© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
Crystal structure of human leukotriene A4 hydrolase, a bifunctional enzyme in inflammation Marjolein M.G.M. Thunnissen1, Pär Nordlund1 and Jesper Z. Haeggström2 1Department of Biochemistry, University of Stockholm, Arrhenius Laboratories A4, S-106 91 Stockholm, Sweden. 2Department of Medical Biochemistry & Biophysics, Karolinska Institutet, S-171 77 Stockholm, Sweden.
Inflammation is a major pathological characteristic of a wide array of severe endemic illnesses potentially affecting almost all tissues and organ systems of the human body. The development and maintenance of inflammation are governed by a complex network of cellular and soluble factors. Among these are the eicosanoids, a class of structurally related paracrine hormones, derived from the metabolism of arachidonic acid, that includes the prostaglandins, the leukotrienes, and the lipoxins. Leukotriene (LT) A4 hydrolase (LTA4H)1 catalyzes the final, rate limiting step in the biosynthesis of LTB4, a potent lipid chemoattractant involved in inflammation2,3, immune responses4, host defense against infection5, platelet activating factor (PAF) induced shock2,6 and lipid homeostasis7. As a bifunctional zinc metalloenzyme, LTA4H integrates a sophisticated epoxide hydrolase activity specific for the fatty acid substrate LTA4 (ref. 8) with an anion dependent aminopeptidase activity in a common active center9. The epoxide hydrolase reaction — that is, the conversion of LTA4 into LTB4 — is unique in that the stereospecific introduction of oxygen occurs at a site distant from the epoxide moiety and it has been suggested that it proceeds via a delocalized carbocation intermediate. The structural requirements for controlling such a reactive species in the active site and for obtaining sufficient reaction specificity are of fundamental biochemical as well as enzymological interest. Here we report the high resolution crystal structure of human LTA4H in complex with the competitive inhibitor bestatin, which provides detailed insight into its mechanism of catalysis and may help in the design of specific anti-inflammatory agents.
Leukotriene (LT) A4 hydrolase/aminopeptidase (LTA4H) is a bifunctional zinc enzyme that catalyzes the biosynthesis of LTB4, a potent lipid chemoattractant involved in inflammation, immune responses, host defense against infection, and PAF-induced shock. The high resolution crystal structure of LTA4H in complex with the competitive inhibitor bestatin reveals a protein folded into three domains that together create a deep cleft harboring the catalytic Zn2+ site. A bent and narrow pocket, shaped to accommodate the substrate LTA4, constitutes a highly confined binding region that can be targeted in the design of specific antiinflammatory agents. Moreover, the structure of the catalytic domain is very similar to that of thermolysin and provides detailed insight into mechanisms of catalysis, in Overall structure description particular the chemical strategy for the unique epoxide Human LTA4H was crystallized in the presence of the competitive hydrolase reaction that generates LTB4. aminopeptidase inhibitor bestatin10 and ytterbium, which enabled
a
b
c Fig. 1 Overall and domain structure of LTA4H. a, Ribbon diagram of the tertiary structure of LTA4H. The N-terminal domain is colored blue (residues 1–207), the catalytic domain green (residues 208–450) and the C-terminal domain red (residues 461–610). A loop containing a highly conserved Pro-rich motif p451-G-φ-P-P-x-k-P-xy460 (φ, hydrophobic residues Phe, Tyr, Trp, Ile, Leu, Val, Met and Ala; capital letter, identical amino acids; small letter, conserved in chemistry) is shown in yellow. The figure was created using MolScript27, Glr (L. Esser and J. Deisenhofer, pers. comm.) and POVRay (http://www.povray.org). b, Stereo view of the superposition of the Cα trace of the catalytic domain (red) on thermolysin (blue). c, Stereo view of a 2Fo - Fc electron density map for the active site, including bestatin, contoured at 1.1 σ.
nature structural biology • volume 8 number 2 • february 2001
131
© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
a
b
Fig. 2 Bestatin binding in LTA4H. a, The Zn2+ site and binding of bestatin (blue) in LTA4H. Hydrogen bonds to bestatin are depicted as thin green lines and the Zn–ligand interactions are depicted as thin black lines. b, Schematic overview of bestatin binding, with distances given in Å.
