In silico design and expression of anti-E1E2 CHIKV scFv in biotinylated form using Escherichia coli Origami B (DE3) for immunochromatographic detection of the Indonesian Chikungunya variant

Homology modelling of E1E2 CHIKV Ind

The E1E2-Ind sequence was obtained from GenBank (AHA87256.1), and the template search for homology modelling was performed using protein BLAST/blastp (https://blast.ncbi.nlm.nih.gov/). Subsequently, the 3D structure of E1E2-Ind was modelled using Modeller9.18 (http://salilab.org/modeller/), and the best 3D structure was selected based on the lowest discrete optimized protein energy score, as a lower value reflects the associated model is the more native-like (Hardianto et al., 2017). The model was further evaluated by the Ramachandran plot generated by the PROCHECK webserver (http://www.ebi.ac.uk/) and then subjected to iterative loop refinement using Modeller9.18 until more than 90% of residues occupied the most favourable regions in the Ramachandran plot.

Design of anti-E1E2 scFv and its complex with E1E2-Ind

The anti-E1E2 scFv was designed and infused with a BAD to form anti-E1E2-scFv-BAD (hereafter referred to as scFv-BAD), the structure of which originated from the African anti-E1E2 CHIKV Fab (Fab CHK152) (PDB ID: 3J30) (Voss et al., 2010). The design of scFv began with the search for conserved domains in 3J30 using the NCBI-conserved domain website (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi) to obtain the conserved VL and VH sequences. Then, a linker (GGGGS)₄ was added between VL and VH, before the attachment of BAD (PDB ID: 2EBV) to the C-terminal to form scFv-BAD. All steps in the design of the scFv-BAD structure were conducted using BIOVIA Discovery Studio Visualizer 2019, and the model was further iteratively refined using Modeller9.18.

The docking of scFv-BAD to E1E2-Ind was adopted from the 3D structure coordinates of African anti-E1E2 CHIKV Fab (PDB ID: 3J30) and CHIKV E1E2 (PDB ID: 3J2W). The scFv-BAD and E1E2-Ind models were superimposed to 3J30 and 3J2W in BIOVIA Discovery Studio Visualizer 2019 to obtain the scFv-BAD and E1E2-Ind complex.

CGMDs simulations

The previously described protocol using the pmemd.cuda programme as implemented in Amber16 was used to conduct the CGMD simulations. The pdb4amber programme in AmberTools16 was employed to handle the disulphide bonds of Cys residues and various charge states of His side chains in the pdb files (Machado et al., 2019). Subsequently, SIRAH Tools was used to map the all-atom structures to CG models. Each resulting CG model was solvated in an octahedral box using WT4 water models, where the minimum distance of the box from the solute was 2.0 nm. The radii of the solvent-solute interaction were scaled by 0.7 to improve the initial hydration of the CG model. WT4 water models were randomly replaced by CG models of Na⁺ and Cl⁻ ions to yield a 0.15 M salt concentration. A 1.2 nm cut-off was used for nonbond interactions, and particle mesh Ewald was used for long-range interactions. The Langevin thermostat and Berendsen barostat were implemented for all simulations. In the initial step of the CGMD simulations, solvent and side chains were relaxed by 5,000-step energy minimization, while the backbone beads were restrained by 1,000 kJ mol⁻¹ nm⁻². In the next stage, 5,000 ns energy minimization was performed without restraint; then, while a 1,000 kJ mol⁻¹ nm⁻² restraint was applied to the CG protein model, the system was equilibrated in NVT conditions at 300 K for 5 ns. The system was further equilibrated for 25 ns of NVT conditions at 300 K, where a 100 kJ mol⁻¹ nm⁻² restraint was imposed on the backbone beads. The production stage was performed in the NPT condition at 1 bar and 300 K, resulting in 2 μs CGMD trajectories. The snapshot frames in CGMD trajectories were written every 1 ns and analyzed using SIRAH Tools.

