Acceleration of protein folding by four orders of magnitude through a single amino acid substitution

Cis prolyl peptide bonds are conserved structural elements in numerous protein families, although their formation is energetically unfavorable, intrinsically slow and often rate-limiting for folding. Here we investigate the reasons underlying the conservation of the cis proline that is diagnostic for the fold of thioredoxin-like thiol-disulfide oxidoreductases. We show that replacement of the conserved cis proline in thioredoxin by alanine can accelerate spontaneous folding to the native, thermodynamically most stable state by more than four orders of magnitude. However, the resulting trans alanine bond leads to small structural rearrangements around the active site that impair the function of thioredoxin as catalyst of electron transfer reactions by more than 100-fold. Our data provide evidence for the absence of a strong evolutionary pressure to achieve intrinsically fast folding rates, which is most likely a consequence of proline isomerases and molecular chaperones that guarantee high in vivo folding rates and yields.

one of the other four prolines in the nonnative cis conformation. Both intermediates I M and I R react to the intermediate I S with all prolines in the trans conformation, and the slow reaction of I S to N (half-live: 280 s) involves the trans-to-cis isomerization of Pro76 as the common rate-limiting folding step for the majority of Trx molecules (Fig. 2). To study the coupling between folding and the trans-to-cis isomerization of a single X-Pro peptide bond in a best-defined experimental set-up, we used the single-proline variant Trx1P, in which the four trans prolines of Trx were replaced by alanines. Trx1P shows WT-like reductase reactivity, and its X-ray structure is virtually identical to that of Trx WT, with the single Pro76 in cis 18 . We show that 95% of the Trx1P molecules accumulate a highly stable intermediate (I trans ) with Pro76 in trans, which reaches N only extremely slowly. Replacement of Pro76 by alanine accelerated folding of the resulting, proline-free variant Trx0P 66'000-fold. The X-ray structure of Trx0P revealed a trans Ala76 bond and only small structural rearrangements relative to Trx1P, but the activity of Trx0P as electron transfer catalyst dropped about 100-fold relative to Trx WT. Our results show that optimal interaction with in vivo electron donors and acceptor substrates dominated over fast folding rates in the molecular evolution of thioredoxins.

Results
The X-ray structure of Trx0P shows how the Trx fold can adapt to a trans 75-76 peptide bond. To test the ability of the Trx fold of tolerating a trans peptide bond at the position of the conserved Pro76, the proline-free variant Trx0P was produced as soluble protein in E. coli, purified with high yields and crystallized. The X-ray structure of the oxidized form of Trx0P was solved at 1.65 Å resolution by molecular replacement using the structure of oxidized Trx1P (Trx1P ox ) (PDB ID 4HU7) 18 as search model. Data collection and refinement parameters are summarized in Supplementary Table 1. Overall, the structure of Trx0P ox is highly similar to that of Trx WT ox and Trx1P ox (RMSDs of 0.724-0.915 Supplementary Table 2, Supplementary Fig. 1a, Supplementary Fig. 2) and shows the classical Trx fold with all regular secondary structure elements preserved (Fig. 1b). However, in contrast to Trx WT and Trx1P, the 75-76 peptide bond in Trx0P (Ile-Ala) adopts the trans conformation.
Notably, the 180° rotation around the 75-76 peptide bond in Trx0P relative to Trx1P is compensated by a structural rearrangement of the hexapeptide segment 70-75 N-terminal to the trans Ala76, which is part of the irregular secondary structure preceding β -strand 4. The position of β -strand 4 (residues 77-82) directly C-terminal to residue 76 is identical in the structures of Trx1P and Trx0P and an essential element of the central β -sheet of the thioredoxin fold, providing a plausible explanation for the fact that the polypeptide main chain around residue 76 only changed its conformation in the segment N-terminal to residue 76 in the Trx0P structure ( Fig. 1b-e). The trans conformation of the 75-76 bond in Trx0P also translates into a slight, but significant conformational change in the polypeptide segment 30-35 harboring the active-site disulfide (Cys32 and Cys35), and structural rearrangements in the segments 39-41 and 92-95 (Fig. 1d). The overall r.m.s. deviation in the main-chain atoms (C α ) between Trx1P and Trx0P is 0.914 Å (Fig. 1c,d), while it reaches values above 2.5 Å for R73, G74, and I75 next to the Pro76Ala replacement. The movement of the loop segment 70-75 likely causes a loss of van der Waals contacts with the active site disulfide, as Ile75 is no longer packed against the segment harboring the active site as a consequence of its 180° side chain flip ( Fig. 1d and Supplementary Fig. 1a). An analogous variant of the periplasmic, thioredoxin-like oxidoreductase DsbA ( Supplementary Fig. 1b) in which the conserved cis Pro151 was replaced by Ala, also resulted in formation of a trans peptide bond (Val150-Ala151) 19 . In contrast to Trx0P, however, the structural rearrangements accommodating the new trans peptide bond were even smaller and limited to the residues 149-152 ( Supplementary Fig. 1b). Together, the structures of Trx0P and DsbA (Pro151Ala) reveal a surprisingly high tolerance of the Trx fold for a trans peptide bond at the position of its conserved cis proline.
