55: Formation and Elimination of Synapses

• Recognition of Synaptic Targets Is Specific

  • Recognition Molecules Promote Selective Synapse Formation

  • Different Synaptic Inputs Are Directed to Discrete Domains of the Postsynaptic Cell

  • Neural Activity Sharpens Synaptic Specificity

• Principles of Synaptic Differentiation Are Revealed at the Neuromuscular Junction

  • Differentiation of Motor Nerve Terminals Is Organized by Muscle Fibers

  • Differentiation of the Postsynaptic Muscle Membrane Is Organized by the Motor Nerve

  • The Nerve Regulates Transcription of Acetylcholine Receptor Genes

  • The Neuromuscular Junction Matures in a Series of Steps

• Central Synapses Develop in Ways Similar to Neuromuscular Junctions

  • Neurotransmitter Receptors Become Localized at Central Synapses

  • Synaptic Organizing Molecules Pattern Central Nerve Terminals

  • Glial Cells Promote Synapse Formation

• Some Synapses Are Eliminated After Birth

• An Overall View

So far we have examined three stages in the development of the mammalian nervous system: the formation and patterning of the neural tube, the birth and differentiation of neurons and glial cells, and the growth and guidance of axons. One additional step must occur before the brain becomes functional: the formation of synapses. Only when synapses are formed and functional can the brain go about the business of processing information.

Three key processes drive synapse formation. First, axons make choices among many potential postsynaptic partners. By forming synaptic connections only on particular target cells, neurons assemble functional circuits that can process information. Usually synapses must be formed at specific sites on the postsynaptic cell; some axons form synapses on dendrites, others on cell bodies, and yet others on axons or nerve terminals. Cellular and subcellular specificity are evident throughout the brain, but we will illustrate the general features of synapse formation with a few well-studied classes of neurons.

Second, after cell-cell contacts have formed, the portion of the axon that contacts the target cell differentiates into a presynaptic nerve terminal, and the domain of the target cell contacted by the axon differentiates into a specialized postsynaptic apparatus. Precise coordination of pre- and postsynaptic differentiation depends on interactions between the axon and its target cell. Much of what we know about these interactions comes from studies of the neuromuscular junction, the synapse between motor neurons and skeletal muscle fibers. The simplicity of this synapse made it a favorable system to probe the structural and electrophysiological principles of chemical synapses, and this simplicity has also helped in the analysis of developing synapses. In this chapter we use the neuromuscular synapse to illustrate key features of synaptic development, and we also apply insights from this peripheral synapse to examine the synapses that form in the central nervous system.

Finally, once formed, synapses continue to mature, often undergoing major rearrangements. One striking aspect of later development is the wholesale elimination of a large fraction of synapses, a process that is usually accompanied by the growth and strengthening of surviving synapses. Like neuronal cell death (see Chapter 53), synapse elimination is a puzzling and seemingly wasteful step in neural development. It is increasingly clear, however, that it plays a key role in refining initial patterns of connectivity. We will discuss the main features of synapse elimination at the neuromuscular junction, where it has been studied intensively, as well as at synapses between neurons, where it also is prominent.

Throughout this chapter we emphasize the interplay of molecular programs and neural activity in shaping synaptic patterns. Synapse formation stands at an interesting crossroads in the sequence of events that assemble the nervous system. The initial steps in this process appear to be "hardwired" by molecular programs. However, as soon as synapses form, the nervous system begins to function, and the activity of neural circuits plays a critical role in subsequent development. Indeed, the information-processing capacity of the nervous system is refined through its use, most dramatically in early postnatal life but also into adulthood. In this sense the nervous system continues to develop throughout life. This discussion will be a useful prelude to Chapter 56, where we discuss how genes and the environment—nature and nurture—interact to customize nervous systems early in postnatal life.

Once axons reach their designated target areas they must choose appropriate synaptic partners from the many potential targets within easy reach. Although synapse formation is thought to be a selective process at both cellular and subcellular levels, few of the molecules that confer synaptic specificity have been identified.

Recognition Molecules Promote Selective Synapse Formation

The specificity of synaptic connections is particularly evident when intertwined axons select subsets of target cells. In these cases axon guidance and selective synapse formation can be distinguished. The first report of such specificity came more than 100 years ago when J. N. Langley, studying the autonomic nervous system, proposed the first version of a chemospecificity hypothesis (see Chapter 54).

Langley observed that autonomic preganglionic neurons are generated at distinct rostrocaudal levels of the spinal cord. Their axons enter sympathetic ganglia together but form synapses with different postsynaptic neurons that innervate distinct targets. Using behavioral assays as a guide, Langley inferred that the axons of preganglionic neurons located in the rostral spinal cord form synapses on ganglion neurons that project their axons to relatively rostral targets such as the eye, whereas neurons that derive from more caudal regions of the spinal cord synapse on ganglion neurons that project to caudal targets such as the ear (Figure 55–1A). He then showed that similar patterns were reestablished after the preganglionic axons were severed and allowed to regenerate, leading him to postulate that some sort of molecular recognition was responsible (Figure 55–1B, C).

Electrophysiological studies later confirmed Langley's intuition about the specificity of synaptic connections in these ganglia. Moreover, this selectivity is apparent from early stages of innervation, even though specific types of postsynaptic neurons are interspersed within the ganglion. The reestablishment of selectivity in adults after nerve damage shows that specificity does not emerge through peculiarities of embryonic timing or neuronal positioning.

