Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Ann N Y Acad Sci. Author manuscript; available in PMC 2012 October 22.
Published in final edited form as:
PMCID: PMC3478106

Recent Structural and Mechanistic Insights into Endplate Acetylcholine Receptors


Voluntary movement mediated by skeletal muscle relies on endplate acetylcholine receptors (AChR) to detect nerve-released ACh and depolarize themuscle fiber. Recent structural and mechanistic studies of the endplate AChR have catalyzed a leap in our understanding of the molecular steps in this chemical-to-electrical transduction process. Studies of acetylcholine binding protein (AChBP) give insight into ACh recognition, the first step in activation of the AChR. An atomic structural model of the Torpedo AChR at a resolution of 0.4 nm, together with single-ion channel recording methods, allow tracing of the link between the agonist binding event and gating of the ion channel, as well as determination of how the channel moves when it opens to allow flow of cations. Structural models of the human AChR enable precise mapping of disease-causing mutations, while studies of the speed with which single AChR channels open and close cast light on pathogenic mechanisms.

Keywords: acetylcholine receptor, acetylcholine binding protein, agonist recognition, binding-gating coupling mechanism, congenital myasthenic syndrome


The chief purpose of the muscle endplate is to detect nerve-released acetylcholine (ACh), and with little delay, create a brief, local depolarization that evokes a regenerating action potential that propagates throughout the muscle fiber. This chemical-to-electrical transduction process is orchestrated by three protein oligomers found at the motor endplate: the acetylcholine receptor (AChR), voltage-gated sodium channel, and acetylcholinesterase. Notably, inherited disorders of neuromuscular transmission, known as congenital myasthenic syndromes (CMS), have been found to originate from mutations in all three of these proteins or in proteins physically associated with them.1 The endplate AChR is the most common molecular target of CMS and is therefore the focus of this review. In recent years, our understanding of the AChR has advanced tremendously owing to the emergence of atomic structural data and the consequent ability to relate structure to function and map disease-causing mutations onto the structure and rationalize their functional consequences.

Overview of AChR Structure

Ever since application of ACh to the motor endplate was found to rapidly increase the electrical conductance of the muscle membrane,2 the AChR has been viewed as two functionally distinct domains within the same macromolecule: a domain that recognizes agonist and a domain that forms the channel. This conclusion was confirmed by electron microscopy of two-dimensional arrays of AChRs from the Torpedo electric organ at a resolution of 0.9 nm,3 which revealed an elongated cylinder composed of five subunits embedded in the cell membrane. Each subunit consists of a large, amino-terminal extracellular domain, four transmembrane domains (TMDs), and a large cytoplasmic domain situated between the third and fourth TMDs. Biochemical and molecular cloning studies showed that two of the five subunits are identical, named α-subunits, while the remaining subunits are products of different genes, and in adult mammals are named β-, δ-, and ε-subunits.4

Three years ago the structure of the AChR from the Torpedo electric organ was solved at a resolution of 0.4 nm, again by electron microscopy of two-dimensional arrays of AChRs.5 It is considered a structural model because the protein main chain, α-carbon atoms, and bulky side chains were defined, whereas small side chains and their orientations were not. The higher resolution images confirmed images at 0.9 nm resolution and revealed the detailed molecular architecture: the extracellular domain of each subunit comprises predominantly β-sheets, the four TMDs are α-helices, and the cytoplasmic domain contains roughly equal amounts of α-helix and unresolved structure.

Further structural insights came from unexpected sources. Glial cells from fresh- and salt-water snails, Lymnaea stagnalis and Aplysia californica, were found to secrete a water-soluble protein that binds ACh, named ACh binding protein (AChBP), which was solved at resolutions from 0.27 to 0.2 nm by X-ray crystallography. 6,7 The amino acid sequences of the AChBPs are homologous to those of AChR subunits, and the structures revealed that AChBP is a homo-pentamer that structurally mimics the AChR extracellular domain. Despite less than 24% homology with the AChR, the AChBP structures conform well to the extracellular domain of the Torpedo AChR. The higher resolution achievable with AChBP enabled resolution of side chains and their orientations, with and without bound agonists and antagonists,7,8 giving insight into ligand recognition. Together, the AChBP and Torpedo AChR structures provide excellent templates for generating structural models of mammalian endplate AChRs (Fig. 1).

