Structures of multidomain proteins adsorbed on hydrophobic interaction chromatography surfaces

Journal of Chromatography A, 1371 (2014) 204–219

Contents lists available at ScienceDirect

Journal of Chromatography A journal homepage: www.elsevier.com/locate/chroma

Structures of multidomain proteins adsorbed on hydrophobic interaction chromatography surfaces Adrian M. Gospodarek ∗ , Weitong Sun, John P. O’Connell, Erik J. Fernandez Department of Chemical Engineering, University of Virginia, Charlottesville, VA 22904 4741, USA

a r t i c l e

i n f o

Article history: Received 11 July 2014 Received in revised form 8 October 2014 Accepted 25 October 2014 Available online 31 October 2014 Keywords: Hydrophobic interaction chromatography Hydrogen exchange mass spectrometry Protein unfolding Surface induced denaturation

a b s t r a c t In hydrophobic interaction chromatography (HIC), interactions between buried hydrophobic residues and HIC surfaces can cause conformational changes that interfere with separations and cause yield losses. This paper extends our previous investigations of protein unfolding in HIC chromatography by identifying protein structures on HIC surfaces under denaturing conditions and relating them to solution behavior. The thermal unfolding of three model multidomain proteins on three HIC surfaces of differing hydrophobicities was investigated with hydrogen exchange mass spectrometry (HXMS). The data were analyzed to obtain unfolding rates and Gibbs free energies for unfolding of adsorbed proteins. The melting temperatures of the proteins were lowered, but by different amounts, on the different surfaces. In addition, the structures of the proteins on the chromatographic surfaces were similar to the partially unfolded structures produced in the absence of a surface by temperature as well as by chemical denaturants. Finally, it was found that patterns of residue exposure to solvent on different surfaces at different temperatures can be largely superimposed. These findings suggest that protein unfolding on various HIC surfaces might be quantitatively related to protein unfolding in solution and that details of surface unfolding behavior might be generalized. © 2014 Elsevier B.V. All rights reserved.

1. Background and introduction Hydrophobic interaction chromatography (HIC) purifies proteins based on their apparent hydrophobicities, making it a valuable tool in downstream purification [1–3]. HIC stationary phases come in a variety of backbone and hydrophobic ligand chemistries leading to differing degrees of hydrophobicity and protein binding capabilities. HIC resins of greater hydrophobicity are attractive options in downstream purification due to their ability to effectively bind proteins with lower salt concentration requirements. Many HIC operations can be fine tuned to provide the desired separation or recovery without negative effects on the target protein’s stability, allowing it to retain its bioactivity [4,5]. In addition, in certain cases these HIC resins can refold proteins in unfolded states, leading to high bioactivity recovery [6–8]. However, these same HIC resins can also lead to protein unfolding with potentially compromised recovery and loss of function [9–11]. Designing HIC processes with high selectivity and limited unfolding effects is a challenge, especially considering the many variables that influence adsorption, such as resin, salt type and concentration, pH, and temperature

∗ Corresponding author. Tel.: +1 302 373 3935. E-mail address: [email protected] (A.M. Gospodarek). http://dx.doi.org/10.1016/j.chroma.2014.10.080 0021-9673/© 2014 Elsevier B.V. All rights reserved.

[12]. Among the goals of predictive methods is identifying the conditions where target proteins would be sufficiently stable during HIC. While there is evidence that resins of higher hydrophobicity destabilize proteins more than those of lesser hydrophobicity [13–15], no generalized empirical or theoretical framework exists to identify the resins and the mobile phase conditions that can destabilize a protein. In this work, we examine the structures of multidomain proteins during HIC. Although most chromatographers would avoid denaturing conditions in developing HIC operating conditions, we have focused on this region in an effort to gain insight into the nature of conformational sensitivity for multidomain proteins. We hypothesize that the local regions of proteins that unfold when adsorbed are related to those that are unfolded in solution, especially in unfolded intermediates. To test these hypotheses, we have examined the structural details of how three multidomain protein systems unfold on three HIC resins of different hydrophobicity by varying denaturing conditions of temperature and concentration of guanidinium hydrochloride. Finally, we obtain the unfolding rates and free energies of unfolding of different HIC surfaces and compare them. We find that the protein conformational changes from residue solvent exposures have similar patterns, including aggregation, but their extents depend on the system conditions and resin. However,

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

the partially unfolded states on the surface of chromatographic media and in solution are quite similar and the solvent exposure pattern changes on different media can be largely superposed at different temperatures. Finally, we show that the free energy of unfolding for a protein adsorbed on the surface can be determined from guanidinium denaturation monitored with HXMS. 2. Materials and methods 2.1. Materials Human serum transferrin (transferrin), human ␣ antitrypsin (antitrypsin), and bovine serum albumin (BSA) were purchased from Sigma–Aldrich (St. Louis, MO, USA). Deuterium oxide, potassium phosphate, ammonium sulfate, calcium chloride, ethylenediaminetetraacetic acid (EDTA), citric acid, formic acid, trifluoroacetic acid (TFA), and guanidinium hydrochloride (GdnHCl) were purchased from Fisher Scientific (Houston, TX, USA) and were of HPLC grade quality or better. Tris(2 carboxyethyl)phosphine hydrochloride (TCEP) was purchased from Thermo Scientific (Rockford, IL, USA). The Tosoh HIC resins used in this study, Phenyl 650M, Butyl 650M, and Hexyl 650C were purchased from Fisher Scientific (Houston, TX, USA). Ultrafree® MC centrifugal filter units were purchased from Fisher Scientific (Houston, TX, USA) for the separation of supernatant liquid from resin particles. 2.2. Sample preparation and measurements 2.2.1. Temperature studies For control experiments without surfaces, 5 ␮L of 20 mg/mL protein solution were mixed and equilibrated for 2 h with 45 ␮L of deuterated buffer at pH 7.0 at room temperature and at temperatures in 10◦ increments up to 82 ◦ C. The compositions of all buffers can be found in Table S1 in supplementary information. Labeling was done at 10, 20, 30, 40, and 60 min. After labeling, 5 ␮L of quench buffer at ice bath temperature was added, bringing the final solution pH to 2.6, near the pH minimum of the hydrogen deuterium exchange reaction [16]. The samples were kept at room temperature for 40 s before adding 147 ␮L of desorption buffer to the solution, after which they were placed on ice for 2 min and then at room temperature for 40 s. Then, 600 ␮L of sample pump solution were added to dilute the protein and GdnHCl concentrations. Solution phase samples were placed at room temperature for 40 s to match the schedule of the adsorbed phase experiments. Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.chroma. 2014.10.080. For adsorbed phase experiments, 35 ␮L of 5 mg/mL protein solution in working buffer was added to 65 ␮L of resin slurry (50:50 dry resin:working buffer) placed in the inner vessel of an Ultrafree® MC centrifugal filter unit. The samples were allowed to equilibrate overnight to assure adsorption equilibrium. Prior to labeling, the sample was centrifuged at 7400 relative centrifugal force (rcf) for 30 s to form a filtrate. To initiate labeling, 90 ␮L of deuterated buffer and 10 ␮L of undeuterated buffer were added to the inner vessel at room temperatures in 10◦ increments to 82 ◦ C. As with the solution phase samples, labeling times of 10, 20, 30, 40, and 60 min were used. After labeling, 10 ␮L of quench buffer was added to the inner vessel and the filter unit was immediately centrifuged at 7400 rcf for 30 s. The inner vessel was transferred to a new empty filter unit on ice and 200 ␮L of desorption buffer was added. The sample was placed on ice for 2 min and then centrifuged at 7400 rcf for 30 s. Finally, 600 ␮L sample pump solution were added to the filtrate.

205

2.2.2. Guanidinium studies Solution and adsorbed phase experiments for the guanidinium studies were done with similar protocols to the temperature studies. The temperature was held constant at 22 ◦ C while the concentration of GdnHCl in the protein, protein resin samples, and deuterated buffer were varied from 0 to 4.5 M in increments of 0.5 M. Samples were equilibrated overnight prior to labeling. Data were collected in increments of 0.1 M GdnHCl for the concentration range in which solvent exposure changed significantly.

2.2.3. HPLC MS studies Samples were injected into a 200 ␮L stainless steel sample loop using a 500 ␮L glass syringe. Sample pump solution and injected sample at 100 ␮L/min were pumped (LabAllianceTM Series I) through the sample loop and into an immobilized pepsin column (2.1 mm inner diameter by 60 mm length) where proteolytic digestion took place. Peptides exiting the column were trapped, desalted, and concentrated on a C8 desalting column (TR1/25109/02, 1 mm inner diameter by 8 mm length, Michrom Bioresources, Inc.). After this desalting step (6 min), flow was switched from the sample pump to a Surveyor MS HPLC pump to elute the peptides off the C8 column. An XBridge C18 column (186003563, 2.1 inner diameter by 50 mm length, 3.5 ␮m pore size, Waters, Inc.) downstream of the C8 column was used for improved resolution of the large number of peptides in these systems. For the solution and adsorbed phase studies, a short gradient run was employed to minimize back exchange but to still effectively resolve the peptides. Peptide desorption was done with a 17 min gradient of 70% solvent A (ddH2 O, 0.1% formic acid, 0.01% TFA) and 30% solvent B (acetonitrile, 0.8% formic acid) to 40% solvent A, followed by a 2 min gradient from 40% solvent A to 10% solvent A, followed by 4 min at 10% solvent A. Peptides were eluted directly to a LTQ linear ion trap mass spectrometer (Thermo Finnigan, San Jose, CA, USA). Data were collected in a positive ion, profile mode with an ESI voltage of 4.3 kV, a capillary temperature of 250 ◦ C, and sheath gas flow rate of 15 units. The peptides identified in MS/MS experiments are shown elsewhere [17].

