Biophysical Characterization for Product Comparability

Published on: 
BioPharm International, BioPharm International-08-02-2010, Volume 2010 Supplement, Issue 6

Spectroscopic methods such as circular dichroism can detect subtle differences in higher order structure before and after changes in process and formulation.


Drug manufacturers must demonstrate the comparability of their products after process and formulation changes to ensure similar quality, safety, and efficacy. Biosimilars also require evaluation of their equivalency to the innovators' products. By complementing traditional biochemical methodologies, biophysical characterization, using a variety of methodologies, can enhance product knowledge in terms of higher order structure, molecular size distribution, and the properties of aggregates. This article presents three case studies that show the advantages of applying state-of-the-art biophysical techniques in comparability assessments.

Changes in process, formulation, or a manufacturing site often are made in late-phase development or after commercialization of pharmaceuticals for various reasons, including meeting increased demand, improving a quality attribute, or reducing cost of goods. However, because of the complexity in structures and the structure–function relationship of biological therapeutics, such changes may lead to changes in molecular structures, which may adversely affect the quality, safety, or efficacy of the drug. For example, the structures may be changed in such a way that the molecules are more prone to aggregation. Large protein aggregates are considered to be potentially immunogenic.1 It is therefore essential to establish comparability in critical attributes between materials before and after production changes. The industry and regulatory authorities around the world have been discussing, adopting, and improving such practices. The FDA and EMA have published several guidance documents in recent years on comparability for biologics, including one for biosimilars.2–4 In-depth characterization of structure and conformation of biomolecules using physicochemical methodologies provides the primary indication for comparability, although ultimate affirmation of comparability in safety and efficacy can only be based on long-term clinical outcomes. Most physicochemical methodologies have limitations and caveats. Therefore, as stated in ICH Q5E, the industry should "apply more than one analytical procedure to evaluate the same quality attribute" to "maximize the potential for detecting relevant differences in the quality attributes of the product that might result from the proposed manufacturing process change."2


Currently established biophysical techniques enable in-depth characterization of biological molecules in higher order structure, molecular size and size distribution, intermolecular interactions, and conformational stability. For example, circular dichroism (CD) spectroscopy is widely used to evaluate secondary and tertiary structures. Tryptophan emission fluorescence spectroscopy also is very useful to probe changes in structure because of tryptophan's sensitivity to its local environment. Other spectroscopic tools used to analyze protein structures include Fourier transform infrared (FTIR), Raman, and nuclear magnetic resonance (NMR) spectroscopy. Applying a combination of these tools, which are based on different physical principles, maximizes the potential to detect structural changes. These methods also can be used to evaluate conformational stability along with other methods such as differential scanning calorimetry (DSC). Multiple techniques based on different separation mechanisms are available to analyze size distribution. For instance, size exclusion chromatography (SEC) and asymmetric flow field-flow fractionation (AF4 or aFFFF) use hydraulic pressure with and without a stationary phase to separate species of different hydrodynamic volume, while analytical ultracentrifugation sedimentation velocity (AUC–SV) separates species by centrifugal force in the solution phase. Dynamic light scattering (DLS), on the other hand, does not physically separate species, but mathematically resolves the size distribution according to their diffusion coefficients. A large variety of methodologies, including many spectroscopic, calorimetric, and sizing methods mentioned above, can be applied to evaluate intermolecular interactions. Also, biosensor-based techniques such as surface plasmon resonance (SPR), particularly the system offered by Biacore, and biolayer interferometry (BLI), are rapidly becoming key tools of in vitro functional characterization in the biotechnology industry.

This article presents three case studies, which show the advantages of applying biophysical techniques in product comparability assessments.

Case Study 1

Analysis of Higher Order Structure Following Process and Formulation Changes

The far and near UV CD spectroscopy techniques were applied to probe the secondary and tertiary structures of an IgG2 antibody after changes were made in both the downstream processes and the formulation matrix. As shown in Figure 1, the far (panel A) and near (panel B) UV CD spectra of the drug substance lot before the changes (Lot 1) are visually different from the spectra of the two lots after the changes (Lot 2 and Lot 3). This indicates that some degree of change in the IgG2 higher order structure might have occurred in response to process or formulation changes. However, the potential structural changes appeared to be reversible because diluting Lot 2 material with the early formulation matrix (referred to as Lot 2*) resulted in far and near UV CD spectra very comparable to those of Lot 1. No difference was found when the drug substance batches from the early and modified processes were tested for function using an ELISA binding assay, which required substantial dilution in the assay buffer.

Figure 1. The far (A) and near (B) UV CD spectra for IgG2 drug substance lots. Lot 1 was made using the early process, and Lots 2 and 3 were made using the new process and with the new formulation matrix. Lot 2* was a significant dilution of Lot 2 material with the early formulation matrix used for Lot 1. All spectra represent the average of three replicates.