Figure 1
Structural model of the AChR from the human motor endplate. The model was constructed using the Torpedo AChR structural model (Protein Data Bank code 2bg9) as a template. The structure is rendered as ribbons with the α-subunits orange and non-α-subunits ...

The endplate AChR contains two ACh binding sites, which are formed at interfaces between extracellular domains of the α-subunits and the adjacent ε- or δ-subunits. The contribution of the α-subunit to the binding site is known as the principal face, while the contribution of the non-α-subunit is called the complementary face.9 The principal face consists of three distinct regions of the linear sequence of the α-subunit, known as loops A–C, while the complementary face consists of four distinct regions, known as loops D–G. Within the binding site interface, a cluster of aromatic residues create an electron-rich cage that stabilizes the cationic ACh. Although the same five aromatic residues are present at each binding site, sequence differences in loops D–G of the non-α-subunits produce affinities of the two sites that differ by as much as 10,000-fold for both agonists and antagonists. Rather than inducing different conformations of the α-subunits, the non-α-subunits produce site-selectivity through direct contributions to the binding sites.9

Overview of AChR Function

Binding of ACh triggers a step increase in current through the receptor channel, which is maintained when ACh is bound, but terminates when the channel closes and ACh dissociates. In the absence of ACh, the channel opens with a very low but detectable probability, but in the presence of ACh the channel opens with a probability approaching unity. Activation by ACh can thus be viewed as an amplification process in which the probability the channel is open increases greatly.10 For the endplate AChR, binding of two molecules of ACh is required for full amplification, as illustrated in the following adaptation of the Monod-Wyman- Changeux (MWC) mechanism to AChR activation11:

An external file that holds a picture, illustration, etc.
Object name is nihms414054f7.jpg

Two molecules of ACh (A) bind to closed AChR channels (C), which in turn open (O) with equilibrium constants (θ) that increase with increasing numbers of bound agonist. Nonliganded receptors open with θ0 of around 10−6, singly-liganded receptors open with θ1 of 10−3, and doubly-liganded receptors open with θ2 of 25.10,12 Channel opening is driven by higher affinity of the agonist for the open relative to the closed state, given by the dissociation constants K* and K, respectively. For wild-type AChRs, the kinetics of activation have been analyzed by fitting the subset of states in blue to single-channel dwell times recorded over a range of agonist concentrations, yielding rate constants for ACh binding to the closed channel and gating of the diliganded channel.1315 (In color in Annals onlines.)

Agonist Recognition

The MWC mechanism predicts that the binding sites of the AChR change structurally when agonist is bound owing to low affinity of the closed state and high affinity of the open state.10,11 The first structural evidence for state-dependent changes came from cryo-electron microscopy of the Torpedo AChR following rapid application of ACh.16 Subsequent analyses revealed agonist-dependent rotation of the inner β-sheets of the binding domain, and tilting of the peripheral β-sheets toward the central water-filled vestibule.17 X-ray structures of AChBP with bound agonist showed a similar tilting of a peripheral β-hairpin known as the C-loop,7,8 but the inner β-sheets showed little change and were similar to those in the Torpedo AChR structural model at 0.4 nm resolution.

Molecular dynamics (MD) simulations also captured agonist-dependent changes of the C-loop of AChBP18 (Fig. 2). Starting with the C-loop in the agonist-bound capped conformation (green), with ACh present in the binding pocket, the C-loop remained capped throughout 45 ns of simulation (blue), but when ACh was not present the C-loop achieved the uncapped conformation within 10 ns and remained uncapped (orange).

Figure 2
Ligand binding site of AChBP in uncapped and capped conformations. Left panel shows superimposed structures of two adjacent AChBP monomers following MD simulation with (blue) or without (brown) ACh bound.18 The flexible C-loop (see text) is highlighted ...

Fluorescence and NMR studies of AChBP in solution also revealed agonist-dependent motion of the C-loop. Without ACh, Trp residues in the binding site were accessible to an extrinsic quencher of Trp fluorescence, whereas with ACh bound, the Trps became inaccessible and quenching was prevented.18 When vicinal cysteines at the tip of the C-loop of AChBP were isotopically labeled with 15N, five spectral peaks were present in the 15N-1H heteronuclear single quantum correlation NMR spectrum, indicating multiple chemical environments of the reporter cysteines. However, when ACh was bound to AChBP, only two peaks were present, one for each cysteine, indicating a uniform environment.19 Thus the collective observations show that the C-loop at the periphery of the ligand binding site changes conformation and traps the agonist within the binding pocket.

