|Additional related information may be found at:|
|Neuropsychopharmacology: The Fifth Generation of Progress|
New Drug Design in Psychopharmacology
The Impact of Molecular Biology
John F. Tallman and Svein G. Dahl
The design of therapeutics for the treatment of neurological and psychiatric disorders has undergone a number of epochs or phases in the modern era. One of the earliest examples in the first generation was the discovery (36) that antihistamines (promethazine) could lead to drowsiness which might be useful in the treatment of psychiatric disorders, including schizophrenia. This serendipitous discovery led to the development of chlorpromazine. By demonstrating that a drug could influence the course of a chronic disorder, chlorpromazine not only revolutionized treatment in psychiatry but also brought legitimacy to the concept of biological psychiatry.
A second phase of progress involved the use of in vivo animal models in the setting of pharmaceutical discovery research that led to the discovery (again serendipitous) of the sedating, muscle-relaxant, and taming effects of chlordiazepoxide, and ultimately to many other benzodiazepines (51). The increasing use of biochemical models (e.g., monoamine reuptake and metabolism) and the incorporation of biochemistry into pharmacological receptor theory (e.g., the discovery of adenylate cyclase) led to a third phase of progress in which in vitro methodology was used to discover drug candidate leads. This list (L-dopa, tricyclic antidepressants, etc.) is quite long and contains many of the therapeutics discussed throughout this volume.
A landmark study for drug identification and design was the discovery of specific radioligand binding to the opiate receptor (47); this study ultimately led to the biochemical identification of the multiple receptor subtypes within a number of neurotransmitter families. Such a multiplicity of subtypes had previously been predicted indirectly through pharmacological measurements of both in vivo and in vitro activities of agonists and antagonists. Over the last 20 years, the precision of drug design has been assisted by the study of receptor binding affinities and receptor localization.
The cloning of receptor molecules and other advances in molecular biology, together with the development of high-powered computers and software for calculation of three-dimensional molecular structures, holds promise for the fourth and perhaps the most precise period where new drug candidates are developed by rational design from the three-dimensional structures of target molecules. The isolation of individual receptor proteins by virtue of their pharmacological and high-affinity binding properties has allowed the determination of portions of their primary amino acid sequence. In turn, this has resulted in the isolation of the primary nucleic acid sequence coding for these receptors and the discovery of hitherto undetected similarities between receptors for distantly related neurotransmitter receptors and differences within what were thought to be single receptor subtypes. It is these studies in this fourth phase of progress that are the focus of this review with the particular perspective of indicating how they will impact molecular design and discovery within the pharmaceutical industry (see Basic Concept and Techniques om Molecular Genetics, Cytology and Circuitry, Neuronal Nicotinic Acetylcholine Receptors: Novel Targets for Central Nervous System, Molecular biology, Pharmacology, and Brain Distribution Subtypes of the Muscarinic Receptor, Cholinergic Transduction, Single Transduction Pathways for Catecholamine Receptors, Seretonin Receptors: Signal Transduction Pathways, and Intracellular Messanger Pathways as Mediators of Neural Plasticity, for related issues).
THE TOOLS OF MODERN DRUG DISCOVERY
Isolation of Cloned Receptors
Four techniques have been of value in identifying the novel lead clones in individual receptor and channel families. Each has its advantages and disadvantages and its proponents. All are characterized by the need for a great degree of cleverness, persistence, and luck for their success. A flow chart for proceeding from pharmacological activity to clone is shown in Fig. 1.
The first technique is to identify a partial amino acid sequence of the receptor protein to be cloned; the immediate limitation is that enough protein must be available for microsequencing. Using degenerate oligonucleotides covering all the possibilities of cDNA coding for this protein sequence, a cDNA library prepared from mRNA of a tissue enriched for the receptor of interest is screened and a number of partial- or full-length clones are identified, isolated, and sequenced. The open reading frames are identified and the full length of the receptor is determined.
The advantage of this approach is that the sequence of the cDNA codes exactly for the protein that is expressed in the tissue of interest. If the screening is done at high stringency, unique clones are identified and the full amino acid structure of the receptor is elucidated. One of the first receptor clones on record, that of the a subunit of the Torpedo californica acetylcholine receptor, was obtained this way over 10 years ago (45).
The disadvantages of this method of screening for clones is that if the known sequence of amino acids is not unique or if the amino acids are coded for by highly degenerate triplets, multiple inconsistent clones are identified. A second problem is that partial sequences from non-full-length clones are identified but full-length sequences cannot be obtained. This problem arises because of the nature of the construction of the library from mRNA by reverse transcriptase and its lack of specific proofreading mechanisms or large amounts of nontranslated RNA that may be in the native mRNA. This is a persistent problem with any type of screen that does not rely on functional activity and particularly hampered the original identification of dopamine D4 receptors (64). The solution chosen by these investigators was to splice a portion of sequence derived from the genomic form of the receptor to the cloned cDNA; while this was a clever approach at the time, recent studies of the AMPA-type glutamate receptor subunits indicate that post-transcriptional modifications may alter critical amino acids due to base modification as the mRNA is processed in the cell (56). This modification is enzymatic and appears to be under specific control separate from that of the gene being expressed; the prevalence and distribution of this enzyme and its function are obscure at the present time. A more vexing problem for the resulting D4 receptor clone was the presence of an intron in the genomic sequence that limited high-level expression of this clone in some cell lines. An intron-deleted form of the dopamine D4 receptor or synthetic gene has led to much higher levels of receptor expression (unpublished results).
The use of functional expression of a receptor to be cloned without significant structural information would appear to circumvent many of the problems described above. Here a constructed library in an expression vector or mRNA itself is used to create a receptor protein in a recipient cell. The cells are then screened for a particular pharmacological activity and the library is subdivided; successive iterations lead to a pure clone. The difficulties of this approach beyond library construction are the need to have very specific assay systems that can determine receptor presence through binding or functional activity. For the biochemical approach, a very-high-affinity ligand of high specific activity, usually iodine-labeled, is needed to allow the detection of a vanishingly small number of positive cells among many possible false leads. For the functional approach, an expression system (such as the Xenopus oocytes) and a specific functional assay are needed. Ionotropic receptor family members for which a single subunit can create a channel are the best candidates for this approach. In this case the response element is built into the protein itself and exogenous transducers, like G proteins, are not required. A real limitation of this approach is that proteins which require other subunits to function (multimeric proteins) or interact with other proteins that are not expressed in the cell to carry out their function are not amenable to cloning in this way. As examples, this approach has led to the cloning of (a) the only serotonin receptor (5-HT3) in the ionotropic or ligand-gated superfamily (41) and (b) the 5HT2c (formerly called 5HT1c) receptor, a G-protein-linked receptor. The 5HT1c receptor was identified by using its specific pharmacology along with multistep activation (G-protein-mediated) of an endogenous Ca2+-gated Cl- channel in Xenopus oocytes (37).
