Bioorthogonal Chemistry Enables Single-Molecule FRET Measurements of Catalytically Active Protein Disulfide Isomerase

Abstract

Folding of newly synthesized proteins in the endoplasmic reticulum is assisted by several families of enzymes. One such family is the Protein Disulfide Isomerases (PDIs). PDIs are oxidoreductases, capable of forming new or breaking existing disulfide bonds. Structural information on PDIs unbound and bound to substrates is highly desirable for developing targeted therapeutics, yet it has been difficult to obtain using traditional approaches because of their relatively large size and remarkable flexibility. Single-molecule FRET (smFRET) could be a powerful tool to study PDIs’ structure and dynamics under conditions relevant to physiology, but its implementation has been hindered by technical challenges of position-specific fluorophore labeling. Here, we overcome this limitation by site-specific engineering of fluorescent dyes into human PDI, the founding member of the family. Proof-of-concept smFRET measurements of catalytically active PDI demonstrate, for the first time, the feasibility of this approach, expanding the toolkit for structural studies of PDIs.

Approximately one-third of all newly synthesized polypeptides enter the endoplasmic reticulum (ER)^[1]. Here, unfolded proteins undergo various posttranslational modifications and interact with a number of molecular chaperones and folding enzymes. This dynamic process ensures correct protein folding and, when appropriate, subsequent release of correctly folded, functional proteins from the ER. One of the most important posttranslational modification, especially for extracellular and membrane proteins, entails the covalent linkage of pairs of cysteine residues (-SH), resulting in a disulfide bond (S-S). Disulfide bonds are not only essential for achieving stable tertiary and quaternary structures but also play important functional roles. Indeed, their transient rupture/formation can lead to protein conformational changes that ultimately modulate complex biological processes^[2], such as hemostasis^[3].

Although disulfide bond formation spontaneously occurs in the presence of molecular oxygen, in the ER, this reaction is facilitated by a family of enzymes known as oxidoreductases^[4]. The active sites of oxidoreductases contain two reactive cysteines that, according to the redox microenvironment, can reduce or oxidize a substrate disulfide bond^[5]. Collectively, humans synthesize more than 20 different oxidoreductases, which differ from one another because of the number and type (catalytic vs non-catalytic) of thioredoxin-like domains^[4].

Protein Disulfide Isomerase (PDI) was first described in 1963^[6]. It is the most abundant and best-characterized oxidoreductase and it is the founding member of the PDI family^[4]. PDI comprises 508 amino acids and has a molecular weight (MW) of ~58 kDa. It is organized in four thioredoxin domains arranged in the order a, b, b’ and a’ and a C-terminal tail (Figures 1A and 1B). Domains a and a’ contain the catalytic motif CGHC, whereas the b and b’ domains are non-catalytic and thought to be responsible for substrate and cofactor recruitment^[7]. Importantly, three linkers connect the four thioredoxin domains, thus making PDI a remarkably flexible enzyme, capable of potentially adopting multiple conformations in response to redox environment and substrate exposure^[8].

Figure 1. Site-specific labelling of human recombinant PDI for smFRET studies.

A) Domain structure of human PDI highlighting the location of the two active sites (-CGHC-) and the residues selected in this study for incorporating the unnatural amino acid (UAA) Prk. B) Structural model of oxidized PDI, location and Cβ-Cβ distances calculated for the pairs 42/467 and 308/467. Shown in parenthesis are the values of energy transfer estimated with the software FPS and a value of R₀=65Å for the FRET pair Atto-550/Atto-647N. C) Schematic of the plasmids (i.e., pBAD hPDI and pEvol PylRS) and strategy used in this study to incorporate UAAs into PDI. D) Expression of PDI 42/467. Prk-containing protein accumulated in the soluble fraction (S) and were successfully purified by immobilized metal affinity chromatography (IMAC). E) SDS-PAGE of PDI wild-type (WT, lanes 1 and 4), PDI 42/467 (P1, lanes 2 and 5) and PDI 308/467 (P2, lanes 3 and 6) after incubation with azide dyes in the absence (−) or in the presence (+) of copper sulfate as a catalyst. The gel was first exposed to 532 nm-laser (bottom panel) and then stained with Coomassie for total protein content (top panel). F) SEC profile of the reaction mixture after labeling. G) SDS-PAGE of the SEC fractions stained with either Coomassie (left panel) or exposed to 532 nm-laser (right panel).

