The Dual-Integrated Multi-Attribute Method: Simultaneous Evaluation of Post-Translational Modifications and Host Cell Proteins for Unpurified Bulk Harvest of Antibody Therapeutics

Abstract

The Multi-Attribute Method (MAM) using liquid chromatography/mass spectrometry (LC/MS) is widely applied to evaluate post-translational modifications (PTMs) in antibody therapeutics. This approach allows the monitoring of multiple PTMs within a single method. In this study, we extended the scope of MAM by incorporating host cell proteins (HCPs) as additional targets and developed a dual, integrated PTMs–HCPs MAM platform. This platform enables the simultaneous evaluation of more comprehensive critical quality attributes through a combination of MS1-based PTMs analysis and data-independent acquisition-based HCP quantification. Although proteomics-based characterization of residual HCPs has been extensively reported, prior studies have primarily focused on purified drug substances or products. Here, we demonstrate the application of our integrated platform to unpurified bulk harvest samples, eliminating the need for purification. While conventional PTMs-MAM and HCP-proteomics are typically conducted separately on purified samples, our dual integrated-MAM approach allows the linking of antibody PTMs with their potential HCP influencers within a single LC/MS injection.

1. Introduction

The Multi-Attribute Method (MAM) has gained significant attention as an advanced quality control strategy for monoclonal antibody (mAb) therapeutics. − MAM is based on peptide mapping and new peak detection (NPD) combined with liquid chromatography/high-resolution mass spectrometry (LC/MS), which together enable quantitative evaluation of post-translational modifications (PTMs) at site-specific resolution. The use of PTMs as critical quality attributes (CQAs) is widely recognized owing to their potential impact on both efficacy and safety. The value of MAM in supporting quality by design (QbD) principles is supported by numerous findings that particular PTMs can strongly influence therapeutic performance. −

A typical MAM workflow begins with the identification and selection of PTMs of interest. Data-dependent acquisition (DDA) is generally applied to construct a PTM reference library. Once the library is established, routine monitoring can be performed using MS1-based quantification and new peak detection (NPD). While MS1-based quantification ensures reliable monitoring of predefined PTMs, the NPD function is designed to detect unexpected changes, such as the appearance of new species or disappearance of known ones, relative to reference or control materials. In practice, achieving reproducible PTM quantitation requires robust and standardized sample preparation, together with optimized LC/MS conditions. , The utility of NPD under good manufacturing practice (GMP) environments is still debated; nevertheless, routine MAM analysis is increasingly expected to allow the monitoring of even subtle PTM fluctuations during product development and quality control.

Like PTMs, host cell proteins (HCPs) are also considered CQAs in mAb therapeutics. Conventional HCP quantification is most often conducted by enzyme-linked immunosorbent assay (ELISA) with polyclonal antibodies. , However, as awareness of the potential safety and stability risks of specific HCPs has grown, attention has focused on the ability of proteomics-based mass spectrometry approaches to provide more detailed characterization. In particular, data-independent acquisition (DIA) has emerged as a powerful technique for improving the depth, sensitivity, and reproducibility of HCP quantitation. − Recent studies have successfully detected trace-level HCPs at concentrations as low as parts per billion through the use of strategies such as nanoelectrospray ionization, antibody depletion through native digestion, and DIA-based proteomics workflows. These methods generally aim to remove the abundant therapeutic antibody from the sample matrix in order to minimize ion suppression and enhance HCP detection.

For patient-centric process development, process conditions that minimize the presence of high-risk HCPs should be established as early in the development process as possible. Applying proteomics-based HCP characterization to in-process samples, such as unpurified bulk harvest (UPBH), provides richer information than limiting analysis to final drug substance or drug product. Several reports recommend evaluating HCP profiles during upstream process development. − The advantages of this strategy include, for example, detecting a high-risk HCP profile early in upstream development, which can guide further optimization of upstream process parameters or inform downstream process development strategies aimed at reducing high-risk HCPs. If neither approach achieves sufficient reduction, returning to cell line development or cell engineering may be considered promptly. Such a streamlined development strategy cannot be achieved by relying solely on ELISA-based HCP profiles or by performing proteomics-based HCP evaluation only at the final drug substance or product stage.

From the perspective of process developers, integrated analytical approaches that simultaneously evaluate multiple CQAs, such as PTMs and HCPs, can significantly reduce both development timelines and costs. Consistent with this, mass spectrometry-based strategies, including MAM for PTM monitoring and proteomics for HCP profiling, have become indispensable tools. Although these strategies both rely on enzymatic digestion and high-resolution LC/MS, they have traditionally been treated as distinct. This is primarily because MAM requires a relatively clean background matrix to achieve accurate PTM quantitation, necessitating the removal of HCPs, whereas HCP characterization often requires depletion of the therapeutic antibody to maximize protein identification. We think these specific sample preparation requirements are the barrier to applying these useful technologies for process development.