The third approach derives intellectually from the old studies of inborn errors of metabolism and the isolation of mutant proteins from patients with such disorders. The major success thus far in the channel area has been the isolation of voltage-dependent potassium channels from a Drosophila mutant Shaker, identified phenotypically because the flies convulsively twitch their legs upon exposure to ether. The identification of the A current and the isolation of genomic clones in the vicinity of the Shaker mutation led (by chromosomal walking) to the Shaker gene. Here the nucleic acid sequence and gene were cloned before the protein was isolated and identified as the potassium channel (29). No novel receptor genes have yet been identified in this precise way.The fourth major method for discovering novel clones is through screening of libraries with previously discovered clones at reduced stringency, by using sequences common to members of the superfamily as probes, or by using these sequences as primers for polymerase chain reaction (PCR)-based techniques (16). These are by far the most widely used techniques and account for the discovery and identification of most monoamine and peptide receptor genes described to date. Here investigators have been assisted by the unexpectedly large similarities across a great variety of receptors in the G-protein-linked superfamily. Such techniques also account for the discovery of most of the GABA, glutamate, and other ionotropic receptor subunits. Most of the later references of this chapter report receptors identified by these methods. Clearly, the methods are efficient at proliferating subtypes, but are dependent upon a starting point within a family and some knowledge of a potential relationship between receptors (see Excitatory Amino Acid Neurotransmission, GABA and Glycine, Cholinergic Transduction, Single Transduction Pathways for Catecholamine Receptor, and Seretonin Receptors: Signal Transduction Pathways).
PCR-based techniques can identify novel orphan clones of unknown sequence (43). However, almost everyone who has worked with these techniques also can report the amplification of unknown sequences from contaminating materials in their libraries or reagents. This can be most vexing because subcloning and sequencing are needed to separate new and interesting sequences from the false leads. Most frequently these PCR-generated sequences are not long enough to contain the entire coding sequence of a receptor, and these sequences are used for further library screens. The list of full-length putative receptor proteins in search of a function, or orphan clones, is increasing and provide unique challenges to those who discover them. Among the receptors that were identified as orphan clones is the cannabinoid receptor, derived first from a tachykinin receptor (27). Some of the mas oncogenes related to the seven-transmembrane spanning region proteins have been found as orphan clones (67).
Expression Systems and Their Applications
A number of methods to express the protein coded for by the cDNA after it has been cloned have been used. While yeast and bacterial systems can yield the largest amounts of protein for study, their use in expression of mammalian receptors has been limited. In an interesting series of experiments, two forms of human b-adrenergic receptor cDNA have been expressed in Escherichia coli and show insertion into bacterial membranes and interactions with G proteins (14). However, bacteria do not generally carry out post-translational modification such as leader sequence removal, intron deletion, glycosylation, and so on, and animal cells are generally much more useful.
Xenopus oocytes are the traditional cells for expression of ion channels and ionotropic receptors. Each oocyte is individually injected with mRNA prepared in vitro from the cDNA; this allows functional studies to be easily carried out, but oocytes are not very useful for biochemical studies. One of us (JFT) has adapted this system to routine drug screening as part of a drug discovery program. Affinity and efficacy at several GABAa receptor subtypes can be determined for a single drug in one day's work; structure–activity relationships can be developed. Large-scale screening of libraries of compounds is not possible with this technique, and it is most useful for lead optimization and modeling.
Mammalian expression systems have been useful for the G-protein-linked family of receptors, and many cell lines permanently expressing single receptors have been developed. Large amounts of cell membrane and passage of cells through a number of generations yield enough material for large-scale screens. Frequently, the specific G protein that the native receptor interacts with is not present in the recipient cell, and this is a limitation of the mammalian cell expression. While adenylate cyclase is used in many cases as a surrogate measure, it is known that a much wider diversity of intracellular messages are controlled by receptors in vivo. Almost any fibroblastoid cell can serve as a recipient, and cookbooks of molecular biology describe the many methods for transfection. The expression of multisubunit proteins is more difficult because of the statistical possibilities for expression of one, two, or more subunits in a single cell. A recent advance in virology and molecular biology of insect viruses has allowed the development of a method based upon the baculovirus infection of insect cells to be applied to receptor expression. The advantages of this system are that very high levels of multimeric proteins can be produced and that control of the absolute levels of infection and ratios of subunits used can be obtained (38). The disadvantage is that the cells that express the receptors are insect cells and may not be identical to mammalian cells in their processing of proteins. This system can easily produce enough receptor protein for library screening of either a pharmaceutical library (archives of compounds synthesized by a company), natural product libraries, or various combinatorial libraries (peptides, nucleic acids, heterocyclic compounds). From these libraries, leads can be obtained for the identification of novel receptor-specific drugs.
Computational Chemistry and Molecular Modeling
When the three-dimensional structure of a receptor molecule is known, new potential drug molecules may be custom-tailored, by molecular modeling techniques, to precisely fit into a binding pocket on the receptor. This approach has been called "structure-based" or "rational" drug design, as opposed to more random synthesis and screening of a series of substances. Over the last 20 years, three-dimensional structures of many proteins have been determined by x-ray crystallographic methods, which has shed new light on their mode of action. Methods of modern molecular biology have made it possible to produce larger quantities of pure receptor material than did previous conventional methods. However, neurotransmitter receptors have several hydrophobic segments in the peptide chain, which probably represent domains of the protein that are embedded into the cell membrane. This makes it difficult to purify and crystallize receptor proteins, and a detailed x-ray crystallographic structure determination of any such receptor has not yet been reported. We and others have therefore used molecular modeling techniques to build three-dimensional receptor models from the amino acid sequences, and have used these models to study the molecular mechanisms of drug–receptor interactions.
