Reprogramming Extracellular Vesicles for Protein Therapeutics Delivery

Delivering protein therapeutics specifically into target cells and tissues is a promising avenue in medicine. Advancing this process will significantly enhance the efficiency of the designed drugs. In this regard, natural membrane-based systems are of particular interest. Extracellular vesicles (EVs), being the bilayer lipid particles secreted by almost all types of cells, have several principal advantages: biocompatibility, carrier stability, and blood–brain barrier penetrability, which make them a perspective tool for protein therapeutic delivery. Here, we evaluate the engineered genetically encoded EVs produced by a human cell line, which allow efficient cargo loading. In the devised system, the protein of interest is captured by self-assembling structures, i.e., “enveloped protein nanocages” (EPN). In their turn, EPNs are encapsulated in fusogenic EVs by the overexpression of vesicular stomatitis virus G protein (VSV-G). The proteomic profiles of different engineered EVs were determined for a comprehensive evaluation of their therapeutic potential. EVs loading mediated by bio-safe Fos–Jun heterodimerization demonstrates an increased efficacy of active cargo loading and delivery into target cells. Our results emphasize the outstanding technological and biomedical potential of the engineered EV systems, including their application in adoptive cell transfer and targeted cell reprogramming.


Introduction
Delivering protein cargoes into target cells remains a significant challenge for modern pharmaceuticals and cell biology still holding a promise of a new era of precision medicine. Targeted delivery technology aims at creating multifunctional carriers that are capable of circulating in the patient's body for a long time and concurrently have low toxicity [1]. In light of this, extracellular vesicles (EVs) seem to be the perfect carriers for protein therapeutics. EVs are classified into two main subtypes: ectosomes (or microvesicles) and exosomes [2]. These two types of vesicles differ in size, origin, composition, and biogenesis mechanism [3]. Generally, exosomes are characterized as nanosized (40-160 nm) EVs, originating from multivesicular bodies and secreted by cells [4]. Exosomes can be isolated by ultracentrifugation at 100,000× g. Ectosomes (microvesicles) are much bigger and have a broader size distribution (100-1000 nm) and plasma membrane origin. They are released during cell-surface budding [5] and can be precipitated at 10,000× g [6]. Apoptotic bodies (>1000 nm) are often classified as the third type of EVs. They are highly specific and make up a minor group of EVs formed during the late stages of apoptosis [7].

Dynamic Light Scattering
All dynamic light scattering (DLS) measurements were obtained using Zetasizer Nano ZS (Malvern Instruments Ltd., WR, Malvern, UK). Purified protein nanocages were measured at 25 • C, then the temperature was ramped up to 60 • C, then ramped back down to 25 • C. Measurements were taken in the presence of TBS: 25 mM Tris, 500 mM NaCl or buffered GITC: 25 mM Tris, 500 mM NaCl, 3 M GITC. Each sample was allowed to equilibrate in the respective buffer for at least 24 h before measurement.

Isolation and Purification of EVs
To optimize EV production, HEK293T and HEK293F cell lines were used. HEK293F cells were seeded in a 24-well plate at 0.5 × 10 6 cells/mL to a final volume 1 mL per well. HEK293T cells were seeded in a 12-well plate. After 24 h, HEK293F reached a final density of 1 × 10 6 cells/mL, and HEK293T reached 80-90% confluence and were transfected with: (i) 0.5 µg of pCMV-NanoLuc-Jun, 0.9 µg of pCMV-EPN-Fos-C, 0.1 µg of pCMV-VSVG or (ii) 0.9 µg of pCMV-EPN-Fos-C, 0.1 µg of pCMV-VSVG or (iii) 0.5 µg of pCMV-NanoLuc-Jun using 293fectin transfection reagent (Gibco, OK, USA) for HEK293F or Lipofectamine 3000 transfection reagent (Invitrogen, MA, USA) for HEK293T. The plasmid ratio was defined in the previously published protocol for the production of engineered EVs [28]. The culture supernatant was collected 48 h and 72 h after transfection. Next, the cell debris and microvesicles were removed after two steps of differential centrifugation at 4 • C (300× g for 10 min, 1000× g for 20 min) and subsequent filtration of supernatant through 0.45 µm filter device (GE Healthcare, UK). EVs were concentrated using Amicon Ultra-0.5 mL 100 kDa Filter Unit (Millipore, Cork, Ireland) and washed with PBS (Gibco, OK, USA) three times. To determine the quantity of NanoLuc luciferase in EVs, Nano-Glo Luciferase Assay (Promega, Madison, WI, USA) was performed under manufacturer's protocol.
For large-scale production of the genetically encoded EVs, HEK293T cells were seeded in T-75 flasks. After 24 h, cells reached 80-90% confluence and were transfected with appropriate plasmids (Table S1) using Lipofectamine 3000 transfection reagent. After 6 h, cells were washed with PBS to remove vesicle residues from FBS and the media were changed to DMEM supplemented with 0.5% BSA (Sigma-Aldrich, St. Louis, MO, USA) and L-Glutamine. Forty-eight hours after transfection, EV-containing media were collected and cell debris and microvesicles were depleted as described above. Purified EVcontaining concentrate was collected through sucrose density gradient ultracentrifugation. Briefly, EVs were mixed with 90% (w/v) sucrose to a final sucrose percentage of 82% (w/v) and then overlaid with a sucrose gradient (10% (w/v)-70% (w/v) sucrose in PBS). Centrifugation was performed at 100,000× g (RCF avg) for 16 h at 4 • C in Beckman Coulter L8-M ultracentrifuge, using Beckman Coulter SW 28 Ti Swinging-Bucket rotor. Fractions, enriched with EVs, were determined using NanoGlo Luciferase Assay, mixed and diluted with PBS, then concentrated using Amicon Ultra-15 centrifugal Filter unit with 100 kDa cut-off (Millipore, Cork, Ireland) and washed with PBS three times. Protein concentrations in probes were measured by CBQCA Protein Quantitation Kit (Invitrogen, Carlsbad, CA, USA). The diameters of EVs were determined in PBS at 25 • C using a Zetasizer Nano ZS.

