Probing the conformational dynamics of thiol-isomerases using non-canonical amino acids and single-molecule FRET

Abstract

Disulfide bonds drive protein correct folding, prevent protein aggregation, and stabilize three-dimensional structures of proteins and their assemblies. Dysregulation of this activity leads to several disorders, including cancer, neurodegeneration, and thrombosis. A family of 20+ enzymes, called thiol-isomerases (TIs), oversee this process in the endoplasmic reticulum of human cells to ensure efficacy and accuracy. While the biophysical and biochemical properties of cysteine residues are well-defined, our structural knowledge of how TIs select, interact and process their substrates remains poorly understood. How TIs structurally and functionally respond to changes in redox environment and other post-translational modifications remain unclear, too. We recently developed a workflow for site-specific incorporation of non-canonical amino acids into protein disulfide isomerase (PDI), the prototypical member of TIs. Combined with click chemistry, this strategy enabled us to perform single-molecule biophysical studies of PDI under various solution conditions. This paper details protocols and discusses challenges in performing these experiments. We expect this approach, combined with other emerging technologies in single-molecule biophysics and structural biology, to facilitate the exploration of the mechanisms by which TIs carry out their fascinating but poorly understood roles in humans, especially in the context of thrombosis.

1. Introduction.

Over one-third of the human proteome encode secretory and membrane proteins. These proteins are translocated into the lumen of the endoplasmic reticulum (ER), where they undergo a series of post-translation modifications [1]. A common and important post-translation modification is the formation of disulfide bonds, an oxidation process in which two cysteine residues are bonded together to form a covalent bond [2]. This process is known as oxidative protein folding [3].

A family of enzymes called thiol-isomerases (TIs), also referred to as protein disulfide isomerases (PDIs) or endoplasmic reticulum proteins (ERps), oversee oxidative protein folding in humans [4]. TIs collectively catalyze the formation, reduction, and isomerization of disulfide bonds in thousands of proteins. They also function as ATP-independent molecular chaperones preventing the aggregation of unfolded proteins [5]. Malfunctioning of TIs is, therefore, a hallmark of several diseases. Overexpression of TIs is documented in many cancer cells in which the synthesis of new proteins is greatly accelerated to meet the high metabolic demands [6–8]. Loss-of-function of TIs caused by mutations and environmental factors, such as oxidative stress and chronic inflammation, can result in neurodegeneration [9] and diabetes [10]. Finally, the sustained extracellular activity of TIs is associated with enhanced thrombus formation [11, 12], which can be a life-threatening condition.

TIs comprise 20+ family members, each containing a combination of catalytically active (a-type) and catalytically inactive (b-type) thioredoxin-like domains [4]. Even though the active domains share the prototypical active CXXC motif, very little sequence identity exists between active and inactive domains in different TIs and within the same TI. Despite this chemical diversity, TIs exhibit extensive overlap in substrate preference and activity, though this redundancy must be incomplete, as organisms lacking specific TIs often exhibit phenotypes which can only be rescued by supplementation with the lacking TI [13, 14]. So, differences in the chemical composition and structures of the domains alone are not always good predictors of substrate specificity and function. Other features, such as structural flexibility and domain-domain interactions, must play a key role [15, 16]. In this context, critical questions in the field are how the individual domains work together to control substrate binding and product release and how TIs respond to the changes in the redox environment and posttranslational modifications.

PDI is the founding member of the TIs’ family. This methodological paper describes the protocols we recently used for elucidating the structural dynamics of PDI in solution [17]. We will discuss the rationale for our choices and the challenges we faced while performing these experiments. We will also provide a new example of how this workflow can be used to study the conformational stability of PDI in different redox states. By sharing this knowledge, we hope to facilitate the exploration of the mechanisms by which TIs carry out their fascinating but poorly understood roles in humans.

2. Site-specific incorporation of non-canonical amino acids into human PDI.

Site-specific labeling of proteins is challenging but necessary for many biophysical studies. Historically, cysteines have been the amino acid of choice for introducing spectroscopic reporters at specific sites because of their swift chemical reactivity at pH 7.4, where most proteins are stable. A common strategy has been engineering cysteine residues at specific sites of the protein of interest and, whenever possible, removing existing cysteines by replacing them with serine, which is isosteric to cysteine but does not react under conditions in which cysteines typically do. Proteins containing engineered cysteines are then carefully reacted with thiol-specific reactants and purified. Maleimide chemistry [18] is often the first choice for targeting cysteines. Between pH 6.5 and 7.5, the reaction is chemoselective; it is ~1,000 times faster than the reaction rate of maleimide with amines. Other chemistries used for cysteine conjugations are iodoacetamide and thiosulfonate [19], yet these options are less common for fluorescent labeling.

While this strategy has proven effective for many systems, labeling enzymes containing cysteine near or at the active site, like PDI, remains problematic. PDI comprises four domains, a-b-b’-a’, connected by three linkers (Figure 1A). The a and a’ domains contain the catalytic motif CGHC [20], totaling four cysteines. Two additional cysteines are in the b’ domain at positions 312 and 343. Thus, genetically eliminating cysteines or labeling them with thiol-specific reagents will likely inactivate PDI, even in carefully controlled conditions.

