Behavior of host-cell-protein-rich aggregates in antibody capture and polishing chromatography

Abstract

Recent work has shown that aggregates in monoclonal antibody (mAb) solutions may be made up not just of mAb oligomers but can also harbor hundreds of host-cell proteins (HCPs), suggesting that aggregate persistence through downstream purification operations may be related to HCP clearance. We have examined this in a primary analysis of aggregate persistence through processing steps that are typically implemented for HCP reduction, demonstrating that the phenomenon is relevant to depth filtration, protein A chromatography and flow-through anion-exchange (AEX) polishing. Confocal laser scanning microscopy observations show that aggregates compete with the mAb to adsorb specifically in protein A chromatography and that this competitive interaction is integral to the efficacy of protein A washes. Column chromatography reveals that the protein A elution tail can have a relatively high concentration of aggregates, which corroborates analogous observations from recent HCP studies. Similar measurements in flow-through AEX chromatography show that relatively large aggregates that harbor HCPs and that persist into the protein A eluate can be retained to an extent that appears to depend primarily on the resin surface chemistry. The total aggregate mass fraction of both protein A eluate pools (~ 2.4 – 3.6%) and AEX flow-through fractions (~ 1.5 – 3.2%) correlates generally with HCP concentrations measured using enzyme-linked immunosorbent assay (ELISA) as well as the number of HCPs that may be identified in proteomic analysis. This suggests that quantification of the aggregate mass fraction may serve as a convenient albeit imperfect surrogate for informing early process development decisions regarding HCP clearance strategies.

1. Introduction

The advent of numerous monoclonal antibody (mAb) therapeutics has expanded the biopharmaceutical market appreciably over the last two decades, partially enabled by the establishment of general platform processes to support mAb downstream purification [1–6]. These platform processes typically begin with clarification of harvest cell culture fluid (HCCF) using centrifugation and/or depth filtration, followed by protein A chromatography for product capture [7,8]. One or more polishing operations using ion-exchange, hydrophobic interaction or multimodal chromatography are also typically employed to clear impurities that could jeopardize the therapeutic’s safety and stability [9,10]. Impurity classes in mAb solutions are often categorized as process- or product-related; process-related impurities include cell debris, lipids, host-cell proteins (HCPs), nucleic acids and adventitious viruses, whereas product-related impurities include mAb aggregates and fragments [11,12]. Although downstream operations typically remove multiple impurity classes simultaneously, the clearance of each one is often analyzed independently.

The distinction between process- and product-related impurities has been blurred by studies that found significant HCP content in mAb solution aggregates [13–21]. This was variously attributed to the electrostatic association of HCPs to chromatin–DNA complexes [13–19], the transient formation of charged mAb clusters [20] and to the unfolded protein response in cell culture [21]. Regardless of the origins, these structures appear to contribute to HCP persistence and we recently studied them in a cross-digest proteomic analysis of high molecular weight (HMW) impurities, showing that more HCPs were conserved among HCCF and protein A eluate aggregates than previously reported [22]. Such HCP-rich aggregates could potentially function like HCP reservoirs, as a typical mAb drug substance contains < 100 ppm HCP but ~0.1–1.0% HMW species [23–26]. How these HMW species may mediate HCP persistence through downstream operations remains an open question.

Some downstream operations are heuristically known to clear either HCPs or HMW impurities but not necessarily both. For instance, protein A chromatography is known to reduce the HCP content of HCCF substantially and several washes have been explored to improve this clearance. High-pH and arginine washes have usually provided meaningful benefits but their effects on aggregate persistence are less well understood [20,27,28]. Protein A chromatography is not necessarily considered beneficial for clearing HMW species because the product concentration effect coupled with the low-pH elution can promote aggregation [29–32]. In polishing, cation-exchange (CEX) and hydrophobic interaction chromatography (HIC) are primarily used to remove HMW impurities but the HCP concentration can be decreased as well [33,34]. Flow-through anion-exchange (AEX) chromatography is known to impact HCP concentrations because the majority of HCPs are more acidic than the typical mAb but this step is not usually considered to be especially effective for aggregate removal [35–37].

Despite some apparent dissimilarities, the relationship between HCP and HMW species may be useful in developing polishing operations for HCP clearance. Enzyme-linked immunosorbent assays (ELISAs) are typically used for estimating HCP concentrations but such measurements are inherently semi-quantitative and do not identify individual HCP species [38,39]. Shotgun proteomic techniques using liquid chromatography-tandem mass spectrometry (LC-MS/MS) could overcome these deficiencies but they are generally time-intensive and prone to false negative identifications, which may preclude confident assertions about the removal of individual HCPs [22,40]. If the HMW content of mAb solutions could serve as an approximate surrogate measure of HCP concentrations, it may be useful for informing early process development decisions such as which resin should be selected for a given polishing operation.

In this article we report a primary analysis of aggregate clearance in processing steps that are typically implemented for HCP reduction. Specifically, factors that contribute to aggregate persistence through protein A and flow-through AEX chromatography are studied on the length scales of both resin beads and the column. Confocal laser scanning microscopy (CLSM) is used to probe aggregate retention profiles and to make qualitative comparisons of adsorption affinity under a variety of conditions, and column studies are used to probe resin capacities for aggregate retention. The total HCP concentration of flow-through AEX polishing pools is compared with their HMW content and a quantitative proteomic analysis is presented. The relevance of aggregate-mediated HCP persistence to depth filtration is also demonstrated in a scale-down format. The observations from this study may be useful in developing process understanding of impurity persistence phenomena.

2. Materials and methods

2.1. Materials

Benzyl alcohol and tris-HCl were purchased from Sigma Aldrich. Tween 80, dimethylformamide (DMF), formic acid (FA), isopropanol (IPA), NaCl, sodium phosphate monobasic monohydrate, sodium carbonate, tris base, glacial acetic acid and acetonitrile (ACN) were purchased from Thermo Fisher Scientific. Dithiothreitol (DTT) was purchased from Bio-Rad, sequencing-grade trypsin was purchased from Promega and pre-digested yeast alcohol dehydrogenase (ADH) was purchased from Waters. AZDye/Alexa Fluor (AF) 488 and 568 NHS esters were purchased from Fluoroprobes and Cyanine5 (Cy5) NHS ester was purchased from Lumiprobe. Fluorophores were dissolved to 10 mg/ml in DMF and stored at −80 °C. The agarose-based protein A and AEX resins MabSelect SuRe LX and Capto Q, respectively, were purchased from Cytiva and an undecorated agarose resin that forms the base matrix of MabSelect SuRe LX was provided by Cytiva. The polystyrene-based AEX resins Poros XQ and Poros 50 HQ were purchased from Applied Biosystems. Other materials were procured from Bristol Myers Squibb as described previously [22], including clarified harvest cell culture fluid (HCCF) from an IgG1 mAb (145 kDa, pI of 8.9) manufacturing process and its downstream protein A eluate that had been viral-inactivated, neutralized and filtered (PAVIN). Both the HCCF and PAVIN had been clarified using depth filtration through an Emphaze AEX Hybrid Purifier (3M) that terminates in 0.2 μm sterile filtration. Buffer solutions were prepared using deionized water from an EMD Millipore Milli-Q system (> 18.2 MΩ cm) and pH and conductivity measurements were made using a Cole-Parmer PC200 meter.

