ABSTRACT
The multi-attribute method (MAM) by liquid chromatography-mass spectrometry peptide mapping has the potential to replace multiple conventional HPLC- and capillary electrophoresis-based purity/impurity assays for release and stability testing of protein biopharmaceuticals such as monoclonal antibodies. Prerequisite is the availability of the new peak detection (NPD) functionality to reliably detect new, absent, and changed peptide species that may impair the quality, safety, and efficacy of the drug. Here, we describe the development, qualification, and application of a highly efficient and robust NPD workflow within the Genedata Expressionist® software. The detection thresholds have been rationally designed, and the NPD workflow has been successfully validated according to ICH Q2 guidelines. Individual case studies, including stability testing of drug product and detection of unknown impurities in drug substance, highlight the workflows’ ability to reliably recognize relevant peptide species below 1% relative abundance without reporting any false positive peaks. The application of this NPD workflow signifies a substantial leap forward in the use of MAM as a quality control tool, as it allows identification of true positive peaks at adequate sensitivity in the absence of false positive peaks.
KEYWORDS: Analytical method validation, biopharmaceutical, mass spectrometry, multi-attribute method, new peak detection, quality control
Introduction
Protein biopharmaceutical drugs such as monoclonal antibodies (mAbs) play an important role in the treatment of various life-threatening diseases and account for the majority of biopharmaceutical drug global sales of more than 300 billion USD in 2021.1 During development of protein biopharmaceuticals, product-related variants that may affect biological activity, pharmacokinetics/pharmacodynamics (PK/PD), immunogenicity, or safety are characterized using liquid chromatography coupled to mass spectrometry (LC-MS) peptide mapping.2 These critical quality attributes (CQAs) are monitored and controlled during product development and commercial manufacturing by setting appropriate process parameter ranges, performing in-process testing, and/or testing at release and during stability in a regulated GMP environment. Owing to the structural complexity and heterogeneity of protein biopharmaceuticals, multiple orthogonal LC- and capillary electrophoresis- (CE) based methods are typically used to comprehensively assess product-related variants that are considered as CQAs.3,4 To justify the use of these methods as purity/impurity assays, individual peaks need to be characterized, which is both time-consuming and often demonstrates the limited capability of these methods to discern product-related variants that have relevance to safety and efficacy from those that have not.5
In recent years, the use of the multi-attribute method (MAM) by LC-MS peptide mapping as a quality control tool has received increased attention because it provides the opportunity to replace multiple conventional HPLC- and CE-based purity/impurity assays, while providing improved specificity to monitor and relatively quantify individual product-related variants.5–14 MAM consists of two functionalities that are both required for its application as a purity assay.6 First, by targeted monitoring of quality attributes, protein modifications that have been previously characterized are relatively quantified by leveraging the mass and retention time (RT) information of respective unmodified and modified peptides documented in a MAM peptide library. Secondly, the new peak detection (NPD) functionality involves the comparison of the sample with a well-characterized, product-specific reference standard to identify new, absent, and changed features. The NPD functionality is considered a key element of the MAM technology to detect any previously unknown modifications or changed features of the protein that may occur as a result of deviations during manufacture and storage of the drug substance (DS) or drug product (DP). For the robust performance of NPD, it is crucial to reliably differentiate true positive from false positive peaks and to achieve sufficient sensitivity to detect species that may impair the safety or efficacy of the drug, i.e., to avoid false negative peaks. Therefore, a robust and sensitive NPD workflow is an essential requirement to replace multiple conventional methods with MAM while mitigating risks to business and patients.7,15
Due to the two-dimensional separation of analytes in an LC-MS experiment, i.e., by retention time and mass, the resulting MS data set is significantly more complex than that of conventional LC- or CE-based methods and may contain more than 10,000 potential peptide signals. Therefore, several mitigation strategies to reduce the number of false positive peaks, while retaining all true positive peaks for NPD have been proposed. Random signal fluctuations were addressed by reference and sample replicate strategies16,17 or more sophisticated statistical approaches to avoid replicate injections.18 Furthermore, the collection of signals that are not considered relevant for reporting in a known-peak list (KPL), or the definition appropriate system-suitability test (SST) criteria have been shown to be successful.17
Several NPD data analysis workflows have been proposed to reveal new peaks, such as unintended sequence variants during clone selection or characterization by LC-MS/MS.16,19–21 In 2017, Griaud et al. applied a differential analysis of all peptide mapping signals across multiple samples to reveal sequence variants consistently present only in batches of an intended biopharmaceutical copy product in the range of 0.5–7.2% and therefore absent from the originator samples.21 This MS workflow was sufficiently sensitive to reveal relevant sequence variants, whereas conventional methods lacked the required resolution and sensitivity. In 2023, Niu et al. published an NPD workflow to reveal sequence variants detected in bioprocess samples with comparable performance in the range of 0.8–38.7%.20 Cao et al. were able to reliably detect new peaks related to another mAb spiked into the sample at 2% relative abundance with no false positive peaks in the majority (20 of 23) of experiments.22 Even lower detection thresholds down to 0.5% (mAb) or even 0.01% (host-cell protein) were achieved with a targeted approach if the signals expected from the spike-in proteins were predefined in a peptide library or the workflow was optimized for the specific sample set.16,23
Besides these strategies, the increase of detection thresholds such as the intensity threshold (IT) and fold-change detection (FCD) threshold has been shown to significantly reduce the number of false positive peaks, albeit at the cost of reduced sensitivity. The IT is typically expressed as percentage relative to the most intense signal base peak chromatogram (BPC) or total ion chromatogram (TIC). IT values in the range of 0.10−1.0% BPC10,15,17 and 0.01−1.0% TIC11,24 have been reported. The minimal FCD threshold has been determined based on the inherent signal variability across multiple injections of the same sample15 and is typically set in the range of 5- to 10-fold,10,11,15,16,20,24,25 although adequate performance with zero false positives has been also shown for FCD thresholds below 2-fold.18 Thresholds are mostly defined empirically and only a limited number of strategies and theoretical considerations regarding the definition of thresholds have been published.17,25 Other relevant NPD parameters are mass and retention time tolerance windows, which are typically defined in the range of 5 to 20 ppm and 0.2 to 0.8 min, respectively.10,15,16,20,24 To exclude non-peptide species, only signals with charge greater than one and more than one isotope are considered.10 The performance of the NPD workflow has been evaluated by calculation of detection and quantification limits or determination of the number of false positives and false negatives for a mAb digest spiked with heavy isotope labeled synthetic peptides.11,18,24 Under optimized conditions, all 15 spiked peptides were accurately detected without any false positive signals at spiking levels as low as 0.2% relative to the half-antibody.18
Here, we describe the development of an NPD workflow using the MS software Genedata Expressionist that combines multiple strategies to detect true new peaks with superior sensitivity while avoiding the detection of false positive peaks. The presented NPD approach integrates previously published concepts to reduce false positives in a single, highly automated workflow and addresses key needs in terms of analysis speed, robustness, user-friendliness as well as data visualization and interpretation tools in a commercially available, MS instrument vendor-agnostic data analysis software.10 Activities and strategies to reduce the number of false positive peaks have been complemented with newly developed automatic artifact detection functionalities that facilitate the elimination of false positives that arise from instrument noise, polymeric species, and in-source adduct formation. Strategies and considerations to define detection thresholds are presented before their validation according to ICH Q2 guidelines. Finally, the applicability of the validated workflow for release and stability testing of protein biopharmaceutical drugs as well as the reliable detection of unknown impurities is demonstrated.
