Beyond the Random Coil: Stochastic Conformational Switching in Intrinsically Disordered Proteins

Summary

Intrinsically disordered proteins (IDPs) participate in critical cellular functions that exploit the flexibility and rapid conformational fluctuations of their native state. Limited information about the native state of IDPs can be gained by the averaging over many heterogeneous molecules that is unavoidable in ensemble approaches. We used single molecule fluorescence to characterize native state conformational dynamics in five synaptic proteins confirmed to be disordered by other techniques. For three of the proteins, SNAP-25, synaptobrevin and complexin, their conformational dynamics could be described with a simple semi-flexible polymer model. Surprisingly, two proteins, neuroligin and the NMDAR-2B glutamate receptor, were observed to stochastically switch among distinct conformational states despite the fact that they appeared intrinsically disordered by other measures. The hop-like intramolecular diffusion found in these proteins is suggested to define a class of functionality previously unrecognized for IDPs.

Introduction

The activity of most proteins is critically dependent on attaining a unique tertiary structure that can position key amino acid residues for molecular recognition and catalysis. However, upwards of 30% of eukaryotic cellular proteins are predicted to be completely or partially intrinsically disordered (ID) (Uversky and Dunker, 2010). These intrinsically disordered proteins (IDPs) are critical for a variety of essential cellular functions like transcription, gating the nuclear pore and membrane fusion.

Despite lacking the “lock and key” interfaces of folded proteins, ID regions often contain sites of molecular recognition, even in proteins that also contain folded domains (Lee et al., 2000). Disorder has been proposed to enable two modes of ligand recognition: 1) ligand induced folding (Dyson and Wright, 2002), in which the IDP adopts a complementary interface after the initial ligand contact and 2) conformational selection (Tsai et al., 2001), in which binding occurs only if the binding site is preformed or unoccluded before ligand contact. These two binding modes depend differently on the timescale of conformational fluctuation within the IDP. Ligand induced folding necessitates fast structural transitions or ligands will diffuse away before the binding interface is formed (Zhou, 2010). To date there have been no physical measurements to support such differences in IDP conformational dynamics.

There exists a continuum of protein structure with stably folded proteins at one extreme and extended random coils at the other (Uversky and Dunker, 2010). Under ideal solvation, the ensemble dimensions of chemically denatured proteins can be described with simple random coil polymer models, despite clear evidence for residual structure and long range interactions in the denatured state (Kohn et al., 2004; Wang et al., 2007). Similarly, low complexity ID sequences have also been described with simple worm-like chain models (Evers et al., 2006). However, most IDPs fall on the continuum between these two extremes. Although IDPs are depleted in the hydrophobic residues that stabilize the structure of folded proteins, they can adopt more compact configurations than expected for a random coil (Marsh and Forman-Kay, 2010). Net charge, proline content and non-specific interactions have all been suggested to affect chain compaction (Mao et al., 2010; Muller-Spath et al., 2010; Vitalis et al., 2007 ). While polymer models adequately describe extended random coils, it is unclear how applicable this is to describe IDP conformation under native conditions, particularly for compact, globular IDPs.

Most experimental methods to characterize IDPs under native conditions are limited to probing for residual secondary structure, identifying sufficiently populated intramolecular contacts, and describing ensemble properties such as the radius of hydration or gyration. However, any molecular details of polymer behavior are lost during ensemble averaging. In contrast, single molecule Fluorescence Resonance Energy Transfer (smFRET) studies have proved useful at examining conformational heterogeneity in chemically denatured proteins (Chen and Rhoades, 2008; Deniz et al., 2000; Hoffmann et al., 2007; McCarney et al., 2005; Michalet et al., 2006; Nettels et al., 2007; Schuler et al., 2002; Talaga et al., 2000) and characterizing IDP conformational transitions such as those occurring in -synuclein upon membrane binding (Ferreon et al., 2009) and in SNARE proteins induced by protein interactions both in vitro (Weninger et al., 2008) and in vivo (Sakon and Weninger, 2010). Here we use smFRET to characterize the conformational behavior of neuronal IDPs under native conditions. The charge/hydropathy (C/H) ratio has been suggested to govern the degree of compaction in IDPs (Mao et al., 2010). To sample IDPs with range of compaction, we have chosen a set of IDPs with differing amino acid composition (Figure 1). It includes soluble and membrane proteins, proteins that do and do not undergo ligand-induced folding, as well as proteins that are completely or partially ID.

Figure 1. Charge Hydropathy Ratio for IDPs used in this Study.

