Multivalent display of membrane proteins using HIV-1 Gag VLPs

Membrane proteins are increasingly becoming attractive targets for biotherapeutics, but their production is challenging due to extraction with detergents often resulting in low yields and purity, and potential changes in conformation. In this study, we evaluated HIV-1 Gag Virus Like Particles (VLPs) for the display of heterologous membrane proteins in their native conformation and lipid environment to serve as reagents for antibody discovery. We have developed a scalable workflow to purify and characterize the VLPs pseudotyped with recombinant membrane proteins. Furthermore, we engineered a dual snorkel–sortase tag for enzymatic, site specific biotin conjugation to enable binding studies. The particle quality was assessed by differential light scattering, analytical HPLC, nanoparticle tracking, and cryo-EM. The antigen presence and display was determined by BLI, Dot -blot, Western blot, and LC-MS/MS. Surface plasmon resonance (SPR) was used to measure binding kinetics with an internal biotherapeutic. Our workflow yields pure, stable, and monodisperse VLPs that present recombinant membrane proteins in their native conformation and lipid environment. Site specific biotinylation via the dual snorkel\x{2013}sortase tag did not alter particle size, purity, or antigen expression/ conformation. Biotinylated, membrane protein pseudotyped VLPs were compatible with SPR and enabled kinetic measurement for the tested biotherapeutic. VLP display of membrane proteins combined with precise surface biotinylation provides a robust platform for displaying membrane antigens in their native conformation and supports antibody discovery against challenging membrane targets.

Membrane proteins constitute more than 30% of FDA-approved drug targets (Santos et al., 2017). However, many membrane proteins that are known to be differentially expressed in pathological states or target cell types remain unexplored (Santos et al., 2017). Membrane proteins are challenging targets because their extraction is cumbersome, time-consuming, requires optimization for every target, results in low yields, and most importantly, can cause denaturation due to the use of detergents (Nakatani et al., 2020; Chipot et al., 2018; Zhou and Cross, 2013). Other challenges include lack of retention of the native lipid environment, small exposed epitopes that lack sufficient diversity, and a lack of multivalent antigen display for immunization (Santos et al., 2017; Jo and Jung, 2016). Enveloped VLPs have emerged as an attractive approach for the multivalent display of heterologous membrane proteins in their most native conformation and lipid environment (Tarres-Freixas et al., 2023; Sabapathy et al., 2019; Vidigal et al., 2018; Kurg et al., 2016; Hoffman et al., 2000; Tucker et al., 2018; Willis et al., 2008). Moreover, the repetitive and particle-like architecture of VLPs makes them naturally immunogenic and readily phagocytosed by antigen-presenting cells (Franco et al., 2011; Deml et al., 2005).

Enveloped VLPs generated by the overexpression of human immunodeficiency virus-1 precursor (HIV-1) Pr55Gag (Gag) has attracted significant interest for the display of membrane proteins, particularly due to their ease of generation from HEK293T (human embryonic kidney 293T antigen) cells (Cervera et al., 2015a; Cervera et al., 2017; Cervera et al., 2013; Cervera et al., 2015b; Lavado-Garcia et al., 2021; Steppert et al., 2016; Gonzalez-Dominguez et al., 2021). The overexpression of Gag protein is sufficient to drive all the steps necessary for VLP particle formation. During the assembly of VLPs, Gag proteins multimerize and insert themselves under the inner leaflet of the plasma membrane through their N-terminal modification of myristic acid (Alfadhli et al., 2009; Gitti et al., 1996). Subsequently, Gag proteins engage the cellular machinery responsible for budding and subsequent secretion, releasing VLPs extracellularly into the media (Freed, 2015; Sundquist and Krausslich, 2012; Briggs et al., 2009; Schur et al., 2015; Wright et al., 2007; Gamble et al., 1997; Summer et al., 1992). Importantly, in the absence of viral proteases and viral nucleic acids, VLPs remain immature and non-infectious.

Previously, secreted VLPs have been purified from culture media (Cervera et al., 2015a; Cervera et al., 2017; Cervera et al., 2015b) by ultracentrifugation on a sucrose cushion (Cervera et al., 2013). While this method yields pure VLPs, its throughput and scalability are low. Recently, many researchers have used either anion exchange (AEX) alone or a subsequent polishing step via desalting on size-exclusion chromatography (SEC) columns (Tarres-Freixas et al., 2023; Gonzalez-Dominguez et al., 2021; Cervera et al., 2019; Gutierrez-Granados et al., 2016). However, analysis of the purity and monodispersity of VLP preparations was limited in these studies. Furthermore, a lack of consistent quality-control steps makes it difficult to compare the VLPs prepared by different methods (Tarres-Freixas et al., 2023; Cervera et al., 2015a; Cervera et al., 2017; Cervera et al., 2015b; Steppert et al., 2016; Gonzalez-Dominguez et al., 2021; Cervera et al., 2019; Gutierrez-Granados et al., 2016). Therefore, there is a need to standardize and develop scalable workflows that result in the production of pure and monodisperse VLPs with high heterologous target membrane protein expression that can be adapted in biologics discovery workflow, including screening and biophysical characterization of antibodies.

This study describes innovations that would expand the utility of the VLP platform for preclinical drug discovery studies. This includes process improvements leading to >99% pure material, a comprehensive quality check on monodispersity, the detection and quantification of heterologous membrane proteins, a comparison of three different cell lines for VLP production and antigen display, and a method for the site-specific conjugation of biotin on VLPs for its subsequent use in the surface plasmon resonance (SPR) assay to determine the binding properties of an internal biotherapeutic. We have used the VLP platform for the native display of several membrane proteins of varying size and number of transmembrane domains, thus showcasing its versatility. Overall, these improvements should lead to a more effective utilization of VLPs for the display of a wide array of membrane proteins during the preclinical discovery of biologics, including immunization, screening for potential binders, and biophysical characterization.

Adaptation to serum-free media of HEK293T

The adherent HEK293T cells (ATCC, CRL-3216) grown in DMEM (Gibco, 10566-016) + 10% FBS were subjected to gradually reducing concentrations of FBS, reducing 5% to 2.5% to 1.25% in DMEM, which were finally transitioned to FreeStyle 293 expression media (Thermo Fisher Scientific, catalog #12338018). The cells consistently grew to 98% viability and a cell density of 4e6/ml with a doubling time of less than 24 h.

