Using extensional flow to reveal diverse aggregation landscapes for three IgG1 molecules

Abstract Monoclonal antibodies (mAbs) currently dominate the biopharmaceutical sector due to their potency and efficacy against a range of disease targets. These proteinaceous therapeutics are, however, susceptible to unfolding, mis‐folding, and aggregation by environmental perturbations. Aggregation thus poses an enormous challenge to biopharmaceutical development, production, formulation, and storage. Hydrodynamic forces have also been linked to aggregation, but the ability of different flow fields (e.g., shear and extensional flow) to trigger aggregation has remained unclear. To address this question, we previously developed a device that allows the degree of extensional flow to be controlled. Using this device we demonstrated that mAbs are particularly sensitive to the force exerted as a result of this flow‐field. Here, to investigate the utility of this device to bio‐process/biopharmaceutical development, we quantify the effects of the flow field and protein concentration on the aggregation of three mAbs. We show that the response surface of mAbs is distinct from that of bovine serum albumin (BSA) and also that mAbs of similar sequence display diverse sensitivity to hydrodynamic flow. Finally, we show that flow‐induced aggregation of each mAb is ameliorated by different buffers, opening up the possibility of using the device as a formulation tool. Perturbation of the native state by extensional flow may thus allow identification of aggregation‐resistant mAb candidates, their bio‐process parameters and formulation to be optimized earlier in the drug‐discovery pipeline using sub‐milligram quantities of material.

One barrier that slows or even halts the process of bringing mAbs to market is aggregation (Buss, Henderson, McFarlane, Shenton, & de Haan, 2012) due to unfolding (partial or complete) or mis-folding (Mahler, Friess, Grauschopf, & Kiese, 2009;Roberts, 2014). While generally deleterious to protein function, aggregation is particularly problematic to the biopharmaceutical industry as aggregates have been linked to immunogenic reactions in patients (Büttel et al., 2011) and to shortened therapeutic half-life (Dobson et al., 2016). In addition, the presence of aggregates during development can lead to a decreased yield and an increase in time to market, due to the need to optimize manufacturing conditions/formulation (Cromwell, Hilario, & Jacobson, 2006;Zurdo et al., 2015). mAb-based biologics are susceptible to aggregation throughout their lifetime, from over-expression in the cell (Kramarczyk, Kelley, & Coffman, 2008) and downstream processing (Skamris et al., 2016;Yu et al., 2016) to the final fill-finish step at high concentration (Cromwell et al., 2006;Rathore & Rajan, 2008). In contrast to the effects of temperature and pH (reviewed in Roberts, 2014), the effects of biopharmaceutical manufacture and transport stresses, such as air-water interfaces (Bee et al., 2012;Maa & Hsu, 1997) and hydrodynamic forces (Biddlecombe et al., 2007;Brückl, Schröder, Scheler, Hahn, & Sonderegger, 2016), are relatively under-studied (Bekard, Asimakis, Bertolini, & Dunstan, 2011;Thomas & Geer, 2011).
Differences in accelerated and innate aggregation propensities (Goldberg et al., 2017) may arise because the acceleration method increases the relative flux through certain pathways which are distinct to those traversed during production or upon storage (Chakroun, Hilton, Ahmad, Platt, & Dalby, 2016;Luo et al., 2011;Phillips et al., 2017;van der Kant et al., 2017). There is thus a need to develop stress tests that more closely replicate the conformational ensemble generated during processing and transport. In light of this, we and others have shown that hydrodynamic extensional flow fields encountered during the nano-filtration, pumping, and fill-finish steps of bio-processing can trigger protein aggregation (Charm & Wong, 1981;Dobson et al., 2017;Simon, Krause, Weber, & Peukert, 2011;Wolfrum, Weichsel, Siedler, Weber, & Peukert, 2017). Using a reciprocating extensional and shear flow device (EFD) (Figure 1a), we showed that extensional flow fields can induce the conformational unfolding/remodeling of bovine serum albumin (BSA, Figure 1b), leading to aggregation that was characterized and quantified by an array of biophysical techniques including DLS, NTA, and TEM (Dobson et al., 2017). By subjecting five other globular proteins that varied in sequence, size, and structure to such flow stresses, we demonstrated that the aggregation propensity of proteins differed based on their fold, sequence, and the fluid fields to which they are subjected (Dobson et al., 2017). For mAb-based biotherapeutic scaffolds we observed a wide-range of sensitivity to flow-induced aggregation under identical conditions (strain rate, pass number, protein concentration and buffer), dependent on the protein sequence (Dobson et al., 2017). Accordingly, the aggregation-prone antibody (MED-I1912_WFL, WFL herein, Figure 1b) was found to be most sensitive to extensional flow, while its rationally engineered aggregationresistant derivative (MEDI1912_STT, STT herein, Figure 1b (Dobson et al., 2016) showed markedly decreased aggregation (∼85 and ∼5% aggregation, respectively). An unrelated mAb, mAb1, showed intermediate behavior (Dobson et al., 2017). For BSA, the extent of aggregation is dependent on the magnitude of the strain rate, the total exposure time to the extensional flow event (controlled by both the number of passes and plunger velocity) and the protein concentration, producing a complex aggregation landscape. These observations suggest that the flow-induced aggregation of mAbs may proceed via a common mechanism, resulting in the formation of insoluble, amorphous aggregates (Figure 1c). Here, using the EFD and a protein pelleting assay, we have mapped in detail the aggregation landscape of WFL, STT, and mAb1 after exposure to defined fluid fields, exposure times, and protein concentrations. These experiments reveal distinct WILLIS ET AL.