Initial Coupling of Binding to Gating

The first step in AChR activation is binding of the agonist. Knowing that the C-loop caps the binding site when the agonist is bound, Mukhtasimova and colleagues asked whether residues physically linked to the C-loop trigger early steps in the activation process.20 A coupled triad of residues, αTyr190, αAsp200, and αLys145, was identified according to the following criteria: (i) proximity of the three residues equivalent to these in the AChBP structure, (ii) conservation of the three residues in allmuscle α-subunits, (iii) marked suppression of gating after mutation of any three residues, and (iv) nonadditive changes in channel gating after mutation of any two residues. In AChBP, this triad corresponds to Tyr185, Asp194, and Lys139, shown in the superimposed capped (green) and uncapped (yellow) conformations (Fig. 3). The recent structure of the ligand binding domain of a muscle α-subunit confirms that αLys145 and αAsp200 form a salt bridge. The collective results suggest that in the resting receptor, the salt bridge formed by αLys145 and αAsp200 stabilizes β-strands 7 and 10 to which they are attached. Upon binding of the agonist, the C-loop changes conformation from an uncapped to a capped conformation, which weakens the αLys145/αAsp200 salt bridge as αTyr190 establishes contact with αLys145. These local changes at the periphery of the binding site displace β-strands 7 and 10, which transmit further changes distally to the junction of the binding and pore domains.

Figure 3
Conserved residues mediate the initial linkage between ACh binding and channel gating. Uncapped (yellow) and capped (green) conformations of the C-loop of AChBP are superimposed with the conserved residues, Tyr185, Asp 194, and Lys139, shown in stick ...

Principal Coupling Pathway

The interface between ligand binding and pore domains is a structural transition zone where β-sheets from the binding domainmeet α-helices from the pore, illustrated by the structural model of the α-subunit (Fig. 4). Within this zone, three loops and an interdomain connector from the binding domain (β1–2 loop, Cys-loop, β8–9 loop, pre-M1 strand) merge with the interhelical M2–M3 loop from the pore domain. Together, these five structural elements transmit structural changes from the ligand binding domain to the pore. A chimera with a binding domain from AChBP and pore domain from the 5HT3 receptor expressed on the cell surface but was not functional, presumably because residues from both AChBP and 5HT3 were present at the binding–pore interface and produced a structural mismatch.21 However, after correcting the mismatch by substituting 5HT3 for AChBP sequences into the three interface loops from the AChBP module, the resulting chimera exhibited ACh-evoked single-channel currents, thus revealing a delicate functional interplay among structural elements at the binding–pore interface.

Figure 4
Principal pathway linking ACh binding to channel gating. Left panel shows an α-subunit from the Torpedo AChR (PDB code 2bg9), with the ligand binding domain in orange and the TMDs in red. The boxed region is expanded in the right panel. Within ...

Inspection of the Torpedo AChR structural model in light of multiple sequence alignments reveals a highly conserved salt bridge buried within the binding–pore interface. An invariant αArg209 from the pre-M1 strand joins a highly conserved αGlu45 from the β1–2 loop, and the resulting interdomain assembly straddles the M2–M3 loop. This salt bridge was recently confirmed by the X-ray structure of a muscle α-subunit ligand binding domain.22 We showed that αArg209 and αGlu45 are key elements of a principal coupling pathway that join three discontinuous regions of the α-subunit: the pre-M1 region and the β1–2 and M2–M3 loops. Charge reversal of either αArg209 or αGlu45 impaired channel gating, but the charge-reversed double mutant restored gating to normal23 (Fig. 6). Mutant cycle analysis applied to the channel gating equilibrium constant revealed strong energetic coupling of −3.1 kcal/mol between αArg209 and αGlu45. The core salt bridge aligns αGlu45 and αVal46 of the β1–2 loop with αPro272 and αSer269 of the M2–M3 loop to create an assembly that engages the ligand binding domain with the pore domain (Fig. 6). αVal46, αPro272, and αSer269 also show interresidue energetic coupling, presumably through hydrophobic interactions.23

Figure 6
Arg209 and Glu45 are functionally interdependent in contributing to channel gating. The salt bridge formed by Arg209 and Glu45 is displayed using the same atomic coordinates as in Figure 4. Left column shows single-channel currents evoked by a high concentration ...