To illustrate the idea of target specificity in more detail we will consider the axons of developing retinal ganglion cells. These neurons differ in their response properties—some ganglion neurons respond to increases in light level (ON cells), others to decreases (OFF cells), others to moving objects, and still others to light of a particular color. The axons of all ganglion cells run through the optic nerve, forming parallel axonal pathways from the retina to the brain.

The response properties of each class of ganglion neurons depend on the synaptic inputs they receive from amacrine and bipolar interneurons, which in turn receive synapses from light-sensitive photoreceptors. All of the synapses from bipolar and amacrine cells onto ganglion cell dendrites occur in a narrow zone of the retina called the inner plexiform layer (see Chapter 26). Axons and dendrites therefore have the daunting task of recognizing their correct partners within a large crowd of inappropriate bystanders.

In the inner plexiform layer the processes of different amacrine and bipolar cell types, as well as the dendrites of functionally distinct ganglion cell types, branch and synapse in different sublayers. The dendrites of ON and OFF cells are restricted to inner and outer portions of the plexiform layer, respectively, and therefore receive synapses from different interneurons (Figure 55–2). There are a dozen sublayers within the inner plexiform layer, each with its own cohort of synapses. Similar lamina-specific connections are found in many other regions of the brain and spinal cord. For example, in the cerebral cortex distinct populations of axons confine their dendritic arbors and synapses to just one or two of the six main layers.

One clue to the establishment of lamina-specific synapses in the retina comes from the finding that subsets of ganglion neurons in the chick embryo express different classes of immunoglobulin-like adhesion molecules (see Chapter 54). These proteins promote homophilic interactions, that is, they bind to the same protein on other cell surfaces. Moreover, the amacrine and bipolar cells that contact a particular ganglion cell type in a specific sublamina often express the same adhesion molecules as their target cell (Figure 55–2B). When the characteristic adhesion molecule of a chick retinal ganglion neuron is removed, the cell's dendrites are no longer confined to a specific lamina. Conversely, the dendrites shift to a new lamina when they are forced to express a new immunoglobulin adhesion molecule. Thus a "like-binds-like" recognition system appears to contribute to this instance of synaptic specificity.

A different type of specificity is evident in the olfactory system. In the olfactory epithelium each neuron expresses just one of approximately 1,000 types of receptors. Neurons expressing one receptor are randomly distributed across a large sector of the epithelium, yet all of their axons converge on the dendrites of just a few target neurons in the olfactory bulb, forming synapse-rich glomeruli (Figure 55–3A). When an individual olfactory receptor is deleted, the axons that normally express the receptor reach the olfactory bulb but fail to converge into specific glomeruli or to terminate on the appropriate postsynaptic cells (Figure 55–3B). Conversely, when neurons are forced to express a different odorant receptor, their axons form glomeruli at a different position within the olfactory bulb (Figure 55–3C).

Together these experiments suggest that olfactory receptors not only determine a neuron's responsiveness to specific odorants but also help the axon to form appropriate synapses on target neurons. Specific olfactory receptors may serve as recognition molecules, but it seems more likely that they influence the expression and activity of other receptor systems that contribute to the guidance of olfactory axons. In support of this view, recognition molecules have been found to regulate the interactions of olfactory axons with each other and with their targets.

Different Synaptic Inputs Are Directed to Discrete Domains of the Postsynaptic Cell

Nerve terminals not only discriminate among candidate targets, they also terminate on a specific portion of the target neuron. In many areas of the brain axons arriving in layered structures often confine their terminals to one layer, even if the dendritic tree of the postsynaptic cell traverses numerous layers. In the cerebellum the axons of different types of neurons terminate on distinct domains of the Purkinje neurons. Granule cell axons contact distal dendritic spines, climbing fiber axons contact proximal dendritic shafts, and basket cell axons contact the axon hillock and initial segment (Figure 55–4).

Such specificity presumably relies on molecular cues on the postsynaptic cell surface. For Purkinje neurons of the cerebellum one such cue is neurofascin, an adhesion molecule of the immunoglobulin superfamily. Neurofascin is present at high levels on the axonal initial segment and directs basket cells to form axons selectively on this axonal domain. Adhesion molecules can therefore also serve as recognition molecules for particular domains of a neuron. Since individual neurons can form synapses with several classes of pre- and postsynaptic cells, it follows that each neuronal subtype must express a variety of synaptic recognition molecules.

Neural Activity Sharpens Synaptic Specificity

So far we have emphasized the role of recognition molecules in the formation of synapses. Once synapses form, neural activity within the circuit plays a critical role in refining synaptic patterns. In the retinotectal system interactions between ephrins and Eph kinases result in formation of a crude retinotopic map in the tectum (see Chapter 54).

But activity-dependent processes refine the axonal arbors of retinal ganglion cells, thus sharpening the tectal map (Figure 55–5). The axons of retinal ganglion neurons initially form broad diffuse arbors, which gradually become denser but more focused. This refinement is inhibited when the activity of synapses is blocked. Likewise, dendritic arbors of some retinal ganglion cells initially span multiple sublaminae in the inner plexiform layer, then become restricted to narrow ones. The molecular mechanisms of this activity-dependent refinement are largely unknown. One idea is that the level and pattern of neuronal activity regulates the expression of recognition molecules.

