Optimizing selectivity of anion hydrophobic multimodal chromatography for purification of a single‐chain variable fragment

Abstract

Single‐chain variable fragments (scFv) are widely used in several fields. However, they can be challenging to purify unless using expensive Protein L‐based affinity adsorbents or affinity tags. In this work, a purification process for a scFv using mixed‐mode (MM) chromatography was developed by design of experiments (DoE) and proteomics for host cell protein (HCP) quantification. Capture of scFv from human embryonic kidney 293 (HEK293) cell feedstocks was performed by hydrophobic charge induction chromatography (MEP HyperCel™), whereafter polishing was performed by anion hydrophobic MM chromatography (Capto Adhere™). The DoE designs of the polishing step included both binding and flow‐through modes, the latter being the standard mode for HCP removal. Chromatography with Capto Adhere™ in binding‐mode with elution by linear salt gradient at pH 7.5 resulted in optimal yield, purity and HCP reduction factor of 98.9 > 98.5%, and 14, respectively. Totally, 258 different HCPs were removed, corresponding to 84% of identified HCPs. The optimized conditions enabled binding of the scFv to Capto Adhere™ below its theoretical pI, while the majority of HCPs were in the flow‐through. Surface property maps indicated the presence of hydrophobic patches in close proximity to negatively charged patches that could potentially play a role in this unique selectivity.

Abbreviations

CIP

Cleaning‐in‐Place

CV

column volume

DoE

design of experiments

EP

electrostatic potential

FT

flow‐through

HCP

host cell protein

HEK293

human embryonic kidney 293

MM

mixed‐mode

pI

isoelectric point

SAP

spatial aggregation propensity

scFv

single‐chain variable fragment

1. INTRODUCTION

Single chain variable fragment (scFv) is a widely used format in both basic and applied clinical research. Its application ranges from therapeutic drugs in tumor therapy and diagnostic probes in imaging to affinity ligands in chromatographic processes 1, 2, 3, 4, 5. Production of scFv is currently particularly challenging from a downstream point of view in terms of achieving high product recovery and purity. Current conventional methods for purification of scFv preparations from feedstocks include the use of Protein L‐based affinity adsorbents or alternatively affinity tags 6, 7, 8. In the former case, the purification is limited to fragments containing the V kappa light chain (I, III, or IV) domain. If the scFv contains kappa light chain II, it is possible to use Protein L‐based affinity adsorbents only by grafting Protein L binding activity 9. However, to avoid additional protein engineering, expensive Protein L‐based adsorbents as well as laborious purification processes, an alternative purification approach is needed.

Mixed‐mode (MM) chromatography has shown to be a highly selective technique for the separation of proteins and commonly used MM adsorbents include cation hydrophobic Capto MMC™ 10, anion hydrophobic Capto Adhere™ 11, 12, 13, ceramic hydroxyapatite 11, 14, and hydrophobic charge induction MEP HyperCel™ 11, 13, 15, 16, 17, 18. Especially for purification of full length mAbs, they have shown to be excellent due to the removal of aggregates, reduction of insoluble aggregation during chromatography, and viral clearance 10, 11, 12, 13, 15, 16, 17. Particularly, Capto Adhere™ in flow‐through (FT) mode and MEP HyperCel™ in binding mode have been reported to reduce the level of host cell proteins (HCPs) in human mAb preparations by 88 12 and 98% 11, respectively.

Nevertheless, there are currently few studies addressing purification of scFvs by MM chromatography. Gagnon et al. 19 highlighted the challenges with purification of mAb‐fragments by conventional methods and concluded that MM chromatography is a valuable alternative when conventional methods are insufficient 19. Furthermore, Lindner et al. 20 developed a four‐step chromatographic purification process for an anti‐epidermal growth factor receptor scFv produced periplasmatically in Escherichia coli. In the Lindner et al. study, capture was performed by cation hydrophobic MM chromatography using Capto MMC™, whereafter intermediate purification and polishing were performed by conventional anion‐ and cation‐exchange chromatography, respectively. Finally, HCPs were removed by ceramic hydroxyapatite chromatography 20.