Intersubunit Interactions

Gating of the AChR channel requires global structural changes, yet how these changes propagate between subunits remains unknown. Low-resolution cryo-electron microscopic images detected agonist-mediated changes in the α-subunits but not in the other subunits.17 However a wealth of data, including analyses of site-directed mutants, rate-equilibrium free energy relationships, and analyses of mutations underlying CMS show that the non-α-subunits contribute profoundly to channel gating2426; the non-α-subunits may move as rigid bodies in response to the α-subunits or as part of an intersubunit network.

Good candidates for intersubunit linkages are residues at the two subunit interfaces that harbor the ligand binding sites. At these interfaces, αTyr127 from each α-subunit juxtaposes εAsn39 or δAsn41 in the Torpedo AChR structural model. These residues were shown to fulfill four criteria expected of structures that couple agonist binding to channel gating: (i) residue proximity, (ii) residue conservation, (iii) impaired channel gating by single mutations, and (iv) nonadditive changes in gating by double mutations.27 Residue proximity was further demonstrated by substituting Cys for the Tyr and Asn at one or both subunit interfaces, and showing impaired channel gating following treatment with the oxidizing agent H2O2 and restored gating following treatment with the reducing agent DTT. The overall analyses indicate that αTyr127 and εAsn39 or δAsn41 associate in the diliganded closed state, priming the receptor for opening, but that the pair dissociates in the open state.

Opening of the Pore

Structural loops from the ligand binding domain of the α-subunit associate closely with the tops of the M1–M3 α-helices. Thus in principle, motions of all three TMDs could contribute to opening of the channel. State-dependent motions of the channel were first elucidated by electron microscopy of the Torpedo AChR at a resolution of 0.9 nm, and suggested the M2s rotate about their long axes upon activation.16 Rotation of the M2s was also suggested by state-dependent disulfide cross-linking of the GABAA receptor.28 Computational studies, including normal mode analysis, MD simulation, and principal component analyses, suggest both a twisting and tilting of the M2s.2931 Lysine substitutions along M2 enable proton block of single-channel currents, with the extent of block giving a picture of residues lining the conduction pathway in the open state.32 This picture concurred with the Torpedo AChR structural model, presumed to be in the closed state, suggesting a simple dilation rather than twisting of the M2s. We observed a 0.1 nm increase of the pore radius following steered MD simulation that changed the C-loop of the neuronal α7 AChR from uncapped to capped conformations, but this resulted from both rotation and tilting of the M2s.33 Our recent MD simulations of single-cation translocation suggest peristaltic motions of M2 and may account for all the proposed motions.34

The speed at which the channel opens and closes depends on the structure of the pore itself. Mutations in many parts of M2 increase the frequency and duration of spontaneous channel openings and prolong ACh-induced openings.3537 The stereo image in Figure 5 shows TMDs of α (magenta) and ε (gold) subunits of the muscle AChR viewed from the pore, and highlights locations of mutations found to cause CMS. Side chains of some mutated residues project between α-helices, while others project into the ion conduction pathway. In M3, however, substitutions of Ileu or Leu for αVal285 shorten mean channel open time and produce two kinetic classes of openings at saturating concentrations of ACh, whereas substitution of Ala increases channel open time and maintains a single kinetic class of openings.38 Because αVal285 articulates with M2, substitution of a larger side chain may impair channel gating by restricting or altering motion of M2.

Figure 5
CMS mutations mapped onto AChR TMDs. Stereo image of the channel domain for two subunits is viewed from the central vestibule, with the α-subunit in magenta and ε-subunit in gold. Residues found to be mutated in different CMS cases are ...