These examples illustrate a widespread phenomenon: Molecular cues control initial specificity but, once the circuit begins to function, specificity is sharpened through neural activity. In the two cases mentioned here, sharpening involves loss of synapses. We will return to the process of synapse elimination at the end of this chapter, and consider its consequences for behavior in the next chapter.

In a few cases neural activity can turn an inappropriate target into an appropriate one. This mechanism has been most clearly demonstrated in skeletal muscle, where mammalian muscle fibers can be divided into several categories according to their contractile characteristics. Muscle fibers of particular types express genes for distinctive isoforms of the main contractile proteins, such as myosins and troponins.

Few muscles are composed exclusively of a single type of fiber; most have fibers of all types. Yet the branches of an individual motor axon innervate muscle fibers of a single type, even in "mixed" muscles in which fibers of different types are intermingled (Figure 55–6A). This pattern, sometimes termed motor unit homogeneity, implies a remarkable degree of synaptic specificity. However, matching does not come about solely through recognition in the motor axon of the appropriate type of muscle fiber. The motor axon can also convert the target muscle fiber to an appropriate type. When a muscle is deinnervated at birth, before the properties of its fibers are fixed, a nerve that normally innervates a slow muscle can be redirected to reinnervate a muscle destined to become fast, and vice versa. Under these conditions the contractile properties of the muscle are partially transformed in a direction imposed by the firing properties of the motor nerve (Figure 55–6B, C).

Different patterns of neural activity are responsible for the switch in muscle properties. Most strikingly, direct electrical stimulation of a muscle with patterns normally evoked by slow or fast nerves leads to changes that are nearly as dramatic as those produced by cross-innervation (Figure 55–6D). Although activity-based conversion of the type used at the neuromuscular junction is unlikely to be a major contributor to synaptic specificity in the central nervous system, it raises the possibility that central axons can modify the properties of their synaptic targets, contributing to the diversification of neuronal subtypes and refining connectivity imposed by recognition molecules.

The neuromuscular junction comprises three types of cells: a motor neuron, a muscle fiber, and Schwann cells. All three types are highly differentiated in the region of the synapse.

The process of synapse formation is initiated when a motor axon, guided by the multiple factors described in Chapter 54, reaches a developing skeletal muscle and approaches an immature muscle fiber. Contact is made and the process of synaptic differentiation begins. As the growth cone begins its transformation into a nerve terminal, the portion of the muscle surface opposite the nerve terminal begins to acquire its own specializations. As development proceeds, synaptic components are added and structural signs of synaptic differentiation become apparent in the pre- and postsynaptic cells and in the synaptic cleft. Eventually the neuromuscular junction acquires its mature and complex form (Figure 55–7A, B).

Three general features of neuromuscular junction development have provided clues about the molecular mechanisms that underlie synapse formation. First, nerve and muscle organize each other's differentiation. In principle, the precise apposition of pre- and postsynaptic specializations might be explained by independent programming of nerve and muscle properties. However, in muscle cells cultured alone acetylcholine (ACh) receptors are generally distributed uniformly on the surface, although some are clustered as in mature postsynaptic membranes. Yet when motor neurons are added to the cultures they extend neurites that contact the muscle cells more or less randomly, instead of seeking out the ACh receptor clusters. New receptor clusters appear precisely at the points of contact with the presynaptic neurites while preexisting uninnervated clusters eventually disperse (Figure 55–8). Thus factors on or released by motor axons exert a profound influence on the synaptic organization of the muscle cell. Likewise, muscles signal retrogradely to motor nerve terminals. When motor neurons in culture extend neurites, they assemble and transport synaptic vesicles, some of which form aggregates similar to those found in nerve terminals. When the neurites contact muscle cells, new vesicle clusters form opposite the muscle membrane, and most of the preexisting clusters disperse.

Second, as these studies showed, motor neurons and muscle cells can synthesize and arrange most synaptic components without each other's help. Uninnervated myotubes can synthesize functional ACh receptors and gather them into high-density aggregates. Likewise, motor axons can form synaptic vesicles and cluster them into varicosities in the absence of muscle. In fact, vesicles in growth cones can synthesize and release ACh in response to electrical stimulation, before the growth cone has reached its target cells. Thus the developmental signals that pass between nerve and muscle do not induce wholesale changes in cell properties; rather they assure that components of the pre- and postsynaptic machinery are organized at the correct time and in the right places. It is useful therefore to think of the intercellular signals that control synaptogenesis as organizers rather than inducers.

A third key feature of neuromuscular junction development is that new synaptic components are added in several distinct steps. The newly formed synapse is not simply a prototype of a fully developed synapse. Although nerve and muscle membrane form close contacts at early stages of synaptogenesis, only later does the synaptic cleft widen and the basal lamina appear. Similarly, ACh receptors accumulate in the postsynaptic membrane before acetylcholinesterase accumulates in the synaptic cleft, and the postsynaptic membrane acquires junctional folds only after the nerve terminal has matured. Several different axons innervate each myotube around the time of birth, but during early postnatal life all but one axon withdraws.

This elaborate sequence is unlikely to be orchestrated by the simple act of contact between nerve and muscle. More probably, multiple signals pass between the cells—the nerve sends a signal to the muscle that triggers the first steps in postsynaptic differentiation, at which point the muscle sends a signal that triggers the initial steps of nerve terminal differentiation. The nerve then sends further signals to the muscle, and this interaction continues.

We now consider retrograde (from muscle to nerve) and anterograde (from nerve to muscle) organizers in more detail.