In this study, we aimed to establish a dual, integrated PTMs–HCPs MAM (di-MAM) platform to address the limitations of conventional approaches. The method combines MS1-based quantitation for PTMs with DIA-based profiling for HCPs, enabling concurrent evaluation within a single workflow. We expected that incorporation of library-based quantitation would facilitate data analysis and improve reproducibility. While previous studies have primarily focused on residual HCPs in purified drug substances or products, profiling HCPs at the unpurified bulk harvest (UPBH) stage provides valuable insights into cell line characteristics, batch-to-batch variability, and upstream process robustness. In particular, we expected that a di-MAM strategy would allow simultaneous monitoring of PTMs and HCPs without additional purification, thereby reducing analytical turnaround time, lower development costs, and improving process understanding. This approach aligns with recent QbD principles, which emphasize evaluating potential CQAs (pCQAs) and assessing the impact of process parameters by monitoring pCQAs during development. Furthermore, the di-MAM platform has the potential to evolve into a process analytical technology (PAT) which is essential for realizing a QbD approach for biotherapeutics considering its feature of broader coverage of CQAs than other methods.

2. Results

2.1. Construction of PTM Libraries and Application for UPBH Samples

To construct the PTM library, DDA was performed on degraded mAb samples, which were expected to exhibit higher levels of PTMs. Peak annotation was conducted using BioPharmaFinder (ThermoFisher). PTM-related peptides identified through this analysis were then exported to Chromeleon (ThermoFisher) for quantification. The library includes peptides with MS1 m/z values corresponding to the most intense charge states, along with their respective retention times. The selected PTMs are summarized in Table and include methionine oxidation (heavy chain: 254, 360, 399, and 430), asparagine deamidation (heavy chain: 346 and 386; light chain: 158), asparagine succinimide (heavy chain: 317, 327, 386, 392, and 436; light chain: 152 and 158), glutamine deamidation (heavy chain: 421), and lysine glycation (heavy chain: 328). The glycan species included in the library comprise high mannose glycans (Man8, Man7, Man6, Man5), fucosylated glycans (G0F, G0F-Gn, G1F, G1F-Gn, G2F), afucosylated species (G0, G0-Gn) and the aglycosylated peptide.

1. Detected PTMs .

	Modification positions
Post-translational modification	Heavy chain					Light chain
Methionine oxidation	254 (248)	360 (354)	399 (393)	430 (424)
Asparagine deamidation	346 (340)	386 (380)				158 (158)
Asparagine succinimide	317 (311)	327 (321)	386 (380)	392 (386)	436 (430)	152 (152)	158 (158)
Glutamine deamidation	421 (415)
Lysine glycation	328 (322)

^aThe number describes the amino acid position from N-terminal of each chain. The number in bracket describes the amino acid position by Kabat numbering.

The established PTM library was then used to verify PTM evaluation of UPBH samples. Stressed UPBH samples and purified mAb subjected to thermal stress (40 °C for 8 weeks) and photostress conditions using a D65 lamp (total illumination of 1208.6 klx-hours at 25 °C) were analyzed. Given that PTM modification rates in mAbs can vary with formulation components, buffer-exchanged UPBH samples from culture media to the same mAb formulation were also stressed. The PTM modification levels are summarized in Table . A significant increase in methionine oxidation was detected in the photostressed samples, while slight increases in asparagine deamidation and lysine glycation were observed in the heat-stressed samples. Methionine 254 (M254) located in the heavy chain showed the largest increase in oxidation, rising from 4.6% in the nonstressed purified sample to 27.0% in the photostressed purified sample, as quantified by MS1. The increase in methionine oxidation was significantly higher in the UPBH sample. Oxidation level in the nonstressed UPBH sample was similar to that of the purified sample (3.7% at M254), but increased to 44.6% under heat stress and 95.9% under photo stress. In contrast, there was no significant difference in methionine oxidation levels between the buffer-exchanged UPBH sample and unexchanged sample.

2. PTM Levels in Forced Degradation Samples.

Average ± SD		M254 Oxidation, %	M360 Oxidation, %	N386 Deamidation, %	N386 NH3 loss, %	N392 NH3 loss, %	M399 Oxidation, %	N152 NH3 loss, %	N158 Deamidation, %
mAb	Non stressed	4.64 ± 14.40	0.60 ± 1.94	0.72 ± 0.40	0.32 ± 0.05	0.75 ± 0.09	0.51 ± 0.04	0.00 ± 0.00	1.87 ± 3.22
	Heat stressed	7.47 ± 0.57	1.00 ± 0.03	1.74 ± 0.72	0.42 ± 0.02	0.81 ± 0.03	3.85 ± 0.29	0.02 ± 0.00	3.21 ± 0.17
	Photo stressed	27.00 ± 8.52	4.11 ± 1.00	0.86 ± 0.45	0.35 ± 0.04	0.82 ± 0.07	0.57 ± 0.03	0.00 ± 0.00	7.43 ± 0.46
UPBH	Non stressed	3.75 ± 0.13	0.66 ± 0.06	1.23 ± 0.33	0.98 ± 0.07	0.34 ± 0.04	1.81 ± 0.27	0.05 ± 0.01	0.26 ± 0.04
	Heat stress	44.58 ± 0.52	11.41 ± 0.68	24.83 ± 17.75	0.83 ± 0.30	0.65 ± 0.27	22.80 ± 0.80	0.05 ± 0.01	2.32 ± 0.30
	Photo stressed	95.88 ± 0.33	58.93 ± 0.43	7.60 ± 9.98	0.99 ± 0.17	0.50 ± 0.17	58.79 ± 1.04	0.06 ± 0.00	0.36 ± 0.11
UPBH (Buffer exchanged)	Non stressed	5.84 ± 0.14	0.87 ± 0.04	2.32 ± 2.13	0.88 ± 0.06	0.29 ± 0.03	2.48 ± 0.79	0.03 ± 0.01	0.25 ± 0.05
	Heat stress	40.76 ± 1.92	7.55 ± 0.70	18.42 ± 23.84	2.39 ± 1.42	0.32 ± 0.14	21.39 ± 24.16	0.04 ± 0.01	0.37 ± 0.04
	Photo stressed	97.12 ± 0.61	70.69 ± 0.45	1.20 ± 0.83	0.89 ± 0.15	0.43 ± 0.09	77.22 ± 0.47	0.06 ± 0.03	0.32 ± 0.11

