Probing interdomain linkers and protein supertertiary structure in vitro and in live cells with fluorescent protein resonance energy transfer.

Abstract

Many proteins are composed of independently-folded domains connected by flexible linkers. The primary sequence and length of such linkers can set the effective concentration for the tethered domains, which impacts rates of association and enzyme activity. The length of such linkers can be sensitive to environmental conditions, which raises questions as to how studies in dilute buffer relate to the highly-crowded cellular environment. To examine the role of linkers in domain separation, we measured Fluorescent Protein-Fluorescence Resonance Energy Transfer (FP-FRET) for a series of tandem FPs that varied in the length of their interdomain linkers. We used discrete molecular dynamics to map the underlying conformational distribution, which revealed intramolecular contact states that we confirmed with single molecule FRET. Simulations found that attached FPs increased linker length and slowed conformational dynamics relative to the bare linkers. This makes the CLYs poor sensors of inherent linker properties. However, we also showed that FP-FRET in CLYs was sensitive to solvent quality and macromolecular crowding making them potent environmental sensors. Finally, we targeted the same proteins to the plasma membrane of living mammalian cells to measure FP-FRET in cellulo. The measured FP-FRET when tethered to the plasma membrane was the same as that in dilute buffer. While caveats remain regarding photophysics, this suggests that the supertertiary conformational ensemble of these CLY proteins may not be affected by this specific cellular environment.

INTRODUCTION

Resolving protein structure has been an important key to understanding protein function. This is no less true for multidomain proteins containing intrinsically disordered regions (IDRs), which adopt a supertertiary ensemble of conformations rather than a single predominant structure [1–3]. Most of our knowledge regarding protein structure has been obtained using purified samples in dilute buffer conditions [4]. In contrast, the cellular environment is crowded with a diversity of macromolecules [5, 6]. Numerous studies have shown that the effects of macromolecular crowding can impact protein folding, stability and association [7–9]. The impact of the cellular environment on protein structure has particular importance for proteins with IDRs [10, 11], whose conformations are generally more sensitive to solvent conditions [12] and are also prone to fuzzy interactions with cellular binding partners [13].

An important class of IDRs are those serving as linkers between folded domains or functional sites [14]. The linker extension helps determine the effective concentration of the tethered entities [15], which can influence rates of association or the kinetics of enzymatic reactions [16]. The high concentration enforced by linkers also strengthens weak interactions between the tethered domains, which can give rise to a dynamic supertertiary structure [1, 17].

Fluorescence Resonance Energy Transfer (FRET) is one of the few methods that can probe protein conformation in vivo [18]. Spectrally distinct FPs, used as donor and acceptor, are fused to a single protein to monitor protein conformation in cells and tissues [19–22]. The combination of enhanced cyan and yellow fluorescent proteins (eCFP and eYFP, respectively) and their derivatives remain one of the most common pairings for Fluorescent Protein FRET (FP-FRET) [23]. Many studies have examined tandem FPs separated by a variety of linker sequences and have confirmed that FP-FRET depends on the length and sequence composition of the linker [24–29]. Interpreting FP-FRET in terms of structural changes has always been a desirable goal [30]. Numerous reports interpret changes in FP-FRET efficiency as directly indicating a protein conformational change, either as an absolute distance or indicating the direction and magnitude [31–44]. However, FPs are not compatible with the common assumptions used in the reported FRET distance calculations [45, 46].

Here, we examined FP-FRET as probe of supertertiary structure using a series of tandem eCFP-Linker-eYFP (CLY) constructs (Figure 1A) that were previously shown to follow polymer scaling of their variable interdomain linker [26]. We used discrete molecular dynamics (DMD), a rapid and predictive molecular dynamics algorithm [47], to probe the conformational landscape. DMD revealed a diverse ensemble with intramolecular contact states that we confirmed with single molecule FRET. These studies show that intramolecular FP interactions biased the linker extension. This makes these CLYs constructs poor probes of inherent linker properties.

Figure 1. Fluorescent Protein Constructs Used in this Study.

A) Representative structural model of the CFP-Linker-YFP (CLY) constructs. The linker contains a variable number of GGSGGS repeats as indicated by the construct numbers. Shown is CLY9. B) SDS-PAGE of CLY constructs using 12% polyacrylamide and stained with Coomassie brilliant blue. The first lane shows molecular weight markers with their formula weights denoted to the left (in kDa). C) Fluorescence emission spectra of CLY1 (blue), CLY5 (orange) and CLY9 (grey) excited at 458 nm. Spectra were normalized to the emission maxima of YFP. D) Fluorescence anisotropy of CFP (cyan) and YFP (yellow) in CLY1, CLY5 and CLY9. Error bars show the standard deviation (SD) from 3 replicate measurements.

Next, we showed that FP-FRET in these CLYs is affected by macromolecular crowding and solvent quality. Having established their environmental sensitivity, we measured FRET from these same CLY constructs in cellulo by targeting them to the plasma membrane of live cells. Interestingly, the observed FP-FRET at the plasma membrane was in agreement with measurements in dilute solution. While caveats remain about photophysics, this similarity in the data suggests that the environment near the plasma membrane may not affect the supertertiary ensemble of the CLY tandem FPs under study.

RESULTS

We characterized three of the CLY tandems from the original series of 9 [26]: CLY1, CLY5 and CLY9. The CLY number indicates how many sequence repeats are present within the linker in addition to N- and C-terminal flanking sequences [26]. Thus, these constructs represent the shortest (23 residues), middle (47 residues) and longest (71 residues) linkers of the much larger published CLY series (Figure 1B). We measured the ensemble fluorescence spectra of the three purified CLYs (Figure 1C). We calculated observed FRET efficiencies (E_obs) of 0.70 ± 0.01 for CLY1, 0.54 ± 0.02 for CLY5 and 0.48 ± 0.04 for CLY9 (Table 1), which agreed well with previous measurements [26]. We also measured the fluorescence anisotropies, which were over 0.3 for both FPs in all CLY constructs (Figure 1D). The high anisotropies are close to the experimentally-determined, fundamental anisotropy of eCFP (0.356) while just below that for eYFP (0.382) [48].

Table 1. Comparison of FRET Efficiency for CLY Constructs In Vitro and in Live Cells.

Data for each CLY protein was organized into rows as indicated in the first column. Subsequent columns contain: observed FRET (E_obs) measured in vitro using purified recombinant proteins in dilute buffer with solution fluorometry and at the plasma membrane in live cells using TIRF microscopy. The absolute FRET efficiency for each experiment was calculated from E_obs using γ –correction.

