Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmaceutical Biotechnology
Relating ProteineProtein Interactions and Aggregation Rates From Low to High Concentrations Ranendu Ghosh 1, Cesar Calero-Rubio 1, Atul Saluja 2, Christopher J. Roberts 1, * 1 2
Department of Chemical & Biomolecular Engineering, University of Delaware, Newark, Delaware 19716 Department of Drug Product Science and Technology, Bristol-Myers Squibb, New Brunswick, New Jersey 08901
a r t i c l e i n f o
a b s t r a c t
Article history: Received 29 October 2015 Revised 22 December 2015 Accepted 6 January 2016
At low protein concentrations (c2), non-native protein aggregation rates are known to be sensitive to changes in conformational stability and “weak” or “colloidal” proteineprotein interactions. Protein eprotein interactions are also known to be strong functions of c2. In the present work, proteineprotein interactions and rates of aggregation were quantified systematically for a monoclonal antibody (MAb) across a broad range of c2 at pH 5.1 and 6.5, with or without 5 wt/wt % sucrose or 100 mM NaCl present. Aggregation rates were determined from initial-rate analysis with size-exclusion chromatography, and interactions were quantified with static and dynamic laser light scattering. A number of hypotheses were tested regarding whether changes in proteineprotein interactions can be predictive of changes in aggregation rates versus c2. Hypotheses were based on (i) changes in thermodynamic activity; (ii) statistical mechanical fluctuation theory; and (iii) surface-contact probabilities. Arguments based on (i) and (ii) were qualitatively inconsistent with experimental rates and scattering. Hypothesis (iii) was reasonably successful and resulted in a semiquantitative correlation between rates and proteineprotein interactions across almost 2 orders of magnitude in c2. However, (iii) requires one to assume that the concentrationdependent proteineprotein KirkwoodeBuff integral is a reasonable surrogate for contact probabilities. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: light scattering (dynamic) biophysical models protein aggregation protein formulation biotechnology
Introduction Proteins are inherently labile molecules that have marginally stable structures and can be degraded and/or inactivated readily under reasonably mild in vitro and in vivo conditions. Non-native protein aggregation (hereafter referred to simply as aggregation) denotes the processes by which natively folded proteins form aggregates via interprotein contacts that involve secondary and/or tertiary structures that are measurably perturbed from the native or folded monomer state.1-3 In this context, aggregates are typically net irreversible under the conditions that they form, and therefore the rates or kinetics of aggregation are a key quantity in determining product quality and shelf life. As aggregation is under kinetic control, the mechanism(s) of aggregation ultimately dictate This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.01.004. Conflict of interest: The authors declare no conflicts of interest for any aspects of the work reported here. Present address for Ranendu Ghosh: Formulation Development Laboratory, Biocon Research Ltd., Bangalore, Karnataka, India, 560099. * Correspondence to: Christopher J. Roberts (Telephone: 302-831-0838; Fax: 302831-1048). E-mail address: [email protected] (C.J. Roberts).
the observed aggregation rates. Aggregation of therapeutic proteins is a long-standing issue during drug product development and commercial manufacturing3,4 and is also problematic to control during protein expression in vivo.2 Studies at relatively low protein concentration (c2) have led to a reasonably common viewpoint, in which aggregation rates or kinetics are mediated by at least 3 main factors:5-7 (1) protein conformational stability or conformational changes that expose stretches of hydrophobic amino acids or so-called aggregation “hot spots”; (2) “weak” or “colloidal” attractions (repulsions) that facilitate (inhibit) reversible protein selfassociation; and (3) intrinsic aggregation propensity, that is, strong interactions between exposed sequence “hot spots” that create net irreversible contacts between proteins, often via formation of interprotein beta sheet structures. Environmental factors that greatly alter conformational stability include temperature, pressure, solution composition, and adsorption to solideliquid interfaces.3,8,9 Protein conformational stability is theoretically predicted to be sensitive to c2, but direct experimental evidence for the magnitude of the effect is very limited.10,11 Similarly, factors that have been found to greatly alter proteineprotein interactions include c2 and solution conditions such as pH and the concentration of commonly employed cosolutes, for example, salts, amino acids, and polyhydroxy
http://dx.doi.org/10.1016/j.xphs.2016.01.004 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
compounds such as sucrose.12,13 The intrinsic aggregation propensity of a protein when it is unfolded is putatively dictated by its primary sequence and is not an easy variable to control once a given protein candidate has been selected for product development and testing in patients.7,14 Accurate prediction of aggregation rates is a long-standing challenge in the biotechnology industry. For low c2 conditions, phenomenological and mechanistic models have been developed in an attempt to provide quantitative prediction of aggregation rates for natively unfolded polypeptides under physiological conditions14,15 and for therapeutic proteins under typical formulation conditions.16-19 In all cases, the models either interpolate or extrapolate rate data from one experimental condition or protein sequence to another. These models have focused primarily on the effects of conformational stability and the inherent aggregation propensity of “hot spot” sequences. Weak or colloidal interactions have been shown to correlate, at least qualitatively, with aggregation rates at low c2.12,20 However, in a number of cases, it is an inverse correlation because changes in conformational stability outweigh changes in proteineprotein interactions as one alters formulation conditions.21-23 In contrast, relatively little has been done to quantitatively or semiquantitatively connect changes in proteineprotein interactions with aggregation rates at high c2.24-26 In part, this may be due to the relatively small amount of published data for aggregation rates across a broad range of concentration as most studies either focus on only high-c2 or low-c2 data or provide only a small number of different c2 conditions.16-19,21,23,26 The present study focuses on a systematic comparison of proteineprotein interactions and accelerated aggregation rates (elevated temperature, 50 C) for a monoclonal antibody (MAb) of the IgG1 subclass, over a broad range of c2 (~1-102 g/L) at pH 5.1 and 6.5, with and without either sucrose (5 wt/wt %) or NaCl (100 mM) present. Aggregation rates are based on loss of monomer, as quantified ex situ with size-exclusion chromatography (SEC). Proteineprotein interactions are quantified in situ with static and dynamic light scattering (DLS) for the same c2 as the measured aggregation rates. To the extent possible within experimental limits, changes in conformational stability are inferred from differential scanning calorimetry at low c2. Concentration-dependent aggregation rates are used to scrutinize general mass-action arguments that have been developed primarily for low-c2 conditions and that encompass a family of models for aggregation kinetics.27 Based on previous work and new arguments presented here, a number of hypotheses are tested with respect to both qualitative and quantitative agreements or “predictability” for experimental aggregation rates in terms of measureable proteineprotein interactions. These include changes in thermodynamic activity or chemical potential, local concentration fluctuations, and the probability of protein surfaces coming into contact. The results highlight a number of limitations for existing mass action models of protein aggregation, as well as hypotheses for how to quantitatively link proteineprotein interactions to aggregation rates at high c2. The proteineprotein KirkwoodeBuff integral (G22) is offered as a surrogate for proteineprotein contact probabilities that are intuitively expected to influence aggregation rates at high concentration. It is found that a reasonable correlation exists between kobs and G22 if one has a reference data set against which to normalize. Finally, outstanding challenges are discussed for predicting and/or measuring the thermodynamic properties of proteins (e.g., native protein chemical potential) and the conformational stability of proteins at elevated concentrations. Overcoming these challenges will require significant advancements in terms of experimental capabilities and modeling approaches.