Average ± SD		N158 NH3 loss, %	K328 Glycation, %	N327 NH3 loss, %	N317 Deamidation, %	N317 NH3 loss, %	M430 Oxidation, %	N436 Deamidation, %	Q421 Deamidation, %
mAb	Non stressed	0.04 ± 0.01	0.04 ± 0.01	0.33 ± 0.25	0.64 ± 0.27	0.05 ± 0.07	0.93 ± 9.64	0.29 ± 0.12	0.12 ± 0.04
	Heat stressed	0.05 ± 0.01	0.04 ± 0.00	0.46 ± 0.07	1.18 ± 0.21	0.61 ± 0.19	2.03 ± 0.10	0.31 ± 0.10	0.39 ± 0.06
	Photo stressed	0.05 ± 0.01	0.04 ± 0.01	0.43 ± 0.26	0.71 ± 0.22	0.16 ± 0.13	15.61 ± 8.48	0.24 ± 0.07	0.09 ± 0.04
UPBH	Non stressed	0.03 ± 0.01	0.76 ± 0.17	0.04 ± 0.01	0.69 ± 0.22	1.87 ± 0.25	0.66 ± 0.02	0.01 ± 0.01	0.02 ± 0.02
	Heat stress	0.03 ± 0.00	2.24 ± 0.04	0.04 ± 0.00	1.87 ± 0.52	2.05 ± 0.45	10.71 ± 0.22	0.34 ± 0.29	0.11 ± 0.11
	Photo stressed	0.03 ± 0.01	0.89 ± 0.04	0.08 ± 0.00	0.95 ± 0.28	1.76 ± 0.33	68.56 ± 10.36	0.07 ± 0.05	0.02 ± 0.02
UPBH (Buffer exchanged)	Non stressed	0.04 ± 0.01	0.49 ± 0.12	0.03 ± 0.01	0.23 ± 0.21	1.57 ± 0.22	1.60 ± 0.04	0.09 ± 0.08	0.03 ± 0.03
	Heat stress	0.04 ± 0.01	0.38 ± 0.04	3.49 ± 2.13	2.20 ± 0.50	4.88 ± 0.87	19.95 ± 15.93	0.48 ± 0.54	0.11 ± 0.06
	Photo stressed	0.05 ± 0.02	0.57 ± 0.11	0.11 ± 0.01	0.49 ± 0.06	1.60 ± 0.33	97.32 ± 0.50	0.00 ± 0.00	0.00 ± 0.00

In contrast to the case with methionine oxidation, the changes in asparagine-related modifications, including deamidation and its intermediate succinimide (NH₃ loss), were relatively small. Deamidation at N386 and N317 in the PENNY peptide increased in the heat-stressed sample, from 0.7% to 1.7% for N386, and from 0.6% to 1.2% for N317. The increase in deamidation in the UPBH sample was also greater compared to the purified sample, with deamidation levels at N386 of 1.2%, 24.8%, and 7.6% in the nonstressed, heat-stressed, and photostressed samples, respectively.

We successfully detected 12 glycan-related PTMs, including the aglycosylated peptide (Table ). Only minor differences were observed in glycan profile among the purified mAb, UPBH, and buffer-exchanged UPBH samples. In nonstressed conditions, the dominant glycan species, G0F, accounted for 77.4% in the purified mAb and 73.7% in the UPBH sample, while the buffer-exchanged UPBH showed a slightly higher level, at 75.6%. G1F glycan ranged between 5.0% and 6.3% across nonstressed samples.

