Profiling Antibody Drug Conjugate Positional Isomers: A System-of

Aug 22, 2012 - Profiling Antibody Drug Conjugate Positional Isomers: A System-of-Equations Approach ... *E-mail: [email protected]. ... inputs f...
12 downloads 11 Views 969KB Size
Download PDF
Article pubs.acs.org/ac

Profiling Antibody Drug Conjugate Positional Isomers: A System-ofEquations Approach Lan N. Le,† Jamie M. R. Moore,† Jun Ouyang,‡ Xiaoying Chen,‡ Mary D. H. Nguyen,‡ and William J. Galush*,† †

Early Stage Pharmaceutical Development and ‡Protein Analytical Chemistry, Genentech, Inc., 1 DNA Way, South San Francisco, California 94080, United States S Supporting Information *

ABSTRACT: Antibody drug conjugates enable the targeted delivery of potent chemotherapeutic agents directly to cancerous cells. They are made by the chemical conjugation of cytotoxins to monoclonal antibodies, which can be achieved by first reducing interchain disulfide bonds followed by conjugation of the resulting free thiols with drugs. This process yields a controlled, but heterogeneous, population of conjugated products that contains species with various numbers of drugs linked to different former interchain disulfide cysteine residues on the antibodies. We have developed a mathematical approach using inputs from capillary electrophoresis and hydrophobic interaction chromatography to determine the positional isomer distribution within a population of antibody drug conjugates. The results are confirmed by analyzing isolated samples of specific drug-to-antibody ratio species. The procedure is amenable to rapid determination of positional isomer distributions and features low material requirements. A survey of several antibody drug conjugates based on the same IgG framework and small molecule drug combination has shown a very similar distribution of isomers among all of the molecules using this technique, suggesting a robust conjugation process.

T

phase clinical studies for HER2-positive metastatic breast cancer. These two examples are emblematic of the promise of antibody drug conjugates: the former takes an antibody with little clinical efficacy on its own16 and creates a powerful treatment through the addition of a drug payload,17 while the latter further empowers a successful, existing therapeutic antibody.18 There are multiple ways to make such ADCs. One type utilizes the eight cysteines of the readily reducible heavy chainheavy chain and heavy chain-light chain disulfides and links them to thiol-reactive forms of a potent cytotoxic drug. The less reactive intrachain disulfides stay unreduced, and the antibody structure remains intact. An example of this class uses a thiolreactive maleimidyl group attached to the toxin monomethyl auristatin E (maleimidocaproyl-valine-citrulline-p-aminobenzyloxycarbonyl-monomethyl-auristatin E, hereafter called vcMMAE.)13 The conjugation process creates a diverse population of antibodies with different numbers of drugs per antibody, but optimized conjugation parameters using this chemistry can accurately and reproducibly control the average number of small molecule drugs per antibody, also known as the drug-to-antibody ratio (DAR). The population of drug-toantibody species may be easily measured by techniques such as

he idea of arming antibodies with cytotoxic agents to form antibody drug conjugates (ADCs) was first demonstrated decades ago.1 This class of molecules seeks to achieve the targeted delivery of a drug payload directly to cancerous cells, resulting in a larger therapeutic window compared to the chemotherapeutic agent alone. Early efforts along these lines failed to lead to successful new drugs; however, since the practical realization of ADCs required extensive development of the antibodies themselves, the drugs to be conjugated to them, and the linkers that bring them together. Until the advent of chimeric,2 humanized,3,4 and fully human5 recombinant monoclonal antibodies (mAbs),6 early attempts at the therapeutic use of murine molecules were often confounded by their immunogenicity and short half-lives in humans.6−8 Similarly, the first attempts to attach standard chemotherapeutic agents to antibodies suffered from the fact that the drug toxins were not potent enough to work well in an antibody-targeted fashion.9,10 More toxic compounds were required.11 Other studies showed the importance of linker stability in circulation and subsequent release upon cellular uptake.12−14 Today, these issues appear to be substantially solved, and many modern ADCs are in development.12,15 Notably, brentuximab vedotin (formerly SGN-35, from Seattle Genetics) has recently been approved by the U.S. Food and Drug Administration (FDA) for Hodgkin lymphoma. Another ADC, trastuzumab emtansine (T-DM1, from Genentech), is in late © 2012 American Chemical Society

