Design of AsLOV2 domain as a carrier of light‐induced dissociable FMN photosensitizer

Abstract

Flavin mononucleotide (FMN) is a highly efficient photosensitizer (PS) yielding singlet oxygen (¹O₂). However, its ¹O₂ production efficiency significantly decreases upon isoalloxazine ring encapsulation into the protein matrix in genetically encoded photosensitizers (GEPS). Reducing isoalloxazine ring interactions with surrounding amino acids by protein engineering may increase ¹O₂ production efficiency GEPS, but at the same time weakened native FMN–protein interactions may cause undesirable FMN dissociation. Here, in contrast, we intentionally induce the FMN release by light‐triggered sulfur oxidation of strategically placed cysteines (oxidation‐prone amino acids) in the isoalloxazine‐binding site due to significantly increased volume of the cysteinyl side residue(s). As a proof of concept, in three variants of the LOV2 domain of Avena sativa (AsLOV2), namely V416C, T418C, and V416C/T418C, the effective ¹O₂ production strongly correlated with the efficiency of irradiation‐induced FMN dissociation (wild type (WT) < V416C < T418C < V416C/T418C). This alternative approach enables us: (i) to overcome the low ¹O₂ production efficiency of flavin‐based GEPSs without affecting native isoalloxazine ring‐protein interactions and (ii) to utilize AsLOV2, due to its inherent binding propensity to FMN, as a PS vehicle, which is released at a target by light irradiation.

1. INTRODUCTION

Photosensitizers (PSs) are small organic compounds that induce cytotoxicity when irradiated with a specific light source (Dougherty et al., 1998; Lee et al., ²⁰²⁰). PSs possess high potential in clinical application in photodynamic therapy (PDT) (O'Connor et al., ²⁰⁰⁹; Zhang et al., ²⁰¹⁸) as well as an alternative therapy against multi‐drug resistant bacteria (Hilgers et al., 2019). However, PSs' applications are significantly affected by their low solubility and stability in aqueous solvents and their inherently low specificity to target cells (Mansoori et al., 2019). These limitations have led to an intensive effort to develop an efficient genetically encoded PSs (GEPSs).

GEPSs are proteins that bind to specific chemical PSs with several biological and medical applications, including immunophotosensitizing, chromophore‐assisted light inactivation (CALI) of proteins and photoablation of cells, photodynamic diagnosis (PDD), antimicrobial photodynamic inactivation, and correlative light electron microscopy (CLEM) (Rodríguez‐Pulido et al., 2016; Souslova et al., ²⁰¹⁷; Hilgers et al., ²⁰¹⁹; Gunaydin et al., ²⁰²¹). GEPSs have several advantages over chemical PSs: (i) protein binding overcomes the low solubility of chemical PSs, (ii) the protein matrix provides an encased phototoxic PS with a constant local environment, regardless of subcellular location, fostering a reliable production of reactive oxygen species, and (iii) conjugation with proper tags (leader peptides, targeted proteins, such as antibodies, and Designed Ankyrin Repeat Proteins (DARPins)) by genetic engineering enables the targeted accumulation of GEPSs at specific subcellular structures or compartments (Proshkina et al., 2015; Serebrovskaya et al., ²⁰⁰⁹; Westberg et al., ²⁰¹⁹). GEPSs are divided into two main classes. The first class includes proteins with a green fluorescent protein‐like structure (Bulina et al., 2006; Gorbachev et al., ²⁰²⁰; Micheletto et al., ²⁰²¹; Sarkisyan et al., ²⁰¹⁵; Takemoto et al., ²⁰¹³) such as KillerOrange, KillerRed, as well as improved versions of KillerRed with enhanced phototoxicity known as SuperNova and SuperNova 2. The second class encompasses flavin‐binding proteins derived from the light‐oxygen‐voltage photoreceptor domain of plants, algae, and bacteria (Petrencakova et al., 2020; Ruiz‐Gonzalez et al., ²⁰¹³; Souslova et al., ²⁰¹⁷; Westberg et al., ^{2015, 2017a}). The main difference between the two classes of GEPS, apart from the chemical nature of PS, is that the second group of PS is dissociable.

The cofactor flavin mononucleotide (FMN), which represents naturally occurring PS as a part of the second group of GEPS, is characterized by a singlet oxygen (¹O₂) production efficiency (Φ_Δ) of ~0.51–0.65 (Baier et al., 2006; Westberg et al., ^2017b). However, upon binding to mini‐singlet oxygen generator (miniSOG) protein (Shu et al., 2011), its Φ_Δ decreases by more than 10‐fold, Φ_Δ ≤ 0.05, demonstrating a major drawback of GEPSs (Endres et al., 2018; Petrencakova et al., ²⁰²⁰; Ragas et al., ²⁰¹¹; Westberg et al., ^{2015, 2017a, 2017b}). Such significant decrease in Φ_Δ is a result of numerous static and dynamic interactions of amino acids surrounding the tightly encapsulated FMN in the protein, including hydrogen bonding, van der Waals, and π–π interactions (van den Berg et al., ²⁰⁰²). Nonetheless, the value of Φ_Δ depends on the lifetime of the FMN triplet state (³FMN), which is, in turn, affected by how the protein matrix (i) quenches ³FMN and (ii) restricts oxygen diffusion toward the PS. Thus, protein engineering methods have been used to reduce the electron affinity of ³FMN by removing intermolecular H‐bonds and precluding photo‐induced electron transfer reactions between the isoalloxazine ring and the protein scaffold, thereby significantly increasing the Φ_Δ to values similar to those of free FMN (Westberg et al., 2015, ^2017a). But this approach is quite complex, requiring carefully removing protein–cofactor interactions that decrease the Φ_Δ value without significantly affecting the protein affinity of the cofactor. Moreover, this approach inevitably weakens the protein–cofactor bond and hence may lead to a undesirable cofactor dissociation.