Mean hydropathy per residue was calculated using normalized Kyte Doolittle scale with a five residue scanning window. This is plotted against the net charge per residue. The C/H ratio has been suggested to dictate the degree of compaction in IDPs (Mao et al., 2010; Muller-Spath et al., 2010). The dotted line represents an empirically-determined charge/hydropathy relationship that distinguishes most globular and disordered proteins (Uversky, 2002). SNAP-25 (solid circle), synaptobrevin (open circle), neuroligin (open square), NMDAR-2B (solid triangle), and complexin (solid square). The C/H ratio was calculated for the polypeptide sequence neglecting the contribution of the dyes, which would introduce systematic shifts for all samples (Muller-Spath et al., 2010).

The presence of ID in these proteins has been confirmed using multiple ensemble methods, but none of these studies suggested differences between the native states of these IDPs. Our smFRET measurements have established differences between these IDPs which we can interpret as variations in the effective stiffness by using simple polymer models but also reflect the degree to which solvation is less than ideal. We found surprising, stochastic conformational transitions on the second timescale in two globular IDPs that were not observed in the extended IDPs. Such differences were not predictable from the primary amino acid sequence or ensemble measurements. This has implications for conformational selection in IDPs (Boehr et al., 2009) because the slow conformational switching would allow even rarely populated conformations to persist on biologically relevant timescales.

Results

Polymer Models for Intrinsic Disorder

The conformation of chemically denatured and low-complexity intrinsically disordered proteins have been described with semi-flexible polymer models to relate the time-averaged separation between points along the polypeptide to the length of a single polymer unit: the Flory virtual bond that defines the separation between sequential alpha carbon atoms (Flory, 1969). For this study, we used 0.36 nm as the virtual length of an amino acid taken from modeling and diffraction studies of peptides in the extended beta-sheet conformation (Pauling and Corey, 1951).

When fully extended, the separation between points on a polypeptide chain, or contour length (L), is given the product of the amino acid dimension (l) and the number of amino acids (N) (Figure 2a). However, full extension is not observed because each section of the polypeptide undergoes intramolecular diffusion. As such, the separation between points is described in terms of the root mean squared (rms) displacement in any direction, $R_{\mathit{rms}}=\sqrt{\langle R^2\rangle }$ (Figure 2a). Interactions or residual structure in polypeptides that resist bending can be described in terms of the persistence length, l_p, commonly defined as the distance along the contour at which the memory of the chain direction is lost. Including this effect in semi-flexible chain modeling in the limit L ≫ l_p gives:

$$R_{\mathit{rms}}^2=2l_{p}L$$ (1)

Figure 2. smFRET measurements of SNAP-25.

(a) Schematic of R_rms measurement in IDPs. Stars represent dye positions. Distance between the two labeling sites when fully extended is termed the contour length (L) while the root mean squared (rms) distance on a diffusing IDP is denoted R_rms. (b) A double-labeled molecule is encapsulated inside a biotinylated liposome which is then immobilized by tethering to a layer of biotinylated BSA and streptavidin deposited on a quartz microscope slide. (c) Representative smFRET traces of S25(177). (d) Histograms of smFRET for S25(119) in open circles and S25(177) in solid circles. Solid line is the fit to a Gaussian distribution. See also Figures S1 and S2.

This model provides a framework to assess relative protein flexibility in the context of polymer physics approaches (Flory, 1969; Grosberg et al., 2002). Ensemble approaches to experimental determination of persistence lengths of unfolded proteins have typically found values from 0.33 to 0.42 nm although some values in the range 0.65–0.80 have been reported (Zhou, 2004) and references therein).

Measuring IDP Conformational Dynamics with smFRET

In this study, all proteins had the endogenous cysteines changed to serine and pairs of unique cysteine residues introduced for fluorescent labeling. We denote the mutants by appending the number of residues between labeling sites in parenthesis following the protein name. Thus, S25(119) denotes SNAP-25 (S25) with the labeling sites 119 residues apart (Supplemental information). Unless otherwise indicated, single molecules were encapsulated in liposomes that were immobilized to allow for extended observation (Boukobza et al., 2001) (Figure 2b).

Our time resolution was 100 milliseconds, which ensures that all of our measurements report time-averaged behavior of the dyes and the protein. FRET efficiency, E, is typically related to static distance between donor and acceptor dyes using Förster’s equation and this approach has been used for disordered and denatured proteins (Deniz et al., 2000; Ferreon et al., 2009; Michalet et al., 2006; Talaga et al., 2000). However, when the interdye distance fluctuates faster than the measurement timescale, improved accuracy when converting measured E to R_RMS dye separation is possible by incorporating probability distributions, p(R), of dye separations, R, based on an assumed model of the motion (O’Brien et al., 2009) as

$$E={\int }_0^L\left(\frac{1}{1+{\left(\frac{R}{R_0}\right)}^6}\right)p(R)dR$$ (2)

where (R₀) is the Forster’s radius. This approach has been used for denatured proteins and short peptides (Hoffmann et al., 2007; Laurence et al., 2005; McCarney et al., 2005; Merchant et al., 2007; Schuler et al., 2002). Typically, a Gaussian probability distribution of distances is assumed but other models have been used (Evers et al., 2006). The probability distribution for distances in a polypeptide containing residual structure has not been established, so we adopted this common assumption when a semiflexible polymer could be reasonably assumed. In all cases, we use an empirically calibrated Förster radius generated by comparing smFRET measurements from doubly-labeled SNARE complexes to their crystal structures (Choi et al., 2010) (Supplemental information).

