Mass spectrometry-based multi-attribute method in protein therapeutics product quality monitoring and quality control

ABSTRACT

The multi-attribute method (MAM), a liquid chromatography-mass spectrometry (LC-MS)-based peptide mapping method, has gained increased interest and applications in the biopharmaceutical industry. MAM can, in one method, provide targeted quantitation of multiple site-specific product quality attributes, as well as new peak detection. In this review, we focus on the scientific and regulatory considerations of using MAM in product quality attribute monitoring and quality control (QC) of therapeutic proteins. We highlight MAM implementation challenges and solutions with several case studies, and provide our perspective on the opportunities to use MS in QC for applications other than standard peptide mapping-based MAM.

KEYWORDS:

• product quality attribute
• quality control
• LC-MS
• multi-attribute method
• Intact MAM
• targeted attribute quantitation
• MS in QC
• cGMP
• GMP
• new peak detection

Introduction

The use of the multi-attribute method (MAM), ¹ a liquid chromatography-mass spectrometry (LC-MS)-based peptide mapping method, has gained increased interest and application in the biopharmaceutical industry.^1–5 Rathore et al. reviewed the role of MS in the characterization of biologic protein products,⁶ and summarized the use of MS in six major areas of biotechnology product analysis: product identity testing, characterization, comparability assessment, adulteration analysis, demonstration of critical quality attribute (CQA) consistency during product development, and in pharmacokinetic and pharmacodynamics studies. While LC-MS-based peptide mapping has been used for many years for characterization of biotherapeutic proteins in biologics license application (BLA) submissions,⁶ companies are now pursuing use of the peptide mapping method for quality control (QC) testing,^1,7

Although MAM incorporates all the capabilities of LC-MS peptide mapping, it also offers improved capabilities for simultaneous detection, identification, quantitation, and QC. As a control system method, MAM offers particular benefits when conventional methods are not suitable for specific CQAs, especially with new modalities and molecule formats. In addition, using MAM as a replacement for multiple relevant conventional technologies (instead of parallel testing, when there is sufficient confidence in MAM) can accelerate product development.

MAM can provide both targeted quantitation of multiple covalent modifications in one method, and new peak detection (NPD) through differential data analysis of LC-MS chromatograms in good manufacturing practice (GMP) batches compared to a reference standard.^2–8−12 Standard, peptide-based MAM can be a platform method with several easy-to-execute options, such as trypsin, Lys-C digest, or non-reduced sample¹³ alternatives. Other protease of choice depends on the type of biotherapeutic proteins to be analyzed and, in certain cases, multi-enzyme digestion is required to achieve full sequence coverage and a reliable post-translational modification (PTM) assessment. For example, trypsin was used in combination with Glu-C¹⁴ or AspN¹⁵ for certain fusion proteins. It is also known that trypsin is not effective in digesting adeno-associated virus capsid proteins.¹⁶

A generic MAM workflow is illustrated in Figure 1. In practice, product quality attributes (PQAs) of a product can be assessed through a comprehensive database search of LC-MS/MS data of the enzyme (typically trypsin) digest of the product. When a panel of PQAs is determined to be CQAs, targeted attribute quantitation (TAQ) will be performed for modified and unmodified peptides that contain the selected CQAs. MAM’s NPD function provides a sensitive approach to detect product degradants or variants that might not be detected from TAQ. It can be performed to identify unexpected product quality changes through a comparative analysis of LC-MS chromatograms from a test sample and a reference standard, as MS data is needed to avoid overlooking variations masked by other peaks when only using ultraviolet (UV) chromatograms. Many CQAs can be easily confounded and masked by co-eluted species and will not be detected by UV detector due to the low sensitivity and specificity. While NPD can be implemented by peptide mapping LC-UV, sensitivity is still a challenge, and LC-UV does not have the same capability as MAM to enable rapid assessment of new peaks to, for example, exclude sample preparation artifacts. At the MAM implementation stage in QC, LC-MS is usually preferred. LC-MS/MS is optional at an early stage, which might provide benefits in streamlining the identification of new peaks with comparable TAQ results since a comprehensive peptide library is not available yet at this stage. A new peak is defined as a new, missing, or changed peak in the test sample after passing the minimum NPD threshold and other critical parameters and data processing filters (e.g., within the defined tolerance for m/z and retention, and defined charge states).¹² This NPD capability is a key part of the rationale that supports the replacement of conventional methods with MAM.¹¹

Figure 1. Generic MAM workflow for biotherapeutic modalities that enables targeted attribute quantitation and new peak detection.

A schematic graph illustrating the basic MAM workflow for biotherapeutic modalities using the LC-MS-based peptide mapping method.

Within the Food and Drug Administration’s Center for Drug Evaluation and Research, MAM is listed as one of the emerging technologies under the FDA’s Emerging Technology Program, where sponsors meet with Emerging Technology Team members to discuss, identify, and resolve potential technical and regulatory challenges regarding the development and implementation of novel technologies, such as MAM, prior to filing a regulatory submission. To facilitate MAM knowledge sharing and implementation, several MAM-focused workshops were organized by the United States Pharmacopoeia, American Society for Mass Spectrometry, and CASSS MS, and an industry-wide MAM Consortium was formed.

MAM can directly measure multiple site-specific product quality attributes and has the potential to replace several conventional QC methods, including hydrophilic interaction liquid chromatography (HILIC) for glycan analysis, ion-exchange chromatography (IEC) for charge variant analysis, reduced capillary electrophoresis-sodium dodecyl sulfate (R-CE-SDS) for clipped variant/fragments analysis, reversedphase (RP), or hydrophobic interaction liquid chromatography for Fc oxidation analysis, and identity test by peptide mapping LC-UV.^1,3,11,17,18 Conventional methods for monitoring and controlling PQAs at the intact level, and whether these PQAs can be monitored or controlled by the standard MAM, are listed in Table 1.

Table 1. General product quality attributes for monitoring or control by conventional methods and by the standard MAM.

	Intact Level Conventional Method	Attributes Contributing to the Product Variant	Peptide-Based Standard MAM (Reduced, Trypsin*)
Product Variant	To monitor or control product variant		For monitoring or control of the PQA?
Charge	ICIEF/IEC	Deamidation	Yes
		Oxidation	Yes
		Hydroxylation	Yes
		Fab glycan	Yes
		Lys succinylation	Yes
		Pyroglutamate	Yes
		Glycation	Yes (might underestimate^Citation19)
		AGEs	Yes
		O-glycan	Yes
		C-terminal lysine	Yes
		Isomerization	Yes
		Succinmide	Yes
		VHS extension	Yes
		High Order Structure	No
		LMWF	Potential (depending on the fragment sequence, e.g., fragments that include tryptic cleavage sites may not be captured due to masking by the digested peptides,^Citation11 and may not be as easy to quantitate ladder cleavages such as the common hinge-region fragmentation^Citation20)
Size	R-CE-SDS	NGHC	Yes
	NR-CE-SDS	LMWF	Potential, see above
	SEC	HMWF	No
Identity	Peptide mapping by UV or QDA	Identity	Yes
Oxidation	HPLC	Oxidation	Yes
Glycosylation Related	CE or HILIC-Glycan	Glycosylation related	Yes
Cysteine Related	HPLC, NR-CE-SDS	Cysteine related	No (NR peptide map might be needed)
Sequence Variant	NGS or LC-MS peptide mapping	Sequence variant	Yes
Host Cell Protein	ELISA or LC-MS peptide mapping	Host Cell Protein	Potential (challenging for low-level host cell proteins)
BsAb-Related Impurities, Homodimer, etc.		BsAb- related impurities, homodimer, etc.	No
New Peak		Purity by MAM	Yes (no new peaks)

*Trypsin is the standard enzyme for peptide-based MAM. Lys-C or another enzyme digest might be needed for some modifications and modalities based on the sequence coverage.

Not only does MAM have the potential to replace multiple conventional methods but it also provides a deeper understanding of products, by means of site-specific identification and quantitation of CQAs. As a peptide-based method with high specificity, sensitivity, and resolution for most products, MAM has great potential to be a platform method. In addition, MAM can provide analytical solutions where conventional methods might not be suitable or adequate, especially for PQAs of new modalities and new molecule formats (e.g., bispecific antibodies, antibody-drug conjugates (ADCs) and other protein conjugates, fusion proteins). NPD, the added function to the peptide mapping method, also enables quick and early detection of unexpected changes that could occur during process development, manufacturing, and long-term storage. Compared to conventional methods, MAM is also ideal for stability testing, as most instability, other than aggregates, comes from covalent modifications, such as deamidation and oxidation, that are readily detectable by MAM.

