Elucidation of Acid-induced Unfolding and Aggregation of Human Immunoglobulin IgG1 and IgG2 Fc

Background: Monoclonal antibodies and Fc fusion proteins contain an IgG Fc moiety, which is associated with various degradation processes, including aggregation. Results: Fc unfolding is triggered by the protonation of acidic residues and depends on the IgG subclass and CH2 domain glycosylation. Conclusion: Fc aggregation in acidic conditions is determined by CH2 stability. Significance: Understanding Fc aggregation is important for improving the quality of Fc-based therapeutics. Understanding the underlying mechanisms of Fc aggregation is an important prerequisite for developing stable and efficacious antibody-based therapeutics. In our study, high resolution two-dimensional nuclear magnetic resonance (NMR) was employed to probe structural changes in the IgG1 Fc. A series of 1H-15N heteronuclear single-quantum correlation NMR spectra were collected between pH 2.5 and 4.7 to assess whether unfolding of CH2 domains precedes that of CH3 domains. The same pH range was subsequently screened in Fc aggregation experiments that utilized molecules of IgG1 and IgG2 subclasses with varying levels of CH2 glycosylation. In addition, differential scanning calorimetry data were collected over a pH range of 3–7 to assess changes in CH2 and CH3 thermostability. As a result, compelling evidence was gathered that emphasizes the importance of CH2 stability in determining the rate and extent of Fc aggregation. In particular, we found that Fc domains of the IgG1 subclass have a lower propensity to aggregate compared with those of the IgG2 subclass. Our data for glycosylated, partially deglycosylated, and fully deglycosylated molecules further revealed the criticality of CH2 glycans in modulating Fc aggregation. These findings provide important insights into the stability of Fc-based therapeutics and promote better understanding of their acid-induced aggregation process.

In order to ensure the safety and efficacy of biotherapeutics, it is critical to understand and prevent protein degradation. The presence of aggregates in therapeutic proteins may jeopardize their safety and efficacy by eliciting unwanted immunogenic responses (1,2). Mitigation of aggregation processes while maximizing biotherapeutic shelf-life remains one of the outstanding challenges in biotechnology.
Monoclonal antibodies (mAbs) continue to represent the leading group of biopharmaceutical products (3)(4)(5)(6). All cur-rently approved therapeutic mAbs belong to the IgG class and have a structure schematically depicted in Fig. 1. Intact mAbs are composed of two identical light chains and two identical heavy chains, which are covalently linked via several inter-and intrachain disulfide bonds. The light chains and heavy chains form two (V L and C L ) and four structurally homologous domains (V H , C H 1, C H 2, and C H 3), respectively. The overall IgG structure consists of two identical Fab domains (V L , C L , V H , and C H 1) and one Fc 3 domain (C H 2 and C H 3) that are connected by a flexible hinge region. The Fc portion harbors one conserved Asn-297 glycosylation site in each of its C H 2 domains. The Fab and Fc regions of mAbs have different biological functions. The Fab regions are responsible for binding to the antigen, whereas the Fc portion plays a role in modulating immune cell activity. In addition to mAbs, there are other classes of biotherapeutics, such as Fc fusion proteins, that also contain Fc. These molecules are composed of therapeutically active peptide or protein moieties that are attached to either the C termini or N termini of an IgG Fc. In such cases, the presence of an IgG Fc moiety may result in improved physiological function, ease of production, solubility, etc. However, the Fc region is also associated with a range of degradation processes, including oxidation (7) and aggregation (8). A detailed understanding of how certain structural changes within the Fc domain lead to aggregation represents an important step toward improving the quality of these therapeutic agents.
Fc-based biologics offer significant manufacturing and physiological advantages. Their purification process is greatly simplified by the available selection of affinity resins targeting the Fc portion (9,10). The presence of a relatively large (ϳ50 kDa) and highly soluble Fc moiety confers increased solubility and half-life (11). In addition, the Fc region engages in specific biologically relevant interactions that may require C H 2 glycosylation (antibody-dependent, cell-mediated cytotoxicity; complement activation; in vivo clearance; etc.) (12)(13)(14)(15). Uncovering the various sources of Fc instability that are connected with particular C H 2 glycoforms will enable production of biologics with enhanced pharmacological properties.
In a typical purification process, mAbs and Fc fusion proteins are exposed to acidic conditions during viral inactivation and elution from affinity resins (9,16). It is well known that low pH conditions may result in protein denaturation and aggregation (17,18). It was shown that acidic pH and high ionic strength can promote formation of nonnative protein structures. Some of the best studied, partially folded, acid-denatured states (A-states or molten globule states) are populated at low pH in the presence of salt. For example, an acid molten globule state of cytochrome c is formed at pH 2.0 -2.5 in the presence of 0.5-1.5 M salt (19 -21). Apomyoglobin, ␤-lactamase, and staphylococcal nuclease also exhibit an acid-and salt-induced formation of A-states at low pH (22)(23)(24)(25). Monoclonal antibodies and their fragments are no exceptions to this rule. Buchner and co-workers (26 -28) demonstrated that intact mAbs, Fab regions, and even isolated C H 3 domains form A-states at acidic pH and high ionic strength. Although the stability and structure of these states are highly dependent on the protein and experimental conditions, their common characteristic is a tendency to aggregate (17,29,30). Unlike small, single-domain proteins, mAbs are complex glycoproteins composed of several independently folded domains. Commercial mAb preparations are rather heterogeneous and may contain differentially processed, incompletely glycosylated, and covalently modified forms (31). The understanding of mAb aggregation is challenged by the intrinsic and extrinsic complexity (not to mention the storage history) of antibody preparations.
Despite the aforementioned issues, significant progress has been made in understanding and preventing aggregation in biopharmaceuticals (for a recent review, see Ref. 32). It was recognized that IgG aggregation can be induced by various factors and proceed through different mechanisms (33)(34)(35)(36); however, the role of individual antibody domains in aggregation remained poorly understood. Recently, we proposed that acidinduced aggregation of mAbs is controlled by the stability of C H 2 domains located in the Fc region (8). At the time, no structural evidence was generated concerning the extent of C H 2 unfolding associated with this aggregation process. We are now filling this gap by gathering all of the necessary structural and stability data to implicate the C H 2 domain. The scope of our study was limited to Fc fragments to allow for the use of high resolution two-dimensional NMR and to reduce the number of differential scanning calorimetry (DSC) transitions. In addition, various forms of Fc (i.e. with respect to IgG subclass and degree of C H 2 glycosylation) were analyzed under conditions promoting acid-induced aggregation. As a result, we revealed the aggregation rank order of the most typical IgG Fc domains currently used in biotechnology.