All-atom MD simulations

All-atom MD simulations were performed with the graphical processing unit-based pmemd programme of Amber16, and the inputs were prepared with the LEaP programme. The cysteine residues that form disulphide bonds were renamed from CYS to CYX, and the neutral histidine residues protonated in the delta and epsilon sections were renamed from HIS to HID and HIE. Neutralizing ions, as well as the model water molecule TIP3P, were added to the system with a minimum radius of 10 Å from the protein surface. It was then minimized, followed by heating to 27°C, and equilibrated for 1,000 ps with a cut-off value of 9 Å. The structure minimization was performed through the steepest descent and conjugate gradient methods. Subsequently, it was gradually heated until it reached temperature stability of 27°C (323°K) for 60 ps before the system density and pressure were equilibrated for 1,000 ps. The restraint was released gradually during equilibrium; then, the production stage proceeded for 10 ns. The calculation for energy interaction was performed using the MMGBSA.py programme, whereas other analyses such as hydrogen bond (H-bond) were performed using cpptraj in AmberTools16.

Construction of recombinant scFv in E. coli Origami B (DE3)

The scFv-BAD amino acid sequence was reverse-translated into a nucleotide sequence using Emboss on Expasy; then, the scFv-BAD codon was optimized based on the E. coli codon preferences using the GCUA software (http://gcua.schoedl.de/), and the guanosine-cytidine (GC) percentage was calculated by an online tool (https://jamiemcgowan.ie/bioinf/gc.html). scFv-BAD was designed with the addition of NdeI and BamH1 restriction sites for insertion into the pET21b (+) vector and synthesized by GenScript (Hong Kong). The pET21b-[scFv-BAD] plasmid was used to transform E. coli Origami B (DE3) by the heat shock method (Froger and Hall, 2007). The positive transformants were grown on a liquid Luria Bertani medium containing 100 μg/ml ampicillin and 40 μg/ml kanamycin and incubated overnight at 37°C. Subsequently, 50 μl of the culture was transferred to a flask containing 450 ml of fresh super broth medium, 100 μg/ml ampicillin, 0.05% glucose (Nacalai, Japan), and 4 M Biotin (Nacalai, Japan) and incubated at 37°Cand 200 rpm until the OD₆₀₀ reached ~0.5. Then, 1 mM isopropyl-L-D-thiogalactoside was added to induce scFv-BAD expression and incubated overnight at 20°C before the cells were collected by centrifugation (8,000 rpm, 20 minutes, 4°C). Cell pellets were resuspended in binding buffer (20 mM phosphate buffer pH 7.4 and 0.5 M NaCl) and sonicated intermittently (40 W, 20 kHz) for 2 minutes on and 2 minutes off, five times. The supernatant containing dissolved protein was separated, while the pellets containing insoluble protein were resuspended in binding buffer and characterized by SDS-PAGE.

Purification of biotinylated scFv-BAD

The soluble protein was purified using the TALON metal affinity resin, washed in varying concentrations of imidazole (100, 250, and 500 mM), and characterized by SDS-PAGE. All supernatants containing biotinylated scFv-BAD were collected and concentrated using ultra cut-off 3K Amicon (Millipore, USA). The protein concentration was determined using the A280 nanodrop (Thermo Scientific, USA) before protein purification using the AKTA explorer (GE Healthcare, USA) via anion exchange and reevaluated by SDS-PAGE.

SDS-PAGE and western blotting

Initially, SDS-PAGE was performed following Laemmli (1970) at 300 V for 95 minutes; then, western blot analysis was conducted according to Kim (2017). An equal concentration of protein sample was separated by 12% SDS-PAGE gel, and then the protein was transferred to a polyvinylidene fluoride membrane (Millipore, USA). The membrane was soaked overnight in blocking buffer (super blocking in tris buffer saline) (Thermo Scientific, USA) at 4°C and then washed three times with TBST (Sigma Aldrich, USA) for 5 minutes. Subsequently, the streptavidin horse radish peroxidase (-HRP) antibody (1:500,000; Sigma Aldrich) was added and then washed six times using TBST for 5 minutes, before incubation at room temperature for 5 minutes with the detection reagent (Sigma Aldrich). The membrane was analyzed with ChemiDoc XRS+ Imager chemiluminescence (BioRad, USA) in a dark room.