While the implementation of MAM for QC testing of therapeutic proteins is of broad interest to the biopharmaceutical industry, many practical challenges and gaps exist. MAM’s TAQ component is not new, as many sponsors have years of experience supporting characterization and process validation studies with non-GMP LC-MS-based peptide mapping methods. However, the method performance of MAM’s TAQ component needs to be demonstrated for suitability in the QC environment.¹¹ And MAM’s NPD function is still new for many sponsors and requires more evaluation and experience in both development and QC. As discussed in the recent position paper by the European Federation of Pharmaceutical Industries and Associations (EFPIA) on the regulatory considerations of using MAM as a QC release and stability tool for biopharmaceuticals,¹⁷ the need for bridging studies for MAM and conventional methods depends on the stage when MAM is introduced as a QC method: prior to first-in-human (FIH) studies, during development post-FIH clinical studies, or as a life-cycle management activity in the commercial phase.¹⁷

In this review, we focus on the scientific and regulatory considerations of using MAM in PQA monitoring and QC of therapeutic proteins.^1,11,12,17 The challenges and solutions for MAM implementation are illustrated by several case studies, and the opportunities to use MS in QC for applications other than the standard peptide mapping-based MAM are discussed.

Part I. Technical and regulatory considerations of implementing MAM as a QC test

In 2019, FDA authors provided their perspective on the four major aspects of MAM implementation in a current good manufacturing practice (cGMP) environment for both new and existing products, risk assessment, method validation, capabilities and specificities of the NPD feature, and comparisons to conventional methods, and suggested approaches to help address potential issues.¹¹ Ren et al.⁷ reviewed advancing MS technology in cGMP environments and summarized technical considerations for MAM implementation in cGMP environments. MAM is expected to replace the conventional QC COA reported results, either directly (e.g., glycans) or indirectly (charge variants). In considering MAM implementation, it is critical to have a thorough understanding of MAM and the conventional methods it replaces, compare MAM and these replaced methods by means of a good study design, and perform risk assessments based on gaps between MAM and the conventional methods. Other technical considerations include method validation, sample preparation, instrumentation, and data analysis.⁷ The regulatory considerations of using MAM as a QC release and stability tool for biopharmaceuticals are discussed in a position paper published by multiple companies from the EFPIA.¹⁷

Overall, to justify its use as a QC test for biotherapeutic proteins, MAM should show performance and capabilities that are comparable or superior to those of the conventional methods that MAM is intended to replace. For a method replacement scenario, method bridging or comparison is required. In the EFPIA-preferred scenario, if MAM is introduced as a QC method prior to FIH, method bridging would not be required.¹⁷ However, until MAM is widely accepted by health authorities, a demonstration of its suitability compared to conventional methods will likely be expected. When implementing MAM at different stages of product development, risk assessment always needs to be performed. Depending on when it is introduced, prior to or post-FIH, different strategies might be used toward implementing MAM as a QC method for the control system.¹⁷ More details of the key perspectives and other recent MAM literature are discussed below.

Differences between MAM and conventional methods, and risk assessment of MAM implementation

According to ICH Q6B,²¹ appropriate purity methods need to be identified. To replace several conventional methods in QC, MAM must have capability at least as good as or better than the conventional methods to measure PQAs with suitable accuracy, precision, specificity, linearity, and sensitivity, and be able to detect relevant variations in the levels of PQAs. The authors of the EFPIA paper¹⁷ thought that continuing to determine charge variant purity by conventional methods (i.e., IEC or imaged capillary isoelectric focusing [iCIEF]) should not be necessary when MAM is in place, and implementation of MAM would require performing the NPD workflow, including appropriate specification setting to control unknown impurities.

To implement and demonstrate the suitability of MAM as a QC method that can replace conventional methods, comparative testing between MAM and the to-be-replaced method(s) needs to be performed during method and product development.¹¹ Any differences in product quality observed between methods should be understood to ensure that MAM is adequate, and to enable the evaluation of potential clinical impacts resulting from product quality differences. Ideally, to enable product quality comparisons to prior clinical experience established with conventional methods, clinical batches from earlier stages of development are assessed in the bridging exercise. Rogstad et al. stated that once a thorough understanding of MAM and the relevant PQAs are attained for a particular product, batch-to-batch comparisons between MAM and other conventional methods should be phased out.¹¹

In the mid-1990s, MS was used to study carbohydrate structures of antibodies.^22,23 Roberts et al.²³ reported an integrated strategy for structural characterization of the protein and carbohydrate components of monoclonal antibodies (mAbs), which directly evaluated the quantitative and qualitative consistency of the MS methods with conventional methods for carbohydrate analysis. Comparison between the HILIC glycan method and MAM is the most direct and well-correlated approaches.¹ While HILIC provides quantitation of released glycans, MAM measures the same species with attached peptide, providing more detailed site-specific information for biotherapeutic proteins with multiple glycosylation sites, and for sites refractory to enzymatic release, which can happen with Fab glycosylation (e.g., cetuximab²⁴ or when O-linked glycans are present. Method comparison becomes challenging when results from MAM and conventional methods may not be directly correlated due to differences in method resolution and specificity, for example, when comparing acidic variants (overall modification at intact level) measured by IEC to the site-specific charge variants measured by MAM. A tryptic peptide mapping-based MAM does not report on charge isoform distribution or could underestimate glycated species, the commonly observed acidic species for a mAb.¹⁹ However, according to authors from multiple companies¹⁷ and Rogstad et al.,¹¹ replacing a charge-based method, including IEC, is deemed acceptable due to MAM’s advantage providing site-specific charge-variant quantitation and better resolution of individual CQAs, compared to the overall charge distribution and co-eluted CQAs in a single IEC peak.

Rogstad et al. discussed the risk assessment when implementing MAM for QC of therapeutic proteins, and suggested approaches to address related issues.¹¹ MAM’s capability and method performance should first be evaluated for use as a control system method for specific CQAs. Risk assessment should consider the mechanism of action of the specific product and address the impact of relevant differences between the bottom-up MAM and conventional methods on product quality, safety, and efficacy. Rogstad et al.¹¹ also highlighted that risk assessment of MAM is likely to be phase appropriate.

As a peptide mapping method, MAM can directly measure the site-specific CQAs in the complementarity-determining region (CDR) or Fc regions (e.g., glycosylation species) that could affect target binding, effector function, or pharmacokinetics. Some of these site-specific modifications could be overlooked or not resolved by conventional methods, and no specific functional assay may be suitable to detect them. In these cases, MAM provides a clear benefit for the control strategy. On the other hand, when using MAM, some information that can be obtained from conventional methods might get lost, such as clipped species that resulted from clipping at arginine or lysine residues (i.e., tryptic cleavage sites).¹¹ If the clipped species are CQAs and need to be controlled by MAM, a risk assessment needs to be performed. Rogstad et al.¹¹ also suggested performing studies to support the use of MAM in these conditions. For example, studies on product degradation pathways and characterization of clipped species under various stress conditions could assess the risk and support the use of MAM to control fragments. If the data demonstrate that the attributes measured by MAM can indirectly but quantitatively control the product charge distribution, it could support replacing IEC with MAM. As a bottom-up method, potential loss of variant distribution information at the intact molecule level might happen and the impact on product quality, safety, and performance should be considered during MAM-focused risk assessment.

To support the future development of MAM as a platform analytical technology, Rogstad et al.¹¹ highly recommended that sponsors include both MAM and orthogonal conventional methods as part of release and stability testing when implementing MAM in an early-stage product with preliminary CQA evaluation status. This would provide both product and method knowledge, and generate sufficient data to support using MAM in the control strategy, at least for the first few products – preferably products with different structural characteristics. Once a biopharmaceutical company has gained sufficient MAM experience, Rogstad et al. stated that prior knowledge from a set of well-characterized precedent molecules could be used to mitigate some risks associated with early-stage candidates.¹¹

Unlike with most of the conventional methods, retrospective analysis can be performed for all MAM data, to obtain historical data for the specific quality attributes identified (e.g., novel impurities or product variants) at any stage of development. MAM enables site-specific CQA data throughout development, which ultimately facilitates key decisions later in development. While peptide mapping has long been a key component of characterization studies throughout development, peptide mapping data is not typically acquired for every clinical batch produced (especially for products with extensive clinical manufacturing), so having MAM on the control system ensures that a more comprehensive data set will be available. As a product progresses through later stages, accumulated MAM data for multiple product batches and stability conditions can be used to help identify and justify CQAs (and proposed acceptance criteria) to be controlled during batch release and stability testing, and establish the control strategies for licensure.

