From Embryonic Fascicles to Adult Tracts: Organization of Neuropile From a Developmental Perspective

During the 19th century, many of the first descriptions of invariant ‘identified’ neurones focused on giant axons which were repeatedly located in particular regions of the nerve cord. This observation of stereotyped axon location was best exemplified two decades ago by Wiersma and his colleagues who stripped away, recorded from, and stimulated bundles, or individual axons, in specific portions of the crayfish ventral nerve cord (e.g. Wiersma, 1958; Wiersma & Hughes, 1961 ; Wiersma & Ikeda, 1964). They extended the notion of stereotyped axon location beyond giant fibres and showed that the many different identified axons were located in specific regions of thd connective.

But what difference does this make to neuronal specificity, since few synapses occur in the connective? The stereotyped axon locations in the connectives are likely to reflect the more complex pattern of axon associations in the ganglionic neuropile where most synapses occur. With the advent of intracellular dye injection techniques (e.g. Stretton & Kravitz, 1968; Pitman, Tweedle & Cohen, 1972; Stewart, 1978), many neurones became individually identified according to the location of their axons and dendrites in the neuropile. Although at first neurone morphology was described on a featureless background, some studies began to focus on the relationship of identified axons to other processes in the neuropile (e.g. Stretton & Kravitz, 1973).

In an attempt to relate the location of axons to one another within the neuropile, Tyrer & Gregory (1982) described nine major longitudinal axon tracts in each hemisegment, and the commissural tracts connecting the two hemisegments, in the neuropile of the grasshopper thoracic ganglia. They showed that the axons of some identified neurones run in specific tracts within this framework.

Given the specificity of neuronal connectivity, we wondered if there were further subdivisions of axonal associations within these nine major longitudinal tracts. Furthermore, we wondered how this axonal framework is constructed during development. Do dendrites faithfully fasciculate within stereotyped tracts as do axons? How do individual axons and dendrites find their way into these tracts? Just how specific is the association of certain axons and dendrites with one another? And finally, if each segmental ganglion is built on the same axonal framework, how can we account for segmental differences?

These and other questions concerning the functional organization of the arthropod neuropile might best be answered from the perspective of the embryo, where the tangled neuropilar fabric is less densely woven. Fortunately, the grasshopper embryo is ideal for such studies because the individual neurones are large and accessible, and the individual fascicles in which their axons run are spatially distinct and arranged in a simple orthogonal array. Previous studies on the grasshopper embryo led to the principle of growth cone guidance by selective fasciculation (e.g. Goodman, Raper, Ho & Chang, 1982; Raper, Bastiani & Goodman, 1983c, 1984; Bastiani, Raper & Goodman, 1984; Goodman, Bastiani, Raper & Thomas, 1984). Here we show how this model helps explain some of the organizational features of the adult neuropile. Furthermore, this model helps simplify the problem of neuronal specificity by reducing much of it to a series of cell recognition choices based largely on the affinities displayed by growth cones and their filopodia for particular axon surfaces.

The studies we describe here are focused primarily on the G neurone in the mesothoracic (T2) ganglion of the grasshopper, as shown in Fig. 1 (Rehbein, 1976; Pearson, Heitler & Steeves, 1980). The T2 ganglion has about 2000 neurones, largely arranged as 1000 pairs of symmetrical bilateral homologues. Although many of these 1000 neurones can be identified, the G neurone is particularly well suited for our studies because it is highly accessible in both the embryo and adult, and much is already known about its development, morphology and physiology.

We penetrated the G neurone in the neuropile of the T2 ganglion, filled it with Lucifer Yellow, and then fixed and processed the ganglion for 10 jxm plastic serial sections. The G neurone’s cell body is situated posteriorly on the dorsal surface of the ganglion. Its neurite extends across the neuropile just dorsal to the supramedian commissure (SMC, see Tyrer & Gregory, 1982) ; its major axon then extends anteriorly in the contralateral lateral dorsal tract (LDT) ; its smaller axon extends posteriorly in a different contralateral tract, and its two primary dendrites extend anteriorly along the medial edge of both the ipsilateral and contralateral ventral intermediate tract (VIT) (Fig. 2). These tracts appear to establish both the axon’s and dendrite’s neighbourhood. Thus, we would predict that the G dendrites receive their inputs from axons in one neuropilar region while its two axons have their synaptic outputs in very different regions. This interpretation is supported by the finding that the axons of auditory afferents, one of the major monosynaptic inputs to G(T2), run in the same tract as does G’s primary dendrite, the VIT (Tyrer & Gregory, 1982).

