Hydrophobic interaction chromatography of proteins

Hydrophobic interaction chromatography of proteins

Journal of Chromatography A, 1141 (2007) 191–205 Hydrophobic interaction chromatography of proteins I. The effects of protein and adsorbent propertie...

1MB Sizes 27 Downloads 368 Views

Journal of Chromatography A, 1141 (2007) 191–205

Hydrophobic interaction chromatography of proteins I. The effects of protein and adsorbent properties on retention and recovery Brian C.S. To a,b , Abraham M. Lenhoff b,∗ a

b

Merck Research Laboratories, Sumneytown Pike, West Point, PA 19486, USA Department of Chemical Engineering, University of Delaware, Newark, DE 19716, USA

Received 2 June 2006; received in revised form 30 November 2006; accepted 5 December 2006 Available online 4 January 2007

Abstract The contributions of protein and adsorbent properties to retention and recovery were examined for hydrophobic interaction chromatography (HIC) using eight commercially available phenyl media and five model proteins (ribonuclease A, lysozyme, ␣-lactalbumin, ovalbumin and BSA). The physical properties of the adsorbents were determined by inverse size exclusion chromatography (ISEC). The adsorbents examined differ from each other in terms of base matrix, ligand density, porosity, mean pore radius, pore size distribution (PSD) and phase ratio, allowing systematic studies to understand how these properties affect protein retention and recovery in HIC media. The proteins differ in such properties as adiabatic compressibility and molecular mass. The retention factors of the proteins in the media were determined by isocratic elution. The results show a very clear trend in that proteins with high adiabatic compressibility (higher flexibility) were more strongly retained. For proteins with similar adiabatic compressibilities, those with higher molecular mass showed stronger retention in Sepharose media, but this trend was not observed in adsorbents with polymethacrylate and polystyrene divinylbenzene base matrices. This observation could be related to protein recovery, which was sensitive to protein flexibility, molecular size, and conformation as well as the ligand densities and base matrices of the adsorbents. Low protein recovery during isocratic elution could affect the interpretation of protein selectivity results in HIC media. The retention data were fitted to a previously published retention model based on the preferential interaction theory, in terms of which retention is driven by release of water molecules and ions upon protein-adsorbent interaction. The calculated number of water molecules released was found to be statistically independent of protein retention strength and adsorbent and protein properties. © 2006 Elsevier B.V. All rights reserved. Keywords: Inverse size exclusion chromatography; Phenyl adsorbents; Pore size distribution; Phase ratio; Adiabatic compressibility; Protein recovery

1. Introduction Although ion exchange chromatography (IEC) remains the most widely used chromatographic method for large-scale separations, hydrophobic interaction chromatography (HIC) is also widely used. HIC is usually assumed to separate proteins by the differences in their surface hydrophobic character. Therefore, it provides an orthogonal chromatography technique to IEC, affinity and size exclusion chromatography for biomolecule purification. HIC and reverse phase chromatography (RPC) are assumed to achieve protein separation based on similar prin-

ciples. However, RPC employs highly hydrophobic surfaces and concomitantly harsh eluents, with the resulting protein denaturation making the method suitable mainly for analytical separations. HIC uses less hydrophobic surfaces, and because of the milder protein-ligand interactions, proteins purified by HIC usually retain their native conformations and biological activities. HIC has been used to purify such products as antibodies [1,2], recombinant proteins [3], and plasmid DNA [4,5]. Most quantitative data and analysis in HIC studies have focused on retention. The retention factor, k , under linear retention conditions can be expressed as k = Keq φ



Corresponding author. Tel.: +1 302 831 8989; fax: +1 302 831 4466. E-mail address: [email protected] (A.M. Lenhoff).

0021-9673/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.chroma.2006.12.020

(1)

where Keq is the adsorption equilibrium constant of a solute on the stationary phase and φ is the phase ratio of the adsorbent.

192

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

Thus, understanding the quantitative basis for linear retention requires probing both the thermodynamic factors captured by Keq and the structural ones reflected by φ. The thermodynamic factors relevant to protein retention in HIC are dominated by the high sensitivity to salt concentration and the chemical nature of the salt. Protein adsorption in HIC is induced by high salt concentrations such as those used in precipitation, and indeed in its nascent stages HIC was sometimes referred to as salting-out chromatography [6]. Later experimental results showed that salts with higher “salting-out” ability increase protein retention in HIC [7]. Thus, analysis of retention has been guided by that of protein solubility, with solvophobic theory [8–12] and preferential interaction theory [13–18] the approaches most commonly used. In addition to the solvophobic theory and preferential interaction theory, other retention models have also been developed [19,20], although they have not been widely used. The model developed by Staby and Mollerup [20] shows that salts that increase the activity coefficient of the protein in the mobile phase improve retention. This emphasizes an important general point, namely that protein hydrophobicity is not the only factor affecting retention in HIC. Eq. (1) indicates a direct dependence of retention on the physical structure of the stationary phase in addition to the thermodynamics of protein solutions and adsorption. This explicit dependence on a quantitative stationary-phase property conceals the possibility that adsorbent physical properties may also affect retention in other ways. For example, it has been shown that cation-exchange adsorbents with pore size distributions (PSD) containing a significant amount of pore space with dimensions similar to those of the protein solute display increased protein retention [21]. Assuming that the phase ratio is the principal stationaryphase physical property of interest, it is the lack of comprehensive information on the physical properties of HIC media that makes it difficult to compare and analyze experimental results generated with different media. Mercury porosimetry and nitrogen adsorption are two methods commonly used to measure the PSD and surface area of porous solids. However, they are performed under conditions that are not relevant to chromatography applications. To obtain more accurate column void volumes, a NMR method was developed [22], but it tends to underestimate the intraparticle porosity for large probe molecules due to pore blockage caused by protein adsorption. Inverse size exclusion chromatography (ISEC) is an alternative method for determining the PSD and accessible surface area of chromatographic media [23–26] that has the advantages of simplicity and being performed under normal chromatographic conditions; this approach is used here. The other contribution to Eq. (1) is the adsorption equilibrium constant, the estimation of which is the objective of retention models. The retention models outlined above are largely conceptual. They typically account for the salt concentration and salt type via parameters such as the molal surface tension increment or the preferential interaction parameter. However, they do not account for properties such as ligand type, ligand density and most notably, the structural properties of the protein. This issue

was partially addressed by a mathematical model developed to predict retention based on protein average surface hydrophobicity [27]. A key additional complicating factor to be considered in HIC is that a protein’s native conformation can change when it is adsorbed onto a hydrophobic surface [18,28–30]. Jones and Fernandez [31] and Wu et al. [18] showed that the adsorption of ␣-lactalbumin onto HIC surfaces could cause the protein to unfold. Therefore, understanding the correlation between protein and adsorbent physical properties and retention can shed light on the mechanisms of protein retention in HIC. It is well known that proteins can undergo conformational changes under high pressure [32]. Dadarlat and Post [33] showed that proteins with high compressibilities have higher heat capacities per residue, Cp , upon unfolding. Murphy et al. [34] showed that the change in entropy per residue, S, of proteins decreases with increasing Cp when proteins unfold. These two results suggest that proteins with high compressibilities also have high conformational entropies, hence flexibilities, in their native states. This is consistent with the correlation that proteins with higher compressibilities are more thermally stable [35]. On the basis of these results, adiabatic compressibilities, ks , of proteins were used to reflect their flexibilities in this study. Studies also indicate that proteins with high ks /Cp tend to have more hydrophobic cores [35,36]. Adiabatic compressibility has been used to probe conformational changes in proteins adsorbed on silica particles [37], polystyrene particles [38] and reverse phase media [39]. It was found that higher adiabatic compressibilities lead to more extensive conformational changes when proteins are adsorbed onto solid surfaces [37,38]. Mahn et al. [40] showed that ribonuclease S has higher retention on a butyl adsorbent than ribonuclease A, ribonuclease T1 wild type and ribonuclease TI variant although they all have almost identical surface hydrophobicity. It was suggested that the higher flexibility of ribonuclease S could be responsible for the stronger retention. On the basis of these results, protein flexibility could have a profound effect on protein retention in HIC. Extensive experimental studies have been conducted to investigate the effects of the salt [30,41,42], ligand type [7,42–47], ligand density [7,48], pH [7,18,42,49–51], and temperature [18,43,49,52,53] on protein retention in HIC. In general, high concentrations of kosmotropic salts promote retention. All these studies suggest that the performance of HIC can be optimized by using appropriate combinations of these factors, but not yet in a predictable fashion. Although many different sets of protein retention data in HIC are available in the literature, they were generated with different adsorbents and proteins under different conditions. The paucity of large sets of retention data in which individual factors are varied systematically makes it difficult to analyze how different factors contribute to protein retention in HIC. Two very large data sets providing retention curves for proteins on commercially available stationary phases have appeared recently [54–56], but little mechanistic analysis was provided. In this paper, the physical structure and PSD of eight commercially available HIC media were determined. The effects of the adsorbent characteristics, as well as of the protein molecular size, surface properties and structural flexibility on retention on