MAM development considerations

To be used as a QC method to quantify PQAs, MAM needs to be robust and have minimal method-introduced artifacts.⁷ As MAM is performed at the peptide level, biotherapeutic proteins must first be digested into peptides, which is usually accomplished using an enzymatic digestion, such as trypsin. Alkaline pH conditions and elevated temperatures used in the sample preparation can enhance trypsin activity, but often induce artificial protein modifications, such as asparagine deamidation and N-terminal glutamine cyclization. Protein oxidation artifacts can also be introduced when formic acid is used to terminate trypsin digest.²⁵ Therefore, MAM should consider and evaluate these parameters to minimize artifacts introduced during sample preparation.

Several options to optimize trypsin digestion conditions to minimize modifications on proteins are available. Ren et al.²⁶ first developed an improved method in 2009 that minimizes trypsin digestion time while maximizing trypsin activity through complete removal of guanidine. Shorter digestion times also reduced trypsin self-digestion and nonspecific cleavages. pH, the temperature used during reduction and alkylation, and digestion time are known to have an effect on deamidation and oxidation.^27–29 Diepold et al.²⁷ were able to significantly reduce the asparagine deamidation and aspartate isomerization induced by the method by performing the tryptic peptide mapping at pH 6.0. Furthermore, the stabilization of the succinimide intermediate product could also be achieved under these conditions. In 2020, Bauer et al.²⁹ studied the tryptic digestion with subsequent LC-MS peptide mapping. Various parameters were evaluated with the aim of reducing the risk of triggering artifacts³⁰ during sample processing and storage. Thus, in addition to pH 6.0, the influence of both the digestion time (comparison between 2 + 2 h and 20 h) and the addition of methionine in all buffers was investigated. As a result, artificial deamidation, in particular, oxidation, could be largely eliminated. Furthermore, the digestion time had no influence on triggering method artifacts under these conditions.²⁹ A qualification of this method showed that it is generic for various types of proteins, robust and precise, and can be used universally for all types of PTMs (e.g., oxidation, deamidation, and glycation) and sequence variants. Finally, Carvalho et al.²⁵ confirmed that using trifluoracetic acid (TFA) instead of formic acid to stop trypsin digestion and a double addition of methionine at a 20 mM concentration at denaturation and after the buffer exchange steps during enzyme digestion greatly reduced the oxidation artifacts induced by sample preparation and ensured the autosampler stability of the oxidation sites.

Another strategy to increase method repeatability and robustness across laboratories and throughput is to use an automated workflow for sample preparation.^{9–29–31–34} Qian et al. developed an automated peptide mapping procedure that features the automation of a “micro-dialysis” step, an efficient desalting approach prior to proteolytic digestion; the manual and automated procedures demonstrated comparable protein sequence coverage, digestion completeness and consistency, and PTM quantification.³¹ Sitasuwan et al.³⁴ and Ogada et al.³² reported a fully automated tip-based MAM sample preparation protocol. Bauer et al. developed and qualified a fully automated sample preparation for LC-MS peptide mapping using protein purification columns for the buffer exchange prior to enzymatic digestion.²⁹ Millan-Martin et al. described an automated and fast trypsin digestion protocol using immobilized trypsin on magnetic beads,⁹ which was incorporated into an optimized peptide mapping workflow by LC-MS. The authors demonstrated the global applicability and robustness of this workflow at four independent sites in Europe. Kristensen et al.³³ described an optimized MAM workflow addressing missed cleavages and chromatographic tailing and carry-over of hydrophobic peptides. A two-step digestion at high and low temperatures was used to reduce the number of missed cleavages and obtain a more complete digestion profile; a C4 column was used to reduce the chromatographic peak tailing and carry-over for hydrophobic peptides.³³ Guan et al.³⁵ showed better stability of digests quenched by 8 M guanidine-HCl, 250 mM acetate, pH 4.7 at a volume ratio of 1:3 compared to those quenched by 2% formic acid at a volume ratio of 1:10 (digestion quenching solution: digests). Finally, Millan-Martin et al. recently published comprehensive MAM workflows for biotherapeutic characterization and cGMP testing³⁶ that included detailed protocols of sample preparations used for MAM from previous publications.^9,26

In addition to minimizing sample preparation-introduced artifacts during MAM development, other key aspects should also be assessed, such as digestion efficiency in terms of levels of peptides with missed or nonspecific cleavages. Based on our experience and accumulated knowledge from the literature, to minimize deamidation and oxidation artifacts and maximize the digestion efficiency, we recommend considerations for each step of a typical trypsin digestion protocol (Table 2). It should be noted that, although the recommended conditions work well for multiple molecules with different formats, critical parameters might require optimization for some specific molecules and for the specific protease used for the digestion step, as MAM is not limited to trypsin digestion.

Table 2. Considerations of MAM trypsin digestion for minimized artifacts.

Step	Considerations	Purpose
Denaturing	Keep Met at 10 to 20 mM	Minimize oxidation artifact
Reduction	Buffer* at pH 7 to 7.5, 37°C	Minimize deamidation artifact while keeping effective reduction
Alkylation	Buffer at pH 7 to 7.5, 37°C or room temperature	Minimize deamidation artifact while keeping effective alkylation
Buffer Exchange	Keep Met at 10 to 20 mM, pH 7 to 7.5 buffer	Minimize oxidation artifact
Trypsin Digestion	Buffer* at pH 7 to 7.5, 37°C, 1 h to 2 h	Minimize deamidation artifact while keeping effective digestion
Quenching	Acidify with TFA (to ~pH 3) and~20 mM Met	Minimize post-digestion oxidation artifact and ensure auto-sampler stability

*Phosphate or bicarbonate buffer should be avoided since it introduces deamidation artifacts.^30,37.

Considerations of instrumentation for QC use

An ideal MS instrument for MAM implementation in QC is expected to be robust and have a low failure rate, a footprint that is practical for cGMP environments, and long-term instrument support commitment from vendors.⁷ In addition, data acquisition and analysis should be performed using software that complies with the FDA’s code of federal regulations, title 21 part 11 (21 CFR Part 11), and data integrity and security should be ensured.⁷ Multiple biopharmaceutical companies have introduced MAM into early-stage development programs. Method transfer, analyst training, commercial QC-suitable instrument and software evaluation and planning all need to be considered at the beginning of the MAM implementation phase to ensure phase-appropriate end-to-end implementation.

Robustness and ease of operation (for both MS and software with fixed methods) are critical for QC use, making them key drivers in the selection of MAM instrument platform, software, and methods. The ease of implementation of MS instruments in the QC environment depends on the type of the instrument itself, on the software used, and on the feasibility of qualification under GMP conditions. MAM includes multiple steps and requires highly qualified operators; therefore, automation of as many steps as possible would be beneficial.

The use of single-quad detectors offers a promising opportunity to implement MS applications in tightly regulated QC environments. For example, Bauer et al.²⁹ performed qualification studies to compare the performance of a single-quadrupole (single-quad) detector with a (high-resolution) QToF mass spectrometer. Other kinds of assays implemented in QC also use single-quad detectors, e.g., the method from Honemann et al.²⁸ for monitoring free fatty acids as degradation products of polysorbate. Single-quad detectors are also already successfully used in commercial QC for the release measurements of an identity test using peptide mapping. Therefore, the implementation of such kinds of MS instruments is feasible under GMP regulations.

With regards to MAM, Zhang et al.³⁸ evaluated the performance of a high-resolution Orbitrap instrument and several low-resolution single-quadrupole and triple-quadrupole instruments to assess the suitability of these instruments for MAM applications. The performance of MAM was evaluated using 184 attributes with abundances varying from 0.002% to 40%. This work showed the superb performance of the high-resolution instruments with a limit of quantitation (LOQ) as low as 0.002%, and near 1% LOQ for single-quadrupole instruments in scan mode.³⁸ The LOQ for each type of attribute was defined as the minimum attribute concentration with relative standard deviation (RSD) ≤10%.³⁹ The authors recommended the single-quadrupole instruments in scan mode for many routine MAM applications when the analysis of attributes below 1% is not critical, and the high-resolution instruments for scenarios where higher performance is desired. Due to the complex method development process and the lack of capability for NPD and retrospective data processing, quadrupole instruments in single ion monitoring or multiple-reaction monitoring modes were not recommended for MAM use.