Transmission Electron Microscopy (TEM)
The carbon-coated TEM grids (Ted Pella, Redding, CA, USA) were treated using a glow discharge device Emitech K100X (Quorum Technologies Ltd., Lewes, UK) at 25 mA current for 45 s. This treatment turned the carbon surface hydrophilic and increased the EVs' adsorption. The suspension of EVs was deposited onto the grid for 2-3 min, and the liquid was blotted using filter paper (OdiChem, Moscow, Russia). The grids were quickly transferred onto the drops of 1% uranyl acetate, stained for 1-2 min, blotted and dried. Images were obtained using a JEM-1011 (Jeol, Tokyo,Japan) transmission electron microscope equipped with Orius SC1000W camera (Gatan Inc., Pleasanton, CA, USA). The magnification was in the range from 30,000× to 250,000×; the acceleration voltage was 80 kV. The sizes of the EVs were measured using ScanEV [32].

Protein Delivery of EV Cargoes to Target Cells
The ability of engineered EVs to deliver NanoLuc-Jun into target cells was determined using NanoGlo Luciferase Assay System (Promega, Madison, WI, USA). EVs were purified by sucrose density centrifugation as described above. EVs were aligned according to the portion of NanoLuc protein in each sample (375 pg NanoLuc), added to 300,000 Jurkat cells and incubated for 2 h in serum-free medium in triplicate. Recombinant NanoLuc-Jun protein was added to control wells. After incubation, the cells were washed with PBS at 300× g for 10 min and then were incubated with 0.005% NP-40 with rotation at room temperature for 10 min. After incubation with detergent, cells were washed twice with PBS and 40,000 cells from each sample were analyzed by Nano-Glo Luciferase Assay (Promega, Madison, WI, USA) to determine the quantity of delivered protein in target cells.
Raw spectra were processed using in MaxQuant 1.6.6.0 (MQ) [34] and Perseus [35]. The data were searched against Human Uniprot Swissprot database, containing canonical and isoform proteins, version from 02.2019, and additionally the sequences of VSV-G, Nano-Luc and EPN were added.
MaxQuant search was performed with the default parameter set, including Trypsin/p protease specificity, max 2 missed cleavages, Met oxidation, Protein N-term acetylation and NQ deamidation as variable modifications and Carbamidomethyl Cys as a fixed modification, max 5 modifications per peptide, 1% PSM and protein FDR. The following options were turned on: second peptide, maxLFQ, match between runs. All runs were analyzed as independent experiments and processed in Perseus.
In Perseus, the protein group results were filtered for contaminants, reverse and "identified only by site" proteins. Only the proteins with maxLFQ values in at least 3 out of 7 LC-MS runs were used. For them, missing values were imputed from normal distribution with 0.3 intensity distribution sigma width and 1.8 intensity distribution center downshift. To quantify proteins in each sample, we used the iBAQ algorithm, implemented into MaxQuant software [36]. Normalization of each protein's iBAQ value to the sum of all iBAQ values generates a relative iBAQ (riBAQ) value corresponding to the mole fraction of each protein [37]. GO analysis was performed in DAVID [38,39].

Molecular Modeling
Visualization of the EVs was created in CellPAINT 2.0 [40]. Individual sprites were created in Illustrate [41] based on 3D structures generated with trRosetta [42].