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

193

Table 1 Physical properties of phenyl media used Stationary phase

Particle size (␮m)

Base matrix

Degree of substitution (␮mol/mL resin)

Tosoh Biosep TSK-gel Phenyl-5PW Toyopearl Phenyl 650S

30 35

Methacrylate Methacrylate

67 [48] –

GE Healthcare Phenyl Sepharose HP Phenyl Sepharose FF LS Phenyl Sepharose FF HS Source 15PHE

24–44 45–165 45–165 15

Agarose Agarose Agarose PS-DVB

25 20 40 –

EM Industries Fractogel EMD Phenyl (S)

20–40

Methacrylate



Applied Biosystems POROS 20 HP2

20

PS-DVB



these HIC adsorbents were systematically studied for five model proteins.

egg white lysozyme and hen egg white ovalbumin were purchased from Sigma (St. Louis, MO, USA).

2. Materials and methods

2.3. Dextran and polyacrylamide standards for media molecular calibration

2.1. Chromatographic stationary phases Eight phenyl stationary phases were selected for this study; their physical properties are summarized in Table 1. TSKgel Phenyl-5PW and Toyopearl Phenyl 650S were purchased from Tosoh Biosep (Montgomeryville, PA, USA). All Phenyl Sepharose media and Source 15PHE were purchased from GE Healthcare (Piscataway, NJ, USA). Fractogel EMD Phenyl (S) was obtained from EM Industries (US associate of E. Merck, Darmstadt, Germany). POROS 20 HP2 was purchased from Applied Biosystems (Foster City, CA, USA). 2.2. Proteins Five globular proteins were used; their physical properties, together with those of proteins used in other studies cited in this paper, are shown in Table 2. The ks values were the average values of the experimental results found in the literature [35,57,58]. Bovine pancreatic ribonuclease A, bovine serum albumin monomer, calcium-depleted bovine ␣-lactalbumin, hen

Thirteen dextran and two polyacrylamide standards were used to calibrate the pore structures of the HIC stationary phases. The peak molecular masses and solution concentrations used for the calibration are shown in Table 3. The viscosity radii, Rη , of the dextran standards were calculated using the equation Rη = 0.0271 × Mp0.498 [59]. For polyacrylamide standards, Rη was calculated as [60]   3[η]Mp 1/3 Rη = (2) 10πNA where η is the intrinsic viscosity in cm3 /g, Mp is the peak molecular mass and NA is Avogadro’s number. Rη captures the dependence of the size exclusion chromatography elution volume on the molecular mass and shape of the probe molecules [61]. The polyacrylamide standards and dextran standards with Mp values of 805,000, 2,285,000 and 9,110,000 were obtained from American Polymer Standards Corporation (Mentor, OH, USA). All other dextran standards were purchased from Polymer Standards Service—USA (Silver Spring, MD, USA).

Table 2 Molecular masses (MW), viscosity radii (Rη ), adiabatic compressibilities (ks ) and surface properties of proteins used in HIC retention experiments Protein

MW (kDa)

Rη (nm) [62]

˚ 2 ) [58] Polar area (A

˚ 2 ) [58] Non-polar area (A

ks (×10−6 cm3 /g/bar)a

Ribonuclease A Cytochrome C Lysozyme ␣-Lactalbumin Myoglobin Trypsin ␣-Chymotrypsin ␣-Chymotrypsinogen A Ovalbumin BSA

13.7 12.4 14.3 14.3 17.8 23.8 21.6 25 46 68

1.75 1.63 1.85 2.02 1.91 2.3 2.2 2.3 2.83 3.62

2401 1503 2548 1874 1702 3869 3991 3936 5180 –

3462 3446 3230 3794 4772 4809 5426 5714 8597 –

0.8 1.2 3.6 4.6 5.8 1.2 2.9 3.7 6.4 6.5

a

Average values of references [35,57,58].

194

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

Table 3 Peak molecular mass (Mp ), viscosity radius (Rη ), and sample concentration of dextran and polyacrylamide standards used for column calibration Mp

Rη (nm)

Concentration (mg/mL)

Dextran 180 830 4,400 9,900 21,400 43,500 124,000 196,000 560,000 805,000 1,450,000 2,285,000 9,110,000 Polyacrylamide 3,965,000 6,500,000

0.36 0.77 1.77 2.65 3.89 5.53 9.32 11.7 19.8 23.7 31.7 39.8 79.2 84.7 114.6

ing 0.5 m AS, 10 mM phosphate, 10 mM CaCl2 and 2 m AS, 10 mM phosphate, 10 mM CaCl2 , pH 7, buffers.

15 15 15 15 15 15 15 15 15 15 15 15 15 5 2

2.4. Experimental procedures 2.4.1. Protein retention measurements AP minicolumns with diameters of 0.5 cm were purchased from Waters Corporation (Milford, MA, USA). Approximately 1 mL of each adsorbent was flow packed into the columns, and the final bed heights were 5.1–5.3 cm. The protein retention experiments were performed with a HPLC system equipped with a GP40 gradient pump and a 100 ␮L injection loop (Dionex Corporation, Sunnyvale, CA, USA). Protein elution was detected by a Rainin Dynamax UV detector (Rainin Instruments, Woburn, MA, USA). Proteins were dissolved in phosphate buffer with the same ammonium sulfate (AS) concentration as the mobile phase to be used to a concentration of 1.2 mg/mL or 10 mg/mL. Isocratic elution was performed with 10 mM phosphate buffer containing different amounts of AS at a flow rate of 0.5 mL/min. The protein injection volume was 50 ␮L. The desired AS concentrations were obtained by blending mobile phases A and B, comprising 10 mM phosphate, pH 7, buffer and 2 molal (m) AS in 10 mM phosphate, pH 7, buffer respectively. Each isocratic elution measurement took 10–360 min, depending on the protein and AS concentration. After each isocratic run, the column was washed with 5–10 column volumes of mobile phase A and then equilibrated with 5 column volumes of buffer containing the appropriate AS concentration prior to the next sample injection. Isocratic elution of ␣-lactalbumin was also performed with mobile phases containing 10 mM CaCl2 to investigate the effect of calcium ions on ␣-lactalbumin retention and recovery. Under normal conditions, Ca2+ forms insoluble precipitates with sulfate and phosphate. However, when 10 mM calcium chloride solution was added into mobile phases containing 0.5 m or more AS, no precipitation was observed. In this set of experiments, proteins were dissolved in phosphate buffer containing the appropriate AS concentration and 10 mM CaCl2 . The desired AS concentration during isocratic elution was obtained by mix-