2.2. Size-exclusion chromatography

A TSKgel G3000SWxl (Tosoh Bioscience) size-exclusion chromatography (SEC) column was used with an Äkta Explorer workstation (Amersham Biosciences) to isolate large-aggregate, small-aggregate and mAb fractions from HCCF and PAVIN feedstocks as described previously [22]. The SEC running buffer was prepared at pH 6.5 and contained 50 mM MES, 200 mM arginine, 5 mM EDTA and 0.05% sodium azide [41]. Isolated fractions were exchanged into a pH 8.3, 100 mM sodium carbonate buffer and concentrated to > 1 mg/ml using Amicon 10 kDa MWCO centrifugal ultrafiltration units (EMD Millipore) at 7500 g. Retentate concentrations were estimated using a NanoDrop Lite spectrophotometer (Thermo Fisher Scientific) using the mAb’s mass extinction coefficient as an estimator for the aggregates.

The TSKgel G3000SWxl column was also used for the offline SEC analysis of fractions that were collected from chromatographic column studies. Samples of 100 μl were analyzed in the pH 6.5 running buffer using a Waters 2695 HPLC equipped with a PDA detector and the absorbance was monitored at 280 and 254 nm. Component concentrations were estimated from SEC peak areas and normalized against their values in the relevant feedstock based on the assumption of a uniform extinction coefficient. As described previously for the fractionation of HCCF [22], SEC components included the aggregate and mAb fractions as well as two small-protein fractions (SPFs), which are denoted SPF 1 and SPF 2.

2.3. Confocal laser scanning microscopy

An illustration of the sample preparation workflow for CLSM studies is provided in Figure 1. Large-aggregate, small-aggregate and mAb SEC fractions were fluorescently labeled at pH 8.3 with AF 488, AF 568 and Cy5, respectively. To visualize adsorption behavior at low concentrations, the equivalent of a 20-fold molar excess of fluorophore to mAb was used to label the aggregates and a 1:1 stoichiometric ratio was used to label the mAb. Conjugation was allowed to proceed for 2 h in the dark at room temperature and the product mixtures were re-fractionated on the TSKgel G3000SWxl SEC column to achieve nearly complete removal of unreacted fluorophore. To avoid excess material loss the purified conjugates were concentrated using only one centrifugation cycle in 10 kDa MWCO ultrafiltration units and the retentates were buffer-exchanged using 10 kDa MWCO Slide-A-Lyzer dialysis cassettes (Thermo Fisher Scientific) according to manufacturer instructions.

Figure 1.

Sample preparation workflow for confocal microscopy experiments. Large-aggregate, small-aggregate and mAb fractions were isolated from harvest cell culture fluid (HCCF) or protein A eluate that had been viral inactivated and neutralized (PAVIN) using size-exclusion chromatography (SEC) and concentrated using centrifugal ultrafiltration. After fluorescent conjugation the product mixtures were re-fractionated using SEC to remove unreacted fluorophore and the purified materials were concentrated, dialyzed, filtered and mixed with chromatography resins of interest. These were imaged in the XZ plane using confocal laser scanning microscopy. (Illustration created using BioRender.com.)

For protein A experiments, each of the labeled HCCF SEC fractions was dialyzed into pH 4.0, 7.0 and 10.0 buffers that had matrices of 25 mM sodium acetate, 25 mM sodium phosphate and 25 mM sodium carbonate, respectively. Unaltered PAVIN feedstock was also exchanged into these buffers using a G-25 Sephadex desalting column (GE Healthcare) to be used as unlabeled mAb. For AEX experiments, labeled PAVIN aggregates were dialyzed into a 20 mM tris buffer at pH 7.8. The dialyzed SEC fractions were filtered using 0.8 μm, low protein binding Acrodisc syringe filters (Pall) and their absorbances were measured on a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific) at both 280 nm and the fluorophore absorption maximum. Conjugation ratios were estimated according to the fluorophore manufacturers’ instructions to be on the order of a 5:1 equivalent of fluorophore to mAb in the aggregates and 0.2:1 in the mAb fraction.

Two experiments were performed in batch with the MabSelect SuRe LX resin; the first probed the pH sensitivity of aggregate adsorption and the second probed the effects of common wash additives. For the pH sensitivity experiment, aliquots of MabSelect SuRe LX were exchanged into the three different pH buffers described above by repeated centrifugation and supernatant aspiration until a dilution factor of ≥ 1000 was achieved for the original solution constituents. Approximately 10 μl of resin was added to a 280 μl Eppendorf tube and 280 μl of each sample was prepared to minimize encapsulated air. Labeled HCCF aggregates were added to approximately 0.03 mg/ml of solution, corresponding to ~0.8 mg/ml of resin. A 1:20 labeled to unlabeled mAb solution was added to 4 mg/ml of solution, corresponding to ~112 mg/ml of resin. Given pipette variability and the small volume of resin used, the reported feed masses per resin volume represent estimates only. Samples were prepared with and without the mAb and controls were prepared with each component individually. A comparable pH 7.0 sample containing mAb was also prepared using the bare agarose base matrix. For the wash additive study, pH 7.0 samples with and without the mAb were prepared that included 1 M urea, 0.5 M NaCl, 0.45 M arginine or 10% IPA.

For the AEX experiment, the three resins of interest were exchanged into 20 mM tris at pH 7.8 and labeled PAVIN aggregates were added at approximately 0.03 mg/ml of solution. AEX samples did not contain the mAb fraction and were prepared with 0 and 100 mM added NaCl. Prior to imaging, protein A and AEX samples were incubated in the dark with end-over-end rotation on a Labquake vertical rotator for 1 h and 4 h, respectively. Following incubation, the Poros samples were exchanged into benzyl alcohol to decrease the refractive index difference between the polystyrene base matrix and the fluid phase for imaging [42,43].

