A Practical Review of Technical Barriers and Solutions on Mass Spectrometry-Based Multi-Attribute Method (MAM) in Biotherapeutics

1. Introduction

1.1. Biotherapeutics and Critical Quality Attributes

Biotherapeutics, dominated by monoclonal antibodies (mAbs) and related formats such as fusion proteins and antibody drug conjugates (ADCs), account for one of the largest and fastest growing categories of approved medicines, with cumulative sales projected to exceed 300 billion USD within the next several years [1]. These molecules are produced in living mammalian systems through complex processes that expose them to a wide spectrum of co- and post-translational modifications (PTMs), including N- and O-linked glycosylation, methionine and tryptophan oxidation, asparagine deamidation, aspartate isomerization, N-terminal pyroglutamate formation, C-terminal lysine clipping, glycation, and succinimide intermediate formation [2]. Individual modifications can directly modulate Fc-receptor binding, effector function, serum half-life, immunogenicity, potency and overall safety, therefore, regulatory authorities require sponsors to identify the critical quality attributes (CQAs) of each product and to demonstrate appropriate control strategy throughout the product lifecycle in accordance with Quality by Design (QbD) principles articulated in ICH Q6B, Q8, Q12, and Q14 [3].

To meet these requirements, the biopharmaceutical industry has deployed orthogonal, single-attribute assays, e.g., ion-exchange chromatography (IEX) and capillary isoelectric focusing (cIEF) for charge variants, size-exclusion chromatography (SEC) and reduced or non-reduced capillary electrophoresis–sodium dodecyl sulfate (CE-SDS) for size variants, hydrophilic interaction liquid chromatography with fluorescence detection (HILIC-FLD) of released N-glycans for glycosylation profiling [4,5]. While these techniques are mature and well-anchored in cGMP environments, they are resource-intensive, produce datasets that can be difficult to reconcile, and largely deliver macro-level descriptions of heterogeneity rather than site-specific information [6].

1.2. Principles and Workflow of the Multi-Attribute Method

The mass spectrometry-based multi-attribute method (MAM) was first introduced in 2015 to address these limitations. The MAM is fundamentally a liquid chromatography–mass spectrometry (LC–MS) based platform that consolidate, in a single experiment, the site-specific quantitation of multiple critical quality attributes (CQAs) together with a non-targeted search for previously unobserved species [7]. In peptide level bottom-up form, MAM consists of three sequential stages: (i) reproducible enzymatic digestion of the therapeutic protein, preceded by denaturation, reduction, and alkylation; (ii) reversed-phase ultra high-performance liquid chromatography (UHPLC) separation of the resulting peptides with high-resolution accurate-mass (HRAM) MS detection; and (iii) two-phase data analysis. The analysis consists of a discovery phase that builds a project-specific peptide library of attributes of interest, followed by a routine monitoring phase that performs targeted attribute quantitation and new peak detection (NPD) [8]. The discovery phase typically uses data-dependent MS/MS acquisition on a high-resolution instrument to assign peptide sequences, localize PTMs, and define the mass, retention time, and charge state of each species of interest. The monitoring phase relies only on MS1 data, which allows deployment on simpler, QC-friendly instrument configurations. The complete workflow, from sample preparation through the discovery and monitoring phases, is summarized in Figure 1 [9]. Most product instability arises from covalent modifications that MS detects with high sensitivity, which makes MAM particularly well suited to stability testing [10].

A distinguishing feature of MAM relative to traditional peptide mapping is NPD, which compares LC-MS features between test and reference samples to identify new, missing, or significantly changed signals. The combination of quantitative attribute readouts and NPD is the foundation of the MAM value proposition: a single LC–MS experiment can, in principle, replace charge-variant assays (IEX/icIEF), released N-glycan analysis (HILIC-FLD), reduced CE-SDS for clipped variants, peptide-mapping identity tests, and host cell protein (HCP) analysis for process-related impurities [11,12]. Accordingly, MAM has been applied across the product lifecycle, including clone selection, upstream and downstream process development, comparability assessment, forced-degradation studies, and HCP analysis [13,14]. MAM does not, however, capture all quality attributes: aggregates, partially assembled species, and higher-order structure fall outside standard reduced peptide mapping and continue to require orthogonal biophysical methods [15,16].

