Introduction of the Mass Spread Function for Characterization of

Nov 30, 2011 - To address this problem, we introduce the mass spread function (MSF) for ... By treating the ESI-MS spectrum of conjugated protein as t...
0 downloads 0 Views 582KB Size
Download PDF
Technical Note pubs.acs.org/ac

Introduction of the Mass Spread Function for Characterization of Protein Conjugates Joseph P. Skinner, Lianli Chi,† Panfilo F. Ozeata, Carol S. Ramsay, Robynn L. O’Hara, Brenda B. Calfin, and Sergey Y. Tetin* Diagnostics Research, Abbott Diagnostics Division, 100 Abbott Park Road, Abbott Park, Illinois 60064, United States S Supporting Information *

ABSTRACT: Traditionally, characterization of protein molecules conjugated with molecular probes is performed by UV− vis spectroscopy. This method determines the average incorporation ratio but does not yield information about the label distribution. Electrospray ionization mass spectroscopy (ESI-MS) allows direct measurement of the fraction of protein containing a given number of labels. However, for a glycosylated protein, this analysis can be severely limited due to spectral overlap of the labels and carbohydrates. To address this problem, we introduce the mass spread function (MSF) for conjugation analysis. By treating the ESI-MS spectrum of conjugated protein as the spectrum before conjugation convolved with the MSF, we are able to quantify the labeled protein population using a binomial distribution function. We first applied this procedure for characterization of labeled antibody F(ab′)2 fragments which do not contain carbohydrates. We then apply the MSF to fit spectra of entire conjugated monoclonal antibodies and quantify the distribution of labels in the presence of glycans.

Conjugation of molecular probes and other small molecules to proteins have important application in diagnostics and therapeutics.1,2 One of the most common methods of producing conjugated macromolecules is through random covalent linking to amino or thiol groups, which results in a mixture of labeled protein molecules with different numbers of labels per protein. Control of the labeling is typically done by adjusting the ratio of label to protein in the reaction mixture. The probe load has direct impact on signal generation or drug related properties, but overload may adversely affect the affinity of the conjugate. Therefore, quantitative analysis of the labeling distribution has been a focus of protein chemists since the introduction of fluorescent and radioactive labeling.3−6 After purification, determination of the average incorporation ratio (IR) of the conjugated probe is typically the first step before use of labeled samples. Immunoglobulin G (IgG) antibodies are one of the most frequently labeled classes of proteins. Stability and robustness of randomly conjugated antibodies have made them excellent biological reagents. It has been shown that loading of an IgG antibody with a probe at an IR up to eight does not reduce its affinity.7 However, the functional activity of the conjugated proteins is always a concern. Use of drug carrying antibodies requires thorough characterization of all antibody−drug species produced during conjugation and elimination of inactive populations.8 Accurate determination of the label distribution is a critical step in conjugate preparation. The probe IR is often obtained from the absorption spectrum of the labeled protein using the predetermined extinction coefficients of the protein and the probe at given wavelengths. This method gives an estimate of © 2011 American Chemical Society

the average IR in the protein population but does not provide any means for characterizing the actual distribution. Modern methods of mass spectrometry can resolve protein populations that carry different amount of the probe. For instance, the use of electrospray ionization mass spectrometry (ESI-MS) for characterizing conjugates of mAb and Fab fragments provides direct measurement of the conjugation profile without the need to rely on UV−vis spectra.9 However, inherent heterogeneity of even moderately glycosylated proteins presents a problem for peak assignment in the mass spectrum due to spectral overlap of labels and carbohydrates. Application of size-exclusion chromatography coupled to the mass spectrometer for characterizing deglycosylated antibody conjugates results in higher quality spectra which can be used for determination of the antibody species with different IRs.10 However, deglycoslylation introduces additional manipulations with the conjugated protein that may influence the determined distribution. It is commonly accepted that, at low and moderate probe to protein ratios, the probe distribution follows binomial or, as a limiting case, Poisson statistics.11,12 In principle, mass spectra of the conjugated protein should fit a given probe distribution predicted by a probability function. Here, we demonstrate that fitting ESI-MS data directly to a function described by a binomial distribution allows for quantification of the labeled protein population. This principle is then further extended to show that such analysis can be applied to intact antibody preparations without the need to perform deglycosylation. Received: August 24, 2011 Accepted: November 30, 2011 Published: November 30, 2011 1172

dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177

Analytical Chemistry

Technical Note

extinction coefficient of ε495 = 70 000 (M cm)−1 was used for Alexa488, as quoted by the manufacturer. A correction factor of ε280/ε495 = 0.16 was used to account for absorbance at 280 nm due to Alexa488 when measuring conjugated protein. The conjugate in this study had an IR = 0.9. NHS-PEO4-biotin and NHS-PEO12-biotin (Pierce, Rockford, IL) was mixed at a 10:1 molar ratio with mouse monoclonal IgG1 in 100 mM phospate buffer containing 150 mM NaCl (pH 7.4). The sample was incubated for 4 h at room temperature followed by dialysis into the same buffer. It should be noted that this antibody is not the same as that used above for labeling with Alexa488 and binds different antigen. ESI-MS Analysis. The samples were desalted using Amicon Ultra centrifugal filter devices (MWCO 10 KD). The ESI-MS experiments were performed on an Applied Biosystems QSTAR Pulsar hybrid quadrupole-TOF LC/MS/MS mass spectrometer using positive mode. A C18 trap column was used to further desalt and concentrate the samples. The mass spectrometer key parameters were set as follows: IS 5500 V, GS1 30, DP 65 V, FP 265 V, and DP2 15 V. The analysis results were processed using Analyst QS 1.1 software. Fitting Analysis. Processed spectra from the Analyst QS software were exported as text files for further analysis. The data was read and analyzed using built-in functions in Mathematica 8.0 (Wolfram Research, Champaign, Illinois). For deconvolution analysis, the Mathematica function ListDeconvolve was applied using the protein spectrum before conjugation as the kernel and the spectrum of the conjugated protein as the convolved data list. The resulting spectrum returned by ListDeconvolve was fit to the MSF defined below using the function NonlinearModelFit. An example of detailed analysis along with Mathematica code is given in the Supporting Information.

For describing heterogeneous spectra of glycosylated proteins, we introduce the concept of a mass spread function (MSF). The mass spread function predicts the fractions of each protein species that contains the specified number of probe molecules according to a probability distribution. Mathematically, the spectrum of the conjugate corresponds to the spectrum of the unlabeled protein convolved with the MSF. The principle of this analysis is analogous to optical imaging, in which an observed image is the convolution of the light signal from the actual object with the point spread function of the optical system.13 At first, we use MSF for characterization of glycan free antibody F(ab′)2 fragments which have been conjugated with the chemiluminescent probe acridinium. Then, we show that the analysis can also be applied to glycosylated mouse monoclonal IgG antibodies conjugated with Alexa488 and biotin.

■

EXPERIMENTAL PROCEDURES Preparation of F(ab′)2 Fragments. Affinity purified mouse IgG2a was digested with Lysyl Endopeptidase enzyme (source: Achromobacter lyticus: Wako 129-02541 (EC 3.4.21.50)) in 50 mM Tris buffer, pH 8.4. Undigested IgG and Fc fragments were removed by passing the solution through a MabSelect Xtra Protein A column (GE Healthcare, Piscataway, NJ). Purity of the F(ab′)2 fragments was confirmed by HPLC using a G3000SWXL column (Tosoh Bioscience, Inc., South San Francisco, CA). F(ab′)2 fragment was dialyzed into 100 mM phospate buffer containing 150 mM NaCl (pH 8.0) prior to conjugation. Preparation of F(ab′)2 Acridinium Conjugate. Antibody F(ab′)2 fragment was conjugated with acridinium-9-carboxamide active ester.2,14 Acridinium was added at 6:1, 8:1, and 10:1 probe to F(ab′)2 molar ratios while mixing at room temperature, followed by overnight incubation. Acridinylated F(ab′)2 conjugates were separated from free probe using prepacked Sephacryl S200 columns (GE Healthcare, Piscataway, NJ). The purified conjugates were evaluated spectrophotometrically at 280 nm and 370 nm to determine protein concentration and acridinium incorporation ratio (IR). An analytically determined molar extinction coefficient of ε370 = 13 430 (M cm)−1 was used for acridinium. A molar extinction coefficient of ε280 = 142 100 (M cm)−1 was used for F(ab′)2, which is based on a protein mass of 98 kDa and A280 = 1.45 at a concentration of 1 mg/mL. The conjugate absorbance at 280 nm was corrected for acridium contribution using ε280/ε370 = 0.247, which was determined by measurement of acridinium alone. Therefore, the corrected absorbance at 280 nm is given by subtracting 24.7% of the absorbance measured at 370 nm from the absorbance at 280 nm. The corrected value at 280 nm is used to determine conjugated protein concentration. Preparation of IgG Conjugates. Alexa Fluor 488 succinimidyl ester (Alexa488) was obtained from Molecular Probes (Eugene, OR). Alexa488 was mixed at a 2:1 molar ratio with mouse IgG1. The sample was incubated in the dark overnight before purification with a NAP-5 column (GE Healthcare, Piscataway, NJ). The sample was then measured using the spectrophotometer to determine the incorporation ratio ϕ based on the absorbance at 495 and 280 nm using a a Cary 3G Spectrophotometer (Agilent Technologies, Palo Alto, CA). A molar extinction coefficient of ε280 = 217 500 (M cm)−1 was used for IgG antibody, given that a 1 mg/mL sample yields an absorbance of 1.45 and has a mass of 150 kD. A molar

