Agrin orchestrates synaptic differentiation at the ... · cholinesterase and postsynaptic infoldings of the mus-cle membrane29–31. No AChR aggregates are formed upon injection of

22 TINS Vol. 21, No. 1, 1998 Copyright © 1998, Elsevier Science Ltd. All rights reserved. 0166 - 2236/98/$19.00 PII: S0166-2236(97)01154-5

R E V I E W

POSTSYNAPTIC DIFFERENTIATION at the vertebrateneuromuscular junction (NMJ) is characterized by

(1) aggregation of numerous proteins, including theacetylcholine receptor, at nerve–muscle contacts, (2) up-regulation of transcription in the myonuclei that underliethese contacts, and (3) downregulation of protein syn-thesis in extrasynaptic regions (reviewed in Ref. 1).While downregulation of protein synthesis is clearlydriven by electrical activity (reviewed in Ref. 2), aggre-gation of proteins at the nerve–muscle contact dependson instructive signals that are released by the motoraxon. Almost 20 years ago, U.J. McMahan and colleaguesidentified the synaptic basal lamina as the carrier of theinformation necessary to induce pre- and postsynapticspecializations during neuromuscular regeneration3,4.Basal lamina extracts from the synapse-rich electricorgan of Torpedo californica were used for subsequentpurification of the protein agrin5. When added to cul-tured myotubes, soluble agrin induces the aggregationof acetylcholine receptors (AChRs) in a dose-dependentway, suggesting that agrin plays a role in inducing post-synaptic differentiation. Agrin-like immunoreactivityremains associated with synaptic basal lamina for sev-eral weeks after degeneration of the cellular elements6.This finding provided circumstantial evidence that theability of the synaptic basal lamina to induce AChRaggregation at former synaptic sites after regenerationof muscle fibers4, might be due to neural agrin that re-mains attached to the synaptic basal lamina. In addition,the use of species-specific antibodies showed that agrinthat is released from motor neurons is responsible forthe aggregation of AChRs in nerve–muscle cultures7.

These observations led McMahan to formulate the‘agrin hypothesis’, which postulates that agrin isreleased from the motor-nerve terminal, binds to areceptor on the muscle-cell surface, and induces post-synaptic specializations. Subsequent binding of agrinto the nascent synaptic basal lamina will then immo-bilize agrin8. At around this time, histochemical stain-

ing showed that agrin-like protein was found at manyplaces throughout the body (for example, extrasynap-tic regions of muscle fibers6,9 and in kidney10). Thisobservation was difficult to reconcile with the postu-lated origin of agrin in motor neurons and its functionin synapse formation. Only the cloning of agrincDNAs and the subsequent generation of mice defi-cient in agrin allowed the resolution of this apparentparadox (see below).

Structure and binding sites of agrin

Agrin is a protein with a predicted molecular weightof 225 kDa, consisting of an array of modules hom-ologous to domains found in other basal lamina proteins (Fig. 1). There are numerous carbohydrate-attachment sites, including several conserved sites thatare occupied by side chains characteristic of heparansulfate proteoglycans12–14 (see also Fig. 2). The mosaicstructure of agrin suggests that it has several functionsand numerous binding interactions, and this predictionhas been borne out experimentally. The region ofagrin necessary and sufficient to bind to the basal laminamaps to the amino-terminus end of the molecule13.The binding to the basal lamina is due to the high-affinity binding of agrin to laminins15. As agrin bindsalso to laminin-2 and laminin-4, the two lamininspresent in the muscle basal lamina, this interaction,together with the local release of agrin from the motoraxon, might be the basis for the localized induction ofpostsynaptic structures. Strong binding of agrin tolaminin-4 might also be the reason why agrin remainsassociated with the synaptic basal lamina after degen-eration of the cellular elements6. At the other end ofagrin, the most carboxy-terminus laminin-G-like domainis necessary and sufficient for its AChR-aggregatingactivity16. The region that confers binding to a-dystro-glycan, the major binding protein that is expressed inmuscle cells, is distinct from the AChR-aggregatingdomain17,18 (see also below).

