Article pubs.acs.org/molecularpharmaceutics

Specific Catalysis of Asparaginyl Deamidation by Carboxylic Acids: Kinetic, Thermodynamic, and Quantitative Structure−Property Relationship Analyses Brian D. Connolly,*,† Benjamin Tran,‡ Jamie M. R. Moore,† Vikas K. Sharma,† and Andrew Kosky§ †

Early Stage Pharmaceutical Development, ‡Purification Development, and §Pharma Technical Development Management, Genentech, Inc., South San Francisco, California 94080, United States ABSTRACT: Asparaginyl (Asn) deamidation could lead to altered potency, safety, and/or pharmacokinetics of therapeutic protein drugs. In this study, we investigated the effects of several different carboxylic acids on Asn deamidation rates using an IgG1 monoclonal antibody (mAb1*) and a model hexapeptide (peptide1) with the sequence YGKNGG. Thermodynamic analyses of the kinetics data revealed that higher deamidation rates are associated with predominantly more negative ΔS and, to a lesser extent, more positive ΔH. The observed differences in deamidation rates were attributed to the unique ability of each type of carboxylic acid to stabilize the energetically unfavorable transition-state conformations required for imide formation. Quantitative structure property relationship (QSPR) analysis using kinetic data demonstrated that molecular descriptors encoding for the geometric spatial distribution of atomic properties on various carboxylic acids are effective determinants for the deamidation reaction. Specifically, the number of O−O and O−H atom pairs on carboxyl and hydroxyl groups with interatomic distances of 4−5 Å on a carboxylic acid buffer appears to determine the rate of deamidation. Collectively, the results from structural and thermodynamic analyses indicate that carboxylic acids presumably form multiple hydrogen bonds and charge−charge interactions with the relevant deamidation site and provide alignment between the reactive atoms on the side chain and backbone. We propose that carboxylic acids catalyze deamidation by stabilizing a specific, energetically unfavorable transition-state conformation of L-asparaginyl intermediate II that readily facilitates bond formation between the γ-carbonyl carbon and the deprotonated backbone nitrogen for cyclic imide formation. KEYWORDS: asparaginyl deamidation, specific catalysis, protein, monoclonal antibody (mAb), peptide, buffer, monocarboxylic acid, dicarboxylic acid, kinetics, thermodynamics, quantitative structure−property relationship (QSPR), GETAWAY descriptors

■

INTRODUCTION Asparaginyl deamidation has been studied extensively due to its biological and biopharmaceutical relevance.1 The conversion of Asn residues to aspartyl and isoaspartyl residues2 has been shown in some instances to affect protein function by altering its charge and structural properties.3−6 Interestingly, the rate of this reaction varies widely depending on the structure of a protein and its environment.7−10 For this reason, it has been proposed that Asn deamidation functions as a genetically specified molecular clock by regulating the timing of biological events.7,8,11 From a biopharmaceutical perspective, it is of interest to design proteins and formulations to minimize the rate of this reaction to ensure consistent drug efficacy, pharmacokinetics, and safety.9,10 Under physiological conditions, nonenzymatic deamidation of Asn residues results in the formation of isoaspartyl (IsoAsp) and aspartyl (Asp) residues in a 3:1 ratio, respectively.2,12−14 It has been proposed that this reaction involves four main steps: (1) deprotonation of the carboxyl-side residue backbone nitrogen; (2) ring formation by reaction of the γ-carbonyl carbon on the Asn side chain with the deprotonated backbone nitrogen; (3) removal of NH3 from the carboxamide group by © 2014 American Chemical Society

proton catalysis; and (4) succinimide ring hydrolysis on either side on the imide nitrogen to give either IsoAsp or Asp (Scheme 1).1 The rate of this reaction depends on protein sequence and structure and on various solution conditions including (1) solvent dielectric, (2) viscosity, (3) pH, (4) ion and ion concentration, and (5) buffer and buffer concentration.1,3,9,10 The effects of carboxylic acids (e.g., acetic acid, succinic acid, citric acid, etc.) on Asn deamidation is of particular interest in the biopharmaceutical industry because of their use as buffers in therapeutic protein formulations. The reported buffer dependence of Asn deamidation has been generally attributed to acid and base catalysis.10 The mechanism of imide formation is believed to involve two sequential general acid- and basecatalyzed reactions; the equilibrium deprotonation of the carboxy terminal backbone nitrogen is a base-catalyzed reaction (Scheme 1A→B), and the formation of intermediate IV from Received: Revised: Accepted: Published: 1345

January 7, 2014 March 10, 2014 March 12, 2014 March 12, 2014 dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Scheme 1a

a

Scheme adapted from ref 1.

III is an acid-catalyzed reaction (Scheme 1C→D). The buffer dependence of Asn deamidation is generally interpreted in this context.1,9,10,15,16 Although the role of acid and base catalysis of Asn deamination has been established, no comprehensive effort has been directed toward studying the specific catalysis of Asn deamidation by buffer molecules. In this study, we measure the effects of 16 structurally similar carboxylic acid buffers and 3 structurally similar nonacid excipients on the Asn deamidation

rates of an IgG1 monoclonal antibody (mAb) and model peptide using high throughput methods. To probe the mechanism of deamidation by carboxylic acid buffers, the importance of various structural motifs is studied using quantitative structure−property relationship (QSPR) methodologies. In addition, thermodynamic analysis is applied to elucidate the nature of the observed buffer catalysis and probe the effects on the energies of activation associated with the reaction. 1346

