Search

Characterizing the effects of mutations on stability is critical for understanding the function and evolution of proteins and improving their biophysical properties. High throughput folding and abundance assays have been successfully used to characterize missense mutations associated with reduced stability. However, screening for increased thermodynamic stability is more challenging since such mutations are rarer and their impact on assay readout is more subtle. Here, a multiplex assay for high throughput screening of protein folding was developed by combining deep mutational scanning, fluorescence-activated cell sorting, and deep sequencing. By analyzing a library of 2000 variants of Adenylate kinase we demonstrate that the readout of the method correlates with stability and that mutants with up to 13 °C increase in thermal melting temperature could be identified with low false positive rate. The discovery of many stabilizing mutations also enabled the analysis of general substitution patterns associated with increased stability in Adenylate kinase. This high throughput method to identify stabilizing mutations can be combined with functional screens to identify mutations that improve both stability and activity., Competing Interests: Declaration of Competing Interest The authors declare that they have no conflict of interest., (Copyright © 2023 The Author(s). Published by Elsevier Ltd.. All rights reserved.)

This Feature Article presents a view of the protein folding transition based on the hypothesis that Nature has built features within the sequences that enable a Shortcut to efficient folding. Nature's Shortcut is proposed to be the early establishment of a set of nonlocal weak contacts, constituting protein loops that significantly constrain regions of the collapsed disordered protein into a native-like low-resolution fluctuating topology of major sections of the backbone. Nature's establishment of this scaffold of nonlocal contacts is claimed to bypass what would otherwise be a nearly hopeless unaided search for the final three-dimensional structure in proteins longer than ∼100 amino acids. To support this main contention of the Feature Article, the loop hypothesis (LH) description of early folding events is experimentally tested with time-resolved Förster resonance energy transfer techniques for adenylate kinase, and the data are shown to be consistent with theoretical predictions from the sequential collapse model (SCM). The experimentally based LH and the theoretically founded SCM are argued to provide a unified picture of the role of nonlocal contacts as constituting Nature's Shortcut to protein folding. Importantly, the SCM is shown to reliably predict key nonlocal contacts utilizing only primary sequence information. This view on Nature's Shortcut is open to the protein community for further detailed assessment, including its practical consequences, by suitable application of advanced experimental and computational techniques.

Folding of proteins to their functional conformation is paramount to life. Though 75% of the proteome consists of multidomain proteins, our knowledge of folding has been based primarily on studies conducted on single-domain and fast-folding proteins. Nonetheless, the complexity of folding landscapes exhibited by multidomain proteins has received increased scrutiny in recent years. We study the three-domain protein adenylate kinase from E. coli (AK), which has been shown to fold through a series of pathways involving several intermediate states. We use a protein design method to manipulate the folding landscape of AK, and single-molecule FRET spectroscopy to study the effects on the folding process. Mutations introduced in the NMP binding (NMPbind) domain of the protein are found to have unexpected effects on the folding landscape. Thus, while stabilizing mutations in the core of the NMPbind domain retain the main folding pathways of wild-type AK, a destabilizing mutation at the interface between the NMPbind and the CORE domains causes a significant repartition of the flux between the folding pathways. Our results demonstrate the outstanding plasticity of the folding landscape of AK and reveal how specific mutations in the primary structure are translated into changes in folding dynamics. The combination of methodologies introduced in this work should prove useful for deepening our understanding of the folding process of multidomain proteins.

By using in-cell NMR experiments, we have demonstrated that the protein folding state in cells is significantly influenced by the cellular health conditions. hAK1 was denatured in cells under stressful culture conditions, while it remained functional and properly folded in cells continuously supplied with a fresh medium.