Transfection of HEK293T with PEI

Cells were split to a cell density of 1e6/mL 24 h before transfection. On the day of transfection, plasmids were diluted at 1 µgm/mL in 1/20th of the total culture media (25 mL for 500 mL culture volume). PEI (Polysciences #23966-2) solution at 1 mg/mL was added at a three-fold molar excess of DNA in 1/20th total volume. After waiting 5 min, DNA and PEI solutions were mixed and allowed to form complexes for 20 min. Following this, DNA-PEI solution was added dropwise into the flasks with cells growing in FreeStyle 293 expression media. The DNA encoding Gag protein was obtained from OriGene (catalog #VC101718).

Transfection of DNA into ExpiCHO and CHO-S

Transfections were conducted according to the manufacturer’s guidelines. In brief, 24 h before transfection, both cell lines were split to a density of 1e6/mL at 100 ml. For ExpiCHO, each of the plasmid DNAs was diluted to 1 μg/mL in 2 ml (1/50th of culture media) of OptiPRO. In another vial, 320 µL of ExpiFectamine (catalog #A29129, Thermo Fisher Scientific) was added to 2 ml of OptiPRO. After 5′ at RT, the DNA-ExpiFectamine complexes were formed for 10′ at RT. Subsequently, the complexes were added to the cells growing in serum-free ExpiCHO media. For CHO-S, on the day of transfection, each of the plasmid DNAs were diluted to 1 μg/mL in 2 ml (1/50th of culture media) of OptiPRO. In another vial, 300 µL of FreeStyle Max (catalog #16447100, Thermo Fisher Scientific) was added to 2 ml of OptiPRO. After waiting 5 min (5′), the contents were mixed and allowed to form complexes for 10′ at RT. Subsequently, the complexes were added dropwise to the cells.

Clarification of lysate, VLP precipitation, and Benzonase digestion

The cells were spun down at 800 g for 10′ RT 72 h post transfection. The clarified medium was mixed with 4× PEG/NaCl solution (40% (W/V) PEG-8000 and 1.2 M NaCl) for 18 h on the bottle roller. The flasks were then spun down at 1,600 g for 1 h at RT. The precipitate was resuspended in 10 mM phosphate buffer at pH = 7.0 and subjected to digestion with 5U of Benzonase (Millipore, catalog #70664-3) at 37 °C for 1 h.

Anion-exchange chromatography (AEX)

A 10 mL column (Cytiva, Tricorn 10/150 Column) containing high capacity Fractogel TMAE resin (MilliporeSigma, catalog #110316) was assembled on AKTA Pure (GE HealthCare) and equilibrated with 10 mM phosphate buffer pH=7.0 according to the manufacturer’s instructions. The following protocol was set up on Unicorn for VLP purification on AKTA: equilibration of column with 5 CVs of phosphate buffer, the resuspended VLPs loaded onto the column in 25 mL of buffer, followed by 5 CVs of washes in phosphate buffer to remove unbound substances. The flow-through was collected through the outlet valve to determine any breakthrough of VLPs. The sample was eluted with a linear salt gradient increasing from 0 M NaCl to 1 M NaCl (in 10 mM phosphate buffer) over 15 CVs, followed by a step gradient to 5 CVs of 2 M NaCl in 10 mM phosphate buffer. The eluates were fractioned into 24-deep well plates. The column was regenerated by 2 CVs of 1 M NaOH, followed by re-equilibration with 10 mM phosphate buffer. UV absorbance, conductivity, and pH were measured using in-line sensors on the AKTA Pure system (GE HealthCare). The fractions were pooled and concentrated to approximately 1–2 mL using Amicon Ultra-15 Centrifugal units for 50,000 MW cutoff (MilliporeSigma, UFC905024).

Size-exclusion chromatography

The pooled and concentrated (to ∼2 mL) AEX eluate was loaded onto the HiLoad 16/600 Superose preparative size-exclusion column (Cytiva, 76308-980). The column was equilibrated with 2 CVs of Dulbecco’s phosphate buffer saline (DPBS), followed by sample loading and isocratic elution for 1.5 CVs onto the 96 deep well plates. The elution was started after 0.2 CV. The column was re-equilibrated with 2 CVs of DBPS and regenerated with 1 M NaOH. The entire purification was done on an AKTA Pure system (Cytiva). UV absorbance, conductivity, and pH were measured using in-line sensors. The peaks corresponding to VLPs were concentrated to approximately 1 mL using Amicon Ultra-15 Centrifugal units for 50,000 MW cutoff (MilliporeSigma, UFC905024).

Data processing for chromatograms

For the AKTA purifications, UV_Abs 280 and conductivity data were exported from Unicorn software and imported into GraphPad Prism 8.0., where the final images were generated.

Protein estimation in VLPs

We denatured 25 µL of VLPs in 10 mM Tris, 2% SDS (sodium dodecyl sulfate), 1× reducing agent in a total volume of 50 µ at 95 °C for 10 min. The total protein content was estimated by detergent-compatible Bradford reagent (Thermo Fisher Scientific catalog #23246) using BSA as a standard. The colorimetric reactions were measured on a microplate reader (Molecular Devices, SpectraMax M5) at 595 nm.

Dynamic light scattering (DLS)

DLS on VLPs was measured on a Wyatt-455-WPR3 plate reader. We aliquoted 30 µL of 500-fold diluted VLP sample onto 384-well DLS plates. Each sample was isothermally imaged ten times for 5 s each at room temperature to calculate the hydrodynamic radii of particles in solution. Each sample was run in duplicate. The 20 measurements per sample were averaged to estimate the final particle size and polydispersity.

Nanoparticle tracking analysis (NTA)

NTA was performed on NanoSight NS300 machine using NTA v3.44 software. The sample was diluted 5000-fold in DPBS and loaded for imaging. A Blue488 laser was used at ambient temperature, and three videos of 1 min each, 25 frames per second, were captured by the software. The software calculated the average concentration by tracking VLP particles in all videos obtained per sample.

