Collapse of a long axis: single-molecule Förster resonance energy transfer and serpin equilibrium unfolding.

The energy required for mechanical inhibition of target proteases is stored in the native structure of inhibitory serpins and accessed by serpin structural remodeling. The overall serpin fold is ellipsoidal with one long and two short axes. Most of the structural remodeling required for function occurs along the long axis, while expansion of the short axes is associated with misfolded, inactive forms. This suggests that ellipticity, as typified by the long axis, may be important for both function and folding. Placement of donor and acceptor fluorophores approximately along the long axis or one of the short axes allows single-pair Förster resonance energy transfer (spFRET) to report on both unfolding transitions and the time-averaged shape of different conformations. Equilibrium unfolding and refolding studies of the well-characterized inhibitory serpin α1-antitrypsin reveal that the long axis collapses in the folding intermediates while the monitored short axis expands. These energetically distinct intermediates are thus more spherical than the native state. Our spFRET studies agree with other equilibrium unfolding studies that found that the region around one of the β strands, s5A, which helps define the long axis and must move for functionally required loop insertion, unfolds at low denaturant concentrations. This supports a connection between functionally important structural lability and unfolding in the inhibitory serpins.

Inhibitory members of the serpin superfamily regulate serine and cysteine proteases required for critical physiological processes, including blood coagulation and inflammation.[1,2] Unlike canonical protease inhibitors that simply bind to the protease active site, inhibition by serpins requires mechanical deformation of the protease active site, mediated by extensive structural rearrangements and repacking of the serpin structure.[3,4] Serpin structural remodeling is triggered when a target protease cleaves the serpin’s solvent-exposed reactive center loop (RCL) leading to insertion of the cleaved RCL into the center of β sheet A in the serpin and translocation of the covalently attached protease ∼70 Å from one end of the serpin to the other (Figure 1).[1,3−6] Because the energy needed for serpin structural remodeling and the associated protease inhibition is stored in the strained, metastable structure of active inhibitory serpins,[7−9] inhibitory serpins must fold to a kinetically trapped, metastable conformation that is not the global energy minimum.

Figure 1

α1AT anatomy and conformational gymnastics. (A) Two 90° views of the native α1AT structure (PDB entry 1qlp(90)) showing β sheets A–C colored yellow, light blue, and blue, respectively, helix F colored turquoise, and the RCL colored magenta. The labeled residues are shown as van der Waals spheres. As shown by the blue ellipses, the separation of residues 232 and 313 approximately coincides with the long axis of α1AT while the separation of residues 232 and 47 is close to one of the short axes. (B) Covalent complex between α1AT and bovine trypsin (green) (PDB entry 1ezx(4)). (C) Inactive, latent serpin conformation of α1AT (PDB entry 1iz2(10)). Protein images were generated using UCSF Chimera.[91]

The ellipsoidal fold of active serpins is characterized by one long axis, ∼70 Å in length for the canonical serpin, α1-antitrypsin (α1AT, also known as α1-proteinase inhibitor), and two approximately equal short axes, ∼45 Å each for α1AT (Figure 1). This watermelon-shaped fold is composed of three β sheets, A–C, and nine α helices in two well-connected domains.[1,2] The watermelon shape is defined by the serpin long axis that is largely coincident with sheet A, the largest of the β sheets and the β sheet most directly involved in protease inhibition as well as other, nonproductive conformations. These lower-energy, nonfunctional conformations include the latent form in which the intact RCL inserts into β sheet A in the absence of cleavage[10,11] and polymeric forms involving insertion of the RCL and/or other β strands into an adjacent serpin.[12−16]

The watermelon shape or asphericity of crystallized serpin conformations may be quantified using the asphericity, Δ, and shape, S, measures developed by Dima and Thirumalai, where S = Δ = 0 for spheres.[17] For active, metastable α1AT, Δ = 0.19 ± 0.01, where a positive Δ indicates that the protein is not spherical, and S = 0.16 ± 0.01, where a positive S indicates a prolate (watermelon) shape. A high value of Δ is also observed for other serpins, making most active serpins more aspherical than >70% of the monomeric proteins studied by Dima and Thirumalai. Interestingly, while functional insertion of the cleaved RCL does not alter the asphericity of α1AT, the values of S and Δ are unchanged, the latent conformation with its intact RCL is significantly less aspherical [Δ = 0.15, and S = 0.11 (Table S1 of the Supporting Information)]. Similar trends are observed for the inhibitory human serpin plasminogen activator inhibitor 1 where the latent form[18,19] is significantly less aspherical than the active[18,20,21] or cleaved[22] forms. Nor is this trend confined to human serpins; the latent state of the inhibitory bacterial serpin tengpin[23] is also significantly less aspherical than the active state (Table S1 of the Supporting Information). These shared structural characteristics suggest that decreased asphericity may be a hallmark of lower-energy, nonproductive serpin conformations.

While permanent alterations in serpin asphericity are associated with inactivation, transient expansion of the serpin structure is required for loop insertion[24] and has also been implicated in polymerization[15,25] and equilibrium unfolding.[25] How then do serpins fold to the aspherical final structure while avoiding more stable and more spherical structures such as the latent state? Serpins have been shown to fold using a mechanism with at least three states[7,26] via a molten globule intermediate[25−28] that is associated with aggregation.[27,29] Previous studies of serpin folding indicate that structures oriented along the long axis unfold first,[25,30−36] emphasizing the importance of this axis for folding, function, and dysfunction. However, to the best of our knowledge, there are no direct measurements of changes in the overall dimensions of a serpin during unfolding.

