CDR clipping-induced heterodimerization: identification of a novel dimerization mechanism in a co-formulated antibody cocktail via a multifaceted mass spectrometry approach

ABSTRACT

Co-formulated antibody cocktails are becoming an increasingly popular therapeutic class; however, they present analytical challenges over traditional single monoclonal antibody (mAb) formulations. One paramount concern is the formation of heteromeric species that have unknown impacts on safety and efficacy. Consequently, effective approaches for identifying and characterizing high-molecular weight (HMW) impurities are critical to the successful development of this therapeutic class. In this study, we used a multifaceted mass spectrometry approach to characterize a unique dimer species formed between two co-formulated mAbs under thermal stress, revealing an intriguing dimerization mechanism that is driven by complementarity-determining region clipping-induced domain swap. Size exclusion chromatography-mass spectrometry, complemented by post-column denaturation, was utilized at both intact and subunit levels to pinpoint the dimerization interface. Additionally, by probing the disulfide bond susceptibility changes via limited reduction and middle-down analysis, the structural changes of the involved domains were studied. These results highlight the critical role of sophisticated analytical methods in comprehending and addressing the complexities linked to co-formulated mAb cocktails.

Introduction

Therapeutic antibodies have emerged as a crucial treatment option for various severe diseases, with over a hundred monoclonal antibody (mAb) products approved by regulatory agencies in the past 35 years.^1,2 This widespread adoption underscores their utility; however, despite their popularity, there are inherent limitations in treatments composed of a single, highly specific molecule with a single target. Mutations in the antigen have the potential to render single therapies ineffective,³ as exemplified by COVID-19⁴ and cancer.⁵ To combat mutation, the immune system relies on multiple antibodies targeting a host of different epitopes.^6,7 Consequently, pharmaceutical companies are using various modalities and strategies to create more effective therapeutic interventions that more closely resemble aspects of natural immune response, such as bispecific antibodies and combination products.^8–10 Combination products offer a variety of approaches for the concurrent administration of multiple therapeutics. These strategies include methods like dual-barrel syringes and multiple infusion bag single lines, which facilitate co-administration of existing formulations. Additionally, there are more integrated techniques such as co-formulation, where therapeutics are combined into a single product for simultaneous delivery.

Co-formulation represents the most integrated approach to combination therapy, where multiple drug substances are combined into a single formulation to deliver therapeutic agents concurrently, maximizing clinical efficacy and patient convenience. This method not only enhances clinical outcomes by harnessing synergistic effects, but also simplifies treatment protocols, reducing administration frequency, and minimizing the risk of medical errors.^8,10 As a result, numerous pharmaceutical companies are investigating co-formulation opportunities. Three fixed-dose combinations (FDCs) are approved in the United States as of 2022.^10–14 There are also several other FDCs that are currently in various stages of clinical trials.¹⁰ While co-formulation offers substantial therapeutic advantages, it also introduces complex analytical challenges that must be addressed to ensure product safety and efficacy.

FDCs present unique analytical challenges due to the increased sample complexity, necessitating additional attention during their development.^15–18 Co-formulated antibodies, similar to their single mAb counterparts, can experience changes in post-translational modifications (PTMs), as well as variations in charge and size profiles, during storage and under stress conditions, resulting from various physical, chemical, or enzymatic degradations.^{16,17,19–22} In particular, the high-molecular weight (HMW) species often pose significant analytical challenges due to their considerable heterogeneity in size, conformation, and interaction mechanisms.²³ Unlike individually formulated mAbs, co-formulated antibodies can potentially interact, forming heteromeric species. These HMW species have unknown impacts on product safety and efficacy,^15,16,24,25 necessitating effective techniques to monitor and characterize them throughout the product lifecycle.

Common techniques for elucidating HMW species in single-mAb formulations include sedimentation velocity analytical ultracentrifugation (SV-AUC),^26,27 size-exclusion chromatography (SEC) coupled with ultraviolet detection (UV),^28,29 and SEC-multiangle light scattering (MALS).³⁰ However, co-formulated mAb products present elevated analytical challenges, as the mAbs are usually similar in size (i.e., hydrodynamic radii), making it nearly impossible to distinguish various dimer species by SEC. Coupling SEC with other techniques such as surface plasmon resonance (SPR) can provide detailed information on specific antibodies in aggregates. For example, aggregates obtained from forced degradation studies of a two-mAb mixture were collected by SEC and analyzed by SPR, demonstrating that only one of the antibodies was present in the aggregates.¹⁶ However, utilizing SPR requires an assumption that the binding epitopes are accessible in the oligomer, which is not necessarily true depending on the geometry of the aggregate species. Another approach to determine oligomer identity is SEC coupled with mass spectrometry (SEC-MS), which can be performed under denaturing conditions³¹ or near native conditions (nSEC-MS).³² Native SEC-MS preserves both covalent and non-covalent oligomeric interactions. In addition, very low levels of HMW impurities (0.01%) can be discerned when using state-of-the-art chromatography systems and high-resolution mass spectrometers.³³ While nSEC-MS preserves both covalent and non-covalent interactions, the technique is not able to distinguish between the two interaction types unless clear mass differences (e.g., from the crosslinks) are detected. Reduced mass accuracy is also experienced during mass deconvolution of HMW species due to their low abundances and relatively poor MS responses. Consequently, unconventional HMW species, which may emerge from co-formulated mAb products, are difficult to decipher. Post-column denaturation (PCD)-assisted SEC-MS (SEC-PCD-MS) alleviates many of these shortcomings by separating the low-abundance HMW species natively and subsequently breaking them into its monomeric constituents prior to MS detection.²⁵ This technique permits unequivocal identification of low levels of HMW species without requiring an enrichment process, which is often time-consuming and prone to introduce artifactual HMW species.

