Recombinant MBP-pσ1 expressed in soybean seeds delays onset and reduces developing disease in an animal model of multiple sclerosis

Abstract

It is estimated that multiple sclerosis (MS) affects over 2.8 million people worldwide, with a prevalence that is expected to continue growing over time. Unfortunately, there is no cure for this autoimmune disease. For several decades, antigen-specific treatments have been used in animal models of experimental autoimmune encephalomyelitis (EAE) to demonstrate their potential for suppressing autoimmune responses. Successes with preventing and limiting ongoing MS disease have been documented using a wide variety of myelin proteins, peptides, autoantigen-conjugates, and mimics when administered by various routes. While those successes were not translatable in the clinic, we have learned a great deal about the roadblocks and hurdles that must be addressed if such therapies are to be useful. Reovirus sigma1 protein (pσ1) is an attachment protein that allows the virus to target M cells with high affinity. Previous studies showed that autoantigens tethered to pσ1 delivered potent tolerogenic signals and diminished autoimmunity following therapeutic intervention. In this proof-of-concept study, we expressed a model multi-epitope autoantigen (human myelin basic protein, MBP) fused to pσ1 in soybean seeds. The expression of chimeric MBP-pσ1 was stable over multiple generations and formed the necessary multimeric structures required for binding to target cells. When administered to SJL mice prophylactically as an oral therapeutic, soymilk formulations containing MBP-pσ1 delayed the onset of clinical EAE and significantly reduced developing disease. These results demonstrate the practicality of soybean as a host for producing and formulating immune-modulating therapies to treat autoimmune diseases.

Introduction

Multiple sclerosis (MS) is an inflammatory, demyelinating disease of the central nervous system (CNS) that results from the breakdown of immune tolerance (Kammona and Kiparissides 2020). MS is thought to be initiated by T-cells that recognize myelin antigens infiltrating the CNS. Once activated T-cells enter the CNS, local antigen presenting cells process and present epitopes of myelin protein synthesized within the CNS. The process of myelin-specific T-cell re-activation results in the recruitment of inflammatory leukocytes and ultimately leads to the destruction of myelin—the protective insulation that wraps around nerve cells in the brain and spinal cord. The major antigens associated with MS disease include myelin basic protein (MBP), proteolipid protein (PLP), and myelin oligodendrocyte protein (MOG). These three proteins comprise ∼75% of the protein composition of myelin (Kammona and Kiparissides 2020).

To date, there is no cure for MS or any of the known autoimmune diseases (Kammona and Kiparissides 2020). The standard treatment for autoimmune disorders typically includes immunosuppressive agents and biologics intended to reduce inflammation and suppress immune responses to self-antigens (Moorman et al. 2021). While such agents can relieve symptoms and reduce the frequency of outbreaks associated with autoimmune diseases, they do not address the root cause of autoimmunity, which is the loss of immune tolerance. Over the past two decades, novel approaches have emerged to establish immune tolerance by inhibiting pathogenic autoantigen-specific immune responses while leaving the rest of the immune system intact. Some of these approaches include the development of tolerogenic vaccines (e.g., protein and peptide-based immunotherapies designed to establish autoantigen-specific immune tolerance), T cell therapies (e.g., utilizing Tregs to establish active immune tolerance or T cells to delete pathogenic immune cells), and IL-2 therapies (e.g., to expand immunosuppressive Tregs). In preclinical studies involving tolerogenic vaccines, administration of myelin peptides by various routes suppressed clinical EAE and offered promise for this type of approach (Moorman et al. 2021). However, when myelin peptides were delivered along with CFA adjuvant in a proinflammatory environment, they instead provoked encephalitogenic priming and induction of EAE. To help overcome this issue, tolerogenic vaccines have been designed with carriers (e.g., monoclonal antibodies, cytokines, cells, adhesion proteins, etc.) to target autoantigens to specific cells within the immune repertoire and concomitantly address issues with protein/peptide stability, half-life, and bioavailability. One such approach took advantage of the adhesion properties of type 3 reovirus protein sigma 1 (pσ1) that binds to microfold (M) cells present on Peyer’s patches in MALT (Rynda et al. 2008). Using animal models of autoimmune diseases, Pascual and colleagues showed that recombinant autoantigens tethered to pσ1 diminished EAE following therapeutic intervention (Rynda et al. 2010, 2008; Rynda-Apple et al. 2011; Suzuki et al. 2008), demonstrating potential utility for pσ1-immunotherapies.

Despite recent successes with protein-based immunotherapies in animal models of autoimmune disease, no therapies have been translated in the clinic. Some of the hurdles that have slowed the development of such therapies include the inability to deliver multi-epitope autoantigens in a tolerizing context and a robust platform that can support the production and formulation of immunotherapeutics at a low cost (Bluestone et al. 2010). In light of the successes with mucosal tolerization therapies reported by Pascual and colleagues (Rynda et al. 2010, 2008; Rynda-Apple et al. 2011; Suzuki et al. 2008), it seems possible that autoantigens fused to Sigma1 could represent a feasible technology for tolerization if practical platforms could be identified for cost-effective production and formulation of such immunotherapies. Unfortunately, most existing recombinant biopharmaceuticals are produced in mammalian cell culture systems, and these systems have high production costs and expensive media and culture condition requirements (Shanmugaraj et al. 2020). However, plants represent a practical alternative for the production of recombinant biotherapeutics and offer advantages of low cost, rapidity, scalability, safety, and ability to assemble complex proteins with eukaryotic-like post-translational modifications (Chen and Davis 2016; Schillberg et al. 2019; Shanmugaraj et al. 2020). Furthermore, plants may be used as bioreactors for the production of recombinant proteins that can be purified, or for direct oral delivery when biotherapeutics are targeted to edible tissues. Various plant hosts have been used to express vaccine antigens, antibodies, bioactive peptides, diagnostic proteins, and oral tolerogens (Merlin et al. 2017; Schillberg et al. 2019; Shanmugaraj et al. 2020). Many of these plant-derived products were shown to be safe and efficacious when evaluated in animal models (Merlin et al. 2017), and numerous other candidates have advanced to clinical trials (Maharjan and Choe 2021; Shim et al. 2019). The few plant-derived products that have been commercialized (Takeyama et al. 2015) or approved for use in humans (Duong and Vogel 2022; Fox 2012) represent purified forms of recombinant proteins. To date, no crude forms of plant-derived therapeutics have been approved for oral delivery in humans.

