Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 1, (2002) Adenine Press (2002)

Express Communication

Loop Fold Structure of Proteins: Resolution of Levinthals Paradox

http://www.jbsdonline.com
Abstract According to Levinthal a protein chain of ordinary size would require enormous time to sort its conformational states before the final fold is reached. Experimentally observed time of folding suggests an estimate of the chain length for which the time would be sufficient. This estimate by order of magnitude fits to experimentally observed universal closed loop elements of globular proteins 25-30 residues. Key words: Levinthals paradox, protein folding, chain conformation, closed loops, loop fold structure

Igor N. Berezovsky1,* Edward N. Trifonov2

1Department

of Structural Biology

The Weizmann Institute of Science P.O.B. 26 Rehovot 76100 Israel

2Genome

Diversity Center

Institute of Evolution University of Haifa Haifa 31905 In 1968 Levinthal in his report of which only brief summary is available (1) noted that reversibly denaturable proteins during transition from random disordered state into a well-defined unique structure have to go through conformational space with immense number of states, so that the time required for visiting all the states would also be very large. Indeed, (e.g. (2)) for a protein chain of length L = 150 (residues) with n = 3 alternative conformations for every residue, the time t required for sorting out all possible conformations of the chain is: t = nL = 3150 10-12 s = 1048 yrs [1] Israel

( = 10-12 s is time for elementary transition (2)). Observed values of t are in the range 10-1 to 103 s (2), that is, the full sorting as above is impossible. Thus, protein folding has to proceed along a certain path that would avoid most of the conformational space. The path should somehow be directed by an as yet unknown sequence-dependent folding rule(s). The size of the short chain for which the observed time span of 10-1 to 103 s would be sufficient for trying every possible state can be calculated from [1], with the same assumptions, as l0 = lg(t/) = 23 to 31 residues. In this case all conformlg n ations could be tried during the given time, and the lowest energy state(s) attended. Being logarithmic this estimate is rather insensitive to the choice of the values for n which according to various authors may change between 1.6 and 10 elementary conformations (3, 4). With these extreme values the above estimate spans the range l0 = 11 to 74 residues. The l0 value may, thus, serve as a rough estimate of the size of the units (chains or chain segments), which could attend all conformational states during observed time. The estimated size of the hypothetical unit is identical to the optimum of loop closure for polypeptide chains, 20 to 50 residues (5), and to the observed size of recently discovered closed loop elements, 25-30 amino acid residues (5-7), of

Phone: 972-8-9343367 Fax: 972-8-9344136 E-mail: Igor.Berezovsky@weizmann.ac.il

6 Berezovsky and Trifonov

which globular proteins are universally built. One can hypothesize that the closed loops are also elementary folding units. In this case their linear arrangement within the protein folds (5, 8) would suggest a straightforward folding path: sequential formation of the closed loop units, along with their synthesis in the ribosome. If such successive formation of the stable folding units in the course of translation is assumed, it will require time proportional to the number of the units, that is, only several fold larger than required for a single unit. The above scenario is consistent with the typical rates of translation, 3 to 20 residues per second (9). Synthesis of the protein of length L = 150 takes, thus, 8 to 50 seconds, which is a fair match to the above range of folding rates. Thus, according to the estimates above the consecutive formation of the loop-like folding units of 25-30 residues is by the order of magnitude time-wise consistent with both folding and translational experiments. Acknowledgement We are grateful to Prof. A. Yu. Grosberg for valuable comments and enlightening discussions.
References and Footnotes 1. 2. 3. 4. 5. 6. 7. 8. 9. Levinthal, C. J. Chim. Phys. Chim. Biol. 65, 44-45 (1968). Branden, C. and Tooze, J. Introduction to Protein Structure, Garland Publishing (1999). Bryngelson, J. D. and Wolynes, P. G. Proc. Natl. Acad. Sci. USA 84, 7524-7528 (1987). Pande, V. S., Grosberg, A. Yu. and Tanaka, T. Reviews of Modern Physics 72, 259-314 (2000). Berezovsky, I. N., Grosberg A. Y. and Trifonov, E. N. FEBS Letters 466, 283-286 (2000). Berezovsky, I. N. and Trifonov, E. N. J. Biomol. Struct. Dyn. 19, 397-403 (2001). Berezovsky, I. N. and Trifonov, E. N. J. Mol. Biol. 307, 1419-1426 (2001). Berezovsky, I. N. and Trifonov, E. N. Prot. Engineering 14, 403-407 (2001). Varenne, S., Buc, J., Lloubes, R. and Lazdunski, C. J Mol Biol. 180, 549-576 (1984).

Date Received: July 6, 2002

Communicated by the Editor M. D. Frank-Kamenetskii

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #1

Cunning Simplicity of a Hierarchical Folding

http://www.jbsdonline.com
Abstract A hierarchic scheme of protein folding does not solve the Levinthal paradox since it cannot provide a simultaneous explanation for major features observed for protein folding: (i) folding within non-astronomical time, (ii) independence of the native structure on large variations in the folding rates of given protein under different conditions, and (iii) co-existence, in a visible quantity, of only the native and the unfolded molecules during folding of moderate size (single-domain) proteins. On the contrary, a nucleation mechanism can account for all these major features simultaneously and resolves the Levinthal paradox. Key words: protein folding, Levinthal paradox, two-state kinetics, mid-transition, co-existence of the native and the unfolded phases, folding nucleus, rate of folding

Alexei V. Finkelstein
Institute of Protein Research Russian Academy of Sciences 142290, Pushchino Moscow Region, Russia

Berezovsky & Trifonov (1) have recently revisited an attractive idea of a hierarchic protein folding (2, 3) in an attempt to resolve the Levinthal paradox (4), i.e., to explain how a problem of sampling the impossibly large number of conformations by the folding protein chain can be avoided. A hierarchic mechanism means that some structures formed at the first stage serve, without any significant reconstruction, as building blocks at the next stage of folding, and then the bigger structures obtained at the second stage serve as the building blocks for the next stage, etc. Specifically, Berezovsky & Trifonov in their clearly written paper (1) assume that local closed loops of 25-30 residues (the smallest folding units) find their lowest-energy structures by an exhaustive search of all their conformations and then stick together, entering the tertiary protein fold in the already found form. In principle, a hierarchic mechanism can help to avoid sampling of the huge conformational space. However, I would like to note that any hierarchic scenario cannot serve as a general solution of the Levinthal paradox. Such a mechanism may work only when the native structure is much more stable than the unfolded or denatured state of protein chain. It cannot work, though, when protein folding occurs near the point of thermodynamic equilibrium between the native and denatured states of the protein. Indeed, any hierarchic mechanism (including the one suggested by Berezovsky & Trifonov) implies that the formed folding unit has to preserve its once found form at least until the next unit will be formed, i.e., for the time comparable with that of its own formation. Long life of the once found form means that it is thermodynamically stable as compared to the initial denatured state of the same piece of the chain. Since the folding units then stick together, this means that the native protein is, in turn, more stable than the sum of these sable folded units, and thus much more stable than the denatured state of protein chain. However, a high stability of the native structure is not obligatory for folding. As demonstrated by numerous chevron plots (5-8), many single-domain proteins (having up to 200250 residues) successfully fold near and even in the point of thermodynamic equilibrium between their native and denatured states. This refers to proteins