Biolayer interferometry and data processing

The kinetics assay for AHC biosensors (Sartorius, catalog #18-5063) was conducted on Octet HTX from Forte Bio. The biosensors were hydrated for 15 min prior to the start of the experiment. The kinetics assay protocol was typically set up as follows: 60 s for equilibration, antibody loading onto the biosensors for 120 s followed by 60 s in buffer, association with VLPs for 600 s (unless otherwise reported), and 600 s for dissociation in buffer. VLPs (and other nanoparticles) generate a negative amplitude in BLI, which is also verified by Sartorius in a technical note (https://www.sartorius.com/download/552362/kinetic-analysis-of-antibody-binding-to-membrane-protein-on-captured-lipoparticles-application-note-en-sartorius-data.pdf). The association/dissociation kinetics data were flipped in the Sartorius data analysis software, and the graph was generated using Prism 8.0. Antibody loading is not shown in BLI graphs. For binding to Streptavidin biosensors (Sartorius, 18-5137), only the association and dissociation kinetics were measured, each for 600 s.

Sample preparation and Cryo-EM imaging

Cryo-EM samples were prepared without dilutions. 3 μL of each sample was applied to holey carbon grids (Quantifoil, 1.2/1.3), incubated briefly at 4 °C with 100% humidity and vitrified in liquid ethane using a Vitrobot IV (Thermo Fisher Scientific). The grids were then transferred to a Glacios microscope operating at 200 kV with a Falcon 4i imaging system, collecting images at a nominal magnification of ×105,000. A calibrated pixel size of 0.91 Å was used for processing.

SEC-HPLC

VLP samples were analyzed with the Superose 6 Increase 5/150-GL (Cytiva Life Sciences, catalog # 29091597) on an Agilent 1100 HPLC system. Typically, 10–20 µg of VLPs were injected into the column running at a flow rate of 0.35 mL/min, maximum pressure cutoff at 30 bar in 0.2 µm sterile-filtered 1× DBPS for 20 min per run. Prior to running the samples, the column was equilibrated with 1× DPBS for 20 CVs.

Generation of conformation-specific antibody for MemPr1

Nucleic acid-based immunization was carried out to identify conformational specific antibodies against MemPr1. Following immunization, the potential hits were identified by hybridoma and expressed as monoclonal antibodies by cloning variable genomic sequences in an IgG format. The purified monoclonal antibodies were then confirmed via binding on overexpressing and knockout cells. The antibody used in current study recognizes conformational specific epitope on native MemPr1 protein.

Western blotting

Western blotting was done using standard procedures. In brief, VLPs were denatured in LDS sample buffer (Invitrogen, NP0001) at 95 °C for 10′ and loaded onto the 4%–12% Bis-Tris NuPAGE gel (NP0336). The proteins were transferred onto methanol-activated 0.2 µm PVDF membrane (Bio-Rad, catalog #1620174) in NuPAGE transfer buffer (Life Technologies, NP0006). After overnight transfer at 25 V, the membrane was washed twice in dH 2 0, followed by blocking in 5% BSA and overnight incubation in primary antibody at 10 ugm/mL. After 3 × 5′ washes in TBS + 0.1% Triton X-100 (TBST, catalog #9997, Cell Signaling Technology), the membrane was incubated in anti-human HRP-conjugated secondary antibody (BD Pharmingen, catalog #555788, Lot #2329349). This was followed by 3 × 5′ washes in 1× TBST and signal development using chemiluminescent substrate (Thermo Fisher Scientific, catalog #34577) and imaged on Bio-Rad Gel Doc system. To detect Gag protein via Western blot, the following antibody was used: Invitrogen, catalog #MA1-7587, Lot #XJ3727441, anti His antibody: Qiagen, catalog #1014992, lot #175024284. The antibodies for detecting MemPr1–MemPr10 for Western blot analysis were commercially sourced.

Design of snorkel–sortase tag, biotinylation reaction, and estimation of biotin molecules

The snorkel–sortase dual-tag construct was derived from Brown et al. (2013). The sequence of the snorkel–sortase dual-tag construct is QKKPRYEIRWKVVVISAILALVVLTVISLIILIMLWGSGSGSSGSGSGSGASSGSSGSGSDYKDDDDKGSGSGSLPETG. The residues “QKKPRYEIRWK” are positively charged residues grafted from the cytoplasmic domain of PDGFR. As described in Brown et al. (2013), the residues “VVVISAILALVVLTVISLIILIMLW” belong to the transmembrane domain of mouse PDGFR, followed by a string of GS residues, flag tag DYKDDDDK, and a terminal recognition site for Sortase (LPETG).The biotinylation reaction was conducted as per manufacturer’s instructions (Active Motif, catalog #13105). In brief, 100 µg of purified VLPs were incubated with sortase enzyme, (Gly) 5 biotin, and buffer at 30 °C for 1 h at 1000 rpm. After the reaction, the enzyme activity was immediately terminated with the stop-solution (Active Motif, catalog #13105). The VLPs were then repurified in 1× DPBS in the void volume of a 5 mL HiScreen Capto Core column 700 to remove reaction intermediates (Cytiva, 17548115) on AKTA Pure (GE HealthCare). The running buffer was 1× DPBS, run at 1.2 mL/min, and the target VLPs were eluted in a clear peak. The biotin molecules per VLPs were then estimated by FluoReporter Biotin Quantitation kit (Invitrogen, catalog #F30751). VLPs were digested with the provided proteases overnight at 37 °C, and biotin molecules were estimated using manufacturer’s guidelines using biocytin standard curve, and fluorescence was measured on a PerkinElmer Envision Plate microplate reader. The biotin molecules per VLP were obtained by the following equation:

N= ((Conc. of biotin in the sample–conc. in unbiotinylated sample (nmoles/mL) dilution factor))/concentration of VLPs (particles/mL)6.023e23*1e-9.

Dot blot

We spot-dried 5–10 µg of VLP samples onto nitrocellulose membrane. Following this, the membrane was blocked in 5% BSA for 10 min, followed by incubation with specified antibodies (streptavidin-HRP, Thermo Fisher Scientific, catalog #N100) or internally generated antibody against MemPr1. For the latter, the blot was washed 3 × 5′ with 1× TBST, followed by incubation in HRP-conjugated secondary antibody for 2 h s (BD Pharmingen, catalog #555788, Lot #2329349), washes with 3 × 5′ 1× TBST, and developed using chemiluminescent substrate (Thermo Fisher Scientific, catalog #34577) and imaged on Bio-Rad Gel doc system.

