Holistic process development to mitigate proteolysis of a subunit rotavirus vaccine candidate produced in Pichia pastoris by means of an acid pH pulse during fed‐batch fermentation

Abstract To meet the challenges of global health, vaccine design and development must be reconsidered to achieve cost of goods as low as 15¢ per dose. A new recombinant protein‐based rotavirus vaccine candidate derived from non‐replicative viral subunits fused to a P2 tetanus toxoid CD4(+) T cell epitope is currently under clinical development. We have sought to simplify the existing manufacturing process to meet these aims. To this end, we have taken a holistic process development approach to reduce process complexity and costs while producing a product with the required characteristics. We have changed expression system from Escherichia coli to Pichia pastoris, to produce a secreted product, thereby reducing the number of purification steps. However, the presence of proteases poses challenges to product quality. To understand the effect of fermentation parameters on product quality small‐scale fermentations were carried out. Media pH and fermentation duration had the greatest impact on the proportion of full‐length product. A novel acidic pH pulse strategy was used to minimize proteolysis, and this combined with an early harvest time significantly increased the proportion of full‐length material (60–75%). An improved downstream process using a combination of CIEX and AIEX to further reduce proteases, resulted in maintaining product quality (95% yield).

Foundation Grand Challenge-"Innovations in Vaccine Manufacturing for Global Markets". The stated aim of which is "to create a new manufacturing platform to enable ultra-low cost, high quality, recombinant subunit vaccines at 15¢ (USD) a dose for global health initiatives." Our initial model system to demonstrate this approach is the production of a recombinant subunit rotavirus vaccine candidate.
Rotaviruses are one of the most common causes of severe diarrheal disease, which prior to the introduction of an effective vaccination program was estimated to result in 500,000 deaths in infants and children worldwide annually. 2 This was found to be reduced to around 200,000 deaths annually in 2013 following introduction of vaccination programs, with the majority of these occurring in low-income countries. 3 Currently, there are two oral live virus rotavirus vaccines available, Rotarix ® (GlaxoSmithKline) and RotaTeq ® (Merck) both of which are recommended for inclusion in routine childhood vaccination programs by the World Health Organisation. 4 Recently, two further vaccines, Rotasil ® (Serum Institute of India) and ROTAVAC ® (Bharat Biotech), have also been pre-qualified for use by the WHO. However, the efficacy of these vaccines has been found to be low in some low-income countries with the highest need. 5 Other concerns with the use of these oral live virus vaccines include an association with a risk of intussusception 6 and a potential for re-assortment of live vaccine stains with circulating wild-type virus stains. 7 In addition, the cost of these vaccines is relatively high and is further increased by the requirement for an extensive cold chain during storage and distribution. 8 To alleviate these issues, new Non Replicative Rotavirus Vaccine (NRRV) based on the VP8 viral proteins from the P [4], P [6], and P [8] serotypes, fused to the P2 universal tetanus toxoid CD4(+) T cell epitope, have been developed and are currently undergoing clinical trials. 9,10 Currently, these NRRV antigens are expressed intracellularly in Escherichia coli, with subsequent purification, requiring three chromatography steps (Figure 1a). 9 Our aim is to move to a secreted system to significantly simplify this process, ultimately to be capable of incorporation into a single integrated process (Figure 1b), 1 thereby reducing cost towards meeting the target of 15¢ per dose or less.
To simplify the production and purification process of NRRV, Pichia pastoris, a methylotrophic yeast, was used as expression system as it provides several advantages over E. coli. For example, P. pastoris can achieve high cell densities in fermentation, and it is capable of secreting high titers of correctly folded heterologous protein. 11,12 In addition, the secretion of endogenous proteins is very low, which is advantageous for the simplification of downstream processing (DSP). 13,14 One potential drawback of using P. pastoris as an expression system for heterologous proteins is the presence of proteolytic enzymes. 15,16 These can have detrimental effects, such as product degradation or truncation, leading to reduced yield, and loss of biological activity. 15,17,18 Several approaches have been used to minimize the damage caused by proteolytic activity. At a fermentation level, supplementation with casamino acids and yeast peptone (substrate competition), 18 addition of protease inhibitors, 19,20 changes in fermentation parameters such as reduction of pH and temperature, 12,15,21,22 as well as utilization of alternate carbon sources, and optimization of induction times have all been used to reduce proteolytic degradation. 17 However, a generalized approach to solving problems of proteolytic degradation has not been established and solutions are case-by-case. Any actions taken at an upstream level to mitigate proteolysis, such as addition of inhibitors or competitors, can have consequences for the subsequent downstream, which will have to be capable of removing any additional components or inhibitors. 23 The introduction of changes to an already defined antigen designed to decrease susceptibility to proteolysis can have regulatory consequences. 24 Therefore, the development of upstream processing (USP) conditions that result in minimal protease activity are of great importance in production of full-length products.
In this study, a whole process approach was taken to identify compatible upstream and downstream conditions, where product quality was improved in fermentation and maintained throughout the purification process, to produce the highest proportion of full-length NRRV ( Figure 2).