Biotinylated scFv ELISA

The sensitivity assay of biotinylated scFv against NS1 DENV and bovine serum albumin (BSA) was analyzed by the ELISA method. All procedures were conducted at room temperature. The E2 CHIKV protein (ProSpec, Japan) was immobilized on a 96-well microtiter plate, rinsed once with phosphate buffer saline-Tween (PBST),, soaked in a blocking buffer (Abcam, USA), and shaken for 1 hour. The blocking buffer was removed by washing three times with PBST. Subsequently, 300 μl of biotinylated scFv, NS1, and BSA in varying concentrations ranging from 0.2 to 5.3 μM was added to each well and incubated for 1 hour before washing three times with PBST. Next, the streptavidin-HRP conjugate was added (Sigma Aldrich) and incubated for an hour, followed by washing four times with PBST. The o-phenylenediamine dihydrochloride substrate (Sigma Aldrich) was kept in a dark room for 5–10 minutes before 100 μl was added to the wells, and the enzyme activity was stopped by the addition of 100 μl of 2 N H₂SO₄. The absorbance was measured using a spectrophotometer (Benchmark Plus, BioRad) at 490 nm.

The attachment of biotinylated scFv to nitrocellulose membranes via the streptavidin-biotin system

Briefly, 0.5 μl streptavidin (1 mg/ml) was dripped onto a nitrocellulose membrane in the test line position and incubated overnight before 0.5 μl of biotinylated scFv (0.8 mg/ml) was dripped onto the test line and incubated overnight. The membrane was placed in a 96-well plate containing 100 μl PBST pH 7.4 [Tween-20 0.05% in phosphate buffer saline (PBS)] and incubated for 20 minutes, then immersed in a staining solution containing 0.025% Coomassie brilliant blue, destained, and dried. The appearance of the test line indicates the presence of biotinylated scFv bound to streptavidin, with streptavidin or biotinylated scFv only used as a negative control.

Conjugation of ScFv-link C with gold nanoparticles

The scFv-Link C used in this study belongs to the Research Centre of Molecular Biology and Bioinformatics, Universitas Padjadjaran. AuNPs were synthesized based on the thermal citrate reduction method with modifications (Kimling et al., 2006). Briefly, 10 ml of hydrogen chloroaurate (HAuCl₄) was heated with constant stirring until the temperature reached 80°C–90°C; then, 1% sodium citrate was added rapidly. The colour of the solution changed from clear, dark purple to dark red, indicating the formation of the AuNP colloid. The AuNP colloid was then cooled at room temperature, and the pH was adjusted to 7.4. The gold nanoparticles were characterized by their absorption using a visible light spectrophotometer (Beckman Coulter, USA).

Fifty microlitre of scFv-Link C (100 μg/ml) was added to 450 μl of AuNP and incubated for 60 minutes before centrifugation to separate the pellets and supernatant. The pellets which contained AuNP-scFv-Link C were resuspended in a wash buffer (1% BSA in Tris pH 7.4), and the AuNP-scFv-Link C conjugate was resuspended in a conjugate diluent before analysis using a visible light spectrophotometer (Keshvari et al., 2016).

The immunochromatographic assay

The nitrocellulose membrane was cut into 2.5 × 0.4 cm sections; then, the control line was marked at the top and the test line at the bottom. Briefly, 0.5 μl (1 mg/ml) of the L protein was dripped on the control line, with 0.5 μl of streptavidin (1 mg/ml) dripped on the test line and incubated overnight. Subsequently, 0.5 μl of biotinylated scFv (0.8 mg/ml) was dripped on the test line and incubated overnight. A conjugate pad (Whatman standard 17; 0.7 × 0.4 cm) was prepared and treated with 10 μl of AuNP-scFv-Link C and dried at room temperature overnight. The sample pad (2 × 0.4 cm) was treated with 0.5 μl (0.9 mg/ml) of E2 CHIKV and then incubated in a desiccator at room temperature overnight. The wick for the absorption pad (1.8 × 0.4 cm) was prepared, and each membrane was fixed to a 6 cm backing card with 1–2 mm gap overlaps in each layer (Fig. 1). The assay was initiated by adding 100 μl of 10 mM PBS buffer in pH 7.4 containing 0.05% Tween-20 (v/v), and then this was incubated until the signal was observed. The specificity of the developed immunochromatographic assay was validated by checking the negative sample, which was NS1 DENV (Abcam, UK) (Sebastian and William, 2016).

