Characterization and Higher-Order Structure ... - ACS Publications

74 downloads 6533 Views 1MB Size Report
Feb 1, 2016 - indicates that DAR6 may form multimers while DAR2 and DAR4 primarily exist in monomeric forms in solution at ambient ... stable to thermal stress such that ADC-1 rapidly forms ...... *E-mail: [email protected]. Phone: ...
Article pubs.acs.org/bc

Characterization and Higher-Order Structure Assessment of an Interchain Cysteine-Based ADC: Impact of Drug Loading and Distribution on the Mechanism of Aggregation Jianxin Guo,† Sandeep Kumar,† Mark Chipley,§ Olivier Marcq,‡,⊥ Devansh Gupta,∥ Zhaowei Jin,† Dheeraj S. Tomar,† Cecily Swabowski,§ Jacquelynn Smith,§ Jason A. Starkey,§ and Satish K. Singh*,† †

Pharmaceutical R&D, §Analytical R&D, and ‡Bioprocess R&D, Pfizer Inc., 700 Chesterfield Parkway West, Chesterfield, Missouri 63017, United States ∥ Department of Chemical and Biological Engineering, Princeton University, Princeton, New Jersey 08544, United States S Supporting Information *

ABSTRACT: The impact of drug loading and distribution on higher order structure and physical stability of an interchain cysteine-based antibody drug conjugate (ADC) has been studied. An IgG1 mAb was conjugated with a cytotoxic auristatin payload following the reduction of interchain disulfides. The 2-D LC-MS analysis shows that there is a preference for certain isomers within the various drug to antibody ratios (DARs). The physical stability of the unconjugated monoclonal antibody, the ADC, and isolated conjugated species with specific DAR, were compared using calorimetric, thermal, chemical denaturation and molecular modeling techniques, as well as techniques to assess hydrophobicity. The DAR was determined to have a significant impact on the biophysical properties and stability of the ADC. The CH2 domain was significantly perturbed in the DAR6 species, which was attributable to quaternary structural changes as assessed by molecular modeling. At accelerated storage temperatures, the DAR6 rapidly forms higher molecular mass species, whereas the DAR2 and the unconjugated mAb were largely stable. Chemical denaturation study indicates that DAR6 may form multimers while DAR2 and DAR4 primarily exist in monomeric forms in solution at ambient conditions. The physical state differences were correlated with a dramatic increase in the hydrophobicity and a reduction in the surface tension of the DAR6 compared to lower DAR species. Molecular modeling of the various DAR species and their conformers demonstrates that the auristatin-based linker payload directly contributes to the hydrophobicity of the ADC molecule. Higher order structural characterization provides insight into the impact of conjugation on the conformational and colloidal factors that determine the physical stability of cysteine-based ADCs, with implications for process and formulation development.



INTRODUCTION Antibody drug conjugates (ADCs) are highly potent targetedtherapy molecules designed to overcome some limitations of conventional antibodies such as low cytotoxicity and weak penetration into tumors. ADCs comprise three primary components: (1) a monoclonal antibody (mAb) that is specific to the target cell-surface tumor antigen and capable of internalization, (2) a highly potent cytotoxic small-molecule drug, and (3) a linker that enables covalent attachment of the cytotoxic drug to the mAb.1,2 The combination has higher tumor-killing ability compared to the antibody alone while also improving the therapeutic index of the cytotoxic drug by reducing off-target effects.3 Lysine and cysteine are the two most common amino acid residues utilized to conjugate the payload to the antibody.4−7 For example, Mylotarg (gemtuzumab ozogamicin) utilizes lysine-based conjugation chemistry to covalently attach calicheamicin to an anti-CD33 mAb. Adcetris (brentuximab © 2016 American Chemical Society

vedotin) is an ADC formed by thiol-maleimide chemistry, where interchain disulfides in an anti-CD30 monoclonal antibody are partially reduced, followed by alkylation with the drug-linker moiety monomethyl auristatin E. Kadcyla (Trastuzumab-DM1, or T-DM1) consists of the therapeutic anti-HER2 monoclonal antibody trastuzumab covalently linked to the maytansine analog DM1 via a two-step linkage, which involves lysine conjugation of mAb to linker followed by maleimide chemistry of drug to linker. Development of these therapeutic candidates can be challenging due to the heterogeneity of the ADC material resulting from the conjugation process,7 which is effectively a pool of antibody molecules conjugated to different levels. With conventional cysteine chemistry where the conjugation process Received: November 5, 2015 Revised: January 6, 2016 Published: February 1, 2016 604

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

Figure 1. Schematic representation of the various drug load and isoforms possible in interchain cysteine-based ADCs.