Received: June 9, 2012 Accepted: August 4, 2012 Published: August 22, 2012 7479

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

hydrophobic interaction chromatography,19 and higher drug-toantibody ratio species are observed to be more cytotoxic in vitro and in vivo.20 However, such measurements do not address the varied physical distribution of drugs across the antibodyin other words, the positional isomerswhich have been shown to modulate pharmacokinetics and activity for related ADCs.21 We have examined whether it is possible to measure the entire positional isomer distribution of vcMMAE within a population of antibodies using routine analytical techniques. For instance, an ADC with two drugs may display them either at the hinge region or on the Fab arms. Other permutations are possible for drug-to-antibody ratios of 4 and 6. The full range of possible positional isomers is shown schematically in Figure 1,

output to preparatively separated samples and evaluate alternative systems-of-equations that take into account RPbased separation. Method precision is evaluated by simulating noise in the inputs, giving an in silico assessment of the sensitivity of the mathematical system to data variability. This approach features small material requirements (99% pure by this method, and DAR6 material was 95% pure, with the major contamination coming from DAR8 molecules.

■

RESULTS AND DISCUSSION Below, we consider multiple combinations of HIC, CE-SDS, and RP that could potentially be used to determine the abundance of the positional isomers. Regardless of the combination, the same basic set of relationships between assays are used: the relative distribution of specific drug-to-antibody ratio species as provided by HIC (0, 2, 4, 6, or 8 drugs per antibody) and a detergent- or organic-mediated dissociation profile of each of the drug-to-antibody ratio species as provided by CE-SDS or RP. The relationships between the outputs of the assays are written as systems-of-equations, as considered below. Of the three systems discussed below, we find two that give satisfactory results but one is simpler to execute than the other. System 1: HIC and CE-SDS. One method to determine positional isomer distributions uses HIC and CE-SDS. The HIC chromatogram (Figure 2A) provides information on the relative distribution of different number of drugs per antibody but not their location. However, the way an ADC dissociates in the presence of SDS detergent in the CE-SDS assay (Figure 2B) is a function of how many drugs are linked to the antibody and where they are attached, since the presence of the drug prevents reformation of an interchain disulfide bond. As shown in Figure 3, the different positional isomers will dissociate to the various fragments, or CE-SDS peaks, depending on the position of the conjugated drugs. This assumes that the conjugation occurs in pairs (i.e., there are no single labeled cysteines), which is supported by the low levels of oddnumbered drug-to-antibody ratio species in the HIC data (Figure 2A). The small, odd-numbered peak areas were integrated with the adjacent, higher even-numbered drug-toantibody species. When present, odd-numbered drug-toantibody ratio species should dissociate in SDS in the same manner as if both cysteines were conjugated. Note that it is not possible to distinguish between mAbs conjugated at the upper hinge disulfide with those conjugated at the lower hinge based on dissociation in the CE-SDS assay. Previous reports indicate that the upper hinge disulfide of IgG1κ antibodies is more susceptible to reduction by DTT and labeling by maleimide moieties.28 There is likely to be a similar situation with the molecules presented, though the reductant used here (TCEP) has been observed to give slightly different positional isomer distributions than DTT.20 Therefore, a different hinge disulfide reduction preference cannot be ruled out. Since the totals of each column of Figure 3 represent the results from CE-SDS, we can write six equations to relate HIC data to the CE-SDS data with the various positional isomers as unknowns (eqs 1.1−1.6). The right-hand side of these six equations need to be weighted by the number of lysines to account for the fact that the FQ dye labels via lysine residues (essentially the same as weighting by the calculated molecular weight of each fragment). Since the left-hand sides of eqs 1.1−1.6 sum to unity, the right-hand sides must be renormalized after weighting for dye labeling. The numerators and denominators on the right-hand side correspond to the

Figure 2. HIC, CE-SDS, and RP chromatograms of mAb1. (A) Chromatogram of the HIC assay showing separation of the peaks based on drug-to-antibody ratio. (B) Electropherogram of the CE-SDS assay where the ADC dissociates in the presence of SDS based on where the drugs are linked to the antibody. (C) The RP chromatogram resolving fragments of the fully reduced ADC showing light chains with 0 or 1 drug and heavy chains with 0−3 drugs. Shoulder peaks are integrated with their nearest neighbor for parts B and C.