In more general terms, do aspherical proteins fold and/or unfold differently along different axes? Previous experiments aimed at addressing structural anisotropies used mechanical force and showed that the force required for protein unfolding can depend on the axis along which the force is applied.[37−40] However, these experiments were performed on relatively spherical proteins (Δ < 0.1, and S < 0.06), and the energy landscapes for mechanical unfolding with its directional perturbation of protein folds can be quite different than those for chemical denaturation, which acts isotropically.[41−44]

We have used single-pair Förster resonance energy transfer (spFRET) to monitor the chemically induced equilibrium unfolding and refolding of active α1AT. spFRET data for two different α1AT variants, one labeled approximately along one of the short axes and the second labeled approximately along the long axis, suggest that chemically denatured α1AT unfolds and refolds via at least two intermediates and that formation of the molten globule intermediates involves collapse of the long axis and an increase in the time-averaged sphericity. In addition, as observed in spFRET experiments with other proteins,[45−47] the unfolded state expands, along all axes, as the denaturant concentration increases. These results indicate the importance of the serpin long axis for both function and folding.

Experimental Procedures

Asphericity (Δ) and Shape (S) Parameters

Δ and S were calculated as described by Dima and Thirumalai[17] using Mathematica (Wolfram Research). The Mathematica scripts for reading the PDB file and performing the calculations were extensively modified versions of the Mathematica notebook Inertiatensor.[48] All of the calculated values are listed in Table S1 of the Supporting Information.

α1AT Preparation and Labeling

Wild-type α1AT contains a single Cys at position 232. To create double-Cys variants for spFRET experiments, a second Cys was introduced into the wild-type background, Ser47Cys (2Cys47) or Ser313Cys (2Cys313), in plasmid pEAT8-137[49,50] using the Quikchange XL (Agilent) site-directed mutagenesis protocol. Control fluorescence experiments required single-Cys variants: wild-type (containing Cys232), Ser47Cys/Cys232Ser (S47C), and Ser313Cys/Cys232Ser (S313C). The previously characterized single-Cys variants, S47C and S313C, do not alter α1AT stability or activity.[31,51] All double- and single-Cys α1AT variants were expressed as inclusion bodies in Escherichia coli BL21(DE3) cells, refolded, and purified as previously described.[51] For all α1AT variants, this procedure yielded protein that was >98% pure as judged by sodium dodecyl sulfate–polyacrylamide gel electrophoresis and gel densitometry.

Single-Cys variants were labeled with the donor Alexa Fluor 488 maleimide (AF488) (Invitrogen) or the acceptor Texas Red maleimide (TR) (Invitrogen) according to the manufacturer’s instructions. Double-Cys variants were first labeled with AF488, using a 1:1 dye:protein ratio. For spFRET measurements, a MonoQ anion exchange column (GE Healthcare) was used to separate singly labeled protein from unlabeled and doubly labeled species. The resulting singly labeled protein was further labeled with TR according to the manufacturer’s instructions. Labeling efficiencies for both fluorophores were measured by absorption according to the manufacturer’s instructions. For the doubly labeled variants, the shorter labeling time used for AF488 (donor), to avoid double labeling, resulted in 80–100% labeling, while longer labeling times used for TR (acceptor) labeling resulted in approximately 100% labeling. Single-Cys variants S47C and S313C were >70% labeled by both AF488 and TR. In contrast, the wild-type, with a single Cys at position 232, was >70% labeled by TR, but only ∼10% labeled using AF488. These results indicate that most of the doubly labeled α1AT variants have the acceptor, TR, attached to Cys232 and the donor, AF488, attached to either residue 47 or 313.

The stoichiometry of inhibition (SI), the number of moles of α1AT needed to inhibit bovine trypsin (Sigma-Aldrich), was determined as previously described.[51] Wild-type α1AT has an SI of 1.0, while the SI is 1.1 for unlabeled and labeled single- and double-Cys variants.

Far-UV Circular Dichroism (CD)

Room-temperature far-UV CD spectra from 200 to 250 nm were recorded on a Jasco J-810 spectropolarimeter using a 0.1 cm path length at room temperature. Samples containing 1.8 μM α1AT variants in 10 mM sodium phosphate (pH 7.4) and 50 mM NaCl with 0–6 M guanidinium chloride (GdmCl) were equilibrated for 2 h. GdmCl concentrations were determined from the index of refraction.[52] In 0 M GdmCl, the spectra of folded, doubly labeled Cys variants are essentially identical to that of wild-type α1AT, indicating that the mutations do not significantly alter the α1AT structure (Figure 2). The apparent fraction of unfolded protein, fapp, was determined from the far-UV CD unfolding data:where θ222 is the measured ellipticity at 222 nm, N is the native, fully folded protein in 0 M GdmCl, and U is the unfolded protein in 6 M GdmCl. fapp was fit to a three-state model for unfolding[53] (Supporting Information and Table S2).

Figure 2

α1AT structure and stability not significantly affected by Cys mutations and fluorescent labeling. (A) Far-UV CD spectra of the native state of wild-type (black), doubly labeled 2Cys47 (blue), and doubly labeled 2Cys313 (red) α1AT in 10 mM sodium phosphate (pH 7.4) and 50 mM NaCl. (B) GdmCl-induced unfolding of α1AT monitored by far-UV CD for the wild-type (black squares), doubly labeled 2Cys47 (blue circles), and doubly labeled 2Cys313 (red triangles). The line is a three-state fit to the wild-type data (Supporting Information and Table S2).