Modern molecular modeling techniques contain four key elements: computer graphics, quantum mechanical calculations, molecular mechanical calculations, and molecular dynamics simulations. Computer graphic visualization of the spatial arrangement of atoms and the distribution of electrostatic potentials in molecules has proven to be a powerful tool for explaining structure–activity relationships of biologically active compounds. High-resolution graphics computers offering stereo viewing, real-time translation, and rotation of molecular models are now available for prices not much higher than those for common workstations. Such computers enable assembly of various parts of a receptor model (such as a seven- or five-helical bundle), as well as inspection of how the various receptor segments fit to each other, to be done by interactive computer graphics techniques.
Energy Minimization of Molecular Models
Quantum mechanical methods allow relatively accurate calculation of molecular structures and electrostatic potentials of small molecules, but require far too much computing time to be applicable for calculations of protein structures. Molecular mechanics calculations are less accurate but a lot faster, and they enable refinement of a protein structure with several hundred amino acid residues to be done on a workstation. The method requires a set of atomic coordinates to start the calculation from. This may be a crystal structure, a molecular model built by interactive computer graphics, or a structure obtained during a molecular dynamics simulation.
Structure refinement by molecular mechanics calculations separates overlapping atoms and usually corrects unrealistic bond lengths and angles. The method does not, however, change a molecular structure across conformational energy barriers, but merely searches for the nearest minimum in the potential energy of the molecule. Once a receptor model has been built by interactive computer graphics or by other methods, it therefore more or less retains its overall structure after molecular mechanical energy minimization.
Molecular Dynamics Simulations
Most of the receptor models that have been proposed up to now were built and examined as static structures. However, proteins in their native state in cells have constantly changing geometries, with movements occurring on a femtosecond (10-15 sec) time scale. Molecular dynamics simulations have been used to (a) study the naturally occurring internal movements in proteins and other biologically active molecules and (b) refine three-dimensional molecular structures. Such simulations, which always are started from an energy-minimized molecular structure, involve a large number of computational steps. Simulations of macromolecules, even over a period of a few picoseconds (10-12 sec), are therefore facilitated by using a supercomputer. In molecular dynamics simulations, kinetic energy is added to the molecular system. A structure may therefore move across conformational barriers and undergo substantial changes during such simulations. For example, a seven-helical protein model may change from an initial circular arrangement of the helices into a more oval and bacteriorhodopsin-like shape during 20–25 psec of molecular dynamics simulation (10, 28)
The classical experiments of Anfinsen (1) suggested that all information required to direct the folding of a protein into its tertiary structure lies in the amino acid sequence. Prediction of secondary and tertiary structures from the amino acid sequence has, however, proven to be extremely difficult, and no generally applicable method for accurate prediction of tertiary protein structures has yet been found. Receptor models therefore have to be based on structural data in addition to the primary structure of the protein. Results from cloning, expression, and ligand binding studies of site-specific mutants and chimeric receptors have been particularly useful for construction and refinement of neurotransmitter receptor models.
Such receptor models are often built from a-helical models of each membrane-spanning segment. These are constructed from the amino acid sequence, and they are assembled by interactive computer graphics or by automated superposition upon an other receptor model. Particular residues postulated to be directly involved in ligand binding are often used to determine the relative positions of individual helices. Several models of G-protein-linked receptors that have been published contain only the seven transmembrane a-helices; these models were constructed by superposition on a model of the membrane-spanning a-helices in bacteriorhodopsin, based on electron diffraction data (22). Other receptor models have also included the loops between helices and the terminal parts (3, 10, 39, 60). The receptor models are usually refined by molecular mechanical energy minimization. Some have also been refined by molecular dynamics simulations, which may lead to a tighter packing of transmembrane (TM) helices and a certain tilt in some of the helices.
However, regardless of the computational procedure used for structure refinement, the features of the starting model are crucial for the final, energy-minimized receptor model. At present, the lack of any detailed ionotropic or metabotropic receptor structure from x-ray crystallographic studies limits the accuracy of all available receptor models. It has now become a trivial task to calculate minimum-energy conformations of a protein and examine how various ligands fit into a known binding site, once the three-dimensional protein structure is known at atomic resolution. Similar calculations may easily be performed on any receptor model, but the validity of the results depends on the accuracy of the model and has to be verified by experimental methods.
STRUCTURE OF NEUROTRANSMITTER RECEPTORS IN TWO MAJOR SUPERFAMILIES
While the discovery technologies simplify the identification of new receptors, and molecular modeling has reached a new level of sophistication, knowledge of the structure of members of the receptor superfamilies is needed to apply these techniques. Some of the common and unique properties of two major receptor superfamilies are discussed next. Chemical communication involves the release of a pulse of neurotransmitter that results in the brief attainment of a very high concentration of transmitter at its receptor in the synaptic cleft. This results in the direct opening of an ion channel in the case of fast transmission mediated by the ionotropic receptors. In the case of slower transmission, this receptor activation of second messenger systems via a G-protein mediator results in metabolic events within the cell or mediation of indirect channel opening.
Ionotropic Receptors and Their Modulators
The transmitters that currently are known as fast transmitters are acetylcholine, glycine, glutamate, GABA, and serotonin. Almost all of these transmitters can also mediate slow transmission through a separate set of G-protein-linked receptors. Structural and functional data that have been derived from molecular biological studies of the sequence and activity of expressed receptors have pointed out the unexpected nature of the phylogenetic relationship of these receptors and allow the information derived from the study of one receptor to be applied to the others. Thus, much of our interpretation of data gathered about glycine, GABA, glutamate, and 5-HT3 receptors is based on the hard-won knowledge about acetylcholine receptors. Some modification of these interpretations will be needed as each receptor and the component proteins are studied in turn, but we can classify the ionotropic receptors together as a superfamily.
Fig. 2 shows a generic representation of how we think the transmembrane organization of a ligand-gated channel subunit occurs. A pentameric structure of the nicotinic cholinergic receptor composed of five of these subunits is thought to be the standard form. Evidence for this is based upon electron microscopic and cross-linking studies (63). By analogy, this structure is thought to be the form of other members of the superfamily.