stable to thermal stress such that ADC-1 rapidly forms aggregates at 40 °C. The aggregates are primarily composed of molecules with an average DAR of six or eight. Pan et al. presented a high-resolution comparison of interchain cysteine linked IgG1 ADCs and the corresponding mAb by hydrogen/ deuterium exchange mass spectrometry (HDX-MS). Results show that conformation and dynamics of ADCs are different from those of the mAbs only in the CH2 domain directly below the hinge and in the CH2-CH3 domain interface. These regions are more structurally dynamic and/or solvent exposed in the ADC with the other parts of the molecule not being significantly perturbed.12 These findings were confirmed by Adem et al., who formulated the ADC pool as well as isolated DAR species in high and low ionic strength buffers and subjected them to thermal incubation at 40 °C.13 The presence of high ionic strength buffer led to time-dependent aggregate and fragment formation of ADC under stress conditions, predominantly from species with high DAR, presumably due to the fact that they have less structural stability with fewer interchain disulfide bonds. The little to no change in secondary and tertiary structure on conjugation observed in the above-discussed studies enables us to explore the quaternary structure change using molecular modeling in this study. In the context of this work, the quaternary structural change refers specifically to the movement of the two CH2 domains with respect to each other. From a formulation perspective, the conformational and colloidal characteristics of the various species are critical to developing a stable product.10 We had also proposed that the high DAR species would drive the instability of the ADC pool and would therefore have to be the focus of the formulation efforts.10 In this report, we extend our work to include a biophysical assessment of the individual DAR species. We have identified and isolated the predominant isomers for the various DAR species. We then examine the impact of conjugation on

relies on the reduction of interchain disulfide bonds, the average drug to antibody ratio (DAR) is controlled largely by the extent to which the bulk antibody interchain disulfide bonds are reduced during the conjugation process and the stoichiometry of the reactants.3,7 The generally nonspecific reduction of these bonds typically produces a distribution of conjugated antibody species with DARs ranging from 0 to 8. The properties of the ADC are therefore determined by that of the pooled individual species which in turn are dependent on their degree of conjugation. To mitigate the above limitations of conventional interchain cysteine residue conjugation, phage display based technologies for site specific conjugation by substituting antibody residues with cysteines have been explored.8,9 A number of other site-specific technologies are also being explored though most are in the early stages of development.3,9 Understanding biophysical and biochemical properties of the ADC is critical to their successful development as products and these properties are strongly related to the construct of the ADC, including the chemistry of conjugation as well as the linker-payload. Our earlier work has compared the physical stability of an ADC that consists of a maleimidocaproyl linker and an auristatin payload, with its parent mAb.10 Spectroscopic analysis shows that while the ADC and mAb have similar secondary and tertiary structures, the ADC is more easily destabilized by thermal stress indicating reduced conformational stability. Molecular modeling calculations also suggest a substantial reduction in the conformational stability of the mAb upon conjugation.10 Beckley et al. have explored temperatureinduced aggregation of an antibody drug conjugate ADC-1, wherein the antibody was linked to the val-cit-Monomethyl Auristatin E (vc-MMAE) linker drug through the reduction of interchain disulfides.11 Thermal analysis shows that for species with high DAR, the conjugation does not measurably alter the secondary structure but it does render the CH2 domain less 605

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

theoretically feasible isomers at a given DAR level may not be energetically stable, while the others may exist transiently, being intermediates for further conjugation to a higher DAR level. [For example, DAR2b may be further conjugated to evolve into DAR4b]. A time course for evolution of different ADC species (different DARs and isomers) during the conjugation reaction is needed to understand this question further. This shall be explored in a future publication. The ADC consists of approximately 3.1% unconjugated mAb, 21.0% DAR2, 22.1% DAR4b and 18.0% DAR4a, 16.8% DAR6 and 3.5% DAR 8. Thermal Analysis. The DSC thermograms show that the stability of the various antibody domains are impacted differently by conjugation. The DSC thermograms shown in Figure 3 indicates two transitions. In general, the first transition

conformational stability as well as on hydrophobicity, i.e., the driver of colloidal instability. Both experimental measurements and computational estimation of the hydrophobicity point to the direct contribution of the linker payload on the biophysical properties of the individual species as well as on the overall hydrophobicity of the ADC. The extent of conjugation (DAR) is the key element. Experimental evidence suggests that the DAR6 species may exist in a multimeric state, while DAR2 and DAR4 species likely exist in monomeric forms under ambient conditions. The study presented here, along with those in the literature, also provides a general roadmap on the techniques for detailed characterization of these complex and heterogeneous products. This characterization or product knowledge, apart from serving the objective of understanding for development as stated above, is also a key regulatory expectation.



RESULTS Characterization of Cysteine-Based ADC Isoforms. ADC conjugation mediated by thiol reduction typically leads to a distribution of species of various DAR and sites of conjugation. The theoretical possibilities are shown in Figure 1. While many potential structural and isomeric species for various DARs are possible using this type of conjugation chemistry, the cysteine-based ADC evaluated as part of this investigation consistently exhibits a preference for certain species. The HIC chromatogram shown in Figure 2 illustrates

Figure 3. DSC thermograms of mAb, ADC (mixture), and isolated drug load species DAR2, DAR4, and DAR6.