While important for its biological function^[9], the structural plasticity of PDI has created significant barriers to our understanding the structural basis of its function. Crystallography and ensemble strategies have produced results that are often at odds with molecular dynamics simulations and biochemical interrogation of PDI in solution^[8]. Inherent limitations of these approaches have prevented resolution of contradictory conclusions. Single-molecule FRET (smFRET) of freely diffusing molecules is a powerful methodology that affords characterization of highly flexible proteins, such as PDI, in solution with high temporal resolution^[10]. By reporting on large data sets of individual, unsynchronized molecules at equilibrium, smFRET overcomes averaging effects and bridges the gap between static atomistic models and ensemble data. However, smFRET studies of PDI and orthologs have not been previously reported, owing to the technical challenges of position-specific fluorophore labeling. In the present study, we overcome this limitation by site-specific engineering of fluorescent dyes into human PDI and report, for the first time, proof-of-concept smFRET measurements of enzymatically active human PDI in solution.

Given that PDI’s active sites contain cysteine residues, fluorescent dyes were attached to the protein by click chemistry after incorporating the non-natural amino acid N-propargyl lysine (Prk) at the desired positions using the AMBER suppressor pyrrolysine tRNA/RS system from Methanosarcina mazei^[11], termed pEvol PylRS. Prk was chosen because of its proven stability in bacteria^[11] and commercial availability. Expression and purification of Prk-containing PDI was attained in five sequential steps. First, the cDNA of human PDI (residues 18–479, corresponding to the fragment abb’a’) was cloned into a pBAD vector expression system (Figure 1C). A C-terminal Avi-tag and an N-terminal His-tag were engineered to facilitate purification of the full-length constructs as well as to enable future biochemical and biophysical applications requiring, for instance, surface immobilization of the labeled protein. Second, amber codons (TAG) were inserted at the desired positions by PCR using appropriate primers. In this study, residues K42, K308 and K467 in the a, b’ and a’ domains were selected based on the currently available X-ray structural data of oxidized human PDI^[12] to obtain two combinations of labeling positions, PDI 42/467 and PDI 308/467 (Figures 1A and 1B). Third, sequence verified PDI variants were co-transfected with the plasmid pEvol PylRS in Top10 cells. Fourth, co-transfected cells were grown until OD₆₀₀=0.4-0.6, induced with 0.02% arabinose and, after overnight incubation at 37°C, harvested by centrifugation and treated with a nonionic detergent to extract the soluble protein. Using this workflow, full-length PDI variants were successfully expressed only in the presence of Prk where they accumulated in the cytoplasm (Figure 1D). Finally, His-tagged PDI was purified by affinity chromatography using a TALON resin. Fractions eluted at 200 mM imidazole were pooled, analyzed by SDS-PAGE and dialyzed overnight in phosphate buffer saline (PBS) to remove imidazole. Protein concentrations were determined by reading at 280 nm with molar extinction coefficients adjusted based on the amino acid sequence. Three milligrams of protein were typically recovered from 1L of cell culture.

Next, affinity purified Prk-containing proteins were stochastically labeled with Atto-550 and Atto-647N azide as a donor and acceptor using copper(I)-catalyzed azide-alkyne cycloaddition (CuAAC)^[11]. The click reaction was quantitative and very specific, as no fluorescent signal was detected in the wild-type enzyme (Figure 1E). Optimal labeling was achieved by reacting 25 μM of PDI in PBS with 4x molar excess of azide fluorophores in presence of 150 μM copper sulphate (CuSO4) and 750 μM tris-hydroxypropyltriazolylmethylamine (THPTA), and 5 mM sodium ascorbate. The reaction mixer was left on slow rotisserie for 1 hour 30 minutes at room temperature, then 30 minutes on ice. The CuAAC labeling reactions were quenched, after 2 hours, by 5 mM ethylenediaminetetraacetic acid (EDTA). The reaction mixture was finally fractionated by size exclusion chromatography (SEC) to separate monomeric doubly labeled PDI from protein aggregates that formed in the presence of copper and were irreversible, even after the addition of EDTA (Figures 1F and 1G). Typical recovery yields starting from 1L of cell culture were in the order of ~300 μg of highly pure monomeric protein, with a labeling efficiency of ~90%.