2.4.2. ISEC procedures The same HPLC system as described above was used for the ISEC measurements, except that detection was by a differential refractometer (Waters Corporation, Milford, MA, USA). With the exception of Source 15PHE and POROS 20 HP2, each adsorbent was washed and decanted five times with 10 mM phosphate buffer, pH 7, to remove preservatives. At the end of the last decantation, the supernatant was removed and 1 M AS in 10 mM phosphate buffer, pH 7, was added to the adsorbent to prepare a 50% slurry. The slurry was then transferred to a GE Healthcare XK 16/70 column equipped with a 300 mL packing reservoir. The adsorbent in the slurry was gravity settled into the glass column and flow packed with 1 M AS in 10 mM phosphate buffer, pH 7, at 12–30 mL/min (360–900 cm/h) until a constant bed height was obtained. Source 15PHE and POROS 20 HP2 adsorbents tended to aggregate in buffer containing AS. Therefore, ethanol was used to prepare the 50% slurry. The slurry was transferred to the columns and flow packed in ethanol at 4 mL/min (60 cm/h). All packed adsorbents were equilibrated with the appropriate mobile phase at 1.5 mL/min until a stable RI baseline was obtained. The bed heights of the packed beds were 49 ± 3 cm. The dextran and polyacrylamide standards were dissolved in the same mobile phases used in the ISEC molecular calibration experiments. Isocratic elution was performed at a volumetric flow rate of 1.5 mL/min. The sample injection volume was 50 ␮L. 2.5. Isocratic elution data analysis The elution peaks obtained from the isocratic elution experiments were non-Gaussian, with long tails observed at high salt concentrations; Fig. 1 shows examples of some peaks, which are discussed in greater detail later. In some cases, the tails did not return to baseline even after 4–5 h of elution. Therefore, the peak maximum is not an accurate representation of the actual protein retention, and the retention times were calculated from the first moments of the elution peaks. The peak area and first moment were estimated by the trapezoidal rule using a time interval of 0.2 s between data points. For elution peaks with tails that did not return to baseline, the last 5–10 min of data were fitted to an exponential function that was extrapolated to the baseline and integrated analytically. In all cases, the areas under the tails contributed less than 5% of the total peak area. Protein recovery during isocratic elution was determined from the ratio of the peak areas obtained during isocratic elution and from an identical injection in the absence of a column. 2.6. ISEC analysis The pore size distributions of the adsorbents were characterized by ISEC. The distribution coefficients of the standards were

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

195

where A is a normalization constant. The parameters rp and sp characterize the PSD of the material, with rp representing the mode and sp the width of the distribution. Eq. (4) expresses Kd in terms of the pore volume accessible to a probe of radius rm . The distribution parameters, rp and sp , were determined by leastsquares fitting of experimental Kd values using the International Mathematical and Statistical Libraries (IMSL) routine DRNLIN, with the integrals evaluated using the routines DQDAG and DQDAGI. The mean pore radius, r¯ , and accessible surface area per unit pore volume, A(rm ), were calculated from ∞ rf (r) dr r¯ = 0∞ (6) 0 f (r) dr ∞ 2 r (2(r − rm )/r )f (r) dr ∞ A(rm ) = m (7) 0 f (r) dr The phase ratio (accessible area per unit accessible mobile phase volume) for a probe of radius rm was found from φ = A(rm ) Fig. 1. Effect of on the shape of elution peak of ␣-lactalbumin during isocratic elution on Phenyl Sepharose HP. (A) With 10 mM Ca2+ ; (B) without Ca2+ . Ca2+

calculated from their retention volumes by Kd =

VR − Vo VT − V o

(3)

where VR is the retention volume of the standard with radius rm , VT is the total mobile phase volume, and Vo is the interparticle void volume. In this study, VT was determined using glucose while Vo was measured using dextran or polyacrylamide standards that were excluded from the pores of the adsorbent. By definition, Kd measures the fraction of total pore volume accessible by a probe molecule with a certain radius. For a material with a volumetric distribution of pore size f(r), the assumption that partitioning is of spheres into cylindrical pores leads to [62] ∞ 2 r f (r)[1 − (rm /r)] dr Kd = m  ∞ (4) 0 f (r) dr In this work, f(r) was assumed to follow a log normal distribution     1 ln(r/rp ) 2 A (5) f (r) = exp − sp r 2

VR − V o VR

(8)

The total intraparticle porosity of the material was calculated from VT − Vo εp = (9) VB − V o where VB is the bed volume. 3. Results and discussion 3.1. HIC media characterization For each adsorbent, ISEC was performed with mobile phases containing 0.2 and 1.2 m AS. The salt concentration had no significant impact on the retention volumes of the probe molecules. This suggests that AS neither promoted interactions between the chromatographic media and the probe molecules nor affected the pore structures of the media. The fitting parameters, porosities, mean pore radii, total mobile phase volumes, and interparticle void volumes determined by ISEC are summarized in Table 4. The calculated PSD and phase ratios of the media are shown in Figs. 2 and 3, respectively. All Sepharose stationary phases have similar porosities and phase ratios. The phase ratios of these media change little for Rη between 0.36 and 1.77 nm and decrease monotonically with

Table 4 PSD fitting parameters (rp and sp ), intraparticle porosities (εp ), total mobile phase volumes (VT ), interparticle void volumes (Vo ), and mean pore radii (¯r ) of HIC media determined by ISEC

rp (nm) sp εp VT Vo r¯ (nm)

Sepharose HP

Sepharose FF LS

Sepharose FF HS

TSK-gel 5PW

Toyopearl 650S

Fractogel EMD

Source 15PHE

POROS 20 HP2

34.7 0.49 0.90 0.93 0.35 39.1

27.3 0.01 0.91 0.94 0.37 27.3

27.0 0.02 0.93 0.96 0.34 27.0

91.2 0.52 0.58 0.75 0.40 104.2

69.8 0.92 0.70 0.81 0.35 106.9

27.1 1.00 0.40 0.64 0.40 44.9

36.0 0.55 0.48 0.69 0.40 41.7

207.5 0.86 0.62 0.77 0.40 300.5

VT and Vo are expressed as fractions of the bed volume.

196

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

Fig. 2. Fitted PSDs of HIC media. (A) Agarose media; (B) polymethacrylate media; (C) polystyrene-divinylbenzene media.

Fig. 3. Calculated phase ratios of HIC media. (A) Agarose media; (B) polymethacrylate media; (C) polystyrene-divinylbenzene media.

probe size in the range Rη = 2.65–19.75 nm. Dextrans with Rη higher than 30 nm were totally excluded. This indicates that the pores in the Sepharose media are accessible to most proteins of practical interest [63]. Sepharose HP has a wider PSD and larger mean pore radius than the Sepharose FF adsorbents. The narrow PSDs of Sepharose FF LS and Sepharose FF HS suggest that they have relatively uniform pore dimensions. These two media have similar mean pore radii but narrower PSD than the cation-exchange adsorbents manufactured on the same base matrix [24]. The difference could be due to lot-to-lot variations in the base media, or the incorporation of different functional groups onto the same base matrix may have some impact on the pore structures of the final media. TSK-gel 5PW and Toyopearl 650S have similar mean pore radii but different PSDs. Both media are manufactured on the same base matrix, but TSK-gel 5PW is more highly cross-linked to improve its rigidity, and this may be responsible for the differences in porosity, PSD and phase ratios. The phase ratio of TSK-gel 5PW is lower than that of Toyopearl 650S, especially at low Rη . The relatively flat region of the phase ratio curve of TSK-gel 5PW at low Rη (<10 nm) extends to higher radii than that of Toyopearl 650S. This suggests that TSK-gel 5PW con-

tains more large pores than Toyopearl 650S, although both media have similar mean pore sizes. This result is consistent with the fitted PSD curves of the two media. Fractogel EMD and Toyopearl 650S are also manufactured on the same base matrix. The phenyl groups in Toyopearl 650S are grafted directly onto the surface, while the functional groups in Fractogel EMD are on polymer “tentacles” attached to the surface [64]. The tentacles are intended to improve the binding capacity via the extra binding sites on the tentacles [65]. The presence of the tentacles reduces both the mean pore size and porosity of the resulting adsorbents and changes the PSD significantly, with more pore volume accessible only to smaller probe molecules. This is consistent with the high sensitivity of the phase ratio to probe size for small Rη . The pores in POROS 20 HP2 are accessible to probe molecules with Rη larger than 100 nm. The POROS medium contains large pores (>500 nm) connected to smaller pores (30–70 nm) [66], which is intended to allow intraparticle convection at high flow rates. The mean pore size of POROS 20 HP2 is an order of magnitude larger than that of Source 15PHE and it was difficult to find a probe molecule large enough to be totally excluded from the POROS adsorbent. As a result, an interparticle