| Protein sample preparation
The proteins used in the study (WFL, STT, and mAb1) were provided by MedImmune Ltd, Cambridge UK, as described previously (Dobson et al., 2016(Dobson et al., , 2017. Buffer reagents were obtained from Sigma-Aldrich (Gillingham, UK), except sodium phosphate dibasic (BDH Lab Supplies, Bristol, UK) and L-arginine (Acros Organics, Geel, Belgium). In all experiments, antibodies were dialyzed into the appropriate 0.22 μm-filtered and de-gassed buffer and subsequently used in extensional flow experiments.
Bovine Serum Albumin (BSA) was prepared as described previously (Dobson et al., 2017). Except for the buffer screen, all mAb experiments were performed in 150 mM ammonium acetate buffer, pH 6.0. For experiments involving BSA, 25 mM ammonium acetate buffer pH 5.1 was used.

| Extensional flow device (EFD) and stress experiments
Full details of the extensional flow device including its validation using computational fluid dynamics (CFD) are described elsewhere  (Majorek et al., 2012), WFL and STT. The surface-view scFv models of WFL and STT were built from the structure of their parent, MEDI_578 (PDB ID: 5jz7) (Dobson et al., 2016). The surface for each protein is color-coded according to the CamSol structurally corrected solubility profiles of WFL and STT to show regions of poor solubility (red) and enhanced solubility (blue) (Supplementary Methods). (c) Proposed mechanism of flow-induced mAb aggregation. (i) The native protein (blue) is perturbed into an aggregation-prone state (red), the relative level of which is dependent on the fluid field and the protein studied. (ii) This can either re-fold to the native state or proceed along the aggregation pathway (iii) to form irreversible, insoluble aggregates (iv). The apparent rate constants (k f and k r ) represent the rate of formation/refolding of the aggregation-prone state, while k f ' and k r ' represent the concentration-dependent rate of oligomer formation and the unimolecular off-rate for oligomer dissociation, respectively. These rate constants remains to be determined (Dobson et al., 2017). Briefly, the EFD consists of two modified Hamilton gas-tight syringes (inner diameter = 4.6 mm) connected by a 0.3 mm inner-diameter borosilicate glass capillary. The capillary length was 75 mm in all experiments except shear-length variation experiments, where a ceramic cutter was used to shorten capillaries to 50 mm (2/3 length) or 37.5 mm (1/2 length) followed by flamefinishing. All protein solutions were 0.22 μm-filtered prior to loading into the device, and any air-bubbles ejected prior to assembling the EFD. The protein solution was shuttled between the syringes at the desired plunger velocity (determining the strain rate and the shear rate) for a given number of passes (determining exposure time); see Supplementary Table S1 for plunger velocities and concomitant center-line strain and shear rates. The plungers were driven by a stepper motor controlled by an Arduino microcontroller. After subjecting the protein to the desired number of passes, the EFD was dissembled and the protein solution removed for quantification of aggregation (insoluble protein pelleting assay, below). All experiments were performed at a concentration of 0.5 mg ml −1 for mAbs and 5 mg ml −1 for BSA unless stated otherwise. As a control, a sample was incubated under ambient conditions (quiescent) alongside the stressed sample for the duration of the experiment and subsequently subjected to the same analysis.