SNAP-25

SNAP-25 (S25) is a SNARE protein that couples a disorder-to-order transition to the fusion of synaptic vesicles with the plasma membrane (Brunger, 2005). The native state of S25 is completely ID (Fasshauer et al., 1997). However, as part of the SNARE complex S25 forms two parallel α-helices (Brunger, 2005). We used two mutants S25(119) and S25(177) expected to show high or no FRET respectively when incorporated into the SNARE complex (Weninger et al., 2008).

Single S25 molecules showed stable donor and acceptor emission intensity until photobleaching occurred (Figures 2c and S1a). Thus FRET efficiency showed no evidence of dynamic changes. For each S25 mutant, FRET efficiency histograms showed a predominant peak with a slight shoulder at higher FRET (Figure 2d and Table 1). The mean FRET, which was taken from a Gaussian fit to the predominant peak, decreased with increasing fluorophore separation as expected from a semi-flexible polymer. The narrow peak widths indicate shot noise limited detection (Table 1 and Figure S2), which is consistent with conformational dynamics much faster than our time resolution (Gopich and Szabo, 2005).

Table 1. Comparison of measured FRET efficiency and model predictions for IDPs.

Table includes only IDPs where the smFRET histogram could be described with a single Gaussian fit. Three independent experiments for each constructs were conducted to calculate the mean FRET efficiencies and widths with standard deviations. smFRET efficiency was converted to distance, R, using the calibrated Förster radius (Supplemental information). R_rms was calculated from the Gaussian weighted probability distributions of smFRET efficiency (supplemental information). The persistence length, l_p, is calculated by combining the empirical R_rms values with equation (1). For SB, * and ** denote full length SB in PC and PC/PS bilayer, respectively. For CX, ^# denotes bound to the SNARE complex. See also Figure S7.

	FRET Efficiency	FRET Peak Width	R (nm)	Rrms (nm)	lp (nm)
S25(119)	0.36±0.018	0.19±0.041	5.8	7.3	0.62
S25(177)	0.26±0.008	0.15±0.003	6.2	8.6	0.58
SB(44)	0.64±0.008	0.17±0.040	4.8	5.0	0.79
	0.69±0.066*	0.22±0.052*	4.6*	4.6*	0.70*
	0.69±0.109**	0.23±0.034**	4.6**	4.6**	0.70**
SB(67)	0.49±0.026	0.19±0.047	5.3	6.1	0.77
	0.42±0.085*	0.19±0.005*	5.5*	6.7*	0.93*
	0.44±0.068**	0.19±0.021**	5.5**	6.5**	0.88**
CX(32)	0.71±0.016	0.19±0.008	4.5	4.5	0.88
	0.52±0.029#	0.15±0.006#	5.2#	----	-----

These FRET values correspond to R_RMS donor-acceptor dye separations of 7.3 nm for S25(119) and 8.6 nm for S25(177) (Table 1). The measured dye separations can be used to calculate the persistence length directly using Eqn. 1, which gives a mean persistence lengths of 0.62 nm and 0.58 nm for S25(119) and S25(177), respectively. Such a short persistence length supports the interpretation of S25 as an unconstrained polymer.

To confirm that the fluorophores were not interacting with the polypeptide, we made ensemble measurements of the hydrodynamic radius (R_H) of S25 with and without dyes. We used dynamic light scattering (DLS), sedimentation velocity analytical ultracentrifugation (AUC) and analytical size exclusion chromatography (SEC), which gave a mean R_H of 3.75±0.38 nm. This value is consistent with the interpretation of S25 as a random coil of 23.3 kDa (Uversky, 2002). Importantly, the measured R_H was not affected by the presence of the dyes (Supplemental information). Thus, dye labeling does not fundamentally change the native state ensemble.

Using the average persistence length of 0.6 nm taken from smFRET measurements and Eqn. 1, we can calculate R_rms for the ends of the entire S25 protein (206 amino acids). This was used to estimate the expected radius of gyration, R_G, as $R_G=R_{\mathit{rms}}/\sqrt{6}$ (Grosberg et al., 2002), which gives R_G = 3.85 nm. Thus the R_G/R_H ratio for S25 in our experiments is 1.03, which is in reasonable agreement with the ratio of 1.06 measured for chemically-denatured proteins (Wilkins et al., 1999).