in the DSC thermogram represents the melting of the conserved heavy chain2 (CH2) domain while the second transition represents melting of the CH3 domain and the Fab fragment.15 A decrease in the Tm for the first transition following conjugation was observed. Additionally, the peak is gradually broadened with loading, indicative of the loosening of the structure and a less cooperative structural transition. There are noticeable differences in the onset and melting point for the first transition, which reflects the unfolding of the CH2 domain. The onset is approximately 42 °C for DAR6, 47 °C for DAR4, 49 °C for the ADC mixture, 55 °C for DAR2, and 63 °C for mAb. For the conjugates, it is expected that the CH2 domain would be disproportionately impacted because of its proximity to the major conjugation site, i.e., the hinge region. Compared to the DAR2 and DAR4 species, the DAR6 species shows a significantly more disrupted CH2 domain with broader peak and much lower heat capacity (Cp). This is because both interchain disulfide bonds in the hinge region in DAR6a are completely reduced to load four of six drug molecules, while DAR2 retains 100% and DAR4 (on average) retains 50% intact hinge region (Figure 1 and Figure 2). Note that the DAR6b with 50% intact hinge is not observed in the mixture, i.e., DAR6 is primarily DAR6a. The ADC mixture also shows a broad first transition that can be deconstructed to at least two species with melting temperatures of approximately 58 and 65 °C, reflecting its combination of different DAR components.

Figure 2. Hydrophobic interaction chromatography (HIC) profile for cysteine-based ADC composed of various DAR species. The species isoform identity corresponds to the nomenclature in Figure 1.

the typical DAR distribution profile and resolution of positional isoforms observed. The individual species were elucidated through two-dimensional liquid chromatography mass spectrometry (2-D LC-MS; Figure S1) in order to ensure proper assignment of the correct DAR and isoform. 2-D LC-MS is a standard approach for the characterization of ADCs, and to discern the structural isoforms generated through cysteinebased conjugation.14 Characterization of the positional isomers observed indicates which cysteine disulfide bonds are conjugated. HIC analyses of the final conjugation reflects the presence of the following as major species in the mixture: DAR0, DAR2a, DAR4a and 4b, DAR6a, and DAR8 (nomenclature in Figure 1). The absence of other (theoretical) isomers in the 2D-LCMS profiles is intriguing. Some of the 606

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry Molecular modeling has shown that conjugation changes the quaternary structure of the CH2 domains, depending upon both the DAR and the site of conjugation. The conformational perturbations in the light chains L and M are restricted to the conjugation sites and have little impact on CH2 domains. However, conjugation at heavy chains leads to structural perturbations in CH2 domains further down along the amino acid sequence of a heavy chain, suggesting propagation of conformational destabilization to antibody structural regions other than the near neighborhood of the conjugation sites. This is clearly apparent in the DAR2 which is limited to DAR2a and not conjugated at the hinge, but the onset of CH2 melting is lower (55 °C instead of 63 °C in Figure 3). The DAR6 species contains a linker-payload at both inter heavy chain disulfide bonds [DAR6a], and therefore shows greater physical instability and propensity to aggregate. The above observation about greater impact on CH2 is in agreement with the HDXMS results of Pan et al. However, our results and the modeling data suggests that the relative perturbation in LC vs HC will be a function of the DAR species. The complete lack of impact on LC as seen by HDX-MS in Pan et al.12 may be due to the fact that the ADC pool was analyzed and the HDX-MS resolution is not high enough for detecting very minor differences. In contrast, the melting point of CH3/Fab domains of the ADC mixture or the isolated species of a specific DAR only decrease by a maximum of 2 °C after conjugation. The CH3 or Fab domains are destabilized to a lesser extent relative to the CH2 domain, which is due in part to the CH3 or Fab domains not containing any conjugation sites. HDX-MS analysis also confirms that intact interchain disulfides reduce conformational dynamics of local protein domains in the CH2 and CH2−CH3 interface of IgG1 mAbs but leave the Fab domains unchanged.12 The second transition is nonreversible especially upon the conjugation with the auristatin linker payload. Furthermore, structural changes at high temperatures such as 80 °C are typically accompanied by the formation of aggregates or precipitates, making it difficult to extract unfolding information from the second transition. Therefore, the data from thermal denaturation studies including thermal unfolding and aggregation experiments is limited to the first transition and thus to the CH2 region. To compensate for the irreversibility observed by thermal denaturation, chemical denaturation was performed to gain better insight into the unfolding dynamics of both transitions, and will be discussed later. Thermal Unfolding. Monitoring dye-based extrinsic fluorescence with SYPRO Orange under thermal stress conditions shows a similar trend observed by DSC. SYPRO Orange binds to hydrophobic residues or patches of a protein and spectral changes reflect changes in the environment around the probe. In an aqueous solution, the fluorescence emission intensity from the probe is negligible. However, in the presence of an unfolded protein, the fluorescence intensity increases significantly due to increased exposure to hydrophobic regions which the probe can bind to. Under thermal scanning, the less stable domain/protein will show an earlier onset of interaction with the probe and is reported as the hydrophobic interaction temperature (Th).16 Results are shown in Figure 4. At 5 °C, SYPRO Orange binds more to DAR6, possibly related to the presence of multimeric species even prior to thermal treatment as demonstrated by the isothermal chemical denaturation experiment discussed later in this study or simply the contribution from the hydrophobic payloads. Upon