Agrin orchestrates synaptic differentiationat the vertebrate neuromuscular junctionMarkus A. Ruegg and John L. Bixby

The synapse is a key structure that is involved in perception, learning and memory.Understandingthe sequence of steps that is involved in establishing synapses during development might also helpto understand mechanisms that cause changes in synapses during learning and memory. Forpractical reasons, most of our current knowledge of synapse development is derived from studiesof the vertebrate neuromuscular junction (NMJ). Several lines of evidence strongly suggest thatmotor axons release the molecule agrin to induce the formation of the postsynaptic apparatus inmuscle fibers. Recent advances implicate proteins such as dystroglycan, MuSK, and rapsyn in thetransduction of agrin signals.Recently,additional functions of agrin have been discovered, includingthe upregulation of gene transcription in myonuclei and the control of presynaptic differentiation.Agrin therefore appears to play a unique role in controlling synaptic differentiation on both sidesof the NMJ.Trends Neurosci. (1998) 21, 22–27

Markus A. Rueggis at the Dept of

Pharmacology,Biozentrum,University of

Basel,Klingelbergstraße

70, CH-4056Basel, Switzerland,and John L. Bixby

is at the Dept ofMolecular &

CellularPharmacology,

University ofMiami School ofMedicine, 1600

NW 10 Ave,Miami, FL 33136,

USA.

TINS Vol. 21, No. 1, 1998 23

Agrin mRNA undergoes alternative splicing at severalsites, two of which influence agrin’s ability to induceAChR clustering on cultured muscle cells. While agrincontaining inserts at these carboxy-terminus sites,termed A and B in chicken or y and z in rodents, in-duces AChR aggregation at picomolar concentrations,no AChR-aggregating activity is observed with agrinisoforms that lack inserts at site A/y and B/z16,19–21.Inserts at the B/z site have a more pronounced influ-ence on agrin’s AChR-aggregating activity than insertsat site A/y.

Splicing at these functionally important sites isregulated in a tissue-specific manner. As summarizedin Table 1, so far only neurons, including motor neur-ons, have been shown to express agrin isoforms thatcontain inserts at site B/z19,22–27. Interestingly, the tissuesthat contain the highest amount of agrin-like protein,such as kidney and lung (Fig. 2), express agrin isoformsthat lack inserts at the B/z site. Hence, alternativesplicing is most likely the explanation for the obser-vation that agrin-like protein is found in many placesoutside of the NMJ, while agrin’s AChR-aggregatingactivity is confined to neurons10. The widespread dis-tribution of agrin isoforms that lack clustering activity,however, raises a new question concerning the func-tion of these ‘inactive’ isoforms. As these agrin isoformsbind with high affinity to a-dystroglycan from severaltissues, agrin that is expressed outside of the NMJmight have a function as a linker molecule betweenthe basal lamina and the cytoskeleton of abutting cells(see below).

Agrin and NMJ formation in vivo

A collaboration among the groups of J. Sanes, J.Merlie and R. Scheller was the first experimental effortto understand the contribution of neural agrin to theformation of NMJs in vivo. In an attempt to inactivateall isoforms that contain amino acid inserts at site B/z(that is, ‘active’ isoforms), they deleted the two exonsthat encode this site by gene targeting in mice. Unex-pectedly, these manipulations not only inactivated all

M.A. Ruegg and J.L. Bixby – Agrin and synapse formation RE V I E W

Fig. 1. Schematic diagram of the structural domains of chick agrin. Names and symbols used for the different structural motifs are taken fromBork and Bairoch11. The NtA domain is required and sufficient for binding to laminins. The second LamG-like domain, if containing the four aminoacid (aa) insert at splice site A/y, binds to heparin, and the third LamG-like domain in conjunction with the B/z site is sufficient to induce aggre-gation of the acetylcholine receptor (AChR) on cultured myotubes. High-affinity binding to a-dystroglycan is observed with an agrin fragment thatlacks this AChR aggregation domain and comprises all EGF-like repeats and the first two LamG-like domains. In addition, agrin can undergo mRNAalternative splicing at several sites, giving rise to several protein isoforms that differ in function and tissue distribution. Importantly, only thoseagrin isoforms that have amino acid inserts at the B/z site are highly active in aggregation of AChR (see also text).

Fig. 2. Agrin expression is not restricted to the neuromuscular junc-tion (NMJ). Western immunoblot analysis of embryonic (A) and adult(B) chick tissue using polyclonal anti-agrin antibodies. In all tissuesexamined, full-length agrin migrates with an Mr between 4 3 105 and6 3 105 (Agrin), consistent with the recent finding that agrin is a highlyglycosylated heparan sulfate proteoglycan. Immunoreactive proteinswith lower Mr are probably degradation products (asterisk). In embry-onic tissue, agrin is detected in brain, spinal cord and skeletal muscle.In adult tissue, the highest amount of agrin-like protein is found in lungand kidney. Agrin is still abundant in adult brain, while only little isfound in skeletal and cardiac muscle. During development, agrinimmunoreactivity becomes restricted to the NMJ (Refs 24,28), consist-ent with the relatively low expression level in adult skeletal muscle. Thewidespread distribution of agrin-like protein suggests multiple functionsof agrin outside of the NMJ. Sizes of standard proteins are indicated.Part A adapted, with permission, from Ref. 13 and part B adapted,with permission, from M. Gesemann and M.A. Ruegg (unpublished).