The ensemble of conformers of globular protein molecules immediately following transfer from unfolding to folding conditions is assumed to be collapsed though still disordered, as the first steps of the folding pathway are initiated. In order to test the hypothesis that long loop closure transitions are part of the initiation of the folding pathway, our groups are studying the initiation of the folding transition of a model protein by time-resolved excitation energy transfer (trFRET) detected fast kinetics experiments. Site-specific double labeling is used to study the timing of conformational transitions of individual loop forming chain segments at the microsecond time regime. Previously, it was shown that at least three long loops in the Escherichia coli adenylate kinase (AK) molecule close within the first 5 ms of folding of AK, while the main global folding transition occurs in a time regime of seconds. In order to enhance the time resolution of the kinetics experiments to the microsecond time regime and determine the rate of closure of the two N terminal loops (loop I residues 1-26 and loop II residues 29-72), we applied a continuous flow based double kinetics experiment. These measurements enabled us to obtain a microsecond series of transient time dependent distributions of distances between the ends of the labeled loops. Analysis of the trFRET experiments show that the N terminal loop (loop I) is closed within less than 60 μs after the initiation of refolding. Loop II is also mostly closed within that time step but shows an additional small reduction of the mean end-to-end distance in a second phase at a rate of 0.005 μs(-1). This second phase can either reflect tightening of a loosely closed loop in the ensemble of conformers or may reflect two subpopulations in the ensemble, which differ in the rate of closure of loop II, but not in the rate of closure of loop I. This study shows the very fast closure of long loops in the otherwise disordered backbone and fine details of the very early hidden pretransition state steps that are essential for the fast and efficient folding of the protein molecule.

The nature of the earliest steps of the initiation of the folding pathway of globular proteins is still controversial. To elucidate the role of early closure of long loop structures in the folding transition, we studied the folding kinetics of subdomain structures in Escherichia coli adenylate kinase (AK) using Förster type resonance excitation energy transfer (FRET)-based methods. The overall folding rate of the AK molecule and of several segments that form native β strands is 0.5 ± 0.3 s(-1), in sharp contrast to the 1000-fold faster closure of three long loop structures in the CORE domain. A FRET-based "double kinetics" analysis revealed complex transient changes in the initially closed N-terminal loop structure that then opens and closes again at the end of the folding pathway. The study of subdomain folding in situ suggests a hierarchic ordered folding mechanism, in which early and rapid cross-linking by hydrophobic loop closure provides structural stabilization at the initiation of the folding pathway.

While fast folding of small proteins has been relatively well characterized by experiments and theories, much less is known for slow folding of larger proteins, for which recent experiments suggested quite complex and rich folding behaviors. Here, we address how the energy landscape theory can be applied to these slow folding reactions. Combining the perfect-funnel approximation with a multiscale method, we first extended our previous atomic-interaction based coarse grained (AICG) model to take into account local flexibility of protein molecules. Using this model, we then investigated the energy landscapes and folding routes of two proteins with complex topologies: a multidomain protein adenylate kinase (AKE) and a knotted protein 2ouf-knot. In the AKE folding, consistent with experimental results, the kinetic free energy surface showed several substates between the fully unfolded and native states. We characterized the structural features of these substates and transitions among them, finding temperature-dependent multiroute folding. For protein 2ouf-knot, we found that the improved atomic-interaction based coarse-grained model can spontaneously tie a knot and fold the protein with a probability up to 96%. The computed folding rate of the knotted protein was much slower than that of its unknotted counterpart, in agreement with experimental findings. Similar to the AKE case, the 2ouf-knot folding exhibited several substates and transitions among them. Interestingly, we found a dead-end substate that lacks the knot, thus suggesting backtracking mechanisms.

Systems biology can offer a great deal of insight into evolution by quantitatively linking complex properties such as protein structure, folding, and function to the fitness of an organism. Although the link between diseases such as Alzheimer's and misfolding is well appreciated, directly showing the importance of protein folding to success in evolution has been more difficult. We show here that predicting success during adaptation can depend critically on enzyme kinetic and folding models. We used a 'weak link' method to favor mutations to an essential, but maladapted, adenylate kinase gene within a microbial population that resulted in the identification of five mutants that arose nearly simultaneously and competed for success. Physicochemical characterization of these mutants showed that, although steady-state enzyme activity is important, success within the population is critically dependent on resistance to denaturation and aggregation. A fitness function based on in vitro measurements of enzyme activity, reversible and irreversible unfolding, and the physiological context reproduces in vivo evolutionary fates in the population linking organismal adaptation to its physical basis.