The trans peptide bond 75-76 in Trx0P leads to a dramatic loss in enzymatic activity. To characterize the functional consequences of the trans 75-76 peptide bond in Trx0P, we first compared the thermodynamic stabilities of the oxidized and reduced forms of Trx0P, Trx1P, and Trx WT at pH 7.0, 25 °C and a ionic strength (I) of 20 mM with guanidinium chloride (GdmCl)-induced unfolding and refolding transitions. All transitions proved to be fully reversible and were evaluated according to the two-state model of protein folding 20 (Fig. 3a, Table 1). While the oxidized form of Trx WT was 12.0 ± 0.3 kJ mol −1 more stable than the reduced form, with a shift in the transition midpoint by 0.74 M GdmCl to higher GdmCl concentration, both redox forms of Trx1P showed essentially the same stability and transition midpoints (Fig. 3a, Table 1). Notably, the situation was reversed for Trx0P relative to Trx WT, where the reduced form was 7.6 ± 0.4 kJ mol −1 more stable than the oxidized form, showing an increase in the transition midpoint by 0.79 M GdmCl relative to Trx0P ox . The more stable oxidized form is a hallmark of Trx WT, which shows a highly reducing redox potential of − 270 mV 21,22 . The formal thermodynamic linkage of the stability difference between the oxidized and reduced form of thioredoxin-like disulfide oxidoreductases with their disulfide exchange equilibrium constants with thiol/ disulfide substrates [23][24][25][26] predicted that Trx1P is more oxidizing than Trx WT, and that Trx0P is more oxidizing than Trx1P. This was indeed confirmed when we measured the redox potentials of Trx0P and Trx1P under identical conditions (pH 7.0, 25 °C, I = 20 mM) by equilibrium titration with mixtures of oxidized and reduced glutathione (GSSG and GSH, respectively). We determined the redox potentials of Trx0P and Trx1P to be − 229 mV and − 240 mV, respectively (Fig. 3b).
Next, we compared the activities of Trx WT, Trx1P and Trx0P as substrates of thioredoxin reductase (TrxR) (Supplementary Table 3, Fig. 3c). All three Trx variants were reduced with very similar k cat values (~10 s −1 ), but Trx0P showed a 6-fold and 8-fold increased K M value compared to Trx WT and Trx1P, respectively. The small structural changes linked to the trans 75-76 bond thus made Trx0P a significantly worse substrate of TrxR. A similar result had been obtained when the cis Pro76 in Trx WT was replaced by Ala 17 . The results indicate that TrxR recognizes the WT-like tertiary structure of Trx with very high specificity. Trx0P also showed a dramatically reduced activity as catalyst of the reduction of insulin disulfides by dithiothreitol (DTT) 27 , which was decreased 17-and 55-fold relative to Trx WT and Trx1P, respectively (Fig. 2d, Supplementary Table 3, Supplementary Fig. 3a). This was clearly due to the slow reduction of insulin disulfides by reduced Trx0P, and not caused by slower regeneration of reduced Trx0P by DTT, as DTT reduced Trx0P even faster than Trx WT (Supplementary Table 3). The diminished reductase activity of Trx0P also agrees qualitatively with the more oxidizing redox potential of Trx0P relative to Trx WT and Trx1P. On the whole, despite the comparably small structural differences to Trx WT and Trx1P, the price to be paid for the trans 75-76 bond in Trx0P is strikingly high: the lower activities of Trx0P as acceptor of electrons from TrxR and, in particular, as donor of electrons to protein substrates predict that the in vivo activity of Trx0P would be about 100-fold lower compared to that of Trx WT.
Trx0P folds four orders of magnitude faster than Trx1P. We next measured the folding kinetics for the oxidized forms of Trx WT, Trx1P and Trx0P at 25 °C and pH 7.0. For Trx0P unfolding and refolding was recorded by stopped-flow tryptophan fluorescence between 0.2 and 4 M GdmCl. Under all conditions, mono-exponential fluorescence traces without burst phases were obtained. Notably, a plot of the apparent rate constants of unfolding/refolding of Trx0P against GdmCl concentration (Fig. 4c) revealed a dynamic two-state folding equilibrium with a free energy of folding (Δ G 0 ) of − 17.4 ± 2.0 kJ mol −1 and a cooperativity (m eq ) of 9.56 ± 0.40 kJ mol −1 M −1 . The values are identical within experimental error to the corresponding parameters obtained with equilibrium unfolding of oxidized Trx0P (− 18.8 ± 0.9 kJ mol −1 and 9.53 ± 0.43 kJ mol −1 M −1 , respectively) ( Fig. 3a, top panel; Fig. 4c, Table 1). The recorded fluorescence change during refolding thus directly reported the formation of native molecules, with an extrapolated rate constant of folding (k F ) at zero denaturant of 7.70 ± 0.76 s −1 , corresponding to half-life of 90 ms. Figure 4a shows the folding kinetics (normalized fluorescence trace) at 0.2 M GdmCl, with k F = 3.5 s −1 and a half-life of 200 ms. The highly complex folding mechanism reported for Trx WT (Fig. 2) thus turned into the simplest possible mechanism through the replacement of all prolines by alanine.