| Insoluble protein pelleting assay
The insoluble protein formed after stress in the flow device was quantified using an insoluble protein assay (Dobson et al., 2017

| Mapping the aggregation behavior of WFL and STT
We showed previously that the extent of flow-induced aggregation of different proteins is sensitive to the strain rate and exposure time (and protein concentration) when modulated independently. For example, for BSA, a center-line strain rate of at least 14,634 s −1 (a plunger speed of 10 mm s −1 ) was required before significant quantities of aggregated protein could be detected after 100 passes (data along green line, Figure 2a and Dobson et al., 2017). This suggests that a force threshold, applied from the flow onto the protein in the extensional flow field, has to be overcome to induce local unfolding and subsequent aggregation of the protein.
However, as partial unfolding is assumingly a stochastic process, the extent of aggregation was also found to be pass number dependent ( In contrast to these data, the response surface of WFL shows only a simple monotonic profile (Figure 2c), where aggregation is dependent solely on pass number above 2 mm s −1 . For example, at 8 mm s −1 aggregation increases from ∼35 to 100% upon increasing from 20 to 100 passes (red line, Figure 2c). Despite the velocity independence, hydrodynamic flow is still necessary to induce aggregation, as minimal aggregation is observed under quiescent conditions. Above the 2 mm s −1 threshold, the number of passes through the device dictates the amount of aggregate observed (∼90-100% after 100 passes). For WFL, these data suggest that (i) the aggregation-prone patches of the protein are exposed to the solvent upon experiencing a relatively small strain rate (6,031 s −1 ) or that (ii) the protein maintains its aggregationprone state after traversing the extensional flow region or that (iii) the affinity/on-rate for WFL oligomerization is higher relative to STT.
For both WFL and STT, not all strain rates bring about aggregation.
When WFL was stressed at 0.5 mm s −1 for 20 passes, minimal aggregation was observed, whereas when STT was stressed for 50 passes at 0.5 mm s −1 no aggregation was observed ( Figure S1). Flowinduced protein aggregation thus appears to be a complex function of the force that the protein is exposed to and the number of times the protein experiences exposure to this force, convoluted with the onand off-rates of aggregation. This multi-factorial dependence results in distinct aggregation landscapes for even very closely related proteins.

| STT and WFL exhibit different concentration dependencies of aggregation
The data above suggest a mechanism whereby aggregation is driven by inter-molecular collisions between activated species and/or with groundstate proteins. Next, we sought to examine whether increased protein (Supplementary Methods). The CDR1 of STT is therefore predicted to be markedly more soluble (greater positive CamSol score) than the same region in WFL (Sormanni et al., 2015). These data suggest either that flow perturbs the structure of these CDRs, exposing aggregation-prone regions to the solvent to different extents for WFL and STT, and/or that the

| Aggregation-prone mAbs are sensitive to high shear
The EFD generates a well-defined region of extensional flow at the point For BSA, this possibility was obviated as aggregation was found to be independent of capillary length, using flow conditions under which BSA is most susceptible to aggregation (plunger velocity of 16 mm s −1 ) (Dobson et al., 2017). The data for WFL and STT above, however, show that the aggregation landscapes for mAbs differ from that of stable, globular proteins such as BSA. To determine the effect of shear flow on mAb aggregation, WFL and STT were subjected to 20 and 100 passes (respectively, to generate a similar aggregate yield) at a plunger velocity of 8 mm s −1 , using capillary lengths of 37.5, 50, and 75 mm (Methods). The extent of aggregation was then quantified using the pelleting assay. Figure 4 shows that the aggregation of STT is independent of capillary length, but that WFL aggregation decreases from 37% to 15% upon halving the capillary length, indicating that shear flow contributes to the aggregation of this protein. Further work is required to identify the underlying mechanism behind this observation. For example, this effect may arise because the key aggregation-prone region of WFL may be exposed at a lower strain rate threshold or because the presence of a shear region may slow the relaxation time of the activated species. Irrespective of the mechanism, the results show that these highly homologous mAbs display diverse behavior under flow.