In this study, we report the identification of a novel dimerization mechanism in a co-formulated mAb cocktail using a multifaceted MS-based approach. Using thermally stressed material, we first observed the rapid growth of a unique truncated dimer species in a co-formulated mAb cocktail through nSEC-MS analysis. Following this observation, SEC-PCD-MS was performed at the intact and subunit levels to identify the truncation site and the domains responsible for dimerization. These analyses uncovered an intriguing dimerization mechanism that was driven by CDR clipping of one mAb, followed by domain swapping between the two mAb molecules. Further, by probing the disulfide bond susceptibility changes via limited reduction and middle-down analysis, the structural changes of the involved domains were studied. Finally, structural analysis and modeling was used to support the proposed domain swapping dimerization mechanism. These findings underscore the importance of advanced analytical techniques in understanding and mitigating the challenges associated with co-formulated biotherapeutics.

Materials and methods

Materials

mAb-1 (hIgG4) and mAb-2 (hIgG4) were produced at Regeneron (Tarrytown, NY). Deionized water was provided by a Milli-Q integral water purification system installed with a MilliPak Express 20 filter (Millipore Sigma, Burlington, MA). Formic Acid (FA, 98–100% Suprapur for trace metal, 1.11670) was purchased from Milli-pore Sigma. Ammonium acetate (MS grade 73,594-100 G-F), and iodoacetamide (IAA; A955–4) were obtained from Sigma-Aldrich (St. Louis, MO). Acetonitrile (ACN; liquid chromatography-MS grade, A955–4), dithiothreitol (DTT, A39255), and Invitrogen UltraPure 1 M Tris-HCl buffer, pH 7.5 (15568–025) were purchased from Thermo Fisher Scientific (Waltham, MA). 2-propanol (IPA 34,965-4X4L) was purchased from Honeywell (Charlotte, NC). PNGase F (P0704L) was purchased from New England Biolabs (Ipswich, MA). FabRICATOR (A0-FR1-020) was purchased from Genovis (Kävlinge, Sweden).

Native SEC-MS and SEC-PCD-MS analysis at intact level

The co-formulated mAb cocktail samples consist of mAb-1 (75 mg/mL) and mAb-2 (75 mg/mL). The single mAb-1 was formulated at 147.9 mg/mL. Both samples were formulated using the same buffer consisting of 10 mM Histidine, 63 mM Arginine, 4.5% Sucrose, and 0.2% PS80 at pH 5.7. Samples were incubated at 40°C for either one month or three months to generate thermally stressed material. The samples were diluted using Tris-HCl, pH 7.5 to 5 mg/mL. Final Tris-HCl concentrations were enzyme specific and listed below. To reduce mass heterogeneity, samples were deglycosylated using PNGase F (1 U/5 µg) at 45°C for 1 hour in 100 mM Tris-HCl pH 7.5. Deglycosylated samples were injected onto either a ACQUITY UPLC Protein BEH (200Å, 1.7 µm, 4.6 mm ×150 mm) or a ACQUITY UPLC Protein BEH (200Å, 1.7 µm, 4.6 mm ×300 mm) column using an isocratic elution method of 150 mM ammonium acetate for 5 minutes (0.6 mL/minute) or 20 minutes (0.2 mL/minute), respectively. If applicable, PCD solvent (60% ACN, 4% FA) was applied via a post-column mixing T at an equal flow rate (1:1 mixing).²⁵ The flow was then split into a microflow ( < 10 μL/minute) for nanoelectrospray ionization (nESI)-MS detection. The remaining high flow was diverted for UV detection (280 nm). A Thermo Q-Exactive UHMR (Thermo Fisher Scientific, Bremen, Germany) equipped with a Microflow-Nanospray Electrospray Ionization (MnESI) Source and a Microfabricated Monolithic Multinozzle (M3) emitter (Newomics, Berkeley, CA) was used to achieve nESI-MS analysis. MS resolution settings of 6,250 or 12,500 (at m/z = 200) were used for either high-sensitivity or high-resolution data acquisition, respectively. A detailed experimental setup and instrument parameters can be found in previous publications.^25,33

Native SEC-MS and SEC-PCD-MS analysis at subunit level

To perform IdeS digestion and deglycosylation, the thermally stressed (40°C, three months), co-formulated mAb cocktail sample was co-treated with FabRICATOR (1 U/ug) and PNGase F (1 U/5 µg) at 37°C for 1 hour in 50 mM Tris-HCl pH 7.5 to generate the F(ab’)₂ and deglycosylated Fc fragments. For subunit analysis, an aliquot of the digests was partially reduced under mild conditions using 5 mM DTT at 37°C for 30 min in 40 mM Tris-HCl pH 7.5. To prevent disulfide bond re-formation, the newly released free thiols were alkylated with 20 mM IAA at 25°C for 30 minutes in the dark. Samples were then analyzed by both nSEC-MS and SEC-PCD-MS methods as described above.

