Anal. Chem. 2008, 80, 2623-2627
Correspondence
Assessment of the Three-Dimensional Structure of Recombinant Protein Therapeutics by NMR Fingerprinting: Demonstration on Recombinant Human Granulocyte Macrophage-Colony Stimulation Factor Yves Aubin,*,†,‡,§,| Genevie`ve Gingras,† and Simon Sauve´†
Centre for Biologics Research, Biologics and Genetic Therapies Directorate, Health Canada, 251 Sir Frederick Banting Driveway, Ottawa, ON, K1A 0K9, Canada, Department of Chemistry, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario, K1S 5B6, Canada, Department of Biochemistry, Microbiology and Immunology, University of Ottawa, 451 Smyth Road, Ottawa, Ontario, K1H 8M5, Canada, and Department of Biochemistry, Universite´ de Montre´ al, 2900 Boulevard Edouard-Montpetit, Montre´ al, Quebec, H3T 1J4, Canada
We describe a simple, powerful, and robust NMR-based method that has the potential to greatly impact the characterization of recombinant protein therapeutics. The method ascertains the bioactive conformational identity of recombinant human granulocyte macrophage-colony stimulation factor (rhGM-CSF) produced in Streptomyces lividans versus Escherichia coli by overlaying their 2D 1H,15N HSQC correlation spectra. An identical match of all resonances implies that rhGM-CSF from both processes share indistinguishable conformations that correlate with in vitro activity. The result of this method is unique among existing methods. It can detect and quantify the active ingredient. Moreover it provides a complete assessment of the conformation with high sensitivity to minor structural changes. The inherently complex nature of biotechnology-derived therapeutics, such as cytokines, hormones, antibodies, or vaccines, makes their overall characterization a very challenging task. The structure of a given protein therapeutic defines its bioactivity. Therefore, assessing the bioactive conformation of a protein is a critical issue in evaluating the bioequivalence between a “biogeneric” and a brand name product. This is especially important in the context of patent expiration on a number of biotechnologyderived therapeutics. For this reason, terms such as subsequent entry biologics (SEB) in Canada, biosimilars in Europe, and followon proteins in the United States have been proposed by regulatory * To whom correspondence should be addressed. E-mail: yves_aubin@ hc-sc.gc.ca. Fax: (613) 941-8933. † Health Canada. ‡ Carleton University. § University of Ottawa. | Universite ´ de Montre´al. 10.1021/ac7026222 CCC: $40.75 Published 2008 Am. Chem. Soc. Published on Web 03/06/2008
agencies and industry.1-3 Currently, bioassays are used to assess the bioactivity, thus providing an inference of the structure. Bioassays often yield false positives and cannot detect structural changes that have little or no effect on bioactivity or that may elicit serious adverse reactions in patients. Thorough assessment of a SEB consequently necessitates elaborate preclinical as well as clinical studies to evaluate safety and efficacy. We demonstrate that NMR spectroscopy, in the form of a fingerprint method, can provide site-specific structural information for recombinant protein therapeutics with only minimal sample preparation. The principle is to assess the bioactive conformation of a recombinant protein therapeutic by comparing its 2D 1H,15NHSQC NMR spectra with those recorded to characterize the reference protein under standardized conditions. Up until the development of cryoprobes, characterizing a therapeutic protein using 1H,15N NMR has been nearly impossible at natural abundance. Methods based on 2D proton NOESY experiments were proposed4 but are very difficult to analyze. As a proof of concept, the strategy was investigated using nonglycosylated recombinant human granulocyte macro-phage colony stimulating factor (rhGMCSF), a cytokine promoting white blood cell proliferation and maturation that is used, among other indications, for the prevention of postchemotherapy infections. MATERIALS AND METHODS NMR Sample Preparation. Singly (nitrogen-15) and doubly (carbon-13 and nitrogen-15) labeled rhGM-CSF were prepared as (1) Woodcock, J.; Griffin, J.; Behrman, R.; Cherney, B.; Crescenzi, T.; Fraser, B.; Hixon, D.; Joneckis, C.; Kozlowski, S.; Rosenberg, A.; Schrager, L.; Shacter, E.; Temple, R.; Webber, K.; Winkle, H. Nat. Rev. Drug Discov. 2007, 6, 437-442. (2) Schellenkens, H. Trends Biotechnol. 2004, 22, 406-410 and references cited therein. (3) Roger, S. D. Nephrology 2006, 11, 341-346. (4) Freedberg, D. I. In State of the Art Analytical Methods for the Characterization of Biological Products and Assessment of Comparability; Mire-Sluis, A. R., Ed.; Karger: Basel, Switzerland, 2005; Vol. 122, pp 77-83.