1087
Materials and Methods Sample Preparation Purified IgG1 was provided by Bristol-Myers Squibb at a starting concentration of ~54 mg/mL. Stock solutions were dialyzed against a given buffer condition (Spectra/Por 7 tubing, 10 kDa MWCO; Spectrum Laboratories, Santa Clara, CA) using four 12 h buffer exchanges of 500 mL each, in refrigerated conditions. Buffer-only conditions were either 10 mM sodium acetate at pH 4.0 or 10 mM histidine HCl at pH 6.35. Glacial acetic acid (Fisher Scientific) and histidine HCl (Sigma) were used for preparation of dialysis buffers. After dialysis, protein solutions were concentrated using Amicon (Millipore, Billerica, MA) ultracentrifugation tubes with a molecular weight cutoff of 10 kDa. The concentrate was collected at different stages of centrifugation so as to monitor solution pH and c2 until the desired pH (pH 5.1 ± 0.1 for acetate buffer and pH 6.5 ± 0.1 for histidine buffer) and concentration (>170 mg/mL) were achieved. Supplementary data include illustrative results from repeated sample preparations. As expected, the solution pH shifted during the protein concentrating step because of self-buffering by the protein.28 The lower starting pH values (before the concentration step) were selected based on trial runs, so as to achieve the desired final pH values in the concentrated stocks. Solutions at lower c2 were subsequently prepared by dilution of the concentrated stock at either pH 5.1 or pH 6.5, using the corresponding buffer solution and independently confirmed to maintain the desired pH after dilution. Samples with either sucrose or NaCl were prepared by gravimetrically diluting concentrated stock solutions to lower concentration. This was done using the appropriate ratios of stock solution, buffer-only solution (pH 5.1 or pH 6.5), and buffer with either 500 mM NaCl or 25% (wt/wt) sucrose. This was done to achieve the desired c2 and a final NaCl concentration of 100 mM or a final sucrose concentration of 5% (wt/wt). Table 1 lists the set of final solution conditions. Final pH was verified for all samples, and all solutions were filtered using 0.22 mm low protein binding filters (Millipore). Final concentrations were confirmed via UV absorbance (Agilent Technologies, Santa Clara, CA) at 280 nm, with an extinction coefficient of 1.54 mL mg1 cm1.
Differential Scanning Calorimetry Differential scanning calorimetry (DSC) experiments were performed using a VP-DSC instrument (Microcal, Northhampton, MA) for antibody solutions with 1 mg/mL protein at each of the different solution conditions in Table 1. Multiple bufferebuffer scans were performed to obtain baseline values and establish thermal history on the instrument immediately before sample scans. Thermal scans were performed over 20 C-90 C with a scan rate of 60 C/h. The average of the buffer scans was subtracted from the subsequent Table 1 Summary of Formulation Conditions Formulation Description pH pH pH pH pH pH a b c
5, buffer onlyb 5, buffer þ NaCl 5, buffer þ sucrose 6.5, buffer onlyc 6.5, buffer þ NaCl 6.5, buffer þ sucrose
Additional Excipient (Final Concentration)
Colors for Curves/ Symbols in Figuresa
none NaCl (100 mM) Sucrose (5% wt/wt) none NaCl (100 mM) Sucrose (5% wt/wt)
Black Blue Red Black Blue Red
Figures 5, 6, and 8 use closed symbols for pH 5, and open symbols for pH 6.5. 10-mM sodium acetate, pH 5.0 ± 0.1. 10-mM histidine HCl, pH 6.5 ± 0.1.