Middle-down subunit analysis

The co-formulated, thermally stressed sample (40°C, three months) was prepared as described above for subunit analysis. The IdeS-digested, partially reduced, and alkylated sample was then subjected to a targeted SEC-PCD-MS/MS analysis using higher-energy collisional dissociation (HCD). A Thermo Fisher Orbitrap Eclipse Tribrid (Thermo Fisher Scientific, Bremen, Germany) with a HESI source was utilized for analysis. MS1 scans were acquired from 900 to 6,000 m/z at resolution setting of 15,000 (at m/z = 200) and AGC setting of 100%. MS2 scans were collected on single charge states using an isolation window of 2 m/z, MS resolution setting of 500,000, scan range of 200–6000 m/z, AGC set to 1900%, and NCE (normalized collision energy) set to 20.

Data analysis

Integration of UV peaks from the nSEC-MS analysis was performed using Thermo Fisher Xcalibur software (version 4.1). The acquired intact MS and subunit MS data were processed and deconvoluted with PMI Intact Mass (Protein Metrics, Cupertino, CA). Major deconvolution parameters were as follows: charge vector spacing 0.2; baseline radius 2; and charge range 5–200. Default settings were used for all other parameters.

Middle-down results were processed using ProSight Native (Proteinaceous, Evanston, IL; version 1.0.24144). Default settings were used for analysis. Fragments of interest were manually validated.

Structural analysis and modeling

VL/VH structures for mAb-1 and mAb-2 were obtained using homology modeling. mAb-1 VL/mAb-2 VH crossover structural model was built via superposing mAb-1 VL to mAb-2 VL. Total interface area, hydrophobic interface area and shape complementarity between VL and VH were calculated for mAb-1, mAb-2 and mAb-1/mAb-2.

‘Triangular’ heterocomplex model was built by superposing mAb-1 VL to mAb-2 VL. Manual adjustment of conformation was performed for VH-FW4 of mAb-1 and CH1 of mAb-2 followed by local energy minimization to allow for proper geometry of the affected residues. All structural analysis and modeling were performed in MOE.³⁴

Results

Detection of a truncated heterodimer in a two-mAb cocktail under thermal stress

Thorough characterization of HMW species is crucial during the development of co-formulated mAb cocktails, as it offers valuable insights into aggregation mechanisms and establishes a foundation for risk assessment. To reveal potential degradation pathways of mAb products within a short period of time, forced degradation studies performed at elevated temperatures are commonly employed. Utilizing SEC-UV, the propensity for HMW formation under thermal stress (40°C) was compared between a co-formulated mAb cocktail (mAb-1 + mAb-2) and its individually formulated counterpart (mAb-1). Importantly, since both mAb-1 and mAb-2 display melting temperatures (T_m: 71.2°C and 75.7°C, respectively) and thermal onset temperatures (T_onset: 59.8°C and 59.2°C, respectively), as determined by differential scanning calorimetry (data not shown), which are significantly higher than 40°C, the selected thermal stress condition is expected to be mild enough to prevent protein unfolding. Interestingly, in the co-formulated sample, the relative abundance of the dimer increased at a much faster rate over three months compared to the individually formulated mAb-1 sample (Figure 1A–B). While these results may suggest a preferential hetero-interaction between mAb-1 and mAb-2 over mAb-1 self-interaction, concrete evidence was lacking. To ascertain which interactions were driving dimer growth, nSEC-MS was performed in the high-sensitivity mode (MS resolution of 6,250). Examination of the deconvoluted mass spectra of the dimer region revealed the presence of an unconventional HMW species in both the co-formulated and the single-mAb samples, which exhibited masses (~280 kDa) slightly smaller than the expected full-length dimers (~290 kDa, including mAb-1 homodimer, mAb-2 homodimer, and mAb-1/mAb-2 heterodimer) (Figure 1C–D). In addition, the abundance of this unusual HMW species was much higher in the co-formulated sample than that in the mAb-1 only sample. As these presumably truncated dimer species were also observed in the mAb-1 only sample, they were likely associated with truncations in mAb-1 molecule. However, detailed characterization was still necessary to elucidate the exact composition of these HMW species. To this end, SEC-PCD-MS is a highly effective tool for in-depth characterization of mAb HMW species.²⁵ By applying PCD solvent after native SEC separation, this technique enabled online dissociation of HMW species into their constituent components before MS detection. Using this technique, the predominant components contributing to the HMW formation in the stressed, co-formulated sample were identified as truncated mAb-1 (mAb-1^T) and full-length mAb-2 (Figure 1E). In contrast, the primary components responsible for the HMW formation in the stressed, mAb-1 only sample were identified as mAb-1^T and full-length mAb-1 (Figure 1F).