Next, we monitored the kinetics of formation of N for Trx WT and Trx1P. Trx WT shows multiple kinetic phases during refolding in stopped-flow fluorescence experiments that however did not allow the quantification of native molecules 28 . We therefore applied interrupted refolding experiments that detect the concentration of N at any refolding time and the fractions of fast and slow refolding molecules [28][29][30] (see Supplementary Fig. 3 for details). Trx WT and Trx1P were unfolded for at least 1.  closed symbols, respectively) of the oxidized (blue circles) and reduced (red squares) proteins. Transitions were followed via the CD signal at 220 nm, fitted to the two-state model of folding (solid lines) and normalized. The deduced parameters D 1/2 and m eq with errors from the fits are given in Table 1. b: Redox potentials of Trx0P (red circles) and Trx1P (blue squares), determined via their equilibrium constants with GSH/GSSG. The fractions of reduced protein were obtained from fluorimetric data, plotted against the [GSH] 2 /[GSSG] ratio and fitted according to a disulfide exchange equilibrium (equation (3), solid lines), yielding equilibrium constants of 0.42 ± 0.05 M and 0.99 ± 0.11 M, and redox potentials (E o ') of − 0.229 ± 2 mV and − 0.240 ± 2 mV for Trx0P and Trx1P, respectively. The indicated errors result from fits according to equation (3). c: Kinetic analyses of Trx variants as substrates of thioredoxin reductase (TrxR). All variants show similar values of v max , but the K M value for Trx0P (red circles) is increased 6.0 and 8.4-fold compared to that of TrxWT (black triangles) and Trx1P (blue squares), respectively (cf. Supplementary Table 3). The indicated errors are from fits according to the Michaelis Menten equation. d: The insulin reductase activity of Trx0P (red circles) is dramatically diminished relative to Trx1P and Trx WT (blue squares and black triangles, respectively). Kinetics of insulin aggregation as a result of Trx-catalyzed reduction by DTT were followed via the increase in optical density at 600 nm. The time of aggregation onset was plotted against catalyst concentration. Data were empirically fitted to a double exponential decay to identify the Trx concentration required to reduce the time of aggregation onset from 44.1 ± 0.3 min (uncatalyzed reaction) to 15 min (dashed line). The estimated error of the deduced Trx concentrations is 10% (see Supplementary Table 3).
Scientific RepoRts | 5:11840 | DOi: 10.1038/srep11840 a fraction of 5 ± 2% fast folders. This is possibly due to residual structure affecting the cis/trans equilibrium of the Ile75-Pro76 peptide bond in the unfolded state. The major fraction of slow Trx1P folders reacted to N extremely slowly with a half-life of 99.1 ± 12.8 min. Thus, the four trans prolines in Trx WT accelerate the rate-limiting cis-to-trans isomerization of Pro76 27-fold relative to Trx1P. This may result from a significant change in the energy landscape through the replacement of the four trans prolines by trans alanines in Trx1P that even affects the late, rate-limiting folding reaction to the energetically most stable, native state 31,32 . More importantly, the results show that the single amino acid substitution Pro76Ala converting Trx1P into Trx0P accelerates folding to the thermodynamically most stable state (N) at zero denaturant by a factor of 66'000.
The folding kinetics of Trx1P at pH 7.0 and 25 °C are consistent with a simple folding mechanism in which U represents a mixture of 5% fast folders with Pro76 in cis (U cis ) that fold directly to N, and 95% slow folders with Pro76 in trans (U trans ) that collapse to a stable intermediate, I trans , with Pro76 still in trans (Fig. 5). We could reproduce the rate-limiting I trans -to-N reaction by two independent experiments within experimental error: the recovery of the activity of Trx1P as substrate of TrxR proceeded with a half-life of 79.5 ± 0.5 min, and the recovery of the tryptophan fluorescence of N (Fig. 6a) showed a half-life of 78.2 ± 0.5 min ( Supplementary Fig. 4).
In addition, we used the strong fluorescence difference between I trans and N ( Fig. 6a) to measure the activation energy of the I trans -to-N reaction (Fig. 4b), which proved to be 101.0 ± 2.8 kJ mol −1 . This compares to activation energies of 67.6 ± 3.2 kJ mol −1 for folding of Trx0P, and 62.4 ± 1.4 kJ mol −1 for the U trans -to-I trans reaction (Fig. 4b).
