3.1. Breakthrough Behavior and Dynamic Binding Capacity
Initial dynamic binding capacity experiments were performed using 5 mg/mL hIgG.
Figure 1A–D shows the resulting breakthrough curves. Analysis of the breakthrough curves can indicate the dominant transport mechanism in each device. For the Purilogics Purexa™ PrA and Cytiva HiTrap Fibro™ PrismA membranes, the breakthrough curves from 5 to 60 s residence time were nearly overlapping, which is characteristic of convective transport of proteins through the large membrane pores to binding sites [
17,
18,
19,
20]. The DBC
10 values were 71.0 ± 1.8 mg/mL and 69.7 ± 1.3 mg/mL. The Sartobind
® Protein A device also showed flow rate independent breakthrough from 12 to 60 s residence time. It had the lowest DBC
10 of 9.6 ± 2.8 mg/mL and did experience early breakthrough at 5 s residence time.
Of the devices tested, the Gore
® Protein Capture Device, showed the greatest variation in performance with flow rate. The dynamic binding capacity increases by 119% from 19.9 to 43.6 mg/mL by increasing residence time from 5 to 60 s. Flow rate-dependent binding implies that pore diffusion is the limiting rate of transport. A recent study of current Protein A resin beads [
13] estimated that the pore radius can vary between 30 and 60 nm, and the particle diameter can vary between 38 and 116 µm, accounting for standard deviation. According to the patent literature, the Gore
® Protein Capture Device utilizes embedded porous silica particles with a particle diameter of 16–24 µm and pore diameter of 100 nm [
21]. Although the small particle diameter reduces the average length of diffusion, the breakthrough data suggest that the Protein A ligand is immobilized in the pores of the silica beads. Minutes of residence time are required to overcome the time scale for diffusion, similar to the limitation observed with Protein A resin beads.
The measured DBC
10 values are comparable to state-of-the-art Protein A resin beads like the MabSelect PrismA that has DBC
10 values of about 80 mg/mL resin with 6 min residence time using hIgG as feed [
13,
22]. Remarkably, the research scale Purilogics Purexa™ PrA and Cytiva HiTrap™ Fibro PrismA membrane columns can reach 70 mg/mL DBC
10 at 5 s residence time. These devices represent breakthrough technologies for the rapid capture step purification of mAbs and related products.
Further analysis of breakthrough curves highlights the impacts of flow distribution and loading concentration on DBC
10.
Figure 1A,B for Cytiva HiTrap™ Fibro PrismA and Purilogics Purexa™ PrA membranes shows an unexpected increase in breakthrough capacity at 120 s residence time. Given that the breakthrough behavior from 5 to 60 s residence time is consistent with convection dominated transport, the observed increase in capacity at 120 s residence time likely is due to better flow distribution in the membrane device, as observed in acetone tracer experiments where sharper peaks are observed at longer residence times (
Figure S2 in Supplementary Materials).
Figure 2A,B shows the effect of feed concentration on DBC
10. For all devices, there is an increase in capacity as concentration increases. The Purilogics Purexa™ PrA and Sartorius Sartobind
® Protein A membranes experience >50% increase in capacity when loading concentration increases from 2 to 5 mg/mL hIgG at 12 s residence time, while the Cytiva HiTrap™ Fibro PrismA and Gore
® Protein Capture Device experience approximately 25% increase. At 120 s residence time, the Sartorius Sartobind
® Protein A membrane does not show a significant increase in DBC
10 while Cytiva HiTrap™ Fibro PrismA and Purilogics Purexa™ PrA show ~40% increase. The Gore
® Protein Capture Device shows the highest increase (80%) with respect to changing load concentration.
The reasons for protein-dependent binding behavior may differ for all membranes tested. For membranes like Purilogics Purexa™ PrA and Cytiva HiTrap™ Fibro PrismA with high dead volume, feed dilution at the inlet may be significant enough to lower the feed concentration into the linear region of Langmuir isotherm. This could result in concentration-dependent breakthrough behavior without affecting the mass transfer significantly. Similar behavior has been observed for other affinity and ion-exchange membrane adsorbers. Grunberg et al. showed that for Convecdiff Protein A membranes from Sartorius, DBC
10 increases at higher titers [
23,
24]. No explanation was given for the behavior. For ion-exchange membranes, similar behavior was found and explained with the hypothesis that higher concentration improves access to ligands by overcoming film diffusion in membrane pores and combating competitive phosphate binding to ligands [
25,
26,
27].
