Relationship between Native-State Solubility and Non-Native

Article pubs.acs.org/molecularpharmaceutics

Relationship between Native-State Solubility and Non-Native Aggregation of Recombinant Human Granulocyte Colony Stimulating Factor: Practical Implications for Protein Therapeutic Development Douglas D. Banks,* Jun Zhang, and Christine C. Siska Department of Process and Product Development, Amgen, Inc., 1201 Amgen Court West, Seattle, Washington 98119-3105, United States S Supporting Information *

ABSTRACT: Prescreening methods are needed in the biotechnology industry for rapid selection of protein therapeutic candidates and formulations of low aggregation propensity. In recent reports solubility measurements have shown promise as one such method, although the connection between protein solubility and non-native aggregation is not well understood. In the present investigation, recombinant human granulocyte colony stimulating factor (rhGCSF) was used to explore this relationship since it was previously shown to rapidly undergo non-native aggregation/precipitation under physiological conditions in a reaction attenuated by the addition of sucrose [Krishnan, S.; et al. Biochemistry 2002, 41, 6422− 6431]. Strong correlations were found between rhGCSF non-native aggregation and both solubility and thermal stability as a function of sucrose concentration. We believe these results make sense in the context of an rhGCSF aggregation mechanism where loss of monomer to insoluble aggregate is limited by association to an observable dimer from a less soluble (and aggregation competent) intermediate species that exists in a temperature sensitive pre-equilibrium with the native monomer. Both solubility and measures of conformational stability report on the position of this equilibrium and therefore the concentration of reactive intermediate. Interestingly, aggregation also correlated with rhGCSF solubility as a function of salting-in concentrations of phosphate since both are dependent on the colloidal stability of the reactive intermediate but not with conformational stability. In lieu of a complete understanding of the aggregation processes that limit protein therapeutic shelf life, these results highlight the potential of using simple solubility measurements as an additional tool in the biotechnology prescreening repertoire. KEYWORDS: protein aggregation, solubility, conformational stability, kinetics, hydrogen/deuterium exchange

■

INTRODUCTION Protein biologics are large complex molecules prone to numerous modes of degradation, that in addition to limiting their efficacy could in principle promote an immunogenic reaction.1,2 Chief among the instabilities that remain an obstacle for the biotechnology industry in maintaining a reasonable protein drug product shelf life is aggregation. To meet this challenge three strategies are often employed. First, protein therapeutics are stored under conditions where their rates of aggregation are lowered; typically, this means storage at refrigerated temperatures. Second, therapeutic proteins may be modified to reduce their intrinsic rates of aggregation; this could be the engineering of their sequence or incorporation of a post-translational modification. Third, the solvent conditions that the protein is formulated in may be optimized to limit aggregation. To realize the full potential of these approaches, i.e., to rationally develop protein based therapeutics, a more complete understanding of the protein aggregation process is needed; a challenge that remains a long-term goal for both industry and academia. © XXXX American Chemical Society

In the meantime, protein biologic development, particularly formulation optimization, remains somewhat of a semiempirical search where predictive aggregation prescreening methods are needed to select which of the many therapeutic candidates or formulations to advance to the low-throughput and resource expensive longer term studies. One such prescreening assay that has gained considerable traction in recent years owing to its potential for automation is the measurement of thermal stability. The rational for using this approach is based on the extended Lumry−Eyring model of non-native aggregation (Scheme 1) where a prerequisite for aggregation, in this case dimerization, is the partial unfolding or structural perturbation of the native-state (N) to an aggregation competent intermediate species (N*).3−5 Received: February 26, 2014 Revised: August 11, 2014 Accepted: August 21, 2014

A

dx.doi.org/10.1021/mp500165j | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

of this aggregate has been shown in past reports to be predominately composed of intermolecular β-sheet;12−14 although, to our knowledge it has not been reported whether it is organized into amyloid filaments or remains amorphous. In the present study, we tested the temperature and concentration dependence of the formulation cosolutes sucrose and potassium phosphate (KPi) on the rhGCSF native-state solubility and whether it, thermal stability, or ΔGunf (for KPi formulations) were predictors of the non-native rhGCSF aggregation behavior at 37 °C. For the use in this report, we define native-state solubility (or simply solubility) as the saturating concentration of monomeric protein in dynamic equilibrium with an amorphous solid phase, as opposed to the growth of irreversible non-native aggregate that eventually precipitates. Sucrose was selected as a cosolute for these comparisons since it has been previously shown to increase rhGCSF conformational stability and limit non-native aggregation.12 Phosphate was chosen because stock solutions of this salting-out kosmotrope can be prepared at neutral pH and at higher concentrations than other common stabilizing salts from the Hofmeister series such as ammonium and sodium sulfate, respectively.20 Moreover, lower concentrations of this salt, like any other, may increase protein solubility due to nonspecific Debye−Hückel shielding of favorable electrostatic interactions between rhGCSF monomers, an effect not possible with uncharged macromolecular crowding agents such as polyethylene glycol (PEG).21 Strong correlations between solubility and aggregation were observed for both formulations; but, only for the sucrose containing formulations was a decrease in aggregation with increasing thermal stability observed. These results, together with the temperature dependence of solubility and conformational stability, revealed additional insight into the mechanism of rhGCSF aggregation and highlight circumstances when the use of a solubility assay may be predictive for proteins that aggregate through non-native pathways.

Scheme 1

In accordance with this model, candidates with high thermal stability are selected for further development with the idea that increasing conformational stability (ΔGunf, the difference in free-energy between the native and unfolded ground states) shifts the equilibrium away from or depopulates the aggregation prone N* state. This link between conformational stability and aggregation propensity has become so entrenched within industry that the term protein “stability” is often used interchangeably to mean both. In practice, however, selecting candidate molecules or formulations on the basis of thermal stability has yielded mixed results.6−8 This may be due to limited unfolding reversibility and association rather than partial unfolding being rate limiting in Scheme 1 or protein association even occurring prior to changes in conformation. Moreover, there is no requirement that a shift in melting temperature (Tm) necessitates a similar change in ΔGunf at temperatures where the protein therapeutic is expected to be stored (e.g., 2− 8 °C).9 In two interesting reports, protein solubility and not conformational stability was shown to be a strong determinate of aggregation/fibrillation.10,11 Motivated by these studies, we recently used a simple solubility screen to rank the effectiveness of formulation cosolutes in solubilizing a handful of monoclonal antibodies under conformationally stabilizing salting-out conditions and found that native-state solubility correlated well to the amounts of antibody dimer formed after nearly a year of static storage at 4 °C.7 On the basis of these results, the primary aim of the current investigation is to test the utility of a similar native-state solubility prescreen extrapolated to mildly destabilizing conditions for proteins that partially unfold prior to undergoing non-native aggregation as depicted in Scheme 1. This is of particular interest since protein biologics are routinely exposed to numerous stresses during production, manufacture, and shipping that may perturb the native-state structure. These include pH extremes during purification and viral inactivation steps and increased contact with the denaturing air−liquid interface during filtration and transportation. The secondary goal of this report is to gain additional insight into the protein aggregation process in general and make progress toward the longer term objective of rational protein/formulation design. Toward these ends, recombinant human granulocyte colony stimulating factor (rhGCSF) was chosen as a model protein since it is perhaps one of the best studied protein therapeutics thought to undergo non-native aggregation following the extended Lumry−Eyring model.12−15 Granulocyte colony stimulating factor is an 18.8 kDa four-helical bundle glycosylated cytokine that induces neutrophil maturation.16,17 Recombinant human GCSF, expressed in Escherichia coli (E. coli), lacks glycosylation and has reached prominent therapeutic importance in the treatment of neutropenia under the trade name Neupogen (Amgen, Inc.). As a therapeutic, rhGCSF is formulated at pH 4.0 and stored at 2−8 °C where it is well folded and shows little tendency toward aggregation.18 At physiological temperature and pH, however, rhGCSF aggregates within days to form insoluble nonreversible aggregate that is to some extent covalently cross-linked due to the presence of a free cysteine thiol at position 17.14,19 The secondary structure