In silico design of an E1E2-Ind model

The electron cryomicroscopy (cryo-EM) structure of African E1E2-CHIKV (PDB ID: 3J2W, hereafter referred to as E1E2-African) had a high sequence homology of 94% to E1E2-Ind (GenBank AHA87256.1); therefore, it was considered a suitable template for E1E2-Ind modelling. Homology modelling of E1E2-Ind was performed using 3J2W as the template, with the best model having 92.4% of its amino acid residues in the most favoured regions in the Ramachandran plot (Fig. S1), 85.4% for E2-Ind (Fig. S2). Since an acceptable model should possess more than 90% residues in the most favoured regions of the Ramachandran plot, the E2-Ind model was subjected to loop refinements using Modeller9.18, increasing the percentage of residues in the most favoured regions to 90.4%. Iterative loop refinements were also applied to the E1-Ind model to improve the number of residues in the generously allowed regions (Fig. S1).

The sequence alignment of E1E2-African and E1E2-Ind (Fig. 2) identified some missense mutations but did not provide information on which mutations affect the binding of specific antibodies. Fortunately, the 3J2W comes with the cryo-EM structure of a neutralizing antibody Fab CHK152 (3J30), so a structural alignment was performed of E1E2-African (3J2W) and the E1E2-Ind model. The Fab has a high neutralizing activity against CHIKV, with an EC₅₀ of 0.013 μg/ml (Sun et al., 2013), and thus can be used to develop an immunochromatographic assay for the detection of CHIKV.

The structural alignment of the E1E2-Ind model and 3J2W identified a mutation of G694S at the epitope of E1E2-Ind binding to neutralizing antibody Fab CHK152 (Fig. 2). The mutation may affect the binding of Fab CHK152 and cause a lower sensitivity in the immunochromatographic assay; thus, the binding of Fab-CHK152-derived scFv (E1E2-scFv) to E1E2-Ind should be evaluated through an in silico approach before conducting wet-lab experiments.

In silico design of the E1E2-scFv complex

The E1E2-scFv complex was designed based on the Fab CHK152 (PDB ID: 3J30) structure. Variable domains in 3J30 were identified using the NCBI-conserved domain website. The VL domain consists of 109 amino acid residues (IVLTQSPASL AVSLGQRATI SCRASESVDS YGNSFMNWYQ QKPGQPPKLL IYRASNLESG IPARFSGSGS RTDFTLTINP VEADDVATYY CQQSNEDPFT FGSGTKLEI), whereas the VH contains 113 amino acid residues (LQQSGPELVK PGVSVKISCK ASGYSFTSFY IYWVKQRPGQ GLEWIGWIFP GSTNTKYNEK FKGKATLTAD TSSSTASMQL SSLTSEDSAV YFCARVDGYA MDYWGQGTSV TVS). Then, the C-terminus of VL was joined to the N-terminus of VH through an oligopeptide linker (GGGGS)₄, resulting in E1E2-scFv. The streptavidin-biotin complex was used to attach E1E2-scFv to the nitrocellulose membrane, requiring the biotinylating of E1E2-scFv at the exposed lysine residues (Fig. S4), which are assisted by BirA. Biotinylating may lead to the blocking of the scFv paratope; however, to reduce the potency of biotinylating on the scFv paratope, we appended the BAD at the C-terminus, yielding scFv-BAD. Lysine residues in the coil structure of BAD (Fig. S4) are highly specific to biotinylating by BirA (Fairhead and Howarth, 2015). The use of BAD and BirA yielded site-specific protein biotinylating, including antibody-avidin conjugation (Asai et al., 2005; Chattopadhaya et al., 2006).