Mass spectrometry on VLPs

The VLP samples were lysed using custom RIPA lysis buffer containing 1% SDS, followed by S-trap digestion with trypsin and analyzed on nano LC-MS/MS with a Waters M-Class LC system interfaced to Thermo Fisher Exploris 480 mass spectrometer. Ten percent of the digested sample containing peptides were loaded onto a trapping column containing XSelect CSH C18 resin (Waters, 3.5 µm particle size). The peptides were eluted over a 75 µm analytical column, with 2.4 µm particles, run at 350 nL/min. The column was heated with a column heater set to 55 °C. The mass spectrometer was operated in the data-dependent acquisition mode, with Orbitrap operating at a resolution of 60,000 and 15,000 for MS and MS/MS, respectively. The instrument was run with a fixed 3 s cycle time for the MS and MS/MS acquisition loop.

The PRM-MS analysis was conducted on a nano Waters M-Class HPLC system interfaced to Thermo Fisher Exploris 480 mass spectrometer. The lyophilized synthetic peptide standards were purchased from SciTide and solubilized in 5% acetonitrile containing 0.1% formic acid to produce 1 pmol/μL stock solutions. Internal peptide standards were added to the digested VLP preparation at 10 fmol/μL. Ten percent of each sample containing 100 fmoles of labeled peptide standard was loaded onto a trapping column, with the mass spectrometer operating in the PRM mode with a quadrupole isolation window of 1.6 Da and an Orbitrap operating at 60,000 and 15,000 FWHM for MS and MS/MS, respectively.

Data were searched using a local copy of Mascot (Matrix Science) with the following parameters: enzyme—Trypsin/P; database—SwissProt Human appended with custom sequence of Gag; fixed modification—carbamidomethyl; variable modifications—oxidation, acetyl, pyro-glu, deamidation; mass values—monoisotopic and peptide mass tolerance of 10 ppm; maximum missed cleavages = 2. Mascot.dat files were filtered with Scaffold for validation and created a non-redundant list per sample. Data were filtered at 1% protein and peptide FDR with at least two unique peptides per protein.

PRM data were analyzed using Skyline 4.2 (University of Washington).

The number of antigen molecules per VLP was calculated using the equation: (antigen molecules/mL)/(VLP molecules/mL). VLP molecules/mL were determined by NTA.

Surface plasmon resonance (SPR) assay

SPR experiments were performed using a Biacore T200 instrument. The experiments were run in 1× DPBS buffer with 0.02% Tween-20 at 25 °C using Cytiva SA sensor chips (Cytiva catalog #BR100531). VLPs were immobilized onto the surface of the Biacore sensor chip by injecting them over a streptavidin-covered sensor chip. VLPs were captured on flow cells 2, 3, and 4 at different RU levels (Figure 9B), while flow cell 1 was used as a reference and had mock VLPs captured on its surface. Antibody was then injected over the sensor chip surface in single-cycle kinetics mode at concentrations of 3.7, 11.1, 33.3, and 100 nM. The association was monitored for 4 min per concentration and dissociation for 15 min, at flow rate of 50 μL/min. The data collected were processed and fitted using Biacore Insight evaluation software. The 95% margin of error (95% MoE) was calculated using the equation: 95% MOE = (t * STDEV)/√n.

Flow cytometry

The transfection into Expi293 cells was done according to the manufacturer’s protocol at 1 μg/μL (Thermo Fisher Scientific, A14635). The cells were subjected to live/dead staining for 15′ at RT (Invitrogen, catalog #L34957). They were stained using standard protocol with primary or isotype antibody at 1 μg/mL for 30′ on ice. After a couple of washes (PBS + 0.1% BSA), the cells were incubated with secondary antibody or buffer (as indicated) for 30′ on ice at 1:500 dilution. After a couple of washes, the cells were subjected to flow cytometry, with the gates set on single and live cells.

Statistics

All one-way comparisons were evaluated for normality, and Students t-test was performed for normal distributions and Mann–Whitney for non-normal distribution. One-way ANOVA was performed for more than two groups. The statistical cutoff for significance was P ≤ 0.05.

Production, purification, and characterization of VLPs

HEK293T cells, adapted into suspension (Cervera et al., 2015a; Cervera et al., 2017; Gonzalez-Dominguez et al., 2021), were chosen to optimize the purification of HIV-1 Gag VLPs (hereafter referred as “VLPs”) (Figures 1A,B). The cells were transiently transfected with the Gag polyprotein (OriGene, VC101718) using polyethyleneimine (PEI). To determine an ideal time-point for harvesting VLPs, cell viabilities were measured for several days post Gag transfection. A sharp reduction in cell viability was observed beyond 72 h post transfection (Cervera et al., 2015a; Cervera et al., 2017; Gonzalez-Dominguez et al., 2021) and was therefore chosen for downstream purification (Figure 1C). The clarified medium containing VLPs was subjected to overnight incubation with polyethylene glycol (PEG)-8000/NaCl and subsequent precipitation by centrifugation. After resuspension in 1× phosphate buffer, the samples were subjected to Benzonase digestion to remove residual nucleic acids (Table 1).

The net negative charge of the membrane envelope (van Dijck et al., 1978) makes AEX chromatography a reasonable first step for purifying VLPs. A strong AEX resin (Steppert et al., 2016; Pereira Aguilar et al., 2019; Pereira Aguilar et al., 2020) with pore dimensions at least ten-fold larger than VLPs was selected to preserve the integrity of the particles during purification. Fractogel® resin, composed of trimethylammoniummethyl and featuring a d90 pore diameter of 63–77 µm (approximately 500-fold larger than the average VLP diameter of ∼140 nm) and a binding capacity of 100–120 mg/mL, was used in the experiments (Steppert et al., 2016; Pereira Aguilar et al., 2019; Pereira Aguilar et al., 2020). The VLPs were eluted by a linearly increasing NaCl gradient from 0% to 1 M NaCl for 15 CVs (1 CV=10 mL), followed by a step gradient to 2 M NaCl for 5 CVs. The VLPs eluted over a broad range of NaCl concentration until 800 mM NaCl, as indicated by the presence of HIV-1 Gag protein in these fractions (representative AEX plot in Figure 1D, SDS-PAGE in Figures 1E,F). Negligible levels of Gag were detected in the flow-through, indicating minimal breakthrough of VLPs during purification (Figure 1F). The AEX eluates from fractions 1–16 (Figure 1D) were concentrated by size-exclusion chromatography.