For Gore
® Protein Capture Device, which has low dead volume, the concentration dependent binding may be attributed to diffusion related transport. Similar behavior has been observed for Protein A resins. Natarajan et al. [
28] evaluated the effect of feed concentration on DBC of Prosep
® Ultra Plus columns, and found higher DBC at higher feed concentrations, but only for longer residence times. This concentration dependent DBC behavior was explained as a result of non-equilibrium mass transfer effects, i.e., there is a maximum in DBC when mass transfer is controlled by surface diffusion rather than pore diffusion. Other studies also have found increasing DBC at higher feed concentrations for different Protein A resins [
29,
30].
The effect of load concentration reported for these chromatography stationary phases indicates that it is an important parameter in determining DBC values for affinity membranes. Furthermore, for future process robustness and productivity determinations, load concentration is an important parameter to consider. Further insight into the exact mass transfer or dispersive causes of concentration dependent binding may be obtained by building computational fluid dynamics models of protein concentration profiles in the membrane adsorbers. Such analysis is beyond the scope of this study.
3.2. Equilibrium Binding Capacity Measured in Static and Dynamic Modes
Static binding capacity experiments were performed and data were fit to the Langmuir isotherm model to determine the maximum binding capacity (q
max) and the apparent dissociation equilibrium constant (K
d) describing protein adsorption. For the Purilogics Purexa™ PrA and Sartobind
® Protein A membranes, these parameters were determined in batch mode. For the Cytiva HiTrap™ Fibro PrismA device and the Gore
® Protein Capture Device, binding capacity studies could not be performed in batch mode. For the Cytiva HiTrap™ Fibro PrismA device, the electrospun membranes were brittle and immediately lost shape upon opening the device. It was not possible to get an accurate measurement of the membrane volume. For the Gore
® Protein Capture Device, we observed much lower binding capacity than expected in static adsorption mode, when compared with DBC results (
Figure S1 in Supplementary Materials). One possible explanation is that poor wettability of the PTFE membrane, noted in academic literature, may increase film resistance for adsorption and require pressure driven flow for protein adsorption [
31,
32]. For all other investigated membrane adsorbers, the hIgG adsorption kinetics showed typical exponential behavior as reported in other studies involving Protein A membranes and resins [
33,
34].
Figure 3 shows adsorption isotherms for Purilogics Purexa™ PrA and Sartobind
® Protein A membranes, and
Table 4 reports the fitted Langmuir isotherm parameters. The K
d values suggest that the membranes have strong affinities for the Fc-region of hIgG and are typical of Protein A affinity devices evaluated in the literature. Boi et al. [
33] reported a K
d value of 9.34 × 10
−2 mg/mL for a recently developed Protein A membrane. Utilizing polyclonal hIgG, Hahn et al. [
10] compared Protein A resins and reported K
d values between 4.5 × 10
−2 mg/mL and 1.20 × 10
−1 mg/mL. Pabst et al. [
13] determined the K
d values for three mAbs across current state of the art Protein A resins. mAbs 1 and 3 were of IgG1 isotype and mAb 2 was an IgG4 isotype. The reported K
d values ranged from 1.37 × 10
−3 to 7.99 × 10
−3 mg/mL for mAb1, 3.31 × 10
−3 to 2.32 × 10
−2 mg/mL for mab2, and 2.61 × 10
−3 to 1.05 × 10
−2 mg/mL for mab3.
Equilibrium binding capacity values also were measured in dynamic experiments.
Figure 4 shows the equilibrium binding capacities at 5 mg/mL hIgG loading concentration. For the Purilogics Purexa™ PrA and Sartobind
® devices, the equilibrium binding capacities are not statistically different than the q
max measured in batch adsorption mode. Purilogics Purexa™ PrA, Cytiva HiTrap™ Fibro PrismA, and Gore
® Protein Capture Device have the highest capacities for Protein A membrane adsorbers reported to date. These three membranes achieve an hIgG EBC value of approximately 88 mg/mL, while the Sartorius Sartobind
® Protein A membrane reached 14 mg/mL.