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Chart 1

Chinese hamster ovary cell lines, and purified at Genentech (South San Francisco, CA). mAb1 material was received at high concentration (>100 mg/mL) in histidine buffer at pH 6.0. The mAb1 material was stored frozen at −70 °C prior to sample preparation. To evaluate the effects of buffer and buffer concentration, mAb1 formulations were prepared with 12 different buffers at 3 buffer concentrations (0 mM, 10 mM, and 25 mM) with 240 mM sucrose and 0.02% polysorbate 20. Additional mAb1 samples were prepared separately at five sodium chloride concentrations (0.01 N, 0.05 N, 0.1 N, and 0.5 N) without buffer at pH 5.5 to evaluate the effects of ionic strength. Automated preparation of mAb1 samples was performed in triplicate by dispensing stock solutions using a Beckman Coulter (Fullerton, CA) Biomek FXp robot. Stock solutions were manually prepared using carboxylic acids (Chart 1) obtained from Sigma Aldrich (St. Louis, MO), compendial grade (USP, NP, EP) hydrochloric acid, sodium hydroxide, sucrose, sodium chloride, and polysorbate 20 materials, and

In addition to probing the underlying mechanism of specific catalysis of Asn deamidation, the results from this study have practical implications as well. Carboxylic acids that increase deamidation rates could be used as a catalyst to identify labile Asn residues. Conversely, carboxylic acids that minimize deamidation could be useful as buffers in protein drug formulations. The findings from this study may inform subsequent efforts for the molecular design of inhibitors to deamidation that could be used as excipients in protein drug formulations. This would be of significant interest as specific inhibitors of Asn deamidation could be used to diminish the rate of this ubiquitous reaction, independent of other solution conditions.

■

MATERIALS AND METHODS mAb1 Sample Preparation Procedures. mAb1, a κ light chain IgG1 monoclonal antibody with a molecular weight of approximately 150,000 g/mol, was cloned, expressed in 1347

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

after every 24 sample measurements using pH 4.0, pH 7.0, and pH 10.0 EMD buffer standards. Reference and pH electrodes are washed with deionized water purified using Elga PureLAB Ultra water purification system between each pH measurement. Determination of Protein Concentration by UV Spectroscopy. Protein concentration was determined using a SpectraMax M2e microplate spectrophotometer (Molecular Devices, Sunnyvale, CA) equipped with SoftMax Pro Software (Molecular Devices, Sunnyvale, CA). Samples were diluted to a nominal concentration of 0.5 mg/mL for analysis using representative buffer solutions for each sample. The UV absorbance of mAb1 and peptide1 samples was measured at 280 and 275 nm, respectively. All samples were analyzed using acrylic copolymer CoStar UV-transparent, 96-well microplates. UV protein concentration determination was calculated using the absorptivities of 1.45 (mg/mL)−1 cm−1 and 2.15 (mg/ mL)−1 cm−1 for mAb1 and peptide1, respectively. Determination of mAb Deamidation Using Ion Exchange Chromatography. The charge heterogeneity of mAb1 samples was determined using a 15 min, pH gradient ion exchange chromatography method. Chromatography was performed at a constant flow rate of 1.0 mL/min with a Dionex (Sunnyvale, CA) U3000 dual-column HPLC using a 4 mm i.d. × 50 mm Dionex ProPac WCX-10 HT analytical column. Prior to injection, samples were stored in the autosampler at 5 °C. For analysis, 100 μL of each sample was injected neat and the column was equilibrated for 2 min with 60% mobile phase A (11.6 mM piperazine at pH 6.0) and 40% mobile phase B (11.6 mM pipierazine at pH 11.0). Separation was achieved using a 9 min gradient from 40% to 70% mobile phase B. The elution was monitored at λ280nm using λ360nm as a reference. Peak integration was performed using Chromeleon software to quantitate the relative percent of charge variants. During analysis, the column compartment was maintained at 30 °C. Determination of Peptide Deamidation Using Reverse Phase UPLC. The Asn deamidation of the model peptide was determined using a 5 min reverse phase UPLC method. Chromatography was performed at a constant flow rate of 1.0 mL/min with a Waters (Waters Corporation, Milford, MA) H-Class Bio UPLC using a 2.1 mm by 100 mm Acquity UPLC BEH C18 column with 130 Å pore size and 1.7 μm particle size. Prior to injection, samples were stored in the autosampler at 5 °C. Twenty microliters of each sample was injected neat, and then the column was equilibrated for 1 min with mobile phase A (10 mM ammonium acetate, 0.01% TFA in water). Separation of deamidation reactants and products was achieved using a 3 min gradient from 0% to 9% mobile phase B (10 mM ammonium acetate, 0.01% TFA in acetonitrile). The 3D UV absorbance spectra of eluting peptides was monitored from 200 to 500 nm at 5 nm intervals using a DAD detector: integration of peak results was performed with Chromeleon software using the 275 nm signal. Peak characterization of IsoAsp and Asp products was performed using Protein L-isoaspartyl methyltransferase (PIMT) and endoproteinase Asp-N, respectively.17,18 PIMT and Asp-N enzymes used for peak characterization were obtained from Roche (Basel, Switzerland) and Promega (Madison, WI), respectively. Following separation, the column was re-equilibrated for 1 min with mobile phase A. During analysis, the column compartment was maintained at 40 °C.