structure determination by multiple wavelength anomalous dispersion (MAD) methods (Table 1). A full model comprising amino acids 1–610 was refined at 1.95 Å resolution to an R-factor of 18.5%. The LTA4H molecule is folded into three domains (N-terminal, catalytic, and C-terminal domains) packed in a flat triangular arrangement with dimensions of ∼85 × 65 × 50 Å3 (Fig. 1a). Although the three domains pack closely and make contact with each other, a deep cleft is created between them. The N- and C-terminal domains Despite the absence of any significant sequence identity, the LTA4H N-terminal domain and the membrane associated protein bacteriochlorophyll (Bchl) a11 share significant structural homology: 111 Cα positions out of 209 in LTA4H and 357 in Bcl a are equivalent (root mean square (r.m.s.) deviation of 2.82 Å). The N-terminal domain is composed of one large sevenstranded mixed β-sheet and two smaller β-sheets that pack on opposite edges of the same face of the seven-stranded sheet so that a kind of envelope is formed, with a large concave surface exposed to solvent (Fig. 1a). The C-terminal domain is formed from two layers of parallel α-helices, five in the inner layer and four in the outer, arranged in an antiparallel manner, with perpendicular loops containing short helical segments on top (Fig. 1a). This unusual coil of helices resembles armadillo repeats or HEAT motif regions12, which are generally larger and create superhelical structures ideally suited for protein–protein interactions. Between the C-terminal and catalytic domain there is an exposed proline-rich loop that resembles an SH3 domain recognition sequence (Fig. 1a). The catalytic domain The catalytic domain is surprisingly similar in structure to the bacterial protease thermolysin13 (Fig. 1b). Although their shared sequence identity is essentially confined to the zinc binding motif (HEXXH-X18-E), their structural homology extends over the whole domain, with 146 Cα positions equivalent (r.m.s. deviation of 1.94 Å). Like thermolysin, this domain consists of two lobes, one mainly α-helical and one mixed α/β. Since this domain contains only 245 amino acids, some truncations are evident relative to thermolysin (which contains 314 residues), especially in the α/β lobe in which the N-terminal extended 132
β structure of thermolysin is truncated to a mixed five-stranded β-sheet in LTA4 hydrolase. The differences in the α-lobe are smaller; the long meandering loop 181–221 of thermolysin has been replaced by an α-helix, and the β-hairpin (residues 245–258) has been deleted in LTA4 hydrolase. The Zn2+ site is between the two lobes and the metal is coordinated by His 295, His 299, one carboxylic oxygen of Glu 318, and the carbonyl and hydroxyl oxygens of bestatin, forming a square based pyramid. Glu 296 and Tyr 383, two residues implicated in the peptide cleaving activity, are located near the Zn2+ and interact with bestatin (Fig. 2a). Although the Zn2+ binding site is formed by residues from the catalytic domain only, bestatin makes interactions with residues from all three domains. The main interactions are made through the carbonyl and hydroxyl oxygens to the Zn2+, as well as many hydrophobic interactions between the phenyl moiety and the protein (Fig. 2b). The other end of bestatin points towards the solvent and makes fewer interactions, whereas the backbone polar atoms are all involved in hydrogen bonds. Putative substrate binding pocket A hydrophobic cavity ∼6–7 Å wide stretches 15 Å deeper into the protein behind the pocket occupied by the phenyl ring of bestatin (Fig. 3a). Most of the residues lining the pocket are conserved and belong to the active site peptide K21, a 21 residue segment (Leu 365–Lys 385) identified by Lys-specific peptide mapping of suicide inactivated LTA4H14 (Fig. 3b). Gln 134, Asp 375 and the hydroxyl of Tyr 267 cluster together to form a hydrophilic patch in the cavity. If LTA4 is modeled into this pocket such that the 5,6-epoxide moiety is bound to Zn2+, then C7–C20 of the fatty acid backbone of LTA4 fit snugly into the deeper cavity, adopting a bent conformation (Fig. 3b). Furthermore, the C1 carboxylate could make direct electrostatic interactions with the positive charges of the conserved Arg 563 and Lys 565, in agreement with the fact that the free carboxylic acid of LTA4 is required for catalysis. Mechanistic implications During the conversion of LTA4 to LTB4, LTA4H acts on a fatty acid containing an unstable allylic trans-epoxide and catalyzes the intriguing chemistry in which the stereospecific introducnature structural biology • volume 8 number 2 • february 2001
© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
a
b
Fig. 3 The putative LTA4 binding cavity. a, Transparent surface representation of the pocket (in green) with the bestatin molecule shown in red and the Zn2+ as a CPK sphere. The cavity was calculated using the program WHATIF28. The protein is depicted in blue. b, Schematic representation of the proposed binding of LTA4 into the cavity. Strictly conserved residues are enclosed in black boxes while conservative mutations are in gray boxes.