Following AEX, the VLPs were fractionated on a preparative 120 mL Superose 6 size-exclusion column. The fractionation range of 5–5,000 kD for Superose 6 resin allowed the separation of high-molecular-weight VLPs from other host-cell proteins and membrane fragments (representative plot in Figure 1G). The peak fraction A from SEC showed the maximum amount of Gag protein on SDS-PAGE (Figure 1H, lanes 3 and 4), while fractions B and C contained minimal amounts of Gag (Figure 1H, lanes 5 and 6). This suggests that peak fraction A, which eluted first during SEC, contained the VLPs—consistent with its high molecular weight.

The monodispersity and purity of VLPs was checked during different purification steps by dynamic light scattering (DLS), analytical HPLC, and nanoparticle tracking analysis (NTA). The polydispersity (PD) score obtained from DLS is an indication of the size distribution of particles, with a score of less than 15% being typically considered a homogeneously sized population for nanoparticles (Danaei et al., 2018; Palmieri et al., 2014). The %PD was multimodal in PEG precipitate, indicating the presence of multiple molecular species (Figure 2A). The % PD decreased to 31.6% in AEX eluates (Figure 2B), and further reduced to 10.06% in SEC-fraction A (Figure 2C); this suggests that our purification methods led to a progressive increase in the monodispersity of the sample (Figure 2D). SEC-fractions B and C contained particles with a high PD score of 38.74% and 45.79%, respectively, and a minimal presence of Gag protein (Figure 1H). The average radii of 69.23 ± 1.93 nm (average ± s.e.m.) and average %PD of 8.7 ± 0.61 was obtained for n = 20 purified preparations (Figures 2O,P, representative size-distribution graph, and table in Figure 2C). As with DLS, the analytical SEC showed a sequential increase in purity from PEG precipitate (Figure 2F) to 68% in AEX eluate (Figure 2G) to > 99% in SEC-fraction A (Figure 2H). The lower elution time is consistent with the larger size of VLPs. SEC-fractions B and C showed multiple peaks at higher elution times in alignment with higher polydispersity scores in DLS (compare Figures 2D,I,J). Next, using NTA, we estimated the concentration of nanoparticles and overall homogeneity in the size of the particles. Similar to DLS and HPLC, NTA revealed an overall increase in the purity and a narrower particle size distribution after two rounds of purification (aggregates indicated by red arrows in Figure 2K are cleared after two rounds of purifications in Figures 2L,M). The average concentration from n=8 purified VLP preparations was found to be 7.11 × 10 11 /mL ± 9.1 × 10 10 particles/mL from 500 mL culture of HEK293T cells (representative plot in Figure 2M, compiled data in Figure 2Q). Altogether, DLS, HPLC, and NTA revealed that our purification workflow results in a sequential increase in purity and monodispersity of the VLP nanoparticles.

Next, we estimated the total protein content of the VLPs by BCA assay. The standard BCA assay on intact VLPs did not detect any soluble proteins in the preparations (data not shown). After denaturation in 1% SDS at 95 °C for 10 min, we estimated the total protein content by detergent-compatible Bradford reagent. As expected, the total protein content reduced as the preparations went through sequential rounds of purifications, from 16 mg in PEG precipitate to 2.52 mgs in the final SEC-fraction A containing VLPs (Figure 2N). An average of 1.12 ± 0.13 mg (n=8) of total protein, which included Gag, was recovered from 500 mL of HEK293T cells (Figure 2R). As another way of assessing the purity of our preparations, we measured DNA content in our VLP preparations (Steppert et al., 2016; Summer et al., 1992). Using our workflow, the DNA amount was reduced from 220 ng/mL to 17.18 ng/mL in the final SEC eluate (Table 1), demonstrating improvement over other published methods (Steppert et al., 2016; Summer et al., 1992). Altogether, we have developed a robust end-to-end purification protocol and quality check methods for VLPs.

The purified VLPs from SEC-fraction A were subjected to electron microscopy to assess the overall structure of the particles (Figures 3A,B). The outer lipid envelope (arrows, Figure 3C) derived from the host-cell plasma membrane enclosing the inner viral capsid (arrowhead, Figure 3C) was clearly observed. An average radius of 72.9 ± 1.46 nm was determined from measuring more than 100 images of VLPs (Figure 3D), consistent with the measurements from DLS (Figure 2O). Altogether, using several orthogonal and high-resolution methods, we showed that our VLP preparations are pure and monodisperse and are suitable for several assays such as biophysical characterization and screening binders.

Multi-spanning membrane proteins are displayed in a native conformation by VLPs

Next, we sought to determine whether heterologous membrane-bound protein could be displayed on the outer lipid envelope of HEK293 Gag VLPs. Since the outer lipid envelope of the VLPs is derived from the host plasma cell membrane, we hypothesized that the target membrane proteins would be present in their native conformation and lipid environment. Additionally, we tested whether heterologous overexpression of membrane protein would affect the purity, homogeneity, or yields of the VLPs. We selected a clinically validated cell adhesion protein (Salvador et al., 2016; Takasawa et al., 2023; Lin et al., 2019) that is localized to the tight junctions of epithelial cells, with four transmembrane domains and a molecular weight of ∼25 kD (henceforth referred to as “MemPr1”). The plasmid encoding Gag and MemPr1 were co-transfected into HEK293T cells (Figure 4A), and VLP purification was performed as described above (Figure 1) using PEG precipitation, AEX, and SEC. The elution time of VLPs, indicated by AEX and SEC, was nearly identical to mock VLPs (Figures 4B,C vs. Figures 1D,G). Monodispersity and purity were analyzed by DLS, analytical SEC, and NTA. The VLP preparations showed an average radius of 61.97 ± 6.25 nm and %PD < 15% (Figures 4D–F, n = 4 samples, representative data in Figure 4D) and > 99% purity in SEC-fraction A (representative data in Figure 4G). The NTA showed an average particle concentration of 4.37 × 10 11 ± 4.3 × 10 6 particles/mL in a similar range as mock VLPs (representative histogram in Figure 4H). Together, these data indicate that pseudotyping with recombinant membrane protein did not affect the size, purity, or number of VLP particles (compare Figures 2,4).