3. Detected Glycan Species .

Average ± SD		G0F	G0F-Gn	G1F	M5	Aglycosylated	G0-Gn
mAb	Non stressed	77.44 ± 1.19	4.56 ± 0.11	4.97 ± 0.09	0.55 ± 0.03	4.69 ± 0.96	0.97 ± 0.12
	Heat stress	79.66 ± 1.21	3.92 ± 0.24	5.04 ± 0.06	0.43 ± 0.05	3.62 ± 0.61	0.81 ± 0.13
	Photo stressed	76.91 ± 2.92	4.15 ± 0.34	5.24 ± 0.49	0.56 ± 0.06	5.82 ± 2.71	0.88 ± 0.20
UPBH	Non stressed	73.71 ± 0.98	4.10 ± 0.13	5.34 ± 0.19	0.40 ± 0.25	8.63 ± 2.00	0.61 ± 0.21
	Heat stress	76.70 ± 1.56	4.09 ± 0.31	5.44 ± 0.13	0.50 ± 0.06	6.44 ± 1.53	0.55 ± 0.11
	Photo stressed	74.79 ± 2.04	4.42 ± 0.17	5.68 ± 0.21	0.53 ± 0.15	7.41 ± 1.96	0.64 ± 0.11
UPBH (Buffer exchanged)	Non stressed	75.62 ± 0.40	4.18 ± 0.40	5.45 ± 0.48	0.36 ± 0.12	8.00 ± 2.47	0.34 ± 0.09
	Heat stress	84.29 ± 2.70	2.44 ± 0.82	0.85 ± 0.60	nd	12.03 ± 2.49	nd
	Photo stressed	80.08 ± 0.66	4.49 ± 0.04	6.26 ± 0.21	0.05 ± 0.03	2.08 ± 0.66	0.08 ± 0.05

Average ± SD		G1F-Gn	G0	G2F	M7	M8	M6
mAb	Non stressed	1.31 ± 0.07	4.48 ± 0.10	0.25 ± 0.04	0.21 ± 0.21	0.42 ± 0.05	0.14 ± 0.03
	Heat stress	1.20 ± 0.05	4.37 ± 0.10	0.27 ± 0.02	0.18 ± 0.18	0.36 ± 0.02	0.14 ± 0.01
	Photo stressed	1.08 ± 0.35	4.53 ± 0.54	0.32 ± 0.11	0.14 ± 0.14	0.28 ± 0.23	0.10 ± 0.09
UPBH	Non stressed	1.12 ± 0.05	4.64 ± 0.13	0.34 ± 0.13	0.26 ± 0.26	0.77 ± 0.23	0.07 ± 0.06
	Heat stress	0.99 ± 0.24	4.79 ± 0.35	0.39 ± 0.17	0.01 ± 0.01	0.03 ± 0.02	0.07 ± 0.03
	Photo stressed	0.96 ± 0.18	4.73 ± 0.56	0.50 ± 0.10	0.14 ± 0.14	0.09 ± 0.07	0.09 ± 0.09
UPBH (Buffer exchanged)	Non stressed	0.63 ± 0.35	4.31 ± 0.29	0.33 ± 0.18	0.12 ± 0.12	0.65 ± 0.30	0.00 ± 0.01
	Heat stress	nd	0.30 ± na	0.08 ± 0.14	nd	nd	nd
	Photo stressed	1.20 ± 0.26	5.02 ± 0.38	0.55 ± 0.05	0.03 ± 0.03	0.14 ± 0.10	0.01 ± 0.01

^and: not detected, na: not applicable (not detected in all three injections).

Minor variations in glycan composition were also observed under stressed conditions. In the purified mAb, G0F ratio increased from 77.4% in the nonstressed sample to 80.0% under heat stress and remained relatively stable at 77.0% under photo stress. Similar trends were observed in UPBH and buffer-exchanged UPBH samples: G0F levels in heat-stressed samples were 76.7% and 84.3%, respectively; nonstressed samples showed the lowest levels, at 73.7% and 75.6%; and photostressed samples exhibited intermediate levels of 74.8% and 80.1%.

An outlier profile was observed in the heat-stressed, buffer-exchanged UPBH sample. In this case, high mannose species (M8, M7, M6, M5) were undetected, and the level of aglycosylated peptide reached 12.0%, which was notably higher than the 2.0–8.0% range seen in other samples.

2.2. Construction of HCP Libraries

To construct the HCP quantification library, DDA data from the tryptic-digested UPBH samples were used in an open spectrum library search. The HCP library was created for DIA analysis to improve specificity, sensitivity and reproducibility for detecting lower-abundance HCPs. To validate the library, the repeatability of retention time and peak area was assessed using triplicate DIA injections of the UPBH sample. Of the 5,759 proteins identified in the spectral library search, 4,031 proteins exhibited reproducible retention times (within a 30-s window across triplicates) for the selected transitions, and 2,505 proteins showed consistent peak areas, with relative standard deviations (RSDs) below 10%. In total, 2,565 proteins satisfied both criteria for retention time and peak area reproducibility. This validated protein library was subsequently used in all downstream experiments.

2.3. Application of MS1/DIA-MAM to Culture Process Development

2.3.1. Cultivation in Various Conditions

We applied the di-MAM approach to UPBH samples obtained under various cultivation conditions using an automated small-scale bioreactor system. Time course of cultivation performance, viable cell density (VCD), viability (Via), mAb titer (Titer), pH, and dissolved oxygen (DO) are shown in Figure . Dissolved oxygen (DO) levels decreased as cultivation progressed, with notable differences from day 5 onward. By day 5, the low DO setting exhibited approximately 60% of the nominal DO level. From day 10, significant divergence was observed between the low DO condition and the low DO condition combined with a low pH setting. Regarding pH control, the intended theoretical ranges were successfully maintained. Similar cell viability was observed across all conditions on day 10. However, the decline in viability during extended cultivation varied considerably between conditions. Notably, the low pH condition exhibited the highest viability on day 14, but by day 17, it had dropped to the lowest level among the tested conditions.