Estimation of Protein Numbers in EVs
The average EV particle mass was estimated, assuming the average EV diameter of 105 nm measured by TEM and EV density of 1.14 g/mL corresponding to 33% (v/w) sucrose fraction.
To determine the abundance of VSV-G, EPN and NanoLuc in EVs, we performed quantitative Western blot and luciferase assay experiments corrected by MS-MS analysis.
Therefore, we estimate the amount of the analyzed protein in a single EV according to the following formula: protein amount in 1 mcg o f vesicle × vesicle mass (6.9 × 10 −16 g) protein molecular weight × N A 6.02 × 10 23 1 mol

Quantification and Statistical Analysis
All experiments were carried out in triplicate (biological repeat). Each biological repeat for DLS and luciferase assay was additionally measured at least in triplicate (technical repeat). Statistical analyses of data were performed with GraphPad Prism 7. Data are represented as mean ± SD. For the comparison of differentially expressed proteins between cell-whole lysates and EVs, as well as across the EV subtypes, an ANOVA test was performed.

Improving Loading Capacity of Protein Nanocages Preserves Their Capacity to Self-Assemble
Spike glycoprotein of the vesicular stomatitis virus (VSV-G) is commonly used for pseudotyping retroviral and lentiviral particles to facilitate membrane fusion with the target cells and enable their entry into mammalian cells. Recently, the usage of VSV-G has been started for producing the engineered vesicles and the subsequent delivery of the exogenous proteins into human cells [43,44]. Active protein loading inside the selfassembling EPN nanocages and their further encapsulation in VSV-G-containing EVs were realized via fusion of the 14 kDa Vpr to the delivered protein [28]. Naturally, Vpr interacts with the p6 domain of the HIV-GAG precursor, thereby packaging Vpr-fusion proteins into viral particles [45]. Upon an attempt to use Vpr for recombinant protein delivery into HIV-1 particles, Vpr-dependent G2 arrest activity and epigenetic disruption of heterochromatin were observed [46,47]. There are several reports of Vpr mutants with lower cytotoxicity [48], but harnessing this protein for therapeutic purposes is still under debate.
Here, the previously reported EVs [28], containing nanocages with the p6 domain, are designated as "EV_p6". To be able to use the EVs as a therapeutic agent, we replaced this toxic and oncogenic protein Vpr with the domain of the leucine zipper of the Jun protein, which forms a leucine zipper heterodimer with the domain of the Fos protein [49]. One of the substantial advantages of the Fos-Jun dimer compared with Vpr-p6 is its small size (4 kDa). Utilizing these domains instead of Vpr full-length protein allows us to fully exploit the loading potential of protein nanocages. We explored two new variants of EPN protein nanocage scaffolds with two different positions of the Fos domain-at the C-end of translated protein (EPN-Fos-C) or between the Interaction domain and the Late budding domain (EPN-Fos-M) ( Figure 1).
Spike glycoprotein of the vesicular stomatitis virus (VSV-G) is commonly used for pseudotyping retroviral and lentiviral particles to facilitate membrane fusion with the target cells and enable their entry into mammalian cells. Recently, the usage of VSV-G has been started for producing the engineered vesicles and the subsequent delivery of the exogenous proteins into human cells [43,44]. Active protein loading inside the self-assembling EPN nanocages and their further encapsulation in VSV-G-containing EVs were realized via fusion of the 14 kDa Vpr to the delivered protein [28]. Naturally, Vpr interacts with the p6 domain of the HIV-GAG precursor, thereby packaging Vpr-fusion proteins into viral particles [45]. Upon an attempt to use Vpr for recombinant protein delivery into HIV-1 particles, Vpr-dependent G2 arrest activity and epigenetic disruption of heterochromatin were observed [46,47]. There are several reports of Vpr mutants with lower cytotoxicity [48], but harnessing this protein for therapeutic purposes is still under debate.
Here, the previously reported EVs [28], containing nanocages with the p6 domain, are designated as "EV_p6". To be able to use the EVs as a therapeutic agent, we replaced this toxic and oncogenic protein Vpr with the domain of the leucine zipper of the Jun protein, which forms a leucine zipper heterodimer with the domain of the Fos protein [49]. One of the substantial advantages of the Fos-Jun dimer compared with Vpr-p6 is its small size (4 kDa). Utilizing these domains instead of Vpr full-length protein allows us to fully exploit the loading potential of protein nanocages. We explored two new variants of EPN protein nanocage scaffolds with two different positions of the Fos domain-at the C-end of translated protein (EPN-Fos-C) or between the Interaction domain and the Late budding domain (EPN-Fos-M) (Figure 1). EPN protein nanocages are highly ordered self-assembled protein dodecahedrons, consisting of 60 monomers, each provided with one p6 domain ("EPN-p6" Figure 1) [27]. To confirm the capability of the modified protein nanocages EPN-Fos-C and EPN-Fos-M to self-assemble into highly ordered structures, we performed the production of EPN in the prokaryotic system. Dynamic light scattering (DLS) measurements of modified protein nanocages with the Fos domain showed a monodisperse population of particles with EPN protein nanocages are highly ordered self-assembled protein dodecahedrons, consisting of 60 monomers, each provided with one p6 domain ("EPN-p6" Figure 1) [27]. To confirm the capability of the modified protein nanocages EPN-Fos-C and EPN-Fos-M to self-assemble into highly ordered structures, we performed the production of EPN in the prokaryotic system. Dynamic light scattering (DLS) measurements of modified protein nanocages with the Fos domain showed a monodisperse population of particles with a hydrodynamic diameter of 25 nm (Figure 2), which corresponds to the size of the original nanocages with only the p6 domain [27]. Additionally, to compare their stability with the original EPN-p6, the purified nanocages were incubated overnight in the presence of 3 M GITC (denaturing conditions) or in non-denaturing conditions. Compared to literature data, the dissociation of the original EPN-p6 and modified nanocages was observed only at 3 M GITC (Figure 2A). No disassembly was observed at 60 • C ( Figure 2B) or in 2 M GITC (data not shown). Thus, we state that the Fos domain does not influence the ability of EPN nanocages to self-assemble. Assuming this with the fact that Fos and Jun domains are biocompatible, we believe that EPN-Fos can be considered to be a promising drug carrier in therapeutics development. 3 M GITC (denaturing conditions) or in non-denaturing conditions. Compared to literature data, the dissociation of the original EPN-p6 and modified nanocages was observed only at 3 M GITC (Figure 2A). No disassembly was observed at 60 °C ( Figure 2B) or in 2 M GITC (data not shown). Thus, we state that the Fos domain does not influence the ability of EPN nanocages to self-assemble. Assuming this with the fact that Fos and Jun domains are biocompatible, we believe that EPN-Fos can be considered to be a promising drug carrier in therapeutics development.