1.3. Intact MAM and Subunit MAM Workflows

While bottom-up peptide-mapping MAM provides the highest site specificity, MAM has been extended to intact MAM (iMAM) and subunit MAM [17,18,19]. iMAM uses native or denaturing intact protein with HRAM detection to provide global proteoform information with minimal sample preparation [20]. It is particularly useful for mass confirmation identity testing, glycoform profiling, glycation quantitation. When combined with cation-exchange or size-exclusion chromatography, it also supports charge- and size-variant analysis with simultaneous structural identification [21,22,23,24]. Subunit MAM, in which the molecule is reduced and optionally cleaved with IdeS or Lys-C into defined fragments before LC-MS analysis, provides intermediate resolution and is especially informative for Antibody Drug Conjugate (ADC).

Several practical implementations illustrate the value of these complementary MAM strategies. A subunit level MAM workflow, qualified for accuracy, precision, linearity, range, and specificity, was shown capable of replacing three conventional release tests: ELISA-based identity, HILIC N-glycan profiling, and the RPLC ratio assay for co-formulated mAbs [25]. An intact-mass MAM workflow optimized for direct analysis of reduced mAbs sampled from bioreactors enabled rapid in-process decisions on identity, N-glycosylation, glycation, and aglycosylation, with results consistent with orthogonal HILIC, CGE, and peptide mapping methods [26]. Building on this, a QC-compliant intact level workflow that requires no reduction step monitored attributes not readily captured by bottom-up MAM, such as correct chain assembly and presence of fragments and aggregates, under both denaturing and native conditions. In addition, intact and subunit level MAM was subsequently extended to ADC characterization, with simultaneous monitoring of conjugation variants, drug-to-antibody ratio (DAR), and site-specific modifications [27,28,29]. Bottom-up MAM and iMAM have also been integrated for the characterization of a GLP-Fc fusion protein, illustrating that the two modes together can replace several orthogonal assays in complex molecule formats [30]. Finally, the MAM platform has been extended beyond mAbs and fusion proteins to viral-vector gene therapies, with the first LC-MS based MAM workflow for adeno-associated virus (AAV) program enabling peptide-level monitoring of capsid-protein modifications and HCP impurities [31,32].

Taken together, these examples illustrate how intact- and subunit-level MAM extensions have been applied across an increasingly diverse set of biotherapeutic modalities and quality attributes. Table 1 provides an overview of the attributes, risks, and monitoring goals of MAM characterization, linking each CQA class to its traditional assays, applicable MAM level, and corresponding monitoring strategy and benefits. The three analytical levels differ in sample preparation burden, site specificity, and the attributes each can resolve, and Table 2 compares them to help select the appropriate level for a given application.

1.4. The MAM Implementation: From Broad Conceptual Acceptance to Practical QC Deployment

Beyond this expansion in scope, MAM has also moved steadily toward regulated use. Over the past decade it has been adopted widely as a characterization tool in non-GMP development laboratories, and regulatory and industry engagement has accelerated its advancement toward QC implementation. The FDA Emerging Technology Program (ETP) has evaluated the use of MAM in quality control environments, the European Federation of Pharmaceutical Industries and Associations (EFPIA) has published a regulatory position paper supporting MAM for QC release and stability testing [65], and the MAM Consortium has conducted interlaboratory studies assessing method performance across participating laboratories [47]. A retrospective review of BLAs approved between 2016 and 2020 identified filings in which MS was used in a QC capacity [64], whereas an earlier analysis of BLAs approved from 2000 to 2015 found MS used for characterization but no such QC use [33,66].

Despite this momentum, the routine deployment of MAM in cGMP QC laboratories, particularly as a replacement for established chromatographic and electrophoretic methods, has proceeded at a conservative pace. The bottleneck is no longer the conceptual or regulatory case for MAM, both of which are now well established. Instead, the obstacles are concrete and technical, and they manifest at every layer of the workflow.