The comparison of CD spectra often is made by visual inspection. It is to some degree subjective, and therefore, may not always be conclusive or straightforward to describe. Here, we demonstrate the use of root mean square deviation (RMSD) to evaluate the similarity of CD spectra. RMSD is widely used in biostatistics and bioinfomatics. It is used to assess the similarity of three-dimensional structures of homologous proteins. The differences in all coordinates of all atoms from the structures in comparison are accounted for with a single RMSD value. The same idea can be applied to spectral comparisons, i.e., the differences in optical signals at all wavelengths can be calculated using the same RMSD formula:

in which NRMSD is the RMSD normalized against the total scale of the reference spectrum, X and Xref are the CD signals for the test and the reference spectrum, respectively, and n is the number of data points. When using the far UV CD spectrum of Lot 2 as the reference, Lot 1 exhibited a higher NRMSD value (6.2%) than Lot 3 (2.3%), suggesting a more significant difference from Lot 2. The buffer change for Lot 2, giving rise to Lot 2*, caused a significant increase in the NRMSD value to 6.6%. A similar trend was observed for the NRMSD analysis on near UV CD spectra (5.1%, 0.9%, and 4.7% for Lot 1, Lot 3, and Lot 2*, respectively). In this case of evaluating the similarity of CD spectra, the result of RMSD analysis is consistent with the visual inspection.

Because a far UV CD spectrum of a protein reflects combined contributions of all secondary structure elements, estimating the structural components by deconvoluting a spectrum potentially can provide another means for semi-quantitative comparison of the far UV CD spectra. There are many algorithms available for the deconvolution based on empirical analysis of model structures.5 It is still debatable how accurate these calculations are.6,7 Several algorithms in the CDPRO package were evaluated to differentiate the far UV CD spectra shown in Figure 1A, and none were found to be sufficiently sensitive to distinguish the differences in the spectra.5 The average calculated fraction of β-sheet is around 45.6%, with a relative standard deviation of 0.3%. We therefore conclude that, even though spectral deconvolution may provide valuable structural information at low resolution, it is not suitable for comparability analyses.


Case Study 2

Analysis of Size Distribution Following Process and Formulation Changes

AUC–SV analysis provides a size distribution profile based on separation by gravitational force in homogeneous solutions, and therefore, is orthogonal to size exclusion chromatography (SEC) when used to evaluate and characterize aggregates. A single speed commonly is used in AUC–SV to separate molecular species. Gravitational sweep AUC, however, applies varying speeds in a single run to expand the dynamic range of analysis to possibly 1.2 μm in the diameter of particles.8 Here, we present an example of applying both approaches to evaluate the comparability of highly heterogeneous samples. Figure 2 shows a profile of single speed AUC–SV for a protein conjugate of ~500 kDa in monomeric form, analyzed using the Sedfit program.9 The conjugate contains a carrier protein, keyhole limpet hemocyanin (KLH), and a peptide, cross-linked to KLH at multiple sites through a specific linker. One drug substance lot (Lot 2) produced by a later process and in a new formulation matrix, is compared to Lot 1 from an early process. The overall profiles of the single-speed AUC–SV are grossly similar, with two main peaks corresponding to the monomer and the dimer. There also are larger species, but in much lesser amounts. However, the sedimentation coefficients of the monomer and dimer in Lot 2 shifted to lower values compared to those in the early lot. The smaller sedimentation coefficients suggest that the monomer and dimer in Lot 2 could be further extended, resulting in larger hydrodynamic radii. This conformational change is likely a response to the changes in formulation conditions, and no impact on efficacy was found in the subsequent preclinical tests.

Figure 2. AUC–SV analysis for the drug substance lots of the KLH conjugate. AUC–SV was performed at a centrifugation speed of 45,000 rpm in the respective formulation matrices. Data were analyzed using the Sedfit program. The sedimentation coefficients were converted to the standard condition at 20 °C in water in the Sedphat program. The profiles are the average of two replicates.