To verify that folding of I trans to N is indeed accompanied by the trans-to-cis isomerization of Pro76, we performed interrupted unfolding experiments in which Trx1P was rapidly unfolded by 5 M GdmCl (unfolding completed within 1 s) and incubated in 5 M GdmCl for different times (5-120 s) at 25 °C prior to refolding. The kinetics of attainment of the cis/trans equilibrium of the Ile75-Pro76 bond in unfolded Trx1P were measured via the increase in the fraction of slow folders with increasing incubation Thermodynamic parameters of GdmCl-dependent unfolding/refolding equilibria of the thioredoxin variants, and folding rates Trx WT ox and Trx1P ox involving the trans-to-cis isomerization of Pro76 *   Fig. 4a and Supplementary Fig. 3). d Kinetic m values, corresponding to the linear dependence of ln k F and ln k U on GdmCl concentration. e Equilibrium m value predicted from kinetic data with the equation m kin = (m U -m F )·RT. f The GdmCl dependence of the observed rate of folding/unfolding of I trans (Fig. 4c) Fig. S3). Trx WT (black triangles) showed 6% fast folding molecules (black dotted line), and 94% of the molecules reached N at a single rate of 3.10 ± 0.18 · 10 −3 s −1 (solid line). Trx1P (blue squares) showed 5% fast folders (blue dotted line), and the residual 95% folded very slowly with a single rate of 1.16 ± 0.15 · 10 −4 s −1 . The indicated errors are standard errors from monoexponential fits. b: Arrhenius plot for folding of Trx0P (red circles), formation of I trans from unfolded Trx1P (green triangles), and formation of native Trx1P from I trans (blue squares), yielding activation energies of 67.6 ± 3.2, 62.4 ± 1.4 and 101.0 ± 2.8 kJ mol −1 , respectively. The indicated errors are standard errors from Arrhenius fits. c: Chevron Plots showing the dependence of the apparent rate constant of unfolding/refolding (k obs ) on [GdmCl] for Trx0P (red circles) and Trx1P I trans (green triangles). Data were fitted according to a two-state model of protein folding in the case of Trx0P (solid, red line). Unfolding/refolding of I trans was evaluated according to a three-state model 55 with a high-energy on-pathway intermediate (solid, green line) (see legend to Table 1 for the deduced kinetic parameters). The unfolding branch of Trx1P is shown for comparison. time ( Supplementary Fig. 5) and yielded an apparent rate constant (k obs ) of 1.80 ± 0.35 · 10 −2 s −1 , consistent with reported rates of cis-to-trans isomerization of prolyl peptide bonds in unstructured peptides 3 . Combination with the known fractions of fast and slow folders yielded half-lifes of 40 ± 7.7 s and 768 ± 382 s for the cis-to-trans and trans-to-cis isomerization of the Ile75-Pro76 peptide bond in unfolded Trx1P, respectively ( Supplementary Fig. 5, Fig. 5).

The intermediate I trans has a highly stable tertiary structure. The rapid formation of I trans during
Trx1P refolding within the dead time of manual mixing, its high population (95% of the molecules) and in particular its extremely slow reaction to N offered the unique possibility to characterize the folding, stability and function of I trans in detail. First, we recorded the GdmCl dependence of its unfolding and refolding kinetics by stopped-flow fluorescence (Fig. 4c). The fluorescence traces monitoring the formation of I trans from unfolded Trx1P could be fitted mono-exponentially with an additional term correcting for the very slow formation of N from I trans and showed no burst phase. Assuming that fluorescence changes resulting from the small fraction of 5% fast folding molecules were negligible, we assigned the first-order rate constants (k obs ) deduced from the fast, main fluorescence change to formation of I trans . For recording the unfolding kinetics of I trans , it was first populated to > 90% by rapid dilution of Trx1P from 4.0 to 0.2 M GdmCl (< 5 min refolding time), and then unfolded again by increasing the GdmCl concentration. As native Trx1P unfolded by more than one order of magnitude slower than I trans in the range of 2-4 M GdmCl (Fig. 4c) and was populated to less than 10% under these conditions, its contribution to the fast fluorescence change during unfolding of I trans was minimal, so that all unfolded traces could be fitted mono-exponentially. Figure 4c shows the combined data on the GdmCl dependence of the apparent rate constant (k obs ) of folding/unfolding of I trans , revealing a clear curvature of the folding branch of I trans . Evaluation according to a mechanism with a high-energy on-pathway folding intermediate 33 yielded folding and unfolding rates of 7.52 ± 0.77 s −1 and 1.14 ± 0.62 · 10 −4 s −1 at zero denaturant concentration, respectively and a free energy of folding (Δ G 0 ) of − 27.5 ± 2.4 kJ mol −1 . In addition, the cooperativity of I trans folding (m kin ) of 12.7 ± 1.89 kJ mol −1 M −1 was identical within experimental error to the m eq value of 13.3 ± 0.29 for Trx1P measured by equilibrium unfolding (Table 1), demonstrating complete tertiary structure formation in I trans . Together, I trans proved to be even 8.7 kJ mol −1 more stable than Trx0P, and only 10.4 kJ mol −1 less stable than N (Table 1, Fig. 7). The tertiary structure of I trans is likely