Each member of the superfamily shows significant amino acid sequence homology in TM1–TM4 (transmembrane segment). The amino acids of TM2 are of particular interest because they are thought to form the pore or ion channel for all members of this receptor superfamily; it is these amino acids that have been highly conserved between and within each family. The secondary structure of the protein segment forming the channel is thought to be helical. These amino acids include a high proportion of hydrophobic amino acids; and at spaced intervals they include neutral amino acids, such as serine and threonine. These may contribute to the inner components of the walls of the ionic channel. Charged amino acids at the ends of these sequences and other amino acids within the channel appear to control ion selectivity both through direct ionic interactions and through control of the geometry of the channel. Specific mutations of these charged amino acids have been carried out in the a7 nAchR (nicotinic acetylcholine receptor), which can form a homomeric channel and thus has been a useful model for mutagenesis because of its simple (a7)5 structure. Some of these mutations alter which cations are allowed to enter (permeability Ca2+ versus permeability Na+), which might be expected. Other mutations of a7 nAChR with homologous amino acids of the glycine receptor actually convert a7 ion channel selectivity from cationic to anionic (19).
Within this superfamily, the amino acids lining the channel form the binding pockets for drugs such as chlorpromazine (nAChR), phencyclidine (NMDA), MK801 (NMDA), some anesthetic barbiturates (GABA and glycine), ondansetron (5HT3), and some insecticides (nAChR and GABA). Because of the great structural similarity of the channel amino acids within a single family, pharmacological subtyping has not been tried and the development of subtype specific drugs would be a difficult effort. However, some interesting naturally occurring point mutations have been identified, including one within the TM2 that may account for pesticide resistance in Drosophila (15). Such molecular studies as these also answer some of the questions of cross-relationships for drugs acting in the channel, but still remaining are the more philosophical questions: When did the nAChR and the other receptors diverge, and why did the ion selectivities change? (See Neuronal Nicotinic Acetylcholine Receptors: Novel Targets for Central Nervous System).
Some of the most divergent parts of the ionotropic receptor superfamily are the extracellular domain and the loop between TM3 and TM4. The extracellular domains contain the neurotransmitter binding sites and some of the modulatory sites. The intracellular loop contains consensus sequences for various protein kinases which control assembly and function of each receptor. Space does not permit us to focus on all the functions of these receptor domains and all the members of this superfamily, so we have chosen to focus on the GABA and part of the glutamate family. These have provided, and continue to provide, some of the most interesting drugs in psychopharmacology.
The GABAa receptor subunits have been classified into five separate groups. For example, the a subunits have been classified together solely based upon homology to one another versus lower homology to the b, g, and d subunits (4). Much similarity is found in the putative transmembrane regions. Comparison of all the N-terminal amino acids of the subunits of the GABAa receptor complex at an amino acid level show little overall homology; however, some of the most interesting binding and modulatory sites are found here. The assembly of homomeric subunits is poorly accomplished, and for GABA it appears that more than one subunit is needed to express the interesting pharmacology of this receptor system. Using the baculovirus expression system, the addition of b subunits to a subunits yields a muscimol binding site and a fully functional channel. The further addition of a g subunit (g2) to the baculovirus infection yields a receptor that also contains a benzodiazepine receptor (20).
The conclusion that we draw is that at least part of the benzodiazepine binding site or structure required for the formation of the benzodiazepine binding site is conferred by the coexpression of the g subunits with a and b. Expression of a with g or b with g in this system does not result in a functional benzodiazepine receptor. Thus we can account for three of the five subunits of the GABAa receptor complex. Is there more than one of each subunit? The baculovirus system allows an approach to this problem by using varying multiplicities of infection. If the ratio of g is increased beyond one, the number of Ro15-1788 sites does not increase, it decreases; one tentative thought is that a single g is part of the complex (20). More than one a or b subunit per receptor still remain the more likely possibilities, while a fourth potential subunit, a d, does not significantly change the pharmacology of the complex. The inclusion of the ab with a g lowered the number of sites for muscimol, indicating the likely interaction of a and b to form a GABA binding site. The whole issue of receptor assembly is still under investigation. In an interesting series of experiments with nAChR subunits (18), the normal routes of receptor assembly were shown to involve the formation of intermediate forms of the receptor complex with fewer number of subunits; these forms of the receptor are not mature; the subunit insertion and conformational changes that allow this to occur may take hours to days. The study of assembly of subunits is a topic of great interest to cell biologists and will be an important future research area.
The GABA binding site appears to be coded for in the N-terminal portion of the GABAa receptor; and photoaffinity labeling with muscimol, a rigid analogue of GABA, has identified phenylalanine 61 in the a subunit as intimately involved in the binding site (57). Because the determinants of GABA that are required for binding are relatively simple, subtyping of this site, by combining different a subunits with b and g, has not yet appeared in the literature. Examination of large libraries of more rigid GABA analogues with receptor subtypes reconstructed from clones should be useful in identifying any subtle differences in the binding site for GABA. If they exist and are substantial, these differences could be exploited by developing a pharmacophore model for each subtype that can be used pharmacologically in the development of directly acting agonists with subtype specificity.
One subtyping of the benzodiazepine response at GABAa receptors preceded the era of molecular biology; that was the type I/II difference originally seen with a Lederle compound (CL218,872) and more recently seen with the hypnotic zolpidem (Ambien) and its weaker analogue, the anxiolytic alpidem (50). For zolpidem, the pharmacology is clear from a binding perspective. Zolpidem is much more potent in displacing from an a1containing construct (e.g., a1b1g2) than from other constructs. A molecular basis of glutamate (a2, etc.) substituting for glycine (a1) at amino acid position 201 accounts for this difference (50). From a functional point of view, some of these differences are less clear. Zolpidem at high doses produces an agonist activity at more than the one subtype, and at higher doses the specificity of the response disappears. Expression of subtypes and screening with individual GABAa receptor subtypes can discover drugs with different receptor subtype specificities. Within the next year, drugs that have been discovered in the current period of discovery that possess substantially greater specificity for one or another GABAa receptor subtype will enter the clinic. Then, we will be able to find out if subtype-specific, rather than subtype-selective, modulatory sites on the GABAa receptor complex analogous to the benzodiazepine site can have similar modes of action or represent improvements over existing medications.