water purified using an Elga PURELAB Ultra (Celle, Germany) water purification system. Stock solutions were pH adjusted to pH 5.5 using either HCl or NaOH. High concentration mAb1 samples were desalted and buffer exchanged into respective solutions using a PD Multitrap G-25 96-well plate. Buffer exchanged samples were collected into 96-well V-bottom polypropylene microplates (Bio-one, Greiner). Samples were then diluted to target concentration of 25 mg/mL using the appropriate stock solution. Sample pH and protein concentration were verified using pH analysis and UV spectroscopy, respectively. After verifying sample composition, each 96-well microplate was heat-sealed with aluminum foil and frozen at −70 °C prior to 40 °C isothermal hold. Following isothermal hold, sample plates were either analyzed immediately or frozen at −70 °C prior to analysis. Peptide Sample Preparation Procedures. A model hexapeptide (peptide1) with sequence YGKNGG was synthesized and purified at Genentech. To enhance stability of the terminal groups, the model peptide was acetylated on the Nterminal amino group and amidated on the C-terminal carboxyl group. The molecular weight of the capped hexapeptide is approximately 1,280 g/mol. The peptide was received as a lyophilized powder and stored at 5 °C temperature prior to sample preparation. The excipients (Chart 1) were obtained from Sigma Aldrich and Santa Cruz Biotechnology (Santa Cruz, CA), compendial grade (USP, NP, EP) hydrochloric acid and sodium hydroxide solutions were obtained from Genentech, and water was purified using an Elga PURELAB Ultra water purification system. Stock solutions were prepared with 25 mM of each of the excipients (Chart 1) in the presence of 5 mM acetic acid at pH 5.5. Additional stock solutions were prepared with oxalic acid, malonic acid, succinic acid, and glutaric acid at the same concentration at various pH intervals (pH range: 4−10). For each stock solution, pH was adjusted using either HCl or NaOH. Each peptide formulation was prepared by reconstituting the lyophilized peptide in each of the respective stock solutions. Sixty microliters of each sample was dispensed in triplicate into 384-well microplates using the Beckman Coulter Biomek FXp liquid handling robot. Sample pH and protein concentration were verified using pH analysis and UV spectroscopy, respectively. After verifying sample composition, each 96-well microplate was heat-sealed with aluminum foil and frozen at −70 °C prior to isothermal hold at 30 °C, 40 °C, 50 °C, and 60 °C. Following isothermal hold at each time interval, sample plates were either analyzed immediately or frozen at −70 °C prior to analysis. Determination of pH. Determination of stock solution pH was performed using a Mettler Toledo Easy Seven pH meter with a Mettler Toledo In-Lab-Micro pH electrode. The Mettler Toledo pH meter was standardized prior to analysis using pH 4.0, pH 7.0, and pH 10.0 buffer standards (EMD). Sample pH analysis was performed with an automated method using a custom, 8-channel pH analysis system. The automated system consists of 8 Beetrode (CAT#NMPH2B) pH electrodes each paired with Dri-ref (CAT#DRIREF-2SH) reference electrodes used to measure the pH of samples in a Greiner Bio One polypropylene 96-well plate. The movement of the 8-channel pH meter is controlled with an automated method using Biomek FXp liquid handling system and Biomek Software. The data collection is controlled using an ELIT 8-channel Ion/pH Analyzer (NICO 2000 LTD, Middlesex, U.K.) and customized Visual Basic (Microsoft, Redmond, WA) software program. The pH probes were calibrated before analysis and recalibrated 1348

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Figure 1. Kinetics plots display first-order fits of the 40 °C mAb1 chemical degradation with 0 mM (green), 10 mM (blue), and 25 mM (red) of (A) formic acid, (B), acetic acid, (C) propionic acid, (D) butyric acid, (E) valeric acid, (F) oxalic acid, (G) malonic acid, (H) succinic acid, (I) malic acid, (J) tartaric acid), (K) phthalic acid, and (L) glutaric acid.

■

⎛ [mAb] ⎞ ln⎜ ⎟ = −k mAbt ⎝ [mAb0 ] ⎠

RESULTS AND DISCUSSION Asparaginyl Deamidation in mAb1. The effects of carboxylic acid buffers on Asn deamidation in mAb1 were first investigated at 40 °C in the presence of 12 carboxylic acid buffers using high throughput ion exchange chromatography. Peak analysis of degraded samples using Asp-N peptide mapping of IEC fractions demonstrated that the primary mechanism of chemical degradation for mAb1 was Asn deamidation of heavy chain, Fc-located Asn-388 (data not shown). The carboxylic acid buffers evaluated include five monocarboxylic acids, namely, (A) formic acid, (B) acetic acid, (D) propionic acid, (F) butyric acid, and (I) valeric acid; and 7 dicarboxylic acids, namely, (K) oxalic acid, (L) malonic acid, (N) succinic acid, (O) malic acid, (P) tartaric acid, (Q) phthalic acid, and (R) glutaric acid. These 12 carboxylic acid buffers were selected to investigate the effects of different structural and physicochemical properties of buffers on mAb1 Asn deamidation rate constants. The chemical degradation of mAb1, as represented by the percent loss of main peak in IEC, in the presence of each carboxylic acid buffer was fitted using the first-order rate law (Figure 1: R2 range, 0.95−1.00). The first-order rate constants, kmAb, were calculated at 40 °C using the integrated first-order rate law:

(1)

wherein kmAb is the reaction rate, and [mAb] is the concentration of unreacted mAb1 (% main peak), and t is time in days. Each mAb1 sample was evaluated using size exclusion chromatography to measure the physical stability of the chemically degraded samples. No measurable change in percent aggregate was observed for the mAb1 samples during the isothermal hold (data not shown). Effect of Buffer Concentration on mAb1 Deamidation Rate. The Asn deamidation rate constants measured for mAb1 vary over a wide range (kmAb range: 0.007 to 0.048 day−1) (Figure 2). For example, the rate constant for 25 mM malonic acid (kmAb = 0.048 day−1) is six times higher than that for 25 mM formic acid (kmAb = 0.008 day−1) at fixed pH, ionic strength, and temperature (Figure 2). Also, the results demonstrate that the number of carboxyl groups (nCOOH) does not solely determine kmAb. For example, the rate constant for 25 mM malonic acid (kmAb = 0.048 day−1, nCOOH = 2) is also six times higher than that for 25 mM oxalic acid (kmAb = 0.008 day−1, nCOOH = 2) (Figure 2). These results are consistent with values for Asn deamidation in mAbs reported in the literature. The observed minimum deamidation rate (0.007 day−1) is comparable to Fc Asn deamidation (0.008 day−1) 1349