very similar to that of native Trx0P, which also has the 75-76 peptide bond in trans. Comparison of the extrapolated rate constants of unfolding and refolding (Table 1) shows that the higher stability of Trx0P compared to I trans is exclusively based on its 120-fold lower rate of spontaneous unfolding, as the proline-independent folding rates are essentially identical for I trans and Trx0P (Fig. 4c, Table 1). We obtained further, independent evidence for a Trx0P-like tertiary structure of I trans . First, the tryptophan fluorescence spectrum of I trans is more similar to that of Trx0P than to the spectrum of native Trx1P, which shows more than 2-fold lower fluorescence intensity than Trx0P (Fig. 6a). Second, the active-site disulfide bond in I trans showed even a slightly higher reactivity with the reductant dithiothreitol (DTT) than that of native Trx1P and Trx0P, but a more than 200-fold higher reactivity with DTT than the disulfide bond in unfolded, oxidized Trx1P (Fig. 6b, Supplementary Table 3). Third, like Trx0P, I trans was only weakly recognized as substrate by TrxR (Fig. 6c). Even though the catalytic parameters of I trans could not be accurately determined due to aggregation at high concentrations during refolding, its K M is similar to that of Trx0P (above 15 μ M) and clearly increased relative to the K M values of native Trx1P and Trx WT (Supplementary Table 3). Fourth, we performed fast analytical gel filtration runs showing that the apparent hydrodynamic volume of I trans was indistinguishable from that of Trx1P and Trx0P within experimental error (Fig. 6d, inset). Together with the notion that folding becomes slower with increasing stability of structured intermediates 29 , the results suggested that the stable tertiary structure of I trans is responsible for its very slow reaction to N. To test whether I trans can fold directly to N or first needs to unfold before the trans-to-cis isomerization of Pro76 and folding to N can take place, we measured the dependence of the kinetics of folding of I trans to N on denaturant concentration. Supplementary Fig. 6 shows that N formed independently of GdmCl concentration between 0.2 and 1.0 M, demonstrating that unfolding of I trans is not required for formation of N. In addition, we recorded refolding kinetics of Trx1P in the presence of a 1.67-fold molar excess of the ribosome-associated proline isomerase trigger factor (TF) from E. coli. TF is a likely catalyst of cis/ Reduction of Trx0P, Trx1P, and Trx1P I trans (2 μ M each; red, blue and green dots, respectively) by 20 μ M DTT, fitted according to second-order kinetics (solid lines) (see Supplementary Table 3 for the deduced rate constants and errors). Reduction of unfolded Trx1P with 100 μ M DTT (black dots) revealed a rate constant of 10.3 ± 1.48 M −1 s −1 ) and was thus more than 100-fold slower than reduction of native Trx1P or I trans (cf. Supplementary Table 3). c: Kinetic analysis documenting the properties of Trx0P (red circles), Trx1P (blue squares) and Trx1P I trans (green triangles) as substrate of TrxR in the presence of 20 mM GdmCl at pH 8.0. Similar to Trx0P, I trans showed an about 10-fold increase in K M relative to that of Trx1P, but its catalytic parameters could not be determined accurately due to unspecific aggregation at higher concentrations (cf. Supplementary Table 3 for the parameters and errors deduced from Michaelis Menten fits). d: I trans has a compact tertiary structure with a buried trans Ile75-Pro76 peptide bond that is inaccessible to catalysis by the PPIase trigger factor. Formation of native Trx1P from I trans (90 μ M) in the absence (black circles) and presence (red squares) of excess trigger factor (150 μ M) was monitored by the increase in TrxR substrate activity and fitted monoexponentially (solid lines). Inset: Size exclusion chromatography on Superdex 75 of I trans , native Trx1P and native Trx0P (oxidized forms), showing that I trans has same hydrodynamic volume as native Trx1P. The retention time of all proteins at ~26.5 min is about ¼ of the half-life of I trans folding (99 min), guaranteeing that only a minor fraction of I trans reacted to N during the chromatography. Retention times of molecular mass standard proteins (in kDa) are indicated on the top.
Scientific RepoRts | 5:11840 | DOi: 10.1038/srep11840 trans isomerization of prolines during folding of thioredoxin in vivo, as it encapsulates nascent polypeptide chains leaving the ribosomal exit tunnel in a cavity that is large enough to accommodate the 12 kDa protein thioredoxin 34,35 . However, no acceleration of formation of native Trx1P by TF was observed, even at high concentrations of Trx1P (90 μ M) and TF (150 μ M) (Fig. 6d). This indicated that the trans Pro76 gets rapidly buried and inaccessible for TF upon the collapse of the polypeptide to I trans and confirmed that N can be formed from I trans directly. The inability of TF to catalyze folding of thioredoxin in vitro does however not question a catalytic function of TF in vivo, because ribosome-associated TF should be able to access prolines in disordered, nascent polypeptide chains that remain folding-incompetent prior to completion of protein synthesis.