Future Prospects

We can ask what questions remain to be answered and how the answers will impact our understanding of the AChR and its function in health and disease states of the neuromuscular junction. The questions fall into two main categories: Structural: Although atomic resolution structural data have emerged, a pressing need remains for more complete structures, structures at higher resolution, and structures of different functional states. Structural advances will directly enable computational studies of the AChR to understand the dynamic changes that drive its function as a ligand-operated ion channel. Atomic resolution structures will also allow structure-based drug design of competitive and noncompetitive activators and inhibitors. Mechanistic: A standard allosteric mechanism describes major features of the receptor activation process, but functional consequences of disease-causing mutations provide clear evidence for additional open and closed states, calling for a more comprehensive mechanism. Mechanistic advances will allow understanding of the detailed mechanisms underlying disease-causing mutations required for rational therapy.


Conflicts of Interest

The authors declare no conflicts of interest.


1. Engel AG, Ohno K, Sine SM. Sleuthing-molecular targets for neurological diseases: the neuromuscular junction as a paradigm. Nat. Rev. Neurosci. 2003;4:339–352. [PubMed]
2. Takeuchi A, Takeuchi N. Active phase of frog’s endplate potential. J. Neurophysiol. 1959;22:395–412. [PubMed]
3. Unwin N. Nicotinic receptor at 9 Å resolution. J. Mol. Biol. 1993;229:1101–1124. [PubMed]
4. Noda M, Takahashi H, Tanabe T, et al. Structural homology of Torpedo californica acetylcholine receptor deduced from cDNA sequence. Nature. 1983;302:528–532. [PubMed]
5. Unwin N. Refined structure of the nicotinic acetylcholine receptor at 4 Å resolution. J. Mol. Biol. 2005;346:967–989. [PubMed]
6. Brejc K, van Dijk W, Klassen R, et al. Crystal structure of an ACh-binding protein reveals the ligand-binding domain of nicotinic receptors. Nature. 2001;411:269–276. [PubMed]
7. Hansen S, Sulzenbacher G, Huxford T, et al. Structures of Aplysia AChBP complexes with nicotinic agonists and antagonists reveal distinctive binding interfaces and conformations. EMBO J. 2005;24:3635–3646. [PubMed]
8. Celie PHN, van Rossum-Fikkert SE, van Dijk WJ, et al. Nicotine and carbamylcholine binding to nicotinic acetylcholine receptors as studied in AChBP crystal structures. Neuron. 2004;41:907–914. [PubMed]
9. Sine SM. The nicotinic receptor ligand binding domain. J. Neurobiol. 2002;3:431–446. [PubMed]
10. Jackson MB. Perfection of a synaptic receptor: kinetics and energetics of the acetylcholine receptor. Proc. Natl. Acad. Sci. 1989;86:2199–2203. [PubMed]
11. Monod J, Wyman J, Changeux JP. On the nature of allosteric transitions: a plausible model. J. Mol. Biol. 1965;12:88–118. [PubMed]
12. Ohno K, Wang H-L, Milone M, Bren N, et al. Congenital myasthenic syndrome caused by decreased agonist binding affinity due to amutation in the acetylcholine receptor ε subunit. Neuron. 1996;17:157–170. [PubMed]
13. Sine SM, Claudio T, Sigworth FJ. Activation of Torpedo acetylcholine receptors expressed in mouse fibroblasts. Single channel current kinetics reveal distinct agonist binding affinities. J. Gen. Physiol. 1991;96:395–437. [PMC free article] [PubMed]
14. Zhang Y, Chen J, Auerbach A. Activation of recombinant mouse acetylcholine receptors by acetylcholine, carbamylcholine and tetramethylammonium. J. Physiol. 1995;486:189–206. [PubMed]
15. Hatton CJ, Shelley C, Brydson M, et al. Properties of the human muscle nicotinic receptor, and of the slow-channel myasthenic syndrome mutant εL221F, inferred from maximumlikelihood fits. J. Physiol. 2003;547:729–760. [PubMed]
16. Unwin N. Acetylcholine receptor channel imaged in the open state. Nature. 1995;373:37–43. [PubMed]