Soybeans represent an ideal system for the production of oral therapeutics and have numerous advantages over other plant-based and traditional expression hosts. For example, soybeans contain ∼40% protein by mass (Liu 1997) and represent the greatest known natural source of protein, making them an ideal bioreactor for the production of recombinant proteins. In addition, soybeans can express large and complex proteins that are recalcitrant to expression in traditional systems (Powell et al. 2011), directly addressing hurdles with myelin autoantigens that are difficult to express. Soy-based products are generally recognized as safe and well-tolerated by most humans and animals, allowing for delivery via oral routes (e.g., soymilk) and the potential to eliminate significant costs associated with downstream purification. Soy-derived therapeutics could also be stored for years as unprocessed seed powder, defatted soy flour, or spray-dried soy protein powder formulations, addressing issues with cold-chain requirements (Oakes et al. 2009). Simply stated, soybean represents a robust and practical host that could support the production and formulation of biotherapeutics, and address many of the hurdles that have slowed the translation of novel immunotherapies in the clinic.

Building on the findings from Pascual and colleagues, we set out to determine whether soybean could be used as a host for the production of efficacious autoantigen-pσ1 chimeras that could be formulated as crude soy formulations and delivered in a tolerizing context. We chose full-length MBP as a model autoantigen for this study since it represents a major multi-epitope protein associated with MS disease. MBP has a long history of use in human MS clinical trials and was shown to be safe when delivered as a systemic (Campbell et al. 1973; Gonsette et al. 1977) or oral (Fukaura et al. 1996; Weiner et al. 1993) antigen-specific therapeutic. Unfortunately, neither purified nor recombinant forms of full-length hMBP are currently available in the quantities needed for practical therapies, as evidenced by the use of bovine-derived MBP in clinical trials involving MS patients (Fukaura et al. 1996; Weiner et al. 1993). Our results demonstrate that soy-derived MBP-pσ1 stably accumulates in transgenic soybean seeds and forms functional trimeric MBP-pσ1 complexes that bind to epithelial cells in cell-binding assays. We also demonstrate that soymilk formulations prepared from seeds expressing MBP-pσ1 are efficacious in delaying the onset of clinical EAE and limiting developing disease in SJL mice when administered as an oral immunotherapy. Together, these results underscore the versatility of soybean as a host for the production and formulation of novel autoantigens that may aid in the future development of antigen-specific immunotherapies.

Materials and methods

Construction of soybean transformation vectors

Sequences encoding the Glycine max Gy1 promoter and signal peptide (SP) (Sims and Goldberg 1989), hMBP, and reovirus type 3 sigma1 protein (pσ1) were synthesized by commercial vendors (Geneart and ATUM). The promoter and SP fragments were designed with nucleotide sequences identical to those present in soybean, while the sequences of hMBP and pσ1 open reading frames were codon-optimized for expression in soy. Synthesized fragments were engineered with unique restriction enzyme cleavage sites to facilitate assembly of the fragments and cloning into pTN200 and pTF101 binary vector backbones. To serve as a base vector for the construction of soybean transformation vectors, the previously-constructed pTN200-hTG plasmid (Powell et al. 2011) was digested with the restriction enzymes HindIII (5′) and NcoI (3′) and ligated with a fragment containing the Gy1 promoter (also digested with HindIII and NcoI). This new plasmid was then digested with NcoI (5′) and XbaI (3′) and joined to the synthetic SP-MBP fragment (also digested with BspHI and XbaI). That plasmid was, in turn, digested with MluI (5′) and XbaI (3′) and ligated with the synthetic pσ1 fragment (also digested with MluI and XbaI) to create a transformation vector referred to as pTN200-MBP-pσ1. A second binary vector was generated by digesting pTN200-MBP-pσ1 with HindIII, and ligating the entire MBP-pσ1 GOI expression cassette into pTF101 (also digested with HindIII) to create pTF101-MBP-pσ1. Restriction enzyme analysis was used to select clones in which the GOI and selectable marker cassettes were arranged in a head-to-tail orientation. The nucleotide sequences of engineered expression cassettes were verified by double-stranded sequencing.

Soybean transformation

Soybean transformation was carried out using Agrobacterium EHA101 strains, glufosinate selection, and the half-seed transformation method previously described (Paz et al. 2006).

Structure predictions

A protein structural homology model of full-length pσ1 was generated using previously solved crystal structures of the pσ1 head (PDB 1KKE) and tail (PDB 6GAP) (Dietrich et al. 2018) and aligned using Chimera X (UCSF). To obtain a protein structural model of MBP-pσ1, the tail domain pσ1 structure (PDB 6GAP), a 6 residue linker sequence, and predicted structure of Homo sapiens MBP (AlphaFold) were combined using Robetta Comparative Modeling (Raman et al. 2009; Song et al. 2013). The resulting MBP-pσ1 structure was further aligned with the full length pσ1 structure using Chimera X. A model of the MBP-pσ1 trimer was generated by mirroring and aligning each monomer based on the pσ1 “head” trimerization interface described by Chappell et al. (Chappell et al. 2002).

Protein extraction, electrophoresis, and western blot analysis

Soluble protein was extracted from seed tissue (cotyledon slivers or ground seed powder) using 1X PBS buffer. Following sonication, soluble protein was separated from insoluble debris by centrifugation (16,100×g, 4°C, 15 min). Supernates were transferred to a fresh tube, and a second centrifugation (16,100×g, 4°C, 15 min) step was carried out. The concentration of extracted seed protein was determined using a Bradford assay with BSA serving as the quantification standard.

For separation of proteins under native conditions, seed protein was mixed with 3X non-denaturing sample buffer (240 mM Tris (pH 6.8), 0.06% bromophenol blue dye, 30% glycerol) and separated in Tris-glycine 4–15% gradient gels (BioRad, Hercules, CA, USA) or native gels prepared in-house. For protein separation under denaturing conditions (SDS-PAGE), seed protein was mixed with 3X denaturing sample buffer (240 mM Tris (pH 6.8), 6% SDS, 0.06% bromophenol blue dye, 30% glycerol) and separated in 7% acrylamide gels using standard SDS-PAGE buffers and conditions. Following electrophoresis, gels were equilibrated in 10 mM N-cyclohexyl-3-aminopropanesulfonic acid buffer containing 10% methanol, and protein was transferred to Immobilon-P membrane (Milliporee, Burlington, MA, USA). Transferred protein was visualized with Coomassie blue dye prior to imaging and western blot probing.

