Effect of mass overloading on binding and elution of unstable proteins in hydrophobic interaction chromatography

Accepted Manuscript Title: Effect of mass overloading on binding and elution of unstable proteins in Hydrophobic Interaction Chromatography ˙ Authors: Renata Muca, Wojciech Marek, Marek Zurawski, Wojciech Pi˛atkowski, Dorota Antos PII: DOI: Reference:

S0021-9673(17)30332-1 http://dx.doi.org/doi:10.1016/j.chroma.2017.02.073 CHROMA 358343

To appear in:

Journal of Chromatography A

Received date: Revised date: Accepted date:

7-12-2016 27-2-2017 28-2-2017

˙ Please cite this article as: Renata Muca, Wojciech Marek, Marek Zurawski, Wojciech Pi˛atkowski, Dorota Antos, Effect of mass overloading on binding and elution of unstable proteins in Hydrophobic Interaction Chromatography, Journal of Chromatography A http://dx.doi.org/10.1016/j.chroma.2017.02.073 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Effect of mass overloading on binding and elution of unstable proteins in Hydrophobic Interaction Chromatography Renata Muca1, Wojciech Marek1, Marek Żurawski2, Wojciech Piątkowski1, Dorota Antos1, Department of Chemical and Process Engineering, Rzeszów University of Technology, Powstańców Warszawy Ave. 6, 35-959 Rzeszów, Poland 2 NanoTemper Technologies Sp.z.o.o. Bobrzyńskiego str. 14, 30-348, Kraków, Poland 1

1. 1

Corresponding author: [email protected], tel.: +48 178651853; fax: +48 178543655

Highlights 

Adsorption behavior of marginally stable proteins on a HIC medium was studied



Influence of mass overloading on the retention pattern of proteins was determined



A mechanistic model was developed and exploited to predict the column dynamics



nanoDSF was used to determine protein stability versus mass overloading

2. Abstract Adsorption

behavior

of

unstable

proteins,

i.e.,

bovine

serum

albumin

and

-lactalbumin, has been studied on a hydrophobic interaction chromatography medium under mass overloading conditions at different kosmotropic salt concentrations in the mobile phase. A mechanistic model has been formulated and used to describe kinetics and thermodynamics of protein interactions with the adsorbent surface. The model assumed two-site binding adsorption and reversible protein unfolding, which allowed predicting the inhibition of protein unfolding at high column loadings. A simplified procedure for the determination of model parameters has been developed, which was based on the inverse method. The model was successfully used to reproduce the pattern of chromatographic elution as well as the course of breakthrough curves. The model formulation was supported by Nano Differential Scanning

1

Fluorimetry measurements, which were exploited to determine the protein stability in the liquid and adsorbed phases at different column loadings and salt concentrations.

Key words: HIC, protein unfolding, mathematical modeling, nanoDSF

3. Introduction Hydrophobic interaction chromatography (HIC) is commonly used for protein purification as an

orthogonal

separation

method

that

complements

affinity and

ion

exchange

chromatography. It exploits interactions between hydrophobic sites of chromatographic matrix and proteins, which are promoted in the presence of structure-enhancing salts (kosmotrope) such as those used in precipitation (e.g. ammonium sulphate) [1,2]. Therefore, HIC can be efficiently used as a subsequent purification stage following precipitation, aqueous two-phase extraction or ion exchange chromatography, where proteins are processed at high salt concentration [3-6]. Interaction of proteins with hydrophobic ligands is reported to be highly selective [7-9], therefore, HIC is often effective in removal of difficult to separate impurities, such as aggregated and misfolded forms of the target protein, which cannot be performed by other separation techniques [1,4, 10-12]. However, structure stability problems, which arise from conformational changes of proteins upon adsorption, limit application of HIC media, especially at very high feed concentrations [13]. Evidence of conformational changes of proteins in HIC has been reported in a number of publications (e.g. [14-17]). To detect changes in the structure of adsorbed proteins various spectroscopic methods have been used, including: circular dichroism [18], fluorescence [19,20], infrared spectroscopy [21], and isothermal titration calorimetry [22,23]. Moreover, hydrogen exchange has proven to be a useful tool for the study of protein conformation both in liquid solution and at solid interfaces (e.g. [24-29]). In chromatographic processes, the phenomenon of protein unfolding usually manifests itself by two-peak elution of pure protein samples. The earlier eluting peak is assigned to native or predominately native protein, whereas the more retained - to partially unfolded protein. The unfolding phenomenon can be used to advantage in amending retention properties of proteins during HIC elution, when pure target protein is recovered in an active form. However, destabilizing conditions of protein separation in HIC are often considered undesirable, as they can adversely affect the process performance by inducing yield losses 2

[13]. This causes that strongly hydrophobic HIC media are often reluctantly used to process unstable proteins, despite high efficiency of the separation could potentially be achieved [29]. The prediction and design of HIC is difficult because the process variables that control retention and selectivity of the separation, such as salt type and concentration, pH, temperature, ligand type and its coverage density, also influence conformational changes. To support the designing procedure, mechanistic models of protein unfolding have been developed. To and Lenhoff [30] suggested a three-state reversible mechanism of conformation change, which assumed that the protein in liquid solution is present only in native state, while both native and unfolding forms can occur on the adsorbent surface. In that study, the underlying kinetic equations were implemented into a heterogeneous dynamic model, i.e. general linear rate model, and used to simulate chromatographic band profiles under linear isotherm conditions. A similar mechanism of protein unfolding was adopted in a few other works [6,31-33] to optimize loading conditions and separation efficiency. The chromatographic bands were simulated there by use of a simplified, pseudo-homogeneous model of column dynamics. In several studies (e.g. [34-36]), the phenomenon of protein unfolding on HIC media was analyzed using a four-state model that assumed reversible conformation change and adsorption of proteins under linear isotherm conditions. However, that model has not been adopted for simulating dynamics of chromatography. A three-state irreversible unfolding model has been proposed by Jungbauer and co-workers [13,21]. The model accounted for the effect of column overloading on the course of protein unfolding. It was used to determine unfolding rate constants and simulate the band profiles in HIC, however, the accuracy of the model predictions was often unsatisfactory. In this study, we modified the latter model by assuming a reversible unfolding mechanism under nonlinear isotherm conditions. Moreover, we considered the possibility of protein aggregation in the adsorbed phase. This approach allowed reproducing the band profiles recorded under different operating conditions. The band profiles were simulated using a pseudo-homogeneous model, similar to that used in our previous studies (e.g. [31-33]). We also proposed a procedure for fast determination of the model parameters. The model formulation was supported by Nano Differential Scanning Fluorimetry measurements (nanoDSF). The nanoDSF technology uses temperature gradient to monitor protein conformational changes over time. Thermal unfolding of proteins usually occurs over a narrow temperature range. The unfolding transition point is determined as a midpoint of the transition from folded to unfolded form. It is referred to as “melting temperature” and can serve as a measure for protein stability. In order to determine the unfolding transition points, 3