Steady State Fluorescence Spectra and Anisotropies

Increasing concentrations of GdmCl and/or protein unfolding can alter the absorption and emission spectra as well as the quantum yields of the fluorophores, potentially changing Ro, the Förster distance at which there is 50% energy transfer. To account for these effects, the steady state absorption spectra (data not shown), fluorescence emission spectra (data not shown), and steady state fluorescence anisotropies (Figure S1 of the Supporting Information) were measured for all three single-Cys variants labeled with AF488 or TR. Absorption spectra were collected using a UV–vis spectrometer (PerkinElmer), while fluorescence spectra and steady state anisotropy data were collected using a Fluorolog fluorometer (ISA) as previously described.[51] Singly labeled α1AT variants were incubated in 50 mM HEPES (pH 7.4), 50 mM NaCl, 0.01% Tween 20 buffer containing 0–6 M GdmCl for 2 h at room temperature to reach equilibrium.[31] Surprisingly, both the AF488 emission spectra and the TR absorption spectra red shift with an increasing GdmCl concentration, resulting in a very slight dependence of Ro on the GdmCl concentration (Figure S2 of the Supporting Information).

The effective value of Ro in the spFRET experiments, Ro,eff, is influenced by differences in quantum yield and detection efficiency between the donor (AF488) and acceptor (TR) as[54]For singly labeled samples with identical optical densities at the donor excitation wavelength, 488 nm, γ is the ratio of the fluorescence intensity of the acceptor-only samples to that of donor-only samples measured on the single-molecule setup[46] and γ was measured as a function of GdmCl concentration (Figure S2 of the Supporting Information). Analytically, γ = (ϕAηA)/(ϕDηD), where A and D refer to the donor and acceptor, respectively, ϕ is the quantum yield, and η is the detection efficiency.[55] The resulting Ro,eff is 51 Å for all GdmCl concentrations (Figure S2 of the Supporting Information).

spFRET Data Collection and Analysis

Doubly labeled α1AT variants were incubated in 50 mM HEPES (pH 7.4), 50 mM NaCl, 0.01% Tween 20 buffer containing 0–6 M GdmCl for 2 h at room temperature to reach equilibrium.[31] After 2 h, the micromolar samples were diluted to <50 pM for spFRET experiments. For refolding experiments, α1AT variants were first unfolded in buffer containing 8 M GdmCl for 2 h and then allowed to refold in 0–6 M GdmCl for 2 h.

spFRET data were collected on a home-built confocal microscope based on an IX-70 inverted microscope (Olympus) that has been previously described[51,55] (and Supporting Information). Data from the donor and acceptor channels were binned at 1 kHz, and a threshold of 30 total (acceptor plus donor) photon counts was used to identify photon bursts arising from α1AT diffusing through the focal volume. The apparent spFRET efficiency, Eapp, for these bursts was calculated using[55,56]where IA is the photon count for the acceptor corrected for background and leakage of donor photons into the acceptor channel (crosstalk) and ID is the photon count for the donor corrected for background. The apparent mean spFRET efficiency, ⟨Em⟩, and the peak half-width, σm, were determined from Gaussian fits to the data. The upper bound for peak half-widths, σsn, due to shot noise were calculated using[56,57]where Pt = 30 photon counts (donor + acceptor) is the threshold used to identify photon bursts arising from single α1AT molecules traversing the observation volume.

Radius of Gyration of the Denatured State Ensemble (DSE)

Assuming that unfolded α1AT behaves as a Gaussian chain, the average radius of gyration of the DSE, Rg, can be calculated from the mean square distance between the dyes, ⟨r2⟩:[58]where Nt is the total number of peptide bonds in the protein (393 for α1AT) and Ndyes is the number of peptide bonds between the donor and acceptor fluorophores. AF488 has a five-carbon linker, while TR maleimide has a two-carbon linker adding the equivalent of three peptide bonds between the donor and acceptor. The mean square distance between the dyes, ⟨r2⟩, can be determined from ⟨Em,γ⟩ for the DSE according to[47,59]where c is the distance of closest approach, lc is the contour length between the fluorophores, assuming 0.38 nm per peptide bond,[60]r is the distance between the fluorophores, E(r) is the energy transfer efficiency at distance r, and P(r) is the probability of observing a particular distance assuming that the unfolded protein behaves as a Gaussian chain. The Gaussian chain approximation assumes that the DSE behaves as an ideal freely jointed polymer with negligible interactions between amino acids. This model allows us to calculate the approximate radius of gyration but neglects confounding factors such as residual structure in the DSE. Note that to obtain distances from the FRET efficiencies, it is necessary to correct for differences in quantum yield and detection efficiency between the donor (AF488) and the acceptor (TR) given by γ, and ⟨Em,γ⟩ is the γ-corrected apparent mean FRET efficiency, ⟨Em,γ⟩ = ⟨Em⟩/[⟨Em⟩ + γ(1 – ⟨Em⟩]. (The FRET efficiency histograms were not γ-corrected because the relationship between the γ-corrected efficiencies and the counts in a histogram bin is not straightforward.[56]) For these experiments, γ varied from 0.85 in 0 M GdmCl to 0.95 in 6 M GdmCl (Figure S2 of the Supporting Information). To determine the value of ⟨r2⟩ associated with ⟨Em,γ⟩ for each concentration of GdmCl, eq 6 was numerically integrated using Mathematica for various values of ⟨r2⟩. For each denaturant concentration, the Rg of the DSE was calculated using the value of ⟨r2⟩ for which the numerical integral matched the experimentally observed values of ⟨Em,γ⟩.