Conformational change is regulating the biological activity of a large number of proteins and enzymes. Efforts in structural biology have provided molecular descriptions of the interactions that stabilize the stable ground states on the reaction trajectories during conformational change. Less is known about equilibrium thermodynamic stabilities of the polypeptide segments that participate in structural changes and whether the stabilities are relevant for the reaction pathway. Adenylate kinase (Adk) is composed of three subdomains: CORE, ATPlid, and AMPbd. ATPlid and AMPbd are flexible nucleotide binding subdomains where large-scale conformational changes are directly coupled to catalytic activity. In this report, the equilibrium thermodynamic stabilities of Adk from both mesophilic and hyperthermophilic bacteria were investigated using solution state NMR spectroscopy together with protein engineering experiments. Equilibrium hydrogen to deuterium exchange experiments indicate that the flexible subdomains are of significantly lower thermodynamic stability compared to the CORE subdomain. Using site-directed mutagenesis, parts of ATPlid and AMPbd could be selectively unfolded as a result of perturbation of hydrophobic clusters located in these respective subdomains. Analysis of the perturbed Adk variants using NMR spin relaxation and C(alpha) chemical shifts shows that the CORE subdomain can fold independently of ATPlid and AMPbd; consequently, folding of the two flexible subdomains occurs independently of each other. Based on the experimental results it is apparent that the flexible subdomains fold into their native structure in a noncooperative manner with respect to the CORE subdomain. These results are discussed in light of the catalytically relevant conformational change of ATPlid and AMPbd.

Most globular protein chains, when transferred from high to low denaturant concentrations, collapse instantly before they refold to their native state. The initial compaction of the protein molecule is assumed to have a key effect on the folding pathway, but it is not known whether the earliest structures formed during or instantly after collapse are defined by local or by non-local interactions--that is, by secondary structural elements or by loop closure of long segments of the protein chain. Stable closure of one or several long loops can reduce the chain entropy at a very early stage and can prevent the protein from following non-productive pathways whose number grows exponentially with the length of the protein chain. In Escherichia coli adenylate kinase (AK), about seven long loops define the topology of the native structure. We selected four loop-forming sections of the chain and probed the time course of loop formation during refolding of AK. We labeled the termini of the loop segments with tryptophan and cysteine-5-amidosalicylic acid. This donor-acceptor pair of probes used with fluorescence resonance excitation energy transfer spectroscopy (FRET) is suitable for detecting very short distances and thus is able to distinguish between random and specific compactions. Refolding of AK was initiated by stopped-flow mixing, followed simultaneously by donor and acceptor fluorescence, and analyzed in terms of energy transfer efficiency and distance. In the collapsed state of AK, observed after the 5-ms dead time of the instrument, one of the selected segments shows a native-like separation of its termini; it forms a loop already in the collapsed state. A second segment that includes the first but is longer by 15 residues shows an almost native-like separation of its termini. In contrast, a segment that is shorter but part of the second segment shows a distance separation of its termini as high as a segment that spans almost the whole protein chain. We conclude that a specific network of non-local interactions, the closure of one or several loops, can play an important role in determining the protein folding pathway at its early phases.

Large conformational changes in the LID and NMP domains of adenylate kinase (AKE) are known to be key to ligand binding and catalysis, yet the order of binding events and domain motion is not well understood. Combining the multiple available structures for AKE with the energy landscape theory for protein folding, a theoretical model was developed for allostery, order of binding events, and efficient catalysis. Coarse-grained models and nonlinear normal mode analysis were used to infer that intrinsic structural fluctuations dominate LID motion, whereas ligand-protein interactions and cracking (local unfolding) are more important during NMP motion. In addition, LID-NMP domain interactions are indispensable for efficient catalysis. LID domain motion precedes NMP domain motion, during both opening and closing. These findings provide a mechanistic explanation for the observed 1:1:1 correspondence between LID domain closure, NMP domain closure, and substrate turnover. This catalytic cycle has likely evolved to reduce misligation, and thus inhibition, of AKE. The separation of allosteric motion into intrinsic structural fluctuations and ligand-induced contributions can be generalized to further our understanding of allosteric transitions in other proteins.

Under the assumption of equivalent heat capacity values, the differential free energy of stability for a pair of proteins midway between their thermal unfolding transition temperatures is shown to be independent of DeltaC(p) up to its cubic term in DeltaT(m). For model calculations reflecting the nearly 30 degrees C difference in T(m) for the adenylate kinases from the arctic bacterium Bacillus globisporus and the thermophilic bacterium Geobacillus stearothermophilus, the resultant error in estimating DeltaDeltaG by the formula 0.5 [DeltaS(T(m1))(1)+DeltaS(T(m2)) (2)] DeltaT(m) is less than 1%. Combined with the analogous thermal unfolding data for the adenylate kinase from Escherichia coli, these three homologous proteins exhibit T(m) and DeltaS(T(m)) values consistent with differential entropy and enthalpy contributions of equal magnitude. When entropy-enthalpy compensation holds for the differential free energy of stability, the incremental changes in T(m) values are shown to be proportionate to the changes in free energy.