Next, we evaluated MemPr1 expression in the VLPs by standard Western blotting (WB), biolayer interferometry (BLI), and dot blot (DB). For WB, we used a commercial antibody that recognizes denatured MemPr1 epitope. In the WB assay, highly enriched target expression was detected in VLPs pseudotyped with MemPr1 (MemPr1-VLPs) versus mock VLPs, which were not pseudotyped with any antigen (Figure 4I) using WB. For BLI and DB assays, which are performed on intact VLPs, a conformation-specific antibody against MemPr1 was used. The conformation-specific MemPr1 antibody was generated from the internal immunization campaign at J&J (see “Materials and Methods”). We confirmed by flow cytometry that this conformation-specific antibody recognizes native antigen on the cells (Supplementary Figure S1) but not the denatured MemPr1 epitope in WB (data not shown). A low level of basal expression was detected (Supplementary Figure S1), consistent with other experiments (Figures 4I,L, 5C–E). The DB assay on intact VLPs showed a marked enrichment of MemPr1 in antigen pseudotyped versus mock VLPs (Figure 4J) using conformation-specific. The BLI assay on the intact VLPs showed increased expression of the target in the SEC-fraction A and no detectable expression in SEC-fractions B and C (Figure 4K) or mock VLPs (Figure 4L) using conformation-specific antibody. The estimated total protein content was an average of 1.16 ± 0.16 mg from 500 ml of HEK293T cells (Figure 4M), indicating that our yields were not substantially affected by antigen co-expression. As of now, we have used the VLP platform to successfully express ten antigens (Supplementary Figure S2 shows nine additional representative examples). These antigens belong to ion channels, transporters, cell junctions, and orphan receptor family of proteins differing in the number of transmembrane domains and molecular weights. Altogether, we conclude that VLP is a robust and versatile platform that is agnostic to the type of membrane-bound protein being displayed; it retains the native conformation of the target while maintaining purity, monodispersity, and yield.

Quantification of antigen molecules per VLP by LC-MS/MS

In the following section, we describe a method for quantifying target antigens in the VLPs by LC-MS/MS. Such a method is critical when no tool antibodies are available, when the exact molarity of antigen is desired, or to quantify and compare target antigens in different VLP preparations. LC-MS/MS was previously used to detect membrane-bound antigens in MLV-Gag particles (Kurg et al., 2016) and subsequently to quantify Gag molecules in the VLPs produced from different cell lines (Lavado-Garcia et al., 2021). With our optimized method, we can both detect and quantitate any heterologous antigen expressed in the VLPs. The lipid envelope of VLPs was solubilized with RIPA lysis buffer, followed by on-column digestion with trypsin and peptide mapping. The peptide mapping data showed that MemPr1 was the only protein that was significantly increased in MemPr1-VLPs compared to the mock VLP samples (Supplementary Table 1). MemPr1-VLPs showed a 12.2-fold increase in spectral counts (Sp_C) over mock VLPs (Figure 5A; Supplementary Table 1, n = 2 biological replicates each, P-value = 0.018). Per Gag molecule, a 50-fold increase in Sp_C of MemPr1 was found in MemPr1-VLPs versus mock VLPs (Figure 5B, 0.039 vs. 0.00074). The peptide mapping analysis revealed a 26% coverage of MemPr1 protein sequence (not shown). Two MemPr1-specific peptides were chosen for the synthesis of stable-isotope-labeled versions to be used as internal standards. The peptides with the following desirable criteria were selected: unique to the target, containing no cysteines and methionine (as they can form disulfide bonds or are often oxidized respectively), and arising from no trypsin mis-cleavages. We compared the chromatographic peak area of the desired peptide with its corresponding stable-isotope-labeled peptide, which is spiked-into the sample, using parallel reaction mode (PRM) in LC-MS/MS. This analysis revealed an average of 10.36 nmol of MemPr1 in antigen-pseudotyped VLPs versus 0.072 nmol in mock VLP for the two peptides per 7.5 µg sample in 100 µL of each (Figure 5C; Table 2). This is equivalent to an average concentration of 6.24 × 10 13 MemPr1 molecules/mL and is approximately a 100-fold higher concentration of MemPr1 than mock VLPs (Figure 5D; Table 2). Furthermore, dividing antigen concentration (MemPr1/ml) with VLP concentration (particles/mL, obtained by NTA) (Table 2) gives MemPr1 molecules per VLP. This analysis showed an average of 246 MemPr1 molecules per MemPr1-VLP versus only 0.4 MemPr1 molecules per mock VLP particle (Figure 5E; Table 2). Notably, the number of antigen molecules could be under-estimated, as the method assumes 100% digestion of all the target proteins with trypsin.

VLPs retain structural integrity and antigen expression up to five freeze–thaw cycles

The stability of VLPs was tested by checking size, %PD, and antigen display upon accelerated stress by repeated freeze–thaw (F/T). The VLP preparations were frozen at −80 °C, followed by thawing at RT (room temperature). The assessment of size distribution with DLS did not reveal any significant difference in the overall size of the mock VLPs (Figures 6A–C,G) or MemPr1-VLPs (Figures 6D–G) after 1F/T, 3F/T, or 5F/T cycles. Consistent with that, the %PD stayed below 15% for all samples, and no significant difference was observed after 1F/T, 3F/T, or 5F/T cycles for both mock and MemPr1-VLPs (Figure 6H). Next, we assessed whether native conformation of MemPr1 was affected by F/T cycles. We did not observe any difference up to three F/T cycles (Figure 6I, maximum BLI signal=3.14 nm for no F/T vs. 2.91 nm for 3F/T). A small reduction in the BLI signal to 2.4 nm was noted after 5F/T cycles in the BLI assay (Figure 6I). Overall, the VLP particles retained size, monodispersity, and antigen display after multiple F/T cycles indicating their high stability. We also tested trehalose as a possible formulation buffer (Lynch et al., 2012). Interestingly, the addition of trehalose increased the polydispersity of the VLP particles (data not shown) and hence was not pursued further.