Phone/Fax: (+7-095) 924 0493 Email: afinkel@vega.protres.ru

311

312 Finkelstein

having a simple two-state folding kinetics, as well as to those that have three-state folding kinetics (i.e., that have detectable folding intermediates far from the equilibrium point). The discovery of this fact in kinetics was actually a re-discovery of wellknown all-or-none thermodynamics of de- and renaturation of single-domain proteins (9). The all-or-none transition means that only the native and denatured proteins are present (close to the mid-transition) in a visible quantity, while others, i.e., semi-native forms, are unstable and therefore present in a very small quantity. An all-or-none transition is a microscopic analog of first order phase transitions (i.e., crystallization) in macroscopic systems (9, 10). Therefore, it is not a surprise that this transition in proteins was shown to follow a nucleation mechanism (8) that is well known (in physics) to be specific for the first order phase transitions (10). Moreover, it has been shown that the nucleation mechanism (that pays main attention to the boundary between the folded and denatured phases within a protein molecule) can resolve the Levinthal paradox (11) and leads to realistic estimates of the protein folding rates (12). The only necessary prerequisite of such a transition is an energy gap between the lowest-energy native fold and misfolded structures (11, 13, 14). On the other hand, the simplicity of a hierarchic folding is cunning. In its strict form, the hierarchic mechanism means that the native protein structure is controlled by kinetics rather than by stability of the whole protein. If so, one has to explain why the same native protein structure results from foldings held under different conditions and having 1000-fold difference in speed (which, according to the logic of Berezovsky & Trifonov (1), has to lead to folding units of rather different sizes); it is noteworthy that this question does not arise at all when the native fold is determined by its stability. Also, one has to explain why the folding with detectable folding intermediates (far from the mid-transition) is not drastically faster than the folding that does not have such intermediates, and why the evidently non native-like intermediates, observed in some cases (15, 16), do not prevent protein from correct hierarchic folding. It is worthwhile to note that the latter fact is easily explained if the choice of the native fold is determined by its stability (rather than by folding kinetics): it is not obligatory that every element of the lowest-energy fold has an enhanced stability, though most of them have to have an enhanced stability for pure statistical reason (17). A hierarchic folding does not provide a general solution of the Levintal paradox since it cannot simultaneously explain for major features, observed for folding of single-domain proteins: (i) folding within non-astronomical time, (ii) independence of the native structure on large variations in the folding rates of given protein under different conditions, and (iii) co-existence, in a visible quantity, of only the native and the denatured protein molecules during the folding near the point of thermodynamic mid-transition between these two states. However, this does not mean that a hierarchic folding scenario is completely inconsistent with all proteins. Specifically, a hierarchic folding seems to take place in large multi-domain proteins whose denaturation is not an all-or-none transition but proceeds as a sum of all-or-none denaturations of their domains (18). The presented considerations give an important criterion of applicability of any protein-folding model to single-domain proteins: whether or not this model can explain protein folding near the point of thermodynamic mid-transition between the folded and unfolded states. A hierarchic folding scheme discussed in this paper cannot satisfy this criterion. The same refers (19) to the simplest versions of a funnel model of protein folding (20-22). On the contrary, the nucleation model of protein folding meets this criterion and resolves the Levinthal paradox. Acknowledgements This work was supported by an International Research Scholars Award from the Howard Hughes Medical Institute and by a grant from the Russian Foundation for Basic Research.

References and Footnotes 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. Berezovsky, I. N. and Trifonov, E. N., J. Biomol. Struct. Dyn. 20, 5-6 (2002). Ptitsyn, O. B., Dokl. Akad. Nauk SSSR 210, 1213-1215 (1973). Baldwin, R. L. and Rose, G. D., Trends Biochem. Sci. 24, 26-33, 77-83 (1999). Levinthal, C., J. Chim. Phys. Chim. Biol. 65, 44-45 (1968). Segawa, S.-I. and Sugihara, M., Biopolymers 23, 2473-2488 (1984). Matouschek, A., Kellis, J. T., Jr., Serrano, L. and Fersht A. R. Nature 340, 122-126 (1989). Kiefhaber, T., Proc. Natl. Acad. Sci. USA 92, 9029-9033 (1995). Fersht, A. R., Curr. Opin. Struct. Biol. 7, 3-9 (1997). Privalov, P. L., Adv. Protein Chem. 33, 167-241 (1979). Landau, L. D. and Lifshitz, E. M., Statistical Physics. London, Pergamon (1959). Finkelstein, A. V. and Badretdinov, A. Y., Folding & Design 2, 115-121 (1997). Galzitskaya, O. V., Ivankov, D. N. and Finkelstein, A. V., FEBS Letters 489, 113-118 (2001). J. B. Bryngelson and P. G. Wolynes, Proc. Natl. Acad. Sci. USA 84, 7524 (1987). Goldstein, R. A., Luthey-Schulten, Z. A. and Wolynes, P. G., Proc. Natl. Acad. Sci. USA 89 4918-4122; ibid., 9029-9033 (1992). Baldwin, R. L., Nature Struct. Biol. 8, 92-94 (2001). Fernandez, A., J. Biomol. Struct. Dyn. 19, 735-737 (2002). Finkelstein, A. V., Badretdinov, A. Ya., and Gutin, A. M., Protein 23, 142-150 (1995). Privalov, P. L., Adv. Protein Chem. 35, 1-104 (1982). Bogatyreva, N. S. and Finkelstein A. V., Protein Engineering 14, 521-523 (2001). Zwanzig, R., Szabo, A. and Bagchi, B., Proc. Natl. Acad. Sci. USA 89, 20-22 (1992). Chan H. S. and Dill K. A., Proteins 30, 2-33 (1998). Bicout, D. J. and Szabo, A., Protein Science 9, 452-465 (2000).

313 Cunning Simplicity of a Hierarchical Folding

Date Received: September 16, 2002

Communicated by the Editor Ramaswamy H Sarma

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #2

Back to Units of Protein Folding

http://www.jbsdonline.com Entia non sunt multiplicanda praeter necessitatem (Occam)
Abstract In response to the criticism by A. Finkelstein (J. Biomol. Struct. Dyn. 20, 311-314, 2002) of our Communication (J. Biomol. Struct. Dyn. 20, 5-6, 2002) several issues are dealt with. Importance of the notion of elementary folding unit, its size and structure, and the necessity of further characterization of the units for the elucidation of the protein folding in vivo are discussed. The criticism (J. Biomol. Struct. Dyn. 20, 311-314, 2002) on the hierarchical protein folding is also briefly addressed. Key words: Levinthal paradox; folding units; protein folding, biological relevance.

