Discussion

There is no evidence for the presence of a refolding intermediate
Curvature or ”roll-over” in the re-folding arm of a chevron is good evidence for the presence of a populated folding intermediate. However, it has been shown that curvature in the refolding arm can indicate the presence of transient aggregation at low concentrations of denaturant, as is observed in FADD DD. Any curvature present due to refolding from an intermediate will be masked by this aggregation. The non-concurrence of equilibrium and kinetic data is another indication of the presence of a refolding intermediate accumulating on the pathway.30 Where a Hammond fit is used, the agreement between kinetic and equilibrium free energies is good (Supplementary Data). Moreover, the average kinetic and equilibrium m-values, taken from WT and mutants, are the same within error (1.3 ± 0.1 kcal mol–1 M–1 and 1.4 ± 0.2 kcal mol–1 M–1, respectively). Furthermore, an amplitude analysis by CD showed no evidence for a dead-time change in amplitude, which would suggest that no helical structure is formed in the dead-time of the stopped-flow experiments (data not shown). Thus, our experiments suggest that, at least above 1.5 M urea where we analyse the kinetic data, the folding of FADD DD is essentially two-state.

The transition state for folding in human FADD DD
In the TS of the human FADD DD, all three cores are partially structured. Helices H3 and H6, which are peripheral to the structure, are essentially detached from the protein. H3 is the only element of secondary structure that is completely unstructured, and H6 is only pinned to H5 via a helix-turn-helix interaction mediated through V173. In contrast, all four central elements of structure are already associated in the TS: three helices (H1, H2 and H5) contribute to central core packing, with the fourth helix (H4) pinned into the core at one end. However, H2 and H4 clearly interact with each other via residues that are packed in the B2 core. Thus, in the TS, helices H1, H2, H4 and H5 are apparently aligned in two parallel pairs (H1/H5 and H2/H4) and the pairs are packed against each other through interactions between H2 and the H1/H5 pair. Long-range packing interactions (H1 – H5, H2 – H4 and H2 – H1/H5) have a predominant role in the formation of structure in the central four-helix bundle, and in B2. Only in the formation of the B1 core do short-range side-chain interactions appear to be significant (in the formation of the helix 5-turn-helix 6 motif).
Analysis of the secondary structure propensity of FADD DD (as determined using the program AGADIR31) suggests that the level of overall helical propensity is very low; only two regions of the protein, the N-terminal region of H4 and the C-terminal region of H6, show any significant helical propensity (> 10%). Interestingly, neither of these regions is involved in forming the central topology-defining core, although the N-terminal region of H4 appears to be important in the folding of B2.

Curvature in the unfolding arm suggests consolidation of the central hydrophobic core in a late transition state
Downwards curvature in the unfolding arm can formally be caused by: (i) changes in the ground (native) state;32 (ii) movement of the transition state along a broad energy barrier;33,34 or (iii) a high-energy intermediate separating two transitions states that switch due to, for example, mutation or an increase in the concentration of denaturant.35 (See Refs 36 and 37 for detailed discussion of different explanations for curvature.) Note that in both models (ii) and (iii), mutation of a residue induces curvature when it is in a region of the protein that is more structured in a late transition state than in an early transition state. Thus, analysis of the chevron plots should, in principle, allow us to infer consolidation of the protein as it crosses the transition state barrier(s).17,38
There is no evidence for changes in the native state at high concentrations of denaturant. WT and some highly destabilised mutants show no evidence of curvature in the unfolding limbs of the chevron plots, and we see no obvious change in kinetic amplitude. To try to make some qualitative sense of our data, we examined chevron plots that had been fit to a linear equation to determine at which concentration of denaturant deviation from linearity occurred. For most mutants, the unfolding limb was either entirely linear (as in WT) or deviated from linearity only at very high concentrations of urea. A few mutants, however, show very clear deviation from linearity at concentrations of urea below 7 M: F101A and I104A in H1; L119M in H2; I129A in H3; W148F in H4; and L161A and L165A in H5 (Fig. 2, and Supplementary Data Fig. S1). These residues are shown in Fig. 4. Some mutants (L119M, I129A and W148F) have low Φ-values in the “early” transition state (determined at 2 M urea), suggesting that they are not significantly structured, and others (F101A, I104A, L161A and L165A) have medium Φ-values. The significant curvature suggests that the Φ-values of all of these mutants increase at higher concentrations of denaturant. Importantly, nearly all these residues form part of the central hydrophobic core, which is formed by the orthogonal parallel helix pairs: H1/H5 and H2/H4; in particular, they are found towards the C-terminal end of these helices (Fig. 4). In the early transition state structure, described above, the N-terminal part of this core is more structured than the C-terminal part. Thus, the central hydrophobic core appears to consolidate as the protein traverses the transition state barrier. Significantly, buried residues in these same helices, H1, H2, H4 and H5, which pack on the opposite side of these helices into the hydrophobic cores formed by the three helix bundles, do not show similar curvature.
Importantly, the experiments cannot provide information about the order of structure formation. The Φ-values do not give sufficient resolution to determine whether the entire structure folds concomitantly, or if there is an order to the packing of the helices. Future simulations might allow us to detect early events.