17. Unwin N, Miyazawa A, Li J, Fujiyoshi Y. Activation of nicotinic acetylcholine receptor involves a switch in conformation of the alpha subunits. J. Mol. Biol. 2002;319:1165–1176. [PubMed]
18. Gao F, Burghardt T, Bren N, et al. Acetylcholine-mediated conformational changes in acetylcholine-binding protein revealed by simulation and intrinsic tryptophan fluorescence. J. Biol. Chem. 2005;280:8443–8451. [PubMed]
19. Gao F, Mer G, Toneli M, et al. Solution NMR of acetylcholine binding protein reveals agonist-mediated conformational change of the C-loop. Mol. Pharmacol. 2006;70:1230–1235. [PubMed]
20. Mukhtasimova N, Free C, Sine SM. Initial coupling of binding to gating mediated by conserved residues in muscle nicotinic receptor. J. Gen. Physiol. 2005;126:23–39. [PMC free article] [PubMed]
21. Bouzat C, Gumilar F, Spitzmaul G, et al. Coupling of agonist binding to channel gating in an AChbinding protein linked to an ion channel. Nature. 2004;430:896–900. [PubMed]
22. Dellisanti CD, Yao Y, Stroud JC, et al. Crystal structure of the extracellular domain of nAChR alpha1 bound to alpha-bungarotoxin at 1.94 Å resolution. Nat. Neurosci. 2007;10:953–962. [PubMed]
23. Lee WY, Sine SM. Principal pathway coupling agonist binding to channel gating in the nicotinic receptor. Nature. 2005;438:243–247. [PubMed]
24. Bouzat C, Gumilar F, Esandi M, Sine SM. Subunit-selective contribution to channel gating of the M4 domain of the nicotinic receptor. Biophys. J. 2002;82:1920–1929. [PubMed]
25. Mitra A, Bailey T, Auerbach A. Structural dynamics of the M4 transmembrane segment during acetylcholine receptor gating. Structure. 2004;12:1909–1918. [PubMed]
26. Sine SM, Engel AG. Recent advances in Cysloop structure and function. Nature. 2006;440:448–455. [PubMed]
27. Mukhtasimova N, Sine SM. An inter-subunit trigger of channel gating in the muscle nicotinic receptor. J. Neurosci. 2007;27:4110–4119. [PubMed]
28. Horenstein J, Wagner D, Czajkowski C, et al. Protein mobility and GABA-induced conformational changes in GABA(A) receptor pore-lining M2 segment. Nat. Neurosci. 2001;4:477–485. [PubMed]
29. Taly A, Delarue M, Grutter T, et al. Normal mode analysis suggests a quaternary twist model for the nicotinic receptor gating mechanism. Biophys. J. 2005;88:3954–3965. [PubMed]
30. Cheng X, Lu B, Grant B, et al. Channel opening motion of alpha7 nicotinic acetylcholine receptor as suggested by normal mode analysis. J. Mol. Biol. 2006;355:310–324. [PubMed]
31. Cheng X, Ivanov I, Wang HL, et al. Nanosecond time scale conformational dynamics of the human alpha7 nicotinic acetylcholine receptor. Biophys. J. 2007;93:2622–2634. [PubMed]
32. Cymes G, Ni Y, Grosman C. Probing ion-channel pores one proton at a time. Nature. 2005;438:975–980. [PMC free article] [PubMed]
33. Cheng X, Wang HL, Grant B, et al. Targeted molecular dynamics study of C-loop closure and channel gating in nicotinic receptors. PLoS Comput. Biol. 2006;2:1173–1184. [PMC free article] [PubMed]
34. Wang HL, Cheng X, Taylor P, et al. Control of cation permeation through the nicotinic receptor. PLoS Comput. Biol. 2007;4:e41. [PubMed]
35. Ohno K, Hutchinson DO, Milone M, et al. Myasthenic syndrome caused by a mutation in the M2 domain of the acetylcholine receptor ε subunit. Proc. Natl. Acad. Sci. 1995;92:758–762. [PubMed]
36. Milone M, Wang HL, Ohno K, et al. Slow channel myasthenic syndrome caused by enhanced activation, desensitization, and agonist binding affinity due to mutation in the M2 domain of the acetylcholine receptor α subunit. J. Neurosci. 1997;17:651–5665. [PubMed]
37. Grosman C. Free-energy landscapes of ion-channel gating are malleable: changes in the number of bound ligands are accompanied by changes in the location of the transition state in acetylcholine-receptor channels. Biochemistry. 2003;42:14977–14987. [PMC free article] [PubMed]
38. Wang HL, Milone M, Ohno K, et al. Acetylcholine receptorM3 domain: Stereochemical and volume contributions to channel gating. Nat. Neurosci. 1999;2:226–233. [PubMed]