EXPERIMENTAL PROCEDURES
The Escherichia coli-derived IgG1 Fc, CHO-derived IgG1 Fc, CHO-derived IgG2 Fc, and the uniformly 2 H, 15 N-labeled, E. coli-derived IgG1 Fc were supplied by the Protein Sciences group at Amgen, Inc. Intact mAbs (IgG1-A, IgG1-B, IgG2-B, and IgG2-C) were supplied by the Process Development group at Amgen, Inc. All purified proteins were verified greater than 95% pure by SDS-PAGE and size exclusion HPLC. Other reagents and chemicals were of analytical grade or better. All solutions were filtered through a 0.22-m filter prior to use.
Protein Preparation, Identification, and Characterization-The E. coli-and CHO-derived Fc were supplied in 10 mM sodium acetate with 9% (w/v) sucrose at pH 5.2. The purity and identity of the Fc were verified by reversed-phase HPLC and mass spectrometry (see supplemental Figs. S1-S3 for details). The E. coli-derived IgG1 Fc material was the most homogeneous. It contained only one minor impurity, a species presumably with an unpaired disulfide. The CHO-derived IgG1 Fc contained fully glycosylated species of expected mass and three minor species: 1) a singly oxidized species, 2) a species with an unpaired disulfide, and 3) a partially glycosylated species. The CHO-derived IgG2 Fc was more heterogeneous, containing some clips and host cell proteins. Its major fraction was composed of two fully glycosylated species that presumably differed in sulfation.
The IgG1-B-derived Fc fragments with differing levels of C H 2 glycosylation were prepared as follows. Two milliliters of IgG1-B at 6 mg/ml were incubated with 24 l of PNGase F (New England Biolabs, Ipswich, MA) in 1ϫ G7 buffer for 45 min at 37°C. Endoproteinase Lys-C (Roche Applied Science) was then added to the reaction mixture at a protein/enzyme weight ratio of 200:1. The sample was incubated at 37°C for an additional 15 min before quenching with 150 mM ammonium ace- tate at pH 4.7 (37). The sample was then cooled and maintained at 4°C for immediate purification or frozen at Ϫ80°C to arrest further enzyme activity.
Glycosylated, partially deglycosylated, and fully deglycosylated Fc were purified using the CEX method described below (also see Ref. 37). Digested IgG1-B material was loaded onto the column in multiple injections. An Agilent 1200 series HPLC system with a 12/13 SelValve external valve (Agilent Technologies, Santa Clara, CA) was used to perform the fractionation. Resulting fractions containing the same species were pooled and concentrated and then reanalyzed by CEX. Purity greater than 90% was achieved for all three Fc species based on reversed-phase HPLC and mass spectrometry (see supplemental Fig. S4).
Aggregation Experiments-Previously, we demonstrated that CEX can be used to measure the aggregation propensity of both intact and fragmented mAbs (8,35). This same method was applied in the current study, where mAb and Fc mixtures were exposed to different solution conditions. In our experiments, each protein was at 0.5 mg/ml final concentration unless noted otherwise. mAb and Fc mixtures were prepared in a native buffer composed of 10 mM sodium acetate at pH 5.2. They were subsequently diluted into various solutions of interest and incubated quiescently at 30°C for up to several days. Sample aliquots were taken at predetermined intervals and analyzed immediately or stored on ice to reduce further aggregation. The loss of soluble monomer was determined for each individual protein relative to an appropriate control stored refrigerated in 10 mM sodium acetate, pH 5.2. Unstable proteins exhibited a faster decrease in the CEX monomer concentration compared with more stable proteins (8,35).
Cation-exchange Chromatography-Aggregation of mAb and Fc mixtures was investigated by cation-exchange chromatography at pH 5.2 (35,37). The method was run on an Agilent 1100 series HPLC system. Chromatography was performed on a ProPac WCX-10 analytical column (weak cation exchange, 4 ϫ 250 mm; Dionex, Sunnyvale, CA) preceded by a ProPac WCX-10G guard column (weak cation exchange, 4 ϫ 50 mm; Dionex) at 25°C. Protein samples were loaded onto the column and analyzed at a flow rate of 0.7 ml/min. The column was equilibrated with Buffer A (20 mM sodium acetate, pH 5.2), and protein was eluted with a linear gradient of Buffer B (20 mM sodium acetate, 300 mM sodium chloride, pH 5.2) from 0 to 100% over 35 min. Following elution, the column was washed with Buffer C (20 mM sodium acetate, 1 M sodium chloride, pH 5.2) for 5 min and then re-equilibrated with Buffer A for 16 min. Absorbance was measured at 215, 235, and 280 nm. Data were analyzed with Dionex Chromeleon software, and the 280 nm signal was integrated to determine protein peak area.
Differential Scanning Calorimetry-DSC measurements were taken using a VP-Capillary DSC system (MicroCal Inc., Northampton, MA) equipped with tantalum 61 cells, each with an active volume of 135 l. Protein samples, typically at 0.5 mg/ml, were scanned from 20 to 110°C at a rate of 60°C/h following an initial 15-min equilibration at 20°C. A filtering period of 16 s was used, and the data were analyzed using Origin 7.0 software (OriginLab Corp., Northampton, MA). Resulting thermograms were corrected by subtraction of buffer control scans. The corrected thermograms were normalized for protein concentration.
Nuclear Magnetic Resonance-Fc NMR measurements were performed at 25°C using a Varian INOVA 800-MHz NMR spectrometer (Varian Inc., Palo Alto, CA) equipped with a 5-mm triple resonance probe. The uniformly 2 H, 15 N-labeled E. coli-derived IgG1 Fc was tested at 5 mg/ml in 10 mM sodium acetate adjusted by HCl to pH 2.5, 3.1, 3.5, and 4.7.
1 H-15 N HSQC spectra were acquired with 64 experiments run in the 15 N dimension (t1) consisting of 16 scans and 1024 data points in the 1 H dimension (t2). The total experimental time for each spectral acquisition was 37 min. Spectra were processed using NMRPipe (38) and analyzed using NMRView (39). The 1 H-15 N cross-peak assignments from Liu et al. (40) were used. The weighted average chemical shift difference was calculated as described previously (41).