Differentiation of Motor Nerve Terminals Is Organized by Muscle Fibers

Soon after the growth cone of a motor axon contacts a developing myotube, a rudimentary form of neurotransmission begins. The axon releases ACh in vesicular packets, the transmitter binds to receptors, and the muscle responds with depolarization and weak contraction.

The onset of transmission at the new synapse reflects the intrinsic capabilities of each synaptic partner. Nevertheless, these intrinsic capabilities cannot readily explain the marked increase in the rate of transmitter release that occurs after nerve-muscle contact is made, nor can they explain the accumulation of synaptic vesicles and the assembly of active zones in the small portion of the motor axon that contacts the muscle surface. These developmental steps require signals from muscle to nerve.

A clue to the source of these signals came from studies on the reinnervation of adult muscle. Although axotomy leaves muscle fibers denervated and leads to insertion of ACh receptors in nonsynaptic regions, the postsynaptic apparatus remains largely intact. It is still recognizable by its synaptic nuclei, junctional folds, and the ACh receptors, which remain far more densely packed in synaptic areas than in extrasynaptic areas of the cell. Damaged peripheral axons regenerate readily (unlike those in the central nervous system) and form new neuromuscular junctions that look and perform much like the original ones.

A century ago, Fernando Tello y Muñóz, a student of Santiago Ramón y Cajal, noted that the new junctions form at preexisting synaptic sites on the de-nervated muscle fibers even though the postsynaptic specializations occupy only 0.1% of the muscle fiber surface (Figure 55–9A). Later, electron microscopy showed that specialization in the axon occurs only in the terminals that contact the muscle. For example, active zones form directly opposite the mouths of the postsynaptic junctional folds. These findings imply that motor axons recognize signals associated with the postsynaptic apparatus.

When regenerating axons reach a muscle fiber they encounter the basal lamina of the synaptic cleft. To explore the significance of this association, muscles were damaged in vivo in a way that killed the muscle fibers but left their basal lamina intact. The necrotic fibers were phagocytized, leaving behind basal lamina sheaths on which synaptic sites were readily recognizable. At the same time that the muscle was damaged the nerve was cut and allowed to regenerate. Under these conditions motor axons reinnervated the empty basal lamina sheaths, contacting synaptic sites as precisely as they would have if muscle fibers were present. Moreover, nerve terminals developed at these sites and active zones even formed opposite struts of basal lamina that once lined junctional folds. These observations implied that components of the basal lamina organize presynaptic specialization (Figure 55–9B).

Several such molecular organizers have now been identified. Among the best studied are isoforms of the protein laminin. Laminins are major components of all basal laminae and promote axon outgrowth in many neuronal types (see Chapter 54). They are heterotrimers of α, β, and γ chains, comprising a family of at least five α, four β, and three γ chains. Muscle fibers synthesize multiple laminin isoforms and incorporate them into the basal lamina. Laminin-211, a heterotrimer containing the α2, β1, and γ1 chains, is the major laminin in the basal lamina, and its absence leads to severe muscular dystrophy. In the synaptic cleft, however, isoforms bearing the β2 chain predominate (Figure 55–10A).

In vitro motor axons that encounter a deposit of β2-containing laminin stop growing, accumulate synaptic vesicles, and acquire the ability to release neurotransmitter. Conversely, the development of nerve terminals and Schwann cells is perturbed in mutant mice that lack the β2 laminin (Figure 55–10B). These laminins appear to act by binding to voltage-sensitive calcium channels that reside in the axon terminal membrane, where they couple activity to transmitter release. Laminins act on the extracellular domain of the channels whereas the intracellular segment recruits or stabilizes other components of the release apparatus.

Because presynaptic specialization proceeds to some extent in the absence of laminins, additional retrograde organizers of axonal specialization must exist. Among these are members of the fibroblast growth factor and collagen IV families, both produced by muscle cells. Thus target-derived proteins from multiple families collaborate to organize the presynaptic nerve terminal.

Differentiation of the Postsynaptic Muscle Membrane Is Organized by the Motor Nerve

Soon after myoblasts fuse to form myotubes, the genes that encode ACh receptor subunits are activated. Receptor subunits are synthesized, assembled into pentamers in the endoplasmic reticulum, and inserted into the plasma membrane. As noted above, some receptors spontaneously form aggregates, but the majority are distributed throughout the membrane at a low density, approximately 1,000 per μm².

Once synapse formation is complete, however, the distribution of the receptors changes drastically. The receptors become concentrated at the synaptic sites of the membrane (to a density up to 10,000 per μm²) and depleted in the nonsynaptic membrane (reduced to 10 per μm² or less). This thousand-fold difference in ACh receptor density occurs a few tens of micrometers from the edge of the nerve terminal.

Appreciation of the critical role of the nerve in the redistribution of ACh receptors inspired a search for factors that might promote their clustering. This quest led to the discovery of a proteoglycan, agrin. Agrin is synthesized by motor neurons, transported down the axon, released from nerve terminals, and incorporated into the synaptic cleft (Figure 55–11A, B). Some agrin isoforms are also made by muscle cells, but the neuronal isoforms are about a thousand-fold more active in aggregating ACh receptors.

