their assigned coordinates accordingly. However, it has yet to be ascertained whether the growth cone possesses the necessary molecular complexity for detecting and integrating ‘positional information’ in three dimensions and responding to it to attain a specified positional value (which would additionally require integration in time since a positional value by itself would not provide the growth cone with any information about the direction in which it should move to attain a desired value unless previous values are known).
The above broad categories of axonal guidance should not be viewed as alternatives. Apart from the evidence that each may operate in different parts of the nervous system, it is possible that a given set of axons may be guided to their targets by more than one mechanism. In the following sections, I shall consider evidence for the involvement of tissues on route to the target field and target-derived chemotropic factors in guiding sensory nerve fibres in development. As almost all of the experimental work on this aspect of sensory neurone development has addressed the peripheral fibres of these neurones, I shall restrict my discussion to sensory axon guidance in the peripheral nervous system.
Sensory axon guidance by intervening tissues
Studies of the innervation of chick limb buds have provided evidence that both sensory and motor nerve fibres are constrained to grow within a system of nonspecific, preformed or permissive pathways which map out the characteristic mixed nerve plexus and its branches at the base of the developing limb. When sensory and motor neurones are experimentally displaced before their axons grow into the region of mesenchyme in which the plexus forms, they trace out a normal plexus even though they may innervate inappropriate targets or reach their normal targets by a variety of novel routes within the plexus (Lance-Jones & Landmesser, 1981b; Honig, Lance-Jones & Landmesser, 1986). Although it is not clear what determines the morphology of this plexus and its branches, it is possible that growing axons may be constrained, in part, by regions which are nonper-missive for growth cone advancement since nerve fibres branch abnormally from the plexus at the site of an experimentally produced deficiency in the juxtaposed precartilaginous pelvic anlage (Tosney & Landmesser, 1984). A recent immunohistochemical study of the distribution of N-CAM in the early limb bud has shown that it is present within and surrounding the presumptive nerve plexus (Tosney, Watanabe, Landmesser & Rutishauser, 1986). Although the distribution of this molecule was not consistent with the precise delineation of pathway boundaries, this does not rule out the possibility that the expression of specific molecular forms of N-CAM within this region may conform more closely to the topography of the presumptive plexus. Laminin immunoreactivity has also been observed in association with early spinal sensory and motor nerve fibres (Rogers, Edson, Letourneau & McLoon, 1986), however, it is unclear whether it governs limb plexus boundaries since it has not been convincingly demonstrated in advance of the growth cones in this region.
Evidence for the involvement of preformed pathways of laminin in guiding sensory nerve fibres is provided by recent immunohistochemical studies of its distribution in the mesenchyme through which trigeminal nerve fibres grow to reach their targets. A punctate array of laminin extending for a short distance from the trigeminal ganglion into the centre of the mandibular process just prior to sensory axon outgrowth has been observed in the chick embryo (Riggott & Moody, 1987), and a similar distribution of laminin is also present in the mandibular and maxillary processes of the mouse embryo at this early stage (A. G. S. Lumsden & J. Cohen, unpublished findings). The possibility that these circumscribed arrays of laminin are important in channelling the growth of early sensory axons is suggested by the finding that laminin is a very effective and preferred substratum for neurite growth in vitro (Sanes, 1985).
Evidence for the selective response of sensory nerve fibres to guidance cues provided by motor axons lying on route to their targets comes from recent studies of the effects of either ablating or removing the ventral part of the neural tube prior to motor axon outgrowth into the fore- or hindlimb buds of chick embryos (Swanson & Lewis, 1986; Landmesser & Honig, 1986). In these embryos, muscle nerves generally failed to form or had a greatly reduced diameter and cutaneous nerves were enlarged. These findings suggest that in the absence of motor axons, proprioceptive nerve fibres are unable to project to muscle but grow along cutaneous nerves instead, implying that proprioceptive nerve fibres selectively respond to guidance cues provided by motor axons. The possibility that these take the form of a selective affinity of proprioceptive nerve fibres (as opposed to cutaneous sensory nerve fibres) for fasciculating with motor axons is supported by the finding that whereas proprioceptive neurites and motor neurites mingle freely in culture, cutaneous sensory neurites and motor neurites tend to remain segregated (A.M.D., unpublished findings).
Although the above studies suggest that cutaneous sensory nerve fibres can grow into the developing limb bud and establish a fairly normal nerve plexus and pattern of major cutaneous nerves in the absence of motor axons, there is some evidence that motor axons may influence the routes taken by individual cutaneous sensory nerve fibres within the nerve plexus. When the rostrocaudal order of a short stretch of DRG is reversed by rotating the corresponding neural crest (Scott, 1986), the cutaneous innervation patterns they establish are dependent on whether or not the underlying ventral neural tube (containing motoneurone precursors) is also rotated. When both the neural crest and neural tube are rotated, the DRG, like the motoneurones of the rotated spinal segments, tend to establish innervation patterns consonant with their original position in the embryo. However, when the neural crest is rotated alone, the DRG tend to establish innervation patterns in accordance with their new position in the embryo. These findings suggest that cutaneous sensory nerve fibres grow in accordance with the innervation pattern of motor axons in their vicinity. Although it is possible that these findings may also result from the neural tube influencing the specificity of DRG to project along a particular pathway, the likelihood that motoneurones influence sorting of cutaneous sensory nerve fibres into the appropriate nerves is also supported by the finding that sensory nerve fibres from different segments cross less extensively in the plexus region when the regional motoneurones are removed (Landmesser & Honig, 1986). Since the segmental contribution of DRGs to cutaneous nerves and the resulting dermatome patterns were not determined in either this study or the similar study of Swanson and Lewis, it is not known whether cutaneous sensory nerve fibres sort out appropriately in the plexus region in the absence of motor axons.
The above studies raise the possibility that much of the sorting behaviour of sensory and motor axons in the plexus region is due to selective fasciculation of related subsets of nerve fibres. Cinemicrography of interactions between sensory and motor axons of various segmental levels in culture may throw light on this issue.
Sensory axon guidance by target-derived chemotropic factors
Several in vivo and in vitro studies of the effects of very high concentrations of nerve growth factor (NGF) on the growth of late-embryonic or newborn sympathetic and sensory nerve fibres have led to the widely held view that NGF attracts developing sympathetic and sensory nerve fibres to their target fields by chemotropism (Levi-Montalcini, 1982). Intracranial injection of neonatal rodents with NGF results in the growth of sympathetic fibres into the spinal cord and along the dorsal columns toward the injection site (Menesini-Chen, Chen & Levi-Montalcini,1978) . Neurite outgrowth from DRG explants is more profuse on the side facing a source of NGF (Chari-wood, Lamont & Banks, 1972; Ebendal & Jacobson, 1977). Sensory neurites in culture show a preferred orientation up a concentration gradient of NGF (Letourneau, 1978) and turn toward a source of NGF within their immediate vicinity (Gundersen & Barrett, 1980).
Quite apart from the implausibility of concentration gradients of a single molecule accounting for the multiple specific routes taken by different sets of sympathetic and sensory nerve fibres to their widely dispersed target fields in the embryo, it has recently been shown that NGF synthesis in developing cutaneous target fields does not occur prior to the arrival of sensory nerve fibres and that these fibres do not express NGF receptors until they reach their targets (Davies et al. 1987a). Thus, the view that NGF attracts nerve fibres to their target fields in development must be abandoned.
The in vivo experimental findings thought to indicate a chemotropic response of nerve fibres to NGF can be attributed to the known trophic effects of this molecule. Sympathetic nerve fibres sprout in response to locally elevated levels of NGF in vivo (Springer & Loy, 1985) and NGF is able to preferentially maintain the branches of sympathetic neurites selectively exposed to NGF in vitro (Campenot, 1982a,b). Thus, it is likely that the growth of sympathetic nerve fibres into the spinal cord after intracranial injection of NGF is due to local sprouting of nerve fibres and the selective maintenance of those branches which by chance grow in the direction of the higher concentration of NGF. Likewise, the increased growth of neurites from the side of a DRG explant facing a source of NGF can be attributed to the higher concentration of NGF promoting the survival and growth of larger numbers of neurones in this part of the explant. It is clear that individual neurites do not show a directional response to NGF since neurites grow radially from the entire perimeter of these explants. The turning response of neurites to a source of NGF can be attributed to the physical properties of the NGF molecule at the massively high concentrations used (Gundersen & Barrett used more than 25 000 times the peak concentration of NGF reached in the most-densely innervated cutaneous target field of the mouse embryo during development, Davies et al. 1978a). At such high concentrations NGF increases the adhesion of neurites to the culture substratum (Schubert & Whitlock, 1977; Gundersen & Barrett, 1980). Since neurites move preferentially from less-to more-adhesive substrata (Letourneau, 1975) it is likely that the turning response of neurites to NGF is simply due to the influence of an unphysiological adhesive gradient.
