Advances in Purification of SARS-CoV-2 Spike Ectodomain Protein Using High-Throughput Screening and Non-Affinity Methods

The spike (S) glycoprotein of the pandemic virus, SARS-CoV-2, is a critically important target of vaccine design and therapeutic development. A high-yield, scalable, cGMP-compliant downstream process for the stabilized, soluble, native-like S protein ectodomain is necessary to meet the extensive material requirements for ongoing research and development. As of June 2021, S proteins have exclusively been purified using difficult-to-scale, low-yield methodologies such as affinity and size-exclusion chromatography. Herein we present the first known non-affinity purification method for two S constructs, S_dF_2P and HexaPro, expressed in the mammalian cell line, CHO-DG44. A high-throughput resin screen on the Tecan Freedom EVO200 automated bioprocess workstation led to identification of ion exchange resins as viable purification steps. The chromatographic unit operations along with industry-standard methodologies for viral clearances, low pH treatment and 20 nm filtration, were assessed for feasibility. The developed process was applied to purify HexaPro from a CHO-DG44 stable pool harvest and yielded the highest yet reported amount of pure S protein. Our results demonstrate that commercially available chromatography resins are suitable for cGMP manufacturing of SARS-CoV-2 Spike protein constructs. We anticipate our results will provide a blueprint for worldwide biopharmaceutical production laboratories, as well as a starting point for process intensification.