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

void volume (Vo ) of 0.4 was assumed in calculating the porosity of this material. This void volume was consistent with those of the other rigid adsorbents used in this study. The porosity calculated based on this assumption is in excellent agreement with the result obtained from mercury porosimetry [67]. Although the POROS media may display improved separation efficiency, its large pore size also results in phase ratios much lower than those of Source 15PHE, and this lack of surface area could result in low protein binding capacity. 3.2. Protein retention To investigate the effect of protein loading, elution experiments were performed with the same injection volumes of protein solutions at 1.2 or 10 mg/mL. The two concentrations gave identical trends although the lower concentration gave larger k values in some cases (data not shown). However, low protein concentrations in the samples limited the salt concentration ranges possible for the experiments due to wide elution peaks and low protein recoveries at high salt concentrations. To maximize the salt concentration range for this study, all elution experiments for which results are presented were performed with an injection protein concentration of 10 mg/mL. The effects of AS concentration on protein retention are summarized in Fig. 4, in which the k values for all the proteins examined are shown on a separate plot for each stationary phase. The ln k curves are, in general, non-linear in the mobile phase salt concentration, consistent with earlier results [17,54,68]. The strength of protein retention on the Sepharose adsorbents generally follows the order BSA > ovalbumin > ␣lactalbumin > lysozyme > ribonuclease A, especially at higher salt concentrations. For the polymethacrylate-based adsorbents, ribonuclease A is the least retained protein, but the differences in retention of the other proteins are less pronounced. This could be due to reduced protein recovery, which is discussed later. The retention curves of ␣-lactalbumin and BSA on the PS-DVB media are not shown due to low protein recoveries. As observed for the other adsorbents, ribonuclease A displays the weakest retention. The retention curves of ovalbumin and lysozyme cross at about 1.2 M AS on all other adsorbents. Similar patterns were also observed for other protein pairs [55]. Eq. (1) shows that k is affected by two factors, the solute adsorption equilibrium constant, Keq , which reflects the intrinsic thermodynamics of adsorption, and the protein-accessible surface area per mobile phase volume, φ, which is a physical property of the adsorbent. The phase ratios of the stationary phases for different proteins were estimated from the ISEC results for dextrans with Rη similar to those of the proteins. As shown in Fig. 3, the phase ratios of some media are as much as an order of magnitude higher than others, so k is not the most reliable parameter for direct comparisons of binding affinities. Keq can be interpreted as the retention factor normalized by the protein-accessible surface area, so it can be used to assess differences in protein affinity among different media. Because k is dimensionless while φ has units of m2 /mL, Eq. (1) indicates that Keq has dimensions of length; this is because it represents the ratio of a surface concentration and a volumetric concentration.

197

The protein retention curves based on Keq are depicted in Fig. 5, and show that the trends of the normalized retention curves are identical to those of the k curves. This demonstrates that the trends in protein retention are similar for all adsorbents although the adsorbents have different physical properties. Therefore, the effects of protein properties on HIC retention are paramount and are discussed in the following sections. 3.2.1. The effect of protein conformation At high salt concentrations, significant peak tailing was observed for some proteins (Fig. 1). The asymmetric peak shape could be caused by the presence of partially unfolded protein, which would have different sorption kinetics and thermodynamics on the HIC surface compared to native protein [18,49,69]. The role of protein conformation in HIC retention can also be studied more specifically by the effect of Ca2+ on ␣-lactalbumin retention [18,31]. ␣-Lactalbumin is a calcium-binding protein, the tertiary structure of which is stabilized by Ca2+ ion [70], so that the removal of Ca2+ promotes conformational change [71]. Crystal structures show that the conformations of the apo and holo ␣-lactalbumin are different in terms of the separation distance of the helical and ␤-loop subdomains, the number of water molecules bound to the protein and the inter- and intraloop hydrogen bonds of Tyr 103 [72]. The ␣-lactalbumin used in this study was calcium-depleted, but for most of the retention experiments, 10 mM Ca2+ was added to both the protein samples and the mobile phases. Isocratic elution of ␣-lactalbumin was also performed without Ca2+ to investigate the effect of conformational change on the shapes of elution peaks and on retention. Fig. 1 shows that Ca2+ has a significant impact on the shapes of the ␣-lactalbumin elution peaks. In the presence of Ca2+ , the elution peaks are asymmetric, but the peak shapes are insensitive to salt concentration, although more peak spreading is observed with increasing retention at higher salt concentrations. Without Ca2+ , in contrast, the elution peaks have long tails and the peak shapes are very sensitive to the AS concentration. In fact, a transition in peak shape is observed in Fig. 1B at 1.1 m AS that may be due to changes in protein conformation. The differences in the shapes of the elution peaks suggest that apo and holo ␣-lactalbumin have different retention behavior on the same HIC media. Two distinct peaks are observed on the chromatograms at AS concentrations >0.5 m for the experiments performed either with or without Ca2+ . In all cases, the first peak eluted at ∼2.5 min, which was close to the dead time of ∼2 min. The areas of all of these first elution peaks account for ∼3% of the total peak area. Wu et al. [18] and Jones and Fernandez [31] also observed two peaks for pure ␣-lactalbumin in HIC columns and both showed that the first peak was native ␣-lactalbumin while the second peak was unfolded ␣-lactalbumin. However, the first peak was a significantly larger fraction of the total peak area than was observed in this study. Similarly, isocratic elution of ovalbumin and BSA also gave two distinct peaks. Since the retention of the first peaks was independent of AS concentration, they were not included in the first moment calculations. These peaks may have represented native ␣-lactalbumin as in the previous reports, or an impurity.

198

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

Fig. 4. Effects of AS concentration on retention factors, k , of proteins on different HIC media. (♦) Ribonuclease A; () lysozyme; () α-lactalbumin with 10 mM Ca2+ ; () ovalbumin; (*) BSA.