Igor N. Berezovsky1,* Edward N. Trifonov2

1Department

of Structural Biology

The Weizmann Institute of Science P.O.B. 26, Rehovot 76100, Israel

2Genome

Diversity Center

Institute of Evolution University of Haifa Haifa 31905, Israel

The main points of our Communication (2) the structural units of folding and closed loops as likely candidates for that role are not challenged by A. Finkelstein (1). But his piece rather invites to a separate chapter in protein folding studies a hierachical folding (3-8). The passage in our note (2) If such successive formation of the stable folding units in the course of translation is assumed, is not a statement, and it is not about hierarchical folding, but rather a simple illustration of how the elementary structural units may partake in the initial stages of folding process during translation, along with the synthesis of the polypeptide chain. This biological dimension is, apparently, a crucial component in protein folding problem. The structural units of folding may or may not remain intact once formed, and their fate during protein folding may follow many different scenaria (6, 7). But again, the discussion on these scenaria is well beyond the scope of our original Communication (2). The very notion of independently folding units is an important concept, and every contribution towards their characterization (9-13) is important. Our study suggests a rather narrow size range for the hypothetical folding units, and the structural component to it - chain-return nature of the units (14). Calculations for the ordinary protein sizes on the basis of Levinthals original idea (15) lead to astronomical times. This initiated decades of theoretical and experimental research (16-20) to figure out what is a time-wise agreeable path of folding of the full-size proteins. The same calculation in reverse as in (2) leads to what we would call the Levinthal limit in other words the size of the polypeptide chain or part thereof for which the observed times of protein folding would be sufficient to sort out all possible conformations. The respective simple formalism provides a rough estimate of the size of such a chain, under the assumption that there are no returns to the same conformations during the sorting. More elaborate approaches should take into consideration both physical factors and biological circumstances that may influence the estimate. It may, thus, become either smaller (e.g. due to long-living native and non-native meta-stable states (5,7)), or larger (e.g. due to

Phone: 972-8-9343367 Fax: 972-8-9344136 Email: Igor.Berezovsky@weizmann.ac.il

315

316 Berezovsky and Trifonov

special sequence organization (4, 21-24), and possible structural guidance within the ribosome or in chaperones (25, 26)). The balance is still to be found. Possible clues are apparent evolutionary fingerprints in protein sequences (23, 24), that indicate the same size range, 25-30 amino acid residues, as the one suggested by the polymer statistics of the polypeptide chains (14) and by the observed rates of translation (2). We believe, that both experimental and theoretical studies on the units of folding is the right way to elucidation of how exactly the protein folds in vivo.
References and Footnotes 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. Finkelstein, A. V., J. Biomol. Struct. Dyn. 20, 311-314 (2002). Berezovsky, I. N. and Trifonov, E. N., J. Biomol. Struct. Dyn. 20, 5-6 (2002). Ptitsyn, O. B,. Dokl. Akad. Nauk SSSR 210, 1213-1215 (1973). Baldwin, R. L. and Rose, G. D., Trends Biochem. Sci. 24, 26-33 (1999). Baldwin, R. L. and Rose, G. D., Trends Biochem. Sci. 24, 77-83 (1999). Fersht, A. R., Proc. Natl. Acad. Sci. USA 97, 1525-1529 (2000). Baldwin, R. L., Nat. Struct. Biol. 6, 814-817 (1999). Fernandez, A., J. Biomol. Struct. Dyn. 19, 735-737 (2002). Abkevich, V. I., Gutin, A. M., Shakhnovich, E. I., Biochemistry 33, 10026-10036 (1994). Panchenko, A. R., Luthey-Schulten, Z. and Wolynes, P. G., Proc. Natl. Acad. Sci. USA 93, 2008-2013 (1996). Panchenko, A. R., Luthey-Schulten, Z., Cole, R. and Wolynes, P. G., J. Mol. Biol. 272, 95105 (1997). Berezovskii, I. N., Esipova, N. G. and Tumanyan, V. G., Biophysics 42, 557-565 (1997). Galzitskaya, O. V., Ivankov, D. N. and Finkelstein A. V., FEBS Lett 489, 113-118 (2001). Berezovsky, I. N., Grosberg A. Y. and Trifonov, E. N. FEBS Letters 466, 283-286 (2000). Levinthal, C. J., Chim. Phys. Chim. Biol. 65, 44-45 (1968). Baldwin, R. L., Nat. Struct. Biol. 6, 814-817 (1999). Fersht, A. R., Curr. Opin. Struct. Biol. 7, 3-9 (1997). Shakhnovich, E. I., Curr. Opin. Struct. Biol. 7, 29-40 (1997). Pande, V. S., Grosberg, A. Yu. and Tanaka, T., Reviews of Modern Physics 72, 259-314 (2000). Fersht, A. R. and Daggett, V., Cell 108, 573-582 (2002). Shakhnovich, E. I. and Gutin, A. M., Proc. Natl. Acad. Sci. USA 90, 7524-7528 (1993). Pande, V. S., Grosberg, A. Yu. and Tanaka, T., Proc. Natl. Acad. Sci. USA 91, 12972-12975 (1994). Trifonov, E. N., Kirzhner, A., Kirzhner, V. M. and Berezovsky, I. N., J. Mol. Evol. 53, 394401 (2001). Berezovsky, I. N. and Trifonov, E. N., Comparative and Functional Genomics, in press (2002). Gething, M. J. and Sambrook, J., Nature 355, 33-45 (1992). Ruddon, R. W. and Bedows, E., J. Biol. Chem. 272, 3125-3128 (1997).

Date Received: October 25, 2002

Communicated by the Editor Ramaswamy H Sarma

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #3

A Few Disconnected Notes Related to Levinthal Paradox

http://www.jbsdonline.com
Abstract We estimate that the longest protein chain capable of exhaustive sampling of all its conformations within a millisecond is shorter than 15 residues. This reinforces the understanding of Levinthal paradox which emerged in the last decade, namely, that cooperative (all-ornone) character of folding and unfolding transition is indicative of the sequences selected, such that reliable folding does not require exhaustive conformation sampling. The opinion is formulated that the discussions of Levinthal paradox should now fly to the new spheres.