Introduction 24
Following the emergence of the SARS-CoV-2 virus in late 2019, a platform approach to betacoronavirus 25 spike protein stabilization in the pre-fusion conformation, along with early solved atomic-level structures of 26 the stabilized spike, allowed for rapid selection of the SARS-CoV-2 spike protein as an antigen for 27 vaccine development [1,2]. Recombinant spike protein constructs, both full length and soluble 28 ectodomain, are the basis of candidates in late-stage clinical trials, including those sponsored by 29 Novavax, Sanofi Pasteur, and GSK [3,4], and have the benefit of robust commercial experience and 30 previous licensure. Thus, recombinant proteins are a worthwhile complement to the novel technologies in 31 parallel development [5]. 32 In addition to vaccine development, numerous efforts to produce large quantities of spike protein are 33 underway in order to supply the high demand for therapeutic, diagnostic, and serosurveillance methods . 34 In therapeutic monoclonal antibody development, standardization of binding assays is important for 35 comparative data analysis. Spike protein binding assays are one method in use by the Coronavirus 36 Immunotherapy Consortium for assessing antibody treatments [6]. Similarly, population-wide serological 37 detection of SARS-CoV-2-specific antibodies with a spike protein ELISA is a useful tool for surveillance 38 and containment, with throughput and cost benefits over PCR-based virus assays [7]. To supply these 39 significant endeavors, a scalable, economical, rapid spike protein production protocol is of critical 40 importance. 41 Various SARS-CoV-2 spike protein production cell types are currently in use and development, including 42 insect [8,9,10], bacterial [11], and, predominantly, mammalian cell lines [12,13,14,15,16,17,18,19]. 43 Mammalian cell lines provide human-or human-like post-translational modifications, including 44 glycosylation, but require longer culture durations to express protein [20]. Glycosylation around the 45 receptor binding domain (RBD) of the spike protein is of specific interest, as it may play an important role 46 in antibody recognition [21]. In early mammalian-cell based production runs of stabilized, soluble spike 47 protein constructs, expression levels of 1 -5 mg of protein per liter of Expi293 cell culture harvest were 48 reported [13]. Yield optimization experiments, focusing mainly on transfection and cell culture conditions, 49 have increased reported upstream titers to between 100 and 150 mg/L in CHO cells [17]. Importantly, all currently reported purification processes employ affinity resins, predominantly featuring 51 immobilized metal affinity chromatography (IMAC) [9,12,13,14,15,16,17] and sometimes StrepTactin 52 [17,19,18], lentil lectin [8,9], immunoaffinity [22], or Anti-FLAG M2 [17] affinity chromatography. Except 53 for lentil lectin, these methods require the inclusion of a tag in the sequence of the molecule and, 54 generally, a protease-mediated cleavage step following purification. While these affinity methods yield a 55 highly pure product and require little optimization or development work, they are difficult to scale to large 56 manufacturing campaigns. Recently, advances have been made in affinity methods for application in 57 cGMP environment, specifically in single-use applications, but cost, ligand supply chain complexities, and 58 productivity remain a challenge [23,24]. Additionally, when size-exclusion chromatography (SEC) is 59 applied as a polish step after affinity chromatography [9,13,18], facility fit challenges arise; the required 60 large column volumes and small load volumes necessitate an extra concentration step prior to 61 chromatography or many cycles when manufactured at large scale. 62 To address these challenges, we employed cutting edge process development methods to find the 63 conditions that enable inexpensive, high-yield purification using non-affinity resins suitable for large-scale 64 manufacturing. Initial studies were performed using CHO-DG44 stable pools expressing the first reported 65 stabilized ectodomain protein, named S_dF_2P, designed from the WA-01 viral sequence [1]. This 66 construct consists of residues 1 -1208 of the spike ectodomain, stabilized by two proline mutations in the 67 S2 fusion machinery region. Additionally, the furin recognition motif, RRAR, at residues 682-685 was 68 mutated to GSAS. In a recently reported Phase 1 clinical trial interim analysis, this construct adjuvanted 69 with CpG 1018 and aluminum hydroxide has been shown to be well tolerated and immunogenic in healthy 70 adults [25]. 71 High-throughput chromatography resin screens using Tecan robotic liquid handlers and Repligen 72 Robocolumns containing 0.1 mL of each respective resin were performed as previously described [26,27,73 28] to select lead candidates for process development. Additionally, to ensure a safety profile meeting 74 regulatory agency guidance, viral clearance methods including low pH treatment and nanofiltration were 75 screened for compatibility with the molecule and purification process [29,30,31]. In sum, a novel process 76 utilizing non-affinity methods was developed in less than four calendar months.
To facilitate rapid product development, analytical and purification methods were developed 78 simultaneously. Initial screening experiments utilized raw binding data, reported in nanometer shift, from 79 the Octet platform. Later, a reference standard became available, and the Octet binding data were fit to a 80 standard curve to report a product-specific concentration. Due to low pH interference with both Octet 81 methods, product quantity was then inferred from GXII purity and A280 results for cation exchange (CEX) 82 step development. The developed process, shown in Figure 1, consists of cell culture harvest clarification by depth filtration 88 followed by ultrafiltration and diafiltration into a suitable buffer for anion exchange (AEX) capture 89 chromatography. Following the AEX step, the material is titrated to pH 3.5 for low pH treatment and then 90 subjected to a CEX step in flow through mode followed by another CEX step in bind-and-elute mode. The 91 purified material is then subjected to nanofiltration and a final concentration/buffer exchange step. 92 Following process development, we applied the developed process with no further optimization to a CHO-93 DG44 stable pool expressing the recently reported stabilized construct, HexaPro, containing four 94 additional proline mutations. The HexaPro construct was selected for the proof of concept run due to 95 previous findings that "HexaPro expressed 9.8-fold higher than [S_dF_2P], had a ~5°C increase in Tm, 96 and retained the trimeric prefusion conformation" [18]. The purification process and analytical methods 97 were applied to the HexaPro stable pool, which yielded 163 mg of purified product per liter harvest. 98 The process described herein is scalable, cost-effective, and provides increased yields of highly pure, 99 well-formed trimers. Moreover, these experiments provide a large dataset of commercially available 100 chromatography resins for further exploration. 101

Capture Resin Screen 103
The anion exchange resin screens yielded heat maps of S_dF_2P binding by mAb118 Octet, reported in 104 raw nanometer shift, as well as total protein concentration by pathlength-corrected A280 ( Figure S1). 105 Importantly, the elution fractions between 100 mM NaCl and 500 mM NaCl showed variations in A280 106 signal. Generally, the Octet binding heat maps ( Figure S1) show a significant portion of S_dF_2P in the 107 flow through and chase fractions, potentially due to high loading density. When comparing S_dF_2P 108 content to total A280, it is clear that successful separation is occurring as there are large A280 peaks but 109 very low S_dF_2P content in fractions > 500 mM NaCl. 110 For a more detailed analysis, pseudo-chromatograms were created by plotting both A280 and Octet nm 111 shift results from the pH 7.0 resin screen against NaCl concentration for each resin (

UF/DF I & Capture Resin Selection and Optimization
Experimental factors such as buffer system, pH, and UF/DF I feed stream conditions were screened for 130 impact on each candidate capture resin. First, the pH 7.0 buffer system used in the resin screen was 131 compared to an MES pH 6.5 buffer system. The load material in each buffer system/pH combination was 132 produced by both a 100 kDa UF/DF I membrane and a 300 kDa UF/DF I membrane to assess the impact 133 of feed stream characteristics on capture step performance. Each chromatography run ( Figure 3A