There is a common belief that surface area is a key predictor of binding capacity. In this study, we tested that perception.
Table 5 shows the specific surface area measurements of all Protein A devices. The measured specific surface areas did not have a strong correlation with the q
max. The specific surface area of the Gore
® Protein Capture Device, 27.72 m
2/g, is approximately 2.6 times higher than the Purilogics Purexa™ PrA device and 5.65 times greater than the Cytiva HiTrap™ Fibro PrismA device. Despite the large difference in specific surface area, these devices have similar q
max values, suggesting that specific surface area is not the key physical property impacting binding capacity. Other factors need to be considered, such as steric effects of Protein A ligand density [
12], the IgG to Protein A binding stoichiometry after immobilization [
12,
34], Protein A immobilization chemistry [
35], and reduction of Protein A activity during the immobilization procedure [
36]. Evaluating these factors are beyond the scope and feasibility of this study.
The specific surface area of the Gore
® Protein Capture Device and the Cytiva HiTrap™ Fibro PrismA device highlight interesting features about the underlying support structure. The specific surface area of the Gore
® Protein Capture Device is comparable to the typical specific surface area of packed resin beds (30–40 m
2/g) because both formats use porous beads. Where packed resin beds are constructed by simply packing a slurry of porous resin beads into a column, the Gore
® Protein Capture Device achieves high specific surface area by embedding porous silica beads into a PTFE matrix. According to the patent literature, the pore diameter of the Davisil silica beads embedded in the Gore
® Protein Capture Device is 100 nm with an estimated specific surface area of 40 m
2/g (manufacturer information). For the Cytiva HiTrap™ Fibro PrismA, the specific surface area is about half of the Purilogics Purexa™ PrA membrane despite similar pore diameter. For this type of support, electrospinning is done to increase the available surface area to volume ratio. There is a common belief that electrospun fiber supports always offer higher surface areas than conventional macroporous membrane supports. However, the higher surface areas are achieved only at small pore diameters and reduced fiber diameter [
37]. In other words, there is a direct correlation between fiber diameter and pore diameter. This would explain the lower specific surface area for the Cytiva HiTrap™ Fibro PrismA device.
3.3. Elution Behavior
Figure 5A–E shows the EVs for all membrane devices at 120 and 12 s residence times. The distinguishing feature of these elution curves is the increased tailing at shorter residence times.
Table 6 shows the calculated tailing ratio using Equation (7) for all Protein A membranes. For all membranes, the tailing ratio is greater than one, which suggests that some degree of tailing occurs during the elution process. This makes sense as the elution is performed by a sharp drop in pH (to 3), which causes a sharp initial front of eluted protein. The tailing ratio for Purilogics Purexa™ PrA, Cytiva HiTrap Fibro™ Prism A, and Sartorius Sartobind
® Protein A show a significant increase at 12 s RT versus 120 s RT. For the Purilogics Purexa™ PrA, the elution curve at 12 s RT presents a shoulder that shifts the peak maximum to the right. This shift effectively reduces the tailing ratio by increasing the symmetry on both sides of the peak maximum. The Gore Protein Device does not show a significant difference in tailing behavior between 12 and 120 s RT.
A few additional observations can be made about the elution peak shapes. The elution curve of the Gore
® Protein Capture device, shown in
Figure 5C, is flattened at the top. This is due to UV-VIS detector saturation, and has been observed in elution profiles of saturated Protein A resins [
13]. There is a shoulder that appears in
Figure 5B for the Purexa™ PrA 12 s RT elution curve. There is currently no explanation for this phenomenon.
For the Purilogics Purexa™ PrA, Cytiva HiTrap™ Fibro PrismA, and the Sartobind
® Protein A membranes, the EV approximately doubles as the residence time decreases from 120 to 12 s (
Figure 5E). The effect of flow rate on elution curves is studied rarely, but a few papers have noted broadened elution peaks at higher flow rates in membrane adsorbers. Hardick [
38] noted that weak anion-exchange electrospun nanofiber membranes, a similar format to the Cytiva HiTrap™ Fibro PrismA support, showed elution peak broadening as a function of flow rate. Boi et al. [
33] found increased tailing as the elution flow rate increased in Sartobind
® Protein A membranes.