2.2.4. Size exclusion chromatography studies Size exclusion chromatography (SEC) was performed to determine if aggregates formed on the surfaces. Protein samples were equilibrated for 1 h on the three different HIC media at a loading of 5.4 mg/mL in a Ultrafree® MC centrifugal filter unit within a 1.5 mL microcentrifuge tube and at temperatures from 22 to 82 ◦ C in increments of 10 ◦ C. The sample was then centrifuged at 7400 rcf for 30 s to remove the supernatant. The filter unit was transferred to a new microcentrifuge tube on ice and 200 ␮L of desorption buffer was added. The sample was placed on ice for 10 min and then centrifuged at 7400 rcf for 30 s. The flowthrough was immediately collected, diluted with 500 ␮L of working buffer and injected into a 500 ␮L sample loop of an AKTA Explorer. This procedure and its volumes and step times were chosen so as to best reproduce the conditions of the MS studies while at the same time ensuring sufficient recovery and detection of protein at 215 nm. A TSK gel G3000SWXL column (7.8 mm inner diameter by 30 cm length, 5 ␮m pore size, TOSOH Bioscience) was equilibrated with 5 column volumes (CVs) of working buffer at 0.5 mL/min prior to sample loading on the column. Sample loading was done at 0.5 mL/min with working buffer at 22 ◦ C and continued until all peaks were detected and the 215 nm signal returned to baseline.

206

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

3. Theory 3.1. Measuring and calculating fractions of labeled peptide Hydrogen deuterium exchange has been used successfully to monitor changes in protein and peptide conformation [18–20]. The extent to which a peptide has been labeled with deuterium is determined by [21] mt − m0 D = N m100 − m0

(1)

where D is the number of deuterated amides, N is the total number of exchange competent residues in a peptide, mt is the mass of a peptide after a given labeling time, m0 is the undeuterated mass of that peptide, and m100 is the fully deuterated mass of that peptide. The value of mt is determined from a first moment analysis of a peptide mass spectrum. In this study, HDExaminer (Sierra Analytics, Modesto, CA) was used. In all cases, Eq. (1) accounts for back exchange during the 12–16 min between sample quenching and introduction into the MS [21].

refolding rate is slow compared to the labeling rate, and molecules that unfold have time to completely label before they become protected again (commonly referred to as the EX1 exchange regime) [21]. For this situation, the observed decrease in D/N protein is limited by the unfolding rate, ku dN = −ku [N] dt

(2)

where [N] is the concentration or fraction of native peptide at a given labeling time. In the mass spectrum, the molecular mass of this peak, MN , is representative of the unlabeled mass plus 1 Da for each residue that is exchange competent under strictly native conditions, such as located on the protein surface. Under the EX1 limit and at concentrations of GdnHCl where the protein molecules transition from all native to a mixture of native and unfolded, a second peak of higher molecular mass appears, representative of unfolded molecules. The molecular mass of this unfolded peak, MU , also stays constant but grows in area with labeling time while the native peak area decreases until it disappears. The weighted mass (centroid) of the two peaks in the transition region, MNU , can then be used to determine [N] from MNU − MN MU − MN

3.2. Unfolding free energies and rates

[N] = 1 −

Solution unfolding free energies and rates can be determined from experimental GdnHCl denaturation curves monitored by HXMS [22–24]. We propose the same method for proteins adsorbed on surfaces. Typically, destabilizing conditions lead to an increase in the population of fully (or partially) unfolded molecules relative to those in native conformation. The steps of the analysis to relate this population shift to unfolding free energies and rates are shown in Fig. 1. First, mass spectra are interpreted to generate the denaturation curves (Fig. 1a). For most regions of tertiary structure the

In the limit where MNU = MU , [N] is 0 and in the limit where MNU = MN , [N] is 1. In this work, experimental D/N values have been measured for different concentrations of GdnHCl using Eq. (1) and mt = MNU to generate the denaturation curves. Next, the denaturation curves are fitted to a sigmoidal equation to allow generation of additional data points by interpolation in the transition region (Fig. 1b). Values of D/N at different GdnHCl concentrations and fixed labeling time are fitted (LSQCURVEFIT, a nonlinear least squares solver in MATLAB® ) to a four parameter sigmoidal equation UD/N − ND/N D = ND/N + N 1 + e−p([GdHCl]−C1/2 )

(3)

(4)

where ND/N is the D/N value of the native peak, UD/N is the D/N value of the unfolded peak, p is a parameter describing the slope of the unfolding transition, and C1/2 is the denaturant concentration at the midpoint of the transition. Next, D/N values generated in MATLAB® from Eq. (4) are used to calculate MNU at specific GdnHCl concentrations using a rearrangement of Eq. (1) with mt = MNU MNU =

D × (m100 − m0 ) + m0 N

(5)

This can then be used to calculate [N] as a function of GdnHCl concentration using Eq. (3) (Fig. 1c). The whole analysis is repeated for each labeling time. Next, generated values of [N] at different labeling times and GdnHCl concentrations can be used to calculate unfolding free energies and rates. Eq. (2) is integrated and solved to obtain ln[N] = −ku t + ln[No ]

Fig. 1. Determination of ku , m, and Gu,o for a reporter peptide. (a) MN and MU are measured along with their peak intensities from the mass spectrum to calculate MNU and D/N. (b) D/N (diamonds) vs. GdnHCl concentration is fitted to a four parameter (ND/N , UD/N , C1/2 , and p) sigmoidal equation (Eq. (4)). (c) [N] (circles) is calculated at various labeling times and GdnHCl concentrations from the sigmoidal equation. (d) For each GdnHCl concentration, [N] is regressed vs. labeling time (Eq. (6)). The slope of the regressed line is −ku and the vertical axis intercept is Gu (squares). (e) Gu is regressed vs. GdnHCl concentration (Eq. (8)). Slope of regressed line is m and vertical axis intercept is Gu,o . 95% confidence intervals of all regressed parameters are propagated to m and Gu,o .

(6)

where [No ] is the fraction of native reporter peptide at zero labeling time. Regression of Eq. (6) with [N] measured at different labeling times, t, yields ku and [No ] (Fig. 1d). Assuming ideal solution, the standard state unfolding free energy, G◦ u , for a given reporter peptide of a protein can then be determined from G◦ u = −RT ln

 [N ]  o 1 − [No ]

(7)

where R is the gas constant and T is temperature. Finally, assuming linear variation of unfolding free energy with GdnHCl concentration [25], the values of Gu calculated at different GdnHCl concentrations can be regressed to determine

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

the unfolding free energy in the absence of guanidinium, G◦ u,o (Fig. 1e) from G◦ u = m[GdnHCl] + G◦ u,o

(8)

where m is the coefficient for the effect of denaturant on stability [26]. Regressed values of G◦ u,o in solution, G◦ u,sol , and on the surface, G◦ u,ads , can be compared to determine the thermodynamic effect, G◦ u , the surface has on the protein from G◦ u = G◦ u,ads − G◦ u,sol

Table 1 Comparisons of Tosoh HIC media characteristics.

DBC (mg/mL)a Retention volume (mL) Hydrophobicity rank Unfolding rankb a b

Phenyl 650M

Butyl 650M

Hexyl 650C

27.5 8.8 3 3

32.2 9.9 2 1

33.2 14.6 1 2

DBC as reported by Tosoh. Based on increase in D/N for the reporter peptides of transferring.

(9)

Regressed values of ku in solution, ku,sol , and on the surface, ku,ads , can also be compared to determine the kinetic effect of the surface, ku , ku = ku,ads − ku,sol

207

(10)

The different values calculated from Eqs. (7)–(10) originate from the fitting of experimental data with Eq. (4). 4. Results and discussion 4.1. Unfolding effects of chromatographic surface chemistry and temperature 4.1.1. Transferrin MS spectra were collected and compared for transferrin labeled in solution and while adsorbed on Phenyl 650M, Butyl 650M, and Hexyl 650C at different temperatures up to 52 ◦ C where considerable precipitation began to occur. The mass spectra for 32 reporter peptides were converted to D/N values using Eq. (1). A total of 33% sequence coverage was obtained for the 679 residues of transferrin. Many of the areas with missing sequence coverage are populated with cysteines involved in disulfide bridges, which interfere with proteolytic digestion. In general, the patterns of transferrin reporter peptide solvent exposures, identified with unfolding, are similar. Unfolding in solution was minimal and occurred only at the highest temperatures, while on the chromatographic surfaces, unfolding occurred at lower temperatures. Of the three surfaces, unfolding on Butyl 650M was seen at the lowest temperatures and it reached the highest D/N values. The temperature and degree of unfolding were next greatest for Hexyl 650C with that for Phenyl 650M being the least. All but one reporter peptide followed this general pattern, with variations seen only in the degree of unfolding. The single exception to this pattern is discussed below. The D/N values for the reporter peptide, residues 67–81, representative of the general pattern in solution and on the three surfaces, are shown in Fig. 2a. The locations of these residues, and those from 202 to 310, 386 to 392, and 424 to 435 are shown in red on the native structure of transferrin on the right of Fig. 2. In solution, no large increases in solvent exposure were observed for these residues over the temperature range studied. The largest increase was between 42 and 52 ◦ C D/N, from 0.37 to 0.43. This is consistent with studies that report transferrin starting to unfold in solution only at temperatures above 60 ◦ C [27]. In contrast, on Butyl 650M, a significant portion of these residues are already labeled at 22 ◦ C. On Hexyl 650C, the largest solvent exposure increase occurred between 32 and 42 ◦ C, with D/N changing from 0.58 to 0.78. On Phenyl 650M, there was also a significant increase in solvent exposure between 32 and 42 ◦ C where D/N increased from 0.41 to 0.59. Reporter peptides spanning were also labeled more completely on Hexyl 650C than on Phenyl 650M. Note that. This labeling behavior demonstrates that all three chromatographic surfaces induced significant unfolding in certain regions at lower temperatures than in solution. While the extent of unfolding varies among the three surfaces, the ordering of their extents is the same.