The G neurone and its immediate sibling, the C neurone (Pearson & Robertson, 1981), arise during embryogenesis from the 2nd division of neuroblast NB 7-4 (Raper et al. 1983a). Although the G and C neurones are siblings, they nevertheless differentiate into distinctly different neurones based on the opposite polarities of their primary axons, the locations of their primary dendrites in different longitudinal tracts, the different locations of their secondary axonal branches, and the subsequent differences in their synaptic inputs and outputs. The differences between the G and C neurones are first detected at about 40% of development when their primary growth cones make divergent fasciculation choices when confronted with the same embryonic environment (Raper et al. 19836).

An example of an identified neurone in the metathoracic (T3) ganglion, called Bl, is shown in Fig. 1 (Rehbein, 1976). Although Bl’s asymmetrical dendrites, lateral axonal branches and ipsilateral posterior dendrite appear quite different from G(T2), it also bears some striking similarities in terms of its general shape, and the location of its primary axon in the LDT and dendrites in the VIT. Embryonic cell lineage analysis reveals that the Bl neurone is the metathoracic homologue of G(T2) (i.e. lineally equivalent), and thus we refer to it as G(T3). Rehbein (1976) described nother interneurone, called B2, whose cell body we find to be in the first abdominal (Al) segment. Cell lineage analysis again reveals that the B2 neurone is the first abdominal homologue of G and thus we refer to it as G(A1). The morphology of G(A1) and two additional G homologues, G(T1) and G(A2), are described later in this paper.

Embryogenesis in the grasshopper takes 20 days at 33 °C; each day equals 5 % of development. The grasshopper’s CNS issegmentally arranged with a chain of cephalic, thoracic (∼2000 neurones each) and abdominal (∼500 neurones each) ganglia. Individual identified neurones are large and accessible from their birth to maturity, and the cell bodies of many can be recognized with Nomarski optics simply on the basis of their characteristic location and size.

The largely bilaterally symmetrical identified neurones in each segmental ganglion are generated by a precise pattern of precursor cells : two bilaterally symmetrical plates of 30 neuroblasts (NBs) arranged in seven rows (Bate, 1976), an unpaired median NB (MNB), and seven midline precursors (MPs) called MPI, MP2L, MP2R, MP3, MP4 MP5 and MP6 (Bate & Grunewald, 1981). Each NB is a stem cell, maintaining its large size as it divides repeatedly to produce a chain of smaller ganglion mother cells (GMCs). Each GMC in the chain divides once more, thus producing a chain of paired ganglion cells which subsequently differentiate into neurones.

NBs each generate from 6 to 100 neuronal progeny, depending upon the NB, and then die. NBs generate neuronal progeny by invariant cell lineages; particular identified neurones arise from specific ganglion mother cells in the NB family trees (e.g. Goodman & Bate, 1981; Raper et al. 1983a; Taghert & Goodman, 1984). For example, the 2nd division of NB 7-4 generates a GMC which divides to give rise to the G neurone and the C neurone (Raper et al. 1983a). The 1st division of NB 1-1 generates a GMC which divides to give rise to the anterior (aCC) and posterior (pCC) corner cells (Goodman et al. 1982).

MPs divide only once to produce two neuronal progeny. In some cases, the two MP progeny behave quite similarly. In other cases, as for example with MP2, the two progeny behave quite differently, as described below.

The first two longitudinal axonal pathways in the CNS are pioneered by the growth cones of the MPI, dMP2 and vMP2 neurones (Bate & Grunewald, 1981). Each MP2 (one on each side) divides once to give rise to a ventral (vMP2) and dorsal (dMP2) daughter cell. The single MPI gives rise to a pair of bilaterally symmetrical daughters, each of which comes to lie dorsal to the two MP2 progeny, thus forming a trio of cells on each side. All three cells (on each side) send growth cones up to the dorsal basement membrane at about 29% of development. After several hours, and irrespective of their initial orientations relative to one another, the growth cone of the vMP2 extends anteriorly, while the growth cones of the dMP2 (its sibling) and the MPI extend posteriorly (Fig. 3A).