intact and fully reduced antibody peaks. Raw peak areas were normalized by migration time into fractional corrected peak areas prior to further analysis.26 Peak elution order was assumed to follow increasing molecular weight between L and mAb.25,26 Labeling of antibodies and their fragments was assumed to be proportional to the number of lysine residues, as has been seen with other proteins.27 Analysis of ADCs by Reversed Phase HPLC (RP). The reversed phase-based separation of reduced ADC fragments was performed on an Agilent 1200 series HPLC (see the Supporting Information). The peaks were detected at 280 nm and eluted in the following order: light (L0), light + 1 drug (L*), heavy (H0), heavy + 1 drug (H*), heavy + 2 drugs (H**), and then heavy + 3 drugs (H***) as shown in Figure 2C. Peak assignments were made using reduced, unconjugated antibody for L0 and H0 and the remaining peaks assigned based on elution time in order of increasing hydrophobicity due to the linker-drug moiety. Similar to the HIC peaks, the RP peaks were corrected for vcMMAE absorbance at 280 nm depending on the number of drugs attached. Collection of Purified DAR Species. Purified DAR2, 4, 6, and 8 species were isolated by fraction collecting HIC eluent. 7481

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

L = (DAR2f + 2· DAR4ff + DAR4fh + 2· DAR6ffh LysL + DAR6fhh + 2·DAR8) · 2·LysH + 2·LysL

(1.6)

1=

DAR2f + DAR2 h DAR2

(1.7)

1=

DAR4ff + DAR4fh + DAR4 hh DAR4

(1.8)

1=

DAR6ffh + DAR6fhh DAR6

(1.9)

LysL = number of lysines on light chain, LysH = number of lysines on heavy chain. These nine equations make up a linear system-of-equations that is analyzed using the lsqnonneg algorithm in MATLAB (R2010b, The Mathworks), which determines the positional isomer distribution (the variables in bold) by performing a least-squares minimization among the possible solutions and constraining variable values to be greater than or equal to zero. Note that DAR0 and DAR8 are directly measured by HIC. The system is overdetermined, meaning that there are potentially multiple equally good solutions. However, replicate computations using the same inputs always return the same output, suggesting that this algorithm finds only one best solution. The values calculated are shown in the first column of Table 1. Table 1. Positional Isomer Distribution for mAb1 Determined Using Four Different Approachesa abundance (%)

Figure 3. System 1 schematic showing the various positional isomers and how they dissociate into fragments in the CE-SDS assay based on the position of the linked drug. The drug-to-antibody ratio is provided by the HIC assay, while the sum of each fragment column is provided by CE-SDS.

weighting and normalization factors, respectively. It can be shown algebraically that the right-hand side of eqs 1.1−1.6 sum to unity, given that all the DAR species fractions do as well. We can also write three additional mass balance equations for the different drug-to-antibody ratios of 2, 4, and 6. Thus, the following nine equations are used to analyze data from HIC and CE-SDS and are referred to as System 1: mAb = (DAR0 + DAR2 h) ·

2·LysH + 2·LysL 2·LysH + 2·LysL

(1.2)

2·LysH HH = (DAR4ff + DAR6ffh) · 2·LysH + 2·LysL

(1.3)

HL = (2·DAR4 hh + DAR6fhh) · H = (DAR6fhh + 2·DAR8) ·

LysH + LysL 2·LysH + 2·LysL

LysH 2·LysH + 2·LysL

system 1

system 1*

CE-SDS*

after Sun et al.*

DAR0 DAR2f DAR2h DAR4ff DAR4fh DAR4hh DAR6ffh DAR6fhh DAR8

5.1 29.5 1.4 33.3 2.8 13.8 0.0 11.5 2.4

5.1 27.7 2.5 30.3 5.7 13.8 1.5 10.0 2.4

5.1 26.1 4.8 30.4 5.0 14.3 0.6 10.9 2.4

5.1 29.0 1.9 30.1 5.0 14.9 1.1 10.4 2.4

a

Asterisks indicate samples prepared from fraction-collected material. The first column shows the positional isomer amounts using System 1 with unfractionated mAb1. The second column shows the positional isomer distribution reconstructed from isolated 2, 4, and 6 drug-toantibody ratio fractions individually analyzed using System 1. The third column shows positional isomers reconstructed from the isolated fractions by associating each positional isomer to unique CE-SDS fragments. The last column shows positional isomers calculated using isolated fractions analyzed by reversed-phase HPLC and CE-SDS according to the method of Sun et al.20.

