Receptor–Receptor Interactions, Receptor Mosaics, and Basic Principles of Molecular Network Organization: Possible Implications for Drug Development

JOURNAL OF

MolecularNeuroscienceEditor-in-Chief: ILLANA GOZES, PhD

JOURNAL OF

MolecularNeuroscienceEditor-in-Chief: ILLANA GOZES, PhD

Volume 26 Nos. 2–3, 2005 ISSN: 0895–8696

HumanaJournals.comSearch, Read, and Download

Proceedings of the Wenner-Gren Foundation Symposium

Receptor–Receptor Interactions Among Heptaspanning Membrane Receptors:From Structure to Function

Special Issue:

JMN_26_2-3_cvr 6/14/05, 12:02 PM1

Journal of Molecular NeuroscienceCopyright © 2005 Humana Press Inc.All rights of any nature whatsoever reserved.ISSN0895-8696/05/26:193–208/$30.00DOI:10.1385/JMN/26:02:193

Journal of Molecular Neuroscience 193 Volume 26, 2005

ORIGINAL ARTICLE

Receptor–Receptor Interactions, Receptor Mosaics, and BasicPrinciples of Molecular Network Organization

Possible Implications for Drug Development

Luigi F. Agnati,*,†,1 Alexander O. Tarakanov,2 Sergi Ferré,3Kjell Fuxe,4 and Diego Guidolin5

1Section of Physiology, Department of Biomedical Sciences, University of Modena, 41100Modena, Italy; 2St. Petersburg Institute for Informatics and Automation, Russian Academy of

Sciences, St. Petersburg, Russia; 3Behavioral Neuroscience Branch, National Institute on DrugAbuse, NIH, DHHS, Baltimore, MD 20817; 4Department of Neuroscience, Division of Cellular

and Molecular Neurochemistry, Karolinska Institute, S-171 77 Stockholm, Sweden;and 5Department of Anatomy and Physiology, University of Padua, Padua, Italy

AbstractThe phenomenon of receptor–receptor interactions was hypothesized by Agnati and Fuxe in the 1980s, and

several indirect proofs were provided in the following years by means of in vitro binding experiments and invivo experiments in physiological and pathological animal models. This paper aims to outline some of the mostimportant features and consequences of this phenomenon in the frame of the structural and functional aspectsof molecular networks. In particular, the concepts of receptor mosaic (RM), and of horizontal and verticalmolecular networks (HMNs, VMNs, respectively) are illustrated. To discuss some aspects of the functionalorganization of molecular networks, not only new data on protein–protein interactions but also the biochemi-cal mechanism of cooperativity will be used. On this basis, some theoretical deductions can be drawn that allowa tentative classification of the RMs and the proposal of the extension of the concept of branching point introducedfor enzymes to the possible switching role of some RMs in directing signals to various VMNs. Finally, thecooperativity phenomenon and the so-called symmetry rule will be used to introduce a proper mathematicalapproach that characterizes RMs as to their receptor composition, receptor topography, and order of receptoractivation inside the RM. These new data on G protein-coupled receptors and molecular network organizationindicate possible new approaches for drug development.

DOI:10.1385/JMN/26:02:193

Index Entries: Receptor–receptor interactions; receptor mosaics; molecular networks; cooperativity.

*Author to whom all correspondence and reprint requests should be addressed. E-mail: [email protected]†This paper is dedicated to Professor Ermanno Manni, former Professor of Human Physiology in Rome.

JMN_040048_FERRE.qxd 01/06/2005 05:36 pm Page 193

194 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

Historical Overview of the Emergence of the Concept of Receptor–ReceptorInteractions and Some DeducedTheoretical ConceptsEmergence of the Concept

of Receptor–Receptor Interactions and RMs

The concept of receptor–receptor interactions(RRIs) has a history of about two to three decades.From a theoretical standpoint it is possible todistinguish cross talk between receptors (or, moregenerally, proteins) from the concept introduced in1980 (Agnati et al., 1980; Fuxe et al., 1981; Fuxe andAgnati, 1985, 1987) of the existence of intramem-brane RRIs, which can take place between completelydifferent receptor types. The proposal as stated inthe Wenner-Gren Center symposium held in Stockholm in 1986 indicated that intramembraneRRI could be a special class of protein–protein inter-action (see below and quotation in Fig. 1). More recentdata have clarified that protein-protein interactionsusually occur between protein domains, whichtypically have 100–250 residues. A domain is anevolutionary unit whose coding sequence canduplicate and/or undergo recombination (Chothiaet al., 2003). Some sequences appear as multidomainproteins and can adopt different linear arrangementsof their domain sets. On average, such domainarchitectures comprise two or three domains;however, some human proteins contain up to 130domains (Wuchty, 2001). Protein domain interactionsare essential to the cell functioning by acting inseveral ways, such as domain–domain interactionsin multidomain polypeptide chains and interchainprotein interactions in multimers and in transientprotein complexes (Park et al., 2001). By now it isclear that RRIs belongs to this vast field of proteindomain interactions. Hence, the individuation anddecoding of the functional domains in the G protein-coupled receptors (GPCRs) is of paramountimportance, and only recently substantial progresshas been made (see Milligan et al., Reynolds et al.,Woods et al., this issue).

Let us start by analyzing Fig. 1. The classicaldefinition of cross-talk between receptors is asfollows: “...interaction between two receptors suchthat one receptor, when activated, modulates theaction of the other. This type of interaction ofteninvolves the production by the first receptor of a second messenger, which then modulates the actionof the second receptor” (Smith, 1997). From the

beginning, we have proposed a more articulate viewby considering RRIs as a different phenomenonfrom receptor cross-talk, by putting forward thehypothesis that receptors could form clusters of func-tionally interconnected proteins at plasma mem-brane level and that the more general frame ofprotein–protein interactions at membrane levelshould be considered (Fuxe and Agnati, 1987). Thus,as shown in the upper panel of Fig. 1, under the head-ing Cross-Talk Between Receptors (proteins), inter-actions are indicated that do not take place directly(physical interaction between proteins or, at most,favored by adapter and/or scaffold proteins) but areinstead mediated by an additional phenomenon—either a bioelectrical phenomenon (changes in mem-brane polarization, left part of the panel) or anenzymatic cascade (phosphorylation/dephospho-rylation, right part of the panel). In several instanceswe have pointed out that the idea of possible cross-talk between membrane proteins (and, hence,between receptors) could be deduced by Hodgkin-Huxley’s model on the molecular mechanism of exci-tation (Hodgkin and Huxley, 1952), which has beenone of the greatest advances in neurosciences and,as such, a potential source of important deductions.