Discussion
Improved biological function is the critical driving force underlying molecular evolution of proteins 36,37 . Molecular evolution of the cytoplasmic reductase thioredoxin is a particularly interesting case because Trx not only needs to interact with its electron donor TrxR with very high efficiency (k cat /K M > 10 6 M −1 s −1 , Supplementary Table 4), but also must donate electrons specifically to numerous substrates in vivo. These multiple functions define very stringent structural prerequisites for a functional Trx fold and, in particular, for the specific conformation and local environment of its active-site cysteine pair. In accordance with this notion, we showed that a single replacement of a buried amino acid (Pro76Ala) in the fully functional Trx variant Trx1P (yielding Trx0P) almost completely abolished the function of the protein as electron transfer catalyst. Although the re-orientation of the 75-76 bond from the cis to the trans conformation in Trx0P could be compensated by minor structural rearrangements and did not affect the overall protein fold, the small structural changes in the segment harboring the active-site disulfide were sufficient for an almost complete loss of function as TrxR-dependent reductase (Fig. 3, Supplementary Table  3). Inversely, the results indicate that the appearance of the cis Pro76 may have been a critical step in the evolution of Trx, in agreement with the finding that the diagnostic cis proline had already been present in the oldest bacterial thioredoxins 38 . But how could the cis proline evolve when it obviously affected the folding rate of Trx so dramatically? Our data on the rates of formation of N in the proline-independent folding of Trx0P and I trans , show that both proteins form N very rapidly with practically identical rates of ~8 s −1 (Fig. 4c, Table 1). Generation of a cis proline in the Trx fold should thus have slowed folding by at least two orders of magnitude. This may have been compensated by a native-like tertiary structure of I trans , proline isomerases as folding catalysts and molecular chaperones that increase the folding yield by preventing aggregation of slow folders. In addition, the cost of incorporation of a cis proline into a tertiary structure is formally comparably low (about 5 kJ mol −1 ), as the trans conformer of a prolyl peptide bond is energetically almost as unfavorable as the cis conformer 39 . Figure 7 summarizes our results on the thermodynamics of Trx1P and Trx0P folding. The fact that Trx0P is less stable than Trx1P I trans suggests a significant stabilization of the Trx fold by Pro76 relative to Ala76, even if Pro76 is in the nonnative trans-conformation. The comparably high free energy of folding of Trx1P I trans also explains the slow formation of N from I trans .
The folding mechanism of Trx1P proved to be strongly simplified relative to that of Trx WT 16 because the accumulation of nonnative cis prolines in U at the four trans prolines of Trx WT was eliminated. A further simplification was observed upon removal of the last proline in Trx0P, which became a perfect two-state folder. Our results are reminiscent of the elimination of one of the two cis prolines in ribonuclease T1, which also dramatically simplified RNaseT1 folding 40 . In addition, the 27-fold faster reaction of I trans to N in Trx WT compared to Trx1P suggests that the four trans prolines in Trx WT destabilize I trans in Trx WT relative to I trans in Trx1P. The extremely slow I trans to N reaction in Trx1P at the level of the intact tertiary structure underlines the strong coupling of folding with proline cis/trans isomerization 4 . We showed that the replacement Pro76Ala in Trx1P accelerated folding of the vast majority (95%) of the molecules by a factor of 66'000. This is, to our knowledge, the largest acceleration of protein folding by a single amino acid replacement reported so far. The high acceleration factor is also a consequence of the extremely slow I trans -to-N reaction in Trx1P folding (t 1/2 = 99 min). Indeed there are only very few examples for comparably slow folding reactions in small, one-domain proteins 28,41 . The very high activation energy of Trx1P folding might even provide the basis for designing a Trx variant that folds under kinetic control and stops folding at the level of I trans . Folding of Trx1P at 4 °C already proved to be close to kinetic control and reached N only with a half-life of 8 hours (Fig. 4b). Thus, even a slight, further stabilization of I trans by additional amino acid replacements might suffice to achieve complete kinetic control of folding at 4 °C. Such thioredoxin variants could e.g. be used for technical applications where reductase activity is switched on by a temperature jump.

Methods
Protein production and purification. Trx WT and Trx1P were produced and purified as described previously 42 . The plasmid for expression of Trx0P was constructed with the QuikChange Site-directed Mutagenesis Kit (Agilent Technologies) using the expression plasmid for Trx1P2 as template. Production of Trx0P in E. coli BL21(DE3) and enrichment with anion exchange chromatography on DE52 cellulose (Whatman) and gel filtration on Superdex 75 (GE Healthcare) under non-reducing conditions was performed exactly as described for Trx WT and Trx1P 18 . After the gel filtration step, the remaining impurities in the combined fractions containing Trx0P were removed by hydrophobic chromatography on Phenyl Sepharose 6 Fast Flow (GE Healthcare) in 20 mM MOPS-NaOH pH 7.0, using a gradient from 1.5 to zero M ammonium sulfate (20 column volumes). Fractions containing pure Trx0P were combined, dialyzed against 5 mM MOPS-NaOH pH 7.0, and stored at − 20 °C. The final yield of pure Trx0P was 11 mg per liter of bacterial culture. Ellman's assay 43 showed that the active-site disulfide bond in Trx0P was formed quantitatively after purification in the absence of reductant, and the correct mass of the protein was verified by ESI mass spectrometry (calculated: 11543. E. coli thioredoxin reductase was purified as described previously, 45 and its concentration was determined via its the specific absorbance at 280 nm (49000 M −1 cm −1 for the monomeric holo-enzyme).
The genetic sequence encoding E. coli trigger factor with N-terminal hexahistidine tag was cloned into pET11a via NdeI and BamHI and expressed in E. coli BL21DE3 at 20 °C for 12 h after induction with 0.5 mM IPTG. After cell lysis in 50 mM potassium phosphate pH 8.0, 300 mM NaCl, 2 mM β -mercaptoethanol, 20 mM imidazole, the protein was purified via Ni-NTA affinity chromatography in the same buffer (elution with 40 mM imidazole-HCl), followed by anion exchange chromatography on Resource Q (GE Healthcare) (10 mM potassium phosphate pH 7.5, elution with an NaCl gradient at ~180 mM NaCl) and a final gel filtration step on Superdex 200 (GE Healthcare) in 20 mM Tris-HCl pH 7.5, 100 mM NaCl, 1 mM EDTA. The concentration of trigger factor (yield: 12 mg per liter of bacterial culture) was determined via its specific absorbance at 280 nm (17420 M −1 cm −1 ), and its identity was confirmed by ESI mass spectrometry (expected mass: 49015.5 Da; found: 49016.0 Da).