| MATERIALS AND METHODS
Chemicals, unless specified otherwise, were obtained from Sigma Chemical Co. Ltd. (Poole, Dorset, UK).

| Cells and seed cultures
Fermentations were performed with Pichia pastoris (Komagataella phaffii NRRL Y-11430) secreting a P [4] or a P [8] serotype of a nonreplicating rotavirus VP8-derived subunit vaccine under control of the F I G U R E 1 Schematic process for the production of NRRV expressed in E. coli and P. pastoris. NRRV production process (a) from product expressed in E. coli 9 and (b) P. pastoris process in fed-batch or perfusion mode with 1 or 2 stage chromatography AOX1 promoter (a gift from the Love Lab at the Koch Institute at MIT) were re-streaked from a frozen clonal stock onto solid yeast extract peptone dextrose media. A single colony was picked and expanded in buffered glycerol complex medium (BMGY) to create working cell banks (WCW).
Inocula from WCW were grown in BMGY medium incubated at

| Fed-batch fermentation
Fermentations were carried out using an ambr ® 250 modular microbial system (Sartorius Stedim Biotech, Royston, UK). The bioreactors and integral reservoirs were aseptically filled with sterile media and feed solutions. Fermentations were carried out following Invitrogen's protocol for Mut + cells. 25 The initial fermentation volume was 120 mL of BSM with 0.52 mL of PTM 1 trace salts. The operating conditions were 30 C, 30% DO (using air or a mix of air and oxygen when required at 0.5 vvm), pH 5.00 ± 0.15 (controlled with 10% [v/v] ammonium hydroxide), and antifoam (polypropylene glycol 2000) was automatically added by the system when required. Reactors were inoculated via the septum to an OD 600 of 1.
Measurements from off-gas CO 2 sensors in the ambr250 modular were used by the system to calculate the carbon evolution rate (CER).
A 20% drop in CER was used to detect the end of batch phase and automatically started the glycerol feed (18.15 mL/hr per liter of the initial fermentation volume [L i Vol ]). Followed by a methanol adaptation phase using feed rate of 3.6 mL/hr/L i Vol . A decrease in CER during methanol adaptation indicated depletion of glycerol and was used as signal by the system to increase methanol feed rate, using an incremental step profile programmed into the ambr250 software (7.3 mL/hr/L i Vol for 2 hr and 10.9 mL/hr/L i Vol for the remainder of the fermentation). Glycerol and methanol feed rates were those specified by Invitrogen's protocol for Mut + cells grown at 30 C. 25 At the end of fermentation, the cell broth was centrifuged at 10,000 × g for 15 min and the supernatant was filtered using 0.80 μm cellulose nitrate and 0.45 μm polyvinylidene difluoride membrane filters (GE Healthcare Life Sciences, Buckinghamshire, UK). Samples were aliquoted into 20 or 30 mL and stored at −20 C until further processing. . Methanol adaptation feed rates were reduced by 30% for 20 C and 50% for 15 C.