Evans et al. recently reported a validated MAM and identity method – ID-MAM – for commercial release and stability testing of a bispecific antibody (BsAb).¹⁸ To address the challenges of operating MAM with high-resolution MS instruments in commercial QC labs and the requirement by several countries for identity testing using a peptide mapping method for a BsAb, the authors used a high-resolution MS instrument for comprehensive characterization during development. They then successfully developed and validated a targeted multi-attribute monitoring method in three development labs and three QC labs using a low-resolution MS that has a fully automated data processing workflow suitable for identity testing, sequence variant control, and quantitation of three selected CQAs (CDR isomerization, Fc Met oxidation, and CDR Met oxidation) in commercial QC labs. This is the first reported MS-based peptide mapping method implemented in GMP-compliant QC labs for commercial release and stability testing of a biotherapeutic drug product. While both studies^18,38 demonstrated the fit-for-purpose use of low-resolution MS for MAM, MAM’s NPD capability compared to high-resolution MS was not evaluated, which is a critical component for full use of MAM capability and should be carefully considered when implementing MAM for QC.¹¹ Low-resolution MS might not have sufficient specificity and sensitivity for NPD. In addition, while Cao et al. developed and qualified a Quadrupole Dalton (QDa)–based focused peptide mapping method to monitor a site-specific deamidation (with claimed 4.5% LOQ) for co-formulated antibodies for which the use of conventional charge-based methods proved to be challenging,⁴⁰ it required selecting a mass charge ratio for selected ion recording channels, optimized MS settings, and using TFA as an ion-pairing reagent in the mobile phases for reduced peak interference. The challenges and general feasibility of using low-resolution MS for sensitive and reliable quantitation of important PTMs (such as deamidation, which is often a charge variant CQA for many products) need to be further assessed. A more QC-suitable high-resolution orbitrap instrument (Exploris MX) designed to deliver high resolution accurate full MS data and is easy to operate in cGMP environments is now available.

Another challenge for and difference between MAM and conventional methods is that the MS-based MAM has the inherent issue of ionization efficiency difference between different peptide isoforms. Zhang et al. proposed a run-time calibration using a well-characterized reference standard as a solution to address the concern over differences in ionization efficiency, as may occur as a result of instrument change.³⁹ The authors developed new mathematical methods to calculate the attribute abundance in the sample, using the reference standard as calibrant to correct and greatly reduce instrument-to-instrument and sample-preparation variabilities.

Attribute selection and targeted attribute quantitation

The identification of potential CQAs (pCQAs) is based primarily on the assessment of attribute impact to safety, immunogenicity, bioactivity, and pharmacokinetics.⁴¹ pCQAs should be evaluated based on attribute type and location on the molecule. Modifications that may occur in binding regions (e.g., CDR and neonatal Fc receptor [FcRn]), and glycans that are part of the mechanism of action should be considered. Molecule-specific attributes (e.g., non-consensus glycosylation,⁴² sequence variants, and conjugation-related ADC variants) should also be investigated and evaluated during development.

Measurements by MAM provide a much greater capability to quantify individual CQAs than conventional methods. Charge attributes that are traditionally monitored in aggregate by IEC would instead be monitored by MAM at a site-specific level, representing an important advancement in quality by design (QbD) biopharmaceutical development. Antibodies contain hundreds of potential modification sites, and not all of these sites will be automatically selected to be controlled by MAM. Filtering based on impact, observed levels, and potential for variation in processing and storage must be applied for the attribute selection. A comprehensive stress panel should be considered for MAM analysis to select the stability-indicating charge- and oxidation attributes that are likely to contribute to IEC changes.¹¹ How to report CQAs should also be carefully considered. The reported MAM values could be for individual sites, or for the sum of all sites with similar modifications, such as deamidation or oxidation, or for the indicator/reporter peptides, which might be considered if the level of the CQA is low for the same type of modification.

New peak detection

MAM’s NPD function compares the three-dimensional (retention time, mass, intensity) result from a test article with the result from a reference standard and enables the detection of new peaks or unexpected changes that are not directly monitored by the TAQ. NPD is required for successful MAM implementation as a purity-indicating assay.¹² To achieve meaningful NPD results, several key parameters of the NPD workflow must be defined, including the MS intensity threshold, fold-change threshold, mass accuracy, and retention time tolerance windows.^12,17 A successful NPD method should allow sensitive and accurate detection of truly changed, product quality – related species compared to the reference standard sample, and minimize false positives and false negatives. To achieve this, empirical determination of phase-appropriate and product-specific NPD settings might be required.

Currently, limited detailed information on the NPD function has been reported in the literature, and the industry recognizes that NPD is a challenging and important function that is required for MAM implementation in QC, especially for its use in monitoring predicted and unexpected product-related impurities/changes under different storage conditions and batch testing throughout the life cycle of a product.^11,12,17

The NPD interlaboratory study done by MAM Consortium reported the industry-wide performance of NPD from 28 participating laboratories, using pre-digested samples of the National Institute of Standards and Technology mAb Reference Material 8671.¹² This work showed that the NPD parameters used across the industry were representative of high-resolution MS performance capabilities. To highlight the need to further refine MAM methodology and accelerate its implementation into QC, common sources of variability that affect NPD and are critical to MAM’s performance as a purity assay were also reported. One major challenge is to find an NPD threshold value that has sensitivity for identifying true, low-level new peaks while preventing or minimizing false-positive NPD.⁹ The authors also highlighted the considerations for empirical determination of NPD thresholds, and suggested that multiple approaches could be used to determine optimal NPD settings.⁹ To determine appropriate NPD thresholds, and measure limits of detection and dynamic range for NPD, the authors also recommended that the test set include peptides representing varied concentrations of “new” species representative of host cell protein peptides or other process impurities.⁹ In addition, different thresholds for NPD detection and reporting can be considered, as the detection level is based on the method, while the reporting level is based on the attribute. Depending on the software platform used, some filtering parameters to reduce the number of new detected components that are not related to PQA changes may be required. Vendor-provided software may or may not be able to do further filtering in the QC environment. Some vendor-neutral software might be able to provide better options to apply NPD within the MAM approach.

Sadek et al. proposed an improved NPD strategy to curate false positives, such as adducts,⁴³ using a “known peak list” concept, and the implementation of an NPD system suitability control strategy. Sequence variant co-mixes were also used to measure the performance of the NPD function.⁴³ Such strategies would reduce or avoid false detection of NPD in QC and mitigate the risk of creating time-intensive discrepancy investigations or unnecessary batch rejection.

Updates of both true and false-positive new peak information in the MAM peptide library throughout the life cycle of the product is expected to be an effective way to achieve the appropriate threshold for NPD, and it supports later targeted monitoring of newly identified PQAs. If MAM is implemented at the post-FIH stage, release and stability samples of clinical batches should be retained so that the clinical exposure of the newly detected impurities can be determined from the characterization of these retained samples.¹⁷ If MAM is implemented prior to a FIH study, then retrospective analysis of the data can be used for these newly identified PQAs.

Method qualification and validation

Method validation is a key scientific and regulatory consideration and requirement for MAM implementation in QC for therapeutic proteins. MAM validation should follow ICH guidelines and FDA guidance for analytical procedure validation as applied to other conventional methods,^21–44–46 and may benefit from considering the enhanced approach principles of ICH Q14,⁴⁷ such as including analytical target profiles (ATPs) for each CQA.¹⁷ According to ICH Q2, several parameters, i.e., accuracy, precision (repeatability/intermediate precision), specificity, quantitation limit, linearity, and range, should be considered during MAM qualification and validation.⁴⁵ Acceptable method performance needs to be demonstrated for MAM’s intended purpose: targeted attribute quantitation and new peak detection.