1.

Time course of cultivation performance, viable cell density (VCD), viability (Via), mAb titer (Titer), pH, and dissolved oxygen (DO). Conditions shown in the legend correspond to Table S1 in the Materials and Methods section. Blue and orange: control condition; green and light blue: low DO condition; magenta and light green: low pH; dark blue and dark orange: low DO and low pH.

Regarding mAb titer, a linear increase was observed under the nominal pH control condition (nominal condition, low DO). In contrast, the rate of increase diminished after day 14 in the lower pH control conditions (low pH and low pH with low DO).

2.3.2. PTM (Oxidation, Deamidation, and Glycation)

PTMs in UPBH were evaluated using samples from an automated cultivation system. Twelve different conditions were tested in duplicate, and MS1 scans were used to quantify PTM levels. Extended cultivation time led to an increase in methionine oxidation, with M254 oxidation showing the highest increase, from an average of 1.88% on day 10 to 2.45% on day 17 (Figure ). In contrast, deamidation levels remained stable, while levels of the intermediate succinimide slightly decreased with longer cultivation times.

2.

PTM (oxidation, deamidation, and glycation) levels (%) in cultivation samples on days 10, 14, and 17. Box plots are shown in red. Green diamonds show the results of ANOVA.

Regarding variations in pH, no significant changes were observed for methionine oxidation or asparagine deamidation levels (Figure ). An elevated deamidation rate was seen in one replicate at standard pH (7.05 ± 0.25) with DO levels of 40% or 15% on day 4. However, another replicate under identical conditions exhibited nominal deamidation, suggesting that the higher rate may have been an outlier or fell within the method’s variance.

3.

PTM (oxidation, deamidation, and glycation) levels (%) in cultivation samples at target pH 6.95 and 7.05. Box plots are shown in red. Green diamonds show the results of ANOVA.

Differences in PTM levels between the low and high DO conditions were minor compared to other cultivation condition variations, with no statistically significant differences observed (Figure ).

4.

PTM (oxidation, deamidation, and glycation) levels (%) in cultivation samples at target dissolved oxygen 15% and 40%. Box plots are shown in red. Green diamonds show the results of ANOVA.

Protein A-purified UPBH samples were also analyzed using the same strategy. Respective PTM levels and correlation between UPBH and the protein A-purified UPBH sample are listed in Table S2 and Figure S1. N158 deamidation was not detected in purified UPBH samples. Statistical analyses with cultivation parameters are shown in Figures S2–S4. Coefficients of determination (R ²) between the UPBH sample and purified sample results ranged from 0.006 to 0.415.

2.3.3. PTM (Glycosylation)

For glycan profiles, distinct trends were observed across different pH levels and harvest days (Figures and ). A slight decreasing trend in G0F level was noted under high pH and extended cultivation conditions, whereas G0F-Gn and G0-Gn levels were higher in these same conditions. Interestingly, other glycan profiles exhibited differing trends between extended cultivation and high pH conditions. For example, the ratio of M5 was lower under high pH but increased over time with longer cultivation, increasing from an average of 1.3% on day 10 to 2.2% on day 17.

5.

PTM (glycosylation) levels (%) in cultivation samples on days 10, 14, and 17. Box plots are shown in red. Green diamonds show the results of ANOVA.

6.

PTM (glycosylation) levels (%) in cultivation samples at target pH 6.95 and 7.05. Box plots are shown in red. Green diamonds show the results of ANOVA.

Aglycosylated and G1F-Gn ratios also showed divergent trends. Although no significant differences were observed between pH 6.05 and pH 7.05, extension of cultivation days resulted in a significant increase in aglycosylated species and significant decrease in G1F-Gn levels, both with small variance. Dissolved oxygen levels, however, did not show any significant impact on glycan profiles.

DO difference only impacted on M7 and M8 (Figure ). Slightly lower ratio of these glycans in 40% DO condition was observed.

7.

PTM (glycosylation) levels (%) in cultivation samples at target dissolved oxygen 15% and 40%.

The glycan profile of respective UPBH samples after protein A purification were also evaluated. The coefficients of determination between UPBH and protein A purified these samples of all glycans ranged between 0.352 and 0.845. Glycan ratios in protein A-purified samples and the results of the same statistical approach are shown in Figures S2–S4. Although the magnitude of changes differed, the same trends as those with unpurified UPBH were observed through different cultivation periods. The different DO conditions had an impact on the ratio of M7 and M8 in protein A-purified samples only, which was again consistent with the changes in unpurified UPBH samples data.

2.3.4. HCP

The expression levels of host cell proteins (HCPs) in UPBH samples cultured under different conditions, cultivation days from 10 to 17, target pH 6.95 or 7.05 and target DO level at 15% or 40%, using automated small-scale bioreactors were analyzed by DIA-MAM, using a preprepared HCP library. A total of 738 HCPs were detected across all UPBH samples. The peak areas of the HCPs were subjected to multivariate analysis, and the score plot of the principal component analysis (PCA) is shown in Figure .

8.