■

EXPERIMENTAL PROCEDURES Materials. Pharmaceutical quality rhGCSF was expressed and purified at Amgen, Inc., from E. coli as described previously22 and stored in 0.1 mM HCl (pH 4.0) at 2−8 °C. Protein concentrations were measured at 280 nm using a theoretical extinction coefficient of 15,720 M−1·cm−1 determined by the methods of Pace et al.23 High purity potassium phosphate (monobasic monohydrate and dibasic dihydrate) and sucrose were purchased from Sigma-Aldrich (St. Louis, MO). Ultrapure urea was from MP Biomedicals Inc. (Solon, OH) and was prepared fresh for each experiment; concentrations were determined by refractive index.24 Solubility Determination. Initial salting-out experiments were performed by preparing 0.5 mL samples at 2 mg/mL rhGCSF in the presence of increasing concentrations of KPi (pH 7.1) added from a 3 M stock solution. Individual saltingout curves were incubated at temperatures ranging from 4 to 37 °C and reached equilibrium rapidly as indicated by the near superimposition of curves stored for 2 and 28 h at 25 °C in Figure 1A. Note, however, that protein in the soluble phase was lost to non-native aggregation/precipitation after 28 h incubation at 37 °C (Figure 1A). For this reason, salting-out experiments conducted at room temperature and greater were equilibrated between 2 and 3 h and the salting-out curves performed at and below 15 °C were equilibrated for 48 h. Following equilibration, samples stored at 15 °C and greater B



Article

Figure 1. continued the data to a single exponential equation: f(x) = b·exp(mx). The temperature and sucrose concentration dependence of m are plotted in the insets of panels B and C, respectively, and fit to a line; note the similar slopes. Conditions for panels B and C are 30 mM KPi (pH 7.1).

were centrifuged at 14,000 rpm for 10 min at their corresponding salting-out equilibration temperatures. For curves stored at colder temperatures, the solutions with higher concentrations of KPi were denser than the precipitated protein phase, and syringe filtration was necessary. Concentrations of soluble rhGCSF remaining in the supernatant or filtrate were determined by absorbance at 280 nm. Below roughly 200 mM KPi, the solubility of rhGCSF decreased and its solubility limit could be determined in the absence of salting-out concentrations of KPi. This was achieved by preparing 0.1 mL samples in 30 mM KPi (pH 7.1) at final concentrations from 75 to 200 mg/mL rhGCSF, spiked-in from protein stock solutions ranging from 100 to 215 mg/mL in 0.1 mM HCl (pH 4.0). From this base formulation the solubility dependence on temperatures between 20 to 32 °C and in the presence of increasing concentrations of sucrose and KPi (added from 2 and 3 M stock solutions prepared in water, respectively) were determined as described for the salting-out experiments. Samples were allowed to equilibrate in these conditions between 2 and 3 h as deemed sufficient from solubility time-course experiments (Figure S1 in the Supporting Information). Near the salting-in/out transition region at roughly 0.2 M KPi, solubility limits could not be reliably determined due to the difficulty in preparing a pH 4.0 stock protein solution at high enough concentration. Because of the large protein demand required to measure solubility below 0.2 M KPi and limited protein supply, the decision was made to collect numerous data points over a broad temperatureformulation space, rather than collecting fewer points in replicate. Aggregation Assay. Sucrose formulations were prepared at a final rhGCSF concentration of 3 mg/mL buffered by 30 mM KPi (pH 7.1) by addition from a 2 M stock sucrose solution to final concentrations ranging from 0.05 to 0.6 M. Potassium phosphate formulations were prepared in a similar manner from a 3 M KPi (pH 7.1) stock to final concentrations ranging from 0.03 to 0.5 M. All formulations were filter sterilized in a laminar flow hood and 0.2 mL aliquots of each were dispensed into sterile 0.5 mL screw-cap microcentrifuge tubes. All formulations were then stored for 4 days at 37 °C. At given time-points during storage a complete set of all formulations were pulled and centrifuged at 14,000 rpm at 10 °C to pellet insoluble aggregate. Following centrifugation 20 μL of each of the supernatants were analyzed by size exclusion chromatography. Size-Exclusion Chromatography. Size exclusion chromatography (SEC) was performed on an Agilent 1100 series quaternary pump liquid chromatography system (Agilent Technologies, Santa Clara, CA) equipped with a single TosoHaas TSK-gel SW2000xl column. Protein was eluted isocratically at a flow rate of 0.5 mL/min with mobile phase consisting of 100 mM sodium phosphate (pH 6.9) and 250 mM sodium chloride. Absorbance was monitored at 280, 230, and 215 nm.

Figure 1. (A) Representative KPi (pH 7.1) salting-out of rhGCSF at 4 (⧫), 15 (▼), 25 (▲), 31 (■), and 37 °C (●) following 2 h incubation. For comparison, samples incubated for 28 h are shown as open symbols; the top of the figure is plotted on a log scale. (B) Sucrose concentration dependence of rhGCSF solubility at 24 (●), 25 (◊), 26 (▽), 28 (△), 30 (□), and 32 °C (○). (C) Temperature dependence of rhGCSF solubility at 0 (○), 0.05 (□), 0.1 (△), 0.15 (▽), 0.2 (◊), 0.25 (●), and 0.3 M sucrose (■). Solid lines in all panels are the fit of C



Article

μL of 0.6 mg/mL rhGCSF formulated in 10 mM sodium phosphate (pH 7.0) and 150 mM NaCl with 24 μL of the same buffer solution prepared in heavy water (D2O, pD 7.0). Following incubation of rhGCSF for 10 s, 30 s, 1 min, 10 min, 1 h, and 4 h at both 37 and 4 °C, the exchange reactions were quenched by mixing 1:1 with a solution consisting of 200 mM sodium phosphate, 4 M guanidinium hydrochloride, and 0.5 M TCEP (pH 2.4). The quenched protein mixtures were then digested online with a custom packed pepsin column described elsewhere.26 Digested peptides were captured on a 2 mm × 1 cm C8 trap column (Agilent Technologies, Santa Clara, CA) and desalted with a 3 min flow. Peptides were then separated on a 2.1 mm × 5 cm C18 column (1.9 μm Hypersil Gold; Thermo Fisher Scientific, Waltham, MA) with a 5 min linear gradient from 4 to 40% acetonitrile in 0.1% formic acid. Protein digestion and peptide separation were carried out in an ice− water bath to reduce back exchange. This H/DX platform has an estimated average deuterium recovery of 70%.27 Mass spectrometric analyses were performed in duplicate on a hybrid Orbitrap Elite mass spectrometer (Thermo Scientific, San Jose, CA) with a resolution of 120,000 for ions with an m/z of 400. MS/MS experiments were performed at the same conditions as described above. Product-ion spectra were acquired in a data-dependent mode, and the ten most abundant ions were selected for product-ion analysis. All data were processed with the software MassAnalyzer28 for the peptide identification and the deuterium level calculation. Deuterium incorporation to each peptide was normalized to its theoretical maximum deuterium uptake and reported without back exchange correction.29 The details describing how comparisons of the temperature-dependent changes in protein conformational dynamics were assessed by H/DX are provided in the Supporting Information.