the shift of intrinsic tryptophan fluorescence at the emission wavelengths of 330 nm and 350 nm are recorded [37]. We exploited that technology to analyze protein stability in both adsorbed and liquid phases versus the protein and salt concentrations, and to verify the model assumptions. For this purpose the unfolding transition data for the protein in liquid solution and resin phase were acquired and examined. A shift in characteristic transition point was attributed to the structure state of the protein in its environment. The experimental and mathematical methodology developed in this work can be used for quick evaluating the unfolding kinetics and efficient design of HIC separations. This aims at exploiting the potential of that technique, while avoiding undesirable effects arising from yield losses, which may contribute to improvement in applicability of HIC in downstream protein processing.

4. 2. Theory 2.1. Kinetic mechanism of protein adsorption and unfolding The protein adsorption and unfolding was assumed to occur according to a simplified mechanism of two-site adsorption and reversible unfolding , which is described as follows. The protein, P, is reversibly bound to a single adsorption site, S*, forming a surface complex Pn S*, according to the second order reaction mechanism: k

a *   * P+S    Pn S

(1)

kd

where: Pn is the protein in the native form; ka, kd denote the lumped kinetic coefficients, which can accommodate the contribution of mass transport as well as adsorption-desorption rates to the overall kinetics of protein adsorption. Both these kinetic effects have been proved to act additively to hinder the progress of adsorption process, while the corresponding kinetic rate equations to be compatible to each other over a wide concentration range [39]. The adsorbed protein in the native form can further interact with another active site, which consequences in its unfolding according to the second order reaction: ku n f

* *    *  Pn S  S    P u S 2

(2)

kf

where: kunf, kf are kinetic coefficients of protein unfolding and folding; Pu, denotes the protein at the solid interface in the unfolded form. At a high protein concentration, a third surface reaction is suggested to be active, when the unfolded protein provides an additional layer of interaction sites in the adsorbed phase, which causes protein aggregation at the solid interface. A simplified mechanism of aggregation is described as follows: 4

k a g ,a

   P P + Pu    ag

(3)

k a g ,d

where: kag,a, kag,d are kinetic coefficients of formation and decomposition of aggregates; Pag denotes the protein at the solid interface in the aggregated form. The protein in the unfolded state, Pu, is strongly bound to the stationary phase and cannot desorb into the liquid solution, which is a cause of mass losses in chromatographic elution. The mechanism given by Eqs (1-3) is represented by the following kinetic equations: -

for the adsorbed phase concentration of the protein in the state Pn:

qn

 kd K a C  t

p

q

qu   (qn  qu )   qn    t



(4)where: kd stands for the lumped kinetic

coefficient, as mentioned above. -

for the adsorbed phase concentration of the protein in the state Pu: qu

 k

t

f

 K q  q    (q  q )   q  n u u   u n

(5)

To account for the protein aggregation in the adsorbed phase, the kinetic model given by Eqs (4) and (5) has to be supplemented by an additional equation, as follows:  q ag t

 k ag ,d

K

ag

C

p

q u  q ag



(6)

For the total protein concentration in the adsorbed phase, q, it holds: q t



qn t



qu

(7)

t

or, if aggregation is included: q t



qn t



qu t



 q ag

(8)

t

where: qn, qu, qag is the protein concentration in adsorbed phase in the native, unfolded and aggregated forms, respectively; q∞ is the binding capacity corresponding to the first adsorption layer; δ is the steric hindrance factor;

Ka 

ka kd

,

Ku 

k unf k

f

,

K ag 

k ag ,a

are the

k ag ,d

equilibrium constants. To simulate the dynamic band profiles, the modified kinetic-dispersive model was employed [6, 31-33]. The model comprises the differential mass balance equation of the protein in the mobile phase: t

C t

u

C  x



q t

 C 2

 e DL

 x

2

(9)

5

where: C [mg mL-1] is the concentration of the protein (BSA or -La) in the mobile phase; q [mg

-1

m L col ]

is the total protein concentration in the adsorbed phase related to the column

volume; u [m s-1] is the superficial velocity; t [s], x [m] are the time and axial coordinates; DL [m2s-1] is the axial dispersion coefficient;  t ,  e [-] are the total and external bed porosity, respectively. To account for the conformational changes of the protein, the underlying kinetic equations (Eqs (4,5,7) or (4-6,8)) are implemented into Eq. (9). To solve the mass balance equations, adequate initial and boundary conditions have to be specified [6, 31-33].

5. 3. Materials and Methods 3.1. Instruments 3.1.1. Chromatographic system The Äkta purifier with a UV detector and a data station (GE Healthcare Life Sciences, Uppsala, Sweden) was used. The injector was a Rheodyne sampling valve with sample loops of different volumes, i.e., 0.1, 0.5, 2.0, 5.0 mL, and a sample Superloop of 50 mL (GE Healthcare Life Sciences). 3.1.2. Differential scanning fluorimetry instrument For NanoDSF measurements, a Prometheus NT.48 (NanoTemper Technologies GmbH, Munich, Germany) was used. It offers unparalleled rapid and accurate analysis of protein folding and thermal stability. The instrument precisely measures changes in tryptophan fluorescence in a unique low sample volume capillary system using dye-free approach.

3.2. Materials As a stationary phase Butyl Sepharose 4FF (GE Healthcare Life Sciences) was used. The resin was packed into a glass column, i.d. 0.66 cm, the bed height in the column was 4 cm, bed volume was 1.37 mL. The proteins were purchased as follows. Bovine serum albumin (BSA) lyophilized powder, (A2153, Sigma, Poland) with the purity greater than 96%, α-Lactalbumin (α-La) from bovine milk type III, calcium depleted, with the purity greater than 85%, lyophilized powder (L6010, Sigma, Poland).