The comparison between the Φ_Δ values of FMN and flavin adenine dinucleotide (FAD), which are 0.51–0.65 and ~0.07, respectively, shows the extreme sensitivity of the isoalloxazine moiety on a close environment (Baier et al., 2006; Westberg et al., ^2017b). Although numerous interactions affect the isoalloxazine moiety when encapsulated in the protein matrix of the flavoprotein miniSOG, the Φ_Δ of miniSOG was increased in two steps, first to 0.19–0.23 in the variant singlet oxygen photosensitizing protein (SOPP) (Westberg et al., 2015) and then to a similar value to that of free FMN (~0.6) in the variant SOPP3 (Westberg et al., ^2017a). Recently, another strategy how to increase miniSOG Φ_Δ value has been identified, namely the photo‐induced transformation of FMN into lumichrome, which increases the accessibility of oxygen to the isoalloxazine ring and makes protein quenching less favorable (Torra et al., 2019). These improved variants of flavin‐based GEPSs were result of replacing amino acids (by mutation or as a result of oxidation) responsible for: (i) steric barriers to oxygen diffusion toward the isoalloxazine ring, (ii) FMN triplet state quenching by electron transfer, and (iii) ¹O₂ quenching by chemical reactions. However, the increased values of Φ_Δ were offset by weakened cofactor–protein interactions, potentially releasing the cofactor, which is due to a reactive nature of free FMN, an unwanted outcome. In fact, the potential deleterious effects of this highly efficient PS are typically minimized by deeply burying the isoalloxazine moiety of flavin cofactors in the protein matrix of flavoproteins or storage proteins (Meissner et al., 2007; Stockwell & Thornton, ²⁰⁰⁶).

Considering the above and the findings of our previous study, in which we observed that the mutation of Cys450 localized close to the isoalloxazine ring for alanine affected both FMN dissociation and Φ_Δ (Petrencakova et al., 2020), we aimed at developing an alternative approach to increasing the Φ_Δ of GEPS system. The term “GEPS system” aims to point to the fact that our approach leads to the creation of two/three‐component system containing a GEPS (flavoprotein) and/or its apoform and free FMN, which arose from a single entity represented by a GEPS (flavoprotein). In the presented approach, we introduced strategic mutations to minimize the effects on protein–cofactor interactions in the native state of the flavoproteins, but which actively trigger FMN dissociation upon irradiation with blue light as a result of oxidation of the designed cysteine(s) in the binding site of isoalloxazine moiety. In such case, the potential of ¹O₂ production by isoalloxazine ring is fully unlocked near the GEPS by avoiding the two major reasons for the decreased Φ_Δ value: (i) quenching of the cofactor ³FMN state and (ii) limited oxygen diffusion due to encapsulation in a protein scaffold.

2. MATERIALS AND METHODS

2.1. Cloning, expression, and purification

The AsLOV2 domain and the corresponding mutants were cloned, expressed, and purified as previously reported (Petrencakova et al., 2020). Briefly, all proteins were expressed in Escherichia coli strain BL21(DE3). The bacterial cells were grown at 37°C in ampicillin containing (100 μg/mL) Terrific Broth (TB) medium to optical density (OD)₆₀₀ ~ 0.6–0.8. Protein expression was induced by adding isopropyl β‐D‐galactopyranoside (100 μM final concentration) following a temperature downshift from 37 to 25°C and expressed in the dark, overnight, at 25°C. The proteins were purified by metal ion affinity chromatography (Ni‐NTA Superflow, Qiagen) followed by purification on a Superdex 75 Increase, 10/300 GL, size exclusion column and concentrated in 20 mM TrisHCl buffer, pH 7.8. All steps were performed in the dark. The 280 nm/477 nm absorbance ratio of the final protein was ~2.6, suggesting the absence of the AsLOV2 apoform (Zayner et al., 2012).

2.2. Time‐resolved singlet oxygen phosphorescence detection

The experimental apparatus used for ¹O₂ detection was described in detail elsewhere (Hovan et al., 2023). Briefly, short 5‐ to 7‐ns‐long laser pulses with 475‐nm wavelength originating from an optical parametric oscillator pumped by an Nd: YAG Q‐switched laser excited 25 μM AsLOV2 samples (2 mL) at 10 Hz frequency. The average laser power was set to 0.8 mW with estimated beam diameter of 50 μm, and the samples were stirred with an overhead glass stirrer to minimize photobleaching. The time‐resolved ¹O₂ phosphorescence signal (at 1270 nm) was measured on a photomultiplier tube (Hamamatsu H10330A‐75), in photon counting mode, connected to a multichannel scaler PCI card (Becker&Hickl, MSA‐300). Spectral background was subtracted from the signal as explained in our previous study (Hovan et al., 2023). Two additional band‐pass filters were used to detect the background in the spectral regions neighboring the 1270 nm ¹O₂ peak. On average, 2500 laser pulses were consecutively detected with each filter throughout the experiments. A single phosphorescence kinetic curve was obtained in 12.5 min (3 × 2500 s pulses). It is important to note that the sample irradiation by the pulsed blue laser light during the ¹O₂ measurements also triggers the FMN dissociation process in our experiments.

2.3. Measurements of FMN triplet state lifetime

To monitor the FMN triplet state, we used an additional 633‐nm cw laser during the time‐resolved ¹O₂ phosphorescence detection, as detailed in our previous study (Petrencakova et al., 2020). Time‐resolved absorption at 633 nm was then recorded on an avalanche photodiode (Thorlabs, APD110A2) connected to a digitizing oscilloscope (Tektronix, DPO 7254). During the experiment, we consecutively collected the average signal from 2500 laser pulses throughout the irradiation process. Over 75 min, we constructed a total of 18 decay curves for each sample due to the utilization of three different filters to correctly subtract background and determine the amplitude of the measured signals. These curves allowed us to monitor the irradiation dependence of the triplet state amplitudes and lifetimes and obtain parameters used for an evaluation of the ¹O₂ phosphorescence data.

2.4. Spectral analysis

Ultraviolet and visible absorption spectra were recorded on ultraviolet (UV)‐2401PC (Shimadzu) and V‐750 (Jasco) UV–Vis spectrophotometers. Protein concentrations were calculated using an extinction coefficient of ε ₄₄₇ = 13,800/M cm for oxidized FMN. Fluorescence emission spectra were recorded on RF‐5301PC (Shimadzu) and FP 8550 (Jasco) spectrofluorophotometers. The emission spectra of FMN were measured using excitation wavelengths at 445 nm, and the fluorescence spectra were measured using a protein concentration of 10 μM. Circular dichroism spectra measurements were performed on a Jasco 810 spectropolarimetry (Jasco) at 10 μM protein concentration. The measurements in the far‐UV and near‐UV spectral regions were performed in a quartz cuvette with 1 mm and 0.5 cm pathlengths, respectively.