■

THEORY We treat the mass spectrum of conjugated protein MSc as the convolution of the unconjugated protein spectrum MSu with a function describing conjugation. We call this function the mass spread function (MSF). It describes how the label is distributed in the conjugated protein population and is applied using the following definition

MSc = MSu ⊗ MSF

(1)

where the symbol ⊗ denotes convolution. To compute the spectrum of the conjugated protein MSc from eq 1, the functional form of the MSF must be defined. The MSF is dependent upon the original mass of the protein to be conjugated, m, and the mass of the probe molecule, μ. The most general form of the mass spread function would be N

MSF(m , μ) =

∑ A jδ(m − jμ) j=0

(2)

where Aj is the fraction of protein containing j labels. The Dirac delta function, δ(m − jμ), in eq 2 ensures application of the amplitude at integer intervals of the conjugate mass. Mathematically, the MSF described in eq 2 is known as the Dirac delta comb. Use of this form of the MSF would be appropriate when labeling is combined with site-directed mutagenesis. For example, if three cysteines were inserted into a protein, N = 3, the analysis would determine the values of A0 through A3. This assumes the protein does not have any 1173

dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177

Analytical Chemistry

Technical Note

Figure 1. (A) Plot of conjugated F(ab′)2 mass spectrum. Absorbance measurement for this sample yielded an IR = 3.0. (B) Fit of F(ab′)2 conjugated with acridinium. A binomial distribution based MSF (solid line) was used to fit the data (gray line). The fit yields an IR = 2.9 (N = 7 and p = 0.41) and a conjugate mass of 569 Da. (C) The data and fit to a lower IR yields IR = 2.2 (N = 6, p = 0.36). (D) The data fit to a higher IR = 3.6 (N = 9, p = 0.41).

parameter, along with N, simply describes the final distribution after labeling and purification. The average IR is calculated as ϕ = Np. This distribution also allows determination of the variance of the labeling distribution according to var = Np(1 − p).

other sites available for labeling except the residues introduced by mutagenesis. For random labeling, the number of conjugate probes on a protein molecule is given by some probability distribution function, pdf, with an average of ϕ. The following general form of the MSF results

■

RESULTS AND DISCUSSION F(ab′)2: No Glycosylation. Data was obtained from a sample of F(ab′)2, which does not contain carbohydrates, before and after conjugation with the probe. The conjugate samples were also analyzed by UV−vis absorbance which enables comparison of the IR. The ESI-MS data of F(ab′)2 is shown as the inset in Figure 1A. This spectrum is used as the input kernel in deconvolution analysis to determine the MSF for a given sample. The ESI-MS of conjugated F(ab)′2 using an input label to protein ratio of 8:1 is shown in Figure 1A. The position of each peak corresponds to the number of labels attached to the protein, and the amplitude describes the relative fraction of each labeled species. On the basis of our description of conjugates in eq 1, the functional form of MSc is given by

N

MSF(m , μ) =

∑ pdf(j)δ(m − jμ) j=0

(3)

where j is the number of conjugates attached to the protein. The first term in the sum, pdf(j), gives the amplitude or fraction of protein containing j labels based on parameters describing the distribution. To fully specify the MSF in eq 3, the number of incorporated labels per protein must be described by a distribution function. Due to the discrete nature of labeling, we describe the probability of a protein containing j labels using a binomial distribution

⎛N ⎞ pdf(j ; N , p) = ⎜ ⎟p j (1 − p)N − j ⎝ j⎠

(4)

MSc =

where the notation for the binomial coefficient has been used for the first term. This assumes that, under the reaction conditions, a maximum of N residues on any protein molecule is labeled and the label can be found on one of these residues with a probability p. It should be noted that p does not describe the probability of labeling a given site during the reaction. This

⎡

2⎤

∫ A exp⎢⎣ − (x −s2M)

⎥ ⎦

N

∑ pdf(j; n , p)δ(x − X + jμ) dx j=0 1174

(5) dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177

Analytical Chemistry

Technical Note

Figure 2. Mass spectrum of IgG, before and after conjugation. (A) The spectrum of the antibody indicates a significant amount of carbohydrates on the antibody. (B) Multiple conjugate states causes overlap after conjugation with Alexa488.