6

3.3. Procedures 3.3.1. Chromatographic elution The samples of proteins were dissolved in the mobile phase and eluted in isocratic mode. The mobile phase was a mixture of phosphate buffer (50 mM sodium phosphate buffer, pH 7 free of AS salt) and the AS salt buffer (AS salt at the concentration of 1.7 M dissolved in phosphate buffer) with different volume content. After each chromatographic run, the column was washed with phosphate buffer to desorb the retained proteins. The protein concentration in samples was changed from 0.5 to 5 [mg mL-1], the AS salt concentration range in the mobile phase was: 0 - 1.02 M AS for BSA, and 0 - 0.68 M AS for α-La. The mobile phase flowrate was: 0.3, 0.5, 1.0 [mL min-1]. The protein profiles were recorded at 280 nm wavelength. Each sample of protein solutions was injected through the injection valve into the Äkta system twice: into the column, and directly into the detector bypassing the column. The detector calibration factor was calculated by the integration of the UV profile of the protein bypassing the column. The relationship between the protein concentration and UV signal in the concentration range investigated was linear. Next, the calibration factor was used to convert the UV signals into the concentration profiles and quantify the incomplete elution of the protein due to its unfolding upon adsorption. 3.3.2. Determination of protein adsorption isotherm by the static method To determine the adsorption isotherm, the protein standard solutions were prepared. The AS salt concentration in the solutions was selected in such a way that the effect of incomplete elution was pronounced, i.e. 0.85 or 1.02 M AS for BSA, and 0.595 or 0.68 M AS for α-La. The protein concentration in samples was changed within the range 0.1 - 5.0 [mg mL-1]. The resin that was used in chromatographic elution experiments was unpacked from the column and washed; at first using deionized water, next by proper AS solution. The resin was separated out of the liquid phase using vacuum filtration and weighted to determine the value of the resin bulk density, which was used to correlate the data obtained by the dynamic and static measurements. Then, a few samples of 0.1 g resin were placed into Eppendorf tubes. A 1.5 mL protein solution was added into each of tubes and stirred for 120 min. at 500 rpm using a magnetic stirrer. Preliminary measurements of the adsorption kinetics confirmed that the duration of experiment was sufficient to establish adsorption equilibrium, which was confirmed by the concentration analysis of supernatants performed at different time intervals. The samples of supernatants for the concentration analysis were acquired using syringe filters PES 0.2 μm (Alchem, Poland). Next, the sample and the corresponding standard solution (i.e. 7

the initial solution) were subsequently injected into the Äkta UV detector through the column bypass. The area of both eluting peaks was compared and used to determine the supernatant concentration. The equilibrium concentration of the protein in the adsorbed phase was calculated as follows: qp  *

where

*

qp

,

VL C

C

0 p

C

p



b

(10)

mS p

are the equilibrium concentrations of the protein in the adsorbed and liquid

phase respectively;

C

0 p

is the protein concentration in the standard solution (initial solution);

VL is the volume of the protein solution; mS is the mass of the resin; b is its bulk density. Each isotherm measurement was repeated twice for each salt concentration in liquid solution. The measurements were found to be reproducible. 3.4.3. Analysis of protein structure stability using nanoDSF method The stability of the protein structures was examined for both liquid and resin phases (i.e. on Butyl Sepharose 4 FF). The analysis was performed for samples of standard liquid solutions with different salt concentrations as well as supernatants and resins acquired during the static measurements. In the latter case, a 0.1 g of resin was contacted and equilibrated with the protein solutions of different volumes (0.3 - 1.5 mL) and protein concentrations (0.5 - 5 [mg mL-1]) in the same way as described in section 3.4.2. Samples of liquid solution and resin slurry were taken using a thin capillary with a volume of 10 µL. The capillaries were placed into trays of Prometheus NT.48 and subjected to the fluorescence analysis. The emission of fluorescent radiation with the wavelengths of 330 nm and 350 nm was measured with the temperature changes from 20 to 95°C, with the rate of 1°C min-1. The fluorescence curves obtained were differentiated, and the ratio of differentials was used to determine the melting temperature of the proteins.

6. 4. Results and Discussion 4.1. Chromatographic elution A series of chromatographic elution experiments was performed in isocratic mode, under different conditions of column overloading and salt concentrations in the mobile phase. Typical retention behavior of proteins is illustrated in Figs 1 and 2. In Fig. 1, elution profiles recorded under the same loading conditions but at various salt concentrations are depicted, whereas the profiles presented in Fig. 2, correspond to the same salt concentration in the mobile phase, but different loading conditions. 8

Insert Fig. 1

Insert Fig. 2 The incomplete protein elution that can be observed for both proteins, BSA and -La, arises from their conformation changes upon adsorption. The unfolded form of the protein is strongly bound to the adsorbent surface and retains in the column, while the native form elutes relatively easily. Such an elution pattern on HIC media is characteristic for unstable proteins, which unfold upon adsorption on hydrophobic surfaces [13,16,19-21]. The phenomenon of incomplete elution is intensified with increasing the salt concentration and decreasing flowrate (Q) of the mobile phase (Figs 1A, B), whereas it is reduced at high column overloading (Figs 2A, B). The salt effect is explained by the enhancement of hydrophobic interactions in the presence of kosmotropic salt. The influence of flowrate, which determines the residence time in the column, stems from slow kinetic rates of protein unfolding and mass transport. The effect of column overloading on the protein stability can be interpreted as inhibition of unfolding due to the competition of protein molecules for free active sites on hydrophobic ligands [13,34]. Another hypothesis was proposed by Fogle et al. [38], who attributed the mass overloading effect to energetic heterogeneity of the adsorbed surface caused by non-uniform distribution of hydrophobic ligands on chromatographic matrix.