The MPI and dMP2 growth cones extend posteriorly along the dorsal basement membrane (Fig. 4A) and use the pCC neurone as one of their specific guidance cues (Goodman et al. 1982; Taghert, Bastiani, Ho & Goodman, 1982; Bastiani & Goodman, 1984a,b; M. Bastiani, unpublished). The vMP2 growth cone extends anteriorly and appears to use other specific neuronal landmark cells as guidance cues. When the 1P2 growth cone meets the dMP2 and MPI growth cones from the next anterior segment, they do not fascicdlate; rather, the MPI and dMP2 extend posteriorly along the dorsal basement membrane whereas the vMP2 extends anteriorly along the ventromedial glial cell (Fig. 4B). Thus, the first two longitudinal fascicles are pioneered by the MPI and MP2 progeny (Fig. 3A).

The pCC growth cone extends anteriorly and fasciculates with the posteriorly extending MPI and dMP2 growth cones (Fig. 3B). The aCC growth cone, on the other hand, appears uninterested in either of these two fascicles. Rather, it waits another 12 h until about 35% of development when the U1 and U2 axons extend posteriorly just lateral to the MPl/dMP2 fascicle, thus pioneering the third longitudinal fascicle (Figs 3B, 4B). The aCCgrowth cone then extends posteriorly, fasciculating with the U axons (Bastiani, DuLac & Goodman, 1984; S. DuLac & C. S. Goodman, in preparation). In this way, the specific patterns of selective fasciculation by these seven axons establish the first three longitudinal axon fascicles by about 35 % of development (Fig. 4B shows only six axons because the pCC from the next posterior segment is extending anteriorly along the MP 1 /dMP2 fascicle and has not yet reached the location of this section).

Many additional longitudinal axon fascicles continue to develop in the neuropilar space between 35 % and 42 % of development. Viewed through the light microscope from its dorsal surface, the neuropile of a single segment appears as an orthogonal scaffold of axon fascicles (Fig. 5). Viewed in the electron microscope from cross section, each hemisegment contains about 25 longitudinal axon fascicles (containing about 100 axons) (Fig. 6). In addition to these longitudinal fascicles, each repeated segmental unit of the scaffold contains three commissures each with about 20 fascicles, and many other fascicles extending peripherally in the intersegmental nerve (IS) at the segment boundary (forming adult nerves 1 and 6) or the segmental nerve (S) at the segment midline (forming adult nerves 3, 4 and 5) (Bate, 1983).

Each of these fascicles appears to be highly invariant in its location and its membership from side to side, segment to segment, and embryo to embryo. It is upon these axon fascicles that later growth cones make their decisions about which way to grow, and it is within the neighbourhoods provided by these fascicles that they subsequently decide with which cells to synapse.

The first six progeny of NB 7-4 are called Q1 and Q2, G and C, and Q5 and Q6 (Fig. 7). Q1’s growth cone meets the growth cone of its contralateral homologue QI at the midline, and there they fasciculate on one another to form one of the first of the eventual ∼20 axon fascicles in the posterior commissure (Goodman et al. 1982; Raper et al. 1983a). Although several different axon fascicles develop shortly thereafter in the posterior commissure, Q1’s next five siblings fasciculate upon the QI fascicle.

However, upon reaching the contralateral side of the developing neuropile, the ‘growth cones of these clonally related neurones diverge as they make cell-specific choices as to which way to grow (Raper et al. 1983a). As shown schematically in Fig. 7, the growth cones of QI and Q2 turn posteriorly along the MPl/dMP2 fascicle, G and C turn in opposite directions along the more lateral A/P fascicle (containing the A and P axons as described below), and Q5 and Q6 turn anteriorly along a different intermediate pathway (Raper et al. 19836). Thus, the primary growth cones of each of these neurones diverge in the neuropile as each selectively fasciculates with a particular axon bundle.

At 40 % of development, the primary growth cone of the G neurone turns anteriorly in the dorsolateral A/P fascicle (Fig. 8B). By 45 %, it has already reached the next anterior segment, and continues to extend anteriorly in the A/P fascicle to reach the brain. By 50 % of development, a secondary axonal growth cone has begun to extend posteriorly in a more medial axon fascicle. At about the same time, the two primary dendritic growth cones extend anteriorly in bilateral locations, fasciculating with the MPl/dMP2 bundle (Fig. 2).