1088
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
protein scan, and the raw DSC data were converted to partial specific heat capacity values as reported previously.29 Reported values of apparent transition temperatures (Tm,app) represent positions of peak maxima determined by visual inspection of the experimental thermograms. As expected based on previous reports for IgG thermal unfolding,21,22,30 the thermograms were not reversible on repeated scanning of the same sample. As such, no thermodynamic analysis of the thermograms is included below. Isothermal Aggregation Rates Isothermal aggregation rates were determined using samples at each formulation condition in Table 1, as a function of initial c2 between 5 and 150 mg/mL. Preliminary tests at a range of incubation temperatures (T) were conducted to identify a single temperature that would satisfy the following criteria: (i) the temperature was at least 10 C below the lowest Tm,app across all conditions in Table 1; (ii) rates were sufficiently slow at the highest concentrations, and sufficiently fast at the lowest concentrations, that is, 1%-10% monomer loss could be quantified for all formulation conditions within practical experimental time scales (hours to months); and (iii) negligible fragmentation occurred within the first 1%-10% monomer loss, as determined by SEC. Based on those initial tests (data not shown), T ¼ 50 C was selected for the detailed comparison of aggregation rates from low to high c2 for all of the conditions in Table 1. Protein solutions at a given formulation condition were pipetted into deactivated borosilicate high-performance liquid chromatography (HPLC) glass vials (Waters, Milford, MA) and hermetically sealed with HPLC screw caps. Vials were incubated in an upright position, and temperature fluctuations in the incubators were verified independently (±0.2 C of the set point) during incubation periods using a digital thermometer. At selected time points, samples were removed and quenched by cooling in an ice-water bath before subsequent analysis or storage. No aggregation was observed to proceed or be reversed after quenching, as evidenced by chromatography, laser scattering, and visual observation; similar to what was found previously with a number of different antibodies.17,21-23,31 All samples (heated and unheated) were stored refrigerated (2-8 C) and used or analyzed within 1 week of preparation. Based on both phenomenological and mechanistic models of protein aggregation,5,27 the rate of monomer loss can be well described by Equation 1 so long as there is not a pronounced lag phase,
dm ¼ kobs ma dt
(Wyatt Technologies, Santa Barbara, CA) and a Wyatt Optilab rEX refractive index (RI) detector unit, were used to quantify monomer concentration and weigh average molecular weight (Mw) for monomer and aggregates. Orthophosphoric acid (0.5 vol/vol %) and 50 mM sodium chloride in distilled, deionized MilliQ water, adjusted to pH 5 with sodium hydroxide was used as the mobile phase with a flow rate of 0.75 mL/min. Alternate mobile phase conditions were also tested to confirm that resolution of the monomer peak was not perturbed by the selected mobile phase (data not shown). A Tosoh TSK-GEL G3000SWxl size exclusion column (Montomeryville, PA) was used for separation of monomer from aggregates and/or fragments. Samples with higher initial c2 were diluted to 1 mg/mL with mobile phase and loaded into individual HPLC vials sealed with preslit screw caps (Waters) for subsequent SEC assay. The injection volume was selected to ensure 100 mg of protein per injection. All samples were visually inspected before loading on SEC. Samples with visible haze or precipitates present were centrifuged at 10,000 g for a minimum of 10 min at room temperature and the collected supernatant was used for chromatographic analysis. Monomer concentration for a given chromatogram was determined based on integrated peak area at 280 nm using ChemStation software (Agilent Technologies) calibrated to external standards. Light scattering signals were collected by an inline MALS detector at each eluting chromatographic slice (1 second average time) and imported into Astra V™ software (Wyatt) for further data processing. An RI detector was used to determine the concentration of eluting c2 for monomer and/or aggregate. Values across a given chromatogram were combined to measure the total or overall weight tot ) of the sample, as described average molecular weight (Mw elsewhere.32
Static Light Scattering (SLS) Batch SLS measurements were carried out using a MALS instrument with a microcuvette accessory from Wyatt. Monomeric samples (without incubation at elevated temperature) were used for SLS measurements and were confirmed to show no dependence on scattering angle. Samples were filtered or centrifuged before measurement, to avoid artifacts from dust and particulates. The excess Rayleigh ratio, Rex 90 at a 90 scattering angle was determined using 31 Equation 2. Pure, filtered toluene was used as a reference solution.
R90 ex ¼ A (1)
where kobs is the observed or effective rate coefficient, m is the fraction of the initial monomer pool that remains at a give time, and a is the effective reaction order. Mathematically, m is defined as the ratio of the monomer concentration (at a given time, t) to the monomer concentration at t ¼ 0. Because all rates were confined to less than 10% monomer loss, the value of a is irrelevant for many kinetic models, as they give essentially zeroth-order kinetics in the “initial rates” regime. kobs follows simply as the negative of the slope of m versus t in that “initial-rate” regime. Uncertainties in kobs values are based on 95% CIs for linear fits of m versus t (see also, Supplementary data). Size Exclusion Chromatography With In-line Multiangle Light Scattering (SEC-MALS) An Agilent 1100 HPLC (Agilent Technologies, Santa Clara, CA) separation module with a variable wavelength detector (VWD), either alone or connected in-line to a DAWN-HELEOS II MALS unit
V V0 n1:983 Vlaser Vdark
(2)
In Equation 2, A is a configuration-specific constant related to scattering geometry of the microcuvette accessory and the toluene calibration standard at a specific laser wavelength (l ¼ 658.9 nm) and temperature (T ¼ 25 C). V and V0 represent voltages for scattering from the protein solution and corresponding buffer solution, respectively. Vlaser and Vdark are the voltages of the incident laser and its dark offset, and n is the RI of solvent. SLS was performed on a series of c2s. For low c2, the excess Rayleigh ratio was regressed with Equation 3 to obtain values of the KirkwoodeBuff integral (G22) for net proteineprotein interactions
M2;app Rex 90 ¼ M2 c2 þ c2 G22 K M2
(3)
app
where M2 represents the apparent molecular weight, M2 is the true molecular weight (146.5 kDa), and K is the instrumental con stant defined by K ¼ 4p2 n2 ðdn=dcÞ2 NA1 l4. dn is the differential dc RI increment of the protein solution, NA represents Avogadro’s
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