Resin samples were transferred to 8-well Lab-Tek II chambered coverglass plates (Nunc) and imaged on an inverted Zeiss 880 confocal laser scanning microscope. The microscope was equipped with water (C-Apochromat 40x/NA 1.2) and oil (Plan-Apochromat 63x/NA 1.4) immersion objectives, laser lines at 488, 561 and 633 nm, a 32-channel ultra-sensitive GaAsP PMT detector and two multi-alkali PMT detectors. The oil immersion objective was used to increase image resolution in the protein A pH sensitivity experiment; otherwise, the water and oil immersion objectives were used for imaging the agarose and polystyrene resins, respectively. Cross-sectional images with a 12-bit color depth were acquired in both vertical (XZ) and horizontal (XY) orientations as shown in Figure 1 and described previously [44], and unloaded resin was used to confirm that background autofluorescence was negligible. Image acquisition settings (laser intensities, pinhole sizes, digital gain, wavelength collection ranges, etc.) were kept the same for all samples and controls within any experimental batch to enable qualitative comparisons of fluorescence intensity profiles.

Post-acquisition brightening was necessary to permit visualization of essential features of the confocal micrographs. A power-law (gamma) transformation with a background intensity cut-off threshold was applied as

$$y=\{\begin{array}{ll}0;\hfill & x\le \epsilon \hfill \\ {\left(\frac{x-\epsilon }{1-\epsilon }\right)}^{\gamma };\hfill & x>\epsilon \hfill \end{array}$$

where x and y are normalized pixel intensities before and after transformation, respectively, ε is the lower-bound intensity threshold and γ is the transformation power. All samples and controls within each experimental batch were brightened using the same parameters. Specifically, the values of (ε, γ) were set to (0.025, 0.30), (0.030, 0.35) and (0.040, 0.30) to brighten the images acquired in the protein A pH sensitivity, protein A wash and AEX studies, respectively. Pure green, red and blue colors were selected to render the detected signals from the channels corresponding to the large-aggregate, small-aggregate and mAb fluorophore conjugates, respectively. Color montage images were constructed to depict the superposition of all color channels alongside the individual signals and XZ images were cropped to focus on a single resin particle.

2.4. Column studies

Diba Omnifit 353 μl columns (3 mm ID × 5 cm, Cole-Parmer) were flow-packed with MabSelect SuRe LX and its bare agarose counterpart in a pH 7.2 buffer of 20 mM sodium phosphate with 150 mM added NaCl. These were validated by injecting a 20 μl pulse of 1% acetone at 150 cm/h (2 min residence time) on an Äkta Pure workstation (GE Healthcare) and verifying that the flow-through peak asymmetry was < 1.5. The AEX resins were flow-packed into 1.06 ml Diba Omnifit columns (3 mm ID × 15 cm) in a pH 7.8 buffer of 20 mM tris with 500 mM NaCl. These were validated by injecting 20 μl of 1 mg/ml lysozyme with 2 M NaCl at 50 cm/h (18 min residence time) into a pH 7.8 running buffer containing 2 M NaCl and verifying that the flow-through peak asymmetry was < 1.8.

The behavior of aggregates on MabSelect SuRe LX was surveyed by loading 10 ml of HCCF (~2× the equilibrium binding capacity) at 75 cm/h (4 min residence time) and collecting 250 μl fractions into 2 ml Whatman 96-well plates (Cytiva). Specifically, the system was (a) equilibrated for 7 column volumes (CVs) in the pH 7.2 phosphate buffer with 150 mM NaCl, (b) loaded with HCCF using a 50 ml Superloop assembly (GE Healthcare), (c) washed in the equilibration buffer for 8 CV, (d) eluted with a 10 CV gradient into a pH 3.6, 40 mM sodium acetate buffer, (e) regenerated by ≥ 15 min of contact with 0.1 M NaOH and (f) re-equilibrated until the pH returned to its neutral set point. The effluent absorbance was monitored at 295 nm and all fractions were analyzed using offline SEC (section 2.2). To ensure that inline neutralization during SEC did not affect the results, the low-pH protein A eluate fractions were analyzed using both the pH 6.5 SEC running buffer and a pH 3.6 running buffer containing 40 mM sodium acetate instead of 50 mM MES. A comparable investigation was performed using the bare agarose matrix to assess the extent of nonspecific adsorption of aggregates but in this study the matrix was equilibrated in a pH 7.0 buffer of 25 mM sodium phosphate without any added NaCl, and ~3 ml of HCCF were loaded, corresponding to the 10% dynamic binding capacity (DBC) of MabSelect SuRe LX.

The effects of common protein A wash additives on aggregate persistence were also studied by (a) equilibrating MabSelect SuRe LX in the pH 7.0 buffer without any added NaCl, (b) loading 3 ml of HCCF, (c) washing in the equilibration buffer for 10 CV, (d) washing in a buffer containing an additive of interest for 10 CV, (e) washing in the equilibration buffer for 5 CV, (f) eluting with a 10 CV step into the pH 3.6 buffer and (g) regenerating and re-equilibrating as described above. The wash additives that were investigated in the confocal microscopy experiment were likewise included in the column study as well as 1% Tween 80 and pH 10 washes. The protein A eluate pools from this study were analyzed offline using SEC, the Bradford assay (Thermo Fisher Scientific) and an F550-1 3G CHO HCP ELISA kit (Cygnus).

The breakthrough of aggregates in flow-through AEX chromatography was surveyed at superficial velocities of 150 and 30 cm/h (6 and 30 min residence times, respectively) by collecting 450 μl fractions while loading PAVIN feedstock that had been titrated to pH 7.8 using 2 M tris base. The AEX columns were (a) equilibrated for 7 CV in a pH 7.8 buffer of 20 mM tris, (b) loaded to 1000 mg mAb/ml column, (c) washed in the equilibration buffer for 7 CV and (d) the adsorbate was eluted using a pH 7.8 buffer of 20 mM tris with 2 M NaCl. A 30 CV gradient elution was used to fractionate the adsorbate in the 150 cm/h runs but adsorbate fractionation was omitted for the 30 cm/h runs. The columns were then (e) regenerated by ≥ 15 min of contact with 1 M NaOH and (f) re-equilibrated until the effluent returned to neutral pH. Selected flow-through fractions were diluted 50% in the pH 6.5 running buffer prior to offline SEC analysis and fractions that were collected upon 98% completion of loading were also analyzed using ELISA. A proteomic analysis was performed of fractions that were collected from the 150 cm/h runs at 5%, 28%, 53% and 99% completion of loading as well as the eluted adsorbate, for which the collected fractions were pooled. Prior to the analysis, samples were buffer-exchanged into 25 mM tris-HCl at pH 8.0 and concentrated to > 5.0 mg/ml using 3 kDa MWCO Amicon centrifugal ultrafiltration units.