Results
Development of an integrated MAM workflow for the robust detection of true new peaks
Several NPD workflow activities were developed and applied in a sequential manner to reliably differentiate relevant peaks, i.e., those derived from the product and its variants, from all other peaks that are considered as non-relevant and would translate into the reporting of false positive peaks if not removed during data analysis (Figure 1, Table 1). True positive peaks could be derived from unmodified/modified peptides of the product or process-related impurities such as host-cell proteins, whereas false positive peaks may originate from the DS/DP matrix such as excipients or may be a result of instrument noise, issues during sample preparation, in-source degradation of peptides or metal adduct formation. To address chemical noise, adduct formation and presence of polymeric excipients (e.g., polysorbates), which are a common source of false positive peaks, an automatic artifact detection capability that leverages the known behavior of these species in an LC-MS experiment for their identification has been developed.
Figure 1.
Sequence of activities within the developed NPD workflow to efficiently flag false positive peaks for subsequent exclusion in downstream analysis.
Table 1.
Individual activities of the developed NPD workflow within Genedata Expressionist that are performed in a sequential manner to reliably detect true positive peaks while removing all false positive peaks.
| NPD workflow steps | Description |
|---|---|
| Experimental design | Preliminary data preparation steps include the following actions:
|
| Automatic artifact detection | Rule-based flagging of artifact species including instrument noise, polymers, and metal adducts:
|
| Intensity and fold-change thresholding | Thresholding activities are used to discard individual features that do not meet the defined intensity and fold-change requirements and therefore are not considered relevant for further downstream profiling. |
| Statistical evaluation | Depending on the number of replicates, the following statistical approaches may be applied to assess the significance of changed species:
|
| KPL search | Known species are annotated and flagged through database searching against a project-specific KPL using user-defined mass accuracy and RT tolerance windows. |
| Advanced filtering of NPD hits | NPD hits are discarded if the fold-change relative to the reference file(s) is insignificant or if the species exhibit at least one of the following characteristics: artifact flag, present only as 1+ species, charge states ≥ 8+. |
| Data evaluation and replicate assessment | Manual review and verification of NPD results include the following steps:
|
| Updating of the KPL | New and changed species may be included in the KPL upon individual risk assessment and identity confirmation in follow-up MS/MS analyses to discard these uncritical species in future NPD analyses of the same molecule. |
For the application of MAM in a quality control (QC) setting and to leverage prior knowledge from product characterization for new peak detection, the newly developed NPD functionality was integrated into an existing targeted attribute monitoring workflow within the Genedata Expressionist software (data not shown). During product development, the MAM peptide library is expected to evolve and ideally will contain all peptides that are related to product-related variants observed at release and during stability above a certain abundance threshold, latest at the time of process performance qualification (PPQ).5 Considering that those product-related variants have been characterized in terms of their impact on product quality, safety, and efficacy, they should not be reported as new peaks. If considered relevant for the overall control strategy of the product, the relative abundance of those attributes could be tested against predefined limits during release and stability or monitored to verify process consistency.
Prior to the application of NPD, a product-specific MAM peptide library and KPL were created. The MAM peptide library was built by analyzing samples of mAb1, that were subjected to thermal, oxidative, glycation, pH, and light stress conditions, in an LC-MS/MS peptide mapping experiment. Relevant unmodified and modified peptides of mAb1 were identified by MS and their amino acid sequence was confirmed by MS/MS prior to documentation of their mass, most abundant charge state(s), and retention time information. Overall, the resulting MAM peptide library comprised 172 entries accounting for a total of 28 unmodified mAb1 peptides and 125 unique degradation products or post-translationally modified peptides such as deamidated, isomerized, oxidized, clipped, glycated, and glycosylated species. For selected modifications (e.g., N-glycosylation), the mass and RT information for multiple charge states were documented in the MAM peptide library.
A KPL was built by comparison of replicate injections of the product-specific reference standard (defined as sample) against the same standard (defined as reference) using the NPD workflow as described above. The IT was set slightly lower as expected for the routine performance of NPD to account for intensity fluctuations and to capture all relevant species. As both sample and reference are identical, it can be assumed that all new peaks detected by the NPD workflow are false positive peaks, and therefore respective peaks were documented in the KPL with their m/z, charge state, and retention time information. After four rounds of LC-MS analysis of nine individual preparations of the product- specific reference standard on two comparable LC-MS setups, no new peaks were identified, indicating that the KPL contained the most commonly observed artifacts. Finally, the KPL of mAb1 comprised 40 unique entries accounting for a total of 28 underalkylated peptide species, 3 multimeric gas-phase complexes, and 9 low-intensity contaminants of unknown origin that exhibited arbitrary intensity fluctuations close to the defined intensity threshold. While the reporting of underalkylated peptides as new and changed species may to some extent be addressed by fine-tuning the alkylation conditions during sample preparation, multimeric gas-phase complexes are formed during electrospray ionization and thus represent technology-inherent artifacts. The low-intensity contaminants were included in the mAb1 KPL as a precautionary measure even though respective signals may be efficiently identified as false positives during NPD analysis by means of statistical evaluation and sample replicate assessment. Although this KPL was used for routine testing in the context of our study, it should be noted that the developed NPD workflow allows to add further species that may be observed during product development.
Selection of detection thresholds to ensure robust and sensitive NPD performance
The selection of appropriate detection thresholds (i.e., the IT and FCD threshold) is crucial for the robust performance of the NPD workflow. The lower the IT the more signals have to be assessed by the NPD workflow, therefore increasing the risk of reporting false positives. On the other hand, a very high IT may render low-abundant modifications on product-related peptides undetectable.
Considering the potential impact of product-related variants on quality, safety and efficacy, a reporting level (RL) of 0.9% was selected arbitrarily. Subsequently, the IT was defined based on the MS intensity distribution of product-related peptides to ensure detection of modifications above the RL for > 95% of all 28 unmodified, primary mAb1 peptides contained in the MAM peptide library (Figure 2). Expressed as relative intensity, the IT was set at 0.05% of the BPC or 0.13% of the median intensity of primary peptides.
Figure 2.