(1.1)

2·LysH + LysL 2·LysH + 2·LysL

HHL = (DAR2f + DAR4fh) ·

isomer

A consistent result is the low abundance of species where only one of the two hinge disulfides is conjugated (DAR2h, DAR4fh, DAR6ffh). In fact, a common output is for the best solution for System 1 to include no DAR6ffh at all. Although we do not believe that this isomer is completely absent, it may be present in very small quantities that are lost in the analysis. Because of the importance of the input data, we use values that are corrected to the best of our scientific knowledge of these assays (see the Experimental Section). However, it is possible that a slight mismatch between HIC and CE-SDS detection or

(1.4)

(1.5) 7482

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

pathway, so this assumption remains unassessed. Related to this, it is imperative that we have the most accurate information for each peak inputted into the system-of-equations. Thus, the HIC data is corrected for the differing vcMMAE absorbance contribution for each drug-to-antibody ratio species. Currently, we do not have an orthogonal method to HIC for the direct assessment of the drug distribution. However, the average drugto-antibody ratio calculated from HIC data agrees well with that calculated by RP-HPLC as discussed in the Supporting Information (both are 3.5 drugs/antibody for mAb1), supporting that the HIC data accurately reflects the abundance of the different drug-to-antibody ratio species in a given sample. We can also assess the results from the solved system-ofequations by entering the resulting positional isomer values back into the equations to determine what the input values would be from CE-SDS. In other words, we compute a selfconsistency check with the positional isomer results of the solver and HIC data values into the right-hand side of eqs 1.1−1.6 and solve for the CE-SDS values, or left-hand side. The measured CE-SDS values from the assay match within a few percent of the calculated CE-SDS values, as expected. Furthermore, using the results from System 1, we can take the entire positional isomer population and calculate the theoretical results of a reversed phase-based separation, which provides information about the distribution of drugs on the heavy and light chains after reduction of the antibody, as seen in Figure 2C. The theoretical RP peak areas match the experimental values (also corrected for vcMMAE absorbance) to within 3% (Table S-1 in the Supporting Information). This gives additional confidence in the accuracy of the positional isomer values using this method. System 2: HIC and RP. Given that the results of the HIC and CE-SDS system predict the reversed-phase fragmentation profile of a sample, it might be expected that a similar systemof-equations relating HIC and RP could also define the positional isomer distribution. Like the HIC and CE-SDS system, this system mathematically relates the drug-to-antibody ratio distribution from HIC to the amount of reduced light or heavy chains and attached drugs provided by RP. Once reduced, the various isomers result in two light chains with 0 or 1 drug (L0, L*, respectively) and two heavy chains with 0, 1, 2, or 3 drugs (H0, H*, H**, H***, respectively), depending on the isomer (Figure S-1 in the Supporting Information). The data is corrected for vcMMAE absorbance at 280 nm based on the number of drugs attached. Since the number of light and heavy chains are stoichiometrically equal, the light (L0, L*) and heavy (H0, H*, H**, H***) are both considered separately and normalized by molarity such that the total light chain and total heavy chain each amount to 50%. Since each antibody gives four fragments, the right-hand side of the system-of-equations needs to be normalized by a factor of 4, to account for each mAb in HIC occurring four times in RP. The same three mass balance equations from System 1 still apply. These equations, System 2, are found in the Supporting Information. This system-of-equations can be evaluated using the same MATLAB solver as the HIC and CE-SDS method with the results for mAb2 shown in the second column of Table 2. Again, there is a large preponderance of species labeled in the Fab regions and double-labeled in the hinge region. The results differ, however, in the abundances of single hinge conjugated species. Solutions to System 2 consistently contain no DAR4fh but do have small amounts of DAR6ffh, a reversal of the results seen for