4.2. Determination of model parameters – dynamic method The choice of mechanistic model for the description of protein unfolding was preceded by screening of various alternative mechanisms of the process. The numerical study excluded the possibility of binding the protein molecule to a single adsorption site, as well as to the number of sites higher than two. The possibility of protein unfolding due to energetic heterogeneity of adsorption sites was also examined. However, the simulations of dynamic profiles and isotherm courses revealed that at low feed concentrations the number of adsorption sites available for protein unfolding had to be of the same magnitude as the total binding capacity. Eventually, the model presented by the set of Eqs (4, 5, 7) was selected for further analysis, as it allowed prediction of chromatographic profiles and breakthrough curves with sufficient accuracy. To determine the model parameters, the inverse method was used, i.e. the dynamic model (Eq. (9)) along with the underlying kinetic equations (Eqs 4,5,7) was implemented into an 9

optimization routine, which was exploited to fit the model predictions to the experimental band profiles. The numerical methods used for solving the model and the optimization problem were similar to those described in our previous works (e.g. [6, 33]). The decision variables of the optimization procedure were the model parameters, i.e., the kinetic (kd, kf) and thermodynamic coefficients (q∞, δ, Ka, Ku). For the sake of simplicity, and to reduce the number of adjustable variables, the steric factor coefficient, δ, was set the same for both native and unfolded form of the protein. The protein concentration in the eluting peaks was relatively low; therefore, the contribution of the reaction of aggregation, which was of the second order with respect to the protein concentration in both mobile and adsorbed phases (Eq. 7), was neglected. Hence, for the simulation of chromatographic band profiles the kinetic term corresponding to the aggregation rate was omitted. The experimental basis for the estimation was a pair of two peaks eluted isocratically at the same salt concentration in the mobile phase. The peaks were selected in such a way that one of them corresponded to a low column overloading, for which phenomenon of incomplete elution was pronounced (only ca. 30 - 40% of the protein mass injected was eluted), while another - to a high column overloading, when most of the protein was eluted from the column (ca. 60 - 70% mass eluted). The objective function in the optimization procedure was the sum of the squared differences between the model predictions and experimental data: nexp 1

Sum 

 i 1

C

t  C exp 1  t   s im 1  

2

1

nexp 2





C

t  C exp 2  t   s im 2  

i 1

2

2

(11)

where: Csim, Cexp are the protein concentration at the same time point of the simulated and experimental profile, respectively; superscripts “1” and “2” denote the experiment number; ε1, ε2 are the weights, which were equal to the mass of the protein injected in each of experiments. A typical experimental pair of BSA peaks and the corresponding results of the model solution with the estimated model parameters are shown in Fig. 3 (peaks no 1,3). Insert Fig. 3 The peaks of α-La were distorted and almost unretained at a higher flowrate (i.e., Q = 1 [mL min-1]) due to strong mass transport resistances (Fig. 2B peak no 3). Therefore, the band profiles obtained at the lowest flowrate used in experiments (0.3 mL min-1) were selected as the basis for the parameter estimation. A similar quality of the optimization results was achieved to that reported for BSA. 10

The model parameters obtained for two different salt concentrations in the mobile phase, i.e., 0.85 and 1.02 M SA for BSA, and 0.595 and 0.68 M for α-La, are summarized in Table 1. The full set of the model parameters: kd, kf, q∞, δ, Ka, Ku, was estimated for BSA at 0.85 M AS and thoroughly verified based on various experimental data acquired under different operating conditions regarding the column loading and flowrate. Next, the binding capacity q∞, which was attributed to the total amount of active sites in the first layer on the adsorbent surface, was set constant, independently of the process conditions and the protein type, and five parameters were estimated (kd, kf, δ, Ka, Ku). The contribution of steric hindrances, specific for the protein type and the salt concertation was accounted for by the steric hindrance coefficient, δ. Insert Table 1 As expected, for both proteins the values of the equilibrium constants of adsorption and unfolding (Ka, Ku) increase with increasing the salt concentration, which arises from the enhancement of hydrophobic interactions between the proteins and adsorbent surface. That trend is particularly pronounced for α-La, which is more strongly adsorbed compared to BSA. The change in the folding rate is opposite, i.e., kf parameter decreases with increasing the salt content, which is in agreement with the observation reported by Deitcher et al. [34]. This indicates the tendency of the adsorbed proteins to remain in the unfolded state at a high salt content in the mobile phase. The lumped kinetic coefficient, kd, is dependent on the salt concentration, and decreases significantly with increase in the salt content in the mobile phase. As mentioned above (section 2.1), that parameter accounts for the contribution of adsorption-desorption rate as well as inter- and intraparticle mass transport resistances to the kinetics of adsorption process for the protein in the native form. Though the molecule of α-La (MW = 14.2 kDa) is much smaller compared to BSA (MW = 66.5 kDa), the value of kd is lower for α-La than for BSA at a similar adsorption strength. This indicates strong resistances to mass transport in adsorption of α-La, which was the cause of peak splitting observed at higher flowrates (Fig.2B, peak no 3). The value of the steric factor coefficient, δ, was found to be very large for BSA and even larger for α-La. As discussed in the next section, both proteins are prone to form aggregates, which particularly holds true for α-La. The presence of high molecular weight aggregates is expected to induce steric hindrances for the mass transport of the protein within the pores of resin particles. It may cause shielding of the adsorption sites and inhibiting protein unfolding, as it is quantified by Eq. (5). The argument 11

in favor of that hypothesis is low value of kd for the proteins, particularly for α-La. The shielding coefficient decreases with increasing the salt concentration. This may be caused by differences in the values of the steric hindrance coefficient for the native and unfolded forms of the proteins, which was not accounted for by the model. However, it should be kept in mind that the mathematical model proposed here is a simplification of the complex phenomena of protein unfolding at the solid interface, therefore, a direct correlation between the values of the model parameters and the course of a real process is not possible. Nevertheless, the model simulation can be used to verify different hypotheses of the process mechanism and to identify the most important trends in adsorption behavior. The accuracy of the model predictions was verified based on various chromatographic elution profiles acquired at different column loadings (e.g., Fig. 2A and B, peaks no 1and 2), flowrates (e.g., Figs 2A and B, peak no 3), as well as breakthrough curves (Fig. 4). Insert Fig. 4 It can be observed that for both proteins the influence of column overloading on the elution pattern was predicted in a good quality. The flowrate effect was also reproduced well (Figs 2A and B, peaks no 3). The predictions of the shapes of breakthrough curves was accurate for the inlet concentration below 1 [mg mL-1]. At higher inlet concentrations the model underpredicted the broadening of the lower part of the concentration front, and the asymmetry of its upper part. The band broadening was attributed to strong concentration dependency of the lumped coefficient kd, which decreased with increase in the protein concentration, which is characteristic for slow rates of mass transport kinetics. The discrepancy between the model predictions and experimental data was observed only for the inlet protein concentrations higher than 1 [mg mL-1]. Also, for the inlet concentrations higher than 1 [mg mL-1], the asymmetry of the upper part of the breakthrough curves could not be reproduced by the model simulations. This may indicate the occurrence of an additional phenomenon contributing to the adsorption mechanism, which was assigned to protein aggregation in adsorbed phase. That mechanism was inactive in chromatographic elution, which was characterized by relatively low residence time and strong dilution of concentration profiles.