The phenotype of mutant mice lacking agrin shows that agrin has a central role in the organization of ACh receptors. Agrin mutants have grossly perturbed neuromuscular junctions and die at birth. The number, size, and density of ACh receptor aggregates are severely reduced in these mice (Figure 55–11C). Other components of the postsynaptic apparatus—including cytoskeletal, membrane, and basal lamina proteins—are also reduced. Interestingly, the differentiation of presynaptic elements is also perturbed. However, the defects in the presynaptic element do not result directly from lack of agrin in the motor neuron, but rather indirectly from the failure of the disorganized postsynaptic apparatus to generate signals for presynaptic specialization.

How does agrin work? Agrin's major receptor is a complex of a muscle-specific tyrosine kinase called MuSK (muscle-specific trk-related receptor with a kringle domain) and a coreceptor subunit called LRP4 (Figure 55–11A). MuSK is normally concentrated at synaptic sites in the muscle membrane, and muscles of mutant mice lacking MuSK do not have ACh receptor clusters (Figure 55–11C). Myotubes generated in vitro from these mutants express normal levels of ACh receptors, but these receptors cannot be clustered by agrin. MuSK therefore appears to be a critical component of the receptor for agrin. LRP4 functions together with MuSK, an adaptor protein Dok-7, and a cytoplasmic protein rapsyn. LRP4 forms a complex with MuSK and binds agrin efficiently. Dok-7 binds MuSK and signals to rapsyn, which is also necessary for ACh receptor clustering. Rapsyn is co-localized with ACh receptors in vivo, is present at ACh receptor clusters soon after they form, and can induce the aggregation of ACh receptors in vitro. In mice lacking rapsyn muscles form normally and ACh receptors accumulate in normal numbers but fail to aggregate at the synaptic sites on the membrane.

Thus an extracellular protein (agrin), transmembrane proteins (MuSK and LRP4), an adaptor protein (Dok-7), and a cytoskeletal protein (rapsyn) form a chain that links commands from the motor axon to ACh receptor clustering in the muscle membrane.

Nevertheless, postsynaptic differentiation can occur in the absence of this transduction pathway. This capacity was apparent in early studies on cultured muscle (see Figure 55–8) and is also seen in vivo: ACh receptor clusters form initially but then disperse in agrin mutants (Figure 55–11C). Clustering also occurs in muscles that lack innervation entirely. Conversely, no clustering occurs in mutant animals lacking MuSK, LRP4, Dok-7, or rapsyn. Thus the signaling pathway that initiates postsynaptic differentiation can be activated without agrin, but agrin is required to maintain clustering of ACh receptors.

The role of agrin is perhaps best understood in terms of the requirement that pre- and postsynaptic specializations be perfectly aligned. Acetycholine receptor aggregates persist in uninnervated muscles but disappear in agrin mutant muscles, suggesting that axons sculpt the postsynaptic membrane through the combined action of agrin and a dispersal factor. One major dispersal factor is ACh itself; clustering persists in mutants that lack both agrin and ACh (Figure 55–11C). Thus agrin may render ACh receptors immune to the declustering effects of ACh. Through a combination of positive and negative factors, the motor neuron ensures that the patches of postsynaptic membrane contacted by axon branches are rich in ACh receptors.

The Nerve Regulates Transcription of Acetylcholine Receptor Genes

Along with redistribution of ACh receptor in the plane of the membrane, the motor nerve orchestrates the transcriptional program responsible for expression of ACh receptor genes in muscle. To understand this aspect of transcriptional control, it is important to appreciate the geometry of the muscle.

Individual muscle fibers are often more than a centimeter long and contain hundreds of nuclei along their length. Most nuclei are far from the synapse, but a few are clustered beneath the synaptic membrane, so that their transcribed and translated products do not have far to go to reach the synapse. In newly formed myotubes most nuclei express genes encoding the ACh receptor α-, β-, δ-, and γ-subunits. In adult muscles, however, only synaptic nuclei express ACh receptor genes; nonsynaptic nuclei do not. This change in pattern occurs in three steps.

During early stages of synapse formation the ACh receptor subunit genes are expressed at higher levels in synaptic nuclei than in their nonsynaptic neighbors (Figure 55–12). Signals acting through MuSK are needed for this specialization. Around the time of birth, ACh receptor gene expression shuts down in nonsynaptic nuclei. This change reflects a repressive effect of the nerve, as originally shown by studies of denervated muscle. When muscle fibers are denervated, as happens when the motor nerve is damaged, the density of ACh receptors in the postsynaptic membrane increases markedly, a phenomenon termed denervation supersensitivity.

This repressive effect of the nerve is mediated by electrical activation of the muscle. Under normal conditions the nerve keeps the muscle electrically active, and active muscle synthesizes fewer ACh receptors than inactive muscle. Indeed, direct stimulation of denervated muscle through implanted electrodes decreases ACh receptor expression, preventing or reversing the effect of denervation (Figure 55–12B). Conversely, when nerve activity is blocked by application of a local anesthetic, the number of ACh receptors throughout the muscle fiber increases, even though the synapse is intact.

In essence, then, the nerve uses ACh to repress expression of ACh receptor genes extrasynaptically. Current that passes through the channel of the receptor leads to an action potential that propagates along the entire muscle fiber. This depolarization opens voltage-dependent Ca²⁺ channels, leading to an influx of Ca²⁺, which activates a signal transduction cascade that reaches nonsynaptic nuclei and regulates transcription of ACh receptor genes. Thus the same voltage changes that produce muscle contraction over a period of milliseconds also regulate transcription of ACh receptor genes over a period of days.