Residues 125–131, shown in Fig. 2b and in blue on the native structure on the right of Fig. 2, are from the single reporter peptide that shows different labeling behavior. No labeling was observed at 22 ◦ C in solution or on any of the three surfaces. As the temperature was increased, there was a dramatic increase in solvent exposure, similar to melting. The apparent melting curves shown in Fig. 2b demonstrate that adsorption to all three surfaces lowers the melting temperature relative to solution for this reporter peptide of transferrin. However, the amount it is lowered depends on the type of surface. It should be noted that these “melting curves” obtained with HXMS are not identical to those typical of thermal denaturation detected by spectroscopic techniques, such as circular dichroism or fluorescence. In general, HXMS denaturation is found at lower temperatures due to the required processing time [28]. However, the lowering of transition temperatures here is much larger than this effect, so much of the effect can be attributed to adsorption. In contrast to the other surfaces, the labeling of transferrin on Butyl 650M displayed a maximum vs. temperature and then decreased between 42 and 52 ◦ C to below the level at 32 ◦ C. Although the labeling of several reporter peptides from the general trend also decreased between 42 and 52 ◦ C on Butyl 650M, none of those decreased to that below 32 ◦ C. As discussed later, we suggest that intermolecular interactions or aggregation may be responsible for this behavior. Finally, three groups of residues: 180–188, 529–557, and 636–666, shown in green on the native structure on the right, did not show increased labeling on any surface. The degree of unfolding of transferrin on the surfaces does not completely follow their relative hydrophobicities determined by Tosoh from dynamic binding capacities (DBC) for lysozyme (see Table 1) which should not be influenced by unfolding. As an alternative measure of hydrophobicity, we did gradient elution experiments at low protein loadings with human serum transferrin. Table 1 compares the hydrophobicity measures with unfolding rank of each surface measured by the increase in D/N for the reporter peptides of transferrin. The hydrophobicity rank of the resins from both measures gives Hexyl 650C > Butyl 650M > Phenyl 650M. The extent of transferrin solvent exposures follows the order: Butyl 650M > Hexyl 650C > Phenyl 650M. The reasons for this difference are not immediately apparent to us. However, the DBC values on Hexyl 650C (33.2 mg/mL) and Butyl 650M (32.2 mg/mL) are quite similar, suggesting that the hydrophobicity ordering might be opposite, and explaining the more apparent unfolding on Butyl 650M.

4.1.2. Antitrypsin MS spectra were collected and compared for antitrypsin labeled in solution and while adsorbed on Phenyl 650M, Butyl 650M, and Hexyl 650C at different temperatures. Unlike transferrin, no apparent precipitation was seen up to 92 ◦ C. The mass spectra for 28 reporter peptides were converted to D/N values using Eq. (1). A total of 53% sequence coverage was obtained for the 392 residues of antitrypsin. This was higher than for transferrin.

208

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

Fig. 2. (Left) Thermal unfolding of different reporter peptides of transferrin in solution (blue, diamonds), on Phenyl 650M (red, squares), on Butyl 650M (purple, Xs), and on Hexyl 650C (green, triangles) in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer at 22–52 ◦ C. Samples were labeled with deuterated buffer for 10 min. The D/N for residues 67–81 is shown in panel (a). The D/N for residues 125–131 is shown in panel (b). (Right) Location of residues 67–81 and the other residues that followed the general unfolding pattern are shown in red on the native structure (PDB ID 2HAV). The exception to the general unfolding pattern, residues 125–131, is shown in blue. Residues that had no enhanced labeling on the surfaces are shown in green. Gray depicts missing sequence coverage. Error bars represent sample 95% confidence intervals of triplicate data points collected on Phenyl 650M at 22 ◦ C. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

As with transferrin, contact of antitrypsin with hydrophobic chromatographic surfaces increased labeling for most, but not all, reporter peptides. Less of an effect is seen on Phenyl 650M than on Butyl 650M and Hexyl 650C. However, for many reporter peptides Hexyl 650C resulted in more labeling than did Butyl 650M, although this was statistically significant for only 6 groups of residues: 78–87, 98–108, 110–119, 278–288, and 304–317. The other difference in labeling patterns between antitrypsin and transferrin is that most antitrypsin reporter peptides were observed with statistically equal increased labeling on all three surfaces: 151–166, 185–189, 238–241, 252–270, 318–343, and 353–375. Two peptides showing this general set of trends are shown in Fig. 3. Four groups of residues: 24–32, 88–92, 131–142, and 197–207 were observed not to have increased surface labeling. For antitrypsin, the extent of unfolding for residues on the surface correlates more directly with increasing DBC hydrophobicity (Table 1). Many reporter peptides that unfold when adsorbed at the same temperature (44% of antitrypsin coverage), do so to statistically equal extents on all three surfaces. In the 25% of antitrypsin coverage where statistically significant differences in reporter peptide labeling exist among all 3 surfaces (as determined by Student’s t-test with Bonferroni corrected ˛ = 0.006), more labeling is observed on Hexyl 650C followed by Butyl 650M and then Phenyl 650M. It is not clear why these reporter peptides of antitrypsin follow this trend while those for transferrin do not. The varying degree of unfolding observed for antitrypsin and the presence of an exception to the general unfolding pattern of transferrin suggest that local stability is important in determining how different groups of residues or domains of a protein will unfold

on a surface. The knowledge and use of transferrin and antitrypsin local stability in solution to explain unfolding trends on the surfaces is presented in the next section. 4.2. Relating partially unfolded states in solution and on chromatographic surfaces Partial unfolding in solution has been observed from increased temperature, from adding denaturants such as urea or GdnHCl, and at extremes in pH [29,30] and on HIC surfaces [31–33]. Chemical denaturation and the mechanism of surface unfolding are not as well understood as are thermal and pH induced denaturation in solution. Recent molecular simulations of urea induced unfolding provide some insight into the complexity of this mechanism [34]. However, the similarity of partially unfolded structures from different denaturation mechanisms is not known. Transferrin is known to unfold via a multistate mechanism in solution [23,27]. Although the structural transition has not been fully characterized, it is known that the N terminal domain is less stable than the C terminal domain, and it unfolds first as denaturant concentration is increased. Fig. 4 shows the effect each chromatographic surface has on the unfolding of these two domains. The difference in D/N between solution and surface, D/N, represents the average reporter peptide unfolding for each domain, weighted by the number of residues in each peptide. In general, almost all D/N values are positive, indicating adsorption increases solvent exposure. Even though the uncertainties are large, at every temperature and for every chromatographic surface, a more positive D/N is observed for the N terminal domain peptides compared

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

209

Fig. 3. (Left) Labeling patterns of different reporter peptides of antitrypsin in solution (blue, diamonds), on Phenyl 650M (red, squares), on Butyl 650M (purple, Xs), and on Hexyl 650C (green, triangles) in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer at 22 to 92 ◦ C. Samples were labeled with deuterated buffer for 10 min. The D/N for residues 185–189 is shown in panel (a). The D/N for residues 266–273 is shown in panel (b). (Right) Location of residues 185–189 and the other residues that unfolded more on Butyl 650M and Hexyl 650C than on Phenyl 650M are shown in blue on the native structure (PDB ID 2QUG). Resides 266–273 and the other residues that unfolded to statistically equal extents on all three surfaces are shown in red. Residues that had no enhanced labeling on the surfaces are shown in green. Gray depicts missing sequence coverage. Error bars represent sample 95% confidence intervals of triplicate data points collected on Phenyl 650M at 22 ◦ C. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

to the C terminal domain peptides, consistent with the solution stability studies. Antitrypsin also is known to unfold via a multistate mechanism in solution [35]. The partially unfolded intermediate is critical for antitrypsin biological function and exists as a small, but significant, subpopulation under physiological conditions [36]. In the presence of low concentrations of GdnHCl, antitrypsin has a well characterized unfolding intermediate structure where at ∼1 M, several ␤ sheet strands and an ␣ helix unfold. The remaining regions stay folded until higher concentrations of GdnHCl, where the protein molecule becomes fully unfolded. The peptide specific solvent exposure patterns for adsorbed antitrypsin in this work compare well with the folded and unfolded regions observed previously for the solution intermediate [35]. The folded regions of the antitrypsin solution intermediate and on the HIC surfaces of this work are shown in Fig. 5. The residues that remain in their native structure in the solution intermediate, and do not unfold on the HIC surfaces, are shown in green. The regions for which no sequence coverage was obtained here are shown in white. The blue arrows identify eight regions that remain folded in the solution intermediate as well as when adsorbed on Phenyl 650M or Butyl 650M. Of these eight, four also do not unfold on Hexyl 650C. The other four (highlighted in yellow in Fig. 5) are the only ones having statistically significant (as determined by Student’s t-test with Bonferroni corrected ˛ = 0.006) enhanced labeling on Hexyl 650C compared to solution. Domains of greater local stability in folding