In this study, an affinity‐free purification process for a scFv, produced in human embryonic kidney 293 (HEK293) cells, was developed using design of experiments (DoE) in combination with a state‐of‐the‐art proteomics approach for identification and quantification of HCPs. Three different MM adsorbents were investigated; MEP HyperCel™, Capto Adhere™, and Capto MMC™. The surface characteristic of the scFv was visualized by both spatial aggregation propensity (SAP) map 21, 22, 23, which was originally developed for analysis of aggregation prone regions of hydrophobic nature 24, and spatial distribution of electrostatic potential (EP) map. SAP and EP maps have previously been shown to be powerful tools for explaining selectivity of different mixed‐mode adsorbents for antigen binding fragments differing in hydrophobicity and surface charge 25, 26. Robinson et al. 27 recently reported that retention of mAbs and their fragments on different mixed‐mode adsorbents can be related to distinct surface properties elucidated by these surface property maps. In this study, SAP and EP maps were used in an attempt to interpret the chromatographic results.

PRACTICAL APPLICATION

A multimodal‐based purification process, consisting of capture by MEP HyperCel and polishing by Capto Adhere, was developed for a scFv using design of experiments and state‐of‐the‐art proteomics approach for mapping of host cell protein (HCP) impurities. Process parameters were screened and successfully optimized for maximum yield (98.9%), purity (>98.5%) and HCP reduction factor (14.3) for the polishing step. The reported approach enables identification of a unique selectivity for proteins that are otherwise challenging to purify.

2. MATERIALS AND METHODS

2.1. Biological material and model scFv

The model protein that was used was a scFv (26.9 kDa, theoretical pI 8.8) with a kappa II variable light chain. Plasmid (pTT5 vector) with DNA sequence encoding the scFv was ordered from Thermo Fischer Scientific (USA). The obtained plasmid was transformed into One Shot® TOP10 chemically competent E. coli cells (Thermo Fischer Scientific). The transformed E. coli cells were spread on LB medium agar plates containing ampicillin (100 μg/mL) and incubated overnight at 37°C. The following day, two to five colonies were inoculated in 50 mL of LB medium, whereafter incubated overnight at 37°C while shaking. The following day, plasmid purification was performed using the PureLink™ HiPure Plasmid FP Maxiprep kit (Thermo Fischer Scientific). Following this, the plasmid DNA was transfected into the Expi293™ Expression System (Life Technologies, USA) according to the instructions of the supplier of the expression system. The transfected HEK293 cells (1‐L scale) were grown for five days at 8% CO₂ and 36.5°C in a shaker incubator. At day five, supernatants were harvested by centrifugation, whereafter the harvest was sterile filtered through a 0.22 μm polyethersulfone (PES) filter system (dead‐ended filtration, EMD Millipore, USA). Capture of scFvs was performed by MM chromatography using MEP HyperCel™ (Pall, USA) column (16 mm × 5.0 cm; column volume, CV, 10 mL) with a load of 3 mg/mL resin and linear flow velocity of 60 cm/h (2 mL/min). The purification was performed on the Äkta Avant purification system (GE Healthcare, Sweden). Running buffers were equilibration buffer (54 mM sodium dihydrogen citrate, 45 mM disodium phosphate; pH 7.5), wash buffer (54 mM sodium dihydrogen citrate, 45 mM disodium phosphate, 1 M sodium chloride; pH 7.5) and elution buffer (54 mM citric acid, 45 mM sodium dihydrogen phosphate; pH 3.0), respectively. The scFvs were eluted by step elution (90% elution buffer) and the collected elution fractions were pooled and adjusted to physiological pH using 0.5 M disodium phosphate (pH 9.0). Finally, the elution pool was centrifuged (3000×g, 30 min) and sterile filtered.

2.2. Analytical methods

2.2.1. SE‐HPLC analysis

Determination of purity, protein concentrations, and relative percentages of high and low molecular weight proteins were carried out by SE‐HPLC analysis (Waters Alliance HPLC system, Waters Corporation, USA) on a BioSep‐SEC‐S3000 (5 μm, 7.8 mm × 30 cm) column from Phenomenex (USA). The flow rate was 1 mL/min and the temperature was 30°C. The SEC running buffer was 200 mM sodium phosphate, 300 mM sodium chloride, 10% isopropanol (pH 6.9), and the run time was 20 min with UV detection at 280 nm. Sample volumes were 5–20 μL. The high and low molecular weight proteins were defined as entities migrating with the retention time lower or higher than retention time for the monomeric protein, respectively. The amount of protein was calculated based on the area under the curve monitored at 280 nm relative to a standard solution.

