High-yield vesicle-packaged recombinant protein production from E. coli

Summary We describe an innovative system that exports diverse recombinant proteins in membrane-bound vesicles from E. coli. These recombinant vesicles compartmentalize proteins within a micro-environment that enables production of otherwise challenging insoluble, toxic, or disulfide-bond containing proteins from bacteria. The release of vesicle-packaged proteins supports isolation from the culture and allows long-term storage of active protein. This technology results in high yields of vesicle-packaged, functional proteins for efficient downstream processing for a wide range of applications from discovery science to applied biotechnology and medicine.

In brief Eastwood et al. describe a system for export of recombinant proteins in membrane-bound vesicles from E. coli. A simple peptide tag allows high-yield production of functional proteins within vesicle packages that simplify purification and enable long-term storage. This approach allows production of insoluble, toxic, and otherwise challenging proteins from bacteria.

INTRODUCTION
Recombinant protein production has led to a revolution in basic research and biotechnology and biotherapeutic industries and plays a key role in the treatment of a wide range of major diseases. Currently, the majority of commercial recombinant proteins are produced using either bacterial or eukaryotic cell expression systems dependent upon the structural complexity and cell-dependent modifications required to obtain functional protein. The Gram-negative bacteria Escherichia coli is an attractive system for recombinant protein production at both academic and industrial scales. It is not only cheap and easy to culture in batches to high densities, but a wide range of strains, reagents, promoters, and tools have been developed to facilitate the production of functional proteins in E. coli. In addition, the application of synthetic biology strategies is now overcoming limitations commonly associated with the application of post-translational modifications and folding of complex proteins. 1 Here, we describe an innovative expression system that induces packaging of a diverse range of recombinant proteins into membrane vesicles in E. coli. We identify a simple peptide tag that results in high yields of vesicle-packaged functional proteins and allows compartmentalization of otherwise toxic, insoluble, and disulfide bond-containing proteins, as well as extracellular release of vesicles into the media for efficient downstream processing. These released protein-packed vesicles support rapid isolation from the media and also provide a micro-environment for stable, long-term storage of functional recombinant proteins. Thus, this system provides significant benefit for a wide range of applications from discovery science to applied biotechnology and medicine.
MOTIVATION The ability to reprogram a cell to direct the packaging of specific molecules into discrete membrane envelopes is one of the major challenges in the fields of synthetic biology and recombinant protein today. We thus set out to develop a system to allow the export of vesicle-packaged proteins from Escherichia coli. The resultant technology, involving a simple peptide tag, not only simplifies subsequent recombinant protein purification but the controlled packaging into membrane vesicles can be applied to the development of numerous technologies and commercializable products within the biotechnology and medical industries, including generation of recombinant bioreactors, environmental dispersion of biomolecules, and vehicles for drug delivery and vaccination, as well as providing a stable environment for isolation and storage of proteins.