Figure 1. Structure of hPDI and plasmids used for site-specific incorporation of ncAAs into hPDI.

(A) Crystal structure of human PDI (4ekz.pdb) showing the arrangement of the four domains a-b-b’-a’ in a U-shape with the two catalytic domains a and a’ facing each other. Minimal interactions between the domains argue for a flexible structure primed for conformational rearrangement in response to changes in the redox environment and ligand binding. (B-C) Vector maps for pEvol and pBAD. (D) Primary sequence of human PDI (gene ID P4HB) cloned into pBAD vector using Notl and NcoI restriction sites. The recombinant protein contains a cleavable N-terminal His-tag (red) and a C-terminal AviTag (blue), which is connected to the protein by a flexible G/S linker (magenta). Cysteines are highlighted in green. Four cysteines are found in the active site motif CxxC. Two are in the b’ domain.

A practical solution to this problem is the site-specific incorporation of non-canonical amino acids (ncAAs) into cysteine-containing proteins. Over the years, various strategies have been developed to introduce ncAAs into proteins, including selective pressure incorporation, stop codon suppression, fragment condensation, protein semisynthesis, and peptidomimetics [21]. Among those, stop codon suppression has become very popular for site-specific incorporation of ncAAs that can be functionalized using biorthogonal chemistries, enabling access to a vast array of biophysical investigations [22]. In Escherichia coli, the ribosome translates mRNA into a polypeptide by complementing triplet codons with matching aminoacylated tRNAs [23]. Three of the 64 triplet codons do not code for an amino acid but cause recruitment of a release factor resulting in the disengagement of the ribosome and termination of the synthesis of the growing polypeptide. These codons are called ochre (TAA), opal (TGA), and amber (TAG). Of the three stop codons, the amber codon is the least used in Escherichia coli (~7%) and rarely terminates essential genes [24]. The amber codon triplet in DNA is (TAG), in mRNA (UAG), and the corresponding tRNA anticodon is (CUA). Stop codon suppression relies on the ability of certain primordial species, such as the archeas Methanococcus jannaschii and Methanosarcina mazei, to not use the amber codon as a stop codon but instead use it as an amino acid coding codon. Over time, engineering the machinery responsible for loading the amino acids at the stop codon (i.e., aminoacyl-tRNA synthetase (aaRS) and tRNA) enabled selective incorporation of the ncAA of choice instead of the original natural amino acid, which is tyrosine for Methanococcus jannaschii and pyrrolysine for Methanosarcina mazei [22].

In our work, site-specific incorporation of clickable ncAAs into human PDI was attained by the stop codon suppression methodology via the amber suppressor pyrrolysine-RS/tRNA^CUA system from Methanosarcina mazei [25, 26] as follows:

1. The amber suppressor pyrrolysine-RS/tRNA^CUA system from Methanosarcina mazei was cloned in the pEvol plasmid (Figure 1B), which is one of the most efficient suppressor plasmids available to date [27]. The pEvol plasmid contains two expression cassettes for the pyrrolysine-RS gene, one with a weak constitutive promoter (glnS) and another one with an inducible promoter (araBAD), and a single expression cassette for the tRNA^CUA driven by the efficient proK promoter. The araBAD promotor drives tightly controlled expression of the gene of interest in response to L-arabinose and is inhibited by glucose. The chloramphenicol resistance gene allows the plasmid to be maintained by chloramphenicol (C) selection in Escherichia coli. We termed this plasmid pEvol-PyIRS.

2. The cDNA of human PDI (protein identifier EC:5.3.4.1, gene ID P4HB, residues 18–479) was cloned into a pBAD vector (ThermoFisher) using the restriction enzymes NcoI and Notl (Figure 1C). A C-terminal Avi-tag and an N-terminal cleavable His-tag were engineered to facilitate the purification of the full-length constructs and enable biochemical and biophysical applications requiring surface immobilization of the labeled protein (Figure 1D). The first 17 amino acids of the PDI gene were not included as they codify for the signal peptide that is removed in the mature form of the protein. A pBAD expression system was preferred over a more commonly used pET expression system for two reasons. First, it allows a tightly controlled, titratable expression of the protein of interest, which is ideal for potentially toxic or insoluble proteins. Second, the gene of interest is placed downstream of the araBAD promoter, the same promotor that drives the expression of the inducible pyrrolysine-RS gene. Hence, adding L-arabinose induces simultaneous expression of PDI and PyIRS, meeting the cells’ metabolic needs. The ampicillin resistance gene allows the plasmid to be maintained by ampicillin (A) selection in Escherichia coli. We termed this plasmid pBAD-hPDI.

3. Potential labeling sites were determined by consideration of the following criteria based on available structural information: 1) the side chains of the amino acids to be mutated should be pointing into the solvent and distal from regions of known structural/functional importance (e.g., active sites, binding sites, etc.), 2) labeling sites should be separated by a distance appropriate for the fluorophores to be used. For Atto550 and Atto647N, as employed here, the R₀ is ~62 Å, so separation distances between 50 and 70 Å is ideal. 3) FRET-assisted structural modeling software like FRET-restrained positioning and screening (FPS) [28] or Olga [29] was used to determine theoretical FRET values based on the labeling positions and fluorophores of choice. 4) Since structures for oxidized and reduced PDI exist, the difference in theoretical FRET values was determined to identify positions expected to show differences in FRET between oxidized and reduced states of PDI. Based on these criteria, positions 88 in the a domain and 467 in the a’ domain were selected for this work.