The quality of scFv-BAD was assessed using the Ramachandran plot. The initial model possessed less than the desired 90% of residues in the most favoured regions (Fig. S4) and therefore was subjected to an iterative loop refinement step to increase the number of residues to 90.8% (Fig. S3).

Figure 1. Schema of paper based immunochromatography for CHIKV detection. [Click here to view]

MD simulations

The E1E2-African crystal structure (3J2W) with Fab CHK152 (3J30) binds to the epitope at the B and arch2 domains of E2 (Fig. 2). Since only one amino acid mutation occurs on the epitope, these structures were adopted for the initial docking position of E1E2-Ind to scFv-BAD by applying a superimposition step. Subsequently, the complex of E1E2-Ind and scFv-BAD was subjected to CGMD simulation to yield a 2 μs trajectory as CGMD simulation can enhance the sampling of the protein conformational space (Machado et al., 2019), thus allowing observation of the dynamic behaviour of E1E2-Ind docked on scFv-BAD. Additionally, the simulation may give information regarding the effect of the linker and BAD introduction. A CGMD simulation of the complex of E1E2-African and scFv-BAD was also performed for comparison.

The analyses of the root mean square fluctuation (RMSF) plots (Fig. 3) revealed the dynamic behaviour of E1E2-Ind docked on scFv-BAD and E1E2-African docked on scFv-BAD. E1-Ind shows greater conformational changes than E1-African within residues 180 and 215 in domain II (Fig. 3A). Nevertheless, the E1-African also exhibits higher fluctuations than the E1-Ind spanning from residues 50–130 in domain II and FL. For E2, the Indonesian predominantly experiences higher fluctuations than the African (Fig. 3A). As shown in Figure 3B and C, E1E2-Ind displays a larger tilt than E1E2-African. Similarly, higher fluctuations are primarily observed on the scFv-BAD interacting with E1E2-Ind more than the E1E2-African (Fig. 3A). These higher fluctuations occur on the light chain and the linker of the scFv-BAD. Meanwhile, in some parts of the heavy chain, scFv-BAD binding to E1E2-African shows higher fluctuations than the binding to E1E2-Ind (Fig. 3A), which are also portrayed on their structure visualizations (Fig. 3B and C).

In the next step, we refined both complexes of E1E2-African and E1E2-Ind binding to scFv-BAD from the last frame of each CGMD simulation through all-atom MD simulations. The intermolecular interactions were analyzed in both complexes showing that the noncovalent bonds are contributed by residues from the E2 chain and the complementarity determining region of scFv-BAD. Also, the tilt of E1E2-Ind on scFv-BAD intensifies the intermolecular interactions, as summarized in Table S1. Salt bridge interaction increased by threefold, twofold for H-bonds. New interactions between E1E2-Ind and scFv-BAD also appear after CGMD and all-atom MD simulations which involve π-cation, π-anion, π–π, and π-alkyl (Table S1). Similarly, the intermolecular interaction profile of E1E2-African also improves after CGMD and all-atom MD simulations. Detailed analysis of H-bonds (Table S2) from 10 ns all-atom MD-trajectory revealed that scFv-BAD forms more H-bonds with E1E2-Ind than E1E2-African. Nonetheless, H-bonds formed by scFv-BAD and E1E2-African are more conserved than those of scFv-BAD and E1E2-Ind (Table S3). All interactions formed by scFv-BAD with E1E2-Ind and E1E2-African are through the E2 chain.

Figure 2. Sequence and structural alignments of E1E2-African (3J2W), binding to Fab CHK152 (3J30), and E1E2-Ind. Purple and orange denote E1E2-African and E1E2-Ind, whereas magenta is Fab CHK152. [Click here to view]

The molecular mechanics generalized-born surface area (MMGBSA) binding energy (ΔG_MMGBSA) from a single 10 ns trajectory from each complex was calculated (Table S4), showing that the E1E2-Ind complex exhibits slightly stronger binding to scFv-BAD (ΔG_MMGBSA = −46.9123 kcal/mol) than E1E2-African (−42.8413 kcal/mol). Nonbonded interaction energy terms, including van der Waals (E_vdW) and electrostatic (E_{electrostatic}), are also stronger for E1E2-Ind than E1E2-African (Table S4).