intermediates also are the most resistant to unfolding on HIC surfaces. Overall, these observations suggest that there is a correspondence between local stability in solution and upon adsorption. The unfolded regions of antitrypsin in the solution intermediate are also similar to the reporter peptides whose D/N values are affected by adsorption on the HIC surfaces, as shown in Fig. 6. The regions that unfold in the solution intermediate and are more solvent exposed on all HIC media are shown in red. The reporter peptides that are more solvent exposed on Butyl 650M and Hexyl 650C, as well as to a lesser extent on Phenyl 650M, are shown in orange. The regions for which no sequence coverage was obtained are shown in white. The blue arrows identify six regions that unfold in the solution intermediate and also unfold when adsorbed on Butyl 650M and Hexyl 650C. Of these six, four also unfold on Phenyl 650M. The combined results of HDX mapping for both of these two domain protein systems support the hypothesis that the domains of a protein which are more stable in solution will also be more stable on hydrophobic surfaces. A relation between solution unfolding and adsorption unfolding should be considered important for developing methods to predict protein unfolding on HIC surfaces from a minimum of measurements. Further, the consistency of the transferrin N terminal domain having lower stability in solution and on the surface supports the hypothesis that individual domain stability in solution guides what groups of residues or domains will be most affected by a chromatographic surface.

210

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

Fig. 6. Unfolded regions of antitrypsin in the guanidinium denatured intermediate (left) and when adsorbed on Phenyl 650M, Butyl 650M, and Hexyl 650C in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer at 22 ◦ C (right). Red represents those regions unfolded in the solution intermediate (left) and on all media (right). Orange represents those regions that unfolded on Butyl 650M and Hexyl 650C to a greater extent than on Phenyl 650M. Blue arrows designate regions that showed similar stability between the two cases. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

4.3. Unfolding effects of surface and GdnHCl

Fig. 4. Difference in D/N, D/N, between human serum transferrin adsorbed on media (Phenyl 650M – blue, Hexyl 650C – green, Butyl 650M – red) and in solution in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer. D/N values represent the average reporter peptide unfolding for each domain weighted by the number of residues in each peptide. Data are averaged and weighted for all reporter peptides in the N terminal domain (top) and C terminal domain (bottom). Error bars represent sample propagated 95% confidence intervals for D/N from triplicate data points collected on Phenyl 650M at 22 ◦ C. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

Fig. 5. Folded regions of antitrypsin in the guanidinium denatured solution intermediate (left) and when adsorbed on Phenyl 650M, Butyl 650M, and Hexyl 650C in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer at 22 ◦ C (right). Green represents those regions folded in the solution intermediate (left) and on all media (right). Yellow represents those regions that unfolded on Hexyl 650C only. Blue arrows designate regions that showed similar stability between the two cases. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

The above results show that (1) the more stable domain of a two domain protein tends to be the more stable domain when adsorbed on a hydrophobic surface and (2) more hydrophobic surfaces tend to be more denaturing. However, there is currently no direct way to predict the stability of a protein upon adsorption on chromatographic media. It seems likely that knowledge of the difference in the unfolding standard state Gibbs energy upon adsorption, G◦ u , along with the kinetics of surface unfolding, ku , combined with these properties in solution, G◦ u and ku , of a solution protein could lead to predictive capability. Here, we determine these properties with HXMS for transferrin labeled in solution, and while adsorbed on Hexyl 650C, as a function of GdnHCl concentration. This method has been used previously to determine the G◦ u of the two domains of transferrin in solution [23]. The sequence coverage, 33%, is the same as the temperature studies with this protein. Many of the reporter peptides, however, do not have large increases in D/N over the concentration range of GdnHCl studied. The limited changes result mainly from reporter peptides that are already mostly solvent exposed in solution and/or on the surface in the absence of GdnHCl (D/N values greater than 0.8). The transition conditions for these peptides are less obvious since the solvent exposure cannot increase significantly. We note nine peptides with well defined transition regions: four from the N terminal domain and five from the C terminal domain. Of these nine, the peptides with lowest G◦ u from the two domains are presented in detail below because they best represent the slowest exchanging residues and the best measure of the global unfolding reaction [37]. The reporter peptide with residues 223–259 illustrates how a surface can affect labeling of the N terminal domain of transferrin in the presence of varying concentrations of GdnHCl as shown in Fig. 7. The D/N values without GdnHCl are greater than zero, as a number of residues are solvent accessible on the protein surface. As GdnHCl is added the D/N value is essentially constant until the concentration where unfolding begins. The D/N value increases with increasing GdnHCl concentration until these residues are fully unfolded and the D/N value reaches an asymptote with increasing GdnHCl concentration.

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

211

Fig. 7. Unfolding of residues 223–259 in the N terminal domain of transferrin in solution (blue, diamonds), and on Hexyl 650C (red, squares) in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , 0 to 5 M GdnHCl, pH 7.0 buffer at 22 ◦ C and various labeling times. The residues locations on the native structure (PDB ID 2HAV) are shown in red. The N terminal and C terminal domain of transferrin are shown in blue and green, respectively. Fits of the experimental data with Eq. (4) are shown as solid lines. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

Table 2 Characteristics of labeling patterns for transferrin residues 223–259 in the presence of varying GdnHCl concentrations. Uncertainties presented were determined by regressing the data to Eq. (4). Parameter

10 min

Solution C1/2 Hexyl 650C C1/2 Solution p Hexyl 650C p

3.6 2.4 2.2 34.9

± ± ± ±

20 min 0.6 0.2 2.0 31.2

2.9 2.0 10 39

± ± ± ±

0.1 0.5 8.2 34.5

40 min 2.9 1.6 3.4 4.5

± ± ± ±

0.4 0.4 3.0 3.1

As Fig. 7 shows, the D/N of these residues 223–259 are statistically the same in solution as when adsorbed on Hexyl 650C at low GdnHCl concentrations. However as GdnHCl concentration increases, the transition to unfolding occurs at lower GdnHCl concentrations for adsorbed transferrin than when in solution. At high GdnHCl concentrations, D/N values plateau for both transferrin in solution and on the surface. The adsorbed D/N asymptote, however, is less than that in solution. This is likely due to the presence of partial solvent protection from this region participating in binding between the protein and the surface. Table 2 lists the midpoints, C1/2 , and slopes, p, obtained from fitting Eq. (4) for the unfolding transitions in solution and when adsorbed on Hexyl 650C for the different labeling times. The transitions are clearly at lower concentrations. Though the p values seem significantly different, their large uncertainties of p from a lack of data points in the transition region prevent definition of any differences between solution and adsorbed species. The parameters, variations, and uncertainties for the peptide with residues 554–589 in the C terminal domain of transferrin are basically the same as in Table 2. Although G◦ u and ku were calculated for all reporter peptides with well defined

transition regions, the uncertainties in these values are too large to make statistically meaningful comparisons [17]. To address the issue of large uncertainties, we made measurements of the multidomain protein, BSA adsorbed on Phenyl 650M, emphasizing data in the transition region. Phenyl 650M was chosen for this study as BSA is nearly completely unfolded on Butyl 650M and Hexyl 650C even at low concentrations of GdnHCl. MS spectra were collected and compared for BSA in solution and adsorbed over increasing concentrations of GdnHCl. Again, not all reporter peptides are suitable for analysis since those mostly solvent exposed (D/N > 0.8) in the absence of GdnHCl do not have well defined transition curves. The reporter peptides with the lowest G◦ u from each of the three domains are presented hereto represent the slowest exchanging residues and global unfolding [37]. The D/N values for residues 127–137 from Domain I of BSA in the presence of varying concentrations of GdnHCl are shown in Fig. 8. In solution, the D/N is limited to 0.69 at 3.8 M GdnHCl and 60 min labeling due to the low solubility of BSA at higher GdnHCl concentrations. In general, the labeling trends for these residues are similar to those of transferrin with the adsorbed residues showing increased solvent exposure at significantly lower concentrations of GdnHCl. Note that the D/N on the surfaces is nonzero with no GdnHCl and reaches the full solvent exposure value of 1.0. Table 3 lists the C1/2 and p values obtained from fitting Eq. (4). It is noteworthy that the C1/2 value is significantly affected by adsorption while the p value appears to be the same. The parameters for the second and third reporter peptides of BSA are presented in Table 4 to illustrate how the surface affects the labeling of Domains II and III with varying concentrations of GdnHCl. Unfortunately, the uncertainties are sufficiently large that

212

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

Fig. 8. Unfolding of residues 127–137 in Domain I of BSA in solution (blue, diamonds), and on Phenyl 650M (red, squares) in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , 0 to 4 M GdnHCl, pH 7.0 buffer at 22 ◦ C and various labeling times. The location of the residues on the native structure (PDB ID 3V03) are shown in red. Domains I, II, and III are shown in green, blue, and yellow, respectively. Fits of the experimental data with Eq. (4) are shown as solid lines. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