Score plot of PCA for detected HCP. Dots are labeled (L to R) with pH, DO, and cultivation period.

The plots for each cultivation condition were positioned closely, and differences in HCP profiles were clearly distinguished by principal components with large contributions. The identical HCP profiles from prolonged cultivation periods were separated by the first principal component. Furthermore, the second and third principal components appeared to distinguish differences in dissolved oxygen levels and pH control. The top- and bottom-ten largest contributing loading factors for each component are shown in Table .

4. Proteins Contributing to Each Principal Component.

	Name	PC1
Positive 10	Elongation factor 2	0.99027
	60S ribosomal protein L12-like protein	0.98861
	Calmodulin	0.98263
	Protein disulfide-isomerase	0.97793
	Heat shock cognate protein	0.97655
	Heat shock cognate 71 kDa protein	0.97385
	PRDX1	0.96845
	Protein disulfide-isomerase	0.96717
	Calnexin	0.96601
	LDLR chaperone MESD	0.96456
Negative 10	Lipoprotein lipase	–0.77384
	CMP-N-acetylneuraminate-beta-galactosamide-alpha-2, 3-sialyltransferase	–0.70426
	Chondroitin sulfate proteoglycan 4	–0.68382
	LRPAP1	–0.63894
	GNAT1	–0.63563
	ARSA	–0.60809
	Guanine nucleotide-binding protein G(T) subunit alpha-1	–0.55862
	Elongation factor 1-alpha 1	–0.52166
	TAX1BP1	–0.51772
	Xanthine dehydrogenase	–0.50509

	Name	PC2
Positive 10	14-3-3 protein eta	0.93206
	Mitogen-activated protein kinase	0.92062
	60S ribosomal protein L7	0.92008
	HSP90AA1	0.91803
	SRI (Fragment)	0.91532
	14-3-3 protein theta	0.91088
	Sorcin-like protein	0.90914
	60S ribosomal protein L27a	0.89535
	RPL7	0.89151
	Isoleucyl-tRNA synthetase (Fragment)	0.88577
Negative 10	RAB14 (Fragment)	–0.89066
	Phospholipase B-like 2 (PLBL2)	–0.88445
	ADP-sugar pyrophosphatase	–0.85046
	SRM (Fragment)	–0.83568
	Hydroxymethylglutaryl-CoA synthase, cytoplasmic	–0.82619
	Protein SET GN = I79_020385	–0.82343
	Protein SET GN = I79_002238	–0.82202
	Protein SET (Fragment)	–0.82102
	Glyceraldehyde-3-phosphate dehydrogenase	–0.78712
	GAPDH	–0.78653

	Name	PC3
Positive 10	StAR-related lipid transfer protein 13	0.88027
	PGM1	0.8789
	26S proteasome non-ATPase regulatory subunit 4	0.86208
	Elongation factor 1-gamma	0.83902
	Keratin, type II cytoskeletal 8	0.82885
	HMGB1	0.81209
	Aldehyde oxidase (Fragment)	0.80662
	High mobility group protein 1	0.78579
	60S ribosomal protein L27	0.77903
	60S ribosomal protein L30	0.77568
Negative 10	Dolichyl-diphosphooligosaccharide-protein glycotransferase (OST)	–0.5402
	Histone H2A (Fragment)	–0.45118
	40S ribosomal protein S27	–0.40899
	MICOS complex subunit	–0.40087
	Procollagen-lysine 5-dioxygenase	–0.39076
	Ran GTPase-activating protein 1	–0.38832
	ZMYND19	–0.37845
	SORD (Fragment)	–0.36842
	Putative ATP-dependent RNA helicase DHX30	–0.36804
	HSP90AA1	–0.36578

For the first principal component, sialyltransferase was noted to have a significant negative contribution. Some heat shock-related proteins made significant contributions. One reported high risk HCP, protein disulfide isomerase, was listed in the top 10 of first principal component. The linearity of each HCP for the top- and bottom-five HPCs listed in component one to three from 10% to 300% of mAb protein load compared to the nominal condition (Figure S5). The coefficient of determination of these 30 proteins ranged from 0.43 to 1.00, and the average was 0.93.

Several endoplasmic reticulum (ER) stress-related proteinssuch as protein disulfide-isomerase, heat shock cognate proteins, and oligosaccharyltransferase (OST)were identified among the proteins with high contributions in the principal component analysis (Table ). Correlation analysis was conducted between these proteins and various cultivation parameters, including cell viability, IgG titer, and major PTMs such as methionine oxidation and aglycosylation. Strong negative correlations were observed between cell viability and the levels of calnexin, heat shock cognate proteins, protein disulfide-isomerase, and peroxiredoxin-1 (PRDX1) (Figure ). In contrast, these proteins showed positive correlations with aglycosylation levels. No significant correlations were detected between any of these proteins and either IgG titer or methionine oxidation (Table S5).

9.

Correlation analysis between ER stress-related proteins and viability or aglycosylation rate.

3. Discussion

In this study, we developed a di-MAM workflow which enables the simultaneous evaluation of PTMs and HCPs in UPBH samples. We first constructed PTM libraries and demonstrated that PTM analysis at the UPBH stage provides valuable information on antibody stability. For the construction of the PTM library, we selected the most intense charge states, considering that UPBH samples or highly stressed samples may show a wide variety of PTM levels and inconsistent detectable charge states.