Fluorescence Correlation Spectroscopy (FCS)

FCS was used to monitor α1AT translational diffusion and to look for conformational changes on time scales faster than that of diffusion through the observation volume. The correlations, G(τ), were calculated by the ISS Vista program according to[61,62]where the broken brackets denote the time average, I(t) is the photon count for channel j, and δI(t) = ⟨I(t)⟩ – I(t). j = k for the autocorrelation, while j ≠ k for the crosscorrelation.

The correlations were calculated using the photon counts from the spFRET experiments, collected with a 100 μm confocal pinhole. Additional FCS experiments were performed as previously described[63] using a 30 or 100 μm confocal pinhole and α1AT variants labeled only with the donor or with the donor and acceptor fluorophores. The ratio of the donor autocorrelation to the donor–acceptor crosscorrelation, GDD(τ)/GDA(τ), for spFRET–FCS experiments and the ratio of the fluorescence autocorrelation of the donor, GDD(τ), in the presence and absence of acceptor fluorophore for FCS-only experiments were calculated using Origin (OriginLab).

To determine diffusion times, τD, correlation curves for α1AT variants labeled only with the donor fluorophore, AF488, were fit to[62]where ⟨N⟩ is the effective number of molecules in the observation volume and the exponential expression accounts for AF488 fluorophores that are nonfluorescent due to quenching interactions. F is the fraction of molecules residing in the dark state at any time, and τdark is the lifetime of the dark state. The last term accounts for fluorescence fluctuations due to translational diffusion with τD = ωo2/D where ωo is the radius of the three-dimensional Gaussian observation volume and S = zo/ωo is the ratio of the axial to the radial extent. S and ωo were determined from FCS experiments using Alexa Fluor 488 hydrazide as a diffusion standard.

The correlation curves also allowed us to easily check for α1AT aggregation because such aggregates will diffuse slowly and have large values of τD. Increases in τD were observed as α1AT expanded due to unfolding, but these increases were within the range expected for increased viscosity at higher GdmCl concentrations and chain expansion. Large deviations and bumps, the hallmarks of significant aggregation, were not observed (data not shown). Using the 100 μm pinhole, the diffusion time of native α1AT in the absence of GdmCl was 240 ± 20 μs and τD increased to 550 ± 50 μs in 5 M GdmCl, reflecting both a 1.3-fold increase in viscosity and α1AT unfolding.

Results

To monitor changes in the shape of α1AT during equilibrium unfolding, the donor (AF488) and acceptor (TR) fluorophores for spFRET were placed approximately along either a short axis or the long axis (Figure 1). α1AT has a single native Cys, residue 232, located at the C-terminal end of β sheet B strand 1 (s1B), and this Cys was labeled in all of the variants. For the short axis variant, 2Cys47, the second Cys, located in a short loop between helix A and s6B, was introduced by mutating Ser47 to Cys, resulting in a donor–acceptor separation of ∼27 Å in the native state. For the long axis variant, 2CyS313, Ser313, located in a flexible region[64] near β sheet A, was mutated to Cys, resulting in a donor–acceptor separation of ∼43 Å in the native state. All of the labeled variants were active, inhibiting bovine trypsin with an SI of 1.1 similar to that of the wild-type.

Stability of the Labeled α1AT Variants

The stability of wild-type α1AT and its labeled variants was monitored using far-UV CD. As expected, on the basis of previous ensemble studies of α1AT unfolding,[7,26,31,65] α1AT unfolding is biphasic as indicated by the plateau around 1 M GdmCl (Figure 2). The Cys mutations and double labeling do not significantly alter the α1AT stability, and fits to the far-UV CD unfolding curves are similar for the wild-type and doubly labeled variants (Figure 2 and Table S2 of the Supporting Information).

Unfolding along the Short Axis

In 0 M GdmCl, spFRET histograms for doubly labeled 2Cys47 show two peaks, a small, “zero” peak arising from proteins containing only AF488 (donor) and proteins in which TR (acceptor) has photobleached as well as a much larger peak with an apparent mean efficiency, ⟨Em⟩, of 0.95 (0.96 when corrected for γ) (Figure 3A). The short distance, ∼27 Å, between residue 47 and Cys232 in the native structure would yield a native state spFRET signal centered at 0.97. However, the five-carbon (AF488) or two-carbon (TR) flexible chains linking the fluorophores to the Cys residues can increase this distance, and the peak centered at 0.95 efficiency with a half-width of 0.04 can easily be assigned to the α1AT native state (Figure 3A).