For western blot analysis, membranes were incubated with blocking buffer (PBS containing 5% nonfat dry milk for 1 h) prior to incubation with primary antibody (e.g., rabbit anti-pσ1 or chicken anti-hMBP, 1 : 5,000) and the appropriate HRP-conjugated secondary antibody (e.g., goat anti-rabbit IgG or goat anti-chicken IgG, 1 : 5,000). Blots were washed 3 times with PBST buffer (PBS containing 0.05% Tween 20) for 10 min between incubations and prior to detection. Immunodetection was carried out using the SuperSignal West Pico substrate kit (Thermo Fisher Scientific, Waltham, MA, USA), and images were captured on X-ray film or the Chemidoc MP (BioRad, Hercules, CA, USA) imaging system.

ELISAs

Approximately 12 T1 seeds derived from each transgenic event were pooled and ground into seed powder using a small coffee grinder. Soy protein was extracted in PBS buffer and quantified as described above. Soluble seed protein was diluted to 0.01 mg ml^－1 in 0.1 M sodium bicarbonate buffer, and 0.1 ml diluted protein was coated onto wells of 96-well ELISA plates. Unbound protein was removed, and wells were washed, blocked with PBS containing 3% BSA, and incubated in succession with primary (rabbit anti-pσ1, 1 : 10,000) and secondary (HRP-conjugated goat anti-rabbit IgG, 1 : 5,000) antibodies. Signal was developed with TMB (3,3′,5,5′-tetramethylbenzidine) One Component reagent (SurModics Inc., Eden Prairie, MN, USA), and reactions were stopped with one-third volume of 0.6 N sulfuric acid. Absorbance readings (405 nm) were obtained using a BioRad iMark microplate reader.

In vitro cell binding assay and confocal microscopy

L929 cells obtained from ATTC (CCL-1, NCTC clone 929) were maintained in RPMI medium 1640 (Gibco) supplemented with 0.1875% sodium bicarbonate (Corning), 0.27% glucose (Corning Inc., Corning, NY, USA), 0.1 mM sodium pyruvate (Cellgro), 25 µg ml^－1 gentamycin (Gibco) and 10% fetal bovine serum (Gemini Bio, West Sacremento, CA, USA). For binding studies, cells were plated onto Lab-Tek 8-well (Permanox coated) chamber slides and incubated at 37°C with 5% CO₂ until 50–80% confluent. Cells were fixed with 4% paraformaldehyde (30 min at 23°C), washed with PBS, and blocked with PBS containing 3% BSA (45 min at 23°C). Soy protein prepared from Wild Type (WT) seeds or transgenic seeds expressing MBP-pσ1 (line 81-15-B) was diluted to 1 mg ml^－1 in PBS buffer supplemented with 1% BSA and incubated with fixed cells for 1 h at 23°C. After washing with PBS, cells were incubated with primary antibody (rabbit anti-pσ1, diluted 1 : 1,000 in PBS containing 1% BSA) for 1 h at 23°C. After washing, cells were incubated with a secondary antibody (goat anti-rabbit IgG-Alexa488, diluted 1 : 1,000 in PBS containing 1% BSA) for 1 h at 23°C in the dark. After a final wash, the complexes were mounted in Flurogel II with DAPI (Electron Microscopy Sciences) overnight at 4°C. Cells were imaged on an Olympus inverted microscope using a 60x water immersion lens and DAPI 405-Alexa Fluor 488-TD1 635 filter set. Image analysis and preparation were carried out using FV10-ASW v.4.2 Viewer software (Olympus, Shinjuku City, Tokyo). Only equal linear adjustments were made to enhance overall contrast, and figures were assembled using Microsoft PowerPoint.

pH manipulation of soymilk formulations

Total seed protein was extracted using a ratio of 20 volumes of PBS buffer/gram seed powder and clarified as described above. Extracted soluble protein was transferred to a small beaker along with a stir bar and pH probe, and dilute HCl was added dropwise with gentle stirring until the desired pH of the solution was achieved. Samples were collected at various times throughout the pH reduction, and insoluble protein was removed by centrifugation. The pH of clarified supernatants was neutralized by the addition of one-tenth volume of 1 M Tris-HCl (pH 8.0), and protein was quantified as described above.

Formulation of oral therapeutics

Formulations for animal studies were prepared simultaneously from WT seeds or transgenic (line 91-3-A) seeds. Total seed protein was extracted from ground seed powder (10 volumes of PBS extraction buffer/gram of soy powder) using 0.05 M Tris buffer (pH 7.4) and sonication. Insoluble debris was removed by passage over cheesecloth followed by centrifugation. HCl was added to lower the pH of the protein solution to 4.6, and insoluble protein was removed by centrifugation. The pH of the soluble protein extract was readjusted to pH 7.4 using dilute NaOH, and extracts were again clarified by centrifugation and passage over a 0.45-micron filter. The extract was concentrated in an Amicon Ultra centrifugal filter units with a 50 kDa MWCO filter.

For quantification of MBP-pσ1 in formulations, known volumes of formulated protein and known masses of purified hMBP were separated in 5% native PAGE gels. Formulated protein was loaded into gels first, and when the dye in the sample buffer reached the bottom of the gel (∼1 h at 120 V), known amounts of hMBP standards were loaded into adjacent wells, and electrophoresis was continued for an additional 15 min. Separated proteins were transferred to membranes, and a chicken anti-MBP antibody was used for the detection of hMBP and hMBP-pσ1 complexes. Densitometry was carried out on scans of western blots using ImageJ software (Schneider et al. 2012), and the resulting data were imported into GraphPad Prism 7. A linear curve was generated by plotting densitometer values (Y-axis) and known masses (X-axis) associated with the hMBP standards, and a best-fit line was generated (typical r²>0.95). The concentrations of MBP-pσ1 in known volumes of formulated protein were determined by extrapolation from the generated curve.