A second question for the drug developer in the GABAa receptor field is whether a single drug can have a multiplicity of actions at different receptor subtypes? Here we return to the concept of agonist, antagonist, and inverse agonist developed originally to encompass the benzodiazepine blocking activities of flumazenil and the anxiogenic and proconvulsant activities of the b-carbolines (61). The answer to this question is yes. For the full agonists, the subtype differences are not interesting because at pharmacological doses in animals all of the subtypes effects are blurred or overwhelmed by the most active form. A more interesting situation is in the case of drugs with agonist activities at one subtype and inverse agonist activities at another. Is there an anxiolytic cognitive enhancer in the group? The answer is perhaps; again we will know when these agents enter the clinic.
What do the clinician and behavioral pharmacologist need to do to adjust to the era where drugs are discovered by their molecular properties rather than by their in vivo activities? Perhaps the focus should be on new ways of assessing pharmacological activities. Because sedation may not be a component of these anxiolytics, tasks where sedation is a measure of component activity are not appropriate. In a similar way, anxiety and proconvulsant activity are not a component of cognitive enhancement; therefore, proconvulsant or anxiogenic activities should not be used to measure preclinical efficacy of inverse agonists as cognitive enhancers. Rating scales related to the disorder and new ways of specifically determining clinical efficacy will be needed. The magic bullets of the near future will possess these activities and will allow the clinician to return with confidence to GABAa mechanisms that have proven relatively safe over the last 30 years, and with these refinements they are likely to be even safer (see also GABA and Glycine).
The opposite of inhibition is excitation. A large number of neurons in the central nervous system use glutamate as a transmitter to produce neuronal excitation; in parallel, glutamate may also influence neuronal plasticity and cause neurotoxicity. The glutamate receptors also consist of two groups: (i) the ionotropic receptors including N-methyl-D-aspartate (NMDA) receptors and the a-amino-3-hydroxy-5-methyl-4 isoxazole propionate (AMPA)-kainate receptor and (ii) the metabotropic glutamate receptors that are coupled to G proteins and modulate the production of intracellular messengers. Over 15 different members of the glutamate family of receptors are known to exist (56).
Within the ionotropic glutamate receptors there are a number of classes. One class consists of four subunits (GluR1-4) and shows high affinity for AMPA, whereas another very similar set consists of two kinds of kainate-selective subunits (Glu 5-7 and KA 1-2). A further related set of clones d1 and d2 has also been found. All of these subunits are substantially larger than the other members of the family (glycine, GABA, nAChR): >100 kD as opposed to 50 kD (56).
Expression of the AMPA subunits show differences in current–voltage relations and permeability to Ca2+. Thus, they are true subtypes controlling different ions in addition to different kinetics in vivo. In particular, GluR2 subunits have particularly weak Ca2+ permeability and dominate the electrical properties of a complex in which they are found. GluR2 has an arginine within TM2 at position, whereas the other subunits contain a glutamate residue. In certain regions of the brain, absence of this form translates to an AMPA-kainate receptor on cells with a permeability to Ca2+, rather than just monovalent cations. More surprising is the finding that the genomic forms of all the AMPA subunits contain glutamate codon (CAG) at the Glu-Arg position even though an arginine codon (CGG) is found in cDNAs for GluR2, GluR5, and GluR6. Apparently, the adenosine-to-guanosine alteration and amino acid change is caused by an RNA editing mechanism, which is enzyme-driven and posttranscriptional. This mechanism occurs with varying efficiency and has an impact on the Ca2+ permeability of the resulting receptor. For the drug discoverer, this type of post-transcriptional editing points out a potential pitfall of clones obtained when molecular biologists use genomic sequences instead of cDNAs for cloning (56).
A further complexity of the AMPA-kainate receptors is increased by alternate splicing between TM3 and TM4. The two alternate splice molecules, flip and flop, import distinct kinetic and amplitudes of response to the AMPA-kainate family. Clearly this family of receptors is diverse and widely distributed in brain and does much of the day-to-day work of glutamate transmission. No drugs in the clinic are yet available to modulate this system, and it takes a back seat to the more glamorous cousins, the NMDA receptors. However, specific modulation of the subtypes of this receptor system could possess many of the advantages (and disadvantages) of the modulators of the GABAa receptors.
NMDA receptors can be reconstituted as heteromeric structures of two subunit types, namely, NR1 along with one of four separate NR2 subunits. In molecular terms, the NR1 are only slightly related to the NR2 subunits (18% sequence identity). The NR2 subunits are also different from the rest of this class because their carboxy-terminal regions have about 550 extra amino acids. Structural models, based upon the nAChR and GABAa receptors, thus may not be applicable here. NR1 is the most common form of NMDA receptor and exists in different splice forms. It can form homomeric receptors but also more efficiently forms heteromeric receptor sites. The important conceptual framework for NMDA receptors is that the NMDA receptor may be viewed as primarily an NR1 receptor widely distributed throughout brain, uniquely modulated in different areas by the formation of heteromeric NR1–NR2 combinations. Here again, we don't know the specific subunits composition of the complex or even the number of subunits per complex. Drug discovery depends upon the mix and match of these subunits (probably in twos) for screening purposes.
Drugs with specificity for the ion channel, like MK-801 and phencyclidine, have been shown routinely to cause psychotomimetic effects at low doses and strong sedation in the therapeutic range (53). The competitive NMDA binding site blockers also seem to share similar properties. Both classes of drug cause a strange transient phenomena called vacuolization in rodent cingulate cortex (46). This finding has slowed regulatory review of most drugs in this class.
The modulatory sites of the NMDA receptors (analogous but not identical to benzodiazepine sites on GABAa receptors) appear to be places to start in drug discovery for the NMDA receptor (33). One such site has been termed the nonstrychnine glycine binding site. This modulatory site normally responds to tonic levels of glycine, and its occupancy by glycine is absolutely required for NMDA agonist activity. Antagonists and agonists for this site already exist (inverse agonist?). All modulatory drugs at the nonstrychnine glycine site studied to date have been free of the vacuolization phenomena and thus are improvements on the competitive and noncompetitive inhibitors.