Figure 3

α1AT unfolding is different along the short and long axes. spFRET histograms for the unfolding of 2Cys47 (A) and 2Cys313 (B) in increasing concentrations of GdmCl as indicated on the plots. The total Gaussian fits to the histograms are indicated by red lines, and the resulting single Gaussians for the native, intermediate, unfolded, and zero peaks are indicated by green, blue, black, and yellow lines, respectively. The distance between residues 232 and 47 (2Cys47) increases as α1AT unfolds, while the distance between residues 232 and 313 (2Cys313) first decreases and then increases. (C and D) Centers, ⟨Em⟩, and half-widths, σm, of the spFRET peaks, determined from the Gaussian fits, as a function of GdmCl concentration for 2Cys47 (C) and 2Cys313 (D). Data for the native, intermediate, and unfolded species are colored green, blue, and black, respectively. Filled symbols were determined from unfolding experiments, while empty symbols were determined from refolding experiments.

As expected from the ensemble experiments, as the concentration of GdmCl increases to 0.5 M, a third peak appears in the spFRET histogram centered at 0.85. Because of its appearance at low GdmCl concentrations, and the subsequent appearance of a fourth lower-efficiency peak at even higher concentrations of denaturant, this ⟨Em⟩ = 0.85 peak must arise from equilibrium unfolding intermediates. The flexibility of the region around the fluorophores can increase the range of interfluorophore distances that can be sampled, thus increasing the width of the spFRET peaks. The spFRET peaks centered at 0.85 have half-widths of 0.08, only slightly higher than the half-width of 0.07 predicted from shot noise.[56,57] The α1AT equilibrium folding intermediate, which has been characterized as a molten globule,[25−28] should be able to access multiple conformations leading to a spFRET peak wider than what would be predicted simply from shot noise. However, the observed difference between the predicted half-width and the measured half-width is quite small and likely arises from experimental noise. The lack of resolvable heterogeneity for the separation between fluorophores at residues 232 and 47 illustrates the limits of the spFRET measurements. Because FRET has an r–6 distance dependence, FRET measurements are most sensitive between efficiencies of 0.8 and 0.2 (40.5 to 64.3 Å for an Ro of 51 Å), and much less sensitive at the extremes. Thus, conformational heterogeneity of the intermediate state(s) that changes the separation between residues 232 and 47 by ∼3.5 Å or less will be masked by noise for an spFRET efficiency peak centered at 0.85.

Around 2.5 M GdmCl, a fourth peak appears centered at 0.65. However, unlike the native and intermediate states, for which the peak centers are independent of denaturant concentration, the center of this peak, arising from the DSE, shifts to lower efficiencies as the GdmCl concentration increases (Figure 3A,B and Table 1). These DSE peaks are considerably wider than what would be predicted from shot noise. Using an Ro,eff of 51 Å to convert the spFRET efficiency peak half-widths to approximate distance changes between the fluorophores reveals that these large half-widths change the fluorophore separation by more than 5 Å. Thus, while experimental noise makes a small contribution to these half-widths, the widths of the spFRET peaks for the DSE likely reflect conformational heterogeneity (Table 1). Such heterogeneity could arise from metastable states separated by low-energy barriers,[66,67] i.e., states with lifetimes longer than the time it takes to diffuse through the observation volume, ∼1 ms.

Table 1

Changes in the Peak Centers and Widths for the α1AT DSEa

[GdmCl] (M)	2Cys47 ⟨Em⟩b	2Cys47 σmb	2Cys47 σsn	2Cys313 ⟨Em⟩b	2Cys313 σmb	2Cys313 σsn
1.5	NAc	NAc	NAc	0.61 (0.02)	0.16 (0.01)	0.09
2	NAc	NAc	NAc	0.61 (0.02)	0.16 (0.01)	0.09
2.5	0.63 (0.02)	0.13 (0.01)	0.09	0.62 (0.02)	0.17 (0.01)	0.09
3	0.61 (0.01)	0.22 (0.01)	0.09	0.62 (0.01)	0.16 (0.01)	0.09
3.5	0.38 (0.01)	0.22 (0.01)	0.09	0.61 (0.01)	0.18 (0.01)	0.09
4	0.26 (0.01)	0.16 (0.01)	0.08	0.52 (0.01)	0.18 (0.01)	0.09
5	0.20 (0.01)	0.13 (0.01)	0.07	0.40 (0.01)	0.18 (0.01)	0.09
6	0.18 (0.01)	0.12 (0.01)	0.07	0.38 (0.01)	0.19 (0.01)	0.09

Peak centers, ⟨Em⟩, and half-widths, σm, were determined from Gaussian fits to the unfolded peaks in the spFRET histograms. The predicted half-widths from shot noise, σsn, were calculated according to eq 4.[56,57]

The number in parentheses is the standard deviation from fits to histograms from at least two independent spFRET experiments.

Not Applicable. For 2Cys47, the DSE is not observed at moderate GdmCl concentrations.

The decrease in the apparent efficiency as the denatured states expand in increasing concentrations of GdmCl is commonly observed in spFRET unfolding experiments[45−47,54,68,69] and is expected due to solvation of the polypeptide chain.[47,69] At low GdmCl concentrations, more compact configurations are favored because of interactions between amino acids, but at higher denaturant concentrations, solvent–amino acid interactions dominate and the unfolded states are, on average, more extended.[45,47,69] The unfolding of 2Cys47 is reversible, and the locations of the spFRET peaks for the native, intermediate, and unfolded ensembles are the same for equilibrium unfolding and refolding (Figure S3 of the Supporting Information). Thus, spFRET measurements approximately along one of the short axes show the expected three-state unfolding and reveal that the intermediate and unfolded ensembles are distinct conformational ensembles separated by an energy barrier.