Synaptobrevin Cytosolic Domain

Synaptobrevin (SB) is also a SNARE protein with a well-characterized disorder to order transition. Unlike S25, SB is an integral membrane protein with an α–helical transmembrane domain. The cytosolic domain of SB (residues 1–96) is largely ID (Fasshauer et al., 1997; Hazzard et al., 1999). We measured smFRET from two doubly-labeled cytosolic domain constructs, SB(44) and SB(67) (Table 1). Like S25, the FRET histograms showed a single Gaussian peak with shot noise limited width (Figures 3a and S2). The mean FRET values were 0.64 for SB(44) and 0.49 for SB(67), which convert to R_rms values of 5.0 nm and 6.1 nm, respectively (Table 1). Assuming a semi-flexible chain model (Eqn. 1) gives persistence lengths of 0.79 for SB(44) and 0.77 for SB(67). These persistence lengths are longer than those found for S25 and suggest increased stiffness, which is consistent with synaptobrevin containing intramolecular interactions that resist bending.

Figure 3. smFRET measurements of synaptobrevin.

(a) Histograms of smFRET for the SB cytosolic domain in liposomes. Solid circles, SB(67). Open circles, SB(44). Solid line, Gaussian fit. (b) Schematic of doubly labeled full length SB reconstituted in PC or PC with 15% PS bilayer. (c) and (d) Comparison of smFRET for full length SB and the SB cytosolic domain for SB(44) and SB(67), respectively. Open circles, SB cytosolic domain free in liposomes. Solid squares, full length SB in 100% PC bilayer. Open triangles, full length SB in 85% PC 15% PS bilayer. See also Figures S1 and S3.

Full-length Synaptobrevin in Lipid Bilayers

Because SB functions in the synaptic vesicle membrane, we investigated the effect of reconstitution in a lipid bilayer on the native state conformation. We measured smFRET between the same labeling sites above but in full-length synaptobrevin (1–116) that was reconstituted into a planar lipid bilayer through its transmembrane domain (Figures 3b and S3). Both SB(44) and SB(67) freely diffused in the deposited bilayer. The diffusion constant was similar in bilayers composed of phosphatydlcholine (PC) or 85% PC with 15% phosphatydlserine (PS) (0.112 and 0.072 μm²/sec, respectively). At each frame of the tracking analysis, we also measured donor and acceptor fluorescence emissions to calculate the smFRET efficiency.

For both full-length SB(44) and SB(67) reconstituted in a bilayer, FRET efficiency levels remained stable with no signs of dynamic behavior and histograms still contained a single Gaussian peak (Figures 3c and 3d). Surprisingly, FRET efficiency values for full-length SB in PC or PS/PC bilayers were unchanged relative to the liposome-encapsulated cytoplasmic domain. The mean FRET efficiencies from SB(44) were 0.64±0.008 (liposome encapsulated), 0.69±0.066 (PC bilayer), and 0.69±0.109 (PS/PC bilayer); and for SB(67) were 0.49±0.026 (liposome encapsulated), 0.42±0.085 (PC bilayer); 0.44±0.068 (PS/PC bilayer). The error bounds reflect standard deviation in mean FRET from three independently repeated experiments under each condition. Thus, neither the helical transmembrane domain nor anchoring to the bilayer affected the conformation or dynamics of the cytoplasmic domain of SB.

Neuroligin Cytoplasmic Domain

Neuroligins (NL) are cell surface proteins that mediate synaptic adhesion (Sudhof, 2008). Ensemble studies found the cytoplasmic domain of NL-3 to be entirely ID (Paz et al., 2008). Our ensemble DLS, AUC and SEC studies of the cytoplasmic domain of NL-1 found mean R_H of 2.53 nm (unlabeled) and 2.63 nm (labeled), which is consistent with their published findings (Supplemental information). This small R_H is consistent with NL-1 cytoplasmic domain forming a compact ID globule rather than being an extended coil like S25 (Uversky, 2002). In addition, circular dichroism spectroscopy of NL showed no signs of secondary structure (Figure 4).

Figure 4. Circular dichroism measurements.

Circular dichroism (CD) spectra of neuroligin (NL), open circles. SNAP-25(S25), open triangles. Glutamate receptor CTD2 (N2B), solid circles.

We measured smFRET from two constructs, NL(59) and NL(105) that used native cysteines for labeling. In contrast to S25 and SB, which showed one Gaussian FRET peak, both NL constructs showed complex distributions spanning the entire range of FRET values (Figure 5a). If these complex distributions arose from sums of several distinct shot noise-limited Gaussian peaks then the semi-flexible chain model would predict dramatically different rigidity parameters for each separate subpopulation. However, the dispersion of FRET values did not arise from stable subpopulations.

Figure 5. smFRET measurements of the neuroligin cytoplasmic domain.

(a) Histograms of smFRET for NL. NL(105), solid circles. NL(59), open circles. (b) Representative smFRET as a function of time for NL(105). Fluorescence intensity, top. Calculated FRET, bottom. (c) Percent of NL(105) and NL(59) molecules displaying transitions in the 10–100 seconds before photobleaching. (d) Lifetimes of the stable FRET states between transitions in NL. NL(105), solid circles. NL(59), open circles. See also Figures S1 and S5.