2.5. SWATH proteomic analysis

Quantitative proteomic analysis was performed using a native digestion method with sequential windowed acquisition of all theoretical fragment ion spectra (SWATH) as described previously [22]. Native digestion was performed under non-reducing conditions [45]; 100 μl samples containing 500 μg of total protein were digested with trypsin at an enzyme to substrate mass ratio of 1:400 for 16 hr at 37 °C. Undigested mAb was reduced and precipitated by adding 1 μl of 50 mg/ml DTT and incubating for 10 min at 90 °C. Samples were centrifuged at 13000 g for 2 min and the supernatants were filtered through 10 kDa MWCO centrifugal ultrafiltration units (VWR) at 5000 g for 3.5 min. The filtration units were rinsed using 100 μl of 25 mM tris-HCl and centrifuged for an additional 4 min. The filtrates were acidified using 3 μl of 20% FA and samples were desalted using OMIX C18 pipette tips (Agilent). The C18 tips were conditioned with 50% ACN, equilibrated with 1% FA, loaded with sample, washed with 0.1% FA and eluted with 0.1% FA in 50% ACN. Three bind-and-elute cycles were completed on the same tip for each sample. Desalted samples were dried using a SpeedVac vacuum concentrator (Thermo Scientific) and redissolved in 49.5 μl of 2% ACN with 0.1% FA. Samples were spiked with pre-digested ADH to a concentration of 6.675 fmol/μl as well as retention time calibrants (iRT, Biognosys, Schlieren, Switzerland) and the volume equivalents of 90.91 μg of digested proteins were analyzed in technical triplicate in LC-MS/MS.

LC-MS/MS was performed as described previously [46]. A TripleTOF 6600 Sciex instrument equipped with an Eksigent NanoLC 425 was operated in microflow mode with a ChromXP C18CL Sciex column (3 μm, 120 Å, 150 mm × 0.3 mm). Mobile phases A and B consisted of 0.1% FA in water and 0.1% FA in ACN, respectively, and a flow rate of 5 μl/min was used. Elution was performed via gradients from 3 to 25% B over 68 min, 25 to 35% B over 5 min and then 35 to 80% B over 2 min. Data-independent acquisition (DIA) was performed using SWATH with an MS1 full scan and 32 variable MS/MS windows [47–49]. When necessary, data-dependent acquisition (DDA) was performed in positive ion mode with a 250 ms MS1 full scan over a mass range of 400–1250 m/z and the top 30 precursor ions were selected for fragmentation followed by a 50 ms MS/MS scan over a mass range of 100–1500 m/z.

A spectral library for SWATH analysis was previously constructed in Skyline (v20.2.0.343; MacCoss Lab, University of Washington) from DDA analyses of HCCF and PAVIN SEC fractions as well as the unfractionated feedstocks [22]. SWATH data extraction was performed with the Skyline command‐ line interface to process triplicate SWATH data for each sample. Peaks were selected and integrated automatically using the mProphet algorithm based on a target decoy approach [50]. A detection 𝑞 value was assigned to each peak and peptides with q > 0.01 in any of the triplicate measurements were removed from the analysis.

For the flow-through samples that were collected from Poros XQ and Poros 50 HQ upon 5% completion of loading, as well as the Poros 50 HQ sample that was taken at 28% completion of loading, an alternate SWATH data processing method was used. The mProphet automatic peak picking algorithm failed in the analysis of these samples because too few peptides were detected relative to the size of the SWATH data extraction target list derived from the library above. To address this, DDA data were collected and used to generate a smaller, more specific target list for each individual sample. Database searches for each sample in triplicate were performed using the Paragon algorithm in the Sciex ProteinPilot software (v5.1) as described previously [46,51], using a local copy of the Chinese hamster RefSeq 2019 database supplemented with trypsin, ADH, retention time calibrants and common contaminants. Proteins identified at a global false discovery rate of 1% were used to generate a sample-specific target list.

Following summarization by Tukey’s median polish, protein peak areas were calculated from peptide peak areas using MSstats (v3.18.5; Olga Vitek Lab, Northeastern University) and normalized against the ADH peak area [52]. The spiked mass of ADH was used to estimate HCP masses from normalized peak areas based on the assumption that all proteins of a given concentration generate the same response, and these were divided by the total injected sample mass to estimate fractional HCP mass concentrations.

The Seaborn Python module was used to interface with SciPy for hierarchical clustering analysis, which was performed using Ward’s minimum variance method based on Euclidean distances. Unidentified HCPs were assigned a concentration of zero and species-specific normalization was applied to HCP concentration vectors. HCP molecular masses and isoelectric points were also estimated from protein sequences by assuming that each residue titrates independently [53]. Seaborn was also used with Scott’s rule for the Gaussian kernel density estimation of property distributions when constructing violin plots [54].

2.6. Depth filtration

Depth filtration was performed using previously clarified HCCF, which was diluted 50% in a pH 7.4, 20 mM sodium phosphate buffer containing 100 mM NaCl. This dilution was performed to mitigate buffer-exchange effects that may occur at the beginning of filtration and to facilitate observation of the phenomena of interest. A 1 cm diameter section was punched from a sheet of 90ZB depth filter media (3M) and inserted into an Äkta online filter assembly as described previously [55]. The depth filter was (a) equilibrated in the pH 7.4 buffer for 75 L/m² (5.9 ml) at 300 L/m²/h (0.393 ml/min), then (b) 10 ml (~125 L/m²) of the 50% HCCF were loaded and 200 μl fractions were collected for offline SEC analysis.

3. Results and discussion

3.1. Aggregate behavior in protein A chromatography

3.1.1. Confocal laser scanning microscopy

After fluorescent labeling of the aggregate and mAb SEC fractions, the conjugation product mixtures were re-fractionated by SEC to remove free fluorophores. This was necessary to enable experiments with AEX resins, as the AF fluorophores used in this study have negatively charged moieties, but it also revealed that some smaller species were present in the aggregate fractions (Supplementary Figure S1). Specifically, a minor peak that appeared to be between the small-aggregate and mAb retention volumes was present in the large-aggregate fraction, and a minor peak at the mAb retention volume was also present in the small-aggregate fraction. This is consistent with recent observations from dynamic light scattering and may be attributable to the incomplete resolution between SEC fractions during the original material isolation or to any dissociation of mAb from the aggregates that may have occurred, as the mAb is expected to be the primary constituent species in the aggregates [22]. While the re-fractionation certainly reduced the relative concentration of these minor components, they were not eliminated entirely and remained present to some small extent in the CLSM studies.

As described in section 2.3, cross-sectional images were acquired in a conventional horizontal (XY) orientation but also in a vertical (XZ) orientation to enable a fair comparison of samples (cf. Figure 1) [44,56]. Absorption and light scattering effects decrease the observed fluorescence intensity as the sample depth (Z) increases, which can confound sample comparisons based on XY images that are taken at different depths. Nonetheless, the XY orientation permits multiple resin particles to be imaged simultaneously and may provide an indication of sample homogeneity. Figure 2 and Supplementary Figure S2 show XZ and XY micrographs, respectively, of MabSelect SuRe LX that was loaded with HCCF-derived samples at different pH values. Single-component controls that were imaged with the same acquisition settings are depicted in Supplementary Figure S3, which shows that color colocalization (i.e. “bleed-through”) artifacts were negligible. Both of the aggregate fractions adsorbed at pH 7.0 but no fluorescence could be observed in a comparable sample that was prepared with the bare agarose base matrix (data not shown), indicating that most aggregates require the protein A ligand to adsorb. This may be due to specific interactions between the ligand and mAb Fc domains that are accessible in aggregates or to nonspecific van der Waals and electrostatic interactions that involve the protein A ligand.