RESULTS AND DISCUSSION
During the development of a fluorescence-based drug screen to identify effectors of alpha-synuclein oligomerisation, 2 we serendipitously discovered that recombinant expression of full-length human a-synuclein (aSyn) in E. coli brought about the release of extracellular aSyn-containing membrane vesicles, frequently containing the bacterial membrane protein OmpA ( Figure 1A). Further analysis revealed that the alpha-helical 3 amino-terminal 38 residues of aSyn are sufficient to bring about the formation and release of OmpA-labeled extracellular membrane-bound vesicles from E. coli cells into the culture media ( Figure 1B). In vitro analysis revealed that this aSyn-derived polypeptide, named here ''vesicle nucleating peptide'' (VNp), interacts with vesicles composed of reconstituted E. coli membrane lipids and subsequently stabilizes its alpha-helical structure 4 (Figure S1). Fluorescence lifetime imaging microscopy (FLIM)fluorescence resonance energy transfer (FRET) revealed that the VNp fusion specifically associates with the inner E. coli membrane in vivo ( Figure S1), which coincides with the formation and release of recombinant VNp-containing vesicles into the growth media ( Figures 1A and 1C-1E). This process occurs without impacting cell growth and so drives large-scale production of vesicles from cells ( Figures 1E and S1) to support isolation of recombinant proteins from growth culture media, as well as from cells harvested upon termination of the culture, thus providing significant savings in both time and resource.
Fusion of sequences encoding VNp to those encoding the monomeric fluorescent protein mNeongreen 5 led to the production and export of large VNp-mNeongreen protein vesicles into the culture media ( Figures 1F and 1G). Immunoelectron microscopy confirmed the exclusive localization of the mNeongreen cargo within the lumen of the vesicles ( Figures 1G and S2). Low-speed centrifugation and subsequent filtration with sterile 0.45 mm polyethersulfone (PES) filters efficiently and effectively isolated the vesicles from bacteria ( Figures 1H, 1I, and S2). Average polydispersity indices from dynamic light scattering (DLS) analysis of isolated VNp-induced vesicles were greater than 1, indicating vesicles with a broad distribution of sizes in the culture media. There was no significant difference in the zeta potential (calculated from peak maxima) between day-old (À10.5 mV) and 4-month-old vesicles (À11.1 mV) and no observable significant loss in vesicle-contained VNp-mNeongreen from vesicles over a 3-month period ( Figure S2). Thus, the isolated vesicles provide a stable environment for effective long-term protein storage of soluble recombinant protein ( Figure S2). The degree of purity of the fusion protein harvested by one-step filtration was determined by mass spectroscopic protein analysis of the isolated vesicles and was found to be sufficient for a very wide range of applications ( Figures 1I and S2) while simultaneously supporting subsequent purification after vesicle sonication where necessary. Together, these data support a model of the VNp fusion interacting with the E. coli membrane and subsequent incorporation into vesicles that release into the culture media ( Figure 1J).
This system provides a simple and attractive mechanism for releasing membrane-packaged recombinant proteins into the media, enabling both enhanced recombinant protein production and subsequent processing. While mNeongreen provided rapid quantification of soluble target protein exported into the media, a wider range of proteins, including a number of model biopharmaceuticals, representing a range of different physical properties and expression challenges (such as membrane binding, disulfide-bond-containing, or otherwise insoluble or toxic proteins; see supplemental information), were used to test the applicability of this technology for the expression of the spectrum of molecules demanded by the life sciences community.
Expression of each protein was tested as VNp, or VNp-mNeongreen amino terminal fusions, and compared to the expression of equivalent non-VNp fusion proteins (Figures 1K-1M and S2; Table 1).
The VNp fusion enhanced the expression of each target protein highly effectively and supports the expression of individual proteins ranging from less than 1 kDa (VNp-His6) to 85 kDa (VNp-mNeongreen-etanercept) in size, as well as protein complexes, as demonstrated by fluorescence from pairs of bimolecular fluorescence complementation (BiFC) VNp fusions 6 within exported vesicles ( Figure S3). Importantly, VNp fusion enhanced the overall yield of each target protein examined, with yields of almost 1 g soluble protein/liter of shaking flask culture obtained in the case for the designed ankyrin repeat protein DARPin Off7 (DARP) ( Table 1). Interestingly, while the addition of the mNeongreen tag was seen to enhance the expression of solubility erythropoietin (EPO), etanercept, and human growth hormone (hGH), the addition of the 25 kDa mNeongreen fluorescent protein tag resulted in a reduction in the overall yield of each model therapeutic protein examined. This is likely to due to a combination of an overall increase in protein size as well as a varying negative effect that mNeongreen can have on the growth of the bacterial cell ( Figure S1). In addition, we observed no   Article ll OPEN ACCESS significant variation in the size or abundance of the VNp-induced vesicles from cultures expressing different VNp fusions; therefore, the differences in abundances (Table 1) are likely due to differences in expression and packaging efficiency within the vesicle. Tobacco etch virus (TEV) protease cleavage of the VNp-mNeongreen tag from VNp-mNeongreen-TEV-DARP and VNp-mNeongreen-TEV-uricase did not impact the solubility of the resultant purified DARP and uricase proteins ( Figure S3), indicating that once expressed, the VNp tag is not necessary for maintaining protein solubility. The VNp expression system was further validated by illustrating its application to larger-volume fermentation cultures ( Figure S3). VNp-DARP expression was induced in E. coli over a 24 h period within 15 L fermentation vessels (see STAR Methods for details). Not only was expression and export of the VNp-DARP fusion sustained over the 24 h period ( Figure S3), yields of VNp-DARP protein greater than 1.4 g/L were reproducibly obtained, which represented 65% of the total protein observed within the cleared medium fraction.