Shepherd (1983) underlines that one of the attrib-utes of a good theory is that it postulates specificand new properties that can be tested experimen-tally (in agreement with the methodology of falsi-fication procedures proposed by Popper [1935]), andHodgkin-Huxley’s model has this characteristic.These investigators had the extraordinary insightnot only to conceive ionic conductances in terms ofchannels that permit movements of ions across themembrane but also to recognize that there are mol-ecular gates that control the opening and closing ofthe channels. Now the view is accepted that chan-nels are generally multimeric proteins and gatinginvolves changes in the channel’s conformation and,in some cases, the conformation change is regulatedby changes in membrane potential (Doyle, 2004).

Thus, it was a reasonable deduction to surmisethat ion fluxes via an ionotropic receptor can causechanges in the membrane potential, which in turncan change the conformation of membrane proteins,such as other receptors (see also Agnati et al., 2005b).As stated above, this view is an obvious consequenceof Hodgkin-Huxley’s hypothesis of the gating of ionchannels (e.g., changes in conformation of proteins)in response to changes in the membrane potential.

In the 1970s, Greengard and collaboratorssuggested the involvement of metabotropic receptors

JMN_040048_FERRE.qxd 01/06/2005 05:36 pm Page 194

Receptor Mosaics 195

Journal of Molecular Neuroscience Volume 26, 2005

that could alter the conformation of other membraneproteins, such as receptors, via a process of phos-phorylation/dephosphorylation (Greengard, 1976).During the same period, Limbird and coworkers(Limbird et al., 1975, Limbird and Lefkowitz, 1976)demonstrated a site–site interaction between β-adrenergic receptors (see Fig. 1, bottom left).

In the 1980s, a new paradigm was proposed (seeFig. 1, bottom right), namely that one receptor (e.g.,for peptides) could change conformation of anotherreceptor for a completely different ligand (e.g., formonoamines) via a direct physical interaction atintramembrane level, and this interaction could alsooccur within a cluster of topographically ordered

receptors (Agnati et al., 1980, 1986, 1995, 2003a; Fuxeet al., 1981; Agnati and Fuxe, 1984; Fuxe and Agnati,1985, 1987). Furthermore, in 1987, it was proposedwhat can be seen as an ante litteram program ofproteomics: “Very likely we will find that some sophis-ticated elaborations of information are performed atmembrane level, via interactions within and amongdifferent classes of macromolecules (such as receptors,ion pumps, ion channels....” (Fuxe and Agnati, 1987).

This assumption was the result of previoustheoretical papers in which deductions were madestarting from the indirect evidence that we had atthat time on RRIs (see Fig. 2). We pointed out thatthe conceptual frame used by McCulloch and Pitts

Fig. 1. A schematic view of the developments of the concept of RRIs, illustrating at least two broad classes: indirectinteractions between membrane proteins (e.g., receptors) involving a bioelectrical (membrane polarization) or an enzy-matic (phosphorylation/dephosphorylation) phenomenon; and direct interactions, based on an intramembrane phe-nomenon of protein–protein interactions. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:36 pm Page 195

196 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

(1943) could be applied at the molecular level leadingto an extraordinary miniaturization of thecomputational circuits (Agnati and Fuxe 1984; Agnatiet al., 2002; 2004b). Furthermore, these membrane-associated microcircuits, when located at pre- andpostsynaptic levels, could represent a mechanismto alter in a prompt way the synaptic weight and,hence, could represent one of the mechanismsinvolved in learning and memory processes, possiblyof paramount importance for short-term memory(Agnati et al., 1982, 2002, 2003b, 2004a). In thesepapers, we have therefore introduced the concept ofthe receptor mosaic (RM), meaning that one canproduce markedly different molecular aggregates ofreceptors by putting together the same buildingblocks to construct functionally different assemblies.As an analogy, by using the same set of tesserae (fromWebster ’s Dictionary: tessera, a small cube or squareof different materials as marble, precious stones,

ivory, glass, wood, etc., used in mosaic work), it ispossible to obtain markedly different mosaics.

These proposals, as well as the new data fromproteomics, led us to view the RM as a special case ofpostgenomic cabled protein mosaic (pgPM). As Paskoand Ringe (2004) underline, it can be surmised thatsome multimeric proteins are the result of the fusionof genes that once coded for separate proteins (seeFig. 3); hence, genomic cabled protein mosaics (gPMs)are expressed. Recently, we have suggested that weshould consider pgPMs, that is, the building of mul-timeric proteins by different, independently expressed,protein domains. The different functional meaning ofthe two types of PMs should be noted. The gPMs canbe thought of as multimeric proteins that cablenanoscale molecular circuits (Agnati et al., 2002) withrelatively low plasticity, as they carry out a fixed impor-tant function and/or represent big building blocks ofmicroscale molecular circuits (i.e., molecular

Fig. 2. Some of the main theoretical deductions drawn from the concept of RRIs are schematically illustrated. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:36 pm Page 196

Receptor Mosaics 197

Journal of Molecular Neuroscience Volume 26, 2005

networks). The pgPMs can be thought of as multi-meric proteins that are assembled in different waysaccording to the transient needs of the cell. Thus,pgPMs show high plasticity that is further enhancedby the fact that the chemical-physical influencesexerted by the microenvironments (ions, pH, tem-perature) affect the single domains. Hence, not onlythese influences can modulate the function of the newlyassembled multimeric complex in asophisticated waybut also can affect the process of domain assemblingin a multimeric complex (Agnati et al., 2004b).

Emergence of the Concept of Horizontal and Vertical Molecular NetworksThe RMs are located at plasma membrane level

and are part of molecular networks; thus, it is animportant aspect to analyze where they are located

and the organization of the molecular network towhich they belong. In the 1990s (Agnati et al., 1990,1995), we stressed the relevance of the plasma mem-brane microdomains for neuron integrative func-tions. Furthermore, it has been suggested that a firstbroad subdivision of the complex tree of moleculesforming molecular networks could be made by indi-cating the cell compartment where they are located.Hence, the following concepts were introduced:

1. Horizontal molecular networks (HMNs)—molecularnetworks made by membrane-associated ormembrane-integral proteins. That is, a HMNrepresents the molecules located within a membranemicrodomain, or associated with the plasmamembrane microdomain, that are physicallyconnected or communicate via a signal released anddiffusing within the membrane (e.g., diacyl glycerol).