A useful in vitro experimental method for investigating whether developing nerve fibres are guided by specific target-derived attractants is explant coculture of neural tissue and target organs. Most of the studies employing this method have been conducted late in development using regenerating neurones and denervated target organs. For example, neonatal rat sympathetic ganglia cocultured with vas deferens or atrium (Chamley & Dowel, 1975) and various mid-embryonic chick sensory, sympathetic and parasympathetic ganglia cocultured with a variety of tissues (Ebendal & Jacobson, 1977). Although neurite outgrowth in these studies was more marked on the side of the ganglia facing the cocultured tissue, it was neither confined to this location nor exclusively orientated towards the tissue but radiated from the entire perimeter of the ganglia. This pattern of neurite outgrowth is similar to that observed from ganglia facing a source of NGF, and is indicative of a concentration-dependent growth response to a neurotrophic factor diffusing from the cocultured tissue. It has since been shown that NGF synthesis in target organs begins with their innervation (Davies et al. 1987a) and increases after denervation (Barth, Korsching & Thoenen, 1984; Heumann & Thoenen, 1986) and that neurones become responsive to NGF after innervating their targets (Davies & Lumsden,1984). If explant coculture is to provide meaningful information on the mechanism of axon guidance it must be done at the stage when nerve fibres are growing to their targets.
The trigeminal ganglion and its cutaneous target field are two well-defined components of the developing sensory nervous system which can be dissected from mouse embryos before the earliest trigeminal nerve fibres have reached their targets. When cocultured at this stage in a collagen gel at a similar spacing as in vivo (Fig. 1), neurites emerge exclusively from the side of the trigeminal ganglion facing the target field (either the maxillary or mandibular process of the first branchial arch) and grow directly towards this tissue (Lumsden & Davies, 1983, 1986). Explants of other cutaneous target fields, including the hyoid process of the second branchial arch which adjoins the trigeminal territory, neither elicit nor direct neurite outgrowth from trigeminal ganglion explants at this stage. The early trigeminal target field has no noticeable influence on neurite outgrowth from explants of other sensory ganglia including the adjacent geniculate ganglion. These findings suggest that the early trigeminal cutaneous target field produces a diffusible factor which specifically directs the growth of early trigeminal nerve fibres. The finding that neurite outgrowth from early trigeminal ganglia is consistently directed towards target field epithelium not mesenchyme when cocultured with these isolated tissue layers suggests that this factor is of epithelial origin.
Head of a 10-day (ElO) mouse embryo showing the trigeminal ganglion (T), geniculate ganglion (G), maxillary (Mx), mandibular (Md) and hyoid processes (H) and the forelimb bud (LB). The arrangement of Mx, TG and LB explants for collagen gel cultures is shown to scale. Below this is a cartoon summary of the characteristic pattern of neurite outgrowth after 48 h incubation in cultures established prior to target field innervation (ElO) and after target field innervation (E12), in cultures supplemented with and without anti-NGF antiserum. For further details see Lumsden & Davies (1983, 1986).
Head of a 10-day (ElO) mouse embryo showing the trigeminal ganglion (T), geniculate ganglion (G), maxillary (Mx), mandibular (Md) and hyoid processes (H) and the forelimb bud (LB). The arrangement of Mx, TG and LB explants for collagen gel cultures is shown to scale. Below this is a cartoon summary of the characteristic pattern of neurite outgrowth after 48 h incubation in cultures established prior to target field innervation (ElO) and after target field innervation (E12), in cultures supplemented with and without anti-NGF antiserum. For further details see Lumsden & Davies (1983, 1986).
A number of observations suggests that the epithelium-derived factor directs the growth of early trigeminal nerve fibres by a chemotropic mechanism. First, the great majority of neurites in these cultures grow directly to the target field from the outset, and those whose initial course is not directly in line with the target generally turn towards it after growing for a short distance. Second, neurite outgrowth from two ganglia grown at different distances from a single maxillary process is exclusively target-directed from both ganglia in the great majority of cases. This indicates that the factor effectively diffuses beyond the nearest ganglion, yet does not elicit neurite outgrowth from the far side of this ganglion as might be expected of a factor with a solely trophic effect.
It is possible that the trigeminal chemotropic factor (or another factor produced by the early trigeminal epithelium) also has a neurite-promoting action since no neurites grow from early trigeminal ganglia grown alone in a collagen gel. Neither the chemotropic nor the neurite-promoting effects are due to laminin. The same specific pattern of neurite outgrowth is observed in cocultures grown in collagen gels containing high levels of laminin, and medium conditioned by early maxillary processes retains its specific neurite-promoting effects when depleted of laminin (Lumsden & Davies, 1987).
It is important to note that when the trigeminal ganglion is cocultured with its cutaneous target field or other cutaneous target fields later in development (when innervation is well underway) neurites grow radially from the entire perimeter of the ganglion. Unlike the specific, target-directed neurite outgrowth observed in earlier cocultures, this undirected outgrowth is completely abolished by anti-NGF antiserum, indicating the production of NGF by these older, denervated target tissues. This not only demonstrates that the trigeminal chemotropic factor is immunochemically distinct from NGF but also highlights the importance of investigating the mechanism of axon guidance with tissues at the correct stage of development.
The above studies of the influence of the trigeminal target field on the growth of early trigeminal nerve fibres together with the previously discussed studies of the distribution of laminin in the vicinity of the early trigeminal ganglion suggest that two distinct mechanisms are involved in guiding trigeminal nerve fibres in development: stereotopic guidance by preformed pathways of laminin and chemotropic guidance by a specific target-derived factor. It is likely that these mechanisms play complementary roles. The short tracts of laminin may constrain the growth of early nerve fibres into distinct nerve trunks in the proximal part of their course, ensuring that they do not radiate directly from the ganglion to their target epithelium. The target-derived chemotropic factor governs the direction in which the fibres grow within the constraints imposed by the laminin.
Although the above experimental studies suggest that the trigeminal epithelium is specified to attract its innervation by the production and release of a specific chemotropic factor, it is unlikely that there is a distinct factor for each of the many cutaneous target fields in the embryo. It is possible, however, that after the cutaneous sensory nerve fibres of DRG have been guided to the correct general region by the various influences which constrain or direct their growth on route, they may be attracted to the adjacent skin by a non-specific chemotropic factor produced generally by cutaneous epithelium.
Regulation of sensory neurone survival by neurotrophic factors
Neuronal death and the role of neurotrophic factors
In the developing vertebrate nervous system, almost all groups of neurones are subject to a phase of cell death which occurs shortly after they begin to innervate their target fields (Oppenheim, 1981). The percentage of neurones that die and the length of time over which cell death occurs is characteristic for each group. In the case of sensory neurones, the magnitude of cell death ranges from 25 % in the vestibular and cochlear ganglia (Ard & Morest, 1984) to over 75 % in the trigeminal mesencephalic nucleus (Rogers & Cowan, 1974) and generally takes place over a period of between 3 and 6 days (Hamburger & Levi-Montalcini, 1949; Carr & Simpson, 1978a; Hamburger, Brunso-Bechtold & Yip, 1981; Davies & Lumsden, 1984).
Studies of the effects of manipulating the size of various target fields prior to their innervation have demonstrated the importance of the target field in regulating neuronal survival (for recent reviews see Oppenheim, 1981; Cowan, Fawcett, O’Leary & Stanfield, 1984). Whereas partial or total removal of a target field causes a proportionate increase in the number of innervating neurones that die during the phase of cell death, target field enlargement by grafting additional target tissue reduces the magnitude of cell death. Likewise, removal of part of a group of neurones prior to its innervation of a normal target field reduces the magnitude of cell death among the remaining neurones. These findings suggest that developing neurones become dependent on their target fields for survival and that a given target field is able to support the survival of only a limited number of neurones. This implies that an important role of cell death in the nervous system is to adjust the size of neuronal populations to the needs of their target fields. It is unclear, however, whether this is simply a quantitative adjustment, that is, neurones are lost at random until the required number is reached, or whether it involves a process of selection, that is, only neurones whose axons make appropriate terminations in the target field are selectively maintained.