1.5. Scope of This Review

Studies have mapped the concept, application, and regulatory landscape of MAM, tracing its evolution from a research stage characterization tool to a QC release and stability method embraced by major biopharmaceutical companies, MS vendors, and an increasingly active regulatory community [67]. MAM is technically feasible, regulatorily acceptable, and operationally advantageous. What is missing from the literature is a focused, problem-driven summary of the technical barriers to MAM implementation and of the practical solutions reported, often as case studies, in primary publications, conference proceedings, vendor application notes, and interlaboratory studies. This review consolidates that scattered technical knowledge into a single, barrier-by-barrier reference. For each barrier, we examine the practical solutions reported in the recent literature, evaluate their relative implementation status and limitations, and identify the gaps where further methodological innovation is most needed. Building on these observations, we offer a perspective on solutions to the technical bottlenecks that will facilitate the transition of MAM from a characterization tool to a routine cGMP QC platform. The literature in this review was assembled from primary studies indexed in PubMed and Web of Science, primarily from the past ten years, with earlier foundational studies included where relevant, supplemented by MAM Consortium interlaboratory reports, and publicly available regulatory documents. Studies were included if they reported quantitative or procedural evidence for MAM sample preparation, LC–MS instrumentation, software, new peak detection, or regulatory implementation. We aim to provide an implementation-oriented resource for scientists actively developing, qualifying, transferring, or troubleshooting MAM workflows in industrial settings.

2. Sample Preparation Barriers and Practical Solutions

Sample preparation is consistently identified as the most influential and most variable step in MAM, since the process (denaturation, reduction, alkylation, buffer exchange, digestion, quench) used to digest the intact proteins into peptides can induce artifacts such as deamidation, oxidation, and incomplete or nonspecific cleavage that directly affect attribute quantitation [68]. MAM aims to quantify the modifications that can be affected by sample handling, therefore, accurate quantitation requires method-induced artifacts to be minimized.

2.1. Enzymatic Digestion: Trypsin, Lys-C, and Beyond

Trypsin remains the primary protease for bottom-up MAM because its C-terminal lysine- or arginine-containing peptides are efficiently ionized in positive mode electrospray ionization and yield predictable MS/MS spectra that support confident peptide identification [69]. Optimization studies over the past decade converge on a set of practical conclusions. Earlier work demonstrated that complete removal of guanidine prior to digestion shortens the digestion time and minimizes trypsin self-cleavage and non-specific activity [70]. One study showed that performing tryptic mapping at pH 6.0 markedly suppressed asparagine deamidation and aspartate isomerization while still stabilizing succinimide intermediates [55]. This work has been extended to a fully qualified protocol combining pH 6.0 digestion, inclusion of methionine in all buffers, and a comparison of 2 plus 2 hours versus 20 hours incubation, showing that artificial deamidation and oxidation could be largely abolished and that digestion time did not introduce additional artifact under those conditions [57]. A subsequent study reported that quenching with trifluoroacetic acid (TFA) rather than formic acid, combined with two methionine additions, one at denaturation, one after buffer exchange, substantially reduced sample preparation-induced oxidation and improved autosampler stability of oxidation sites [39]. Digestion efficiency has also been addressed directly: a two-step digestion at distinct temperatures reduced missed cleavages while preserving recovery of hydrophobic peptides [40].

However, trypsin is not universally adequate. Tryptic mapping inherently underestimates clipped variants generated at lysine or arginine residues as the clipped peptide collapses into a tryptic fragment, can underreport glycation which blocks tryptic cleavage at modified lysines, and risks isomerization at digestion pH greater than 7.5 [51]. For these reasons, alternative or supplementary enzymes have been adopted: Lys-C is favored for its tolerance of guanidine-HCl, extensive digestion within 30 min in denaturing buffer, and a cleaner peptide fingerprint that benefits NPD. A Lys-C based MAM workflow eliminates desalting entirely and demonstrated consistent PTM quantitation between manual and automated protocols [41]. Glu-C has been used for the heavily glycosylated erythropoietin beta molecule, and studies have exploited chymotrypsin, Asp-N, pepsin, or dual-enzyme strategies for fusion proteins and adeno-associated virus (AAV) capsid proteins where tryptic coverage is insufficient [30,31,49,71]. For middle-down approaches, IdeS combined with reduction or Lys-C is now an established option for subunit-level MAM analysis [42].