Protein	Linker Length	Eobs In Vitro	Eobs Live Cell	FRET In Vitro	FRET Live Cell
CLY1	23	0.70 ± 0.01	0.74 ± 0.05	0.61 ± 0.01	0.65 ± 0.05
CLY5	47	0.54 ± 0.02	0.58 ± 0.01	0.44 ± 0.02	0.48 ± 0.01
CLY9	71	0.48 ± 0.04	0.44 ± 0.04	0.37 ± 0.04	0.35 ± 0.04

Molecular Dynamics Simulations of the CLY Proteins

We used discrete molecular dynamics (DMD) simulations [47] to map the conformational ensembles for the CLY proteins and elucidate their dependence on the interdomain linker. For comparison, we also performed DMD simulations of the linkers lacking the FP tags (L1, L5 and L9). A comparison of the autocorrelation plots reveals that the presence of the FP tags slowed the conformational dynamics of the CLYs, relative to the bare linkers, by more than an order of magnitude (Figure 2A). We used implicit solvent model in our simulations, which resulted in faster reconfiguration time than in experiments. The implicit solvent in simulations removes the dampening due to collision with solvent and accelerates the sampling of the phase space by at least 2 orders of magnitude of faster dynamics [47]. However, the relative values are self-consistent and show that the added mass of the FPs skews the linker dynamics away from that of a bare polypeptide. Due to the exponentially increased conformational space with increasing linker length and the slowed dynamics with FP tags, our DMD simulations of CLY9 failed to converge and were not included in the analyses.

Figure 2. Discrete Molecular Dynamics Simulations of the CLYs and Their Interdomain Linkers.

(A) Autocorrelation of linker end-to-end distances in CLY1 (blue) and CLY5 (orange) along with those for the bare linkers of these two variants (L1, L5; dashed lines). (B) Probability distribution of the linker length (end-to-end distance) in the control linker sequence, Linker1, Linker5, Linker9 and in CLY1 and CLY5 FPs (C) Probability distribution of interchromophore distance (D) Probability density distribution function (PDDF) of the average orientation factor (κ²) between donor-acceptor pairs in CLY1 and CLY5 compared to an ideal isotropic distribution.

Simulation of the linkers without FPs revealed the probability distributions of linker end-to-end distance with root mean squared (RMS) extensions of 2.5 nm for L1, 3.7 nm for L5 and 4.5 nm for L9 (Figure 2B). A plot of the RMS end-to-end distances of the linkers against their length in residues followed a power-law relationship with a scaling exponent of ~0.51 (Figure S1), which is consistent with the polymer scaling of a polypeptide under theta conditions [49, 50]. The presence of the FPs shifted the probability distribution of the linker end-to-end distance towards longer values for both CLY1 and CLY5. The RMS linker end-to-end for CLY1 increased ~24% to 3.1 nm while CLY5 showed a broad distribution with a RMS length of 4.4 nm, a ~19% increase.

We modeled the chromophores into eCFP and eYFP in each of the CLY conformations to relate the DMD simulations to the FRET data, which yields the RMS interchromophore distance. In contrast to the linker end-to-end distance distribution, the simulated interchromophore distance distribution contained two distinct peaks for both CLY1 and CLY5. A small population at shorter distance was centered at 3.7 nm and 4.0 nm for CLY1 and CLY5, respectively (Figure 2C). These closed states involved contact between the FPs (Figure S2). The predominant population was a broad peak centered near 6.1 nm for CLY1 and 7.0 nm for CLY5. The change in the relative areas of the two peaks in the interchromophore distance distribution suggests that the longer linker decreased the occupancy of the contact state from 37% in CLY1 to 16% in CLY5 and also broadened the interchromophore distance distribution outward to longer distances.

To aid interpretation of the FP-FRET as distance, we computed the orientation factor (κ²) using the angles relating the transition dipole moments of eCFP and eYFP within CLY1 and CLY5 (Figure 2D). The calculated, instantaneous κ² values spans the entire range. This agrees with predictions, based on our measured anisotropy values, which estimate a minimum κ² of 0.02 and a maximum κ² of 3.81 [45, 48, 51]. The κ² distributions for CLY1 and CLY5 were similar to the ideal distribution of κ², estimated from completely random orientations, but had a slightly increased probability of high κ² values (Figure 2D). The weak deviation from the isotropic distribution is likely due to the contact states involving the FPs. As a result, the mean κ² was 0.703 for CLY1 and 0.694 for CLY5, which were slightly higher than the value of 2/3 for a dynamic isotropic distribution and well above the value of 0.476 for a static isotropic distribution.

To identify the intramolecular contacts between eCFP and eYFP in the compact state observed in DMD, we used a cutoff interchromophore distance of 50 Å. There are minimal contacts in states with longer separation. Within this subpopulation with FPs in contact, the relative orientation of eCFP with respect to eYFP was highly variable resulting in a diverse ensemble that could not be described by a single predominant conformation (Figure S2). We computed the binding probability of each residue in eCFP and eYFP in the contact states using a cutoff distance of 8.5 Å between Cα atoms to define a contact. A plot of the binding probability per residue (using the residue numbering of the monomeric FPs) reveals contacts throughout the primary sequence (Figure 3A & C). Because they share β-barrel structures with nearly identical solvent-exposed residues, eCFP and eYFP have similar inter-domain contact profiles in both CLY1 (Figure 3A) and CLY5 (Figure 3C). Mapping the frequent contacts onto the structures of eCFP and eYFP reveals a broad interaction interface spread across one face of the FP barrels (Figure 3 B & D for CLY1 and CLY5, respectively). Residues with high inter-domain contact frequencies include hydrophobic residues (e.g., V176, L178, A206) and aromatic residues (e.g., Y39, Y151, F223) along with polar and charged residues (e.g., E142, K156, D173, D197, D210 in YFP), suggesting that the inter-domain interfaces are stabilized by a mixture of electrostatic and hydrophobic interactions.

Figure 3. Intramolecular- Contacts between Fluorescent Proteins in DMD Simulations.

(A) Average binding probability of FPs in CLY1. Residue numbering is that of monomeric FPs to facilitate comparisons. A binding probability cutoff of 0.05 (grey dashed line) was used to define residues making frequent contacts. Location of frequent intramolecular contacts in CLY1 shown in side and top views of (B) eCFP and (C) eYFP with residues are rainbow colored according to their binding probabilities. (D Average binding probability of FPs in CLY5. Residue numbering is that of monomeric FPs. Location of intramolecular contacts in CLY5 shown in side and top views of (E) eCFP and (F) eYFP.

Single Molecule Examination of the FRET Distribution

DMD simulations suggested that CLY proteins existed in a heterogeneous mixture of contact and non-contact states, which should have different FRET efficiency. To qualitatively examine the distribution of FRET efficiency, we measured single molecule FRET using TIRF microscopy with camera detection. Purified CLY proteins were encapsulated in liposomes, which were then attached to a passivated microscope slide and imaged with TIRF microscopy under 458 nm illumination. To confirm that liposome encapsulation does not affect CLY FRET, we measured ensemble E_obs for protein encapsulated in liposomes, which found no significant differences in E_obs attributable to the liposomes (Fig S3).

With a single laser, we were unable to use alternating excitation to identify molecules with active donor and acceptor fluorophores. To circumvent this issue, raw traces from single CLY molecules were filtered based on the presence of non-zero emission in both channels and single step photobleaching, which is only possible for a single CLY tandem with two active chromophores (Figure 4A). Only those molecules lasting longer than one second were selected for this analysis. E_obs was calculated for each frame until the first dark state or photobleaching occurred.

Figure 4. Single Molecule Fluorescence Microscopy of CLY Constructs.

A) Representative traces of fluorescence emission over time for CLY1 (top panel), CLY5 (middle panel) and CLY9 (bottom panel) with CFP emission in cyan and YFP emission in yellow. B) Distribution of observed FRET efficiency for CLY1 (top panel, n=523), CLY5 (middle panel, n=625), and CLY9 (bottom panel, n=325). FRET efficiency was calculated independently for each 100 millisecond time bin.