By examining the emission from individual molecules, we found that both NL(59) and NL(105) showed surprising dynamic behavior (Figures 5b and S1d) that was not observed in S25 or SB (Figures S1a, S1b and S1c). NL showed stochastic transitions between different FRET efficiency values. Control experiments verified the switching was not due to protein aggregation (Figure S4) or photophysical artifacts including effects on quantum yield or dye mobility (Supplemental information). The switching behavior was common. About 50% of NL(105) and 30% of NL(59) showed transitions in the 10–100 seconds before photobleaching (Figure 5c). Increasing the temperature to 37 °C increased the percentage of molecules undergoing transitions to about 90% (Figure S5).

All transitions occurred within a single 100 msec frame followed by variable periods of stable FRET. We used a Gaussian derivative kernel algorithm to detect transitions (Sass et al., 2010) and compiled histograms of the dwell times between transitions (Figure 5d). The resultant histogram could be well fit by a single exponential with rate constants of 1.37 s⁻¹ and 0.99 s⁻¹ for NL(59) and NL(105), respectively. We found no relationship between FRET values and dwell times (Figures S5d and S5e). Analysis of the sequence of FRET transitions found a very complex pattern of transitions between particular conformations (Figures S5f and S5g).

The NL cytoplasmic domain interacts with the synaptic scaffold protein PSD-95, which binds to the C-terminus (Sudhof, 2008). To examine the effect of protein interactions on NL dynamics, we coencapsulated doubly labeled NL with 20-fold excess unlabeled PSD-95 (Figure 6). Coencapsulation experiments with acceptor labeled NL and donor labeled PSD-95 produced high FRET efficiency (Figures S4b and S4d) as expected from published results (Irie et al., 1997). The presence of PSD-95 had minimal effect on the FRET distribution (Figure 6a) and no effect on the dwell time histogram (Figure 6c). Thus, PSD-95 binding does not induce a folded state in the ID cytoplasmic domain of NL and transitions persist (Figures 6d and S6).

Figure 6. Effect of PSD-95 interactions on smFRET transitions in neuroligin.

(a) Histogram of smFRET measurements of NL(92) alone, open circles, and NL(92) co-encapsulated with 20-fold molar excess of unlabeled PSD-95, solid squares. (b) Schematic of a double-labeled NL co-encapsulated with unlabeled PSD-95 inside a surface tethered liposome. (c) Histogram of the of dwell times of distinct FRET states for NL(92) alone, open circles, and NL(92) with PSD-95, solid circles. Solid lines are single exponential fits with rate constants of 1.19 s⁻¹ for NL(92) alone and 1.05 s⁻¹ for NL(92) co-encapsulated with PSD-95. (d) Percent of NL(92) molecules observed to have at least one FRET state transition before dye bleaching during a 100 sec. observation window in the absence and presence of PSD-95. See also Figures S1, S4 and S6.

Glutamate Receptor Cytoplasmic Domain

N-methyl-D-asparate receptors are glutamate gated ion channels in the post synaptic membrane of excitatory synapses that play an important role in development, learning and memory (Lau and Zukin, 2007). The cytoplasmic domain of the N2B subtype is predicted to be ID (Ryan et al., 2008). Within the 75 kDa N2B cytoplasmic domain, palmitoylation defines an intervening membrane binding region (Hayashi et al., 2009) that partitions the CTD into two ‘domains’ that we term CTD1 and CTD2. While the full cytoplasmic domain was found to be insoluble, we were able to express and purify to homogeneity CTD2 (residues1259–1482) which contains the CAMKII and PSD-95 binding sites (Traynelis et al., 2010). Analytical SEC of N2B gave an R_H of 3.27±0.021 nm for an apparent molecular weight of 70.4±1.2 kDa compared to the actual molecular weight of 25.3 kDa. Circular dichroism of N2B showed no signs of secondary structure (Figure 4). This is the first experimental conformation that the cytoplasmic domain is disordered.

We measured smFRET in three constructs, N2B(15), N2B(121), and N2B(172) (Figure 7a). Upwards of 80% of N2B(121) molecules showed stochastic transitions between FRET efficiency similar to NL (Figure 7b). Transitions were not detected in N2B(15) and N2B(172), which showed high and low FRET efficiency, respectively. The transition rate constant for NR2B(121) was 1.38 s -1, which is similar to NL(59) (Figure 7c). As with neuroligin all FRET states were of similar lifetime (Figure S8c). The FRET efficiency histogram was well fit by two Gaussian functions with mean FRET efficiencies of 0.46 and 0.65, and widths of 0.27 and 0.17, respectively. The transition density plots show a limited set of transitions approximating a two state system (Figure 7e).