Unfolding along the Long Axis

In 0 M GdmCl, the spFRET histogram for doubly labeled 2Cys313 displays two peaks, the zero peak and a peak centered at an apparent efficiency of 0.80 (0.82 when corrected for γ) corresponding to the native state (Figure 3B). The half-width of the native state peak is 0.14, considerably wider than the half-width of 0.07 that would be predicted from shot noise. This large half-width indicates significant flexibility around and/or between residues 232 and 313. Cys232 is at the end of s1B, and the region from residue 232 to 313 contains four β strands and three α helices (s2B–s3B−αG–αH–s2C–s6A−αI). Data from mass spectrometry experiments including hydrogen–deuterium exchange mass spectrometry (HDXMS) and oxidative labeling[25,33,64] as well as PEGylation of single introduced buried Cys residues as a function of GdmCl concentration[32] suggest that β sheet B is quite stable and that the region from residue 275 to 313 (end of αH to 313) is more labile with large fluctuations likely in the loopy hI–s5A region (residues 299–331).

The effects of loop motions on the spFRET peak shape, location, and width depend on the time scale of the motion relative to the time it takes α1AT to diffuse through the observation volume (∼475 μs) and the bin time used for the histogram (1 ms).[56] On the basis of these time scales, loop conformational changes that are ∼200 μs or longer would likely increase the peak width. Modeling long-time scale protein fluctuations using coarse graining combined with Monte Carlo methods as implemented by the CABS-flex Web server[70,71] suggests that the hI–s5A loop may occasionally fold back on itself, allowing it to populate both less flexible and more flexible metastable states perhaps accounting for the large width of the 2Cys313 native spFRET peak (data not shown).

Surprisingly, as the GdmCl concentration is increased, a third peak appears to be centered at a higher spFRET efficiency of 0.95, corresponding to a decrease in the distance between the two fluorophores. This contraction is not due to fluorescence artifacts, because the absorption and emission characteristics of AF488 and TR do not significantly change from 0 to 3 M GdmCl (Figures S1 and S2 of the Supporting Information). This signal must, therefore, arise from intermediates in equilibrium folding for which the long axis has contracted relative to the native, folded structure. Note that the relative insensitivity of FRET to distance changes at such high efficiencies results in very narrow peak widths despite the conformational heterogeneity observed for the 2Cys313 native state and expected for the intermediate ensemble.

For the 2Cys47 α1AT variant, the location of the native state peak is independent of denaturant concentration (Figure 3). However, when the 2Cys313 α1AT variant was incubated in higher concentrations of GdmCl, expansion of the intermediate ensemble made it difficult to distinguish between spFRET signals arising from intermediate conformations and those arising from the native state with its intact long axis. At 0.85 M GdmCl, the center of the 2Cys313 peak center assigned to the native state shifts from an apparent efficiency of 0.80 to 0.86. The location of this peak, between the intermediate peak at 0.95 and the original native peak at 0.80, suggests that two intermediate ensembles may be populated.

Unlike 2Cys47, in which an unfolded state peak does not appear until ∼2 M GdmCl, for 2Cys313 an unfolded peak appears at 1.5 M GdmCl, revealing that the region between residues 232 and 313 likely unfolds at denaturant concentrations lower than those for the region between residues 232 and 47. As observed for the 2Cys47 unfolded peak, the apparent spFRET peak efficiency for the 2Cys313 unfolded peak decreases with increasing GdmCl concentrations, and the half-widths are wider than those predicted from shot noise (Figure 3D and Table 1). These results are consistent with previous ensemble fluorescence experiments indicating that the region around residue 313 unfolds in low GdmCl concentrations,[31] and that the region near Trp238 in strand 2B, within 10 Å of Cys232, unfolds only at high GdmCl concentrations.[34] In addition, equilibrium unfolding studies of α1AT variants in which buried residues were mutated to Cys and the exposure of buried residues was probed by reactivity with high-molecular weight poly(ethylene glycol)-maleimide (PEG-Mal) revealed that residue 302 in helix I (15 Å from residue 313) and residue 332 in strand 5A (13 Å from residue 313) are fully accessible to PEG-Mal at GdmCl concentrations of <1 M while residue 237 in strand 2B (6 Å from residue 232) is fully accessible only at GdmCl concentrations of >2.5 M.[32] Thus, multiple probes of folding show that the helix I to strand 5A region containing residue 313 is quite labile while β sheet B containing residue 232 is much more stable.

The unfolding of 2Cys313 is reversible, and peak centers and half-widths similar to those observed upon unfolding are observed upon refolding (Figure S3 of the Supporting Information). The 2Cys47 and 2Cys313 results suggest that the α1AT long axis and short axes behave differently during equilibrium unfolding and refolding. The monitored short axis shows the expected expansion as the protein unfolds via the intermediate; in contrast, the long axis contracts in the initial folding intermediates and then expands upon further unfolding.