Stressed UPBH samples exhibited markedly higher PTM levels than purified antibodies, even after buffer exchange into the same formulation. This suggests that matrix components in UPBH, such as HCPs, DNA, and cell debris, may accelerate modifications beyond buffer-related effects. Moreover, the loss of minor glycan species in heat-stressed UPBH samples likely reflects antibody instability, resulting in precipitation or degradation during stress treatment or sample preparation. Since some hold time at the UPBH stage is inevitable during manufacturing, these findings indicate the need to monitor and minimize degradation under upstream conditions. Our results reveal that MAM can directly capture such PTM liabilities in UPBH samples.

For HCP evaluation, we constructed a DIA-based quantification library using DDA data from UPBH. The number of reproducibly quantified HCPs was comparable to previous reports. However, fewer HCPs were detected in small-scale automated cultures than in large-scale production runs, which likely reflected scale-dependent differences in HCP expression. Although library-free DIA methods have been widely applied in proteomics, we show that a library-based approach improves quantitative repeatability and robustness, which is particularly important for monitoring process consistency. In the application study for HCP evaluation of the sample from different cultivation conditions. The reduced level of phospholipase B-like 2 (PLBL2) in cultures harvested at day 14 was observed. PLBL2 is a well-studied immunogenic HCP which is typically monitored in purified products. Our results indicate that its relative abundance to antibody varied with cultivation period, suggesting that culture duration could be optimized not only based on productivity metrics (titer, viability, doubling time) but also with consideration to patient safety–relevant HCPs. This, in turn, suggests the patient-centric perspective enabled by dual integrated MAM.

The dual PTM–HCP analysis provided deeper insights into the relationship between cell conditions and mAb quality. Principal component analysis revealed that several endoplasmic reticulum (ER) stress–related proteins, such as protein disulfide isomerase, calnexin, and dolichyl-diphosphooligosaccharide–protein glycotransferase (OST), were major contributors to the differentiation of samples cultivated under different conditions. In addition, peroxiredoxin 1, a key protein involved in maintaining the intracellular redox balance, was also identified as a high-contributing protein.

ER stress and redox imbalance are known to cause protein aggregation, increases in aglycosylated or misfolded proteins, and eventually apoptosis. − Consistent with these mechanisms, we observed strong negative correlations between ER stress-related protein abundance and cell viability, and positive correlations with aglycosylation levels (Figure ). These results suggest that decreased cell viability, likely associated with apoptosis, induces ER stress-related protein expression. Consequently, monitoring these stress-related proteins could provide a useful indicator for maintaining appropriate glycosylation and protein folding during production.

In contrast, no correlation was observed between ER stress-related proteins and methionine oxidation (Table S5), suggesting that the increase in oxidation observed at longer cultivation times may result from heat or chemical stress rather than ER stress or redox imbalance within the cell. We believe that this ability to generate such insights is considered the true significance of di-MAM which allows simultaneous evaluation of PTMs and HCPs.

Importantly, HCP evaluation at the UPBH stage should not be interpreted as predictive of residual HCP levels in purified drug substance. Instead, it provides actionable insights to guide both upstream process adjustments and downstream purification strategies. Moreover, as cell line engineering advances toward reducing high-risk HCP expression, di-MAM offers a powerful feedback tool for linking cell biology, process conditions, and product quality. We propose utilizing di-MAM for process understanding and monitoring processes, in conjunction with previously established HCP proteomics platforms that enable absolute quantitation of HCPs in final products, thereby allowing the establishment of acceptance criterion for each HCP.

In conclusion, we established an MAM approach that enables concurrent, library-based quantitation of PTMs and HCPs in UPBH samples without purification steps. While methods for PTM evaluation and HCP profile monitoring have traditionally been developed and applied independentlyprimarily for final product analysisour integrated approach provides a unified framework for investigating both critical quality attributes (CQAs) within a single workflow. This integration can significantly reduce the time and cost associated with complex mass spectrometry analyses during process development and monitoring. Furthermore, broader application of this method may elucidate the interplay among cultivation conditions, antibody quality attributes, and HCP expression. By accelerating process understanding and decision-making, di-MAM represents a valuable advance for bioprocess development and quality evaluation.

4. Materials and Methods

4.1. Materials

The following reagents and materials were utilized: 2-amino-2-hydroxymethyl-1,3-propanediol (Tris), formic acid (FA), and 0.1 mol/L hydrochloric acid were purchased from Fujifilm Wako (cat. 207-06275, 067-04531, and 7647-01-0, respectively). An 8 M guanidine-HCl solution was obtained from ThermoFisher (cat. 24115). Dithiothreitol (DTT) and sodium iodoacetate (IAA) were purchased from Nacalai Tesque (cat. 14112 and 19305-24, respectively). Trypsin/Lys-C Mix was obtained from Promega (cat. V5073). Trifluoroacetic acid (TFA) and acetonitrile were obtained from Kanto Chemical (cat. 40578-30 and 01031-2B, respectively). Dialysis cassettes (Slide-A-Lyzer G2) for buffer exchange of UPBH samples were purchased from ThermoFisher Scientific.