Figure 3. Numerical deconvolution of IgG labeled with Alexa488. (A) The MSF obtained from deconvolution is shown as the solid line. A fit to eq 5 (dashed line) yields an IR = 0.73 (p = 0.36, L = 2). (B) The distribution of labels as determined by fitting to the deconvolved spectrum.

where a Gaussian with amplitude A and width s is used for MSu. In the above equation, convolution is performed over the mass variable x. The convolved spectrum will be plotted along the new conjugate mass axis denoted by X. It should be noted that in the analysis below, eq 5 is also used as the MSF when fitting deconvolved IgG spectra. This is done because eq 5 properly positions the peaks as Gaussians along the mass axis. By examining the data in the plot (Figure 1A), eight peaks can be observed, so we expect conjugates with up to seven labels and the unlabeled fraction. The fit, shown as the dashed line in Figure 1B, yields a value of N = 7 and p = 0.40. These values indicate that, at most, seven residues are labeled on any given protein molecule, each with a probability of 0.40. The first peak observed in Figure 1A at a mass of ∼98 kDa corresponds to unlabeled F(ab′)2 (N = 0). The mean is Np = ϕ = 2.8 with a variance of 1.7. The fitted probe mass is 566.8 Da, which agrees with the expected value of 567 Da. For reference, the incorporation ratio measured by UV−vis absorbance was 3.0. The same procedure was applied to two more conjugated F(ab′)2 samples. The data and resulting fit for the second sample (input dye to protein ratio of 6:1) is shown in Figure 1C. The values yielded for this sample are N = 6 and p = 0.36. This gives an incorporation ratio of ϕ = Np = 2.2 and a variance of 1.4. The third sample measured (input dye to protein ratio of 10:1) gave N = 9 and p = 0.41. The fit, shown in Figure 2D (dashed line), gives an incorporation ration of ϕ = Np = 3.6

and variance = 2.2. The IR values measured using absorbance for these two samples are 2.2 and 3.6, respectively. IgG: Glycosylation. To determine the incorporation ratio in the presence of carbohydrates, intact antibody was conjugated with Alexa488. ESI-MS data obtained before conjugation with the fluorophore is shown in Figure 2A. In this case, glycosylation results in a spectrum with a more complicated structure. The spectrum contains multiple peaks with different heights and widths due to various numbers and masses of carbohydrates attached to the antibody. The range of the mass covered by the spectrum is important when considering a conjugate sample. In this case, the minimum mass is 148.4 kDa and the maximum is 149.0 kDa. Alexa488 has a mass of 515.5 Da, less than the spread of the spectrum, and therefore, overlap of conjugate with differing numbers of probes is expected due to the presence of the carbohydrates. Conjugation of the antibody with Alexa488 resulted in the spectrum shown in Figure 2B. Conjugation has shifted some of the peaks along the x-axis; however, the fraction of conjugates with a given number of labels cannot be assigned to peaks due to overlap of the spectra. One may indentify the largest peak in Figure 2B as antibody containing a single label. However, this peak is positioned within the range of the spectrum before conjugation and cannot be used to determine the relative amount based upon the amplitude. The spectrum presented in Figure 2B was deconvolved using the spectrum before conjugation shown in Figure 2A. The 1175

dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177

Analytical Chemistry

Technical Note

Figure 4. Deconvolution analysis on IgG labeled with biotin. (A) Antibody conjugated with PEO4-biotin. The spectrum of the antibody before labeling is shown in the inset. (B) The same antibody conjugated with PEO12-biotin. (C) The MSF obtained from decovolution of the IgG conjugated with PEO4-biotin. The fit (dashed line) coincides with the data and yields IR = 3.16 (p = 0.45 and L = 7). (D) The MSF obtained from the IgG labeled with PEO12-biotin. Fitting to eq 5 (dashed line) yields IR = 2.63 (p = 0.38, L = 7).