Optimization of scFv-BAD codon and construction of the expression vector

The scFv-BAD codon was optimized using the E. coli K12 codon table available in GCUA, resulting in 90% of optimum codons and 53.4% GC content. Subsequently, scFv-BAD was constructed in the pET21b(+) vector with the order: VL (109 aa) - (GGGGS)₄ (20 aa) - VH (113 aa) - BAD (45 aa) and 6× His-Tag at the C-terminal (Fig. 4).

Expression of scFv-BAD protein

The characterization of the expressed protein using SDS-PAGE showed that scFv-BAD with a molecular weight of 30 kDa was expressed in E. coli Origami DE3, but scFv-BAD was more expressed in the insoluble fraction (Fig. 5). Overexpression of exogenous proteins in E. coli may lead to protein aggregation that causes the formation of insoluble inclusion bodies (Gupta et al., 2020). For the convenience of the purification process, only protein in the soluble fraction was collected for further purification.

Figure 3. (A) RMSF graphs of the complexes of E1E2-Ind and E1E2-African binding to scFv-BAD. (B) The first and the last CGMD trajectory snapshots of the complex of E1E2-Ind (top panel) and scFv-BAD (lower panel). (C) The first and the last CGMD trajectory snapshots of the complex of E1E2-African (top panel) and scFv-BAD (lower panel). These analyses were obtained from 2 μs CGMD simulation trajectories. Colour bars in (A) represent domains of E1 (top panel), E2 (middle panel), and scFv (lower panel). [Click here to view]

Figure 4. Construction of scFv-BAD expression vector. (A) Map of pET 21b(+)-scFv-BAD. (B) Map of scFv-BAD and the amino acid sequences. [Click here to view]

Figure 5. Characterization of scFv-BAD fusion with SDS-PAGE. (1) Soluble fraction after induction and sonication. (2) Insoluble fraction after induction and sonication. (3) Insoluble fraction after induction without sonication. (4) Insoluble fraction before induction. (5) Protein marker. [Click here to view]

Purification and characterization of scFv-BAD

Firstly, scFv-BAD was purified based on the ability of His-Tag to interact with the cobalt column; however, there were still many other protein bands; therefore, the purification was continued using the anion exchange method to yield a single band, as shown in Figure 6. Characterization of biotinylated scFv by western blotting using streptavidin-HRP showed that biotinylated scFv with a molecular weight of ~30 kDa was expressed (Fig. 6).

Sensitivity of biotinylated scFv

The interaction of biotinylated scFv with 10 μM of the commercial E2 CHIKV, NS1 DENV, and BSA was assessed by ELISA. The results in Figure 7 and Table S5 indicate that scFv-BAD was more sensitive to E2 CHIKV when the scFv concentration was above 0.6 μM, indicating that the sensitivity of biotinylated scFv to E2 CHIKV is not too high compared to NS1 and BSA.

Specificity of the immunochromatographic assay

The specificity of the interaction of scFv to E2 CHIKV using the NS1 DENV antigen for comparison (Fig. 8) revealed sensitivity to E2 CHIKV until 0.2 μl of 0.9 mg/ml. However, if the concentration of NS1 DENV was higher (0.6 μl of 1 mg/ml), it gave a positive result; thus, false positive results will be obtained if the concentration of NS1 is sufficiently high. Given that the NS1 protein is structurally distinct from CHIKV E2, this finding requires additional investigation and is a limitation of the qualitative immunochromatographic approach.

Figure 6. Purification and characterization of scFv-BAD; (A) SDS-PAGE result of purification scFv-BAD; (1) with His-tag affinity chromatography; (2) with anion exchange chromatography; (3) protein molecular weight marker. (B) Characterization of scFv-BAD with western blot; (1) the result of His-tag affinity chromatography; (2) the result of anion exchange chromatography; (3) protein molecular weight marker. [Click here to view]