2.2.2. SDS‐PAGE analysis

The purity of scFv in a sample was visualized by SDS‐PAGE gel. Precast NuPAGE 4–12% Bis–Tris Gels (1.0 mm × 10 and 12 wells), NuPAGE MES SDS Running Buffer 20×, Sample Buffer 4×, Sample Reducing Agent 10× and Mark12 Unstained Standard from Thermo Fisher Scientific were used according to the instructions of the manufacturer. Coomassie blue and silver staining of gels were performed using Instant Blue Protein Stain (Expedeon, UK) and SilverQuest staining kit (Thermo Fisher Scientific), respectively. The gels were scanned using Gel Doc™ EZ System (Bio‐rad, USA), whereafter qualitative analyses were performed using Image Lab™ Software (Bio‐rad).

2.2.3. Nano LC‐MS/MS analysis

Estimation of relative ppm of host cell protein (HCP) was performed according to the following procedure.

Enzymatic digestion procedure: Volume equivalent to 75 μg protein sample was added into an Amicon filter (10 000 molecular weight cut‐off, 0.5 mL), with additional 300 μL of denaturation buffer (6 M guanidinium hydrochlorid, 0.5 M Tris‐HCl, 5 mM EDTA; pH 7.6) and filters were centrifuged at 14 000×g for 15 min. Reduction of disulfide bridges was achieved by adding 100 μL of denaturation buffer and 5 μL of dithiothreitol solution (1.4 M in water), and incubated for 30 min at 56°C. After the reduction step, blocking of reduced Cys was achieved by addition of 20 μL of iodoacetamide (1 M in water), for 30 min at room temperature, protected from light. After the alkylation step, 300 μL of digestion buffer (2 M urea, 50 mM Tris, 8 mM methylamine, 2 mM calcium chloride; pH 8.0) was added and the filter was then centrifuged at 14000×g for 10 min. Finally, digestion was performed by addition of 100 μL digestion buffer and 1.5 μg trypsin and Lys‐C over night at 37°C. Digestion was stopped by acidification with 5 μL of 10% TFA and subsequently spun down. Twelve microliters of digest was desalted in PorosOligoR3 stage tip and redissolved in 12 μL of 0.1% formic acid (FA) in water (MS grade) prior to MS analysis.

Mass spectrometry analysis: LC‐MS/MS analysis of peptides was performed on an Easy‐nLC1000 nano HPLC system (Thermo Fischer Scientific) connected to a Q‐Exactive mass spectrometer (MS) (Thermo Fischer Scientific). Analytical columns were in house made by packing PicoTip Emitters (OD 360 μm, ID 75 μm) with Reprosil‐Pur Basic C18 material (particle size of 1.9 μm). Typically, column length ranged from 25 to 28 cm.

Nano LC‐MS/MS analysis was carried out as follows. Samples (2 μL) were loaded into the nanoLC column in 0.1% FA in water (solvent A). Peptides were then eluted by increasing the concentration of 0.1% FA in 90% acetonitrile (solvent B). Nano LC‐MS/MS blank runs (injection of solvent A) were run in between. MS was operated in the data dependent mode to automatically switch between full scan MS and MS/MS acquisition. Survey full scan MS spectra (m/z 300−1750) were acquired in the Orbitrap analyzer with resolution of 70 000 (at m/z 200) after accumulation of ions to a maximum automatic gain control target of 1 × 10⁶ counts. The 12 most intense multiple charged ions (z ≥ 2) were sequentially isolated and fragmented in the octopole collision cell by highenergy collisional dissociation with normalized high‐energy collisional dissociation collision energy 29% and acquired with 35 000 resolution and maximum injection time of 120 ms in the Orbitrap analyzer. The MS/MS ion selection threshold was set to automatic gain control target of 1 × 10⁵ counts and 2 Da isolation widths was used for precursor selection. Duplicate or triplicate MS analyses were performed in all cases.