MS intensity distribution of mAb1 product-specific unmodified peptides that are contained in the MAM peptide library (primary peptides) multiplied by the reporting level (RL) of 0.9%. Other peptides that are derived from the digestion of the product with trypsin/Lys-C were not included as they are either too small/hydrophilic to be retained on the RP column or are not expected to bear any modification site. The NPD intensity threshold (IT) is shown as red line. Assuming comparable ionization efficiencies of unmodified and modified peptides, the intensity distribution plot allows to rationally define the IT to enable detection of protein modifications above the RL. For modifications on the lowest ionizing peptide (peptide #1) the selected IT results in an effective RL of 1.6%, whereas for all other peptides the effective RL is below 0.9%. For modifications on the highest ionizing peptide the effective RL is 0.05% and therefore identical to the IT, expressed in % relative to the BPC.
The minimal FCD threshold to enable robust NPD performance was estimated based on the inherent variation of MS intensity observed within a single LC-MS sequence.15 The variation was expressed as the fold-change in MS intensity of product-related peptides obtained from replicate injections of individual preparations of the product-specific reference standard. For each peptide and injection, the worst-case fold-change was calculated by considering the lowest intensity observed across all injections as denominator. Finally, the average and standard deviation (SD) of the fold-change from 10 injections within the same analytical sequence was calculated. Considering all product-related peptides in the MAM peptide library, the calculated worst-case fold-change (average +3 SD) was in the range of 1.1 to 2.0. The minimal FCD threshold for further NPD analysis of mAb1 was set to exceed the worst-case fold-change of the most variable peptide and defined as 3.0.
Validation of the NPD workflow and analytical control strategy
The RL, IT, and FCD threshold were qualified by spiking mAb1 digest with synthetic peptides, i.e., the Pierce™ Retention Time Calibration (PRTC) standard, similar to the approach described by others (see Table S1).11,18,24 The equimolar PRTC calibration mixture contains 15 heavy isotope-labeled peptides with varying hydrophobicity covering the retention time range of the LC method, and therefore are considered representative of potential new species the NPD workflow should detect during routine use. As the MAM NPD workflow is considered a limit test, specificity as well as detection and quantitation limits (DL/QL) were validated according to ICH Q2 guidelines.
Specificity was demonstrated by absence of interference. Five replicate injections of mAb1 (defined as sample) were compared against another injection of mAb1 digest (defined as reference) within the same analytical sequence using the NPD workflow and KPL as described above. No new peak was detected in any of the five replicate injections, confirming specificity through absence of interference from matrix components, impurities, and product-related peptide species.
For the validation of DL and QL, PRTC peptides were spiked at levels of 0.1, 0.3, 0.4, 0.5, and 1.0 pmol into mAb1 digests. Considering a column load of 4 μg mAb1 (i.e., equivalent to approximately 56 pmol of light and heavy chain peptides), the selected PRTC spiking amounts translate into 0.2, 0.5, 0.7, 0.9, and 1.8% relative abundance vs product-related peptide species. Each spiking level was injected six times and the resulting MS intensities of the PRTC peptides were plotted against the spiking level in pmol. For each of the 15 peptides, a linear response was observed across the entire tested range (R2 ≥0.99, see Figure S1). DL and QL were calculated based on the slope and the standard error of the y-intercept of the regression line. Considering all 15 PRTC peptides, the DL was estimated to be in the range of 0.013–0.032 pmol (i.e., corresponding to approximately 0.02–0.06% relative abundance). Accordingly, the QL was found to be in the range of 0.040–0.097 pmol (i.e., equivalent to approximately 0.07–0.18% relative abundance). As the determined DL and QL are below the defined IT and RL (see Table S2), it can be assumed that product-related variants present above the RL can be reliably detected and quantified. Consequently, both the IT and RL are considered qualified. This was confirmed by six replicate injections of mAb1 digest spiked with PRTC peptides at the RL (0.5 pmol spiking level corresponding to 0.9% relative abundance) and performance of NPD using the unspiked mAb1 digests as reference. In each of the replicate injections, all 15 PRTC peptides were consistently detected without introducing false positives (see Table S3). At the highest tested spiking level, however, two false positive peaks were reported. These additional peaks included the sodium adduct of PRTC 10 as well as a heterodimer involving PRTC 7 and a mAb1 peptide. The sodium adduct was not flagged during automatic artifact detection due to faulty isotope clustering of the parent feature. Manual editing of the respective PRTC 10 isotope cluster resulted in the correct flagging of the metal adduct during respective artifact detection activities. The reporting of the PRTC 7-based heterodimer could not be prevented on workflow level in the present study. The current NPD workflow does not allow for the systematic flagging of artificial gas-phase complexes. However, the risk of repetitive reporting of this artifact in future NPD analyses of mAb1 may be substantially mitigated by documenting the identified and confirmed heterodimeric species in the project-specific KPL.
The performance of the fold-change detection was evaluated using the same data set that was used for the qualification of the IT. The mAb1 digests spiked with PRTC peptides at a level of 0.1 pmol were defined as references, while the other spiking levels of 0.3, 0.4, 0.5, and 1.0 pmol were used as samples. Given the experimental design, each sample is expected to feature 15 increased PRTC-related peptide species that are present at fold-change ratios of approximately 3, 4, 5, and 10 depending on the spiking level. Accordingly, NPD analysis revealed the expected increase of all PRTC-related species for each measured mAb1 sample and spiking level. The average calculated fold-change ratios from six replicate injections generally corresponded well with the expected values, except for PRTC 7, for which significantly higher fold-change ratios were obtained (see Table S4). The root cause for the aberrant PRTC 7 fold-change ratios was found to be related to data preprocessing of the reference files. That is, the PRTC 7 peptide almost co-elutes with a mAb1 species with isobaric isotope masses. These isobaric signals may negatively impact data preprocessing outcomes, resulting in a slight underestimation of the PRTC 7 intensity at the 0.1 pmol spiking level. Consequently, a minor absolute change in the reference signal intensity translates into significant differences in the calculated fold-change ratios for PRTC 7. Given the acceptable accuracy (recovery, in the range of 103% to 146%) and precision (%CV ≤ 25%) of the fold-change detection for all other PRTC peptides, a FCD threshold of 3-fold or higher is considered qualified.
Finally, adequate specificity and sensitivity during routine use will be confirmed by performance of an SST using the NPD workflow in conjunction with a negative and positive control (see Table 2). Comparison of bracketing injections of the product-specific reference standard serves as negative control, whereas the comparison of the reference standard with the same reference standard spiked with PRTC peptides at the RL (0.5 pmol spiking level) serves as positive control.5
Table 2.
NPD SST and respective requirements.
| SST | Approach | Requirement |
|---|---|---|
| Negative control | Comparison of bracketing injections of the project specific reference standard. | No new, absent, or changed peak |
| Positive control | Comparison of the reference standard spiked with 15 synthetic peptides at the RL with the unspiked reference standard. | All 15 peptides are identified |
The overall approach to rationally design, validate, and confirm NPD-related detection thresholds for the routine use is summarized below:
Define preliminary IT based on intensity distribution of product-related peptides, desired RL, and peptide coverage. Estimate minimal FCD threshold based on intensity variation of product-related peptides.