24 TINS Vol. 21, No. 1, 1998

B/z-containing agrin isoforms but also affected theexpression of other agrin isoforms. Consequently, the‘knockout’ mice are severe hypomorphs for all formsof agrin. Nevertheless these mice have provided strongevidence in support of the hypothesis8 that agrin isessential for forming functional NMJs. Homozygousmutants die perinatally and they have no normalNMJs; muscle fibers exhibit only small AChR clusters,which are distributed over the entire muscle fiber28

(Fig. 3A). One interpretation of the presence of thesesmall clusters is that they represent ‘spontaneous’ AChRaggregates, like those observed when muscle cells aregrown in culture without neurons. The knockout micealso show additional, unexpected defects in that tran-scription of AChR subunits in myonuclei and the formation of presynaptic specializations is abrogated.

Important complements to the agrin knockoutexperiment are studies in which the instructive role ofneural agrin was tested directly. If muscle fibers areengineered to secrete neural agrin in extrasynapticregions by transfecting or microinjecting them withexpression constructs, they form a postsynaptic appa-ratus that contains most, if not all, the hallmarks ofpostsynaptic differentiation, including high-densityAChR clusters (Fig. 3B), accumulations of acetyl-cholinesterase and postsynaptic infoldings of the mus-cle membrane29–31. No AChR aggregates are formedupon injection of cDNA constructs that encode a non-neuronal agrin isoform that lacks inserts at sites A/yand B/z31. This provides additional evidence that onlyagrin isoforms that are synthesized by motor neuronsare capable of inducing postsynaptic differentiation.

Notably, experiments by H.R. Brenner and colleaguesshowed that injection of cDNA constructs that encodeneural agrin of chick can induce postsynaptic differ-entiation even in fully innervated muscle30,31 (Fig. 3B).Aggregates of AChR that are induced by the ectopicdeposition of neural agrin cause a rapid membrane de-polarization as shown by iontophoresis of ACh. More-over, the aggregated AChRs have the short open timethat is indicative of the channel subtype of the adult.Consistent with these observations, mRNA that en-codes the (adult) ε-subunit of the AChR accumulates

M.A. Ruegg and J.L. Bixby – Agrin and synapse formationRE V I E W

at the ectopic sites30,32 (Fig. 3B). Taken together, theseexperiments show that, in electrically active musclefibers, neural agrin is sufficient to organize postsynapticdifferentiation, including the regulation of myonucleartranscription, in the absence of other factors that arereleased locally by motor neurons.

Agrin receptors and signaling pathways

The ability of neural agrin to initiate postsynapticdifferentiation requires activation of ‘second messen-gers’ inside the muscle fiber8. Aggregation of AChR ispreceded by tyrosine phosphorylation of the AChR b-subunit33 and inhibitors of tyrosine kinases abrogateagrin-induced aggregation of AChR (Refs 34–36).Although it has not been demonstrated directly thatphosphorylation of the AChR b-subunit is required foraggregation of AChR, these experiments suggest thattyrosine kinases participate in the signaling that isactivated by neural agrin. Over the past few years,major efforts by many laboratories have provided uswith a clearer picture of the steps that are involved inthis process. a-Dystroglycan

The studies of J.R. Fallon and colleagues, who usedextracts of the electric organ of T. californica to iden-tify the signaling receptor for agrin biochemically, ledto the isolation of the Torpedo homolog of a-dystro-glycan37. In a remarkable convergence, three other lab-oratories also reported the binding of agrin to a-dystroglycan from other species38–40. a-Dystroglycan isa peripheral membrane protein that assembles with several other molecules of the sarcolemma and thecytoskeleton of muscle fibers to form the dys-trophin–glycoprotein complex (reviewed in Ref. 41).Because agrin binds to a-dystroglycan, and becauseclustering of AChR can be somewhat perturbed withan antibody to a-dystroglycan, it was suggested that a-dystroglycan is the protein that mediates agrin-induced formation of postsynaptic specializations37–39.However, a-dystroglycan does not fulfill the criteriafor an agrin-signaling receptor. First, it is difficult toenvisage how a peripheral membrane protein wouldactivate second-messenger systems, and second,a-dystroglycan binds tightly to both AChR-aggregat-ing and inactive forms of agrin17,40. Moreover, thesmallest fragment of chick neural agrin that was foundto be sufficient for aggregation of AChR (Ref. 16) andphosphorylation of the AChR b-subunit42 does notbind to a-dystroglycan at all17. These experimentsquestion the involvement of a-dystroglycan in agrinsignaling. It appears more likely that dystroglycanplays a role as a structural component that is necessaryto assemble the postsynaptic apparatus. The recentobservation that dystroglycan-deficient mice aredefective in formation of basement membranes43, suggests that dystroglycan might also be involved inorganization of the synaptic basal lamina.MuSK and rapsyn