4. Amber codons (TAG) were introduced into PDI at one or multiple positions by PCR using appropriate primers (Integrated DNA technologies). The plasmids were verified by sequencing (Genewiz) using pBAD forward and reverse primers.

5. The sequence-verified pBAD-hPDI plasmid (70 ng) was co-transformed with pEvol-PyIRS (70 ng) using the heat shock method in TOP10 cells (Invitrogen), which is a host strain lacking genes for L-arabinose catabolism. This allows efficient and stable gene activation without reduction in the L-arabinose concentration over time.

6. For protein expression, a single colony containing the pBAD-hPDI and pEvol-PyIRS plasmids was grown overnight at 37°C (225 rpm) in 5 ml of Luria-Bertani (LB) broth (Sigma-Aldrich) containing 100 μg/ml ampicillin (A) and 34 μg/ml chloramphenicol (C) (LBAC). 1 ml of the overnight solution was then used to inoculate 100 ml LBAC in a shake flask. TOP10 cells were grown for ~2 h to reach an OD₆₀₀ of ~0.25 and then supplemented with 1 ml of 100 mM propargyl lysine (PrK, SciChem - SC-8002) (stock solution 100 mM in 0.1 M sodium hydroxide (NaOH)) to attain a concentration of 1 mM. 1 ml of 0.1 M hydrochloric acid (HCl) solution was added to neutralize the 0.1M NaOH. Cells were grown until OD₆₀₀ of ~0.55 was reached. A 0.2 ml solution of 10% (w/v) L-arabinose (Sigma-Aldrich) was then added to induce expression. Cells were grown overnight at 37°C, 225 rpm.

7. For protein purification, cells were harvested by centrifugation and resuspended in a nonionic detergent. Recombinant PDI protein was purified by immobilized metal affinity chromatography using a TALON resin (Takara, Japan). After extensive washing (Tris 20 mM pH 7.4, 300 mM NaCl, 20 mM imidazole) and elution (Tris 20 mM pH 7.4, 300 mM NaCl, 200 mM imidazole), the fractions containing the purified protein were pooled together and buffer exchanged to 1x Phosphate-Buffered Saline (PBS), pH 7.4 ± 0.1 (Corning, 46-013-CM) by extensively dialyzing at 4°C using a 10 kDa cutoff to remove imidazole and Tris, which is not compatible with click chemistry reactions. After dialysis, the purified protein was collected by centrifugation, concentrated to ~2.5 mg/ml (~45 μM), aliquoted, and stored at −80°C. On average, 0.40 mg of ~80% pure protein was typically obtained from a 100 ml culture.

3. Fluorescent labeling of human PDI by click chemistry.

Having established a robust recombinant system for site-specific incorporation of Prk into PDI at one or multiple positions, we next developed a protocol for fluorescent labeling PDI for smFRET experiments. Because Prk is a linear alkyne, we used copper(I)-catalyzed azide–alkyne cycloaddition (CuAAC) to conjugate azide-containing fluorescent dyes into PDI (Figure 2A) [30]. In this reaction, copper ions in the +1 oxidation state (Cu^I) serve two purposes. First, they increase the reaction rate up to 7 orders of magnitude compared to the uncatalyzed process. Second, they impart regioselectivity of the cycloaddition reaction, with 1,4-disubstituted 1,2,3-triazole being the only product formed. In our protocol, the catalytic Cu^I was generated in situ by reacting copper sulfate (CuSO₄) with sodium ascorbate. Since byproducts of this reaction can damage proteins, we also added tris-hydroxypropyltriazolylmethylamine (THPTA) as a scavenger for protecting PDI. In general, for our downstream applications, Atto-550 (Figure 2B) and Atto-647N (Figure 2C) azide dyes (Atto-TEC, Germany) were preferred over spectrally similar Cy-3/Cy-5 and AlexaFluor-550/AlexaFluor-647 dyes because of their high fluorescent quantum yield, relatively long lifetime, and high thermal, pH and photo-stability. Furthermore, in these dyes, the reactive azido group is placed at the end of a hydrophilic, chemically inert pegylated linker, which, when fully extended, can exceed 12 Å. This minimizes the likelihood of protein-dyes interactions, which can be a source of FRET artifacts [31].

Figure 2. Click-chemistry and structure of the fluorescent dyes.

(A) N-Proparagyl-L-lysine incorporated into PDI (yellow) specifically reacts with an Atto-azide (orange) in the presence of copper (I) to form a 1,4 triazole. (B-C) Chemical structures of Atto-550 azide (Atto-TEC, Germany, AD 550) and Atto-647N azide (Atto-TEC, Germany, AD 647N) used to label PDI. Note how the azido reactive group is linked to the fluorophore by a pegylated linker composed of 8 atoms of carbon and 3 atoms of oxygen. MW of Atto-550 azide is 794 g/mol. MW of Atto-647N azide is 846 g/mol.