Fig. 3. Schematic illustration of the structural organization of a HMN and the possible functional role of an integratedinput unit of a RM. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:36 pm Page 197

198 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

In some cases, molecules can also bulge out from theplasma membrane toward the extracellular spacewhere they may interact with other molecules (e.g.,ectoenzymes such as adenosine deaminase; see Francoet al., this issue).

2. Vertical molecular networks (VMNs)—molecularnetworks that deepen into the cytoplasm, oftenreaching into the nucleus.

These concepts are illustrated in the lower panelof Fig. 2, with more details in Fig. 4. Horizontalmolecular networks (HMNs) are two-sidedinput/output networks, that is, they can receivesignals from the extracellular as well from theintracellular environment of the cell and can giveteleological responses—responses aimed at main-taining cell homeostasis, which affect both the extracellular and the intracellular environment of

the cell. Thus, HMNs work as complex interfacesbetween the extracellular and intracellularenvironments (Agnati et al., 2002, 2003a) andhave been indicated as a target of electronic andchemical signals present at the local circuit level(Agnati et al., 2005b).

The concept of HMNs is also supported by thedemonstration in the plasma membrane of special-ized microdomains in which molecules involved inextracellular signal detection and transduction areconcentrated (Okamoto et al., 1998). Among these, ofparticular relevance are the so-called lipid rafts, whichdenote dynamic assemblies of cholesterol and sphin-golipids scattered within a fluid, disordered phase ofthe lipid bilayer (Simons and Ikonen, 1997). Thus,lipid rafts concentrate special classes of lipids andproteins that function in transmembrane signaling

Fig. 4. Multimeric proteins can be assembled at the genome level or postgenomically. The basic similarity betweenpostgenomically assembled multimeric proteins and RMs is schematically illustrated. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 198

Receptor Mosaics 199

Journal of Molecular Neuroscience Volume 26, 2005

events. Hence, lipid rafts have emerged as special-ized membrane platforms for HMNs. Our proposalof HMNs has similarities with Lisanti’s proposal ofsignalosomes (Okamoto et al., 1998) and the view ofthe RM as an integrated input unit to the molecularnetworks with Bockaert’s proposal of GPCRs orga-nized in a functional unit (Bockaert et al., 2003).

It should be noted that lipid rafts are not simplyplatforms for HMNs, but lipids in the lipid rafts havea role both in the structural organization of theHMNs and in the elaboration of the information car-ried out by the HMNs. In particular, lipid–proteininteractions are of importance, as lipids tend to adopt a superlattice distribution and the phospholipiddistribution is affected by the charge of the headgroups (Virtanen et al., 1998). It is clear that theplasma membrane is not a uniform bilayer butcontains islands of markedly different chemicalcomposition and structural characteristics (Gil et al.,1998; Paratcha and Ibanez, 2002). Lipid-mediatedprotein–protein interactions should also bementioned (Gil et al., 1998). There is a selectiveaccumulation of the lipid species that hydrophobi-cally matches the protein hydrophobic domain at theprotein–lipid interface and, hence, wets this inter-face (lipid annulus). A wetting layer can be sharedby two or more proteins giving rise to proteinclusters; hence, it can contribute to the formation ofRMs. In the frame of our previous discussion, it isimportant to mention data suggesting that the exactcomposition and nature of the α-helices will sub-stantially influence how and in which phospho-lipid environment this polypeptide chain can exist(Lewis et al., 2001). In conclusion, there are prob-ably very strict relationships between the lipid raftsand the HMN that uses lipid rafts as a platform,as (1) the lipid environment affects the proteinsthat can be inserted into the membrane; (2) the pro-teins inserted in the membrane can be organizingcenters for distribution and arrangements of lipidsin the raft; (3) the proteins inserted in the mem-brane can attract and repulse other charged mole-cules (lipids and proteins) in the raft, allowing acontinuous reshuffling of the molecular distribu-tion in the raft; and (4) both lipids and proteins inthe lipid raft can have a structural and functionalrole in the HMN.

If these assumptions are correct, it would be ofimportance to discover not only the protein domaininteractions involved in the assembly and functionof a HMN but also protein–lipid and lipid–lipidinteractions that could affect these aspects.

Furthermore, data on the oligomerization of recep-tors, as well as on receptor trafficking and on thebiochemical mechanisms controlling these aspects,are important in understanding RM formation andplasticity and how HMNs can affect cell function.An aspect is obviously the permanence of receptorsat membrane level in different functional conditions.In particular, in basal conditions, the time ofpermanence of receptors at membrane level is rele-vant for the organization and the plastic adjustmentsof the RM and, hence, for the integrative functionsof HMNs. In the case of agonist stimulation, whichcan induce receptor clustering (Franco et al., 2003),the time of permanence of receptors at membranelevel can give hints to better understand thedesensitization phenomenon (see also Genedaniet al., this issue).

It can be surmised that HMNs are formed by dif-ferent classes of molecules, and some of them aretargets for other cytoplasmic and extracellularmolecules. Of obvious importance are those actingon the GPCRs (hence on RMs) as they regulate thelevel of inputs to the cell biochemical machinery, thatis, to the VMNs. Among these are (1) moleculescontrolling GCPR function (e.g., regulators of Gprotein-signaling proteins [Ishii and Kurachi, 2003]);(2) molecules controlling GPCR internalization (e.g.,β-arrestins and caveolins [Gainetdinov et al., 2004;Genedani et al., Franco et al., and Lefkowitz et al.,this issue]); or (3) molecules controlling GPCR desen-sitization (e.g., GPCR kinases, which are key mod-ulators of GPCR signaling and can also interact withcaveolins [Penela et al., 2003]).

These are some of the molecules that likely keepunder control the inputs that reach the VMNs,address the signal impinging on the cell only to someof the VMNs among all those that are potentiallyinterconnected with the HMN receiving theextracellular signals, and can also give off their ownsignals to the intracellular biochemical machinery(Miller and Lefkowitz, 2001).