Recent experiments have provided strong evidencefor the involvement of two other proteins in agrin sig-naling. These are the muscle-specific receptor tyrosinekinase, MuSK (Refs 44–46), and the cytoplasmic pe-ripheral membrane protein rapsyn47. The best evidencefor their being involved in NMJ formation comes fromknockout experiments. Mice that lack MuSK have a verysimilar phenotype, both post- and presynaptically, tothat of agrin knockout mice48. Moreover, muscle cells

TABLE I. Alternative mRNA splicing of agrin transcripts is tissue specific

Splice insert

Tissue A/y site B/z site Refs

PeripheryLung 0 0 22,23Kidney 0 0 22–24

Central nervous systemGlia 0<<4 0 22–26Neurons 4 0, 11, 19, 8 22–26

Neuromuscular junctionMotor neurons 4 0, 11, 19, 8 23,26,27Skeletal muscle 0>>4 0 19,22,24Schwann cells 0 0 19,22

The summarized data are derived from different developmental stages and differentspecies. Although there are other sites in agrin that undergo alternative mRNA splic-ing, only sites A/y and B/z are indicated in the table. Amino acid inserts at site B/z arerequisite to render agrin highly active in aggregation of the acetylcholine receptor(AChR). Insertions at site A/y improve the AChR-aggregating activity of agrin onlymoderately.

TINS Vol. 21, No. 1, 1998 25

taken from these animals are incapable of aggregatingAChRs in response to neural agrin49. These resultsshow convincingly that MuSK is required for neuralagrin signaling. However, MuSK is not, by itself, theagrin receptor, as MuSK and neural agrin do not bindto each other. Nevertheless, addition of neural agrin tocultured muscle cells causes rapid phosphorylation ofMuSK, but no activation of MuSK is observed when it isexpressed in fibroblast-like cells49. These results suggestthat an additional protein is expressed in muscle fibersthat selectively recognizes neural isoforms of agrin, andthat this binding is requisite for the rapid phosphoryl-ation of MuSK (Fig. 4).

Like the agrin and the MuSK knockouts, rapsyn-deficient mice have a perturbed NMJ. Presynaptic spe-cializations still form, although the motor terminalsarborize poorly and grow more extensively than inwild-type mice52. However, AChRs and several otherpostsynaptic molecules are not clustered at thesenerve–muscle contacts. As rapsyn causes the clusteringof AChRs when coexpressed in nonmuscle cells53,54,the knockout data suggest that rapsyn is required forlinking most postsynaptic molecules to the underly-ing cytoskeleton.

Recent experiments now suggest that rapsyn is notmerely structural, but that it also participates in signaltransduction that is activated by agrin. In culturedmyotubes from rapsyn-deficient mice, neural agrinstill induces rapid phosphorylation of MuSK, but theAChR b-subunit does not become tyrosine phosphoryl-ated, nor do AChRs form aggregates55. This suggeststhat rapsyn serves as an adapter molecule to linkMuSK activation to further downstream events thatare important for the localization of molecules to thepostsynaptic apparatus. An interaction of rapsyn andMuSK is suggested by the finding that, when they arecoexpressed in fibroblasts, rapsyn and MuSK codistrib-ute56. Interestingly, the colocalization of rapsyn andMuSK in these cells depends on the extracellular, andnot the cytoplasmic, domain of MuSK55, suggestingthat fibroblasts express a linker molecule that confersthe binding (see Fig. 4B). One interesting idea is thatAChRs themselves might link the extracellular domainof MuSK and rapsyn since a small percentage of MuSKco-immunoprecipitates with AChRs in C2 myotubes57.

On the other hand, MuSK remains localized tonerve–muscle contacts in rapsyn-deficient mice55, sug-gesting that the first step in NMJ development is theformation of a primary scaffold of MuSK by motorneuron-derived factors such as agrin (Fig. 4A). Thisprimary scaffold would then serve as a docking site forthe adapter molecule rapsyn, which in turn wouldmediate aggregation of other molecules at the post-synaptic site (Fig. 4B).