Fig. 6 shows the effect of Ca2+ on ␣-lactalbumin retention. In Sepharose HP, the removal of Ca2+ from the protein sample and mobile phase increased the retention of ␣-lactalbumin, consistent with what has been reported [18]. In TSK-gel 5PW, the removal of Ca2+ ions had no effect on retention, while the removal of Ca2+ reduced retention in Fractogel EMD and Source 15PHE. This trend can be explained by the fraction of protein recovered. With the exception of Source 15PHE, ␣-lactalbumin recovery on all media was 90–100% in the presence of Ca2+ (Fig. 7). When the experiments were performed without Ca2+ , ␣lactalbumin recovery remained constant at ∼80% in Sepharose

HP but was significantly lower in the other three media, especially when the salt concentration was above 1 m. With similar recoveries, ␣-lactalbumin with lower conformational stability is more strongly retained on Sepharose HP. This is consistent with the observation that a less stable conformation exposes the protein interior for interaction with the adsorbent [31]. In general, the adsorbents with low ␣-lactalbumin recoveries in the absence of Ca2+ displayed stronger retention of other proteins compared to Sepharose HP. This is reflected by the adsorption equilibrium constants summarized in Table 5. The strong media–protein interaction may result in irreversible binding of a portion of the

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

199

Fig. 5. Effects of AS concentration on adsorption equilibrium constants, Keq , of proteins on different HIC media. (♦) Ribonuclease A; () lysozyme; () ()lactalbumin with 10 mM Ca2+ ; () ovalbumin; (*) BSA.

injected protein that cannot be eluted under isocratic elution conditions on the time scale of the experiment. When the recovery is low, the elution peaks represent the retention of only the less retained protein fraction. As a result, the values of the first moments calculated from the elution peaks could be equal to or lower than those obtained from experiments with higher recoveries. This explanation is consistent with the Keq values shown in Fig. 6A and B. Therefore, special attention to protein recovery is needed when comparing protein retention in HIC. This analysis also suggests that Sepharose HP appears to cause less conformational change than the other media for which data are presented in Fig. 6. Without information on the ligand density of the adsor-

bents, it is difficult to identify conclusively why Sepharose HP causes less conformational change. A possible cause is interaction with the base matrices of the adsorbents, which also affects protein recovery (Section 3.3). 3.2.2. The effects of protein adiabatic compressibility Norde suggested that protein adsorption onto solid surfaces is driven partially by an increase in entropy related to protein structural changes [73]. A more expanded structure lowers the free energy of the protein molecule when it interacts with an adsorbent. The extent of protein conformational change would be expected to be related to the flexibility of the protein tertiary

200

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

Fig. 7. Effect of Ca2+ on ␣-lactalbumin recovery during isocratic elution. Squares: Sepharose HP; Triangles: TSK-gel 5PW; Diamonds: Fractogel EMD; Circles: Source 15PHE. Solid symbols: without Ca2+ ; open:10 mM Ca2+ .

Fig. 6. The effect of Ca2+ on ␣-lactalbumin retention. () Without Ca2+ ; (): with 10 mM Ca2+ .

structure, which in turn is reflected by the adiabatic compressibility [33,34]. Therefore, the correlation between ks and protein retention can shed light on the role of protein flexibility in protein retention mechanisms in HIC. As shown in Figs. 4 and 5, for proteins of similar size, ribonuclease A had the lowest retention. The retention of ␣-lactalbumin and lysozyme was similar, with that of ␣-lactalbumin lower at low salt concentration but higher at high salt concentration. The

same trend was observed on all the media examined (except POROS 20 HP2, on which retention data of ␣-lactalbumin and BSA were not obtained due to low protein recovery), indicating that the differences in retention among these three proteins are due mainly to differences in protein properties. Since both hydrophobic and hydrophilic amino acids are involved in interactions with a HIC surface [51], the effect of protein surface properties in HIC retention was investigated. The polar and non-polar surface areas of atomic groups on 12 globular proteins were calculated using the “contour buildup” algorithm [58]. Results show that ribonuclease A, lysozyme and ␣-lactalbumin have similar polar and non-polar surface areas (Table 2), although their retention extents are quite different. Therefore, the retention behavior of these three proteins cannot be fully explained by their surface properties. However, there is a clear trend of increasing retention with higher ks at high salt concentration. This is consistent with the results that more flexible ribonuclease S has higher retention on a HIC adsorbent than other ribonuclease variants although they all have identical surface hydrophobicity [40]. From this analysis, protein retention in HIC appears to be correlated with protein flexibility. For a protein with low ks , its rigid tertiary structure can limit conformational change when it binds to a HIC surface. As a result, only some of the external surface area may be available to interact with the adsorbent. On the other hand, proteins with high ks are more flexible, so when they come into contact with the HIC surface, they can spread, unfold or expose internal surface to maximize the interaction with the surface, resulting in higher retention. In fact, conformational changes have been observed when proteins were adsorbed onto ion-exchange membranes [74], HIC media [31], reversephase media [75–77], silica [37,78] and negatively charged polystyrene particles [38]. Isocratic elution experiments with trypsin, ␣-chymotrypsin and ␣-chymotrypsinogen A [12] and lysozyme and ␣-lactalbumin [54] also showed that for proteins with similar molecular masses, those with higher ks were more highly retained, although these experiments were not designed to correlate protein retention and adiabatic compressibility. The adiabatic compressibility is actually the sum of the intrinsic compressibility of the protein and the compressibility effect of hydration [58]. This suggests that ks may be a function of salt type and salt concentration. It is not clear if the AS concentra-

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

201

Table 5 Comparisons of retention factors, k , and adsorption equilibrium constants, Keq , of different HIC media at selected ammonium sulfate concentrations Salt conc. (m)

Sepharose HP

Sepharose FF LS

TSK-gel 5PW

Toyopearl 650S

Fractogel EMD

Source 15PHE

POROS 20 HP2

RNA LYS ALA OVA BSA

2 1.4 1.4 1.4 1

16.1 9.25 21.0 32.9 6.35

15.6 11.5 21.6 36.8 16.7

14.1 6.86 25.4 15.9 0.64

14.6 12.3 12.8 36.7 0.58

15.0 20.8 32.8 64.9 2.18

16.8 7.25

5.18 3.17

18.8

4.50

Keq (␮m) RNA LYS ALA OVA BSA

2 1.4 1.4 1.4 1

0.44 0.26 0.58 0.97 0.21

1.29 0.63 2.32 1.50 0.06

0.67 0.56 0.59 1.80 0.03

0.53 0.73 1.15 2.79 0.12

0.72 0.31

0.80 0.49

0.88

0.71

k

0.39 0.29 0.54 0.99 0.50

RNA: ribonuclease A; LYS: lysozyme; ALA: ␣-lactalbumin with 10 mM Ca2+ ; OVA: ovalbumin.

tions used for the elution experiments affected the ks , and if so, whether all proteins were affected to the same extent. However, experimental results obtained from this study and in the literature clearly show that at the same salt concentration, proteins with high ks are more strongly retained when protein recoveries during isocratic elution are comparable. Therefore, ks is a promising parameter for use in predicting retention in HIC. 3.2.3. The effects of protein molecular mass Molecular mass is another protein property that may affect retention in HIC. With the same elution gradient, trypsin has been shown to be more strongly retained than ribonuclease A and cytochrome C, ovalbumin than myoglobin, and ␣chymotrypsinogen A than lysozyme [48]. Given the adiabatic compressibilities (Table 2), these results suggest that for proteins with similar ks values, the larger ones are more strongly retained. To investigate if the same trend holds with other proteins and adsorbents, the retention of ovalbumin and BSA, with similar ks values, in seven phenyl adsorbents with different base matrices were compared (Figs. 4 and 5). On all the agarose-based media, BSA was more strongly retained than ovalbumin. Similar results were also reported for agarose-based media with other HIC ligands [54]. The retention of BSA and ovalbumin was similar in polymethacrylate-based TSK-gel 5PW and Toyopearl 650S, and ovalbumin was slightly more strongly retained in polymethacrylate-based Fractogel EMD and PSDVB-based Source 15PHE. Therefore, correlations between molecular mass and retention observed previously [48] on TSK-gel 5PW columns were not reproduced in this study for ovalbumin and BSA on the same adsorbent. This again can be explained in part by protein recovery. Kato et al. obtained similar (90–100%) recoveries for all of the proteins used in their study [48]. However, in this study, the recovery of BSA was lower than that of ovalbumin, especially at high salt concentration (Fig. 8). The low BSA recovery could skew the retention factors, as discussed earlier. BSA had higher retention than ovalbumin on all Sepharose media even though the BSA recoveries observed in these media were comparable to those on the polymethacrylate media. The reason for this observation is not clear. These results indicate

that for proteins with similar ks values, larger proteins are generally more strongly retained in HIC media, but this correlation depends on the adsorbent base matrix. The relation between molecular mass and retention could be contributed by the larger non-polar and total surface areas available for adsorption on the

Fig. 8. Protein recovery during isocratic elution. (♦): Sepharose HP; (): Sepharose FF LS; (): Sepharose FF HS; (): TSK-gel 5PW; (): Toyopearl 650S; (): Fractogel EMD; (+): Source 15PHE; ( ) POROS 20HP2.