tion of water occurs distant (C12) from the epoxide ring (C5/C6) and a thermodynamically unfavorable cis double bond is introduced at ∆6 (between C6 and C7) in LTB4. The resulting 12R-hydroxyl group and the 6-cis double bond are both critical for the biological activity of LTB4. This unique epoxide hydrolase reaction requires the prosthetic zinc and appears to follow an SN1 mechanism involving an unstable carbocation intermediate15. If LTA4 is modeled into the putative binding pocket with Arg 563 and Lys 565 forming the carboxylate recognition site, the catalytic zinc is proximal to the labile allylic epoxide, suggesting that Zn2+ acts as a weak Lewis acid to activate and open the epoxide ring (Fig. 4). A carbocation would thus be generated whose charge would be delocalized over the conjugated triene system (C6 to C12), leaving the planar sp2 hybridized C12 open for nucleophilic attack from either side of the molecule. In this model, the conserved Asp 375 is ideally positioned to polarize a water molecule for attack at C12 and could thus control the positional and stereospecific insertion of the 12R-hydroxyl group in LTB4. In the structure, a water molecule is positioned at hydrogen bonding distance from Asp 375 (Fig. 2a,b). The shape and curvature of the LTA4 binding cavity also suggest the chemical strategy by which the 6-cis double bond in LTB4 is created. Since there is free rotation between C6 and C7 of LTA4, the enzyme may keep this bond in a ‘pro-cis’ configuration in the transition state, which would promote the formation of a cis double bond from the carbocation intermediate (Fig. 4). The entire modeled LTA4 molecule would then adopt a bent shape that fits very well with the architecture of the binding pocket (Fig. 3b). In addition, Tyr 378 and Tyr 383 are hydrogen bonded to each other and in our model they are in close contact with the conjugated triene system just at the angle of the L-shaped binding cavity (Fig. 3a,b). Their positions would be ideal for assisting optimal substrate alignment and promoting a specific double bond configuration. Thus, the critical double bond geometry at ∆6 in LTB4 seems to be controlled by the exact binding conformation of LTA4 at the active site. nature structural biology • volume 8 number 2 • february 2001
Tyr 378 has been identified as the major residue that binds LTA4 covalently during suicide inactivation of LTA4H16, a selfregulatory side reaction that limits the formation of the chemotactic LTB4. In our model of substrate binding, the position of Tyr 378 in the LTA4 binding pocket is near the C6 atom of the modeled LTA4. One may thus speculate that the phenolic hydroxyl can attack the carbocation intermediate to possibly form an ether adduct at C6 (compare Fig. 4). The aminopeptidase activity The aminopeptidase reaction catalyzed by LTA4H involves Glu 296 and Tyr 383, presumably as general base9 and proton donor17, respectively; the equivalent residues in thermolysin are Glu 143 and His 231 (ref. 13). In the LTA4H structure, the positions of Glu 296 and Tyr 383 as well as their interactions with bestatin agree well with these postulated functions. The position of the conserved Glu 271 suggests that it may contribute to the exopeptidase specificity, as discussed for the corresponding Glu 350 in aminopeptidase N18. Based on mutational analysis and the proteolytic mechanism of thermolysin, the catalytic mechanism of aminopeptidase N was proposed to involve a pentavalent transition state in which the free α-amino group of the substrate interacts with Glu 350. In the structure of LTA4H, the free amino group of bestatin makes a similar interaction with Glu 271 (Fig. 2a,b). Furthermore, in LTA4H, Glu 271, Gln 136 and the N-terminal domain fill out the space into which the upstream peptide moiety (towards the N-terminus) of a substrate binds in thermolysin. These residues may thus contribute to the exopeptidase function of LTA4H, in contrast to the endopeptidase function displayed by thermolysin. Based on its zinc signature and proteolytic activity, LTA4H has been classified as a metallopeptidase of the M1 family19. From sequence alignments and information provided by the structure of LTA4H, it seems that members of this family of diverse enzymes share a highly conserved catalytic domain that includes part of the N-terminal domain, as we see it in LTA4H, as well as the thermolysin-like domain. In contrast, there is no 133
© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
Fig. 4 Proposed reaction mechanism for the epoxide hydrolase activity of LTA4H. The carboxylate of LTA4 is bound to Arg 563 and Lys 565. The catalytic Zn2+ acts as a Lewis acid and activates the epoxide to form a carbocation intermediate according to an SN1 reaction. Water is added at C12 in a stereospecific manner directed by Asp 375. The double bond geometry is controlled by the binding conformation of LTA4.