Control of gene transcription

How can neural agrin specify transcriptional activ-ity in myonuclei when deposited in extrasynapticregions of muscle fibers? Gene expression in subsyn-aptic myonuclei is thought to be regulated by ARIA(for acetylcholine receptor-inducing activity), a mem-ber of the neuregulin family (reviewed in Ref. 58). Inmuscle cells, ARIA increases synthesis of AChR pro-tein59 and induces transcription of genes that encodesome AChR subunits60. Neuregulin/ARIA signaling istransduced through the receptor tyrosine kinasesErbB2, ErbB3 and ErbB4, and both neuregulins and

these receptors accumulate at the NMJ (Refs 61,62).Most importantly, mice that are heterozygous for adeletion of a neuregulin isoform have significantlyfewer AChRs in the subsynaptic muscle membrane63.

The current model predicts that neuregulins arereleased from the motor neuron and are subsequentlyincorporated into the synaptic basal lamina (Fig. 4C).The observation that ectopic neural agrin also altersgene transcription in myonuclei30,32 clearly shows that

M.A. Ruegg and J.L. Bixby – Agrin and synapse formation RE V I E W

Fig. 3. Neural agrin is required and sufficient for postsynaptic differentiation. (A) Deletionof the B/z site in the mouse gene that encodes agrin prevents formation of functional neuro-muscular junctions (NMJs). Schematic representation and electron micrographs of the NMJphenotype in heterozygous (+/–; top row) and homozygous (–/–; bottom row) agrin knockoutmice. In +/– control mice, motor nerves contact muscle in an innervation band, where axonsbranch in an orderly manner to innervate single muscle fibers. In response to innervation, apostsynaptic apparatus is assembled in the muscle fiber as indicated by the aggregation ofacetylcholine receptors (AChRs are in red) at the nerve–muscle contact. The formation ofsynaptic specializations is already visible at embryonic day 18 as seen in the electron micro-graph (top right panel). Nerve terminals that contact muscle are vesicle-rich and occasionally,postsynaptic infoldings are found. The formation of both pre- and postsynaptic specializationsis abrogated in the –/– mice as AChR clusters (red) are small and are distributed over the entiremuscle fiber. Presynaptic differentiation is grossly perturbed and motor axons grow extensively.The failure to form synaptic specializations can be seen in the electron micrograph (rightpanel). In addition, the muscle basal lamina in the –/– mice is not continuous but appearspatchy (arrows). (B) In innervated adult rat muscle, injection of expression constructs thatencode full-length neural agrin is sufficient to induce postsynaptic specializations. Expressionconstructs are injected into single muscle fibers (top drawing). After 4–6 weeks, muscle fibershave deposits of chick neural agrin in muscle basal lamina at ectopic sites (green; bottomdrawing). Staining of the muscle with anti-chick agrin antibodies reveals the deposits of neuralagrin as visualized in optical, longitudinal sections through the injection site (upper row;Agrin). In this area, ectopic neural agrin induces aggregation of AChRs (upper row; AChR).Aggregation of AChRs is also readily visible in cross sections through ectopic sites (lower row;AChR). As visualized by in situ hybridization, transcripts that encode the e-subunit of the AChRaccumulate at these sites (lower row; e-mRNA). Arrowheads indicate the position of AChRaggregates shown in the left micrograph. Scale bar in upper row, 25 mm, in lower row, 15 mm.Pictures in A adapted, with permission, from Ref. 28, and pictures in B adapted, with permission, from Ref. 29.

26 TINS Vol. 21, No. 1, 1998

the nerve is not required. One explanation for theeffect of agrin on gene transcription could be thatmuscle fibers themselves synthesize neuregulins (Fig.4C). In this model, agrin-induced reorganization ofmolecules at ectopic sites, both in the muscle mem-brane and in the basal lamina, would create a ‘sink’ formuscle-derived neuregulins (Fig. 4C). Indeed, muscle

cells contain neuregulin-like protein64 and neuregulin-like activity can be extracted from cultured myotubes(Th. Meier and H.R. Brenner, pers. commun.). In ad-dition, ErbB receptors are aggregated at ectopic AChRclusters that are induced by neural agrin31,32. However,rapsyn-deficient mice lack aggregates of ErbBs atnerve–muscle contacts, but show normal transcrip-tional regulation of AChR subunits52. This leaves openthe possibility that activation of gene transcription canbe a direct consequence of MuSK signaling, becauseMuSK remains concentrated at nerve–muscle contactsin rapsyn-deficient animals55.