143
The elution fractions from the chromatography runs were assessed for recovery by octet titer and purity 144 by HP-SEC and SDS-PAGE. Across all resins, the 300 kDa load material yielded an elution of higher 145 purity than the 100 kDa load material (POROS 50D data shown in Figure 3). Furthermore, the pH 6.5 146 MES condition provided better resolution of the main S_dF_2P band from impurities in the flow 147 through/chase than the pH 7.0 Sodium Phosphate condition, based on SDS-PAGE ( Figure 3B). Thus, the 148 300 kDa-produced load material buffered in 25 mM MES, 25 mM NaCl pH 6.5 was selected. 149 All four runs on QAE-550C yielded overall low levels of the protein of interest compared to the other 150 resins and was thus not considered for further optimization ( Figure 3C). Both POROS 50 D and Gigacap 151 Q650M had 200 mM NaCl elution fractions with about 50% purity by HP-SEC. By octet titer, these 152 fractions yielded 69% and 78% recovery, respectively. The flow through/chase fraction for POROS 50 D 153 contained 30% recovery, compared to 10% for Gigacap Q650M. Both resins had a negligible amount of 154 S_dF_2P in the 100 mM NaCl fraction. Despite the higher recovery loss in the flow through/chase 155 fraction, POROS 50 D was selected as the capture step because the total mass balance was closer to 156 100%, so modulation of residence time and loading density were paths forward to reduce loss in the flow 157 through. Gigacap Q650M could also be chosen as a capture step to fit inventory or other laboratory-158 specific concerns. 159 Additionally, load material produced from a 300 kDa UF/DF I process, buffered in Sodium Phosphate or 160 MES with 25 mM NaCl at pH 6.5 were assessed on POROS 50 D. MES was confirmed as the buffer 161 system because the product band at ~200 kDa was more concentrated in fractions 7 through 10, 162 compared to the Sodium Phosphate buffer system, where the band of interest was found in fractions 4 163 through 9 ( Figure S3). The final process parameters can be found in Table S1.

Polish Step Resin Screen 165
Due to interference with Octet titer at pH ≤ 5.0, the CEX polish resin screen samples were assessed for 166 S_dF_2P content by using a concentration controlled GXII result. The purity of samples with an A280 167 greater than or equal to the median A280 value were reported to eliminate samples with high purity but 168 unacceptably low yield. Additionally, total protein content as measured by corrected A280 were reported 169 in the A280 heat map ( Figure S2). The A280 and GXII results were plotted against NaCl concentration for Measuring mAb118 binding of each neutralized sample on the Octet platform relative to control, the 195 relative binding was 96% after a 30-minute hold, 95% at 60 minutes, 89% at 90 minutes, and 99% at 120 196 minutes. These results indicate that low pH treatment for 60 minutes is a viable unit operation for 197 implementation in a cGMP process when assessed by binding to mAb118.  Figure 5). This design space has been explored previously with regard to 220 parvovirus clearance [32]. Both conditions showed adequate mass throughput, as measured by load 221 A280, for selection and scale up in a cGMP process and either could be selected for fit into a process. 222

224
Flux through the 20 nm filter is plotted against Mass throughput, measured by load A280 and volume.

225
The Viresolve Shield Prefilter and Viresolve Pro Nanofilter at pH 4.0 were chosen as the pH condition for 226 20 nm filtration due to higher mass throughput. Although low pH treatment and 20 nm filtration are 227 general industry practices, further experimentation such as live virus spike studies will be necessary to 228 confirm viral inactivation and clearance for implementation into a cGMP process. 229

UF/DF II 230
Flat sheet membranes with 300 kDa and 100 kDa pore sizes were tested for UF/DF II. The 300 kDa 231 membrane retentate contained no protein as measured by A280 and was therefore not analyzed further. 232 The 100 kDa membrane was able to retain the protein, and the intermediate samples were assessed for 233 HCP clearance. Peak HCP clearance, a 71-fold reduction, was identified to occur at the 20X DF sample 234 point. The sample taken after the chase was pooled with the 20X DF material showed only a 17-fold 235 reduction in HCP ppm from the load, so the chase was not pooled moving forward. 236

Proof of concept 237
The developed process, listed in Figure 1, was applied to the HexaPro construct. The upstream process 238 in CHO-DG44 cells yielded 737.8 mg/L of HexaPro in day 14 cell culture as measured by Octet titer. The 239 cell culture harvest was purified as described in the previous methods and yielded 163 mg of highly pure, 240 well-formed trimer per liter of cell culture harvest for a 22% purification yield. The final product contained 241 acceptable process-related impurity levels: 740 ppm HCP, and < 6 pg/mL (1.3 ppb) residual Host Cell 242 DNA. Additional characterization data for the proof of concept run is displayed in Figure 6.