After several trials, optimal labeling was achieved by reacting 25 μM of PDI in PBS with 4x molar excess of Atto-550 and Atto-647N azide fluorophores in the presence of 150 μM CuSO₄ and 750 μM THPTA, and 5 mM sodium ascorbate. The reaction mix was left on a slow rotisserie for 90 minutes at room temperature, then 30 minutes on ice. The CuAAC labeling reaction was then quenched, after 2 hours, with 5 mM ethylenediaminetetraacetic acid (EDTA). The reaction mixture was finally fractionated by size exclusion chromatography (Superdex 200 Increase 10/300 GL, Cytiva) in Tris 20 mM, 150 mM NaCl, 2 mM EDTA, pH 7.4 to separate monomeric fluorescently labeled copper-free PDI species from higher-order PDI oligomers that are known to form in the presence of copper [32]. Following this protocol, typical recovery yields starting from 100 mL of cell culture were in the order of ~50 μg of highly pure monomeric protein, with a labeling efficiency of ~50% for each site. As we showed in our recent publications [33, 34], reductase activity assay, far-UV CD, and tryptophan fluorescence confirmed that the labeling protocol does not perturb the structure and functional properties of PDI unless the dyes are located next to active sites, in which case, we expect reduced enzymatic activity but no evident structural perturbations. Finally, Ellman’s reaction was used to determine the redox state of monomeric PDI after labeling and purification. We consistently found that PDI is in the oxidized state, consistent with PDI forming two disulfide bonds, one in each active site.

Importantly, the concentration and order by which each species was added to the mixture significantly altered the labeling efficiency and quality of the recovered protein. While increasing the concentration of copper from 150 μM to 250 μM led to slightly higher labeling efficiency (from 50% to 65%), the amount of monomeric protein significantly dropped from ~50 μg to ~5 μg. By contrast, significantly more monomeric protein could be recovered (from ~50 μg to ~150 μg) by decreasing the concentration of copper from 150 μM to 75 μM, but the labeling efficiency was poor, approximately 15% for each site. The addition of CuSO₄, THPTA, and ascorbic acid to Prk-containing PDI a few minutes before the addition of the azide dyes did not improve labeling efficiency nor monomer recovery, indicating that the generation of Cu(I) occurs very rapidly under these conditions and that Prk remains accessible to azide dyes in oligomeric PDI species which form in the presence of copper. Finally, the chemical nature of the dyes affected the final yield and quality of the protein, with negatively charged dyes (i.e., Atto-488) generally reacting poorly with Prk-containing PDI compared to positively charged dyes (i.e., Atto-550 and Atto-647N). The complete workflow to produce and characterize doubly labeled PDI proteins is summarized in Figure 3.

Figure 3. Workflow used in our laboratory to attain double labeled PDI suitable for smFRET studies.

The workflow is divided into four sequential steps. In Step 1, TOP10 cells co-transformed with pEvol-PyIRS and pBAD-PDI are grown, supplemented with the ncAAs of choice, and then recombinant protein is induced by the addition of arabinose. Cells are harvested by centrifugation. In Step 2, the recombinant protein is purified by metal affinity chromatography and purity is assessed by SDS-PAGE. In Step 3, dialyzed protein in PBS is subjected to click chemistry, and monomeric labeled PDI is purified by size exclusion chromatography. In Step 4, labeled PDI, alongside unlabeled PDI and PDI WT, is subjected to biochemical and biophysical measurements to determine possible structural and functional effects caused by the mutations and labeling procedure. A detailed description of these assays is provided in the material and method section of our recently published work [33, 34].

4. Conformational dynamics of PDI by multiparameter fluorescence confocal smFRET.

The availability of catalytically active PDI in which donor and acceptor fluorophores have been introduced at specific sites opens the door for various biophysical investigations, most notably, single-molecule FRET (smFRET).

FRET, which stands for Förster resonance energy transfer, is a widely adopted method for studying protein conformations and dynamics. In a FRET experiment, the FRET efficiency (E) between donor and acceptor fluorophores is quantified with respect to their proximity (r) according to the equation E=1/(1+r/R₀)⁶, with R₀ being the Förster distance of a donor/acceptor pair [35, 36]. Accordingly, FRET is often used as a spectroscopic ruler to measure distances on the molecular scale. While FRET can be measured in bulk using a fluorimeter, the combination of FRET with microscopy provides the most valuable information when studying biological systems, enabling real-time observation of single biomolecules. By overcoming ensemble- and time-averaging effects characteristic of bulk measurements, smFRET can reveal otherwise hidden states of biomolecules and transient intermediates, illuminating how biomolecules work at the microscopic level [37].