Thermal and Conformational Stability Measurements. Thermal denaturation experiments were performed on a Jasco J-815 circular dichroism (CD) spectrophotometer (Jasco, Tokyo, Japan) interfaced with a Peltier temperature controlled cell holder (Jasco). Samples were heated from 25 to 80 °C, while monitoring the far ultraviolet (far-UV CD) signal at 222 nm using a bandwidth of 2 nm. Because of a pronounced shift in the thermal unfolding curve as a function of the heating rate, ranging from 45 to 120 °C/h (Figure S2A in the Supporting Information), and lack of reproducibility from the same sample following thermal denaturation and cooling (Figure S2B in the Supporting Information), only an apparent melting temperature (Tapp m ) was determined as a function of sucrose and KPi concentration using a heating rate of 45 °C/h. This apparent lack of reversibility was observed for samples spanning a concentration range of 7 to 1 μM measured in cuvettes with 0.2 and 1 cm path lengths, respectively. The conformational stabilities of rhGCSF in the presence of 30 and 188 mM KPi (pH 7.1) were determined by isothermal urea induced equilibrium unfolding at temperatures ranging from 4 to 37 °C. Individual 6.8 μM rhGCSF samples of both formulations were equilibrated in final urea concentrations ranging from 0 to 9.8 M. Equilibration times of 72, 48, and 24 h were used for unfolding curves incubated at 4, 10−15, and 20− 30 °C, respectively. Unfolding curves monitored at 37 °C were equilibrated for only 3 h, as deemed sufficient from kinetic unfolding studies performed near the folding/unfolding transition region at this temperature. Equilibrium refolding experiments at 25 °C were conducted in a similar manner starting from an unfolded stock equilibrated overnight in 8 M urea and diluting to final urea concentrations ranging from 7.8 to 0.5 M. Following equilibration, all samples were analyzed by far-UV CD collected at 222 nm using a bandwidth of 2 nm and a path length of 0.2 cm at the same temperatures used for sample equilibration. Equilibrium unfolding and refolding curves at 25 °C were also monitored by intrinsic tryptophan fluorescence with a Varian Carey Eclipse spectrofluorometer (Varian Inc., Palo Alto, CA) using an excitation wavelength of 295 nm and monitoring emission from 310 to 400 nm. The isothermal equilibrium unfolding transitions monitored by far-UV CD and fluorescence spectroscopy were fit locally and globally (described in the Supporting Information) to a two-state model described previously25 using Igor Pro 6.21 (WaveMetrics Inc., Lake Oswego, OR). This model uses the linear extrapolation method between the free energy of unfolding and denaturant concentration24 according to eq 1. ΔG = ΔG H2O − m[urea]

■

(1)

Here, ΔGH2O represents the free energy of unfolding in the absence of urea and m is the dependence of the free energy of unfolding on the denaturant concentration. For comparison of the data collected with different optical probes or at different temperatures, data were normalized to the fraction unfolded (Fapp) according to eq 2. Fapp =

(YF − Yi ) (YF − YU)

RESULTS

rhGCSF Solubility. In order to make meaningful comparisons between the formulation dependence of nativestate solubility and non-native aggregation, our basic approach was to correlate the aggregation data collected at 37 °C with the solubility data extrapolated to these same conditions from temperatures less than 37 °C, where little non-native aggregation was observed during the time required for solubility equilibration. Working at some of the high rhGCSF concentrations necessary for solubility determination, it was essential to be able to distinguish native-state solubility from non-native irreversible aggregation and precipitation. The distinction between the two was based on two criteria. First, equilibrium was rapidly reached for the native-state solubility experiments conducted near room temperature, as demonstrated in a representative solubility time-course experiment (Figure S1 in the Supporting Information). This was not the case for non-native aggregation where aggregation/precipitation proceeded until the entire soluble monomer population was depleted. Second, the pelleted precipitate from the solubility experiments could be easily resolubilized after dilution and overnight soak in cold 0.15 M KPi; recovery was on average about 95%. These samples remained monomeric and showed the same amount of alpha helical structure by farUV CD spectroscopy as the nonprecipitated control (Figure S3 in the Supporting Information). In contrast, non-native rhGCSF aggregation led to irreversible aggregation/precipitation that has been shown in past reports to be predominantly

(2)

Here, Yi is the observed signal at a given urea concentration, and YF and YU are the extrapolated signals of the folded and unfolded baselines at the same urea concentration, respectively. Hydrogen/Deuterium Exchange (H/DX) Mass Spectrometry. Deuterium exchange was initiated by dilution of 6 D



Article

composed of intermolecular β-sheet structure and contain some degree of disulfide covalent cross-links.12,19 Initial solubility data was collected by salting-out the rhGCSF native state with increasing concentrations of the kosmotrope KPi. Extrapolation of salting-out curves are typically done by fitting the log of the solubility data (S) to the empirical Cohn equation: log S = β − KsI. Here, β represents the theoretical solubility limit in the absence of salt, Ks is the slope, and I is ionic strength or concentration.30,31 Representative rhGCSF salting-out data conducted at neutral pH by addition of increasing concentrations of the KPi and fit instead to a single exponential equation are shown in Figure 1A. Perhaps the most striking feature of this plot is the unusual solubility temperature dependence. Typically, protein solubility decreases as the temperature is lowered;32 however, this appears to be the opposite for rhGCSF, where for instance nearly twice as much KPi was required to precipitate the same amount of rhGCSF at 4 °C compared to 37 °C. This inverse temperature dependence appeared fully reversible, as turbid solutions of rhGCSF stored near 37 °C for short-term cleared upon lowering the storage temperature. Another interesting feature of Figure 1A is the extrapolated high solubility limit of rhGCSF in the absence of phosphate (β). Contrary to this extrapolation; however, the solubility limit of rhGCSF rapidly decreased below 0.2 M KPi and showed the same inverse temperature dependence as the salting-out curves. Importantly, this phenomenon allowed for the effect of sucrose and KPi on the solubility of rhGCSF to be measured directly at more physiological relevant ionic strengths, rather than relying on lengthy extrapolations from either salting-out curves or from solubility curves generated from macromolecular crowding agents such as PEG. Representative rhGCSF solubility data (formulated in 30 mM KPi pH 7.1) as a function of sucrose concentration are shown in Figure 1B and demonstrate that increasing concentrations of sucrose increase rhGCSF solubility, particularly at lower storage temperatures. These data, as well as the temperature dependence of solubility at a given sucrose concentration (Figure 1C), were well described by an empirical single exponential equation similar to the KPi salting-out data. Rather than fitting the data twice to extrapolate to 37 °C and some of the higher sucrose concentrations used for the aggregation studies, the entire rhGCSF solubility surface was analyzed in a single global fit (Figure 2A) to an exponential function of the type shown in eq 3 where X and T represent the sucrose concentration and temperature, respectively; m are the exponential constants, and b is the common intercept. f (X , T ) = b·exp(mX X + m TT )

(3)

Practically, eq 3 was recast to the base-point form (eq 4) so that the solubility limit (Y1) at a specified sucrose concentration and temperature (X1 and T1) could be determined directly from the fit, rather than calculating it from the fitted parameters. This avoided the need of propagating the error of Y1. f (X , T ) = Y1·exp[mX (X − X1) + m T(T − T1)]