Perform a linearity experiment with PRTC peptides spiked into product digest slightly above, at, and below RL.
Calculate DL/QL for PRTC peptides based on linearity data. Show that DL is below IT and RL. Show reproducible recovery of all 15 PRTC peptides at the RL.
Calculate fold-change using lowest spiking level as reference and other spiking levels with intensities above IT as sample. Show adequate recovery and precision of determined fold-change ratios.
Set SST based on PRTC spike recovery at qualified RL for continuous method performance qualification.
NPD use-cases
The efficiency and robustness of the validated NPD workflow is demonstrated through two relevant use-cases, which were selected to demonstrate its performance in a QC setting.
Use-case 1 – stability testing
In this use-case, mAb1 DP stability samples were subjected to MAM targeted attribute monitoring and NPD analysis. Samples stored under long-term (24 and 36 months at 2–8 °C) and accelerated (3, 6, and 24 months at 25 °C) conditions were selected and analyzed in triplicate. The IT and FCD thresholds were set to 0.05% of the BPC and 3-fold, respectively. In addition, the KPL, which was established as described above, was used. Targeted attribute monitoring revealed that, following storage for up to 36 months at 2–8°C, product quality attributes such as N-terminal pyroglutamate, asparagine deamidation, iso-aspartate formation, as well as methionine and tryptophan oxidation were only slightly increased, indicating that the product is stable at the intended long-term storage condition (see Figure 3). On the contrary, following prolonged storage at 25°C, the relative abundance of the above-mentioned quality attributes increased significantly and additional degradation pathways, such as deacetylation of a FA2 glycan structure, were triggered. Other attributes such as high-mannose species remained stable under both conditions, as expected. In line with the results from targeted attribute monitoring, NPD revealed no new or changed peaks in the samples stored under long-term conditions, while changed peaks were detected already after 3 months at 25°C and the number of changed peaks increased with increasing storage duration (see Figure 4).
Figure 3.
Targeted attribute monitoring using mAb1 stability samples stored for up to 36 months at 2–8°C and 24 months at 25°C, respectively. The relative abundance of selected product quality attributes (PQAs) that are defined by the MAM peptide library is plotted against the storage time in months.
Figure 4.
NPD analysis using mAb1 stability samples stored for up to 36 months at 2–8°C and 24 months at 25°C, respectively. Each sample was prepared and injected in triplicate. Results for each replicate (R) are shown. (A). Shows the number of false positive peaks (B). Shows the number of true positive peaks after performance of NPD. In each inset, the number of peaks without the application of a KPL, with the KPL, and with the KPL and consideration of a sample replicate strategy is shown.
The combined use of the developed automatic artifact detection, statistics, KPL-based annotation of known species as well as IT and FCD thresholding was observed to facilitate the efficient filtering of true positive peaks (see Figure 5). Without KPL applied, up to nine false positive peaks (all related to underalkylation) were detected in individual injections. By leveraging the KPL information, the number of false positive peaks was reduced to zero in all 15 injections. No replicates had to be considered to further reduce the number of false positive peaks (e.g., due to random appearance of new peaks in any of the replicate injections). On the other hand, the number of true positive peaks was not affected by the use of a KPL (see Figure 4).
Figure 5.
NPD workflow performance to filter true positive peaks as exemplified by the analysis of stability samples stored for 24 months at 25°C. Initially 17,191 groups (peptide candidate species or peaks) were identified in the complete dataset comprising three sample and three reference replicates, respectively. Applying the developed workflow activities, artifacts were flagged, and false positive peaks were subtracted from the total number of peaks, as indicated by the horizontal arrows. The remaining number of peaks after each subset of activities is shown in bold with 16 true positive peaks being finally reported. The overall analysis run-time for the parallel analysis of 18 samples was approximately 20 s.
Overall, this example highlights the efficiency of the developed MAM workflow to monitor multiple individual product quality attributes at the site-specific level and to detect major changes in product quality by NPD. No false positives were detected that may lead to an investigation or even unnecessary batch rejection.
Use-case 2 - impurity detection
In this use-case, the capability of the NPD workflow to detect unknown impurities at or even below the validated RL was tested. Therefore, mAb1 DS was spiked with another IgG2 antibody (mAb2) at levels below and slightly above the validated RL, i.e., at 0.1, 0.5, 1.0 and 2.0% (mol/mol). The mixture was subjected to the MAM sample preparation procedure in triplicate and analyzed by LC-MS. For NPD the mAb1/mAb2 digest was treated as sample, while a mAb1 digest without addition of mAb2 was used as reference. Considering that both mAb1 and mAb2 belong to the same IgG isotype, but bear lambda and kappa light chains, respectively, up to 16 light chain peptides and up to 10 heavy chain peptides derived from mAb2 would be expected as new peaks. Excluded are very small and/or hydrophilic peptides that are expected to elute in the void volume of the RP column.
The NPD workflow was performed using the established KPL, the mAb1-specific MAM peptide library, and the validated IT thresholds and FCD settings as described above (i.e., IT of 0.05% BPC and FCD threshold of 3-fold). The samples of each spiking level were analyzed independently against the defined reference. The results are summarized in Table 3. At all spiking levels, including the lowest level of 0.1%, mAb2-specific peptides were reliably identified as new peaks without a preconceived notion of their identity. No false positive signals were identified across all spiking levels. The total number of changed and new features in this dataset was 60, four species were excluded based on charge state characteristics, 29 were considered true positives, six underalkylated peptides were annotated by the KPL, and 21 species were flagged as adducts with the newly developed artifact detection functionalities.
Table 3.
NPD results from the analysis of mAb1 samples spiked with increasing amounts of mAb2.
| mAb2 spiking level | mAb2 light chain peptides | mAb2 heavy chain peptides | False positive peaks |
|---|---|---|---|
| [%mol/mol] | expected/detected | expected/detected | |
| 0.1 | 16/1 | 10/0 | 0 |
| 0.5 | 16/11 | 10/7 | 0 |
| 1.0 | 16/13 | 10/9 | 0 |
| 2.0 | 16/181 | 10/112 | 0 |
1two additional, non-expected peaks were identified, which could be assigned to mAb2 light chain-specific peptides bearing an isomerized aspartate residue.
2one additional, non-expected peak was identified, which could be assigned to a mAb2 heavy-specific peptide with multiple missed cleavages.
Furthermore, two modified mAb2 light chain peptides and one heavy chain peptide were additionally detected at the highest spiking level of 2.0%. While the heavy chain peptide was assigned to a species carrying multiple missed cleavages, the two light chain peptides corresponded to the fully cleaved and a corresponding missed cleaved peptide carrying an iso-aspartate instead of the expected aspartate residue. Based on the MS intensity of the unmodified and modified peptide, the relative abundance of the iso-aspartate modification in the mAb2 light chain was estimated at approx. 12.5%, which aligns very well with the level determined during mAb2 DS characterization (11.8%), which was performed independently from this study using another LC-MS setup.