A major challenge in MAM validation is the demonstration of suitable method precision despite the complexities and inherent signal fluctuations and variations of the sensitive MS measurement.¹⁰ Other challenges may include determining the detection and quantitation limits for both TAQ and NPD measurements, and establishing appropriate system suitability testing (SST) for method performance control.¹¹ Furthermore, we need to consider what the QC reported value(s) should be for MAM. The method can readily be validated for site-specific results, but if the reported value is a sum of similar variants or if an indicator/reporter system is used, then the actual reported values should be validated. In addition to regular precision measurements (repeatability and intermediate precision) using the same test sample from a single batch, assessing multiple batches of a product is suggested. This would ensure that MAM can assess low levels of batch-to-batch variability for individual CQA measurements, and increase understanding of MAM’s discriminatory power by means of precision comparison between a single batch and inherent batch-to-batch variability.¹¹ For the TAQ, since setting detection and quantitation limits of each quality attribute may not be practical, selecting a carefully designed subset of peptide peaks that could be more representative of all intended quality attributes is desirable. Evans et al.¹⁸ discussed their strategy for selecting CQAs to monitor using a targeted MAM, with the goal of covering all critical degradation pathways for the specific BsAb product. Based on the considerations of structure/function relationships, risk assessments, clinical batch history, and product stability trends, the TAQ included only monitoring of CDR isomerization, Fc Met oxidation, and CDR Met oxidation. Additional BsAb attributes, such as deamidation and glycosylation, that were monitored by alternative control strategies were not included in the low-resolution MAM to simplify global filings. The authors emphasized this CQA selection as a general strategy to use MAM in QC.¹⁸ Rogstad et al. discussed that ideal limit testing would include modified and unmodified peptide pairs with different size, retention time, and abundance to represent the range of characteristics of attributes being measured.¹¹ Similarly, the EFPIA regulatory considerations paper¹⁷ suggested that certain quality attributes may be used as surrogates for others, and grouping certain attributes could be considered. For more complicated NPD limit determination, Rogstad et al. suggested that an approach to determine generalized limits can be applied, where other known peptides are spiked into the product sample and the best peptides to be introduced will be the potential impurities and expected protein modifications that may not be actively monitored.¹¹ Similarly, the authors of the EFPIA paper also suggested spiking varying amounts of synthetic peptides, or peptides resulting from a different enzyme digest (e.g., chymotryptic peptides in a tryptic digest), for the empirical determination of NPD thresholds.

SST is needed to ensure that the performance of an analytical method such as MAM is reliable and consistent and that measurements from a testing session are suitable for their intended purpose. With the appropriate design and adequate acceptance criteria, SST will account for fluctuations from sample preparation and LC-MS analysis using the sensitive and complex MAM. The standard SST options include premade peptide standard mixtures, digests of known proteins or reference standard samples, with or without a low-level spiked protein or peptides. To ensure the data integrity of MAM analysis, in addition to sample preparation and LC-MS analysis, SST should also consider the data processing to account for the entire MAM workflow. In addition to this SST for each testing session and before MAM is run, a separate evaluation should also be routinely performed to evaluate the LC-MS system for optimum performance using a protein digest standard, especially any time the instrument is “modified” or being “serviced.”

Given the complexity and breadth of data produced by MS, MAM validation is more complicated than that of conventional LC-based methods. Xu et al. reported a QDa-based MAM for product characterization, process development, and QC of therapeutic proteins.⁵ Gosh et al.⁴⁸ fully validated a sensitive LC-MS/MS antigen identification pipeline following the current FDA and European Medicines Agency guidelines. Hao et al.¹⁰ recently described the desired MAM performance profile and addressed the major scientific challenges of MAM validation. The authors reported a platform method validation strategy and the performance of an optimized MAM workflow to support accelerated product development. MAM performance for common and representative PQAs (N-deamidation, D-isomerization, M-oxidation, sequence variant, and several forms of N-glycosylation) was demonstrated using three mAb products. The ATP concept and relevant regulatory guidelines were used to guide the design of and acceptance criteria for method validation. Key method characteristics, including specificity, accuracy, linearity, precision, quantitation limit, and robustness were evaluated. A comprehensive SST with carefully designed acceptance criteria was developed to evaluate method performance and final data quality.¹⁰ A well-characterized immunoglobulin 1 (IgG1) antibody, prepared alongside samples of interest in each analysis, was used as an assay control to demonstrate the performance of both sample preparation and LC-MS analysis. To allow the control of specific quality attributes that are unique to the product, a product-specific reference standard was also included in each assay. Both qualitative and quantitative criteria were used to assess the holistic performance of the workflow, including reduction, alkylation, and sample digestion, specific peptide retention time and LC separation, column age, mass accuracy, MS signal intensity, and relative quantitation. Acceptance criteria were designed to ensure the expected method performance and the validity of the results, while accommodating small fluctuations in instrument performance and sample preparation.¹⁰ To replace visual inspection¹⁰ and further streamline the SST evaluation, recommendations on the SST criteria and the rationale when a reference standard is used as SST and bracketing samples for each test session are available (Table 3).

Table 3. Considerations of SST controls of bracketing reference standards for MAM targeted variant analysis*.

SST Evaluation Criteria	Rationale for Provided Control
1) Mass accuracy for designated peptides	Detector mass accuracy performance across the retention time range of the method and the run sequence.
2) The absolute peak height of designated peak heights is within the given ranges	Assures reference standard sample preparation, chromatographic separation, and mass spectrometry performance across the retention time range of the method and the run sequence.
3) The relative peak height of designated peptides is within the given ranges	Replaces visual inspection^Citation10 for overall pattern consistently (peak intensity relative to the most abundant peak intensity). Covers the remaining peaks not covered by SST evaluation criteria 1) and 2). Ensures overall pattern is consistent with historical pattern.
4) The relative abundance of designated peptides is within the given ranges	Assures the quantitation of attributes. Specific peptides are identified with acceptance criteria that assure method sensitivity, column separation/column age, sample preparation (alkylation and oxidation), and glycan quantitation.

*The acceptance criteria can be established based on the statistical analysis of many data points acquired during method development/qualification/validation.

Setting specifications

As stated in the recent EFPIA position paper,¹⁷ when using MAM, release and shelf-life specifications and acceptance criteria should be set only for those quality attributes determined as CQAs, and should follow the same principles as applied for any other purity method according to ICH Q6B.²¹ Several key aspects should be considered for setting a CQA specification limit, including preclinical and clinical experience, method performance, process capability (batch-to-batch variations) and stability profile. Dose escalation studies also need to be supported by MAM, where possible. If this is not possible, a bridging study will be key to link MAM values to the conventional method data that were used to support the clinical phases. According to this EFPIA paper, specifications could be set based on early indicator peptides representative of a certain PQA class (e.g., the peptide that contains the site most susceptible to methionine oxidation, usually DTLMISR in the Fc part of an IgG,⁴⁹ to represent the overall oxidation status of a mAb) prior to pivotal studies. To reduce complexity, additional recommendations from the paper¹⁷ were:

• Report the sum of relative abundance of the same type of modifications with similar criticality to reduce complexity (e.g., oxidation in the CDR).

• For an attribute that is defined as a CQA during late-stage development and cannot be measured by conventional methods, define the acceptance criteria of this new CQA. qualifying the MAM method for this CQA and analyze retention samples or conduct retrospective analysis of MAM data from development and stability.

• Use a two-tiered NPD specification approach, with an alert and action limit, for batch disposition. The authors provided examples on this approach and indicated the alert and action limit definitions would be company-specific. While exceeding the alert limit will only trigger peak characterization and does not affect batch disposition, exceeding the action limit will put the batch on hold and trigger an investigation.

Part II. MAM implementation strategies and case studies

With the capabilities and advantages discussed above, MAM implementation in QC is expected to require fewer resources than the preparation and execution of multiple replaced conventional QC methods, while enhancing deep understanding of products. However, successful industry examples for MAM implementation in QC are still quite limited, and few details of the implementation approaches, challenges, and strategies have been reported thus far. In the following sections, we review our MAM implementation, the challenges, and solutions through case studies and strategies.

Challenges and solutions of MAM implementation

Like other biopharmaceutical companies that are considering or have implemented MAM in QC, our major focus has initially been on implementing MAM during early-stage clinical development. Through these practices, we have gained experience in instrument and software qualification and implementation in QC, and optimized sample preparation and LC-MS parameters for consistent method performance for multiple attributes, including glycosylation, oxidation, deamidation, non-glycosylated heavy chain (NGHC), clips, glycation, and isomerization.

