articlesCommunication between distant sites in proteins is fundamentalto their function and often defines the biological role of a proteinfamily. In signaling proteins, it represents information transfer —the transmission of signals initiated at one functional surface to adistinct surface mediating downstream signaling. For example,ligand binding at an externally accessible site in G protein–coupled receptors (GPCRs) reliably triggers structural changes atdistant cytoplasmic domains that mediate interaction with het-erotrimeric G proteins1,2. Studies in many other protein systemsindicate that long-range interactions of amino acids also areimportant in binding (and catalytic) specificity. Substrate recog-nition in the chymotrypsin family of serine proteases3,4, the tun-ing of antibody specificity through B-cell maturation5and thecooperativity of oxygen binding in hemoglobin6–9all depend notonly on residues directly contacting substrate, but also on distantresidues located in supporting loops and other secondary struc-tural elements. Crystallographic studies in all of these sys-tems5,9–11indicate that the distant residues participating insubstrate recognition do so by acting through intervening posi-tions to control the structure of the substrate-binding site. Theselong-range interactions are remarkable because many other sites,even if closer to active site residues, show little contribution tofunction. Taken together, these studies indicate that proteins arecomplex materials in which perturbations at sites — for example,substrate binding, covalent modification or mutation — maycause conformational change to happen in a fracture-like mannerthat is not obvious in atomic structures. From a biological pointof view, these fractures represent the energy transduction mecha-nisms that mediate signal flow, allosteric regulation and speci-ficity in molecular recognition.How can we globally map energetic interactions betweenamino acid residues in protein structures? Although methodssuch as the thermodynamic double mutant cycle12–14provideexcellent tools for estimating such interactions, practical limita-tions restrict these techniques to small studies in specific modelsystems. An alternative approach is suggested by a newsequence-based statistical method for estimating thermo-dynamic coupling between residues in proteins15. The basis ofthis method is that the coupling of two sites in a protein, whetherfor structural or functional reasons, should cause those twopositions to co-evolve16–18. In principle, this might be revealed inan analysis of a large and diverse multiple sequence alignment(MSA) of a protein family. Application of this method for oneactive site residue in a small protein interaction domain (thePDZ domain) family predicted energetic coupling to a small setof other residues that were organized into a chain-like networkthrough the protein core, linking the active site residue with dis-tant sites15. These predictions were verified through mutagene-sis, suggesting that the statistical measurement of couplingthrough sequence analysis is a good reporter of thermodynamiccoupling.These results suggest the possibility that we can visualize theglobal network of energetic interactions between pairs of aminoacids and explain long-range energetic interactions in proteins.Here, we describe this mapping for three protein families thatrepresent completely distinct folds and biological activities: (i) atransmembrane signaling receptor family (GPCRs), (ii) anenzyme family that has served as a model system for catalyticspecificity (the chymotrypsin class of serine proteases) and (iii) amulti-subunit protein family that is the classic model system forallosteric regulation (hemoglobin).A statistical mapping of interactions in proteinsTo illustrate the sequence analysis, we consider four positions ofa hypothetical protein (i, j, k and l) and a corresponding MSA ofthe protein family (Fig. 1a,b). If the MSA is sufficiently large anddiverse that it describes the evolutionary constraints on the fam-ily, we can make the following two postulates about the aminoacid frequencies observed at specific sites. First, if site l con-tributes nothing to either the folding or function of the protein,Evolutionarily conserved networks of residuesmediate allosteric communication in proteinsGürol M. Süel1,2, Steve W. Lockless1,2, Mark A. Wall2and Rama Ranganathan2Published online 16 December 2002; doi:10.1038/nsb881A fundamental goal in cellular signaling is to understand allosteric communication, the process by which signalsoriginating at one site in a protein propagate reliably to affect distant functional sites. The general principles ofprotein structure that underlie this process remain unknown. Here, we describe a sequence-based statisticalmethod for quantitatively mapping the global network of amino acid interactions in a protein. Application ofthis method for three structurally and functionally distinct protein families (G protein–coupled receptors, thechymotrypsin class of serine proteases and hemoglobins) reveals a surprisingly simple architecture for amino acidinteractions in each protein family: a small subset of residues forms physically connected networks that linkdistant functional sites in the tertiary structure. Although small in number, residues comprising the networkshow excellent correlation with the large body of mechanistic data available for each family. The data suggestthat evolutionarily conserved sparse networks of amino acid interactions represent structural motifs forallosteric communication in proteins.1These authors contributed equally to this work. 2Howard Hughes Medical Institute and Department of Pharmacology, The University of Texas Southwestern MedicalCenter, 5323 Harry Hines Boulevard, Dallas, Texas 75390-9050, USA.Correspondence should be addressed to R.R. e-mail: [email protected] structural biology • volume 10 number 1 • january 2003 59© 2003 Nature Publishing Group http://www.nature.com/naturestructuralbiologyarticlesthe corresponding amino acid frequencies in the MSA should beunconstrained and, therefore, should approach their mean val-ues in all proteins. However, if sites i, j and k make some contri-bution, the amino acid distributions at these sites should deviatefrom these mean values, and the extent of this deviation shouldprovide a quantitative measure of the underlying evolutionaryconstraint (conservation). Second, the functional