The view that the influence of the target field on neuronal survival is mediated by its production of a neurotrophic factor which the innervating neurones require for their survival has received substantial support from work on NGF. NGF supports the survival of sympathetic neurones and certain kinds of sensory neurones in culture during defined periods of their development (Levi-Montalcini & Angeletti, 1968; Thoenen & Barde, 1980; Davies & Lindsay, 1985) and prevents loss of these neurones in vivo if administered throughout the phase of natural neuronal death (Thoenen & Barde, 1980; Hamburger & Yip, 1984). Conversely, the administration of anti-NGF antibodies during or shortly after the phase of cell death leads to the loss of substantial numbers of these neurones (Levi-Montalcini & Angeletti, 1968; Johnson, Gorin, Brandéis & Pearson, 1980). NGF and its messenger RNA are found in limiting amounts in the target fields of sympathetic and sensory neurones in vivo, the amount present correlating with innervation density (Korsching & Thoenen, 1983a; Heumann, Korsching, Scott & Thoenen, 1984; Shelton & Reichardt, 1984; Davies et al. 1987a). NGF-sensitive neurones possess specific cell-surface receptors for NGF (Sutter, Riopelle, Harris-Warrick & Shooter, 1979) which mediate its uptake in the target field. From here, NGF is conveyed by axonal transport to the cell bodies of the innervating neurones where it promotes their survival (Korsching & Thoenen, 19836; Palmatier, Hartman & Johnson, 1984; Davies et al. 1987a). Preventing the uptake of NGF from the target field in neonates by destroying adrenergic terminals with 6-hydroxy dopamine or interrupting axonal transport with vinblastine or axotomy leads to the death of sympathetic neurones which can be prevented by the concomitant administration of exogenous NGF (Levi-Montalcini et al. 1975; Hendry & Campbell, 1976; Menesini, Chen, Calissano & Levi-Montalcini, 1977). .
In the following sections, I shall consider aspects of the synthesis, distribution and specificity of neurotrophic factors and the nature of the competition of neurones for these factors which underlie the regulation of the number and connectivity of sensory neurones.
Site of neurotrophic factor synthesis
The recent isolation of a cDNA clone encoding NGF (Scott et al. 1983) and the subsequent development of sufficiently sensitive hybridization assays for detecting the extremely low levels of NGF mRNA in vivo (Heumann et al. 1984; Shelton & Reichardt, 1984) were essential prerequisites for identifying the cells that synthesize NGF during normal development. This was ascertained in a detailed developmental study of the most densely innervated cutaneous target field in the mouse embryo, the maxillary process of the first branchial arch (Davies et al. 1987a). The feasibility of enzymically separating the maxillary process into its principal cellular components, the epithelium (presumptive epidermis) and mesenchyme (presumptive dermis and subcutaneous tissue), permitted measurements of NGF mRNA in these components at intervals throughout the early stages of sensory innervation. Both components contain NGF mRNA, the concentration begin fivefold higher in the epithelium. This suggests that both the epithelium and mesenchyme of developing skin synthesize NGF and that the level of synthesis is highest in epithelial cells. This finding accords with the pattern of innervation of mature skin; whereas sensory fibres terminate in both the epidermis and dermis, the epidermis is more densely innervated, receiving large numbers of fibres which terminate on intraepidermal Merkel cells (Iggo & Andres, 1982). This finding is also supported by the localization of NGF mRNA by in situ hybridization in sections of the maxillary target field during the early stages of whisker follicle development (Davies et al. 1987a). In addition to showing that both the surface and follicular epithelium express higher levels of NGF mRNA than the mesenchyme, these in situ hybridizations also revealed that NGF mRNA expression in the mesenchyme is not uniform but is higher in the region next to the epithelium. Since this region gives rise to the dermis, this finding accords with higher innervation density of the dermis compared with subcutaneous tissues.
Immunohistochemical studies of the distribution of NGF in the denervated iris have led to the proposal that NGF is synthesized exclusively by Schwann cells (Rush, 1984; Finn, Ferguson, Renton & Rush, 1986). This proposal is invalid since NGF synthesis occurs in cutaneous epithelium (where there are no Schwann cells) and NGF mRNA expression in mesenchyme is not associated with bundles of nerve fibres. The recent finding that transection of the adult sciatic nerve triggers NGF receptor expression and NGF synthesis in Schwann cells (Taniuchi, Clark & Johnson, 1986; Heumann, Korsching, Bandtlow & Thoenen, 1987) suggests that the appearance of NGF immunoreactivity in Schwann cells in the denervated iris is the consequence of the response of these cells to axonal injury. Furthermore, recent in situ hybridization data suggest that the distribution of NGF mRNA in the intact normal adult iris is not consistent with exclusive expression in Schwann cells (Bandtlow, Heumann, Schwarb & Thoenen, 1987).
Although the studies of NGF synthesis in the maxillary process indicate the kinds of cells that synthesize NGF during development, they provide no direct information on the distribution of newly synthesized NGF in the target field and its availability to the innervating neurones. This information is essential for understanding the nature of the competition of neurones for NGF. There are two basic alternatives, either NGF is freely diffusible within the target field or its distribution and availability are restricted to the cells that synthesize it.
Free diffusion implies widespread availability of NGF in the target field and a means of regulating neuronal survival which is essentially quantitative - the number of neurones supported being dependent on the ambient concentration of NGF. Evidence for the release and diffusion of newly synthesized NGF comes from the finding that adult rat irides placed in culture release NGF into the culture medium (Ebendal, Olson, Seiger& Hedlund, 1980; Barth et al. 1984) and that anti-NGF antiserum considerably reduces collateral branching of local sympathetic fibres occurring after septal denervation (Springer & Loy, 1985). NGF synthesis in denervated target fields, however, occurs at a far higher rate than in development and takes place, at least in part, in glial cells. Thus, it may be misleading to draw parallels between regeneration and development in this respect.
Restriction of NGF synthesis and availability to certain target field cells is an attractive mechanism not only for regulating the number of innervating neurones but also for selectively supporting neurones on the basis of their axon terminations in the target field. In this model, NGF-producing cells can be regarded as specific target cells that only support the survival of those NGF-sensitive cells that encounter them. If correct, this model implies that the initial excess of neurones in development is required to increase the probability that all such target cells are encountered. One way NGF might be restricted to the cells that synthesize it is if these cells also express low-affinity NGF receptors (Kd= 10–9M, Sutter et al.1979), newly synthesized NGF could then be held on the cell surface until encountered by an NGF-sensitive nerve fibre and sequestered with its high-affinity receptors (Kd = 10–11). The finding that Schwann cells both synthesize NGF (Heumann et al. 1987) and express low-affinity NGF receptors after axotomy in adult rats (Taniuchi et al. 1986) suggests that a similar mechanism may operate in regeneration. Here, however, it would have a different function –stimulation of axon regrowth by presenting NGF to the regenerating nerve fibres. For these reasons, it will be of interest to see if the NGF receptor gene (Radeke et al. 1987) is expressed in developing cutaneous epithelium.
Regulation of neurotrophic factor levels
The dependence of developing neurones on a supply of a neurotrophic factor from their target field implies that the level of this factor in the target field during innervation is crucial in determining the number of neurones that survive. Knowledge of changes in neurotrophic factor levels in the target field during development and recognition of the factors that influence these levels will lead to a more complete understanding of how the target field regulates its innervation and how regional .differences in innervation density are established.
An important recent advance that has facilitated investigation of the above issues is the development of a very sensitive two-site enzyme immunoassay for NGF capable of measuring the very low levels of this protein present in normal development (Korsching & Thoenen, 1983a). Application of this technique and a sensitive hybridization assay for NGF mRNA (Heumann et al. 1984) to the cutaneous target field of the trigeminal ganglion has provided detailed information on the regulation of NGF levels in the target field during development (Davies et al. 1987a). NGF synthesis commences with the arrival of the earliest nerve fibres in the target field, and after a steady increase in the level of NGF during the first two days of innervation there is a marked fall in both the concentration and total amount of NGF coincident with the onset of neuronal death in the trigeminal ganglion (Fig. 2).
The development of the trigeminal ganglion and its innervation of the periphery. (A) Number of neurones in the trigeminal ganglion from the 10th embryonic day (E10) to postnatal day 4 (PND 4). E10 and Ell (joined by interrupted lines) are estimates from axon counts. The phases of nerve fibre recruitment to the trigeminal nerve, nerve fibre encounter with the peripheral target field epithelium (MD, mandibular process; MX, maxillary process) and neurone death and fibre degeneration are shown by the horizontal bars (data from Davies & Lumsden, 1984). (B) Total amount of NGF in the maxillary process (solid circles) and trigeminal ganglion (open circles) from E10 to E15. (C) Concentration of NGF in the same. (D) Total amount of NGF mRNA in the maxillary process from E10 to E15. (E) Concentration of NGF mRNA in the same (data from Davies et al. 1987a).