Recent theoretical work suggests that protein folding involves an ensemble of pathways on a rugged energy landscape. We provide direct evidence for heterogeneous folding pathways from single-molecule studies, facilitated by a recently developed immobilization technique. Individual fluorophore-labeled molecules of the protein adenylate kinase were trapped within surface-tethered lipid vesicles, thereby allowing spatial restriction without inducing any spurious interactions with the environment, which often occur when using direct surface-linking techniques. The conformational fluctuations of these protein molecules, prepared at the thermodynamic midtransition point, were studied by using fluorescence resonance energy transfer between two specifically attached labels. Folding and unfolding transitions appeared in experimental time traces as correlated steps in donor and acceptor fluorescence intensity. The size of the steps, in fluorescence resonance energy transfer efficiency units, shows a very broad distribution. This distribution peaks at a relatively low value, indicating a preference for small-step motion on the energy landscape. The time scale of the transitions is also distributed, and although many transitions are too fast to be time-resolved here, the slowest ones may take >1 sec to complete. These extremely slow changes during the folding of single molecules highlight the possible importance of correlated, non-Markovian conformational dynamics.

The refolding transition of Escherichia coli adenylate kinase (AK) was investigated by monitoring the refolding kinetics of a selected 20 residue helical segment in the CORE domain of the protein. Residues 169 and 188 were labeled by 1-acetamido-methyl-pyrene, and by bimane, respectively. The experiment combines double-jump stopped-flow fast mixing initiation of refolding and time-resolved Förster energy transfer spectroscopy for monitoring the conformational transitions (double-kinetics experiment). Two kinetic phases were found in the denaturant-induced unfolding of AK. In the first phase, the fluorescence quantum yields of both probes decreased. The distribution of the distances between them transformed from the native state's narrow distribution with the mean distance corresponding to the distance in the crystal structure, to a distribution compatible with an unordered structure. In the second, slow step of denaturation, neither the fluorescence parameters of the probes nor the distance distribution between them changed. This step appeared to be a transformation of the fast-folding species formed in the first phase, to the slow-folding species. Refolding of the fast-folding species of the denatured state of AK was also a two-phase process. During the first fast phase, within less than 5ms, the fluorescence emission of both probes increased, but the distance distribution between the labeled sites was unchanged. Only during the second slow refolding step did the intramolecular distance distribution change from the characteristic of the denatured state to the narrow distribution of the native state. This experiment shows that for the case of the CORE domain of AK, the large helical segment of residues 169-188 was not formed in the first compaction step of refolding. The helical conformation of this segment is established only in the second, much slower, refolding phase, simultaneously with the completion of the native structure.

Adenylate kinase (AKe) from E. coli is a small, single-chain, monomeric enzyme with no tryptophan and a single cysteine residue. We have constructed six single-Trp mutants of AKe to facilitate optical studies of these proteins and to specifically examine the interrelationship between their structure, function, dynamics, and folding reactions. In this study, the effects of hydrostatic pressure on the folding reactions of AKe were studied. The native structure of AKe was transformed to a non-native, yet pressure stable, conformation by hydrostatic pressure of about 300 MPa. This pressure lability of AKe is rather low for a monomeric protein and presumably may be attributed to substantial conformational flexibility and a correspondingly large volume change. The refolding of AKe after pressure-induced denaturation was reversible under ambient conditions. At low temperature (near 0 degrees C), the refolding process of pressure-exposed AKe mutants displayed a significant hysteresis. The observation of a slow refolding rate in the 193 region and a faster folding rate around the active site (86, 41, 73 regions) leads us to suggest that in the folding process, priority is afforded to functional regions. The slow structural return of the 193 region apparently does not hinder the more rapid return of enzymatic activity of AKe. Circular dichroism studies on the pressure-denatured Y193W mutant show that the secondary structure (calculated from far-UV spectra) returned at a rapid rate, but the tertiary structure alignment (calculated from near-UV spectra) around the 193 region occurred more slowly at rates comparable to those detected by fluorescence intensity. Denaturation of AKe mutants by guanidine hydrochloride and subsequent refolding experiments were also consistent with a much slower refolding process around the 193 region than near the active site. Fast refolding kinetic traces were observed in F86W, S41W, and A73W mutants using a fluorescence detection stopped-flow rapid mixing device, while only a slow kinetic trace was observed for Y193W. The results suggest that the differences in regional folding rates of AKe are not derived from the specific denaturation methods, but rather are inherent in the structural organization of the protein.