The versatility of the system was further demonstrated by the production of correctly folded (e.g., mNeongreen) and membrane-binding (FGF21) as well as enzymatically active (uricase) proteins ( Figure S3). The vesicle-isolated VNp-uricase was not only as enzymatically active as uricase purified from a cell pellet, but this activity was maintained to a higher degree by VNpuricase stored within isolated vesicles for 2 months at 4 C when compared with purified protein stored at 4 C in buffer over the same period ( Figure S3), highlighting the stable environment the vesicles afford their protein cargo.
The VNp fusion allows production of soluble proteins that are otherwise insoluble or reduce the viability of bacterial cells (e.g., DNase, etanercept, EPO, and hGH) (Table 1; Figure S3). In the case of the disulfide-bond-containing proteins etanercept and hGH, 7,8 the majority of the soluble recombinant protein remained within the cell (Table 1). Electron microscopy (EM) data show that fusing VNp-mNG to etanercept, an anti-inflammatory therapeutic consisting of a fusion between a tumor necrosis factor and immunoglobulin G1 (IgG1), impacts VNp remodeling of the inner membrane to induce VNp-fusion-containing internalized cytosolic membrane structures (Figures 2A and 2B). This tumor necrosis factor (TNF)-IgG1 therapeutic fusion was not only dimeric (disulfide-bond dependent) but also exhibited appropriate ligand-binding properties when isolated from VNpinduced cytosolic vesicles. The ability to bind protein A was maintained upon TEV protease-dependent proteolytic removal of the VNp-mNeongreen fusion ( Figures 2C-2E). Similarly, proteolytic cleavage of VNp-mNG from VNp-mNG-TEV-DARP and VNp-mNG-TEV-uricase did not impact the solubility of DARP or uricase ( Figure 2F), indicating that VNp is not required to maintain solubility once expressed.
To explore whether dimerization was sufficient to induce internalization of a VNp-fusion protein, stable alpha-helical VNp  (Figures 3 and S3). Expression of the VNp-LZ fusion induced formation of cytosolic VNp-LZ-fusion-filled vesicular structures, which form from the CydAB 10 -containing inner membrane (Figures 3B-3D; Video S1). While the precise molecular basis is not yet understood, these data illustrate that dimerized VNp fusions promote inward, rather than outward, curvature of the bacterial membrane to provide an attractive method for generating recombinant proteins within cytosolic membrane-bound structures to further facilitate the production of disulfide-bond-containing and otherwise insoluble or toxic proteins from E. coli. Spurred on by the success of this approach, we asked whether simple modifications to the VNp amino acid sequence to modulate the ability to form VNp-fusion-containing vesicles would enhance the exported protein yields. We therefore systematically tested equivalent VNp sequences from the band g-synuclein isoforms (Table 1) as well as a series of constructs generated through modifying charges and side-chain lengths of targeted residues along the helix surface and found that we could not only enhance vesicular export over a wide range of culture temperatures (VNp6) but could also reduce the size of the VNp to 20 residues in length (VNp15) to enhance the export of the target model biopharmaceuticals DARP and stefin A at yields of more than 2.5 g soluble recombinant protein/liter of bacterial flask culture (Table 1). These high yields were reproducible in both academic and industrial environments.
The VNp system exhibits flexibility, as vesicle-packaged proteins can be generated in different E. coli strains (e.g., BL21, l, JM109, and K12 lineages; Figure S4), making it perfectly suited for the production of synthetic proteins with modifications supported by specialist E. coli hosts. For example, the VNp system functions in W3110 cells, which allows generation of recombinant-protein-filled vesicles with a reduced immunogenic response. 11 VNp fusions can be expressed from a variety of plasmids (including pUC19-and pBR322-based derivatives), and modulated VNp-fusion expression can be driven from diverse promoters (e.g., T7, rhamnose) and induction levels ( Figure S4; Table 1), making this a truly versatile system.
Unlike native outer membrane vesicles that occur naturally in E. coli, which form spontaneously in the absence of recombinant protein expression, 12-14 the VNp system described here nucleates vesicle formation through interactions with the inner membrane. In addition, while recombinant proteins are absent from native vesicles released into the media when expressed in E. coli cells, the VNp system allows a simple tagging mechanism for targeted recombinant proteins into vesicles (Table 1). This simple peptide fusion increases yields and simplifies downstream processing of a wide range of recombinant proteins from E. coli. Importantly, the ease with which otherwise insoluble or toxic proteins can be isolated in milligram or gram quantities suggests that this approach is an attractive starting point for the expression of any recombinant protein of interest. It should be noted that the isolation of protein from VNp-fusion-induced  vesicles is unlikely to provide a route to avoid endotoxin entirely, as proteins are always wrapped in endotoxin during normal secretion processes or homogenizations. Therefore, depending upon the downstream application, the enriched VNp-derived proteins may require further purification. The method has the potential to allow continuous release of protein from extended period cultures in appropriately genetically modified stable expression strains. Another beneficial aspect of this innovation is the stability of proteins and preservation of enzymatic activity when the vesicles are maintained at 4 C. As the use of this system is more broadly adopted and further enhancements and adaptations emerge, its impact can be anticipated to be highly significant. We therefore predict rapid adoption of this versatile system into a wide range of downstream processes and applications.