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

(kmAb = 0.010 day−1, nCOOH = 1) despite their same length alkyl chains (nC = 3). Also, kmAb of monocarboxylic acids is slightly higher for buffers with longer alkyl chain lengths (Figure 2). For example, valeric acid (kmAb = 0.012 day−1, nC = 5) has a rate constant 50% higher than formic acid (kmAb = 0.008 day−1, nC = 1). Similarly, the topological distance between carboxyl groups on dicarboxylic acid buffers appears to strongly influence deamidation rates. For example, kmAb is higher for malonic acid (kmAb = 0.047 day−1, nC = 3) compared to both succinic acid (kmAb = 0.0219 day−1, nC = 4) and oxalic acid (kmAb = 0.008 day−1, nC = 2). This suggests that a specific topological distance between carboxyl groups can result in large (up to 6fold) increases in deamidation rate. The observed importance of the topological distance between carboxylic acid groups suggests that effective catalysis of the deamidation reaction requires a single buffer molecule to simultaneously interact at two sites separated by a specific topological distance (nC = 3). A concomitant decrease in deamidation rate is observed if the carboxyl groups are separated by longer (succinic acid, nC = 4) or shorter (oxalic acid, nC = 2) distances than the malonic acid structure (nC = 3). This is consistent with the increase in deamidation rate observed with increasing alkyl chain length for the monocarboxylic acids. Thus, if a monocarboxylic acid is too short (e.g., formic acid or acetic acid) to interact at the two putative sites separated by larger topological distances (nC ≥ 3), then it will not likely increase the deamidation rate. Furthermore, the addition of hydroxyl groups to mono- and dicarboxylic acids resulted in higher rate constants than those of the unhydroxylated analogues (Figure 2). For example, kmAb for tartaric acid (kmAb = 0.042 day−1, nOH = 2) is significantly higher than that for malic acid (kmAb = 0.028 day−1, nOH = 1) and succinic acid (kmAb = 0.022 day−1, nOH = 0). Presumably, the addition of hydroxyl groups to succinic acid is effectively increasing the number of potential pharmacophoric points capable of two-site interactions with the topological separation distance (nC = 3). Conversely, the addition of an aromatic ring to a dicarboxylic acid buffer does not appear to have a significant effect on kmAb (Figure 2). Phthalic acid and succinic acid have the same number of carboxylate groups (nCOOH = 2) with the same topological distribution (nC = 4) between carboxylic acids. Phthalic acid (kmAb = 0.026 day−1, nBnz = 1) has only a slightly higher rate constant than succinic acid (kmAb = 0.022 day−1, nBnz = 0). This suggests that as long as the carboxyl and hydroxyl groups have the same topological distribution, there is no contribution by aromatic groups to the overall deamidation rate. QSPR Analysis of mAb1 Deamidation by Carboxylic Acids. Qualitative interpretations of kmAb suggest that the structure of carboxylic acid buffers is important for the catalysis of deamidation, independent of ionic strength and pH. As discussed previously, there is a complex relationship between the number of carboxyl and hydroxyl groups and their topological distribution on a buffer molecule and the observed rate of deamidation for mAb1. To screen for the specific structural and physicochemical properties that influence deamidation rate and to elucidate the mechanism of action for deamidation catalysis by carboxylic acid buffers, we implemented quantitative structure property relationship (QSPR) analysis.

Figure 2. Bar chart plots of the first-order rate constants, kmAb, at 40 °C for mAb1 formulated with each of the 12 carboxylic acid buffers at 0 mM (green), 10 mM (blue), and 25 mM (red). The results are sorted by both buffer and buffer concentration.

reported by Pace et al. for Asn deamidation in a mAb Fc fragment.10 Also, the results demonstrate that Asn deamidation rates for mAb1 typically increase at higher buffer concentration (Figure 2). The experimentally determined kinetics (Figure 1) and rate constants (Figure 2) emphasize the unique buffer concentration dependence of kmAb on each of these 12 carboxylic acid buffers. With the exception of formic and oxalic acid, kmAb measurably increases linearly as a function of buffer concentration in the range evaluated (≤25 mM). Notably, the buffer concentration dependence established is significant even at much lower concentrations (≤0.025 M) than buffer concentrations (0.1 to 2 M) commonly studied and reported in the literature.16 Similar buffer dependences of Asn deamidation have been reported for mAbs at low buffer concentrations (≤0.025 M) in the presence of acetate, succinate, tartrate, citrate, phosphate, and TRIS buffers.10,19 Effect of Ionic Strength on mAb1 Deamidation Rate. To investigate the contribution of ionic strength on the observed buffer concentration dependence of kmAb, formulations were evaluated at various NaCl concentrations. Comparable deamidation rates were observed for samples that contain both low concentrations (0.01M, kmAb = 0.005 day−1) and high concentrations (0.50M, kmAb = 0.005 day−1) of NaCl. Additionally, regression analysis demonstrates that there is a weak (R2 = 0.24, n = 11) and statistically insignificant (p = 0.131) linear correlation between kmAb and the theoretical ionic strength of each sample. These combined results demonstrate that the small differences in ionic strength (I range: 0.002 M and 0.004 M) for samples evaluated in this study do not account for the observed differences in kmAb for each buffer at the various buffer concentrations evaluated. This establishes that the observed differences in kmAb cannot be attributed solely to ionic strength or attributed to the number of carboxylic acid groups on each molecule. Rather, the results suggest that the differences in kmAb can be attributed to properties that are unique to each of the carboxylic acid buffers. Effect of Carboxylic Acid Structure on mAb Deamidation Rate. Several rank-order trends are observable within each subgroup of carboxylic acids, suggesting that various structural motifs are important for catalysis of deamidation. In general, k mAb is higher for dicarboxylic acid buffers than for monocarboxylic acid buffers (Figure 2). For example, the rate constant for malonic acid (kmAb = 0.048 day−1, nCOOH = 2) is approximately 4.8 times higher than that for propionic acid 1350