2.5. Determination of a light‐induced released FMN

Relative amounts of released FMN from AsLOV2 wt and its variants were determined by FMN fluorescence after filtration using 10 kDa cut‐off filter tubes. Each sample, that is, before and after irradiation with AsLOV2 wt and its variants, 900 μL of 10 μM protein, was loaded into an Amicon Ultra Centrifugal filter tube and centrifuged for 5 min at 7500g. After the spin, the collected flow‐through of each sample was measured for FMN fluorescence.

2.6. Thermal stability of AsLOV2 variants

The thermally induced transitions of AsLOV2 variants were measured on an RF‐5301PC spectrofluorophotometer by monitoring the fluorescence emission at 495 nm (λ _ex = 450 nm) while increasing the temperature, controlled with a Peltier block, from 25 to 90°C and recording data at a 1.5°C/min scan rate. All measurements were performed in 25 mM phosphate buffer, pH 7.9.

Thermal denaturation was analyzed by fitting normalized experimental data according to the following equation:

$$y_{\mathrm{obs}}=\frac{\exp \left[\left(∆H_{\mathrm{vH}}/R\right)\times \left(\left(1/T\right)-\left(1/T_{\mathrm{m}}\right)\right)\right]}{1+\exp \left[\left(∆H_{\mathrm{vH}}/R\right)\times \left(\left(1/T\right)-\left(1/T_{\mathrm{m}}\right)\right)\right]}$$ (1)

where y _obs is the experimentally measured parameter, ΔH _vH is the van't Hoff enthalpy change, R is the gas constant, and T _m is the transition temperature.

The thermal stability of the study proteins was also analyzed by differential scanning calorimetry (DSC). DSC measurements were performed on a VP‐DSC differential scanning microcalorimeter (Microcal, USA), and the stabilities of the variants were assessed by determining the temperatures at the maxima of the heat capacity change curves of the corresponding proteins.

2.7. Molecular modeling

The Schrödinger suite of programs (Schrödinger, 2022) was used to process the AsLOV2‐related protein structures downloaded from the Protein Data Bank (PDB) database (2V1A.pdb; 2V1B.pdb) (Halavaty & Moffat, 2007). The protein structures were treated and prepared for calculations in Maestro (2022). The updated structures were then single (V416C, V416M, T418C, and T418M) and double mutated (V416C/T418C, V416C/T418M, V416M/T418C, and V416M/T418M). The geometries of the mutated structures were optimized and then solvated with water molecules added with a 1 nm buffer at neutral pH around the proteins. The geometries of the resulting solvated protein structures were then minimized again and equilibrated. After equilibration, the final structures were submitted for 500 ns constant‐temperature, constant‐pressure ensemble (NPT) (pressure at 1.01325 bar) molecular dynamics (MD) simulations with the Desmond program (Shaw, 2020) at 300 K. Molecular geometries resulting from simulations were saved at 100 ps intervals and used for further analysis. Molecular complexes were visualized in the Maestro program.

2.8. Free energy of binding

We calculated the binding free energy of FMN to the AsLOV2 domain, that is, potential of mean force (PMF), by pulling an FMN with the constant velocity (v _z = 1 Å/ns) through steered molecular dynamics (SMD) simulation from z = 7 (inside of the AsLOV2 domain) to z = 25 (outside of the AsLOV2 domain) (see Figure S1 for further details on the SMD). At each window, the system qA equilibrated for 1 ns while constraining the position of FMN and the α‐atoms of the protein at 310 K using the CHARMM36m force field (Huang et al., 2017). The initial structure of the complex of FMN and AsLOV2 domain was adapted from the crystallographic structure (PDB ID: 2v0u) (Halavaty & Moffat, 2007). We also adapted three mutants (V416C, T418C, and V416C/T418C) of the AsLOV2 domain to compare the effect of the position of cysteine residue (Figure S2) on FMN binding. For this purpose, we divided the path from z = 7 to z = 15 into eight smaller windows (the length of each window is 1 Å), performing eight independent SMD simulations in each window. From z = 15 to z = 25, we divided the path into five smaller windows (2 Å window length), also performing eight independent SMD simulations in each window. PMF was constructed from SMD simulations based on the Jarzynski equality equation (Jarzynski, 1997; Park & Schulten, ²⁰⁰⁴).

$$∆A=-{\beta }^{-1}\ln ⟨\exp \left[-\mathit{\beta W}\right]⟩,$$ (2)

where ∆A is a free energy difference, ß is the product of the Boltzmann factor and temperature, and W is the non‐equilibrium work derived from SMD simulations. The non‐equilibrium work of the pulling force was assessed using the following equation:

$$W=-\mathit{kv}{\int }_0^{t}d{t}^{\prime }\left[z\left(t^{\prime }\right)-z_0-v{t}^{\prime }\right],$$ (3)

where k and v are the force constant (418.4 kJ/mol/Ų) and velocity of pulling (1 Å/ns), and z(t′) and z ₀ are the reaction coordinate at t′ in the simulation and the initial position of the center of mass of FMN, respectively. Using a second‐order cumulant expansion of Equation (2), we derived the following equation:

$$\mathrm{\Delta }A=⟨W⟩-\frac{\beta }{2}\left[⟨W^2⟩-{⟨W⟩}^2\right].$$ (4)

The system with 66 Å × 66 Å × 78 Å dimensions was filled with 10,612 modified TIP3P water (Figure S3) (Price & Brooks, 2004), and the pressure of 1 atm was maintained using the Langevin piston method with a piston period of 100 fs, a damping time constant of 50 fs, and a piston temperature of 310 K (Feller et al., 1995; Martyna et al., ¹⁹⁹⁴). Full electrostatics was applied using the particle‐mesh Ewald method with a 1 Å grid width (Darden et al., 1993). A group‐based cut‐off was used to calculate nonbonded interactions with a switching function, updating every tenth time‐step. Using the SHAKE algorithm, the hydrogen bond was kept rigid, with a 2 fs time‐step (Andersen, 1983). All simulations were performed using nanoscale molecule dynamics (Phillips et al., 2005), and the graphics shown in this report were prepared using visual molecular dynamics (Humphrey et al., 1996).