Case study 1. Phase 1 MAM implementation, mAb1

The first Phase 1 (clinical stage) project to implement MAM on the Certificate of Analysis (COA) was mAb1. Selecting stability-indicating attributes by MAM was a unique challenge for this molecule because it lacked pCQAs that were susceptible to change upon stress. The CDRs had no degradation hotspots, and Fc binding was not part of the mechanism of action. For example, by MAM, a pCQA methionine oxidation site was below clinically impactful levels in the reference standard and forced degradation levels. While this pCQA did not provide much information on the stability condition, a more sensitive and quantitative approach was to monitor a different, non-pCQA methionine oxidation site, which had a greater than 10-fold change in the stressed material, compared to unstressed material. Moreover, it could serve as a surrogate for other methionine residues as they trend similarly under stress. In this way, because they show a higher response to stress than pCQAs, non-pCQAs can serve as reporter sites to provide the worst-case condition of the protein. These reporter sites also have higher starting levels in the reference standard and show acceptable method performance. Reporter sites are more sensitive and quantitative than monitoring pCQAs and provide better control of the product consistency. Therefore, based on the stability of the molecule, it may be a more suitable strategy to monitor reporter sites rather than pCQAs.

Case study 2. A multimeric fab conjugate, mAb2

Another Phase 1 project to implement MAM on the COA was a multimeric Fab conjugate. Due to the structural complexity, conventional ICIEF and IEC are either dominated by peaks that represent the hydrolysis product at the conjugation site, which masks potential protein-related charge variants, or display poor resolution due to the complexity of overlapping, minor variants from Fab intermediates, along with hydrolysis. MAM is therefore selected as suitable for charge heterogeneity characterization and testing MAM also provides quantitation of charge-related attributes at the peptide level for the conjugate, while IEC still remains the charge assay for the Fab intermediate.

Case study 3. Preparation of MAM GMP implementation, mAb3

In this case study, we describe insights from preparing for GMP use of MAM and the performance of an MAM validation feasibility study for a Phase 1 molecule, mAb3.

In line with information technology (IT) systems used at other sites for MAM implementation, we installed an enterprise solution that offers access to data from different locations (e.g., personal laptops, different sites) and enables simplified connection of new (GMP-qualified) instruments. Centralization and security of data storage is realized by instantaneously moving the acquired raw data and associated metadata to a combination of a network-attached storage share and a structured query language server. This is done by a so-called domain controller, which controls the data flow. The data is protected from loss through regular backups. Data evaluation is performed by users who access the stored data mostly through a terminal server.

The IT system allows the generation of both discovery and QC data. Methods are developed in “discovery mode” and are then transferred to the QC part of the system for generation of QC data. Appropriate handling of data as well as segregation of duties is secured by defined user roles and associated (access) rights. The system allows the import of data from other instruments and data sources to the discovery part and the export of working copies of raw data to external sources, while the original raw data remains in the system. This allows flexible data evaluation using various third-party software solutions used in discovery applications. In summary, the established solution greatly facilitates data connectivity from discovery to QC.

We also established a system that enables the connection of additional instruments without changing the validated IT system itself. The Chromeleon software implementation required substantial resources and time to complete validation, generate extensive life-cycle and computer system validation documentation and operating procedures, and provide dedicated support by users and applications. We recommend that sponsors plan for sufficient time if they are considering a QC-suitable software implementation for MAM.

MS and high-performance liquid chromatography (HPLC) instruments were qualified at the same time of validation as the Chromeleon software IT system. Data integrity for instruments was already ensured by connecting them to this validated IT system. As for the IT system, sponsors should consider the effort of maintaining and operating the instruments, e.g., GMP-compliant documentation, instrument logbook review, and firmware updates on control computers. The LC and MS parts are viewed as one instrument, which enables faster qualification and is more streamlined from a data integrity perspective. However, flexibly substituting the connected LC part, which is frequently done in case of technical issues or for experimental design reasons during non-GMP extended characterization analyses, requires substantial effort under GMP conditions and practically is not feasible.

Other challenges we faced during the initial method validation were the resources and experience required to successfully perform typical GMP work due to the greater complexity of the MAM system compared to conventional methods. Another unexpected challenge was that mid-validation, we had to substitute the LC part of the GMP-qualified LC-MS systems due to observed issues, which required time-consuming requalification and further delayed validation completion. The complexity of the instrumentation and method also require extensive analyst training. In addition, sponsors should arrange to have replacement equipment available and promptly resolve unplanned instrument-related events/changes.

MAM including TAQ for seven attributes and NPD using the parameters described in our recent work⁴³ was successfully validated. Workflows, infrastructure readiness, expert knowledge, lessons learned, and validation results will be leveraged for future GMP applications of MAM.

Case study 4. A late-stage MAM implementation pilot for replacing conventional methods, mAb4

We conducted a pilot study for mAb4 with the intention of using MAM to potentially replace several conventional control system assays for drug substance, including IEC, CE-Glycan, identity test by UV peptide mapping, NGHC by reduced capillary electrophoresis sodium dodecyl sulfate (CE-SDS), and total Fc oxidation by RP-UHPLC. With this intended purpose, we identified 47 attributes that provide a detailed molecular profile of mAb4, including 5 oxidation attributes, 25 charge attribute – related PTMs and clips, and all relevant glycosylation attributes. The list of 47 attributes includes non-CQAs as potential surrogates to establish correlation and indirect control of CQAs, and was further evaluated through sample testing and data analysis. A full stress panel, including pH stress (acidic and basic pH), thermal stress, light exposure (ICH light and ambient light), and oxidation stress by AAPH, was analyzed. In addition, an intermediate precision study of the reference standard material, including different sample preparation sessions, columns, and instruments, was conducted to evaluate the method performance.

The 47 quality attributes were iteratively refined based on the following four considerations: 1) % abundance of attributes, relative to both unstressed and stress panels; 2) stability-indicating capabilities; 3) method performance–identifying and selecting those pCQAs with acceptable precision performance and quality attributes that can serve as the reporter attributes for all other (p)CQAs; and 4) for glycosylation-related attributes, a product-specific criticality assessment determined which attributes were moved forward for the validation, although the peak areas of all of the glycosylation-related attributes were used as the denominator for the calculation of % abundance. As a result, following the initial MAM evaluation, 19 attributes were identified that are most representative of product quality or could serve as reporter attributes for other pCQAs with robust quantitation. Some attributes were also included in the validation, as they are considered informative for platform knowledge. This list of 19 attributes will be further filtered strategically after method validation and product-specific process validation and bridging studies, and only a subset of this list is expected to be reported on the drug substance COA. The mean value, relative deviation (SD), and % relative standard deviation (RSD) of a subset of representative attributes for mAb4 from the intermediate precision study are listed in Table 4.

Table 4. Mean value, standard deviation (SD), and % relative standard deviation (RSD) of % abundance of a subset of representative attributes for mAb4 from the intermediate precision study (n = 24, using three instruments, three column resin lots, with six independent sample preparations by two analysts on two different days).

Attribute	Mean (%)	SD	% RSD
Methionine Oxidation in Fc	2.3	0.2	10.3
N-terminal Pyroglutamate	1.0	0.1	13.5
N-terminal VHS Extension	0.8	0.1	7.2
Tryptophan Oxidation	1.7	0.2	13.7
Aspartic acid Succinimide formation	0.7	0.1	8.9
Internal clip	0.5	0.1	13.0
Lysine Glycation	2.9	0.5	17.7
Asparagine Deamidation	1.1	0.1	11.6
NGHC	2.6	0.3	10.2
M5	1.6	0.2	12.9
G0F	66.7	1.1	1.7

We also performed a preliminary comparison between MAM and IEC data, since MAM will potentially replace the conventional method for mAb4. Extended characterization studies of isolated IEC fractions of mAb4 revealed that the major variant forms in the acidic region are deamidation, glycation, and low-molecular weight forms (LMWF). The major variants eluting in the basic region are high-molecular weight forms (HMWF), oxidation, C-terminal lysine, proline amidation, and N-terminal VHS. Using the acidic region as an example, we assessed the method comparison between MAM charge-attribute quantitation and IEC. LMWFs, which are part of the acidic region by IEC, are measured by non-reduced CE-SDS and are therefore not included for control by MAM. The major acidic variants eluting in the acidic region, including deamidation on the four most susceptible Asn sites and glycation on the six most susceptible Lys sites, which are identified through the analysis of the stressed models, are summed.

The comparison to the % acidic region by IEC is shown in Figure 2. The plot included results from nine real-time and accelerated stability and stress samples, derived from four different drug substance batches. Peptide-level MAM data for deamidation and glycation was combined using binomial correction⁵⁰ to reflect values at the intact protein level. Sialylated glycans were excluded from this analysis due to low and consistent levels. Although an equivalency between MAM and IEC was not expected, a strong relationship between MAM and IEC was observed. The methods were in good agreement with the known chemical changes occurring during stress and storage. The comparison enabled strategic filtering and surrogate selection for the final MAM peptides reported on the COA in lieu of the IEC acidic region.