2.2. Controlling Method-Induced Modifications

To obtain accurate attribute values to meet regulatory requirements, several practical recommendations are now widely adopted across industry. Methionine is included in denaturing and buffer-exchange steps to scavenge reactive oxygen species and suppress methionine oxidation [43]; phosphate and bicarbonate buffers are avoided as they accelerate asparagine deamidation [56]; reduction and alkylation are performed at pH 7-7.5 to balance reaction efficiency with artifact suppression; digestion is held at pH 7-7.5 for reasonable length to minimize deamidation while ensuring complete cleavage; and the digest is quenched by acidification with TFA together with methionine to prevent post-digestion oxidation in the autosampler [72].

Artificial oxidation, however, is not confined to sample preparation and remains a critical technical challenge across the entire MAM workflow. An investigation traced an abrupt, up to 5-fold increase in apparent methionine oxidation of an Fc-fusion protein reference standard, observed during seven-year trending, to a single bag of microcentrifuge tubes used to collect buffer-exchange samples. This study further showed that ADC oxidation levels depended on the sample preparation procedure, and on-column oxidation, which increased with column age and produced broad, noisy features that disrupted automated peak integration, was substantially reduced by flushing the LC system and column with 20 mM EDTA, implicating metal ions from acid-driven corrosion of the flow path [43]. These observations underscore the need for control strategies to minimize artificial oxidation, including qualification and change control of oxidation-sensitive consumables, system-suitability trending against a well-characterized reference standard, and metal management across the LC flow path and column lifecycle.

2.3. Solubility, Acid-Catalyzed Artifact, and Hydrophobic Peptide Handling

Beyond chemical modification, another source of bias is the differential solubility of hydrophobic tryptic peptides and their modified counterparts. Hydrophobic peptides may precipitate preferentially during acidic stop conditions, producing apparent oxidation that depends strongly on guanidine concentration in the final digest and on the hold time in autosampler [54]. This effect is particularly pronounced for tryptophan dioxidation, for which artifact-driven levels were several-fold higher under unfavorable conditions than the stable values obtained when solubility was maintained. Acid-catalyzed deamidation, often unnoticed in stored digests, can similarly affect the results. Harsh acidification (e.g., 1% TFA) promotes time-dependent deamidation during autosampler storage, whereas milder acidification (e.g., 1% formic acid, 0.1% TFA, or 0.1% difluoroacetic acid) or omitting the acidification step substantially reduces this artifact and improves confidence in deamidation results [54].

2.4. Automation for Reproducibility and Throughput

Manual sample preparation introduces considerable analyst-to-analyst and day-to-day variability, which can make it difficult to achieve the precision expected for a QC release method. Automation has therefore emerged as a pivotal part of MAM. A Hamilton-based, fully walk away peptide mapping procedure that automated a 96-well microdialysis desalting step has been developed and showed PTMs quantitation reproducibility comparable to the manual workflow, validated against forced-degradation data [63]. Another study described a tip-based, fully automated MAM sample preparation workflow [73,74]. A subsequent comparison of centrifugal, gravity-driven, and positive pressure buffer exchange tips showed that fully automated formats reduced variability relative to manual processing [75]. Other groups have implemented automated MAM sample preparation for process development and QC [76]; for example, a fast, automated trypsin digestion procedure using immobilized trypsin on magnetic beads, demonstrated cross-site robustness at four laboratories [62]. Together, these developments point to a clear industry shift toward fully automated, walk-away MAM sample preparation on robotic liquid-handling platforms such as the Hamilton Microlab STAR, Tecan Fluent, and Agilent AssayMAP Bravo, offering improved throughput, reproducibility, and inter-laboratory comparability, as well as tighter integration with LC–MS acquisition. The sample preparation barriers discussed in this section, together with their mechanistic origins and the practical solutions reported, are summarized in Table 3.

3. LC-MS Instrumentation and System Suitability

Once artifacts introduced during sample preparation are controlled, the reliability of MAM depends on the LC–MS system, which must separate closely related peptides reproducibly and detect them with sufficient resolution and mass accuracy. Most published MAM workflows use C18 reversed-phase columns; however, C4 columns have been shown to reduce tailing and carry-over of hydrophobic peptides without compromising sequence coverage [40]. Additionally, column temperature, mobile phase composition, and gradient design must be optimized to retain very hydrophilic peptides such as the deamidation hotspot peptide VSNK.