2.9. Top–down mass spectrometry

To prevent light‐induced oxidation, the denatured protein samples were desalted in the dark. Protein samples were denatured by mixing them with 0.1% formic acid (FA) in a 1:1 (v:v) ratio. Denatured protein samples (10 μg) were loaded onto a reverse‐phase microtrap column (Optimize Technologies, USA), desalted using 0.1% FA (3 × 250 μL). Subsequently, the samples were eluted with 80% methyl cyanide (MeCN), 0.1% FA. The desalted samples were then diluted 50× using 30% MeCN, and 0.1% FA before being sprayed into a solariX‐XR mass spectrometer equipped with a 15 T magnet (Bruker Daltonics, Billerica, USA) and an electrospray ionization (ESI) source and operated in positive ionization mode. Mass spectrometry (MS) spectra were acquired in a broadband mode ranging from 207.20 to 3000 m/z. The collision voltage was set to −5.0 V at 150°C desolvation temperature, 0.200 s ion accumulation time, and 1 ms time of flight, acquiring 128 scans using a 2 M data point transient starting at 207.20 m/z. For Tandem mass spectrometry method (MSMS), oxidized clusters were isolated in a quadrupole and transferred into the collision cell for collision‐induced dissociation (CID). Oxidized ion clusters were isolated for four charge states using multi‐continuous accumulation of selected ions (multi‐CASI) mode at the following nominal masses: 947.00 amu (21+), 996.00 amu (20+), 1046.00 amu (19+), 1108.00 amu (18+). CID was induced by setting a quadrupole selection window of 15.00 amu for each charge state and applying −13.00/−15.00/−17.00/−19.00 V collision voltage for +21/+20/+19/+18 charge states, respectively, setting ion accumulation at 0.5 s. MSMS spectra were acquired by collecting 256 scans with a 2 M data point transient starting at 207.20 m/z.

2.10. Bottom–up mass spectrometry

For the analysis was taken an aliquot from each sample corresponding to 0.5 µg of the protein. The samples were reduced and alkylated by adding TCEP [tris(2‐carboxyethyl)phosphine] and chloroacetamide at a concentration of 10 mM and 30 mM, respectively. Then, the protein samples were incubated at 90 °C for 2 min and cooled down on the bench. Further, samples were diluted using 50 mM 4‐ethylmorpholine buffer pH = 8.3 containing 5% acetonitrile (ACN) and digested by trypsin (ratio 2:1) at 37 °C for 2 h. Digestion was terminated by adding trifluoroacetic acid (TFA) to a final concentration of 0.5%. Samples were desalted using StageTip approach and dried. Prior LC‐MSMS analysis samples were dissolved in containing 2% ACN and 0.1% TFA. Shotgun proteomic analysis was performed using the Vanquish liquid chromatography system (Thermo Scientific) coupled to the timsToF SCP mass spectrometer equipped with Captive spray (Bruker Daltonics) and operated in a positive data‐dependent mode. Peptides were trapped using C18 trap column (PepMap Neo C18 5µm, 300 µm × 5 mm, Thermo Scientific) and separated on a C18 analytical column (DNV PepMap Neo 75 µm × 150 mm, 2 µm, Thermo Scientific) using a linear gradient of 5% to 35% ACN over 50 minutes at a flow rate of 350 nL/min. Both the trap and analytical columns were heated to 50 °C. The timsTOF SCP settings were based on the standard proteomics PASEF method. Only light‐induced samples were measured in DDA mode to prepare spectral library. Quantification of modification extent was done using data acquired by DIA mode on the timsTOF SCP. DIA analysis was done using the same LC setup. All samples were run in triplicates.

2.11. Data processing

Top–down raw data were deconvoluted in Data Analysis software, version 5.3 (Bruker Daltonics) and processed as previously described (Polak et al., 2022; Yassaghi et al., ²⁰²²). Fragment ions from raw spectra were assigned using an in‐house software tool, and a search was conducted against an in silico library of either wild type or mutant protein sequences with a mass accuracy of 2 ppm. The results were validated manually.

DDA data were processed by FragPipe search workflow (Kong et al., 2017; Yu et al., ²⁰²⁰). The search was performed against a database containing sequences of wild type and respective mutants of AsLOV2 and cRAP contaminant sequences. Precursor ion tolerance was set at 12 ppm, and the mass tolerance for MS/MS fragment ions was set at 0.05 Da. Carbamidomethylation of cysteine and various oxidative modifications of Met, Phe, Trp, Tyr, His, Pro, and Cys (+O, +2O, +O₂‐H₂O, etc.) were set as a variable, the false‐discovery ratio was set to less than 1%. Obtained results were exported to the Skyline software (MacLean et al., 2010), where fragmentation spectra were manually validated. Quantification was done using DataAnalysis 5.3 software and in‐house python scripts. Data were corrected by subtracting background oxidation. The extent of oxidative modifications and statistical analysis were performed according to the work of Loginov et al. (2022).

3. RESULTS AND DISCUSSION

3.1. Design of efficient GEPS system from AsLOV2 wt

The AsLOV2 domain belongs to the superfamily of Per‐Arnt‐Sim (PAS) domains, which are found in various sensor proteins, in organisms ranging from archaea to eukaryotes (Taylor & Zhulin, 1999). The crystal structure of AsLOV2 displays the typical PAS fold, flanked by a conserved N‐terminal turn‐helix‐turn motif and a C‐terminal flanking region containing an amphipathic Jα helix. These regions dock on the LOV2 core domain and bury several hydrophobic residues of the central β‐sheet of the core domain (Figures 1a and S3) (Halavaty & Moffat, 2007). The isoalloxazine moiety is buried deep in the protein, connected to solvent molecules through several tunnels, as shown in our recent study (Petrencakova et al., 2020). Based on our previous results, we hypothesized that suitably replacing small amino acids near the isoalloxazine ring with ¹O₂ oxidation‐prone amino acids, such as cysteine or methionine, could lead to irradiation‐triggered flavin cofactor dissociation. To identify suitable amino acids for replacement, we applied the following exclusion criteria: (i) conservative amino acids—to avoid adversely affecting flavin binding, (ii) large amino acids—replacing large with small amino acids is sterically meaningless as the increase of their volumes by oxidation may not be sufficient to create “crowding effect” in the isoalloxazine moiety binding site, and (iii) positively charged amino acids—they may stabilize negatively charged flavin. As a result, only five amino acids (Val416, Thr418, Asn425, Thr458, and Val463) were left close to the isoalloxazine ring (Figures 1c and S4). After visual inspection, we decided to replace Val416 and Thr418 by cysteinyl and methionyl residues. Of the eight possible variants, only three containing cysteines, that is, V416C, T418C, and V416C/T418C, were expressed in E. coli in ample concentration for biophysical characterization. These observations matched our MD simulations (Figure S4), thus indicating that replacing any of the two positions by methionyl residues perturbs helices and most likely decreases the stability of the whole protein structure.