In view of the concept of RMs it will be importantto assess which receptors form heterodimers or high-order hetero-oligomers and how the receptors areorganized within the RM. Some of these aspects arediscussed below, together with some aspects of thefunctional organization of molecular networks thatcan be related to RM organization. To this aim notonly new data on protein–protein interactions butalso the biochemical mechanism of cooperativity willbe used. On this basis, some theoretical deductionscan be drawn that allow a tentative classification

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 199

200 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

for the RMs and the proposal of the extension of theconcept of branching point introduced for enzymes(Koshland and Hamadani, 2002) to the possibleswitching role of some RMs in directing signals tovarious VMNs. These deductions might representa contribution to a possible better understanding ofthe functional organization of the molecularnetworks. Great efforts have been made and excellentpapers have been published on possible formalapproaches to the molecular networks (Wattsand Strogatz, 1998; Weng et al., 1999; Lauffenburger2000; Kitano, 2002a, 2000b; Wutchy, 2001). Hence, itmight be of interest to briefly discuss a mathematicalapproach that has been introduced recently (Agnati et al., 2005a) to characterize a limited partof the molecular networks, namely the RMs. The present formal approach allows the RMcharacterization as to the receptor composition, recep-tor topography, and order of activation of the recep-tors inside the RM.

Biochemical Mechanisms Involved in the Assembling of RMsChemical Aspects of Protein–Protein Interactions

Proteins are the main constitutive elements ofmolecular networks, especially of RMs, as proteinspossess the Lego property, that is, the capability of astereochemical directionality with regard to other pro-teins resulting in high-molecular-weight complexesthat are capable of emergent functions (Agnati et al.,2002, 2003a). Allosteric modulation of a protein (or areceptor) often represents a preliminary and neces-sary step for the appearance of the Lego property andthus of protein complexes endowed with emergentcharacteristics.

Evaluation of the protein–protein interactionshas also been studied by means of quantitativeapproaches, and it has been shown that the map ofinteractions between protein families has the formof a scale-free network; that is, most protein familiesinteract with one or two other families, whereasa few families are extremely versatile in their inter-actions and are connected to many families (Parket al., 2001). Obviously, it is of paramount importanceto carry out such analysis also for GPCRs, althoughit should be kept in mind that the functionalrepertoire of proteins and likely GPCRs is furtherenhanced by the formation of complexes with othertypes of molecules such as membrane lipids andcarbohydrates.

Structural and functional aspects of protein–protein interactions can be analyzed in a chemicalgeneral frame (Borgan and Thorn, 1998; Kortemmeand Baker, 2002):

1. Most interaction interfaces are composed of relativelylarge protein surfaces (larger than 600 Å2), withcomplementary conformations and electrostatic saltbridging for enhanced stability.

2. A small set of hot-spot residues at the interfacescontribute significantly to the free energy binding ofthe protein-protein interaction.

3. Hot spots are mainly clustered at the centers of theinterfaces.

4. Hot spots are protected from contact with bulk solventby peripheral residues that do not contribute signif-icantly to the binding energy of the protein–proteininteraction. This ring of peripheral residues excludessolvent from the center, where a reduction of theeffective dielectric constant strengthens electrostaticand hydrogen bonding interactions.

It is proposed that sometimes it will be possibleto give to interacting protein domains a functionalmeaning, besides a chemical (structural) meaning.For example, some interacting domains might beimportant primarily for allosteric interactions;furthermore, different allosteric modulations mightoccur according to which interacting domains areeither tightened or weakened. This view is inagreement with the so-called moonlightingphenomenon that refers to a protein exerting adifferent function by using parts of the proteinthat are different than its original active site (Jamesand Tawfik, 2003).

In 1995, it was suggested that RRIs could occurvia interactions of extracellular, intramembrane, andintracellular domains of the GPCRs (Agnati et al.,1995). Although some of these interactions are thepreferred ones for a certain GPCR family (see, e.g.,the interactions via the C-terminals for GABAB1 andGABAB2 receptors belonging to family III; Parmen-tier et al., 2002), it should be considered that severalconformations are possible and can address the RRI.It has been suggested that a protein has a ruggedenergy landscape with many local minimacorresponding to an ensemble of preexistingstructures with similar but discrete energy levels(James and Tawfik, 2003). This view is in agreementwith our proposal of plasticity of the building ofpgPMs also in response to microenvironmental chem-ical-physical conditions (see above, Agnati et al.,2005b, and Fig. 3). Within this frame we can discusssome structural and functional aspects of GPCRs and

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 200

Receptor Mosaics 201

Journal of Molecular Neuroscience Volume 26, 2005

of the oligomerization process (see also Agnati et al.,2005b). The following might be surmised:

1. Receptor C-terminus and N-terminus might not be ofcrucial importance for the formation of a stable oligomer,in view of the small area of the interfaces, unless largeadapter proteins interacting with the C-terminus andN-terminus and with several loops of the receptors areinvolved. In the case of the involvement of large adapterproteins, the existence of strong interaction spots clus-tered at the core of the interacting surfaces should bedemonstrated, together with a peripheral ring ofresidues for the exclusion of the solvent.

2. Intramembrane RRIs can be of high importance in viewof the low dielectric constant and, hence, the very effec-tive electrostatic and hydrogen bonding interactions.

3. Intramembrane RRIs might rely on an interactionsurface large enough to account for the stabilityobserved for receptor dimers and oligomers. The α-helices are at least 30 Å long (20 residues), span-ning the thickness of the hydrophobic portion of thelipid bilayer, and about 5 Å large. From these data itcan be deduced that usually more than one helix isinvolved in the RRI to overcome the minimal size (600Å2) necessary to achieve a stable interaction.

Because, in several cases, receptor oligomers areformed before reaching the plasma membrane (Lee etal., 2003), it might be suggested that in some instancesother biochemical mechanisms (e.g., molecules oper-ating as chaperones) can stabilize the oligomer beforeits insertion into the plasma membrane. Furthermore,it is likely that in most cases, extracellular, intramem-brane, and intracellular domains of the GPCRs areinvolved in RRIs, even if on the basis of the availabledata the stability of the macromolecular complex mightbe achieved above all because of interactions amongthe intramembrane helices and/or the bridging actionof adapter proteins. The possibility that proteins caninteract also with the extracellular domains of theGPCRs to help the stabilization of receptor oligomersshould not be overlooked.