protein in the leukotriene cascade and important structural features of its active center most likely pertain to other enzymes of this pathway, as well as to a variety of metalloproteases related to LTA4H. Moreover, we have obtained insight into a unique active site that harbors two distinct activities, each using zinc but different amino acids, to hydrolyze lipid and peptide substrates respectively. The chemical properties of this active site and its location deeply buried in the protein provide excellent opportunities for the structure based design of anti-inflammatory drugs. Methods
homology for residues in the C-terminal domain, which we thus believe is unique to LTA4H. We propose that all metallopeptidases of class M1 function in a manner similar to thermolysin. Their common architecture would thus have been preserved over a large evolutionary distance to serve as a catalyst in a variety of metal assisted proteolytic reactions. In conclusion, the crystal structure of the bifunctional LTA4H reveals a protein folded into three domains with a central catalytic domain structurally similar to thermolysin. The architecture of the putative active site allows us to propose a mechanism for the synthesis of LTB4, a proinflammatory and immune modulating lipid mediator. It is the first structure of a
Protein purification and crystallization. Human recombinant LTA4H was expressed in Escherichia coli and purified to homogeneity as described17,20. It was further purified by chromatography on hydroxyapatite (TSKgel HA-1000, Tosohaas) followed by anion-exchange (Mono-Q HR5/5). Plate-like crystals could be obtained by using liquid–liquid diffusion in capillaries; 5 µl of 28% (w/v) PEG 8000, 0.1 mM Na acetate, 0.1 mM imidazole buffer, pH 6.8, and 5 mM YbCl3 was injected into the bottom of a melting point capillary and an equal volume of LTA4H (5 mg ml-1) in 10 mM Tris-HCl, pH 8, supplemented with 1 mM bestatin, was layered on top. Crystals with an average size of 0.6 mm × 0.4 mm × 0.05 mm grew in 3–4 weeks at 22 °C. The crystals belonged to space group P21212 with cell dimensions a = 67.6 Å, b = 133.5 Å, c = 83.4 Å, α = β = γ = 90° at 100 K.
Data collection and structure determination. For data collection, crystals were soaked in 15% (w/v) PEG 8000, 50 mM Na acetate, 50 mM imidazole buffer, pH 6.8, 2.5 mM YbCl3 and 25% (v/v) glycerol. Three data sets, peak (PK), point of inflection (PI) and remote (RM) were collected at the ytterbium LIII edge (λ = 1.3862 Å) from a single crystal at beamline BM14 of the European Synchrotron Radiation Facility (ESRF), Grenoble, France. Statistics on data collection and quality are given in Table 1. Data were integrated using the program Denzo, scaled to each other using Scalepack21 and further Table 1 Data collection statistics data analysis was performed using pro22 Crystal 1 Crystal 2 grams from the CCP4 package . Heavy atom parameters were refined using PK PI RM MLPHARE23 and SHARP24 (for details see Diffraction limit (Å) 2.5 2.5 2.15 1.95 Table 1). The final figure of merit was 0.57 λ (Å) 1.3838 1.3842 0.8856 0.992 to 2.15 Å. Phase information was further Rmerge (I) (%)1 3.0 3.0 3.0 4.3 improved at 2.15 Å by solvent flipping using SOLOMON25. All model building was Ranom (I) (%) 8.1 4.7 3.3 4.4 performed using QUANTA (Molecular Completeness (%) 97.7 97.7 97.8 98.2 Simulations, Inc.). Data to 1.95 Å from a Mean I / σ (I) 6.59 10.74 13.16 13.3 second crystal were used for refinement Multiplicity of observation 3.3 6.2 3.8 4.7 with TNT26 during which a Zn2+, bestatin, an Riso (F) (%) – 2.4 7.0 acetate, an imidazole molecule and 550 Phasing power water molecules were identified. The Rfree was 24.7% and the working R-factor was Isomorphous (centric / acentric) –/– 3.7 / 5.4 3.5 / 5.4 18.8% for all data between 25 and 1.95 Å. Anomalous (centric / acentric) – / 3.2 – / 4.7 – / 3.8 In a final round of refinement, all data Rcullis between 25 and 1.95 Å were included, 2 Isomorphous (centric / acentric) –/– 0.37 / 0.38 0.33 / 0.36 yielding a final R-factor of 18.5%. Most of 2 Anomalous (centric / acentric) – / 0.60 – / 0.40 – / 0.53 the model is in good density except for the loop encompassing residues 179–184. The 1R merge(I) = Σj,k| Ijk - j| / Σj,kIjk, where Ij,k are the k individual observations of each reflection j and model has good stereochemistry (r.m.s devij is the value after weighted averaging. ation from ideal geometry in bonds of 2R cullis = Σεiso / Σ∆iso for acentric reflections and Σεano / Σ∆Bijvoet for anomalous differences, where εiso and εano are the isomorphous and anomalous lack of closure, respectively, and ∆iso is the isomor- 0.010 Å and in angles of 2.2°) and 91.7% of the residues lie in the most favored part of phous difference and ∆Bijvoet is the Bijvoet difference. the Ramachandran plot.