Reversed-phase Chromatography and Mass Spectrometry-Reversed-phase analysis of the E. coli-derived IgG1 Fc, CHOderived IgG1 Fc, and CHO-derived IgG2 Fc was carried out on a Waters (Milford, MA) Acquity system, equipped with a Diphenyl 3 m, 1 ϫ 50-mm column (Varian Inc.) as described previously (42). Typically, 5 g of protein was injected on the column. The column was held at 95% solvent A (0.1% TFA in water) and 5% solvent B (90% acetonitrile and 0.085% TFA in water) for 5 min followed by a gradient from 5% B to 38% B over 13 min. Fc elution was achieved with a linear gradient from 38% B to 46% B in 40 min at a flow rate of 0.05 ml/min. The column temperature was maintained at 75°C throughout the run, and detection was at 214 nm.
Reversed-phase analysis of the IgG1-B-derived Fc was carried out on a Waters Acquity UPLC system as previously described (42). Typically, 5 g of sample was injected onto an Acquity BEH 1.7 m 1 ϫ 50 mm phenyl column. The column was held at 72% solvent A (0.1% TFA in water) and 28% solvent B (90% acetonitrile and 0.085% TFA in water) for 0.7 min. Solvent B was increased to 31.4% at 0.9 min, to 49.4% at 3.4 min, and to 90% at 3.5 min. At 4.10 min, solvent B returned to the starting level (28%) and remained constant until the end of the assay at 5 min. The column temperature was maintained at 80°C throughout the run, the flow rate was kept constant at 0.35 ml/min, and the detection was at 214 nm.
The mass spectrometric analysis was carried out in positive ion mode on a Waters Q-TOF Premier or LCT Premier mass spectrometer equipped with an electrospray ionization source. The capillary and cone voltages were set at 3200 and 60 V, respectively. The desolvation and source temperatures were set at 350 and 80°C, respectively. All other voltages were optimized to provide maximal signal intensity. The instrument was calibrated in the m/z range of 1500 -4000 using multiply charged ions of a standard antibody with a calculated molecular mass value of 148,251.2 Da or commercial trypsinogen with a mass of 32,300 Da. All raw data were processed using Waters MassLynx MaxEnt 1 software to obtain the deconvoluted mass.

Effect of Acidic Conditions on IgG1 Fc Structure via NMR
Analysis-Although Fc is a relatively large protein (ϳ50 kDa), recent studies demonstrated that it is amenable to high resolu-tion two-dimensional NMR analysis (7,43). Furthermore, resonance assignments from Liu et al. (40) made investigation of pH effects on IgG1 Fc structure straightforward. In the current study, the Fc conformation was assessed between pH 2.5 and 4.7 by acquiring a series of 1 H-15 N HSQC spectra of the uniformly 2 H, 15 N-labeled E. coli-derived IgG1 Fc. Due to the low ionic strength of the protein solutions (see "Experimental Procedures"), no evidence of aggregation was seen in any of the Fc samples throughout the experiment. At pH 4.7, the amide peaks of Fc were highly dispersed, which was consistent with a folded conformation (see Fig. 2A). A similar degree of dispersion was seen at pH 3.5, although a number of peaks were reduced in intensity (Fig. 2B). In addition, some new, low intensity peaks emerged that were not present at higher pH. At pH 3.1, a subset of native resonances disappeared, whereas a different set of peaks appeared (Fig. 2C). Spectral properties of these new peaks were characteristic of a disordered, largely unfolded protein conformation. They resembled the minor resonances that were barely visible at pH 3.5. The remaining native resonances disappeared at pH 2.5, where NMR showed limited peak dispersion, consistent with an unfolded state ( Fig. 2D).
At pH 3.5 and 4.7, the number of assigned amide resonances available for analysis was 116 and 117, respectively. This represented 51% of the 227 Fc amino acid residues. Many of the missing peaks originated from the vicinity of the hinge region or were due to peak overlap between the different pH spectra. The number of native resonances dropped to 46 at pH 3.1, and none were present at pH 2.5.
The weighted average, residue-specific, chemical shift changes between pH 3.5 and 4.7 and between 3.1 and 3.5 are shown in Fig. 3. The most prominent changes between pH 3.5 and 4.7 were clustered around residue positions 250 -255 and 310 -315 (Fig. 3A). These regions overlap with two short C H 2 ␣-helices that interface with the C H 3 domains (see "Discussion"). In addition, notable chemical shift changes (Ն0.05 ppm) were associated with positions corresponding to Asp-280, Gln-295, Leu-306, and Thr-335 of C H 2 and Gly-385 and Lys-447 of C H 3. Similar regions produced peaks with reduced intensity at pH 3.5, which probably reflected changes in the C H 2 conformational dynamics. Moreover, some of the native resonances that were present at pH 4.7 apparently disappeared at pH 3.5, among them resonances from Lys-290 and possibly Trp-277 and Val-412 (see Table 1).
Peaks indicating the presence of a folded C H 2 domain were virtually non-existent at pH 3.1, suggesting that major unfolding had occurred (Fig. 2C). Therefore, estimates for the chemical shift changes between pH 3.1 and 3.5 were only available for C H 3 domains (Fig. 3B). At least three residue positions showed significant chemical shift changes at pH 3.1: Arg-344, Trp-381, and Lys-447. Of interest is the Arg-344 residue, the residue near  Table 1). Both of these segments contain residues forming the C H 2-C H 3 domain interface, which probably gets disrupted because of C H 2 unfolding (see "Discussion").
A further reduction in pH from 3.1 to 2.5 resulted in the disappearance of all remaining folded resonances (Fig. 2D). The NMR spectrum of Fc at pH 2.5 was now consistent with an unfolded protein conformation devoid of stable tertiary or secondary structure. The high affinity C H 3-C H 3 interaction was probably disrupted also, as demonstrated by the lack of native resonances originating from the domain contact area. Specifically, this was reflected by the absence of native amide peaks from the following residues: Leu-351, Glu-357, Ser-364, Leu-368, Lys-370, Thr-394, Asp-399, Phe-405, and Lys-409. Because all of these positions were in a native-like environment at higher pH, the data were consistent with a scenario where dissociation and unfolding of the C H 3-C H 3 interchain complex occurred simultaneously (see "Discussion").