In general, smFRET experiments are performed on molecules tethered to surfaces or freely diffusing in a medium of choice [36]. In experiments on surface-immobilized molecules, the molecules are usually excited by total internal reflection fluorescence (TIRF) and detected by cameras. In contrast, in experiments on freely diffusing molecules, the molecules are excited and detected by confocal optics coupled with single-photon avalanche diode detectors (SPADs). While an advantage of performing experiments on surface-immobilized molecules is the ability to observe the kinetic behavior of single molecules for a relatively long period of time (typically seconds to several minutes), which is then analyzed to determine the number of FRET states present at equilibrium, the transition pathways and rate of transitions between the FRET states, some proteins are not amenable to immobilization, as their structures may be altered or rigidified once immobilized. By contrast, even though the observation time of freely diffusing molecules is much shorter compared to that of surface-immobilized molecules, typically limited to ~1–5 ms, experiments using freely diffusing molecules can be used to identify conformational states that a protein adopts in a native-like environment while tumbling in solution. Furthermore, unlike camera-based systems, SPADs have a high time resolution. Paired with picosecond pulsed lasers and time-correlated single photon counting (TCSPC) modules, this setup enables accurate measurements of the fluorescence lifetimes for each molecule, which occur on the ns timescale. Thus, smFRET measurements of freely diffusing single molecules, unlike camera-based systems, enable the recording of multiple fluorescence parameters, namely fluorescence intensity, fluorescence lifetime, and anisotropy, providing access to a wealth of information that spans from the ns to the ms timescale. This capability, known as multiparameter fluorescence detection [38], has proven extremely useful in studying the conformational dynamics of different proteins in solution and remains one of the methods of choice for studying particularly malleable proteins that can be easily damaged when immobilized onto a surface or proteins whose motions are too fast to be characterized using traditional camera-based systems, such as intrinsically disordered proteins [39, 40] but also membrane proteins [17].

In our experiments, multiparameter fluorescence confocal smFRET studies of PDI were performed on a customized MT200 microscope (PicoQuant) equipped with pulsed interleaved excitation (PIE) and 4 SPADs detectors. The schematics of the instrument are shown in Figure 4. Briefly, polarized excitation light at 532 and 638 nm originating from an LDH-P-FA-530L pulsed laser diode head and an LDH-D-C-640 dual model laser diode head is used to excite the donor and acceptor fluorophores, respectively. Lasers are electronically delayed with a pulse rate of 20 MHz to alternate the donor and acceptor excitation. A dichroic mirror (ZT405/488/532/640rpc-XT, Chroma) reflecting at 532 and 638 nm guided the light to a high numerical aperture apochromatic objective (60x, N.A. 1.2, water immersion, Olympus) that focused the light to a confocal volume of ~1 fl. Fluorescence from excited molecules is collected with the same objective and focused onto a 50-μm diameter pinhole. After the pinhole, a polarizing cube is used to split the light 50:50 into perpendicular and parallel directions. In the perpendicular path, the donor and acceptor emissions are separated via a dichroic filter with a dividing edge at 620 nm (620DCXR, Chroma). Suited bandpass filters (HQ580/70m, Chroma and HQ690/70m, Chroma) were inserted to eliminate the respective excitation wavelength and minimize spectral crosstalk. The fluorescence was detected with two single-photon avalanche diode detectors (t-SPAD, Perkin Elmer) using Time-correlated Single Photon Counting with the TimeHarp 200 board (HydraHarp 400, PicoQuant). In the parallel path, the donor and acceptor emissions are separated via a dichroic filter ZT633rdc-UF1 (Chroma), donor emission filter ET585/65m (Chroma), and acceptor emission filter ET700/75m (Chroma). The fluorescence was detected with two single-photon avalanche diode detectors (Excelitas Technologies) using Time-correlated Single Photon Counting with the TimeHarp 200 board (HydraHarp 400, PicoQuant). Fine-tuning of the system is performed such that very similar fluorescent intensity values (~5% difference) were obtained for the two sets of detectors.

Figure 4. Four-channel confocal microscope for smFRET-PIE measurements.

m=mirror; Pol=polarizing beamsplitter; f=bandpass filters. In our experiment, the lasers’ repetition rate was set to 20 MHz, which is >10 times faster than the Atto dyes’ rate of fluorescence decay.

For data collection, measurements of a solution of ~50 pM doubly labeled PDI in 20 mM Tris, 150 mM NaCl, 2 mM EDTA, pH 7.4 (TBSE), and 0.003% Tween 20 were performed at 25 μm in the solution above the glass surface of the imaging chamber using a laser power of ~30 μW. Total acquisition time was ~40 minutes per sample. Combined with a confocal volume of ~1 fl, the low concentration of PDI minimizes the probability of multiple molecules being present simultaneously. The confocal volume is determined experimentally before our measurements by fluorescence correlation spectroscopy (FCS) using solutions of 0.1–10 nM Atto-655 carboxy derivative in water-0.01% Tween 20. This dye has the advantage of having negligible triplet state contributions. Our laboratory also uses a set of well-standardized double-stranded DNA constructs labeled with Atto-550/647N dyes [33]. These reagents are helpful for monitoring parameters like lasers’ intensities and detection efficiency, which can fluctuate over time. Data recording uses Sympho-Time Software 64, version 2.4 (PicoQuant).