EAE experiments

All animal experiments were conducted under protocols that were reviewed and approved by the University of North Carolina at Charlotte IACUC. Female SJL mice (6 weeks old) were obtained from Charles River Laboratories and allowed to acclimate ∼2 weeks prior to initiation of studies. Induction of EAE was carried out as previously described (Miller et al. 2010). On day 0, groups of mice (n=10/group) were immunized subcutaneously with 0.1 mg of purified hMBP protein emulsified in Freund’s Complete Adjuvant containing heat-inactivated M. tuberculosis at a final concentration of 4 mg ml^－1. As part of the immunization schedule, animals also received 400 ng pertussis toxin intraperitoneally on days 0 and 2. On days 1, 4, and 7, mice were gavaged with 0.1 ml of a ∼75 mg ml^－1 soy formulations prepared from either WT seeds or transgenic seeds. The amount of MBP-pσ1 in transgenic soymilk formulations was estimated to be ∼50 µg/dose. Animals were monitored by the same pair of investigators for EAE using a standard clinical scoring system of 0–4, where grade 0 represents no symptoms; grade 1 represents mildly decreased activity, weak grip with fatigability; grade 2 represents weakness, hunched posture at rest, and/or tremor; grade 3 represents severe generalized limb and body weakness, and grade 4 represents moribund (Miller et al. 2010).

Statistics

For the EAE animal experiments, data are presented as the mean clinical score±SE. Significance between clinical scores in the two groups was determined by the Mann–Whitney U test using GraphPad Prism software (version 9.4.1).

Results

Construction of the MBP-pσ1 expression cassette and in silico structure predictions

Sequences encoding the open reading frames of human myelin basic protein and reovirus pσ1 were optimized for expression in soybeans in an effort to increase stable accumulation in seeds. Synthesized fragments were joined to create a MBP-pσ1 open reading frame and ligated downstream of the soybean 11S Gy1 seed-specific promoter that targets expression to maturing seeds. The MBP-pσ1 gene of expression (GOI) cassette was cloned adjacent to a bar selectable marker cassette in the binary vectors pTN200 and pTF101 (Figure 1A). While pTN200 and pTF101 both utilize the bar gene for selection during transformation, the vectors utilize different regulatory elements to drive bar transcription and translation. For example, a nopaline synthase promoter drives constitutive expression of bar in the pTN200 selectable marker expression cassette, while the double-enhanced 35S promoter and tobacco etch virus translational enhancer drive expression of bar in pTF101 (Figure 1A). Both binary vectors are routinely used for selection during soybean transformation, and the use of both vectors in this study allowed us to determine whether substantial differences in GOI expression were correlated with vector choice. Prior to soybean transformation, the integrity of the expression cassettes was verified by double-stranded sequencing, and no errors were detected. The open reading frame of MBP-pσ1 is shown in Figure 1B.

Figure 1. Design of MBP-pσ1 and soybean transformation vectors. (A) Binary vectors used for soybean transformation. The pTN200 and pTF101 vector backbones were used for assembly of the MBP-pσ1 expression cassette, and the resulting binary vectors were used for soybean transformation. RB, right border; LB, left border; P_11S, soybean Gy1 promoter; SP_11S, soybean Gy1 signal peptide; mbp-pσ1, chimeric gene encoding human myelin basic protein fused in frame to reovirus pσ1; T_35S, cauliflower mosaic virus 35S terminator; P_nos, nopaline synthase promoter; bar, gene conferring resistance to bialophos; T_nos, nopaline synthase terminator; P_2x35S, enhanced 35S promoter; TEV, tobacco etch virus translational leader sequence; Tvsp, vegetative storage protein terminator. (B) Amino acid sequence of MBP-pσ1 open reading frame. The gy1 SP sequence is upper case underlined; the hMBP sequence is upper case green text; an artificial linker connecting the hMBP and pσ1 domains is lower case underlined; the pσ1 sequence is upper case black text. (C) MBP-pσ1 trimeric model. Comparative protein structural homology modeling software was used to predict the structure of the chimeric hMBP-pσ1 protein. Crystal structures of the pσ1 head (PDB 1KKE, teal) and tail (PDB 6GAP, blue), predicted structure of MBP (AlphaFold, green) are highlighted as well as a 6-residue linker sequence connecting MBP to the tail region of pσ1 (red). Bottom left inset, a de novo surface model of a single MBP monomer (green) connected by a 6 residue linker (red) to the α-helix tail of pσ1 (blue). Top right inset, a surface model shows the trimeric pσ1 head domains remain intact as indicated by the interaction interface (red), as described by Chappell et al. (2002).

The crystal structure of pσ1 has been determined, and essential regulatory sequences have been previously characterized (Chappell et al. 2002). Native pσ1 is comprised of a long fibrous tail that inserts into the virion and a globular head that protrudes from the virion and binds host cells. Regulatory elements within the tail include a sialic acid-binding domain (SIAB) and a trimerization domain (Zlotkowska et al. 2012). Thus, it is important that autoantigen sequences tethered to pσ1 do not interfere with the trimerization, SIAB, or head domains, as these are essential for trimerization and cell binding. To determine whether MBP could interfere with essential pσ1 domains, we used comparative structural homology software to predict the trimeric structure of MBP-pσ1. Since high resolution crystal structures of the pσ1 trimer have been solved, we were able to build high confidence homology models using the predicted structure of human MBP obtained from Alphafold, a 6-residue linker, and tail domain pσ1 structure (PDB 6GAP). X-ray crystallography studies have determined the trimerization interface of the pσ1 head region, and have shown that the alpha-helix tail of pσ1 will form a coiled-coil structure when forming a trimer (Chappell et al. 2002; Dietrich et al. 2018). We combined these crystal structures to form a full-length pσ1, which was then aligned with the predicted structure of MBP, resulting in a MBP-pσ1 trimeric model (Figure 1C). Our model indicates MBP is distal to the tail and head of pσ1 (Figure 1C), and closely associated with additional MBP domains. Interestingly, previous studies have suggested that MBP may also form homodimers, therefore MBP may be able to undergo self-interaction as a result of close proximity due to the trimerization of the pσ1 tail region. While the predicted structures do not exclude interference between the MBP and pσ1 domains in vivo, they nonetheless predicted that the hMBP sequences present at the N-terminus of this novel chimeric protein had a low probability of interfering with pσ1 domains necessary for trimerization and cell binding, and provided sufficient confidence to move forward with soybean transformation experiments.