CEX Analysis of Fc Aggregation-Previously, we demonstrated the utility of CEX in measuring the aggregation propensity of both intact and fragmented mAbs (8,35). Similar to size exclusion HPLC, CEX is a nondenaturing chromatographic technique that can effectively separate aggregates from monomers. However, in contrast to size exclusion HPLC, CEX can resolve complex mixtures composed of similarly sized proteins. Working with protein mixtures allows us to monitor the aggregation of different molecules simultaneously and under identical conditions. Hence, CEX was selected to establish the rank order of Fc aggregation as a function of C H 2 glycosylation and subclass (IgG1 versus IgG2).
In addition to separating aggregates from monomers, CEX is useful in detecting degraded or chemically modified proteins (31,37). This was an added benefit because our goal was to measure Fc aggregation with minimal interference from chemical degradations. Our initial studies were focused on finding conditions to induce Fc aggregation within a short period of time at moderately elevated temperatures. First, we performed a pH screening experiment using protein solutions buffered with 10 mM sodium acetate to mimic the conditions that were used for NMR. The samples contained a mixture of four different molecules: three full-length mAbs (IgG1-A, IgG2-B, and IgG2-C) and E. coli-derived IgG1 Fc. The choice to use the three mAbs was dictated by our previous experience with these molecules (8,35). They served as internal controls to optimize solvent composition and incubation time to assess Fc aggregation. In agreement with the NMR results, aggregation was not observed in these low ionic strength solutions at pH 3-5 even after 2 days of storage at 30°C (data not shown). This was consistent with the important role of the ionic strength and acid concentration in low pH mAb aggregation (8). Subsequently, the pH screening was repeated in the presence of high (100 mM) sodium acetate with and without 50 mM NaCl (the corresponding solutions are abbreviated as 100Ax and 100AxN, where A represents sodium acetate, x is the pH, and N is NaCl). All four molecules were premixed in 10 mM sodium acetate at pH 5.2 prior to being exposed to the low pH conditions. Fig. 4, A and B, shows CEX chromatographic traces for samples incubated in 100A37N and 100A34, respectively. Fig. 5 summarizes results from various conditions in terms of percentage of monomer recovery based on CEX. It is evident that IgG1-A aggregated only at pH 3.4 (Fig. 5, A and E), whereas aggregation of Fc and the two IgG2s occurred at pH 3.7-4.1 (Fig. 5, B and F, C and G, and D and H, respectively). Previously, we observed that low pH aggregation of mAbs was dependent on C H 2 glycosylation and the IgG subclass (8). In particular, glycosylated IgG1 mAbs were more resistant to aggregation compared with their glycosylated IgG2 counterparts, whereas an aglyco-IgG1 (an IgG1 mAb devoid of C H 2 glycosylation) was the least stable molecule tested (8). Consistent with these findings, Fig. 5 reveals the following aggregation rank order of the four molecules (listed from the highest aggregation propensity to the lowest): E. coli-derived IgG1 Fc (i.e. aglyco-IgG1 Fc) Ͼ IgG2-C Ͼ IgG2-B Ͼ IgG1-A. Aggregation propensity of Fc in 100A34 and 100A34N was particularly high and resulted in the loss of 30 -40% of monomer at t ϭ 0 (Fig.  5, D and H, respectively). Thus, the CEX data demonstrated an increased instability of aglyco-IgG1 Fc compared with glycosylated mAbs. Furthermore, the aggregation rank order for these molecules was the same in either the 100Ax or 100AxN conditions, which indicated the following: 1) the underlying aggregation mechanism was largely unaffected by NaCl, and 2) the rate and extent of Fc aggregation could be appropriately modulated by varying the ionic strength. Because covalent modification and fragmentation were not evident in these experiments (see Fig. 4), protein aggregation was the major degradation process. In summary, sufficient evidence was gathered to support the low pH approach for generating Fc aggregation data.
Our next experiment was performed on a mixture composed of three different Fc moieties: E. coli-derived IgG1 Fc, CHOderived IgG1 Fc, and CHO-derived IgG2 Fc. This mixture was subjected to aggregation in the 100A31N and 100A35N conditions as outlined above. The CEX overlays corresponding to aggregation in 100A35N are shown in Fig. 6A. Quantitative aggregation results in 100A31N and 100A35N are summarized in Fig. 6, B and C, respectively. Despite the heterogeneous nature of the Fc samples, the CHO-derived IgG1 Fc (including all minor forms) was evidently more resistant to aggregation compared with its aglycosylated (E. coli) variant or the CHOderived IgG2 Fc. Aggregation of the latter two molecules appeared similar in 100A35N but differed in 100A31N. In particular, aglyco-IgG1 Fc lost ϳ55% of monomer at t ϭ 0 but aggregated more slowly afterward (gray symbols in Fig. 6C). The initial monomer loss of the CHO-derived IgG2 Fc was less than 40%, but the remaining monomer disappeared rapidly (open symbols in Fig. 6C). Because all three molecules were premixed at pH 5.2 prior to the low pH exposure, this result indicated a lack of stability of the aglyco-IgG1 and glyco-IgG2 Fc. Consequently, the rank order of Fc aggregation was found to be as follows: aglyco-IgG1 Fc Ն IgG2 Fc (CHO) Ͼ IgG1 Fc (CHO). This was consistent with our earlier findings (8) as well as the aggregation rank order that was drawn from Fig. 5. Therefore, a conclusion was made that the 100A31N and 100A35N conditions primarily promoted a C H 2-dependent aggregation mechanism.
Our last aggregation experiment utilized differentially glycosylated Fc fractions generated from another IgG1 mAb, IgG1-B. The success of this experiment depended on 1) the ability of CEX to resolve Fc fragments with different levels of C H 2 glycosylation and 2) the optimization of PNGase F treatment to achieve an optimal ratio of glycosylated, partially deglycosylated, and fully deglycosylated Fc for purification. The ability of CEX to separate differentially glycosylated Fc was verified by analyzing PNGase F-treated and -untreated IgG1-B following Lys-C limited proteolysis (data not shown). Subsequently, IgG1-B and PNGase F concentrations were varied along with incubation temperature and duration to achieve an optimal rate of digestion and desired ratio of glycosylated, partially deglycosylated, and fully deglycosylated Fc. Storage temperature and duration after digestion were also assessed to ensure that this ratio was sufficiently maintained over the course of purification (see "Experimental Procedures"). The resulting Fc fractions were verified by reversed-phase HPLC and mass spectrometry (see supplemental Fig. S4), mixed together, and subjected to an aggregation process in 100A31N. The corresponding CEX results are shown in Fig.  7, A and B. In agreement with the data in Fig. 6C, glycosylated Fc was more resistant to aggregation compared with its    lated Fc. Such results provided compelling evidence for the importance of C H 2 glycosylation in determining the rate and extent of Fc aggregation.