To rule out potential artifacts arising from the incorporation of the fluorescent dyes and CuAAC chemistry, we performed several control experiments. Figure 2 shows the results of PDI 42/467, but identical results were obtained for PDI 308/467 (Figure S1 in the Supporting Information). First, we used circular dichroism (Figure 2A) and the redox-dependent variation of the intrinsic fluorescence upon addition of GSH and GSSH (Figure 2B) as a proxy of proper PDI folding^[13] and redox sensitivity^[14]. In these measurements, PDI 42/467 displayed identical secondary structure before and after labeling, and similar fluorescence changes when compared to the wild-type, thus documenting structural integrity and unaltered redox-sensitivity. Second, we reacted labeled PDI 42/467 against the sulfhydryl reactive probe Alexa Fluor-488 maleimide in the absence and presence of the reducing agent TCEP. Since monomeric PDI failed to react with the maleimide dye in the absence of TCEP (Figure 2C), we concluded that 1) all the natural cysteines are oxidized after purification and 2) the cysteines can be efficiently reduced by the reducing agent. Finally, we tested the reductase activity of labeled and unlabeled enzymes using the insulin assay^[15] and compared their catalytic efficiency with PDI wild type (Figure 2D). In this assay, the reduction of two-chain insulin by PDI promotes the aggregation of insulin B chain and is followed by increase in turbidity (650 nm). We found that the catalytic activity of the unlabeled variants as well as fluorescently labeled PDI 42/467 and PDI 308/467 was minimally perturbed compared to PDI wild type proving that the effect of the mutations was negligible and that incorporation of fluorescent dyes at positions 42, 308 and 467 does not affect the reductase activity of the enzyme. Taken together, these results demonstrate that bioorthogonal chemistry allows site-specific labeling of PDI and that, after purification, fluorescently labeled PDI variants are properly folded, catalytically active and have fully oxidized active sites.

Figure 2. Structural and functional characterization of labelled PDI.

A) Far-UV CD spectra of PDI WT (gray dotted line), unlabeled (black) and fluorescently labelled (red) PDI 42/467 obtained in 100 mM potassium phosphate (pH 7.4), 2 mM EDTA, at 25°C. Buffer only is shown as a black dotted line. B) Fluorescence spectra of PDI WT (200 nM, gray) and fluorescently labelled PDI 42/467 (200 nM, red) obtained under reducing (1 mM GSH, solid lines) and oxidizing (1 mM GSSG, dotted lines) conditions. C) Reactivity of monomeric doubly labeled PDI 42/467 (5 μM) toward AlexaFluor-488 maleimide (25 μM, top panel) in the absence (−) and in the presence of the thiol reducing agent TCEP (+, 500 μM). The gel was sequentially irradiated with 488-nm, 532-nm and 633-nm lasers to document the incorporation of AlexaFluor-488, Atto-550 and Atto-647N. D) Reductase activity of PDI monitored by the insulin assay. Shown are the progress curves for PDI WT (gray) and PDI 42/467 before (black) and after labeling (red). No absorbance at 650 nm was observed in the absence of enzyme (black dotted line).