As each molecule transverses the confocal volume, fluorescence photons are detected and stored in the Time-tagged Time-resolved Mode in PTU file format. After binning the data set to 0.5 ms, bursts with more than 40 counts and 5 photons per time window are searched with the All Photon Burst Search (APBS) algorithm using the software PAM [41]. From each burst, two bursts-integrated quantities can be retrieved. The first quantity is the FRET efficiency (E), which is defined by the equation:

$$E=\frac{F_{DA}}{F_{DA}+F_{DD}}$$ (Eq. 1)

where F_DA and F_DD are the signal intensities referring to the number of background-corrected fluorescence photons detected in the donor (D) and acceptor (A) channels after donor (D) excitation. The second quantity is stoichiometry (S), which is defined by the equation:

$$S=\frac{F_{DA}+F_{DD}}{F_{DA}+F_{DD}+F_{AA}}$$ (Eq. 2)

where F_AA is the signal intensity referring to the number of background-corrected fluorescence photons detected in the acceptor (A) channel after acceptor (A) excitation. Since lasers are adjusted so that (F_DA+F_DD) ≈ F_AA, stoichiometry theoretically predicts a value of 1 for D-only species, a value of 0 for A-only species, and a value of 0.5 for species with one donor and one acceptor, D/A species. In reality, because of shot noise, a distribution of D, A, and D/A only species is found in the interval 0.8–1 for D only, 0–0.20 for A only, and 0.25–0.75 for D/A species [42]. Importantly, while the FRET efficiency is a distance-dependent quantity as F_DA and F_DD are dictated by the distance between D and A, stoichiometry is a distance-independent quantity as it is the ratio of the total number of photons after donor excitation divided by the total number of photons after donor and acceptor excitations [42]. Hence plotting of FRET efficiency versus stoichiometry enables the selection of molecules that contain a 1:1 ratio of donor and acceptor regardless of their FRET efficiency. Additional filters, such as ALEX-2CDE [43] and |TDX-TAA| [44], are usually applied to identify and eliminate photobleached molecules that could falsify the FRET distributions. The detailed description of these filters and examples of how to use them is well described in the literature [43–45].

Since fluorophores and optical systems are not ideal, accurate estimation of FRET efficiencies (E) and populations (S) requires correction of raw donor and acceptor intensities using following equations [38, 44].

$$E=\frac{F_{DA}-\alpha F_{AA}-\beta F_{DD}}{F_{DA}-\alpha F_{AA}-\beta F_{DD}+\gamma F_{DD}}$$ (Eq. 3)

$$S=\frac{F_{DA}-\alpha F_{AA}-\beta F_{DD}+\gamma F_{DD}}{F_{DA}-\alpha F_{AA}-\beta F_{DD}+\gamma F_{DD}+F_{AA}}$$ (Eq.4)

In these equations, α is the direct excitation of the acceptor at the donor excitation wavelength, β is leakage of the donor in the acceptor channel, and γ is the detection correction factor, which captures the different sensitivities of the different detection channels. Since a sample is never 100% labeled with 100% photoactive donor and acceptor molecules, using PIE, correction factors can be measured by analyzing photoactive D-only and A-only species [44]. Due to possible quenching effects, this strategy is preferred over the use of fluorophores alone.

Retrieving E and S for each molecule is advantageous for all systems and is critical for proteins in which donor and acceptors are introduced stochastically, thus producing a complex mixture of D-only, A-only, and D/A PDI molecules. For example, Figure 5A shows the smFRET analysis of PDI doubly labeled with Atto-550/647N at positions 88 and 467, which results in three populations of molecules, D-only, A-only, and D/A. Using stoichiometry as a filter, we could identify and further analyze molecules containing one donor and one acceptor (Figure 5B) while discarding D-only and A-only species. The broad FRET distribution resulting after selection documents that PDI 88/467 is heterogeneous, adopting multiple conformational states at equilibrium, which are characterized by a lower FRET (LF) and a higher FRET (HF).

Figure 5. smFRET studies of PDI.

2D plots of FRET efficiency versus stoichiometry of smFRET data collected for oxidized PDI 88/467 before (A) and after selecting D/A molecules (B). Two major FRET populations, lower FRET (LF) and higher FRET (HF) are visible in the FRET efficiency plots. PDI (50 pM) was solubilized in 20 mM Tris, 150 mM NaCl, 2 mM EDTA, pH 7.4 (TBSE), and 0.003% Tween 20. Data were collected at room temperature (~20°C). Selection criteria include 0.25<S<0.75, ALEX-CDE<12, |TDX-TAA|<0.5. Correction factors are γ=0.85, β=0.08, and α=0.05. Low-intensity bursts and potential PDI dimers/aggregates were also eliminated by limiting the count rate of DD, DA, and AA to 0.5–80 kHz. 2D plots of the lifetime of the donor versus FRET efficiency (C) and the lifetime of the acceptor versus FRET efficiency (D) of the same dataset in panel A, that is, before clean-up. Note how D-only, A-only, LF, and HF populations are easily identifiable in these plots. For panel C, the blue line represents the static FRET line, which has been modified to consider the flexibility of the linker connecting the two dyes (R₀=62 Å and apparent linker length =5 Å) according to [41], under the assumption that the dynamics of tethered dyes are much faster than the dynamics of PDI. Two populations, O₁ and O₂, make up the LF population. These populations exchange with C, which has a higher FRET.