A. Grosberg1,2
1Department

of Physics

University of Minnesota Minneapolis, MN 55455, USA

2Institute

of Biochemical Physics

Russian Academy of Sciences Moscow 117977, Russia

A long time ago, well before the breakup of Soviet Union, a large biophysics meeting was held in the then Soviet Republic of Georgia. The site was a rural place in the center of a famous wine producing region, the month was October, and the major event was the all-Georgian wine testing festival, advertised as a merry traditional peasant holiday. When biophysicists arrived, they found a large plaza with dozens of pavilions, each representing a particular village, and each offering (for free!) a glass of young wine. Many biophysicists deem themselves experts, and their intent was to test systematically all different sorts of wine. Before long, however, the cloud of biophysicists seemed perfectly obeying the diffusion equation, with each individual in the cloud undergoing random walks. The few participants (perhaps bad biophysicists) who were able to continue scientific observations realized soon that wine testing continued for an unexpectedly long time. Assuming visit to one pavilion takes time , and assuming there were some M pavilions, one could have naively expected that after time close to M the testing would be over. Such expectation was proved to be terribly wrong, and the real time was much longer than that. One possible reason is trivial: wine testers, even if they were to complete the exhaustive testing, are unlikely to realize the completion of the task and to stop at that. But there is also another more interesting reason: the time necessary for the random walk to visit all M sites does not scale as M, it can be significantly larger than that, simply because random walk visits some sites great many times before the first visit to some other sites. This simple property of random walks and diffusion processes seems to be underappreciated in the current discussions revolving around the celebrated Levinthal paradox (1), particularly in the ones initiated by the recent very clearly written article (2) by I. Berezovsky and E. Trifonov (IBET for brevity). In the most standard formulation, Levinthal paradox arises from the idea that the time required for a protein molecule to sample all of its conformations is at least M, where M is the number of distinct conformations, and is the time necessary to sample one conformation. Then, the paradox goes, unguided folding into one particular (native) state requires at least time of order M which is far too long, because M is astronomically large. In fact, M is so large that, for instance, H. Fraunfelder (3) suggests to

Email: grosberg@physics.umn.edu

317

318 Grosberg

call it a biological number, where biological numbers dwarf astronomical ones. Clearly, M is so large because it is exponential in the degree of polymerization of the protein chain, N: M = esN, where s is a constant close to unity (see, e.g., (4)). I. Berezovsky and E. Trifonov (2) turned the Levinthals argument up-side-down. They argued that while exhaustive sampling of all conformations is out of reach for large N, it is definitely possible for sufficiently small N. Thus, they decided to estimate the length of protein chain N0 such that at N < N0 protein can exhaustively sample all of its conformations within some specified time T, say, one millisecond. Assuming exhaustive sampling time M, they wrote esN0 = T and obtained N0 = (1/s) 1n (T/) . [1]

This number, according to (2), turns out to be somewhere between 23 and 31. This led IBET to a series of attractive speculations. Leaving speculations aside for a moment, let us discuss the estimate of N0 more closely. First of all, the very question is perfectly legitimate: what is the maximal length of the chain which can exhaustively sample all of its conformations within the specified time interval T, such as a millisecond to a second? Comparison with wine testing makes it immediately clear that the answer depends on how the sampling is organized. Indeed, in terms of wine testing, M is the time required to visit all pavilions in an orderly fashion, one pavilion after another, never returning to the already visited place. Of course, a sober person can do that, but sober model is unrealistic for the wine tester. Equally unrealistic is the model of protein chain dynamics which orderly samples all conformational states, never returning to the once visited conformation. For all other sampling strategies, the exhaustion time is larger than M. In fact, Levinthal did not say that the time of exhaustive conformation sampling (or wine testing) was M - he said it was at least M; in other words, he said it is M. This was sufficient for him to conclude that exhaustive sampling is impossible for realistic N, such as N = 150. The opposite, and perhaps more realistic, model of protein dynamics (and also of wine testing) would be purely random walk in the space of conformations. To begin with, suppose it is an unbiased random walk, which means there is no conformation dependent (free) energy landscape involved. (Note that energy landscape, when present, will increase the exhaustion time (because exhaustion requires visiting all the tops of all energy barriers) unlike folding time, which, of course, can be decreased). How can we estimate the exhaustive sampling time for the unbiased random walk model? Consider first that wine testing pavilions arranged along a line. Then random walk of longevity t brings us as far as about t/, which means we cover all M sites when t/ M, and the time of exhaustive sampling is t 2. Needless saying, the difference between M and M2 is very significant. Based on M2, IBET would have obtained N0 = (1/2s) 1n (T/) , [2]

twice smaller than (1). This is the length between 9 and 15, which seems to rule out most of the speculations by IBET. In fact, accurate estimate of exhaustive sampling time by a random walk is not completely trivial. More sophisticated estimate for one dimensional case, which will not be derived here, reads t M2 / 1nM. This increases the result for N0 to between 11 and 16. For the random walk in the space of higher dimension d, the result depends on d. When d crosses over 2, the mechanism of sampling changes, because random walk tends to leave behind large unvisited regions. At d > 2, exhaustive sampling is only possible because the overall volume is restricted, and random walk is forced to come back.

What is d in reality is anybodys guess. Please do not forget that d here is the dimension of the abstract space of protein conformations, not the real three-dimensional space. The relevant dimension was measured for the vicinity of the native conformation of a lattice model protein (5). Measurements in different regions of conformation space yield results for d between 1.4 and 4.5. The result close to (1) would be correct for the dimension d as high as M; in this case, exhaustion time would have scaled as M ln M, yielding N0 between 18 and 25. However, this estimate is completely unrealistic, because d = M corresponds to the situation where each conformation (site in conformation space, or wine pavilion) can be equally probably reached from every other conformation in just one step . Clearly, real protein chains are nowhere near this extreme. There are quite a few other factors (5, 6), all reducing the result for N0, and all fundamentally arising from the fact that we are dealing with a polymer chain in which all units are linearly connected. To conclude this part, it appears that IBET significantly overestimated the length of a protein capable of exhaustive conformation sampling. It seems safe to say that this length is smaller than 15. We leave it for the reader to decide whether the concept of folding should be applied to such a short chain. At the next stage, IBET suggested that exhaustively sampling blocks combine together to form hierarchically folding large proteins. This possibility was critically reviewed recently by Finkelstein (7). His analysis seems quite convincing. His major point is that reasonably fast folding is frequently observed (see (8) and references therein) under the conditions where native state is not significantly lower in free energy (or not lower at all!) than fully denatured state that is, in the point of thermodynamic equilibrium between native and denatured states. Under such conditions, even the globule as a whole is not particularly stable, while the parts supposedly the units of hierarchical scenario are not stable at all. In addition to this convincing argument by Finkelstein, the small value of N0 makes the stability of the presumed folding units questionable, and the whole of the hierarchical scheme even more difficult to imagine. It should be noted that the understanding of Levinthal paradox has progressed very significantly since it was first formulated (1). First of all, it is found that the folding time, under the conditions of thermodynamic equilibrium between folded and unfolded states, scales as exp (sN2/3) (9-11), which is very significantly smaller than Levinthal time proportional to exp (sN). This estimate, as already said, is valid under the conditions of thermodynamic equilibrium, which means that it relies on the transition between denatured and native forms being highly cooperative, of all-or-none type. Indeed, high cooperativity is a well established experimental fact (12). It is also well understood that high cooperativity is the property of proteins which is due to their peculiar selected sequences. Among random sequences, vast majority would not have exhibited any signs of cooperativity, as it was first established by Shakhnovich and Gutin (13). This latter fact has been extensively tested using lattice models (as described, e.g., in the review article (4); see also references therein). Since everything related to the lattice models is perceived with a large dose of (healthy?) skepticism in protein community, it is important to emphasize that the fact of non-cooperative folding in the majority of sequences is well understood beyond lattice models. Actually, it was foreseen by Bryngelson and Wolynes a long time ago (14). Speaking about the relation between sequence selection, the all-or-none cooperative mechanism of folding, as the scaling of folding time, it is interesting to mention that experimental observations do not provide any evidence on the folding (under equilibrium conditions) time dependence on the chain length. That means,