To study the conformational dynamics of human PDI in solution, we performed smFRET experiments using a confocal microscope equipped with pulse interleaved excitation (PIE) and SPAD detectors, as described earlier^[16]. Thanks to PIE, single monomeric PDI molecules containing an active donor/acceptor FRET pair can be easily isolated from molecules containing donor and acceptor only, which are irrelevant for our goal (Figure S2 in the Supporting Information). Briefly, the fluorescence intensity of single PDI molecules at 100 pM concentration was recorded as they freely difuse through the confocal volume, on a timescale of ~0.5 ms. Next, fluorescence bursts with >40 photons were extracted and used to calculate FRET efficiency values. FRET efficiency histograms were finally created from thousands of events using the Matlab-based software PAM^[17]. Measurements were performed at 15 μW laser power and 20 MHz repetition rate for each laser line, and 32 ps time resolution with a total acquisition time of 40 minutes. All experiments were repeated fresh up to three times on each protein sample with a buffer consisting of 20 mM Tris, 150 mM NaCl, 2 mM EDTA, pH 7.4. EDTA was added to chelate bivalent metal ions. Tween 0.003% or 200 nM unlabeled proteins were added to the buffer to minimize non-specific absorption of the labeled proteins to the imaging chambers. Using these experimental conditions, labeled PDI was stable for at least 12 hours at room temperature (Figure S3 in the Supporting Information).

Since large-scale redox dependent conformational changes are expected for PDI^[12], smFRET measurements of PDI 42/467 (Figure 3A) and PDI 308/467 (Figure 3B) were performed in the absence and presence of the reducing agent 1,4-Dithiothreitol (DTT, 1 mM). Significant FRET changes were observed in the presence of DTT, as expected, documenting structural changes involving relocation of residues 42/467 and 308/467 away from each other. Unexpected, however, was the structural heterogeneity observed for both redox states as well as the difference between theoretical and experimental FRET values, suggesting that PDI adopts multiple conformations in solution which are significantly different from what have thus far been captured crystallographically. Importantly, DTT had no effect on the photo-physical properties of the dyes since the same FRET changes were observed using different reducing agents, such as GSH (Figure 3C, top panel). This demonstrates that the FRET changes induced by the reducing agents are due to PDI flexibility and not the result of experimental artifacts. To further exclude non-specific binding of the fluorescent dyes to PDI, we performed two additional experiments. First, we treated PDI with guanidinium chloride (Gnd-HCl) (Figure 3C bottom panel). In the presence of 3M Gnd-HCl, a concentration sufficient to unfold PDI^[13], we observed a remarkable shift of the signal toward lower FRET values. This increasing separation of the two dyes within the protein is consistent with site-specific covalently labeled PDI undergoing unfolding. Second, we measured fluorescence quantum yield and anisotropy values of PDI variants that were singly labeled at position 42, 308 and 467. The values of fluorescence quantum yield were 0.79±0.02, 0.78±0.02 and 0.79±0.02 for Atto-550 and 0.64±0.02, 0.64±0.02 and 0.65±0.02 for Atto-647N. The values of anisotropy were 0.18±0.02, 0.19±0.02 and 0.21±0.02 for Atto-550 and 0.20±0.02, 0.18±0.02 and 0.21±0.02 for Atto-647N. These values are consistent with freely rotating dyes attached to PDI experiencing similar chemical environments, thus excluding strong interactions between the dyes and the protein.

Figure 3. Confocal single-molecule FRET experiments of oxidized and reduced PDI.

FRET efficiency histogram for PDI 42/467 (A) and PDI 308/467 (B) in the absence (top panel, yellow) and presence of 1 mM DTT (bottom panel, cyan) as a reducing agent. FRET efficiency histograms were fitted with three Gaussian functions (black lines). The mean FRET efficiency for each population (S for oxidized PDI and R for reduced PDI) is shown in parenthesis. C) Control experiments. Reduced PDI 308/467 obtained by treatment with GSH (top panel). Chemical denaturation of PDI 308/467 induced by 3M Gnd-HCl (bottom panel)

In summary, we report a novel method for site-specific labelling of human recombinant PDI that enables smFRET measurements of the full-length protein under conditions relevant to physiology. We anticipate this method will be useful to study the structure and conformational dynamics of PDI, and other oxidoreductases, free and bound to substrates thus advancing our structural and functional understanding of this family of enzymes. Considering the established role of PDI in cancer^[18], neurodegeneration^[19] and thrombosis^[3], the workflow reported here may also have important ramifications for the development and characterization of novel, selective PDI inhibitors.