Analysis of T1 seeds for expression of MBP-pσ1

Transformation of soybean with binary vector pTN200-MBP-pσ1 resulted in eight independent T0 transgenic events that were taken to maturity and produced T1 seeds. Transformation with pTF101-MBP-pσ1 also produced eight independent T0 transgenic events, but only seven of these events produced sufficient T1 seed for subsequent characterization. As a first screen for MBP-pσ1 expression, ELISAs and western blot analyses were carried out using seed protein extracted from pooled samples of T1 seeds. ELISA absorbance values were plotted to allow for a relative comparison of transgenic protein from each transformation event (Figure 2A). Anti-pσ1 antibodies reacted with protein derived from transgenic samples but not WT protein, suggesting that MBP-pσ1 protein accumulated in T1 seeds. Varying absorbance values were observed for events transformed with both transformation vectors, and the ranges of absorbance values for each transformation were similar. Transgenic events with the greatest absorbance values were 81-15, 91-1, and 91-37, while protein derived from 91-16 exhibited the lowest absorbance value. There was no obvious correlation between MBP-pσ1 expression in transformation events generated with the pTN200 and pTF101 vector backbones.

Figure 2. Characterization of pooled T1 protein extracts. (A) ELISA results of pooled T1 protein. Equal amounts (1 µg) of pooled T1 seed protein or WT protein were coated onto 96-plate wells and incubated with polyclonal anti-pσ1 (primary) and anti-IgG (secondary) antibodies. Events designated 81- and 91- were transformed with pTN200-MBP-pσ1 and pTF101-MBP-pσ1, respectively. (B) western blot of pooled T1 protein separated by SDS-PAGE. Equal amounts (10 µg) of pooled T1 seed protein and WT protein were separated in 7% SDS-PAGE gels, transferred to a membrane, and probed with the antibodies used in ELISAs. A gel loading standard was included to allow comparison of samples run on different gels. Sizes of MW standards are shown in kilodaltons (kDa). The open arrow indicates transgenic protein with a MW consistent with intact hMBP-pσ1 (∼71 kDa). (C) western blot of pooled T1 protein separated by native PAGE. Equal amounts (10 µg) of pooled T1 seed protein and WT protein were separated in 7% native PAGE gels and probed with the antibodies used in ELISAs. The filled arrow indicates novel MBP-pσ1 complexes observed in transgenic protein extracts. (D) In vitro cell binding assays. Mouse L cells were fixed onto slides and incubated with WT or MBP-pσ1 soymilk. Fluorescence was detected following incubation with anti-pσ1 (primary) and Alexa 488-conjugated (secondary) antibodies. Images were captured using a 60x water immersion lens.

To visualize transgenic protein and verify that extracted MBP-pσ1 is intact, pooled samples were subjected to SDS-PAGE followed by western blot analysis. Probing of blots with anti-pσ1 identified one predominant protein band in samples prepared from transgenic seeds that were absent in WT seeds (Figure 2B). The novel protein had an observed molecular mass consistent with that predicted for monomeric MBP-pσ1 (∼71 kDa), suggesting that soy-derived MBP-pσ1 was intact. Notably, there was a direct correlation between the intensity of the novel 71 kDa protein observed in western blots (Figure 2B) and the absorbance readings of ELISAs (Figure 2A). The western blots also revealed the greatest levels of MBP-pσ1 in events 81-15, 91-1, and 91-37 and the least level of MBP-pσ1 in event 91-16, consistent with results observed in ELISAs (Figure 2A). The anti-pσ1 antibody also detected smears of protein migrating slower than the predominant 71 kDa band, suggesting that aggregates of MBP-pσ1 and/or interactions with endogenous soy proteins may also exist in seed protein extracts.

To determine whether multimeric (e.g., trimeric) MBP-pσ1 complexes are present in extracted protein, pooled T1 protein samples were subjected to native PAGE separation followed by western blot analysis. Under native protein separation conditions, a novel protein complex was detected in transgenic samples (Figure 2C). The signal intensity in the detected complexes directly correlated with signal patterns observed in SDS-PAGE western blots (Figure 2B) and absorbance values observed in ELISAs (Figure 2A). Again, the greatest levels of detectable protein complexes were observed in extracts derived from events 81-15, 91-1, and 91-37, with the lowest level of detectable complex observed in event 91-16 (Figure 2C). These assays indicate that MBP-pσ1 accumulates in transgenic seeds and that lines 81-15, 91-1, and 91-37 express the greatest levels of transgenic protein relative to the other events. Furthermore, all three assays (ELISA, SDS-PAGE, and native PAGE followed by western blot analysis) revealed identical information with respect to MBP-pσ1 expression levels, demonstrating that any of these assays can be used to detect and compare relative amounts of the accumulated target protein.

To determine whether functional trimeric MBP-pσ1 complexes are present in seed extracts, in vitro cell binding assays were carried out. Functional trimeric pσ1 will bind to receptors present on a variety of epithelial cells (e.g., M cells, mouse L cells, Hela cells, etc.) while monomeric pσ1 will not bind. For cell binding assays, mouse L cells were fixed onto slides and incubated with protein derived from transgenic events expressing MBP-pσ1 or WT seeds. Binding was detected using anti-pσ1 (primary) and fluorescently-labeled (secondary) antibodies and visualized using confocal microscopy. Soymilk prepared from seeds expressing MBP-pσ1 contained trimeric complexes that bound to L cells, as indicated by fluorescence of cell membranes (Figure 2D). In contrast, fluorescence was absent when fixed cells were incubated with WT extracts prior to confocal detection. These results demonstrate that transgenic seed extracts contain functional forms of MBP-pσ1 capable of binding L cells in vitro and by extrapolation, should also bind M cells in the mouse gut.