For all three proteins, the distributions showed E_obs across the range of possible values (Figure 4B). Previous single molecule analysis of FP-FRET has also reported broad or irregular distributions [52, 53] and observed the coexistence of multiple states within the conformational ensemble [54]. For CLY1, ~44% of the population had E_obs > 0.8 while this dropped to ~33% for CLY5 and 23% for CLY9. However, removal of apparent donor-only and acceptor-only molecules would also remove some actual low and high FRET molecules. For these reasons, the smFRET distributions are only interpreted qualitatively to confirm the presence of contact and non-contact states.

Hydrodynamic Measurements of the CLYs

As another estimate of the effect of linker length on the molecular dimensions of the CLY proteins, we used analytical size exclusion chromatography (SEC) to measure the hydrodynamic radii of the purified CLY proteins (R_SEC). We measured elution volumes of 1.60 ± 0.01 mL for CLY1, 1.50 ± 0.02 mL for CLY5 and 1.40 ± 0.01 mL for CLY9 (Mean ± SD; Figure 5A). From these elution volumes, we calculated R_SEC relative to a series of globular molecular weight standards [55]. This gave R_SEC values of 3.75 ± 0.01 nm for CLY1, 3.90 ± 0.03 nm for CLY5 and 4.03 ± 0.01 nm for CLY9 (Table 3).

Figure 5. Molecular Dimensions of CLY Constructs.

A) Representative elution profiles for CLY1 (blue), CLY5 (orange) and CLY9 (black) from analytical size exclusion chromatography. Elution volume was normalized to the void volume of the column (V_e/V_o). B) Sedimentation velocity analysis of CLY1 (blue), CLY5 (orange) and CLY9 (black). The c(s) continuous sedimentation coefficient distributions were produced using SEDFIT and fit to a single ideal species model using SEDANAL. Data were normalized by peak height. C) Comparison of estimates for the hydrodynamic radii. From SEC (open circles), AUC (open squares) and DMD (magenta). D) Interchromophore distances calculated from FRET using different assumptions about dynamics. The details of each model are provided in the main text. E) Comparison of interchromophore distances from the DMD ensemble. Shown for the population mean (cyan), the sub-ensemble with FPs in contact (grey), and the sub-ensemble in non-contact states (black). These are compared to selected FRET distances as indicated. Simple linear fits are included to visually highlight the differences in the datasets. Polymer scaling exponents ranged from 0.34 to 0.47 depending on the assumptions of dynamics (Fig. S4).

Table 3. Comparison of Measured and Calculated Hydrodynamic Dimensions for CLY Constructs.

Data for each CLY protein is organized into rows as indicated in the first column. Distances are reported in nanometers (nm). The Stokes radii were measured with analytical ultracentrifugation (R_AUC). The hydrodynamic radius was measured with analytical size exclusion chromatography (R_SEC). The error bars indicate the propagated experimental SD. The effective hydrodynamic radius was calculated from the DMD ensemble using HYDROPRO (R_DMD) [56]. The error bars indicate the standard deviation of observed radius within the conformational ensemble.

Protein	Linker Length	RAUC	RSEC	RDMD
CLY1	23	3.13 ± 0.03	3.75 ± 0.01	3.61 ± 0.31
CLY5	47	3.31 ± 0.02	3.90 ± 0.03	3.67 ± 0.54
CLY9	71	3.46 ± 0.02	4.03 ± 0.01	ND

To provide a complementary estimate of the hydrated dimensions of the CLY proteins, we used sedimentation velocity analytical ultracentrifugation (AUC). The hydrodynamic properties of the three CLYs were similar with all three sedimentation coefficients near 4 S (Figure 5B). Lengthening the linker would be expected to increase the sedimentation coefficient because the molecular weight has increased. However, the opposite trend was observed with a slight, but statistically-significant decrease in s_20,w. Sedimentation velocity experiments recovered the frictional ratio (F/f₀), which provides an estimate of the deviation from a sphere. This gave frictional ratios of 1.24 ± 0.01, 1.30 ± 0.01 and 1.35 ± 0.01 for CLY1, CLY5, and CLY9, respectively, corresponding to slightly more extended conformations for the longer linkers. We calculated Stokes Radii (R_AUC) of 3.13 ± 0.03 nm for CLY1, 3.31 ± 0.02 nm for CLY5 and 3.46 ± 0.02 nm for CLY9 (Table 3).

Both methods estimate the same physical parameter and should be considered equivalent. We have adopted the different nomenclature according to the standard practice associated with each technique as a means of distinguishing the results. Nonetheless, we observed a systematic difference with SEC reporting a radius 6 ± 0.2 Å larger than AUC (Fig 5C). To compare DMD simulations with experimental estimates, we used the program HYDROPRO to predict the hydrodynamic coefficients using a surface-shell based model [56] for the simulated conformational ensembles of CLY1 and CLY5, which included both contact and non-contact states. Based on the DMD-derived structural ensembles, we obtained the average hydrodynamic coefficients with standard deviations of 3.61 ± 0.31 nm for CLY1 and 3.67 ± 0.54 nm for CLY5. The mean values from HYDROPRO agreed reasonably well with estimates from AUC given the large range within the simulated ensemble (Figure 5C).

Calculation of FRET Distances

The first step to calculating reliable distances from ensemble solution E_obs is to obtain the absolute FRET efficiency. This requires γ normalization to correct for differences in the quantum yields of donor and acceptor fluorophores [57–59]. We used published values for the quantum yields of eCFP and eYFP [60]. In addition to distance, the FRET efficiency depends strongly on fluorophore orientation and dynamics, which is encompassed by the orientation factor κ² [46]. Assigning a value to κ² requires that one choose a model for the averaging behavior of the FPs. We calculated RMS distances using the common assumption that the FPs are dynamically reorienting on the timescale of fluorescence emission where κ² = 2/3, which we termed <D_isotropic>. Next, we chose the well-known, but less used, assumption that the FPs are static where κ² = 0.476 for a static isotropic distribution [51], which we termed <D_static>. Along these same lines, we related FP-FRET to RMS distance using a published look-up table generated from Monte Carlo simulations of a static ensemble of FP-FRET pairs [61], which we termed <D_MC>. Finally, we used the mean values of κ² from each static conformation in the simulated DMD ensemble, which we termed <D_DMD>. Moving away from the isotropic assumption decreased the calculated RMS interchromophore distances while the mean κ² from DMD gave similar RMS distance estimates to the isotropic assumption (Table 4).

Table 4. Comparison of Calculated Interchromophore Distances for CLY Constructs.

Data for each CLY protein is organized into rows as indicated in the first column. The interchromophore distances (in nm) were calculated from the FRET efficiency using several approaches: the κ²= 2/3 approximation for isotropic rotation (<D_isotropic>); the κ²= 0.475 approximation for a static ensemble (<D_static>); a published lookup table relating FP-FRET to distance based on Monte Carlo calculations of a static FP ensemble [61] (<D_MC>); using the average κ² from the ensemble of conformations observed in DMD (<D_DMD>). These are compared to the RMS population-mean interchromophore distance observed in DMD simulations (<D_{DMD Mean}>). ND, not determined as the DMD simulations were not successful for CLY9.