In addition to its molecule-sorting power, a significant advantage of this setup over other systems in which lasers are not pulsed or pulsed in the microsecond timescale is the ability to measure the fluorescence lifetime for each burst. This enables the identification of donor and acceptor quenching, which may lead to misinterpretation of the data. It also enables a qualitative description of the system in terms of conformational dynamics [46, 47]. In fact, the FRET theory says that the FRET efficiency (E) is linearly related to the fluorescent lifetime (τ) by the equation:

$$E=1-\frac{{\tau }_{D(A)}}{{\tau }_D}$$ (Eq. 5)

where τ_D(A) is the lifetime of the donor in the presence of the acceptor, and τ_D is the lifetime of the donor without acceptor. Thus, all FRET populations arising from different interdye distances should be located along this line, which represents the theoretical relationship between the FRET efficiency and the donor lifetime. In contrast, any deviation from this line would suggest that additional photophysical processes are occurring. Besides photo-induced dye chemistry and physics, FRET species that are not interconverting or interconverting ~10 times slower than diffusion (typically a few ms) will lie on this line. This is because the molecules will be in the same state while they diffuse through the confocal volume. In contrast, FRET species that interconvert at a comparable timescale or faster than diffusion will deviate from this line, resulting in a smear between species for slower exchanges and a well-defined population shifted toward the right for faster exchanges. This is because, in burst-wise analysis, E values derived from fluorescence intensities are averaged per molecular species fractions, whereas fluorescence lifetimes are averaged per brightness [47]. Thus, observing where FRET species lie with respect to the “static FRET line” enables a qualitative assessment of whether FRET species traversing the confocal volume are in a static or dynamic equilibrium [46, 47].

For example, burst-wise lifetime analysis of PDI 88/467 led to the observation that most of the LF and HF species do not lie on the static FRET line, which was calculated using a donor lifetime of 3.3 ns, but they are instead shifted toward the right (Figure 5C). Since the lifetime of the acceptor (τ_A) in the D/A population was identical to the lifetime of the A-only species (Figure 5D), we interpreted this phenomenon as PDI 88/467 undergoing rapid exchange between three conformationally distinct ensembles, which we called C, O₁ and O₂ [46].

5. Thermodynamic stability of PDI monitored by smFRET.

PDI is a redox-regulated enzyme. We recently published that the addition of the reducing agent dithiothreitol (DTT) to oxidized PDI favors the opening of the structure [33]. Here, we apply smFRET to investigate how oxidized and reduced PDI responds to chemical denaturation by guanidium hydrochloride (Gnd-HCl), a well-established method to gain information on the thermodynamic stability of macromolecules [48]. For this experiment, smFRET data of PDI 88/467 were collected at increasing concentrations of Gnd-HCl in the absence and presence of 1 mM DTT. Data in Figure 6 document that Gnd-HCl causes oxidized (Figure 6A) and reduced (Figure 6B) PDI to transition toward lower FRET (~0.05) in a dose-dependent fashion. Unfolding of PDI or quenching of the donor by Gnd-HCl could explain these results. Since 2D plots of τ_D(A) versus FRET efficiency (Figures 6C and 6E) ruled out the latter possibility, we concluded that FRET changes caused by Gnd-HCl are due to conformation changes; specifically, PDI transitioning from more compact (higher FRET) to more elongated conformations (lower FRET). And even though the 2D plots of τ_D(A) versus FRET efficiency (Figure 6D and 6F) show small but significant acceptor quenching at 3M Gnd-HCl, this phenomenon does not change our structural interpretation since acceptor quenching does not influence E and τ_D(A), which depend only on the transition rates from the donor excited state.

Figure 6. Chemical denaturation experiment of PDI monitored by smFRET.

smFRET histograms of PDI 88/467 in the absence (A) and presence (B) of 1 mM DTT collected at increasing concentrations of guanidinium hydrochloride (Gnd-HCl) (from 0 to 3 M). PDI (50 pM) was solubilized in 20 mM Tris, 150 mM NaCl, 2 mM EDTA, 0.003% Tween 20, pH 7.4 and different concentrations of Gnd-HCl for 1 hour at room temperature (~20°C) before data collection. pH-buffered 7.4 M Gnd-HCl was prepared by combining 10 X TBSE stock with 8 M Gnd-HCl (ThermoFisher, 24115). Folded (F) PDI populates C, O_1, and O₂. Unfolded PDI (U) adopts an elongated structure characterized by very low FRET. (C-F) 2D plots of the lifetime of the donor in the presence of acceptor versus FRET efficiency (C, E) and the lifetime of the acceptor versus FRET efficiency (D, F) of oxidized and reduced PDI 88/467 at 0 and 3M Gnd-HCl. The arrow shows the transition between folded and unfolded states. (G, H) Denaturation profiles of oxidized (G) and reduced (H) PDI were monitored by smFRET and tryptophan (Trp) fluorescence. Values of mean FRET (black dots) and scaled fluorescence intensity (white dots) were plotted versus the concentration of Gnd-HCl. The solid red and blue lines represent the best global fit to a two-state model using the equation Y=(N+sn*x+(U+su*x)*exp(−g+m*x)/R*T)/(1+exp(−g+m*x)/R*T), where N and U are the intensities of the native and unfolded states, sn and su are the baseline slopes for the native and unfolded regions, g is the free energy change for unfolding reaction at the reference temperature, m is denaturant concentration index, R is the molar gas constant, and T is the temperature in Kelvin. Values of N and U were fixed at 0.69 and 0.05 for oxidized PDI and 0.53 and 0.05 for reduced PDI, respectively. R was 1.9872 cal/K mol and T=293.15K. For Trp fluorescence experiments: 200 nM PDI in 20 mM Tris, 150 nM NaCl, 2 mM EDTA, pH 7.4 was incubated in increasing Gnd-HCl concentrations for one hour at 293.15 K with and without 100 μM DTT and then fluorescence intensity was measured with excitation at 295 nm and emission at 337 nm (slits 5/10 nm) with a Horiba FluoroMax-4 (0.1 s integration time) in a 3×3 mm quartz cuvette. Fluorescence intensity values (F) were normalized by the equation ΔF/ΔF_max = (F_obs-F_min)/(F_max-F_min) and then re-scaled to top and bottom FRET efficiency values (E) to facilitate comparison.