Figure 1.

Detection of a truncated heterodimer species in a two-mAb cocktail under thermal stress. SEC-UV analysis of (A) co-formulated and (B) single-mAb (mAb-1) samples after thermal stress, highlighting the dimer (red, 15x zoom) and monomer regions (green). Percentages were calculated using UV peak areas of dimer and main peaks. Deconvoluted mass spectra derived from the dimer region of nSEC-MS analysis for (C) co-formulated and (D) single-mAb (mAb-1) samples. Deconvoluted mass spectra derived from the dimer region of SEC-PCD-MS analysis for (E) co-formulated and (F) single-mAb (mAb-1) samples. Deconvoluted mass spectra derived from the monomer region of SEC-PCD-MS analysis for (G) co-formulated and (H) single-mAb (mAb-1) samples. Percentages represent the area fraction of mAb-1^T relative to the area of full-length mAb-1. Red circles highlight the clipped species (mAb-1^T). Deconvoluted mass values represent the molecular weights of the analytes following treatment with PNGase F. Cyan: mAb-1 LC; blue: mAb-1 HC; red: mAb-2 HC; Pink: mAb-2 LC.

It was also observed that the mAb-1^T species (133 kDa) was significantly enriched in the dimer region compared to the monomer region in both the co-formulated and the single-mAb samples. Using the full-length mAb-1 as a comparator, the relative abundance of the mAb-1^T was enriched from 8% in the monomer region to ~ 1548% in the dimer region in the stressed, co-formulated sample, representing a ~ 195-fold enrichment (Figures 1E,G). In contrast, the relative abundance of the mAb-1^T increased from 10% in the monomer region to ~ 46% in the dimer region in the stressed, mAb-1 only sample (Figures 1F, 1H), representing a ~ 5-fold enrichment. Notably, comparing the abundances of mAb-1^T in both the monomer and dimer region relative to that of the intact mAb-1, the clipping rate was largely comparable between the co-formulated sample (15.9%) and the mAb-1 only samples (12.8%), suggesting the clipping event is independent to the concentration of mAb-1. These results indicated that the mAb-1^T exhibited a greater propensity for dimerization compared to the full-length mAb-1. Most intriguingly, this species also showed a clear preference for forming heterodimer with mAb-2 over self-interacting with mAb-1. This preference explained the faster dimer growth in the co-formulated sample compared to the individually formulated mAb-1 under identical thermal stress conditions.

The clipped VH domain remains associated prior to dimerization, forming native-like mAb structure

To identify the truncation site within the mAb-1 molecule, SEC-PCD-MS analysis of the stressed (40°C, 3 months), individually formulated mAb-1 sample was further performed in the high-resolution mode (MS resolution of 12,500). In the absence of PCD, analysis of the monomer region primarily identified the full-length mAb-1, with only a minimal presence of the mAb-1^T species detected (Figure 2A, left panel). In contrast, with PCD enabled, the analysis clearly revealed an increased level of the mAb-1^T species, as well as the emergence of its complementary fragment (11,326 Da) within the same spectrum (Figure 2A, right panel; Table S1). These observations suggested the presence of a non-covalent, native-like mAb clipped product, which either underwent partial dissociation during ionization in the gas phase (without PCD) or complete dissociation due to the denaturing solvent (with PCD). By matching with the cDNA-predicted sequence, the truncation site was localized to the N-terminal of Ser103 within the heavy chain (HC) CDR3. This cleavage site was also confirmed by peptide mapping analysis (data not shown). It was hypothesized that, after clipping, this HC variable region fragment (VH, ~11 kDa) likely continued to interact with the remaining mAb molecule (i.e., mAb-1^T, 133 kDa) through VH-VL non-covalent interaction, forming a native-like mAb structure. This hypothesis can be effectively demonstrated by comparing the elution profiles of each fragment using the extracted ion chromatograms (XICs) from the SEC-PCD-MS analysis. As shown in Figure 2B and Table S1, the XICs of the VH fragment (~11 kDa) and the mAb-1^T species (~133 kDa) exhibited nearly identical elution times and profiles, despite their distinct size differences. This indicated that the VH fragment and the mAb-1^T species were associated during SEC separation and were only dissociated under PCD-MS conditions. In addition, this analysis also revealed that the clipped, native-like mAb species eluted slightly earlier than the full-length mAb during SEC separation. This observation is consistent with several recent reports highlighting that CDR3-clipped mAbs often elute slightly before the SEC main peak.^35,36 The authors proposed that the early shift of the retention time of the clipped species was likely caused by weaker secondary interactions with SEC column matrix. Furthermore, under native MS detection conditions, the clipping event should theoretically result in an 18 Da mass increase due to the hydrolysis of the amide bond, which involves the addition of one water molecule. This shift, however, would be very difficult to observe due to the low abundance of the clipped species and its co-elution with the full-length mAb. Nevertheless, an examination of the deconvoluted mass spectra across the main peak (from nSEC-MS data) highlighted a slight mass increase in the leading edge of the peak (data not shown), indicating the presence of the clipped species. These results highlighted the effectiveness of the SEC-PCD-MS approach in identifying low levels of protein backbone clippings in mAb samples.