PEO4-biotin and PEO12-biotin are 473.22 Da and 825.64 Da according the manufacturer.

resulting MSF is shown in Figure 3 (solid line) along with a fit to eq 5 (dashed line). The location and heights of the peaks correspond to the distribution of labels on the IgG molecules. The IR based on the fit is 0.73 (N = 2, p = 0.36) with a variance of 0.46. The incorporation ratio from UV−vis absorbance measurement resulted in 0.9. Figure 3B shows the distribution of the number of labels per protein molecule based on the values obtained from fitting the data above. Using the plot for reference, one can see that ∼40% of the sample is unlabeled antibody, ∼48% is single labeled, and ∼12% contain two labels. In total, ∼60% of the antibody is labeled. A step-by-step analysis of this data can be found in the Supporting Information. Two more conjugated antibody samples were prepared with two different lengths of PEO-linked biotin. These samples were analyzed following the numerical deconvolution algorithm described above. The mass spectra for the antibody after conjugation with PEO4-biotin (Figure 4A) and PEO12-biotin (Figure 4B) are shown in Figure 4. The inset of Figure 4A displays the antibody spectrum before conjugation with the biotin reagents. The deconvolved spectra and fits to eq 5 are shown in Figure 4C,D. In each instance, the mass of the conjugate was allowed to vary. The PEO4-biotin sample resulted in an IR = 3.1 (variance =1.5) with a conjugate mass of 472.7 Da. The PEO12-biotin sample yields IR = 2.6 (variance =1.6) and a conjugate mass of 827.5 Da, which is in good agreement with the expected masses. The net mass addition of

■

CONCLUSIONS We have described a method which allows quantification of the distribution of labels on antibody subjected to common labeling techniques by fitting ESI-MS data before and after conjugation. We introduced a mass spread function which describes the random labeling as a binomial distribution due to the discrete nature of protein labeling. The work shows that such a process allows quantitative characterization of conjugated samples by describing the protein conjugate in terms of a probability distribution. The method may be applied to glycoproteins without the need for deglycosylation.

■

ASSOCIATED CONTENT

S Supporting Information *

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

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Present Address †

National Glycoengineering Research Center, Shandong University, 27 Shanda Nan Road, Jinan Shandong, China 250100.

1176

dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177

Analytical Chemistry

■

Technical Note

ACKNOWLEDGMENTS We would like to thank Prof. Don Lamb at LudwigMaximilians-Universitat, Munich, Germany, for suggesting we attempt this convolution approach.

■

REFERENCES

(1) AlleyS. C.OkeleyN. M.SenterP. D.Curr. Opin. Chem. Biol.2010 (2) Adamczyk, M.; Gebler, J.; Shreder, K.; Wu, J. Bioconjugate Chem. 2000, 11, 557−563. (3) Creech, H. J.; Oginsky, E. L.; Cheever, F. S. Cancer Res. 1947, 7, 290−296. (4) Hughes, W. L. Jr. Cold Spring Harb. Symp. Quant. Biol. 1950, 14, 79−84. (5) McFarlane, A. S. Biochem. J. 1956, 62, 135−143. (6) Weber, G. Biochem. J. 1952, 51, 155−167. (7) Ruan, Q.; Skinner, J. P.; Tetin, S. Y. Anal. Biochem. 2009, 393, 196−204. (8) Sievers, E. L. Blood Cells, Mol., Dis. 2003, 31, 7−10. (9) Bennett, K. L.; Smith, S. V.; Lambrecht, R. M.; Truscott, R. J.; Sheil, M. M. Bioconjugate Chem. 1996, 7, 16−22. (10) Lazar, A. C.; Wang, L.; Blattler, W. A.; Amphlett, G.; Lambert, J. M.; Zhang, W. Rapid Commun. Mass Spectrom. 2005, 19, 1806−1814. (11) Boukari, H.; Nossal, R.; Sackett, D. L.; Schuck, P. Phys. Rev. Lett. 2004, 93, 098106. (12) Muller, J. D.; Chen, Y.; Gratton, E. Biophys. J. 2000, 78, 474− 486. (13) Hecht, E. Optics, 4 ed.; Addison Wesley: San Francisco, 2002. (14) Adamczyk, M.; Chen, Y.-Y.; Mattingly, P. G.; Pan, Y.; Rege, S. J. Org. Chem. 1998, 63, 5636−5639.

1177

dx.doi.org/10.1021/ac202239j | Anal. Chem. 2012, 84, 1172−1177