The transition between folded and unfolded states is best captured by plots in Figures 6G and 6H, in which the mean FRET is plotted versus the concentration of Gnd-HCl. Fitting these plots with a two-state model [48–50], we estimated free energy change for the unfolding reaction at 293.15K (ΔG^o_un,o) of 7.89±1.66 kcal/mol and 7.79±0.64 kcal/mol, and slopes of the transition region of 2.91±0.69 kcal-liter/mol² and 3.32±0.31 kcal-liter/mol² for oxidized and reduced PDI respectively, suggesting oxidized and reduced PDI have similar stability to Gnd-HCl denaturation and similar unfolding cooperativity. These conclusions align with what we obtained by monitoring the chemical unfolding of oxidized (Figure 6G) and reduced PDI (Figure 6H) using tryptophan fluorescence, a technique orthogonal to smFRET. It is indeed curious that the chemical stability of the two states of PDI is similar. However, the cysteines which form disulfide bonds in oxidized PDI are separated by only two amino acids in the primary sequence (i.e., CxxC), so their stabilizing effect may be small. The proximity of the two cysteines could also explain why the unfolded state monitored by smFRET appears to be equally relaxed in reduced and oxidized PDI despite the presence of disulfide bonds in oxidized PDI but not in reduced PDI. Alternatively, unfolded states could look alike in oxidized and reduced PDI because the FRET efficiency is very low, outside the range in which differences in inter-fluorophore distances may be inferred.

Another exciting feature captured by smFRET is the behavior of PDI at Gnd-HCl=0.25. Low concentrations of Gnd-HCl seem to favor compaction of the structure and accumulation of O₂, which translates into higher FRET. This phenomenon is particularly evident in oxidized PDI (Figure 6A), but it is also visible in reduced PDI (Figure 6B). A possible explanation is that a dynamic H-bonding network stabilizes open PDI forms like O₁. In contrast, closed forms may rely more heavily on hydrophobic interactions, which are favored by the proximity of the catalytic domains and coordinated interaction between adjacent domains. Future studies will address whether structural changes induced by low concentrations of Gnd-HCl are linked to catalysis. An intriguing possibility would be that shifting the conformational ensemble toward O₂ may enhance reductase activity.

6. Advantages of the current method and areas of improvement.

In conclusion, our work shows that it is possible to attain site-specific incorporation of ncAAs into PDI, opening the door for innovative ways to explore its mechanism of action. Continuing applying this workflow to PDI and other TIs, especially those involved in thrombosis [11], will illuminate the mechanisms through which TIs quickly adapt to changes in the extracellular redox environment and how they exploit their innate structural flexibility to interact with substrates and release products favoring coagulation.

An advantage of the pyrrolysine-RS/tRNA^CUA system from Methanosarcina mazei used in our studies is that the catalytic promiscuity of the PyIRS enzyme [51] enables the recognition, activation, and tRNA loading of ncAAs with different functional properties with high efficiency. While the linear alkyne Prk was chosen in this study, we recently successfully incorporated the cyclic alkyne cyclooctyne-lysine (SCO, SiChem SC-8000) into PDI by using the PyIRS variant Y306A/Y384F (unpublished data). This major improvement and commercial availability of tetrazine and methyl-tetrazine dyes should enable the transition to copper-free click chemistry in the following years, likely simplifying protein purification, reducing costs, and minimizing the probability of introducing artifacts that could arise when copper is introduced into the system. Coupling the currently available pyrrolysine-RS/tRNA^CUA system from Methanosarcina mazei with another aaRS/tRNA system, such as the tyrosyl–RS/tRNA^CUU from Methanococcus jannaschii or the recently developed pyrrolysine-RS/tRNA^CUU system from Methanogenic archaeon ISO4-G1 [52] will provide the basis for introducing two ncAAs with different chemical reactivities at different sites, further improving the accuracy of our smFRET measurements.

In addition to enabling smFRET measurements, incorporating ncAAs into TIs will also open the door for structural investigations of TIs using techniques complementary to smFRET, such as double electron–electron resonance spectroscopy, crosslinking-based mass spectrometry methods, atomic-force microscopy, and cryo-EM. With proper design, covalent trapping of TIs bound to low-affinity substrates and transient intermediates will become possible, permitting determination of the structures of TIs bound to different substrates or to the same substrate in different states along its folding trajectory. The emerging picture from these studies will transform our basic understanding of TIs with important implications for both basic and translational sciences.

Abbreviations:

PDI

human protein disulfide isomerase

smFRET

single-molecule FRET

TIs

thiol-isomerases