The long isoform of eukaryotic adenylate kinase has a dual subcellular location in the cytoplasm and in the mitochondrial intermembrane space. Protein sequences and modifications are identical in both locations. In yeast, the bulk of the major form of adenylate kinase (Aky2p) is in the cytoplasm and, in the steady state, only 5-8% is sorted to the mitochondrial intermembrane space. Since the reasons for exclusion from mitochondrial import are unclear, we have constructed aky2 mutants with elevated mitochondrial uptake efficiency of Aky2p in vivo and in vitro. We have analyzed the effect of the mutations on secondary structure prediction in silico and have tested folding velocity and folding stability. One type of mutants displayed decreased proteolytic stability and retarded renaturation kinetics after chaotropic denaturation implying that deterioration of folding leads to prolonged presentation of target information to mitochondrial import receptors, thereby effecting improved uptake. In a second type of mutants, increased import efficiency was correlated with an increased probability of formation of an alpha-helix with increased amphipathic moment at the N-terminus suggesting that targeting interactions with mitochondrial import receptors had been improved at the level of binding affinity.

Techniques from graph theory are applied to analyze the bond networks in proteins and identify the flexible and rigid regions. The bond network consists of distance constraints defined by the covalent and hydrogen bonds and salt bridges in the protein, identified by geometric and energetic criteria. We use an algorithm that counts the degrees of freedom within this constraint network and that identifies all the rigid and flexible substructures in the protein, including overconstrained regions (with more crosslinking bonds than are needed to rigidify the region) and underconstrained or flexible regions, in which dihedral bond rotations can occur. The number of extra constraints or remaining degrees of bond-rotational freedom within a substructure quantifies its relative rigidity/flexibility and provides a flexibility index for each bond in the structure. This novel computational procedure, first used in the analysis of glassy materials, is approximately a million times faster than molecular dynamics simulations and captures the essential conformational flexibility of the protein main and side-chains from analysis of a single, static three-dimensional structure. This approach is demonstrated by comparison with experimental measures of flexibility for three proteins in which hinge and loop motion are essential for biological function: HIV protease, adenylate kinase, and dihydrofolate reductase., (Copyright 2001 Wiley-Liss, Inc.)

A method for determination of transient (on the millisecond timescale) intramolecular distance distributions (IDDs) by time-resolved dynamic non-radiative excitation energy transfer measurements was developed. The time-course of the development of the IDD between residues 73 and 203 in the CORE domain of Escherichia coli adenylate kinase throughout refolding from the GuHCl-induced denatured state was determined. The mean of the apparent IDD reduced to a value close to its magnitude in the native protein, within 2 ms (the dead-time of the instrument). At that time the width of that distribution was rather large (16+/-2 A). The large width implies that the intramolecular diffusion coefficient of the labeled segment does not exceed 10(-7) cm(2)/second. In a second slower phase of the refolding transition, the width was reduced to its native value (6+/-4 A)., (Copyright 2000 Academic Press.)