Figure 4. Extrinsic fluorescence intensity profile as a function of temperature. (A) Sypro Orange fluorescence intensity profile. (B) DCVJ fluorescence intensity profile.

unfolding, a dramatic increase in fluorescence intensity is observed for all the samples (Figure 4A). Two temperaturedependent transitions are observed for the various molecules, indicative of two domains that have different conformational stabilities. The onset of hydrophobic interaction (Th) from the first transition follows the order: DAR6 < DAR4 < ADC < DAR2 < mAb. These results agree with the DSC results that conformational stability decreases with the increase in DAR. Thermal Aggregation. DCVJ belongs to a class of viscosity-sensitive fluorescent dyes, commonly referred to as molecular rotors.17−19 When DCVJ binds to aggregates, the dye becomes partially immobilized and is accompanied by an increase in fluorescence quantum yield.20,21 Studies have shown that DCVJ responds to the early stages of aggregation and, thus, has a strong preference for oligomeric aggregates.21 At time 0, the DAR6 solution shows that the highest initial DCVJ fluorescence signal is in agreement with SYPRO Orange observations. The fluorescence intensity decrease upon heating up to 40 °C is observed in all the solutions including DCVJ buffer solution which was not plotted in the figure, and thus could be attributed to the inherent property of the fluorophore.22 On heating the solution beyond 40 °C, DCVJ intensity increases and reveals at least one transition which coincides with the first transition detected by DSC and SYPRO Orange (Figure 4B). Again, the onset of aggregation follows order: DAR6 < DAR4 < ADC < DAR2 < mAb. The transition for the mAb is less obvious, indicating little to no aggregation detected in the mAb solution at the low concentration. DCVJ study gives a preliminary indication that DAR6 is more aggregation prone compared to DAR4, DAR2, and mAb. Aggregation Detected by SE-HPLC. Thermal incubation also provides confirmatory evidence that DAR6 aggregates the fastest and to greatest extent while parent mAb and DAR2 are relatively stable. Incubation at stress temperatures of 50 and 40 °C (Figure 5) shows that the initial rate of aggregation increases with DAR. DAR2 and DAR4 species are shown to have a much lower rates of aggregation than the DAR6. The latter formed over 76% HMMS in 9 h, while DAR4 formed 32% and almost no HMMS is detected in DAR2 in the same time frame (Figure 5). The onset of thermal unfolding and aggregation is approximately 42 °C for DAR6, which explains the significant increase in aggregate level observed at 50 °C and even at 40 °C. On the other hand, the onset of thermal unfolding and aggregation is at approximately 55 °C for DAR2 and 63 °C for mAb. Therefore, almost no aggregation is 607

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

C1/21 (midpoint of guanidine hydrochloride concentration required for complete unfolding) were generated from the first transition, while ΔG2 and C1/22 from the second transition after mathematical fitting to a three-state transition model with the instrument software.24 According to the literature, the transition occurring within the concentration range of 1−2 M guanidine hydrochloride is mostly related to a CH2 domain change.25 Any other transitions occurring beyond this concentration range can have potential contributions from multiple individual domains including the CH3 and Fab region. The mAb is mostly stable and shows the highest ΔG and C1/2. For the different DAR species, the data from the first transition correlates well with the observation from thermal denaturation. It takes more guanidine hydrochloride (higher C1/21) to denature DAR2, compared to DAR4 and DAR6 species. The apparent free energy of unfolding (ΔG1) also changes in the same order of DAR, that is DAR6 < DAR4 < DAR2. A smaller free energy barrier requiring lower guanidine hydrochloride concentrations enables the unfolding of the CH2 domain of DAR6 as opposed to DAR2 and DAR4. This is not surprising considering 0%, 50%, and 100% intact hinge disulfides remain for DAR6, DAR4, and DAR2 respectively. DAR6 also exhibits the lowest onset for denaturation. Data from the second transition, on the other hand, shows higher ΔG2 and C1/22 for DAR6 compared to DAR2 and DAR4. We were intrigued by these results and did extensive literature research. The latest publication from Schön et al. suggests that ΔG could change if the molecule forms native state self-association or denatured state aggregates. The publication also offers a method to test this hypothesis by determining the dependencies of ΔG as a function of protein concentration using ICD.26 In the absence of aggregation or self-association, the equilibrium of monomeric protein between the native and denatured states is independent of the protein concentration. The situation is different if aggregation occurs, since aggregation is a concentration dependent association phenomenon. In the presence of native protein self-association, the native state is stabilized. The degree of native protein self-association is concentration dependent, and hence the apparent ΔG will increase as the protein concentration is increased. Similarly in the case of denatured protein aggregation, the denatured state is stabilized, and hence the apparent ΔG will decrease as the protein concentration is increased.26 Using the same approach as elaborated by Schön et al.,26 we determined ΔGs as a function of protein concentration for DAR2, DAR4, and DAR6 species (Figure 7). For DAR2 and DAR4, no concentration dependency of ΔG (Figure 7A) was