4.3. Static measurements of adsorption isotherms To gain a better insight into the adsorption mechanism, the adsorption isotherm was measured for both proteins at two different salt concentrations in the liquid solution, for which the phenomenon of incomplete elution was pronounced. The measurements were performed using 12

the static method (section 3.3.2). The stirring of the resin slurry in the static measurements accelerated mass transport, therefore, the adsorption equilibrium established in a shorter time compared to the dynamic measurements. The concentration profiles became time-invariant in no longer than two hours (section 3.3.2), regardless of the column loading conditions and the salt concentration. This proved that all surface reactions were reversible. The results of the isotherm measurements are depicted in Fig 5. The course of the isotherm curves indicates continuous increase in the protein concentration in the adsorbed phase with increasing the liquid phase concentration. Similar isotherm shapes have been reported in a few studies, where adsorption of BSA and other unstable proteins on Sepharose-type resins was investigated [40-43]. All curves are convex, which is reflected in the peak shapes recorded under overloading conditions - with a sharp front and a diffuse rear profile. It is also evident that the isotherms are nonlinear in very wide concentration range. The curves bend at low protein concentration, i.e. even below 0.1 mg/mL. This means that nonlinear isotherm conditions can be reached at relatively low column loadings. A rough information on the range of isotherm nonlinearity in chromatographic elution can be provided by matching the value of the protein concentration at the peak maximum with the corresponding point on the isotherm course. For instant, the peaks no 1, 2 in Fig. 2A and no 1 in Fig. 2B are eluted under nonlinear isotherm conditions. Though the isotherm are convex, they do not correspond to the Langmuir type, which is demonstrated using the Scatchard linearization, q*/C= f(q*) (Fig. 5B). This stems from complex adsorption behavior of proteins that involves conformational changes of their structure. To analyze the interplay between the dynamic (i.e., the inverse method) and static methods of determining the isotherm parameters, the isotherm course was predicted by the kinetic model comprising Eqs (4), (5) and (7) at the steady state. The calculation results are compared to the experimental isotherm data in Fig.5. A significant discrepancy between the model predications and equilibrium data is evident. In case of BSA, the agreement between both methods exists at low protein concentration, whereas for -La the discrepancy occurs almost over the whole investigated concentration range. Moreover, the experimental isotherm data obtained for -La at two different salt concentration in the liquid solution (0.595 and 0.68 M AS), coincide, which was confirmed by repetitive measurements, while the corresponding dynamic elution profiles and simulations of the isotherm course depend strongly on the salt concentration. Insert Fig. 5 13

We hypothesized that the inconsistency between the static and dynamic methods for determining the isotherm was attributed to aggregation of protein molecules at the solid interface. Carta and co-workers have already evidenced the aggregation of protein molecules that unfolded in the adsorbed phase [28, 44, 45]. To confirm that suggestion, the SEC analysis was performed for samples of supernatants obtained in the static measurements as well as after desorption of the proteins from the resin. In case of -La a small amount of aggregates was found in the supernatants, which was not present in the standard solutions (Fig. S1 supplementary materials). In case of BSA, no presence of aggregates in the supernatants was observed, however, it is possible that they were reversibly formed only in the solid phase, and decomposed in the liquid phase (Eq. 3). To evaluate roughly aggregation phenomenon, we extended the model with a simplified mechanism presented by Eq. (6), which was the second order reaction with respect to the protein concentration in the liquid and adsorbed phases. The equilibrium constant of aggregation was evaluated by fitting the model simulation to the isotherm data obtained by the static method. Next, the whole model was used to simulate the breakthrough curves. The kinetic aggregation coefficient kag,d, was adjusted based on the shape of breakthrough curves at the highest inlet concentration. The additional coefficients obtained are presented in Table 2. The extended model was used to repeat all simulations of chromatographic elution. As expected, no influence of aggregation on peak shape in chromatographic elution and on the start of the breakthrough curves, was observed. The latter can be correlated to the dynamic binding capacity. Insert Table 2 In case of -La, the aggregation mechanism given by Eq. (6) was oversimplified; the model simulations indicated that the number of adsorbed molecules of proteins participating in aggregation is much higher compared to BSA, which was not predicted correctly by the model. The contribution of aggregation was expected to dominate in the adsorption mechanism of -La, which can be a cause of strong discrepancy between the isotherm courses determined experimentally and those predicted based on the inverse method. (Fig. 5 D, isotherm data for 0.595 and 0.69 M AS). Nevertheless, for both proteins, the additional adsorption mechanism was active outside of the range of the column loadings practically applicable. It occurred when most of active sites on the adsorbent surface was occupied by protein molecules, when chromatographic peaks were very weakly retained. 14