Figure 2. (A) rhGCSF solubility dependence on temperature and sucrose concentration (buffered with 30 mM KPi, pH 7.1). This data is also shown rotated horizontally (B) to highlight the slight twist in the surface. (C) The rhGCSF temperature−KPi (pH 7.1) concentration solubility surface. Data (●) were fit globally (gray surfaces) to a single (A and B) or sum of two (C) exponential equations of the type shown in eq 4, incorporating the equation of a line for mX and mT as described in the text. Parameters of these fits are given in Table 1

(4)

It was also noted that mX showed a small temperature dependence and m T showed the same small sucrose concentration dependence (Figure 1B,C, insets). This added a slight twist to the entire surface (Figure 2B) and was accounted for by substituting the equation of a line: bX + mT for mX and bT + mX for mT; note that the slopes are common to both. In all, four parameters were used to fit the surface shown in Figure 2A,B and used to calculate the sucrose concentration

solubility dependence of rhGCSF at 37 °C. To account for both the salting-in and salting-out phases of the rhGCSF solubility with KPi concentration, the entire KPi-temperature solubility surface was fit in a similar manner as the sucrose data, E



Article

Table 1. Fitted Parameters of the Sucrose and KPi Solubility Surfacesa Y1 (mg/mL) bX (M−1) bT (°C−1) m (M−1 °C−1) X1 (M) T1 (°C)

sucrose

KPi salt-in

KPi salt-out

29 (3) 3.4 (2) −0.137 (0.003) −0.04 (0.01) 0.6 37

21 (2) 40 (3) −0.038 (0.008) −0.7 (0.2) 0.03 37

1194 (320) −5.9 (0.6) 0.04 (0.03) −0.11 (0.03)

Globally fit parameters of the sucrose/KPi−temperature solubility surfaces shown in Figure 2 to eq 4, incorporating the equation of a line for mX and mT as described in the text. X1 and T1 are fixed, and the values in parentheses are the standard deviations of the fitted parameters. For correlation with the sucrose aggregation data, T1 was fixed at 37 °C and X1 was incrementally changed to the corresponding sucrose concentration used for the aggregation experiments, and the surface was refit to determine Y1. The same procedure was used for the KPi solubility-aggregation correlations, with the exception that KPi solubility is equal to [(Y1salt‑in)−1 + (Y1salt‑out)−1]−1. a

slowest, respectively. In all cases a clear protein concentration dependence on monomer loss was observed and the slopes of the log−log plots of initial rates vs concentration for all formulations were between 1.8 (±0.1) and 2.02 (±0.06), indicative of a second order reaction (Figure S6 in the Supporting Information). In any event, the apparent relaxation times derived from the double exponential fits of the dimerization reaction were used only as a convenient empirical measure for correlation with solubility data, similar to the halflives from the sigmoidal fits, and are not intended for the purpose of deriving mechanistic detail. As demonstrated previously,12 sucrose had a remarkable effect on the rate of rhGCSF aggregation. In the absence of sucrose, nearly the entire starting amount of rhGCSF precipitated after about 3 days storage at 37 °C (Figure 3A) while roughly 75% of the protein remained soluble at the highest concentration of sucrose tested (0.6 M). Increasing concentrations of sucrose also had a pronounced effect on the dimer rising (τ1) and falling (τ2) phases, but virtually none on the maximum amount of dimer populated (Figure 3B). As shown in Figure 3C, both measures of rhGCSF aggregation for the sucrose containing formulations, half-life and the apparent dimer relaxation times, correlated well with the native rhGCSF solubility data that was calculated in the same conditions used for the aggregation experiments. In contrast to the sucrose formulations, rhGCSF aggregation showed a nonmonotonic dependence on KPi concentration similar to that observed for the solubility experiments. Increasing the KPi concentration from 30 to 188 mM slowed the aggregation process, while concentrations greater than 188 mM increased the rate of aggregation (Figure 4A,B). This same trend was also observed for the effect of KPi on the kinetics of rhGCSF dimerization (Figure 4C,D). Qualitatively, there was an excellent agreement between the KPi concentration dependence of aggregation/dimerization and solubility (Figure 5). Here the rhGCSF solubility limit calculated at 37 °C showed the same trend observed for the evolution in KPi concentration dependence of non-native aggregation (Figure 5A) and dimer τ2 (Figure 5B). Correlations between solubility and aggregation half-life and apparent dimer relaxation times are shown in Figure 5A,B insets, respectively. They are not quite as strong as the correlations for the sucrose formulations, which is likely consequence of the greater scatter in the KPi− temperature solubility surface. In order to assess the robustness of the above results, nearly the entire sucrose and KPi aggregation data sets were repeated (Figures S7 and S8 in the Supporting Information). The same trends and correlations to solubility as shown in Figures 3−5

but to the sum of two exponential equations according to [(eq 4)−1 + (eq 4)−1]−1 (Figure 2C). The fitted parameters of both solubility surfaces are given in Table 1. rhGCSF Aggregation Kinetics. Aggregation experiments were conducted at 37 °C at a protein concentration below the protein solubility limit calculated at this temperature (Table 1). Of the remaining soluble protein injected onto the SEC column, the main species eluting were monomer and dimer (as determined by multiangle light scattering); a representative SEC chromatogram is given in Supporting Information, Figure S4. To quantify the effects of sucrose and KPi on the rhGCSF non-native aggregation kinetics for correlation with the solubility data, the time required for half of the initial concentration of rhGCSF to be lost to insoluble aggregate, or half-life, was determined by fitting the time dependence of the total SEC integrated area to a sigmoidal equation. Since the rhGCSF aggregate has been shown previously to be composed of intermolecular β-sheet,12,13 the fact that we found evidence of Thioflavin-T binding (Figure S5 and details in the Supporting Information) and reports that the rate of rhGCSF aggregation did not increase in the presence of preformed aggregate seed,14 we chose to fit the total SEC area time course to a sigmoidal equation describing fibrillation through a primary nucleation and elongation pathway provided elsewhere.33 We note that a more rigorous investigation into the fibrillation mechanism of rhGCSF is currently underway; but for the purpose of the current report, this fit was used simply as a means of determining the rhGCSF half-life. The kinetics of the observed SEC dimer were reasonably fit to either the sum of a second order equation and first order exponential, or just the sum of two exponential equations (eq 5). Here, ΔAi is the change in SEC area associated with a particular kinetic phase and τi is its corresponding apparent relaxation time. N

A (t ) =

∑ ΔAi ·exp(−t /τi) i=1

(5)

We do not believe this indicates that dimer formation is limited by a first order unfolding process, but simply results from the limited number of data points and the inability to distinguish between a second and first order reaction at the early timepoints of the association process. To test this hypothesis the initial rates of monomer loss were determined during the early stages of aggregation (where only dimer appeared to grow and no appreciable loss in the total soluble protein was detected) as a function of starting rhGCSF concentration at pH 7.1 in the presence of 0.03 and 0.2 M KPi and 0.03 M KPi/0.5 M sucrose, conditions where the aggregation rates were the fastest and F



Article

Figure 3. Non-native aggregation kinetics of rhGCSF buffered with 30 mM KPi (pH 7.1) and formulated in 0 (●), 0.05 (■), 0.1 (▲), 0.2 (▼), 0.3 (⧫), 0.4 (○), 0.5 (□), and 0.6 M sucrose (△) and stored at 37 °C. Changes in the total integrated SEC area (A) and dimer area (B) were fit to a sigmoidal equation given elsewhere33 and the sum of two exponentials (eq 5), respectively (solid lines). (C) Both the halflife of total protein loss (●) and the apparent relaxation times of the dimer rising (τ1,○) and falling (τ2, □) phases correlated well to the rhGCSF solubility calculated at 37 °C. Error bars for panels A and B are smaller than the symbols and are the standard deviation of triplicate injections and represent the precision of the method. Error bars for panel C are the standard deviations of the fitted solubility, half-life, and relaxation time parameters.