202

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

larger proteins (Table 2), but the absence of a correlation for the smaller proteins, discussed in the previous section, indicates that such a relation must be treated with caution unless the compressibility is considered as well. It is possible that the distribution of non-polar surface area is a hidden underlying factor [40].

study were fitted to Eq. (10). The estimated parameters are summarized in Table 6 and are consistent with the model formulation in terms of which the interaction between protein and adsorbent results in the release of water molecules and salt ions. The number of water molecules estimated as being released is 1–2 orders of magnitude higher than that of ions. Although the same trends were observed by Perkins et al., the numbers of water molecules and ions released as calculated in this work are consistently 2–3 times higher than those of Perkins et al. for the lysozymeSepharose and ovalbumin-Sepharose pairs. Comparison of the retention factors from the two studies shows the k values of both proteins obtained in this work to be consistently lower. Comparison of the shapes of the retention (ln k versus AS concentration) curves from the two studies shows the curvatures of the curves to be comparable for lysozyme, while the retention of ovalbumin in this work is more sensitive to salt. These differences could be due to variations in the adsorbents. The salt concentration ranges used for the studies (0.2–1.1 M for Perkins et al. compared with 0.5–1.6 m in this study) could be another explanation for the differences observed in the estimated number of ions and water molecules released. The differences in the salt concentration ranges can change the shape and hence the fitting parameters of the ln k versus salt concentration plots. The results listed in Table 6 also show that, within the model framework, the number of water molecules released by the same protein upon interaction with different adsorbents remains statistically the same. All five proteins used in this study released similar numbers of water molecules, although their retention factors were quite different at the same salt concentrations. This suggests that the number of water molecules released as calculated from Eq. (10) is not sensitive enough to allow the prediction of adsorption selectivity in HIC.

3.2.4. Interpretation of retention data in terms of the preferential interaction model On the basis of the preferential interaction theory, protein retention in HIC can be correlated to salt concentration by [17] ln(k ) = C + and g=



∂ ln m ∂ ln a

n ν1 (ν+ + ν− ) ln(m) − m g m1 g

(10)

 (11) T,P

where C is a fitting parameter, a is the mean ionic activity, n is the number of ions associated with the electrolytes used in the elution buffer, m1 is the molal water concentration, and (ν+ + ν− ) and ν1 are the changes in the number of ions and water molecules per adsorbed protein molecule, respectively. In the system with AS in the mobile phase, m1 = 55.5, n = 3 and g = 1.6 [17]. The retention model based on the solvophobic theory has the same functional dependence on m as Eq. (10) although the fitting parameters are interpreted differently [10]. Perkins et al. fitted their lysozyme and ovalbumin retention data on agarose-based adsorbents to Eq. (10) and suggested that protein retention in HIC is dominated by the release of water molecules upon adsorption [17]. To extend the analysis to other proteins and adsorbents, the retention data generated from this

Table 6 Changes in numbers of ions and water molecules upon protein-adsorbent interaction, calculated from Eq. (10) RNA

LYS

−ν1 Sepharose HP Sepharose FF LS Sepharose FF HS TSK-gel 5PW Toyopearl 650S Fractogel EMD Source 15PHE POROS 20 HP2

264 260 294 311 276 303 195 210

± ± ± ± ± ± ± ±

−(ν+ + ν− ) 24 49 25 15 17 15 22 54

12 12 12 16 12 15 7 9

± ± ± ± ± ± ± ±

2 4 2 1 1 1 2 4

ALC (without Ca2+ )

Sepharose HP Sepharose FF LS Sepharose FF HS TSK-gel 5PW Toyopearl 650S Fractogel EMD Source 15PHE POROS 20 HP2

ALC

−ν1 206 202 123 251 226 218 195 192

± ± ± ± ± ± ± ±

−(ν+ + ν− ) 11 33 440 46 21 45 10 18

6 5 1 7 5 6 5 5

± ± ± ± ± ± ± ±

1 2 15 2 1 2 1 1

OVA

−ν1

−(ν+ + ν− )

−ν1

376 ± 43 – – 387 ± 205 – 250 ± 26 – –

10 ± 2 – – 11 ± 9 – 6±1 – –

316 338 630 418 462 293 335 307

Values reflect best fit estimates with 80% confidence intervals.

± ± ± ± ± ± ± ±

−ν1 247 ± 189 ± – 353 ± 276 ± 221 ± – –

−(ν+ + ν− ) 56 129 35 54 58

6± 3± – 9± 6± 3± – –

3 6 2 4 3

BSA −(ν+ + ν− ) 58 41 192 56 59 57 75 36

7 8 15 13 14 5 9 9

± ± ± ± ± ± ± ±

3 2 7 3 3 3 4 2

−␯1 616 ± 533 ± 417 ± 457 ± 481 ± 418 ± – –

−(ν+ + ν− ) 142 189 105 94 96 255

18 ± 14 ± 3± 15 ± 18 ± 11 ± – –

6 8 1 4 5 12

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

3.3. Protein recovery Since proteins can change conformation when interacting with the adsorbent, sometimes resulting in irreversible binding, it is important to understand how protein and media properties affect recovery. Ribonuclease A and lysozyme showed essentially 100% recovery in all media at all salt concentrations used in this study. The recoveries of ␣-lactalbumin, ovalbumin, and BSA are depicted in Fig. 8. Protein recoveries from the agarose media varied only slightly with salt concentration. Similar trends were also observed for ␣-lactalbumin (with Ca2+ ) and ovalbumin on Toyopearl 650S and TSK-gel 5PW. For BSA, the recoveries from all polymethacrylate and PS-DVB media were more sensitive to salt concentration. On the PS-DVB media, the recoveries of ␣-lactalbumin, even with Ca2+ , and BSA were lower than those on the other adsorbents. The recoveries of BSA on Source 15PHE and POROS 20 HP2 were about 75 and 55% respectively, even when the mobile phase contained only 10 mM sodium phosphate (Fig. 8C). PS-DVB is a highly hydrophobic base matrix, especially when it is not hydrophilized, so low ␣-lactalbumin and BSA recoveries suggest that the combination of high salt concentration and a hydrophobic base matrix can promote strong protein-adsorbent interactions. In fact, Krisdhasima et al. showed that both ␣-lactalbumin and BSA yielded irreversibly adsorbed forms on hydrophobic surfaces [78]. The experimental data also show that ␣-lactalbumin and BSA adsorbed more strongly on POROS 20 HP2 than on Source 15PHE (Fig. 8A and C). On Source 15PHE, ␣-lactalbumin and BSA that were not eluted during isocratic elution could be recovered using mobile phases with low salt concentrations. However, proteins bound to POROS 20 HP2 could not be eluted even with 10 mM phosphate buffer. Because of the strong proteinadsorbent interaction, the POROS 20 HP2 column could not be regenerated properly. These results suggest that proteins can interact with the base matrix in addition to the phenyl ligand, and highly hydrophobic base matrices can adversely affect recovery. In addition to the base matrix, it appears that the tentacles on Fractogel EMD can also affect protein recovery because the recoveries of ovalbumin and BSA in Fractogel EMD were considerably lower than on the other polymethacrylate-based media. The recoveries of ␣-lactalbumin, ovalbumin, and BSA on Sepharose FF HS were considerably lower than those on Sepharose FF LS, suggesting that ligand density has an impact on protein recovery as well. This is consistent with the results from experiments performed with polymethacrylate media [48]. Apart from the stationary-phase properties (base matrix and ligand density), the physical properties of the proteins that affect retention (see Sections 3.2.1–3.2.3) also have an impact on recovery. In general, small proteins (i.e., ribonuclease A, lysozyme, and ␣-lactalbumin in the presence of Ca2+ ) had high recoveries. Protein conformation also plays an important role, as reflected by the lower ␣-lactalbumin recovery in the absence of Ca2+ (Fig. 7). The recovery of BSA was considerably lower than those of the other proteins used in this study, with flexibility