Effect of Acidic Conditions on Fc Stability via DSC Analysis-Thermostability of E. coli-derived IgG1 Fc, CHO-derived IgG1 Fc, and CHO-derived IgG2 Fc was assessed under conditions mimicking the CEX aggregation experiments (see above). DSC samples were made by diluting protein stock solutions into different 100Ax buffers, except for pH 7 controls that were prepared in 100 mM sodium phosphate (100NaP i 70). Fig. 8 shows DSC traces for the three Fc variants. A summary with all of the DSC data as a function of pH and apparent T m is given in Fig. 9 and Table 2. Although IgG Fcs are known to consist of two different, independently folded domains, C H 2 and C H 3, some of the DSC profiles contained an additional high temperature peak. We will refer to this peak as an "A-state" in accordance with Buchner et al. (26). At pH 7.0, the DSC profile of the E. coli-derived IgG1 Fc was characterized only by the presence of C H 2 and C H 3 transitions (Fig. 8A). Between pH 3.5 and 5.2, an additional A-state transition was present, giving rise to a characteristic three-peak profile (Fig. 8, B-E). Below pH 3.5, the aglycosylated C H 2 domain was unfolded and no longer produced a peak. As a consequence, the corresponding DSC profile contained only C H 3 and A-state transitions (Fig. 8F).  Results for CHO-derived IgG1 Fc were similar to those for E. coli-derived IgG1 Fc with respect to the C H 3 and A-state transitions (Fig. 8). The latter transition was now seen across the entire pH range, including pH 7.0 (Fig. 8A). As expected, glycosylated C H 2 had a higher T m compared with its aglycosylated counterpart, confirming the important stabilizing role of C H 2 glycans (see Fig. 9A and Table 2). The magnitude of this difference increased as pH decreased; the initial T m difference of ϳ6°C reached ϳ13°C when the pH dropped from 7.0 to less than 4.0. C H 3 of the CHO-derived Fc also appeared to have a higher T m but only at a pH below 3.5 (Fig. 9A). Such a result indicated that C H 3 domains in the aglyco-Fc were affected by the C H 2 instability (unfolding) at pH Ͻ 3.5, in agreement with the NMR data.
The DSC profiles of the CHO-derived IgG2 Fc (Figs. 8 and 9) generally resembled those of the CHO-derived IgG1 Fc. The IgG2 Fc differed notably from IgG1 Fc in the following ways: 1) the A-state transition was less frequently observed; 2) the C H 3 transition had a lower T m across the entire pH range; and 3) the T m of the C H 2 transition was comparable at pH 5.2-7.0, but a further reduction in pH revealed a stability difference. Specifically, the T m of the IgG2 C H 2 was much more dependent on pH and decreased significantly. The maximal difference seen in this study (ϳ7°C at pH 3.5) agreed with our previously reported value obtained at pH 3.5 in the presence of 500 mM NaCl (8).
The following domain properties were the same across all three Fc types (see Fig. 9): 1) the T m of C H 2 was higher at pH 7.0 compared with 5.2, and it decreased in a non-linear fashion upon solution acidification; 2) the T m of C H 3 was nearly identical between pH 7.0 and 5.2 but decreased in a non-linear fashion upon further acidification; and 3) the T m of the A-state seemed to be lower at pH 7.0 than at 5.2. It proceeded through a maximum at pH ϳ5.2 before undergoing a non-linear decrease upon further pH reduction.
In order to understand the nature of the A-state transition, the concentration dependence of the thermograms for CHOderived IgG1 Fc was investigated. For this purpose, 0.1, 0.2, 0.3, and 0.4 mg/ml protein samples were prepared in 100A35 and  Table 2. Note that in F, the E. coli-derived IgG1 Fc was at pH 3.3, whereas both of the CHO-derived Fc were at pH 3.0.
subsequently analyzed by DSC. Although the signal at low protein concentrations was weak, triplicate runs reproducibly revealed the concentration-dependent nature of the A-state.
The characteristic three-peak profile was seen in all cases, but the enthalpy of the A-state peak visibly decreased at 0.1 mg/ml protein (see supplemental Fig. S5). 15 N HSQC NMR is a powerful approach that allows a detailed understanding of protein structural changes. Changes in chemical shifts may originate from changes in hydrogen bonding because there is a strong correlation between hydrogen bond energies and amide proton or amide nitrogen chemical shifts (45). Protein unfolding is expected to result in chemical shift and peak intensity changes, but some regions of a molecule may exhibit greater changes in conformational dynamics or in the electronic/electrostatic environment than others. This may especially be the case with proteins composed of structurally distinct domains, such as mAbs and their Fab and Fc fragments.

High Resolution Structural Analysis of Fc Unfolding-Analysis of chemical shift perturbations via 1 H-
An IgG Fc molecule is composed of two types of independently folded domains, C H 2 and C H 3. Thus, denaturation of Fc is expected to be a multistage process, where changes in one domain are not fully coupled with changes in another. Until now this view was supported by evidence from differential scanning calorimetry and optical spectroscopy studies (46,47). It has also been known that C H 2 domains tend to be less stable than C H 3 under a large variety of conditions. Our residue-specific two-dimensional NMR analysis provides direct evidence for both of these notions. We demonstrate that structural changes in C H 2 and C H 3 are largely uncoupled and that unfolding of C H 2 precedes that of C H 3. This mainly follows from the following: 1) the chemical shift and peak intensity changes at pH 3.5, indicating increased susceptibility of C H 2 to structural perturbations (Fig. 3A), and 2) the absence of native peaks from C H 2 at pH 3.1, in contrast to the presence of at least 46 assigned resonances from C H 3 (Figs. 2C and 3B).