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Table 1 chemical identifier (buffer) buffer buffer buffer ··· buffer

activity to be modeled (degradation rate)

buffer descriptor 1

buffer descriptor 2

buffer descriptor 3

···

bufffer descriptor k

Kmab‑1 Kmab‑2 Kmab‑3 ··· Kmab‑n

Xbuffer‑1,1 Xbuffer‑2,1 Xbuffer‑3,1 ··· Xbuffer‑n,1

Xbuffer‑1,2 Xbuffer‑2,2 Xbuffer‑3,2 ··· Xbuffer‑n,2

Xbuffer‑1,3 Xbuffer‑2,3 Xbuffer‑3,3 ··· Xbuffer‑n,3

··· ··· ··· ··· ···

Xbuffer‑1,k Xbuffer‑2,k Xbuffer‑3,k ··· Xbuffer‑n,k

1 2 3 n

values (hii and hjj) from this matrix describe the accessibility of atoms on the molecule. More accessible, external atoms have higher leverage values compared to central backbone atoms and result in higher R-GETAWAY values. The geometric distance matrix, also derived from the Cartesian coordinate matrix, describes the geometric distance between atom pairs. Thus, higher R-values are calculated for atom pairs with shorter geometric, interatomic distances. The atomic weight matrices (wi and wj) apply weights to account for atomic properties such as mass (m), electronegativity (e), polarity (p), or volume (v). Of all R-GETAWAY (Rk: k = 0 to 7) descriptors, R4(m) has the highest correlation with kmAb (R2 = 0.81). Scatter plots display the correlation between Rk(m) descriptors with kmAb (Figure 3). The R4(m) descriptor encodes the number of atoms separated by 4 to 5 Å on the carboxylic acid molecules, weighted by atomic mass. Thus, the high correlation obtained for R4 descriptors demonstrate that the interatomic distance of 4−5 Å between atom pairs is important for catalytic activity. Conversely, the weak correlations (R2 < 0.2) for other Rk descriptors demonstrate that other interatomic spatial distributions (5 Å) do not contribute to observed differences in kmAb. Interestingly, using C-depleted molecular matrices for calculation of R4(m) further improves the correlation with kmAb (R2 = 0.97). The improved correlation with C-depleted R4(m) descriptors indicates that the spatial distribution of hydrogen and oxygen atoms is important for catalytic effect. Stepwise QSPR analysis of Rk(m) descriptors for each individual atom pair demonstrates that the spatial distribution of O−O (R2 = 0.93), and O−H (R2 = 0.52) atom pairs has the largest contribution to the observed correlations compared to C−O (R2 = 0.12), H−H (R2 = 0.01), C−C (R2 = 0.00), and C−H (R2 = 0.00) (Figure 4). Additional stepwise regression analysis demonstrates that the carbonyl oxygen atom pairs on the carboxyl and hydroxyl group C(O)OH−C(O)OH atom pairs contribute the most to the catalytic effectiveness. The second highest contribution results from the distribution of carbonyl oxygen on carboxyl groups and hydroxyl oxygen atom pairs C(O)OH−OH. To lesser extents, there are significant contributions to the observed correlation that also result from C(O)OH−C(O)OH, C(O)OH−C(O)OH, C(O)OH−OH, C(O)OH−OH, and C(O)OH−CH atom pairs. These findings suggest that the amount of accessible hydroxyl and carboxyl groups separated by 4 Å to 5 Å can be a determinant for catalysis of the deamidation reaction. The importance of the spatial distribution of hydrogen and oxygen atoms suggests that direct interactions (i.e., hydrogen bonding, charge−charge interactions, etc.) between the carboxylic acid and the potential pharmacophoric points on the protein substrate could be involved in the specific catalysis of this reaction. Interestingly, the QSPR analyses demonstrate the structural specificity required for catalytic effectiveness. This could also indicate a high level of conformational specificity for the transition state. Although statistical analysis identifies specific

Quantitative information encoding various physicochemical and structural properties was accomplished by calculating molecular descriptors for each buffer compound in an energyminimized conformation. Energy-minimized, molecular structures for each carboxylic acid buffer molecule used in the mAb1 study were optimized using Ghemical 2.10 and Empirical Potential Structure Refinement (EPSR) software programs. More than 1600 molecular descriptors were calculated for each of the optimized buffer structures using the E-Dragon 5.4 software.20 QSPR analysis was performed by using regression analysis to relate various molecular descriptors for each of the 12 carboxylic acid buffers with the corresponding mAb deamidation rates (Table 1). To understand the relationship of specific structural and physicochemical properties of buffers on kmAb, the correlation between molecular descriptors for buffers and kmAb was determined by calculating the correlation coefficient (R2) for each data set using linear regression analysis. The results of this QSPR analysis establish that the 3D descriptors, which encode information about the molecular geometry in 3D-space (R2 max: 0.81), outperform 2D- (R2 max: 0.57) and 1D-descriptors (R2 max: 0.56), which encode information about the topology and atomic properties of each molecule, respectively. The results from the QSPR analysis identify various GETAWAY (GEometry, Topology, and Atom-Weights AssemblY) descriptors as the highest performing determinants of kmAb based on the strength of the calculated correlation. Among the 3D descriptors, GETAWAY (R2 max: 0.81), 3D-MoRSE (R2 max: 0.63), and RDF (R2 max: 0.59) based descriptors resulted in the highest correlation with kmAb. Of the various GETAWAY descriptor classes evaluated, R-GETAWAY (Rk) descriptors had the strongest correlation (R2 max: 0.81) with kmAb. The autocorrelation, R-GETAWAY descriptor, is calculated by summing the information from each unique pair of atoms i and j using various matrix elements which describe the molecular structure and properties. R-GETAWAY descriptors are defined by the following equation:21 A−1