Figure 4. Non-native aggregation kinetics of rhGCSF formulated in 0.03 (●), 0.06 (■), 0.09 (▲), 0.12 (▼), 0.2 (⧫), 0.3 (○), 0.4 (□), and 0.5 M KPi (pH 7.1) (△) and stored at 37 °C. Changes in the total SEC integrated area (A,B) and dimer area (C,D) were fit to the same equations used to describe the sucrose data (solid lines). Because of the nonmonotonic changes in SEC areas with increasing KPi concentration, the change in total SEC area and dimer area are split into panels A,B and C,D, respectively, for clarity. For reference, the KPi concentration with the slowest rate of SEC area change (⧫) is shown in both plots. Error bars are described in the legend of Figure 3.

were observed, although the half-lives and relaxation times for these repeated experiments were somewhat slower than our initial experiments. We believe this is a result of the starting concentration of the repeated experiments being about 0.2 mg/ mL lower than our initial experiments. rhGCSF Conformational Stability. Despite the lack of thermal unfolding reversibility, unfolding data collected in the

presence of increasing concentrations of sucrose were well fit to a two-state thermal denaturation model and the Tapp m values derived from these fits correlated well to the sucrose rhGCSF 37 °C aggregation half-lives and dimer relaxation times (Figure S9A in the Supporting Information). Thermal denaturation of rhGCSF in the presence of increasing KPi concentrations was possible but, like the sucrose formulations, was not a reversible G



Article

formulations and fit globally to determine ΔGH2O values at 37 °C; representative fits are given in Figure S11 in the Supporting Information. Similar to the solubility measurements, this was done to ensure that the ΔGH2O value calculated at 37 °C was not skewed by aggregation at this higher temperature. The thermodynamic parameters derived from these fits are provided in Table 2 and used to construct the Table 2. Thermodynamic Parameters for the Unfolding of rhGCSFa parameter

30 mM KPi

188 mM KPi

Hm (kcal mol−1) ΔCpb (kcal mol−1 K−1) Tm (K) h (kcal mol−1 M−1) s (kcal mol−1 K−1 M−1) c (kcal mol−1 K−1 M−1)

178 (13) 3.9 (0.5) 328 (2) 15 (3) 0.044 (0.009) 0.16 (0.08)

185 (7) 4.0 (0.3) 324.4 (0.6) 18 (1) 0.053 (0.005) 0.19 (0.05)

a

Results are from isothermal urea induced equilibrium unfolding experiments conducted as a function of temperature and globally fit to a two-state model incorporating eqs S1 and S2; all parameters are defined in the Supporting Information. Errors represent the standard deviations of the fitted parameters. bThe ΔCp values were about 1.5 times greater than expected based on the size of rhGCSF.9 Incorporating a linear temperature dependence of ΔCp, as the strong temperature dependence of the m value may warrant, decreased its value to 3 kcal mol−1 at 25 °C without significantly altering the other thermodynamic parameters.

stability curves shown in Figure 6 where the ΔGH2O and m values derived from both the local and globally fit isothermal urea denaturation curves are overlaid for comparison. Unlike the sucrose formulations, where increasing concentrations led to an increase in the Tapp m and conformational stability,12 increasing the KPi to a concentration where aggregation was slowest (near 0.2 M) slightly lowered the

Figure 5. Comparison of the KPi (pH 7.1) concentration dependence of rhGCSF solubility calculated at 37 °C, as described in the text (solid lines), and non-native aggregation monitored by SEC. (A) The decrease in total soluble protein following storage at 37 °C for 0 (●), 1.5 (■), 4.0 (▲), 20.25 (▼), 45 (⧫), 69 (○), and 93 h (□) are in good qualitative agreement with solubility. Inset, correlation between solubility and aggregation half-life (○) or amount of soluble protein remaining after 45 h storage (●). (B) The apparent relaxation time for the dimer falling phase (τ2) also shows good qualitative agreement with solubility and a reasonable correlation was made (inset, ●; no correlation with τ1 was observed, ■). Error bars are described in the legend of Figure 3.

process and resulted in a decrease in the Tapp m (Figure S9B in the Supporting Information). This decrease in Tapp m , however, likely reflects the loss in CD signal due to aggregation with increasing KPi concentration as indicated by an increase in the CD dynode voltage, increase of the post-transition baseline ellipticity, and sample turbidity following thermal denaturation. For these reasons, conformational stability was determined by urea denaturation at low (30 mM) and intermediate (188 mM) KPi concentrations where aggregation was fast and slow, respectively. Representative urea induced equilibrium unfolding and refolding curves monitored by far-UV CD and intrinsic tryptophan fluorescence are shown in Figure S10A in the Supporting Information. Unfolding and refolding curves monitored by both optical probes were fit with the two-state equilibrium unfolding model and their normalized Fapp signals overlaid well, in agreement with the two-state model assumption (Figure S10B in the Supporting Information). In practice, multiple isothermal urea denaturation curves were collected over a range of temperatures for both KPi

Figure 6. Comparison of the locally fit ΔGH2O and m values (upper inset) plotted as a function of temperature for rhGCSF formulated in 30 mM KPi (pH 7.1) (○) and 188 mM KPi (pH 7.1) (□) determined from isothermal urea-induced denaturation experiments monitored by far-UV CD at 222 nm. Solid lines were calculated from the globally fit parameters given in Table 2. Error bars represent the standard deviation of the locally fit unfolding curves and are smaller than the data points. Lower inset, calculation of the rhGCSF solubility at 4 (solid line) and 37 °C (long dashed line). For reference, a short dashed line is drawn at 188 mM KPi. H



Article

rhGCSF conformational stability at 37 °C. At refrigerated conditions, the opposite was the case, 188 mM KPi slightly stabilizes rhGCSF compared to the 30 mM KPi formulation. This result is likely a consequence of the temperature dependence of the KPi salting-in/out transition. The rhGCSF solubility limit was calculated at 4 and 37 °C from the parameters in Table 1 (Figure 6, bottom inset), and it appears that the 188 mM KPi formulation falls within the salting-in region at 37 °C, but in the stabilizing salting-out region at 4 °C. Hydrogen/Deuterium Exchange Mass Spectrometry. The time-course of deuterium incorporation for all 118 peptic peptides derived from rhGCSF are displayed in Figure S12 of the Supporting Information; redundant peptides are not shown, and the sequence coverage was 100%. Comparison of the H/ DX data collected at 4 and 37 °C were informative and revealed much greater exchange at 37 °C consistent with a more dynamic and open native state structure in agreement with the temperature dependence of the equilibrium unfolding m values, discussed later. Unlike the global unfolding studies, however, the H/DX data was collected at the peptide level and offers far greater resolution. It revealed greater conformational dynamics practically throughout the entire four-helical core structure as opposed to more local conformational changes (Figure 7). Note that rhGCSF is still quite stable at 37 °C (7−8 kcal/mol), a temperature some 24 °C below the rhGCSF Tapp m and 14 °C below even the onset of thermal unfolding.