Figure 2. Comparison between acidic variants measured by MAM and IEC in stability and stress samples of mAb4.

A schematic graph illustrating the comparison of MAM and IEC methods in measuring acidic variants in stability and stress samples of mAb4.

A challenge for applications of MAM’s non-targeted approaches (i.e., NPD) in QC is that NPD goes beyond confirmation of expected product quality and instead interrogates product quality outside of conventional expectations of laboratory capabilities and quality system requirements. This creates barriers for technology implementation. Greater flexibility in the MAM testing framework is needed to overcome these barriers. As mentioned in Part I above, NPD should be applied in a phase-appropriate manner during drug development and for extended characterization for comparability studies supporting manufacturing changes and transfers. NPD can also be applied at process control points, which go before drug product release where unexpected changes to product quality (i.e., potential new peak findings) can be detected early in the process and investigated. Following a risk-based approach, both TAQ and NPD measurements could be applied for drug substance release and stability testing, and only TAQ measurements could be applied for drug product release. These control strategies would provide a pathway for global deployment and help actualize the benefit of MAM technology in the commercial QC environment.

Phase I MAM strategy

Sponsors should consider the resource burden when MAM is added to the control system prior to method replacement, resulting in dual-control system methods, i.e., both MAM and the conventional methods would be in use. Three considerations for Phase 1 MAM implementation are as follows. First, consider selectively using MAM as a non-GMP method by default, only requiring method qualification. Project selection should be based on a product-specific risk and benefit assessment (e.g., the potential to replace >2-3 assays by MAM) and aim to cover different modalities, attributes, and structures to evaluate MAM as a platform method in the future. Second, implement MAM for GMP/COA use when conventional methods are not suitable or adequate to detect CQAs. Third, consider applying MAM as a non-GMP method for peptide characterization, from molecule assessment for a manufacturing developability study, to process development and validation, and comparability studies. In these situations, both the TAQ and NPD functions can be applied, and no method qualification is required, which enables more MAM performance evaluation and retrospective data analysis with minimal front-end investment. In this case, MAM and conventional methods are used in parallel during molecule assessment, which is expected to provide enough supporting data and evidence to use MAM as part of a control system directly prior to FIH studies, without the need for further method bridging. Dialogue should be maintained with health authorities as more experience is gained, while making further adjustments to the strategy as needed.

Part III. Non-standard peptide mapping-based MAMs and MAM applications beyond QC release and stability testing

Most current MAM applications are based on LC-MS analysis of a reduced trypsin digestion with C18 reversed-phase LC separation. Alternative protein digestion using different enzymes such as Lys-C, or Glu-C, and LC columns (e.g., C4³³ might be more desirable for specific applications, to obtain complete sequence coverage for enhanced MAM monitoring and address challenges with current MAM workflows. For example, as our current trypsin digestion protocol does not provide complete protein sequence coverage or for the low-concentration drug product of a new modality molecule by LC-MS analysis, an alternative enzyme digestion protocol has to be developed using a different enzyme and procedures to accommodate a low protein concentration. Buettner et al.¹⁴ performed multi-attribute monitoring of complex erythropoietin beta glycosylation by Glu-C LC-MS peptide mapping. Li et al.⁵¹ developed an improved robust sample preparation using Lys-C for high-throughput MAM testing without a desalting step, using a HSS T3 column for enhanced PQA monitoring for hydrophilic peptides, including optimized peptide retention and separation of a stability-indicating VSNK peptide. The analytical performance of identity, quality attribute monitoring, and NPD with this method was evaluated.⁵² In addition, when mAbs require assessing PQAs such as disulfide bond linkages, free thiol or cysteine-related modifications, and trisulfides, a non-reduced peptide mapping-based MAM might be needed.⁵³ For example, disulfide-related variants can be identified by assessing peptide map differences between non-reduced and reduced samples. NR-CE-SDS allows separation of chains that lack inter-chain disulfide bonds but does not identify unpaired cysteine sites. HIC can be used to resolve thiol variants, including intended and unintended conjugation sites for ADCs, also without site identification. RPLC can separate disulfide isoforms of different types of IgGs.⁵⁴ Release and/or stability test methods are expected to include MAM for site-specific monitoring plus chromatographic and electrophoretic methods for lower-resolution structural confirmation.

MAM represents an advancement in testing methodology. While we focused on MAM use as a QC release and stability assay in this work, using MAM as part of an extended characterization method is already common in other areas, such as process development and eventually process validation, comparability studies before and after process changes, or between innovative biotherapeutics and biosimilars, and for advanced process control. The use of MAM as a process analytical technology for advanced process control and real-time monitoring is an emerging application that could enable real-time release testing (RTRT) strategies. Adaptive control of glycans and charge variants by MAM would lead to validated process control and can justify removing glycan and IEC/ICIEF from drug substance release. Defining CQAs and understanding their correlation with critical process parameters is key to ensuring drug safety and efficacy, when applying QbD and RTRT concepts to the manufacturing process. Song et al.⁵⁵ demonstrated global harmonization on automated MAM platforms and inter-site comparability, including for automated sample preparation and LC-MS instruments, as well as MAM support for cell-line development and cell culture, and downstream process development. Liu et al.⁵⁶ performed simultaneous monitoring and comparison of multiple PQAs throughout the 17-day cell culture process at different scales (3-liter development mini-bioreactor and 2000-liter GMP single-use bioreactor) using MAM, which provided critical information to understand any potential process performance differences during scale-up and/or technology transfer, and laid a solid foundation for possibly using MAM for real-time in-process control and release of biotherapeutics in the future. Subsequently, a fully integrated end-to-end automated online MAM system was developed for automated real-time online process monitoring.⁵⁷ A recent paper discussed⁵⁸ using MAM in three independent case studies, to track the behavior of mAb PQAs over the course of a 12-day cell culture experiment, to test the purity and identity of a product through analysis of samples spiked with host cell proteins, and to compare a drug product and a biosimilar with known sequence variants. All of these case studies demonstrated the wide application range of the MAM workflow.

Direct online connectivity of MAM to upstream and downstream processing unit operations could be an important future direction, as this would enable even more timely measurement of PQAs or CQAs at relevant points in the production process. The industry-wide BioPhorum Development Group recently conducted a multi-company survey on process analytical technology, and respondents identified MAM as having one of the highest potential business values for both upstream and downstream processing.⁵⁹ Continued maturity in the areas of online sampling, automated sample preparation, and data analysis will be key factors in enabling MAM-driven process control.

Part IV. MS in QC: beyond peptide-based MAM

MAM implementation in QC also brings many opportunities to use MS in QC for applications other than the current peptide mapping-based MAM. Carillo et al.⁶⁰ reported on the QC-compliant intact multi-attribute method (iMAM) as a flexible tool for the analysis of mAbs. The authors discussed using iMAM to monitor certain critical information that is missing from the current peptide mapping-based MAM workflows, such as correct-chain assembly; the presence of fragments, dimers, and larger aggregates; and the drug-to-antibody-ratio (DAR) of ADCs, with workflows using several intact analysis approaches under both denaturing and native conditions. In addition, peptide-based MAM is great for the detection of single-site modifications (e.g., Fc glycans and DTLMISR oxidation), but is less suitable for, and often underestimates, a low-level modification at multiple sites.¹⁹ Intact-level analysis might be needed to quantify such low-level multi-site modifications. Absolute quantitation of BsAb homodimers,⁶¹ previously done in a development lab, can now be performed in a QC environment as needed, with QC compliant instruments and software. Current examples and potential applications for MS in QC beyond peptide mapping-based MAM are summarized in Table 5.

Table 5. Current examples and potential MS applications in QC beyond the peptide mapping-based MAM.