number, and the other symbols were defined previously. Values of dn were determined in the presence and absence of sucrose as dc 0.191 mL/g and 0.205 mL/g, respectively. Net proteineprotein interactions can also be represented by the zero-q static structure factor Sq¼0, which is related to G22 by Sq ¼ 0 ¼ 1 þ c2G22.33,34 For conditions when solution nonidealities are sufficiently small to be neglected,34,35 the ratio M2,app/M2 is effectively equal to 1, and therefore, one can replace the term in brackets in Equation 3 with Sq ¼ 0. In the limit that c2 approaches zero, and/or the interactions are sufficiently weak, G22 ¼ 2B22, where B22 is the commonly reported second osmotic virial coefficient.34,35 Dynamic Light Scattering DLS was performed as a function of c2 for each solution condition using a QELS accessory (Wyatt Technologies, Santa Barbara, CA) inline with the MALS instrument described previously and was performed simultaneously on the same samples as those for SLS. The intensity autocorrelation function was nonlinearly regressed in MATLABTM (Mathworks, Natick, MA) to the cumulant expansion given by Equation 4,36
h m i2 g ð2Þ ðtÞ ¼ a þ b exp 2q2 Dc t 1 þ t2 2
(4)
where g(2) is the normalized autocorrelation function, t denotes the delay time, a is the constant for short delay time baseline intensity, and b is an instrument-specific constant. q is the magnitude of the scattering vector, which is related to laser wavelength (l), scattering angle (q), and RI of the solution (n), q ¼ 4pn l1sin(q/2). Dc denotes the translational collective diffusion coefficient. m is the second cumulant and can be related to the polydispersity index (p2) via p2 ¼ m/(Dcq2)2.36,37 In dilute conditions (c2 ~ 1-10 mg/mL), Dc values were linear with respect to c2 and were regressed with Equation 5
Dc ¼ D0 ð1 þ kD c2 Þ
(5)
to give the infinite-dilution or tracer diffusion coefficient (D0) and the so-called interaction parameter (kD).12 Monte Carlo Simulations Radial distribution functions were computed using Monte Carlo (MC) simulations in a cubic box with 100 nm for each side, with periodic boundary conditions in the grand-canonical ensemble.38 Two different structural models for the proteins were used: a spherical model and a coarse-grained (CG)-MAb model. The spherical model treated each protein as a hard sphere of 10.5 nm in diameter. The CG-MAb model corresponded to 6 hard spheres, each of 4.3 nm in diameter, located as follows: one centered on the CH3 domain, one on the CH2 domain, one on each CH1/CL domain, and one on each FV domain. The domains in the Fab were in contact with each other, as were the domains in the Fc. The Fab and Fc regions were connected by linkers that maintained a center-tocenter distance of 2.6 nm between the Fc and each Fab, in keeping with published crystal structures.39 For the calculations needed here, only steric interactions were included (i.e., no van der Waals, hydrogen bonding, hydrophobic, or electrostatic interactions). c2 values of 15 mg/mL and 150 mg/mL were simulated in each case, by appropriate selection of the chemical potential input to the grandcanonical ensemble. Each MC simulation consisted of an initially empty box and was performed in 2 stages. The first stage consisted of 106 MC steps to allow the system to reach the equilibrium
1089
concentration.38,40 The second stage consisted of 105 MC cycles, each one with 105 MC steps. At the end of each MC cycle, a temporary histogram of the positions of the center of mass of each particle was constructed with a discrete value of 0.2 nm. This temporary histogram was normalized to the current concentration and added to a collective histogram. The final radial distribution function was obtained by normalizing the collective histogram at the end of the simulation.38 Results Figure 1 shows DSC thermograms for 1 mg/mL MAb solutions at pH 5.1 (panel A) and pH 6.5 (panel B) for buffer-only conditions (black) and those including 5 wt/wt % sucrose (red) or 100 mM NaCl (blue) in each panel. The same color scheme is used throughout the figures in the following sections and is indicated in Table 1 for easier reference when comparing data sets across figures in the following sections. The conditions without added NaCl show 3 endothermic peaks at pH 5.1 and at least 2 peaks at pH 6.5. For both pH values, conditions with added NaCl show only a single endothermic peak that culminates in a large exotherm. The pH 6 conditions also showed exotherms at the highest temperatures. Such exotherms are indicative of precipitation within this type of DSC configuration,41 and it was confirmed by visual inspection that solutions were cloudy on removal from the DSC for the 100 mM NaCl conditions at both pH 5.1 and pH 6.5 (data not shown). Even for samples that did not show exotherms, the proteins in solution after heating were highly aggregated based on SEC analysis (not shown). Figure 2 shows SLS (Fig. 2a) and DLS (Fig. 2b) as a function of MAb concentration at low c2 for pH 5.1 (main panels) and pH 6.5 (insets) with buffer-only conditions (black) as well as those including 5 wt/wt % sucrose (red) or 100 mM NaCl (blue) in each case. Symbols are experimental values for the excess Rayleigh ratio (panel A) or the collective diffusion coefficient normalized by the tracer diffusion coefficient (panel B). Values for the tracer diffusion coefficient (D0) are provided in Table S1 of Supplementary data. Solid curves are fits to Equation 3 in panel A or Equation 5 in panel B. Qualitatively, the SLS results for both pH values indicate net repulsions between proteins for the buffer-only and 5 wt/wt % sucrose conditions, based on the downward curvature in Figure 2a. Conversely, the SLS results show slight upward curvature for the 100-mM NaCl conditions for both pH values in Figure 2a. Interestingly, the DLS results show a positive slope for all conditions reported in Figure 2b. If only DLS data were available, one could erroneously conclude that proteineprotein interactions are net repulsive at all solution conditions reported in Figure 2. Figure 3a shows the values of B22 from the fits in Figure 2a, assuming the interactions are sufficiently weak that B22 ¼ ½ G22 holds. This identity holds only when jc2G22j
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmaceutical Biotechnology
Relating ProteineProtein Interactions and Aggregation Rates From Low to High Concentrations Ranendu Ghosh 1, Cesar Calero-Rubio 1, Atul Saluja 2, Christopher J. Roberts 1, * 1 2
Department of Chemical & Biomolecular Engineering, University of Delaware, Newark, Delaware 19716 Department of Drug Product Science and Technology, Bristol-Myers Squibb, New Brunswick, New Jersey 08901
a r t i c l e i n f o
a b s t r a c t
Article history: Received 29 October 2015 Revised 22 December 2015 Accepted 6 January 2016
At low protein concentrations (c2), non-native protein aggregation rates are known to be sensitive to changes in conformational stability and “weak” or “colloidal” proteineprotein interactions. Protein eprotein interactions are also known to be strong functions of c2. In the present work, proteineprotein interactions and rates of aggregation were quantified systematically for a monoclonal antibody (MAb) across a broad range of c2 at pH 5.1 and 6.5, with or without 5 wt/wt % sucrose or 100 mM NaCl present. Aggregation rates were determined from initial-rate analysis with size-exclusion chromatography, and interactions were quantified with static and dynamic laser light scattering. A number of hypotheses were tested regarding whether changes in proteineprotein interactions can be predictive of changes in aggregation rates versus c2. Hypotheses were based on (i) changes in thermodynamic activity; (ii) statistical mechanical fluctuation theory; and (iii) surface-contact probabilities. Arguments based on (i) and (ii) were qualitatively inconsistent with experimental rates and scattering. Hypothesis (iii) was reasonably successful and resulted in a semiquantitative correlation between rates and proteineprotein interactions across almost 2 orders of magnitude in c2. However, (iii) requires one to assume that the concentrationdependent proteineprotein KirkwoodeBuff integral is a reasonable surrogate for contact probabilities. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: light scattering (dynamic) biophysical models protein aggregation protein formulation biotechnology