Analytical Chemistry, Vol. 80, No. 7, April 1, 2008 2623
described elsewhere.5 The buffer was exchanged by ultrafiltration (Amicon, MWCO 5K) with the reference buffer (2 mM sodium phosphate pH 7.4), concentrated down to 300 µL, and transferred to a Shigemi (Allison Park, PA) tube. NMR samples contained a protein concentration of 0.4 mM, 5% deuterium oxide for field frequency lock and 1 mM of 3-(trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS) as an internal chemical shift reference. Sample Preparation of Leucotropin. Samples were provided by Cangene (Winnipeg, Canada) as lyophilized powder meant to be resuspended with 1 mL of sterile water. The resulting solution contained 0.5 mg of Leucotropin, 10 mM sodium phosphate, 40 mM sodium chloride, and 165 mM of mannitol. The sample was prepared by dissolving eight vials in the NMR buffer (2 mM sodium phosphate pH 7.4). The buffer was exchanged to remove the excipients as above. The final sample contained 4 mg of Leucotropin in 250 µL of reference buffer, and 5% of deuterium oxide was added for field frequency lock. The final protein concentration was 1 mM rhGM-CSF. NMR Spectroscopy. Resonance assignment of 13C,15N-rhGMCSF has been reported elsewhere.5 Two-dimensional 1H-15Nheteronuclear single quantum correlation (HSQC) experiments with water flipback pulse, gradient coherence selection, and sensitivity enhancement6 were recorded on a Bruker AVANCE III 600 MHz and AVANCE 700 MHz spectrometers (Milton, Canada) equipped with a cryogenic probehead. Spectra were recorded on doubly labeled rhGM-CSF-wild-type, singly labeled rhGM-CSF-N17D mutant (both produced in Escherichia coli) and natural abundance on Leucotropin. A two-dimensional data matrix of 512 by 32 complex points was collected with spectral windows of 14 and 32 ppm for proton and nitrogen, respectively. Two scans per FID were collected for 13C,15N-rhGM-CSF for a total acquisition time of 5 min, while 1280 or 5120 scans per FID were collected for Leucotropin for total acquisition times of 24 and 96 h, respectively. All spectra were recorded at 25 °C. The residual water signal was deconvoluted with a polynomial function using the software nmrPipe,7 and each vector was multiplied by a shifted sine-square function prior to zero filling and Fourier transform. The indirect dimension was linear predicted prior to multiplication of a cosine function followed by zero filling and Fourier transformation. Visualization of the spectra was carried out using NMRView.8 Limit of Detection. The limit of detection (LOD) was measured in the following fashion. Spectra were recorded on freshly prepared samples containing 0.25 mM (sample A) and 1 mM (sample B) of Leucotropin, as described above. Twodimensional 1H,15N-HSQC experiments were collected on an AVANCE III 600 MHz spectrometer with the identical spectral windows in the proton and nitrogen dimensions. Total acquisition times for sample A were 24 and 96 h and for sample B were 1.5, 6, 12, and 24 h. Only the number of transients per FID was modified, the matrix being recorded and processed as described above. (5) Sauve´, S.; Gingras, G.; Aubin, Y. Biomol. NMR Assign. [Online early access] DOI:10.1007/s12104-007-9070-8. Published Online: Dec 8, 2007. (6) Kay, L. E.; Keifer, P.; Saarinen, T. J. Am. Chem. Soc. 1992, 114, 1066310665. (7) Delaglio, F.; Grzesiek, S.; Vuister, G. W.; Zhu, G.; Pfeifer, J.; Bax, A. J. Biomol. NMR 1995, 6, 277-293. (8) Johnson, B. A.; Blevins, R. A. J. Biomol. NMR 1994, 4, 603-614.