Biochemical Mechanisms Involved in the Assembling of RMsA basic multifaceted problem is the understand-

ing of how a RM will be located in the propermembrane microdomain and the mechanisms thatallow a RM to be organized as far as composition(i.e., types of receptors involved) and topography ina suitable way. According to our analogy of themosaic, the receptor assembly must be made withthe proper tesserae, which have to be properlylocated, only so the assembly has a meaning, that is,it forms a mosaic.

A general principle in logic is the so-calledOccam’s razor (frustra fit per plura quod potest fieri perpauciora [William of Occam, 1290–1350]), which statesthat it is vain to do with more what can be done withfewer elements. In this frame it can be observed thatGPCRs are rather large molecules (heptaspanningmolecules of about 50 kDa). In some cases, only fivetransmembrane domains appear sufficient to havea functional GPCR (Ling et al., 1999); furthermore,GCPRs form multimers. These characteristics wouldbe without a clear teleological meaning if the mul-tifunctional role of GPCRs that is attributable to theircapability of participating in several interactions (seethe data presented by Milligan at this congress) evensimultaneously with other proteins (and possiblylipids and carbohydrates) would not be considered.This means that it is not vain to have rather largeGPCRs that form multimers, as these multimers areactually microcircuits (RMs) and the RMs have a greatversatility to take part in molecular networks. It istherefore possible to discuss the assembly of recep-tors in RMs at least under the following aspects:

1. The assembly of the tesserae and quality control ofthese building blocks at the level of ER and Golgiapparatus (see Bouvier, this issue).

2. The possible function of chaperones for GCPRs. Thismight be the possible role of some orphan receptors(see Agnati et al. 2004b).

3. The permissive and/or attractive function of somelipids in the lipid raft. Aspecial case might be the lipidannulus (see above) that represents a wetting layerthat might be shared by two or more proteins givingrise to protein clusters and, hence, contributing to theformation of RMs.

4. The fishing function of the GCPRs within a RM.Actually this might be a further role of some orphanreceptors.

All of these aspects are of paramount importancenot only for the understanding of structural andfunctional organization of molecular networks, butalso for a new strategy in drug development (seebelow and Agnati et al., 2003a).

Classification of RMs and FunctionalOrganization Principles of MolecularNetworks

As discussed in several reviews (George et al., 2002;Agnati et al., 2003a; Terrillon and Bouvier, 2004), amajor question remains unanswered: What are the dif-ferent effects of receptor ligands on oligomerizationand trafficking? It might be speculated that to face this

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 201

202 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

problem it is important to consider (1) the membranemicrodomain and the HMN where the receptors arelocated; and (2) the composition of the RM and thetopography of receptors within the RM. Hence, a basicaspect to be clarified is the understanding of the struc-tural and functional principles according to whichmolecular networks are organized also at membranelevel. A possible clarification might be found in a ten-tative classification of RMs and in a mathematicalmodel to evaluate the relevance of composition, topog-raphy, and order of activation of single receptors inthe RM. A possible biochemical criterion might becooperativity, which is related to the binding of oneand the same ligand to a multimeric protein.

Three conditions should be met for cooperativity(Koshland and Hamadani, 2002): (1) The binding ofthe ligand induces a conformational change in the pro-tein subunit; (2) the conformational changes areintramolecularly transferred to other subunits of themultimeric protein via their interfaces; and (3) the sitesare initially essentially identical to each other. We willbriefly present a possible classification as well as amathematical model for RM characterization. Both arebased on cooperativity. A more detailed analysis ofthese aspects can be found in our submitted papers(see, e.g., Agnati et al., 2005a).

RM Classification and Functional Organizationof Molecular Networks Based on theCooperativity CriterionIn principle, three types of RMs can be distin-

guished:

1. RM1s (RM type 1) are formed by one type of recep-tor (homo-oligomers) or by isoreceptors (i.e., hetero-oligomers formed by different subtypes of the samereceptor, e.g., subtypes of GABAB receptors [Parmentier et al., 2002]).

2. RM2s (RM type 2) are hetero-oligomers formed bydifferent types of receptors. The same type or sub-types of receptors can be present, but not in contact.

3. RMms (RMs of a mixed type) are hetero-oligomersformed by different types of receptors. However, thesame type or subtypes of receptors can be present andin contact forming an RM1-island within the recep-tor assembly. The suggestion of the possible existenceof this type of RM arises from the assumption thathigh-order oligomers might exist.

Receptor mosaic 1 (RM1) can be surmised to show cooperativity, as we are dealing with a receptor complex that binds the same ligand (e.g., GABA). Receptor mosaic 2 (RM2) does not show

cooperativity (the ligand is different), but to describethe RRI (also in this case), we can employ the broaderconcept of allosteric interaction. Mixed receptormosaic (RMm) can show cooperativity within the RM1-islands. This proposal is based on what hasbeen discovered in, for example, hemocyanin, inwhich a small set of subunits are present and showcooperativity. Thus, allosteric cooperative unitsmight be present in large molecular assemblies (VanHolde et al., 2000). In analogy with this datum it issurmised that in a high-order RM (basically of theRM2 type), small RM1-islands might be present andwork as cooperative units greatly affecting thefunction of the entire RM.

Hence, the following can be deduced from GPCRoligomerization: Not only are RMs working asintegrated input units but also allosteric interactionsare in operation in these macromolecular assemblies(Agnati et al., 1982, 2002), and in some instances(RM1 and RMm) a cooperative behavior might bepresent. According to Koshland and Hamadani(2002), cooperativity in terms of enzymes is corre-lated with being a branch point enzyme. We suggestthat RM1 (and RM1-islands) can represent a branchpoint controlling the activation pattern within theHMN and VMN (see Fig. 5), and such branch pointsmight show regulated cooperativity, that is, thecooperativity is not operating according to a fixedpattern but might be modulated by allostericinteractions with other membrane-associatedproteins or even by ions (see also Armstrong andStrange, 2001). In the characterization of molecularnetworks, the role of hubs has been discussed (Albertet al., 2000; Kitano, 2002a; Agnati et al., 2004b). Thesenodes of a network are characterized by being highlyinterconnected, that is, by having several inputs fromother nodes and several outputs toward other nodes(Jeong et al., 2001). It can be speculated that in someRMs there are hub receptors, and we propose thatdopamine D2 receptors can have such a role in viewof the neuropsychopharmacological and molecu-lar data of an involvement of these receptors inschizophrenia and of possible alterations in themolecular structure of the D2-type of receptors inthis and other neuropsychological disorders as wellas in drug addiction (Agnati et al., 1986; Ferré etal., 2003). From the concepts presented, it can bededuced that hubs and branching points in mole-cular networks can be located also at plasma mem-brane level, which is also at the level of HMNs oreven of RMs.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 202