Table 3 Characteristics of labeling patterns for BSA residues 127–137 (Domain I) in the presence of varying GdnHCl concentrations. Uncertainties presented were determined by regressing the data to Eq. (4). Parameter

10 min

Solution C1/2 Phenyl 650M C1/2 Solution p-value Phenyl 650M p-value

3.7 1.7 3.6 3.4

± ± ± ±

0.1 0.3 1.3 2.9

30 min 3.7 1.5 3.2 3.0

± ± ± ±

0.1 0.4 0.8 3.7

60 min 3.5 1.4 2.6 4.4

± ± ± ±

0.1 0.2 0.7 4.1

all that can be said definitely is that the transition to unfolding occurs at lower concentrations for adsorbed BSA than in solution. The parameters of Eq. (4) were used to interpolate D/N values in Eq. (5) to obtain MNU and then to get [N] from Eq. (3). These [N] values were fitted to Eq. (6) as a function of labeling time to determine No and ku . The No values were then used to calculate G◦ u using Eq. (7) for the various denaturant concentrations, and the obtained denaturant dependence of G◦ u in Eq. (8) was

used to determine m and G◦ u,o . Values of G◦ u , m, and ku for residues 127–137, 333–340, and 529–543 in solution and adsorbed are shown in Table 5. The magnitudes of G◦ u in solution suggest that domain stabilities are I > II > III, although experimental uncertainty prevents definite ordering. The literature differs over which domain is most stable in solution but agrees that Domain III is probably least stable and unfolds first in chemical denaturation [38–41]. The order of domain stability of adsorbed BSA is the same but the Gibbs energy of unfolding is significantly decreased, as measured by G◦ u . Since the values for all domains are statistically the same, a surface has the same effect on all domains of BSA, regardless of their solution stability, and domain unfolding order on the surface is determined only by solution stability. The m values we find for BSA shown in Table 5 compare favorably with values determined by a correlation of m with domain specific changes in accessible surface area upon unfolding, ASA, and by the number of disulfide crosslinks present [42]. Though it might be expected that some of the protein surface area would be

Table 4 Characteristics of labeling patterns for BSA residues 333–340 and 529–54 in the presence of varying GdnHCl concentrations. Uncertainties presented were determined by the MATLAB program used to fit the data to Eq. (4). Residues

Parameter

10 min

333–340 (Domain II)

Solution C1/2 Phenyl 650M C1/2 Solution p-value Phenyl 650M p-value

3.7 1.7 3.3 4.4

± ± ± ±

0.1 0.3 1.6 4.5

30 min 3.8 1.5 2.3 6.2

± ± ± ±

0.2 0.3 1.1 12.4

529–543 (Domain III)

Solution C1/2 Phenyl 650M C1/2 Solution p-value Phenyl 650M p-value

3.1 1.5 3.9 3.5

± ± ± ±

0.2 0.3 3.2 4.4

2.7 1.6 4.5 1.9

± ± ± ±

0.1 0.7 2.9 3.1

60 min 3.5 1.3 2.5 5.8

± ± ± ±

0.1 0.2 0.9 7.0

2.9 0.75 4.0 8.9

± ± ± ±

0.2 0.32 2.7 10.6

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219 Table 5 Unfolding rates and free energies of BSA on Phenyl 650M. Domain Ia d

Gu,sol (kcal/mol) Gu,ads (kcal/mol)d Gu (kcal/mol)e msol (kcal/mol M)d mads (kcal/mol M)d mref (kcal/mol M)f ku,sol (min−1 )g ku,ads (min−1 )g ku (min−1 )h

10.9 6.4 −4.5 2.3 2.0 3.7 0.012 0.015 0.003

± ± ± ± ± ± ± ± ±

5.5 3.1 6.3 1.1 1.1 1.1 0.008 0.010 0.012

Domain IIb 7.8 4.8 −4.0 2.1 2.7 3.5 0.012 0.017 0.005

± ± ± ± ± ± ± ± ±

3.4 2.6 4.3 1.1 1.3 1.3 0.008 0.009 0.012

Domain IIIc 7.2 ± 3.2 ± −4.0 ± 2.4 ± 1.8 ± 3.5 ± 0.030 ± – –

3.0 1.2 3.2 1.0 0.7 1.3 0.012

a

Data from GdnHCl denaturation curve of residues 127–137 used. Data from GdnHCl denaturation curve of residues 333–340 used. c Data from GdnHCl denaturation curve of residues 529–543 used. d Calculated with Eqs. (6)–(8). e Calculated with Eq. (9). f Calculated with Eq. (12) from [43]: m = (0.28 ± 0.03) × [ASA−(792 ± 780)(# crosslinks)]. g Calculated with Eq. (6). h Calculated with Eq. (10). b

on the chromatographic surface and change the m value, it is the same in both solution and adsorbed. 4.4. Solvent protection increase on the chromatographic surface MS spectra were collected and compared for BSA labeled in solution and while adsorbed on Butyl 650M and Hexyl 650C at different temperatures. The mass spectra for the reporter peptides were converted to D/N values using Eq. (1). A total of 34% sequence coverage was obtained for the 585 residues of BSA. Many of the areas with missing sequence coverage are populated with cysteines involved in disulfide bridges, which reduce proteolytic digestion. Fig. 9 shows the native structure of the three domains of BSA (left) and the location of the reporter peptides and cysteines of BSA (right). Fortunately, reporter peptides were obtained for all three domains, with 40% in Domain I, 37% in Domain III, and 25% in Domain II. Unlike transferrin and antitrypsin, some BSA peptides show solvent protection increase with temperature. The D/N values for two reporter peptides where solvent protection on the surface

213

increases with temperature are shown in Fig. 10. One of these peptides, residues 307–313, shown in panel (a), have (1) substantially more labeling on all surfaces at lower temperatures, (2) labeling increased in solution with an increase in temperature, consistent with the reported melting temperature of 42 ◦ C [43], and (3) there was less labeling on the surface at higher temperatures. At 22 ◦ C, peptide is fully labeled on Butyl 650M and mostly unfolded on Hexyl 650C. On the Butyl 650M surface, the solvent exposure decreases from 52 to 82 ◦ C. In addition to residues 307–313, this trend is also observed for residues 20–31, 39–45, 49–70, 127–137, 154–164, 169–182, 200–209, 219–226, 307–313, 324–329, 331–340, 422–436, 440–460, and 530–547. These residues are shown in red on the native structure on the right of Fig. 10. The other reporter peptide of Fig. 10, residues 3–14 in panel b, also showed decreased labeling with increasing temperature on the surfaces, but no increase in solution. The location of the peptide on the native structure is in blue on the right of the figure. In addition to residues 3–14, this trend is also observed for residues 559–584; it may be significant that these residues are located at the termini of BSA. As mentioned in Section 4.3, Domain III of BSA is usually considered the least stable. Our lower value of Gu for Domain III relative to the other domains in solution and on Phenyl 650M is consistent with these observations. However, the different domains of BSA do not show distinctive behaviors on Butyl 650M and Hexyl 650C. Rather, certain reporter peptides in each of the three domains show decreased labeling. Two scenarios may explain this decrease in labeling on the surface relative to solution. First, regions of the protein bound to the hydrophobic ligands, can provide enhanced protection. In one study, hydrogen deuterium exchange rates were measured by NMR for bovine ␣ lactalbumin in solution and adsorbed on hydrophobic, polystyrene nanospheres [32]. While some adsorbed residues had less protection, several had less solvent exposure when adsorbed. The authors hypothesized that these residues with enhanced protection, located in one specific region, directly interact with the polystyrene surface. A second possibility for increased protection is that protein protein contacts on the surface reduce solvent accessibility. Hydrogen deuterium exchange rates have been shown to be reduced by

Fig. 9. (Left) Native structure of BSA (PDB ID 3V03) highlighting Domain I (red), Domain II (green), and Domain III (blue). (Right) Location of reporter peptides (highlighted according to domain) obtained in this study. Orange depicts cystines of BSA. Gray depicts missing sequence coverage. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