319 Disconnected Notes Related to Levinthal Paradox

320 Grosberg

the above mentioned theoretical prediction, exp (sN2/3), although sufficient to rule out any paradoxes, may be still an overestimate. The role of sequence selection is also well understood from a different view point, namely, related to the mutation stability (see also in the review article (4)). In the majority of sequences, every mutation breaks the stability of the native state with the probability very close to 100%. By contrast, the selected sequences the same ones which exhibit highly cooperative folding-unfolding transition! are reliable in the sense that their native state with high probability survives and remains stable even after several mutations. While the real mechanisms of evolutionary sequence selection remain unknown, and while the computational models of sequence selection keep improving since the first suggestions (15, 16), it is getting increasingly clear that there are many sequences which meet the sufficient criteria of reliable folding. To summarize, in the light of all the findings of the last decade, it seems clear (to the present author, at least) that the discussions about Levinthal paradox must now move forward to the new spheres. How does the sequence selection work (or worked) in real evolution? What are the specific scenario of folding dynamics for selected sequences how specific is the nucleation, how many and which conformations belong to the transition state, what is the reaction coordinate associated with folding; in other words, how precisely do these selected sequences slide down their folding funnels (17-20)? What are the physical principles behind the selection of certain spatial structures, or folds and fold families (21)? What are the general physical principles behind the enzymatic, motor and other functions of proteins, and do they have any relation to the principles involved in folding? What are the mechanisms of aggregation, or mechanisms preventing aggregation, of proteins? There are very many works on these subjects, to make the list of them is a daunting task far beyond the framework of the present note. However many questions remain open, it seems that the Levinthals question how can protein sample biologically large number of conformations has been answered: protein does not sample them. Most importantly, there are sufficiently many good sequences for the evolution to select from, where every good sequence is capable of folding, and does not need exhaustive conformation sampling to do so. Understanding this was a remarkable achievement of the last decade. I am indebted to I. Berezovsky and E. Trifonov. Their paper (2) initiated the present note, and my personal discussions with them were useful and pleasant. I thank also V. Pande for critical reading of the first draft of this manuscript.
Reference and Footnotes 1. Levinthal, C., J. Chim. Phys. Biol. 65, 44-45 (1968). 2. Berezovsky I., Trifonov E., J. Biomol. Struct. Dyn. 20, 5-6 (2002). 3. H. Fraunfelder, Colloquium talk at the University of Minnesota Physics Department, October, 2002. 4. Pande, V. S., Grosberg A. Yu. and Tanaka, T., Reviews of Modern Physics 72, 259-314 (2000). 5. Du, R., Grosberg, A., Tanaka, T., Phys. Rev. Letters 84, 1828-1831 (2000). 6. Scala, A., Nunes Amaral, L. A., Barthlmy, M., Europhys. Lett. 55, 594-600 (2001). 7. Finkelstein, A. V., J. Biomol. Struct. Dyn. 20, 311-314 (2002). 8. Fersht, A. Curr. Opin. Struct. Biol. 7, 3-9 (1997). 9. Finkelstein, A., Badretdinov, A., Folding & Design 2, 115 (1997); Folding & Design 3, 67 (1998). 10. Du, R., Grosberg, A. Yu., Tanaka, T., Phys. Rev. Letters 83, 4670-4673 (1999). 11. Finkelstein, A. V., Ptitsyn, O. B., Protein Physics, Academic Press, 2002. 12. Privalov, P. L., Adv. Protein. Chem. 33, 167-241 (1997). 13. Shakhnovich, E., Gutin, A., Biophys. Chem. 34, 187, 1989. 14. Bryngelson, J. D., Wolynes, P. G., Proc. Natl. Ac. Sci. 87, 7524-7528 (1987). 15. Shakhnovich, E. I., Gutin, A. M., Proc. Nat. Acad. Sci., USA 90, 7195 (1993). 16. Pande, V. S., Grosberg, A. Yu., Tanaka, T., Proc. Nat. Acad. Sci. USA 91, 12972 (1994). 17. Shakhnovich, E. I., Folding and Design 1, 50-52 (1996); Shakhnovich, E. I. Curr. Opin. Struct. Biol. 7, 29-40 (1997).

18. Pande, V. S., Grosberg, A. Yu., Tanaka, T., Rokhsar, D. S., Curr. Opin. Struct. Biol. 8, 68-79 (1998). 19. Onuchic, J. N., Socci, N. D., Luthey-Schulten, Z., Wolynes, P. G., Folding and Design 1, 441-450 (1996). 20. Dill, K. A., Chan, H. S., Nat. Struct. Biol. 4, 10-19 (1997). 21. Li, H., Helling, R., Tang, C., Wingreen, N., Science 273, 666 (1996).

321 Disconnected Notes Related to Levinthal Paradox

Date Received: October 27, 2002

Communicated by the Editor Ramaswamy H Sarma

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #4

Loop Folds in Proteins and Evolutionary Conservation of Folding Nuclei

http://www.jbsdonline.com
Abstract We show that loops of close contacts involving hydrophobic residues are important in protein folding. Contrary to Berezovsky and Trifonov (J. Biomol. Struct. Dyn. 20, 5-6, 2002) the loops important in protein folding usually are much larger in size than 23-31 residues, being instead comparable to the size of the protein for single domain proteins. Additionally what is important are not single loop contacts, but a highly interconnected network of such loop contacts, which provides extra stability to a protein fold and which leads to their conservation in evolution. Key words: protein folding, Levinthal paradox, loop fold structure, closed loops, evolutionary conservation, folding nucleus

Andrzej Kloczkowski Robert L. Jernigan*

Baker Center for Bioinformatics and Biological Statistics Iowa State University 123 Office and Lab Bldg. Ames, IA 50011-3020