| mAb1 exhibits traits of both WFL and STT in response to flow
While comparing the biophysical properties of STT and WFL under quiescent and flow conditions is an extremely powerful method to delineate the mechanism and determinants of flow-induced aggregation, these proteins may represent extremes of the mAb scaffolds usually encountered during bio-processing. We thus complemented the above data by examining the (plunger velocity (2-16 mms −1 ), pass number (0-100) and shear exposure time (37.5, 50, and 75 mm) dependencies of mAb1, an unrelated IgG1 that has ∼72% sequence identity to WFL and STT (Dobson et al., 2016(Dobson et al., , 2017. The data in Figure 5a show that, similarly to WFL and STT, the extent of mAb1 aggregation is directly dependent on pass number and that mAb1 displays intermediate sensitivity to pass number (rank order: WFL>mAb1>STT). Like STT (Figure 4), the aggregation of mAb1 is independent of the length of the shear region (31.1 ± 1% and 29.6 ± 2% for full-and half-length capillaries, respectively) (Figure 5b). mAb1 also displays a STT-like response in aggregation level to an increase in plunger velocity, with low levels of aggregation observed at 2 mm s −1 and increased levels of aggregate with increasing strain rate (Figure 5c). Comparison of these three IgG1s reveal that their primary sequence modulates flow-induced aggregation, producing complex and distinct landscapes dependent on the sequence and flow conditions used.

| Using the EFD as a buffer screening tool
The  Shear-length dependence on WFL and STT aggregation. Decreasing the length of the borosilicate glass capillary decreases the length of time proteins are exposed to high shear downstream of the extension region. WFL was stressed for 20 passes at a plunger velocity of 8 mm s −1 , while STT was stressed for 100 passes at the same plunger velocity. The capillary wall shear rate in the capillary = 50,375 s −1 . Bars show percentage insoluble WFL (black bars) and STT (gray bars) formed following stress using capillaries of differing length. Capillary length was either 75 mm (full-length), 50 mm (2/3), or 37.5 mm (half-length). The quiescent samples were not exposed to flow stress. Both proteins were stressed at a concentration of 0.5 mg ml −1 in 150 mM ammonium acetate buffer, pH 6.0. Error bars indicate the error from two independent experiments assess the effect of buffer composition on stability under flow, STT, WFL, and mAb1 were dialyzed into four buffers in which the quiescent stability of STT and WFL is known, as well as arginine + succinate buffer (Methods) (Dobson et al., 2016). Each mAb (0.5 mg ml −1 ) was then stressed for 100 passes at 8 mm s −1 and the resulting aggregation quantified. The results show that the ability of each buffer to modulate aggregation is dependent on the mAb sequence ( Figure 6). For example, the high flow sensitivity of WFL aggregation was largely maintained in all buffers, with arginine + succinate a noteworthy exception (aggregation decreased from ∼86% in histidine, acetate, succinate, and phosphate buffers to 20% in arginine + succinate buffer). Interestingly, both STT and mAb1 displayed buffer-dependent levels of flow-induced aggregation. These effects were mAbdependent, with STT showing a greater extent of aggregation in histidine compared with phosphate and vice-versa for mAb1. Given arginine's widespread use as a stabilizing excipient (Baynes, Wang, & Trout, 2005;Kim, Hada, Thapa, & Jeong, 2016), it is also of note that all mAbs exhibit greatly suppressed aggregation in arginine + succinate buffer. Extensional flow-induced mAb aggregation is thus sensitive to both the buffer used and the mAb sequence. Previous work to identify the buffer which afforded STT the greatest stability used dynamic light scattering. This required higher protein concentrations (4 mg ml −1 ) (Dobson et al., 2016) and showed limited ability to differentiate between buffers. The results presented here suggest that extensional flow stress tests could be used to screen the stability of mAbs in different buffer environments and hence to optimize conditions in the pipeline for biologics production, using as little as 0.25 mg of protein per experiment.