The mass spectrometers used for MAM must reconcile two competing demands: sufficient resolving power and mass accuracy to confidently assign modifications and differentiate closely related species, and the robustness, ease of operation, and footprint required in a regulated QC environment. The MAM Consortium study confirmed that the industry has converged on HRAM platforms, with linear ion-trap Orbitrap (LIT-Orbitrap), quadrupole-Orbitrap (Q-Orbitrap), and quadrupole-time-of-flight (Q-TOF) as the dominant instrument families [77]. Orbitrap analyzers offer resolving power up to 480,000 (at m/z 200) and sub-ppm to low-ppm mass accuracy with internal or external calibration, respectively. Combined with automated calibration, low maintenance requirements, and compliance-ready data systems, these features have supported their deployment in both R&D and GMP QC environments. Q-TOF instruments trade some resolving power for fast, scan-rate-independent acquisition and, when paired with compliance-ready software, have been increasingly implemented for MAM in QC settings.

Low-resolution instruments offer operational simplicity, a smaller footprint, and lower cost, which can nonetheless fill defined roles in the monitoring phase, where a fixed panel of attributes is quantified and untargeted discovery is not required [35]. A systematic comparison of instrument types quantified this trade-off: a high-resolution instrument achieved a quantitation limit of 0.002%, whereas single-quadrupole instruments in scan mode reached approximately 1%, which should nonetheless be sufficient for many routine MAM applications [52]. Consistent with this, single-quadrupole (QDa) methods have been qualified or validated per ICH for monitoring PTMs in a therapeutic mAb and for site-specific methionine and tryptophan oxidation in the heavy-chain CDRs of two co-formulated mAbs [53]. Another example is ID-MAM, in which high-resolution peptide mapping supported characterization during development and a targeted QDa method was then validated across six laboratories in four countries for identity testing, sequence variant control, and CQA quantitation in commercial release and stability testing of a bispecific antibody [34]; the same platform subsequently replaced a profile-based ion-exchange method in a commercial drug product specification, a change approved by several health authorities [37]. A multi-laboratory study conducted at the FDA illustrates what such a transfer requires. Using a single lot of rituximab subjected to long-term and accelerated stability testing as well as oxidative, pH, and heat stress, the authors quantified glycosylation, deamidation, oxidation, and N- and C-terminal variants on two Orbitrap systems and a benchtop Q-TOF across two sites, benchmarked against orthogonal HILIC-FLD and CEX-UV methods. Full validation was performed on one Orbitrap, and suitability of the other two systems was established by demonstrating comparable results. All three met the system suitability criteria for retention time and fractional peak area, although the mass-error tolerance was raised from 5 to 10 ppm for the Q-TOF, consistent with its specifications [78]. Transfer of a validated MAM between instruments and laboratories is therefore feasible, but it requires an understanding of instrument-specific capabilities and, in a regulated setting, evaluation of any differences for their potential impact on product specifications, in line with ICH Q14. Table 4 summarizes the chromatographic and mass spectrometric barriers discussed above, along with their mechanistic basis and the corresponding solutions.

4. Data Processing and Software

When instruments are matched across sites, results depend on how the data are processed. The data-processing layer is where the regulatory-compliance burden is most concentrated [83]. Several vendor software platforms are now used routinely: Thermo Fisher Scientific’s BioPharma Finder and Chromeleon CDS, Waters’ UNIFI and waters_connect with its Peptide MAM application, SCIEX OS with BioPharmaView, Genedata Expressionist, and the Protein Metrics Byos platform with its Byonic and Byologic modules. The freely available, open-source Skyline application has also been used in research settings, but it is not 21 CFR Part 11-compliant and therefore cannot be used for regulated data processing [9,82,83,84].

Inter-laboratory comparisons show that data-processing choices can contribute to between-site variability. In a three-laboratory study of nivolumab, digests prepared by a single laboratory were distributed to the participants and processed in waters_connect, UNIFI, and Skyline on nominally identical benchtop LC-MS platforms. Agreement across laboratories and software was good for most attributes, with glycoform repeatability between 0.1% and 1.1% RSD, but oxidation results diverged sharply for one peptide, ranging from 3.5% to 16% across the nine instrument and software combinations. The authors attributed this to how each package handled a broad, tailing, noisy peak: automatic integration algorithms differ between platforms, and charge state selection is user adjustable in Skyline and waters_connect but fixed in UNIFI, so different charge states were sometimes summed for the same peptide. The practical implication is that where multiple sites use different software, comparability should be evaluated early in method development, with particular attention to poorly shaped peaks, and methods validated in parallel on each platform where necessary, ideally against a well-characterized reference standard such as NISTmAb [47,82,85]. These software-related limitations, and the advances that address them, are summarized in Table 5.