Figure 2.

Color montage confocal micrographs from XZ (vertical) cross-sectional imaging of MabSelect SuRe LX loaded with HCCF aggregates ± mAb at different pH values (columns). The top row shows all color channels combined and the bottom three rows depict the individual fluorophore color channels. The microscope objective (Plan-Apochromat 63x/NA 1.4) was located at the zenith; some spherical aberration is present due to the refractive index mismatch between the oil immersion medium and the samples. Acquisition settings were the same for all samples and images were cropped to one resin particle and brightened as described in the text (ε = 0.025, γ = 0.30).

The samples shown in the left half in Figure 2 were prepared with an excess of mAb to saturate the resin particles and those in the right half were prepared without the mAb to delineate multi-component interaction effects. In both samples at pH 7.0 the large aggregates were localized to the resin periphery, which may be due to strong adsorption or steric exclusion. The apparent pore radius of MabSelect SuRe LX has been reported to be ~40 nm [57] but the resin structure is expected to contain a distribution of pore sizes. Based on DLS measurements [22], a pore radius of ~40 nm would be small enough to exclude some of the largest aggregates but large enough to permit inward diffusion of the smaller species in the large-aggregate fraction. This may partially explain why the large-aggregate fraction appears to penetrate into the resin interior at pH 4.0, albeit incompletely, as evidenced by a subtle annulus at the resin periphery in the XY adsorption profiles (Supplementary Figure S2). However, any dissociation that may occur of mAb molecules from the large aggregates could also contribute to this observation.

Comparison of mAb-containing samples shows that the retention of aggregates decreases at low pH, suggesting that aggregates may co-elute with the product in protein A chromatography. Relative to pH 7.0, the pH 10.0 sample shows a clear decrease in aggregate retention without a substantial impact on the mAb, consistent with the successful implementation of high-pH washes for the capture step [27]. A comparable decrease in aggregate retention was not observed between the pH 7.0 and pH 10.0 samples that were devoid of free mAb, suggesting competitive adsorption between the free mAb and aggregates in which the relative adsorption strength of these components is modulated by the solution conditions. This is corroborated by the pH 4.0 samples, where the one devoid of mAb has brighter intensity profiles in the aggregate fluorophore channels than its mAb-containing counterpart.

The effects of other protein A washes were probed at neutral pH and comparable micrographs are given in Figure 3 and Supplementary Figures S4 and S5. Relative to the no-wash-additive comparator, 1 M urea and 10% IPA additives did not affect the adsorption profiles observably. The addition of 0.5 M NaCl and 0.45 M arginine did have observable effects, with the addition of arginine appearing more consequential and thus consistent with its relatively broad application in protein A washes [28]. As observed with the pH 10.0 systems in Figure 2, the wash additives in Figure 3 did not appear to reduce aggregate adsorption in the absence of mAb, which evidently must compete for the resin surface if the wash is to have an optimal impact. However, given the excess of mAb that was used to saturate the particles this may not be an accurate representation of process behavior because there is relatively little mAb in a protein A column’s interstitial volume during the wash step.

Figure 3.

Color montage of confocal micrographs from XZ (vertical) cross-sectional imaging of MabSelect SuRe LX loaded with HCCF aggregates ± mAb in the presence of common protein A wash additives at pH 7.0. The microscope objective (C-Apochromat 40x/NA 1.2) was located at the zenith, acquisition settings were the same for all samples and images were cropped and brightened as described in the text (ε = 0.030, γ = 0.35).

3.1.2. Column studies

To probe solution conditions that are more directly relevant to downstream processing, aggregate behavior on MabSelect SuRe LX was investigated in column format using HCCF as the feed with offline SEC analysis. Figure 4A shows the breakthrough and co-elution of the components that were quantified by SEC, where the component concentrations are normalized against those in the HCCF feed. A subpopulation of the aggregates broke through immediately but the majority appeared to be retained and their breakthrough was shallower than that of the mAb. This is perhaps due to heterogeneity in aggregate adsorption equilibria and the inevitably slower intra-particle uptake rates of aggregates relative to the mAb, but the effect would hardly be observed in practice when loading to below the DBC. Most of the SPF impurities broke through immediately but a small subpopulation of SPF 1 was retained along with the mAb and HMW species, and the retained components co-eluted with the product. The elution tail appeared to have a relatively high aggregate concentration and it is unclear whether this is primarily due to thermodynamic or transport effects. This phenomenon seems consistent with the recent finding of higher HCP concentrations in the elution tails of some mAb products that exhibit recalcitrant purification behavior [20].

Figure 4.

(A) Breakthrough and elution behavior of HCCF aggregates, mAb and small protein fraction (SPF) components on MabSelect SuRe LX as analyzed using offline SEC with a pH 6.5 running buffer. Concentrations are normalized relative to the feedstock and the absorbance at 295 nm (A295) is normalized against its maximum value at breakthrough. Conductivity and pH traces are plotted on the right-hand ordinate scales. (B) Aggregate and SPF 1 content (left ordinate) and ELISA-determined HCP concentrations (right ordinate) of protein A eluate pools after the application of different washes.

Since the running buffer for the SEC analysis was prepared at pH 6.5, and because neutralization can promote aggregation, there was a concern that the SEC measurement may have affected the results in the analysis of eluate fractions [32]. This was addressed by performing a comparable SEC analysis of the eluate fractions in a pH 3.6 running buffer, which showed that inline neutralization during SEC had negligible effects (Supplementary Figure S6). Also, negligible nonspecific adsorption of aggregates onto the bare agarose base matrix was corroborated by a comparable column study that showed a nearly complete breakthrough of all components at the start of loading (Supplementary Figure S7).

The effects of common protein A wash additives were probed by measuring the aggregate and HCP content of protein A eluate pools after subjecting the adsorbate to different washes (Figure 4B). The aggregate and HCP content are generally correlated, consistent with previous ELISA observations [22]. The eluate material subjected to the 1 M urea wash was an exception, where HCP content was measured to be relatively higher than the aggregate content would suggest. This may be due to the disruption of aggregates by urea and consequent liberation of HCPs or it may be simply an artifact of measurement inaccuracies. In general, all the wash additives reduced the aggregate and HCP concentrations in the eluate pool but the effects were moderate, and the pH 10.0 wash appeared to have the greatest benefit for HCP reduction in this system. The pH 10.0 wash was followed by the NaCl and arginine washes, which agrees broadly with CLSM observations except for the order of effects, but such a discrepancy may be attributable to the excess of mAb that was used in the CLSM samples.