The methodology that went into their discovery is traditional medicinal chemistry in which the requirements for an antagonist and agonist pharmacophore are defined and then subtlely modified to enhance the fit in the glycine site of the NMDA receptor (33). The first agonists were simple derivatives of glycine, lacking enantioselectivity. The antagonists (such as kynurenic acid) and weak partial agonists (including R-(+)-HA-966 and L 687414) also contain the structural elements of glycine, but with additional groups that confer stereospecificity on the resulting derivatives. Some of the partial agonists have been shown to have potent anticonvulsant effects, and some of the antagonists have been shown to be of interest as neuroprotective agents. Probably for these uses, relatively non-subtype-selective drugs are needed. However, anxiolytic activity could require subtype-specific agents, and a general cognitive enhancing function of agonists would certainly need some greater degree of specificity. Before we see the subtype-specific generation of compounds, we will see the relatively nonspecific drugs in the clinic. Those that are useful will continue in development. Some modulation of mesolimbic dopamine by drugs at the nonstrychnine glycine site may indicate a use for nonstrychnine glycine site modulators in schizophrenia. Active subtyping would follow in the next generation of drugs much faster than in the case of the GABAa modulators (see also Excitatory Amino Acid Neurotransmission and GABA and Glycine).
More than 200 different G-protein-linked receptors have now been cloned, and the number of cloned receptors belonging to this superfamily continues to increase at a seemingly exponential rate. As for the ionotropic receptors, the metabotropic receptors show highest sequence similarity in the putative membrane-spanning regions.
Structure of Membrane-Spanning Regions
The generally accepted concept of G-protein-linked receptors as proteins with seven transmembrane a-helices was originally based on the pioneering work of Henderson and Unwin (21). They demonstrated that the membrane protein bacteriorhodopsin in Halobacterium halobium, which has been characterized as a "photon-driven proton pump" because it translocates protons in response to light, contains seven membrane-spanning a-helices, closely packed in an oval arrangement with the chromophore, retinal, located in a central pore between the helices. This structure was later confirmed by more accurate electron microscopic and electron diffraction studies by the same group (22).
Bacteriorhodopsin is not linked to any G protein, and it has been questioned whether G-protein-linked receptors may have the same overall architecture as bacteriorhodopsin, because their amino acid sequences have only 10–15% identity with that of bacteriorhodopsin. However, the available information from a number of structural, biochemical, and biological studies has suggested a seven-membrane-spanning a-helical topography as a general structure also of G-protein-linked receptors. This has recently been confirmed by Henderson and collaborators (55) for visual rhodopsin, the photoreceptor in retinal rod cells, which activates transducin (Gt) in the visual signal transduction process (34). Two-dimensional projection maps of crystalline rhodopsin at 9-Å resolution were interpreted to show seven transmembrane helices, four being nearly perpendicular to the membrane plane.
The approximate locations of the putative membrane-spanning domains in a receptor sequence are usually determined from hydropathy indices, which may be calculated by different methods (26, 35). We have calculated such indices for the photosynthetic reaction center of Rhodopseudomonas viridis, a membrane protein for which the three-dimensional structure is known from x-ray crystallographic experiments (6). The hydropathy indices predicted the correct number of putative membrane-spanning a-helices in the protein, but not their exact length or localization in the sequence. It seems that the exact start and end points of each TM helix in a membrane protein may only be determined by x-ray crystallographic, nuclear magnetic resonance (NMR)-spectroscopic, or other physical chemistry methods. However, molecular biological alteration of amino acid sequences and expression in cells has provided valuable information about the organization of synaptic, membrane-spanning, and cytoplasmic domains of various G-protein-linked receptors (49, 54). In a review of his structure–function studies of bacteriorhodopsin and rhodopsin, Khorana (34) suggested that the membrane-spanning a-helices in bacteriorhodopsin have 20–27 residues (average 23) and that the seven a-helices in rhodopsin have 24–30 residues with an average length of 27 residues.
Structure of Synaptic and Cytoplasmic Domains
While there is at least indirect information about the three-dimensional structure of the membrane-spanning domains, there is virtually no available information about the tertiary structure of the synaptic and cytoplasmic domains of G-protein-linked receptors. Molecular biology studies with rhodopsin have suggested that the extracellular domains may have a crucial role in the folding and stabilization of the protein structure (34). A disulfide bond between cysteine residues in the extracellular domain has a functional role in rhodopsin (31) and in the a2-adrenergic receptor (8). These two cysteine residues are conserved and may have a similar function in most G-protein-linked receptors (49).
Molecular dynamics simulations have been used to model the loops between the TM helices and the N- and C-terminal parts in G-protein-linked receptors (3, 10, 39). In principle, such simulations may lead to more energetically favorable conformations than do models built only by secondary structure predictions from the amino acid sequence. It is difficult to judge how well the folding of loops and terminal parts after such simulations mimic the real structures, due to the lack of tertiary structural data for these receptor domains.
Addition of synaptic and cytoplasmic loops and terminals to receptor models has, however, demonstrated that dopamine and 5-HT receptors have a bipolar overall structure, with mainly negative electrostatic potentials at the synaptic side and positive electrostatic potentials in the cytoplasmic domains (3, 10, 60). This is illustrated in Fig. 3 for the rat dopamine D2 receptor. From this it seems likely that negative electrostatic fields around certain domains at the synaptic side may guide the positively charged neurotransmitter molecules and the protonated receptor antagonists to their receptor binding sites. Similar mechanisms have been demonstrated for substrate–enzyme interactions (66). Rational drug design from three-dimensional receptor models must therefore take into consideration that both the molecular electrostatic potentials and the three-dimensional structure of ligand molecules must fulfill certain requirements for receptor recognition and binding.
Ligand Binding Sites
Individual G-protein-linked receptors are similar in overall structure and function, but differ in key amino acid residues. Shortly after the cloning of a b2-adrenergic receptor (7), site-specific mutagenesis experiments demonstrated that residues within the postulated membrane-spanning domains were involved in ligand binding and signal transduction (58). From this, and from the proposed seven-membrane-spanning topograpny of the receptor, it was suggested that the b2-adrenergic receptor has a bacteriorhodopsin-like arrangement of seven a-helices, each with (a) a length of 20–28 amino acids and (b) a ligand binding pocket in the central pore between the helices (9, 59). The suggestion from site-directed mutagenesis studies that the antagonist binding site lies within the membrane-spanning domain of the b2-adrenergic receptor (9) has since been confirmed by fluorescence emission spectroscopic experiments (62), which have shown that the antagonist binding site lies within the membrane-spanning segment, about 12 Å from the synaptic membrane surface.