Introduction Proteins are inherently labile molecules that have marginally stable structures and can be degraded and/or inactivated readily under reasonably mild in vitro and in vivo conditions. Non-native protein aggregation (hereafter referred to simply as aggregation) denotes the processes by which natively folded proteins form aggregates via interprotein contacts that involve secondary and/or tertiary structures that are measurably perturbed from the native or folded monomer state.1-3 In this context, aggregates are typically net irreversible under the conditions that they form, and therefore the rates or kinetics of aggregation are a key quantity in determining product quality and shelf life. As aggregation is under kinetic control, the mechanism(s) of aggregation ultimately dictate This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.01.004. Conflict of interest: The authors declare no conflicts of interest for any aspects of the work reported here. Present address for Ranendu Ghosh: Formulation Development Laboratory, Biocon Research Ltd., Bangalore, Karnataka, India, 560099. * Correspondence to: Christopher J. Roberts (Telephone: 302-831-0838; Fax: 302831-1048). E-mail address: [email protected] (C.J. Roberts).
the observed aggregation rates. Aggregation of therapeutic proteins is a long-standing issue during drug product development and commercial manufacturing3,4 and is also problematic to control during protein expression in vivo.2 Studies at relatively low protein concentration (c2) have led to a reasonably common viewpoint, in which aggregation rates or kinetics are mediated by at least 3 main factors:5-7 (1) protein conformational stability or conformational changes that expose stretches of hydrophobic amino acids or so-called aggregation “hot spots”; (2) “weak” or “colloidal” attractions (repulsions) that facilitate (inhibit) reversible protein selfassociation; and (3) intrinsic aggregation propensity, that is, strong interactions between exposed sequence “hot spots” that create net irreversible contacts between proteins, often via formation of interprotein beta sheet structures. Environmental factors that greatly alter conformational stability include temperature, pressure, solution composition, and adsorption to solideliquid interfaces.3,8,9 Protein conformational stability is theoretically predicted to be sensitive to c2, but direct experimental evidence for the magnitude of the effect is very limited.10,11 Similarly, factors that have been found to greatly alter proteineprotein interactions include c2 and solution conditions such as pH and the concentration of commonly employed cosolutes, for example, salts, amino acids, and polyhydroxy
http://dx.doi.org/10.1016/j.xphs.2016.01.004 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
compounds such as sucrose.12,13 The intrinsic aggregation propensity of a protein when it is unfolded is putatively dictated by its primary sequence and is not an easy variable to control once a given protein candidate has been selected for product development and testing in patients.7,14 Accurate prediction of aggregation rates is a long-standing challenge in the biotechnology industry. For low c2 conditions, phenomenological and mechanistic models have been developed in an attempt to provide quantitative prediction of aggregation rates for natively unfolded polypeptides under physiological conditions14,15 and for therapeutic proteins under typical formulation conditions.16-19 In all cases, the models either interpolate or extrapolate rate data from one experimental condition or protein sequence to another. These models have focused primarily on the effects of conformational stability and the inherent aggregation propensity of “hot spot” sequences. Weak or colloidal interactions have been shown to correlate, at least qualitatively, with aggregation rates at low c2.12,20 However, in a number of cases, it is an inverse correlation because changes in conformational stability outweigh changes in proteineprotein interactions as one alters formulation conditions.21-23 In contrast, relatively little has been done to quantitatively or semiquantitatively connect changes in proteineprotein interactions with aggregation rates at high c2.24-26 In part, this may be due to the relatively small amount of published data for aggregation rates across a broad range of concentration as most studies either focus on only high-c2 or low-c2 data or provide only a small number of different c2 conditions.16-19,21,23,26 The present study focuses on a systematic comparison of proteineprotein interactions and accelerated aggregation rates (elevated temperature, 50 C) for a monoclonal antibody (MAb) of the IgG1 subclass, over a broad range of c2 (~1-102 g/L) at pH 5.1 and 6.5, with and without either sucrose (5 wt/wt %) or NaCl (100 mM) present. Aggregation rates are based on loss of monomer, as quantified ex situ with size-exclusion chromatography (SEC). Proteineprotein interactions are quantified in situ with static and dynamic light scattering (DLS) for the same c2 as the measured aggregation rates. To the extent possible within experimental limits, changes in conformational stability are inferred from differential scanning calorimetry at low c2. Concentration-dependent aggregation rates are used to scrutinize general mass-action arguments that have been developed primarily for low-c2 conditions and that encompass a family of models for aggregation kinetics.27 Based on previous work and new arguments presented here, a number of hypotheses are tested with respect to both qualitative and quantitative agreements or “predictability” for experimental aggregation rates in terms of measureable proteineprotein interactions. These include changes in thermodynamic activity or chemical potential, local concentration fluctuations, and the probability of protein surfaces coming into contact. The results highlight a number of limitations for existing mass action models of protein aggregation, as well as hypotheses for how to quantitatively link proteineprotein interactions to aggregation rates at high c2. The proteineprotein KirkwoodeBuff integral (G22) is offered as a surrogate for proteineprotein contact probabilities that are intuitively expected to influence aggregation rates at high concentration. It is found that a reasonable correlation exists between kobs and G22 if one has a reference data set against which to normalize. Finally, outstanding challenges are discussed for predicting and/or measuring the thermodynamic properties of proteins (e.g., native protein chemical potential) and the conformational stability of proteins at elevated concentrations. Overcoming these challenges will require significant advancements in terms of experimental capabilities and modeling approaches.