The increase in transcription of ACh receptor genes in nuclei beneath the synapse, along with the decrease in nuclei distant from synapses, leads to localization of ACh receptor mRNA and thus preferential synthesis and insertion of ACh receptors near synaptic sites. This local synthesis is reminiscent of that seen at postsynaptic sites on dendritic spines in the brain. Local synthesis in muscle is advantageous since ACh receptors synthesized near the ends of fibers would never reach the synapse without degradation.

We have used the ACh receptor as an example of postsynaptic differentiation, but many components of the postsynaptic apparatus are regulated in similar ways—their aggregation depends on agrin and MuSK, and their transcription is enhanced in synaptic nuclei and repressed in extrasynaptic nuclei by electrical activity. Thus synaptic components have tailor-made regulatory mechanisms, but many of these components are regulated in parallel.

The Neuromuscular Junction Matures in a Series of Steps

The adult neuromuscular junction is dramatically different in its molecular architecture, shape, size, and functional properties from the simple nerve-muscle contact that initiates neurotransmission in the embryo. Maturation of the nerve terminal, the postsynaptic membrane, and the intervening synaptic cleft occurs in a complex series of steps. We illustrate this step-wise synaptic construction with a continued focus on the development of ACh receptors.

As we have seen, ACh receptors aggregate in the plane of the membrane as the neuromuscular junction begins to form, and receptor gene transcription is enhanced in synaptic nuclei. A few days later activity begins to decrease the level of extrasynaptic receptors and the stability of receptors changes. In embryonic muscle ACh receptors are turned over rapidly (with a half-life of approximately 1 day) throughout the membrane, whereas in adult muscle they are relatively stable (with a half-life of approximately 2 weeks). The metabolic stabilization of ACh receptors helps concentrate them at synaptic sites and stabilize the postsynaptic apparatus.

During the first few postnatal days the composure of the ACh receptor changes, shuts off the δ gene and activates the ε gene. As a result, new ACh receptors inserted in the membrane are composed of α-, β-, δ-, and ε-subunits rather than α-, β-, δ-, and γ-subunits. This altered subunit composition tunes the receptor in a way that is suited to its mature function. However, although it occurs at the same time as the metabolic stabilization, the two changes are not causally linked.

These molecular changes in the postsynaptic membrane are accompanied by changes in its shape (Figure 55–13). Soon after birth junctional folds begin to form in the postsynaptic membrane and ACh receptors become concentrated at the crests of the folds, along with rapsyn, whereas other membrane and cytoskeletal proteins are localized in the depths of the folds. The initial aggregate of ACh receptors appears to have a plaque-like appearance. Perforations that undergo fusion and fission eventually transform the dense plaque into a pretzel-shape that follows the branches of the nerve ending. Finally, the postsynaptic membrane enlarges and eventually contains many more ACh receptors than were present in the initial cluster. Each of these changes occurs while the synapse is functional, implying that ongoing activity plays an important role in synaptic maturation.

Synapses in the central nervous system are structurally similar to neuromuscular junctions, and they function in a similar way. Their formation also adheres to the principles of development of the neuromuscular junction: Pre- and postsynaptic elements regulate each other's differentiation by organizing presynthesized synaptic components rather than by inducing expression of specific genes, and synapses develop in a progressive series of steps (Figure 53–14).

Resolving whether this cellular parallel extends to the molecular level has proven difficult, because of the small size and relative inaccessibility of central synapses. But studies of cultured neurons suggest that the cellular logic of synapse formation is indeed conserved between neuromuscular junctions and central synapses, although different organizers are involved.

Neurotransmitter Receptors Become Localized at Central Synapses

The concentration of neurotransmitter receptors in the postsynaptic membrane is a feature shared by many synapses. In the brain, receptors for glutamate, glycine, GABA (γ-aminobutyric acid), and other neurotransmitters are concentrated in patches of membrane aligned with nerve terminals that contain the corresponding transmitter.

The processes by which these receptors become localized may be similar to those at the neuromuscular junction. In cultures of dissociated hippocampal neurons, for example, both glutamatergic and GABA-ergic nerve terminals appear to stimulate clustering of appropriate receptors in the postsynaptic membrane (Figure 55–15). The mediators of these effects are unknown. Moreover, nerves can induce expression of genes encoding glutamate receptors in central neurons, much as occurs for ACh receptors in muscle. Finally, electrical activity also regulates expression of neurotransmitter receptors in neurons.

In forming receptor clusters, central neurons face an obvious challenge that myotubes do not: They are contacted by axon terminals from distinct classes of neurons that use different neurotransmitters. Thus the nerve terminal probably has an instructive role in the clustering of receptors. In cultures of hippocampal neurons, glutamatergic and GABAergic axons terminate on adjacent regions of the same dendrite. Initially glutamate and GABA receptors are dispersed, but soon each type becomes selectively clustered beneath terminals that release that neurotransmitter. This observation implies the existence of multiple clustering signals with parallel pathways of signal transduction.