Comparison with other members of the DD superfamily
Clark and co-workers have examined the folding of a number of WT DDs, all of which appear to have complex folding pathways characterised by multiphasic folding and unfolding.39–42 Complex folding behaviour has been proposed, and they have suggested that this might be an intrinsic consequence of the complex all-alpha Greek key topology. This would, if true, be in stark contrast to the relatively simple, conserved folding mechanism observed in all-beta Greek key proteins, where populated intermediates are observed only in the more stable Ig-like domains.43 Our results for FADD DD demonstrate clearly that these domains do not have intrinsically complex folding pathways. FADD DD folds in a simple two-state manner with no stable kinetic or equilibrium intermediate. Nonetheless, as is true for most of these other DDs, the observed rate constant of folding is somewhat lower than might be expected for such all-alpha proteins with relatively low contact orders.44

Comparison with the all-beta Greek key Ig-like domains
The folding of FADD DD can be compared with that of the all-beta Greek key Ig-like domains studied earlier, including TNfn3,11,45 FNfn10,13 CAfn2,14 and TI I279,10 in our laboratory, and CD2d1.15 In the Ig-like domains, tertiary structure is the dominant factor influencing the folding mechanism. These domains exhibit a near-classical nucleation-condensation mechanism where long-range key residues, in the central BCEF strands, interact in the TS to set up the complicated topology; the transition state is an expanded version of the native state, with the folding nucleus involving secondary and tertiary interactions, centred around the structural core. Peripheral regions pack late.
Unlike the Ig-like domains, FADD DD has three discrete hydrophobic cores. The central core is composed of residues from the central helices H1, H2, H4 and H5; the hydrophobic residues interacting in these helices can be regarded as the structural core of this four-helix bundle. The parallel helices H1-H5 and H2-H4 have to be organised relative to each other to form the Greek key topology; this is analogous to the organisation of the anti-parallel beta strands B-E and C-F relative to each other, in the core of the Ig-like domains. In the Ig-like domains, we infer from the Φ-value analysis that the alignment of strands within the sheet and the packing of the two sheets together occurs concomitantly; these are the critical nucleating events for Ig-like domain folding. Although it is not possible to determine the order of events in the folding of the DD from the Φ-value analysis alone, we can show that helices H2 and H4 come together via the core of B2, whereas helices H1 and H5 require the formation of the central core to come together. In both Greek key structures, many long-range, tertiary interactions are involved in formation and stabilisation of the central core; however, local interactions important for helix formation are also involved in the DD. Interestingly, simulations of the TS for folding in the Ig-like protein TNfn3 suggested that the four strands form an “open horseshoe” structure, with all four central strands connected, but with the “ring” of structures incomplete.45 This is reminiscent of what we observe in FADD DD with all four central helices in contact, but where the packing of H4 into the central core is only marginal. Another similarity with the Ig-like domains is the observation that the elements of secondary structure most peripheral to the central core are relatively unimportant in formation of the TS structure. In the Ig-like domains, the two beta strands at the N- and C-termini are not involved in forming the structural core, and these fold late. In the DD fold, the N-terminal helix H1 is involved in forming the central core, but H3 and the C-terminal helix H6 are both peripheral to the structure and not involved in packing the central hydrophobic core. The Φ-value analysis suggests that these peripheral helices are either completely unfolded (H3) or attached only loosely (H6) in the TS.
Although FADD DD and TNfn3 have similar stabilities, and appear to have similar folding mechanisms, FADD DD folds significantly faster than TNfn3 (∼ 1000 s- 1 compared to 6 s- 1). This is not unexpected; FADD DD is an all alpha-helical protein with a significantly lower relative contact order than the all-beta TNfn3, and it has been shown that proteins with low contact orders generally have higher folding rate constants.44

Comparison with other helix bundles
A number of helix-bundle proteins have been studied. FADD DD differs from all of these helical proteins in terms of structure, as three distinct hydrophobic cores can be identified. Here we compare the folding of FADD DD with simple helix bundle proteins. We largely ignore the cytochrome c proteins where the haem provides essential stability to the protein,46 and the much larger globins, which have more complex hierarchical folding mechanisms.47–50
Formation of the central core of FADD involves packing of a four-helix bundle. Several other four-helix bundle proteins have been studied in detail by Φ-value analysis; two members of the ACBP family51 and apocytochrome b56252 (which are up-down helical bundles) and Im7, the homologous Im98 and the FF domain of HYPA/FBP1153 (which have three long helices plus a shorter helix). In all these cases, formation of the TS involves packing of three helices with one helix being essentially unstructured. It has been suggested that the early, obligatory stages in nucleation of folding will be formation of the interactions that are necessary to establish the overall topology.54–56 Packing of three helices in a simple four-helix bundle is sufficient to establish the topology. It seems probable that the more complex Greek key topology of the DD requires all four central elements to be assembled, as was found in the Ig-like Greek key domains.
The folding of the two three-helix bundles of FADD DD can be compared with the folding of other three-helix bundle proteins. A number of these have been studied extensively, using Φ-value analysis. In some cases, one well-formed helix is observed in the TS with other elements of structure packed against it (as, for example, in protein A27,57), whereas in others, two elements of structure come together with the third helix relatively unstructured (e.g. spectrin domains17,18 and peripheral subunit binding domains58). The three-helix bundles of FADD DD fall into this second category. Interestingly, as in the spectrin domains, the helices that are in contact are those that are separated in sequence (H1 with H5 and H2 with H4). This suggests that formation of these long-range interactions is the important step for folding these bundles.