5. NPD Applications, Barriers and Practical Solutions

New peak detection is the feature that elevates MAM from a quantitative peptide-mapping assay to a purity-indicating method capable of replacing conventional impurity tests. In one study, two host cell proteins, cathepsin L and recombinant human lipoprotein lipase (rhLPL), were spiked into a commercial IgG1 drug product and detected by NPD without any change to the LC–MS parameters used for attribute monitoring. At 1000 ppm, twelve new peaks were flagged and assigned to tryptic peptides of the two proteins; at 100 ppm, eight were still detected, three from cathepsin L and five from rhLPL, while spiking at 10 ppm fell below the detection limit [44]. The same acquisition therefore supports both site-specific attribute quantitation and detection of low-level process-related impurities.

Robust NPD depends on four parameters: the MS intensity threshold (IT), the fold-change detection (FCD) threshold, the mass accuracy tolerance, and the retention time tolerance window. The IT may be absolute, based on signal intensity or peak area, or relative, based on the TIC or the most abundant peak; a relative IT is generally preferred because it adapts to instrument response drift over time. The retention time window is equally decisive, since a peak that drifts outside it is reported as a new peak in one run and as a missing peak in another, so reliable alignment is a prerequisite for interpretable NPD output [48].

The intensity threshold determines whether NPD is informative or unusable, as illustrated by a platform Lys-C MAM developed for an IgG4 mAb and its co-formulations. NPD was performed in Chromeleon at intensity thresholds of 0.01%, 0.1%, and 0.3% of TIC, comparing six preparations of mAb-1 drug substance from two different bioprocesses against a single Process 2 replicate as reference. At 0.01% TIC, 22 to 24 new peaks were flagged in Process 1, but 1 to 3 were also flagged between replicate preparations of the same Process 2 material, a false-positive rate that makes the setting impractical. At 0.3% TIC, no new peaks were detected in either process, so a genuine process difference would have been missed entirely. At 0.1% TIC, the same three new peaks were consistently flagged in Process 1 with none between Process 2 replicates; they were identified as one oxidized peptide and two N-glycosylated peptides, consistent with conventional HIC and HILIC N-glycan profiling [60]. These results show that IT and FCD thresholds must be set empirically for each product and instrument configuration rather than adopted from a platform default. Controlling the conditions that generate run-to-run artifacts, including sample hold time, autosampler stability, and chromatographic carry-over, is a prerequisite for reliable NPD, because any artifact that differs between the test sample and the reference above the detection thresholds will be flagged as a new peak. Even with appropriate thresholds, false positives remain the main obstacle to routine NPD deployment. One study curated and categorized commonly observed false positives, then introduced a known peak list, a pairwise analysis approach, and a qualified NPD system suitability control strategy to address the false positive obstacle of routine NPD deployment. Using sequence-variant co-mixes as a controlled test of detection performance, the authors showed that NPD outperformed conventional control-system methods in detecting an unexpected change relative to the reference [35]. In routine use, performance can be safeguarded by bracketing analytical runs with reference standard injections and by including a spike-recovery check in system suitability [48]. Table 6 consolidates the NPD limitations discussed in this section with the practical solutions reported for each.

6. MAM Applications Across the Product Lifecycle

6.1. Cell-Line Development and Manufacturing Support

MAM is increasingly applied in early-stage development where a comprehensive CQA profile must be generated from very limited material. A nano-flow MAM platform quantified up to 23 attributes, including deamidation, oxidation, succinimide formation, terminal composition, and N-glycosylation, from low-microgram amount of 96-well clone culture material while achieving high sequence coverage [90]. Clones can therefore be selected based on product quality rather than titer alone, derisking downstream development.