Overall, the results confirm the specificity and sensitivity of the NPD workflow to reliably detect unknown impurities at levels as low as 0.1% relative abundance. In addition, the data set nicely demonstrates the opportunity to apply MAM including NPD as an identity test for DS release, as the inadvertent contamination of mAb1 DS with mAb2 DS could be consistently detected at patient-relevant levels.
Discussion
Here, we summarize the development, qualification, and application of a MAM NPD workflow using the commercially available MS software Genedata Expressionist. The newly developed NPD workflow combines previously described, separate strategies to reduce the number of false positive peaks, such as intensity and fold-change thresholding,10,11,15–18,20,24,25 consideration of sample or reference replicates,16,17 application of a KPL and peptide library,16,17 and use of statistical approaches.18 It also leverages the innovative automatic identification and removal of known artifacts, including instrument noise, polymer species, and in-source artifacts, to increase efficiency. To our knowledge, this artifact-removal strategy has not been described previously. In principle, more than one NPD workflow activity could address an individual false positive peak. Therefore, the high degree of built-in redundancy provides a very robust and reliable workflow to detect true positive peaks in the absence of any false positive peaks. Furthermore, the developed NPD workflow allows efficient assessment of multiple samples in parallel within a single NPD analysis. The overall analysis run-time for the parallel analysis of 18 samples was approximately 20 seconds in use-case 1, where other software solutions rely on computationally expensive and lengthy pairwise comparisons. When taking into account the time required for an operator to review the NPD assignments made by the automated workflow, the total duration for detecting, reviewing, and reporting new peaks in up to approximately 20 samples is under 30 minutes. This approach addresses the issue of analysis time, which has been identified as a limitation in current NPD workflows.10,17
As performance of NPD in conjunction with targeted attribute monitoring may in the future replace multiple conventional purity/impurity methods in QC laboratories,7,15 the detection thresholds should address potential risks to patients, i.e., the presence of any newly introduced impurities at relevant levels. Therefore, we rationally selected the IT considering the MS intensity distribution of product-related peptides and a RL of 0.9%. Expressed as relative intensity, the IT used in this study is defined as 0.05% of the BPC, which is at the lower end of sensitivity reported for other published NPD workflows.10,11,15–18,24,25 The RL defines the lowest relative abundance a product-related variant will be detected and reported as new peak. Due to the different ionization efficiencies of product-related peptides, the effective RL for modifications on low intensity peptides will be higher compared to the most intense peptide for a given IT. In previous publications, the IT has been solely defined based on the most intense signal (BPC or TIC), which does not account for the different ionization efficiencies of product-related peptides and hence does not allow assessment of the risk to patients. On the contrary, the selection of the IT presented here provides a risk-based approach, which may support the justification of using NPD as purity assay in replacement of conventional methods. Therefore, addressing the request of Rogstad et al. that “MAM capacity and performance should be evaluated specifically in the context of the CQAs of the candidate protein product and its overall control strategy.”7 Irrespective of these considerations, the presented NPD workflow is designed to detect and identify any unexpected peptides or molecular species present, whether they are modifications of expected peptides, non-product related impurities or contaminants as exemplified by the validation experiments and use-case 2.
Prior to full validation of the NPD workflow for commercial release and stability testing, it is expected that NPD will be used in a development laboratory, which may apply different LC-MS instrumentation. Therefore, the IT has to be defined as relative intensity to account for differences in absolute intensity values generated by different mass spectrometers. Adequate sensitivity of the LC-MS instrumentation used in the commercial QC laboratory will be confirmed by determining the DL that needs to be below the set RL and resulting IT, as discussed above. Nevertheless, a lower IT may be applied during development to identify new peaks that could be characterized and subsequently documented in the KPL or MAM peptide library for knowledge preservation.
The FCD threshold has been set at 3-fold considering the inherent variability of the MS signal. Albeit setting lower FCD thresholds without increasing the risk for false positive signals is possible by using statistics,18 the FCD threshold is considered less critical in terms of patient risk compared to the IT. Other than the IT, which is key to detect new peaks at relevant levels, the FCD threshold will only address changed signals, i.e., species that are present in the product-specific reference standard and thus have been exposed to patients already. In case any of these species are considered highly critical for ensuring the quality, safety, and/or efficacy of the drug, the sponsor may decide to include them in the MAM peptide library and subject them to targeted monitoring applying patient-relevant limits.
The RL, IT, and FCD threshold have been qualified, and the overall workflow has been validated as limit test according to ICH Q2 guidelines by spiking DS with commercially available heavy isotope-labeled peptides. The determined DL and QL were well below the predefined RL and align well with results published by others using a similar experimental design and LC-MS setup.24 The SST and acceptance criteria were defined and confirmed using the validated workflow to ensure the appropriate sensitivity and selectivity throughout the lifecycle of the method in line with ICH Q14 principles. Therefore, the definition of a RL allows for the first time to assess the adequacy of the detection limits as determined during method validation and to justify system-suitability-test criteria for continued method performance verification.
The performance of the fully validated workflow has been demonstrated considering two use-cases that were designed based on the potential application of MAM in a QC lab. The first use-case demonstrates the stability indicating capability of MAM targeted attribute monitoring as well as NPD, whereas the second use-case showcases the capability of the NPD workflow to reliably detect unknown impurities that may have contaminated the product. In both use-cases, relevant new peaks were identified in the absence of any false negative peaks. The majority ( > 99.9%) of non-relevant peaks could be excluded by the automatic artifact detection, product-centric IT and FCD thresholding, as well as statistical evaluation. In this context it should be noted that in use-case 2 the newly developed automatic artifact detection capabilities were essential to reduce the number of false positives to zero. Without automatic artifact detection, a total of 21 adducts would have passed thresholding, statistics, and replicate assessment criteria, and therefore the number of false positive peaks would be significantly increased. Adducts of new species display similar characteristics as true positive species and thus cannot be eliminated by means of thresholding, statistics, and/or replicate assessment. This may become particularly relevant for product-related impurities that are absent from the reference material, but are present in the sample. In such a scenario, NPD analysis would report adducts of the modified, product-related peptide as new peaks, thus triggering in-depth investigation/deviation and/or even unnecessary batch rejection.
Remaining non-relevant peaks introduced by the sample preparation procedure or other random variation could be addressed by a KPL and sample replicate strategy. The developed NPD workflow allows knowledge obtained during product development, as documented in the MAM peptide library and KPL, to be leveraged to reduce the number of false positive peaks. While the MAM peptide library contains information related to product-specific unmodified and modified peptides, the KPL can be used to document the RT and m/z values of any other peaks that are irrelevant for the quality, safety and efficacy of the product. In this context, it is expected that the peptide library will be built by targeted LC-MS/MS characterization of the product and its variants, while the KPL will be populated considering predicted or observed false positives from development or validation data.24 A potential challenge with this approach is that RT information will vary from one LC-MS to another LC-MS setup, and hence a KPL established in a development environment may not be applicable for a commercial QC laboratory although it is essential for adequate NPD performance. Therefore, the KPL we used was built by replicate analysis of the product-specific reference standard – an approach that could be easily integrated into the method transfer protocol from development to commercial QC and executed within a short period of time.