1087
Materials and Methods Sample Preparation Purified IgG1 was provided by Bristol-Myers Squibb at a starting concentration of ~54 mg/mL. Stock solutions were dialyzed against a given buffer condition (Spectra/Por 7 tubing, 10 kDa MWCO; Spectrum Laboratories, Santa Clara, CA) using four 12 h buffer exchanges of 500 mL each, in refrigerated conditions. Buffer-only conditions were either 10 mM sodium acetate at pH 4.0 or 10 mM histidine HCl at pH 6.35. Glacial acetic acid (Fisher Scientific) and histidine HCl (Sigma) were used for preparation of dialysis buffers. After dialysis, protein solutions were concentrated using Amicon (Millipore, Billerica, MA) ultracentrifugation tubes with a molecular weight cutoff of 10 kDa. The concentrate was collected at different stages of centrifugation so as to monitor solution pH and c2 until the desired pH (pH 5.1 ± 0.1 for acetate buffer and pH 6.5 ± 0.1 for histidine buffer) and concentration (>170 mg/mL) were achieved. Supplementary data include illustrative results from repeated sample preparations. As expected, the solution pH shifted during the protein concentrating step because of self-buffering by the protein.28 The lower starting pH values (before the concentration step) were selected based on trial runs, so as to achieve the desired final pH values in the concentrated stocks. Solutions at lower c2 were subsequently prepared by dilution of the concentrated stock at either pH 5.1 or pH 6.5, using the corresponding buffer solution and independently confirmed to maintain the desired pH after dilution. Samples with either sucrose or NaCl were prepared by gravimetrically diluting concentrated stock solutions to lower concentration. This was done using the appropriate ratios of stock solution, buffer-only solution (pH 5.1 or pH 6.5), and buffer with either 500 mM NaCl or 25% (wt/wt) sucrose. This was done to achieve the desired c2 and a final NaCl concentration of 100 mM or a final sucrose concentration of 5% (wt/wt). Table 1 lists the set of final solution conditions. Final pH was verified for all samples, and all solutions were filtered using 0.22 mm low protein binding filters (Millipore). Final concentrations were confirmed via UV absorbance (Agilent Technologies, Santa Clara, CA) at 280 nm, with an extinction coefficient of 1.54 mL mg1 cm1.
Differential Scanning Calorimetry Differential scanning calorimetry (DSC) experiments were performed using a VP-DSC instrument (Microcal, Northhampton, MA) for antibody solutions with 1 mg/mL protein at each of the different solution conditions in Table 1. Multiple bufferebuffer scans were performed to obtain baseline values and establish thermal history on the instrument immediately before sample scans. Thermal scans were performed over 20 C-90 C with a scan rate of 60 C/h. The average of the buffer scans was subtracted from the subsequent Table 1 Summary of Formulation Conditions Formulation Description pH pH pH pH pH pH a b c
5, buffer onlyb 5, buffer þ NaCl 5, buffer þ sucrose 6.5, buffer onlyc 6.5, buffer þ NaCl 6.5, buffer þ sucrose
Additional Excipient (Final Concentration)
Colors for Curves/ Symbols in Figuresa
none NaCl (100 mM) Sucrose (5% wt/wt) none NaCl (100 mM) Sucrose (5% wt/wt)
Black Blue Red Black Blue Red
Figures 5, 6, and 8 use closed symbols for pH 5, and open symbols for pH 6.5. 10-mM sodium acetate, pH 5.0 ± 0.1. 10-mM histidine HCl, pH 6.5 ± 0.1.