134
nature structural biology • volume 8 number 2 • february 2001
© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
Coordinates. Coordinates have been deposited at the PDB (accession code 1HS6).
Acknowledgments We thank A. Thompson at BM14 and D. Logan for help during data collection and setting up SHARP refinement, M. Andberg for the plasmid pT3-MB4 and E. Ohlsson for technical assistance, as well as A. Wetterholm and B. Samuelsson for helpful discussions and advice. We also would like to thank personnel at beamline X25 of the NSLS, Brookhaven, and I711 of MAX-Lab, Lund, for preliminary data not used in this article. The work was funded by the Swedish Natural Sciences Research Council, the Swedish Medical Research Council, The European Union, and Konung Gustav V:s 80-Årsfond.
Correspondence should be addressed to M.M.G.M.T. email:
[email protected]
7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.
Received 25 August, 2000; accepted 30 November, 2000. 22. 1. Samuelsson, B., Dahlén, S.-E., Lindgren, J.-Å., Rouzer, C.A. & Serhan, C.N. Science 237, 1171–1176 (1987). 2. Chen, X.S., Sheller, J.R., Johnson, E.N. & Funk, C.D. Nature 372, 179–182 (1994). 3. Yokomizo, T., Izumi, T., Chang, K., Takuwa, Y. & Shimuzu, T. Nature 387, 620–624 (1997). 4. Griffiths, R.J. et al. J. Exp. Med. 185, 1123–1129 (1997). 5. Bailie, M.B. et al. J. Immunol. 157, 5221–5224 (1996). 6. Byrum, R.S., Goulet, J.L., Snouwaert, J.N., Griffiths, R.J. & Koller, B.H. J. Immunol. 163, 6810–6819 (1999).
A collapsed state functions to self-chaperone RNA folding into a native ribonucleoprotein complex Amy E. Webb and Kevin M. Weeks Department of Chemistry, University of North Carolina, Chapel Hill, North Carolina 27599-3290, USA.
Most large RNAs achieve their active, native structures only as complexes with one or more cofactor proteins. By varying the Mg2+ concentration, the catalytic core of the bI5 group I intron RNA can be manipulated into one of three states, expanded, collapsed or native, or into balanced equilibria between these states. Under near-physiological conditions, the bI5 RNA folds rapidly to a collapsed but non-native state. Hydroxyl radical footprinting demonstrates that assembly with the CBP2 protein cofactor chases the RNA from the collapsed state to the native state. In contrast, CBP2 also binds to the RNA in the expanded state to form many non-native interactions. This structural picture is reinforced by functional splicing experiments showing that RNA in an expanded state forms a non-productive, kinetically trapped complex with CBP2. Thus, rapid folding to the collapsed state functions to self-chaperone bI5 RNA folding by preventing premature interaction with its protein cofactor. This productive, self-chaperoning role for RNA collapsed states may be especially important to avert misassembly of large multi-component RNA–protein machines in the cell. Prior to assembly with protein cofactors, many cellular RNAs do not fold into their active, native structures. Native RNA structure is acquired during a dynamic process1 of structural biogenesis that includes formation and rearrangement2–4 of RNA helices, consolidation of RNA tertiary structure5 and nature structural biology • volume 8 number 2 • february 2001
23. 24. 25. 26. 27. 28.