Figure 2.

CDR3-clipped species retain native-like mAb structure. (A) deconvoluted mass spectrum of mAb-1 (thermally stressed at 40°C for 3 months) obtained under native (left) or PCD (right) conditions. (B) UV and extracted ion chromatograms (XICs) of mAb-1^T, N-terminal fragment, and full-length mAb-1 from SEC-PCD-UV/MS analysis. The deconvoluted mass values correspond to the molecular weights of the analytes after PNGase F treatment. XICs were generated using the most abundant charge state of each species.

Finally, since only two fragments were identified, the clipping in mAb-1 appeared to be highly specific under thermal stress conditions. Indeed, protein backbone fragmentation occurring N-terminal to serine (Xaa-Ser) is not unusual and is well-documented in literature,³⁷ which is proposed to occur via a two-step mechanism. First, the neighboring N-terminal peptide bond undergoes nucleophilic attack from the serine alcohol, forming an oxazolidine intermediate. This intermediate then rearranges to an ester intermediate that is hydrolyzed by water to complete peptide-bond cleavage.³⁸ Due to the high specificity observed within mAb-1, the local microenvironment (e.g., solvent accessibility, pKa) likely plays a pivotal role in cleavage propensity.

Identification of domains involved in heterodimerization

While previous analyses clearly showed the preference for heterodimerization between the CDR3-clipped mAb-1 molecule (mAb-1^T) and the full-length mAb-2 molecule, the underlying biochemical root cause remained unidentified. To ascertain the domains involved in the heterodimerization, several strategies were systematically used to deconstruct the interaction. The stressed, co-formulated mAb sample was first subjected to IdeS digestion to localize the interaction to either the fragment crystallizable region (Fc) or the F(ab’)₂ arms (Figure 3), and then the IdeS-treated sample was analyzed using SEC-PCD-MS. In the absence of PCD, a truncated F(ab’)₂ heterodimer species (M.W., ~183 kDa) was observed as the major HMW species in the dimer region (Figure S1A and S1B, Table S1). Subsequent PCD-MS analysis confirmed that the truncated F(ab’)₂ heterodimer was composed of the truncated F(ab’)₂ from mAb-1 and the intact F(ab’)₂ from mAb-2, confirming a Fab-Fab interaction (Figure S1D). Similar to the observations at the intact mAb level, enrichment of the truncated mAb-1 F(ab’)₂ species was evident in the dimer region compared to the monomer region (Figure S1C and S1E). However, it remained unclear whether and how the truncated Fab arm was involved in the hetero-interaction.

Figure 3.

Schematic of sample treatment strategies to identify domains involved in heterodimerization.

To improve resolution, the IdeS-treated sample was subsequently subjected to partial reduction under native conditions to selectively reduce the inter-chain disulfide bonds and further localize the interaction to the constituent subunits (Figure 3). The newly formed free thiols (1x from light chain (LC) and 3x from Fd (i.e., VH and CH1 domains)) were subsequently capped by iodoacetamide (IAA)-based alkylation reaction to prevent disulfide bond re-formation. Importantly, within a properly folded mAb molecule, the intra-domain disulfide bonds are highly resistant to reduction under native conditions due to their poor solvent accessibility.^39,40 This feature could also be used to probe the potential structural changes of domains involved in dimerization. Following partial reduction and alkylation, nSEC-MS analysis, both with and without PCD, revealed a non-covalent heterocomplex (~63 kDa) that was composed of the truncated Fd (Fd^T) from mAb-1, LC from mAb-1, and Fd from mAb-2 (Figure 4, Table S1). Interestingly, in contrast to the expected Fab-Fab heterodimer, LC from mAb-2 was absent from the observed heterocomplex (Figures 3 and 4). To ensure this was not caused by gas phase dissociation of mAb-2 LC from the non-covalent complex, the elution profile of each component was reconstructed and compared based on the XICs from the SEC-PCD-MS analysis. This comparison clearly demonstrated that the Fd^T from mAb-1, LC from mAb-1, and Fd from mAb-2 all exhibited nearly identical elution profiles, which also closely matched the elution profile of the heterocomplex without PCD (albeit with minor retention time shift due to the PCD flow) (Figure 4C). In contrast, the XIC of mAb-2 LC displayed a markedly different elution profile, indicating its absence from the heterocomplex in the solution phase. This observation likely suggested the hetero-interaction disrupted both the VH-VL and the CH1-CL non-covalent interactions within the interacting Fab arm of mAb-2. Hence, after reducing the inter HC and LC disulfide bond, the LC of mAb-2 was no longer retained in the heterocomplex.

Figure 4.

Localization of the domains involved in heterodimerization. (A) UV and XICs of the heterocomplex and the expected Fab complexes from nSEC-UV/MS analysis. (B) deconvoluted mass spectra of the heterocomplex (top) and the expected Fab complexes (bottom) from nSEC-MS analysis. (C) XICs of subunits originating from the heterocomplex and Fab-2 complex from SEC-PCD-MS analysis. (D) deconvoluted mass spectra of the dissociated subunits from the heterocomplex (top) and the expected Fab complexes (bottom) from SEC-PCD-MS analysis. XICs were generated using the most abundant charge state of each species. Deconvoluted mass values represent the molecular weights of the analytes following treatment with PNGase F, IdeS and alkylation with iodoacetamide. Gray circle: expected alkylation event; yellow circle: unexpected alkylation event.