Limitations of the study
While there is no guarantee this system will enhance production for all proteins, its use resulted in significant increase in yield and solubility for each of the proteins we have tested to date (n > 60). While some of the recombinant-protein-filled vesicles remain cytosolic, it is our experience that these tend to be either dimeric, disulfide-bond-containing, or toxic proteins, which may reflect differences in localized membrane conformation and/or membrane affinity. However, in each of these cases, we also observed enhanced expression and/or functionality of the subsequently expressed proteins. The model depicted in Figure 1J represents a model of how the system works based on current biochemical and imaging data presented in this study. Elucidating the membrane composition of the vesicles and further in vitro studies will provide insight into the precise mechanism underlying the formation of the recombinant-protein-filled vesicles described here.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled upon reasonable request by the lead contact Dan Mulvihill (d.p.mulvihill@kent.ac.uk).

Materials availability
Plasmids generated in this study have been deposited to Addgene. Plasmid ID #s 182386-182425.
Data and code availability d All the raw datasets generated during this study have been deposited at Kent Data Repository and are publicly available as of the date of publication. Microscopy data reported in this study will be shared by the lead contact upon request.
d This paper does not report original code. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
Bacterial cell culture and protein induction All bacterial cells were cultured at 37 C using LB (10 g Tryptone; 10 g NaCl; 5 g Yeast Extract (per litre)) and TB (12 g Tryptone; 24 g Yeast Extract; 4 mL 10% glycerol; 17 mM KH 2 PO4 72 mM K 2 HPO 4 (per litre) media. 5 mL LB starters from fresh bacterial transformations were cultured at 37 C to saturation and used to inoculate 100-500 mL volume TB media, flask cultures that were incubated overnight at 37 C with 200 rpm orbital shaking. Recombinant protein expression from the T7 promoter was induced by addition of IPTG to a final concentration of 20 mg/mL (except etanercept where 10 mg/mL was used) once the culture had reached an OD 600 of 0.8-1.0). Growth curves were generated from 96 well plate cultures, prepared from late log-phase cultures, diluted into fresh media to an OD 600 of 0.1 nm at the start of the growth analysis experiment. OD 600 absorbance values were obtained using a Thermo Scientific Multiscan Go 1510-0318C plate reader and recorded using the SkanIt Software 4.0. at OD 600 values were taken every 15 minutes for the duration of the experiment, and growth curves generated from averages of 4 individual biological repeats.

METHOD DETAILS
Soluble protein extracts Cell pellets from 50 mL of culture were resuspended in 5 mL of soluble extract buffer (20 mM TRIS, 500 mM NaCl, pH 8.0), sonicated for a total of 2 min (6 3 20 sec pulses), and cell debris removed by centrifugation at 18,000 rpm (4 C) for 30 min. Target protein concentration was determined using fluorescence of mNeonGreen fusion or gel densitometry. Both techniques were compared directly on the same samples to determine equivalence.