FIGURE 1.

(a) Crystal structure of LOV2 domain of Avena sativa (AsLOV2; PDB: 2v0u). The N‐terminal turn‐helix‐turn motif is shown in dark blue, and the Jα helix is shown in red. (b) Detailed view of the binding site of the isoalloxazine moiety, indicating a tight arrangement around the isoalloxazine ring. (c) Isoalloxazine ring‐binding site, highlighting the five amino acids that may be replaced based on the selection criteria (see main text). The amino acids Val416 and Thr418 were modified in this study.

3.2. Conformational and stability analysis

To assess the possible impact of the mutations on the stability and conformational properties of the proteins, we performed (i) in silico analysis of FMN binding to AsLOV2 variants, (ii) circular dichroism (CD) measurements in the far‐ and near‐UV spectral regions, and (iii) thermal denaturation experiments, by following the variation of FMN fluorescence as a function of temperature, and DSC measurements of all AsLOV2 variants. FMN binding to the proteins was analyzed in silico by calculating the PMF by pulling an FMN at constant velocity in SMD simulations (Figure S1). No significant differences in dissociation constants of the FMN cofactor were found when comparing the free energy of pulling between wild type and mutant AsLOV2 variants (Figure 2a).

FIGURE 2.

(a) Free energy profiles of flavin mononucleotide (FMN) binding to wild type and mutant LOV2 domain of Avena sativa (AsLOV2) domains are shown as a function of the position of the center of FMN. The origin of the z‐coordinate (z _FMN = 0) corresponds to the center of the AsLOV2 domain (Figure S1). Snapshots at z _FMN = 7, 11, 15, 19, and 23 Å are shown. The position of FMN in the crystal structure (PDB ID = 2v0u) corresponds to z _FMN ≈7.8 Å. (b) and (c) far‐ and near‐UV circular dichroism spectra of the AsLOV2 variants expressed as the mean residue ellipticity and molar ellipticity, respectively. (d) Thermal denaturation of AsLOV2 monitored by FMN fluorescence. The symbols and lines correspond to experimental data and fits according to Equation (1), respectively. (e) DSC thermograms of AsLOV2 variants. AsLOV2 variants are colored as follows: WT (black), V416C (blue), T418C (red), and V416C/T418C (green) (left box).

The mutations in positions 416 and 418 induced only small changes in the secondary structure of AsLOV2, as shown by CD spectra in the far‐UV region. In two variants, V416C and T418C, the ellipticity in the far‐UV region at 220 nm, reflecting a contribution of the α‐helical conformation, slightly increased by ~20% and 15%, respectively. By contrast, in the double mutant, V416C/T418C, the ellipticity decreased by ~15% at 220 nm (Figure 2b).

In turn, the near‐UV spectra of AsLOV2 wt, V416C, and T418C significantly overlap, indicating unperturbed tertiary structure around the aromatic amino acid residues (Figure 2c) (Kelly et al., 2005). The double mutant showed a slightly decreased signal in the near‐UV region, which may either reflect small changes in the tertiary structure of the protein as indicated by decreased thermal stability of the double mutant and FMN fluorescence analysis.

The thermal denaturation of the proteins was analyzed by monitoring the variation of flavin fluorescence as a function of temperature to assess protein stability upon flavin binding (Figure 2d) and by DSC analysis to assess the global stability of the tertiary conformation (Figure 2e). Both approaches indicate the thermal stabilities of the single variants decreased similarly and only slightly (<4°C), on both local and global levels, indicating unperturbed cooperativity interactions in the protein variants. The double variant transition temperature decreased by nearly 8°C, as determined by both methods, suggesting moderate destabilization of the protein structure. Nevertheless, the cooperativity of the transition was not perturbed. Ultimately, we measured the following transition/melting temperatures (CD/DSC) of the AsLOV2: WT (57.2°C/59.3°C), V416C (53.4°C/54.9°C), T418C (56.4°C/60.6°C), and V416C/T418C (47.7°C/53.4°C).

3.3. FMN triplet state lifetimes

The analysis of the transient absorption data showed double exponential curves describing the decay of the absorption signal, except for the T418C variant of AsLOV2, in line with our previous findings for the wild type of AsLOV2 (Petrencakova et al., 2020). Based on these results, two populations of FMN were identified, one located inside the protein matrix and the other in the solvent. For the T418C variant, transient absorption was adequately represented by a single exponential curve, suggesting either one population of FMN or, more realistically, two populations with the same or very similar lifetimes. The experimental data were fitted by the following equation:

$$\text{Absorbance}=A_1{\mathrm{e}}^{-\left(t/{\tau }_{\text{FMNin}}\right)}+A_2{\mathrm{e}}^{-\left(t/{\tau }_{\text{FMNout}}\right)}.$$ (5)

For T418C, we only used one term of Equation (5). The experimental data and the corresponding fits are shown in Figure S5. The resulting fits reproduce the measured data well. Based on our previous study (Petrencakova et al., 2020), the amplitudes of the FMN inside the proteins should decrease and the amplitudes of the free FMN in the solvent should increase with the irradiation time. These correlations were observed in all AsLOV2 variants, except for T418C because only one amplitude was assessed for this variant. However, the amplitude of T418C remained constant throughout the experiment (Figure S6). As shown by the lifetimes of individual triplet states listed in Table S1 (Figure S7) and the ¹O₂ lifetime (see below), the free FMN lifetime values of all variants were close to 2.7 μs, which is the lifetime of free FMN in air‐saturated water at 37°C (Westberg et al., ^{2017a, 2017b}). These results support the hypothesis that FMN dissociates from the protein matrix upon irradiation.

The lifetime of FMN inside the WT protein was similar to that reported in our previous study (1.57 μs) (Petrencakova et al., 2020). Conversely, the corresponding value of the V416C variant was half of the WT value. This finding indicates that substitution with cysteine provides an efficient pathway to quench the FMN triplet state or to produce an adduct similar to C450.