2624
Analytical Chemistry, Vol. 80, No. 7, April 1, 2008
Bioassay. Bioactivity of rhGM-CSF prepared for the NMR characterization was confirmed using a TF-1 cell proliferation assay. TF-1 cells (American Type Culture Collection no. CRL-2003) were maintained as suspension cultures in RPMI 1640 medium (Invitrogen, Burlington, Ontario) containing 2 mM glutamine, 10% fetal bovine serum (Hyclone, Logan, UT), and 2.0 ng/mL purified recombinant GM-CSF (R&D Systems, Minneapolis, MN). Aliquots of 105 cells/mL were transferred in triplicate into 24-well microplates and incubated in the absence or presence of various concentrations of rhGM-CSF from S. Lividans and E. coli and control (rhGM-CSF from R&D Systems). Cell numbers and viability were determined using a hemeocytometer and scoring for trypan blue exclusion. RESULTS AND DISCUSSION At present, the only three-dimensional structure of rhGM-CSF reported in the literature was determined by X-ray crystallography9 (PDB code: 2GMF). The NMR solution structure of rhGM-CSF is being completed in our laboratory and will be reported elsewhere.10 Reference rhGM-CSF was produced in E. coli in the presence of carbon-13 and nitrogen-15 isotopes in order to allow application of multidimensional NMR techniques for structure determination. These data provided the reference spectra for the assessment of the bioactive conformation of rhGM-CSF. Production of rhGM-CSF in E. coli in our laboratory required refolding to assume its bioactive conformation. Correct folding was then confirmed by a TF-1 cell proliferation assay in a dose response fashion. This assay was used by Cangene to assess the activity of Leucotropin. Mass spectrometric (MS) analysis of unlabeled rhGM-CSF by MALDI-TOF yielded a molecular weight of 14 476 ( 2 amu (rhGM-CSF MW ) 14 477 amu). MS analysis also revealed that all amino acid residues were present and intact. There was no evidence of chemical modifications such as hydrolysis or oxidation of side chain moieties. Absence of asparagine deamidation could not be ruled out by MS but could be inferred by NMR. All side chain amide resonances have been observed and assigned but two5 (Q86 and N109). These two are most probably undergoing slow conformational exchange. Nevertheless, any significant deamidation would show very different 2D spectra, as was observed for mutants N17D (see Figure 2b) and N37D (data not shown). Among the numerous experiments carried out to determine the structure of rhGM-CSF, the 2D 1H,15N-HSQC is of particular interest. The pulse sequence uses a flipback water pulse to minimize water perturbation, gradient echo coherence selection, and a sensitivity enhancement scheme.6 This experiment correlates backbone amide 1H,15N resonance pairs. Such data sets are straightforward to collect and are robust to minor setting errors. These two qualities are especially important for industry. Because 1H,15N correlations are highly sensitive to the local environment of each amide pair, a 2D contour map from the HSQC experiment spreads resonances in both axes and offers a high level of resolution. Consequently, the processed data provide a fingerprint, analogous to a biometric measurement, which in turn is unique to the structure of the protein under a given set of solution conditions (pH, ionic strength, etc.). A comparison of a 2D 1H,15N-HSQC of a SEB at natural abundance against a (9) Rozwarski, D. A.; Diederichs, K.; Hecht, R.; Boon, T.; Karplus, P. A. Proteins 1996, 26, 304-313. (10) Sauve´, S.; Gingras, G.; Aubin, Y. Manuscript in preparation.
Figure 1. Overlay of 2D 1H,15N-HSQC spectra of 0.4 mM 13C,15N-rhGM-CSF produced in E. coli (red) and 1 mM (natural abundance) Leucotropin (black) produced in S. lividans. The 15N isotopic ratio between the labeled and natural abundance samples is 133:1; therefore, the contour level used to plot the red spectrum was set for best clarity, but a few peaks are not showing. All resonances are observed and superimposed, when a lower contour level is selected. Whereas a partial assignment is indicated for clarity purposes, only S69 could not be observed. Peaks labeled with an asterisk are impurities only observed in the 13C,15N-sample used for resonance assignment.