Quality Attribute	Method	Reference
scFc Oxidation	Samples were treated with IdeS, EndoS, and dithiothreitol to generate three individual IgG subunits, LC, Fd’ and single chain Fc, RP-MS in commercial QC lab	Sokolowska et al.^Citation2
NGHC LC and HC Identity LC Glycation N-Linked Glycosylation Profile	A rapid intact MS-based MAM, directly out of bioreactors	Lanter et al.^Citation62
CDR Oxidation Sites	Used the IdeS protease-digested and reduced co-formulated mAbs	Xu et al.^Citation63
Conjugated Species of ADC DAR	Subunit and intact MS-based MAM	Martelet et al.^Citation64
Chain Assembly Fragments Aggregates Protein Identification	QC-compliant iMAM, through RP-MS, CEX-MS, and SEC-MS	Carillo et al.^Citation60
BsAb Homodimer	Standard addition-based absolute quantitation, intact MS, potential use in QC	Macchi et al.^Citation61

Part V. MAM outlook, remaining challenges

Above we reviewed and discussed the key considerations and steps taken for MAM implementation in QC and presented our current strategies for increasing experience with MAM performance and end-to-end (from early to commercial stage of protein drug development) MAM implementation workflows. While substantial progress has been made in implementing MAM technologies for QC, there are several remaining implementation questions and challenges. The main goal of MAM includes 1) enhanced process and product knowledge and direct control of CQAs (safety and efficacy to patients) and 2) potential to replace several conventional analytical methods for control of CQAs, which would result in increased speed to patients and cost saving in QC. While parallel testing may be needed at earlier stages, as regulators and sponsors continue their understanding of MAM, it may be treated like a more conventional method in terms of implementation as part of the control strategy in the next 5–10 years.

As an innovation approach, the general use of MAM for early-stage clinical phase and FIH specifications may overburden the QC capacity for supporting a large portfolio pipeline. Conventional methods and new technologies including MAM need to provide a cohesive product testing regimen that includes in-process, end-product, and stability testing, and also support product characterization during comparability exercises, reference standard qualification, and continuous process verification. Therefore, a balanced approach to using MAM for specific elements should be guided by ensuring the best control of product quality while enabling more efficient use of relevant control methods and technology.

As shown in Table 1, while MAM can measure a multitude of covalent modification sites, replacement of some methods (e.g., charge-based methods) may require substantial method bridging work. In addition, identification and validation of indicator (reporter, surrogate) results could add additional layers of complexity and work, especially when the indicator result indirectly controls a CQA. Therefore, for standard mAbs, using MAM in QC might not be a clear-cut option, considering the required investment and value returned when conventional methods remain suitable and adequate. However, the capability of performing MS analysis in QC labs creates many opportunities for future control systems. Future biotech products will be not just IgG1 mAbs, and thus more diverse and complex, rapid platform methods that can be used for any protein-based product are needed, and MAM is the best way to accomplish this. We can rapidly identify degradation hotspots and covalent molecular variants, enabling targeted QC testing using MAM.

One major challenge is deciding when, where, and how to implement MAM. From a knowledge-based perspective, the preferred option is to introduce MAM in the initial clinical control strategy. This approach might avoid intensive bridging studies for the replacement of multiple conventional QC methods, as bridging studies are required when MAM is introduced at late-stage development or during product life-cycle management. Although the application of MAM potentially requires extra product characterization to establish a meaningful list of relevant CQAs for specification setting, the reduced efforts for extended characterization studies might balance out this investment. Numerous reports demonstrate that MAM is beneficial for the accurate determination of identity, glycosylation, and oxidation compared to conventional QC methods. However, when MAM is used starting with FIH programs, parallel testing with MAM and conventional methods is often required, and therefore, an alternative strategy might be more practical. That is, starting with the peptide-based MAM for identity, glycan, and oxidation testing, then moving on to replacing other existing methods such as IEC when the value of MAM is demonstrated, and the overall workflow is optimized. For example, the charge-variant profile needs to be carefully assessed by an orthogonal testing method to ensure that all relevant proteoforms are still covered by the initially established peptide attribute list. The most obvious approach is to select a conventional test method (e.g., cation-exchange chromatography with UV detection) as the extended characterization method for comparability exercises and process characterization/validation studies. However, identifying new charge variants with MS detection by the application of native/intact LC-MS⁶⁵ or multi-dimensional LC-MS (mD-LC-MS)⁶⁶ (both with charge-variant separation prior to MS analysis) represents an alternative and innovative approach to maintain the validity of peptide mapping-based MAM approaches for complex antibody formats.

Another advantage of MAM is that MAM can not only detect new product variants via NPD but can also directly assess the identity of undesired variants by accurate mass determination and further MS/MS fragmentation for structural elucidation. In conclusion, the LC-MS peptide mapping-based MAM provides the basis for simultaneously monitoring quality attributes at sensitive amino acid levels, and, following comprehensive product characterization and adequate method qualification, in principle allows the replacement of currently applied QC assays for determination of identity, charge variants derived from post-translational (e.g., Fab glycosylation) or chemical degradation (e.g., oxidation, deamidation, isomerization, or glycation), and fragmentation. Since high-resolution ESI-MS will be introduced in the regular QC control strategy, further extension of MAM application field should be considered for complex protein formats in the future. iMAM-based approaches such as native LC-MS (with size-exclusion and/or cation-exchange separation prior to MS analysis) are required anyway to validate the peptide mapping-based MAM attribute list for completeness. Thus, the combined application of the intact (native) and peptide mapping-based MAM not only enables a holistic monitoring of quality attributes with high resolution, but also allows the future replacement of low-resolution QC methods for size-variant determination (e.g., SEC-UV and non-reduced CE-SDS). This strategy can focus specifications on identified CQAs and might sufficiently address the higher product variant diversity of co-formulations, bispecific antibodies, and other less-conventionally structured molecule types.

Abbreviations

ADC	=	antibody-drug conjugate
AGE	=	advanced glycation end product
ATP	=	analytical target profile
BLA	=	biologics license application
BsAb	=	bispecific antibody
CDR	=	complementarity-determining region
CE	=	capillary electrophoresis
CE-SDS	=	capillary electrophoresis sodium dodecyl sulfate
CEX	=	cation-exchange chromatography
cGMP	=	current good manufacturing practice
COA	=	certificate of analysis
CQA	=	critical quality attribute
DAR	=	drug-to-antibody ratio
EFPIA	=	European Federation of Pharmaceutical Industries and Associations
ELISA	=	enzyme-linked immunosorbent assay
EMA	=	European Medicines Agency
ESI	=	electrospray ionization
FcRn	=	neonatal Fc receptor
FDA	=	United States Food and Drug Administration
FIH	=	first-in-human
GMP	=	good manufacturing practice
HC	=	heavy chain
HILIC	=	hydrophilic interaction liquid chromatography
HMWF	=	high-molecular weight forms
HPLC	=	high-performance liquid chromatography
ICH	=	International Council for Harmonisation
ICIEF	=	imaged capillary isoelectric focusing
IEC	=	ion-exchange chromatography
IgG1	=	immunoglobulin G1
iMAM	=	intact multi-attribute method
IT	=	information technology
LC	=	light chain
LC	=	liquid chromatography
LC-MS	=	liquid chromatography-mass spectrometry
LC-MS/MS	=	liquid chromatography-tandem mass spectrometry
LMWF	=	low-molecular weight forms
LOQ	=	limit of quantitation
mAb	=	monoclonal antibody
MAM	=	multi-attribute method
mD	=	multi-dimensional
MS	=	mass spectrometry
NGHC	=	non-glycosylated heavy chain
NGS	=	next-generation sequencing
NPD	=	new peak detection
NR	=	non-reduced
NR-CE-SDS	=	non-reduced capillary electrophoresis sodium dodecyl sulfate
pCQA	=	potential critical quality attribute
PQA	=	product quality attribute
PTM	=	post-translational modification
QbD	=	quality by design
QC	=	quality control
QDa	=	quadrupole dalton
R-CE-SDS	=	reduced capillary electrophoresis sodium dodecyl sulphate
RP	=	reversed-phase
RSD	=	relative standard deviation
RTRT	=	real-time release testing
scFc	=	single-chain fragment crystallizable
SEC	=	size-exclusion chromatography
SST	=	system suitability testing
TAQ	=	targeted attribute quantitation
TFA	=	trifluoracetic acid
UHPLC	=	ultra-high-performance liquid chromatography
UV	=	ultraviolet
VHS	=	Valine-Histidine-Serine

Acknowledgments

We thank Riley Togashi, Liz Johnson, Qinjingwen Cao, Jia Guo, Delia Li, John Guan, Belen Tadesse, Rachel Liu, Susan Janes, Benjamin Moore, Chengfeng Ren, Frank Macchi, Lindsay Yang, Jack Harris, Laura Yee, Emily Liu, Vanessa Tran, Jennifer Moore, Jack Yim, John De Los Santos, Anthony Molano, Yunyi Lee, Tim Spirakes, Mary Zhu, Susan Janes, Irina Astafieva, and Kimia Rahimi for their contributions to MAM implementation. We also thank Nina Davis for the editorial review, Vikas Sharma, and Guoying Jiang for their strategic input.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.