quantification or an inefficiency in the sample dissociation scheme shown in Figure 3 leads to a solution where all DAR6 species are forced to DAR6fhh. Evidence of this is that the lefthand side of eq 1.5 is sometimes observed to be slightly greater than the right-hand side for a set of HIC and CE-SDS inputs, even though DAR6fhh is set to 100%. Verification of Accuracy. The accuracy of the system-ofequations approach can be assessed by comparing its results to data generated by preparatively isolating drug-to-antibody ratio 2, 4, and 6 fractions from a single parent batch of material using HIC. These fractions are analyzed to determine the relative abundance of each isomer within the purified sample, which may be done by solving System 1 for each fraction, as performed above. The isomer distribution of the parent material can be recreated by recombining and properly weighting the data from the HIC-purified fractions. The isomer distribution determined by this method for mAb1 shows good agreement with that calculated using unfractionated mAb1 material, as shown in the first two columns of Table 1. The isolated drug-to-antibody ratio species may also be analyzed for their positional isomer content without using a simultaneous system-of-equations. Instead, we can utilize the fact that within a given, isolated drug-to-antibody ratio fraction, different isomers dissociate to unique peaks as measured by CE-SDS. For drug-to-antibody ratio 2 species, DAR2f is the only isomer that leads to HHL, and only DAR2h remains as an intact mAb (Figure 3). For a drug-to-antibody ratio of 4, only DAR4ff leads to HH, DAR4fh to HHL, and DAR4hh to two HL fragments. For six drugs per antibody, DAR6ffh uniquely creates HH, and DAR6fhh creates HL. By examining CE-SDS data of each isolated drug-to-antibody ratio fraction and converting peaks to mole percent (normalizing the area of each peak for migration time and the number of lysines of each fragment and setting the sum of all peaks to 100%), the amount of each unique peak named above corresponds to the abundance of the parent isomer. Each isolated drug-to-antibody ratio fraction is analyzed in turn. The results as weighted by the amount of each drug-to-antibody ratio agree well with those found by solving System 1 (shown in the third column of Table 1). A similar approach based on reversed-phase HPLC data of isolated fractions for all but one species (DAR4fh, which requires CE) has been used previously for measurements on another IgGκ antibody.20 This same approach may be applied here, using absorbance coefficients calculated29 for the different fragments with corrections added for drug absorbance. Again, the results compare very favorably with those found using System 1 with unfractionated mAb1 (last column, Table 1). The output of these different approaches vary by at most 3.4% between them, suggesting that all of them point to almost the same quantitative values for each positional isomer. We find only 0.61.4% DAR6ffh using the fraction collected methods, which supports the view that the best solution to System 1, which includes no DAR6ffh, is relatively accurate. Method precision, which is relevant to discerning differences between samples, is discussed below. There are several factors that are likely important to the accuracy of this system-of-equations approach. The equations mathematically relate the drug load distribution (from HIC) by how each positional isomer will mechanistically dissociate in the presence of detergent. It is possible that not every molecule dissociates the exact way proposed in Figure 3 or with the implied ideal efficiency. We currently do not have a way to test the quantitative accuracy of each proposed dissociation 7483

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

the first hinge region disulfide is slow but is very quickly followed by reduction of the second hinge disulfide. Indeed, the DAR2h and DAR4fh positional isomers are also in low abundance in our data, suggesting that during the reduction process their unconjugated forms are short-lived intermediates on the way to those corresponding to DAR4hh and DAR6fhh, respectively. Similarly, the reduced species that leads to DAR6ffh rapidly becomes that leading to DAR8. Precision. Although the inputs into the systems-ofequations contain error due to assay variability, this cannot be easily propagated through to the solution to calculate an overall estimate of output precision. Instead, in silico simulations can be performed where the HIC and CE-SDS inputs are randomly changed by known amounts for System 1 to simulate experimental noise in the data and thus assess the precision of the system. Imagine that there are three possible noise perturbations for each piece of information fed into the system: it can increase in value, decrease in value, or remain the same. The various perturbations sum to zero so the inputs need not be rescaled. The results from the perturbed inputs are used to calculate a standard deviation due to the added noise for each positional isomer, which yields a measure of the precision of the system. The result of the in silico simulations show that when HIC and CE-SDS have estimated variabilities of 0.2% and 0.5%, respectively, solving System 1 is precise to within 0.5%. Similar simulations can be performed for System 3, where the inputs are changed by known amounts for HIC, CE-SDS, and RP. Like the previous System 1 precision assessment, System 3 is also precise to within 0.5% given variability in RP of 0.5%, HIC of 0.2%, and CE-SDS of 0.5%. Additional simulations can be conducted to assess the sensitivity of each positional isomer across a range of perturbed inputs to System 1. The standard deviation of each positional isomer for perturbations of the HIC and CE-SDS inputs ranging from 0 to 3.0% are shown in Figure S-2 in the Supporting Information for several mAbs. The red dots indicate the calculated precision at the estimated assay variability. These graphs show how the various positional isomers have different sensitivities to the inputs. For situations where there are zero DAR6ffh species detected for the given CE-SDS and HIC inputs, the calculated values of this isomer do not deviate from this value until the noise level reaches 1−2%, depending on the molecule. This likely corresponds to the level of mismatch between the ideal and actual dissociation and detection efficiencies of the HIC and CE-SDS techniques discussed above. Varying Average Drug-to-Antibody Ratio. We can easily assess how the positional isomer percentages of a population change with variation in average drug-to-antibody ratio with this method. For mAb1 there are three different batches with average drug-to-antibody ratios of 2.0, 3.5, and 5.5. Figure 4 shows how the amounts of the isomers change with an increasing average drug-to-antibody ratio by solving System 1, with uncertainty bars showing the estimated assay precision discussed above. Lower average drug-to-antibody ratio samples have abundant 0 and 2 drug-to-antibody ratio species, and higher average samples have more 6 and 8 drug-to-antibody ratio species. Upon inspection of the relative percentage of each isomer within a drug load amount, the different average drugto-antibody ratio batches have different relative positional isomer distributions, particularly for antibodies with four drugs. This is shown in the table of values in Figure 4, where there is