reference fingerprint spectrum thus eliminates the need for complete structure determination of the SEB. The 2D 1H,15NHSQC also filters away any signals (e.g., signals arising from excipients) that do not bear a proton-nitrogen pair. An NMR fingerprint method allows one to record 2D NMR spectra for a protein at natural abundance in the low millimolar (0.25-1.00 mM) range in a reasonable amount of time (0.25-4 days). This is quite feasible given the increasing availability of modern spectrometers equipped with cryogenic probes. These provide a significant increase in sensitivity compared to conventional room temperature probes. NMR being a very precise technique will allow the precise characterization of biologics that can surpass bioassays. The assay was tested on Leucotropin from Cangene (Winnipeg, MB, Canada), a 127 amino acid nonglycosylated polypeptide that shares an amino acid sequence identical to that of human GMCSF. Leucotropin is produced in a patented Streptomyces-based expression system that generates secreted rhGM-CSF in its active conformation. Two-dimensional 1H,15N-HSQC NMR experiments were collected on Leucotropin at natural abundance on a Bruker AVANCE 700 MHz equipped with a cryogenic probehead. Spectra were collected at 24 and 96 h of total acquisition time. Very good signal-to-noise ratios were already obtained after 12 h. The overlay of spectra from rhGM-CSF produced in E.coli and Leucotropin is shown in Figure 1. All resonances from both spectra are perfectly matched, thereby demonstrating that both proteins have identical conformations. Moreover, our studies of rhGM-CSF have shown5 two sets of resonances for residues 31-35 indicating that the small R-helix formed in the X-ray structure is in slow conformational exchange in solution between the helical and random coil
conformations at a ratio of 2:1.5 This behavior was also observed on the 2D map of Leucotropin recorded in 96 h. It is noteworthy that this segment is right next to a N-glycosylation site. It is reasonable to say that it is not involved in receptor binding because the carbohydrate chain masks this region in the glycosylated form of the protein. We are studying the effects of the glycosidic chain on the conformational behavior of this peptide segment.10 A major chemical degradation pathway of proteins during production and storage is asparagine deamidation.11 The degradation consists in exchanging the amino group (-CONH2) of an asparagine side chain to an oxygen atom (-COO-, at pH ) 7.4). This chemical process produces an aspartate mutant. In order to investigate the ability of the NMR fingerprint method to resolve such modifications of a product, a mutant (rhGM-CSF-N17D) was prepared, 15N-labeled, and studied by NMR. The choice of N17 was based on the relatively high probability of its side chain to undergo deamidation11 and its strategic position in the tertiary structure. The conformation of the side chain of N17 is fully extended toward the solvent (see Figure 2a), and therefore does not interact with any other residues of the protein. In addition, the N17D mutation is an example of a very small structural perturbation. The overlay of the NMR spectra of the wild-type and N17D mutant is shown in Figure 2b. The simple substitution of a NH2 moiety (N17) for an oxygen atom (D17) modifies the magnetic environment resulting in a number of chemical shift differences (11) Wakankar, A. A.; Borchardt, R. T. J. Pharm. Sci. 2006, 95, 2321-2336.
Analytical Chemistry, Vol. 80, No. 7, April 1, 2008
2625
Figure 2. (a) X-ray structure of human GM-CSF (PDB code 2GMF), (b) overlay of portion of the 2D 1H,15N-HSQC spectra of rhGM-CSF produced in E. coli, the wild-type (red), and N17D (cyan).
in the immediate vicinity (V16 and A18) and on the neighboring strands (Q86, C88, V116, and I117). This results in a distinct fingerprint for the mutant, clearly demonstrating that very small perturbations of the local structure can be unambiguously detected by this method. The current limit of detection (LOD) of the method was determined by recording 2D-HSQC on Leucotropin samples of 0.25 mM (sample A) and 1 mM (sample B) of protein concentration. A 2D spectrum recorded on sample B in 1.5 h already shows the main features of the fingerprint of rhGM-CSF. This suggests that, in some cases, it may be possible/useful to assess the conformation of a recombinant protein with such partial data because changes in the conformation, such as partial unfolding or chemical hydrolysis, will induce a number of chemical shift changes. However, it takes a 2D map recorded in 6 h (sample B) to allow an unambiguous conclusion on the structure. This dataset shows an average signal-to-noise ratio (S/N) of 3. Dividing the sample concentration by a factor of 2 requires an increase of the total acquisition time by a factor of 4. Thus, equivalent results on a sample of protein concentration of 0.25 mM were expected by acquiring during 96 h. This was indeed verified with sample A, and therefore, we consider this to be the limit of detection of the method when applied to rhGM-CSF. The ability to record 2D spectra of rhGM-CSF at natural abundance in relatively short periods of time is made possible by the use of cryogenic probeheads. These offer a gain in sensitivity by a factor of 3-4 compared to conventional probeheads. However, this gain diminishes as the ionic strength of the sample increases.12 Whereas many proteins have been studied by NMR at very low ionic strength, most proteins require salt concentrations of 100-150 mM that are more biologically relevant. Under such conditions, the sensitivity advantage of cryogenic probeheads is only a factor of 2 but does not compromise the method. (12) Kelly, A. E.; Hou, H. D.; Withers, R.; Do¨tsch, V. J. Am. Chem. Soc. 2002, 124, 12013-12019.
2626 Analytical Chemistry, Vol. 80, No. 7, April 1, 2008
Finally, experiments to test potential interactions between excipients and rhGM-CSF were carried out. Spectra of rhGMCSF in the presence of mannitol (up to 250 mM), used as an excipient in the formulation of Leucotropin, did not show any chemical shift perturbations (data not shown). In view of this LOD, it is important to note that the method will provide site-specific information of the active ingredient of a SEB but not of low-level impurities (