Discussion & Conclusion 252
To rapidly respond to the SARS-CoV-2 pandemic, large quantities of soluble, stabilized spike ectodomain 253 protein are needed as a vaccine candidate and as a reagent for therapeutic and diagnostic development. 254 To date, purification of such proteins has required costly and difficult-to-scale processes, including affinity 255 and size-exclusion chromatography. This publication details the first known work to utilize high-throughput 256 robotics to select commercially available, inexpensive chromatography media to purify coronavirus S 257 proteins. We have demonstrated that the process presented herein is suitable for cGMP production of a 258 The HexaPro product produced by the novel process was assessed by various analytical methods to be 268 good quality with low levels of process-and product-related impurities. By DSC, the Tm of the HexaPro 269 product was found to be 59.3°C, an increase over previously reported Tm for S_dF_2P [18]. Binding data, 270 measured on the Octet platform, show differential binding curves to three SARS-CoV-2 specific 271 antibodies (RBD-binding mAb109, S2-binding mAb112, and mAb118, which was utilized for all other 272 Octet datasets herein and binds the NTD) [23]. Host cell protein was successfully cleared throughout 273 each unit operation to a final level of 740 ppm. 274 The methods and datasets presented provide a strong basis for further optimization. the pools recovered to ≥ 80%, the medium was replaced with ActiCHO P medium containing 6 mM L-295 glutamine and 100 nM MTX. 296

Clarification and Concentration/Buffer Exchange 297
For harvest volumes less than 5 L, the harvest material was clarified of whole cells and cell debris by 298 centrifugation at 3000 rpm for 30 minutes, followed by 0.8/0.2 µm sterile filtration (Sartorius Stedim, 299 Germany). Alternatively, for larger volumes, the harvest was subjected to a depth filtration train consisting 300 of Clarisolve 20MS followed by Millistak+ F0HC filters (MilliporeSigma, Burlington, MA) with a subsequent 301 0.8/0.2 µm sterile filter. The depth filters were arranged in series and equilibrated with 1X PBS. The cell 302 culture harvest was pumped through the filters at a 60 LMH feed flux based on the F0HC filter area and 303 chased with 1X PBS. Clarified harvest was stored at 2-8°C for further development activities. 304

UF/DF1 305
The clarified harvest was buffer exchanged using 100 kDa or 300 kDa Millipore Pellicon flat sheet 306 membranes (MilliporeSigma, Burlington, MA) with a five-fold ultrafiltration and a five-fold diafiltration into 307 various buffers as needed for capture chromatography. The feed flux was set to 330 LMH with a trans-membrane pressure of 10 psi. The 300 kDa flat sheet method was scaled up to a 1 m 2 filter, with loading 309 densities constant at around 10 L/m 2 . 310

Capture Resin Screen 311
Thirty-two anion exchange resins ( Figure S1) were screened in duplicate at two pH conditions (pH 7.0 312 and pH 8.0) with a step gradient of NaCl elution conditions, in 50 mM NaCl increments ranging from 100 313 mM to 500 mM NaCl, followed by a 1 M NaCl strip. The resin screen was performed using the TECAN 314  PA) were selected and further tested at pH 6.5 in an MES buffer system and pH 7.0 in a Sodium 331 Phosphate buffer system to assess the impact of lower pH and buffer system on recovery (as measured 332 by Octet titer) and purity (measured by HP-SEC) while including a head-to-head comparison to previous 333 experiments performed in Sodium Phosphate pH 7.0. Subsequently, POROS 50 D was tested at pH 6.5 in both buffer systems listed above to investigate the impact each factor individually (i.e., buffer system 335 and pH) ( Figure S3).    response values for all sample and calibrator replicates was ≤ 20% for all points above 0.78 µg/mL. 413

GXII
nominal magnification was 100,000x, corresponding to a pixel size of 2.2 Å, and the defocus was set at -467 1.0 µm. Data was collected automatically using SerialEM [35]. Particles were picked from the 468 micrographs automatically using in-house written software (YT, unpublished). 2D classification was 469 performed using Relion 1.4 [36]. DLS and DSC 477 DLS and DSC methods were performed as previously reported [37]. 478 479 480