Receptor Mosaics 203

Journal of Molecular Neuroscience Volume 26, 2005

Mathematical Characterization of RMs Based on a Simple Energy Function and on the Symmetry Rule for CooperativityThe Koshland-Nemethy-Filmer (KNF) model of

cooperativity discusses how individual subunits ofmultimeric proteins will switch states in response toligand binding. Consequently, some subunits ina multimeric protein can exist in the weak bindingstate and others in the strong binding state. The KNFmodel can adequately describe both positive and neg-ative cooperativity (Koshland et al., 1966). A morerecent model has been proposed (Goh et al., 2004),but for the present mathematical approach the KNFmodel will be considered together with the symmetryrule proposed by Ackers and collaborators (1992).The symmetry rule (KNF model) has been provenfor hemoglobin (Ackers et al., 1992), and it maintains

that a quaternary switching from tense form (thedeoxy, low-affinity state) to relaxed form (the oxy,high-affinity state) occurs whenever heme-site bind-ing creates a tetramer with at least one ligated sub-unit on each dimeric half-molecule. Thus, in the caseof hemoglobin, the symmetry rule maintains that thequaternary switching between low- and high-affin-ity state occurs whenever there is at least one ligatedsubunit on each side of the interface that separatesthe two dimeric half-molecules.

The mathematical approach is summarized inFigs. 6 and 7, where the main steps of the formu-lation of the approach are indicated. In Fig. 6 theformal space of a two-dimensional RM is given,which can be described by (n • m) matrix. Each ele-ment of the matrix is a monomer or it is an emptyspace; the monomer can be linked to another

Fig. 5. Schematic representation of a possible deduction that can be drawn by the proposed classification of RMsand by the concept of branching point, suggested by Koshland and Hamadani (2002), for organization of metabolicpathways. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 203

204 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

monomer to form a dimer. Each monomer has avalue equal to 1 if it is in the bound state; other-wise, it has a value equal to 0. The interactionsamong monomers of different dimers (bonds) areindicated, and a simple energy function is appliedtogether with the symmetry rule to calculatethe value with which each monomer contributes tothe summation that expresses the cooperativityfunction.

A very simple case is illustrated in Fig. 7. In a sep-arate paper (Agnati et al., 2005a), it has been shownthat the present mathematical approach allows giv-ing a numerical index to characterize effects of recep-tor composition, receptor topography, and order ofreceptor activation on the cooperativity shown by theentire RM. This mathematical model can be furtherdeveloped to consider continuous states of monomers(instead of the binary 0 and 1) and thus can provide

a more realistic and experimentally testable descrip-tion of RRI within a RM in terms of free energy.

Possible New Approaches for DrugDevelopment

It has been reported that about 50% of all mod-ern drugs modulate GPCR function (George et al.,2002). Until recently, the models of signal transduc-tion maintained that the successive steps of thedecoding process were organized substantially in alinear fashion. By this, we mean that the ligand inter-acts with the receptor and the ligand receptor com-plex triggers specific signals in the cell (e.g., changesin the second messenger levels) that cause multiplecellular chemical-physical changes representing thecell response. It should be noted that according tothis view, the process becomes divergent only from

Fig. 6. Schematic representation of the two-dimensional space to describe a RM and definitions of the parametersused for its characterization. Abbreviations: x, y correspond to different monomers. For further details, see text.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 204

Receptor Mosaics 205

Journal of Molecular Neuroscience Volume 26, 2005

changes of second messenger levels on, that is, onlyafter the second messenger signaling a branchedorganized process is in operation. Later on it wasrealized that the complex ligand-GPCR could leadto the subsequent release of the components ofthe trimeric subunits assembled into the G proteins,and not only the α-subunit, but also the β- and γ- subunits, could trigger cell responses. All of thesesubunits could change the activity of effectors suchas protein kinases and ion channels.

Hence, at the G protein level a branched organizedprocess was considered. Furthermore, GPCRs can alsotransmit signals to cell biochemical machinerythrough mechanisms that function independentlyof G protein coupling such as Ca2+ influx and ERK2activation (Brzostowski and Kimmel, 2001). Thus, atan earlier stage of the G protein transduction,

abranched organized process can occur. As discussedin this paper, a branched organized process occurseven at an earlier phase, at the ligand-receptorrecognition level as multiple receptors could interactin the plasma membrane leading to a concertedactivation of multiple transduction systems. Thus, thedecoding process in its entirety is a highly branchedorganized process (see Agnati et al., 2003a, 2004a).

It is therefore important to change paradigm fordrug development by taking the following into account(see also Agnati and Fuxe, Concluding Remarks, thisissue):(1) the processes involved in the receptorsynthesis and trafficking, particularly the building ofthe RM (composition, topography, membranemicrodomain where it is located); (2) the peculiar bio-chemical characteristics of the GPCR that depend interalia on the RM to which it belongs and the proteins

Fig. 7. Schematic representation of the possible application of the symmetry rule to obtain an objective, even if arbi-trary, criterion to describe a tetramer. It should be noted that this mathematical approach can be used for any RM, andit allows characterization, as to composition, topography, and order of activation of receptors, in the RM. For furtherdetails, see text and Agnati et al., 2005a.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 205

206 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

and lipids with which it interacts; and (3) the compo-sitions of the lipid rafts. Hence, there is also the pos-sibility of acting on the GPCR indirectly via an actionon the platform where the respective RM is located.

These new pathways for drug development shouldallow the discovery of more selective and highly effec-tive drugs. In turn, these new drugs will be a verypowerful tool to better understand the extraordinarycomplexity of cell biology, which represents the basicstep in shedding some light on the computation capa-bility of the central nervous system.

AcknowledgmentsThe work reported in this paper has been

supported by grant no. QLG3-CT-2001-01056 fromthe European Community.