Comparison of VLPs produced in HEK293T, ExpiCHO, and CHO-S cell lines

To further optimize our VLP production, we evaluated whether cell lines other than HEK293T could enhance yield and antigen expression, which could lead to the production of more effective immunogens for downstream applications. We tested VLP production in ExpiCHO and CHO-S cells, which are commonly used in the preclinical discovery of biologics. The Gag and MemPr1 encoding plasmids were transfected at identical cell density and culture volumes into ExpiCHO and CHO-S, according to the manufacturer’s guidelines (Thermo Fisher Scientific, catalog #A29129, catalog #16447100, Materials and Methods). As before, the VLPs were recovered from media by PEG/NaCl precipitation, followed by purification with AEX (Figure 7A) and SEC (Figure 7B). When overlaid, the elution times of ExpiCHO and CHO-S VLPs were identical to VLPs derived from HEK293T, albeit the latter showed increased mAU (milli absorbance unit), indicating more protein content than both CHO cell lines (Figure 7B). ExpiCHO and CHO-S VLP preparations were determined to be monodisperse, as %PD stayed below 15% by DLS (Figures 7C,D). However, the particles were smaller for both ExpiCHO and CHO-S VLPs versus HEK-VLPs, presumably due to cell-line-specific differences in VLP production (Figures 7C,D vs. Figure 4D). The analytical HPLC showed comparable purity for VLPs derived from ExpiCHO, CHO-S, and HEK293T-VLPs (Figure 7E). Importantly, the VLP yields from CHO-S and ExpiCHO were 8- and 16-fold lower than HEK293T, with the preparations done from identical volume and cell density (Figure 7F). Furthermore, conformation-specific antibody against MemPr1 in BLI assay revealed maximum antigen to be displayed on HEK293T-VLPs, followed by ExpiCHO-VLPs where only half the binding was observed (Figure 7G), followed by CHO-S VLPs which failed to display any antigen (Figure 7G). Importantly for these experiments, identical amounts of 5 µg, equivalent to 5 × 10 8 particles determined from NTA, were used for HEK293T, ExpiCHO, and CHO-S VLPs. Comparable results were observed for two other antigens (data not shown). Altogether, among the cell lines evaluated, the VLPs produced from HEK293T displayed maximum antigen and highest VLP yields.

Dual-tag method for site-specific biotinylation of VLPs

To expand the utility of VLPs for preclinical studies, we embarked upon developing technology for the site-specific biotinylation of VLPs which would enable antigen immobilization, via streptavidin, for several downstream assays such as biophysical characterization and screening of potential binders. To this end, we engineered a recognition site for the sortase enzyme (LPXTG) (Theile et al., 2013) displayed extracellularly through a transmembrane snorkel tag (Brown et al., 2013). The snorkel tag contained the transmembrane domain from the single-pass mouse-platelet-derived growth factor receptor PDGFRB (UniProt #P05622). Consistent with the “positive-inside rule” for membrane proteins (Dowhan and Bogdanov, 2009; Nilsson et al., 2005), the snorkel construct contains engineered sequences to retain correct topology in the membrane. Additionally, we incorporated GS residues proximal to the membrane for flexibility and an increasing accessibility of Sortase to its recognition sequence, LPXTG, which is displayed terminally on the VLPs (see Materials and Methods) (Brown et al., 2013) (Figure 8A). The plasmid encoding the dual snorkel–sortase tag was then co-transfected into HEK293T cells along with Gag and plasmid-encoding MemPr1. The VLPs were purified using our standard protocol (representative chromatograms in Figures 8B–E). The VLPs were eluted at the expected volume for both mock and MemPr1-VLPs. We subjected 100 µg of each of the VLPs to in vitro biotinylation reaction with purified sortase and poly-glycine conjugated biotin (Gly 5 -biotin) using the manufacturer’s guidelines (Figure 8A see Materials and Methods). After reaction, the biotinylated VLPs were collected in the void volume of HiScreen Capto Core column 700 (Tarres-Freixas et al., 2023), separating the remaining reaction constituents (Figures 8F,G). The biotin conjugation on VLPs was detected via two methods: dot blot on the intact VLPs using streptavidin-HRP (Figure 8H) and binding of VLPs to the Streptavidin (SAX) biosensor in a BLI assay (Figure 8I). The degree of biotinylation was determined by a FluoReporter Biotin Quantitation kit (See Materials and Methods). An average of 4,200 biotin molecules/VLP were detected in MemPr1-VLPs (Table 3), far exceeding the number of antigen molecules per VLP (Table 2); this indicated that biotinylation was more abundant than the membrane protein expression. We then assessed whether biotin conjugation had any impact on antigen detection. Similar levels of MemPr1 were detected before and after the biotinylation reaction by BLI (Figure 8J). Analytical SEC did not reveal any differences in the overall purity after biotinylation (Figure 8K, red arrows). A small peak corresponding to lower molecular weight, probably from unreacted biotin, was detected in the analytical HPLC profiles. DLS showed no difference in the overall size, and polydispersity stayed below 15% after biotinylation reaction (Figure 8L). In conclusion, we have developed a robust and reproducible method for site-specific biotin conjugation of VLPs that does not mask any epitopes on antigens or affect the size or monodispersity of VLPs.

Determination of binding affinity of antibody to VLP-displayed antigen via surface plasmon resonance (SPR)

Using biotin-conjugated MemPr1-VLPs, we determined the binding kinetics of an internally generated antibody against MemPr1 using SPR. MemPr1-VLPs were immobilized on a SPR sensor chip via streptavidin–biotin coupling. A bispecific antibody, monovalent for MemPr1, was then injected over the sensor chip at progressively increasing concentrations in a single-cycle kinetics assay and analyzed by SPR (representative sensogram in Figure 9A). A significantly increased binding signal specific for MemPr1 versus mock VLPs was observed (Figure 9A). The mock VLP values were subtracted, with the resultant curve shown in Figure 9A. These data were subsequently analyzed using a 1:1 binding model to determine binding kinetics and dissociation constant. Figure 9B shows that binding parameters for VLPs immobilized at three different densities (see immobilized response unit (RU) levels in the table) were similar. An average dissociation constant of 14.1 nM was obtained for the bispecific antibody with a monovalent binding site for MemPr1 (Figure 9B). Altogether, these results validate the use of the VLP platform for the biophysical characterization of therapeutic antibodies against complex membrane targets.