The title Loop Fold Structure of Proteins: Resolution of Levinthal paradox of the communication by Igor N. Berezovsky and Edward N. Trifonov (1) follows a long line of work, looking for order or regularity in proteins. It suggests that the Levinthal paradox really does exist and has yet not been resolved, which is untrue. There is significant literature on this problem, with probably the most fundamental paper on this paradox written ten years ago by Zwanzig et al. (2). The common opinion in the protein folding community is that the Levinthal paradox (3) of finding a needle in a haystack doesnt exist because proteins do not fold by randomly searching all possible (extremely large) numbers of conformations. The folding landscape does not resemble a flat golf course with a single hole corresponding to the native state. Rather it looks more like a bumpy funnel, so that the ball rolls almost always (except for kinetic traps) downhill (4, 5). The funnel-like landscape of folding is, according to the most popular theories, the result of hydrophobic collapse, which greatly reduces the total conformational space (6-8). A protein does not fold from random coil conformations, and the assumption of a random coil as a starting point for the folding process is the basis of the Levinthal paradox. Proteins fold to the native state from non-native (denaturated) states that are already substantially structured and make all the arithmetic supporting the Levinthal paradox irrelevant. In the letter Cunning simplicity of a hierarchical folding Alexei V. Finkelstein (9) does not support this simplest funnel-like mechanism of folding. Instead he explains the folding pathways through formation of a folding nucleus and a delicate thermodynamic balance between the native and denaturated states of the protein (10). He also classified the idea of Berezovsky and Trifonov as a hierarchical folding and shows that such hierarchical folding would lead to a native state that is too stable to unfold. In our opinion the terminology hierarchical folding used by Finkelstein is improper. Hierarchical folding usually refers to the old views on protein folding, that the primary structure (sequence) leads to the formation of the protein secondary strucPhone: (515) 294-3833 Fax: (515)294-3841 Email: jernigan@iastate.edu

323

324 Kloczkowski and Jernigan

ture, which in turn leads to the formation of the protein tertiary structure. Such terminology leads him to conclude that the closed loops with nearly standard size segments of 23-31 residues reported by Trifonov and coworkers cannot be combined with more modern mechanisms of folding kinetics, such as a nucleation mechanism. As a matter of fact closed loops, but not necessarily only those of almost regular size 23-31 as reported by Berezovsky et al. in the series of their earlier papers (1113), might be quite important in protein folding. The formation of such closed loops in the protein native state does not mean that residues forming such loop contacts always remain intact during the folding process, but there is a possibility that these loop contacts inside a protein core might be a part of the folding nucleus (14), especially if several of such loop contacts involving hydrophobic residues are located near one another inside the core. We studied the non-functional evolutionarily conserved residues in several subfamilies of proteins. The non-functional conserved residues are located inside a protein core and their detection might give us valuable information about the mechanism of protein folding. This work was initiated in our Lab by the late Oleg Ptitsyn who studied different subfamilies of c-type cytochromes (15). He came to the conclusion that there is a common folding nucleus composed of four residues. Using the horse cytochrome c (1hrc) (which has the total length on 105 residues) as the reference for the numbering, the conserved residues are Gly(6), Phe(10), Leu(94) and Tyr(97). All these residues are hydrophobic and form a network of five conserved contacts 6-94, 6-97, 10-94, 10-97 and 94-97. According to Ptitsyn this network of conserved contacts could be the folding nucleus for cytochrome c, which is of critical importance for its folding. Note that the sizes of these loops are mostly substantially larger than the 23-31 proposed by Berezovsky et al. The second subfamily of proteins studied by Ptitsyn and Kai-Li Ting was globins (16). Using sperm whale myoglobin (1mbd) as the reference the structurally conserved residues are Val(10), Trp(14), Ile(111), Leu(115), Met(131) and possibly Leu(135). All these hydrophobic residues form a network of conserved contacts which is possibly the folding nucleus. Recently Kai-Li Ting and Jernigan (17) studied the conservation of the non-functional residues in the lysozyme/-lactalbumin family and identified the possible folding nucleus. They identified 19 conserved hydrophobic non-functional residues. That is probably too many due to a lack of substantial evolutionary diversity in the studied protein family. Another possibility is that there are present three conserved subclusters, where each subcluster is composed of a network of conserved contacts. Importantly, all of these conserved residues are obtained only by including phylogenetically diverse cases, pointing up the uncertainties involved in sequence comparisons. Kloczkowski and Jernigan (18) developed a theory of evolutionary conserved residues based on the hydrophobic-polar HP lattice model of protein and showed that a highly connected network of hydrophobic contacts provides extra stability to a protein fold and may be crucial in reaching the lowest energy native state. We have developed a model for locating the evolutionarily conserved residues in proteins. We first find a core of a protein with the known structure based on the packing of residues, measured by number of contacts. Then we find the so called supercore inside the core, by maximizing the number of hydrophobic contacts. The predicted conserved residues for 1hrc and 1mbd are exactly the same as those reported by Ptitsyn and Ptistyn and Ting, respectively. The theory was also applied with success for identifying the conserved residues reported by Mirny and Shakhnovich (19) who used COC (conservatism of conservatism) method. These results show that looping contacts are indeed important for protein folding. What is however important are not the single loop contacts of regular size 23-31

residues, reported by Berezovsky et al., but rather an interconnected network (or cluster) of such contacts of loops of much larger size, but possibly involving also short helical contacts. In the case of 1hrc such a loop is of size 90 residues (which is nearly the total length 105 of the protein), while in the case of 1mbd the size of the loops is 100-120, also approaching the total length 153 of the protein. These are more consistent with the frequently remarked upon feature of proteins that the ends of the chain are close together (20).
References and Footnotes 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. Berezovsky, I. N. and Trifonov, E. N., J. Biomol. Struct. Dyn. 20, 5-6 (2002). Zwanzig, R., Szabo, A. and Bagchi, B., Proc. Natl. Acad. Sci. USA 89, 20-22 (1992). Levinthal, C., J. Chim. Phys. Chim. Biol. 65, 44-45 (1968). Bryngelson, J. D. and Wolynes, P. G., Proc. Natl. Acad. Sci. USA 84, 7524-7528 (1987). Dill, K. A., Protein Sci. 8, 1166-1180 (1999). Lau, K. F. and Dill, K. A., Proc. Natl. Acad. Sci. USA 87, 638-642 (1990). Chan, H. S. and Dill, K. A., J. Chem. Phys. 95, 3775-3787 (1991). Yue, K. and Dill, K. A., Proc. Natl. Acad. Sci. USA 92, 146-150 (1995). Finkelstein, A. V., J. Biomol. Struct. Dyn. 20, 311-314 (2002). Finkelstein, A. V., Badretdinov, A. Y. and Gutin, A. M., Proteins 23, 142-150 (1995). Berezovsky, I. N., Grosberg A. Y. and Trifonov, E. N., FESB Letters 466, 283-286 (2000). Berezovsky, I. N. and Trifonov, E. N., J. Mol. Biol. 307, 1419-1426 (2001). Berezovsky, I. N. and Trifonov, E. N., J. Biomol. Struct. Dyn. 19, 397-403 (2001). Ptitsyn, O. B., Dokl. Akad. Nauk SSSR 210, 1213-1215 (1973). Ptitsyn, O. B., J. Mol. Biol. 278, 655-666 (1998). Ptitsyn, O. B. and Ting, K.-L., J. Mol. Biol. 291, 671-682 (1999). Ting, K.-L. and Jernigan, R. L., J. Mol. Evol. 54, 425-436 (2002). Kloczkowski, A. and Jernigan, R. L., Evolutionary Conserved Residues and Protein Folding, to be published. 19. Mirny, L. A. and Shakhnovich, E. I., J. Mol. Biol. 291, 177-196 (1999). 20. Bahar, I. and Jernigan, R. L., Biophys. J. 66, 454-466 (1994).