4.4. NanoDSF measurements of protein stability Finally, the unfolding transition points (“melting temperatures”) of the proteins were determined using the nanoDSF technique. The proteins that exhibit a stable structure melt at higher temperatures compared to those occurring in unfolded, or partly unfolded forms. Therefore, measurement of melting temperature can be used to detect the protein state in its environment. For this purpose, the samples of standard protein solutions, supernatants, and slurry of resins obtained after contacting with various liquid solutions of proteins, which differed in the salt content, volume and protein concentration, were subjected to the measurements. In all cases, the fluorescence curves were measured at two different wavelengths (330 and 350 nm). Next, the curves were differentiated and the ratio of the differentials was calculated to determine the melting temperature. The ratio of differentials is presented in the form of a peak, which maximum corresponds to the average melting temperature of all protein molecules present in the solution (Fig. 6). Typical results of measurement are presented in Figs 7 A-C). The difference in the peak sign, i.e. negative or positive, does not indicate different melting phenomena for the two proteins; it just depends on the result of differentiation of the fluorescence curves at 330 and 350 nm. Insert Fig. 6 Insert Fig. 7 In Fig. 7A, the influence of the salt concentration on the melting temperature of BSA is presented. The melting temperature of the protein in the supernatant was found to be the same as in the standard solutions, independently of the salt concentration used in experiments. This indicated that protein was stable in the supernatant and unfolded form was accumulated only in the adsorbed phase. This also confirmed strong adsorption of the unfolded form and its absence from the liquid phase, in agreement with the model assumptions. As it can be observed, the protein in the adsorbed phase was significantly less stable than in the liquid solution. The melting temperature peak shifted toward lower temperatures with increasing the salt content in the solution used for the resin equilibration. This reveals the enhancement of hydrophobic interactions with increase in the salt content, which was reflected in the reported values of the model parameters.

15

The effect of column loading on the stability of BSA is presented in Fig. 7B. It can be observed that increase in the column loading causes protein stabilization in the adsorbed phase. The peaks in each pair no 2, 3 and 4, 5 in Fig. 7B correspond to the same protein concentration, but different volumes of the solution used for the equilibration of resin. It is evident that the effect of protein stabilization enhances at the higher mass overloading , for which the protein concentration at equilibrium corresponds to strong nonlinearity of the isotherm, in agreement with the dynamic and thermodynamic measurements and simulations reported above. In case of -La, strong destabilization of the adsorbed protein can be observed (Fig. 7C). The melting temperature of the protein was shifted toward 20C. This confirms the model predictions, which indicated a tendency of the protein to unfold and aggregate. Nevertheless, the protein structure in the supernatant was stable, with the melting temperature the same as for the standard liquid solutions. This proved that unfolded form of the proteins was accumulated mainly in the adsorbed phase, similarly as it was reported above for BSA.

7. 5. Conclusions A mechanistic model was proposed to describe the retention behavior of unstable proteins BSA and -La on a Butyl Sepharose medium under nonlinear isotherm conditions. A two-site binding adsorption and reversible unfolding surface reaction were assumed, where the unfolded form was present only in the adsorbed phase. The unfolding phenomenon enhanced with increasing the salt content in liquid solution and reduced with increasing column loading. The latter was explained by the reduction in availability of the adsorption sites due to isotherm nonlinearity, which aroused from steric hindrances to mass transport at the solid interface. The kinetic rate equations of the surface reaction were implemented into the model of the column dynamics and used to predict band profiles and estimate the isotherm course. The model was found to be efficient in reproducing the shapes of band profiles. However, the isotherm courses predicted based on the dynamic method of the isotherm determination were in disagreement with the results obtained by the static method. This was particularly pronounced for -La. The discrepancy was attributed to protein aggregation in the adsorbed phase, which was a low-rate process, active at high protein concentrations. The phenomenon of aggregation was additionally accounted for in mathematical modelling, however, it was found to have no influence on the shape of chromatographic peaks and the dynamic binding capacity. 16

The model predictions were supported by nanoDSF measurements of protein stability in liquid and solid phase solutions. The results of those measurements revealed high stability of protein structure in the liquid phase at adsorption equilibrium, and its destabilization in the adsorbed phase. This proved that the unfolded form of the protein mainly accumulated in the adsorbed phase. The instability of -La was suggested as a cause of its aggregation in the adsorbed phase. The presence of aggregates might result in pore clogging on the adsorbent surface, which could explain high values of steric hindrance factors and slow rate of mass transport kinetics predicted by the mathematical model. The nanoDSF technique can be exploited as an efficient tool for identifying the presence of the unfolded protein in both liquid and adsorbed phases, estimating the range of the isotherm nonlinearity, which can support the formulation of mathematical model. It also allows simultaneous measurements of protein stability in a huge number of small-volume samples. The combination of that technique with mathematical modelling can bring a benefit in fast evaluation of adsorption behavior of unstable proteins on HIC media.

8. Acknowledgments Financial

support

of

this

work

by

National

Science

Center

(project

UMO-

2015/18/M/ST8/00349) is gratefully acknowledged. NanoTemper Technologies Sp. z o.o., Kraków, Poland is greatly acknowledged.

17

9. 10.

5. References

[1] S. Hjerten, Some general aspects of hydrophobic interaction chromatography, J. Chromatogr. A 87 (1973) 325–331. [2] S. Pahlman, J. Rosengren, S. Hjerten, Hydrophobic interaction chromatography on uncharged sepharose derivatives: Effects of neutral salts on the adsorption of proteins, J. Chromatogr. A 131 (1977) 99–108. [3] J.A. Queiroz, C.T. Tomaz, J.M.S. Cabral, Hydrophobic interaction chromatography of proteins, J. Biotech. 87 (2001) 143–159. [4] A.M. Azevedo, P.A.J. Rosa, I.F. Ferreira, M.R. Aires-Barros, Integrated process for the purification of antibodies combining aqueous two-phase extraction, hydrophobic interaction chromatography and size-exclusion Chromatography, J. Chromatogr. A 1213 (2008) 154– 161. [5] W. Marek, R. Muca, S. Woś, W. Piątkowski, D. Antos, Isolation of monoclonal antibody from a CHO supernatant. I. Assessment of different separation concepts, J. Chromatogr. A 1305 (2013) 55–63. [6] W. Marek, R. Muca, S. Woś, W. Piątkowski, D. Antos, Isolation of monoclonal antibody from a CHO supernatant. II. Dynamics of the integrated separation on IEC and HIC column, J. Chromatogr. A 1305 (2013) 64–75. [7] A. Mahn, J.A. Asenjo, Prediction of protein retention in hydrophobic interaction Chromatography, Biotech. Adv. 23 (2005) 359–368. [8] A. Mahn, G. Zapata-Torres, J.A. Asenjo, A theory of protein-resin interaction in hydrophobic interaction chromatography, J. Chromatogr. A 1066 (2005) 81–88. [9] J.C. Salgado, I. Rapaport, J.A. Asenjo, Predicting the behavior of proteins in hydrophobic interaction chromatography 1: Using the hydrophobic imbalance (HI) to describe their surface amino acid distribution, J. Chromatogr. A, 1107 (2006) 110–119. [10] Y. Kato, K. Nakamura, T. Kitamura, M. Hasegawa, H. Sasaki, Hydrophobic interaction chromatography at low salt concentration for the capture of monoclonal antibodies, J. Chromatogr. A 1036 (2004) 45–50. [11] 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.