R k(w) =

∑∑ i=1 j>i

hii ·hjj rij

k = 1, 2, 3, 4, ..., D

·wi ·wj ·δ(dij ; k) (2)

wherein Rk(w) is the w-weighted kth order autocorrelation index, rij and dij are the geometric and topological distance between atoms i and j, and hii and hjj are the atomic leverages of atoms i and j obtained from the molecular influence matrix. RGETAWAY descriptors encode information about accessibility, spatial distribution, and atomic properties of each atom pair using the molecular influence matrix (Hij), the connectivity matrix (Dij), the geometry matrix (Rij), and atomic weights (wi and wj). The molecular influence matrix is calculated from the Cartesian coordinate matrix of atoms in 3D-space. The leverage 1351

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Figure 4. 2D-contour plot displays the correlation coefficient (R2) between R-GETAWAY descriptors (Rk) for carboxylic acids and mAb1 deamidation rate (kmAb) for individual atom pairs at various spatial distributions (k). R-GETAWAY descriptors were calculated for each atom pair with eq 2 using stepwise calculations. The color map and contour on the z-axis represent the correlation (R2) between kmAb and the R-GETAWAY descriptor for each combination of atom pair (xaxis) and spatial distribution (y-axis).

a simplified protein. Herein, a model hexapeptide (peptide1) with sequence YGKNGG was used to study the temperature dependence of Asn deamidation rates in the presence of various excipients. In this study, 18 excipients (Chart 1) were evaluated including 7 monocarboxylic acid buffers, namely, (A) formic acid, (B) acetic acid, (C) glycolic acid, (D) propionic acid, (E) 3-hydroxypropionic acid, (F) butyric acid, and (G) 3hydroxybutanoic acid; 7 dicarboxylic acid buffers, namely, (K) oxalic acid, (L) malonic acid, (N) succinic acid, (O) malic acid, (P) tartaric acid, (Q) phthalic acid, and (S) terephthalic acid; and 3 nonacids, namely, (H) butane-1,3-diol, (J) 3-hydroxybutan-2-one, and (M) pentane-2,4-dione. These 17 excipients were selected to investigate the effects of different structural and physicochemical properties on Asn deamidation rates using a model peptide. The deamidation rates of the model hexapeptide in the presence of each excipient were measured at 30 °C, 40 °C, 50 °C, and 60 °C using reverse phase ultraperformance liquid chromatography (RP-UPLC). Fits of the chemical degradation of peptide1 in the presence of each carboxylic acid buffer were performed using the first-order rate law (Figure 5: R2 range, 0.96−1.00). The first-order rate constants, kpeptide, for peptide deamidation were calculated using the integrated first-order rate law:

Figure 3. Scatter plot panels display the linear regression analysis of the Rk(m) descriptors (A) R0(m), (B) R1(m), (C) R2(m), (D) R3(m), (E) R4(m), (F) R5(m), (G) R6(m), (H) R7(m), and kmAb (day−1) for each of the 12 carboxylic acid buffers evaluated.

oxygen and hydrogen atom pairs as the strongest determinants of deamidation, it is important to note that the R-GETAWAY descriptor includes the summed contribution from all atom pairs with the proper spatial distribution. This indicates that, for each carboxylic acid buffer, multiple atom pairs can contribute to the overall catalytic effect. Thus, the buffer molecule could potentially be interacting with the transition state using different potential pharmacophoric points, in a range of conformations. Asparaginyl Deamidation in a Model Peptide. To determine the energies of activation associated with the buffer catalyzed deamidation reaction observed in the mAb-1 study, classical temperature dependence studies were performed using

⎛ [Asn] ⎞ ln⎜ ⎟ = −k peptidet ⎝ [Asn 0] ⎠

(3)

wherein t is time in days, kpeptide is the reaction rate constant at the various temperatures, and [Asn] is the concentration of unreacted Asn residues. Effect of Carboxylic Acid Structure on Peptide Deamidation. Overall, the rate constants for the hexapeptide in the presence of the 14 carboxylic acid buffers and 3 excipients evaluated vary over a wide range at each temperature (Figure 1352

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Figure 5. Kinetics plots display first-order fits of peptide deamidation at 30 °C (blue), 40 °C (green), 50 °C (orange), and 60 °C (red) formulated with (A) formic acid, (B), acetic acid, (C) glycolic acid, (D) propionic acid, (E) butyric acid, (F) valeric acid, (G) oxalic acid, (H) malonic acid, (I) succinic acid, (J) malic acid, (K) tartaric acid, (L) phthalic acid, (M) terephthalic acid, (N) 3-hydroxybutan-2-one, (O) 3-hydroxybutanoic acid, (P) 3-hydroxypropanoic acid, (Q), butane-1,3-diol, and (R) pentane-2,4-dione.