Bioinformatics analysis of raw mass spectrometry (MS) data: Raw MS data were analyzed with the Proteome Discoverer software v2.1 (Thermo Scientific Inc.) using SequestHT (node in PD 2.1, in‐house server) as search engine. Data were analyzed against internal database containing the protein of interest and the available Uniprot database for Cricetulus griseuss. False discovery rates lower than 1% for peptide‐spectrum matches and lower than 5% for protein groups were calculated by Percolator v2.05 (validation node in PD 2.1). Only peptides with PD2.1 high‐confidence were taken into consideration. Moreover, only proteins with two or more identified peptides were accepted, in order to avoid false positive identifications. Relative protein quantitation was done by Proteome Discoverer software v2.1 (Thermo Fischer Scientific) by using a precursor area calculation workflow.

2.3. Capture test of scFv using Capto MMC™

The load was prepared according to the following. scFv from HEK293 feedstock was adjusted to pH 5.0 with 1 M hydrochloric acid, whereafter the sample was centrifuged (3000×g for 20 min) in order to remove precipitate. The obtained supernatant was sterile filtered through a 0.22 μm PES filter system (dead‐ended filtration, EMD Millipore). Seven milligrams of scFv sample was loaded onto a Capto MMC™ (GE Healthcare) column (7 mm × 2.5 cm; CV 1 mL) with linear flow velocity of 78 cm/h (0.5 mL/min). Subsequently, the column was washed for 5 CVs with A‐buffer (17.2 mM sodium dihydrogen citrate, 4.2 mM disodium phosphate, 100 mM Tris base, 23.8 mM citric acid, 5.8 mM sodium dihydrogen phosphate, 100 mM sodium chloride; pH 5.0), whereafter linear gradient elution (0–100% B‐buffer over 50 CVs) was performed; salt gradient B‐buffer (17.2 mM sodium dihydrogen citrate, 4.2 mM disodium phosphate, 100 mM Tris base, 23.8 mM citric acid, 5.8 mM sodium dihydrogen phosphate, 1.5 M sodium chloride; pH 5.0) and pH gradient B‐buffer (41 mM sodium dihydrogen citrate, 10 mM disodium phosphate, 50 mM Tris base; pH 8.5). Collected samples were analyzed by SE‐HPLC.

2.4. Screening design of scFv purification using Capto Adhere™

The statistical JMP software version 12.2 (SAS Institute Inc., USA) was used for screening design for identification of significant parameters for yield and purity in the purification step using Capto Adhere™ (GE Healthcare). The screening design is illustrated in Supporting Information Table 1. The load consisted of captured scFv from MEP HyperCel™ (see Section 2.1 Biological material and model scFv), and was buffer‐exchanged (PD‐10 desalting columns, GE Healthcare) into different A‐buffers (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES, 0 or 150 mM sodium chloride; pH 6.0 or 7.5) reflecting the conditions described in screening design (Supporting Information Table 1). For each screening experiment, 6.5 mL of buffer‐exchanged scFv sample (0.54 mg/mL) was loaded onto a Capto Adhere™ (GE Healthcare) column (5 mm × 5.0 cm; CV 1 mL) with linear flow velocity of 150 cm/h (0.50 mL/min). Subsequently, the column was washed for 5 CVs with each corresponding A‐buffer and elution was performed in linear gradient‐mode, 0–100% of B‐buffers (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES, 0.15 or 1.0 M sodium chloride; pH 3.5 or 6) over 50 CVs. The combination of the different buffer components for achieving a linear elution gradient was selected based on their linear titration curve calculated using the in silico method described in 28. All eight purifications were performed in scouting‐mode using the Äkta Avant purification system. Cleaning‐in‐Place (CIP) procedure of the column was performed by 1 M acetic acid (5 CVs), followed by 0.5 M NaOH (5 CVs). The following samples were collected; flow‐through (FT), wash, and elution fractions (0.5 mL‐fractions). Pooling of elution fractions was performed from front (3.5% of maximum peak height) to tail (3.5% of maximum peak height) of main elution peak, unless impurity peaks were overlapping with main elution peak. All collected samples were analyzed by the analytical methods as described in the previous section.