203

Table 7 Effect of ligand density on protein retention at 0.8 m AS in Sepharose media Sepharose FF LS ln Keq (␮m) RNA −4.90 LYS −3.60 ALA −3.10 OVA −4.12 BSA −2.53

Sepharose FF HS

ln Keq HS − ln Keq LS

−4.31 −2.01 −1.36 −1.44 4.08

0.60 1.59 1.74 2.68 6.61

Keq HS / Keq LS 1.8 4.9 5.7 14.6 745

ln Keq of ribonuclease A on both Sepharose FF LS and Sepharose FF HS and ln Keq of BSA on Sepharose FF HS were calculated from polynomial fits of the experimental data.

again a likely reason. However, the high molecular mass may also be implicated, consistent with the findings of Goheen and Gibbins that protein recovery was linearly proportional to the log of protein molecular mass [79]. The recovery results indicate that proteins can bind to HIC adsorbents irreversibly under some conditions. Since retention models developed based on the preferential interaction theory, solvophobic theory, and stoichiometric displacement assume reversible adsorption, these models may not be valid for HIC systems with low protein recovery. 3.4. Effect of media properties on protein retention The k and Keq values of the model proteins in different HIC media at selected salt concentrations are presented in Table 5 to facilitate the comparison of protein binding affinities of different media. The k and Keq values on Sepharose FF HS are not included in this table because protein retention at the salt concentrations listed in Table 5 was very strong. However, Sepharose FF HS had the highest k and Keq values among all media for all the proteins used in this study (Figs. 4 and 5). 3.4.1. The effect of ligand density The effect of ligand density on protein retention is reflected by the adsorption equilibrium constants of Sepharose FF LS and Sepharose FF HS shown in Figs. 4 and 5 and Table 7. Since these two stationary phases have almost identical mean pore radii, porosities, base matrices, phase ratios and PSD, the differences observed should be due solely to the difference in ligand density. When the ligand density was doubled, the retention of all proteins increased, with the order of protein binding affinity remaining unchanged. The increase in ligand density affected retention of the proteins to different extents. With similar molecular mass, the impact of increased ligand density was more pronounced for proteins with higher ks . On the other hand, with similar ks , the retention of larger proteins was more sensitive to changes in ligand density. Similar results were also obtained in experiments performed with gradient elution [48]. Since the effect of ligand density on protein retention appears to depend on protein flexibility and molecular mass, protein selectivity in HIC may be improved by manipulating the ligand density of the adsorbent.

204

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205

3.4.2. The effect of adsorbent physical structure The physical structure of the adsorbent can also have an effect on retention. Sepharose HP has a similar ligand density to Sepharose FF LS, but a different PSD. These two media showed similar retention values for the same proteins, with the exception of BSA (Table 5). This suggests that, in this case, the difference in PSD did not have a significant impact on protein retention, although Sepharose HP contains more pore volume with dimensions close to those of the protein solutes. This is inconsistent with what was found in cation-exchange media, where adsorbents with PSDs containing a significant amount of pore space with dimensions of the same order of magnitude as the protein solute showed increased protein retention [21]. A possible explanation for this discrepancy is that, for the HIC media discussed above, the majority of the smaller pores had radii between 10–20 nm, an order of magnitude larger than the radii of the proteins used in this study. These pores may be too large to show the PSD effect. Another possible explanation is that the effect seen in cation exchangers is due to non-contact electrostatic interactions, which would be screened at the high salt concentrations used in HIC. Toyopearl 650S and Fractogel EMD were manufactured on the same base matrix, but protein retention and binding affinity on Fractogel EMD are stronger than on Toyopearl 650S. A comparison of the mean pore radii of Toyopearl 650S and Fractogel EMD suggests that the tentacles in Fractogel EMD may be longer than 10 nm, providing extra surface for protein–media interaction. This is also reflected in the PSD of Fractogel EMD, which contains a significant amount of pore space with dimensions similar to the radii of the proteins used in this study. As mentioned above, this type of PSD increased protein retention in cation-exchange adsorbents. Therefore, either the tentacle flexibility or difference in apparent PSD can explain the higher k and Keq in Fractogel EMD, although the potential difference in ligand density cannot be ignored. 4. Conclusions The results presented here show that protein retention in HIC is affected by, in addition to salt, the physical properties of the protein such as structural flexibility, molecular mass and conformation stability. These findings suggest that protein flexibility and conformational stability can be used as alternative adjustable parameters to optimize the efficiency of HIC. It may also be possible to alter the retention of flexible proteins by changing the column operating pressure, which can easily be achieved by regulating the column pressure at the column outlet. For metalbinding proteins, the addition or removal of the appropriate metal ions could help to improve the separation performance of HIC. In addition to protein properties, the properties of HIC adsorbents also play an essential role. In this study, we identified ligand density, base matrix as well as polymer tentacles as the important factors that can affect HIC performance. The base matrices of the HIC adsorbents can interact with the protein and affect selectivity and protein recovery. Hydrophilic base matrices are preferred for HIC adsorbents as they can minimize non-specific protein adsorption. Changes in ligand density

affect protein retention to different extents depending on protein properties and the salt concentration, but high ligand densities can result in yield loss due to irreversible binding, which may necessitate difficult and harsh adsorbent regeneration cycles. It is also important to check the recoveries of retention experiments as results with low recovery can provide misleading trends. The HIC retention models available in the literature were developed based on reversible protein-ligand interactions. The results of this study suggest that this assumption may not be valid for some proteins. Although these models can fit the experimental results adequately, they may not capture the actual physics of the protein-adsorbent interaction. Acknowledgement We are grateful for support from the National Science Foundation (grant numbers CTS-9977120 and CTS-0350631) and from Merck Research Laboratories for a fellowship to B.C.S.T. References [1] O. Manzke, H. Tesch, V. Diehl, H. Bohlen, J. Immunol. Methods 208 (1997) 65. [2] K.C. O’Connor, S. Ghatak, B.D. Stollar, Anal. Biochem. 278 (2000) 239. [3] J.O. Kvassman, J.D. Shore, Fibrinolysis 9 (1995) 215. [4] M.M. Diogo, J.A. Queiroz, G.A. Monteiro, S.A.M. Martins, G.N.M. Ferreira, D.M.F. Prazeres, Biotech. Bioeng. 68 (2000) 576. [5] M.M. Diogo, J.A. Queiroz, G.A. Monteiro, D.M.F. Prazeres, Anal. Biochem. 275 (1999) 122. [6] C.C. Shepard, A. Tiselius, Disc. Farad. Soc. 7 (1949) 275. [7] J.L. Fausnaugh, L.A. Kennedy, F.E. Regnier, J. Chromatogr. 317 (1984) 141. [8] W. Melander, C. Horv´ath, Arch. Biochem. Biophys. 183 (1977) 200. [9] W.R. Melander, D. Corradini, C. Horv´ath, J. Chromotogr. 317 (1984) 67. [10] W.R. Melander, Z. El Rassi, C. Horv´ath, J. Chromatogr. 469 (1989) 3. [11] A. Katti, Y.F. Maa, C. Horv´ath, Chromatographia 24 (1987) 646. [12] L. Szepesy, C. Horv´ath, J. Chromatogr. 26 (1988) 13. [13] T. Arakawa, S.N. Timasheff, Biochemistry 21 (1982) 6545. [14] S.N. Timasheff, T. Arakawa, J. Cryst. Growth 90 (1988) 39. [15] S.N. Timasheff, Ann. Rev. Biophys. Biomol. Struct. 22 (1993) 67. [16] T. Arakawa, Arch. Biochem. Biophys. 248 (1986) 101. [17] T.W. Perkins, D.S. Mak, T.W. Root, E.N. Lightfoot, J. Chromatogr. A 766 (1997) 1. [18] S.L. Wu, A. Figueroa, B.L. Karger, J. Chromatogr. 371 (1986) 3. [19] X. Geng, L. Guo, J. Chang, J. Chromatogr. 507 (1990) 1. [20] A. Staby, J. Mollerup, J. Chromatogr. A 734 (1996) 205. [21] P. DePhillips, A.M. Lenhoff, J. Chromatogr. A 993 (2001) 57. [22] T.W. Perkins, T.W. Root, E.N. Lightfoot, Anal. Chem. 69 (1997) 3293. [23] L.Z. Vilenchik, J. Asrar, R.C. Ayotte, L. Ternorutsky, C.J. Hardiman, J. Chromatogr. 648 (1993) 9. [24] P. DePhillips, A.M. Lenhoff, J. Chromatogr. A 883 (2000) 39. [25] M. Ousalem, X.X. Zhu, J. Hradil, J. Chromatogr. A 903 (2000) 13. [26] H. Guan, G. Guiochon, J. Chromatogr. A 731 (1996) 27. [27] M.E. Lienqueo, A. Mahn, J.A. Asenjo, J. Chromatogr. A 978 (2002) 71. [28] D. Corradini, D. Capitani, L. Cellai, J. Liquid Chromatogr. 17 (1994) 4335. [29] T. Zoungrana, G.H. Findenegg, W. Norde, J. Coll. Interf. Sci. 190 (1997) 437. [30] S. Oscarsson, J. Chromatogr. B 666 (1995) 21. [31] T.T. Jones, E.J. Fernandez, J. Colloid Interface Sci. 259 (2003) 27. [32] K. Heremans, L. Smeller, Biochim. Biophys. Acta 1386 (1998) 353. [33] V.M. Dadarlat, C.B. Post, PNAS 100 (2003) 14778. [34] K.P. Murphy, P.L. Privalov, S.J. Gill, Science 247 (1990) 559. [35] K. Gekko, Y. Hasegawa, Biochemistry 25 (1986) 6563. [36] A. Yang, K.A. Sharp, B. Honig, J. Mol. Biol. 227 (1992) 889.