dependence of kmAb and kpeptide demonstrates that the carboxylic acids have the same rank order effects on the deamidation rate in both proteins (R2 = 0.86). This indicates that the structural motifs on carboxylic acids that were shown previously to increase the deamidation rate on mAb1 appear to have the same effects on a model peptide. These results suggest that the

6). For example, the rate constant for 25 mM malonic acid (kpeptide = 0.39 at 60 °C) is approximately 2.4 times higher than that for 25 mM formic acid (kpeptide = 0.16 day−1 at 60 °C) at fixed pH, ionic strength, and temperature (Figure 6). Although the 40 °C rate constants are higher for the model peptide compared with mAb1, regression analysis of the linear 1353

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

Figure 6. Bar chart displays the first-order rate constants of peptide deamidation, kpeptide, at 30 °C (blue), 40 °C (green), 50 °C (orange), and 60 °C (red) formulated with 25 mM of each of the 17 excipients. The results are sorted by both buffer and temperature.

Figure 7. Bar chart displays the enthalpy of activation (blue) and the entropy of activation (red) for peptide samples formulated with 17 different excipients.

acid and base catalysis, deamidation of peptide1 was measured at 30 °C in the presence of oxalic, malonic, succinic, and glutaric acid buffers at various pH values. The pH dependence of kpeptide in the presence of oxalic, malonic, succinic, and glutaric acids is consistent with reported pH dependences of deamidation rates.23 For all samples analyzed, kpeptide is at a minimum around pH 5 and increases at both lower and higher pH due to acid catalyzed hydrolysis and base catalysis, respectively (Figure 8A). However, at each pH, the magnitude of kpeptide is different for each buffer. Interestingly, the results demonstrate that the general rank order trend of kpeptide for buffers (malonic > succinic > glutaric > oxalic) is consistent at all the pH values evaluated (Figure 8A: pH range, 4−10). For example, kpeptide for malonic acid is approximately 300% higher than for oxalic acid at pH 4 (Figure 8B), ∼100% higher at pH 6 (Figure 8C), ∼50% higher at pH 8 (Figure 8D), and ∼60% higher at pH 10 (Figure 8E). These results establish that the buffer-specific differences in kpeptide observed in the presence of each carboxylic acid buffer persist across a broad pH range. These findings suggest that the unique buffer dependence of Asn deamidation observed in the presence of carboxylic acids is independent of the buffer-specific acid and base catalysis that is expected to predominate at lower pH values ( 3) or smaller (nC < 3) alkyl chain length can still interact with the potential pharmacophoric points on the protein substrate, but have reduced catalytic activity due to suboptimal geometries: perhaps due to weaker hydrogen bonding (Scheme 2D). In this way, we propose that some carboxylic acids facilitate the reaction of intermediate II to tetrahedral intermediate III by providing a specific geometric alignment (i.e., interatomic distance) on the transition state between the γ-carbonyl carbon and the deprotonated backbone nitrogen on the carboxyl side residue (Scheme 1C). This interpretation is consistent with the negative ΔS⧧ obtained from the thermodynamic analysis, which suggests that catalysis involves the binding of the carboxylic acids to the transition state. Additionally, this is also consistent with the QSPR analysis, which suggests that there is specific geometrical distribution of hydrogen bond donors/ acceptors and charge groups on hydroxyl and carboxyl groups required for optimal binding between the carboxylic acids and the transition state.

potential to design such inhibitory excipients for pharmaceutical applications.

■

AUTHOR INFORMATION

Corresponding Author

*Early Stage Pharmaceutical Development, Genentech, Inc., 1 DNA Way, South San Francisco, CA 94080. E-mail: connolly. [email protected]. Tel: 650.467.4813. Fax: 650.225.3613. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors would like to acknowledge Kevin Ng for support of high throughput sample preparation activities and Jasper Lin and John Wang for their careful review of this manuscript.

■

REFERENCES

(1) Robinson, N. E.; Robinson, A. B. Molecular Clocks: Deamidation of Asparaginyl and Glutaminyl Residues in Peptides and Proteins; Althouse Press: Cave Junction, OR, 2004. (2) Geiger, T.; Clarke, S. Deamidation, isomerization, and racemization at asparaginyl and aspartyl residues in peptides. Succinimide-linked reactions that contribute to protein degradation. J. Biol. Chem. 1987, 262 (2), 785−94. (3) Robinson, A. B.; Scotchler, J. W. Sequence dependent deamidation rates for model peptides of histone IV. Int. J. Pept. Protein Res. 1974, 6 (5), 279−82. (4) Ibrahim, H. R. On the novel catalytically-independent antimicrobial function of hen egg-white lysozyme: a conformationdependent activity. Nahrung 1998, 42 (3−4), 187−93. (5) Westall, F. C. Released myelin basic protein: the immunogenic factor? Immunochemistry 1974, 11 (8), 513−5. (6) Solstad, T.; Flatmark, T. Microheterogeneity of recombinant human phenylalanine hydroxylase as a result of nonenzymatic deamidations of labile amide containing amino acids. Effects on catalytic and stability properties. Eur. J. Biochem./FEBS 2000, 267 (20), 6302−10. (7) Robinson, N. E.; Robinson, A. B. Molecular clocks. Proc. Natl. Acad. Sci. U.S.A. 2001, 98 (3), 944−9. (8) Robinson, N. E.; Robinson, A. B. Amide molecular clocks in drosophila proteins: potential regulators of aging and other processes. Mech. Ageing Dev. 2004, 125 (4), 259−67. (9) Wakankar, A. A.; Borchardt, R. T. Formulation considerations for proteins susceptible to asparagine deamidation and aspartate isomerization. J. Pharm. Sci. 2006, 95 (11), 2321−36. (10) Pace, A. L.; Wong, R. L.; Zhang, Y. T.; Kao, Y. H.; Wang, Y. J. Asparagine deamidation dependence on buffer type, pH, and temperature. J. Pharm. Sci. 2013, 102 (6), 1712−23. (11) Robinson, A. B. Molecular clocks, molecular profiles, and optimum diets: three approaches to the problem of aging. Mech. Ageing Dev. 1979, 9 (3−4), 225−36. (12) Cournoyer, J. J.; Lin, C.; Bowman, M. J.; O’Connor, P. B. Quantitating the relative abundance of isoaspartyl residues in deamidated proteins by electron capture dissociation. J. Am. Soc. Mass Spectrom. 2007, 18 (1), 48−56. (13) Cournoyer, J. J.; Lin, C.; O’Connor, P. B. Detecting deamidation products in proteins by electron capture dissociation. Anal. Chem. 2006, 78 (4), 1264−71. (14) O’Connor, P. B.; Cournoyer, J. J.; Pitteri, S. J.; Chrisman, P. A.; McLuckey, S. A. Differentiation of aspartic and isoaspartic acids using electron transfer dissociation. J. Am. Soc. Mass Spectrom. 2006, 17 (1), 15−9. (15) Capasso, S.; Mazzarella, L.; Sica, F.; Zagari, A. Deamidation via cyclic imide in asparaginyl peptides. Pept. Res. 1989, 2 (2), 195−200. (16) Capasso, S.; Mazzarella, L.; Zagari, A. Deamidation via cyclic imide of asparaginyl peptides: dependence on salts, buffers and organic solvents. Pept. Res. 1991, 4 (4), 234−8.