Regulation of presynaptic differentiation

Proteins capable of inducing (1) specific adhesion ofmotor axons to muscle cells, (2) inhibition of furtheraxon growth, and (3) assembly of nerve terminal-specific proteins, have been called ‘stop signals’, andpostulated to be present in muscle during inner-vation65. The fact that stop-signal functions are severelycompromised in the agrin knockout mouse indicatesthat agrin is itself a stop signal, that it acts indirectlyby regulating the expression of a stop signal, or both.

One candidate for a motor neuron-selective stopsignal is the b2 chain of laminin, previously called s-laminin, whose localization to NMJs is regulated byagrin28. A recombinant fragment of the b2 chain hasthe ability in vitro to cause selective adhesion of motor

M.A. Ruegg and J.L. Bixby – Agrin and synapse formationRE V I E W

Fig. 4. Model of the steps that are involved in agrin-induced for-mation of the neuromuscular junction (NMJ). (A) Neural agrin causesthe formation of a primary scaffold that contains MuSK. Growth conesof motor axons release neural agrin isoforms (nA) that are active inaggregation of acetylcholine receptor (AChR). Binding of neural agrin toan unknown protein (?) causes phosphorylation of MuSK (P). Althoughthe mechanism of MuSK phosphorylation is not yet understood, exam-ples from other receptor systems suggest a model where MuSK becomesactivated by being an integral part of a receptor complex. The lack ofany NMJs in the MuSK knockout mice shows that the formation of thisprimary scaffold is requisite for all downstream events that inducesynaptic differentiation. (B) Binding of agrin to laminins immobilizesthe primary MuSK scaffold and rapsyn links other postsynaptic mol-ecules to this scaffold. Binding of neural agrin to laminin-2 and laminin-4 in the muscle basal lamina immobilizes agrin near its release site andthus immobilizes the MuSK scaffold. Co-distribution of rapsyn andMuSK in transfected fibroblasts suggests that rapsyn becomes linkedsubsequently to the MuSK scaffold. This linkage depends on the extra-cellular domain of MuSK and thus is mediated by a transmembranelinker molecule (Rapsyn Associated Transmembrane Linker, RATL).Rapsyn in turn is required for the phosphorylation of the AChR b-subunit (P). Phosphorylation of AChR subunits might be mediated bythe AChR-associated tyrosine kinases of the src family50 or Grb2 (Ref.51). Rapsyn might also mediate linkage of other molecules, such as theproteins of the dystrophin–glycoprotein complex (DGC), to postsynap-tic sites. Accumulation of the DGC might help to accumulate agrin isoforms that are synthesized by muscle fibers (mA) to the site ofnerve–muscle contact. (C) Regulation of presynaptic specialization andtranscriptional control in myonuclei. Accumulation of neural and mus-cle agrin at synaptic sites and the aggregation of the DGC might sup-port the assembly of synaptic basal lamina (~~) and might induce theformation of presynaptic specializations in the nerve terminal. As anadditional consequence of the organizational activity of agrin, myo-nuclei (MN) accumulate in subsynaptic regions and become transcrip-tionally activated. Activation of transcription is likely to be based on the ‘trapping’ of muscle-derived (brown) and nerve-derived (green)neuregulins in synaptic basal lamina and subsequent activation of ErbBreceptors. To simplify the scheme, molecules that are not activated in aparticular step are kept in gray and the molecules that link the DGC tothe cytoskeleton of muscle fibers are omitted.

TINS Vol. 21, No. 1, 1998 27

neurons and to inhibit motor-neurite growth65.However, native laminin-4 (containing the β2 chain)has recently been shown to be neurite-promoting, andnot inhibitory66. Additionally, the b2 chain is notpresent when NMJs are first formed67, and knockoutmice that lack the b2 chain, despite some aberrantsynaptic morphology, have normal numbers of func-tional NMJs (Ref. 68).

Could agrin itself be a stop signal? Agrin that is syn-thesized by muscle fibers is found in basal lamina ofmuscle prior to the ingrowth of the first motor axonsduring development; it later colocalizes with AChRclusters9,69 (see also Fig. 4). In vitro, agrin fulfils the cri-teria for a motor neuron-selective stop signal. First, agrinis adhesive for neurons, and might be selectively adhe-sive for motor neurons70–72. Second, agrin is inhibitoryto neurite growth for several populations of neurons,including motor neurons14,71,72. Finally, agrin expressedon the surface of nonmuscle cells causes the accumu-lation of synaptic vesicles in motor neurites at sites ofcontact71. Importantly, the latter activity is also ex-erted by muscle-derived agrin73, which lacks the abilityto cluster AChRs and to signal through MuSK42,49.