| Chromatography: resin selection
To assess the suitability of resins for use in bind, elute chromatography as the first stage in the purification process automated micro-scale chromatography was carried out on a TECAN Evo liquid handling platform (Tecan UK Ltd, Reading, UK), using 96 well filter plates with either repeatedly diluted threefold and re-concentrated to original volume. This was repeated six times until the buffer exchange was >99.5%. The best performing resin from each class was then scaled-up.

| Combined cation and anion exchange chromatography
A two-step chromatography procedure was used for routine purification of all NRRV products. This was developed to maximize product purity and quality while minimizing the requirements for buffer exchange and thus ease of incorporation into a future integrated production system. was employed to elute P2-VP8-P [4]. This eluate was then applied directly to a 5 mL HiTrap Capto Q ImpRes column (GE Healthcare Life Sciences) pre-equilibrated in 50 mM Tris, pH 7.5. P2-VP8-P [4] was recovered in the flow-through, with remaining HCP and DNA binding to the column.

| Analytical methods
Cell growth was measured off-line using optical density measurements at 600 nm, wet, and dry cell weight (WCW, DCW).

| Impact of media and pH control
Initial fermentations were carried out using a rich defined media (RDM), and pH was controlled with potassium hydroxide (Figure 3a, [RDM 1]). 26 Cell growth and protein production were lower than in BSM 1 and 2, which could have been caused by nitrogen limitation. 30 Despite the low titer (0.55 g/L), product quality was much improved to 55% fulllength P2-VP8-P [4] at 46 hr; however, it was not stable over time, decreasing to 20% at 76 hr. Ammonium hydroxide was used to control pH either during induction (RDM 2) or throughout the whole fermentation (RDM 3 & 5). 31 Both strategies proved to be an efficient way to overcome the limitations in cell growth and protein production observed in RDM 1. To ease bioreactor operation, NH 4 OH was used to control pH throughout the whole fermentation in all further experiments. In changing pH from 6.5 to 7.0 (RDM 3) again, approximately 55% full-length P2-VP8-P [4] was observed at 46 hr; however, this was not stable, and by 76 hr, P2-VP8-P [4] was fully truncated.

| Identification of product truncation
It is known that proteases can be problematic during the production of heterologous proteins in P. pastoris. 17 For these reasons, we chose not to modify the media but focus on USP control strategies to inhibit protease activity.

| pH and temperature to control product truncation
Reduction of pH and temperature during fermentation have been used to decrease protease activity in fermentation supernatant. 12,15,21,22 Therefore, we tested these strategies to reduce protease activity and therefore product truncation to improve product quality.
Low pH strategies were tested: pH 4.0 throughout the fermentation (RDM 9 & 10) or a pH reduction strategy during induction from pH 6.5 to pH 3.0 (RDM 11& 12) with amino acid supplementation to avoid nitrogen limitation. 26  full-length was observed; however by 77 hr, the proportion of fulllength P2-VP8-P [4] decreased to 20% (Figure 3b).
A reduction of temperature in combination with a pH pulse was tested in an attempt to maintain the highest proportion of full-length P2-VP8-P [4]. However, reducing induction temperature did not improve product quality (RDM 17 & 18) (Figure 3b). When induction temperature was reduced to 15 C, due to low cell metabolism fermentation, the pH only reached 5 instead of 3. The pH pulse strategy was also tested with BSM (BSM 3) where 60% of the product was full-length P2-VP8-P [4] (Figure 3b), which was a significant improvement on fermentation BSM 1 where only truncated species were present (Figure 3a).
Zymograms were used to detect the presence of proteolytic enzymes in the fermentation supernatant throughout methanol induction (RDM 12) (Figure 4a). Protease activity at~60 and~40 kDa were visible at 37 and 39 hr when pH was greater than 6.1. When pH was reduced to 3.4, proteolytic activity was reduced and was neither detectable in the zymogram nor when the pH was increased to 6.4 (45 to 47 hr). Suggesting that a decrease on pH during methanol induction was an efficient strategy to reduce proteolytic activity in fermentation supernatant.
Based on these findings, a pH pulse during methanol induction was proposed and tested with BSM and RDM, leading to 60 and 75% full-length P2-VP8-P [4], respectively ( Figure 3b). However, depending on the media used, the reduction in pH was conducted at different times during the fermentation, at 60 and 30 hr for BSM and RDM, respectively (BSM 4 & 5 and RDM 15; Figure 3b).