Equilibrium guanidine HCl unfolding of isolated C H 2 was previously shown to be consistent with a two-state process (48). Results from our study suggest a more complex unfolding sce-  nario that invokes partially denatured conformations. Feige et al. (48) derived their conclusion based on evidence provided by intrinsic fluorescence and circular dichroism. Because of limited sensitivity and structural resolution, optical spectroscopy may not always reveal nearly native protein conformations accessible by NMR. The observed discrepancy may also be explained by the following: 1) differences in the experimental pH (the guanidine HCl unfolding studies were performed at pH 7.5), 2) differences in the mode of protein denaturation (chaotropeversus acid-induced), or 3) differences in the unfolding mechanism of isolated C H 2 domains versus C H 2 domains within the context of an IgG Fc. All of these aspects could possibly be addressed by performing NMR-based unfolding experiments on isolated C H 2. However, such studies were beyond the scope of our investigation. Structural characteristics of C H 2 at pH 3.5 are indicative of partial denaturation caused by protonation of acidic side chains. Particular suspects are the Glu-380 and Glu-430 residues that form salt bridges with Lys-248 and Lys-338, respectively. Both glutamic acid residues reside in C H 3, and their side chains are expected to be largely protonated at pH 3.5 (pK ␣ of free glutamic acid is 4.07). Their C H 2 partners, Lys-248 and Lys-338, lay in close proximity to the two C H 2 ␣-helices that undergo the most significant perturbation at pH 3.5 (Fig. 3A). The crystal structure of IgG1 Fc (Protein Data Bank entry 1HZH (49)) provides evidence that the Lys-248 -Glu-380 and Lys-338 -Glu-430 ion pairs may be required for stabilization of the C H 2-C H 3 interaction area. Therefore, it is plausible that the loss of such interactions due to the Glu-380 and Glu-430 protonation could modulate the structure of the C H 2-C H 3 interface. It is important to note that the same region represents the Protein A-binding site, and its impairment causes acid elution of mAbs from Protein A resins. Some C H 2 destabilization may also be brought about by the protonation of Asp-312, a residue that resides in one of the two C H 2 helices and interacts with Lys-317. Indeed, the amide peak of Asp-312 shows a significant chemical shift perturbation and intensity loss at pH 3.5 (Fig.  3A). Another interesting observation is related to the Leu-306 and Trp-277 side chains that are packed against each other in the hydrophobic core of the C H 2 domain. At pH 3.5, the Leu-306 resonance undergoes a notable chemical shift change, whereas the peak for Trp-277 possibly disappears (Table 1). Moreover, chemical shift changes are evident for the amide peak of the Gln-295 residue that is located near the hinge region and is far removed from the aforementioned helices. This indicates that changes in the C H 2 domain structure may extend beyond the C H 2-C H 3 interface (see Fig. 10, A and B). The potential outcome of these changes is the exposure of buried hydrophobic surfaces that can prime Fc for aggregation. The scenario and regions involved appear to be different from the computationally predicted aggregation-prone regions in C H 2 (50). However, because of the differences in the experimental pH (SAP analysis was performed at pH 6.5), a direct comparison of the data may not be appropriate.
In contrast to C H 2, structural changes in C H 3 are limited at pH 3.5. We only note the chemical shift changes for Gly-385 and Lys-447 and the potential absence of the Val-412 resonance (Table 1). At pH 3.5, native resonances from C H 3 remain rela-tively unperturbed, indicating that it maintains an overall folded structure (Fig. 3A).
The reduction in pH from 4.7 to 3.5 is associated with only a partial denaturation of C H 2, whereas pH reduction from 3.5 to 3.1 results in a major loss of tertiary and secondary structure. In contrast, unfolding of C H 3 only takes place below pH 3.1, as evidenced by the dispersion of its amide resonances. Although many of the native C H 3 resonances are missing at pH 3.1, this is probably related to the denaturation of C H 2 rather than C H 3. This is supported by the location of the majority of the missing peaks at the C H 2-C H 3 interface (positions 366 -380 and 428 -438; see Fig. 10, C and D, and Table 1). However, the absence of some of the native C H 3 peaks may simply be caused by an overlap with resonances originating from denatured C H 2. Furthermore, there is indirect DSC evidence suggesting only a slight difference in C H 3 T m between the E. coli-and CHO-derived IgG1 Fc at pH below 3.5 (Fig. 9A). In the case of these two Fc moieties, the difference in stability is probably determined by the presence or absence of C H 2 glycans. Because the magnitude of the C H 3 T m difference is rather small, we can deduce that C H 2 denaturation has only a minor impact on the stability of the adjacent C H 3. The C H 3-C H 3 contact area also remains relatively unperturbed, but some changes within these domains do take place, as evidenced by the Trp-381 and Tyr-391 resonances. The side chains of these two residues form a tertiary contact by packing against each other in the hydrophobic core. At pH 3.1, the Trp-381 resonance exhibits notable changes in its chemical shift and intensity, whereas the peak for Tyr-391 disappears entirely (Fig. 3B and Table 1). Furthermore, Val-369 is another deeply buried C H 3 residue that is missing its native resonance at pH 3.1 (Table 1). Therefore, it can be concluded that although the C H 3 domains are folded at this pH, their tertiary structure is not fully native.
A further reduction in pH from 3.1 to 2.5 brings about major unfolding of C H 3, as evidenced by the disappearance of resonances from its folded state (Fig. 2D). An important feature of the C H 3 domain is the formation of a tightly bound C H 3-C H 3 dimer. Upon inspection of the Fc crystal structure, a number of C H 3-C H 3 interdomain interactions can be found. Among them, the ion pairs Glu-357-Lys-370 and Asp-399 -Lys-409 clearly stand out. Although no direct NMR evidence is currently available, one can hypothesize that protonation of the Glu-357 and Asp-399 residues might contribute to C H 3 unfolding at pH 2.5. Protonation of these side chains would be expected to result in destabilization of the C H 3-C H 3 dimer due to the loss of four interdomain salt bridges. Another interaction that may be influenced by the same mechanism is the ion pair formed between Asp-356 and Lys-439. A more detailed analysis of the specific role of acidic side chains in Fc stability will be presented elsewhere.

C H 2/C H 3 Stability and Fc Aggregation in Acidic
Conditions-Direct investigation of Fc aggregation by 1 H-15 N HSQC NMR is precluded by the large size of products of this reaction. To forestall unwanted aggregation, our NMR measurements were performed under low ionic strength conditions. These experiments provided information on the location and extent of structural changes in C H 2 and C H 3, which is necessary for improved understanding of Fc aggregation. Our DSC experi-ments allowed us to extend this analysis to the CHO-derived IgG1 and IgG2 Fc, for which isotopically enriched material was not available. The intrinsic susceptibility of C H 2 to pH-induced changes is reflected by its T m versus pH profile. For example, the pH profile of C H 2 from the E. coli-derived IgG1 Fc is less linear compared with its CHO-derived counterpart (Fig. 9A). Similarly, the CHO-derived IgG2 Fc shows less linear dependence on pH compared with the CHO-derived IgG1 Fc (Fig. 9B). The increased susceptibility of aglycosylated C H 2 is probably caused by the absence of stabilizing interactions from carbohydrates; the reasons for the higher susceptibility of IgG2 C H 2 are less obvious. Previously, we speculated on the role of the C H 2 sequence variations between the IgG1 and IgG2 subclasses (8).