rhGCSF Aggregation Mechanism. A prerequisite to understanding this correlation is a basic knowledge of the initial steps leading to rhGCSF aggregation. The loss of native rhGCSF has been proposed in past reports to follow the general nature of Scheme 1,12,13 where association is believed to be initiated from a partially unfolded species (N*), the exact nature of which remains unclear. Assuming that association operates at the diffusion limit, N* has been suggested to represent a high energy rarely populated (1 in a million) transition state species that, compared to native monomer, exposes about 15% the surface area as the unfolded state.12,13 On the basis of the temperature dependence of the urea equilibrium unfolding and solubility experiments in the present report and the rhGCSF concentration dependence of monomer loss, we propose a slightly different interpretation of the early processes leading to rhGCSF aggregation that help explain the native-state solubility-aggregation connection. The denaturant dependence of unfolding, or m value, along with the change in heat capacity of unfolding (ΔCp) have been shown to correlate well with the change in solvent accessible surface area between the native and unfolded state for proteins that unfold by an equilibrium two-state process.9 Accordingly, the strong decrease in the rhGCSF unfolding m value with increasing temperature (Figure 6, inset) signifies either a less well folded and more solvent accessible native-state or a more compact unfolded-state. The latter case seems less plausible and does not agree with the H/DX data that clearly show greater exchange rates throughout nearly the entire helical core at 37 °C compared to 4 °C. This shift to a more dynamic and solvent accessible structure at 37 °C does not seem to be rate limiting as previously thought.14 Instead, the rhGCSF concentration dependence of monomer loss suggests that dimerization is controlled by association. This was the case even in the presence of 0.5 M sucrose where conformational stability is high and partial unfolding is expected to be limiting.13 We speculate that this shift to a more open structure with increasing temperature is also the likely cause of the anomalous solubility temperature dependence, perhaps due to greater exposure of hydrophobic surface area, as shown in Figure 7. These results are consistent with an aggregation mechanism where the native monomer exists in a highly temperature sensitive pre-equilibrium with a well populated intermediate species (N*) of greater solvent exposure. This interpretation is also in agreement with a study on the aggregation temperature dependence of the bovine homologue (b-GCSF) by Roberts et al., where the equilibrium between N and N* was also shown to be established on a much faster time scale than the association process, a reaction that was considerably slower than the diffusion limit, i.e., an activated process.34 We note here that N* may alternatively be viewed simply as the gradual temperature-dependent increase in the structural dynamics of the native state ensemble, rather than the result of a cooperative transition to a distinct intermediate species. Since both models account for our experimental findings and the distinction between these two possibilities is beyond the scope of the current investigation, we choose to describe N* as an intermediate populated in a pre-equilibrium with native monomer in keeping with Scheme 1. This brings us to the role of the observed SEC dimer. This species has been described as off-pathway and not involved in higher order non-native rhGCSF aggregation based on decreasing dimer amounts at longer storage times, reversibility upon dilution, and the monomer−dimer equilibrium being

Figure 7. Hydrogen/deuterium exchange mass spectrometry shows the regions of the rhGCSF structure (pdb 1RHG,17 rendered with pyMOL 1.6, Schrödinger LLC) with greater conformational dynamics at 37 °C (magenta) compared to 4 °C.

■

DISCUSSION Previously, we used a simple solubility screen to rank the effectiveness of formulation cosolutes at solubilizing a handful of monoclonal antibodies and found that native-state solubility correlated well to the amounts of antibody dimer formed after nearly a year of static storage at 4 °C in the presence of the same cosolutes.7 This connection between the thermodynamics of native-state solubility and the rates of dimer formation may be understood since solubilizing the native-state likely lowers its chemical potential to a greater extent than the less solvent exposed dimer or monomer−dimer transition state. What is intriguing in the current report, however, is the correlation between the apparent rhGCSF native-state solubility limit and the rhGCSF rate of non-native aggregation. A better understanding of the underlying cause of this correlation is needed if solubility assays are to be used as prescreening tools for protein biologics that aggregate through non-native pathways. I



Article

unaffected by the presence of increasing sucrose concentration.12 On the basis of this reasoning, the existence of an additional “on-pathway” dimer was proposed that cannot be detected due to coelution with the observed SEC dimer.12 From our perspective, irreversibility is not a prerequisite for the observed SEC dimer to be on-pathway. Moreover, a minimally populated dimer that decreases at longer storage times is consistent with a mechanism where the rate of dimer formation is slower than its subsequent rate of loss to higher order aggregate following an initial induction period. As way of demonstration, kinetic simulations of a simple aggregation mechanism based on Scheme 1 were performed using the program KINSIM.35 For this mechanism the N to N* preequilibrium was set much faster than both the association reaction to dimer (k2) and the following step from dimer to tetramer (k3) that was added as a simple means to account for dimer loss for the purpose of demonstration. If k2 > k3, then dimer accumulates (Figure S13A in the Supporting Information); however, when k2 is rate limiting, very little dimer is populated during the course of the reaction (Figure S13B in the Supporting Information), similar to the rhGCSF experimental results (Figures 3 and 4). Solubility and Thermal/Conformational Stability as Aggregation Prescreening Methods. Since association appears rate limiting, the rate of soluble monomer loss is dependent on the concentration of reactive monomer N*, which may change depending on temperature and solvent conditions. For example, the addition of preferentially excluded cosolutes that interact unfavorably with surfaces normally found in the native-state interior, e.g., the peptide backbone, should drive the equilibrium toward the less solvent exposed nativestate and limit protein association (Figure 8). Returning to the kinetic simulations, shifting the pre-equilibrium away from the N* species, as increasing concentration of sucrose would be expected to do, results in a decrease in the rates of dimer growth and loss, as well as the total amount of dimer formed

(Figure S13C in the Supporting Information). Assuming that the dimer must reach some critical concentration prior to its loss to higher order aggregate, and the dimer is allowed to grow to this amount, the entire curve shifts to the right (Figure S13C in the Supporting Information) consistent with our experimental results (Figure 3B). Any measure proportional to the N to N* ratio should in principle, therefore, correlate to the rhGCSF aggregation rate and be useful as a prescreening tool. In the case of conformational stability, the titration of sucrose will incrementally increase the chemical potentials of both the unfolded and N* states by an amount proportional to their solvent exposed surface areas (Figure 8). The ΔGunf between the native and unfolded-state will therefore correlate to the difference in free energy between the native and N* state, and by extension with the rate of non-native aggregation. In the case of rhGCSF, the apparent temperature at which the unfolded and native states are in equal concentration (Tapp m ) also appears to correlate with the rhGCSF aggregation (Figure S9A inset in the Supporting Information). We anticipate that solubility will also be proportional to the position of the N to N* pre-equilibrium if both states have different solubility limits. This seems likely given the concomitant decrease in solubility and increase in solvent exposed surface area with increasing temperature. Note that the stabilizing contacts between N* monomers during the initial steps of the aggregation process do not necessarily have to be the same as those responsible for native-state precipitation at lower temperatures for the solubilityaggregation correlation; its only necessary that solubility be proportional to the N* concentration. The correlation between solubility and non-native aggregation as a function of KPi concentration is perhaps more interesting. Increasing KPi from 0.03 to roughly 0.2 M greatly increased rhGCSF solubility and decreased its rate of dimerization and subsequent aggregation. Unlike sucrose, however, this probably is not the result of a significant depopulation of the reactive N* state since increasing KPi concentration to 0.188 mM had little effect on both rhGCSF conformational stability at 37 °C (Figure 6) and H/DX exchange rates (data not shown). If this is the case, then any solubility−aggregation dependence must therefore be a result of KPi destabilizing the same or similar attractive contacts/ interactions formed in the native-state precipitate and between monomers at the early stages of aggregation. These attractive forces are likely electrostatic since rhGCSF has been reported to have a highly asymmetric charge distribution at neutral pH,13 and increasing ionic strength would be expected to screen said charges and thereby increase the energy of the dimer and monomer−dimer transitions state ensemble (Figure 8). Beyond about 0.2 M KPi, all of the attractive surface charges are shielded and additional KPi is excluded, salting-out the protein and increasing its rate of aggregation. Practical Implications and Limitations of the Solubility Assay. To meet the challenge of timely protein therapeutic development, aggregation prescreening methods are needed to select candidate proteins with low intrinsic rates of aggregation as well as the extrinsic solvent conditions that further minimize this rate prior to initiating costly longer term studies. Solubility screens such as the one described in the current report and others that utilize high concentrations of kosmotrope salts and PEG to induce native-state precipitation and liquid−liquid phase separation are well suited to fill this role.36,37 Compared to other measures of colloidal interactions