| Effect of fermentation time on product truncation
For most of the conditions tested, it was found that of the variables

| Resin screening
We aimed to use a rational design approach to the development of DSP protocols, leading to the simplest process possible, ultimately being capable of incorporation into an integrated process To select the best performing matrix for P2-VP8-P [4] purification, a total of 18 resins were tested in a combination of static binding and dynamic binding studies using automated 96-well filter plate assays 27 and 0.6 mL RoboColumns 28 (Table 1). Resins were tested to define optimal binding and elution conditions for both pH and ionic strength. Resin suitability was also determined by product quality using SDS-PAGE to assess the level of truncation during the purification. The best performing resin from each of the classes tested were then scaled up to 1 and 5 mL columns for further optimization. These were: HIC -Capto Phenyl (High Sub); multimodal -Nuvia C Prime; cation exchange -Capto S ImpAct.

| Hydrophobic interaction chromatography
HIC is used as a bind-elute step in the published protocol based on E. coli expressed material, after the removal of HCP by anion exchange. 9 Therefore, it was expected that yeast expressed P2-VP8-P [4] could be purified by a bind-elute strategy using HIC resins. This was indeed the case. However, to obtain efficient binding, the clarified medium must first be subject to buffer exchange and addition of 1 M ammonium sulfate. P2-VP8-P [4] is eluted by decreasing the ammonium sulfate concentration to 0.8 M (Figure 5a). Initially, clarified medium was buffer exchanged to Tris at pH 7.5 (pre-treatment A, Figure 5a) as this was the binding condition shown to work for bacterial P2-VP8-P [4]. However, as shown in Figure 5a, bringing the pH to 7.5 leads to a high level of truncation of P2-VP8-P [4] by residual protease activity. This could decrease the yield of full-length protein due to proteolysis during handling/purification. To alleviate this, clarified medium was instead buffer exchanged to a pH of 4.0 (pre-treatment B, Figure 5a). This was found not to change the bind-elute characteristics of P2-VP8-P[4] (Figure 6a).
However, performing the chromatography at this reduced pH did not have the desired effect of reducing protease activity during purification, rather it resulted in a higher levels of truncation compared to P2-VP8-P [4] after HIC purification at pH 7.5 (Figures 5a and 6a). Use of HIC alone also resulted in unacceptably high levels of both HCP (>1,000 ppm) and residual DNA (16-24 ng/dose) irrespective of chromatography pH (Figure 5a). Although this protocol produced full-length material, it was not stable to further proteolysis, becoming completely truncated within 4 days at room temperature. Showing that significant levels of protease remained after HIC purification (data not shown).