ReferencesAckers G. K., Doyle M. L., Myers D., and Daugherty

M. A. (1992) Molecular code of cooperativity inhaemoglobin. Science 255, 54–63.

Agnati L. F. and Fuxe K. (1984) New concepts on thestructure of the neuronal networks: the miniaturiza-tion and hierarchical organization of the central nervoussystem. Biosci. Rep. 4, 93–98.

Agnati L. F., Ferré S., Cortelli P., and Fuxe K. (1995) A briefappraisal on some aspects of the receptor-receptorinteraction. Neurochem. Int. 27, 139–146.

Agnati L. F., Ferré S., Leo G., Lluis C., Canela E. I., FrancoR., and Fuxe K. (2004a) On the molecular basis of thereceptor mosaic hypothesis of the engram. Cell. Mol.Neurobiol. 24, 501–517.

Agnati L. F., Ferré S., Lluis C., Franco R., and Fuxe K.(2003a) Molecular mechanisms and therapeuticalimplications of intramembrane receptor/receptorinteractions among heptahelical receptors withexamples from the striatopallidal GABA neurons.Pharmacol. Rev. 55, 509–550.

Agnati L. F., Franzen O., Ferré S., Leo G., Franco R., andFuxe K. (2003b) Possibile role of intramembranereceptor-receptor interactions in memory and learningvia formation of long-lived heteromeric complexes:focus on motor learning in the basal ganglia. J. NeuralTransm. 65, 195–222.

Agnati L. F., Fuxe K., Zini I., Lenzi P., and Hökfelt T. (1980)Aspects on receptor regulation and isoreceptoridentification. Med. Biol. 58, 182–187.

Agnati L. F., Fuxe K., Zoli M., Merlo Pich E., Benfenati F.,Zini I., and Goldstein M. (1986) Aspects on the information handling by the central nervous system:focus on cotransmission in the aged rat brain. Coexis-tence of neuronal messengers. Prog. Brain Res. 68,291–301.

Agnati L. F., Fuxe K., Zoli M., Rondanini C., and ÖgrenS. O. (1982) New vistas on synaptic pasticity: mosaichypothesis on the engram. Med. Biol. 60, 183–190.

Agnati L. F., Guidolin D., Genedani S., Ferré S., BigianiA., Woods A., and Fuxe K. (2005b) How proteins cometogether in the plasma membrane and function inmacromolecular assemblies: focus on receptor mosaics.J. Mol. Neurosci., in press.

Agnati L. F., Santarossa L., Benfenati F., Ferri M., MorpurgoA., Apolloni B., and Fuxe K. (2002) Molecular basis oflearning and memory: modeling based on receptormosaics, in From Synapses to Rules: Discovering SymbolicRules from Neural Processed Data, Apolloni, B., and Kurfess, F., eds., Kluwer Academic/Plenum Press,New York, pp. 165–196.

Agnati L. F., Santarossa L., Genedani S., Canela E. I., LeoG., Franco R., et al. (2004b) On the nested hierarchicalorganization of CNS: basic characteristics of neuronalmolecular networks, in Lecture Notes in ComputerSciences, Springer, Berlin, pp. 24–54.

Agnati L. F., Tarakanov A. O., and Guidolin D. (2005a) Asimple mathematical model of cooperativity in recep-tor mosaics based on the “Symmetry rule.” Biosystems80(2), 165–173.

Agnati L. F., Zoli M., Merlo Pich E., Benfenati F., and Fuxe K. (1990) Aspects of neural plasticity in the cen-tral nervous system. VII. Theoretical aspects of brain communication and computation. Neurochem. Int.16, 479–500.

Albert R., Jeong H., and Barbasi A.,L. (2000) Error andattack tolerance of complex networks. Nature 406,378–382.

Armstrong D. and Strange P. G. (2001) Dopamine D2receptor dimer formation: evidence from ligandbinding. J. Biol. Chem. 276, 22621–22629.

Bockaert J., Marin P., Dumuis A., and Fagni L. (2003)The ‘magic tail’ of G protein-coupled receptors: ananchorage for functional protein networks. FEBS Lett.546, 65–72.

Borgan A. U. and Thorn K. S. (1998) Anatomy of hot spotsin protein interfaces. J. Mol. Biol. 3, 1–9.

Brzostowski J. A. and Kimmel A. (2001) Signaling at zero-G: G-protein independent functions for 7-TM recep-tors. Trends Biol. Sci. 26, 291–297.

Chothia C., Gough J., Vogel C., and Teichmann S. A. (2003)Evolution of protein repertoire. Science 300, 1701–1703.

Doyle D. (2004) Structural changes during ion channelgating. Trends Neurosci. 27, 298–302.

Ferré S., Ciruela F., Woods A. S., Canals M., Burgueno J.,Marcellino D., et al. (2003) Glutamate mGlu5-adenosine A2A-dopamine D2 receptor interactions inthe striatum. Implications for drug therapy in neu-ropsychiatric disorders and drug abuse. Curr. Med.Chem. Central Nervous Systems Agents 33, 1–26.

Franco R., Canals M., Marcellino D., Ferré S., Agnati L.,Mallol J., et al. (2003) Regulation of heptaspanning-membrane receptor function by dimerization andclustering. Trends Biochem. Sci. 28, 238–243.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 206

Receptor Mosaics 207

Journal of Molecular Neuroscience Volume 26, 2005

Fuxe K. and Agnati L. F. (1985) Receptor-receptorinteractions in the central nervous system. A newintegrative mechanism in synapses. Med. Res. Rev. 5,441–482.

Fuxe K. and Agnati L. F. (1987) Receptor-Receptor Interac-tions. A New Intramembrane Integrative Mechanism,Macmillan Press, London, UK.

Fuxe K., Agnati L. F., Benfenati F., Cimino M., Algeri S.,Hökfelt T., and Mutt V. (1981) Modulation by chole-cystokinins of 3H-spiroperidol binding in rat striatum:evidence for increased affinity and reduction in thenumber of binding sites. Acta. Physiol. Scand. 113,567–569.

Gainetdinov R. R., Premont R. T., Bohn L. M., LefkowitzR. J., and Caron M. G. (2004) Desensitization of G pro-tein-coupled receptors and neuronal functions. Annu.Rev. Neurosci. 27, 107–144.

George S. R., O’Dowd B. F., and Lee S. P. (2002) G-protein-coupled receptor oligomerization and itspotential for drug discovery. Nat. Rev. Drug Discov. 1,808–820.