Figure 8. Proposed free energy diagram showing the shift in equilibrium from the native monomer (N) to the more solvent exposed reactive intermediate (N*) at higher temperatures and its subsequent association to dimer (N*2). The addition of sucrose (long dashed line) destabilizes N*, shifting the equilibrium back to N. The low salting-in concentrations of KPi (3000 in vitro translated E. coli proteins (about 70% of the proteome),38,39 a result expected if a subset of the stabilizing interactions found in non-native aggregate were also present in the transition state ensemble of a rate limiting association step. General Considerations. Although destabilizing the N* species by the addition of sucrose seems to be the best strategy for minimizing aggregation of rhGCSF in the present study, we feel that it would be a mistake to follow only this approach in light of the fact that proteins have evolved to be only marginally conformationally stable for reasons related to biological function and regulation. Increasing the colloidal stability of N* by other means may still prove to be an attractive method for reducing the non-native aggregation rate of rhGCSF and potentially for other therapeutically important proteins. For instance, the structurally related hematopoietic cytokine erythropoietin that folds by a similar mechanism as rhGCSF40−42 is only about half as thermodynamically stable,42 shows the same strong m value temperature dependence (unpublished results), but does not form insoluble aggregate at physiological conditions due to the presence of glycosylation at four sites.43,44 Interestingly, the addition of a single N-terminal polyethylene glycol molecule to rhGCSF (as perhaps a glycosylation equivalent) was reported to render non-native rhGCSF aggregates soluble, but have little effect on the conformational stability or the initial steps of the rhGCSF aggregation process.19 Therefore, solubility may yet be an important attribute that protein therapeutics are selected/ engineered for and optimal solvent and storage conditions are screened to enhance, even for proteins that aggregate through non-native pathways.

■

■

AUTHOR INFORMATION

Corresponding Author

*(D.D.B.) E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors thank Michael Treuheit for encouraging these studies.

■

ABBREVIATIONS rhGCSF, recombinant human granulocyte colony stimulating factor; KPi, potassium phosphate; far-UV CD, far-ultraviolet circular dichroism spectroscopy; HPLC, high pressure liquid chromatography; Tapp m , apparent melting temperature; SEC, size exclusion chromatography; H/DX, hydrogen/deuterium exchange; ΔGunf, the difference in free energy between the native and unfolded ground states

■

REFERENCES

(1) Frokjaer, S.; Otzen, D. E. Protein drug stability: a formulation challenge. Nat. Rev. Drug Discovery 2005, 4 (4), 298−306. (2) Maas, C.; Hermeling, S.; Bouma, B.; Jiskoot, W.; Gebbink, M. F. B. G. A role for protein misfolding in immunogenicity of biopharmaceuticals. J. Biol. Chem. 2007, 282 (4), 2229−2236. (3) Lumry, R.; Eyring, H. Conformation changes of proteins. J. Phys. Chem. 1954, 58 (2), 110−120. (4) Roberts, C. J., Nonnative Protein Aggregation. In Misbehaving Proteins, Protein (Mis)Folding, Aggregation, and Stability; Murphy, R., Tsai, A., Eds.; Springer Science+Buisiness Media, LLC: New York, 2006; pp 17−46. (5) Roberts, C. J. Kinetics of irreversible protein aggregation: analysis of extended lumry−eyring models and implications for predicting protein shelf life. J. Phys. Chem. B 2003, 107 (5), 1194−1207. (6) Schaefer, J. V.; Plückthun, A. Engineering aggregation resistance in IgG by two independent mechanisms: lessons from comparison of Pichia pastoris and mammalian cell expression. J. Mol. Biol. 2012, 417 (4), 309−335. (7) Banks, D. D.; Latypov, R. F.; Ketchem, R. R.; Woodard, J.; Scavezze, J. L.; Siska, C. C.; Razinkov, V. I. Native-state solubility and transfer free energy as predictive tools for selecting excipients to include in protein formulation development studies. J. Pharm. Sci. 2012, 101 (8), 2720−2732. (8) Ishibashi, M.; Tsumoto, K.; Tokunaga, M.; Ejima, D.; Kita, Y.; Arakawa, T. Is arginine a protein-denaturant? Protein Expression Purif. 2005, 42 (1), 1−6. (9) Myers, J. K.; Nick Pace, C.; Martin Scholtz, J. Denaturant m values and heat capacity changes: Relation to changes in accessible surface areas of protein unfolding. Protein Sci. 1995, 4 (10), 2138− 2148. (10) Schmittschmitt, J. P.; Scholtz, J. M. The role of protein stability, solubility, and net charge in amyloid fibril formation. Protein Sci. 2003, 12 (10), 2374−2378. (11) Topping, T. B.; Gloss, L. M. The impact of solubility and electrostatics on fibril formation by the H3 and H4 histones. Protein Sci. 2011, 20 (12), 2060−2073. (12) Krishnan, S.; Chi, E. Y.; Webb, J. N.; Chang, B. S.; Shan, D.; Goldenberg, M.; Manning, M. C.; Randolph, T. W.; Carpenter, J. F. Aggregation of granulocyte colony stimulating factor under physiological conditions: characterization and thermodynamic inhibition. Biochemistry 2002, 41 (20), 6422−6431.

ASSOCIATED CONTENT

S Supporting Information *

Brief description of how the isothermal equilibrium denaturation experiments were globally fit, the details describing how comparisons of the temperature-dependent changes in protein K