Devchand, P.R. et al. Nature 384, 39–43 (1996). Corey, E.J. Experientia 38, 1259–1381 (1982). Wetterholm, A. et al. Proc. Natl. Acad. Sci. USA 89, 9141–9145 (1992). Örning, L. et al. J. Biol. Chem. 266, 16507–16511 (1991). Matthews, B.W., Fenna, R.E., Bolognesi, M.C., Schmid, M.F. & Olson, J.M. J. Mol. Biol. 131, 259–285 (1979). Groves, M.R. & Barford, D. Curr. Opin. Struct. Biol. 9, 383–389 (1999). Holmes, M.A. & Matthews, B.W. J. Mol. Biol. 160, 623–639 (1982). Mueller, M.J. et al. Proc. Natl. Acad. Sci. USA 92, 8383–8387 (1995). Blomster Andberg, M., Hamberg, M. & Haeggstrom, J.Z. J. Biol. Chem. 272, 23057–23063 (1997). Mueller, M.J. et al. Proc. Natl. Acad. Sci. USA 93, 5931–5935 (1996). Blomster, M., Wetterholm, A., Mueller, M.J. & Haeggström, J.Z. Eur. J. Biochem. 231, 528–534 (1995). Luciani, N. et al. Biochemistry 37, 686–692 (1998). Barret, A.J., Rawlings, N.D. & Woessner, J.F. In Handbook of proteolytic enzymes. (eds Barret, A.J., Rawlings, N.D. & Woessner, J.F.) 994–996 (Academic Press, London, San Diego; 1998). Wetterholm, A. et al. Biochim. Biophys. Acta 1080, 96–102 (1991). Otwinowski, Z. In Data collection and processing. Proceedings of the CCP4 study weekend. 56–62 (SERC Daresbury Laboratory, Warrington, UK; 1993). Collaborative Computing Project Number 4. Acta Crystallogr. D 50, 760–763 (1994). Otwinowski, Z. In Isomorphous replacement anomalous scattering. Proceedings of the CCP4 study weekend. 80–85 (SERC Daresbury Laboratory, Warrington, UK; 1991). de La Fortelle, E. & Bricogne, G. Methods Enzymol. 276, 472–494 (1997). Abrahams, J.P. & Leslie, A.G.W. Acta Crystallogr. D 52, 30–42 (1996). Tronrud, D.E. Acta Crystallogr. A 48, 912–916 (1992). Kraulis, P.J. J. Appl. Crystallogr. 24, 946–950 (1991). Vriend, G. J. Mol. Graphics 8, 52–56 (1990).
assembly with cofactor proteins6–8. Thus, protein cofactors are likely to encounter and bind to cellular RNAs at multiple stages of structural maturity. What roles do thermodynamically favored, but non-native RNA folding intermediates play in native ribonucleoprotein assembly? Recent work on the bI5 (ref. 9) and Tetrahymena10 group I intron RNAs and on the catalytic domain of RNase P11 emphasizes that large RNAs can readily form collapsed but non-native structures. In these cases, the collapsed state is ∼10–15% larger than the native state and more compact than expanded states formed in the absence of divalent ions. Identification of a collapsed, non-native RNA folding state raises the immediate and previously unresolved question of whether this intermediate plays any productive role in the assembly of a native RNA or RNA–protein complex. The relevance of non-native states that fold slowly to the native state for both protein12–14 and RNA15–17 folding reactions has been widely discussed. A productive collapsed intermediate would lie on the pathway for formation of the native state (Fig. 1a). If off-pathway intermediates become populated, overall folding can be slow, limited by escape from (or rearrangement of) these kinetically trapped states in parallel pathways on the folding landscape (Fig. 1a). In the case of the Tetrahymena group I intron, a large fraction of molecules in the collapsed state is in a kinetic trap that rearranges to form the catalytically active native state10,15,18. In contrast, the collapsed state identified for the bI5 group I intron is not a kinetic trap. The observed slow folding of bI5 RNA from the collapsed state is characterized by a negligible activation enthalpy and is not accelerated by the addition of the denaturant urea9. Thus, this slow folding to the native state from the collapsed state is consistent with a relatively fluid folding transition state. Like most cellular RNAs, the bI5 RNA achieves its biologically functional structure under physiological ion conditions only upon assembly with an obligatory19 protein cofactor, cytochrome b precursor messenger RNA (pre-mRNA) processing protein 2 (CBP2). In this and many other cases, including assembly of the ribosome and pre-mRNA processing complexes, RNA folding 135