Gil T., Ipsen J. H., Mouritsen O. G., Sabra M. C., SperottoM. M., and Zuckermann M. J. (1998) Theoretical analysisof protein organization in lipid membranes. Biochim.Biophys. Acta 1376, 245–266.

Goh C. S., Milburn D., and Gerstein M. (2004) Confor-mational changes associated with protein-proteininteractions. Curr. Opin. Struct. Biol. 14, 1–6.

Greengard P. (1976) Possible role for cyclic nucleotidesand phosphorylated membrane proteins in postsy-naptic actions of neurotransmitters. Nature 260,101–108.

Hodgkin A. L. and Huxley A. F. (1952) A quantitativedescription of membrane current and its application toconduction and excitation in nerve. J. Physiol. 117,500–544.

Ishii M. and Kurachi Y. (2003) Physiological actions ofregulators of G-protein signaling (RGS) proteins. Life Sci. 74, 163–171.

James J. C. and Tawfik D. S. (2003) Conformational diver-sity and protein evolution: a 60-year old hypothesisrevisited. Trends Biochem. 28, 361–368.

Jeong H., Mason S. P., Barabasi Z. N., and Oltvai Z .N.(2001) Lethality and centrality in protein networks.Nature 411, 41,42.

Kitano H. (2002a) Computational systems in biology.Nature 420, 206–210.

Kitano H. (2002b) Looking beyond the details: a rise insystem-oriented approaches in genetics and molecu-lar biology. Curr. Genet. 41, 1–10.

Kortemme T. and Baker D. (2002) Asimple physical modelfor binding energy hot spots in protein-protein com-plexes. Proc. Natl. Acad. Sci. U. S. A. 99, 14116–14121.

Koshland D. E., Jr., and Hamadani K. (2002) Proteomicsand models for enzyme cooperativity. J. Biol. Chem.277, 46841–46844.

Koshland D. E., Nemethy G., and Filmer D. (1966) Com-parison of experimental binding data and theoretical

models in proteins containing subunits. Biochemistry 5,365–385.

Lauffenburger D. A. (2000) Cell signalling pathways ascontrol modules: complexity for simplicity? Proc. Natl.Acad. Sci. U. S. A. 97, 5031–5033.

Lee S. P., O’Dowd B. F., and George S. R. (2003) Homo-and hetero-oligomerization of G protein-coupled recep-tors. Life Sci. 74, 173–180.

Lewis R. N., Zhang Y. P., Hodges R. S., Subczynski W. K.,Kusumi A., Flach C. R., et al. (2001) Apolyalanine-basedpeptide cannot form a stable transmembrane alpha-helix in fully hydrated phospholipid bilayers. Bio-chemistry 40, 12103–12111.

Limbird L. E. and Lefkowitz R. J. (1976) Negative coop-erativity among β-adrenergic receptors in frog ery-throcyte membranes. J. Biol. Chem. 251, 5007–5014.

Limbird L. E., Mayts P. D., and Lefkowitz R. J. (1975) Beta-adrenergic receptors: evidence for negative coopera-tivity. Biochem. Biophys. Res. Commun. 64, 1160–1168.

McCulloch W. S. and Pitts W. H. (1943) A logical calculusof the ideas imminent in nervous activity. Bull. Math.Biophys. 5, 111–133.

Miller W. E. and Lefkowitz R. J. (2001) Expanding rolesfor -arrestins as scaffolds and adapters in GPCR sig-naling and trafficking. Curr. Opin. Cell. Biol. 13,139–145.

Okamoto T., Schlegel A., Scherer P. E., and Lisanti M. P.(1998) Caveolins, a family of scaffolding proteins fororganizing “preassembled signaling complexes” at theplasma membrane. J. Biol. Chem. 273, 5419–5422.

Paratcha G. and Ibanez C. F. (2002) Lipid rafts and thecontrol of neurotrophic factor signaling in the nervoussystem: variations on a theme. Curr. Opin. Neurobiol.12, 542–549.

Park J., Lappe M., and Teichmann S. A. (2001) mappingprotein family interactions: intramolecular and inter-molecular protein family interaction repertoires in PDBand yeast. J. Mol. Biol. 307, 929–938.

Parmentier M. L., Prezeau L., Bockaert J., and Pin J. (2002)A model for the functioning of family 3 GPCRs. TrendsPharmacol. Sci. 23, 268–274.

Pasko G. A. and Ringe D. (2004) Protein Structure and Func-tion, Sinauer Associates, Sunderland, MA.

Penela P., Ribas C., and Mayor F., Jr. (2003) Mechanismsof regulation of the expression and function of G pro-tein-coupled receptor kinases. Cell Signal. 15, 973–981.

Popper K. (1935) Logik der Forshung, Springer, Vienna. Ravasz E., Somera A. L., Mongru D. A., Oltvai Z. N., and

Barabasi A. L. (2002) Hierarchical organization of mod-ularity in metabolic networks. Science 297, 1551–1555.

Shepherd G. (1983) Neurobiology, Oxford University Press,New York.

Simons K. and Ikonen E. (1997) Functional rafts in cellmembranes. Nature 387, 569-–72.

Smith A. D. (1997) Oxford Dictionary of Biochemistry andMolecular Biology, Oxford University Press, Oxford, UK.

Terrillon S. and Bouvier M. (2004) Roles of G-protein-coupled receptor dimerization. EMBO Rep. 5, 30–34.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 207

208 Agnati et al.

Journal of Molecular Neuroscience Volume 26, 2005

van Holde K. E., Miller K. I., and van Olden E. (2000)Allostery in very large molecular assemblies. Biophys.Chem. 86, 165–172.

Virtanen J. A., Cheng K. H., and Somerharju P. (1998) Phos-pholipid composition of the mammalian red cell mem-brane can be rationalized by a superlattice model. Proc.Natl. Acad. Sci. U. S. A. 95, 4964–4969.

Watts D. J. and Strogatz S. H. (1998) Collective dynamicsof “small-world” networks. Nature 393, 440–442.

Weng G., Bhalla U. S., and Iyengar R. (1999) Complex-ity in biological signalling systems. Science 284, 92–96.

Wuchty S. (2001) Scale-free behavior in protein domainnetworks. Mol. Biol. Evol. 18, 1694–1702.

JMN_040048_FERRE.qxd 01/06/2005 05:37 pm Page 208