Protein	FRET Efficiency	<Disotropic>	<Dstatic>	<DMC>	<DSAW>	<DDMD>	<DDMD Mean>
CLY1	0.61 ± 0.01	4.46 ± 0.03	4.21 ± 0.03	3.81 ± 0.05	4.57 ± 0.02	4.5 ± 0.03	5.2 ± 1.5
CLY5	0.44 ± 0.02	5.01 ± 0.1	4.74 ± 0.1	4.56 ± 0.1	5.60 ± 0.04	5.0 ± 0.1	6.2 ± 1.8
CLY9	0.37 ± 0.04	5.24 ± 0.1	4.95 ± 0.1	4.85 ± 0.14	6.15 ± 0.15	ND	ND

The CLY proteins undergo conformational dynamics that are faster than the timescale for ensemble spectroscopy. Thus, our <D> represents the time average of a conformational distribution. For denatured proteins and unstructured polypeptides, the form of the conformational distribution is frequently obtained from simplified polymer models [62]. This has proven valuable for experiments involving small, organic fluorophores attached to polypeptides [63–65]. DMD and smFRET suggest that the CLY ensemble contains a mixture of bound and unbound states so the linker scaling was distorted from that of bare polypeptides. Both the dynamic timescale and underlying conformational distribution likely deviate from any ideal polymer model. Nonetheless, we also estimated the interchromophore distance from the FRET efficiency using a self-avoiding random walk model for the averaging <D_SAW> [63]. This treatment had the largest effect particularly at low FRET where distance estimates based on FRET were greatly increased (Figure 5D).

An important parameter to characterize polypeptide expansion is the scaling coefficient for the radius of gyration (R_g) of the polypeptide, which can be complicated to obtain from the end-to-end distance measured with FRET even with small organic dyes [65, 66]. In contrast to studies with organic dyes, the linkers account for only 1.7%, 4.5% and 7.1% of the molecular weight of CLY1, 5 and 9, respectively. The linker contribution to R_g distribution of CLY proteins would be small. Obtaining the end-to-end distance from FRET requires a model that relates the chromophore positions to the linker termini, which is complicated for the CLYs. The chromophore is buried with the FP core and each FP has hindered rotation about its terminus. Previously, the distribution of the chromophore relative to the linker termini was calculated as an ideal sphere, but this was noted to be unrealistic [26, 67]. Our DMD now confirmed this simple model to be inaccurate (Table 2). Because we lack an appropriate model to determine linker extension from FRET data, we are only able to report the apparent polymer scaling in terms of the interchromophore distance (Figure S4). Notably, the SAW model had the effect of generating an apparent scaling coefficient of 0.47, while all the other treatments resulted in apparent scaling coefficients ≤ 0.4.

Table 2. Mean Intramolecular Distances for the CLY Constructs from Discrete Molecular Dynamics.

Distances are reported in nanometers (nm). <V_c> is the RMS distance between the eCFP chromophore and the linker N-terminus. <V_y> is the RMS distance between the eYFP chromophore and the linker C-terminus. <R_c> is the RMS distance between the two chromophores. <R_e> is the RMS end-to-end distance for the linker. Model <R_e> is the calculated end-to-end distance for the linker using a published geometrical model to relate the interchromophore separation to the linker end-to-end distance as $\langle R_e\rangle =\sqrt{\left(\langle V_c^2\rangle +\langle V_y^2\rangle +\langle R_c^2\rangle \right)}$ [26]. This geometrical calculation using these parameters does not recover the observed RMS end-to-end distance for the linker.

Parameter	CLY1	CLY5
<Vc>	2.0 ± 0.1	2.0 ± 0.1
<Vy>	2.2 ± 0.1	2.4± 0.1
<Rc>	5.2± 1.5	6.2 ± 1..8
<Re>	3.1 ± 1.0	4.4 ± 15
Model <Re>	4.4	5.6

Sensitivity to Macromolecular Crowding.

Having measured FP-FRET from the CLYs in the dilute solution, we next examined their sensitivity to macromolecular crowding and solvent quality. The concentration of biological molecules in the cytoplasm of cells has been estimated to be in the range of 0.08–0.4 gm/ml [68]. Crowding in living systems can have a dramatic effect on protein structure and function [69], which is not captured by measurements in dilute solution. The presence of additional macromolecules creates inaccessible space in the solution, which is called the excluded volume [70]. To estimate the sensitivity of the CLYs to crowding, we used polyethylene glycol (PEG) because of its inert nature and previous use in studies of crowding [12]. We measured ensemble fluorescence spectra of the three CLYs in increasing concentrations of PEG and compared two different molecular weight PEGs (PEG 400 and PEG 3000) to understand the role of molecular size in crowding.

In the presence of PEG 3000, all three CLYs showed an increase in the observed FRET efficiency (E_obs) with increasing PEG concentrations (Figure 6A). At high concentration of PEG 3000, all three CLYs showed E_obs close to 0.8, which would likely correspond to the contact state observed in DMD (Figure S5). The concentration dependence of these transitions was well fit to a sigmodal function. The transition midpoints were similar but showed a trend towards higher PEG concentrations required with increased linker length (Figure 6A). In contrast to the compaction induced by PEG 3000, PEG 400 gave a linear decrease in E_obs for all three CLYs suggesting increased FP separation (Figure 6B). For CLY1, E_obs decreased by 25% in 0.4g/ml PEG 400, whereas CLY5 and CLY9 each showed greater than 50% decrease in E_obs. The contrasting observations in the presence of PEG 3000 and 400 reveals the importance of the size and shape of the crowding agents on protein conformation in a crowded environment [68].

Figure 6: Sensitivity of the CLY Constructs to Solvent Conditions.

(A) The percent change in observed FRET efficiencies (% E_obs) in increasing concentrations of PEG 3000. CLY1 (blue), CLY5 (orange) and CLY9 (black), respectively. The final E_obs in buffer was set to 100% to facilitate comparisons between CLYs. For the remaining panels that show decreasing E_obs, the starting E_obs in buffer was set to 100% (B) The change in E_obs as a function of increasing concentration of PEG 400. (C) The change in E_obs as a function of increasing concentration of urea. (D) The change in E_obs as a function of increasing concentration of sodium chloride. Error bars represent SD from 3 replicate experiments under each condition. Values from the fits along with the raw E_obs curves are shown in supplemental material (Figure S5).

Intramolecular interactions within a protein are sensitive to solvent quality. As solvent quality increases, the residue interactions with solvent are favored over interactions with other residues within the protein [71]. Here, we used low concentrations of urea to enhance the solvent quality relative to buffer and examined the effect on linker extension in the CLYs. We measured ensemble fluorescence spectra of the CLYs in an increasing concentration of urea up to 4M, which is still below denaturing concentrations for the FPs. All CLYs showed a decrease in E_obs with increasing urea concentration (Figure 6C). CLY1 showed a slight decrease in E_obs that was linear across the entire concentration range. CLY5 and CLY9 showed a similar sensitivity to urea that was greater than observed for CLY1. However, the expansion of CLY5 and CLY9 became non-linear above 2M urea and appeared to level off at different % E_obs values with CLY9 decreasing further than CLY5. However, both proteins leveled off at similar E_obs values (Figure S5).