Characterizing the DSE

Fully denatured proteins can be modeled as unstructured, random polymer chains with mean square end-to-end distances that follow a Gaussian distribution.[45,47,59,72−74] In this model, a Gaussian distribution also describes the root-mean-square distance between donor and acceptor fluorophores separated by ≥80 amino acids. The Gaussian chain model and eqs 5–8 can then be used to determine the radius of gyration, Rg, which should be the same for both the short and long axis spFRET variants if the denatured states are truly unstructured polymers. Above 3.5 M GdmCl, the radii of gyration determined from the γ-corrected centers of the unfolded peaks are similar for both 2Cys47 and 2Cys313, confirming that the time-averaged α1AT conformation is no longer asymmetric (Figure 4). As observed for other proteins, the Rg of the DSE increases with increasing concentrations of GdmCl.[75,76] In 6 M GdmCl, the Rg is 58 ± 2 Å, a 145% increase relative to the Rg of 23.7 Å measured by small angle neutron scattering for the native state.[77]

Figure 4

Radius of gyration (Rg) tracks α1AT expansion with increasing GdmCl concentrations. Rg calculated from 2Cys47 spFRET histograms (circles) and 2Cys313 spFRET histograms (triangles). The filled symbols were calculated from unfolding experiments, and the empty symbols were calculated from refolding experiments.

Limits on the Kinetics of Interconversion between States

While the equilibrium spFRET data provide information about conformational distributions, the time scales of fluorescence fluctuations due to translational diffusion and fast structural fluctuations may be determined using FCS.[61,62,78,79] At low protein concentrations, diffusion of fluorescent proteins in and out of the observation volume leads to fluorescence fluctuations on the time scale of diffusion (hundreds of microseconds) that are reflected in the decay of the correlation function (Figure S4A of the Supporting Information).[61,62] Fluorescence fluctuations on time scales faster than that of diffusion, arising, for example, from significant structural fluctuations that alter the donor to acceptor distance, will lead to exponential decays of the correlation with time constants of microseconds or longer in addition to the contributions from diffusion and fluorophore photochemistry (eq 10).[62,78,80−82] In the absence of dynamic changes in distance between the donor and acceptor, the fluorescence correlation can be expressed as G1(τ) = Gpc(τ)Gdiff(τ), which has contributions from diffusion, Gdiff(τ), and photochemistry such as excursions into the triplet states and quenching by Trp residues, Gpc(τ) (see eq 10). In the presence of structural dynamics that alter the distance between the donor and acceptor fluorophores, an additional term, Gdynamics(τ), contributes to the fluorescence fluctuations and the correlation function can be expressed as G2(τ) = Gdynamics(τ)Gpc(τ)Gdiff(τ). The ratio between these functions will reveal the time scale(s) of the structural fluctuations:[78,80−82]This ratio may be calculated in two ways: (i) the donor autocorrelation for protein labeled with both the donor and acceptor may be divided by the crosscorrelation between the donor and acceptor channels,[82] or (ii) the donor autocorrelation for protein labeled with both the donor and acceptor may be divided by the donor autocorrelation for protein labeled only with the donor. In both of these cases, the diffusion and photochemistry contributions should cancel out and the contribution of fluorescence fluctuations due to conformational changes should be obvious.[80,82]

To put limits on how fast α1AT can interconvert between states, we performed FCS experiments on α1AT labeled only with the donor, AF488, and on α1AT labeled with both the donor, AF488, and the acceptor, TR, for both 2Cys47 and 2Cys313 and calculated both types of ratios. The resulting ratios are essentially flat and similar for all concentrations of GdmCl, including 0 M GdmCl (Figure S4 of the Supporting Information). These results indicate that the average interconversion between states occurs more slowly than diffusion through the observation volume. This puts a lower limit of ∼250 μs for interconversion between the folded and intermediate ensembles and of ∼500 μs for interconversion between the intermediate ensemble and the DSE. (The approximate doubling of the diffusion time arises from both the increase in viscosity at higher GdmCl concentrations and expansion associated with unfolding.)

Discussion

Native α1AT is a watermelon-shaped ellipsoid. In 2Cys47, the two fluorophores are located approximately along one of the short axes of the ellipsoid, while in 2Cys313, the two fluorophores are close to the long axis (Figure 1). By combining these labeling schemes with spFRET, we tracked the overall symmetry of α1AT during equilibrium unfolding and refolding. The spFRET histograms were calculated for data binned at 1 kHz, and the shape of α1AT is therefore averaged over the time (hundreds of microseconds) it takes for the molecules to traverse the observation volume. This is particularly important for the DSE at high GdmCl concentrations where the instantaneous conformation of unfolded α1AT can be quite extended and aspherical, but the lack of preferred conformations leads to a relatively spherical time-averaged conformation. Our time-averaged results indicate that the first events in equilibrium unfolding involve contraction of the long axis and expansion of the short axes, resulting in a more spherical overall conformation, followed by overall expansion of α1AT as shown in Scheme 1, in which the dots indicate the approximate locations of the fluorophores; N, I1, I2, and U indicate the native, intermediate, and denatured state ensembles, respectively; and the braces denote the heterogeneity of the intermediate ensembles and the DSE.

Scheme 1

α1AT Equilibrium Folding Showing the Initial Compaction of the Long Axis and Expansion of the Short Axes

A number of α1AT equilibrium unfolding studies suggest that the unfolding intermediates are classic molten globules[25,27,28] with a compact conformation, and significant, but fluctuating, secondary structure.[83] For α1AT, GdmCl-induced equilibrium intermediates show significant secondary structure as determined by far-UV CD[7,28,31] (Figure 2), and the backbone amide hydrogens throughout the protein show no significant protection from hydrogen exchange [with the exception of a peptide in sheet B (residues 241–251) that shows some protection up to 4 M GdmCl], indicating that the structure fluctuates, and no significant heat capacity peak is observed for the transition between the intermediates and the unfolded states in differential scanning calorimetry (DSC) experiments.[25] Collapse of some structural constraints is also supported by the accessibility of buried Cys residues in helix I, s5A, and helix F regions to PEGylation even at low GdmCl concentrations.[32] Residue 313 is in the helix I–turn–s5A region, and the 2Cys313 spFRET results, which report on the lability of the long axis in this region, clearly agree with the HDXMS[25] and PEGylation[32] results at low GdmCl concentrations (<1 M) as well as the significant stabilization afforded by mutating Lys335 in s5A to small residues (Ala, Gly, and Val).[8,84,85] Overall, the decrease in asphericity observed for the intermediates is entirely consistent with the intermediates’ molten globule character because relaxation of the anisotropic constraints imposed by the tertiary structure will increase the overall, time-averaged symmetry during serpin unfolding.