In the second use-case, impurities were detected at levels as low as 0.1% relative to the active pharmaceutical ingredient without any false positive peaks, which is considered an appropriate level of sensitivity in the context protein biopharmaceuticals. No peptide library of the impurities as reported by others16 was required to achieve the reported sensitivity. In other published studies where NPD was applied to relevant development samples, either a higher IT was used to achieve zero false positive peaks11 or the analysis was limited to pre-defined modifications such a specific methionine oxidation sites.26 Comparable and even lower sensitivity in the range of 0.1% to 0.01% has been reported for samples spiked with HCP impurities, but the number of false positive peaks remains unclear, or the workflow had to be optimized for the specific sample set.23 Notably, our NPD workflow was capable of detecting product variants of the antibody that was spiked as an impurity, indicating the potential use of NPD for product characterization, such as the streamlined identification of product-related variants in forced-degradation studies.
Overall, we believe that this study demonstrates for the first time the application of a fully validated NPD workflow in a QC-like setting with relevant performance in terms of selectivity and sensitivity to ensure patient safety and minimize business risks.
Materials and methods
Materials and chemicals
Experimental work was performed on two different IgG2 monoclonal antibodies (mAb1 and mAb2) that were produced in-house. The heavy chains of mAb1 and mAb2 were paired with a lambda- and kappa-type light chain, respectively.
Dithiothreitol (DTT, # D9163), iodoacetamide (IAM, #I1149-5 G), and L-histidine (His, #H8000) were purchased from Sigma. LC-MS grade acetonitrile (ACN, #1.00029.1000) and 30% hydrochloric acid (HCl, #1.00318.0500) were obtained from Merck. The Pierce peptide retention time calibration (PRTC, #88320) mixture, sequencing grade trifluoroacetic acid (TFA, #28904), 8 M guanidine hydrochloride (Gnd-HCl, #24115), and 1 M tris-hydrochloride (Tris-HCl, #1000291000), pH 8.0, were ordered from Thermo Fisher Scientific. Calcium chloride dihydrate (CaCl2, # C1099-500 GM) was purchased from VWR. MS grade modified porcine trypsin (#V5280) and MS grade lysyl-endopeptidase C (LysC, #125–05061) were obtained from Promega and Fujifilm Wako, respectively.
Combined LysC and trypsin digestion
The sample preparation for LC-MS and LC-MS/MS peptide mapping analyses was based on a combined LysC and trypsin digestion protocol and was performed in an automated fashion using a Hamilton STARlet liquid handling system in combination with a Biometra TRobot II PCR thermal cycler for sample heating.
Briefly, the concentration of each protein sample was normalized to 2.5 mg/mL. A 19 µg protein sample was reduced for 1 h at 37°C using 5 mM DTT under denaturing and slightly alkaline solution conditions (4.7 M Gnd-HCl and 38 mM Tris-HCl, pH 8.0). Carbamidomethylation of reduced cysteine residues was accomplished in the presence of 9 mM IAM and by incubating the sample solution in the dark for 1 h at 25°C. Following quenching of the alkylation reaction with DTT, approximately 7 µg of the reduced and alkylated protein was diluted 8.5-fold in 25 mM His-HCl buffer, pH 6.0. The sample was subsequently subjected to enzymatic proteolysis at 37°C for 20 h using a combination of LysC and trypsin, each at a 1:50 (w/w) enzyme-to-protein ratio. The protein digest was acidified with TFA to a final concentration of 0.1% (v/v) and the sample was stored at ≤ −60°C until LC-MS analysis.
Individual samples from forced degradation studies were prepared similarly to the method described above. The sequential order of sample preparation steps remained consistent, with increased protein quantity and reagent concentrations to allow for higher column loads during subsequent LC-MS/MS peptide mapping analyses for in-depth protein characterization.
LC-MS/MS data acquisition
In-depth characterization to elucidate the protein main degradation pathways was performed on stressed material and respective control samples using a Thermo Vanquish UHPLC system coupled to a high-resolution Thermo Orbitrap Fusion Lumos mass spectrometer with a heated electrospray ionization source. The UHPLC system was equipped with a Waters Acquity UPLC Peptide HSS T3 column (100 Å pore size, 1.8 µm particle size, and 2.1 × 150 mm column dimensions). Approximately 22 µg of digest was injected for each analysis. Mobile phase A and B were composed of 0.1% TFA in milliQ water and ACN, respectively. Tryptic peptides were separated at a constant flow rate of 0.20 mL/min and 40°C column temperature using the following gradient: 0% B for 5 min, 0–10% B in 1 min, 10–47.5% B in 69 min, which was followed by a 34 min column cleaning and re-equilibration phase. The mass spectrometer was operated in positive ionization mode with the following source parameters: 3.8 kV spray voltage, 275°C ion transfer tube temperature, sheath gas flow rate of 35, and auxiliary gas flow rate of 10. MS data were acquired with the Xcalibur software using a data-dependent acquisition method with a 3 s duty cycle. Full MS scans were recorded at 120’000 resolution and with the AGC target set to standard, 60% RF lens, 50 ms maximum injection time, and 300–2000 m/z scan range. Precursor ions were fragmented using a combination of collision-induced dissociation (CID, collision energy: 30%, activation time: 10 ms, activation Q: 0.25, and 500 ms maximum injection time) and higher-energy collisional dissociation (HCD, collision energy: 33.25–36.75% and automatic maximum injection time selection). The isolation window was set to 1.6 m/z and fragment ions were acquired at 30’000 resolution.
LC-MS data acquisition
Data acquisition related to targeted attribute monitoring and new peak detection analyses was conducted in Chromeleon 7.2 using a Thermo Vanquish UHPLC system coupled to a Thermo Exactive Plus mass spectrometer with a heated electrospray ionization source. About 4 µg of digest was injected for each analysis. The LC conditions and the gradient elution of tryptic peptides matched the above-described procedure for LC-MS/MS analysis. The mass spectrometer was operated in positive ionization mode with the following source conditions: 3.7 kV spray voltage, 250°C capillary temperature, 250°C auxiliary gas temperature, 55 S-lens RF level, sheath gas flow rate of 45, and auxiliary gas flow rate of 10. MS data were acquired in full scan mode at 140’000 resolution and with a 3e6 AGC target, 200 ms maximum injection time, and 300–2000 m/z scan range.