Figure 7. smFRET measurements of glutamate receptor cytoplasmic domain.

(a) Histograms of smFRET in N2B. Three different contour lengths between donor and acceptor dye molecules were measured. Open circles, N2B(15). Solid circles, N2B(121). Open triangles, N2B(172). (b) Representative smFRET as a function of time for N2B(121). Fluorescence intensity, top. Calculated FRET, bottom. (c) Lifetimes of the stable FRET states between transitions in N2B(121). (d) Plots of dwell times in distinct FRET states as a function of FRET efficiency for single molecules of N2B(121). (e) The FRET state before a transition (y-axis) plotted against the FRET state after that transition (x-axis) for N2B(121).

Complexin

Complexin (CX) is a soluble presynaptic protein that is critical for evoked neurotransmitter release. CX contains a central α-helix flanked by ID regions with no tertiary structure (Brunger, 2005). Thus, CX differs from the other proteins in that it contains a mixture of stable secondary structure and ID.

We measured smFRET from two doubly labeled constructs. In CX(32) the fluorophores were placed at the ends of the central helix, while in CX(69) the C-terminal fluorophore was attached to the native cysteine within the ID region (Supplemental information). For CX(32), the FRET distribution contained a single Gaussian peak centered at 0.71 (Figure 8a and Table 1), with a shot noise limited width (Figure S2). The l_p value for CX(32) is the largest we have measured and is consistent with the polypeptide between the fluorophores adopting partial α-helical conformation with increased rigidity. The CX(69) was poorly fit by a single Gaussian but could be fit by two Gaussian functions giving mean FRET ± widths of 0.45 ± 0.30 and 0.59 ± 0.15 (Figure 8a). The peak widths for smFRET histograms compiled from ensembles of CX(69) are wider than for the other IDPs, but we have confirmed that the data from individual molecules each shows shot noise limited width. This suggests the presence of static heterogeneity in the ID region or conformational switching on a time scale much slower than the observation time of 10¹–10² seconds. The complicated distributions leave unclear the applicability of simple polymer models so no distance calculations were made.

Figure 8. Effect of binding to SNARE complex reconstituted in a lipid bilayer on the smFRET histograms of complexin.

(a) Histogram of smFRET for CX encapsulated in liposomes. CX(69), solid circles. CX(32), open circles. (b) Schematic of doubly labeled CX binding to unlabeled SNARE complexes reconstituted into PC bilayer. (c) and (d) Comparison of smFRET in CX when encapsulated in liposomes (open circles) or bound to the SNARE complex on lipid bilayers (solid squares). (c) CX(32). (d) CX(69)

Structural transitions in Complexin upon SNARE binding

SNARE binding lengthens the central helix in CX (Chen et al., 2002) and membrane interactions may induce α-helices in the ID regions (Seiler et al., 2009). To examine these effects, we measured smFRET within CX bound to SNARE complexes that were reconstituted into planar lipid bilayers (Figure 8b). In our hands, CX does not interact with PC bilayers but binds the SNARE complex with 10⁻⁸ M affinity (Bowen et al., 2005). Thus CX retained on a protein-containing PC bilayer is bound to an unlabeled SNARE complex. For CX(32), SNARE binding decreased the mean FRET to 0.52 meaning that the dyes moved apart (Figure 8c and Table 1). This increase in dye separation relative to the unbound state is in agreement with extension of the central α-helix. Our smFRET derived dye separation for CX(32) using the Forster relation for static distances compares favorably to the crystal structure of complexin bound to SNARE complex (Chen et al., 2002) so these sites were used in the calibration of the Förster radius.

Intramolecular smFRET from CX(32) bound to SNARE complex was still described by a single Gaussian peak (Figures 8c and 8d). Neither CX construct displayed evidence of dynamic behavior (Figure S1f). smFRET histograms for CX(69) bound to the SNARE complex still required two Gaussian functions (mean FRET ± widths of 0.30 ± 0.24 and 0.66 ± 0.19 ) to fit the distribution, which suggests heterogeneity in the conformation of the c-terminal region. This multi-state behavior persists from the unbound state (Figure 8a) and was also seen previously in intermolecular FRET between the c-terminal region of complexin and to the SNARE complex (Bowen et al., 2005).

Discussion

Characterizing the native state of IDPs and the effect of ligand interactions on IDP conformation remains a major challenge (Eliezer, 2009). The possibility that IDPs fall into distinct classes (Vucetic et al., 2003) remains controversial in part because of the lack of methods to characterize properties of the disordered native state. Describing the native state ensemble is critical for understanding the folding and binding mechanisms of IDPs.