In particular, we hypothesized that the region surrounding the Y IgG1 300F IgG2 substitution (Table 3) could be associated with IgG2 C H 2 instability at low pH. Our NMR data at pH 3.5 for E. coli IgG1 Fc lend support to this idea. As revealed by the chemical shift analysis, the structural environment of Gln-295 undergoes a significant change at this pH (Fig. 3A). Gln-295 is located near a tertiary contact formed by the side chains of His-268, Glu-294, and Y IgG1 300F IgG2 . It is reasonable to presume that protonation of Glu-294 at pH 3.5 would destabilize the His-268 -Glu-294 charge-charge interaction. The specific residue located at position 300 may further influence the energy  C and D). In all cases, the hinge region is at the top, and the C H 3 domains are at the bottom. In B and D, the molecule is rotated by 90°along the vertical axis parallel to the plane of the page. Unperturbed residue positions for which resonance data are available are colored blue. Residues colored red undergo weighted average chemical shift changes equal or exceeding 0.05 ppm (see "Results" and Fig. 3). Positions colored magenta correspond to residues that lost their native resonances (see "Results" and Table 1). Residue positions for which resonance data are not available are shown in gray. The figures were generated using PyMOL (Schrodinger, LLC, New York). of this contact. Indeed, the T m versus pH profiles of C H 2 shown in Fig. 9 are generally consistent with protonation of a glutamic acid residue(s). It would then follow that the mechanism of destabilization of the region, including Gln-295, is similar to the one proposed for the two C H 2 helices at the C H 2-C H 3 interface (see above). Such scenarios tie together the widespread structural perturbations in C H 2 at pH 3.5 with the location of key acidic residues in the Fc structure. Other mechanisms, including anion binding, may contribute to this destabilization, especially when the ionic strength is increased.
The above considerations are consistent with our previous studies on low pH mAb aggregation. Ten different IgG2s that were studied earlier showed an increased aggregation propensity compared with five different IgG1s (8). This subclass dependence correlated with reduced thermostability of IgG2 C H 2 relative to IgG1 C H 2. In addition, regardless of the subclass, C H 2 glycosylation was found to be an important attribute that determined stability and aggregation of Fc and intact mAbs. Some of these conclusions have already been confirmed outside of our group (50).
It is interesting to note that the T m versus pH profiles for the C H 3 domain of CHO-derived IgG1 and IgG2 Fc are very similar but offset by 6 -7°C (Fig. 9B). There are only three sequence positions that differ in these domains between the two molecules (see Table 3): D IgG1 356E IgG2 , L IgG1 358M IgG2 , and V IgG1 397M IgG2 . Therefore, some or all of these residues must be responsible for the reduced stability of IgG2 C H 3. Commercial IgGs tend to contain either Asp-356/Leu-358 or Glu-356/ Met-358 substitutions (8,51). Thus, an opportunity exists to assess the impact of these residues without the need for new mutant construction. Our preliminary experiments on an IgG1 containing Asp-356/Leu-358 versus an IgG1 with Glu-356/ Met-358 revealed no difference in their C H 3 domain stability (data not shown). Given that this finding is confirmed in the case of other IgGs, this result leaves the V IgG1 397M IgG2 mutation as the culprit for the reduced stability of IgG2 C H 3.
The CHO-derived IgG1 Fc provided the most complete pH profile for the A-state transition (Fig. 9). For reasons not fully understood, the E. coli-derived IgG1 Fc and CHO-derived IgG2 Fc failed to produce this transition at some of the pH values. One possible explanation for this is the narrow temperature range (20 -110°C) that was probed in our experiments. Because the A-state transition tends to occur at Ն100°C, it may have been necessary to extend the DSC measurements up to at least 120°C. Another possibility is to use higher protein concentrations for improved detection of this low enthalpy transition. Nevertheless, three important conclusions can be made based on the available data. First, the A-state transition does not seem to be affected by C H 2 glycosylation (or C H 2 (de)stabilization as a result thereof), suggesting that it forms independent of C H 2 (Fig. 9A). This dovetails nicely with earlier findings that show that the A-state is formed by isolated C H 3 domains (28). Such an agreement serves to confirm that the A-state formation might be C H 3-dependent even in the case of a full-length Fc. Second, the T m versus pH profiles of the A-state for CHO-derived IgG1 and IgG2 Fc appear similar but are offset by several°C (Fig. 9B). The reason for this is currently unknown, and further studies in this area are warranted. Third, the protein concentration dependence of the A-state suggests that its formation requires the association of several Fc molecules (supplemental Fig. S5). This is again consistent with the previous report from Thies et al. (28). In particular, they showed that C H 3 domains formed a defined oligomer consisting of 12-14 subunits at pH 2 in the presence of salt. Although an assessment of the Fc oligomeric structure was beyond the scope of our investigation, the appearance of a well defined DSC transition is consistent with the formation of a distinct protein complex or aggregate.