Peptide library creation
LC-MS/MS data were analyzed using Genedata Expressionist software 2025.1 and earlier versions. Raw files underwent instrument-specific preprocessing including chromatogram smoothing, chemical noise subtraction, RT alignment, peak detection, and isotope clustering. Processed data were searched against the expected product sequence with ≤10.0 ppm mass accuracy; carbamidomethylation of cysteine residues was fixed, while common post-translational modifications (e.g., methionine/tryptophan oxidation, glycation, deamidation) were set as variable modifications. Each sequence assignment was manually verified based on MS/MS evidence. RT, mass, and charge state(s) of identified and relevant species were compiled into a project-specific peptide library for targeted attribute monitoring.
Targeted attribute monitoring
LC-MS data analysis for targeted attribute monitoring relied on a staggered process including the following steps in the Genedata Expressionist software version 2025.1: (i) data preprocessing, (ii) peak detection and isotope clustering, (iii) data-driven calibration of library RT values, (iv) refinement of integration boundaries, (v) library-based annotation of peptide species, and (vi) relative quantification of quality attributes.
Briefly, LC-MS raw files were collectively subjected to background subtraction, RT alignment, peak detection, and isotope clustering using instrument-specific parameter settings. Data-driven calibration of library RT values was initiated by annotating predefined reference peptides in the LC-MS dataset with ≤10.0 ppm mass accuracy and a relaxed RT tolerance window of ≤1.0 min. A custom software module evaluated the observed versus theoretical RT values of the annotated reference peptides and subsequently calibrated the remaining library RT entries based on these assessments. Following manual refinement of integration boundaries, annotation of targeted species was accomplished through database searching against the RT-calibrated, project-specific peptide library. Peptide candidates were matched to library entries using a mass accuracy threshold of 10.0 ppm and a ≤0.20 min RT tolerance window. The relative abundance for each quality attribute was calculated as a percentage of the total integrated maximum intensity (i.e., area under the curve defined by the peak boundaries and the maximum intensity data point at each MS scan) for all related forms of a given peptide and charge state(s).
New peak detection
NPD analysis was performed in the Genedata Expressionist software version 2025.1 using a staggered workflow that encompassed the following key characteristics: (1) automatic artifact detection, (2) product-centric IT and FCD thresholding, (3) statistical evaluation, (4) KPL-based annotation of known species, and (5) manual evaluation/verification of NPD results.
Annotated LC-MS files from targeted attribute monitoring were loaded into the NPD workflow and subjected to charge state grouping. Subsequent artifact detection involved the rule-based flagging of signals that were likely related to chemical noise, polyethylene glycol (PEG)-based polymers, and metal adduct ions. That is, features were flagged as potential chemical noise species if the same mass (±3 ppm mass accuracy tolerance) was observed ≥25 times throughout the LC-MS dataset. PEG-related signals were efficiently identified and flagged by detecting the characteristic 44 Da mass differences (±2.5 ppm mass tolerance) related to the consecutive ethylene oxide (C2H4O) monomer species. A polymer series had to consist of ≥ 8 individual features that shared at least one common charge state. Different adduct ions (i.e., sodium-, potassium-, and iron-related gas-phase complexes) and in-source losses (i.e., dehydration) were searched using an inclusion (−18 to +55 Da) and an exclusion delta mass window (−16 to +20 Da). Potential adduct species had to co-elute (±0.05 min RT tolerance) and share at least one common charge state with the parent feature. The IT and FCD thresholds were set to 0.05% BPC and 3.0, respectively. Differences in feature abundance were statistically evaluated using a two-sample t-test by comparing the measured intensities of a given feature across multiple reference injections to a single sample intensity. The statistical evaluation was performed for each feature and each sample individually with a defined alpha value of ≤0.001. Features that did not meet the defined prerequisites in terms of IT, FCD threshold, and statistical significance were excluded from further NPD analysis. NPD-relevant features were searched against a project-specific KPL with a mass accuracy threshold of 10.0 ppm and a ≤1.0 min RT tolerance window. Finally, NPD results were manually reviewed and verified. Species were invalidated if they had KPL-related annotations, artifact flags, appeared only as 1+ species, or had charge states ≥8+. Remaining NPD hits were checked for consistency and reproducibility across replicate measurements.
Supplementary Material
Acknowledgments
The authors would like to thank Dr. Christoph Bächler for his support in exploring novel technologies for biopharmaceutical development.
Funding Statement
The author(s) reported there is no funding associated with the work featured in this article.
Disclosure statement
D. M., C. S., R. O., C. S., M. Z., A. S., E. W., P. G., N. D., and M. E. are employed by Genedata.
Supplementary Information
Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2025.2572411
References
- 1.Walsh G, Walsh E.. Biopharmaceutical benchmarks 2022. Nat Biotechnol. 2022;40(12):1722–16. doi: 10.1038/s41587-022-01582-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Mans J, Oyugi M, Asmelash B, Sommers C, Rogstad S. The use of mass spectrometry in therapeutic protein biologics license applications: a retrospective review revisited. J Am Soc Mass Spectrom. 2023;34(11):2575–2584. doi: 10.1021/jasms.3c00286. [DOI] [PubMed] [Google Scholar]
- 3.Ruesch MN, Benetti L, Berkay E, Cirelli DJ, Frantz N, Gastens MH, Kelley WP, Kretsinger J, Lewis M, Novick S, et al. Strategies for setting patient-centric commercial specifications for biotherapeutic products. J Pharm Sci. 2021;110(2):771–784. doi: 10.1016/j.xphs.2020.09.048. [DOI] [PubMed] [Google Scholar]
- 4.Kretsinger J, Frantz N, Hart SA, Kelley WP, Kitchen B, Novick S, Rellahan B, Stranges D, Stroop CJM, Yin P, et al. Expectations for phase-appropriate drug substance and drug product specifications for early-stage protein therapeutics. J Pharm Sci. 2019;108(4):1442–1452. doi: 10.1016/j.xphs.2018.11.042. [DOI] [PubMed] [Google Scholar]