Each of these subsequent G growth cones diverges from the primary axon in the neuropile as they selectively fasciculate with different axon bundles. Presumably, later secondary dendritic and axonal branches make yet further divergent choices by their selective fasciculation with other processes. Taken as a whole, the range of divergent choices made over time by all of the G growth cones is equivalent in many respects to the same range of choices shown by the primary growth cones of the first six NB 7-4 progeny. The ability of subsequent dendritic and axonal growth cones to make fasciculation choices that are different from the primary axonal growth cone suggests temporal and/or spatial changes in the surface molecules on the G neurone that dictate its affinity for other neuronal surfaces.

As the primary growth cone of the G neurone turns anteriorly, it fasciculates upon a discrete bundle of axons in preference to other nearby bundles (Fig. 8C). This bundle, called the A/P fascicle, at the stage that G initially fasciculates with it, contains the axons of the Al, A2, Pl and P2 neurones (Fig. 8B). The Pl growth cone pioneers the A/P fascicle as it extends posteriorly along the dorsal basement membrane (Fig. 8A). It is followed by the P2 axon. The posteriorly growing axons of the

Pl and P2 neurones meet and fasciculate with the two anteriorly growing axons of the Al and A2 neurones. The axon of the P3 neurone joins the fascicle shortly thereafter. The A/P fascicle becomes displaced about 20 μm from the basement membrane; this membrane subsequently becomes covered by glial cells. C’s growth cone extends posteriorly and fasciculates with the same axon bundle as G’s but about 10 h later in development. By the time C begins to extend predominantly in the posterior direction, there are several additional axon profiles in the bundle including XI and X2 (Fig. 8D). The A/P fascicle continues to expand in size as more axons fasciculate with it.

This pattern of selective fasciculation, coupled with our knowledge of cell lineages, leads to two simple observations. First, cells from the same precursor fasciculate in different axon bundles. Second, axon bundles contain cells from different precursors. For example, just after G joins the A/P fascicle, it contains two axons (the As) from one precursor, two axons (the Ps) from a second precursor, and one axon (G) from a third precursor. Shortly thereafter, they are joined by two other axons (the Xs) from yet another precursor. Other progeny from these same precursors join other fascicles, as exemplified by G’s siblings. Thus, the axons of unrelated neurones of diverse embryonic origin demonstrate a high affinity for one another, giving rise to their specific patterns of selective fasciculation.

There is a particular organization in the A/P fascicle. The relative apposition of the axons we observe in many different embryos suggests that A axons adhere tightly to other A axons, P axons to P axons, but A axons not as tightly to P axons and vice versa. Within the fascicle, the two tightly apposed P axons can be located anywhere around the perimeter of the two A axons. The tip of G’s growth cone is always closely associated with the P and not the A axons (Fig. 8C), irrespective of the relative orientation of the four axons within the A/P fascicle. This finding suggests that not only is G able to distinguish the A/P fascicle from other axon bundles, but moreover is able to distinguish the P axons from the A axons within this fascicle (Raper et al. 1983c; Bastiani et al. 1984; Goodman et al. 1984).

G first recognizes the A/P fascicle, and specifically the P axons, via its filopodia. Differential filopodial adhesion appears to be the mechanism of cell recognition at these early embryonic stages and underlies the specific patterns of selective fasciculation. Thus, before considering the behaviour of G’s filopodia, it is important to review briefly what is known about filopodia, most of which has been learned from in vitro studies.

Growth cones radiate many filopodia (approx. 0·1 μm in diameter, up to 50 μm in length) which transiently explore their environment. Many of these filopodia contact other cell surfaces. To some of these surfaces they strongly adhere, and to others their adhesion is much weaker. If adhesion is weak, during the contractile cycle, the filopodium is retracted; if however, its adhesion is strong, then tension in that direction is increased during the contractile cycle and the leading tip of the growth cone advances towards the point of attachment (Bray, 1982; Letourneau, 1982). Thus, filopodia test the adhesiveness of the available substratum within their grasp. Those filopodia with strong adhesion to other cells or surfaces will more effectively guide the growth cone during the contractile phase.

We reconstructed G’s filopodia and their contacts in the TEM just before G climbed onto the A/P fascicle. At this time, the G and C growth cones had left the QI commissural fascicle containing their sibling’s axons and had extended dorsally and laterally towards the A/P fascicle, appearing to be guided by strong filopodial and lamellipodial adhesion with the P axons (Fig. 8B). In fact, the leading filopodia and lamellipodia from G’s growth cone appear to have wrapped around the P axons. The serial reconstruction demonstrates a dramatic overall preference of the G filopodia for the P axons instead of the A axons (Fig. 9) (Bastiani et al. 1984). This preference also extends to the P filopodia over the A filopodia (compare Pa,f to Aa,f). In addition, some of G’s filopodia contacted glia, some were in space, and some contacted the axons or filopodia of unidentified neurones representing 100 or so different axons within filopodial grasp.