1088
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
protein scan, and the raw DSC data were converted to partial specific heat capacity values as reported previously.29 Reported values of apparent transition temperatures (Tm,app) represent positions of peak maxima determined by visual inspection of the experimental thermograms. As expected based on previous reports for IgG thermal unfolding,21,22,30 the thermograms were not reversible on repeated scanning of the same sample. As such, no thermodynamic analysis of the thermograms is included below. Isothermal Aggregation Rates Isothermal aggregation rates were determined using samples at each formulation condition in Table 1, as a function of initial c2 between 5 and 150 mg/mL. Preliminary tests at a range of incubation temperatures (T) were conducted to identify a single temperature that would satisfy the following criteria: (i) the temperature was at least 10 C below the lowest Tm,app across all conditions in Table 1; (ii) rates were sufficiently slow at the highest concentrations, and sufficiently fast at the lowest concentrations, that is, 1%-10% monomer loss could be quantified for all formulation conditions within practical experimental time scales (hours to months); and (iii) negligible fragmentation occurred within the first 1%-10% monomer loss, as determined by SEC. Based on those initial tests (data not shown), T ¼ 50 C was selected for the detailed comparison of aggregation rates from low to high c2 for all of the conditions in Table 1. Protein solutions at a given formulation condition were pipetted into deactivated borosilicate high-performance liquid chromatography (HPLC) glass vials (Waters, Milford, MA) and hermetically sealed with HPLC screw caps. Vials were incubated in an upright position, and temperature fluctuations in the incubators were verified independently (±0.2 C of the set point) during incubation periods using a digital thermometer. At selected time points, samples were removed and quenched by cooling in an ice-water bath before subsequent analysis or storage. No aggregation was observed to proceed or be reversed after quenching, as evidenced by chromatography, laser scattering, and visual observation; similar to what was found previously with a number of different antibodies.17,21-23,31 All samples (heated and unheated) were stored refrigerated (2-8 C) and used or analyzed within 1 week of preparation. Based on both phenomenological and mechanistic models of protein aggregation,5,27 the rate of monomer loss can be well described by Equation 1 so long as there is not a pronounced lag phase,
dm ¼ kobs ma dt
(Wyatt Technologies, Santa Barbara, CA) and a Wyatt Optilab rEX refractive index (RI) detector unit, were used to quantify monomer concentration and weigh average molecular weight (Mw) for monomer and aggregates. Orthophosphoric acid (0.5 vol/vol %) and 50 mM sodium chloride in distilled, deionized MilliQ water, adjusted to pH 5 with sodium hydroxide was used as the mobile phase with a flow rate of 0.75 mL/min. Alternate mobile phase conditions were also tested to confirm that resolution of the monomer peak was not perturbed by the selected mobile phase (data not shown). A Tosoh TSK-GEL G3000SWxl size exclusion column (Montomeryville, PA) was used for separation of monomer from aggregates and/or fragments. Samples with higher initial c2 were diluted to 1 mg/mL with mobile phase and loaded into individual HPLC vials sealed with preslit screw caps (Waters) for subsequent SEC assay. The injection volume was selected to ensure 100 mg of protein per injection. All samples were visually inspected before loading on SEC. Samples with visible haze or precipitates present were centrifuged at 10,000 g for a minimum of 10 min at room temperature and the collected supernatant was used for chromatographic analysis. Monomer concentration for a given chromatogram was determined based on integrated peak area at 280 nm using ChemStation software (Agilent Technologies) calibrated to external standards. Light scattering signals were collected by an inline MALS detector at each eluting chromatographic slice (1 second average time) and imported into Astra V™ software (Wyatt) for further data processing. An RI detector was used to determine the concentration of eluting c2 for monomer and/or aggregate. Values across a given chromatogram were combined to measure the total or overall weight tot ) of the sample, as described average molecular weight (Mw elsewhere.32
Static Light Scattering (SLS) Batch SLS measurements were carried out using a MALS instrument with a microcuvette accessory from Wyatt. Monomeric samples (without incubation at elevated temperature) were used for SLS measurements and were confirmed to show no dependence on scattering angle. Samples were filtered or centrifuged before measurement, to avoid artifacts from dust and particulates. The excess Rayleigh ratio, Rex 90 at a 90 scattering angle was determined using 31 Equation 2. Pure, filtered toluene was used as a reference solution.
R90 ex ¼ A (1)
where kobs is the observed or effective rate coefficient, m is the fraction of the initial monomer pool that remains at a give time, and a is the effective reaction order. Mathematically, m is defined as the ratio of the monomer concentration (at a given time, t) to the monomer concentration at t ¼ 0. Because all rates were confined to less than 10% monomer loss, the value of a is irrelevant for many kinetic models, as they give essentially zeroth-order kinetics in the “initial rates” regime. kobs follows simply as the negative of the slope of m versus t in that “initial-rate” regime. Uncertainties in kobs values are based on 95% CIs for linear fits of m versus t (see also, Supplementary data). Size Exclusion Chromatography With In-line Multiangle Light Scattering (SEC-MALS) An Agilent 1100 HPLC (Agilent Technologies, Santa Clara, CA) separation module with a variable wavelength detector (VWD), either alone or connected in-line to a DAWN-HELEOS II MALS unit
V V0 n1:983 Vlaser Vdark
(2)
In Equation 2, A is a configuration-specific constant related to scattering geometry of the microcuvette accessory and the toluene calibration standard at a specific laser wavelength (l ¼ 658.9 nm) and temperature (T ¼ 25 C). V and V0 represent voltages for scattering from the protein solution and corresponding buffer solution, respectively. Vlaser and Vdark are the voltages of the incident laser and its dark offset, and n is the RI of solvent. SLS was performed on a series of c2s. For low c2, the excess Rayleigh ratio was regressed with Equation 3 to obtain values of the KirkwoodeBuff integral (G22) for net proteineprotein interactions