Site-directed mutagenesis studies have shown that two conserved serine residues in TM helix 5 are involved in agonist binding and activation of b2-adrenergic (59), dopamine D1 (48), and dopamine D2 (40) receptors. It has been postulated that these serine residues are involved in ligand binding and may form hydrogen bonds with the hydroxyl groups in catecholamines (59).
Site-directed mutagenesis studies have also demonstrated that aspartic acid residues in TM helices 2 and 3 have functional roles in metabotropic receptors. An aspartic acid residue in TM3, which is conserved in G-protein-linked neurotransmitter receptors, is required for antagonist and agonist binding, and it has been postulated that this residue acts as a counterion in the binding of protonated ligands (54). A conserved aspartic acid residue in TM2 plays an important role in agonist-induced signal transduction in G-protein-linked neurotransmitter receptors. It is not clear, however, whether this residue is directly involved in agonist binding. It has been suggested that the corresponding residue acts as a counterion for agonist binding to b2-adrenergic (65) and 5-HT1A receptors (24) and for binding of Na+ to a2-adrenergic and dopamine D2 receptors (54).
It is possible that agonists and antagonists may have different but overlapping binding sites in G-protein-linked neurotransmitter receptors. Many receptor models place the conserved Asp residue in TM2 further down in the central core of the receptor than the conserved Asp residue in TM3. This arrangement offers a steric explanation of how the Asp in TM2 may be involved in signal transduction, and possibly also in agonist binding, without affecting antagonist binding. In this mode, the neurotransmitter moves down the central core of the receptor, as its own positive electrostatic potentials are attracted to negative potentials created by aspartate residues in TM3 and TM2, to "dock" the transmitter in the receptor. This attraction between the transmitter and the receptor changes the conformation of the receptor protein, leading to G-protein stimulation and subsequent intracellular biochemical changes.
Acetylcholine, dopamine, 5-HT, norepinephrine, and other receptor ligands have flexible structures, which move rapidly between various conformations as the ligand approaches the receptor. Our molecular dynamics simulation of ligand–receptor complexes suggests that transmitters as well as antagonists bind to metabotropic receptors by a "zipper" mechanism, in which the ligand adjusts its conformation as it approaches and binds to the receptor. These simulations also suggest that the amine groups of dopamine and serotonin may interact with more than one site in the receptor.
Because of their flexibility and ability to change conformation as a result of binding, the interaction of neuropeptides with their receptors, most of which are in the G-protein-linked family, are likely to be quite complex to model using the above techniques. However, a few unifying principles are beginning to emerge from studies of the molecular biology of their receptors. The first is the continuation of the discovery of unexpected relationships between seemingly unrelated peptide receptors. This is exemplified by the high degree (37%) of identity of delta opiate receptors to somatostatin, angiotensin (31% identity), and the chemotactic receptors (about 20%). This may indicate that they control similar processes in the cell, and more importantly for the drug discoverer, they partly explain observations like the interactions of somatostatin analogues with the opiate receptor (12). That common motifs exist within the peptide family is supported by studies that implicate an amino–aromatic interaction between his-197 of the NK-1 receptor and a low-molecular-weight antagonist, CP 96,345. This amino acid is located at the extracellular surface of the fifth transmembrane helical domain and may indicate a more superficial area of binding for peptides to their receptors than the biogenic amines (13). It also indicated that there may exist common areas on peptide receptors similar to those on the biogenic amine receptors that are hot spots for drug design (see also Cholinergic Transduction, Single Transduction Pathways for Catecholamine Receptors, Seretonin Receptors: Signal Transduction Pathways, Corticotrepin-Releasing Factor: Physiology, Pharmacology, and Role in Central Nervous Systemand Immune Disorders, Monomine Oxidase: Basic and Clinical Perspectives, Vasopressin and oxytocin in the Central Nervous System, Thyrotropin-Releasing Hormone: Focus on Basic Neurobiology, Somatastatin in the Central Nervous System, Galanin: A Neuropeptide with Important Central Nervous System Actions, The Neurobiology of Neurotensin, Cholecystokinin, and Intracellular Messanger Pathways as Mediators of Neural Plasticity, for related issues).
The three-dimensional structures of two guanine nucleotide binding proteins, EF-Tu and p21ras, have been determined by x-ray crystallography (5, 30). These structures, together with data from various biochemical experiments, have been used to model a subunits of G proteins (25, 42), which are coupled to membrane receptors (32). A model of transducin (Gta), constructed from the crystal structure of Cu–Zn superoxide dismutase (52), was used to propose a possible mechanism for interaction between transducin and rhodopsin (23).
Biological experiments have produced substantial information about which receptor domains are interacting with G proteins. As the receptor models become more refined, and more information about the structure of various G proteins becomes available, a next step may be to build models of receptor–G-protein complexes, use these models to suggest new site-directed mutagenesis experiments, and use the results of such experiments to refine the models of G-protein–receptor complexes.
RECEPTOR MODELING AND DRUG DESIGN
One of the aims of receptor modeling is to explain the specificity of various ligands and to make correct predictions about receptor binding affinities of ligands which have not yet been examined in receptor binding assays. Despite the theoretical effort to fold G-protein-linked receptors based upon a knowledge of their amino acid sequences, the determination of protein structure requires the application of additional techniques to confirm such folding. X-ray crystallographic techniques and protein NMR are two of the most advanced analytical techniques currently being implemented in drug design; however, the hurdles blocking their application to the study of receptors for psychotherapeutics are formidable.