To test whether protein A washes could be more effective under conditions of column saturation, a multicolumn experiment (Supplementary Figure S8) was performed as described in the Supplementary Material. Columns of 177 and 1760 μl were loaded in series until the 177 μl column was saturated and the two columns were separately washed and eluted. For comparison both columns were washed and eluted together without disconnection. The aggregate content of the eluate pools (Supplementary Figure S9) shows that column saturation did not have an appreciable effect on the efficacy of either pH 10 or 0.45 M arginine washes, suggesting that the competitive adsorption effect that was observed in CLSM samples requires an excess of mAb beyond what is required to saturate the resin.

An alternative strategy for possibly improving HCP clearance in the capture step could be to exploit transport rate differences between the mAb and HCP-rich aggregates, which was tested by using a protein A membrane adsorber at different flow rates as described in the Supplementary Material. It was hypothesized that using lower residence times (i.e., higher flow rates) could improve aggregate clearance due to slow aggregate mass transfer to the adsorbent surface but no such effect was observed in the initial breakthrough profile (Supplementary Figures S10 and S11) for residence times as low as 1.5 s. This emphasizes the need for clearance strategies based on polishing operations.

3.2. Aggregate behavior in AEX chromatography

3.2.1. Confocal laser scanning microscopy

To elucidate the role of aggregate-mediated HCP persistence in flow-through AEX chromatography, a comparable CLSM study to that on MabSelect SuRe LX was performed using PAVIN aggregates and three strong AEX resins of interest. Capto Q is an agarose-based resin with dextran surface extenders and quaternary amine chemistry that has a nominal mean particle diameter of 90 μm and the highest ionic capacity among the resins studied. The 50 μm particle diameter polystyrene-based Poros XQ and Poros 50 HQ resins are not functionalized with dextran and have quaternary amine and partially-quaternized polyethyleneimine chemistries, respectively, and Poros XQ has a higher ionic capacity than Poros 50 HQ. Adsorption profiles of PAVIN aggregates on these resins are shown in Figure 5 and Supplementary Figure S12; mAb was not included in these systems because the mAb is not intended to bind under the chosen solution conditions, which are relevant to flow-through polishing. Due to the resin base matrices and the immersion media, different acquisition settings were used for imaging Capto Q and Poros samples, which precludes a direct comparison of their fluorescence intensity profiles. Single-component controls for the Capto Q image acquisition settings are shown in Supplementary Figure S13; these were prepared by loading single aggregate fractions onto MabSelect SuRe LX at pH 7.8 to ensure strong aggregate adsorption, and they show that color colocalization artifacts were negligible. No such controls were prepared for the Poros resins due to material constraints.

Figure 5.

Color montage of confocal micrographs from XZ (vertical) cross-sectional imaging of AEX resins of interest for flow-through polishing applications. The particles were loaded with PAVIN aggregates ± 100 mM NaCl at pH 7.8. The microscope objective (water immersion C-Apochromat 40x/NA 1.2 for Capto Q, oil immersion Plan-Apochromat 63x/NA 1.4 for Poros resins) was located at the zenith and images were cropped and brightened as described in the text (ε = 0.040, γ = 0.30). Acquisition settings were the same for the two Poros resins but differed for Capto Q, so the fluorescence intensities in the Capto Q samples cannot be directly compared with those in the Poros samples.

Comparison of Supplementary Figure S13 with Figure 5 and Supplementary Figure S12 shows that PAVIN large aggregates bound with high affinity to the Capto Q surface and were localized to the periphery, whereas small aggregates appeared to bind weakly throughout the resin volume. As on Capto Q, the large aggregates appeared to be localized primarily to the Poros resin peripheries and the small aggregates adsorbed throughout the resin volume, although the adsorption of small aggregates on Poros 50 HQ was biased toward the periphery as well. Samples were prepared with 0 and 100 mM added NaCl to probe the resin’s salt sensitivity, which appeared to be low for Capto Q in this ionic strength range. A comparably low salt sensitivity was observed for the Poros 50 HQ samples, which appeared to bind the large and small aggregates weakly, but this was not the case for Poros XQ. Both aggregate fractions appeared to bind to Poros XQ strongly without any added salt, but the addition of 100 mM NaCl substantially reduced the adsorption strength, which was unexpected given Poros XQ’s marketed use in salt-tolerant applications. As the primary constituent of the aggregates is mAb, this behavior may be partially explained by the proximity of the solution pH to the mAb’s pI, which is ~1 pH unit higher than that of the solution, and the difference between the Poros resins may be attributed to their different surface chemistries. However, the consistency of adsorption to the three AEX resins indicates sufficient negative charge at pH 7.8 to effect charge-driven binding.

As in the CLSM study of protein A wash additives, these results provide microscale information but may not be completely representative of column behavior because the PAVIN feed to AEX columns contains mAb and a more complex buffer matrix. The negatively charged moieties in the conjugated fluorophores could also have contributed to stronger apparent adsorption in the AEX CLSM micrographs than what may be manifested by unlabeled samples.

3.2.2. Column studies

The column behavior of PAVIN aggregates during flow-through AEX chromatography was probed at two superficial velocities. Figure 6 shows the breakthrough of SEC fraction components where component concentrations have been normalized against those in the PAVIN feed. Aside from pH and conductivity transients that occurred at the beginning and end of loading due to buffer changes, the pH remained within a range of 0.2 units and the conductivity remained < 5 mS/cm. The buffer change at the start of loading caused a few early flow-through fractions to phase-separate, resulting in a leading spike in the apparent aggregate content of some profiles, but the separated phase may be expected to redissolve if the flow-through volume were pooled rather than fractionated. Figure 6 shows that the large aggregates were somewhat retained and there was a moderate effect of superficial velocity on large-aggregate retention, but the much more pronounced observation is that of appreciable capacity differences among the resins. The large aggregates were most extensively retained on Poros 50 HQ, followed by Poros XQ and then Capto Q, which is in the inverse order of the resin ionic capacities. This may be due in part to Poros 50 HQ’s partially-quaternized polyethyleneimine chemistry, which may have a more multimodal character than the other surface chemistries because the quaternization reagent is expected to be more hydrophobic than a methyl group, and additional hydrophobic character may result from the ethyl moiety in the polyethyleneimine chain. The greater retention on the Poros resins may also be due to pore structure; unlike Capto Q the Poros resins are purported to have relatively large throughpores that may enable the large aggregates to access a larger surface area near the resin periphery [58].

Figure 6.