Article

(13) Chi, E. Y.; Krishnan, S.; Kendrick, B. S.; Chang, B. S.; Carpenter, J. F.; Randolph, T. W. Roles of conformational stability and colloidal stability in the aggregation of recombinant human granulocyte colonystimulating factor. Protein Sci. 2003, 12 (5), 903−913. (14) Raso, S. W.; Abel, J.; Barnes, J. M.; Maloney, K. M.; Pipes, G.; Treuheit, M. J.; King, J.; Brems, D. N. Aggregation of granulocytecolony stimulating factor in vitro involves a conformationally altered monomeric state. Protein Sci. 2005, 14 (9), 2246−2257. (15) Thirumangalathu, R.; Krishnan, S.; Brems, D. N.; Randolph, T. W.; Carpenter, J. F. Effects of pH, temperature, and sucrose on benzyl alcohol-induced aggregation of recombinant human granulocyte colony stimulating factor. J. Pharm. Sci. 2006, 95 (7), 1480−1497. (16) Metcalf, D.; Nicola, N. A. Proliferative effects of purified granulocyte colony-stimulating factor (G-CSF) on normal mouse hemopoietic cells. J. Cell. Physiol. 1983, 116 (2), 198−206. (17) Hill, C. P.; Osslund, T. D.; Eisenberg, D. The structure of granulocyte-colony-stimulating factor and its relationship to other growth factors. Proc. Natl. Acad. Sci. U.S.A. 1993, 90 (11), 5167−5171. (18) Herman, A. C.; Boone, T. C.; Lu, H. S. Characterization, Formulation, and Stabiltiy of Neupogen (Filgrastim), a Recombinant Human Granulocyte-Colony-Stimulating Factor. In Formulation, Characterization, and Stabilty of Protein Drugs; Pearlman, R., Wang, J. Y., Eds.; Plenum Press: New York, 1996; pp 303−328. (19) Rajan, R. S.; Li, T.; Aras, M.; Sloey, C.; Sutherland, W.; Arai, H.; Briddell, R.; Kinstler, O.; Lueras, A. M. K.; Zhang, Y.; Yeghnazar, H.; Treuheit, M.; Brems, D. N. Modulation of protein aggregation by polyethylene glycol conjugation: GCSF as a case study. Protein Sci. 2006, 15 (5), 1063−1075. (20) Butler, A. M.; Blatt, H.; Southgate, H. The solubility of the plasma proteins: II. dependence on pH, temperature, and lipid content in concentrated solutions of potassium phosphate and application to their separate precipitation. J. Biol. Chem. 1935, 109 (2), 755−767. (21) Arakawa, T.; Timasheff, S. N. Theory of Protein Solubility. In Methods in Enzymology; Academic Press: New York, 1985; Vol. 114, pp 49−77. (22) Souza, L.; Boone, T.; Gabrilove, J.; Lai, P.; Zsebo, K.; Murdock, D.; Chazin, V.; Bruszewski, J.; Lu, H.; Chen, K.; et al. Recombinant human granulocyte colony-stimulating factor: effects on normal and leukemic myeloid cells. Science 1986, 232 (4746), 61−65. (23) Pace, C. N.; Vajdos, F.; Fee, L.; Grimsley, G.; Gray, T. How to measure and predict the molar absorption coefficient of a protein. Protein Sci. 1995, 4 (11), 2411−2423. (24) Pace, C. N. Determination and Analysis of Urea and Guanidine Hydrochloride Denaturation Curves. In Methods in Enzymology; Academic Press: New York, 1986; Vol. 131, pp 266−280. (25) Santoro, M. M.; Bolen, D. W. Unfolding free energy changes determined by the linear extrapolation method. 1. Unfolding of phenylmethanesulfonyl.alpha.-chymotrypsin using different denaturants. Biochemistry 1988, 27 (21), 8063−8068. (26) Zhang, J.; Chalmers, M. J.; Stayrook, K. R.; Burris, L. L.; GarciaOrdonez, R. D.; Pascal, B. D.; Burris, T. P.; Dodge, J. A.; Griffin, P. R. Hydrogen/deuterium exchange reveals distinct agonist/partial agonist receptor dynamics within vitamin D receptor/retinoid X receptor heterodimer. Structure 2010, 18 (10), 1332−1341. (27) Chalmers, M. J.; Busby, S. A.; Pascal, B. D.; He, Y.; Hendrickson, C. L.; Marshall, A. G.; Griffin, P. R. Probing protein ligand interactions by automated hydrogen/deuterium exchange mass spectrometry. Anal. Chem. 2006, 78 (4), 1005−1014. (28) Zhang, Z.; Zhang, A.; Xiao, G. Improved protein hydrogen/ deuterium exchange mass spectrometry platform with fully automated data processing. Anal. Chem. 2012, 84 (11), 4942−4949. (29) Wales, T. E.; Engen, J. R. Hydrogen exchange mass spectrometry for the analysis of protein dynamics. Mass Spectrom. Rev. 2006, 25 (1), 158−170. (30) Cohn, E. J. The physical chemistry of the proteins. Physiol. Rev. 1925, 5 (3), 349−437. (31) Englard, S.; Seifter, S.; Murray, P. D. Precipitation Techniques. In Methods in Enzymology; Academic Press: New York, 1990; Vol. 182, pp 285−300.

(32) Green, A. A.; Hughes, W. L. Protein Fractionation on the Basis of Solubility in Aqueous Solutions of Salts and Organic Solvents. In Methods in Enzymology; Academic Press: New York, 1955; Vol. 1, pp 67−90. (33) Cohen, S. I. A.; Vendruscolo, M.; Dobson, C. M.; Knowles, T. P. J. Nucleated polymerisation in the presence of pre-formed seed filaments. Int. J. Mol. Sci. 2011, 12 (9), 5844−5852. (34) Roberts, C. J.; Darrington, R. T.; Whitley, M. B. Irreversible aggregation of recombinant bovine granulocyte-colony stimulating factor (bG-CSF) and implications for predicting protein shelf life. J. Pharm. Sci. 2003, 92 (5), 1095−1111. (35) Barshop, B. A.; Wrenn, R. F.; Frieden, C. Analysis of numerical methods for computer simulation of kinetic processes: Development of KINSIM: a flexible, portable system. Anal. Biochem. 1983, 130 (1), 134−145. (36) Yamniuk, A. P.; Ditto, N.; Patel, M.; Dai, J.; Sejwal, P.; Stetsko, P.; Doyle, M. L. Application of a kosmotrope-based solubility assay to multiple protein therapeutic classes indicates broad use as a highthroughput screen for protein therapeutic aggregation propensity. J. Pharm. Sci. 2013, 102 (8), 2424−2439. (37) Wang, Y.; Latypov, R. F.; Lomakin, A.; Meyer, J. A.; Kerwin, B. A.; Vunnum, S.; Benedek, G. B. Quantitative evaluation of colloidal stability of antibody solutions using PEG-induced liquid−liquid phase separation. Mol. Pharmaceutics 2014, 11 (5), 1391−1402. (38) Agostini, F.; Vendruscolo, M.; Tartaglia, G. G. Sequence-based prediction of protein solubility. J. Mol. Biol. 2012, 421 (2−3), 237− 241. (39) Niwa, T.; Ying, B.-W.; Saito, K.; Jin, W.; Takada, S.; Ueda, T.; Taguchi, H. Bimodal protein solubility distribution revealed by an aggregation analysis of the entire ensemble of Escherichia coli proteins. Proc. Natl. Acad. Sci. U.S.A. 2009, 106 (11), 4201−4206. (40) Banks, D. D.; Scavezze, J. L.; Siska, C. C. Kinetic folding mechanism of erythropoietin. Biophys. J. 2009, 96 (10), 4221−4230. (41) Brems, D. N. The kinetics of G-CSF folding. Protein Sci. 2002, 11 (10), 2504−2511. (42) Banks, D. D. The effect of glycosylation on the folding kinetics of erythropoietin. J. Mol. Biol. 2011, 412 (3), 536−550. (43) Narhi, L. O.; Arakawa, T.; Aoki, K. H.; Elmore, R.; Rohde, M. F.; Boone, T.; Strickland, T. W. The effect of carbohydrate on the structure and stability of erythropoietin. J. Biol. Chem. 1991, 266 (34), 23022−23026. (44) Narhi, L. O.; Arakawa, T.; Aoki, K.; Wen, J.; Elliott, S.; Boone, T.; Cheetham, J. Asn to Lys mutations at three sites which are Nglycosylated in the mammalian protein decrease the aggregation of Escherichia coli-derived erythropoietin. Protein Eng. 2001, 14 (2), 135− 140.

L


Relationship between Native-State Solubility and Non-Native

Recommend Documents