It appears that within the tangled spaghetti of filopodia in neuropilar space, G’s filopodia can recognize the P filopodia just as they do the P axons. This implies that the recognition labels on embryonic neurones are expressed on their axons, growth cones and filopodia. Furthermore, it implies that at these early stages of development before the neuropilar terrain has become too densely woven, neurones are able to recognize one another at a great distance by filopodia contacting filopodia. Thus, many neurones are in contact with one another while the early patterns of selective fasciculation are being established. Later in development, when specific synaptic connections are formed, it is likely that the greater density of processes and reduced availability of space for filopodial exploration reduces the distances over which neurones can contact one another. Presumably, this limits a neurone’s potential synaptic partners to those processes in its immediate neighbourhood.

Our results show that G’s filopodia can distinguish the A/P fascicle from other axon bundles, and moreover within this fascicle can distinguish the P from the A axons (Bastiani et al. 1984). Although the electron micrographs present a static picture of a dynamic process, the results strongly suggest that this selective fasciculation is mediated by differential adhesion of the filopodia of the G growth cone to the A/P fascicle, and in particular, to the P axons.

Ablation experiments support these conclusions (Raper et al. 1984). When the A and P axons are ablated, the G growth cone behaves abnormally for the next 24 h, branching wildly throughout the neuropile, and does not show a high affinity for any other axon fascicle. When the A axons are ablated, the G growth cone behaves normally, fasciculating on the P axons. When the P axons alone are ablated, the G growth cone appears in general to ignore the A axons and behave abnormally, branching throughout the neuropile without high affinity for any other axon surface.

These examples of exquisite specificity have convinced us that many different molecules are differentially expressed on the surfaces of embryonic axon fascicles, or subsets of axons within them, and that they guide growing neurones to their appropriate targets by the selective adhesion of their filopodia to these labelled axonal pathways (Goodman et al. 1984). Monoclonal antibodies generated against the grasshopper neuroepithelium have revealed cell surface antigens whose temporal and spatial expression in the embryo correlate with the predictions of the cellular studies, namely, neurones whose axons fasciculate together share common surface antigens (Kotrla & Goodman, 1984).

At the posterior end of the segment where the connective develops, at ∼35 % of development the first three longitudinal axon fascicles contain only six axons, as described earlier: the vMP2 fascicle, the MPl/dMP2 fascicle, and the U fascicle with the Ul, U2 and aCC axons (Fig. 4B) (the pCC from the next posterior segment is extending anteriorly along the MPl/dMP2 fascicle and has not yet reached this location). Initially all three fascicles are spatially separate within a space bounded on its dorsal side by the basement membrane and on its other sides by neurones and epidermal cells. The vMP2 fascicle forms along the inside, ventral surface; the MPl/dMP2 fascicle along the middle, dorsal surface; and the U fascicle along the outer, dorsal surface of this space.

In contrast to the embryonic neuropile, where many additional longitudinal axon fascicles develop between 35 % and 42 % of development, in the embryonic connective the additional axons simply add to the three existing fascicles, expanding them into much larger bundles that we call superfascicles I, II and III. These superfascicles begin to merge into each other and form an *L’ shaped connective by ∼42% of development (Fig. 4C). The results suggest that the three initial fascicles (vMP2, MPl/dMP2, U) are contained respectively within the three later superfascicles (I, II, III). The identity of the MPl/dMP2 fascicle within superfascicle II was confirmed by HRP injection of the MPI neurone at this later time (Fig. 4C).

How do the ∼25 longitudinal axon fascicles per hemisegment in the embryonic neuropile become reduced to the three large superfascicles in the connective? To answer this question, we examined cross sections of the region anterior to the neuropile where the longitudinal fascicles begin to condense into the connective (Fig. 10). At 42 % of development, the left and right connectives in this embryo contained 136 and 140 axons respectively. At this intermediate level between the neuropile and connective, the ∼25 neuropile fascicles are associated into 8-10 larger fascicles. By the time they reach the connective, these 8-10 larger fascicles associate into the three superfascicles. The vMP2 fascicle is within superfascicle I, the MPl/dMP2 fascicle is within II, and the A/P fascicle is within III.