Compared to the region around residue 313, the helix A–strand 6B region containing residue 47 is more stable. It is accessible to PEGylation only above 1 M GdmCl[32] and gains 40–60% protection from hydrogen–deuterium exchange[35] and oxidative labeling[33] after refolding for ∼10 min. The 2Cys47 spFRET histogram even shows a small native state peak for 2Cys47 at 3 M GdmCl (Figure 3A), suggesting that some regions of α1AT retain small but significant amounts of native state character even in the presence of moderate to high denaturant concentrations. This retention of local tertiary constraints is supported by HDXMS data for the peptide containing residues 241–251 that is not fully exchanged even at 4 M GdmCl, indicating the presence of some hydrogen bonding in β sheet B[25] and the lack of complete PEGylation at 3 M GdmCl for the Cys residues introduced at positions 55 and 65, near Ser47, and at position 237, near Cys232.[32] In addition, fluorescence from Trp238, near Cys232, shows only one transition with unfolding midpoints at 5.3 M for urea denaturation and 1.8 M for guanidinium sulfate denaturation, again indicating the retention of local native state character in β sheet B even at high denaturant concentrations. The higher local stability reported by 2Cys47 is also evident in the intermediate ensemble(s); the 2Cys47 intermediate ensemble shows significant population even at 4 M GdmCl where the 2Cys313 histogram shows only the DSE (Figure 3).

These differences in the spFRET histograms are consistent with the greater stability of the short axis monitored in the 2Cys47 experiments relative to the long axis monitored in the 2Cys313 experiments. Differences in stability between the 2Cys313 and 2Cys47 intermediate ensembles suggest that the s1B–s2B region probed by Cys232 in both constructs is quite stable. As discussed above, the stability of this region is further supported by fluorescence data from the Trp residue in s2B (Trp238),[34] PEGylation data,[32] and the early acquisition of structure in this region during kinetic experiments.[33,35] Thus, the B/C barrel region is the most stable part of α1AT. All of these data, along with fragment complementation studies that lead to stable, folded serpins for both α1AT[86] and ovalbumin,[87] support a model of α1AT equilibrium folding in which secondary structure formation at moderate concentrations of GdmCl results in a molten globule, and while significant structure is present between 1 and ∼3 M GdmCl, particularly in the B/C barrel region, the entire ellipsoidal structure does not snap into place and is not stable until low GdmCl concentrations (<0.5 M) are reached.

How Different Are the Intermediate Ensembles and the DSE?

Far-UV CD unfolding experiments reveal a gradual transition between the intermediate ensemble and the DSE (Figure 2),[7,28,31] while in DSC experiments, no heat capacity peak is observed for the intermediate to unfolded transition.[25] This behavior would be expected in the absence of an energy barrier between these conformational ensembles resulting in a continuous expansion of the α1AT structure.[25] In spFRET experiments, continuous expansion from the intermediate to the denatured ensembles would be detected as a GdmCl-dependent change in the mean energy transfer efficiency of the intermediate peaks, as observed for the DSE (Figure 3). However, the peak centers for the intermediates are not dependent on GdmCl (Figure 3), and the peaks for the intermediate ensembles and the DSE are distinct, indicating that these are two different conformational ensembles separated by an energy barrier. In contrast to the DSC, which might not detect a small change in heat capacity, the spFRET data reveal that there is likely a small energy barrier between the intermediate and denatured state ensembles.

The DSE

Trp fluorescence studies of serpin unfolding suggest that serpins have residual secondary structure even in high concentrations of chemical denaturants, and that this residual structure helps serpins refold.[34,88] This residual structure is reported to be in the B/C barrel,[34,88] a region that is particularly stable in the native α1AT conformation.[64] In spFRET experiments, such residual structure should reduce the Rg of the DSE. Above 5.5 M GdmCl, the Rg for α1AT levels out around 60 Å (Figure 4), which is significantly lower than the mean Rg value of 69 Å predicted for a 394-amino acid protein based on small angle X-ray scattering data from a number of proteins but is within the 95% confidence limits.[89] It is likely that the relatively low Rg arises from residual structure because Cys232 (used in all of our labeling schemes) is located in the B/C barrel and would therefore be affected by any residual structure.

These results demonstrate the utility of using multiple labeling locations for spFRET studies to probe global protein symmetry, particularly for larger proteins. They also suggest that the ellipticity of the serpin fold, which likely facilitates the conformational changes required for protease inhibition, may also play a role in serpin folding. Interestingly, insertion of the RCL into β sheet A increases the internal symmetry of α1AT and can result in two-state unfolding,[7] indicating that both internal and global symmetry can play a role in serpin folding.