Stability of MBP-pσ1 over generations

To verify MBP-pσ1 stability over generations, T1 plants from multiple different lines were germinated and taken to maturity in a growth chamber. During outgrowth, the T1 plants appeared phenotypically similar to nontransgenic plants and did not exhibit signs of abnormal growth or seed yield. 12 progeny T2 seeds from each T1 plant were then screened for expression of MBP-pσ1 using native PAGE and western blot analysis. A representation of the western blot results obtained are shown in Figure 3. Multiple types of MBP-pσ1 expression patterns were observed in the T2 progeny screens. The most predominant expression pattern showed MBP-pσ1 expression in all 12/12 progeny tested, as observed in events 81-15-1, 81-15-3, 81-15-B, and 91-1-1. A second observed MBP-pσ1 expression pattern was similar to that anticipated for the expression of a gene following Mendelian segregation (e.g., MBP-pσ1 expression in 9/12 progeny and no MBP-pσ1 expression in 3/12 progeny), as observed in line 91-37-2. A third expression pattern was a hybrid of the two, as observed in line 91-1-B, where 11/12 progeny showed expression of MBP-pσ1 with one event presumably segregating away the MBP-pσ1 gene. Regardless of the different T2 expression patterns, all events expressed MBP-pσ1 in the T2 generation, demonstrating the stability of transgene expression between the T1 and T2 generations.

Figure 3. Characterization of MBP-pσ1 in T2 seed screens. Western blot results from six representative T2 screens are shown. T2 seeds were collected from T1 parent plants designated 81-15-1, 81-15-3, 81-15-B, 91-1-1, 91-1-B, and 91-37-2. Equal amounts of seed protein prepared from cotyledon slivers of 12 individual T2 seeds of the indicated lineage were separated by native PAGE and western blot analysis. WT seed protein was included as a negative control. Equal exposures are shown for relative comparison of signal intensity.

Formulation of therapeutic extracts

Previously published literature demonstrated that 50 µg doses of pσ1-based tolerogen were efficacious in preventing or reducing EAE when delivered via mucosal routes (Huarte et al. 2011; Rynda-Apple et al. 2011). Based on these studies, a similar 50 µg dose of soy-derived MBP-pσ1 was chosen for evaluation in this study. However, obtaining the desired 50 µg dose in crude soymilk formulations presented some challenges due to the solubility of soy protein, the expression level of MBP-pσ1, and the volume restrictions associated with oral gavage experiments in mice. Therefore, methods that enrich MBP-pσ1 in soymilk formulations were explored. Given the relatively high isoelectric point of MBP-pσ1 (pI=9.7) along with the acidic nature of soymilk proteins, it seemed likely that MBP-pσ1 could be partially purified by decreasing the pH of extracted protein solutions and removing precipitated proteins. To test this possibility, dilute HCl was added to seed protein extracts (soymilk) to reduce the pH of the solution from neutral down to a final pH of 3.0. At various steps during acid addition, protein samples were collected, and insoluble protein was removed. After neutralization, equal volumes of soluble protein were analyzed by native PAGE and western blot analysis (Figure 4A). Coomassie-stained blots showed an apparent reduction of total seed protein, concomitant with the decrease in pH of the extracted soymilk solution (Figure 4A, top panel). A Bradford assay was used to quantify protein in each sample, and the amount of protein precipitated at each pH step was calculated as a percentage of protein relative to the starting concentration. Only ∼5% of total soy protein was precipitated as the pH of MBP-pσ1 soymilk was decreased from pH 7.6 to pH 6.5. In contrast, ∼84% of total soy protein was precipitated as the pH was decreased to pH 4.5. The correlation between loss of protein and decreasing pH is consistent with the precipitation of 11S and 7S protein components, which represent major classes of protein in the seed proteome. To determine whether the drop in pH impacted MBP-pσ1, protein blots were subjected to western blot analysis followed by densitometry (Figure 4A, bottom panel). Reduction from pH 7.6 to pH 6.5 resulted in <2% loss of MBP-pσ1 signal, while further reduction to pH 4.5 resulted in ∼30% loss of MBP-pσ1 signal. Since T2 seed biomass expressing MBP-pσ1 was not limiting, a 30% loss of target during pH manipulation was deemed acceptable as the overall process resulted in a net 4.7-fold enrichment of MBP-pσ1. Ammonium sulfate precipitation, ion-exchange chromatography, and size exclusion chromatography were also explored as potential methods to enrich soymilk formulations for the MBP-pσ1 target. While those approaches also resulted in partial purification and enrichment of MBP-pσ1 (data not shown), none were as straightforward or time-efficient as the pH manipulation of soymilk extracts. Based on the pH reduction experiment results, we developed a soymilk formulation process that included pH manipulation as a first step, followed by clarification and concentration (Figure 4B).

Figure 4. Formulation of therapeutic protein extracts. (A) pH manipulation of protein extracts to enrich for MBP-pσ1. Dilute HCl was used to decrease the pH of extracted protein solutions in a stepwise manner. Equal volumes of soluble protein collected from each pH step were separated under native conditions in 4–15% gradient gels and visualized with Coomassie blue staining (top panel). A Bradford assay was used to quantify total soluble protein in each sample, and total protein lost was calculated as a percentage of extracted protein at pH 7.6. Sizes of molecular weight standards are shown in kDa. Membranes were also subjected to western blot analysis and densitometric analysis of detected signal (bottom panel). The amount of MBP-pσ1 lost at each pH step was calculated as a percentage of the densitometric signal in protein extracted at pH 7.6. The arrow represents MBP-pσ1 complexes. (B) Schematic outline of the process used to generate oral gavage formulations. (C) Densitometric quantification of MBP-pσ1 in therapeutic formulations. Known volumes of MBP-pσ1 formulations and known masses of hMBP standards were subjected to native PAGE and western blot analysis using an anti-hMBP antibody for detection. Descending triangles indicate decreasing volumes of samples from two different preparations. Asterisks indicate specific samples with densitometric signals that fell within the linear range of the standard curve and were used for MBP-pσ1 quantification. The arrow indicates MBP-pσ1 complexes.