The fact that MuSK-deficient mice have aberrantpresynaptic differentiation48 has been used to arguethat agrin’s effect on the nerve is indirect. However,MuSK is also expressed at low levels in nonmuscle tis-sue45. Although there is no major decrease in the levelof agrin in MuSK-deficient mice (J.R. Sanes, pers. com-mun.), it is still possible that the lack of an MuSK-dependent primary scaffold55 (Fig. 4) affects the accu-mulation of agrin at nerve–muscle contacts. Insummary, agrin clearly interacts directly with neur-ons, and agrin has functional activities that are con-sistent with its being a stop signal. The extent towhich agrin acts alone to initiate nerve-terminal dif-ferentiation or, as seems more likely, acts both directlyand indirectly, remains to be determined.

Concluding remarks

Evidence is now strong for agrin’s being the keyplayer in the formation of the NMJ. Although recentfindings have provided us with a better knowledge ofthe molecules that are involved in agrin-induced sig-naling at the NMJ, many crucial details are still miss-ing. In particular, it will be important to identify themolecules that mediate the binding of neural agrin tothe signaling complex (Fig. 4A) and to understandhow agrin induces differentiation of the presynapticnerve terminal. Because agrin also participates in theregulation of gene transcription, another importantissue is to establish the relationship between agrin andthe signaling that is activated by the neuregulins.

A major reason for studying NMJ development is tolearn more about mechanisms that might also be usedin the formation of other synapses. As agrin-likeimmunoreactivity is also found at synaptic sites in thechick retina74, another focus of future studies will beto understand the role of agrin in the formation ofinterneuronal synapses.

Selected references1 Hall, Z.W. and Sanes, J.R. (1993) Cell/Neuron 72/10, 99–1212 Duclert, A. and Changeux, J-P. (1995) Physiol. Rev. 75, 339–3683 Sanes, J.R., Marshall, L.M. and McMahan, U.J. (1978) J. Cell

Biol. 78, 176–1984 Burden, S.J., Sargent, P.B. and McMahan, U.J. (1979) J. Cell

Biol. 82, 412–425

5 Nitkin, R.M. et al. (1987) J. Cell Biol. 105, 2471–24786 Reist, N.E., Magill, C. and McMahan, U.J. (1987) J. Cell Biol.

105, 2457–24697 Reist, N.E., Werle, M.J. and McMahan, U.J. (1992) Neuron 8,

865–8688 McMahan, U.J. (1990) Cold Spring Harbor Symp. Quant. Biol. 55,

407–4189 Fallon, J.R. and Gelfman, C.E. (1989) J. Cell Biol. 108,

1527–153510 Godfrey, E.W. (1991) Exp. Cell Res. 195, 99–10911 Bork, P. and Bairoch, A. (1995) Trends Biochem. Sci. 20, poster12 Tsen, G. et al. (1995) J. Biol. Chem. 270, 3392–339913 Denzer, A.J. et al. (1995) J. Cell Biol. 131, 1547–156014 Halfter, W. et al. (1997) J. Comp. Neurol. 383, 1–1715 Denzer, A.J. et al. (1997) J. Cell Biol. 137, 671–68316 Gesemann, M., Denzer, A.J. and Ruegg, M.A. (1995) J. Cell

Biol. 128, 625–63617 Gesemann, M. et al. (1996) Neuron 16, 755–76718 Hopf, C. and Hoch, W. (1996) J. Biol. Chem. 271, 5231–523619 Ruegg, M.A. et al. (1992) Neuron 8, 691–69920 Ferns, M. et al. (1992) Neuron 8, 1079–108621 Ferns, M.J. et al. (1993) Neuron 11, 491–50222 Ma, E., Morgan, R. and Godfrey, E.W. (1994) J. Neurosci. 14,

2943–295223 Stone, D.M. and Nikolics, K. (1995) J. Neurosci. 15, 6767–677824 Hoch, W. et al. (1993) Neuron 11, 479–49025 O’Connor, L.T. et al. (1994) J. Neurosci. 14, 1141–115226 Smith, M.A. and O’Dowd, D.K. (1994) Neuron 12, 795–80427 Ma, E., Morgan, R. and Godfrey, E.W. (1995) J. Neurobiol. 26,

585–59728 Gautam, M. et al. (1996) Cell 85, 525–53529 Cohen, I. et al. (1997) Mol. Cell. Neurosci. 9, 237–25330 Jones, G. et al. (1997) Proc. Natl. Acad. Sci. U. S. A. 94, 2654–265931 Meier, T. et al. (1997) J. Neurosci. 17, 6534–654432 Rimer, M. et al. (1997) Mol. Cell. Neurosci. 9, 254–26333 Wallace, B.G., Qu, Z. and Huganir, R.L. (1991) Neuron 6,

869–87834 Wallace, B.G. (1994) J. Cell Biol. 125, 661–66835 Meier, T., Perez, G.M. and Wallace, B.G. (1995) J. Cell Biol.