2.5. Optimization design of scFv purification using Capto Adhere™ in binding‐mode

The statistical JMP software version 12.2 was used for the design of optimization of significant parameters for yield, purity, and elution volume in the purification step using Capto Adhere™ (GE Healthcare). The optimization design is illustrated in Supporting Information Table 3. As previously described, the load consisted of captured scFv from MEP HyperCel™ (section 2.1 Biological material and model scFv), and was buffer‐exchanged (PD‐10 desalting columns, GE Healthcare) into A‐buffer (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES; pH 7.5). In scouting‐mode on the Äkta Avant purification system, 4–20 mL of buffer‐exchanged scFv samples (0.70 mg/mL) were loaded onto a Capto Adhere™ column (5 mm × 5.0 cm; CV 1 mL) with linear flow velocity of 50–150 cm/h (0.17–0.50 mL/min). Subsequently, the column was washed for 5 CVs with A‐buffer, whereafter elution was performed in linear gradient‐mode, 0–100% of B‐buffers (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES, 1.0 M sodium chloride; pH 3.5 or 7.5) over 50 CVs. CIP procedure of the column, sample collections, pooling of elution fractions and analysis were performed as described in the previous section.

2.6. Control test of scFv purification using Capto Q

The load consisted of captured scFv from MEP HyperCel™, see section 2.1 Biological material and model scFv, and was buffer‐exchanged (PD‐10 desalting columns, GE Healthcare) into A‐buffer (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES; pH 7.5) reflecting the optimized condition for Capto Adhere™. 6.5 mL of buffer‐exchanged scFv sample (0.70 mg/mL) was loaded onto a Capto Q column (5 mm × 5.0 cm; CV 1 mL) with linear flow velocity of 50 cm/h (0.17 mL/min). Subsequently, the column was washed for 5 CVs with A‐buffer, whereafter elution was performed in linear gradient‐mode, 0–100% of B‐buffer (24 mM formic acid, 21 mM acetic acid, 26 mM MES, 30 mM HEPES, 1.0 M sodium chloride; pH 7.5) over 50 CVs. Collected samples were analyzed by SE‐HPLC.

2.7. Homology modelling and in silico surface property characterization

A homology model of the scFv was generated using the RosettaCM protocol 29 in RosettaScripts 30. Briefly, structural homologs were detected using HHsearch 31 and the 10 closest PDB homolog structures were selected and served as templates for homology modelling in Rosetta 32. A total of 100 variants were generated followed by a restraint minimization 33 using the flag beta 34 and the lowest scoring model was selected. The surface properties of the model scFv were evaluated by SAP 35 and spatial distribution of electrostatic potential (EP) maps 25, 26 in the BIOVIA Discovery Studio 2018 software (Accelrys, USA). SAP values were calculated using a radius of 10 Å as it has been shown to be the most appropriate for identifying hydrophobic patches 23. EP values were calculated using the CHARMm PBEQ module in the BIOVIA Discovery Studio 2018 software.

3. RESULTS AND DISCUSSION

3.1. Capture of scFv from HEK293 feedstock

Two different mixed‐mode adsorbents, MEP HyperCel™ and Capto MMC™, were investigated for direct capture of scFv from HEK293 cell feedstocks (see experimental results in Supporting Information S1‐S2). In case of MEP HyperCel™, binding of scFv was performed by hydrophobic interaction at physiological pH. Elution was performed by electrostatic charge repulsion at pH < 4.8, which is the pK _a of the MEP HyperCel™ ligand. Yield and purity were estimated to 82 and 86%, respectively according to SE‐HPLC analyses (data not shown) (visualization of purity can be seen in Supporting Information Figure 1). In case of Capto MMC™, the binding of scFv was performed at pH 5.0, i.e. below its theoretical pI. Two different elution conditions were investigated; linear salt and pH gradients. In the former case of the linear salt gradient (0.1–1.5 M sodium chloride over 50 CVs), elution of scFv was not observed as binding to Capto MMC™ was very strong. The same phenomenon was observed in the latter case of the linear pH gradient (pH 5 to 8.5 over 50 CVs). No scFv protein was found in the FT and recovery from the resin was only possible at both high conductivity and high pH such as during CIP procedure with 0.5 M sodium hydroxide. This was in contrast to what was observed by Lindner et al. 20 for anti‐epidermal growth factor receptor scFv with pI of 7.4, which likely has a different spatial distribution of electrostatic potential and spatial hydrophobicity. As a result of poor recovery from Capto MMC™, MEP HyperCel™ was chosen as a candidate for capture.