M2;app Rex 90 ¼ M2 c2 þ c2 G22 K M2
(3)
app
where M2 represents the apparent molecular weight, M2 is the true molecular weight (146.5 kDa), and K is the instrumental con stant defined by K ¼ 4p2 n2 ðdn=dcÞ2 NA1 l4. dn is the differential dc RI increment of the protein solution, NA represents Avogadro’s
R. Ghosh et al. / Journal of Pharmaceutical Sciences 105 (2016) 1086e1096
number, and the other symbols were defined previously. Values of dn were determined in the presence and absence of sucrose as dc 0.191 mL/g and 0.205 mL/g, respectively. Net proteineprotein interactions can also be represented by the zero-q static structure factor Sq¼0, which is related to G22 by Sq ¼ 0 ¼ 1 þ c2G22.33,34 For conditions when solution nonidealities are sufficiently small to be neglected,34,35 the ratio M2,app/M2 is effectively equal to 1, and therefore, one can replace the term in brackets in Equation 3 with Sq ¼ 0. In the limit that c2 approaches zero, and/or the interactions are sufficiently weak, G22 ¼ 2B22, where B22 is the commonly reported second osmotic virial coefficient.34,35 Dynamic Light Scattering DLS was performed as a function of c2 for each solution condition using a QELS accessory (Wyatt Technologies, Santa Barbara, CA) inline with the MALS instrument described previously and was performed simultaneously on the same samples as those for SLS. The intensity autocorrelation function was nonlinearly regressed in MATLABTM (Mathworks, Natick, MA) to the cumulant expansion given by Equation 4,36
h m i2 g ð2Þ ðtÞ ¼ a þ b exp 2q2 Dc t 1 þ t2 2
(4)
where g(2) is the normalized autocorrelation function, t denotes the delay time, a is the constant for short delay time baseline intensity, and b is an instrument-specific constant. q is the magnitude of the scattering vector, which is related to laser wavelength (l), scattering angle (q), and RI of the solution (n), q ¼ 4pn l1sin(q/2). Dc denotes the translational collective diffusion coefficient. m is the second cumulant and can be related to the polydispersity index (p2) via p2 ¼ m/(Dcq2)2.36,37 In dilute conditions (c2 ~ 1-10 mg/mL), Dc values were linear with respect to c2 and were regressed with Equation 5
Dc ¼ D0 ð1 þ kD c2 Þ
(5)
to give the infinite-dilution or tracer diffusion coefficient (D0) and the so-called interaction parameter (kD).12 Monte Carlo Simulations Radial distribution functions were computed using Monte Carlo (MC) simulations in a cubic box with 100 nm for each side, with periodic boundary conditions in the grand-canonical ensemble.38 Two different structural models for the proteins were used: a spherical model and a coarse-grained (CG)-MAb model. The spherical model treated each protein as a hard sphere of 10.5 nm in diameter. The CG-MAb model corresponded to 6 hard spheres, each of 4.3 nm in diameter, located as follows: one centered on the CH3 domain, one on the CH2 domain, one on each CH1/CL domain, and one on each FV domain. The domains in the Fab were in contact with each other, as were the domains in the Fc. The Fab and Fc regions were connected by linkers that maintained a center-tocenter distance of 2.6 nm between the Fc and each Fab, in keeping with published crystal structures.39 For the calculations needed here, only steric interactions were included (i.e., no van der Waals, hydrogen bonding, hydrophobic, or electrostatic interactions). c2 values of 15 mg/mL and 150 mg/mL were simulated in each case, by appropriate selection of the chemical potential input to the grandcanonical ensemble. Each MC simulation consisted of an initially empty box and was performed in 2 stages. The first stage consisted of 106 MC steps to allow the system to reach the equilibrium
1089
concentration.38,40 The second stage consisted of 105 MC cycles, each one with 105 MC steps. At the end of each MC cycle, a temporary histogram of the positions of the center of mass of each particle was constructed with a discrete value of 0.2 nm. This temporary histogram was normalized to the current concentration and added to a collective histogram. The final radial distribution function was obtained by normalizing the collective histogram at the end of the simulation.38 Results Figure 1 shows DSC thermograms for 1 mg/mL MAb solutions at pH 5.1 (panel A) and pH 6.5 (panel B) for buffer-only conditions (black) and those including 5 wt/wt % sucrose (red) or 100 mM NaCl (blue) in each panel. The same color scheme is used throughout the figures in the following sections and is indicated in Table 1 for easier reference when comparing data sets across figures in the following sections. The conditions without added NaCl show 3 endothermic peaks at pH 5.1 and at least 2 peaks at pH 6.5. For both pH values, conditions with added NaCl show only a single endothermic peak that culminates in a large exotherm. The pH 6 conditions also showed exotherms at the highest temperatures. Such exotherms are indicative of precipitation within this type of DSC configuration,41 and it was confirmed by visual inspection that solutions were cloudy on removal from the DSC for the 100 mM NaCl conditions at both pH 5.1 and pH 6.5 (data not shown). Even for samples that did not show exotherms, the proteins in solution after heating were highly aggregated based on SEC analysis (not shown). Figure 2 shows SLS (Fig. 2a) and DLS (Fig. 2b) as a function of MAb concentration at low c2 for pH 5.1 (main panels) and pH 6.5 (insets) with buffer-only conditions (black) as well as those including 5 wt/wt % sucrose (red) or 100 mM NaCl (blue) in each case. Symbols are experimental values for the excess Rayleigh ratio (panel A) or the collective diffusion coefficient normalized by the tracer diffusion coefficient (panel B). Values for the tracer diffusion coefficient (D0) are provided in Table S1 of Supplementary data. Solid curves are fits to Equation 3 in panel A or Equation 5 in panel B. Qualitatively, the SLS results for both pH values indicate net repulsions between proteins for the buffer-only and 5 wt/wt % sucrose conditions, based on the downward curvature in Figure 2a. Conversely, the SLS results show slight upward curvature for the 100-mM NaCl conditions for both pH values in Figure 2a. Interestingly, the DLS results show a positive slope for all conditions reported in Figure 2b. If only DLS data were available, one could erroneously conclude that proteineprotein interactions are net repulsive at all solution conditions reported in Figure 2. Figure 3a shows the values of B22 from the fits in Figure 2a, assuming the interactions are sufficiently weak that B22 ¼ ½ G22 holds. This identity holds only when jc2G22j