Crystallization and structure determination of Trx0P ox . Crystals of Trx0P ox were grown via sitting drop vapor diffusion at 4 °C after mixing 175 nl of Trx0P ox solution (15.5 mg/ml in 20 mM MOPS-NaOH, pH 7.0) with 100 nl of precipitant solution (10% (w/v) PEG 1000, 10% (w/v) PEG 8000). Cryoprotection was achieved by shortly dipping the crystal into a solution of mother liquor with 20% ethylene glycol. Data to a resolution of 1.65 Å were collected at beamline X06DA of the Swiss Light Source (SLS, Villigen, Switzerland) at a wavelength of 1.0 Å. A complete dataset was obtained from a single crystal. Diffraction images were indexed and integrated with XDS 46 in space group P2 1 (a = 34.67 Å, b = 48.57 Å, c = 89.94 Å, α = β = 90°, γ = 101.01°), scaled with XSCALE and converted to amplitudes with XDSCONV and programs of the CCP4 suite 47 .
The three molecules of Trx0P in the asymmetric unit were positioned by molecular replacement with PHASER 48 , using the crystal structure of Trx1P ox (PDB 4HU7)2 as search model. Refinement was carried out using PHENIX 49 . After the first round of refinement, clear difference electron density at the mutation site of P76A mandated the replacement of the cis-proline of Trx1P by a trans-alanine. MolProbity 50 was used for structure validation; secondary structure elements were assigned using phenix. ksdssp. Diffraction data and refinement parameters are summarized in Supplementary Table 1 Interrupted refolding of Trx1P ox and Trx WT ox . Interrupted refolding experiments (N-tests) were used to monitor formation of native molecules during refolding of Trx WT ox or Trx1P ox , as they detect spectroscopically silent folding steps and quantitatively monitor the fraction of N at any time during refolding 30 . Proteins were first unfolded in 4.0 M GdmCl, 50 mM MOPS-NaOH pH 7.0, 1 mM EDTA at 25 °C for at least 2 h before initiation of refolding by rapid, 20-fold dilution with 50 mM MOPS-NaOH pH 7.0, 1 mM EDTA at 25 °C. The final concentrations in the refolding reactions were 35 μ M Trx and 0.2 M GdmCl. Samples were removed after different refolding times and mixed with an equal volume of GdmCl solution in 50 mM MOPS-NaOH pH 7.0, 1.0 mM EDTA. The final GdmCl concentrations during this unfolding step were 2.88 M and 3.45 M for Trx WT ox and Trx1P ox , respectively, and chosen such that native proteins unfolded with half-lives of about 30 s and folding intermediates (including I trans ) unfolded within the dead time of manual mixing. Unfolding reactions at 25 °C were monitored via the increase in the far-UV CD signal at 220 nm with a J715 CD spectrometer (Jasco) and unfolding of N was fitted mono-exponentially (cf. Figure S4). The corresponding amplitudes were plotted against refolding time and normalized, yielding the kinetics of formation of N. The y-axis intercept yielded the fraction of fast folders (folding within the dead time of manual mixing). The folding kinetics of the slow folders could be fitted with a single exponential function for both TrxWT ox and Trx1P ox , and were interpreted as the rate-limiting I trans -to-N transition.  Table 1 for the deduced kinetic parameters) 33 .
Fluorescence and CD spectroscopy. Fluorescence emission spectra of oxidized Trx0P, Trx1P, and Trx1P I trans immediately after start of refolding (rapid dilution from 4.0 M GdmCl to 0.2 M GdmCl) were recorded at 25 °C in 50 mM MOPS-NaOH pH 7.0, 1 mM EDTA containing 0.2 M GdmCl using a temperature-controlled Quantum Master 7 fluorescence spectrometer (PTI). The excitation wavelength was 280 nm in all measurements.
Far-UV and near-UV CD spectra of oxidized Trx WT, Trx0P, and Trx1P were recorded in 5 mM KH 2 PO 4 -KOH pH 7.0 at 25 °C using a temperature-controlled J715 CD spectrometer (Jasco). The protein concentrations in far-UV and near-UV CD measurements were 0.2 mg/ml and 0.4 mg/ml, respectively. CD signals were converted to mean residue ellipticity. The far-UV CD spectrum of Trx1P I trans was recorded immediately after the initiation of refolding of Trx1P at 25 °C in 50 mM MOPS-NaOH, 1 mM EDTA, 0.2 M GdmCl pH 7.0. Activity of Trx0P, Trx1P, and Trx1P I trans as a substrate of thioredoxin reductase (TrxR). The activities of Trx0P, Trx1P and Trx1P I trans as substrates of TrxR at pH 8.0 and 25 °C were measured as described previously 45 at constant TrxR concentrations of 40 nM (monomer), with the exception that 1 mM DTNB (5,5'-dithiobis-(2-nitrobenzoic acid), Ellman's reagent) and 500 μ M NADPH in a final reaction volume of 360 μ l were used and all reactions contained 20 mM GdmCl. Samples of Trx1P ox I trans were obtained immediately after refolding of Trx1P ox in 0.2 M GdmCl, 50 mM MOPS-NaOH pH 7.0, 1 mM EDTA). The reactions were initiated by addition of TrxR and followed by the absorbance increase at 412 nm as a result of the reduction of DTNB by reduced thioredoxin. The initial velocities were plotted against substrate concentrations and evaluated according to Michaelis-Menten kinetics. For calculation of k cat values, a molar extinction coefficient of 28300 M −1 cm −1 was used for the two TNB molecules formed per catalytic cycle.