Peak tailing is tied to flow distribution and the specific format of the membrane adsorber device [
39,
40]. All membrane adsorbers evaluated in this study operate in direct-flow format. Madadkar et al. [
41] evaluated the flow distribution in typical direct-flow (dead-end) membrane chromatography devices and concluded that the high aspect ratio (bed diameter to bed height) and the large dead volumes lead to poor fluid residence time distribution in the device. Poor flow distribution then leads to early breakthrough and peak broadening during elution, which is exacerbated at higher flow rates.
As predicted by that study, the device with the highest dead volume to membrane volume ratio, the Purilogics Purexa™ PrA membrane device, showed the highest EV (9.0 CVs at 12 s RT) followed by the Cytiva HiTrap™ Fibro PrismA membrane device (7.4 CVs at 12 s RT), and then the Sartobind
® device (1.9 at 12 s RT). The Gore
® Protein Capture Device showed low EVs that did not change with flow rate, suggesting that the Gore
® device has effective flow distribution. At 12 and 120 s RT, the Gore
® Protein Capture Device maintained an EV of 2.0 and 1.9 CV. By comparison, commercial Protein A resins show EVs between 1.8 and 3.8 CVs at 6 min residence time [
13]. Importantly, the research-scale devices tested are all direct-flow columns. It is anticipated that commercial scale devices will use cassette formats with improved flow distribution properties.
One potential solution to increased tailing is to reduce the flow rate at elution for the membranes with high dead volume. As shown in
Figure 5A–D, a reduction to 120 s residence time in the elution step led to reduction in EV by half for Purilogics Purexa™ PrA, Cytiva HiTrap™ Fibro PrismA and Sartorius Sartobind
® Protein A membranes. The reduction in EV is also mirrored by acetone-pulsing experiments where reduced tailing was observed at higher residence times (
Figure S2 in Supplementary Materials).
Another alternative is to alter the membrane holder design to allow for a better flow distribution. Ghosh et al. [
42] investigated one new device design where the incoming fluid flows laterally across the membrane stack, similar to crossflow filtration. Computational fluid models showed more uniform residence time distributions for the fluid [
43]. Results from the models were supported by experimental results showing increased DBC
10 and smaller EVs for a laterally fed membrane of the support, bed volume and dead volume to a direct-flow membrane chromatography device or a radial flow membrane chromatography device [
41,
44].
3.5. Permeability and Pressure Drop
Figure 7A shows absolute pressure drop versus residence time for Protein A membranes. The absolute pressure drop values of these research scale devices are low, remaining below 2 bar at residence times as short as 12 s.
The pressure profile for resin columns typically is characterized by graphing the pressure drop normalized by bed height versus the linear velocity, and the permeability typically is calculated from the slope using Darcy’s Law. For this study,
Figure 7B shows pressure drop per thickness versus linear velocity for membrane adsorbers, and
Table 5 presents the calculated permeabilities. Permeabilities measured in this study ranged from 0.62 × 10
−15 m
2 to 6.19 × 10
−15 m
2, with Cytiva HiTrap™ Fibro PrismA being the lowest permeability device followed by Purilogics Purexa™ PrA, Gore
® Protein Capture Device, and Sartorius Sartobind
® Protein A. Lower permeabilities were observed for smaller pore diameter membranes (
Table 5). Comparatively, Protein A resin permeabilities, calculated from pressure profiles given in a recent study by Pabst et al., are in the range from 4.27 × 10
−12 m
2 to 1.22 × 10
−12 m
2.