Breakthrough behavior of PAVIN aggregate, mAb and small protein fraction 1 (SPF 1) components during flow-through AEX chromatography on different resins at superficial velocities of 150 and 30 cm/h. Ordinate scales are as described in Figure 4A. Vertical purple bars in the 150 cm/h plots denote the volume midpoints of 450 μl fractions that were taken for proteomic analysis from the flow-through at 5%, 28%, 53% and 99% completion of loading.

On all three resins a subpopulation of the large-aggregate fraction broke through immediately, which is presumably indicative of heterogeneity within this pseudo-component. Conversely, the small aggregates almost completely flowed through, with a minor subpopulation appearing to be retained on the Poros resins only. However, this is to some extent an artifact of the incomplete resolution between the large- and small-aggregate peaks in the SEC analysis, which are shown in Supplementary Figure S14 for the 150 cm/h runs. The flow-through of PAVIN small aggregates supports the speculation that mAb oligomers may be the primary constituent of this fraction, whereas the large-aggregate fraction contains a higher content of HCPs that may promote retention [22]. Reconciling these observations with the CLSM data requires knowledge of the PAVIN feedstock’s ionic strength, which was estimated to be ~55 mM based on a correlation between ionic strength and the instrument’s conductivity response, and this falls between the ionic strengths of the 20 mM tris buffers prepared with 0 and 100 mM added NaCl. The greater retention of large aggregates on the Poros 50 HQ column may therefore agree with the CLSM results but proving this would require finer salt steps between CLSM samples and careful matching of the PAVIN buffer matrix.

The measurement of the SPF 1 concentration is also influenced by artifacts arising from the incomplete resolution between component peaks in the SEC analysis, because overlap with the mAb peak may lead to overestimation of the SPF 1 concentration. To remedy this deficiency partially, the mAb-deficient adsorbate was fractionated by salt-gradient elution in each 150 cm/h experiment. The eluate profiles (Supplementary Figure S15) show the adsorbed and eluted masses to agree within 15% of the fed mass for each component, which was considered acceptably accurate given the low signals measured in the SEC analysis and the uncertainties in using a lumped extinction coefficient for each SEC pseudo-component. The elution profiles confirm that SPF 1 components were largely unretained and reveal a slight tail in the elution of large aggregates from Poros XQ, which is further indicative of the aggregate heterogeneity that was inferred based on the breakthrough results.

Flow-through AEX polishing steps have been heuristically associated with HCP removal but not necessarily aggregate reduction and this may be partially attributable to the flow-through of small-aggregate impurities as well as the large load and capacity differences among resins. The relationship between the persistence of HMW and HCP impurities is explored directly in Figure 7A using ELISA and SEC data for AEX fractions that were collected upon 98% completion of loading. The fraction at 98% loading was selected for ELISA analysis because the end of the flow-through should in principle be the least pure. The large-aggregate and HCP content of these fractions are plotted in Figure 7A and the correlation between those quantities is depicted in Figure 7B, in which the total HMW content is used to facilitate comparison with data from the column study of protein A washes. Despite the one exception of the urea wash sample, the two metrics correlate across experiments and suggest that aggregate breakthrough in flow-through AEX chromatography may be a possible surrogate for monitoring HCP persistence conveniently using SEC.

Figure 7.

(A) Large-aggregate (left ordinate) and ELISA-determined HCP content (right ordinate) of AEX flow-through fractions that were collected upon 98% completion of loading. (B) Correlation between the ELISA-determined HCP concentrations and the total high molecular weight (HMW) contents of samples from the protein A wash and flow-through AEX column studies.

3.2.3. Proteomic analysis

To investigate HCP clearance in flow-through AEX more thoroughly, fractions were selected from the 150 cm/h experiments as well as the eluted adsorbates and analyzed using quantitative SWATH proteomics. An overview of the number of HCP species identified and the total HCP concentration in each sample is shown in Figure 8, as well as the relationship of those two metrics to the HMW mass fraction. The number of HCP species that were identified in each flow-through fraction appears to be generally consistent with the breakthrough of large aggregates (Figure 6) and was smaller than the number of species that were previously identified in the SWATH analysis of the PAVIN feedstock [22]. In principle the number of species and the total HCP concentration should be monotonically increasing across flow-through fractions for a given resin; deviations from this expectation are attributed to measurement variability, low HCP concentrations near the limit of detection and the inherent ambiguities of shotgun proteomic analysis that have been discussed elsewhere [22]. With few exceptions the number of HCPs increases as expected across the flow-through fractions and suggests that Poros 50 HQ provided the best clearance. This is corroborated by the adsorbate samples, which show that the highest number of detectable HCPs and the greatest HCP mass adsorbed to Poros 50 HQ.

Figure 8.

Number of HCP species identified in the SWATH proteomic analysis of AEX samples (top row) and the total HCP concentration of those samples (bottom row), both on logarithmic ordinates. Individual samples are shown in the bar plots (left column) and PAVIN feedstock data that were acquired in a previous analysis are included for reference [22]. Numbers above bars in the top left plot refer to the number of detected HCP species and error bars in the bottom left plot represent 2× the standard deviation. HCP concentrations were computed in units of ppm on a total mass basis (i.e., ng HCP / mg total). The relationship between the HMW mass fraction and the number of HCP species (top right) or the total HCP concentration (bottom right) of the flow-through samples is also shown. The HMW content was not quantified for samples taken upon 5% completion of loading for Capto Q and Poros 50 HQ due to phase separation that was observed at the start of loading.

The number of HCP species correlates generally with the HMW content but, unlike the ELISA data, the SWATH-determined total HCP concentration does not. The data show that Poros 50 HQ offered superior clearance through 28% completion of loading but the trend in the concentration data of flow-through fractions is unclear thereafter. Given this ambiguity, emphasis is placed on the interpretation of data from the adsorbate samples, which represent measurements of cumulative rather than instantaneous behavior and naturally exploit mAb depletion coupled with HCP enrichment prior to the proteomic analysis. These data are therefore expected to be more reliable indicators of polishing performance and they corroborate that Poros 50 HQ provided the best HCP clearance.

The clustered heat map in Figure 9 reveals broad overlap in the HCP profiles that were observed among the flow-through fractions. For reference Supplementary Figure S16 shows only the flow-through data to facilitate fraction comparisons and provide more granularity in the depiction of relative concentrations, and Supplementary Figure S17 provides Venn diagrams to depict explicitly the overlap in the number of HCP species that were identified across resins for each AEX fraction. Figure 9 shows that most of the HCP species that broke through on the Poros resins also broke through on Capto Q but most of the flow-through species were also present in the adsorbate and were therefore partially retained. This is corroborated by the observation that the concentration of most HCPs in the flow-through samples was lower than in the parent PAVIN feedstock (Supplementary Figure S18). Supplementary Figure S18 also shows that individual HCP concentrations tended to be higher in the adsorbate pools than in the PAVIN feed, consistent with the total HCP concentrations shown in Figure 8.