Upon close examination of the deconvoluted mass of the processed heterocomplex (composed of mAb-1 Fd^T, mAb-1 LC, and mAb-2 Fd), a mass deviation of ~ +114 Da was observed (expected: 62530.5 Da, observed: 62652.0 Da) (Figures 3, 4B, Table S1). PCD-MS analysis successfully localized the mass deviation to the Fd of mAb-2 (expected mass with 3 × IAA addition: 24804.5 Da, observed mass: 24920.1 Da) (Figures 3, 4D, Table S1). Based on the XIC analysis, this mass addition was not observed on the mAb-2 Fd originating from the Fab fragment, which exhibited the expected number of IAA modifications (Figure 4C). Given that Cys alkylation by IAA introduces a mass increase of approximately +57 Da per site, it was hypothesized that an intra-domain disulfide bond within mAb-2 Fd from the processed heterocomplex was unexpectedly reduced and subsequently alkylated. To confirm this hypothesis, middle-down MS analysis using higher-energy collisional dissociation (HCD) was conducted on the mAb-2 Fd fragment (i.e., with 5 ×IAA addition) originating from the processed heterocomplex. Rich HCD fragmentation was observed from an originally disulfide-bonded region within the CH1 domain, suggesting the reduction of this disulfide bond in this species (Figure 5A and Figure S2). Further analysis of the b- and y-ions also localized the additional two IAA modifications to Cys147 and Cys203 residues within the mAb-2 CH1 domain (Figure 5B). In contrast, fragmentation of the mAb-2 Fd fragment (i.e., with 3 × IAA addition) originating from the Fab region resulted in no fragmentation within the CH1 disulfide-bonded region (Figure S3A-C). This observation aligns well with the fact that the HCD fragmentation is highly inefficient at generating b and y ions within the disulfide-bonded regions of IgG subunits, as it requires the concurrent cleavage of both the peptide amide bond and the disulfide bond.^40–42 These data confirmed the unexpected reduction (and alkylation) of mAb-2 CH1 domain disulfide bond in the processed heterocomplex under the limited reduction conditions. The significantly increased susceptibility of this originally well-buried disulfide bond can be attributed to its increased solvent accessibility, resulting from the absence of mAb-2 LC in the processed heterocomplex, which normally protects the CH1 through its cognate CH1-CL pairing (Figure 4C, pink). Because the mAb-2 VH domain disulfide bond did not exhibit a similar susceptibility increase, it is plausible that it remained shielded from reduction by interacting with the mAb-1 VL domain, suggesting a VH-VL domain swapping within the heterocomplex (Figure 3). Together, these findings indicated that the truncated heterodimer, formed specifically between the CDR3-clipped mAb-1 and the full-length mAb-2, is largely maintained through the mAb-1 VL and mAb-2 VH interaction, which is driven by a specific mAb-1 CDR clipping event.

Figure 5.

Study the structural impacts of domains involved in heterodimerization using disulfide bond susceptibility changes probed by middle-down analysis. (A) HCD MS2 fragmentation patterns for mAb-2 Fd (5x iodoacetamide) derived from the heterocomplex. Observed y-ion fragments are shown in blue. (B) zoom-in view of the signature fragment ions that localized the additional two IAA modifications. Expected isotope envelopes are red.

Structural analysis and modeling support a stable complex formed via mAb-1 LC spontaneous domain swap with mAb-2 HC

Structural analysis and modeling were conducted to uncover the mechanisms underlying VH-VL domain swapping between CDR3-clipped mAb-1 and full-length mAb-2. This analysis revealed two potential complementary mechanisms driving the observed behavior. The first mechanism involved biophysical parameters such as interface area and complementarity, which explain the preferential pairing of mAb-1 VL with mAb-2 VH. Using a structural model based on mAb-1 VL/mAb-2 VH, the VL/VH interface area, hydrophobic interface area, and shape complementarity were calculated to compare the relative strength between different VL/VH interactions (Table 1). Analysis of properly paired chains, specifically mAb-1 Fd with mAb-1 LC and mAb-2 Fd with mAb-2 LC, revealed similar shape complementarity (i.e., the degree to which the interface contour boundaries of VL and VH align with each other) among VL/VH interfaces (0.38–0.43) (Table 1). However, mAb-2 exhibited a greater hydrophobic surface area of 670 Ų compared to 530 Ų for mAb-1. Minor differences in total surface area also favored mAb-2 interactions (1795.1 Ų) over mAb-1 (1758.4 Ų). These findings suggested stronger interactions between VL and VH in mAb-2. Interestingly, the swapped mAb-1 VL/mAb-2 VH interface had both larger buried total surface (1814.3 Ų) and hydrophobic surface (650 Ų) areas than properly paired mAb-1 chains, indicating a stronger interaction for mAb-2 VH than mAb-1 VH toward mAb-1 VL binding (Table 1). This demonstrated the preferential binding of mAb-1 VL to mAb-2 VH over its own cleaved cognate mAb-1 VH in the presence of mAb-2. However, these values characterizing the VL-VH interaction alone did not fully explain the preferential interaction between mAb-1 VL and mAb-2 VH compared to the cognate pairing of VH and VL in mAb2. Therefore, additional mechanisms are likely involved in driving the preferential heterocomplexing behavior observed between truncated mAb-1 and mAb-2.