214

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

Fig. 10. Thermal unfolding and solvent protection increases of different regions of BSA in solution (blue, diamonds), on Butyl 650M (red, squares), and on Hexyl 650C (green, triangles) in 25 mM PO4 , 1.5 M (NH4 )2 SO4 , pH 7.0 buffer at 22–82 ◦ C. Samples were labeled with deuterated buffer for 10 min. The D/N for residues 307–313 is shown in panel (a). The D/N for residues 3–14 is shown in panel (b). (Right) Location of residues 307–313 and the other residues with similar unfolding pattern are shown in red on the native structure (PDB ID 3V03). Residues 3–14 and the other residues where no increased labeling in solution are shown in blue. Error bars represent sample 95% confidence intervals of triplicate data points collected on Butyl 650M at 22 ◦ C. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)

intermolecular interactions formed in bovine insulin dimers [44], diphtheria toxin oligomers [45], and from aggregation of amyloid beta peptides [46]. To investigate if BSA aggregation was occurring on the surface, SEC was performed. Fig. 11 shows SEC chromatograms of BSA in solution (left) and adsorbed on Butyl 650M (right) incubated at temperatures of 52, 62, 72, and 82 ◦ C. At temperatures of 52 ◦ C (Fig. 11a) and lower (data not shown), two distinct peaks are observed in solution (left) and eluted from Butyl 650M (right). These larger and smaller peaks have been characterized previously as monomers and dimers, respectively [47]. Although the same peaks are observed both in solution and from protein eluted from Butyl 650M, the dimer peak is larger for the protein sample exposed to the HIC surface. As temperature is increased, two additional higher molecular weight peaks appear for Butyl 650M at 62 ◦ C (Fig. 11b, right) while no such peaks begin to appear until 72 ◦ C in solution (Fig. 11c, left). On the surface the two additional peaks become larger at 72 ◦ C (Fig. 11c, right). While they are smaller at 82 ◦ C (Fig. 11d, right) this reduction is probably from limited recovery of irreversibly bound protein, as indicated by the comparatively low 215 nm signal. SEC chromatograms obtained for BSA adsorbed on Phenyl 650M at various temperatures showed the same trends (data not shown). It is likely that these two additional peaks are higher order oligomers, since their SEC retention times are less than those of the monomer and dimer peaks. Apparently, the HIC surface shifts the distribution of the species toward oligomers at elevated temperatures. Aggregates on the surface probably do not have native structure since there was increased solvent exposure relative to

solution before it begins to decrease at higher temperatures (see Fig. 10). Surface induced aggregation has been observed previously [48–51]. One proposed cause is surface destabilization of the protein on the surface by increasing the unfolding and the aggregation rates, perhaps by eliminating a required precursory unfolding for nonnative aggregation of a protein. Thus, if unfolding were the rate limiting step, the surface could facilitate aggregation. Alternatively, if the adsorbed protein molecules are mobile, the 2-dimensional adsorption surface would simply provide a higher rate of molecular collisions than in a 3-dimensional solution. Only a few reporter peptides of transferrin and antitrypsin showed an increase in solvent protection but not on all surfaces (data not shown). SEC was also performed for transferrin and antitrypsin samples (data not shown) to evaluate if irreversible aggregation on the surface might cause decreased solvent exposure. However, only one peak was found at all temperatures and the retention times were the same as those for solution transferrin and antitrypsin, indicating no aggregation. Although these studies were performed at elevated temperatures, these findings may still have important implications in typical HIC process design. We have shown that oligomeric species can be formed during HIC processes at conditions under which the solution protein is native. We examined [20] whether the residue groups of BSA that unfold on Butyl 650M (D/N > 0.5) at lower temperatures, and become mostly solvent protected (D/N < 0.1) at 80 ◦ C would be identified as aggregation “hot spots” by the aggregation calculators AGGRESCAN and PASTA. Of the seven, five were predicted by AGGRESCAN though three groups predicted to be

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

52°C

mAU at 215nm

800

Solution

600 400 200 6 8 10 12 Retention volume [mL]

(b)

62°C mAU at 215nm

4

mAU at 215nm

0

600 500 400 300 200 100 0

4

200 100 0

6 8 10 12 Retention volume [mL]

4

6 8 10 12 Retention volume [mL]

4

6 8 10 12 Retention volume [mL]

mAU at 215nm

72°C

4

6 8 10 12 Retention volume [mL]

300 250 200 150 100 50 0

(d)

82°C

600

100

400 200 0

6 8 10 12 Retention volume [mL]

300

mAU at 215nm

600 500 400 300 200 100 0

4

400

mAU at 215nm

mAU at 215nm

(c)

600 500 400 300 200 100 0

Butyl 650M

mAU at 215nm

(a)

4

6 8 10 12 Retention volume [mL]

75 50 25 0

4

6 8 10 12 Retention volume [mL]

Fig. 11. SEC chromatograms of BSA in solution (left) and eluted from Butyl 650M (right) at temperatures of 52 ◦ C (a), 62 ◦ C (b), 72 ◦ C (c), and 82 ◦ C (d). Samples were incubated for 1 h, eluted with 200 ␮L of desorption buffer, diluted with 500 ␮L of working buffer, and loaded on a TSK gel G3000SWXL column at 22 ◦ C for analysis.

hot spots did not show solvent protection to D/N less than 0.2. PASTA was much less successful in prediction. However, the overlap between the hot spots and solvent protected regions suggests using an aggregation calculator to flag aggregation prone regions early in development whereas proteins of higher stability, like transferrin and antitrypsin, could be expected to remain monomeric under a wider range of process conditions. 4.5. Superposition of surface effects on solvent protection We have shown that temperature and HIC surface type have significant effects on the solvent protection. As seen in the D/N values of Figs. 2 and 3, the thermal unfolding of transferrin and antitrypsin are different on Phenyl 650M, Butyl 650M, and Hexyl 650C. In addition, the effects are not uniform across all reporter peptides as discussed in Sections 4.1 and 4.2. Further, the effects of temperature and surface for BSA are different again. As seen in the labeling patterns of Fig. 10 and SEC chromatograms of Fig. 11, BSA

215

unfolds at low temperatures on the surface and then aggregates at higher temperatures. One of the hypotheses of this work is that proteins unfold in similar ways under different denaturing environments. A relation between the unfolded state of proteins on the surface and their solution intermediates was already discussed in Section 4.3. Here we explore if the effects of temperature can produce the same labeling patterns on different chromatographic surfaces. Solution surface comparisons for the reporter peptides of transferrin can show differences in labeling and variability at certain temperature offsets, Ts. Fig. 12 shows differences in D/N values, D/N, for most transferrin reporter peptides adsorbed on Butyl 650M and Hexyl 650C when compared at zero and 10 ◦ C offsets. Error bars represent propagated uncertainties in D/N for individual reporter peptides. First, when compared at the same temperature most regions have positive D/N values corresponding to more unfolding on Butyl 650M. However, when Hexyl 650C is compared to Butyl 650M at a temperature 10 ◦ C higher, as shown on the bottom of Fig. 12, there are no statistical differences between deuterium labeling patterns for essentially all regions. For the comparison of Hexyl 650C at 32 ◦ C to Butyl 650M at 22 ◦ C (Fig. 12c), 22 of the 32 reporter peptides have effectively zero differences. Of the 10 that do not, only 1 peptide (reporter Peptide 20) has a D/N magnitude that could be as high as 0.1. For the comparison of Hexyl 650C at 42 ◦ C to Butyl 650M at 32 ◦ C (Fig. 12d), 25 of the 32 reporter peptides have essentially no differences. Only 1 peptide (reporter Peptide 13) has a D/N magnitude that could be as high as 0.1. This suggests that the effect on protein stability by changing from one alkyl surface to another can be quantitatively described by a simple superposition of shifted temperatures. Comparisons of labeling between Butyl 650M on Phenyl 650M at different temperature offsets (data not shown) are not as satisfactory with average D/N values of ≥0.1 and only 15 of the 32 reporter peptides having differences of zero within uncertainty at a 20 ◦ C offset. Superposition patterns were also tried for the differences in surface and solution labeling at different temperatures (data not shown) with similar less satisfactory comparisons for Butyl 650M and Phenyl 650M. The unfolding patterns for antitrypsin among temperatures and surfaces are similar to those for transferrin. Solution surface comparisons for the reporter peptides of antitrypsin show that the differences in labeling and variability are smaller when T is 10 ◦ C or greater. Fig. 13 shows differences between deuterium labeling patterns for antitrypsin adsorbed on Phenyl 650M and on Hexyl 650C for a 20 ◦ C difference. At this temperature, differences between deuterium labeling patterns for most regions are zero within uncertainty. For the comparison of Phenyl 650M at 52 ◦ C to Hexyl 650C at 32 ◦ C (Fig. 13, left), 27 of the 31 reporter peptides have differences of zero. Further, the other four have D/N which could be less than 0.1. Comparison of differences on Phenyl 650M at 72 ◦ C to those on Hexyl 650C at 52 ◦ C (Fig. 13, right), 27 of the 31 reporter peptides have differences of zero. Of the four that are larger, three have D/N magnitudes that are probably greater than 0.1, indicating that superposition does not hold as well at these higher temperatures. Comparisons of labeling of antitrypsin between Butyl 650M and Hexyl 650C at T = 20 ◦ C are very similar as shown in Fig. 14. Here, D/N for most regions is zero within uncertainty. For the comparison of Butyl 650M at 72 ◦ C to Hexyl 650C at 52 ◦ C (Fig. 14, left), 27 of the 31 reporter peptides have no differences. However of the four that are significant, all have a D/N magnitude that probably equal to or greater than 0.1. Comparing Butyl 650M at 92 ◦ C with Hexyl 650C at 72 ◦ C (Fig. 14, right), 29 of the 31 reporter peptides have no differences. Of the two that are significant, only reporter Peptide 19 has a D/N magnitude that is probably equal to or greater than 0.1.