18

[12] 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. [13] E. Haimer, A. Tscheliessnig, R. Hahn, A. Jungbauer, Hydrophobic interaction chromatography of proteins IV. Kinetics of protein spreading, J. Chromatogr. A 1139 (2007) 84–94. [14] S.C. Goheen, S.C. Engelhorn, Hydrophobic interaction high-performance liquid chromatography of proteins, J. Chromatogr. A 317 (1984) 55–65. [15] S.L. Wu, K. Benedek, B.L. Karger, Thermal behavior of proteins in high-performance hydrophobic-interaction chromatography On-line spectroscopic and chromatographic characterization, J. Chromatogr. A 359 (1986) 3–17. [16] T.T. Jones, E.J. Fernandez, α-Lactalbumin tertiary structure changes on hydrophobic interaction chromatography surfaces, J. Colloid Inter. Sci. 259 (2003) 27–35. [17] A. Jungbauer, C. Machold, R. Hahn, Hydrophobic interaction chromatography of proteins: III. Unfolding of proteins upon adsorption, J. Chromatogr. A 1079 (2005) 221–228. [18] N. J. Greenfield, Using circular dichroism collected as a function of temperature to determine the thermodynamics of protein unfolding and binding interactions, Nat. Protoc. 1 (2006) 2527–2535. [19] M.F.M. Engel, C.P.M. van Mierlo, A.J.W.G. Visser, Kinetic and structural characterization of adsorption-induced unfolding of bovine α-lactalbumin, J. Biol. Chem. 277 (2002) 10922–10930. [20] P. Oroszlan, R. Blanco, X.M. Lu, D. Yarmush, B.L. Karger, Intrinsic fluorescence studies of the kinetic mechanism of unfolding of α-lactalbumin on weakly hydrophobic chromatographic surfaces, J. Chromatogr. A 500 (1990) 481–502. [21] R. Ueberbacher, E. Haimer, R. Hahn, A. Jungbauer, Hydrophobic interaction chromatography of proteins V. Quantitative assessment of conformational changes, J. Chromatogr. A 1198–1199 (2008) 154–163. [22] R. Ueberbacher, A. Rodler, R. Hahn, A. Jungbauer, Hydrophobic interaction chromatography of proteins: Thermodynamic analysis of conformational changes, J. Chromatogr. A, 1217 (2010) 184–190. [23] W.Y. Chen, H.M. Huang, C.C. Lin, F.Y. Lin, Y.C. Chan, Effect of temperature on hydrophobic interaction between proteins and hydrophobic adsorbents: Studies by isothermal titration calorimetry and the van't Hoff equation, Langmuir 19 (2003) 9395–9403.

19

[24] Y.W. Bai, T.R. Sosnick, L. Mayne, S.W. Englander, Protein folding intermediates: Native-state hydrogen exchange, Science 5221 (1995) 192–197. [25] Z.Q. Zhang, D.L. Smith, Determination of amide hydrogen exchange by mass spectrometry: A new tool for protein structure elucidation, Protein Sci. 2 (1993) 522–531. [26] J. Buijs, C.C. Vera, E. Ayala, E. Steensma, P. Hakansson, S. Oscarsson, Conformational stability of adsorbed insulin studied with mass spectrometry and hydrogen exchange, Anal. Chem. 71 (1999) 3219–3225. [27] J.L. McNay, E.J. Fernandez, How does a protein unfold on a reversed-phase liquid chromatography surface?, J. Chromatogr. A 849 (1999) 135–148. [28] J. Guo, G. Carta, Unfolding and aggregation of monoclonal antibodies on cation exchange columns: Effects of resin type, load buffer, and protein stability, J. Chromatogr. A, 1388 (2015) 184–194. [29] Y. Xiao, T. Tibbs-Jones, A.H. Laurent, J.P. O’Connell, T.M. Przybycien, E.J. Fernandez, Protein instability during HIC: Hydrogen exchange labeling analysis and a framework for describing mobile and stationary phase effects, Biotechnol. Bioeng. 96 (2007) 80–93. [30] B.C.S. To, A.M. Lenhoff, Hydrophobic interaction chromatography of proteins: III. Transport and kinetic parameters in isocratic elution, J. Chromatogr. A, 1205 (2008) 46–59. [31] R. Muca, W. Marek, W. Piątkowski, D. Antos, Influence of the sample-solvent on protein retention, mass transfer and unfolding kinetics in hydrophobic interaction chromatography, J. Chromatogr. A 1217 (2010) 2812–2820. [32] W. Marek, R. Muca, W. Piątkowski, D. Antos, Multiple-injection technique for isolating a target protein from multicomponent mixtures, J. Chromatogr. A 1218 (2011) 5423–5433. [33] R. Bochenek, W. Marek, W. Piątkowski, D. Antos, Evaluating the performance of different multicolumn setups for chromatographic separation of proteins on hydrophobic interaction chromatography media by a numerical study, J. Chromatogr. A 1301 (2013) 60– 72. [34] R.W. Deitcher, Y. Xiao, J.P. O’Connell, E.J. Fernandez, Protein instability during HIC: Evidence of unfolding reversibility, and apparent adsorption strength of disulfide bondreduced α-lactalbumin variants, Biotechnol. Bioeng. 102 (2009) 1416–1427. [35] Y. Xiao, A.S. Freed, T. Tibbs-Jones, K. Makrodimitris, J.P. O’Connell, E.J. Fernan- dez, Protein instability during HIC: Describing the effects of mobile phase conditions on instability and chromatographic retention, Biotechnol. Bioeng. 93 (2006) 1177–1189.

20

[36] Y. Xiao, T. Tibbs-Jones, A.H. Laurent, J.P. O’Connell, T.M. Przybycien, E.J. Fernandez, Protein instability during HIC: Hydrogen exchange labeling analysis and a framework for describing mobile and stationary phase effects, Biotechnol. Bioeng. 96 (2007) 80–93. [37] G. Senisterra, I. Chau, M. Vedadi, Thermal denaturation assays in chemical biology. Assay, Drug Dev. Technol. 10 (2012) 128–136.