■

CONCLUSIONS Collectively, the results from the kinetics, thermodynamics, and QSPR studies provide compelling evidence that specific catalysis of Asn deamidation by carboxylic acids, rather than acid and base catalysis, describes the observed differences in kmAb and kpeptide. The weak correlation between the pKa of carboxylic acids and both kmAb (R2 = 0.38) and k2 (R2 = 0.13) demonstrates that acid and base catalysis do not describe the observed differences in kmAb and kpeptide. These results are consistent with the results from the thermodynamic analysis that demonstrate there is no correlation between increased deamidation rates and decreased enthalpies of activation. Conversely, the strong rank-order correlation between increased kpeptide and decreased entropies of activation emphasizes the role of associative interactions between the carboxylic acids and the transition state in the activated complex. Elucidation of the structural aspects of this reaction using QSPR analysis demonstrates that the spatial distribution of specific functional groups (e.g., carboxyl and hydroxyl groups) on buffer molecules can determine the rate of chemical degradation reactions in proteins. The higher correlations obtained with the R-GETAWAY descriptor class demonstrate the importance of 4−5 Å geometric distances between atom pairs capable of forming hydrogen bonds or charge−charge interactions. The coincidence of these important structural motifs on the carboxylic acids and the pharmacophoric points on the transition state suggest that specific catalysis by direct interactions (i.e., hydrogen bonding and charge−charge interactions) between the carboxylic acid and the deamidation reactive site could facilitate reaction of intermediate II to III by providing alignment between the reactive γ-carbonyl carbon and backbone nitrogen atoms. These findings are important because they indicate that the rate of Asn deamidation can be modulated via two-site, simultaneous interactions on the Asn side chain and backbone. Although the carboxylic acid buffers evaluated in this study tended to increase rather than decrease deamidation rates, it is possible that the findings from this study could be applied to design molecules with similar geometric dimensions capable of diminishing the rate of this ubiquitous reaction by facilitating less reactive conformations of the Asn side chain. Future work in this group will focus on exploring the 1357

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358

Molecular Pharmaceutics

Article

(17) Johnson, B. A.; Aswad, D. W. Optimal conditions for the use of protein L-isoaspartyl methyltransferase in assessing the isoaspartate content of peptides and proteins. Anal. Biochem. 1991, 192 (2), 384− 91. (18) Zhang, W.; Czupryn, J. M.; Boyle, P. T., Jr.; Amari, J. Characterization of asparagine deamidation and aspartate isomerization in recombinant human interleukin-11. Pharm. Res. 2002, 19 (8), 1223−31. (19) Zheng, J. Y.; Janis, L. J. Influence of pH, buffer species, and storage temperature on physicochemical stability of a humanized monoclonal antibody LA298. Int. J. Pharm. 2006, 308 (1−2), 46−51. (20) Tetko, I. V.; Gasteiger, J.; Todeschini, R.; Mauri, A.; Livingstone, D.; Ertl, P.; Palyulin, V. A.; Radchenko, E. V.; Zefirov, N. S.; Makarenko, A. S.; Tanchuk, V. Y.; Prokopenko, V. V. Virtual computational chemistry laboratory–design and description. J. Comput.-Aided Mol. Des. 2005, 19 (6), 453−63. (21) Consonni, V.; Todeschini, R.; Pavan, M. Structure/response correlations and similarity/diversity analysis by GETAWAY descriptors. 1. Theory of the novel 3D molecular descriptors. J. Chem. Inf. Comput. Sci. 2002, 42 (3), 682−92. (22) Eyring, H. The activated complex in chemical reactions. J. Chem. Phys. 1935, 3, 107. (23) Capasso, S.; Mazzarella, L.; Sica, F.; Zagari, A.; Salvadori, S. Kinetics and mechanism of succinimide ring formation in the deamidation process of asparagine residues. J. Chem. Soc., Perkin Trans. 2 1993, No. 4, 679−82.

1358

dx.doi.org/10.1021/mp500011z | Mol. Pharmaceutics 2014, 11, 1345−1358