| DISCUSSION
Examining the effect of both the magnitude and duration of defined hydrodynamic stresses on protein solutions has been carried out on several globular proteins (Dobson et al., 2017;Simon et al., 2011).
Performing such experiments on proteins of more complex topology and inherent commercial value, such as mAbs, is useful to both the biotechnology industry and the wider scientific community. We show here that the aggregation response of WFL and STT to flow are distinct, despite these mAbs differing by only three residues per V H domain. WFL formed large amounts of insoluble protein under flow, even when exposed to low strain rates for as few as 20 passes. By contrast, the STT aggregation landscape shows a much larger region where few aggregates form. Such an analysis could inform Using the EFD to screen optimal buffers for WFL, STT, and mAb1. WFL (black bars), STT (gray bars), and mAb1 (white bars) were dialyzed into 10 mM L-histidine pH 6.0, 10 mM sodium acetate pH 6.0, 10 mM sodium succinate pH 6.0, 10 mM sodium phosphate pH 7.2, and 125 mM L-arginine + 20 mM sodium succinate pH 6.0 and stressed for 100 passes in the device at a plunger velocity of 8 mm s −1 . Error bars indicate the error from two independent experiments manufacturing practice through relating the strain rates and exposure times experienced in the EFD to plant equipment (e.g., the multiple passes experienced in a tangential flow filtration device (Rosenberg, Hepbildikler, Kuhne, & Winter, 2009)). To do this, studies of plant equipment need to report strain rates due to extensional flow alongside the more commonly reported shear rates (Bee et al., 2009;Charm & Wong, 1981).
It has been suggested that shear flow does not cause damage to the tertiary structure of proteins (Bee et al., 2009), and/or that the shear rates required to do so may be hard to achieve experimentally (Jaspe & Hagen, 2006). We show here, however, that reducing the amount of time the protein is exposed to high shear (by decreasing the length of the capillary), WFL aggregation is diminished, indicating a role for shear in the process (note: at a plunger velocity of 8 mm s −1 , proteins spend ∼18 µs in the extension region, 40 ms in the capillary, and ∼5 s in the syringes (Dobson et al., 2017)). Whether shear alone or an initial extensional flow event prior to shear is required to induce WFL aggregation remains to be seen. The latter scenario could cause a localized unfolding of a protein molecule (e.g., as seen for von Willebrand factor (Lippok et al., 2016) and for BSA in our previous study (Dobson et al., 2017), which may then be susceptible to "tumbling" events under shear flow in the capillary (Smith, Babcock, & Chu, 1999) and rapid self-association in a concentration and timedependent manner.
The mAb1 antibody exhibited "intermediate" behavior when subjected to the same experiments as WFL and STT. mAb1 showed a linear pass number dependence (similar to WFL and STT) but a strainsensitivity and shear-insensitivity akin to BSA and STT (Figures 2a and   2b). Taken together, our experiments point toward a common mechanism of mAb aggregation induced by the flow fields present in our device (Figure 1c) centered on the formation of activated, aggregation-prone species that readily self-associate, forming soluble, and then insoluble aggregates (Dobson et al., 2017;Roberts, 2007;Wang, Nema, & Teagarden, 2010). The flux through the pathway is governed by the ability of extensional flow to activate each native mAb into a perturbed structural state (van der Kant et al., 2017), the affinity of the exposed APR, the rate of relaxation from the activated species (which may itself be modulated by force [Bustamante, Chemla, Forde, & Izhaky, 2004]) and the productive collisional frequency. Consequently, mechanically robust proteins such as BSA (seventeen intramolecular disulfide cross-links in the native state) or, like STT, those with reduced aggregation propensity, require the application of high strain rates and/or pass number to induce appreciable aggregation.
The aggregation behavior of mAbs under hydrodynamic stress is clearly affected by a variety of parameters. The defined nature of the flow environment in the EFD can allow the effects of protein sequence and concentration, surface chemistry (Biddlecombe et al., 2007), strain rate, shear rate and the total exposure time on the observed aggregation to be determined in the absence of other confounding factors such as air-water interfaces or the action of stirring (Fleischman et al., 2017;Joubert et al., 2011;Kiese et al., 2008;Tamizi & Jouyban, 2016;Zhao & Cieplak, 2017). Furthermore, by subjecting mAbs to hydrodynamic stress in different buffer conditions, we have highlighted that the choice of excipient can be another crucial factor in influencing the extent of aggregation, opening up the possibility of using the EFD as a formulation tool.
In summary, by mapping the response of three mAbs to defined hydrodynamic flows, we have demonstrated that the aggregation of these proteins can be minimized by changing processes (e.g., operating under low strain conditions), changing sequence (WFL vs. STT), or by changing buffer conditions.

| CONCLUSION
There is an unmet need to predict the aggregation propensity of proteins of pharmaceutical interest, at an early stage in the development pipeline, in order to maximize their chances of success of transitioning from the bench to the market (Jain et al., 2017).
Various accelerated stress methodologies, such as stirring, heating, and shaking, are employed to make such assessments (Tamizi & Jouyban, 2016). Nonetheless, ranking the probability of failure during bioprocessing remains a significant challenge. Here, we use a well-characterized extensional and shear flow device to subject three mAbs to defined fluid fields (strain rate, shear rate, and exposure time). Quantifying the resultant aggregation, allows estimation of the likelihood of biopharmaceutical aggregation under flow and the identification of buffers that minimize the effects of flow. This method will be a useful addition to the repertoire of tools available to the biopharmaceutical industry to distinguish "manufacturable" proteins from poorer candidates, requiring only small quantities of protein and thus allowing assessment early in the development pipeline.

CONFLICTS OF INTEREST
The authors declare no conflicts of interest.