3.2. Screening DoE of polishing step: investigation of binding‐ and FT‐modes

Following capture of scFv using MEP HyperCel™, polishing was investigated using Capto Adhere™. A fractional factorial screening DoE was designed for investigation of different loading and elution conditions (linear salt and pH gradients), covering both binding‐ and flow‐through (FT)‐modes, on yield and purity. The measured properties of the obtained elution pools (1.E1–1.E8) can be found together with the design in Supporting Information Table 1 and 2. Overall, yield and purity ranged from 83.7 to 98.7% and 93.3 to 96.1%, respectively. Lowest purity was obtained for experiments in FT‐mode, i.e. at 0.15 M sodium chloride in the load. Impurities with MW higher than 27 kDa that were visible by Coomassie stained SDS‐PAGE gel were fully removed in binding‐mode relative to FT‐mode, with an exception to binding‐mode at loading pH 6 (band with MW of 66.3 kDa is present in lane 2 of Supplementary material S3). In addition, proteomic data from nano LC‐MS/MS analyses showed that lowest HCP levels were obtained in binding‐mode relative to FT‐mode (Figure 1A and Supporting Information Table 1). All chromatographic profiles of FT‐ and binding‐modes are shown in Figure 2.

Figure 1.

Bar charts showing (A) yield, purity, and total HCP in elution pools 1.E1‐1.E8 from screening design of Capto Adhere™, and (B) yield, purity, and elution volume in elution pools 2.E1‐2.E9 from optimization design of Capto Adhere™. Total HCP in load sample is illustrated as a red triangles, whereas total HCP in elution pool is illustrated as black dots

Figure 2.

Chromatograms from screening DoE study; (A–H) illustrate experiments 1–8, respectively, in Supporting Information Table 1. CIP 1 corresponds to acidic cleaning with 1 M acetic acid, whereas CIP 2 corresponds to basic cleaning with 0.5 M NaOH

The parameters (Supporting Information Table 1) were screened by standard least square and analysis of variance for effects on yield and purity, whereafter a combined linear model was adapted for the significant parameters. Data from elution pool 1.E6 was excluded from the statistical analysis due to incomplete elution of scFv (elution buffer contained 0.15 M sodium chloride only). In the case of yield, loading pH, and second‐effect of loading pH and concentration of sodium chloride in the load were statistically significant (p < 0.05), whereas in the case of purity, loading pH was statistically significant. In the combined model, the significant parameters relating to yield and purity were adapted to the experimental data with accuracy of 93 and 87%, respectively (Supplementary material S4). The predicted plots for the responses showed that the experimental data points fitted closer to the model rather than the null hypothesis (mean value of experimental data) (Supporting Information Figure 4), thus the adapted model can be considered as significant. The root mean square errors for the yield and purity were determined to be 2.4 and 0.5%, respectively. The model showed that highest yield and purity were obtained in binding‐mode (without addition of sodium chloride in load) when loading at pH 7.5.

3.3. Optimization DoE of polishing step: Maximizing yield and purity in binding‐mode

Full factorial optimization DoE was designed for Capto Adhere™ based on the optimal conditions in the screening DoE, i.e. loading at pH 7.5 in binding‐mode (without addition of sodium chloride in the load). The optimization DoE was designed according to the following: the load consisted of captured scFv from MEP HyperCel™ as previously described, whereas the static conditions were selected as loading at pH 7.5 and elution with linear salt gradient (0–1 M sodium chloride over 50 CVs). Furthermore, three parameters, load, flow rate, and pH elution gradient, were selected for the full factorial DoE with one center point. In addition, three responses, yield, HCP reduction factor, and elution volume, were defined. The design can be found in Supporting Information Table 3 and 4 together with measured properties of obtained elution pools (2.E1–2.E9).

A graphical overview of the experimental obtained data can be seen in Figure 1B. Overall, yield, HCP reduction factor, and elution volume ranged from 71.1 to 98.9%, 6.8 to 22.9 and 7 to 11 CVs, respectively. Silver stained SDS‐PAGE gel showed one band corresponding to scFv protein only for all elution pools (2.E1–2.E9) (see Figure 3A). The purity was estimated to >98.5% for all elution pools according to SE‐HPLC analyses (see Supporting Information Table 3), thus the optimization was based on HCP reduction factor. Further on, the parameters were screened for significance by standard least square and analysis of variance for effect on the responses. In the case of yield, load was statistically significant (p < 0.05), whereas in the case of HCP reduction factor, load, pH elution gradient, and second‐effect of flow rate and pH elution gradient were statistically significant. In the case of elution volume, load, and flow rate were statistically significant. The significant parameters relating to yield, HCP reduction factor and elution volume were adapted to the experimental data in a combined model with accuracy of 92, 92, and 90%, respectively (Supplementary material S5). Predicted plots for the responses showed that the experimental data points were fitted closer to the model rather than the null hypothesis (Supporting Information Figure 5), suggesting that the obtained model is significant. Root mean square errors for the yield, HCP reduction factor and elution volume were determined to 5.2%, 2.1 units in HCP reduction factor and 0.7 CVs, respectively.