The development of the trigeminal ganglion and its innervation of the periphery. (A) Number of neurones in the trigeminal ganglion from the 10th embryonic day (E10) to postnatal day 4 (PND 4). E10 and Ell (joined by interrupted lines) are estimates from axon counts. The phases of nerve fibre recruitment to the trigeminal nerve, nerve fibre encounter with the peripheral target field epithelium (MD, mandibular process; MX, maxillary process) and neurone death and fibre degeneration are shown by the horizontal bars (data from Davies & Lumsden, 1984). (B) Total amount of NGF in the maxillary process (solid circles) and trigeminal ganglion (open circles) from E10 to E15. (C) Concentration of NGF in the same. (D) Total amount of NGF mRNA in the maxillary process from E10 to E15. (E) Concentration of NGF mRNA in the same (data from Davies et al. 1987a).
NGF mRNA appears in the target field just before NGF, and the levels of these two molecules increase in parallel during the early stages of target field innervation. This suggests NGF synthesis in development is governed by the level of its messenger. The continuous increase in the total amount of NGF mRNA in the target throughout the period of innervation suggests that the fall in the level of NGF midway through this period is not due to decreased synthesis. Rather, removal by the innervating sensory nerve fibres appears to be the major if not the sole cause, as increasing levels of NGF unaccompanied by NGF mRNA appear in the trigeminal ganglion from shortly after the commencement of NGF synthesis in the target field.
Since numerous in vitro studies have shown that the survival of developing NGF-sensitive neurones is dependent on the ambient concentration of NGF, it is possible that the fall in the concentration of NGF in the target field contributes to the coincident onset of neuronal death in the trigeminal ganglion. Prior to the fall it is conceivable that the relatively high concentration of NGF in the target field is able to delay the onset of neuronal death in the ganglion until a large proportion of the neurones have innervated the target field, thereby ensuring that the majority of neurones compete for survival during the same period of development (i.e. after the fall in the level of NGF). If the level of NGF were limiting from the earliest stages of target field innervation it is possible that too many neurones would be eliminated before the capacity of the growing target field has increased to support the required number of neurones.
One possible consequence of a delay in the onset of neuronal death due to an initially high level of neurotrophic factor in the target field is that experimental removal of the target field prior to innervation would cause a small advance in the timecourse of neuronal death within the innervating population of neurones. Most studies of the effects of early target field removal, however, have reported that the accompanying neuronal death occurs over much the same period of development as naturally occurring neuronal death. For example, this has been observed in the lateral motor column (Prestige, 1967; Oppenheim, Chu-Wang & Maderdrut, 1978), trochlear nucleus (Cowan & Wenger, 1967), isthmo-optic nucleus (Cowan & Wenger, 1968) and ciliary ganglion (Landmesser & Pilar, 1974). However, in a recent detailed study of neuronal death among the ventrolateral sensory neurones of DRG it was found that the peak degeneration clearly occurs one day earlier in ganglia deprived of their peripheral target field (Hamburger & Yip, 1984).
The demonstration that sensory neurones influence the level of NGF in the target field during development raises the possibility that different NGF-sensitive neuronal populations innervating the same target field modulate each others survival and connectivity by influencing the level of NGF. Consonant with this possibility is the finding that removal of either of the three classes of fibres that innervate the iris (sensory, parasympathetic and sympathetic) increases the levels in the iris of molecules specifically associated with each of the two remaining classes of fibres (Kessler, Bell & Black, 1983; Kessler, 1985). These studies, however, were carried out in the adult iris in which axonal damage leads to the production of large quantities of freely diffusible NGF and possibly other neurotrophic factors by Schwann cells. Thus, it is possible that the changes observed in these studies are consequent upon the responses of Schwann cells to axonal damage and do not necessarily reflect the occurrence of reciprocal interactions among different classes of fibres during normal development.
Two findings raise the possibility that NGF synthesis is initiated by the arriving nerve fibres: the onset of NGF synthesis and target field innervation are temporally related and the total amount of NGF mRNA in the target field increases until the last fibres arrive (Fig. 2). If nerve fibres do trigger NGF synthesis, this does not imply that these fibres control the overall production of NGF by the target field. Rather, there may be a limited number of presumptive NGF-producing cells in each target field that only synthesize NGF when encountered by the appropriate nerve fibres. If so, regulation of the number of these cells in different target fields would provide a mechanism for varying innervation density in different regions.
Stage of neurotrophic factor dependence
Several studies suggest that the survival of sensory neurones is dependent on neurotrophic factors for only a limited period of their development. As with most aspects of neurotrophic factor research, the most extensive data comes from studies of influence of NGF on sensory neurones.
The stage at which developing sensory neurones become responsive to NGF has been ascertained from studies of the mouse trigeminal ganglion, which can be dissected prior to the emergence of nerve fibres from it in vivo (Davies, Lumsden, Slavkin & Burnstock, 1981). Neurite outgrowth from ganglia explanted at this early stage is neither augmented by NGF nor reduced by anti-NGF antiserum (Davies & Lumsden, 1984), and these neurites do not bind iodinated NGF, suggesting that they lack NGF receptors (Davies et al. 1987a). In ganglia explanted at later ages, the appearance of NGF-promoted neurite outgrowth and receptor expression coincides with that stage at which the earliest trigeminal nerve fibres reach their peripheral targets in vivo. These findings suggest that developing sensory neurones lack NGF receptors when their axons are growing to their targets and that they express receptors and become responsive to NGF when they encounter their targets. Early neurones cultured prior to innervating their targets express receptors after 48 h in culture (by which time they would have encountered their targets in vivo). This suggests that expression of NGF receptors does not depend on an inductive interaction with the target field but is an intrinsic feature of the development of sensory neurones.
Does the appearance of NGF receptors and NGF responsiveness indicate the acquisition of NGF dependence by early sensory neurones? To investigate whether NGF is necessary for neuronal survival at this stage, it is necessary to study these neurones in dissociated culture. This is complicated by the difficulty of distinguishing and separating very early neurones from other cells at the outset. Nonetheless, in low-density cultures of trigeminal ganglion cells established prior to target field innervation, neither NGF nor anti-NGF influence neuronal survival. In cultures established at later stages, increasing numbers of neurones survive in the presence of NGF compared with cultures supplemented with anti-NGF (A.M.D., unpublished results). This provides some evidence that the survival of sensory neurones also becomes dependent on NGF when they encounter their targets. A phase of NGF-independent survival has also been described for early sympathetic neurones grown in dissociated culture (Coughlin & Collins, 1985). In view of the parallel with developing sensory neurones, it will be of interest to ascertain whether target encounter and the transition between NGF-independent survival and NGF dependence are related in sympathetic neurones.
Do the findings on the appearance of NGF dependence apply to other neurotrophic factors? There is some indirect evidence that another purified neurotrophic factor for sensory neurones, brain-derived neurotrophic factor (BDNF, Barde, Edgar & Thoenen, 1982), acts during the same stage of development as NGF. The neurite-promoting effects of BDNF on DRG neurones begin at the same stage of development as NGF responsiveness (Davies, Thoenen & Barde, 1986a) and show a similar developmental timecourse (Lindsay, Thoenen & Barde, 1985b; Davies et al. 1986a).
Is the survival of sensory neurones independent of all neurotrophic factor support prior to target field innervation? The finding that newly differentiated trigeminal neurones survive and grow for about a day in low-density culture on a laminin substratum (whereas later embryonic trigeminal neurones degenerate within several hours under these conditions) provides some evidence for neurotrophic factor independence at this stage (A.M.D., unpublished findings). Furthermore, there are marked differences in the survival and growth of newly differentiated neurones of different cranial sensory ganglia in culture (e.g. nodose neurones survive far longer and extend neurites at a much faster rate than vestibular neurones). Since these early neurones were grown under identical conditions, these intrinsic differences in survival and growth are unlikely to be due to differences in the availability of an unidentified neurotrophic factor in the serum supplements of the culture medium. Rather, it is possible that they are a reflection of the distances the axons of these neurones have to grow to reach their targets in vivo (nodose fibres have to grow many times further than vestibular fibres).