Except for the models described earlier, x-ray crystallographic study of receptors is at a primitive level. While the steps needed in a structure determination are clear, the first step of obtaining receptor protein crystals is the major limitation for these studies to be successful. Because of the extensive transmembrane structure required for their function, even the solubilization of many receptors has not been possible; this includes several receptors whose amino acid sequences are known. There is little question that x-ray crystallography can be powerful in producing atomic resolution of proteins and their complexes with ligands or transmitter substances. This information can be used either (a) a posteriori to rationalize structure–activity relationships as has been done in the design of human rhinovirus inhibitors or (b) a priori as has been done in the production of inhibitors of human immunodeficiency viral proteases (HIV PR) that are required for viral replication (11). Here crystallographic information led to the realization that HIV PR was structurally related to aspartic proteases such as renin and that a reduced or simplified renin inhibitor could serve as a starting point for discovery efforts. A second starting point was the search of a database for structures that show complementarity to the active site; from this study (11), the surprising finding that haloperidol could inhibit HIV PR was made. A third approach attempted to mimic the pharmacophoric aspects of the inhibitor enzyme interactions (11); from these studies a central OH (hydroxy), two hydrogen bonds, and an aromatic group along with their spacing were identified.
Besides x-ray crystallography, NMR is the only other technique with the ability to produce details at the atomic level. An advantage is that crystals are not required. However, at present, NMR is limited to proteins with molecular weight less that about 40 kD due to the broad signals obtained for proteins of this size. An additional limitation is the need for high concentrations and solubility (about 1 mM) of proteins being studied. The results with solution NMR when it has been applied to suitable problems are spectacular. The interactions of the immunosuppressant cyclosporin with its target cyclophilin have been visualized, and conformation of cyclosporin bound to cyclophilin has been determined. This is quite different from the conformation previously determined in solution both by x-ray crystallography and by NMR spectroscopy (11). This also can provide a starting point for additional structure–activity determination and development of additional drug candidates.
While advances like these are still in the future for the discoverer of psychotherapeutics, the ability to obtain human receptor subtypes and to express these subtypes in cell lines at high levels provides a powerful screening tool for existing and novel drug libraries. In addition to conventional libraries of drugs that most pharmaceutical companies have prepared in the course of drug discovery, new libraries based upon small building blocks have been prepared (44). These blocks can be nucleic acids, amino acids, or heterocyclic precursors. What is unique about all these approaches is the random nature of the compounds formed; these libraries can be classified as diversity libraries, and the approach is a combinatorial approach. Because no a priori assumptions are made about the compounds, their structures can theoretically fill all or any space. With a high-throughput screen using a cloned receptor expressed in a cell line, it is literally possible to screen millions of possible drug candidates for a lead structure. While not as intellectually satisfying as a purely rational approach, it has resulted in the discovery of important drug candidates and the identification of a number of useful templates for solving diverse problems in drug design. One such template is the 1,4-benzodiazepine template (Fig. 4); closely related compounds have solved problems that cross a diverse group of receptor families, including GABA, cholecystokinin, opiate, PAF receptor, and HIV PR (2). Perhaps common motifs of binding crossing families of receptors exist; and there are common sites for these templates to interact with, along with unique elements that are specific to each receptor.
The use of cloned receptors as targets for drug discovery is both an evolution and a revolution. It is an evolution because their use is a logical outgrowth of the historical progress in the pursuit of rational drug design by pharmaceutical firms. While we have not yet reached a point where a purely rational approach is possible, the use of cloned expressed receptors allows a previously unattainable level of chemical precision in drug design. It allows not only the attainment of high affinity for the target receptor, but also specificity through lowering affinity for other receptor subtypes through the use of reverse screening techniques. This may result in the diminution of unwanted side effects in the next generation of drugs. The importance of side effects can be underscored by the focus that continues to be placed on them and their interference with dosing both acutely and during chronic treatment. These are major themes of several chapters in this volume.
Another evolution in drug design is in the use of human receptor homologues for design of new therapeutic entities. The recent experience of the Pfizer group in the discovery and development of nonpeptidic substance P antagonists underscores the importance of having activity at human receptors for potential human therapeutics. The high affinity of CP 96,345 as a potential human therapeutic would not have been discovered through the use of substance P receptor assays with rat brain membranes, where its affinity is 100-fold lower (or rat clones for that matter), nor could it have even been easily identified though traditional pharmacological models in rodents. These differences have been explored through the use of chimeric receptors and have a clear molecular basis (17). Enough similar experiences are emerging to indicate that human receptor testing should be the norm in choosing a clinical candidate in the future.
The use of receptor clones has been revolutionary also. It has led to the discovery of unexpected relationships between the receptors for distinct transmitter families (e.g., the ionotropic receptors) and a realignment of traditional pharmacological classifications of subtypes within the monoamine families of receptors (e.g., the reclassified 5-HT2c receptors are more related to 5-HT2a than type 1 forms). An emerging theme (exemplified by the relation of d-opiate receptors to somatostatin receptors) is that commonality of biochemical function (whether channel opening or activation of second messenger system) may indicate closer phylogenetic relationships than the transmitter that interacts to activate the receptor; this has implications for the starting point of drug design and choice of targets. The complexity of the second messenger systems also remains for further exploration.
The use of receptor clones most significantly expands our horizons for drug discovery. It identifies new members within a family where the older drugs may have exerted parts of their therapeutic actions (e.g., clozapine's action at dopamine D4 receptors) and thus new and specific targets for therapeutics. It allows totally novel receptors to be identified within a family where no therapeutics yet exist [e.g.: the many new serotonin (5HT4-?) receptors] and should spur interest in novel mechanisms. Finally, the technology allows sufficient amounts of rare receptors to be produced to allow them to be explored for their biological and future therapeutic importance.
Receptor modeling techniques have provided new insight into receptor mechanisms, and they have been used to design biological experiments. Modeling of receptor molecules, along with molecular dynamics simulations of ligand–receptor interactions, has provided some clues both to the nature of ligand–receptor associations and to the structural transitions that occur upon agonist binding. Until a detailed crystal structure and related biological activity of an ionotropic or metabotropic receptor molecule is available, molecular modeling techniques may not be expected to provide information about the fine structure of such proteins and detailed receptor mechanisms. Molecular modeling of receptor molecules, based on the amino acid sequences of the protein and other available structural data, may still provide information which may be used in a rational design of potential drug molecules. The rapid development of molecular modeling applications indicate that such techniques will play an increasingly important role in future drug design.
The presence of advanced technologies for drug discovery and design can lead to the discovery of magic bullets. How these magic bullets can be used in the clinic and what disorders can be treated with these receptor-specific drugs still remains the domain of the astute and observant clinician. Perhaps we have come full circle with a new generation of precise agents whose ultimate utility in the clinic will be determined not only rationally, but also serendipitously.