325 Loop Folds in Proteins and Conservation of Folding Nuclei

Date Received: November 6, 2002

Communicated by the Editor Ramaswamy H Sarma

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #5

What is Paradoxical about Levinthal Paradox?

http://www.jbsdonline.com
Abstract We would be tempted to state that there has never been a Levinthal paradox. Indeed, Levinthal raised an interesting problem about protein folding, as he realized that proteins have no time to explore exhaustively their conformational space on the way to their native structure. He did not seem to find this paradoxical and immediately proposed a straightforward solution, which has essentially never been refuted. In other words, Levinthal solved his own paradox.

Marianne Rooman* Yves Dehouck Jean Marc Kwasigroch Christophe Biot Dimitri Gilis
Ingnierie Biomolculaire Universit Libre de Bruxelles CP 165/64, 50 avenue Roosevelt B-1050 Bruxelles, Belgium

During a meeting held in 1969 in Monticello, Levinthal estimated the number of different conformations accessible to a 150-residue protein to be roughly of the order of 10300, whereas the number of conformations sampled by a natural protein before reaching its final state is of the order of 108 (1). In the same report (usually incorrectly referenced), Levinthal himself solved what could at that time appear as nonintuitive, even paradoxical. Indeed, he proposed that protein folding is speeded and guided by the rapid formation of local interactions which then determine the further folding of the peptide; this suggests local amino acid sequences which form stable interactions and serve as nucleation points in the folding process. Of course, there are two words in Levinthals sentence that are questionable: local and stable. Let us start with the last one. Clearly, a few residues can only present a very marginal stability. Anfinsen clarified this point in 1973 by introducing the concept of flickering equilibria (2): it seems reasonable to suggest that portions of a protein chain that serve as nucleation sites for folding will be those that can flicker in and out of the conformation that they occupy in the final protein, and that they will form a relatively rigid structure stabilized by a set of cooperative interactions. Moreover, Anfinsen did not restrict nucleation to local interactions along the protein chain, since he stated that the nucleation centers might be expected to involve substructures as helices, pleated sheets or -bends. It was realized at about the same time that the folding of a protein into its native structure essentially occurs independent of the initial conditions; this conclusion was reached on the basis of the experimental observation that small denatured proteins were able to refold in vitro (3), under suitable conditions. This ruled out the hypothesis, to our knowledge originally due to Chantrenne (4), that the only way of achieving correct folding is by the sequential growth of the polypeptide chain on the ribosome. In the 70s, the prevailing view was that folding follows pathways along which nucleation events take place. The number of pathways was (and is still) a subject of debate. In particular, according to Honig et al. (5) in 1976, proteins fold by following a multiply branched pathway. In good agreement with this assumption, most current scenarios involve a huge number of parallel pathways possibly sharing a number of key steps. Another point of discussion concerned the question of whether protein folding is under thermodynamic or kinetic control. According to Anfinsen (2), native protein

*Phone:

32-2-650 2067/5572 Fax: 32-2-650 3606 Email: mrooman@ulb.ac.be

327

328 Rooman et al.

structures correspond to global free energy minima. But Levinthal (1) argued that the final conformation has not necessarily to be the one of lowest free energy. It obviously must be a metastable state which is in a sufficiently deep energy well to survive possible perturbations in a biological system. These chosen extracts show that the folding problem was well understood in the early 70s. What have we learned since then? Many new folding models, theories and concepts have been proposed, often based on interesting ideas or experimental observations, and indisputably provide valuable precisions and clarifications on the mechanisms of protein folding. We feel however that the originality of these new concepts is often somewhat overrated by comparing to a hypothetical classical view of protein folding which would involve well defined pathways, predominance of local interactions and an absolute necessity for stable intermediates. As stressed by Honig (6), this simplistic view is very unlikely to correctly reflect the thoughts prevailing in those pioneering days. Curiously however, the lack of contradiction between old and new views has not prevented ongoing, passionate, debates around Levinthals paradox, frequently opposing the partisans of different models. First, there has been a continuing dispute between the supporters of nucleation and those of hierarchic folding, where typically some secondary structure elements or loops form first. These two views easily reconcile when considering that small structure elements can only flicker in and out, as nicely pointed out by Anfinsen (2). Therefore, hierarchic folding units must not be viewed as rigid but rather as flickering entities. On the other hand, it is obvious that all residues along the chain are not equally prone to constitute nucleation centers. This is supported by the experimental observation that some peptides are more structured in solution than others (7), which actually means that they flicker in and out of a specific conformation. Moreover, some specific protein residues have been observed to form native tertiary contacts earlier than others and to stabilize folding nuclei (8). Hence, we do not feel that hierarchic folding must be opposed to nucleation; rather, these approaches should be considered as complementary. In the context of hierarchic folding, it has been recently suggested that proteins are made up of closed loops that fold separately (9). It is difficult to believe that all current proteins exhibit this property. However, the idea that original proteins were small closed-loop peptides with flickering stability, which have been assembled during evolution, and that some trace of this evolution is left in the current proteins, is quite attractive. Again, with this softened view, much of the controversy vanishes. Another much debated issue is whether nucleation centers consist of local interactions along the chain or of tertiary contacts. The answer seems obvious: it is proteindependent. Indeed, numerous experimental and theoretical studies have revealed protein sequences exhibiting a strong signal towards the native structure, encoded locally along the sequence or, on the contrary, in specific tertiary contacts. Both tendencies are probably often conjugated, with for example small flickering secondary structure elements forming a tertiary contact and thereby inducing nucleation. To achieve rapid folding towards the native state, it has been suggested that the energy gap between the native conformation and the other conformations that are structurally unrelated to it must be sufficiently large (10). This is probably, in general, a necessary condition, at least in a slightly modified form taking into account the existence of proteins adopting several folded structures, depending or not on the environmental conditions. Another concept proposed a few years ago is that protein energy landscapes have the shape of a funnel (11). This unquestionably yields a very nice, intuitive, vision of the folding mechanism, where folding is funneled towards low energy states, which are much less numerous than high energy states.