As an alternative to the N-test, the rate of formation of native Trx1P ox from I trans was monitored by measuring the increase in the activity as substrate of TrxR as a function of refolding time. Refolding of unfolded Trx1P ox was started as described above (10 μ M Trx in the refolding reaction containing 0.2 M GdmCl) at 25 °C and pH 7.0. After different refolding times, samples were removed and diluted 1:10 with TrxR reaction buffer (0.1 M Tris-HCl pH 8.0, 0.25 mM EDTA, 0.25 mg/ml BSA, 1 mM DTNB, 0.5 mM NADPH), resulting in final concentrations of 1 μ M Trx and 20 mM GdmCl in the TrxR activity assay. The initial reaction velocities were plotted against refolding time and the rate of formation of native Trx1Pox was determined assuming Michaelis-Menten type substrate activity for I trans and native Trx1P according to equation (1), . TrxR substrate activity of Trx1P ox during refolding was recorded as described above and the activity increase as a function of refolding time was fitted mono-exponentially. As a consequence of formation of trans Pro76 in unfolded Trx1P ox , the amplitudes of the slow activity increase of Trx1P ox as TrxR substrate during refolding increased with unfolding time (cf. Figure S5A). The amplitudes were plotted against unfolding time ( Supplementary Fig. 5B) and yielded the apparent rate constant (k obs ) of attainment of the cis/trans equilibrium of Pro76 in unfolded Trx1P ox .

Kinetics of reduction of Trx WT and variants by DTT. Kinetics of reduction of oxidized
Trx0P, Trx1P, or Trx1P I trans by DTT were recorded at 25 °C in 50 mM MOPS-NaOH, pH 7.0 under pseudo-first-order conditions. Oxidized thioredoxins (2.0 μ M) were mixed with 200, 100, 50 or 20 μ M DTT in a temperature-controlled Quantum Master 7 fluorescence spectrometer. For each Trx variant, the increase in tryptophan fluorescence at 345 nm (excitation at 280 nm) upon reduction was fitted globally to pseudo-first-order kinetics. The kinetics of reduction of unfolded Trx1P could also be followed by an increase in tryptophan fluorescence, which is most likely caused by quenching of the fluorescence of Trp31 by the Cys32-Cys35 disulfide in the unfolded protein.
Redox equilibria with glutathione. The redox equilibrium between Trx0P or Trx1P and GSH/GSSG (eq. 2) was determined essentially as described previously 42 , making use of the increase in tryptophan fluorescence upon reduction of Trx and assuming that mixed disulfides between Trx and glutathione are not significantly populated at equilibrium. Trx1P ox or Trx0P ox (2 μ M) were mixed with GSSG (10 μ M) and variable concentrations of GSH (0.015-33 mM). After incubation under argon atmosphere for 5 h at 25 °C in 100 mM KH 2 PO 4 /K 2 HPO 4 pH 7.0, 1 mM EDTA, the fluorescence at 345 nm was recorded (excitation at 295 nm). As the GSH preparation contained small impurities of GSSG, the true GSSG concentration in all samples was determined immediately after each fluorescence measurement via glutathione reductase-catalyzed NADPH oxidation as described previously 53 . The equilibrium constant (K eq , eq. 2) was determined by fitting the fluorescence data according to equation 3, where F is the measured fluorescence and F ox and F red are the fluorescence intensities of oxidized and reduced Trx, respectively. Redox potentials were calculated with the Nernst equation (eq. 4), using a value of − 240 mV for the standard redox potential of glutathione 54  Thioredoxin-catalyzed reduction of insulin by DTT. The activity of Trx0P and Trx1P as catalysts of the reduction of horse insulin (130 μ M) by DTT (1 mM) was measured at 25 °C and pH 7.0 as described previously 18 , except that a final DTT concentration of 1.0 mM was used. The final concentration of Trx was varied between zero and 20 μ M. The aggregation of insulin as a consequence of reduction of insulin disulfides was followed via the increase in optical density (OD) at 600 nm. The time required to reach an OD 600 nm value of 0.1 was defined as the time of aggregation onset and plotted against Trx concentration. The time of aggregation onset was 44.1 ± 0.3 min for the uncatalyzed reaction.
Data analysis and statistics. The quantitative evaluation of the data provided in Table 1 and If not stated otherwise, the given errors result from the fits to the indicated equations/models.