Recombinant vesicle isolation
Vesicles were isolated directly from bacterial cell cultures by passing the culture through a 0.45 mm PES filter. Typical purity and concentration from equivalent volume of culture and filter flow through are shown in Figure 1. Exclusion of viable cells from the vesicle containing filtrate was routinely tested by plating onto LB plates lacking antibiotics and incubating overnight at 37 C (example shown in Figure S2).

Protein concentration determination
Fluorescence scan was used to determine concentration of mNeongreen fusion proteins in vesicle containing media and soluble protein extracts. Absorbance was measured at 506 nm using a Varian Cary 50 Bio UV-Vis spectrophotometer, with measurements from an equivalent empty vector culture used for baseline correction, and concentration determined using an extinction coefficient of 116,000 M À1 cm À1 . Concentration of non-mNeongreen labelled proteins was determined by gel densitometry analysis of triplicate samples run alongside BSA loading standards on Coomassie stained SDS-PAGE gels. Gels were scanned and analysed using Image J software. Concentration was determined by both UV and densitometry for three independent VNp-mNeongreen samples to confirm parity between techniques. Average yields in Figure 1 & Table S1 were calculated (mg target protein/litre culture) from a minimum of 3 independent biological repeats from cultures of BL21 DE3 E. coli cells grown in TB media.

Protein isolation from vesicles
Purified VNp induced vesicles were resuspended in ice cold 1xPBS before being sonicated to disrupt vesicle membrane, and release the VNp-fusion protein. In order to further purify carboxyl His 6 tagged recombinant VNp-fusion protein (all recombinant proteins expressed during this study contain carboxyl-terminal His 6 affinity tags), this solution was then mixed in a 1 in 5 dilution of 5 x binding buffer (250 mM TRIS 2.5 M NaCl 5% Triton-X 50 mM Imidazole pH 7.8) before passing over a Ni 2+ -agarose resin gravity column. Cytosolic recombinant protein was purified by passing soluble protein extracts (supplemented with Imidazole to 20 mM) over the Ni 2+ -agarose resin gravity column. In both cases matrix bound His-tagged protein was washed, eluted (using imidazole), and dialysed into appropriate storage or assay buffer. Protein identity and amino-terminal acetylation of isolated proteins was confirmed by electrospray mass-spectroscopy.