Production of Engineered EVs in Mammalian Cells
The self-assembled nanocages loaded with protein cargoes could be delivered to target cells via engineered genetically encoded EVs. For a more reliable analysis of cargo encapsulation and delivery, we used NanoLuc as a model for potential protein pharmaceutics. First of all, we confirmed that fusion with the Jun domain will not interrupt the encapsulation of self-assembled nanocages loaded with NanoLuc-Jun in EVs. Using pCMV-VSV-G, pCMV-EPN-Fos-C and pCMV-NanoLuc-Jun for transient co-transfection of eukaryotic cells, we obtained the EVs loaded with NanoLuc and optimized EV production ( Figure 3A). It was shown that both tested cell lines are able to produce EVs. Nevertheless, HEK293T cells are more suitable, as they have a higher level of EV production in a shorter time (48 h). Further, we showed that the use of EPN-Fos-M nanocages allows the production of the same level of EVs loaded with NanoLuc comparable to EPN-Fos-C (Figure 3B).

Production of Engineered EVs in Mammalian Cells
The self-assembled nanocages loaded with protein cargoes could be delivered to target cells via engineered genetically encoded EVs. For a more reliable analysis of cargo encapsulation and delivery, we used NanoLuc as a model for potential protein pharmaceutics. First of all, we confirmed that fusion with the Jun domain will not interrupt the encapsulation of self-assembled nanocages loaded with NanoLuc-Jun in EVs. Using pCMV-VSV-G, pCMV-EPN-Fos-C and pCMV-NanoLuc-Jun for transient co-transfection of eukaryotic cells, we obtained the EVs loaded with NanoLuc and optimized EV production ( Figure 3A). It was shown that both tested cell lines are able to produce EVs. Nevertheless, HEK293T cells are more suitable, as they have a higher level of EV production in a shorter time (48 h). Further, we showed that the use of EPN-Fos-M nanocages allows the production of the same level of EVs loaded with NanoLuc comparable to EPN-Fos-C ( Figure 3B).

EV Production by Lentiviral Construction
In case of the genetically encoded EVs used as a therapeutic agent, it is advisable to use the single genetic construct. The co-transfection of eukaryotic cells with several genetic constructs leads to non-uniform distribution of vectors between cells. To ensure a broader standardization of EVs, we created the multicistronic construct based on a lentiviral vector pLX ( Figure 4A). Other advantages of a lentiviral vector are the possibility to deliver gene information into dividing and non-dividing cells and to create a stable cell producer line for EVs, which is significantly cost-effective. These features provide the ability to mediate long-term therapeutic transgene expression not only in vitro but also in a patient's body. We analyzed the total yield of NanoLuc encapsulated into EVs, both those produced in co-transfected cells and with a lentiviral construct utilized. Although the overall yield of the encapsulated NanoLuc is falling in the case of lentiviruses (Figure 4), lentiviral transduction can be applied for further EV production as it allows for the theoretically maximum cargo filling of each individual EV. Each experiment was performed in triplicate. Mean ± SD is shown. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, ns-not significant, two-tailed Student's t-test. Utilization of each of the genetic constructs (pCMV-VSVG, pCMV-EPN-Fos-C, pCMV-EPN-Fos-M, pCMV-EPN-p6, pCMV-NanoLuc-Jun, pCMV-NanoLuc-Vpr) for the production of EVs is indicated as "+".