Commercial VLP-based vaccines such as PreHevbrio for hepatitis (Vesikari et al., 2021) and Cervarix (Paavonen et al., 2007), Gardasil, and Gardasil 9 (Joura et al., 2015) for human papillomavirus demonstrate that VLPs are efficacious and robust immunogens. In this study, we have described methods for the scalable and reproducible production of HIV-1 Gag based VLPs for the multivalent display of membrane proteins in their native conformation as a key tool reagent during preclinical drug discovery. The systematic characterization showed that our VLP preparations were pure and monodisperse and displayed antigens in a native conformation, setting a benchmark in the field (Figure 10). The site-specific conjugation of biotin on the VLP surface using a snorkel–sortase dual tag preserved the integrity of the antigen and facilitated the immobilization of natively displaying membrane protein in a SPR assay, enabling the determination of kinetic parameters for an internal biotherapeutic (Figure 10). Altogether, our process improvements and innovations will lead to the production of high-quality VLP preparations that can be used to display a wide variety of valuable, disease-relevant membrane proteins for several downstream applications. These applications include immunization, biophysical characterization, panning antibody libraries, affinity maturation, screening binders from antibody generation campaigns, and pharmacokinetic analysis of therapeutic antibodies (Figure 10). In contrast, other methods of purified membrane protein preparations such as micelles and many types of nanodiscs require extensive optimization, resources, and time (Zhou and Cross, 2013).

Previously, researchers have used ultracentrifugation to condense VLPs from the clarified media (Cervera et al., 2013), while others have purified VLPs with AEX (Steppert et al., 2016) and then a subsequent elution in smaller SEC columns (Tarres-Freixas et al., 2023; Gonzalez-Dominguez et al., 2021; Cervera et al., 2019; Gutierrez-Granados et al., 2016). Ultracentrifugation lacks throughput, and performing AEX purification alone did not result in material of high purity—only ∼70% purity (Figure 2D) and PD=31% after AEX (data not shown). Moreover, the quality of purified VLPs was difficult to assess due to limited QC on purity and monodispersity. Here, we show a reproducible and scalable method for obtaining pure and monodisperse VLP particles by following a four-step protocol: clarification of media, precipitation of VLPs with PEG/NaCl, purification with AEX, and subsequent fractionation by preparative SEC. Our workflow involves standard AEX elution with Fractogel® TMAE resin, followed by fractionation on a 120 mL Superose SEC column (Figures 1–3). We performed an extensive quality check with analytical SEC-HPLC, DLS, NTA, Western blotting, antigen detection by conformation-specific antibody in BLI, and dot blot assays (Figure 4). The electron micrographs of our preparations did not reveal the presence of exosomes and show a clear lipid envelope surrounding the internal capsid, further providing evidence that our preparations are pure and monodisperse (Figure 3). We demonstrated the versatility of the VLP platform to incorporate a wide array of membrane-bound proteins of different classes and numbers of transmembrane domains (Supplementary Figure S2); the methods described herein are agnostic to the type of membrane protein being expressed. However, other reports suggest preferential budding of VLPs from cholesterol and glycosphingolipid-rich lipid rafts, which could lead to the selective enrichment of specific membrane proteins in the VLP membrane (Lingwood et al., 2010; Nguyen and Hildreth, 2000; Llewellyn et al., 2010; Llewellyn et al., 2013; Hogu et al., 2012; Grover et al., 2015; Favard et al., 2019; Thali, 2009; Nydegger et al., 2006). Many researchers have increased the valence of antigens in the VLP membrane by fusing heterologous proteins with Gag-compatible cytoplasmic tail of Env proteins (Zhang et al., 2023) or by fusing the antigen itself to Gag proteins (Tarres-Freixas et al., 2023). The display of a wide variety of membrane proteins on VLPs (Supplementary Figure S2) suggests that their overexpression may override the barriers of selective localization into distinct membrane compartments (Doyle and Wang, 2019).

In addition to streamlining purification workflows for VLPs, the technical improvements highlighted in this study will allow for better adaptation of VLPs in the preclinical discovery of biologics. First, site-specific biotin labeling is ideal for immobilization antigens for downstream assays such as SPR and the high-throughput screening of binders. This is in contrast with the chemical conjugation of biotin using reactive amines or cysteines, which, while being efficient, can compromise epitope detection by antibodies (Tantipanjaporn and Wong, 2023). These biotinylated MemPr1-VLPs were used in the SPR assay to determine the dissociation constants for a monovalent bispecific antibody which could be adapted for other membrane proteins as a plug-and-play approach. Second, we developed methods for quantifying target protein molecules per VLP by LC-MS/MS. Our calculation of the number of MemPr1 molecules per VLP is 246, which is in accord with the multivalent display of antigens in VLPs. This number is likely an under-estimation since the process relies on the efficiency of trypsin digestion and often results in poor coverage in transmembrane regions of membrane proteins. Nevertheless, our method represents a significant advance in the field and provides methods for quantitation of antigen molecules in VLPs. Lastly, we found that, among the cell lines tested, HEK293T cells were the best in terms of antigen expression and VLP yields. The primary reason could be that HIV-1 Gag engages the cellular machinery of HEK293T more efficiently to produce VLPs than ExpiCHO and CHO-S.

Overall, this study describes methodologies for the purification, characterization, antigen quantitation, and site-specific biotin labeling of VLPs. The extensive characterization of antigen pseudotyped VLPs and site-specific labeling set a benchmark for VLP antigen display and will expand its potential applications. The methods described here will significantly advance the applications of VLPs in several downstream applications in therapeutic antibody discovery workflows, vaccines, and drug nanocarrier production (Nooraei et al., 2021).