DMD simulations identified charged residues involved in intramolecular contacts between FPs. Increasing the salt concentration should screen such charge interactions between the FPs and consequently affect the conformation of the CLYs. To test this, we measured ensemble fluorescence spectra of the CLYs in an increasing concentration of sodium chloride. All CLYs showed decreasing E_obs with increasing salt that was well fit to a single exponential decay (Figure 6D). As with urea, the sensitivity of CLY1 was much lower than CLY5 and CLY9, which were very similar. The expansion of CLY5 was more sensitive to salt than observed in urea. This suggests that screening the charge interactions within the FP contact state allows the FPs to increase their separation.

The monotonic and non-monotonic changes of E_obs for the CLYs in response to PEG, urea and salt demonstrate the unique impacts of solvent on protein structure. In particular, the occupancy of the intramolecular contact state in these CLYs appears highly sensitive to solvent conditions. Thus, these CLYs can be considered potent probes that undergo conformational changes in response to their environment. The sensitivity of the CLYs to this broad range of solvent conditions and crowding agents is similar to that recently observed for FP tandems composed of mTurquoise2 and mNeonGreen [72].

Comparison of FRET Measured in Live Cells to Solution Measurements

Having established that the CLY proteins were sensitive to the chemical environment, we examined the effect of the cellular environment on FP-FRET. To measure FRET in cellulo, we expressed the CLY constructs in CHO cells and targeted them to the plasma membrane (Figure S6) by appending the N-terminus of Lyn-kinase, which is both myristoylated and palmitoylated [73, 74]. Fluorescence emission was collected using through-objective Total Internal Reflection Fluorescence (TIRF) to confine illumination to the plasma membrane. Four replicate images were used to estimate direct excitation and to calculate E_obs: 1) eCFP and 2) eYFP channels with 458 nm excitation, 3) eCFP and 4) eYFP channels with 514 nm excitation (Figure 7A).

Figure 7. Live Cell Fluorescence Microscopy of CLY Constructs.

A) Representative series of image for CHO cells expressing plasma membrane-targeted CLY9. The first and second panels show replicate images after ratiometric separation of emission from CFP and YFP under TIR excitation at 458 nm. The third and fourth panels show replicate images after ratiometric separation of emission from CFP and YFP under TIR excitation at 514 nm. B) Measured E_obs is plotted against the linker length in residues using the values from Table 1. Ensemble solution fluorometry (orange); Live cell microscopy (cyan). Simple linear fits are included to visually highlight the similarity between the datasets.

For each cell, we calculated E_obs from three randomly-selected regions on the plasma membrane. As each region contains multiple copies of CLY, these represent ensemble FRET. The mean E_obs values for plasma membrane targeted CLYs were 0.74 ± 0.05 for CLY1, 0.58 ± 0.01 for CLY5 and 0.44 ± 0.04 for CLY9 (Mean ± SD; Table 1). The fluorescence intensity of the selected cells varied by over an order of magnitude, which allowed measurements across an order of magnitude range in protein concentration. We found that E_obs was constant over the observed range of protein expression levels. These values are in good agreement with the E_obs measured for recombinant CLY proteins in vitro. Thus, E_obs calculated from the plasma membrane of live cells and solution fluorometry gave very similar estimates of FRET efficiency (Figure 7B).

DISCUSSION

The eCFP-Linker-eYFP (CLY) constructs were created to study the properties of their interdomain linkers [26]. Our DMD simulations (Figure 2) and single molecule FRET (Figure 4) measurements showed that these constructs sample a complex distribution with intramolecular contacts between the FPs playing a major role in the conformational ensemble (Figure 3). As such, while the bare linkers showed scaling consistent with a polymer in ideal solvent (Figure S1), the linkers in the CLYs were distorted to longer lengths (Figure 2). This was due to fuzzy FP interactions, which sampled a wide variety of conformations with contacts across the surface of the proteins (Figure 3). These contacts did involve A206, which is typically mutated to lysine in later generations of these FPs to further reduce their affinity for other FPs [73]. This mutation presumably decreases these specific intramolecular contacts but would not eliminate the other widespread interactions that occur when the FPs are tethered at such high local concentrations. In agreement with this notion, FP-FRET measurements of FP association in live cells found that A206K decreased interactions between non-tethered FPs in the cytoplasm but did not eliminate FP-FRET when crowding was increased [75].

Both DMD and smFRET suggest that occupancy of the contact state depended on the linker length. Thus, the effects of the linker on contact formation contribute to the overall FRET dependency on linker length. Additionally, the increased linker length allows for somewhat increased chromophore separation in the non-contact states (Figure 2). Thus, the FRET in these CLY constructs depends on both the occupancy of contact states and the linker length distribution of non-contact states. Unfortunately, the true solution occupancy remains unknown because the smFRET histograms had to be filtered. The utility of the CLYs as probes of inherent linker properties is diminished because the tethered FPs adopt a supertertiary ensemble based on protein interactions. Nonetheless, the CLYs are still useful to explore the relationship between FP-FRET and distance, and as environmental sensors.

The relationship between FRET and distance depends on chromophore dynamics [51]. The slow rotation of large FPs [76, 77] has been used to suggest that FP-FRET falls into the static averaging regime [45, 46]. However, both DMD and fluorescence anisotropy suggested that the CLYs do undergo some conformational dynamics. This places FP-FRET into the intermediate case, which would require knowledge of the orientational distribution [46]. We used mean <κ²> from the DMD conformational ensembles, which was close to the isotropic value for κ², and recovered the isotropic distance estimates (Table 4). Comparing distance estimates using different assumptions about dynamics revealed that the static assumption leads to shorter distances without a large change in the distance dependency on linker length. In contrast, using published look up table based on Monte Carlo calculations of a static, isotropic ensemble steepened the distance dependency by further lowering distance estimates at high FRET. Using a self-avoiding random walk model [63] to estimate the FRET distance relationship had the greatest effect on the distance scaling particularly at low FRET.

Polypeptide dimensions are characterized by the polymer scaling exponent that relates the linker end-to-end distance back to the number of polymer units. For the CLYs, we lack a model to relate the interchromophore distance estimated from FRET to the linker end-to-end distance. Thus, we cannot report an experimental estimate for linker dimensions. Previously, the chromophore was assumed to rotate freely about the linker termini following a spherical distribution [26]. However, our DMD simulations did not recover a regular distribution that can easily relate these points using a simple model. Nevertheless, we can still fit the interchromophore distances as a best estimate of the apparent polymer scaling. This reveals that the SAW model gives a polymer scaling exponent of 0.47 for the CLY series while all other estimates yield an apparent polymer scaling exponent ≤ 0.4 (Figure S4), which is more consistent with the CLYs acting as globular proteins. This highlights that the tandem FPs represent a supertertiary ensemble determined by the balance of linker properties and interactions between folded domains.

The DMD simulations are useful in describing the forces underlying the conformational ensemble. However, these simulations were not set up to fully reproduce protein-protein interactions in experiments, primarily due to treating eYFP as a rigid body along other approximations to reduce computational cost (see EXPERIMENTAL PROCEDURES). Thus, while general aspects of the two state model are likely correct, it does not provide an exact estimate of the occupancy of the bound state in experiments. One can see that the distance dependence on linker length in DMD varies between the subensembles in contact and non-contact states (Figure 5E). The experimental FP-FRET distances fall between these two extremes regardless of which model is used to estimate the FRET distance. Nonetheless, these results show that FP-FRET correctly recovers the qualitative differences between the CLY series within the great uncertainty of which model to use for the dynamics.