EV Production by Lentiviral Construction
In case of the genetically encoded EVs used as a therapeutic agent, it is advisable to use the single genetic construct. The co-transfection of eukaryotic cells with several genetic constructs leads to non-uniform distribution of vectors between cells. To ensure a broader standardization of EVs, we created the multicistronic construct based on a lentiviral vector pLX ( Figure 4A). Other advantages of a lentiviral vector are the possibility to deliver gene information into dividing and non-dividing cells and to create a stable cell producer line for EVs, which is significantly cost-effective. These features provide the ability to mediate long-term therapeutic transgene expression not only in vitro but also in a patient's body. We analyzed the total yield of NanoLuc encapsulated into EVs, both those produced in co-transfected cells and with a lentiviral construct utilized. Although the overall yield of the encapsulated NanoLuc is falling in the case of lentiviruses (Figure 4), lentiviral transduction can be applied for further EV production as it allows for the theoretically maximum cargo filling of each individual EV.  For each analysis, samples were concentrated with 100 kDa membrane and washed three times with PBS to remove NanoLuc which is not encapsulated in EVs. NanoLuc-Jun amounts in EVs were measured via NanoGlo assay, utilizing prokaryotic NanoLuc-Jun protein for calibration. ** p < 0.01, two-tailed Student's t-test (n = 3).

Isolation and Characterization of Genetically Encoded EVs
To further characterize the engineered EVs and compare them with the previously published data, we obtained the EVs with modified EPN-Fos-C, EPN-Fos-M nanocages and EVs produced with a lentiviral construction (EPN_C(lenti)) ( Table S1). As a control For each analysis, samples were concentrated with 100 kDa membrane and washed three times with PBS to remove NanoLuc which is not encapsulated in EVs. NanoLuc-Jun amounts in EVs were measured via NanoGlo assay, utilizing prokaryotic NanoLuc-Jun protein for calibration. ** p < 0.01, two-tailed Student's t-test (n = 3).

Isolation and Characterization of Genetically Encoded EVs
To further characterize the engineered EVs and compare them with the previously published data, we obtained the EVs with modified EPN-Fos-C, EPN-Fos-M nanocages and EVs produced with a lentiviral construction (EPN_C(lenti)) ( Table S1). As a control for the non-specific encapsulation of the targeted protein, we analyzed co-transfection without EPN nanocages (EV_EPN(-)). As a control for the natural EVs constitutively produced by cells, where NanoLuc can be encapsulated non-specifically, we used transfection with pCMV-NanoLuc-Jun only (designated NLuc). Vesicles obtained in eukaryotic producer cells are often contaminated with various proteins and self-vesicles from the media serum. Further analysis of the genetically encoded EVs requires ultra-purified preparations, i.e., density separation of EVs being "a gold standard" for vesicle isolation [50]. Prior to gradient centrifugation, we performed several steps of differential centrifugation, concentration of the final supernatant and its washing at 100 kDa membrane to remove media proteins including soluble NanoLuc, which is not encapsulated into EVs. The presence of the genetically encoded EVs was checked with a luciferase assay. For all types of vesicles, except "NLuc", luciferase was enriched in fractions with 25-40% (w/v) sucrose ( Figure S2). We measured the size of the highly purified EVs utilizing the dynamic light scattering method ( Figure S3 and Table 1). The average diameter of EVs loaded with EPN was 170 nm and did not differ significantly (p > 0.5), while unloaded EV_EPN(-) were smaller (p < 0.01) and had an average diameter of 150 nm. The TEM images of the samples EV_C and EV_C(lenti) are shown in Figure 5A,B. Most particles demonstrated the cup-shaped morphology, which is typical of the EVs imaged by negative staining TEM. Up to one-third of the particles had visible VSV-G spikes protruding from the vesicle membranes, as shown in the insets. The mean thickness of this "shell" was 12 ± 2 nm (mean ± SD). The mean sizes of the EV_C were 110 ± 40 nm and of EV_C(lenti) were 100 ± 40 nm. The histograms of the size distributions are shown in Figure 5C.
To calculate the amounts of VSV-G and NanoLuc-Jun in each EV, we first performed a quantitative measurement of VSV-G by immunoblotting and NanoLuc measurement by luciferase assay using the purified recombinant proteins as standards. According to this model, we estimated the quantity of key recombinant proteins and nanocages in EVs ( Table 2). The amount of NanoLuc molecules in every single EPN nanocage was significantly higher in EV_C as compared with EV_p6 (15 and 1.5 molecules, respectively), which may be due to the small molecular size of the Jun domain and its flexibility. However, EV_C is inferior in terms of loading capacity to EV_C(lenti). In EV_C(lenti), almost all sites on the surface of EPN nanocages were occupied with NanoLuc cargo (40 out of 60). This observation confirms the importance of the uniform distribution of genetic constructs among the cells for producing the vesicles loaded with cargo proteins. To illustrate different EVs, filled with cargo protein (NanoLuc) and nanocages, we obtained a 2D model (Figure 6).
The TEM images of the samples EV_C and EV_C(lenti) are shown in Figure 5A,B. Most particles demonstrated the cup-shaped morphology, which is typical of the EVs imaged by negative staining TEM. Up to one-third of the particles had visible VSV-G spikes protruding from the vesicle membranes, as shown in the insets. The mean thickness of this "shell" was 12 ± 2 nm (mean ± SD). The mean sizes of the EV_С were 110 ± 40 nm and of EV_C(lenti) were 100 ± 40 nm. The histograms of the size distributions are shown in Figure 5С. To calculate the amounts of VSV-G and NanoLuc-Jun in each EV, we first performed a quantitative measurement of VSV-G by immunoblotting and NanoLuc measurement by luciferase assay using the purified recombinant proteins as standards. According to this model, we estimated the quantity of key recombinant proteins and nanocages in EVs (Table 2). The amount of NanoLuc molecules in every single EPN nanocage was significantly higher in EV_C as compared with EV_p6 (15 and 1.5 molecules, respectively), which may be due to the small molecular size of the Jun domain and its flexibility. However, EV_C is inferior in terms of loading capacity to EV_C(lenti). In EV_C(lenti), almost all sites on the surface of EPN nanocages were occupied with NanoLuc cargo (40 out of 60). This observation confirms the importance of the uniform distribution of genetic constructs among the cells for producing the vesicles loaded with cargo proteins. To illustrate different EVs, filled with cargo protein (NanoLuc) and nanocages, we obtained a 2D model ( Figure 6).  Table 2. The estimated number of recombinant protein molecules per single vesicle. The numbers of VSV-G proteins are derived from the quantitative Western blotting results ( Figure S4), the numbers of NanoLuc proteins are derived from the luciferase assay measurement ( Figure S5) and the numbers of EPN proteins are derived from the EPN/NanoLuc molar ratio estimated by MS/MS analysis (Table S2). Data represent the mean of three biological replicates ± SD.