In vivo studies of the effects of anti-NGF antibodies and in vitro studies of the influence of NGF on neuronal survival have provided evidence that the dependence of developing sensory neurones on NGF is lost as they mature. Transplacental transfer of anti-NGF antibodies to fetal rodents causes degeneration of NGF-sensitive sensory neurones (Johnson et al.1980)whereas administration of anti-NGF antiserum to neonates causes negligible loss of these neurones (Levi-Montalcini & Angeletti, 1968; Yip, Rich, Lampe & Johnson, 1984). The survival of midem-bryonic chick DRG neurones grown in dissociated neurone-enriched culture is promoted by NGF whereas the survival of these neurones cultured under similar conditions later in development is independent of NGF (Greene, 1977). These findings accord with data indicative of a marked fall in the number of NGF receptors on chick DRG neurones from the midembryonic period onwards (Herrup & Shooter, 1975; Rohrer & Barde, 1982), suggesting that dependence may be regulated by NGF-receptor expression. Similar in vivo and in vitro studies of developing sympathetic neurones suggest that these neurones also pass through a phase of NGF dependence (Levi-Montalcini & Angeletti, 1968; Chun & Patterson, 1977). In all of the above cases, the phase of NGF dependence extends throughout and just beyond the period of neuronal death. This is consistent with the importance of target-derived NGF in regulating neuronal survival during this critical period of development.
Is the survival of mature sensory neurones independent of all neurotrophic factor support? This issue has been approached in two ways: by studying the effects of depriving adult neurones of target-derived trophic support by axotomy in vivo and by studying the survival and growth of adult neurones in vitro. Cutting either the peripheral branches of adult guinea pig DRG or the central branches in animals autoimmunized against NGF leads to the death of about a quarter of the neurones (Johnson & Yip, 1985). This suggests that the survival of at least some adult DRG neurones is dependent on trophic support from their target fields. However, the possibility that adult sensory neurones die, in part, as a consequence of axonal injury or become dependent on neurotrophic factors only after injury cannot be discounted in these experiments. In a recent study of adult rat DRG neurones in culture, Lindsay (1986) found that the great majority of these neurones survived for many weeks and extended long processes in the absence of possible sources of neurotrophic factors such as serum supplements and ganglion-supporting cells (which have been shown to synthesize neurotrophic factors in vitro, Burnham, Raiborn & Varón, 1972). Since only a relatively small number of neurones were lost in setting up these cultures, it seems unlikely that the findings apply to an unrepresentative subpopulation of neurones. Thus, in contrast to the findings of Johnson and Yip, these in vitro studies suggest that the survival of adult DRG neurones is not dependent on neurotrophic factors. Interestingly, although neither NGF nor BDNF significantly increased the number of surviving neurones, these factors enhanced the regeneration of neuronal processes during the first few days in culture. This may be an important feature of the response of mature sensory neurones to injury.
Specificity of neurotrophic factors for sensory neurones
Several experimental observations suggest that the survival of developing sensory neurones is dependent on trophic support from both their peripheral and central target fields. Dependence on the peripheral target field is demonstrated by the finding that extirpation of a limb bud prior to its innervation by dorsal root ganglia (DRG) substantially increases the magnitude of neuronal death in these ganglia (Hamburger & Levi-Montalcini, 1949; Prestige, 1967; Carr & Simpson, 19786; Hamburger & Yip, 1984). Although the effects of similar early manipulations of the central target fields of sensory ganglia have not been studied, the finding that about half of the neurones in newborn rat dorsal root ganglia die after their central branches are cut (Yip & Johnson, 1984) provides some indirect evidence that at least some sensory neurones depend on the CNS for survival. Further circumstantial evidence for the dependence of developing sensory neurones on their peripheral and central target fields comes from studies of the embryonic chick cochleovestibular ganglion in culture (Ard, Merest & Hauger, 1985). Explants comprising the ganglion with or without its central or peripheral target field were cultured from the stage at which the earliest nerve fibres are growing to their targets to the end of the phase of natural neuronal death in vivo. Although neuronal survival was substantially lower in these cultures than in vivo, there were two-to fourfold more neurones in ganglia cultured with either their central or peripheral target field than in ganglia cultured alone.
Until recently, almost all of the work on the neurotrophic factor requirements of developing sensory neurones has been done on embryonic DRG neurones. In culture, the survival of these neurones is promoted to varying extents by several purified neurotrophic factors, namely NGF (Levi-Montalcini & Angeletti, 1968; Greene, 1977), BDNF (Barde et al. 1982), ciliary neuronotrophic factor (Barbin, Manthorpe & Varón, 1984) and neuroleukin (Gurney, Heinrich, Lee & Yin, 1986) and by extracts of a variety of tissues including heart, liver (Lindsay & Tarbit, 1979), gut (Riopelle & Cameron, 1981), spinal cord (Lindsay & Peters, 1984) and skeletal muscle (Hsu, Natyzak & Trupin, 1984). DRG, however, are composed of a variety of cutaneous sensory, visceral sensory and proprioceptive neurones, and since these different classes of sensory neurones cannot be separately identified by their morphology in culture it has not been possible, using DRG neurones, to ascertain whether these classes (or possibly functionally distinct subsets of neurones among these classes) have different neurotrophic factor requirements.
The segregation of cranial sensory neurones (Fig. 3) into functionally distinct groups which are derived from either neural crest or placode (Table 1) together with the feasibility of dissecting these groups from chick embryos has facilitated investigation of the specificity of various neurotrophic factors for different kinds of sensory neurones. In the following sections, I shall consider whether specificity is related to sensory neurone connectivity and function (i.e. cutaneous sensory, visceral sensory and proprioceptive or functional subsets of these) or whether it is simply related to sensory neurone ontogeny (neural crest versus placodal derivation). I shall consider first,the specificity of neurotrophic factors from the periphery and second, the specificity of neurotrophic factors from the CNS.
Function, development and neurotrophic factor responsiveness of populations of sensory neurones in the chick embryo

Head of a day-10 (E10) chick embryo showing the location of the populations of cranial sensory ganglia. Trigeminal ganglion (T), trigeminal mesencephalic nucleus (TMN), vestibulo-acoustic ganglion (VA), geniculate ganglion (G), jugular ganglion (J), petrosal ganglion (P) and nodose ganglion (N).
Head of a day-10 (E10) chick embryo showing the location of the populations of cranial sensory ganglia. Trigeminal ganglion (T), trigeminal mesencephalic nucleus (TMN), vestibulo-acoustic ganglion (VA), geniculate ganglion (G), jugular ganglion (J), petrosal ganglion (P) and nodose ganglion (N).
Neurotrophic factors from the periphery
Several findings suggest that sensory neurones obtain NGF predominantly, if not exclusively, from their peripheral target fields. NGF is present in cutaneous target fields (Davies et al. 1987a) and in spinal nerves distal to DRG (peripheral branches of sensory neurones) but is undetectable in spinal nerve roots (central branches of sensory neurones) and in the spinal cord (Korsching & Thoenen, 1985). NGF mRNA is present in appreciable amounts in cutaneous target fields (Davies et al. 1987a) but only low levels are present in the spinal cord (Shelton & Reichardt, 1986; Goedert, Fine, Hunt & Ullrich, 1986). Yip & Johnson (1984) have recently proposed that NGF may provide trophic support to sensory neurones via their central branches. This was based on the demonstration that 125I-labelled NGF is conveyed by axonal transport from the spinal cord to DRG and that the loss of DRG neurones which accompanies dorsal rhizotomy in newborn rats can be prevented by the concomitant administration of NGF. These findings merely indicate that NGF receptors are distributed over the whole surface of sensory neurones and that a sufficiently high concentration of NGF can compensate for loss of central neurotrophic support (see section on interaction of peripheral and central neurotrophic factors).