Gravitational sweep AUC has been shown to include much larger species in one sedimentation run.8,10 Although it is still difficult to obtain quantitative information, the method is proven to be valuable for qualitative assessment of size distribution, which also provides an estimate of the appropriate speed to be used in a single speed run for highly heterogeneous samples. Subjected to speeds ranging from 3,000 to 45,000 rpm, the two lots of the KLH conjugate, before and after the process changes, were analyzed using the wide distribution analysis of the SedAnal program, and exhibited similar profiles with regard to very large aggregates (Figure 3).11 There seem to be three pools of large species with the sedimentation coefficients centered at ~120 S, ~400 S, and ~3,000 S, respectively. These peaks are broad, so they may contain a very heterogeneous population of aggregates. The lower limits of the hydrodynamic radius and the molecular weight can be calculated, assuming they are compact spheres, using the following equations:

in which D is the diffusion coefficient, k is the Boltzmann constant, T is the temperature in Kelvin, h is the viscosity of the solvent, Rs is the radius of anhydrous particle, s is the sedimentation coefficient in Svedberg, M is the molecular weight, v bar is the partial specific volume, r is the solvent density, R is the gas constant, and N0 is Avogadro's number. The calculated parameters for the three aggregate populations are shown in Figure 3. It is important to point out that the calculated numbers represent underestimated values. That is because the change in rotor speed can broaden the peaks, and therefore, the diffusion coefficient no longer corresponds to the peak width. For the same reason, the relative abundance cannot be determined by peak area either. However, it is obvious that the very large species have very low quantities in the drug substance.

Figure 3. Gravitational sweep AUC analysis for the KLH conjugate. The centrifugation speed varied from 3,000 to 45,000 rpm. Data were analyzed using the wide distribution analysis in SedAnal.

Another example of applying gravitational sweep AUC is in formulation development for a Qβ virus-like-particle (VLP) conjugated with a peptide at multiple sites. In a formulation matrix with higher salt concentration, SEC analysis suffered significant loss in total peak area, and AF4 suggested the presence of very large aggregates (data not shown). To confirm, the VLP conjugate in low and high salt formulations were analyzed with gravitational sweep AUC (Figure 4). The distribution profiles are markedly different. The protein formed very large and heterogeneous aggregates in the high salt formulation that peaked at ~665 S. The minimum radius and the minimum molecular weight for the equivalent anhydrous sphere are ~32 nm and ~111 MDa, respectively. Particles of this size may not be recoverable by the matrix used in SEC chromatography.

Figure 4. Gravitational sweep AUC analysis for a VLP conjugate formulated with low and high salt concentrations. The centrifugation speed was varied from 3,000 to 20,000 rpm. The size distribution profiles were obtained using SedAnal.

DLS is another powerful biophysical technique that offers different advantages in characterizing aggregates. DLS measurement is simpler and offers higher throughput than AUC, and requires less substantial sample dilution, and is more sensitive to trace amounts of very large particles than SEC and AUC, which makes DLS very useful in detecting large aggregates. Figure 5 presents a DLS analysis of the bacterial Qβ VLP samples to evaluate a downstream process change. The process change did not generate any large aggregates. However, despite the fact that the size distribution is monomodal, the peak has shifted significantly to a larger size, suggesting the presence of small aggregates that are not resolvable by DLS. AUC–SV analysis showed similar results (Figure 6). On the other hand, no substantial difference was detected by either DLS or AUC in samples produced at different manufacturing sites using the original process, indicating that the original process is robust with regard to site change.

Figure 5. Dynamic light scattering for drug substance batches produced at different manufacturing sites or by different manufacturing processes. The scattered light intensity was measured at the scattering angle of 173°. The size distributions were analyzed using the CONTIN algorithm, and averaged over multiple measurements.

Figure 6. AUC–SV for drug substance batches produced at different manufacturing sites or by different manufacturing processes. The samples were analyzed at a centrifugation speed of 20,000 rpm and data were analyzed using Sedfit. Each profile is the average of two replicates.

Case Study 3

Functional Analyses Following Process Changes Using a Biosensor Technology

Ion exchange chromatography is commonly used to indicate charge heterogeneity of protein products. The anion exchange chromatogram of an Fc fusion protein has a single broad peak. The peak elution time of the drug substance produced with a new cell line and new processes was different from that of earlier material by 1 min. The peak width and range remained unchanged. A variety of biochemical and mass spectrometry techniques were used to determine the cause of the peak shift. Only low level variations in the COOH terminal lysine and deamidation in the Fc domain were observed, but their relation to the peak shift was not clear. No change in higher order structures was found by spectroscopic methods (data not shown). An SPR biosensor method (Biacore) was then used to analyze the ligand binding of two drug substance lots, Lot 1 and Lot 2, from the early and the new processes, respectively. Multiple concentrations of the ligand were used in triplicates. The association (ka) and dissociation (kd) rate constants were obtained by global fitting of the sensorgrams to the 1:1 kinetic model using the manufacturer-provided software, and used to calculate the affinity constant, KD = ka/kd. Table 1 lists the results for all three sets, and for clarity, Figure 7 shows the global fitting for one set as an example. The affinity constants of the two lots are within the experimental error. Even though the Fc effector function has not been shown to be involved in the mode of action of this biotherapeutic, binding of the two lots with the high affinity Fcγ receptor, FcγRI, was compared mainly to assess the structural integrity of the Fc domain. As shown in Figure 8, the two lots were comparable.