To quantify MBP-pσ1 in oral gavage formulations, known amounts of commercially-purified human MBP protein standards and different volumes of formulated protein were separated in 5% native PAGE gels and then subjected to western blot analysis and densitometry of the resulting blots. To account for the size differences of MBP-pσ1 trimeric protein (∼213 kDa) and purified hMBP standards (∼18.5 kDa) on low percentage acrylamide gels, the formulated protein was loaded into gels and separated prior to a second loading and separation of hMBP standards. A polyclonal anti-hMBP antibody was then used to detect the MBP-pσ1 complexes and MBP standards in western blot assays (Figure 4C). The staggered loading of formulated protein followed by hMBP standards in native 5% gels resulted in sharper bands relative to the simultaneous loading of samples in gradient gels. Densitometry of signal detected in hMBP standards was used to generate standard curves, and the concentration of MBP-pσ1 in known volumes of soymilk formulations was ultimately determined by extrapolation from the generated curves.

EAE efficacy experiments

For EAE induction, groups of susceptible SJL mice were immunized with purified recombinant hMBP in Freund’s adjuvant on day 0. On days 0 and 2, mice were injected with pertussis toxin. Oral therapy was performed on days 1, 4, and 7, with control groups receiving WT soymilk formulations and experimental groups receiving similar soymilk formulations containing ∼50 µg MBP-pσ1. Mice were monitored daily for disease progression using a standard clinical scoring system of 0 to 4. Therapeutic treatment with formulations containing MBP-pσ1 significantly suppressed the development of clinical disease (Figure 5). The average peak clinical score in animals treated with MBP-pσ1 formulations was 0.2, while animals receiving WT milk developed disease with an average peak clinical score of 1.3 by day 16. All animals treated with WT soymilk exhibited significant disease with clinical scores ≥1.0 by day 16. Treatment with MBP-pσ1 formulations delayed the onset of disease from day 11 (WT formulation groups) to day 15 (MBP-pσ1 formulation groups). Collectively, these results demonstrated the efficacy of soymilk formulations containing recombinant MBP-pσ1 in suppressing clinical EAE and delaying the onset of disease in an animal model of MS when administered as an oral therapeutic.

Figure 5. Evaluation of MBP-pσ1 efficacy in EAE animal studies. On day 0, groups of female SJL mice (n=10/group) were immunized subcutaneously with 0.1 mg of purified hMBP adjuvanted with CFA (4 mg ml^－1 heat-inactivated M. tuberculosis). On days 0 and 2 (open arrows), mice were also immunized intraperitoneally with 400 ng of pertussis toxin. On days 1, 4, and 7 (filled arrows), mice were gavaged with 0.1 ml control (WT) soymilk or soymilk containing ∼50 µg of MBP-pσ1. Animals were monitored by the same pair of investigators for development of disease and assigned a score of 0 (no symptoms), 1 (mildly decreased activity, weak grip with fatigability), 2 (weakness, hunched posture at rest, and/or tremor), 3 (severe generalized limb and body weakness and/or lethargy) or 4 (moribund). Bars represent the standard error of the mean. Asterisks denote significance (p<0.05) between the two groups.

Discussion

The concept of using plants for the production of recombinant proteins has been explored for several decades with successful production of vaccine antigens, antibodies, and other therapeutic proteins in a wide variety of hosts. While many plant-derived recombinant proteins have advanced to evaluation in clinical trials (Maharjan and Choe 2021; Shim et al. 2019), only two plant-derived products have been approved by regulatory agencies for use in humans: The first was recombinant human glucocerebrosidase produced in carrot suspension cells and approved in 2012 for the treatment of Gaucher’s disease (Fox 2012), and the second was recombinant SARS-CoV spike protein produced as a virus-like particle in N. benthamiana and approved in 2022 for use in Canada as a two-dose adjuvanted COVID-19 vaccine (Duong and Vogel 2022). In both commercialized products, the target protein was purified prior to formulation. To our knowledge, there are no plant-derived products on the market or in clinical testing that have been formulated as a crude extract.

Some of the hurdles that have slowed the development of practical oral immunotherapies include the inability to deliver immunotherapeutics in a tolerizing context and a robust platform that can support the production and formulation of products at a low cost (Bluestone et al. 2010). In this study, we addressed these hurdles by using soybean to produce a novel multi-epitope autoantigen fused to pσ1. MBP is a highly charged, unstructured protein that is difficult to express in traditional protein expression systems. While relatively small amounts of MBP have been isolated from human tissue (Chen and Lai 1990; Tigyi et al. 1984) or purified following overexpression in bactieria (Nye et al. 1995), neither source would sustain the amounts of MBP needed in a clinical setting (Smith and Miller 2006). In contrast, we demonstrated that soybean could support the production and formulation of a novel MBP-based oral therapeutic. In past studies, we expressed a variety of recombinant proteins in soy ranging in size from relatively small vaccine antigens (Hudson et al. 2013) to large glycosylated proteins capable of self-assembly into multimeric complexes (Powell et al. 2011). This latter attribute is especially relevant for potential immunotherapeutics utilizing pσ1 since that protein must self-trimerize for functional binding to M cells on Peyer’s patches in MALT. Since soybeans have evolved to store significant amounts of relatively large glycinin and conglycinin protein complexes in seeds, it is not surprising that they also provide a favorable environment for the expression and accumulation of large and complex recombinant proteins such as ∼660 kDa human thyroglobulin (Powell et al. 2011) and the ∼213 kDa trimeric MBP-pσ1 complexes reported in this study.

MS is a complex disease, and while MBP represents the predominant myelin antigen associated with the disease, some MS patients possess antibodies to MOG and/or PLP proteins. In past efforts to develop efficacious immune-specific therapies, investigators utilized bacterial recombinant expression systems to produce chimeric proteins containing full-length antigens and/or epitopes derived from major myelin proteins for systemic delivery and evaluation in animal models. Intravenous delivery of such purified recombinant proteins suppressed the development of active and passive EAE when delivered prophylactically and ameliorated EAE when administered in multiple doses during ongoing disease (Elliott et al. 1996; Kaushansky et al. 2011; Zhong et al. 2002). While these and similar studies represent incremental advances made towards protein-based immunotherapies using EAE models over past decades, inherent issues with bacterial production hosts, a requirement for protein purification, and less-than-ideal routes for delivery are not practical for use in a clinic.