We selected five neuronal proteins that represent a range of sequence diversity of IDPs (Figure 1). We used smFRET to characterize the molecular dimensions (R_rms between labeling sites) of IDPs undergoing free diffusion inside liposomes. smFRET for three of the proteins showed single Gaussian peak of shot noise limited width (Figure S2), consistent with rapid dynamics. Using a semi-flexible chain model (Eqn. 1), we extracted the persistence length, l_p as a metric to compare “stiffness” in IDPs. Although such models are known to be an abstraction, they provide a framework to compare the relative flexibility of IDPs under native buffer conditions.

S25 is known to contain some residual helicity under native conditions, yet l_p was only 0.6 nm, which is similar to that of chemically or mechanically denatured proteins (Zhou, 2004). Our measured R_H values are also consistent the interpretation of S25 as an extended random coil. Chemical denaturants approximate an ideal solvent for polypeptides (Dill and Shortle, 1991), so our measurements suggest that standard native buffer conditions constitute an ideal solvent for S25.

The cytoplasmic domain of SB has been shown to be largely ID with minimal residual helicity (Hazzard et al., 1999). The C/H ratio of the SB SNARE motif is similar to N2B, while including the proline-rich N-terminal domain shifts the C/H ratio into the range of folded proteins. In addition the R_H measured by SEC showed SB to be a compact globule, with the apparent weight only 25% larger than the actual molecular weight. Nonetheless, SB smFRET showed single Gaussian peaks of shot noise limited width. The polymer model gives l_p values only slightly higher than S25, consistent with a slight increase in intramolecular interactions that resist bending. Although detergent binding to the SB SNARE motif induces helix formation (Ellena et al., 2009), reconstitution of full-length SB in a lipid bilayer did not change the smFRET values (Figure 3) suggesting that the conformation and l_p is the same. These R_rms values were calculated using the assumption of rapid dynamics that are described by a Gaussian probability distribution for R_rms, which may not hold true for structured regions.

The same uncertainty applies CX, which is known to contain helicity in the region separating the dyes. In the unbound state, this region is dynamically sampling helical conformations (Chen et al., 2002). SEC found CX to be the most extended IDP of the set with a 5 fold increase in the apparent molecular weight, so the Gaussian assumption was still applied. Once bound to the SNARE complex, the entire intervening region in CX(32) forms a stable α-helix (Chen et al., 2002). Calculations of interdye separation using the standard FRET equation are in excellent agreement with molecular dynamics simulations of the dyes attached to the crystal structure suggesting that the polymer model would not improve accuracy.

In contrast, our smFRET measurements involving a label site in the c-terminal domain revealed multiple conformations in both the isolated state and the SNARE bound state. CX has no tertiary structure and our SEC measured R_H confirmed that CX adopts an extended conformation. The nature of conformational heterogeneity in a random coil is unclear and no interconversion was observed even in the unbound state. The semiflexible polymer model would assign different rigidity to each population. Rather than select for one conformation, SNARE complex binding increased the FRET difference between the two populations, increasing the high FRET and decreasing the low FRET states, respectively.

Experiments with S25, SB resulted in histograms with single Gaussian smFRET peaks even when SB was reconstituted in lipid bilayers. We interpret stable FRET and shot noise limited Gaussian peaks as arising from averaging of motions faster than our 100msec timescale (Gopich and Szabo, 2005). In contrast, NL and N2B showed more complex smFRET distributions. Single NL and N2B molecules showed evidence of stochastic switching between discrete FRET states. To our knowledge this is the first demonstration that IDPs can have distinct conformational behavior within the native state ensemble. Some compact globular IDPs do not undergo rapid continuous diffusion.

The interpretation of this switching behavior is difficult. FRET transitions are often interpreted as a change in structure. However the nature of the intervening stable FRET states remains a mystery. It is not clear if the stable FRET represents rapid averaging about a subset of possible structures or the presence of a single stable structure. Paradoxically, these extremes of behavior are indistinguishable in the current experiments. Our circular dichroism measurements for both proteins are consistent with ID, so any stable conformation does not involve regular secondary structural elements. Our analysis found no evidence for a preferential conformation as all sampled FRET states were of similar lifetime.

The free energy landscape of IDPs has been suggested to be nearly flat such that all conformations are of equivalent energy and the transition barriers are small (Figure 9a). As the height of the transition barriers increases, intramolecular diffusion can become transiently restricted in localized free energy minima (Figure 9b). The slow rate of transitions seen in NL and N2B would suggest that the height of the barriers between local free energy minima are much higher than those of the other IDPs resulting in the stochastic behavior observed. We term this phenomenon hop-diffusion in analogy to the behavior of membrane components subject to local confinement (Ritchie et al., 2003).

Figure 9. Schematic representation of energy landscapes that could generate steady and hop-like intramolecular diffusion in the native state of IDPs.

Conformational space is represented on the x-axis while free energy is represented on the y-axis. (a) SNAP-25 where all conformations are of roughly equal energy with small barrier separating conformations. (b) Neuroligin where all conformations are still equal but the height of the barrier is increased, which would slow transitions. (c) Hypothetical restricted chain where physiological conditions change the energy landscape to favor a subset of possible conformations.