For an independent estimate of the molecular dimensions of the CLYs, we measured the hydrodynamic radius for each CLY protein with both SEC and AUC (Figure 2, Table 3). The hydrodynamic radii showed a more shallow dependency on linker length than the interchromophore distances. This was true for both experimental data and DMD simulations using HYDROPRO to calculate the predicted hydrodynamic radius from representative models [56]. The HYDROPRO estimates were in better agreement with AUC than SEC, which show similar scaling but systematically differ. The absolute values for the hydrodynamic dimensions are systematically higher than the interchromophore distances, but this fails to account for the chromophore being located within the protein interior [78] (Figure S7). The FPs are β-barrels of roughly 2.5 by 4 nm [79]. Depending on the relative orientation of the FPs, the chromophores are 1 nm or more from the surface (Table 2). This additional distance to the protein surface comes close to accounting for the systematic deviation.

Numerous reports have characterized the FP-FRET dependence on linker properties in dilute buffer conditions [25, 26, 80] while others have characterized the effects of crowded solutions [54, 72, 81–84]. To understand the environmental sensitivity of the CLYs, we measured the dependence of FP-FRET on solvent conditions. This revealed that the CLYs are sensitive to macromolecular crowding, which may affect occupancy of the contact states (Figure 6). Previous studies of FP-FRET in live cells showed that increased crowding enhances weak protein interactions [75]. A previous study of FP-FRET also reported a slight sigmoidal character to the crowding transition induced by Ficoll [83]. The large molecular weight PEG may be excluded from the intramolecular contact interface such that increasing [PEG] forces all the CLYs into similar contact states (Figure 6A). In contrast, the smaller PEG reduces the occupancy of the crowded state and drives the FPs apart (Figure 6B). Previous FP-FRET studies of crowding have also found that small molecules favor expansion [83]. DMD revealed intramolecular interactions between the FPs (Figure 3). Solvent quality determines the balance between intrachain interactions and interactions with solvent [50]. Increasing the solvent quality with urea appears to disfavor the contact state leading to expansion (Figure 6C). Similarly, high salt concentrations screened electrostatic interactions identified by DMD resulting in lower FRET (Figure 6D). The varied response of the CLY to different solvents establishes their broad environmental sensitivity and is similar to other FP-FRET sensors [72].

We expressed the CLY proteins in live cells and targeted them to the plasma membrane (Figure S6) to measure FP-FRET with TIRF microscopy. The protein was lipid modified at the N-terminus, which is more than 20 residues from the start of the FP. Thus, the protein should be extended some nanometers from the membrane surface. We obtained good agreement between in vitro and in cellulo measurements of γ-corrected FRET efficiency for all three CLYs (Table 4). It should be noted that parameters in the Förster radius, such as the spectral overlap, quantum yield and refractive index, could differ between live cells and dilute buffer and were not measured by us in cellulo. However, the molecular brightness of the progenitor eGFP was found to be identical in the nucleus, the cytoplasm, and in vitro [85]. It is still possible that photophysical effects could differ in such a way as to give the same γ-corrected FRET efficiency despite a different underlying conformational distribution. Nevertheless, a simple possible conclusion is that the CLY proteins adopt the same conformational distribution in vitro and in cellulo. Since we only measured FP-FRET at the plasma membrane, which is a specialized environment, it is not possible to extrapolate these results into a general statement. This lack of response may not be conserved throughout the cell. Other FP-FRET sensors developed to detect cellular crowding have reported differences in FP-FRET between proteins in dilute solution and proteins diffusing in the cytoplasm [81, 83]. The magnitude of this effect was estimated to be the equivalent of 20% crowding agent, which yielded significant changes in the “FRET ratio” of their reporter. The transitions induced by crowding in these CLY constructs showed little responsiveness until concentrations above 20% so we may not be sensitive in this range. Additionally, we measured FRET from spectroscopy and cell imaging using the same spectral band passes and corrected for differences in leakage and detection efficiency. Applying similar corrections could lower this estimate of the crowding effect in live cells [81, 83].

We note that this is not the first case where the cellular environment had no discernable effects on molecular structure [86–88]. A similar lack of structural effects was also reported for fluorescently-labeled polyethylene glycol (PEG) in live cells and dilute aqueous buffer [89]. Thus, the effects of molecular crowding and the cellular environment are clearly molecule specific. The low complexity GGS repeats used in the present study are unlikely to be a target of posttranslational modifications or protein interactions that might alter the linker conformation. Additionally, the intramolecular interactions between the FPs had a predominant effect on the conformational ensemble. Given that the high local concentration of the tandem FPs imparted by the linkers persists in cellulo, such supertertiary interactions between folded protein domains may be less sensitive to the cellular environment.

EXPERIMENTAL PROCEDURES

Expression and purification of recombinant proteins

The CLY constructs were a generous gift from Dr. Maarten Merkx [26]. From these plasmids, we created a 6 Histidine-tagged eCFP by introducing a stop codon before the linker. We also created a 6-histidine tagged eYFP by PCR amplifying the gene for eYFP and recloning it back into the empty vector. All constructs were expressed in E. coli at 30 °C. Constructs were purified using nickel affinity, anion exchange and size exclusion chromatography. Before all experiments, the 6-His tags were removed by protease cleavage. Absorbance measurements for purified CLY proteins suggested an equimolar ratio of eCFP and eYFP (1.0 ± 0.2).

Ensemble fluorescence spectroscopy

Ensemble fluorescence spectra were recorded on an ISS PC1 spectrofluorometer using a 1.0 mm excitation slit and a 0.5 mm emission slit. For eCFP and CLYs, samples were excited at 458 nm. For eYFP alone and eYFP within the CLYs, samples were excited at 514 nm. Raw spectra were background corrected using a buffer-only control spectrum. Anisotropy measurements were collected using Glan Thompson polarizers in the L conformation.

Calculation of Apparent FRET Efficiency for Solution Fluorescence

To maintain consistency with microscopy experiments, we calculated FRET from ensemble fluorescence using the summed emission intensity within the wavelength ranges of the emission spectra that correspond to the transmission bands of the Donor and Acceptor band pass filters used in live cell microscopy. As such, the donor intensity (D) was taken as the summed, background-subtracted emission between 470–500 nm. The raw acceptor intensity (A_raw) was taken as the summed background-subtracted emission between 540–600 nm, which was corrected 1) to remove the contribution from direct excitation of YFP at 458 nm and 2) to remove the contribution from leakage of CFP emission into the YFP bandpass (A_corr).