Figure 9.

Clustered heat map of HCP concentrations measured in the proteomic analysis of all AEX samples. HCPs (rows) are clustered, whereas samples (columns) are not. The color represents a normalized measure of the HCP concentration and white represents no identification (for which a concentration of zero was assigned to perform the clustering analysis). HCP concentrations were computed in units of ppm on a total mass basis (i.e., ng HCP / mg total) and species-specific normalization was used (based on the minimum and maximum concentrations observed for the given HCP across all samples).

Around 160 HCPs that were identified in this work have also been observed in prior studies and may be generally considered difficult to remove [9,10,28,59–63]; clustered heat maps for this subset of HCPs are shown in Supplementary Figures S19 and S20 (with and without adsorbate pool data, respectively). Among those species there are 36 HCPs that are considered to represent a relatively high risk to therapeutic safety or product stability [10], which are listed in Figure 10 and Supplementary Figure S21. As was observed with the full set of HCPs, the highest concentrations of the high-risk species were generally observed in the Poros 50 HQ adsorbate and in the Capto Q samples when the analysis is restricted to the flow-through fractions (Supplementary Figure S21). Up to 53% loading (~530 mg mAb/ml column) there were four and six detectable high-risk HCPs that were observed to break through on Poros XQ and Poros 50 HQ, respectively, and the safety concerns for all but one of these HCPs are related to immunogenicity. All of the species that were detected in the Poros 50 HQ flow-through were also detected in the Poros XQ flow-through in similar patterns. HCPs that were detected in earlier flow-through fractions but not in later fractions emphasize the ambiguities of false negatives in the proteomic analysis of mAb solutions that contain dilute impurities that may be near the limit of detection [22].

Figure 10.

Clustered heat map of HCP concentrations for species that are considered to represent a relatively high therapeutic risk [10]. Each HCP (row) is annotated with its known or expected deleterious impact type on product quality. Rows, columns and heat map colors are as described in Figure 9.

Distributions of HCP concentrations and computed isoelectric points in each AEX sample are shown in Figure 11. Comparable molecular mass distributions are given in Supplementary Figure S22 for reference. Many of the species that were detected in flow-through fractions were present in sub-ppm quantities. Only two detectable species (elongation factor 1-α1 and peroxiredoxin-1) were present in the initial breakthrough on Poros 50 HQ and only one other detectable species (serine protease HTRA1 isoform X3) was present in the corresponding Poros XQ sample. The computed isoelectric point for each of these species is > 9, which is consistent with their early breakthrough in AEX chromatography and suggests that some quantity of these species may exist in free solution apart from aggregates. The concentration profiles (Figure 10 and Supplementary Figures S19 – S21) indicate that these species were nonetheless partially retained by the Poros resins. Other species that were detected in the adsorbate samples had comparably high isoelectric points; aggregates may be implicated in the retention of such species that would not otherwise be expected to adsorb to AEX resins at the load pH of 7.8.

Figure 11.

Violin plots of the distributions of logarithmically transformed HCP concentrations (top) and computed isoelectric points (bottom) for all HCPs in each AEX sample. PAVIN feed data that were acquired in a previous analysis are also shown for reference [22]. Horizontal lines represent individual observations and grey shaded regions represent smoothed probability distributions that were truncated at the observation extrema. Smoothed probability distributions were estimated using a Gaussian kernel with Scott’s rule [54] as implemented in Seaborn and should be interpreted with caution (i.e., used only as a visual tool to facilitate comparisons) for samples that contain only a few HCPs.

3.3. Relevance to depth filtration

A scale-down experiment was performed with HCCF and 90ZB depth-filter media to illustrate the potential utility of depth filtration for aggregate clearance (Figure 12), analogous to the protein A and AEX experiments. As in the flow-through AEX experiments, the mAb and SPF components broke through immediately but the large aggregates were partially retained, which may be due to the AEX-like functionality of the media’s polymeric binder [64]. Unlike the AEX experiments, the small aggregates were initially retained beyond what could be explained by incomplete SEC resolution, which may be due to a combination of the polymeric binder and the CEX-like functionality of the filter aid. The capacity for aggregate retention was nonetheless fairly low, which partially motivated the 50% dilution of the HCCF feedstock to facilitate breakthrough observations, and the HCCF had already been clarified before its use in this experiment. Cellular debris and nucleic acids in unclarified HCCF are expected to compete for adsorption sites that could potentially retain aggregates, which would decrease capacity further, but hybrid purifiers with explicit AEX functionality would likely have greater capacities for the retention of large aggregates. Regardless, this demonstrates that aggregate-mediated HCP persistence is relevant to depth filtration and attention to this phenomenon may be useful for improving HCP clearance. As suggested previously this could potentially be advantageous for maximizing both the performance and lifetime of protein A columns [13,19].

Figure 12.

Breakthrough behavior of HCCF aggregates, mAb and small protein fraction (SPF) components during depth filtration on 90ZB media. The HCCF feed had been diluted 50% in the equilibration buffer and loading proceeded to ~125 L/m². Ordinate scales are as described in Figure 4A.

4. Conclusions

Aggregates partially mediate HCP persistence through several downstream operations including depth filtration, protein A chromatography and flow-through AEX. Depth filtration has the potential to remove some of these HMW impurities but with limited capacity. Aggregates adsorb specifically in protein A chromatography and co-elute to a significant extent with the mAb product; the aggregate concentrations can also be relatively high in the elution tail, which could possibly be leveraged to improve clearance using altered pool windows and recycling techniques. The effectiveness of additives for washing aggregates off the protein A column depends partially on competitive adsorption between the aggregates and the mAb, which may result in trade-offs between HCP clearance and preserving the mAb yield. Large aggregates that persist into the protein A eluate can be mostly cleared in flow-through AEX chromatography if the resin is selected appropriately, which appears to depend primarily on the resin surface chemistry and pore structure, but the AEX step appears to offer negligible clearance of small aggregates. The aggregate content correlates broadly with the HCP concentration measured using ELISA and the number of HCPs that may be identified in proteomic analysis and may serve as a convenient albeit imperfect surrogate for assessing HCP persistence through polishing operations. Moreover, the linkage between the aggregates and HCP content and the disparate behavior of the aggregates compared to the mAb in different downstream steps reinforces prior suggestions regarding optimal paths to improved control strategies for HCP clearance [13].

Highlights.

• HCP persistence is partially mediated by aggregates in mAb solutions

• Aggregates compete with mAb to adsorb specifically in protein A chromatography

• Aggregates may co-elute with the mAb in protein A chromatography

• Aggregate and HCP clearance may be correlated in flow-through AEX polishing

• Relatively large aggregates are differentially retained by AEX resins