B.C.S. To, A.M. Lenhoff / J. Chromatogr. A 1141 (2007) 191–205 [37] A. Kondo, S. Oku, K. Higashitani, J. Coll. Interf. Sci. 143 (1991) 214. [38] A. Kondo, F. Murakami, K. Higashitani, Biotech. Bioeng. 40 (1992) 889. [39] T. Przybycien, S. Sane, E. Shapiro, S. Cramer, The structural impacts of bioseparation operations on target proteins, Presented at Recovery of Biological Products IX, Whistler, BC, 1999. [40] A. Mahn, M.E. Lienqueo, J.A. Asenjo, J. Chromatogr. A 1043 (2004) 47. [41] G. Rippel, L. Szepesy, J. Chromatogr. A 664 (1994) 27. [42] M.N. Schmuck, M.P. Nowland, K.M. Gooding, J. Chromatogr. 371 (1986) 55. [43] K. Benedek, J. Chromatogr. 458 (1988) 93. [44] J.Y. Gao, P.L. Dubin, Biopolymers 49 (1999) 185. [45] P.P. Berna, N. Berna, J. Porath, S. Oscarsson, J. Chromatogr. A 800 (1998) 151. [46] F.Y. Lin, W.H. Chen, R.C. Ruaan, H.M. Huang, J. Chromatogr. A 872 (2000) 37. [47] F.Y. Lin, W.Y. Chen, M.T.W. Hearn, Anal. Chem. 73 (2001) 3875. [48] Y. Kato, K. Nakamura, T. Kitamura, H. Moriyama, M. Hasegawa, H. Sasaki, J. Chromatogr. A 971 (2002) 143. [49] X. Geng, L. Guo, J. Chang, J. Chromatogr. 507 (1990) 1. [50] M.L. Heinitz, L. Kennedy, W. Kopaciewicz, F.E. Regnier, J. Chromatogr. 443 (1988) 173. [51] J.L. Fausnaugh, F.E. Regnier, J. Chromatogr. 359 (1986) 131. [52] S.C. Goheen, S.C. Engelhorn, J. Chromatogr. 317 (1984) 55. [53] W.Y. Chen, H.M. Huang, C.C. Lin, F.Y. Lin, Y.C. Chan, Langmuir 19 (2003) 9395. [54] C. Machold, K. Deinhofer, R. Hahn, A. Jungbauer, J. Chromatogr. A 972 (2002) 3. [55] R. Hahn, K. Deinhofer, C. Machold, A. Jungbauer, J. Chromatogr. B 790 (2003) 99. [56] F. Xia, D. Nagrath, S. Garde, S.M. Cramer, Biotech. Bioeng. 87 (2004) 354. [57] D.P. Kharakoz, A.P. Sarvazyan, Biopolymers 33 (1993) 11. [58] T.V. Chalikian, M. Totrov, R. Abagyan, K.J. Breslauer, J. Mol. Biol. 260 (1996) 588.

205

[59] L. Hagel, in: P.L. Dubin (Ed.), Aqueous Size-exclusion Chromatography, Elsevier, Amsterdam, 1988, p. 119. [60] C. Tanford, Physical Chemistry of Macromolecules, Wiley, New York, 1967, p. 391. [61] M. Potschka, Anal. Biochem. 162 (1987) 47. [62] E.M. Renkin, J. Gen. Physiol. 38 (1954) 225. [63] M. Briˇssˇov´a, M. Petro, I. Lac´ık, A.C. Powers, T. Wang, Anal. Biochem. 242 (1996) 104. [64] W. M¨uller, J. Chromatogr. 510 (1990) 133. [65] R. Janzen, K.K. Unger, W. Muller, M.T.W. Hearn, J. Chromatogr. 522 (1990) 77. [66] D.C. Nash, H.A. Chase, J. Chromatogr. A 807 (1998) 185. [67] D. Whitney, M. McCoy, N. Gordon, N. Afeyan, J. Chromatogr. A 807 (1998) 165. [68] F. Xia, D. Nagrath, S.M. Cramer, J. Chromatogr. A 989 (2003) 47. [69] T.T. Jones, D. Nagrath, Y. Xiao, J.P. O’Connell, S.M. Cramer, E.J. Fernandez, Modulating the impact of conformational changes on hydrophobic interaction chromatography (HIC), Presented at AIChE Meeting, San Francisco, CA, 2003. [70] E.A. Permyakov, L.A. Morozova, E.A. Burstein, Biophys. Chem. 21 (1985) 21. [71] M.J. Kronman, S.K. Sinha, K. Brew, J. Biol. Chem. 256 (1981) 8582. [72] E.D. Chrysina, K. Brew, K.R. Acharya, J. Biol. Chem. 275 (2000) 37021. [73] W. Norde, Adv. Colloid Interface Sci. 25 (1986) 267. [74] H. Yang, M.R. Etzel, Ind. Eng. Chem. Res. 42 (2003) 890. [75] S.U. Sane, S.M. Cramer, T.M. Przybycien, J. Chromatogr. A 849 (1999) 149. [76] J.L.M. McNay, E.J. Fernandez, Biotech. Bioeng. 76 (2001) 224. [77] J.L.M. McNay, J.P. O’Connell, E.J. Fernandez, Biotech. Bioeng. 76 (2001) 233. [78] V. Krisdhasima, P. Vinaraphong, J. McGuire, J. Coll. Interf. Sci. 161 (1993) 325. [79] S.C. Goheen, B.M. Gibbins, J. Chromatogr. A 890 (2000) 73.