For the T418C variant, we assessed either a single triplet state lifetime or two with almost identical values. One way or the other, the cysteine at position 418 of the amino acid sequence likely affects the FMN position in the protein matrix, preventing C450 from creating a covalent adduct with FMN. Therefore, substitution with cysteine increases the efficiency of ¹O₂ production (see below) by promoting the dissociation of FMN to the surrounding solvent.

For V416C/T418C, we found practically the same lifetime as for WT. This suggests that the effects of the V416C and T418C mutations somehow cancel each other. Based on the FMN triplet state lifetime values, either the electron is transferred from FMN to one of the cysteines or the competition between interactions is not detectable.

3.4. Singlet oxygen phosphorescence measurements

Singlet oxygen phosphorescence kinetics at different stages of irradiation are shown in Figure 3. The time between individual curves was 12.5 min. The data were fitted using the following equation (Jimenez‐Banzo et al., 2008; Ragas et al., ²⁰¹³):

$$P\left(t\right)=\begin{array}{l}{A}_{\text{in}}\frac{{\tau }_{∆\text{in}}{\tau }_{\text{FMNin}}}{{\tau }_{∆\text{in}}-{\tau }_{\text{FMNin}}}\left({\mathrm{e}}^{-\left(t/{\tau }_{\text{FMNin}}\right)}-{\mathrm{e}}^{-\left(t/{\tau }_{∆\text{in}}\right)}\right)\\ +A_{\text{out}}\frac{{\tau }_{∆\text{out}}{\tau }_{\text{FMNout}}}{{\tau }_{∆\text{out}}-{\tau }_{\text{FMNout}}}\left({\mathrm{e}}^{-\left(t/{\tau }_{\text{FMNout}}\right)}-{\mathrm{e}}^{-\left(t/{\tau }_{∆\text{out}}\right)}\right).\end{array}$$ (6)

FIGURE 3.

Upper row—time‐resolved singlet oxygen phosphorescence of the LOV2 domain of Avena sativa (AsLOV2) variants, with the same scaled y‐axis for all plots. The color scheme represent the following accumulated incident energy: black open squares 0.2 J, red open circles 0.8 J, blue triangles 1.4 J, green reverse triangles 2 J, purple diamonds 2.6 J, and yellow ochre left triangles 3.2 J. The black lines correspond to fits by Equation (6). Lower row—fluorescence spectra of the AsLOV2 variants before irradiation (red curves), free flavin mononucleotide (green curves), and AsLOV2 variants after irradiation (black dotes). The blue line is the linear combination of the red and green curves. ¹O₂, singlet oxygen.

Even though it is possible to fit the phosphorescence kinetics just with the second term of the Equation (6), like it was shown in our previous study (Petrencakova et al., 2020), both FMN populations contribute to ¹O₂ phosphorescence. To minimize the number of free parameters in Equation (6), the lifetimes of individual FMN populations were set to the lifetimes obtained from transient absorption measurements and the value of ${\tau }_{∆\text{out}}$ to was set to be 3.5 μs, which corresponds to the ¹O₂ lifetime in pure water (Westberg et al., ^{2017a, 2017b}). The corresponding fits in Figure 3 reproduce the measured kinetics. The resulting ¹O₂ lifetimes measured at different stages of the experiment are listed in Table S1. In each protein variant, the effective lifetime ${\tau }_{∆\mathit{\text{in}}}$ increased toward the ¹O₂ lifetime in pure water (3.5 μs) (Westberg et al., ^{2017a, 2017b}). It is important to note, that ${\tau }_{∆\text{in}}$ represents an effective lifetime value, which reflects the changes due to gradual oxidation of ¹O₂ quenchers (i.e., amino acids) and also the diffusion of ¹O₂ out of the protein matrix.

Singlet oxygen phosphorescence increased over time, as shown in the kinetics curves (Figure 3). This result supports our assumption, based on our recent work (Petrencakova et al., 2020), that FMN dissociates to the solvent upon irradiation, producing ¹O₂. The wild type and V416C variants displayed similar trends. After the first round of irradiation, a significant portion of FMN was already dissociated from V416C/T418C and found in the solvent.

The amount of FMN released into the solvent upon irradiation was assessed by FMN fluorescence. Because FMN fluorescence spectra after irradiation can be linearly combined with the corresponding spectra of the protein before irradiation (red color) and with the fluorescence spectrum of free FMN (green color) (Petrencakova et al., 2020), we obtained closely correlative fits of the fluorescence spectra of the protein variants after irradiation, as shown in Figure 3 (lower row). The close fits (blue lines) and the presence of isosbestic points in the fluorescence spectra of the studied AsLOV2 systems validate our approach.

Based on this analysis, we were able to order the AsLOV2 variants by their efficiency of FMN release to the solvent upon 75 min irradiation (expressed as a percentage in brackets) as follows: WT(33%) < T416C(41%) < T418C(51%) < V416C/T418C(65%). The percentage of FMN dissociation from AsLOV2 wt in this study was lower than in our previous study (Petrencakova et al., 2020). This difference may be related to minor beam diameter adjustment in comparison with the previous experiments (Petrencakova et al., 2020).

The results shown in Figure 3 were further analyzed by plotting the curves of the amplitudes obtained from Equation (6), corresponding to the amount of ¹O₂ production as a function of time (Figures 4a and S8). Assuming that most ¹O₂ is produced only by free FMN (Petrencakova et al., 2020), the slopes provide the relative rates of FMN dissociation to the solvent. Accordingly, the slopes can be ordered as follows: WT < V416C < T418C < V416C/T418C. Variant V416C/T418C releases FMN faster than any other variant (Figures 3 and 4). In Figure S8, the irradiation dependence of [A _in] is also visible for each protein variant. These changes (minor increase) might be attributed to changing quantum yield of the proteins as the proteins become gradually oxidized (Pimenta et al., 2013; Torra et al., ²⁰¹⁹). As it is clearly indicated in absorbance spectra of AsLOV2 wt and its variants measured after irradiation (Figure S9), in our experimental setup, there is no evidence of phototransformation of FMN to lumichrome as it has been previously observed by Torra et al. (2019). For completeness, we note that the formation of lumichrome in AsLOV2 systems studied was observed but only upon prolonged light irradiation suggesting the phototransformation of the free FMN.

FIGURE 4.