Table 2. Positional Isomer Amounts for mAb2 Determined by System 1, System 2, and System 3 abundance (%) isomer

system 1

system 2

system 3

DAR0 DAR2f DAR2h DAR4ff DAR4fh DAR4hh DAR6ffh DAR6fhh DAR8

6.5 29.6 2.1 29.3 4.2 13.3 0.0 13.4 1.5

6.5 30.0 1.8 31.0 0.0 15.8 3.0 10.5 1.5

6.5 29.9 1.9 29.1 4.1 13.6 0.5 12.9 1.5

mAb2 with System 1 (Table 2, first column). Both of these single hinge labeled species are of low abundance in fractioncollected samples discussed above, so it is difficult at first to determine which combination of assays yields a system-ofequations with more consistent and accurate results. Several analyses cast more doubt on the robustness of System 2 than on System 1. The most compelling demonstration of this is shown by creating different, idealized, model data sets of constant average drug-to-antibody ratio for isomer distributions both near-to and far-from those observed (3.5, here). We define a population of positional isomers and manually calculate the idealized results of the HIC, CE-SDS, and RP assays as if the ADCs dissociated exactly according to the schemes shown in Figure 3 and Figure S-1 in the Supporting Information. We take this HIC, CE-SDS, and RP data, solve Systems 1 and 2, and compare the results back to the predefined isomer population. While solving System 1 always returns the correct distribution of isomers (including DAR6ffh), the solution to System 2 fails to do so. Notably, System 2 often gives unsatisfactory and inconsistent positional isomer results for drug-to-antibody ratio 4, which is likely attributed to the paired DAR4 variables in eqs 2.1, 2.2, 2.4, and 2.5 in the Supporting Information. An alternative mathematical solver also proved unsatisfactory, as discussed in the Supporting Information. System 3: HIC and CE-SDS and RP. We can evaluate whether there is a benefit to adding more information to the HIC and CE-SDS system, System 1. A combined system-ofequations may be developed where the RP equations are added to the HIC and CE-SDS equations to determine the positional isomers. This combined system of 15 equations uses information from all three assays to determine the positional isomer amounts. As with System 1, the solutions of the combined equations are consistent using all tested solvers. Table 2 compares the results of System 1 to System 3 for mAb2. The positional isomer values for drug-to-antibody ratio 2 and 4 are consistent between System 1 and System 3, and the combined system-of-equations gives a very small amount for the DAR6ffh species, though this is not consistently observed. A practical conclusion is that the HIC and CE-SDS method (System 1) is useful if a rapid estimate of positional isomer values is needed while only evaluating the sample with two assays. The combined system-of-equations (System 3) may give slightly more detailed information, mainly for DAR6 isomers but at the cost of running an additional assay. Both methods find that the DAR6ffh isomer is almost nonexistent, and the DAR6 species is overwhelmingly the DAR6fhh isomer. This is consistent with a cooperative situation where the reduction of 7484

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

controlled conjugation process targeting a specific average drug-to-antibody ratio leads to consistent positional isomers.

■

CONCLUSIONS A distinct advantage of this method is that it allows quick positional isomer analysis of quantity-constrained samples, which are generally not well suited to the isolation and analysis of individual drug-to-antibody ratio fractions. Furthermore, there is some flexibility in the exact techniques used to obtain the inputs of this system-of-equations approach. Although we present data obtained using conventional CE-SDS, we have also obtained very similar results using a microchip-based CE separation (Bioanalyzer, Agilent Technologies). Indeed, part of the appeal of this method is that it extracts further, detailed data about a sample from straightforward and relatively common assays. We expect that this method can prove especially useful for process characterization and development to easily profile positional isomer distributions of cysteine-linked ADCs. This type of approach is well suited to testing many small-scale samples representing a wide range of process parameter space. The apparent precision capabilities of this approach mean that small differences in samples should be resolvable. This type of approach can enable investigations of the roles that positional drug distributions may play in ADC manufacturing, storage, and stability as well as their clinical safety, pharmacokinetics, and efficacy. Building a greater understanding of such ADCs will hopefully help bring more of these promising therapeutics to patients.