Figure 3.

(A) Silver stained SDS‐PAGE gel for the samples from the optimization DoE with 100 ng load of model scFv; lane (M) molecular weight standard (Mark12), lane (1) scFv harvest from HEK293 feedstock, lane (2) captured scFv from MEP HyperCel™ step, lanes (3‐11) elution pools 2.E1–2.E9, see Supporting Information Table 3. Visualization of proteomic data generated by nano LC‐MS/MS for the optimization DoE of Capto Adhere™ polishing step is illustrated by 3D bubble charts displaying molecular weight by theoretical pI of identified proteins in (B) load sample to Capto Adhere™ and (C) elution pool 2.E7, respectively. The bubble size corresponds to protein level in relative ppm‐MS. The purple bubbles illustrate the model scFv, whereas grey bubbles illustrate the identified HCPs

Highest purity was obtained in elution pool 2.E2 at low load (3 mg/mL), low flow rate (50 cm/h) and no pH elution gradient, i.e. elution by linear salt gradient only. However, the yield was only 71.1% (Supporting Information Table 3). The conditions for optimum yield, HCP reduction factor, and elution volume were high load (13 mg/CV), low flow rate (50 cm/h) and elution by linear salt gradient only (see elution pool 2.E7 in Supporting Information Table 3). Contour plots of the significant parameters in relation to the responses can be found in Figure 4. In case of yield, optimum conditions were high load and low flow rate (Figure 4A), whereas low load and low flow rate were optimum conditions for both elution volume and HCP reduction factor (Figure 4B and C). Furthermore, dual salt/pH elution gradient correlated with a higher level of HCPs in the elution pool relative to salt elution gradient only, suggesting that more HCPs coelute with the scFv. This could also be observed by the larger CIP peaks for experiments with linear salt gradients as compared to experiments with linear pH gradients in the screening DoE in Figure 2. Overall, optimum yield, HCP reduction factor and elution volume, based on experimental data, were 98.9%, 14.3 and 9 CVs respectively, corresponding to results for elution pool 2.E7 (Supporting Information Table 3). In these conditions, a total of 258 different HCPs were removed, corresponding to 84% of all HCPs identified in the load sample for the polishing step with Capto Adhere. The full list of identified HCPs can be found in Supplementary material S6‐S7. The relative ppm‐MS levels of the identified HCPs, in relation to their properties in terms of MW and theoretical pI, is visualized for the load sample (Capto Adhere step) and elution pool with optimum yield in Figure 3B and C, respectively. Most of the HCPs, which were removed in the experiments in binding‐mode, have pI below the one of the scFv (8.8), indicating that they were separated based on differences in their surface net charge relative to the one of scFv. The two most abundant HCPs, radixin and nucleoside diphosphate kinase, have a pI of 6.3, which can also explain their retention at pH 7.5 on the anion exchanger of the Capto Adhere™ resin. A subsequent polishing step with cation‐exchange chromatography in FT‐mode for the HCPs at pH 7.5, i.e. pH > 6.3, could potentially be explored in future studies to further remove the two most abundant HCPs as well as those with pI < 7.5 (Figure 3B and C). Tao et al. 36 showed that conventional cation‐exchange chromatography, although in binding‐mode for full‐length mAbs, can successfully be used as an alternative to affinity chromatography for reduction of HCPs by up to a factor of 44.

Figure 4.

Contour plots for significant parameters identified in the optimization DoE of Capto Adhere™. (A) Flow rate and load related to yield, (B) flow rate and load related to elution volume, (C) flow rate and load related to HCP reduction factor, (D) pH elution gradient and flow rate related to HCP elution factor and (E) pH elution gradient and load related to HCP reduction factor. The black points illustrate the experimentally obtained data in Supporting Information Table 3

Overall, the capture step with MEP HyperCel resulted in a purity of 86%. The polishing step with Capto Adhere resulted in a further improvement in purity to >98.5%. Silver stained SDS‐PAGE gel showed high level of purity for all elution pools (see lanes 3–11 in Figure 3A). Further assessment based on relative HCP ppm‐MS resulted in the optimal HCP reduction factor of 14 for the polishing step at the conditions of high load (13 mg/CV), low flow rate (50 cm/h) and elution by linear salt gradient (see experiment 7 in S5 Supporting Information Table 3).