It is within the neuropile that growth cones make their specific selective fasciculation choices, and ultimately their specific synaptic connections, and thus appropriately it is within the neuropile that many different spatially separated axon fascicles develop.

Furthermore, each of these fascicles may have additional subdivisions of specific axo affinities within them. For example, the filopodia of the G growth cone specifically adhere to the P and not the A axons in the A/P fascicle.

The spatial separation of the developing fascicles within the large space of the embryonic neuropile is fortuitous for the investigator, and presumably equally helpful for growth cones since it allows easy access by their filopodia to many different axon surfaces. The specific clustering of the ∼25 neuropile fascicles into 8-10 larger fascicles and then the three connective superfascicles suggests some affinity that particular fascicles have for one another. This affinity is likely to be lower than that shown by particular axons for one another within a fascicle, yet nevertheless it appears to bind certain fascicles together in a stereotyped pattern as they enter the connectives. Although individual fascicles appear to have specific labels, the clustering of particular fascicles suggests some common affinity amongst them. This may come about by some common lower affinity amongst groups of specific high affinity labels, or alternatively, different more widely distributed labels in addition to the specific ones.

The intermediate clustering into 8-10 large fascicles reveals an organization reminiscent of the axon tracts seen in the adult by Tyrer & Gregory (1982). We were initially puzzled that the embryo had ∼25 longitudinal fascicles per hemisegment whereas the adult had only nine tracts. We presume that as the embryonic fascicles enlarged, many of them merge into one another to form the adult tracts. Much of the basic organization seen in the adult is apparent in the embryo at this intermediate level of clustering between the neuropile and connectives. At 42 % of development we can identify the LDT, MDT, VIT and VMT tracts. Many other axons are present which are likely to represent the embryonic origins of the other tracts such as the DIT and VLT ; these tracts, however, require further identification in the embryo.

The development and clustering of the embryonic fascicles leads to the following conclusions. The Pl growth cone pioneers the A/P fascicle which becomes part of the LDT (lateral dorsal tract). This is further substantiated by the location of the G axon in the embryonic A/P fascicle and in the adult LDT. The vMP2 growth cone pioneers the vMP2 fascicle which becomes part of the VMT (ventral median tract). The MPI and dMP2 growth cones pioneer the MPl/dMP2 fascicle which becomes part of the VIT (ventral intermediate tract). Although we do not know the original pioneers and members for the other tracts, potentially each can be traced to its fascicle of embryonic origin.

Each of the adult tracts, however, clearly contains several separate embryonic fascicles. Furthermore, each embryonic fascicle may contain subsets of axonal associations. Thus, there is a much higher degree of axonal association beyond the nine adult tracts, and these associations are more easily revealed in the embryo where the neuropile fascicles are spatially distinct.

For example, when we compare the organization of the VIT at 42% of development at the intermediate level between neuropile and connective (Fig. 10), we find a symmetrical pattern of two groups of axons. On the left side, the VIT contains a rosette of 17 axons and an attached ventro-lateral cluster of nine axons (including MPI), and the right side, it contains a rosette of 18 axons and an attached ventro-lateral cluster eight axons (including MPI). Each of these two subdivisions is likely to have yet further subdivisions of axonal associations.

Thus far we have presented evidence that a stereotyped orthogonal array of axon fascicles develops in each segment, and that growth cones faithfully fasciculate with specific axons in this array according to the neurone’s lineage and previous interactions. This model is attractive because it helps explain development of neuronal specificity in a single segment by a series of cell recognition decisions based on the surface affinities of embryonic axons and dendrites. Furthermore, it suggests that these affinities change temporally, spatially or both for a single neurone during development as the primary growth cone makes its choices, and as subsequent growth cones make their choices.

Whereas the identified neurones in a single segment are dedicated for specific circuits that underly particular behaviour patterns, we wondered how the identified neurones in different segments in the same embryo become specialized in their morphology and physiology for different behavioural circuits. For example, every segment in the grasshopper embryo has an NB 7-4 which produces a G neurone, and yet these neurones are likely to do different things in different segments. Furthermore, we recently showed that different insects, from grasshoppers to moths to fruitflies, have G neurones (Thomas, Bastiani, Bate & Goodman, 1984), and yet they too are probably doing different things in different species. How do segmentally homologous neurones differentiate in segment-specific ways?