Table 1.

Structural analysis on interface features among native mAb-1, mAb-2 and swapped mAb-1/mAb-2 VL/VH pairs.

Interface Features	mAb-1 VL	mAb-2 VL	mAb-1 VL
	mAb-1 VH	mAb-2 VH	mAb-2 VH
Total VL/VH interface area (Å2)	1758.4	1795.1	1814.3
Hydrophobic interface area (Å2)	530	670	650
Shape Complementarity	0.38	0.43	0.42

In this context, the second mechanism we explored highlighted potential structural characteristics of the heterodimer that may influence the preference of mAb-1 VL for mAb-2 VH. Analysis of the available structures of mAb-1 and mAb-2 revealed that the VH-truncated mAb-1 Fd^T contained a long framework 4 (FW4) chain preceding the CH1 domain (Figure 6). It is known that the VH-CH1 interaction can adopt various conformations and elbow angles.⁴³ Consequently, both the heavy chain framework 4 (HC-FW4) of mAb-1 and the CH1 domain of mAb-2 can exhibit rotational flexibility around the linker region connecting the VH and CH1 domains (Figure 6, left). In addition to conformational flexibility, increased temperatures during thermal stress conditions were also reported to reduce the native contact fractions between LCs and HCs, thereby inducing structural instability.⁴⁴ Together, these structural dynamics can expose new hydrophobic regions on both the FW4 of mAb-1 and the CH1 of mAb-2, facilitating their association and leading to the stabilization of a non-covalently linked heterocomplex (mAb-1 Fd^T, mAb-1 LC, and mAb-2 Fd). Notably, these interactions were likely less pronounced between mAb-1^T and mAb-1, as the preferential formation of truncated dimer in mAb-1 single formulation sample was less evident. Further, the proposed structural model also illustrated that FW4 has incomplete coverage of the whole CL-CH1 interface in mAb-2 CH1 domain (Figure 6), corroborating the increased reduction susceptibility of intra-CH1 disulfide bond because of partial exposure of this domain in the processed heterocomplex (Figure 5 and 6). Lastly, from the perspective of pairing frequency among human germline families, both VL of mAb-1 and mAb-2 belong to hVK3(D). This results in an equal prevalence of mAb-2 VH/mAb-2 VL and mAb-2 VH/mAb-1 VL pairings, with a frequency of 16% according to OAS database.⁴⁵ These data suggested that the observed domain-swapping mechanism is not driven solely by repertoire-level pairing tendencies or natural frequencies, but is also influenced by specific biophysical interactions and structural features.

Figure 6.

‘Triangular’ heterocomplex model formed by mAb-1 LC, mAb-1 Fd^T (FW4 and CH1) and mAb-2 Fd (VH-CH1). Left: cleaved mAb-1 Fab. The dashed box highlights the VH FW4 region following the cleavage. Right: mAb-2 Fd. The dashed box highlights CH1 from mAb-2. Potential rotations of highlighted regions needed to form the complex are indicated by the curved arrows. Middle: structural model of possible heterocomplex formation between cleaved mAb-1 Fab and mAb-2 Fd. Hydrophobic patches are depicted with matching colors to their corresponding Fab chains. Cyan: mAb-1 LC; blue: mAb-1 Fd^T; red: mAb-2 Fd.

Discussion

Heterodimers represent a novel impurity and may serve as critical quality attributes in co-formulated therapeutic antibody products. In this study, we present the discovery of a unique truncated heterodimer in a co-formulated mAb cocktail drug product, formed during forced degradation studies conducted at 40°C. A multifaceted MS approach was used to identify heterodimer formation and elucidate the underlying dimerization mechanisms. Our findings suggest that the heterodimerization was initiated by CDR3 clipping on mAb-1, followed by domain swapping between truncated mAb-1 and intact mAb-2. Importantly, the formation of clipped species and heterodimer was negligible under normal storage conditions (5°C 3 months, data not shown), highlighting its temperature-dependent nature. This observation aligned well with previous analyses using synthetic peptides containing the same Xaa-Ser motif, which demonstrated a cleavage rate at 37°C that was ten times slower compared to 60°C.⁴⁶ Stability studies at elevated temperatures, as recommended by regulatory guidelines from agencies such as the European Medicines Agency, US Food and Drug Administration, and World Health Organization, offer critical insights into the behavior of therapeutic antibodies under stress conditions.^47–49 While the degradation observed under these conditions may not replicate normal storage scenarios, such studies are indispensable in revealing the potential product degradation pathways. They provide critical data to inform the development, manufacturing, and storage strategies for therapeutic antibodies. These findings emphasize the importance of temperature regulation in mitigating impurity formation throughout the product lifecycle of therapeutic antibodies, ensuring their safety and efficacy.