Comparing the permeabilities of resins versus membranes, Protein A resin bed permeabilities are higher than Protein A membranes. This finding is surprising due to the typical flow profile of Protein A membranes versus resins. Protein A membranes are typically operated at seconds of residence time where their absolute pressure drop is still low (<2 bar at residence time of 10 s). On the other hand, resins are operated at minutes of residence time because higher flow rates result in even higher pressure drops due to bed compaction. For comparison, a typical research-scale protein A resin having 10 cm bed height shows 3–4 bar of absolute pressure drop at 120 s residence time, whereas the Protein A membranes in this study with bed heights as high as 0.4 cm show less than 0.6 bar of absolute pressure drop at the same residence time/linear velocity [
13]. Two important conclusions can be drawn from the discrepancy between absolute pressure drops and permeabilities. Firstly, that the higher absolute pressure drop found in Protein A resins is a function of the bed height rather than the permeability of the resin bed itself. This conclusion agrees with a study by Herigstad et al. that showed that the permeability of a monolith, 5.74 × 10
−15 m
2, was lower than that of MabSelect, 9.5 × 10
−12 m
2 [
15], despite the lower absolute pressure drop for the monolith bed. Secondly, for membranes, a lower absolute pressure drop is a function of shorter bed height rather than the permeability of the material; and, even more importantly, the ability to operate at seconds of residence time despite lower permeability is due to the device format, i.e., a large pore structure and an aspect ratio with large diameter to bed height.
This important effect of device format is also highlighted by the unique nature of the composite Gore
® Protein Capture Device. It embeds silica beads in the PTFE matrix in a way that does not allow compression at high flow rates. The use of silica beads themselves adds rigidity, as it is harder than materials like agarose and polymethacrylate, and increases surface area for immobilization and binding [
45]. The resulting device is a composite with a flow profile that allows for lower back pressure at seconds of residence time (where resin beds cannot operate) but a binding profile like a resin bed, as it is flow dependent.
3.6. Capture from Clarified Cell Culture Harvest
Membranes were challenged by purifying an IgG1 mAb directly from clarified cell culture supernatant. The mAb concentration was 0.9 mg/mL as measured by a Cedex Bioanalyzer.
Figure 8A–D shows the results of the process metrics: yield, EV, HCP clearance and dsDNA clearance. All devices had yields at or above 80%. Measured EVs using mAb followed the same trend as using the model hIgG. HCP LRVs were between 1.37 and 1.87 from a feed that had 513,333 ± 62,186 ng/mg mAb. Using EV and yield, final HCP concentrations (in ng HCP/mg mAb) were calculated to be 1476 ± 108 for Purexa™ PrA, 3385 ± 372 for Cytiva HiTrap™ Fibro PrismA, 4373 ± 287 for Gore
® Protein Capture Device, and 3182 ± 243 for Sartorius Sartorbind
® Protein A. The dsDNA LRVs were between 1.46 and 1.78 from a starting concentration of 27,333 ± 3456 ng/mg mAb. According to FDA regulations, the final product should contain less than 100 pg of host cell DNA per dose and 100 ppm of HCP per mg of antibody product [
46,
47]. High clearance in the Protein A step is critical towards meeting those criteria.
Evaluating all metrics comparatively, no one membrane performs best in all categories. The Gore® Protein Capture Device has a low EV but is inferior to Purilogics Purexa™ PrA, Cytiva HiTrap™ Fibro PrismA, and Sartorius Sartorbind® Protein A in all other metrics. The Cytiva HiTrap™ Fibro PrismA membrane has the highest yield but is inferior to the Purilogics Purexa™ PrA, and Sartorius Sartobind® Protein A devices in HCP clearance, and has poorer EV than Gore® Protein Capture Device and Sartorius Sartobind® Protein A. The Purilogics Purexa™ PrA and Sartorius Sartobind® Protein A membranes have the best HCP clearance but the Purilogics Purexa™ PrA membrane has the highest EV and the Sartorius Sartobind® Protein has the lowest binding capacity, which means that each run results in about 85% less mass of mAb processed compared to the Purilogics Purexa™ PrA, and Cytiva HiTrap™ Fibro PrismA devices.
Outside of EV performance, which is tied to dead volume and flow distribution (as previously discussed), the yield and impurity clearance metrics cannot be linked directly to the measured physical characteristics of each device. Nevertheless, the yield and overall clearance capacity of the membrane devices can be compared to typical Protein A resin columns. According to a recent study, yields from current state-of-the-art Protein A resins are between 80 and 90%. HCP and DNA LRVs can vary widely depending on the molecule and the use of in-house assays versus generic commercial HCP kits. In this study, the HCP and DNA kit used were generic and may result in lower values. In the cited study, in-house assays with antibodies raised against the HCP from their CHO cell lines were used, and the resulting HCP and DNA LRVs were between 2 and 3 and 2.5 and 3.5 respectively [
13].