Figure 4. Positional isomer amounts determined using System 1 for mAb1 samples with average drug-to-antibody ratios (DARavg) of 2.0, 3.5, and 5.5. Samples with low average drug-to-antibody ratios have more unconjugated antibody and DAR2 species, while higher average drug-to-antibody ratio samples contain more DAR4, 6, and 8 species. The error bars are precision bars as discussed in the text. The values beneath the chart show the relative amount of each positional isomer within a drug load.

an increasing percentage of the DAR4ff positional isomer with increasing average drug-to-antibody ratio. Results Across Multiple ADCs. Seven different ADCs with an average drug-to-antibody ratio of close to 3.5 were assessed by solving System 1 to determine the positional isomer amounts. Though the IgG1κ antibodies considered here are highly homologous in sequence, it is conceivable that the positional isomer distribution could vary between them, even for similar average drug-to-antibody ratios. The results show that the absolute distribution of isomers is quite consistent across multiple ADCs at a similar average drug-to-antibody ratio, as shown in Table 3. This result supports the view that

■

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We thank Dana Olson, Matt Hutchinson, Charles Morgan, Brian Connolly, Daren Nelson, Nia Beckley, Pervina Kei, and Yilma Adem for materials, data, and technical assistance.

abundance (%) isomer

mAb1

mAb2

mAb3

mAb4

mAb5

mAb6

mAb7

5.1 29.5 1.4 33.3 2.8 13.8 0.0 11.5 2.4 3.5

6.5 29.6 2.1 29.3 4.2 13.3 0.0 13.4 1.5 3.4

5.7 28.4 1.8 29.1 4.9 14.8 0.0 13.5 1.7 3.5

7.2 28.0 0.3 26.7 5.3 16.6 0.0 13.3 2.7 3.5

4.3 24.4 2.1 28.6 7.0 16.3 0.6 13.4 3.4 3.7

4.2 25.7 1.8 29.3 5.3 16.6 0.0 14.3 2.9 3.7

5.9 30.2 1.8 29.5 4.1 15.3 0.0 11.3 1.7 3.4

AUTHOR INFORMATION

Corresponding Author

Table 3. Positional Isomer Values Shown for mAbs 1−7 Calculated Using System 1

DAR0 DAR2f DAR2h DAR4ff DAR4fh DAR4hh DAR6ffh DAR6fhh DAR8 DARavg

ASSOCIATED CONTENT

■

REFERENCES

(1) Moolten, F. L.; Cooperband, S. R. Science 1970, 169, 68−70. (2) Morrison, S. L.; Johnson, M. J.; Herzenberg, L. A.; Oi, V. T. Proc. Natl. Acad. Sci. U.S.A. 1984, 81, 6851−6855. (3) Verhoeyen, M.; Milstein, C.; Winter, G. Science 1988, 239, 1534− 1536. (4) Jones, P.; Dear, P.; Foote, J.; Neuberger, M.; Winter, G. Nature 1986, 321, 522−525. (5) Lonberg, N. Curr. Opin. Immunol. 2008, 20, 450−459. (6) Almagro, J. C.; Fransson, J. Front. Biosci. 2008, 13, 1619−1633. (7) Hwang, W. Y. K.; Foote, J. Methods 2005, 36, 3−10. (8) Carter, P. Nat. Rev. Immunol. 2006, 6, 343−357. (9) Lambert, J. Curr. Opin. Pharmacol. 2005, 5, 543−549. (10) Tolcher, A. W.; Sugarman, S.; Gelmon, K. A.; Cohen, R.; Saleh, M.; Isaacs, C.; Young, L.; Healey, D.; Onetto, N.; Slichenmyer, W. J. Clin. Oncol. 1999, 17, 478−484.

defining the average drug-to-antibody ratio of a conjugation batch also controls the positional distribution of those drugs to within a few percent for ADCs built off of a common antibody framework. Changing the framework, conjugation solution conditions, or other process parameters may result in different positional isomers, but these data show that a robustly 7485