In central neurons several distinct proteins have been found to play a role similar to that of rapsyn at the neuromuscular junction. One, gephyrin, is highly concentrated in the synaptic densities at glycinergic and some GABAergic synapses (Figure 55–16). Gephyrin is not structurally related to rapsyn but appears to have similar functions: it links the receptors to the underlying cytoskeleton. Its overexpression in nonneural cells leads to clustering of glycine receptors. Moreover, in gephyrin-deficient mutant mice glycine receptor clusters fail to form at inhibitory synapses. Similarly, a class of proteins that share conserved segments called PDZ domains—the prototypes being PSD-95 or SAP-90—facilitates clustering of NMDA-type (N-methyl-D-aspartate) glutamate receptors and their associated proteins (Figure 55–16). Still other PDZ-containing proteins interact with AMPA-type (α-amino-3-hydroxy-5-methylisoxazole-4-propionate) and metabotropic glutamate receptors. Different pre-synaptic signals may activate pathways that lead to the expression and localization of gephyrin, PSD-95, and other functionally related proteins.

Synaptic Organizing Molecules Pattern Central Nerve Terminals

Nerve terminals at neuromuscular junctions and central synapses are quite similar, reflecting the fact that the motor axon is part of a central neuron. Most of the major protein components of synaptic vesicles have now been isolated and appear to be identical at both types of synapses. Likewise, the mechanisms of transmitter release differ only quantitatively, not qualitatively.

However, the synaptic cleft differs dramatically. Whereas muscle fibers are ensheathed by a basal lamina that has a distinctive molecular structure at the neuromuscular junction, central neurons do not have a prominent basal lamina. Central synaptic clefts contain no detectable laminin or collagen. Instead, intercellular adhesion at central synapses may involve the interaction of matched adhesion molecules on pre- and postsynaptic membranes, with no intermediate matrix.

Several adhesion molecules link the pre- and postsynaptic membrane and also pattern presynaptic differentiation as synapses form. Among them is a class of proteins in postsynaptic membranes, the neuroligins. Their receptors on axonal membranes are the neurexins. The ability of neuroligins to promote presynaptic differentiation was first revealed by culturing neurons with nonneural cells engineered to express neuroligins. In culture synaptic vesicles form clusters at sites of contact with the neuroligin-expressing cells and they are capable of releasing neurotransmitter when stimulated (Figure 55–17). Conversely, neurotransmitter receptors in dendrites aggregate at sites that contact nonneural cells engineered to express neurexins. Thus neurexin-neuroligin interactions facilitate precise apposition of pre- and postsynaptic specializations. Differences among neurexin and neuroligin protein isoforms may also contribute to the matching of excitatory and inhibitory nerve terminals with excitatory and inhibitory neurotransmitter receptors.

How do neurexins and neuroligins work? Part of the answer is that their carboxy terminal tails bind to PDZ domains in proteins such as PSD-95 (Figure 55–16). Indeed, a remarkable number of proteins in both pre- and postsynaptic membranes have PDZ domain-binding motifs, notably adhesion molecules, neurotransmitter receptors, and ion channels. Moreover, many cytoplasmic proteins that possess PDZ domains are present in nerve terminals and beneath the postsynaptic membrane. Thus PDZ-containing proteins can serve as scaffolding molecules that link key components on both sides of the synapse. Interactions of proteins such as neurexins and neuroligins may provide a means of coupling the intercellular interactions required for synaptic recognition to the intracellular interactions required to cluster synaptic components within the cell membrane.

An indication of their critical role is that mutations of neurexin and neuroligin genes have been found in a small subset of patients with autism. However, even though neurexin-neuroligin interactions can organize synapses in culture, it remains unclear whether this is their primary role in vivo. Genetic deletion of neurexins or neuroligins has little effect on synapse size or number. Thus other transsynaptic ligand-receptor pairs may organize synaptic differentiation in the brain, with neuroligins and neurexins consolidating these synapses at a later time and specifying their particular properties. Candidate synaptic organizers include adhesion molecules of the cadherin and immunoglobulin superfamilies, as well as ephrins and Eph kinases, and soluble members of the fibroblast growth factor and Wnt families of morphogens. Which of the many proteins capable of influencing synaptic differentiation in culture are actually crucial for synapse formation in vivo remains to be determined.

Glial Cells Promote Synapse Formation

We have focused on the pre- and postsynaptic partners at synapses, but the organization of synapses involves a third cellular element: glial cells. Schwann cells are the glia at neuromuscular junctions, and astrocytes are the glia at central synapses. Neurons form few synapses when cultured in isolation but many when glia are present (Figure 55–18A).

Glial cell surface and secreted molecules are both required for optimal synapse formation. Several glial-derived molecules that enhance synapse function have been isolated. One in particular is a large matrix protein, thrombospondin, and another a lipid, cholesterol. Others presumably remain to be discovered.

In adult mammals each muscle fiber bears only a single synapse. However, this is not the case in the embryo. At intermediate stages of development several axons converge on each myotube and form synapses at a common site. Soon after birth all inputs but one are eliminated.

The process of synapse elimination is not a consequence of neuronal death. It occurs long after the period of naturally occurring cell death (see Chapter 53). Each motor axon withdraws branches from some muscle fibers but strengthens its connections with others, thus focusing its increasing capacity for transmitter release on a decreasing number of targets. Moreover, axonal elimination is not targeted to defective synapses; all inputs to a neonatal myotube are morphologically and electrically similar, and each can activate the postsynaptic cell (Figure 55–19).

What is the purpose of the transient stage of polyneuronal innervation? One possibility is that it ensures that each muscle fiber is innervated. A second is that it allows all axons to capture an appropriate set of target cells. A third, intriguing idea is that synapse elimination provides a means by which activity can change the strength of specific synaptic connections. We begin to explore this idea in Chapter 56.