216

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

(a)

(b) 0.4 0.3 0.2

0.2

0.1

0.1

0.0 -0.1

1

6

11

16

21

26

31

0.0 --0.1

-0.2

--0.2

-0.3

--0.3

-0.4

Butyl @ 42C - Hexyl @ 42C

0.3

ΔD/N

ΔD/N

0.4

Butyl @ 32C - Hexyl @ 32C

1

6

11

16

--0.4

Peptide

21

26

31

Peptide

(d)

(c) 0.4 0.3 0.2

0.2

0.1

0.1

0.0 -0.1

1

6

11

16

21

26

31

0.0 --0.1

-0.2

--0.2

-0.3

--0.3

-0.4

Hexyl @ 42C - Butyl @ 32C

0.3

ΔD/N

ΔD/N

0.4

Hexyl @ 32C - Butyl @ 22C

1

6

11

16

--0.4

Peptide

21

26

31

Peptide

Fig. 12. Differences in D/N, D/N, between transferrin adsorbed on Butyl 650M at 32 ◦ C and Hexyl 650C at 32 ◦ C (a), adsorbed on Butyl 650M at 42 ◦ C and Hexyl 650C at 42 ◦ C (b), adsorbed on Hexyl 650C at 32 ◦ C and on Butyl 650M at 22 ◦ C (c), and adsorbed on Hexyl 650C at 42 ◦ C and on Butyl 650M at 32 ◦ C (d).

Superposition also works for BSA on the HIC surfaces. Fig. 15 shows differences between deuterium labeling patterns for BSA adsorbed on Hexyl 650C and on Butyl 650M for T = 20 ◦ C. Many regions in the 4 comparisons have differences of zero within uncertainty with more small differences at higher temperatures. For example, 13, 15, 18, and 24 reporter peptides have D/N of zero within uncertainty as shown in panels (a), (b), (c), and (d), respectively. Recall that aggregation probably occurs at these

temperatures, so the agreement may arise from aggregation being initiated on the Hexyl 650C surface at a higher temperature than on Butyl 650M. The patterns found by temperature superposition demonstrate that a large portion of the regions that unfold or become more solvent protected from surface interactions, or from aggregation, are generally the same on different surfaces. Superposition works best when comparing media that have only alkyl ligands, suggesting

0.4

0.4

Phenyl @ 52C - Hexyl @32C

0.2

0.2

0.1

0.1

0

1

6

11

16

21

26 6

0

-0.1

--0.1

-0.2

--0.2

-0.3

--0.3

-0.4

Peptide

Phenyl @ 7 72C - Hexyl @ 52C

0.3

Δ D/N

Δ D/N

0.3

--0.4

1

6

11

16

21

26

Peptide

Fig. 13. Differences in D/N, D/N, between antitrypsin adsorbed on Phenyl 650M at 52 ◦ C and on Hexyl 650C at 32 ◦ C (left), and adsorbed on Phenyl 650M at 72 ◦ C and on Hexyl 650C at 52 ◦ C (right).

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

0.4

0.4

Butyl @ 72C - Hexyl @ 52C

0.2

0.2

0.1

0.1

-0.1

6

1

11

16

2 21

26

31

0.0

-0.2

-0.3

-0.3

6

1

-0.1

-0.2

-0.4

Butyl @ 9 92C - Hexyl @ 72C

0.3

ΔD/N

ΔD/N

0.3

0.0

217

11

-0.4

Peptide

16

21 1

26

31

Peptide

Fig. 14. Differences in D/N, D/N, between antitrypsin adsorbed on Butyl 650M at 72 ◦ C and on Hexyl 650C at 52 ◦ C (left), and adsorbed on Butyl 650M at 92 ◦ C and on Hexyl 650C at 72 ◦ C (right).

that unfolding data on one surface type could be used to anticipate unfolding on other similar surfaces. The differences between the alkyl and phenyl surfaces indicate that ligand chemistry can be important in predicting the regions of proteins that may unfold when adsorbed, though many similarities exist. For example, HIC surfaces with aromatic ligands have pi pi interactions in addition to alkyl hydrophobic interactions. The presence of these pi pi interactions for Phenyl 650M adsorption may explain why superposition is

(a)

less successful when comparing the Phenyl 650M D/N data to that of Butyl 650M or Hexyl 650C. It is likely that additional thermal denaturation data collected on media of alternate hydrophobicity (e.g. Ethyl 650M, Phenyl Sepharose, etc.) would provide more information on how far superposition might be extended to other surfaces. Again, the potential value of this approach would be to leverage limited amounts of structural data to describe other media and proteins of interest.

(b) 0.4

0.4

Hexyl @ 42C - Butyl @ 22C

0.2

0.2

0.1

0.1

0 -0.1

1

6

11

16 6

21

0 -0.1

-0.2

-0.2

-0.3

-0.3

-0.4

Hexyl @ 52C - Butyl @ 32C

0.3

ΔD/N

ΔD/N

0.3

1

6

-0.4

Peptide

(c)

11

16

21

Peptide

(d) 0.4

0.4

Hexyl @ 62C - Butyl @ 42C

0.2

0.2

0.1

0.1

0 -0.1

1

6

11

16 6

21

0 -0.1

-0.2

-0.2

-0.3

-0.3

-0.4

Peptide

Hexyl @ 72C - Butyl @ 52C

0.3

Δ D/N

ΔD/N

0.3

-0.4

1

6

11

16

21

Peptide

Fig. 15. Differences in D/N, D/N, between BSA adsorbed on Hexyl 650C at 42 ◦ C and on Butyl 650M at 22 ◦ C (a), adsorbed on Hexyl 650C at 52 ◦ C and Butyl 650M at 32 ◦ C (b), adsorbed on Hexyl 650C at 62 ◦ C and on Butyl 650M at 42 ◦ C (c), and adsorbed on Butyl 650M at 72 ◦ C and on Hexyl 650C at 52 ◦ C (d).

218

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219

5. Conclusions In this study, we showed that unfolding at the peptide level for an adsorbed protein is related to its solution stability and on the chemistry of an HIC surface. We demonstrated that the regions of antitrypsin and transferrin that unfold in their solution intermediate also unfold when adsorbed on the HIC surfaces, and do so more readily. For multidomain proteins, the surface effects were essentially the same for all domains, so any differences in domain stability would originate in solution. Further, Butyl 650M and Hexyl 650C both affect similar regions of the proteins, but to different extents. Yet when the unfolding behaviors of the two media are superimposed at different temperatures, the patterns become very similar. We observed that Phenyl 650M unfolds the proteins to a lesser extent and affects different regions, probably due to its different ligand chemistry. Butyl 650M and Hexyl 650C have been shown to unfold BSA at lower temperatures while facilitating its aggregation at higher temperatures. As with antitrypsin and transferrin, these surfaces both unfold BSA, but to different extents. It appears that hydrophobic surfaces can act as catalysts for aggregation for proteins that are aggregation prone, as shown by the formation of BSA aggregates on the surface at conditions where none are observed in solution. Further, most regions with the largest increase in solvent protection were predicted to be “hot spots” by an aggregation calculator. The identifying of regions that unfold or aggregate from solution stability data or calculators alone can be useful in a process development setting. New types of therapeutic proteins with structures and chemistries different than that of classical monoclonal antibodies continue to emerge within the pharmaceutical industry. Information about solution stability or aggregation propensity for new structures would be of great aid in anticipating potential problems in HIC operations. Finally, we demonstrated that the thermodynamic and kinetic effects of HIC surfaces on protein unfolding can be determined by HXMS. Together with solution stability data, the relative Gibbs energies of unfolding and of rates in solution and adsorbed are quantitative information that could aid in HIC media selection and protein modification for protein purification without deleterious unfolding. A library of thermodynamic and kinetic information on resins would be valuable in the process development setting. Choosing appropriate media based on this information would prevent potential bottlenecks in the process development that arise from protein unfolding and/or aggregation due to improper media selection. Acknowledgments This research was financially supported by NSF Grant no. CBET 1134256. The authors would also like to thank André Dumetz (GlaxoSmithKline, King of Prussia, PA, USA), Michael Shirts and Scott Acton (University of Virginia, Charlottesville, VA, USA) for helpful discussions on protein surface interactions. References [1] J. Queiroz, Hydrophobic interaction chromatography of proteins, J. Biotechnol. 87 (2001) 143–159. [2] S. Hjerten, Some general aspects of hydrophobic interaction chromatography, J. Chromatogr. 87 (1973) 325–331. [3] J.T. Mccue, Theory and use of hydrophobic interaction chromatography in protein purification applications, Methods Enzymol. 463 (2009) 405–414. [4] J. Valliere-Douglass, A. Wallace, A. Balland, Separation of populations of antibody variants by fine tuning of hydrophobic-interaction chromatography operating conditions, J. Chromatogr. A 1214 (2008) 81–89. [5] J.T. Mccue, P. Engel, A. Ng, R. Macniven, J. Thommes, Modeling of protein monomer/aggregate purification and separation using hydrophobic interaction chromatography, Bioprocess Biosyst. Eng. 31 (2008) 261–275.