Delivery of Protein Cargo to Target Cells
To confirm the functionality of EVs with the novel Fos domain and lentiviral EVs, we analyzed their ability to deliver cargo proteins into target cells. Each sample was diluted according to NanoLuc portion before the incubation with target cells. Delivery efficiency was quantified via luciferase assay. We found that EV_C and EV_M vesicles delivered NanoLuc-Jun more effectively than EV_p6 (Figure 7), indicating the improvement of the initial EV capacity. It should be mentioned that the lentiviral construct showed the same efficiency as the EVs obtained via a co-transfection when adding an equal amount of NanoLuc, loaded into EVs ( Figure S6). However, EVs_C(lenti) contained more than twice as much NanoLuc in the equal amount of EV preparation. Thus, considering the total amount of the EVs supplemented to the target cells, it is evident that EV_C(lenti) allow the delivery of more NanoLuc. Thereby, using the lentiviral construct is the most effective method of loading a protein cargo into engineered EVs and delivering it to the target cell.

Delivery of Protein Cargo to Target Cells
To confirm the functionality of EVs with the novel Fos domain and lentiviral EVs, we analyzed their ability to deliver cargo proteins into target cells. Each sample was diluted according to NanoLuc portion before the incubation with target cells. Delivery efficiency was quantified via luciferase assay. We found that EV_C and EV_M vesicles delivered NanoLuc-Jun more effectively than EV_p6 (Figure 7), indicating the improvement of the initial EV capacity. It should be mentioned that the lentiviral construct showed the same efficiency as the EVs obtained via a co-transfection when adding an equal amount of NanoLuc, loaded into EVs ( Figure S6). However, EVs_C(lenti) contained more than twice as much NanoLuc in the equal amount of EV preparation. Thus, considering the total amount of the EVs supplemented to the target cells, it is evident that EV_C(lenti) allow the delivery of more NanoLuc. Thereby, using the lentiviral construct is the most effective method of loading a protein cargo into engineered EVs and delivering it to the target cell. Highly purified EVs containing 50 pg of NanoLuc were applied to 40,000 target cells. The amount of added EVs was normalized to NanoLuc portion in each EV sample. Data are mean ± SD, 3 biological repeats were used. * p < 0.05, *** p < 0.001, **** p < 0.0001, two-tailed Student's t-test.

Proteome Profiling of Engineered EVs
To better understand the protein composition of the obtained vesicles, proteome analysis of highly purified vesicles was performed. An equal amount of input proteins for three independent preparations for each type of vesicle was trypsinized and analyzed by Figure 7. The level of delivered NanoLuc in cells after 2 h of incubation. Highly purified EVs containing 50 pg of NanoLuc were applied to 40,000 target cells. The amount of added EVs was normalized to NanoLuc portion in each EV sample. Data are mean ± SD, 3 biological repeats were used. * p < 0.05, *** p < 0.001, **** p < 0.0001, two-tailed Student's t-test.