(a) Variation of amplitude [A _out] derived from Equation (6) as a function of time, reflecting the singlet oxygen phosphorescence of each LOV2 domain of Avena sativa (AsLOV2) variant shown in Figure 3 (upper row): wt (black squares), V416C (blue triangles), T418C (red circles), and V416C/T418C (green reverse triangles). (b) Correlation between the amplitude [A _out] derived from Equation (6) after the last round of irradiation and flavin mononucleotide (FMN) dissociation, as calculated from Figure 3. The correlation is described by linear equation y = 2.06x + 22.12 with R ² = 0.998. (c) Fluorescence intensity of the irradiation‐released FMN of the corresponding AsLOV2 after filtration experiment. No fluorescence was present in the filtrates of AsLOV2 wt and its variants before irradiation.

The ¹O₂ phosphorescence of the V416C/T418C variant indicates the most efficient FMN release. However, corresponding slope in Figure 4a is the lowest, suggesting the presence of free FMN after the first round of irradiation, is most likely due to ambient light exposure induced FMN dissociation before the experiment. This result underscores the high sensitivity of the variant V416C/T418C to irradiation.

The phosphorescence values derived from Figure 3 (upper row) were also plotted as the highest amplitude of the ¹O₂ phosphorescence as a function of the percentage of FMN dissociation from the corresponding proteins (Figure 4b). The resulting plot showed a strong correlation, suggesting a causal relationship. Combined, these findings support our model based on ¹O₂ production by free FMN.

To assess the relative extent of FMN dissociation induced by the light irradiation, we performed filtration experiments as in our previous report (Petrencakova et al., 2020). In these experiments, the released FMN passes through the filter, while the protein is retained. Obtained results show that the relative release of FMN correspond to the presented results from phosphorescence and fluorescence experiments (Figures 3 and 4a,b).

3.5. Oxidative modification of the AsLOV2 variants

FMN release is accompanied by ¹O₂ production and modification/oxidation of solvent‐accessible protein residues, including residues in the dissociation channel, as it has been previously shown by mass spectrometry (Petrencakova et al., 2020). Halved samples were first analyzed using the top–down approach. Before and after irradiation (Figure S10), desalted protein samples were sprayed via an ESI source in broadband m/z mode to confirm the mass and thus the site‐specific mutation, and to show the homogeneity of the samples, meaning that the site‐specific mutations had no further effect on protein charge for ESI spraying. The zoom on 21+ charge state (Figure S11) shows predominant N‐terminal methylation (+14.0156 amu) even before irradiation (gray spectra). After irradiation, small but, nevertheless, distinct changes in oxidation pattern appeared, thereby confirming ¹O₂ oxidation of protein residues. Oxidized clusters were isolated and fragmented in multi‐CASI mode by CID generating complementary b‐ and y‐fragment ions. The extent of their oxidation was calculated for singly and doubly oxidized fragment ions (Figures 5 and S12; Table S2), and 7 and 11 b‐ and y‐ions were simultaneously found in MSMS replicated spectra, respectively. We omitted the b‐ions from the final interpretation for the reason that N‐terminal methionine is extensively methylated (+14.0156 amu) and thus biases with the extent the oxidation of b‐type fragment ions (+15.9949 amu).

FIGURE 5.

(a) The results of the extent of oxidation of selected b‐ and y‐fragment ions of irradiated samples are expressed as the mean ± standard deviation (SD) of three independent measurements. The inset in the y‐ions plot indicates the extent of doubly oxidized y127 and y148 fragment ions. (b) The sequence of wild type LOV2 domain of Avena sativa (AsLOV2) protein with the denoted fragment ions is displayed at the top of the figure. The His‐tag, which is not part of the AsLOV2 sequence, is colored in gray and underlined in red. The positions of the mutations V416C and T418C are shown in blue and red squares, respectively.

In the y‐ions, the first oxidized fragment is y32, which covers and pinpoints the oxidation to His519 residue. Fragment ion y31 also covers His519 but was not observed as oxidized. We manually validated spectra and found that unmodified y31 fragment ion was just at the level of noise that is why its oxidized form was below the level of detection. The next major increase of oxidation was observed between y36 and y127, covering P420–I510 region. The oxidation of residues is significant in this region (Figure 5; Table S2), with the highest extent observed in V416/T418 mutant. This is in accordance with the DSC studies where this mutant displays moderate destabilization of the protein structure, which likely leads to an exposing of some residues in this region to solvent and their subsequent oxidation. Another increase in an oxidation extent can be observed between y127 and y148 fragment ions, covering both mutations and F415 located inside the FMN binding pocket. As can be seen from Figure 5 and Table S2, single and double oxidations follow the increasing trend WT ~ V416C/T418C < V416C ~ T418C. The overall extent of oxidation between y127 and y148 fragment ions is driven by F415, C416, and C418 residues for all proteins.

The second half of the samples was analyzed using the bottom–up approach. Samples were in‐solution digested by trypsin for a short time to prevent oxidation during long incubation times.

The peptides were then identified by a searching engine, including their singly, doubly, and triply oxidized forms. Out of the identified peptides and residues, several residues throughout the sequence were observed as oxidized but at a low and nonsignificant extent, mostly <1%. Nevertheless, peptides containing embedded cysteine mutations were identified and their extent of oxidation was calculated for all three mutated variants. We also identified two peptides with the sequence ⁴⁴³EEILGRNCR and ⁴⁴⁹NCRFLQGPETDR containing C450. However, generating two peptides during trypsin digestion resulted in a decreased peptide signal in liquid chromatography‐mass spectrometry (LCMS) trace, and the calculated extent of oxidation, which differs in both peptides, was observed just above the level of detection. Thus, the calculated extent of oxidation for C450 in Table 1 is a weighted arithmetic average of the extents from both peptides. The extent of modification was then calculated for wt and all three variants (Table 1). The bottom–up dataset is in general in high agreement with our top–down dataset. As can be seen from Table 1, the F415 and C450, located near to and inside the binding pocket, respectively, were both observed as oxidized. When cysteine mutation is introduced inside the binding pocket, the extent of F415 oxidation decreased as cysteines, more sensitive to ¹O₂, were observed predominantly as triply oxidized. However, the effect of mutation results in different outcomes for particular mutation. While the V416C oxidation leads to FMN stabilization and the C416 scavenges ¹O₂, the T418C and V416C/T418C mutations more likely destabilize the FMN binding, which is supported by protecting the F415 from oxidation and not observing compensation of oxidation on C450 and mutated residues (Table 1).