[6] X. Geng, X. Chang, High-performance hydrophobic interaction chromatography as a tool for protein refolding, J. Chromatogr. 599 (1992) 185–194. [7] J. Li, Y. Liu, F. Wang, G. Ma, Z. Su, Hydrophobic interaction chromatography correctly refolding proteins assisted by glycerol and urea gradients, J. Chromatogr. 24 (2004) 193–199. [8] X. Geng, Q. Bai, Y. Zhang, X. Li, D. Wu, Refolding and purification of interferongamma in industry by hydrophobic interaction chromatography, J. Biotechnol. 113 (2004) 137–149. [9] Y. Kato, T. Kitamura, T. Hashimoto, Operational variables in high-performance hydrophobic interaction chromatography of proteins on TSKGEL Phenyl-5PW, J. Chromatogr. 298 (1984) 407–418. [10] S.-L.W. Wu, A. Figueroa, B.L. Karger, Protein conformational effects in hydrophobic interaction chromatography, J. Chromatogr. 371 (1986) 3–27. [11] K. Benedek, Thermodynamics of alpha-lactalbumin denaturation in hydrophobic-interaction chromatography and stationary phase comparison, J. Chromatogr. 458 (1988) 93–104. [12] P. Gagnon, E. Grund, T. Lindbäck, Large scale process development for hydrophobic interaction chromatography. Part 1: gel selection and development of binding conditions, J. Honours Biol. Pharmacol. Program 8 (1995) 1–9. [13] A. Jungbauer, C. Machold, R. Hahn, Hydrophobic interaction chromatography of proteins, J. Chromatogr. A 1079 (2005) 221–228. [14] R.W. Deitcher, J.P. O’Connell, E.J. Fernandez, Changes in solvent exposure reveal the kinetics and equilibria of adsorbed protein unfolding in hydrophobic interaction chromatography, J. Chromatogr. A 1217 (2010) 5571–5583. [15] R.W. Deitcher, J.E. Rome, P.A. Gildea, J.P. O’Connell, E.J. Fernandez, A new thermodynamic model describes the effects of ligand density and type, salt concentration and protein species in hydrophobic interaction chromatography, J. Chromatogr. A 1217 (2010) 199–208. [16] Y. Bai, J.S. Milne, L. Mayne, S.W. Englander, Primary structure effects on peptide group hydrogen exchange, Proteins 17 (1993) 75–86. [17] A. Gospodarek, Are Unfolding Pathways of Multi-domain Proteins on Chromatographic Surfaces Related to Solution Phase Unfolding Pathways? University of Virginia, Charlottesville, VA, USA, 2013. [18] J.R. Engen, Analysis of protein conformation and dynamics by hydrogen/deuterium exchange MS Anal. Chem. 81 (2009) 7870–7875. [19] A.J. Percy, M. Rey, K.M. Burns, D.C. Schriemer, Probing protein interactions with hydrogen/deuterium exchange and mass spectrometry—a review, Anal. Chim. Acta 721 (2012) 7–21. [20] I.A. Kaltashov, S.J. Eyles, Studies of biomolecular conformations and conformational dynamics by mass spectrometry, Mass Spectrom. Rev. 21 (2002) 37–71. [21] Z. Zhang, D.L. Smith, Determination of amide hydrogen exchange by mass spectrometry: a new tool for protein structure elucidation, Protein Sci. 2 (October) (1993) 522–531. [22] K.D. Powell, M.C. Fitzgerald, Accuracy and precision of a new H/D exchangeand mass spectrometry-based technique for measuring the thermodynamic properties of protein–peptide complexes, Biochemistry 42 (2003) 4962–4970. [23] L. Tang, P.L. Roulhac, M.C. Fitzgerald, H/D exchange and mass spectrometrybased method for biophysical analysis of multidomain proteins at the domain level, Anal. Chem. 79 (2007) 8728–8739. [24] G.M. West, L. Tang, M.C. Fitzgerald, Thermodynamic analysis of protein stability and ligand binding using a chemical modification- and mass spectrometrybased strategy, Anal. Chem. 80 (2008) 4175–4185. [25] K.C. Aune, C. Tanford, Thermodynamics of the denaturation of lysozyme by guanidine hydrochloride II: dependence on denaturant concentration at 25 ◦ C, Biochemistry 8 (1968) 4586–4590. [26] F. Greene, C.N. Pace, Urea and guanidine hydrochloride denaturation of ribonuclease, lysozyme, alpha-chymotrypsin, and beta-lactoglobulin, J. Biol. Chem. 249 (1974) 5388–5393. [27] Z.M. Shen, J.T. Yang, Y.M. Feng, C.S. Wu, Conformational stability of porcine serum transferrin, Protein Sci.: a Publication of the Protein Society 1 (1992) 1477–1484. [28] S. Ghaemmaghami, M.C. Fitzgerald, T.G. Oas, A quantitative, high-throughput screen for protein stability, Proc. Natl. Acad. Sci. U. S. A. 97 (2000) 8296–8301. [29] M.K. Santra, A. Banerjee, O. Rahaman, D. Panda, Unfolding pathways of human serum albumin: evidence for sequential unfolding and folding of its three domains, Int. J. Biol. Macromol. 37 (2005) 200–204. [30] B. Ahmad, R.H. Khan, Urea induced unfolding of F isomer of human serum albumin: a case study using multiple probes, Arch. Biochem. Biophys. 437 (2005) 159–167. [31] J. Mcnay, E. Fernandez, How does a protein unfold on a reversed-phase liquid chromatography surface? J. Chromatogr. A 849 (1999) 135–148. [32] M.F.M. Engel, A.J.W.G. Visser, C.P.M. van Mierlo, Conformation and orientation of a protein folding intermediate trapped by adsorption, Proc. Natl. Acad. Sci. U. S. A. 101 (2004) 11316–11321. [33] A.M. Gospodarek, M.E. Smatlak, J.P. O’Connell, E.J. Fernandez, Protein stability and structure in HIC: hydrogen exchange experiments and COREX calculations, Langmuir: ACS J. Surf. Colloids 27 (2011) 286–295. [34] D. Horinek, R.R. Netz, Can simulations quantitatively predict peptide transfer free energies to urea solutions? Thermodynamic concepts and force field limitations, J. Phys. Chem. A 115 (2011) 6125–6136. [35] B. Krishnan, L.M. Gierasch, Dynamic local unfolding in the serpin ␣-1 antitrypsin provides a mechanism for loop insertion and polymerization, Nat. Struct. Mol. Biol. 18 (2011) 222–226. [36] M. Yamasaki, T.J. Sendall, L.E. Harris, G.M.W. Lewis, J.A. Huntington, Loop-sheet mechanism of serpin polymerization tested by reactive center loop mutations, J. Biol. Chem. 285 (2010) 30752–30758.

A.M. Gospodarek et al. / J. Chromatogr. A 1371 (2014) 204–219 [37] Y. Bai, J.J. Englander, L. Mayne, J.S. Milne, S.W. Englander, Thermodynamic parameters from hydrogen exchange measurements, Methods Enzymol. 259 (1995) 344–356. [38] M.Y. Khan, S.K. Agarwal, S. Hangloo, Urea-induced structural transformations in bovine serum albumin, J. Biochem. 102 (1987) 313–317. [39] S. Tayyab, N. Sharma, M. Mushahid Khan, Use of domain specific ligands to study urea-induced unfolding of bovine serum albumin, Biochem. Biophys. Res. Commun. 277 (2000) 83–88. [40] D.M. Togashi, A.G. Ryder, D. O’Shaughnessy, Monitoring local unfolding of bovine serum albumin during denaturation using steady-state and time-resolved fluorescence spectroscopy, J. Fluoresc. 20 (2010) 441–452. [41] H. Wu, P. Wang, X. Hu, Z. Dai, X. Zou, Site-selective probe for investigating the asynchronous unfolding of domains in bovine serum albumin, Talanta 84 (2011) 881–886. [42] J.K. Myers, C.N. Pace, J.M. Scholtz, Denaturant m values and heat capacity changes: relation to changes in accessible surface areas of protein unfolding, Protein Sci. 4 (1995) 98. [43] V.J. Lin, J.L. Koenig, Raman studies of bovine serum albumin, Biopolymers 15 (1976) 203–218. [44] K. Tokihiro, T. Irie, F. Hirayama, K. Uekama, Mass spectroscopic evidence on inhibiting effect of matosyl-a-cyclodextrin on insulin self-association, Pharm. Sci. 2 (1996) 519–522.

219

[45] P. Man, C. Montagner, H. Vitrac, D. Kavan, S. Pichard, D. Gillet, et al., Accessibility changes within diphtheria toxin T domain when in the functional molten globule state, as determined using hydrogen/deuterium exchange measurements, FEBS J. 277 (2010) 653–662. [46] W. Qi, A. Zhang, D. Patel, S. Lee, J.L. Harrington, L. Zhao, et al., Simultaneous monitoring of peptide aggregate distributions, structure, and kinetics using amide hydrogen exchange: application to Abeta(1–40) fibrillogenesis, Biotechnol. Bioeng. 100 (2008) 1214–1227. [47] E.J. Suda, K.E. Thomas, T.M. Pabst, P. Mensah, N. Ramasubramanyan, M.E. Gustafson, et al., Comparison of agarose and dextran-grafted agarose strong ion exchangers for the separation of protein aggregates, J. Chromatogr. A 1216 (2009) 5256–5264. [48] V. Sluzky, J.A. Tamada, A.M. Klibanov, R. Langer, Kinetics of insulin aggregation in aqueous solutions upon agitation in the presence of hydrophobic surfaces, Proc. Natl. Acad. Sci. U. S. A. 88 (1991) 9377–9381. [49] T. Vermonden, C.E. Giacomelli, W. Norde, Reversibility of structural rearrangements in bovine serum albumin during homomolecular exchange from AgI particles, Langmuir 17 (2001) 3734–3740. [50] W. Norde, C.E. Giacomelli, BSA structural changes during homomolecular exchange between the adsorbed and the dissolved states, J. Biotechnol. 79 (2000) 259–268. [51] A. Sethuraman, G. Belfort, Protein structural perturbation and aggregation on homogeneous surfaces, Biophys. J. 88 (2005) 1322–1333.