The timescale of structural transitions has been proposed to determine the binding mechanism in IDPs (Zhou, 2010). Fast transitions are needed for induced folding, while slow transitions would result in conformational selection. The shot noise limited FRET peaks for S25 and SB are entirely in accord with their known mechanism of induced-folding. The stochastic switching in NL and N2B would be more compatible with a conformational selection mode of binding. That the transition lifetime approaches the second timescale would mean that even rarely sampled conformations can have very long lifetimes with respect to the speed of synaptic transmission. This may ensure that synaptic signaling proteins are not “active” unless the correct conformation is induced by local conditions or interactions (Figure 9c). We found that protein interactions with PSD-95 did not induce folding in NL but did affect the structural transitions, with different transitions being favored in the bound and unbound state (Figure 6).

Interestingly, previous work found similar stochastic FRET switching in S25 upon binding the SNARE protein syntaxin (Weninger et al., 2008). Thus, an extended random coil capable of ligand induced folding like S25 can be changed to mode of conformational selection by protein interactions. The binary SNARE complex would represent a state of conformational selection awaiting the arrival of the synaptic vesicle containing SB. The correct conformation would need to precede SB binding to insure efficient membrane fusion, a notion supported by the stabilizing effect of accessory proteins on the binary complex (Weninger et al., 2008).

The physical origin of the transitions in NL and N2B is unknown. Compositional profiling (Vacic et al., 2007) showed that both are enriched in aromatic residues and prolines relative to the three other proteins, which show no significant compositional differences. Our SEC results agree with the notion that the C/H ratio governs compaction but a low C/H ratio does not appear sufficient to elicit this phenomenon as SB was similar to N2B. A direct comparison of their compositional profiles shows that SB is enriched in bulky amino acids relative to N2B, which may limit access to collapsed states despite the lower C/H ratio. The deviation from simple polymer behavior for NL and N2B suggests that these proteins are poorly solvated by standard native buffer conditions. Weak, presumably non-specific, intramolecular interactions must stabilize these transiently populated states.

Annotated sequence analysis has broken IDPs into distinct flavors (Vucetic et al., 2003), but experimental demonstrations of differences in IDP behavior has been lacking. Our characterization of hop-like diffusion in NL and N2B revealed kinetic differences in conformational sampling of otherwise indistinguishable IDPs. Our results demonstrate that smFRET is a tool uniquely suited to reveal differences in the disordered state of proteins. We expect that smFRET can be effectively applied to characterize transiently stabilized conformations that may fundamental for the functions of many other IDPs.

Experimental Procedures

Protein Constructs

All proteins were cloned from Rattus norvegicus. Briefly, proteins were expressed as 6-His fusions or GST-fusions and purified by affinity, ion exchange, and gel filtration chromatography. All proteins were examined by SDS-PAGE for purity (>90%) before labeling. Double cysteine mutants were created by the QuickChange method (Agilent Technologies) from cysteine-free templates wherein native cysteines were mutated to serine. Proteins were randomly labeled with a mixture of maleimide derivatives of Alexa 555 and 647 (Invitrogen, Carlsbad, CA) using the manufacturer’s protocols as described previously (Weninger et al., 2008). The labeling efficiencies were greater than 80% except for the full length synaptobrevin which was closer to 50%.

Soluble proteins were encapsulated in biotinylated PC liposomes (Avanti Polar Lipids) for immobilization to a biotinylated BSA coated quartz surface as described (Boukobza et al., 2001; Choi et al., 2010). Studies of complexin bound to the SNARE complex were performed as previously described (Bowen et al., 2005). Full-length synaptobrevin in lipid bilayers was reconstituted into liposomes using detergent assisted insertion as previously described (Weninger et al., 2008).

FRET Microscopy

FRET observations were performed in custom-built quartz flow cells as described previously (McCann et al., 2010; Weninger et al., 2008). Observations included oxygen scavengers (glucose oxidase, 20 units per ml and catalase, 1,000 units per ml) and 100 μM cyclooctatetraene. Data was recorded using prism-type total internal reflection microscopy, detected by EMCCD cameras (either Cascade 512B or Andor iXon) and illuminated by an alternating laser sequence of 635 nm and 532 nm to distinguish the number of donor and acceptor dye molecules in each liposome. Data analysis was as described previously (McCann et al., 2010). Briefly, FRET efficiency (E) is calculated from the background subtracted intensity time traces. We applied global gamma correction with the mean gamma value determined from single molecule photobleaching (McCann et al., 2010). Histograms of FRET efficiency for hundreds of molecules were characterized by non-linear least squared fitting with Gaussian functions. Analysis of full-length synaptobrevin diffusion in lipid bilayers used a custom tracking algorithm written in MATLAB. Additional methodology is available in supplemental information.