Circular dichroism (CD)
Measurements were made in 2 mm quartz cuvettes using a Jasco 715 spectropolarimeter. VNp protein and 100 nm extruded vesicles were diluted in CD buffer (10 mM potassium phosphate, 5 mM MgCl 2 pH 7.0) to a concentration of 0.4 mg/mL and 0.2 respectively. Broad negative peaks at 208 and 222 nm and a positive peak at < 200 nm are consistent with an a-helical structure.
Electrospray LC-MS of proteins Electrospray mass spectra were recorded on a Bruker micrOTOF-Q II mass spectrometer. Samples were desalted on-line by reverse-phase HPLC on a Phenomenex Jupiter C4 column (5 mm, 300 Ǻ, 2.0 mm 3 50 mm) running on an Agilent 1100 HPLC system at a flow rate of 0.2 mL/min using a short water, acetonitrile, 0.05% trifluoroacetic acid gradient. The eluant was monitored at 214 nm & 280 nm and then directed into the electrospray source, operating in positive ion mode, at 4.5 kV and mass spectra recorded from 500-3,000 m/z. Data was analysed and deconvoluted to give uncharged protein masses with Bruker's Compass Data Analysis software.
In-gel tryptic digest and proteomic analysis of recombinant vesicles Sample of purified VNp-DARP induced vesicles (shown in Figure S5B) were run on SDS-PAGE, which was subsequently coomassie stained, and the whole sample lane cut out, cut into small pieces, which were subsequently transferred to a 1.5 mL microfuge tube and stored in distilled water at 4 C until processing. The gel particles were washed with 150 mL of freshly made 50 mM NH 4 HCO 3 : acetonitrile (1:1 ratio) for 15 mins. Liquid was removed and gel fragments resuspended in 150 mL acetonitrile for 15 mins, before liquid was again removed, and gel pieces were resuspended in 100 mL of 10 mM DTT in 50 mM NH 4 HCO 3 , and incubated for 30 min at 56 C. Gel pieces were centrifuged, and excess liquid removed before incubating for 1 min with 100 mL of acetonitrile, which was again removed and gel fragments were suspended in 100 mL of 55 mM chloroacetamide in 50 mM NH 4 HCO 3 and incubated for 20 min at room temp in the dark. Pellets were then centrifuged, the chloroacetamide solution was removed. Gel pieces were subject to subsequent 15 min washes in 150 mL of 50 mM NH 4 HCO 3 :acetonitrile (1:1), and then150 mL of 50 mM NH 4 HCO 3 for 15 min, and liquid was removed by centrifugation after each wash. Gel pieces were then washed for 15 mins with 200 mL of acetonitrile, and then rehydrated in 50 mL of digestion buffer (12.5 mM NH 4 HCO 3 , 10% acetonitrile) containing 5 ng/mL of trypsin, which was left overnight at room temperature. Upon completion of digestion, 15 mL acetonitrile was added to the sample, where was then sonicate in an ultrasound bath for 15 mins. Gel fragments were isolated by centrifugation and the supernatant collected in a fresh 0.5 mL microfuge tube (A). The gel fragment pellet was resuspended in 30 mL 50% acetonitrile with 5% formic acid, and sonicate in an ultrasound bath for 15 mins, and pellet again isolated by centrifugation and supernatant collected in a fresh 0.5 mL microfuge tube (B). Contents of tube A and B were combined, vacuum dried, and subsequently resuspended in 20 mL of 5% acetonitrile, 0.1% TFA. Samples were run through Pierce C18 Spin Tips and analysis by nano-LCMS.
Gel filtration assay 500 mL of protein samples were loaded to a Superdex 200 Increase 10/300 GL size-exclusion column (GE Healthcare Life Sciences) equilibrated at room temperature in PBS and run at 0.75 mL/min flow rate. Eluted proteins were measured by Viscotek Sec-Mals 9 and Viscotek RI detector VE3580 (Malvern Panalytical).
Lipid binding assay Affinity of VNp for E. coli membrane lipids was established using a thermal shift fluorescence binding assay adapted from. 15 Equivalent assay samples, made up of: 65 mL 3 mg/mL of VNp-mNeongreen, 65 mL 1 mM of 100 nm extruded vesicles composed of the lipid mixture to be tested; 15 mL, 10%OGP; and 5 mL 20 mM Tris-HCl pH 7.0, were prepared in PCR tubes and held at the defined temperature in a gradient PCR machine for 10 minutes. Samples were centrifuged at 18,000 xg, and supernatant fluorescence was determined in black 96 well plates (BRAND, Germany) using a BMG Clariostar (BMG Labtech). Fluorescence readings were normalised and used to create a melting curve, where the melting temperature (Tm) was determined using Origin software (OriginLab). The final Tm value was an average (± s.d) calculated from three independent sample repeats.
Uricase assay 500 mL of 100 mM Tris pH 8.5 with 200 mM Uric acid was placed in a cuvette and OD 293 measurements were taken over for 4 or 5 minutes. Subsequently either 500 mL of 4.5 mg/mL purified VNp2-Uricase (dialysed into 0.1 M Tris pH 8.5) or dialysis buffer alone was added to the cuvette and OD 293 measurements taken for 25 mins. (Adapted from 16 ).
Widefield fluorescence microscopy Cells were mounted onto coverslips under <1 mm thick circular LB-agarose(2%) pads, and attached with appropriate spacers onto glass slides, before being visualised on an inverted microscope. 17 All live cell imaging for each sample was completed within 30 mins of mounting cells onto coverslips. Structured Illumination Microscopy (SIM) was undertaken using a Zeiss Elyra PS 1 microscope with a 100x NA 1.46 oil immersion objective lens (Zeiss a Plan-Apochromat) as described previously. 18,19 Briefly, cells were mounted under thin LB-agarose pads onto high precision No.1.5 coverslips (Zeiss, Jenna, Germany). 488 nm and 561 nm laser were used to illuminate mNeongreen and