Proteome Profiling of Engineered EVs
To better understand the protein composition of the obtained vesicles, proteome analysis of highly purified vesicles was performed. An equal amount of input proteins for three independent preparations for each type of vesicle was trypsinized and analyzed by LC-MS/MS. To evaluate protein abundance, we further used intensity-based absolute quantification (iBAQ) [51]. For this analysis, the sum of peptide intensities referring to one protein is normalized to the number of the theoretically possible peptides. The relative iBAQ then represents the relative proportion of each protein in the sample (Table S2). We first compared the relative iBAQ values of recombinant proteins VSV-G, EPN and NanoLuc. The amount of VSV-G was slightly elevated in EV_p6 and EV_EPN(-) vesicles as compared to the improved vesicles containing the Fos domain (~20% versus~10%). At the same time, the EPN protein, as expected, was not detected in EV_EPN(-) and was significantly decreased in these improved vesicles assembled with the Fos domain (4-10% versus 15-25%). Apparently, the presence of an additional domain partially obstructs EPN expression, nanocage assembly or its packaging into vesicles. However, the amount of the delivered NanoLuc was significantly elevated in the newly designed vesicles (~2-4%), as compared to EV_p6 and EV_EPN(-) vesicles (<0.5%). These data are in good agreement with Western blot and luciferase assay analysis, revealing an increased NanoLuc loading and a decreased amount of VSV-G within vesicles obtained with the newly developed system utilizing Fos-Jun interaction instead of Vpr-p6.
To gain further functional insight into the proteomic cargo in purified EVs, we performed gene ontology (GO) analysis using the DAVID database v6.8. Firstly, we obtained the whole-cell proteomes of all cells, producing the investigated vesicles. Secondly, using whole-cell proteomes as a background for vesicle proteomes, we obtained the statistically significant enrichment only for 5-12 proteins for EV_C, EV_M, EV_C(lenti), EV_EPN(-) and 37 proteins for initial EV_p6 (Table S3). The identified proteins were strongly enriched in extracellular exosome, extracellular matrix, and extracellular space cellular components, further confirming the efficient EV isolation. It should be noted that since the initial EV_p6 had a greater amount of significantly elevated proteins compared to the whole-cell lysate, more GO categories were identified for them. For EV_p6, mainly cellular components related to extracellular exosome, matrix, space, and vesicle were identified, but the adhesion, cytoskeleton, and membrane categories were also significantly enriched.
Next, we selected 135 high confidence proteins that were quantified in all three independent proteomic assays performed on different biological batches for at least one type of vesicle (Table S2). Interestingly, among the enriched GO terms pointed out by this analysis, several biological processes were highlighted. They were related not only to membrane organization, transport, cell-cell adhesion, cytoskeleton and cell movement, but also cell killing, biological regulation, response to stimulus, modification-dependent proteolysis, signal transduction, viral process, developmental process, metabolic process, gene expression and protein folding. Based on the relative number of protein molecules calculated by iBAQ analysis, there was no statistically significant difference between all types of the investigated vesicles, neither in individual protein abundance nor in the distribution of proteins classified into GO categories (Figure 8). At the same time, surprisingly, the majority of abundant proteins were enriched in biological processes, related not only to cell-cell adhesion, but also to biological regulation, cell killing, metabolic process, and protein folding.

Discussion
In this study, we put forward an improved method of protein pharmaceutics delivery enveloped in the engineered EVs. The proposed method aims to provide an easy to modulate, effective and low-cost machinery to envelope protein cargoes into EVs and to deliver them precisely to target cells. We demonstrate that active loading of the engineered EVs applying modified EPN-Fos nanocages increases the efficacy of loading cargo inside EVs and delivering it to target cells.
The development of novel protein delivery systems is at the focus of many research groups. Two major tasks yet to be resolved are to prevent protein degradation and to