131, 441–45136 Ferns, M., Deiner, M. and Hall, Z.W. (1996) J. Cell Biol. 132,

937–94437 Bowe, M.A. et al. (1994) Neuron 12, 1173–118038 Gee, S.H. et al. (1994) Cell 77, 675–68639 Campanelli, J.T. et al. (1994) Cell 77, 663–67440 Sugiyama, J., Bowen, D.C. and Hall, Z.W. (1994) Neuron 13,

103–11541 Henry, M.D. and Campbell, K.P. (1996) Curr. Opin. Cell Biol. 8,

625–63142 Meier, T. et al. (1996) EMBO J. 15, 2625–263143 Williamson, R.A. et al. (1997) Hum. Mol. Genet. 6, 831–84144 Jennings, C.G., Dyer, S.M. and Burden, S.J. (1993) Proc. Natl.

Acad. Sci. U. S. A. 90, 2895–289945 Ganju, P. et al. (1995) Oncogene 11, 281–29046 Valenzuela, D.M. et al. (1995) Neuron 15, 573–58447 Frail, D.E. et al. (1988) J. Biol. Chem. 263, 15602–1560748 DeChiara, T.M. et al. (1996) Cell 85, 501–51249 Glass, D.J. et al. (1996) Cell 85, 513–52350 Fuhrer, C. and Hall, Z.W. (1996) J. Biol. Chem. 271, 32474–3248151 Colledge, M. and Froehner, S.C. (1997) J. Neurosci. 17, 5038–504552 Gautam, M. et al. (1995) Nature 377, 232–23653 Froehner, S.C. et al. (1990) Neuron 5, 403–41054 Phillips, W.D. et al. (1991) Science 251, 568–57055 Apel, E.D. et al. (1997) Neuron 18, 623–63556 Gillespie, S.K.H. et al. (1996) Neuron 16, 953–96257 Fuhrer, C. et al. (1997) EMBO J. 251, 568–57058 Lemke, G. (1996) Mol. Cell. Neurosci. 7, 247–26259 Falls, D.L. et al. (1993) Cell 72, 801–81560 Chu, G.C. et al. (1995) Neuron 14, 329–33961 Jo, S.A. et al. (1995) Nature 373, 158–16162 Zhu, X.J. et al. (1995) EMBO J. 14, 5842–584863 Sandrock, A.W., Jr et al. (1997) Science 276, 599–60364 Moscoso, L.M. et al. (1995) Dev. Biol. 172, 158–16965 Porter, B.E., Weis, J. and Sanes, J.R. (1995) Neuron 14, 549–55966 Brandenberger, R. et al. (1996) J. Cell Biol. 135, 1583–159267 Chiu, A.Y. and Sanes, J.R. (1984) Dev. Biol. 103, 456–46768 Noakes, P.G. et al. (1995) Nature 374, 258–26269 Lieth, E. and Fallon, J.R. (1993) J. Neurosci. 13, 2509–251470 Burg, M.A., Halfter, W. and Cole, G.J. (1995) J. Neurosci. Res.

41, 49–6471 Campagna, J.A., Ruegg, M.A. and Bixby, J.L. (1995) Neuron 15,

1365–137472 Chang, D. et al. (1997) Dev. Biol. 181, 21–3573 Campagna, J.A., Ruegg, M.A. and Bixby, J.L. Eur. J. Neurosci.

(in press)74 Mann, S. and Kröger, S. (1996) Mol. Cell. Neurosci. 8, 1–13

M.A. Ruegg and J.L. Bixby – Agrin and synapse formation RE V I E W

AcknowledgementsWe thank all ourcolleagues who havecommunicated theirmost recent resultsto us. We aregrateful to W.B.Adams, R.Bookman and Th.Meier for criticallyreading themanuscript. Ourwork is supported bythe Swiss NationalScience Foundation,the SwissFoundation forResearch on MuscleDiseases, theRentenanstalt/SwissLife and theSandoz-Stiftung(MAR) and the USNational Institutesof Health and theUS National ScienceFoundation (JLB).