Corrections for direct excitation and donor leakage into the acceptor channel were determined using the measured emission spectra of purified, recombinant eYFP and eCFP. To estimate direct excitation of YFP at 458 nm, we measured emission spectra of purified, recombinant eYFP and calculated the ratio of emission in the Acceptor bandpass under excitation at 458 nm and 514 nm $\left(\frac{\text{YFP}{\left(458\right)}_A}{\text{YFP}{\left(514\right)}_A}\right)$, which was 7.3 ± 0.3%. To estimate donor leakage, we measured the emission spectra of purified, recombinant eCFP and calculated the ratio of the emission transmitted by the Donor and Acceptor bandpasses during 458 nm excitation $\left(\frac{\text{CFP}{\left(458\right)}_{\text{A}}}{\text{CFP}{\left(458\right)}_{\text{D}}}\right)$ to be 22.5 ± 0.2%. Thus, the fully corrected acceptor emission intensity (A_corr) was calculated using these correction factors as:

$${\text{A}}_{\text{corr}}=\text{CLY}{\left(458\right)}_{\text{A}}-\underset{\text{Direct Excitation }}{\left[\text{CLY}{\left(514\right)}_{\text{A}}\times \left(\frac{\text{YFP}{\left(458\right)}_{\text{A}}}{\text{YFP}{\left(514\right)}_{\text{A}}}\right)\right]}-\underset{\mathit{\text{Donor Leakage}}}{\left[\text{CLY}{\left(458\right)}_{\text{D}}\times \left(\frac{\text{CFP}{\left(458\right)}_{\text{A}}}{\text{CFP}{\left(458\right)}_{\text{D}}}\right)\right]}$$

where the three letters indicate the construct, the number in parentheses indicates the wavelength of excitation and the subscript indicates the bandpass used to obtain the intensity.

To account for the fraction of the emission spectrum captured by our use of bandpasses to represent the entire spectrum (i.e. detection efficiency), the summed intensity of emission within the band pass was divided by the summed emission intensity across the entire spectrum. This gave a detection efficiency for CFP (η_D) of 47 ± 1 % and detection efficiency for YFP (η_A) of 40.4 ± 0.8 %. This gave a relative detection efficiency (η_A/D) of 0.85. Thus, the observed FRET efficiency was calculated as follows:

$$E_{obs}=\frac{{\text{A}}_{\text{corr}}}{{\text{A}}_{\text{corr}}+\left({\eta }_{A/D}\times \text{ D}\right)})$$

Calculation of Distances from FRET efficiency

Calculation of accurate distances from energy transfer experiments requires the absolute FRET efficiency, which requires correction by the full gamma factor (γ) to account for the differences in the quantum yields of donor (φ_D) and acceptor (φ_A) in addition to relative detection efficiency as follows:

$$\gamma = \left(\frac{{\eta }_A}{{\eta }_D}\right)\times \left(\frac{{\phi }_A}{{\phi }_D}\right)= {\eta }_{A/D}\times {\phi }_{A/D}$$

We used values of 0.61 and 0.4 for φ_A and φ_D, respectively [60], which gave a value of 1.525 for φ_A⁄D. This gave γ values of 1.3 for solution fluorescence and 1.6 for live cell imaging. Thus, absolute FRET efficiency for distance calculations was calculated as follows:

$$E_{abs}=\frac{{\text{A}}_{\text{corr}}}{{\text{A}}_{\text{corr}}+\left(\gamma \times \text{ D}\right)}$$

Previously, Evers et al. calculated the Förster radius (R_o) for the CLYs to be 48 Å under the assumption of dynamic, isotropic rotation of the fluorophores with κ²=2/3 [26]. The slow rotation of FPs may better described as a static ensemble on the timescale of fluorescence emission with κ²= 0.475. This static assumption reduces R_o to 45.4 Å. Thus, we calculated the root mean square distance between chromophores <D> based on the γ-corrected, absolute FRET efficiency (E_abs) and the calculated Förster radius as follows:

$$\langle D\rangle = R_0{\left(\frac{1}{E_{abs}}-1\right)}^{1/6}$$

Analytical size exclusion chromatography (SEC)

Purified CLY proteins were run on a silica-based analytical size exclusion column (KW-802.5, Showa Denko America, New York, NY) and their elution times were compared to globular gel filtration standards (Bio-Rad, Hercules, CA) to determine the apparent molecular weight. The hydrodynamic radius was then calculated from the apparent molecular weight [55].

$$LOG\left(R_H\right) =\left(-0.204+0.357*LOG \left(M{W}_{app}\right)\right)$$

Analytical Ultracentrifugation (AUC)

Sedimentation velocity samples were loaded into double sector cells equipped with quartz windows. Measurements were carried out in a Beckman-Coulter XL-I analytical ultracentrifuge at 20 °C and 45,000 RPM. Hydrodynamic properties were calculated using Sednterp [90]. Sedimentation coefficient distributions were produced using SEDFIT with a regularization of 0.95 [91]. Data were fit to a single ideal species model using SEDANAL [92]. Confidence intervals were obtained using the F-statistic to define a statistically significant increase in the variance upon adjusting each parameter from its best-fit value. CLYs were monomeric but a small amount of a higher order species was detected in CLY5 (< 1%).

Discrete Molecular Dynamics

Discrete molecular dynamics (DMD) is a rapid and predictive molecular dynamics algorithm that has been widely used to study protein folding [93], conformational dynamics [94], and super-tertiary structure [95]. The interatomic potentials in the united-atom DMD simulations include solvation, van der Waals, hydrogen bond, and electrostatic interactions. The solvation energy in the implicit solvent simulations was calculated using the Lazaridis–Karplus EEF1 (effective energy function 1) model [96]. The screened electrostatic interactions between charged atoms were computed by the Debye-Hückel approximation, where a Debye length of 1 nm was used by assuming a water dielectric constant of 80 and a monovalent electrolyte concentration of 0.1 M. The distance- and angular-dependent hydrogen bond interaction was modeled using a reaction-like algorithm [97]. The DMD program is available via Molecules in Action, LLC (http://www.moleculesinaction.com/).

Atomistic DMD simulations were performed at 300 K to systematically probe the conformational ensemble of CLY FP tandems and elucidate the influence of linker length on the structure and conformational dynamics of CLYs. For comparison, we also performed DMD simulations of the three linkers without FP tags (L1, 23 residues; L5, 47 residues; L9, 71 residues). For each bare linker, we started from fully extended conformations and performed 10 independent simulations starting from randomized velocities. Simulations of L1 and L5 lasted for 400 ns. Simulations of L9 linker were extended to 700 ns to have adequate sampling of the conformational space, as evaluated by the convergence of end-to-end distance distributions.

To insure sufficient conformational sampling, we generated different starting conformations of the CLYs with randomized interchromophore distances and orientations. Specifically, eCFP (PDB ID: 2YDZ) and YFP (PDB ID 1YFP) were randomly rotated and positioned away from each other such that the distance, between C-terminus of eCFP and N-terminus of eYFP, followed a Gaussian distribution with a mean length of an ideal polypeptide under theta conditions. Then, for these non-clashing conformations, we used our in-house loop modeling tool to insert the linker sequence between the two termini. We constructed 200 starting structures for CLY1 and 300 structures for CLY5 and CLY9. For each of the starting structures, an independent DMD simulation was performed. To reduce the computational cost, we kept the β-barrel of eYFP static in our simulations, while the eCFP and linker were free to move. In addition, all buried residues in eCFP were substituted by glycine to save some extra computational cost. Finally, a structure-based Gō constraint was assigned between Cα atoms in the eCFP β-barrel to ensure that the folded structure was maintained during simulations. These mutations did not affect the intramolecular interactions between FPs. Each of the independent simulations lasted for 500 ns. In the case of CLY1 and CLY5, the convergence in terms of linker end-to-end distance and inter-domain center-of-mass distance distributions was achieved. CLY9 simulations failed to converge due to the increased conformational space made accessible by the longer linker along with the slowed conformational dynamics imparted by FPs. Therefore, we only present the results for CLY1 and CLY5.