Figure 5. Aggregation profiles of IgG1 mAb, ADC pool, DAR2, DAR4, and DAR6 species at (A) 50 °C and (B) 40 °C.

observed for both samples at 50 °C for 24 h and 40 °C for 2 weeks. Isothermal Chemical Denaturation. Isothermal chemical denaturation (ICD) provides a quantitative measure of protein conformational stability. Unlike thermal denaturation which can lead to irreversible unfolding, chemical denaturation is mostly reversible and can be employed at relevant temperatures, and it is commonly observed that proteins do not aggregate in solutions that contain denaturant at concentrations that result in complete unfolding.23 In order to evaluate the unfolding behavior without the complication of aggregation, isothermal chemical denaturation study was performed by adding various concentrations of guanidine hydrocholoride from 0 to 5.5 M to alter protein conformation such that the protein structure ranged from native to partially to fully unfolded. Equilibrium denaturation curves for mAb, ADC, DAR2, DAR4, and DAR6 were measured (Figure 6) and analyzed

Figure 6. Equilibrium unfolding curves for IgG1 mAb, ADC pool, DAR2, DAR4, and DAR6 species.

(Table 1). Two transitions were observed for all the samples. The parameters, ΔG1 (Gibbs free energy of unfolding) and Table 1. Thermodynamic Parameters Generated from Fitting the Equilibrium Unfolding Curvesa ΔG1 (kCal/ mol) ΔG2 (kCal/ mol) C1/21 (M) C1/22 (M) a

mAb

ADC

DAR2

DAR4

5.8, 5.8

5.1, 4.8

6.4, 6.2

3.6, 3.8

2.82, 3.15

11.5, 11.4 6.4, 6.4

6.5, 6.4

5.5, 5.7

7.80, 7.43

1.75, 1.82 3.15, 3.24

1.72, 1.76 3.58, 3.57

1.86, 1.83 1.72, 1.63 1.87, 1.80 3.42, 3.40 3.15, 3.15 3.21, 3.19

DAR6

Figure 7. Dependency of ΔG as a function of protein concentration for (A) DAR2, DAR4, and (B) DAR6. (Each data point is an average of two measurements.)

Results from two measurements are presented. 608

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

Figure 8. Hydrophobicity assessment by (A) ANS fluorescence intensity profile, (B) ANS binding rate constant, and (C) surface tension.

detected. However, for DAR6, ΔGs decrease with concentration and the change is more substantial for ΔG2 (Figure 7B). This implies that the perturbed structure of DAR6 species results in formation of multimers, possibly through reversible weak association. This conclusion is supported by visual observation of the DAR6 solution which was slightly cloudy at 0.3 mg/mL but turns increasingly clear on dilution to 0.1 mg/mL. When the DAR6 sample is diluted into the SEC mobile phase (0.02 M phosphate buffer with 0.4 M sodium chloride), the solution also becomes clear, but if added to phosphate buffer without salt, the solution turns cloudy. This shows the reversibility of the weak association, and helps to explain why the measured DAR6 HMMS level at time 0 is very low (Figure 5). Hydrophobicity. Isothermal chemical denaturation reveals the possible presence of multimers for DAR6 at ambient temperature. The hydrophobicity of various DAR species was subsequently assessed in order to evaluate if there is any connection between physical state and physical properties. The fluorescent dye ANS (8-Anilino-1-naphthalene sulfonic acid) is known to have an affinity for the apolar regions of proteins and has been used to compare hydrophobicity of the molecules. ANS by itself exhibits weak fluorescence in the aqueous solution. When the dye binds to the hydrophobic portion of the protein molecule, its fluorescence quantum yield increases. ANS fluorescence in the mAb solution is weak (Figure 8A). DAR2 shows stronger fluorescence intensity while DAR4 and DAR6 increase fluorescence intensity further to 250 A.U. and 500 A.U. (Figure 8A). The trend holds true with ANS binding rate constant determined from the stopped-flow measurements. The binding (rate) of mAb with ANS is negligible. However, the binding rate constant increases with DAR and almost doubles when DAR increases from 4 to 6 (Figure 8B). The ANS data in Figure 8A,B reflects the increase in both the number of hydrophobic payload as well as the perturbation in conformation on conjugation. The steeper change in ANS binding data between DAR4 and DAR6 reflects change from monomeric DAR4 to multimers of DAR6. Fluorescence intensity and ANS binding rate constant for ADC mixture fall between the DAR4 and DAR6 values. [Note: A control experiment using a nonconjugated mixture of mAb with linkerpayload could not be performed due to the very limited aqueous solubility of linker-payload in buffer.] The impact of a solute on the surface tension of its aqueous solution is a good measure of its hydrophobicity since surface tension is lowered by the migration of the solute molecules to the air−water interface. The higher the hydrophobicity, the larger the concentration at the interface and the greater the