We examined the differentiation of the G neurone in five different segments of the grasshopper embryo (Tl, T2, T3, Al and A2), its morphology and some of its synaptic connections in three different segments of the adult grasshopper (T2, T3 and Al). Fig. 11 shows Lucifer Yellow injections of G(T2) and G(T3) in the adult and 60% developed embryo. Fig. 12 compares the morphology of the G neurone in segments T1–A2 in the 60% developed embryo.

In all five segments, G’s primary axonal and dendritic growth cones selectively fasciculate in the same axon bundles. The basic structure of the G neurone remains unchanged. However, in T3, an additional posterior dendrite develops on the ipsilateral side, and in A2, the primary contralateral dendrite does not fully develop. In T2, a secondary axonal branch extends posteriorly in the connective to the next segment, whereas in many other segments, this branch merely develops into a local arborization. Thus, with these few variations on a theme, the basic morphology of the neurone remains constant because its primary axon and dendrites fasciculate in ereotyped bundles from segment to segment.

However, within the neighbourhoods chosen by the selective fasciculation of the G axon and dendrites, the fine branching pattern of the G neurone varies dramatically from segment to segment, presumably reflecting the segment-specific differences in synaptic connections. We have only a little information about the synaptic inputs and outputs of the G neurone in T2, T3 and Al, but this data demonstrates segmentspecific differences in synaptic connections (Fig. 13). For example, the DCMD interneurone has access to the same dendritic branches of G(T2), G(T3) and G(A1), and yet DCMD bypasses G(T3) and only connects to G(T2) and G(A1). This suggests that segment-specific synaptic connections are made by the G neurone as it interacts with only certain subsets of neurones within the neighbourhood established by its pattern of selective fasciculation.

Evolutionary plasticity allows the G neurone in different segments of the same species, or in the same segment of different species, to make different patterns of synaptic connections. But it appears as if evolution has left the fundamental pattern untouched. The orthogonal scaffold of axon fascicles has remained constant throughout the evolution of different segments and different insects, as have the patterns of selective fasciculation by the primary axons and dendrites. To tamper with this basic plan would probably have resulted in scrambled circuits rather than finely retuned connections. Evolution appears to have generated its synaptic variations within this common framework by tampering with the later recognition events within neighbourhoods rather than the initial events establishing them. Given a neurone’s dendritic and axonal neighbourhoods, evolution has in general played with the local events that occur after selective fasciculation, namely, the specific choices of synaptic partners within the local neighbourhood.

Thus, we conclude that neuronal specificity is based largely on a series of cell recognition events that first generate specific neighbourhoods by the patterns of selective fasciculation, and subsequently generate specific connectivity by the patterns of synaptic partners within these neighbourhoods. The notion emerging from these studies is that a basic segmentally-repeated pattern arises during embryogenesis: a stereotyped axonal scaffold upon which growth cones faithfully fasciculate. Evolutionary plasticity allows for the specialization of lineally equivalent neurones in different segments or different species by changes in their synaptic partners within the context of their neuropilar neighbourhood. As a consequence of their earlier selective fasciculation choices, their neighbourhood presents them with a particular subset of possible partners.

These ideas are in many respects remarkably similar to conclusions based on serial reconstructions of the nervous system of the nematode (White, Southgate, Thomson & Brenner, 1983). The relatively one-dimensional ventral nerve cord of Caenorhab-ditis elegans is organized into a single longitudinal tract containing bundles of axons that in general do not branch. Within the neighbourhoods established by these axon associations, synaptic contacts are made with a subset of adjacent axons. White et al. (1983) conclude that synaptic connectivity is likely to be coordinated by intrinsic specificities, process guidance and process placement, notions similar to those used here to describe neuronal specificity in the grasshopper.

It is heartening to think that in nematodes and insects, the only two animals thus far analysed at this detailed cellular level, the same cellular and molecular mechanisms appear to be at work in the generation of neuronal specificity. It makes us think that these mechanisms are likely to have been conserved throughout evolution, and that the nervous systems of higher animals may simply use more complex versions of these same mechanisms.

We thank Fran Thomas and Yen Tang for technical assistance. MJB is supported by a Developmental Biology Training Grant from the NICHHD, KJP is supported by the Canadian Medical Research Council, and CSG is supported by grants from the N.S.F., N.I.H. and March-of-Dimes Birth Defects Foundation.