Almost all studies of the specificity of NGF for different kinds of sensory neurones have focused on differences in the response of populations of neural-crest-derived and populations of placode-derived sensory neurones. In culture, NGF promotes the survival and growth of neurones in neural-crest-derived sensory ganglia, namely, DRG (Levi-Montalcini & Angeletti, 1968; Greene, 1977), the dorsomedial part of the trigeminal ganglion (Levi-Montalcini, 1962; Ebendal & Hedlund, 1975; Davies & Lumsden, 1983) and the jugular ganglion (Davies & Lindsay, 1985). In contrast, NGF does not support the survival of the placode-derived neurones of the ventrolateral part of the trigeminal ganglion (Levi-Montalcini, 1962;Ebendal & Hedlund, 1975; Davies & Lumsden,1983), the nodose (Lindsay & Rohrer, 1985), petrosal, geniculate and vestibular ganglia (Davies & Lindsay, 1985). Consonant with these findings is the demonstration that passive placental transfer of anti-NGF antibodies to fetal guinea pigs causes a marked depletion of neurones in neural-crest-derived sensory ganglia but has no effect on the survival of placode-derived sensory neurones (Pearson, Johnson & Brandéis, 1983). Furthermore, DRG neurones possess NGF receptors (Herrup & Shooter, 1975; Sutter et al. 1979; Rohrer & Barde, 1982; Raivich, Zimmermann & Sutter, 1985) whereas nodose ganglion neurones do not (Lindsay & Rohrer, 1985) and neural-crest-derived sensory ganglia, but not ganglia in which the neurones are of placodal origin, are heavily labelled by 125I-NGF in sections of chick embryos (Raivich, Zimmermann & Sutter, 1987).
The above findings have led to the widely held view that NGF-dependence is directly related to sensory neurone ontogeny, in that all neural-crest-derived sensory neurones require NGF for survival whereas placode-derived neurones do not. The finding that the neural-crest-derived sensory neurones of the embryonic chick trigeminal mesencephalic nucleus (TMN) are not supported by NGF in culture but do survive when grown under similar conditions in the presence of BDNF (Davies, Lumsden & Rohrer, 1987b) clearly shows that not all neural-crest-derived sensory neurones are dependent on NGF for survival. Straznicky & Rush (1985) have also shown that NGF administered to chick embryos reduces naturally occurring cell death in the trigeminal ganglion but has negligible effect on preventing cell death in the TMN. However, so entrenched is the view that NGF dependence is an invariable feature of neural-crest-derived sensory neurones, these workers interpreted their findings as indicating that TMN neurones are of placodal origin. Cell-tracing studies in quail-chick chimaeras (Narayanan & Narayanan, 1978) clearly indicates that these neurones are derived from neural crest.
In addition to TMN neurones, is the survival of other neural crest-derived proprioceptive neurones independent of NGF? There is some evidence that not all neurones in DRG require NGF for survival. Johnson and coworkers (1980, 1983) have shown that no more than 80 to 85 % of DRG neurones degenerate in fetal rodents exposed transplacentally to maternal anti-NGF antibodies, compared with the loss of almost all sympathetic neurones. Although these workers concede that their findings may indicate that some embryonic DRG neurones are not dependent on NGF, they favour the interpretation that DRG neurones that are bom early pass through a phase of NGF-dependence before maternal antibody is able to reach the fetus. The finding that a greater number of embryonic DRG neurones are supported in culture by NGF plus BDNF than are supported by either NGF or BDNF alone (Barde et al. 1982; Lindsay et al. 1985b), together with the demonstration that sensory neurones respond to NGF and BDNF during the same stage of development (Lindsay et al. 1985b; Davies et al. 1986a), suggests that some DRG neurones do not depend on NGF for survival.
Several observations suggest that proprioceptive neurones comprise at least some of the NGF-independent neurones in developing DRG. Rodent pups exposed to a high titre of maternal anti-NGF antibody in utero are unresponsive to a variety of painful stimuli but their proprioceptive function is unaltered (Johnson et al. 1983). Anti-NGF antibodies injected into fetal rats results in a depletion of more than 90 % of unmyelinated fibres and 35 % of myelinated fibres in spinal nerve dorsal roots examined 4 months after birth (Goedert et al. 1984). The loss of myelinated nerve fibres was restricted to small-diameter fibres with no change in large-diameter fibres suggesting that proprioceptive neurones (from which large-diameter myelinated fibres arise) are not dependent on endogenous NGF. The death of small sensory neurones, but not large sensory neurones, which occurs after crushing the nerve to a muscle in newborn rats can be prevented by the concomitant administration of NGF, suggesting that proprioceptive neurones are not supported by NGF after being isolated from their peripheral target by axotomy (Miyata, Kashihara, Homma & Kuno, 1986). NGF injected over the gluteus maximus muscle in neonatal mice causes prolific growth of the small diameter sensory axons which ramify over the surface of the muscle but does not affect the axons of muscle spindles and tendon organs (Hopkins & Slack, 1984). Whereas the levels of NGF and its messenger RNA are high in developing cutaneous target fields (Davies et al. 1987a) and in a variety of other peripheral tissues, these molecules are present in very low amounts in skeletal muscle (Korsching & Thoenen, 1983a; Heumann et al. 1984; Shelton & Reichardt,1984) .
Although NGF does not promote the survival of TMN neurones in culture, the great majority of these neurones express NGF receptors (Davies et al. 1987b). Likewise, whereas more than 90 % of embryonic chick DRG neurones express NGF receptors, only half of these neurones are supported by NGF in culture (Rohrer & Barde, 1982), suggesting that at least a proportion of the NGF-independent neurones in DRG express NGF receptors. Several recent studies have also shown that NGF binds specifically to several non-neuronal cell types and other NGF-independent neurones during a brief period of development. These comprise the supporting cells of sensory and autonomic ganglia (Rohrer, 1985; Zimmermann & Sutter, 1983), developing muscle cells (Raivich et al. 1985) and the neurones of the lateral motor column (Raivich et al. 1985). This raises the possibility that the NGF receptors on these cells (and on proprioceptive neurones) may be involved in functions other than maintaining neuronal survival.
What are the common features of NGF-dependent sensory neurones? The demonstration that exogenous NGF increases the substance P content of DRG (Schwartz & Costa, 1979; Kessler & Black, 1980) and that anti-NGF antibodies have the opposite effect (Otten, Goedert, Mayer & Lembeck, 1980; Goedert et al. 1984) together with the growing body of evidence implicating substance P sensory neurones in nociception (Nicoll, Schenker & Leeman, 1980) suggest that nociceptive neurones comprise a functionally distinct subset of sensory neurones that respond to NGF. This is consistent with the finding that anti-NGF antibodies cause degeneration of a substantial number of unmyelinated sensory nerve fibres (Goedert et al. 1984) since nociceptive afferents comprise a large proportion of these fibres. Large numbers of substance P neurones are only found in NGF-responsive sensory ganglia (Fontaine-Perus, Chaconie & Le Douarin, 1985) and, although a small number of substance P neurones are present in the nodose ganglion (the neurones of which do not respond to NGF, Lindsay & Rohrer, 1985), there is some evidence that these neurones subserve sensory functions other than nociception (Heike, O’Donohue & Jacobowitz, 1980). Since there are more NGF-dependent sensory neurones in DRG than neurones that express substance P (Jessell & Dodd, 1986), it is clear that other kinds of sensory neurones in addition to nociceptive neurones are dependent on NGF for survival.
Is the response of sensory neurones to other neurotrophic factors from the periphery related to embryonic derivation or to sensory function? The first evidence that responsiveness is unrelated to embryonic derivation came from a comparative study of the influence of NGF and tissue extracts on the survival and growth of two populations of neural-crest-derived sensory neurones: the dorsomedial part of the trigeminal ganglion (DM-TG) and lumbosacral DRG (Davies & Lindsay, 1984). Whereas NGF promotes the survival of most of the neurones in both populations, extracts of liver and eye promote the survival of about a third of the DRG neurones but have negligible effect on the survival of DM-TG neurones. This suggests that these tissue extracts contain a factor or factors to which a proportion of DRG neurones, but not DM-TG neurones, respond.
Since the DM-TG contains mainly cutaneous sensory neurones whereas DRG contain several other kinds of sensory neurones in addition to cutaneous sensory, it is possible that the response of DRG neurones to NGF and several tissue extracts is related to broader sensory function of DRG.
Further evidence that the neurotrophic factor requirements of developing sensory neurones are related to the kinds of sensory structures that they innervate comes from an in vitro study of the TMN neurones of the chick embryo (Davies, 1986). In dissociated, glia-free culture these neurones die within 24 h unless supplemented with a soluble extract of their peripheral target tissue, skeletal muscle. This extract promotes the long-term survival of the great majority of these neurones. Skeletal muscle extract elicits neurite outgrowth from TMN, VL-TG and DM-TG explants in accordance with the proportion of proprioceptive neurones in these populations. This suggests that skeletal muscle contains a neurotrophic factor to which proprioceptive, but not cutaneous, sensory neurones respond. The recent isolation of neuroleukin, a 56×103Mr protein present in skeletal muscle which promotes the survival of motoneurones and NGF-independent sensory neurones in DRG (Gurney et al. 1986), raises the possibility that this neurotrophic factor may have been responsible for the specific trophic effects observed in the above study.