- 5.Pohl T, Gervais A, Dirksen EHC, D’Alessio V, Bechtold-Peters K, Burkitt W, Cao L, Greven S, Lennard A, Li X, et al. Technical considerations for the implementation of the multi-attribute-method by mass spectrometry in a quality control laboratory. Eur J Pharm Biopharm. 2023;188:231–242. doi: 10.1016/j.ejpb.2023.04.024. [DOI] [PubMed] [Google Scholar]
- 6.Rogers RS, Abernathy M, Richardson DD, Rouse JC, Sperry JB, Swann P, Wypych J, Yu C, Zang L, Deshpande R. A view on the importance of “multi-attribute method” for measuring purity of biopharmaceuticals and improving overall control strategy. Aaps J. 2017;20(7). doi: 10.1208/s12248-017-0168-3. [DOI] [PubMed] [Google Scholar]
- 7.Rogstad S, Yan H, Wang X, Powers D, Brorson K, Damdinsuren B, Lee S. Multi-attribute method for quality control of therapeutic proteins. Anal Chem. 2019;91(22):14170–14177. doi: 10.1021/acs.analchem.9b03808. [DOI] [PubMed] [Google Scholar]
- 8.Ren D. Advancing mass spectrometry technology in cGMP environments. Trends Biotechnol. 2020;38(10):1051–1053. doi: 10.1016/j.tibtech.2020.06.007. [DOI] [PubMed] [Google Scholar]
- 9.Hao Z, Moore B, Ren C, Sadek M, Macchi F, Yang L, Harris J, Yee L, Liu E, Tran V, et al. Multi-attribute method performance profile for quality control of monoclonal antibody therapeutics. J Pharm Biomed Anal. 2021;205:114330. doi: 10.1016/j.jpba.2021.114330. [DOI] [PubMed] [Google Scholar]
- 10.Millan-Martin S, Jakes C, Carillo S, Rogers R, Ren D, Bones J. Comprehensive multi-attribute method workflow for biotherapeutic characterization and current good manufacturing practices testing. Nat Protoc. 2023;18(4):1056–1089. doi: 10.1038/s41596-022-00785-5. [DOI] [PubMed] [Google Scholar]
- 11.Li X, Pierson NA, Hua X, Patel BA, Olma MH, Strulson CA, Letarte S, Richardson DD. Analytical performance evaluation of identity, quality-attribute monitoring and new peak detection in a platform multi-attribute method using Lys-C digestion for characterization and quality control of therapeutic monoclonal antibodies. J Pharm Sci. 2023;112(3):691–699. doi: 10.1016/j.xphs.2022.10.018. [DOI] [PubMed] [Google Scholar]
- 12.Gervais A, Dirksen EHC, Pohl T, Bechtold-Peters K, Burkitt W, D’Alessio V, Greven S, Lennard A, Li X, Lössner C, et al. Compliance and regulatory considerations for the implementation of the multi-attribute-method by mass spectrometry in a quality control laboratory. Eur J Pharm Biopharm. 2023;191:57–67. doi: 10.1016/j.ejpb.2023.08.008. [DOI] [PubMed] [Google Scholar]
- 13.Yang F, Zhang J, Buettner A, Vosika E, Sadek M, Hao Z, Reusch D, Koenig M, Chan W, Bathke A, et al. Mass spectrometry-based multi-attribute method in protein therapeutics product quality monitoring and quality control. MAbs. 2023;15(1):2197668. doi: 10.1080/19420862.2023.2197668. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Evans AR, Mulholland J, Lewis MJ, Hu P. Targeted CQA analytical control strategy for commercial antibody products: replacing ion-exchange chromatography methods for charge heterogeneity with multi-attribute monitoring. MAbs. 2024;16(1):2341641. doi: 10.1080/19420862.2024.2341641. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Mouchahoir T, Schiel JE, Rogers R, Heckert A, Place BJ, Ammerman A, Li X, Robinson T, Schmidt B, Chumsae CM, et al. New peak detection performance metrics from the MAM consortium interlaboratory study. J Am Soc Mass Spectrom. 2021;32(4):913–928. doi: 10.1021/jasms.0c00415. [DOI] [PubMed] [Google Scholar]
- 16.Cao Q, Guan J, Li D, Zhang J, Togashi R, Johnson EJ, Chan W, Guo J, Liu P, Liang Y, et al. A robust and sensitive new peak detection and identification method for mass spectrometry-based differential analysis in biologics characterization. Anal Chem. 2025;97(2):1246–1253. doi: 10.1021/acs.analchem.4c04913. [DOI] [PubMed] [Google Scholar]
- 17.Sadek M, Moore BN, Yu C, Ruppe N, Abdun-Nabi A, Hao Z, Alvarez M, Dahotre S, Deperalta G. A robust purity method for biotherapeutics using new peak detection in an LC-MS-based multi-attribute method. J Am Soc Mass Spectrom. 2023;34(3):484–492. doi: 10.1021/jasms.2c00355. [DOI] [PubMed] [Google Scholar]
- 18.Zhang Z, Richardson J, Shah B. Method for detecting rare differences between two LC-MS runs. Anal Biochem. 2023;674:115211. doi: 10.1016/j.ab.2023.115211. [DOI] [PubMed] [Google Scholar]
- 19.Griaud F, Denefeld B, Lang M, Hensinger H, Haberl P, Berg M. Unbiased in-depth characterization of CEX fractions from a stressed monoclonal antibody by mass spectrometry. MAbs. 2017;9(5):820–830. doi: 10.1080/19420862.2017.1313367. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Niu B, Lu Y, Chen X, Xu W. Using new peak detection to solve sequence variants analysis challenges in bioprocess development. J Am Soc Mass Spectrom. 2023;34(3):401–408. doi: 10.1021/jasms.2c00292. [DOI] [PubMed] [Google Scholar]
- 21.Griaud F, Winter A, Denefeld B, Lang M, Hensinger H, Straube F, Sackewitz M, Berg M. Identification of multiple serine to asparagine sequence variation sites in an intended copy product of Lucentis(R) by mass spectrometry. MAbs. 2017;9(8):1337–1348. doi: 10.1080/19420862.2017.1366395. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Cao M, Xu W, Niu B, Kabundi I, Luo H, Prophet M, Chen W, Liu D, Saveliev SV, Urh M, et al. An automated and qualified platform method for site-specific succinimide and deamidation quantitation using low-pH peptide mapping. J Pharm Sci. 2019;108(11):3540–3549. doi: 10.1016/j.xphs.2019.07.019. [DOI] [PubMed] [Google Scholar]
- 23.Jakes C, Millán-Martín S, Carillo S, Scheffler K, Zaborowska I, Bones J. Tracking the behavior of monoclonal antibody product quality attributes using a multi-attribute method workflow. J Am Soc Mass Spectrom. 2021;32(8):1998–2012. doi: 10.1021/jasms.0c00432. [DOI] [PubMed] [Google Scholar]
- 24.Oyugi M, Wang X, Yang X, Wu D, Rogstad S. Method validation and new peak detection for the liquid chromatography-mass spectrometry multi-attribute method. J Pharm Biomed Anal. 2023;234:115564. doi: 10.1016/j.jpba.2023.115564. [DOI] [PubMed] [Google Scholar]
- 25.Numao E, Yanagisawa K, Yagi Y, Tsuchida D, Yamazaki K. Comparison of a new peak detection function for selecting a phase-appropriate multi-attribute method system. J Biosci Bioeng. 2025;139(2):156–163. doi: 10.1016/j.jbiosc.2024.10.005. [DOI] [PubMed] [Google Scholar]
- 26.Carvalho SB, Gomes RA, Pfenninger A, Fischer M, Strotbek M, Isidro IA, Tugçu N, Gomes-Alves P. Multi attribute method implementation using a high resolution mass spectrometry platform: from sample preparation to batch analysis. PLOS ONE. 2022;17(1):e0262711. doi: 10.1371/journal.pone.0262711. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.