| Cation exchange and multimodal chromatography
The HIC procedure while being significantly simplified over that used previously is still too complex for incorporation into a fully integrated process, 1 due to the requirement for buffer exchanges. Ideally, a process that requires no buffer exchange to be performed between the steps is desirable. This situation is most easily met by the use of a cation exchange or multimodal resin as the capture step. Examples of both classes of resin were found to be able to purify P2-VP8-P[4] (Table 1).
However, the use of a cation exchanger (Capto S ImpAct) was found to provide the most robust protocol. To achieve efficient binding, it is sufficient simply to adjust the clarified medium to pH 4.0 and conductivity to 10 mS/cm by direct addition of dilute citric acid (pre-treatment D, Figure 5b). This removes the need for any buffer exchange between fermentation output and column purification, thereby increasing Positive controls: thermolysin (37 kDa) and trypsin (19 kDa) compatibility with an integrated USP-DSP process. 1 The residual protease, which also binds the column, is efficiently removed by a pre-elution wash step at pH 4.0 with 0.2 M NaCl (Figure 4b). P2-VP8-P [4] is then eluted by increasing the salt to 0.4 M NaCl. The material produced by this single-column purification was found still to contain residual protease activity, which was only detected by MS analysis of material stored F I G U R E 5 Downstream processing and purification of NRRV from best performing fed-batch fermentations. Proportion of full-length P2-VP8-P [4] and truncated species determined by intact mass spectrometry. Required pre-column treatment is shown and bind-elution conditions indicated. Product recovery was determined by mass balance calculation using densitometry of SDS-PAGE. HCP and DNA content were determined for final purified NRRV (using the highest dose amount of 60 μg, currently under clinical evaluation).
(a) One-stage chromatography by HIC.
(b) Two-stage chromatography by CIEX followed by AIEX. For cases were insufficient levels of full-length, NRRV was recovered after initial CIEX step no further processing was performed at room temperature for 2 days. This revealed that further truncated species were being formed. To remove this activity and other residual HCP and nucleic acid impurities, a second anion exchange column (Capto Q ImpRes) was included in flow through mode (Figure 5b). This resulted in final HCP levels of <500 ppm and residual DNA of 26 pg/ dose. To alleviate the need for a buffer exchange between cation and anion exchange columns, the elution procedure was modified such that following the pH 4.0, 0.2 M NaCl wash the column is brought back to 0 M NaCl. P2-VP8-P [4] is then eluted by elevating the pH to 7.5 (Figures 5b and 6b). This eluate is then applied directly to a Capto Q ImpRes column, where the HCPs and DNA are bound and the P2-VP8-P [4] is in the flow-through (Figures 5b and 6b). This simplified protocol also resulted in final HCP levels <500 ppm and DNA levels of 30 pg/dose ( Figure 5b) and is compatible with incorporation into an integrated USP-DSP methodology. 1 Material produced in this manner was found to be stable at pH 7.5 for up to 8 days at room temperature when analyzed by intact MS (data not shown).

| USP-DSP interactions
Having significantly improved the individual elements of both USP and DSP a whole process approach was used to identify the best integrated process to give the highest yield of full-length P2-VP8-P [4] (Figure 7a). USP-DSP protocols were tested with a third independent batch achieving 76% full-length P2-VP8-P [4] determined by intact MS analysis with 81% total product recovery. In addition, an alternative P [4] expression plasmid which failed to yield any full-length products under standard BSM fermentation conditions (Section 2.2) was also tested. Under the improved process, this produced 93% full-length product (Figure 7b).
To serve as a platform process for the production of NRRV vaccine candidates, the USP-DSP must be capable of producing fulllength product from multiple virus serotypes. This USP-DSP protocol was successfully tested with cells expressing different product constructs P2-VP8-P [4] and P2-VP8-P [8]. The MS results showed that for both serotypes the greater proportion of product was full-length being 72-93% for P [4] and 91% for P [8] (Figure 7b). Proving that it is feasible to use this process as a platform for the production of different NRRV antigens.

| CONCLUSION
Proteases can be problematic during the production of secreted heterologous proteins in P. pastoris. 17,18 Here, we show that product truncation can be addressed at a process level to improve the quality of NRRV antigen. Fermentation process conditions have been improved to minimize proteolysis using a novel acidic pH pulse strategy. This has resulted in a fermentation harvest containing the majority of material as full-length, as For future studies, the identification of the specific proteases involved will help in the more rapid design and implementation of combined USP-DSP protocols for other secreted products in P. pastoris.
Although the two-step chromatography used reduces HCP and nucleic acid levels to within specified tolerances, it is possible that removal of the small amounts of residual product related impurities would require inclusion of an additional step if this proves necessary for clinical grade vaccine.