[38] J. L. Fogle, J. P. O’Connell, E. J. Fernandez, Loading, stationary phase, and salt effects during hydrophobic interaction chromatography: -Lactalbumin is stabilized at high loadings, J. Chromatogr. A 1121 (2006) 209–218. [39] G. Guiochon, A. Felinger, D. G. Shirazi, Fundamentals of preparative and nonlinear chromatography, Academic Press 2006, Amsterdam, The Netherlands. [40] Q. Meng, J. Wang, G. Ma, Z. Su, Isotherm type shift of hydrophobic interaction adsorption and its effect on chromatographic behavior, J. Chromatogr. Sci. 51 (2012) 173– 180. [41] M. R. Mirani, F. Rahimpour, Thermodynamic modelling of hydrophobic interaction chromatography of biomolecules in the presence of salt, J. Chromatogr. A 1422 (2015), 170– 177. [42] J. Chien, St. M. Cramer, Protein behavior in hydrophobic interaction chromatography, J. Chromatogr. A 1165 (2007) 67–77. [43] J. Chien, St. Y. Sun, Modeling of the salt effects on hydrophobic adsorption equilibrium of protein. J. Chromatogr. A 992 (2003) 29–40. [44] J. Guo, S. Zhang, G. Carta, Unfolding and aggregation of a glycosylated monoclonal antibody on a cation exchange column - part I. Chromatographic elution and batch adsorption behavior, J. Chromatogr. A 1356 (2014) 117–28. [45] J. Guo, G. Carta, Unfolding and Aggregation of a glycosylated monoclonal antibody on a cation exchange column - part II. Protein structure effects by hydrogen deuterium exchange mass spectrometry, J. Chromatogr. A 1356 (2014) 129–37.

21

11.

Captions

Fig.1. Effect of the salt concentration on the elution pattern of the proteins. A) BSA; peak no 1 - buffer free of AS; no 2 - 0.34 M AS; no 3 - 0.68 M AS; no 4 - 0.85 M AS; B) α-La; peak no 1 - buffer free of AS; no 2 - 0.50 M AS; no 3 - 0.595 M AS; no 4 - 0.68 M AS. the injection volume Vinj = 0.1 mL, Cp,inj denotes the protein concentration in the injected sample; Q is the mobile phase flowrate. Fig.2. Effect of column overloading and flowrate on the retention pattern of the proteins. The percent amount indicates the ratio of the mass of the protein eluted to the mass injected into the column. Symbols - experimental data, lines - model simulations; Cinj [mg mL-1], Vinj [mL], Q [mL min-1]; A) BSA; peak no 1: Cinj = 2, Vinj = 2, Q = 1; peak no 2: Cinj = 5, Vinj = 0.1, Q = 1; peak no 3: the same as peak no 2 but Q = 0.5; B) α-La; peak no 1: Cinj = 5, Vinj = 2, Q = 0.3, peak no 2: Cinj = 5, Vinj = 0.1, Q = 0.3; peak no 3: the same as for peak no 2 but Q = 1. Fig.3. Illustration of the peak fitting procedure. Cinj [mg mL-1], Vinj [mL]: peak no 1: Cinj = 2, Vinj = 2, and peak no 3: Cinj = 5, Vinj - 0.1, were used as the experimental basis for the model parameter estimation; peak no 2: Cinj = 2, Vinj = 0.5 – an illustration of the model verification. Symbols – experimental data, lines – model simulations. Fig 4. Typical courses of the breakthrough curves for BSA; symbols - experimental data, lines model simulations; line no 1 (Cinj = 0.3 mg mL-1) and no 2 (Cinj = 0.5 mg mL-1): simulations with the model parameters presented in Table 1; line no 3 (Cinj = 3 mg mL-1): the model was extended with the aggregation kinetics (section 4.3), the kd value was threefold lower than that presented in Table 1; line no 4: the same as line no 3 but no aggregation was included. Fig. 5. Adsorption isotherms: (A), (C) BSA, and (D) -La; B) the Scatchard plot q*/C= f(q*) for BSA corresponding to the plots depicted in (A); symbols - experimental data, lines - the model simulations. Fig. 6. Illustration of the course of the fluorescence curves and determination of the melting temperature. The sample concentration 1 mg mL-1 of BSA in 0.85 M salt solution. Fig.7. Illustration of the protein stability in the liquid solution and resin based on nanoDSF measurements. A) Effect of the salt concentration on the stability of BSA in resin: peak no 1; buffer free of AS or supernatant regardless of the salt content in the resin slurry; no 2: 0.51 M AS; no 3: 0.85 M AS; no 4: 1.02 M AS; B) effect of the column loading on the stability of BSA in resin, the initial protein concentration in the solution, Cp [mg mL-1], and the solution volumes, V [mL], were used for the resin equilibration: peak no 1: buffer free of AS or supernatant; no 2: Cp = 5, V = 1.5; no 3: Cp = 5, V = 0.3; no 4: Cp = 1, V = 1.5; no 5: Cp = 1, V = 0.3; C) effect of the column loading on the stability of -La; peak no 1: buffer free of AS; no 2: a supernatant (from the initial solution of Cp = 5, V = 0.3); no 3: Cp = 5, V = 0.5; no 4: Cp = 5, V = 0.3.

22

23

24

25

26

Table 1. Values of kinetic parameters estimated for the kinetic model consisting of Eqs 4 and 5. BSA 2 2 Csalt Ka x 10 kd x 10 Ku x 102 kf x 104 q∞  -1 [s-1] [mL mg-1] [s-1] [-] [M] [mL mg-1] [mg m L c o l ] 0.85 1.02

3.34 16.9

1.66 0.597

0.595 0.68

10.45 23.2

0.430 0.229

1.45 0.335 α-La 0.0541 0.0908

2.00 1.42

106.6a 20.0 106.6b 14.0

1.51 0.907

106.6b 81.2 106.6b 58.0

a) estimated b) set the same as in a)

27

Table 2 Parameters of the aggregation reaction for BSA. Csalt M AS

Kagg

kag,d x 103 -1

[mL mg ]

0.85 1.02

0.51 1.61

[s-1]

2.1 2.0

28