The safety implications of this truncated heterodimer, particularly concerning immunogenicity and potency, are not yet fully understood. Hence, the significance of this quality attribute and the establishment of appropriate control strategies (e.g., QC specifications) are uncertain and require thorough evaluation. Further, while the Xaa-Ser clipping observed in this study was present at negligible levels under normal storage conditions, other CDR3 clippings, have been reported at significantly higher levels. When considering co-formulation of these mAb molecules with other mAbs, heterodimerization through similar mechanisms becomes a possibility and necessitates thorough investigation.^35,50 Additionally, beyond the clipping and heterodimer formation observed in the drug product, it is necessary to account for the prolonged duration therapeutic antibodies remain in the body post-administration- typically 18–21 days for most IgG antibodies- at physiological temperatures of approximately 37°C.⁵¹ During this time, the dynamic physiological environment may promote clipping and aggregation, either with itself or with endogenous antibodies, suggesting that the risk of dimerization cannot be entirely ruled out. This underscores the importance of designing antibodies and formulations that not only minimize impurity formation during manufacturing and storage, but also preserve structural stability throughout their in vivo lifecycle. One potential strategy to mitigate clipping risks involves amino acid substitution. For example, Lyons et al. demonstrated that replacing the amino acid residue N-terminal to serine residues with Lys and Pro effectively reduced the rate of Xaa-Ser cleavage in the presence of zinc.⁵² However, altering the CDR3 sequence poses significant challenges, as it may compromise the structural integrity and functional efficacy of the antigen-binding site. This highlights the need for a balanced approach that minimizes risks without affecting therapeutic performance. Optimizing formulation conditions, such as incorporating stabilizing excipients to reduce conformational flexibility or adjusting pH, could suppress clipping and heterodimer formation. For example, studies have shown that fragmentation at Xaa-Ser motif exhibits pH dependency and can be effectively controlled by a slightly acidic formulation pH range of 3–6.5.³⁸ However, significant deviations from the optimal pH range of 5–7, known to optimize antibody stability and minimize aggregation, could destabilize the antibodies, reducing their efficacy and shelf life. These considerations underscore the importance of carefully calibrating formulation adjustments to balance colloidal stability with long-term storage requirements.

While the effects of CDR3-clipping on mAbs have been extensively studied regarding potency, hydrophobicity, thermal stability and other properties, our work is the first to demonstrate the formation of CDR3-clipped heterodimers in FDCs.^35,36 The current analysis could be expanded or refined through alternative techniques such as hydrogen-deuterium exchange (HDX), which offers unique insights into protein dynamics and conformational changes. Although HDX has been used to study the structural and dynamic properties of native-like CDR3-clipped antibodies,³⁶ its application to investigating CDR3 clip-induced heterodimers remains unexplored, presenting a promising direction for future research. An orthogonal approach, such as MS-based cross-linking, could also be used to map spatial relationships and interaction sites within dimers, providing structural insights that are otherwise difficult to obtain.⁵³ Additionally, integrating rational design strategies with predictive models for heterodimer formation could streamline development processes, reducing costs and enhancing patient safety by identifying problematic formulations early. By elucidating the molecular interactions and structural stability of co-formulated mAbs, this research provides critical insights for optimizing FDC design and improving therapeutic efficacy and safety. Furthermore, the multifaceted MS approach presented here offers a broadly applicable framework for extended characterization and analysis of high molecular weight species in mAb FDCs, supporting their successful development in clinical applications.

Funding Statement

This work was supported by the Regeneron Pharmaceuticals.

Abbreviations

mAb

monoclonal antibody

mAb-1^T

truncated mAb-1

Fd^T

truncated Fd

mAb-1 Fd^T

truncated mAb-1 Fd

HMW

high molecular weight

CDR

Complementarity-Determining Region

PCD

post-column denaturation

SEC

size exclusion chromatography

MS

mass spectrometry

PTMs

post-translational modifications

SV-AUC

Sedimentation Velocity Analytical Ultracentrifugation

UV

ultraviolet

MALS

multi-angle light scattering

SPR

surface plasmon resonance

FDC

fixed dose combination

HPLC

high-performance liquid chromatography

Da

Daltons

HC

heavy chain

VH

variable heavy

XIC

extracted ion chromatogram

Fc

Fragment Crystallizable

MW

molecular weight

CH1

first constant heavy chain domain

LC

light chain

IAA

Iodoacetamide

HCD

Higher-Energy Collisional Dissociation

FW4

Framework 4

FA

Formic Acid

ACN

Acetonitrile

TFA

Trifluoroacetic Acid

TCEP

Tris(2-carboxyethyl)phosphine

IPA

Isopropyl Alcohol

U

Units

mL

milli Liters

nESI

nano-Electrospray Ionization

UHMR

Ultra-High Mass Range

w/w

Weight/Weight

Disclosure statement

The authors declare the following competing financial interest(s): A.H., Y.S., T.N.T, Y.Y., S.W., and N.L. are full-time employees and shareholders of Regeneron Pharmaceuticals Inc.

Supplementary Information

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2025.2546074