The same capacity to monitor multiple attributes in parallel from a single in-process sample makes MAM well suited to QbD-driven process development [91]. In one application, daily samples taken across a 12-day fed-batch cell culture were analyzed against a workbook of 24 CQAs, revealing that most attributes remained stable while the N-glycoform distribution shifted noticeably over the course of the run, information that can be fed back into feeding strategy and harvest timing [44]. A second study demonstrated global harmonization of an automated MAM platform across sites and its capacity to support both upstream and downstream process development [92,93]. A third study applied MAM to daily in-process samples from a development mini-bioreactor and a 2000 L GMP single-use bioreactor across a 17-day process, monitoring deamidation, oxidation, glycosylation, N-terminal cyclization, and C-terminal clipping in parallel and enabling attribute trending at manufacturing scale that would otherwise require several orthogonal assays [46]. Extending this further, fully integrated online MAM systems deliver near-real-time attribute readouts that can be linked directly to process levers within a process analytical technology (PAT) control loop [50,94,95].

6.2. Comparability and Biosimilarity Assessment

Analytical comparability underpins post-change process validation and biosimilar submissions, and MAM suits this task by combining site-specific attribute quantitation with NPD-based detection of unexpected differences in a single analysis. In an assessment of seven adalimumab biosimilars against the reference product, the workflow proceeded in three steps: a discovery phase characterized the most relevant modifications of the reference product and generated the target peptide workbook; targeted monitoring of reference batch-to-batch variability defined statistical intervals from which similarity ranges were established; and biosimilarity was then evaluated against those ranges, with NPD flagging new or modified peaks relative to the reference [96,97]. This three-step structure is the transferable element, since the same logic applies to any post-change comparability assessment: establish the attribute panel, define acceptance ranges from the pre-change material, then test the new material against both the panel and NPD.

7. Regulatory Considerations for Implementing MAM in QC

The regulatory landscape for MAM is evolving rapidly. Several existing frameworks support MAM implementation: ICH Q2(R2) for analytical procedure validation, ICH Q5E for comparability, ICH Q6B for specifications, ICH Q8/Q9/Q10 for pharmaceutical development, quality risk management, and the pharmaceutical quality system, ICH Q12 for lifecycle management, and ICH Q14 for analytical procedure development. In addition, a dedicated USP general chapter for MAM now exists, covering software compliance, qualification, validation, and analytical transfer [98]. Applying the enhanced approach described in ICH Q14, in which an analytical target profile is defined for CQAs and the analytical technology is selected against that profile using prior knowledge and risk assessment, is expected to facilitate MAM development and implementation, particularly when MAM is introduced during the product lifecycle [99]. While no fundamental regulatory barrier to incorporating MAM into an overall control strategy has been identified, successful implementation requires addressing several regulatory and operational considerations, including assay bridging, New Peak Detection (NPD) validation, regional testing requirements, post-approval change management, and the development of regulatory familiarity through early and continued engagement with health authorities. In the longer term, the effort required to implement MAM may be comparable to, or potentially less than, that required for developing and maintaining multiple conventional methods.

A foundational FDA perspective on MAM implementation in cGMP environments identified four key areas for evaluation: risk assessment, method validation, NPD capability and specificity, and comparison against conventional methods [33]. FDA also supports MAM through its Emerging Technology Program, in which sponsors meet the Emerging Technology Team to identify and resolve technical and regulatory challenges before filing [100]. A milestone was reached in 2019 when, after several rounds of Emerging Technology Team review of MAM for control purposes, FDA agreed to a sunset strategy for conventional methods for one applicant: a defined plan for phasing out the conventional methods once MAM was established in the control strategy. Several companies have since established MAM in QC [4,61]. In addition, FDA awarded the U.S. Pharmacopeia a grant to create a MAM knowledge base aimed at enhancing biosimilar and interchangeable biosimilar product development [101].

The EFPIA Manufacturing and Quality Expert Group founded a "MAM as QC tool" working group in 2021, which published a regulatory position paper aiming to support MAM for QC release and stability testing. The group has identified three areas requiring further alignment: software solutions, best practices for NPD threshold definition and validation, and common standards and system suitability criteria enabling comparison across instruments and laboratories [65].