In this article we focus on presenting a broad range of examples illustrating low-energy transitions via hinge-bending motions. The examples are divided according to the type of hinge-bending involved; namely, motions involving fragments of the protein chains, hinge-bending motions involving protein domains, and hinge-bending motions between the covalently unconnected subunits. We further make a distinction between allosterically and nonallosterically regulated proteins. These transitions are discussed within the general framework of folding and binding funnels. We propose that the conformers manifesting such swiveling motions are not the outcome of "induced fit" binding mechanism; instead, molecules exist in an ensemble of conformations that are in equilibrium in solution. These ensembles, which populate the bottoms of the funnels, a priori contain both the "open" and the "closed" conformational isomers. Furthermore, we argue that there are no fundamental differences among the physical principles behind the folding and binding funnels. Hence, there is no basic difference between funnels depicting ensembles of conformers of single molecules with fragment, or domain motions, as compared to subunits in multimeric quaternary structures, also showing such conformational transitions. The difference relates only to the size and complexity of the system. The larger the system, the more complex its corresponding fused funnel(s). In particular, funnels associated with allosterically regulated proteins are expected to be more complicated, because allostery is frequently involved with movements between subunits, and consequently is often observed in multichain and multimolecular complexes. This review centers on the critical role played by flexibility and conformational fluctuations in enzyme activity. Internal motions that extend over different time scales and with different amplitudes are known to be essential for the catalytic cycle. The conformational change observed in enzyme-substrate complexes as compared to the unbound enzyme state, and in particular the hinge-bending motions observed in enzymes with two domains, have a substantial effect on the enzymatic catalytic activity. The examples we review span the lipolytic enzymes that are particularly interesting, owing to their activation at the water-oil interface; an allosterically controlled dehydrogenase (lactate dehydrogenase); a DNA methyltransferase, with a covalently-bound intermediate; large-scale flexible loop motions in a glycolytic enzyme (TIM); domain motion in PGK, an enzyme which is essential in most cells, both for ATP generation in aerobes and for fermentation in anaerobes; adenylate kinase, showing large conformational changes, owing to their need to shield their catalytic centers from water; a calcium-binding protein (calmodulin), involved in a wide range of cellular calcium-dependent signaling; diphtheria toxin, whose large domain motion has been shown to yield "domain swapping;" the hexameric glutamate dehydrogenase, which has been studied both in a thermophile and in a mesophile; an allosteric enzyme, showing subunit motion between the R and the T states (aspartate transcarbamoylase), and the historically well-studied lac repressor. Nonallosteric subunit transitions are also addressed, with some examples (aspartate receptor and BamHI endonuclease). Hence, using this enzyme-catalysis-centered discussion, we address energy funnel landscapes of large-scale conformational transitions, rather than the faster, quasi-harmonic, thermal fluctuations.

A new approach for evaluating the secondary structure of proteins by CD spectroscopy of overlapping peptide segments is applied to porcine adenylate kinase (AK1) and yeast guanylate kinase (GK3). One hundred seventy-six peptide segments of a length of 15 residues, overlapping by 13 residues and covering the complete sequences of AK1 and GK3, were synthesized in order to evaluate their secondary structure composition by CD spectroscopy. The peptides were prepared by solid phase multiple peptide synthesis method using the 9-fluorenylmethoxycarbonyl/tert-butyl strategy. The individual peptide secondary structures were studied with CD spectroscopy in a mixture of 30% trifluoroethanol in phosphate buffer (pH 7) and subsequently compared with x-ray data of AK1 and GK3. Peptide segments that cover alpha-helical regions of the AK1 or GK3 sequence mainly showed CD spectra with increasing and decreasing Cotton effects that were typical for appearing and disappearing alpha-helical structures. For segments with dominating beta-sheet conformation, however, the application of this method is limited due to the stability and clustering of beta-sheet segments in solution and due to the difficult interpretation of random-coiled superimposed beta-sheet CD signals. Nevertheless, the results of this method especially for alpha-helical segments are very impressive. All alpha-helical and 71% of the beta-sheet containing regions of the AK1 and GK3 could be identified. Moreover, it was shown that CD spectra of consecutive peptide content reveal the appearance and disappearance of alpha-helical secondary structure elements and help localizing them on the sequence string.

It is investigated whether protein segments predicted to have a well-defined conformational preference in the absence of tertiary interactions are conserved in families of homologous proteins. The prediction method follows the procedures of Rooman, M., Kocher, J.-P., and Wodak, S. (preceding paper in this issue). It uses a knowledge-based force field that incorporates only local interactions along the sequence and identifies segments whose lowest energy structure displays a sizable energy gap relative to other computed conformations. In 13 of the protein families and subfamilies considered that are sufficiently homologous to have similar 3D structures, at least one region is consistently predicted as having the same preferred conformation in virtually all family members. These regions are between 4 and 26 residues long. They are often located at chain ends and correspond primarily to segments of secondary structure heavily involved in interactions with the rest of the protein, suggesting that they could act as nuclei around which other parts of the structure would assemble. Experimental data on early folding intermediates or on protein fragments with appreciable structure in aqueous solution are available for more than half of the protein families. Comparison of our results with these data is quite favorable. They reveal that each of the experimentally identified early formed, or independently stable, substructures harbors at least one of the segments consistently predicted as having a preferred conformation by our procedure. The implications of our findings for the conservation of folding pathways in homologous proteins are discussed.