A number of studies have shown that other populations of NGF-independent sensory neurones are also supported in culture by extracts of their peripheral target tissues. For example, liver extract promotes the survival of nearly half of the neurones of the nodose and petrosal ganglia (Lindsay & Tarbit, 1979; Lindsay & Rohrer, 1985; Davies & Lindsay,1985)and an extract of membranous labyrinth supports the survival of vestibular neurones (A.M.D., unpublished findings). However, the specificity of these extracts for other kinds of sensory neurones has yet to be examined in detail, and the relationship of the neurotrophic factor(s) in these extracts to the factor in skeletal muscle has yet to be ascertained. Significant progress will only be made when the neurotrophic factors in these extracts have been purified. The availability of monoclonal antibodies for a variety of neuropeptides, cell-surface carbohydrate antigens and cytoskeletal components expressed by different subsets of sensory neurones (Jessell & Dodd, 1986) provides the opportunity to further delineate which kinds of sensory neurones respond to a given neurotrophic factor.
In summary, it is becoming apparent that the neurotrophic factor requirements of developing sensory neurones are related to the kinds of structures they innervate. This is in contrast to the prevalent view that sensory neurones are differentially responsive to peripheral target-derived neurotrophic factors depending on whether the neurones are of neural crest or placodal origin (Lindsay, Barde, Davies & Rohrer, 1985a; Lindsay et al. 1985b). Detailed studies of the specificity of neurotrophic factors for the neurones of the chick trigeminal system suggest that there may be at least three different neurotrophic factors from the periphery. The survival of DM-TG neurones (small cutaneous sensory neurones including many substance P neurones) is promoted by NGF, TMN neurones (proprioceptive) are supported by a skeletal muscle-derived factor and VL-TG neurones (large cutaneous sensory) are supported by neither of these factors, thereby raising the possibility of a third factor. In the light of these findings, it will be of interest to ascertain the kinds of structures innervated by DM-TG and VL-TG neurones in mature skin.
Neurotrophic factors from the central nervous system
One candidate for a neurotrophic factor that mediates the trophic support of the CNS on developing sensory neurones is BDNF, a basic protein of relative molecular mass 12 300 purified from pig brain (Barde et al. 1982). This protein promotes the survival of embyronic DRG and nodose ganglion neurones in culture but is inactive on sympathetic and parasympathetic neurones (Lindsay et al. 1985b). In addition to DRG and nodose ganglion neurones, a recent study of the influence of BDNF on explants and dissociated cultures of all other populations of sensory neurones in the chick embryo (Table 1) has shown that BDNF-responsive neurones are present in each of these populations (Davies et al. 1986a).
The issue of whether BDNF promotes the survival of all sensory neurones or only a certain subpopulation was resolved in a comparative study of the influence of BDNF and NGF on the survival of embryonic DM-TG and VL-TG neurones in culture (Davies et al. 1986a). BDNF and NGF have reciprocal effects on the survival of these neurones; BDNF promotes the survival of the majority of VL-TG neurones and NGF promotes the survival of the majority of DM-TG neurones. This indicates that BDNF does not support the survival of all sensory neurones and that the proportion of BDNF-responsive neurones varies in different populations of sensory neurones. These findings also raise the possibility that BDNF-responsive and NGF-responsive sensory neurones comprise distinct subsets. The finding that the combined effect of NGF and BDNF on the survival of midembryonic chick DRG neurones in culture is additive (Lindsay et al. 1985b) provides further evidence in support of this possibility. The finding, however, that the combined effect of NGF and BDNF is less than additive earlier in development suggests that NGF-responsive and BDNF-responsive subsets may be partially overlapping at earlier stages.
Are there other neurotrophic factors for sensory neurones present in the CNS? An opportunity to screen for such factors is provided by the predominantly BDNF-insensitive neurones of the DM-TG. Mouse brain extract promotes the survival of over half of these neurones in the presence of anti-mouse NGF antiserum (A.M.D., unpublished findings). This suggests that the CNS contains a sensory neurone neurotrophic factor that is neither BDNF nor NGF. Purification of this factor will determine whether it is specific for BDNF-insensitive sensory neurones.
Interaction of peripheral and central neurotrophic factors
Is the survival of each sensory neurone dependent on neurotrophic factor support from both its peripheral and central target fields or are there separate subsets of sensory neurones which only require trophic support from one or other of these target fields? To resolve this issue it was necessary to study the influence of neurotrophic factors from the periphery and CNS on the survival of a functionally distinct subset of sensory neurones under stringent tissue culture conditions. For this study (Davies, Thoenen & Barde, 1986b), TMN neurones, which innervate skeletal muscle and certain groups of neurones in the brainstem, were chosen since they can be prepared completely free of glial cells and were known from previous studies to respond to a factor present in skeletal muscle and to BDNF (Davies, 1986; Davies et al. 1986a). Although the factor in skeletal muscle extract (SMx) has yet to be purified, it was shown to be distinct from BDNF by the demonstration that, unlike BDNF, it is inactive on VL-TG neurones.
Virtually all TMN neurones die within several hours unless grown with SMx or BDNF. In the presence of saturating concentrations of either of these factors, over 70% of the neurones survive and can be sustained in culture for over a month with regular supplements of either factor. The finding that there is negligible additional survival in the presence of saturating concentrations of both factors in combination (Fig. 4) indicates that all surviving neurones respond to both factors. The combination of subsaturating concentrations of SMx and BDNF is additive at concentrations that promote half-maximal survival alone and is greater than additive at very low concentrations (e.g. the percentage survival in the lowest concentration combination was greater than the percentage survival in cultures supplemented with a fourfold higher concentration of either factor alone,see Fig. 4). These findings not only indicate that all TMN neurones respond to two different neurotrophic factors present in their peripheral and central target fields, but also suggest that these factors potentiate each other at low concentration. .
Percentage survival of TMN neurones after 48 h incubation with BDNF alone (light hatched bars), skeletal muscle extract alone (stippled bars) and BDNF plus skeletal muscle extract (dark hatched bars). BDNF, 1X =25pgm–1; skeletal muscle extract, lX = 625ngml–1; BDNF+extract, lX = 25ngml–1 + 625ngmr1. Saturation is 256X.
Percentage survival of TMN neurones after 48 h incubation with BDNF alone (light hatched bars), skeletal muscle extract alone (stippled bars) and BDNF plus skeletal muscle extract (dark hatched bars). BDNF, 1X =25pgm–1; skeletal muscle extract, lX = 625ngml–1; BDNF+extract, lX = 25ngml–1 + 625ngmr1. Saturation is 256X.
The neurite-promoting effects of SMx and BDNF on TMN explants have a very similar developmental time course. This suggests that these factors are active during the same stage of development. The magnitude of neurite outgrowth in response to each factor is most marked during the period of natural neuronal death in vivo, suggesting that TMN neurones are most sensitive to the neurotrophic effects of SMx and BDNF at the same time as their survival is critically dependent on forming appropriate peripheral and central connections.
Developing sensory neurones respond to distinct neurotrophic factors from the periphery and CNS during the critical stage of target field innervation. This forms the basis of mechanism for selectively supporting the survival of only those neurones that make appropriate terminations in both targets fields. If the availability of each neurotrophic factor in the respective central and peripheral target fields lies below the critical threshold for maintaining survival, it would be necessary for a developing sensory neurone to make appropriate terminations in both target fields to procure sufficient trophic support to ensure survival. The evidence that peripheral and central neurotrophic factors potentiate each other at low concentrations implies that the available concentration of these factors at specific termination sites could be kept well below the threshold at which either would promote neuronal survival alone, thereby ensuring that the mechanism of selection operates effectively with a reduced margin of error.
In the light of the above findings, it will be of interest to ascertain whether the, as yet, uncharacterized CNS-derived factor for DM-TG neurones cooperates with NGF in regulating the survival of these and other NGF-dependent sensory neurones, and whether, in addition to sensory neurones, other neurones which innervate more than one target field require a different neurotrophic factor from each target field for survival. Furthermore, given the evidence suggesting that the survival of various populations of developing neurones is dependent on the afferents they receive in addition to their target fields (Levi-Montalcini, 1949; Sohal, 1976; Cunningham, Huddelston & Murphy, 1979; Okada & Oppenheim, 1984), it will be of interest to ascertain whether neuronal survival in these populations is regulated by different neurotrophic factors from their target fields and afferents.