DMD Analysis

We computed the linker end-to-end distances using T234 as the linker N-terminus while the C-terminus varies as S253, S277 and S301 for CLY1, CLY5 and CLY9, respectively. We also used these points to calculate the autocorrelation functions of the linker end-to-end distances with and without FP tags. The chromophore was not included in simulations, which instead used the three native amino acids before cyclization. To obtain the interchromophore distances, we aligned the crystal structures of eCFP and eYFP to their corresponding domains in CLY, for each CLY conformation in the ensemble, and used the resulting chromophore positions. The interchromophore displacement and corresponding distance were calculated between the centers of the five-atom ring. The relationship between modeled chromophores in eCFP and eYFP was also used to calculate the orientation factors (<κ²>) for each conformation [46, 98]. For each chromophore, the dipole was defined as the vector connecting the centers of the five-atom ring and six-atom ring. Figure S6 in the supplemental materials illustrates the relationship between linker end-to-end distance and the interchromophore distance. To determine the interdomain contact frequencies, we used a cutoff distance of 8.5 Å between Cα atoms to define a contact between two residues. For a given residue in both eCFP and eYFP, an interdomain contact was defined as those residues that make at least one contact with residues in the other FP. The hydrodynamic coefficients for each structure in the conformational ensembles of CLY1 and CLY5 including both contact and non-contact states were calculated using HYDROPRO [56] to compare with the experimentally determined values. Structure figures were prepared using PyMol molecular visualization program [The PyMOL Molecular Graphics System. Schrödinger, LLC].

Single Molecule Microscopy

Fluorescent proteins were encapsulated by extrusion in 200 nm liposomes composed of egg phosphatidylcholine with 0.1% biotinylated phosphatidylethanolamine (Avanti Polar Lipids, Alabaster, AL). Unencapsulated proteins were removed by desalting with Sepharose CL-4B (GE Healthcare Bio-Sciences, Pittsburgh, PA). Liposomes were attached to a passivated quartz slide coated with biotinylated bovine serum albumin and streptavidin. Single molecule fluorescence was collected with a prism-based TIRF microscope. The eCFP and eYFP emission was separated using an Optosplit image splitter containing a 515 nm dichroic mirror (Chroma, Bellows Falls, VT), a 483/32 eCFP band pass and a 571/72 eYFP band pass (Semrock, Rochester, NY). Emission was collected with an Andor iXon EMCCD camera (Andor Technologies, Belfast, UK) at a frame rate of 10 Hz. Samples were excited using the 458 nm line from an argon Ion laser. All experiments were conducted in 25 mM Hepes at pH 7.4 with 100 mM NaCl. Oxygen scavengers and triplet state quenchers were omitted as these did not improve fluorophore performance. Data was analyzed using home written MATLAB scripts. Single molecule traces of fluorescence emission were filtered based on having non-zero intensity in both channels and the presence of single step photobleaching. As with the ensemble FRET experiments, the Donor Leakage of 25%, the Acceptor Direct Excitation of 7% and the Relative Detection Efficiency of 1.1 were determined from single FP control experiments.

Live Cell Microscopy

Plasmids were transfected into Chinese Hamster Ovary (CHO) cells and imaged 24–36 hours after transfection at room temperature in Hanks’ balanced salt solution using a Nikon Eclipse Ti-S microscope equipped with the TIRF slider and a fiber-coupled argon ion laser (Lasos, Jena, Germany). Fluorescence emission was collected with a 60x oil-immersion TIRF objective and transmitted through a dual band 458/514 TIRF filter cube with a CFP Donor band pass of 485/30 and a YFP Acceptor band pass of 570/60 (Chroma, Bellows Falls, VT). Emission was then spectrally-separated using an Optosplit image splitter (Cairn Research, Faversham, England) containing a 500 nm dichroic mirror (Chroma, Bellows Falls, VT) and relayed to an iXon DU897 emCCD camera (Andor Technologies, South Windsor, CT). Nikon Elements was used to align the replicate images and to calculate the mean pixel intensity within 1 μm² regions of interest (ROI). Each cell was sampled with 3 spatially-resolved ROI. All intensity measurements were background subtracted using the mean pixel intensity of ROI located on untransfected cells. FRET was calculated separately for each ROI. Intensity values for eYFP under 514 nm excitation (and thereby protein concentration) varied by ~40 fold. Each construct was measured using 3 separate transfections on different dates. For each transfection, 20 cells were measured resulting in N=60 cells for each construct.

Calculation of Apparent FRET Efficiency for Live Cell Imaging

The donor intensity (D) was taken as the background-subtracted emission in the CFP channel after the ratiometric image splitter. The measured acceptor intensity (A_meas) was taken as the background-subtracted emission in the YFP channel after the ratiometric image splitter. The measured intensity was corrected (A_corr) as described above to remove the contribution from direct excitation of YFP and to remove CFP leakage into the YFP channel. To estimate direct excitation of YFP, we imaged cells transfected with plasma membrane-targeted eYFP (eYFP-MEM) and measured the ratio of emission in the Acceptor channel under excitation at 458 nm and 514 nm $\left(\frac{\text{YFP}{\left(458\right)}_A}{\text{YFP}{\left(514\right)}_A}\right)$, which was 5 ± 1 % with the filter sets in use. To estimate donor leakage, we imaged cells transfected with plasma membrane-targeted eCFP (PMT-CFP) and measured the ratio of the emission transmitted by the Donor and Acceptor channels during 458 nm excitation $\left(\frac{\text{CFP}{\left(458\right)}_{\text{A}}}{\text{CFP}{\left(458\right)}_{\text{D}}}\right)$, which was calculated to be 48 ± 3 % with the filter sets in use.

The transmission properties of the optical filters limit how much of the spectrum of fluorescence emission is collected, resulting in differential detection efficiency for the donor and acceptor. We estimated the detection efficiencies by multiplying the emission spectra of the fluorophores by the transmission properties of the optical filters and detector efficiencies of the EMCCD cameras as provided by the manufacturers [59]. Using our spectrum of purified eCFP gave a detection efficiency for CFP (η_D) of 36 %. Using our spectrum of purified eYFP gave a detection efficiency for YFP (η_A) of 37 %. Thus, the relative detection efficiency of the donor and acceptor (η_A/D) was 1.02.

ABBREVIATIONS:

AUC

analytical ultracentrifugation

CHO

Chinese hamster ovary cells

CLY

CFP-linker-YFP

DMD

discrete molecular dynamics

eCFP

enhanced cyan fluorescent protein

eYFP

enhanced yellow fluorescent protein

FRET

fluorescence resonance energy transfer

FP

fluorescent protein

IDR

intrinsically disordered region

PEG

polyethylene glycol

PMT

plasma membrane targeted

RMS

root mean squared

SDS-PAGE

sodium dodecyl sulfate–polyacrylamide gel electrophoresis

SEC

size exclusion chromatography

SD

standard deviation

TIRF

total internal reflection fluorescence illumination