Discussion
In this study, we put forward an improved method of protein pharmaceutics delivery enveloped in the engineered EVs. The proposed method aims to provide an easy to modulate, effective and low-cost machinery to envelope protein cargoes into EVs and to deliver them precisely to target cells. We demonstrate that active loading of the engineered EVs applying modified EPN-Fos nanocages increases the efficacy of loading cargo inside EVs and delivering it to target cells.
The development of novel protein delivery systems is at the focus of many research groups. Two major tasks yet to be resolved are to prevent protein degradation and to achieve its specific delivery into the target cells. However, despite a wide palette of protein drug carriers existing today, many of them are not approved for human therapy due to their possible toxicity, while others are used very rarely due to their high cost. Utilizing viral proteins for pharmaceutic envelopes seems to be a highly effective strategy, but it may raise some concerns, largely due to VSV-G immunogenicity. Research investigating the potential to reduce this immunogenicity via pseudotyping with different glycoprotein subtypes is currently underway [52][53][54], but it still remains a challenge [55]. Alternatively, new strategies for the efficient production of designer exosomes, for example, the utilization of recombinant CD63 instead of VSV-G, are proposed [56]. Thus, the future nanocagecontaining EVs could be obtained with a non-toxic and non-immunogenic analog of VSV-G. Another biosafety question is concerned with the Vpr protein used for the active load of cargo inside the previously described nanocage-containing EVs.
Here, we replaced this toxic protein with the leucine zipper domain of the Jun protein, which is present in normal human cells. Although the whole Jun protein is marked as a proto-oncogene [57], we took only a small, 4 kDa part of this protein-its leucine zipper. The investigated engineered EVs deliver their cargo into the cytoplasm of target cells due to VSV-G interaction with low density lipoprotein receptors on the cell surface and subsequent envelope fusion with early endosome membrane [58]. Noncovalent Fos-Jun binding allows for a dissociation of a nanocage and cargo protein in a recipient cell without nanocage disassembling, providing the targeted delivery of protein pharmaceuticals. This property turns out to be useful, for example, for the delivery of transcription factors. Comparison with the recently published results, reporting the development of Gectosomesplit GFP system [44], suggests that the engineered nanocage-containing EVs allow more pharmacological protein molecules loaded into the same vesicle. The EVs developed here demonstrate that the ratio of VSV-G to delivered protein reaches 1:2 in the case of utilizing the lentiviral construct, while the ratio of VSV-G to delivered protein is 6:1 in the case of Gectosomes [44]. The total amount of NanoLuc molecules in every single EPN nanocage was significantly higher in EV_C as compared with the initial EV_p6.
One possible reason for such an improvement might lie in the Fos domain structure. This domain, together with its glycine-linker, forms a "flexible hand", which allows it to direct the binding site of loading protein to the outer surface of EPN and therefore facilitate a more efficient interaction with the Jun domain and cargo protein than the p6 region. Fos-linker flexibility and length allow it not to load all the delivered protein inside the nanocage, but to place it outside. Therefore, in the case of Fos-Jun nanocage-containing EVs, we are not limited with the volume of the EPN nanocage, which is significantly less than the volume of the total vesicle. Still, the possibility to deliver larger proteins utilizing the modified nanocage-containing EVs requires further studies. Meanwhile, the investigated EVs containing EPN nanocages seem to be a promising platform for the targeted delivery of small protein therapeutic agents, such as peptides, required to enhance antigen presentation. If the highly specific delivery of engineered EVs is required, there are two major strategies to achieve it: (i) antibodies, anchored into the EV membrane [59]; (ii) VSV-G replacement with other viral glycoproteins, such as RVG, which will redirect EV tropism [60]; (iii) using the targeting peptides, which are mostly fused with transmembrane proteins [26], and, in the case of 5-9 a.a. peptides, they could be fused with VSV-G molecules without any loss of activity [61]. It should be mentioned that in spite of more efficient cargo loading and protein delivery, the use of lentiviral genetic constructs with two IRES sequences resulted in a significant decrease in EV yield compared to the co-transfection system. In the future, it may be necessary to optimize EV production by constructing lentiviral vectors with several independent internal promoters. The EPN sequence of the late budding domain comprising the p6 domain could also be modified and partially reduced. Originally, this domain recruited ESCRT machinery to catalyze the final membrane fission step required for release from the cell. The utilization of VSV-G can reduce dependence on ESCRT machinery recruiting.
In general, EVs are described as a very heterogeneous population with hundreds of associated markers [62]. We obtained mass spectrometry data for the investigated nanocagecontaining vesicles and gene ontology analysis revealed that the majority of abundant proteins were enriched in a variety of biological processes, such as membrane organization, transport, cell-cell adhesion, cytoskeleton, cell movement, cell killing, biological regulation, response to stimulus and signal transduction. The presence of some proteins like cell-cell adhesion, cytoskeleton and membrane organization are essential for vesicle formation, while other non-structure proteins, like elements with protein folding activity, can be very useful for further delivery applications, since they could be helpful in maintaining the right folding of therapeutic proteins inside the engineered EVs.
Additionally, we show that a uniform distribution of genetic material in the case of a lentiviral construct instead of co-transfection with several plasmids leads to a more effective loading of therapeutics into the engineered EVs. This is an auspicious observation for future clinical applications of EVs. Utilizing the autologous long-living cells for transduction with lentiviral constructs makes it possible to obtain a cell population capable of producing engineered EVs in vivo. Future advances could spur the shift towards personalized medicine and give a chance to more effective therapy of autoimmune diseases and cancer.
Data Availability Statement: The mass spectrometry proteomics data presented in this study are available as Supplementary Tables S2 and S3. All other data are available on request from the corresponding author.