The establishment of somatotopic projections
Since the pioneering studies of Woolsey & Van der Loos (1970), the trigeminal system of rodents has become an important and extensively studied experimental model for investigating the establishment of somatotopic projections in the developing nervous system (for recent reviews see Durham & Woolsey, 1984; Van der Loos & Welker, 1985). The characteristic pattern of whisker follicles on the snout is represented by a similar and functionally corresponding arrangement of multineuronal units in the brainstem trigeminal nuclei, thalamus and somatosensory cortex. In development, the appearance of the whisker follicle pattern on the snout precedes the sequential emergence of the homomorphic neuronal arrays in the brainstem, thalamus and cortex. This developmental sequence, together with the demonstration that snout skin cultured prior to innervation is capable of generating the whisker follicle pattern and that early lesions of either whisker follicles or their nerve supply disrupt the establishment of the central patterns, suggest that development of the central patterns is dependent on spatial information provided by the periphery. This developmental principle appears to apply to the entire somatosensory system. Dawson & Killackey (1987) have recently shown that lesions of limbs or their nerve supply during a short critical period in neonatal rats disrupt the establishment of the corresponding cortical representations.
What cellular mechanisms are involved in transferring the peripheral pattern to the CNS? In a light microscopic study of the embryonic rat maxillary nerve (which innervates the whisker follicles), Erzurumlu & Killackey (1982, 1983) concluded that this nerve is highly ordered from the earliest stages of development into fasciculi which innervate single rows of whisker follicles and that these fasciculi remain discrete throughout their course. These workers proposed that the ordered growth of trigeminal nerve fibres could alone account for the transfer of the whisker follicle pattern to the brainstem. They argued that the earliest trigeminal nerve fibres grow to their targets in straight lines and that later fibres are guided by their predecessors resulting in the establishment of the somatotopic projection by the maintenance of near-neighbour relationships between growing fibres. However, in a recent study of detailed reconstructions of the embryonic mouse maxillary nerve from serial light and electron microscopic sections at stages prior to, and after, the appearance of early whisker follicles (Davies & Lumsden, 1986), it was shown that there is neither correspondence nor obvious pattern in the arrangement of fasciculi in the developing nerve to the emerging pattern of whisker follicles. Fasciculi both merge and branch to form an intricate plexus, the arrangement of which is not consistent from one embryo to another. Analysis of the numbers of sensory nerve fibres in merging and branching fasciculi revealed that this plexus is not simply comprised of discrete collections of associated fibres which merge and separate, but that nerve fibres are freely exchanged between fasciculi. This indicates that spatial information cannot be transferred from the periphery to the CNS by the maintenance of near-neighbour relationships between growing nerve fibres. Williams & Rakic (1985) have also shown that nerve fibres move freely between fasciculi in the developing optic nerve. Thus, we must look for an alternative mechanism to account for the development of somatotopic projections.
Several electron microscopic studies have demonstrated the occurrence of collateral branching along the course of developing nerves (Landmesser & Pilar, 1976; Sohal & Weidman, 1978). There is some indirect evidence that collateral branching in the embryonic mouse maxillary nerve also begins shortly after nerve fibres reach their targets (Davies & Lumsden, 1986). A possible consequence of collateral branching together with the random exchange of growing fibres between fasciculi in this nerve is that each neurone innervates not one but several whisker follicles during development. This proposal is in general agreement with several recent studies which have shown that the connections of many different kinds of neurones are more diffuse during development than in the mature nervous system. As development proceeds, excess terminal or collateral branches are eliminated, and the available evidence suggests that impulse activity is necessary for this transition to occur (for a recent review see Cowan et al. 1984). It is possible that somatotopic sensory projections develop along similar lines with perhaps reverberating impulse activity in sensory neurones involved in selecting appropriate peripheral and central collaterals. This mechanism would explain one observation in the trigeminal system which is difficult to reconcile with a model based on the maintenance of near-neighbour relationships, that is, whereas the whisker pattern is accurately replicated in the brainstem, somatotopy in the ganglion is not so well defined (Arvidsson, 1982; Erzurumlu & Killackey, 1983). In the model proposed here, the establishment of the whisker follicle pattern in the brainstem results from the selective maintenance of topographically related collaterals in the peripheral and central target fields of the trigeminal ganglion that reach these corresponding loci by an element of random growth in the plexus of the early nerve. Thus, the perikarya of neurones that innervate a single whisker follicle would inevitably be scattered to varying extents within the ganglion.
Conclusions
The growth of sensory nerve fibres is channelled within a system of stereotyped pathways in at least the proximal part of their course to peripheral targets. Although the factors which govern the location and dimensions of these pathways are unclear, there is evidence that in some locations either physical barriers to growth cone advancement or preformed pathways of laminin may play a role. In addition to these relatively nonspecific influences on axon growth, there is increasing evidence for the existence of various kinds of specific guidance cues to which subsets of sensory fibres selectively respond. It is apparent that motor axons guide proprioceptive axons to muscle, possibly by selective fasciculation of these two classes of fibres. Several indirect observations suggest that motor axons may also influence pathway selection by cutaneous sensory nerve fibres of different spinal ganglia. There is compelling evidence that trigeminal nerve fibres are guided, in part, to their peripheral target field by a specific, epithelium-derived, chemotropic factor. It seems unlikely, however, that region-specific, chemotropic guidance operates generally in the developing sensory nervous system. The view that NGF guides developing sensory nerve fibres to their target fields by chemotropism can be discounted on account of the demonstration that sensory nerve fibres lack NGF receptors when they are growing to their targets and NGF is not synthesized in the target field until the arrival of the earliest nerve fibres.
Sensory neurones become dependent on neurotrophic support from their peripheral and central target fields when their fibres reach these locations. The onset of dependence, at least in the case of those neurones that require NGF for survival, appears to be regulated by the expression of specific cell-surface receptors occurring as part of an intrinsic developmental programme which is independent of target field encounter. The onset of NGF synthesis (and possibly other neurotrophic factors) in sensory neurone target fields commences with the arrival of the earliest nerve fibres - whether these fibres initiate synthesis has yet to be ascertained. It is clear that synthesis occurs in cell types (e.g. epithelial cells) from which specific receptor cells later differentiate (e.g. Merkel cells) and does not exclusively occur in Schwann cells in the early stages of target field innervation. Are the cells which synthesize a particular neurotrophic factor the precursors from which specific receptors cells later differentiate and is the availability of the factor restricted to the nerve fibres that encounter these cells?
Recent experimental evidence suggests that the specific neurotrophic factor requirements of sensory neurones are related to the kinds of sensory structures and second-order neurones they innervate rather than their derivation from neural crest versus placode. The significance of this finding is that it provides a means by which a given target field can independently regulate its innervation by functionally distinct kinds of sensory neurones.
The survival of each sensory neurone appears to depend on two distinct neurotrophic factors: one from its peripheral target field, the other from its central target field. Both factors are active on sensory neurones during the same critical period of target field innervation and potentiate each others action at low, possibly physiological concentrations. The significance of this finding is that it provides the means for selectively supporting the survival of neurones that make appropriate terminations in both target fields. One implication of the operation of this mechanism is that the neurotrophic factor requirements of sensory neurones are determined prior to target field innervation. In accordance with this possibility is the recent finding that nerve fibres of the same modality tend to be clustered together in adult cutaneous nerves (Roberts & Elardo, 1986) which implies that the modality of sensory neurones is determined prior to innervation rather than induced by specific cells encountered by chance.
In the establishment of somatotopic projections of sensory receptive fields to the CNS, it is clear that the periphery provides the spatial information that organizes the central representations. Although the means by which spatial information is transferred from the periphery to the CNS is unclear, it is apparent that spatiotemporal models based on the ordered growth of nerve fibres do not provide a convincing explanation since near-neighbour relationships are not maintained between growing nerve fibres. Mechanisms based on the selective maintenance of topographically related central and peripheral terminations in an initially diffuse nerve fibre projection pattern provide a more-likely explanation. The selective stabilization of topographically related collaterals by the release of neurotrophic factors from target field cells in relation to pattern of impulse activity in the innervating neurones (Lichtman & Purves, 1981) is an attractive though, as yet, unsubstantiated hypothesis.
ACKNOWLEDGEMENTS
My thanks to Chris Wylie for many helpful suggestions. It has not been possible to include all of the relevant references in this review because of limited space; my apologies for this. Grant support by the Medical Research Council, Wellcome Trust and Nuffield Foundation is acknowledged.