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486

Analytical Chemistry

Article

(11) King, H. D.; Staab, A. J.; Pham-Kaplita, K.; Yurgaitis, D.; Firestone, R. A.; Lasch, S. J.; Trail, P. A. Bioorg. Med. Chem. Lett. 2003, 13, 2119−2122. (12) Alley, S. C.; Okeley, N. M.; Senter, P. D. Curr. Opin. Chem. Biol. 2010, 14, 529−537. (13) Doronina, S. O.; Toki, B. E.; Torgov, M. Y.; Mendelsohn, B. A.; Cerveny, C. G.; Chace, D. F.; DeBlanc, R. L.; Gearing, R. P.; Bovee, T. D.; Siegall, C. B.; Francisco, J. A.; Wahl, A. F.; Meyer, D. L.; Senter, P. D. Nat. Biotechnol. 2003, 21, 778−784. (14) Chari, R. V.; Martell, B. A.; Gross, J. L.; Cook, S. B.; Shah, S. A.; Blättler, W. A.; McKenzie, S. J.; Goldmacher, V. S. Cancer Res. 1992, 52, 127−131. (15) Sapra, P.; Hooper, A. T.; O’Donnell, C. J.; Gerber, H.-P. Expert Opin. Invest. Drugs 2011, 20, 1131−1149. (16) Forero-Torres, A.; Leonard, J. P.; Younes, A.; Rosenblatt, J. D.; Brice, P.; Bartlett, N. L.; Bosly, A.; Pinter-Brown, L.; Kennedy, D.; Sievers, E. L.; Gopal, A. K. Br. J. Hamaetol. 2009, 146, 171−179. (17) Younes, A.; Bartlett, N.; Leonard, J.; Kennedy, D. A.; Lynch, C. M.; Sievers, E. L.; Forero-Torres, A. N. Engl. J. Med. 2010, 363, 1812− 1821. (18) Phillips, G. D. L.; Li, G.; Dugger, D. L.; Crocker, L. M.; Parsons, K. L.; Mai, E.; Blattler, W. A.; Lambert, J. M.; Chari, R. V. J.; Lutz, R. J.; Wong, W. L. T.; Jacobson, F. S.; Koeppen, H.; Schwall, R. H.; Kenkare-Mitra, S. R.; Spencer, S. D.; Sliwkowski, M. X. Cancer Res. 2008, 68, 9280−9290. (19) Hamblett, K. J. Clin. Cancer Res. 2004, 10, 7063−7070. (20) Sun, M. M. C.; Beam, K. S.; Cerveny, C. G.; Hamblett, K. J.; Blackmore, R. S.; Torgov, M. Y.; Handley, F. G. M.; Ihle, N. C.; Senter, P. D.; Alley, S. C. Bioconjugate Chem. 2005, 16, 1282−1290. (21) Shen, B.-Q.; Xu, K.; Liu, L.; Raab, H.; Bhakta, S.; Kenrick, M.; Parsons-Reponte, K. L.; Tien, J.; Yu, S.-F.; Mai, E.; Li, D.; Tibbitts, J.; Baudys, J.; Saad, O. M.; Scales, S. J.; McDonald, P. J.; Hass, P. E.; Eigenbrot, C.; Nguyen, T.; Solis, W. A.; Fuji, R. N.; Flagella, K. M.; Patel, D.; Spencer, S. D.; Khawli, L. A.; Ebens, A.; Wong, W. L.; Vandlen, R.; Kaur, S.; Sliwkowski, M. X.; Scheller, R. H.; Polakis, P.; Junutula, J. R. Nat. Biotechnol. 2012, 1−8. (22) McDonagh, C. F.; Turcott, E.; Westendorf, L.; Webster, J. B.; Alley, S. C.; Kim, K.; Andreyka, J.; Stone, I.; Hamblett, K. J.; Francisco, J. A.; Carter, P. Protein Eng. Des. Sel. 2006, 19, 299−307. (23) Wakankar, A.; Chen, Y.; Gokarn, Y.; Jacobson, F. S. MAbs 2011, 3, 161−172. (24) Michels, D. A.; Brady, L. J.; Guo, A.; Balland, A. Anal. Chem. 2007, 79, 5963−5971. (25) Hunt, G. G.; Nashabeh, W. W. Anal. Chem. 1999, 71, 2390− 2397. (26) Salas-Solano, O.; Tomlinson, B.; Du, S.; Parker, M.; Strahan, A.; Ma, S. Anal. Chem. 2006, 78, 6583−6594. (27) Wojcik, R.; Swearingen, K. E.; Dickerson, J. A.; Turner, E. H.; Ramsay, L. M.; Dovichi, N. J. J. Chromatogr., A 2008, 1194, 243−248. (28) Liu, H.; Chumsae, C.; Gaza-Bulseco, G.; Hurkmans, K.; Radziejewski, C. H. Anal. Chem. 2010, 82, 5219−5226. (29) Bewley, T. Anal. Biochem. 1982, 123, 55−65.

7486

dx.doi.org/10.1021/ac301568f | Anal. Chem. 2012, 84, 7479−7486