There are still many questions regarding the A-state that need to be addressed in order to fully understand Fc aggregation. Semantically, it is unclear whether it is appropriate to call it an acid-denatured state if it is present even at neutral pH ( Fig.  9). Our findings to date favor the idea that Fc aggregation in acidic conditions is primarily determined by its C H 2 domains. Nevertheless, questions on the possible contributing role of the A-state (or C H 3 domains) in Fc aggregation remain unanswered. There is also a necessity to discuss the value of calorimetrically determined stability in predicting Fc aggregation. For instance, is there a correlation between DSC and increased aggregation of aglyco-IgG1 Fc or glyco-IgG2 Fc compared with glyco-IgG1 Fc? Fig. 9 shows that thermostability of the A-state does not correlate with aggregation. For example, the T m versus pH profiles of the A-states for aglyco-and glyco-IgG1 Fc overlay almost perfectly (Fig. 9A), yet aggregation propensities of these two Fc moieties are dramatically different (Fig. 6). The A-state transition of the glyco-IgG2 Fc occurs at higher temperatures than that of the glyco-IgG1 Fc (Fig. 9B), yet it is the CHO-derived IgG2 Fc that aggregates more readily (Fig. 6). The same reasoning applies to C H 3 domains. Their stability in the aglyco-and glyco-IgG1 Fc is similar until pH Ͻ3.5 (Fig. 9A), yet aggregation of the aglyco-Fc is already under way at pH 3.7 (Fig. 5, D and H). As far as the IgG1 and IgG2 Fc, their C H 3 profiles are substantially shifted along the x axis (Fig. 9B). Assuming that C H 3 denaturation was driving aggregation, aggregation of IgG2 Fc would be higher compared with both glyco-and aglyco-IgG1 Fc. Is this really the case? Fig. 9C shows an overlay of the T m versus pH profiles for E. coli-derived IgG1 Fc and CHO-derived IgG2 Fc. Analysis of these profiles along with the results from Fig. 6 helps to identify domains that correlate with aggregation. The aglyco-IgG1 Fc experiences a much larger loss of monomer at t ϭ 0 in 100A31N (Fig. 6C), which is consistent with the lower stability of its C H 2 domains (Fig. 9C). If Fc aggregation was primarily dependent on C H 3, one would have predicted the opposite result. Thus, we can conclude that C H 2 stability differences from calorimetry are generally reflective of Fc aggregation propensity. However, a prediction based solely on T m is not recommended, because aggregation may also be influenced by structural perturbations or kinetic aspects that escape DSC detection.

Relationship of C H 2 Glycosylation and Product Quality-
The overall goal of biopharmaceutical production is the achievement of the highest possible yield and purity with reduced aggregate levels. The recently introduced concept of "quality by design" provides a description of the desired state for manufacturing (52). Understanding the relationship between product quality and manufacturing is one of the main expectations of quality by design. Here we will discuss how knowledge of Fc aggregation mechanisms may help improve quality of Fc-based therapeutics.
In the past few years, understanding of the importance of C H 2 glycosylation has improved dramatically (14,15). Apart from its biological function, C H 2 glycosylation may now be viewed as one of the attributes determining mAb aggregation during production and storage (8,50). The majority of therapeutic mAbs are produced in a glycosylated form via the CHO cell expression system. Subsequent purification steps are streamlined and tightly controlled to ensure consistency and reproducibility of product quality. However, existing purification platforms rarely result in bulk material that is devoid of incompletely glycosylated species. The presence of such molecules is routinely assayed via rCE-SDS (14). Disulfide reduction and denaturation of IgGs forces them to dissociate into constituent light and heavy chains that can be separated based on size. Migration of the latter is influenced by the presence or absence of C H 2 glycans, resulting in the detection of two distinct heavy chain peaks. Typical rCE-SDS estimates for non-glycosylated heavy chain are on the order of 1% for different mAb preparations. 4 Assuming that there is only one non-glycosylated heavy chain per antibody molecule, this would translate into 2% of partially glycosylated species in the purified bulk. Is there a particular role for these species in mAb and Fc aggregation? Is their presence a quality attribute that reflects risks of aggregation in biopharmaceutical preparations?
Recent studies showed that aglycosylated mAbs and Fc are unstable in acidic conditions and at elevated temperature (8,50). However, no information was available on the aggregation propensity of a fully glycosylated molecule relative to its aglycosylated and partially glycosylated forms. To the best of our knowledge, Fig. 7 provides the first illustration of the effect of all three levels of C H 2 glycosylation in the case of Fc. The data shown therein were generated from a protein mixture to ensure identical conditions for establishing aggregation propensity. It is evident that Fc aggregation correlates strongly with the degree of C H 2 glycosylation. Because IgG aggregation in acidic conditions proceeds predominantly via a C H 2-dependent pathway (see above), we can argue that such results are relevant also for intact mAbs and Fc fusion proteins.
The fate of partially glycosylated species is determined by their exposure to stress conditions triggering nonnative aggregation. A typical manufacturing process is a matrix of conditions, some of which are known to be denaturing (acidic pH during viral inactivation, freezing and thawing, stirring, etc.). Therefore, it is possible that biopharmaceutical preparations containing increased levels of such species are at a higher risk of forming aggregates. Even a fraction of denatured, partially glycosylated species that could form immunogenic aggregates may negatively impact product quality. Moreover, the probability of aglycosylated species (species that lack both heavy chain carbohydrates) may be low, but it is far from negligible. Such species are of even greater concern because of their increased aggregation propensity (Fig. 7) and the inability of most current techniques to detect them. In particular, their direct assessment is complicated by the prevalence of partially glycosylated mAbs.
The CEX traces in Fig. 6A attest to the fact that C H 2 glycosylation and IgG subclass must be the main determinants of Fc aggregation. This follows from their overriding effects on stability, even when considering all of the major and minor species that are associated with Fc heterogeneity. Therefore, considering all of the evidence generated in this and preceding work (8), there is sufficient knowledge to be applied in a practical test. It is likely that opportunities exist to improve the quality of biotherapeutics by increasing the level of C H 2 glycosylation, assuming that this does not interfere with functional requirements (Fc effector functions, in vivo clearance, etc.). This may be achieved by improving cell culture and/or purification processes. Alternatively, C H 2 aggregation may be mitigated by finding appropriate formulation and/or protein engineering strategies. If tested, some of these ideas may prove beneficial in reducing aggregation issues associated with mAb-and Fc-based therapeutics.
The discussion above should not be viewed as an argument against the production of aglycosylated biologics. To the best of our knowledge, they can be successfully developed given that sufficient efforts are spent to optimize manufacturing, storage, and delivery conditions. Our goal was to highlight that the aggregation propensity of fully glycosylated, partially glycosylated, and aglycosylated mAbs can vary significantly and that the cause of protein instability may not necessarily be associated with the main product but rather with some of its unstable forms that escape detection.

CONCLUSIONS
A combined use of two-dimensional NMR, DSC, and CEX proved useful for gaining a detailed understanding of the mechanisms of unfolding and aggregation of IgG Fc. Fc aggregation under acidic conditions was found to be primarily determined by the C H 2 domain stability. This process appeared to be triggered by C H 2 unfolding associated with the protonation of specific acidic residues. The rate and extent of Fc aggregation were shown to be highly dependent on the subclass (IgG2 Fc was less stable than IgG1 Fc) and the degree of C H 2 glycosylation. The ionic strength of the solution played an important role in Fc aggregation under acidic conditions.