The numbers of identified HCPs in the load samples for the screening and optimization DoE studies were 556 and 314, respectively (Supporting Information Table 1 and Table 3). This difference is due to batch‐to‐batch variability in HCP level during expression of the scFv. The level of HCPs, as well as their number, depends on the number of HEK293 cells that lyse during expression and is therefore difficult to control. However, the deviation between the total levels of HCPs between the batches was regarded as acceptable (576 000 ± 150 800 ppm; mean ± σ). Despite this limitation, we show in both DoE studies that the HCP level could be significantly decreased by using MEP HyperCel™ for capture and Capto Adhere™ for polishing in binding‐mode as compared to FT‐mode.

3.4. Importance of hydrophobic interaction for scFv binding

The importance of the hydrophobic interaction for scFv binding to Capto Adhere™ was investigated by a control run with Capto Q, using the optimum chromatographic conditions from the optimization DoE. No binding of scFv to Capto Q was observed (data not shown), verifying the importance of surface hydrophobicity in relation to negative charge for binding to Capto Adhere™ at pH below theoretical pI. Surface property maps indicated the presence of two potential hydrophobic patches in close proximity to negatively charged patches; (I) Leu, Thr, and Asp residues located in the variable heavy chain (VH) (front view in Figure 5A–C) and (II) Phe, Ile and Glu residues located in the variable light chain (VL) (back view in Figure 5A–C). To investigate whether these residues play a role in the binding, further experiments are required. Furthermore, in the case of cation hydrophobic Capto MMC™, the binding of scFv was too strong at the tested conditions (pH 5.0) and as a result, recovery was only possible at both high pH and high conductivity using 0.5 M sodium hydroxide, potentially following protein denaturation.

Figure 5.

Surface property characteristics of model scFv. (A) The heavy variable (VH) and light variable (VL) chains of the scFv are shown as grey and light blue, respectively. (B) SAP map visualizing hydrophobic and hydrophilic patches. (C) EP map visualizing spatial distribution of electrostatic potential

4. CONCLUDING REMARKS

In this paper, we demonstrated that an affinity‐free purification process for a scFv could be developed by fractional and full factorial DoE of MM chromatography in combination with a proteomics approach. Capture of a model scFv (pI 8.8) was investigated using MEP HyperCel™ through a hydrophobic charge induction mechanism and Capto MMC™ using a cation hydrophobic mechanism. MEP HyperCel™ was selected as a candidate for capture as Capto MMC™ was shown not to be suitable as the scFvs could not be eluted due to strong binding. Polishing was investigated using Capto Adhere™ through a hydrophobic anionic binding mechanism. Using DoE, chromatographic parameters were screened and successfully optimized for maximum yield (98.9%), purity (>98.5%), HCP reduction factor (14.3), and elution volume (9 CVs). In addition, the optimum conditions resulted in the removal of 258 HCPs, which corresponded to 84% of all HCPs identified in the load sample. The importance of the hydrophobic interaction as one of the main driving forces for binding was tested by the chromatographic experiment using the pure anion exchange adsorbent Capto Q. No binding of scFv to Capto Q was observed at the optimum conditions determined by optimization DoE for Capto Adhere™. Taken together, MM chromatography enables high selectivity based on surface property characteristics of the protein in question and the process parameters, which can be optimized by a statistical DoE approach in combination with a proteomics approach. Future studies aim for understanding the ratio of hydrophobicity to negative charge, mapping the residues that are pivotal for this binding, as well as investigating whether this information can be transferred to other scFv proteins. This knowledge will enable new possibilities for purification of other challenging proteins.

CONFLICT OF INTEREST

The authors have declared no conflict of interest.

Supporting information

Supporting information

Sakhnini LI, Pedersen AK, León IR, et al. Optimizing selectivity of anion hydrophobic multimodal chromatography for purification of a single‐chain variable fragment. Eng Life Sci. 2019;19:490–501. 10.1002/elsc.201800207