Building on the successes of immune-specific therapies in EAE models, human trials were carried out to evaluate the potential of myelin basic protein, myelin peptides, and altered peptide ligands as systemic immunotherapies for patients with MS. While intravenous delivery of basic protein and myelin peptides were safe and did not exacerbate disease, clinical outcomes remained disappointing (Campbell et al. 1973; Gonsette et al. 1977). As an alternative to systemic delivery, oral tolerance therapies utilizing myelin proteins were also evaluated in human clinical trials. In one study, MS patients consumed a bovine-derived myelin preparation (e.g., Myloral™) containing MBP and PLP (Fukaura et al. 1996; Weiner et al. 1993). Presumably, the difficulty and expense of expressing and purifying recombinant human MBP and human PLP precluded their use in this trial. Unfortunately, a Phase III clinical trial concluded that oral delivery of myelin proteins was not efficacious in treating MS (Benson et al. 1999). It is unclear why this therapy was ineffective in patients. However, one contributing factor could be the clearing of myelin proteins prior to recognition by effector cells responsible for initiating tolerogenic responses, suggesting a need for higher doses or novel ways to target autoantigenic proteins to effector cells.

Over the past decades, much has been learned about the gut mucosa and the requirements for the induction of oral tolerance, including the importance of the Peyer’s patches as an inductive site (Fujihashi et al. 2001; Song et al. 2008). Furthermore, regulatory dendritic cells (Matta et al. 2010; Scott et al. 2011) and regulatory T cells (Mizrahi and Ilan 2009; Tsuji and Kosaka 2008) present in Peyer’s patches and gut mucosa have been shown to play a critical role in tolerance induction. Thus, targeting autoantigens to Peyer’s patches and gut regulatory dendritic cells seems necessary for optimal oral tolerance induction. Given the involvement of M cells in immune tolerance, Pascual and colleagues showed that antigen uptake could be enhanced using M-cell targeting proteins such as reovirus pσ1. In one study, treatment with yeast-purified MOG:pσ1 suppressed MOG_35-55-induced EAE in C57BL/6 mice when administered prophylactically or during peak disease (Rynda et al. 2010). Similar protection was reported in the SJL model following nasal treatment with yeast-purified PLP:OVA:pσ1 (Rynda et al. 2010). The absence of disease suppression following treatment with pσ1 by itself demonstrated the antigen-specific nature of the suppression, while the lack of immune responses to pσ1 demonstrated the tolerogenic context of antigen presentation to effector cells. Notably, the reversal of clinical disease following oral administration of MOG-pσ1 demonstrated the efficacy of pσ1-based therapeutics delivered by practical mucosal routes. In light of these data, immune therapies that can be delivered by mucosal routes and target autoantigens to Peyer’s patches may show relevance in the clinic.

One of the goals of this study was to evaluate the efficacy of crude soymilk formulations containing the MBP-pσ1 tolerogen. While volume constraints associated with oral gavage experimentation in mice (e.g., 0.1 ml/gavage) and selection of a 50 µg therapeutic dose prevented us from using unfractionated soymilk for the gavage studies, we nonetheless demonstrated efficacy using soymilk formulations enriched for the MBP-pσ1 protein. Given the relatively high isoelectric point of MBP-pσ1 and the relatively low pI of the soybean proteome, a simple reduction of pH to 4.5 removed up to 85% of extracted protein and resulted in a 4.7-fold enrichment of MBP-pσ1 (Figure 4). Given the acidic nature of soy proteins, pH manipulation represents a powerful tool that can be used to provide significant enrichment of recombinant proteins with a pI >7–8. While we do not believe the pH manipulation and removal of non-target soy proteins impacted the efficacy of MBP-pσ1 formulations, this will need to be explored experimentally. One way to overcome gavage volume limitations in mice is to place soymilk formulations in feeder bottles and allow animals to consume soymilk formulations at will. We found that mice will consume significant volumes (up to 10 ml) of soymilk, though it can be challenging to synchronize consumption such that all animals consume the same formulation volume within a designated time period. A second way to overcome issues with gavage is to use rat models of EAE (Mannie et al. 2009). The larger body size of rats allows for larger gavage volumes and would eliminate the need for therapeutic enrichment in crude soymilk formulations. Finally, we have been exploring methods for spray-drying soymilk formulations—essentially converting them from liquid formulations to shelf-stable protein powders that can be reconstituted at desired therapeutic concentrations.

While our study evaluated the efficacy of MBP-pσ1 soymilk formulations administered prior to disease development, future studies will address the efficacy of MBP-pσ1 administered during ongoing EAE disease and in complex models of EAE. In this regard, it will be interesting to determine whether treatment with soy-based MBP-pσ1 reduces or eliminates secondary disease in the relapsing SJL model. Furthermore, while our selection of a 50 µg dose of MBP-pσ1 was based on an extensive body of work from similar studies (Huarte et al. 2011; Rynda et al. 2010, 2008; Rynda-Apple et al. 2011), Kaushansky et al. demonstrated that systemic administration of decreasing doses (i.e., 200, 100, 50, 20, and 10 µg/animal) of recombinant Y-MSPc were similarly efficacious in reversing ongoing PLP_139-151-induced EAE (Kaushansky et al. 2011). Thus, soy-derived pσ1-based therapies may also be efficacious when administered at relatively low (i.e., <50 µg) doses. We believe that the use of M-cell targeting ligands combined with a robust and cost-effective soybean platform that supports the production and formulation of protein therapeutics that can potentially bypass purification represents a practical strategy for the development of immune-modulating therapies to treat autoimmune diseases.

Abbreviations

CFA

complete Freund’s adjuvant

CNS

central nervous system

EAE

experimental autoimmune encephalomyelitis

MALT

mucosal-associated lymphatic tissue

MBP

myelin basic protein

MOG

myelin oligodendrocyte protein

MS

multiple sclerosis

PAGE

poly-acrylamide gel electrophoresis

PLP

proteolipid protein

pσ1

reovirus protein sigma 1

Conflict of interest

The authors declare there are no conflicts of interest.

Author contribution

K.J.P. designed the research, assisted with experiments, assembled figures, and wrote the manuscript. L.M.R. and L.H.R. performed most of the experiments, with L.H.R. assisting with the EAE animal studies. J.G. and R.C. carried out the protein modeling and structure prediction studies.

Funding

This work was supported in part by award number R43NS065505 from the National Institute of Neurological Disorders and R43AI155401 from the National Institute of Allergy and Infectious Diseases. The content is solely the responsibility of the authors and does not necessarily represent the official views of the individual institutes or the National Institutes of Health.