TABLE 1.

List of selected oxidation products and their extent of oxidation.

Residue	Modification a	The extent of modification, b %
		WT	V416C	T418C	V416C/T418C
F415	1O	20.37	10.17	12.38	9.22
V416C	3O		21.61		5.04
T418C	3O			0.80	0.38
C450	3O	3.42	3.35	1.71	1.50

^a Single modifications were identified using ion mobility.

^b Data represent values with subtracted background oxidation level (S _irradiated − S _control).

Based on our MS data, we reached the following conclusion: (i) the modifications of AsLOV2 variants are light‐dependent as the overall extent of the oxidative modifications is much higher in irradiated samples of all AsLOV2 variants than in non‐irradiated samples (Figures S10 and S11), (ii) the same amino acids are oxidized in all AsLOV2 variants (Figures 5 and S12), albeit to a different extent, and (iii) the residues are oxidized at a different extent due to the conformational changes and mild protein destabilization, observable mostly on both T418C and V416C/T418C protein variant. Conclusions (i) and (ii) strongly indicate that the AsLOV2 variants share similar conformational properties, and conclusion (iii) supports more efficient FMN dissociation from the AsLOV2 variants.

Because FMN is less efficiently dissociated from AsLOV2 C450A than from AsLOV2 wt, the oxidation damage is larger in the C450 variant than in the WT protein (Petrencakova et al., 2020). The extent of oxidation (both singly and doubly) between y127 and y148 ions, both of which contain extra cysteine(s) in position(s) 416 and 418, increased similarly in WT, V416C, and T418C, but not in the V416C/T418C double mutant (Figure 5; Table S2). This suggests that FMN is released more quickly from the double mutant than from the single variants and produce ¹O₂ in the solvent, in consent with the fluorescence spectra shown in Figure 3.

4. CONCLUSIONS

Our approach to the design of efficient flavin‐based GEPS system relies on controlled cofactor dissociation as a direct result of irradiation at a suitable wavelength. In our particular case, it depends on a natural propensity of flavoprotein AsLOV2 to bind highly efficient PS represented by FMN, which upon blue light irradiation oxidizes close ¹O₂ prone oxidation cysteinyl residue(s) and this chemical modification triggers its dissociation. Assuming tight encapsulation of the isoalloxazine ring in its binding site, the design method is based on an identification of amino acid positions occupied by nonconservative amino acids with small side residues near the isoalloxazine moiety. Replacement of the small amino acid by amino acid, which becomes upon ¹O₂ oxidation larger than the original one leads to a compensatory steric effect, accompanied by cofactor release. Applying this approach, we were able to increase the Φ_Δ values in all prepared AsLOV2 variant systems. In particular, the Φ_Δ value of AsLOV2 V416C/T418C system after irradiation is ~16 times higher than that of AsLOV2 wt system before irradiation without significantly compromising protein stability and FMN binding affinity. The presented proof‐of‐principle study demonstrates that this approach is a viable alternative to the traditional design of the efficient GEPSs. Our GEPS system fulfills two critical properties of GEPS: (i) ability to target PS by suitable tags and (ii) increased Φ_Δ value.

The potential limitations of our GEPS system is related to the limited spatial specificity due to a diffusion of the irradiation‐triggered release of PS. Spatial specificity is very critical for the CLEM and PDD. As such, our AsLOV2 system is unlikely suitable for use in CLEM and PDD applications. However, we believe that AsLOV2 system might be useful in CALI of proteins and photoablation of cells as discussed for example in Souslova et al. (2017). In fact, our AsLOV2 system is related to currently developing drug carrier systems for PDT based on liposomes and polymer nanoparticles, which release PS loads at or in the target cells due to a change in the physico‐chemical property of solution environment such as pH, ionic strength, or temperature (Akasov et al., 2022; Debele et al., ²⁰¹⁵). In the case of AsLOV2 system, the spatial specificity of FMN action is determined by the applied light as only the incident light triggers the release of PS and a production of ¹O₂. The efficiency as well as the mechanism of action of the GEPS system proposed in this work needs to be determined in an analogous study as the one very recently performed by Mogensen et al. (2021).

AUTHOR CONTRIBUTIONS

Erik Sedlák: Conceptualization; formal analysis; supervision; funding acquisition; writing – review and editing. Kristína Felčíková: Investigation; formal analysis; writing – original draft. Andrej Hovan: Investigation; formal analysis; writing – original draft. Marek Polák: Investigation; formal analysis; writing – original draft. Dmitry S. Loginov: Investigation; formal analysis. Veronika Holotová: Investigation; formal analysis. Carlos Díaz: Investigation; formal analysis. Tibor Kožár: Investigation; formal analysis; writing – original draft. One‐Sun Lee: Investigation; formal analysis; writing – original draft. Rastislav Varhač: Investigation; formal analysis. Petr Novák: Formal analysis; writing – original draft; conceptualization. Gregor Bánó: Formal analysis; writing – original draft; conceptualization.

Supporting information

Data S1. Supporting Information.

Figure S1. A schematic view of the complex of the AsLOV2 domain and FMN.

Figure S2 and S3. The top and side views on AsLOV2 variants and WT, respectively.

Figure S4. MD simulation structures of AsLOV2 variants.

Figure S5. Transient absorption measurements.

Figure S6. Time evolution of FMN triplet states amplitudes.

Figure S7. Time evolution of ³FMN amplitudes.

Figure S8. Time evolution of FMN amplitudes related to singlet oxygen production.

Figure S9. Absorption spectra of ASLOV2 wt and its variants.

Figure S10. Broadband ESI‐MS spectra of ASLOV2 wt and its variants.

Figure S11. A zoom into the ESI‐MS spectra 21+ charge state.

Figure S12. The sequence of wild type AsLOV2 protein with the denoted fragment ions.

Table S1. The ³FMN and singlet oxygen lifetimes at different stages of the irradiation experiment.

Table S2. The extent of oxidation for b‐ and y‐fragment ions generated by multiCASI‐CID. This material is available free of charge via the Internet at http://pubs.acs.org.

Felčíková K, Hovan A, Polák M, Loginov DS, Holotová V, Díaz C, et al. Design of AsLOV2 domain as a carrier of light‐induced dissociable FMN photosensitizer. Protein Science. 2024;33(4):e4921. 10.1002/pro.4921