However, this basically corresponds to requiring the independence towards initial conditions and the absence of insurmountable energy minima on the ways to the native state, and to translating the old view in the framework of statistical mechanics. Besides, a funnel-like shape has been shown to constitute an insufficient condition for ensuring consistent folding (12). Finally, with the finding of proteins that polymerize, aggregate, exhibit domain swapping or several folded structures, it was realized that folding is even more complex to predict and simulate than originally hoped for, and that there is probably also not a unique answer to the question whether native structures correspond to relative or absolute free energy minima. These considerations bring us to think about the folding mechanism that evolution tends to favor. This issue is probably related to the possible biological role played by the multiplicity of conformations, or to its pathological consequences. But most certainly, evolution singled out a tiny fraction of possible amino acid sequences, having adequate folding and functional properties. Pretending that nothing has evolved since the early 70s is certainly an exaggeration. The development of potent experimental and theoretical techniques have led to support and clarify many of the proposed views. We have, however, the impression that some earlier contributions in the disciplne have been forgotten or misinterpreted and that in light of these, recent advances should somewhat be relativized. Acknowledgments D. G. and M. R. are Research Assistant and Research Director, respectively, at the Belgian National Fund for Scientific Research (FNRS). C. B. is supported by a BioVal research program of the Walloon Region, and Y. D. by a grant from the Fonds de la Recherche pour lIndustrie et lAgriculture (FRIA).
References and Footnotes 1. Levinthal, C. How to Fold Graciously. In Mossbauer spectroscopy in biological systems. Proceedings of a meeting held at Allerton House, Monticello, Illinois. Edited by Debrunner P., Tsibris J. & Munck E. (University of Illinois Press, Urbana, Illinois, 1969), pp 22-24. 2. Anfinsen, C. B., Science 181, 223-230 (1973). 3. Anfinsen, C. B., Haber, E., Sela, M., and White F. H., Proc. Natl. Acad. USA 47, 1309-1314 (1961); Anfinsen, C. B. General remarks on protein structure and biosynthesis. In Informational Macromolecules, edited by Vogel, H. J., Bryson, V. and Lampen, J. O. (Academic Press, NewYork & London, 1963), pp 153-166; Schechter, A. N., Chen, R. F., and Anfinsen, C. B., Science 167, 886-887 (1970). 4. Chantrenne H., The Biosynthesis of Proteins (Pergamon Press, Oxford, London, New-York & Paris, 1961), p 122. 5. Honig, B., Ray, A., and Levinthal, C., Proc. Natl. Acad. Sci. USA 73, 1974-1978 (1976). 6. Honig B., J. Mol. Biol. 293, 283-293 (1999). 7. Dyson, H. J., Cross, K. J., Houghten, R. A., Wilson, I. A., Wright, P. E., and Lerner, R. A. Nature 318, 480-483 (1985); Dyson, H. J., Rance, M., Houghten R. A., Wright, P. E., and Lerner, R. A., J. Mol. Biol. 201, 201-217 (1988); Wright, P. E., Dyson, H. J., and Lerner, R. A., Biochemistry 27, 7167-7175 (1988). 8. Itzaki, L. S., Otzen D. E., and Fersht, A. R., J. Mol. Biol. 254, 260-288 (1995). 9. Berezovsky, I. N., Grosberg, A. Y., Trifonov, E. N., FEBS Lett. 466, 283-286 (2000); Berezovsky, I. N., and Trifonov, E. N., J. Mol. Biol. 307, 1419-1426 (2001); J. Biomol. Struct. Dyn. 20, 5-6 (2002). 10. Sali, A., Shakhnovich, E., and Karplus, M., J. Mol. Biol. 235, 1614-1636 (1994); Gutin, A. M., Abkevich, V. I., and Shakhnovich, E. I., Proc. Natl. Acad. Sci. USA 92, 1282-1286 (1995). 11. Bryngelson, J. D., Onuchic, J. N., Socci, N. D., and Wolynes P. G., Proteins 21, 167-195 (1995). 12. Bogatyreva, N. S., and Finkelstein, A. V., Protein Eng. 14, 521-523 (2001).

329 What is Paradoxical about Levinthal Paradox?

Date Received: October 29, 2002

Communicated by the Editor Ramaswamy H Sarma

Journal of Biomolecular Structure & Dynamics, ISSN 0739-1102 Volume 20, Issue Number 3, (2002) Adenine Press (2002)

An Opinion Piece: Conversation on Levinthal Paradox & Protein Folding #6

Protein Folding: Where is the Paradox?

http://www.jbsdonline.com
Abstract In this contribution we shall try to argue that no folding scenario be it hierachical, nonhierarchical, nucleation, etc. needs to be invoked to solve Levinthals paradox: It fails on its own grounds.

Ariel Fernndez1,* Alejandro Belinky2 Mara de las Mercedes Boland3

1Institute

for Biophysical Dynamics

The University of Chicago Chicago, IL 60637

2Finance

Since we could not find a satisfactory definition of paradox, we decided to coin one of our own: A paradox is a logically consistent construction which sprouts from a false premise but one which is not too obviously so, and produces a logical conclusion which is very ostensibly false, causing surprise and forcing us to revise the starting premise. In this way, paradoxes help us to think more rigorously and be more critical of our own thoughts. Here is, we believe, where the Socratic or moral content of the paradox resides. Thus, the layman might not know that the sum of infinite rational numbers may yield a finite result, and for that reason, he might be surprised at some of Zenos conclusions, like an arrow never reaches the target. Since the layman would find the latter statement more striking than the original premise (we assume he has not been exposed to Calculus), he will be forced to think carefully about something he might have never thought about otherwise, i.e. that the sum of infinite rational numbers might be finite. And, after this thinking induced by the paradox, he might be a bit less of a layman. Not quite in the ancient tradition of Zenos paradoxes, Levinthal argued that since the number of possible conformations of a protein chain may be estimated to be exponential in the number (N) of aminoacids, the exhaustive exploration of conformation space in a finite time of biological relevance is practically impossible since it is also exponential in N. He assumed the number of possible conformations for each individual aminoacid to be small and fixed, say from 2 to 100, and that the conformations available to each individual aminoacid may be assigned constant and equal probabilities (although the equality condition may be relaxed) at all times throughout the exploration of conformation space. This premise is false: the conformations of an individual aminoacid are not equally probable in time, nor are their probabilities constant in time. Thus, the number of a-priori possible conformations of the chain might indeed be exponential in N, but this does not imply that the time to exhaustively visit all accessible conformations of the chain is also exponential in N. The probability of a conformation of an individual aminoacid within the chain depends on the geometric or structural constraints and basins of attraction the chain generates as each aminoacid picks its coordinates. Thus, the repulsive LennardJones terms in the intramolecular potential energy to name a single contribution dramatically reduce the probability of certain conformations (i.e. those producing an over-all structure at odds with excluded volume), while the attractive pairwise contributions dramatically increase the probability of other conformations.

and Economics Program

Columbia University Business School 3022 Broadway #8G New York, NY 10027
3100

Morningside Drive

New York, NY 10027

*Phone:

773 834 4782 Fax: 773 702 0439 Email: ariel@uchicago.edu

331

332 Fernndez et al.

In a nutshell: The very existence of an intramolecular potential renders the starting premise of the Levinthal paradox false. Moreover, even the idea that the protein chain exhaustively explores all conformations available along a successful folding pathway is false. Why is the thermodynamic limit even relevant to protein folding? Where does the idea that the native structure is the free energy minimum come from? To the best of our knowledge, these are baseless hypotheses. Thus, we fail to see why the Levinthal paradox attracts attention: Its premises are too obviously false and the striking conclusion it purports to reach is uncalled for. The former might not be much of a defect in a paradox, the latter certainly is.

Date Received: October 5, 2002

Communicated by the Editor Ramaswamy H Sarma