drop in the surface tension. Steady state surface tension of the various species was measured and shows a similar trend as hydrophobicity measured by ANS binding (Figure 8C). The surface tension of buffer is 71.9 ± 0.6 mN/m. mAb reduces surface tension to 67.2 ± 0.3 mN/m. DAR2 and DAR4 decrease the surface tension further to 65.1 ± 0.3 and 58.4 ± 0.3 mN/m, respectively. The surface tension of ADC mixture (63.1 ± 0.07 mN/m) is between DAR2 and DAR4. However, the most significant surface tension drop occurs with DAR6 at 28.6 ± 0.1 mN/m. It has been previously hypothesized and confirmed by surface tension analysis that monoclonal antibodies are surface active and can form a thin protein layer at the air−water interface.27 MAbs at the interface expose their hydrophobic core to air leading to unfolding and multiple nonspecific intermolecular interactions. DAR6 with its higher payload is expected to be the most hydrophobic among these samples. The significant drop in surface tension for DAR6 supports the conclusion about its higher hydrophobicity, which also explains the reduced colloidal stability and consequently greater tendency to form multimers in solution. Molecular Models of ADCs at Different DAR Loads Structural Consequences of Conjugation: Perturbation of CH2 Domain Quaternary Structure. To probe structural consequences of conformational destabilization in ADCs, the ADC molecular models at different DAR levels were superposed onto the molecular model for the parent IgG1 mAb via structural alignments. Only the protein portions of these models were superposed. The alignments were performed both globally (whole structures) and at the level of individual heavy and light chains. The Cα atom Root Mean Square Deviation (Cα-RMSD) values were computed in each case. Table 2 shows the results from these calculations. It can be seen that structural perturbations in the mAb structure caused by the conjugation are not proportional to the DAR load. A visual examination of the superposed models revealed that the conformational perturbations in the light chains L and M are restricted to the conjugation sites at the C-termini of these chains (Table 2). Interestingly, the heavy chain H shows the largest deviation for each ADC model superposed. Within the heavy chain H, the conformations of hinge region (sites of conjugation) and CH2 domain are perturbed to the greatest extents. Since the hinge region contains the conjugating Cys residues, large deviations in its conformation are expected. However, structural perturbations observed for CH2 domain, which is further down along the amino acid sequence of a heavy chain, suggest propagation of conformational destabilization to antibody structural regions other than the immediate neighborhood of the conjugation sites. Further examination of the surperposed structures suggested that structural perturbations are limited to 609

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

the changes in quaternary structure formed by the two CH2 domains in an antibody molecule. The conformations of the individual CH2 domains are not perturbed. Similar observations on impairment of CH2 domain quarternary structure due to movement of the two CH2 domains were also made in a molecular dynamics study that explored the consequences of disulfide breakage and scrambling in an IgG2 antibody.28 The HDX-MS studies on the interchain cysteine-linked IgG1 ADC by Pan et al.12 have subsequently confirmed the above28 insights that it is the loss of disulfide bonds that causes the most local conformation changes. The conjugation and the payload (whether hydrophobic or hydrophilic) do not cause any further signficant impact. The change in quaternary structure of the CH2 domains due to conjugation depends upon both DAR load and location of the conjugation site. The distance between two CH2 domains was measured by computing the distance between the Cα atom positions of Asn 299 residues (glycosylation sites) in each CH2 domain. Table 2 shows that this distance increases with the DAR load and the CH2 domains in DAR 8 ADC model are further apart by >2 Å in comparison to the parent mAb. The increases in CH2 domain distance are smaller when the conjugation sites fall in the upper hinge region (see the values for DAR2a_1 and DAR4b models) involving breakage of the disulfide bonds between light and heavy chains. If the conjugations involve the heavy chain Cys residues in the lower portion of the hinge region, the changes the CH2 domain distance are larger (Table 2). This observation can be rationalized by noting that the Cys residues in the lower hinge are closer to the CH2 domain. Overall, these results help rationalize the decrease in melting termperature for CH2 domains and early onset of aggregation in ADCs studied here (Figure 3). Molecular Models of ADCs at Different DAR Loads Increased Hydrophobicity of the ADCs Comes from the LP Molecules. Conjugation changes the molecular surface properties of the parent IgG1 mAb. Figure S2 shows solvent exposed polar (pink) and nonpolar (green) surfaces on parent mAb and ADCs at different DAR levels. Qualitatively this figure shows that the amount of solvent accessible nonpolar surface in an ADC increases with its DAR load. To uncover the molecular origins of increase in solvent accessible nonpolar surfaces on ADCs as compared to the parent IgG1 mAb, accessible surface area (ASA) calculations were performed and the results are summarized in Table 3 and Table 4. Table 3 shows that nonpolar and polar surface areas of the parent mAb accessible to solvent decrease when it becomes part of an ADC. The decrease in nonpolar ASA of the parent antibody is correlated with the DAR loads in the ADCs. On the other hand, a similar relationship between decrease in polar

Table 2. . Computed Structural Perturbations in the Parent mAb upon Conjugations ADC model superposed on to parent mAb Parent mAb DAR2a_1 Conformer DAR2a_1 Conformer DAR2a_1 Conformer DAR2a_1 Conformer DAR2a_1 Conformer DAR4a Conformer DAR4a Conformer DAR4a Conformer DAR4a Conformer DAR4a Conformer DAR4b Conformer DAR4b Conformer DAR4b Conformer DAR4b Conformer DAR4b Conformer DAR6a_2 Conformer DAR6a_2 Conformer DAR6a_2 Conformer DAR6a_2 Conformer DAR6a_2 Conformer DAR8 Conformer DAR8 Conformer DAR8 Conformer DAR8 Conformer DAR8 Conformer