Three scenarios for introducing MAM as a QC tool have been proposed: introduction during development, after first-in-human studies but before registration; introduction in the commercial phase as a lifecycle management activity; and introduction prior to first-in-human studies for release and stability testing in place of conventional methods. All three require risk assessment, method validation, specification setting, and NPD validation. However, only the first two additionally require bridging studies against the conventional methods being replaced [38]. Applicants may also implement MAM gradually, first as an additional method and then as a replacement. The earlier MAM is introduced, the lower the bridging burden [65,85].

Deployment in a regulated environment also requires computer system validation (CSV), because the LC-MS platform is computer-controlled and the control software performs both acquisition and analysis [86,87]. CSV is resource-intensive and is performed against a specific software version, so version changes may trigger re-validation. Sponsors should therefore plan sufficient time for QC-suitable software implementation, including lifecycle and CSV documentation and operating procedures [88]. The MS and LC components must be qualified together as an integrated system, and replacing a component later in the lifecycle, such as substituting an LC stack, is feasible but requires substantial requalification [89]. The regulatory and lifecycle considerations described above are summarized in Table 7 as a reference for sponsors planning QC implementation.

8. Conclusions

MAM provides two analytical capabilities within a single LC–MS experiment: quantitation of a predefined panel of at-tributes at specific amino acid residues, and untargeted detection of species not included in that panel. Conventional assays measure categories of product-related variants indirectly, whereas MAM measures the underlying modifications directly and, through new peak detection, retains sensitivity to unanticipated changes. This combination accounts for the transition of MAM from a characterization tool to a method applied to release and stability testing, although MAM complements rather than replaces orthogonal methods for aggregates and higher-order structure.

Many of the major technical barriers examined in this review now have demonstrated mitigation strategies, although several remain insufficiently harmonized or broadly validated for routine multi-site QC deployment. Method-induced artifacts arising during denaturation, reduction, alkylation, digestion, and quenching are understood mechanistically and can be suppressed through defined control of buffer composition, pH, temperature, incubation time, and scavenger addition. Automated sample preparation has reduced analyst-dependent variability to levels consistent with QC release requirements and has demonstrated cross-site reproducibility. High-resolution accurate-mass instrumentation meets the performance requirements for targeted quantitation and new peak detection, and lower-resolution detectors have been qualified for defined monitoring applications in which the attribute panel is fixed. Compliance-ready software platforms are available for the complete workflow, and inter-laboratory studies have established benchmark values for system suitability, retention-time reproducibility, mass accuracy, and detection thresholds. Applications from clone selection through process development and biosimilar comparability demonstrate that these solutions already support decisions across the product lifecycle, and existing regulatory frameworks accommodate MAM without modification, with no regulatory impediment to incorporating it into a control strategy identified to date.

Readiness nevertheless varies by workflow domain, and technical solvability does not equate to adoption readiness. Artifact control and automated sample preparation are established and widely implemented, as is HRAM instrumentation for both targeted quantitation and NPD. Data processing and software remain at an emerging stage, since reported attribute values still depend measurably on the processing platform and no common convention exists for quantitation denominators, reporting limits, or data exchange. NPD sits between the two: performance has been benchmarked across laboratories and false-positive control strategies have been demonstrated, but threshold setting remains product- and platform-specific and no harmonized validation practice exists. Regulatory implementation is established in principle and emerging in practice, with defined introduction scenarios and accepted filings, but unresolved questions around regional import testing, NPD specification setting, and transfer between instrument platforms. Across all five domains the reported solutions remain distributed across individual studies, vendor platforms, and organization-specific practices, so each implementation continues to require substantial independent development, alongside lifecycle governance, workforce capability, and a defensible operational and economic case. The synthesis presented here is narrative and problem-driven rather than a formal systematic review; some barriers may therefore be under-represented if their solutions have not yet reached the published literature accessible to the authors.

Four areas warrant further investigation. Artificial oxidation remains incompletely controlled, since its sources include consumables and flow-path materials rather than the analytical procedure alone. The contribution of data-processing algorithms to reported attribute levels requires systematic characterization, so that software-dependent differences can be quantified. Shared contaminant libraries and standardized validation approaches for new peak detection would reduce the product-specific empirical optimization currently required. Finally, a system suitability standard that reflects sample preparation performance, in addition to instrument performance, would address a recognized limitation of current practice. The central question for MAM has shifted from whether the method can meet quality control requirements to how efficiently the field can consolidate what has already been demonstrated.