Table 1. Comparison of binding kinetics and affinity


We have demonstrated that spectroscopic methods such as circular dichroism (CD) can detect subtle differences in higher order structure before and after changes in process and formulation. To facilitate the analysis and presentation of the comparability of CD spectra, the RMSD calculation is proposed to semi-quantify the spectral comparison with a single numeric indicator. This approach is not limited to CD and potentially can be applied to other types of spectra. We also presented different types of AUC analysis for evaluating size distribution in the context of comparability assessments. The results demonstrate that size distribution, as an important property, can be sensitive to process change. Our case studies also show that it is beneficial to use multiple techniques for comprehensive characterization of aggregates, and it is convenient to use AUC and DLS as orthogonal methods to enhance the analysis. Finally, we presented the use of biosensor technology to help address the question of the likelihood that the function of a molecule will be compromised in the event that changes in some properties are detected by other physiochemical assessments.

Figure 7. Overlaid sensorgrams for the ligand-binding of the Fc fusion protein lots from the early (Lot 1) and the new (Lot 2) processes. The Fc fusion protein was captured by preimmobilized protein A in sample flow cells. A ligand sample with a concentration indicated in the figure was injected at 0 sec. The dissociation phase began at 280 s when the ligand injection was switched to the flow of running buffer. The sensorgrams of one set of the triplicates for Lot 1 (pink) and Lot 2 (cyan), as well as the fitted curves (black), are overlaid. All sensorgrams were "double corrected" for the reference flow cell signals and the buffer injection responses.

The biophysical techniques used in these case studies are not validated and are not intended for release tests. However, these methods enable in-depth characterization and are useful for establishing product comparability in physicochemical attributes, which often are related to quality. As suggested in ICH Q5E, significant changes in product quality attributes must be evaluated for their impact on product safety and efficacy.

Figure 8. Overlaid sensorgrams for the FcγRI-binding of the Fc fusion protein from the early (Lot 1) and the new (Lot 2) processes. FcγRI with a His tag at its C-terminus was captured by an anti-His antibody preimmobilized in the sample flow cells. Lot 1 and Lot 2 were injected at 5 and 10 nM in triplicates, and the sensorgrams are overlaid. All sensorgrams were "double corrected" for the reference flow cell signals and the buffer injection responses.


We thank James Carroll, John Steckert, and Ned Mozier for scientific discussion, and Zhaojiang Lu, Monica Brzezinski, Justin Sperry, Mike Dupuis, Min Huang, and Philip Boyle for technical assistance and material supply.

QIN ZOU, PHD, is a senior principal scientist in analytical research and development, BioTherapeutics Pharmaceutical Sciences, Pfizer Inc., Chesterfield, MO, 636.247.1257, Yin Luo, PhD, is a director in analytical research and development, BioTherapeutics Pharmaceutical Sciences, Pfizer Inc., Andover, MA,


1. Rosenberg AS. Effects of protein aggregates: an immunologic perspective. AAPS J. 2006;8:E501–07.

2. US Food and Drug Administration. Guidance for Industry: Q5E Comparability of biotechnological/biological products subject to changes in their manufacturing process. Rockville, MD; 2005.

3. European Medical Association. Guideline on comparability of medical products containing biotechnology-derived proteins as active substance: quality issues. Geneva; Switzerland. 2003.

4. EMA. Guideline on similar biological medicinal products containg biotechnology-derived proteins as active substance: quality issues. Geneva, Switzerland; 2005.

5. Sreerama N, Woody RW. Estimation of protein secondary structure from circular dichroism spectra: Comparison of CONTIN, SELCON, and CDSSTR methods with an expanded reference set. Anal Biochem. 2000;287:252–60.

6. Tetin SY, et al. Accuracy of protein secondary structure determination from circular dichroism spectra based on immunoglobulin examples. Anal Biochem. 2000;321:183–7.

7. Khrapunov S. Circular dichroism spectroscopy has intrinsic limitations for protein secondary structure analysis. Anal Biochem. 2009;389:174–6.

8. Mächtle W. High-resolution, submicron particle size distribution analysis using gravitational-sweep sedimentation. Biophys J. 1999;76:1080–91.

9. Schuck P. Size-distribution analysis of macromolecules by sedimentation velocity ultracentrifugation and Lamm equation modeling. Biophys J. 2000;78:1606-19.

10. Stafford WF, Braswell EH. Sedimentation velocity, multi-speed method for analyzing polydisperse solutions. Biophys Chem. 2004;108:273–9.

11. Stafford WF, Sherwood PJ. Analysis of heterologous interacting systems by sedimentation velocity: Curve fitting algorithms for estimation of sedimentation coefficients, equilibrium and rate constants. Biophysical Chem. 2004;108:231–43.