Like synapse formation, synapse elimination results from intercellular interactions. Every muscle fiber ends up with exactly one input: None have zero, and very few have more than one. It is difficult to imagine how this could occur without feedback from the muscle cell. Moreover, the axons that remain after partial denervation at birth have a larger number of synapses than they did initially. Thus synapse elimination appears to be a competitive process.

What drives the competition and what is the reward? There is good evidence that neural activity plays a role: Paralysis of muscle reduces competition, whereas direct stimulation enhances it. These findings show that activity is involved, but they do not show that activity determines the outcome of the process, as all axons are affected similarly. Because the essence of the competitive process is that some synapses gain territory at the expense of others, differential activity among axons may be a determinant of axon winners and losers. Changing the activity of only a subset of axons in a living animal has been a technical challenge, but genetic approaches have made this possible in mice. In fact, when the activity of one of the inputs to a muscle fiber is decreased, that axon is highly likely to withdraw.

If the more active axon wins the competition, there is a new problem. Because all synapses made by an axon have the same activity pattern, one might predict that the least active axon in the muscle would eventually lose all of its synapses and the most active would retain all of its synapses. Yet this does not happen. Instead, all axons win at some sites and lose at others, so that every axon ends up innervating a substantial number of muscle fibers.

One possible resolution to this paradox is that the outcome of competition may not be determined by the number of synaptic potentials from the winning axon at a synapse but rather by the total amount of synaptic input that the axon provides to the muscle—a product of the number of impulses and the amount of transmitter released per impulse. In this case an axon that loses at several synapses might redistribute its resources (for example, synaptic vesicles) so that the remaining terminals would be strengthened and more likely to win at their synapses. Conversely, an axon that wins many competitions might find itself with insufficient vesicles to generate large synaptic potentials, and would eventually lose to competitors at some synapses. In this way the number of muscle fibers innervated by individual axons would vary little among axons, as observed.

If activity drives the competition, what is the object of the competition? One idea is that the mechanisms are similar to those that determine whether neurons live or die. The muscle might produce limited amounts of a trophic substance for which the axons compete. As the winner grows it either deprives the loser of its sustenance or gains enough strength to mount an attack on its competitor. Alternatively, the muscle might release a toxic or punitive factor. In these scenarios, although the muscle does contribute a factor in the competition, the outcome is entirely dependent on differences between axons. These differences could be related to activity. The more active axon might be better able to take up trophic factor or resist a toxin. Such positive and negative competitive interactions have been demonstrated at nerve-muscle synapses in culture although not in vivo.

The muscle, however, could play a selective role in synapse elimination rather than just providing a broadly distributed signal. This idea is based on studies in which individual neuromuscular junctions were observed at close intervals during the process of synapse elimination. Acetylcholine receptors and components of the postsynaptic cytoskeleton begin to disappear from parts of the maturing neuromuscular junction before nerve terminals withdraw. This suggests that differences in activity among competing axons may elicit different responses in postsynaptic ACh receptors. For example, the more active axon might trigger a signal from the muscle fiber that strengthens its adhesive interactions with the synaptic cleft, whereas the less active axon might elicit a signal that weakens those interactions.

The complexity of the brain makes direct demonstration of synapse elimination problematic, but electrophysiological evidence from many parts of the central nervous system indicates that synapse elimination is widespread. In autonomic ganglia synapse elimination has been documented directly and their rules seem similar to that found at neuromuscular junctions. Individual axons withdraw from some postsynaptic cells while simultaneously increasing the size of the synapses they form with other neurons.

The formation of synapses completes the hardwiring of the nervous system. To form a functional network, synaptic connections must be exquisitely specific: nerve terminals must recognize the proper target cell from numerous potential partners and often even a specific portion of the target cell's surface. Specificity starts with molecular recognition between partners and is enhanced by patterns of electrical activity.

The requirements for the synapse are stringent. The postsynaptic membrane must be responsive to the neurotransmitter that the nerve terminal releases. The apposition of pre- and postsynaptic elements must be precise at a molecular level, so that responses can occur on a time scale of milliseconds. And synaptic structure must be sufficiently stable to last a lifetime, yet sufficiently plastic to change with experience.

To meet these demands, synaptogenesis needs to be a highly interactive process. Although the pre- and postsynaptic cells can each synthesize synaptic components on their own, they exchange numerous signals to coordinate their activities spatially and temporally. In this chapter we have illustrated these interactions primarily with a focus on the neuromuscular junction. Motor nerve terminals use a combination of electrical and chemical signals to sculpt the postsynaptic apparatus of the muscle fiber. One key signal is agrin, which acts with the neurotransmitter ACh to sculpt the postsynaptic membrane. In turn, the muscle fiber provides signals that organize synaptic specialization in the nerve terminal, signals that include both soluble trophic factors and matrix-associated proteins such as the laminins.

Although central synapses are less accessible, it is becoming clear that they develop by the same general rules as neuromuscular junctions. Some developmentally important molecules found at the neuromuscular junction also regulate differentiation of central synapses, but the latter also use different signals such as the neurexins and neuroligins.

Finally, some of the synapses formed by a particular axon are eliminated, while others prosper. Axons begin by making small synapses on many targets, and end up with large synapses on relatively few targets. This rearrangement is often triggered by activity providing a means through which experience, translated into neural impulses, can modify neural circuits during early life.