RMSD (Å)

number of residues superposed per molecule or chain

H:N299− K:N299 distance (Å)

All chains

0 4.730

1332

25.29 25.80

Heavy chains H

2.210

448

Light chains L

0.768

218

Heavy chains K

0.792

448

Light chains M

0.711

218

All chains

3.640

1332

Heavy chains H

2.150

448

Light chains L

1.176

218

Heavy chains K

1.048

448

Light chains M

1.363

218

All chains

4.192

1332

Heavy chains H

4.706

448

Light chains L

0.814

218

Heavy chains K

1.702

448

Light chains M

0.538

218

All chains

3.906

1332

Heavy chains H

4.248

448

Light chains L

0.752

218

Heavy chains K

1.501

448

Light chains M

1.247

218

All chains

2.465

1332

Heavy chains H

1.982

448

Light chains L

0.647

218

Heavy chains K

1.731

448

Light chains M

0.977

218

polypeptide chains superposed

1 1 1 1 1 25.82

1 1 1 1 1 26.29

2 2 2 2 2 26.54

2 2 2 2 2 27.31

4 4 4 4 4

Table 3. Nonpolar and Polar Accessible Surface Areas in the Most Stable Conformers for Different ADC Models1 parent antibody 2

LP molecules 2

2

whole ADC molecule 2

ADC model

ASAnp (Å )

ASAp (Å )

ASAnp (Å )

ASAp (Å )

ASAnp (Å2)

ASAp (Å2)

DAR2a_1 Conformer1 DAR4a Conformer 1 DAR4b Conformer 2 DAR6a_2 Conformer 2 DAR8 Conformer 4

27540.6 27037.6 27125.9 26954.6 26398.9

21686.9 21478.5 21745.0 20803.5 20871.3

767.2 1607.8 1868.7 2061.2 3079.6

162.0 295.8 410.4 491.0 642.6

28307.8 28645.4 28994.6 29015.8 29478.5

21848.9 21774.3 22155.4 21294.5 21513.9

a

ASAnp stands for nonpolar accessible surface area (ASA) and ASAp stands for polar ASA. Note that antibody conformations may be slightly different in different conformers of the ADC models for the specific DAR. All LP molecules were grouped together for the purpose of ASA calculations. 610

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Article

Bioconjugate Chemistry

developed to mitigate the above-mentioned challenges, a number of ADC candidates currently in advanced drug development stages still utilize conventional non-site-specific conjugation technologies. Therefore, it is important for researchers engaged in drug product development to fully understand the molecular basis of instability and drug product heterogeneity in such ADCs. Some of the experiements in this report were carried out at stress conditions (40 and 50 °C) in the liquid state. For biotherapeutic products formulated in the liquid state, such stress studies are a quick way to assess viability of long-term storage under refrigerated conditions. ADCs have (to date) primarily been formulated in the lyophilized state where liquid state stability is less of an issue. On the other hand, it is not unusual for the conjugation processes to be carried out under conditions approaching these stress conditions, e.g., 30−40 °C (see e.g., Sun et al.;7 Strop et al.29). However, the general value of stress studies, irrespective of the final product presentation or process condition, lies in their ability to help elucidate the relationship between structure and property (exemplified by this work), as well as to gain product and process knowledge and understanding. Regulatory bodies require such studies as part of the dossier for formulation justification, stability, and comparability. It is known that this interchain cysteine conjugation chemistry leads to a range of DARs. The 2-D LC-MS analysis shows that there is a preference for certain isomers within the various DARsnot all the theoretical isoforms are actually formed. This suggests that there are steric/conformational and energetic factors involved in the conjugation process that remain to be elucidated. Molecular modeling and dynamic simulations can be useful in finding answers to this question. While some ADC isomers may not be feasible due to steric hindrance, others may exist transiently over the course of the conjugation reaction, if they serve as intermediates for higher DAR loading. A time course 2-D LC-MS analysis of different ADC species as the conjugation reaction proceeds would be required to detect such intermediates. The insights gained from molecular modeling and time course experiments will offer ways to control the distribution of ADC species during the conjugation process. This study has explored the impact of drug load and its distribution on higher order structural properties and related aggregation behavior of an antibody drug conjugate made by conventional cysteine chemistry. In general, protein aggregation can arise from conformational instability as well as colloidal interactions.30−32 In the case of the ADC studied here, both of these sources of aggregation are present to a greater extent with increasing DAR. Thermal analysis (DSC) as well as chemical denaturation results show decreased conformational stability (primarily at the CH2 domain) for the conjugated species, with the level of destabilization directly related to DAR level. When examined from the perspective of properties that govern colloidal stability, the general trend indicates that the higher the DAR, the more hydrophobic the species. There is a gradual and steady change in these properties from DAR 2 to DAR 4. The trend is however amplified when moving from DAR 4 to DAR 6 with the doubling of the ANS binding rate constant and halving of the surface tension in the solution. Additionally, the substantially high initial values for Sypro Orange and DCVJ fluorescence also support the hydrophobic nature and the likely multimeric state of DAR 6 even at 2−8 °C. Such multimers

Table 4. Nonpolar and Polar Accessible Surface Areas of Antibody and LP Molecules That Get Buried upon Formation of the ADCs1 antibody ADC DAR2a_1 Conformer1 DAR4a Conformer 1 DAR4b Conformer 2 DAR6a_2 Conformer 2 DAR8 Conformer 4

LP molecules

ΔASAnp (Å2)

ΔASAp (Å2)

ΔASAnp (Å2)

ΔASAp (Å2)

number of LP molecules

826.3

535.6

1401.5

493.3

2

1508.5

1141.9

2413.2

767.0

4

1352.4

978.7

2088.9

574.6

4

2353.4

1714.1

3506.7

1159.1

6

2424.6

1879.5

3337.3

1257.3

8

See legend to Table 3 for explanation of symbols. Δ denotes the change in different ASAs.

a

ASA of the antibody and the DAR load was not observed (Table 3). The drop in nonpolar ASA of antibody is ably compensated by the LP molecules, which expose four to six times greater amounts of nonpolar surface areas than polar ones at every DAR load. In the end, the overall nonpolar ASA of the whole ADC molecule becomes greater than that of antibody alone by approximately 3−12% (Table 3). In each ADC model, LP molecules form both hydrophobic and electrostatic contacts with several amino acid residues in the parent antibody. Table 4 summarizes the results obtained by computing the changes in polar and nonpolar ASAs of the antibody and the LP molecules when they are part of an ADC with reference to their corresponding ASAs alone. These are the molecular surfaces that get buried when antibody and LPs interact. It can be seen that the LP molecules interact with the parent mAb by burying both polar and nonpolar surfaces. On average, the antibody contributes comparable amounts of polar and nonpolar surfaces for interaction with the LP molecules. However, interaction of the LP molecules with the antibody is predominantly hydrophobic in nature. At each DAR level, the LPs bury two to three times more nonpolar surface area than the polar surface area. Despite this, the amount of nonpolar surface area of LP molecules that remains solvent accessible is greater than their polar surface area (Table 3), pointing to the strongly hydrophobic character of the payload drugs. In summary, the above observations on molecular structures of the ADCs help explain the experimental observations on increased hydrophobicity and aggregation tendencies of the ADCs at every DAR load (Figure 4 and Figure 5).



DISCUSSION The promise of bioconjugation has been long recognized in the field of biotherapeutic drugs and vaccines. In recent years, the availability of improved methods to produce, purify, and analytically characterize the conjugated biologics has renewed our interest in these therapeutics. Antibody drug conjugates combine the desirable attributes of the antibodies (selectivity, long circulation half-life) with those of small molecule drugs (potency, cytotoxicity) to yield highly potent anticancer therapeutics. However, ADCs also present greater drug product development challenges because of reduced physicochemical stability and increased product heterogeneity, in comparison to their parent monoclonal antibodies. While several site specific conjugation technologies with well-defined DAR are being 611

DOI: 10.1021/acs.bioconjchem.5b00603 Bioconjugate Chem. 2016, 27, 604−615

Bioconjugate Chemistry



could easily break up upon dilution during the analysis by SEC which would result in low initial T0 values of HMMS. The insights generated here show that the strategy for formulation of the ADC will need to address both the conformational (reduced structural stability) and colloidal (increased hydrophobicity) aspects. The DAR for conventional cysteine conjugation can range from 2 to 8 (preferentially even numbers),7 i.e., multiple DAR species are present in the drug substance mixture after purification. If the high DAR species form multimers due to their colloidal instability, even at low concentrations under unstressed conditions, these higher DAR species likely drive the instability of the overall pool, and therefore should become the focus of the process and formulation efforts. Formulation would likely require the presence of structure stabilizers such as sucrose (increase free energy of unfolding) in a low ionic strength buffer (increase second virial coefficient) to provide stabilization against aggregation.33 The enhanced hydrophobicity will also make the ADC more susceptible to aggregation under interfacial stresses such as agitation and require the addition of surfactants to ameliorate this concern.34 Clearly, the hydrophobicity of the linker payload itself has a strong contribution to this aggregation behavior and changing to a hydrophilic linker− payload will reduce this colloidal instability, even if the conformational perturbation does not differ between the two types of payloads.

Article

MATERIALS AND METHODS

Materials. The IgG1 parent mAb, ADC mixture (conventional cysteine chemistry with vc-auristatin as the linkerpayload; DARs ranging 0−8) were produced by Bioprocess R&D at Pfizer Inc. (Pearl River, NY). DAR2, DAR4, and DAR6 species used in this study were isolated by preparative hydrophobic interaction chromatography. The DAR8 species could not be studied since it immediately precipitated on isolation. The free auristatin levels in the ADC samples used in this study were below detection limit (