4.2. Monoclonal Antibody

In this study, a Chinese Hamster Ovary (CHO)-expressed IgG2 antibody produced by Astellas Pharma Inc. was used. UPBH samples and final purified mAb were collected from a manufacturing-scale system. These samples were utilized for library construction and a forced degradation study.

4.3. Thermal Stress

To produce thermal stressed samples for both the mAb and UPBH, samples were incubated for 8 weeks at 40 °C. A buffer exchanged UPBH sample, adjusted to the same formulation buffer as the purified mAb (10 mM sodium succinate, 5% sorbitol, pH4.5), was also incubated under the same conditions.

4.4. Photo Stress

To produce photo stressed mAb, UPBH and buffer exchanged UPBH, samples were exposed to a D65 lamp until total illumination of 1208.6 lx-hours at 25 °C. Light-protected samples covered by aluminum foil were also stored under the same conditions.

In the forced degradation study, both the mAb and UPBH samples were subjected to thermal stress (40 °C for 8 weeks) and photostress conditions using a D65 lamp (total illumination of 1208.6 lx-hours at 25 °C). Additionally, a buffer-exchanged UPBH sample, adjusted to the same formulation buffer as the purified sample, was subjected to degradation under the same conditions.

4.5. Automated Cultivation System

Small-scale cultivation experiments were conducted using an automated microbioreactor system (Ambr 15 Cell Culture System, Cytiva) to evaluate various cultivation conditions using the same cell line. To evaluate the impact of cultivation parameters on PTM level and HCP profile, the following parameters were intentionally varied: harvest days (10, 14, and 17 days), dissolved oxygen level control (15% and 40%), and pH control (6.95 and 7.05). These combinations of parameters are summarized in Table S1. Two biological replicates were prepared.

The following reagents and materials were utilized: 2-amino-2-hydroxymethyl-1,3-propanediol (Tris), formic acid (FA), and 0.1 mol/L hydrochloric acid were purchased from Fujifilm Wako (cat. 207-06275, 067-04531, and 7647-01-0, respectively). An 8 M guanidine-HCl solution was obtained from ThermoFisher Scientific (cat. 24115). Dithiothreitol (DTT) and sodium iodoacetate (IAA) were purchased from Nacalai Tesque (cat. 14112 and 19305-24, respectively). Trypsin/Lys-C Mix was obtained from Promega (cat. V5073). Trifluoroacetic acid (TFA) and acetonitrile were obtained from Kanto Chemical (cat. 40578-30 and 01031-2B, respectively). Dialysis cassettes (Slide-A-Lyzer G2) for buffer exchange of UPBH samples were purchased from ThermoFisher Scientific.

4.6. Protein a Purification

Protein A purification for UPBH samples was conducted under the following conditions. 3–5 mg of protein was loaded on a protein A affinity column (POROS, A20 4.6 × 50 mm, ThermoFisher Scientific). After short gradient to wash out medium components and impurities, column-retained antibody was eluted by longer gradient. Each gradient program is shown in the supplemental table. For mobile phase A, 10 mmol/L sodium acetate buffer including 100 mmol/L of sodium chloride was used. For mobile phase B, 40 mmol/L of sodium acetate, pH3.7 was used. After elution, antibody solution was neutralized by 1 mol/L of Tris-HCl pH8.0 from Nippongene.

4.7. Sample Preparation

The purified sample was diluted to a concentration of 2 mg/mL using a denaturing buffer composed of 7 mol/L guanidine-HCl and 100 mmol/L Tris-HCl (pH 8.4). For UPBH samples with protein concentrations lower than 4 mg/mL, a 2-fold dilution was consistently performed to achieve the desired conditions. To reduce disulfide bonds, 10 μL of 1 mol/L dithiothreitol was added to 2100 μL of the sample, followed by incubation at room temperature for 30 min. Next, 10 μL of 1 mol/L sodium iodoacetate was added, and the mixture was incubated at room temperature for 25 min, protected from light, to alkylate the reduced cysteines.

Buffer exchange was performed using a Zeba spin desalting column (ThermoFisher Scientific) to replace the denaturing buffer with a digestion buffer (50 mmol/L methionine, 50 mmol/L Tris-HCl). Trypsin/Lys-C Mix was then added at an enzyme-to-substrate ratio of approximately 1:10. The mixture was incubated at 37 °C for 1 h to facilitate digestion, and the reaction was quenched by adding 5% trifluoroacetic acid.

4.8. LC/MS/MS

Digested samples were separated using a Waters ACQUITY Premier BEH C18 column (1.7 μm, 2.1 mm × 100 mm) on a Vanquish or Ultimate 3000 (ThermoFisher) system. The aqueous mobile phase (A) consisted of 0.1% formic acid (FA), while the organic mobile phase (B) was acetonitrile with 0.1% FA. The following gradient program was employed: an isocratic flow at 2% B for 2.5 min, a linear increase to 40% B over 65 min, an isocratic flow at 90% B from 70 to 80 min, and re-equilibration at 2% B until 90 min.

The LC system (Vanquish or Ultimate 3000, ThermoFisher) was coupled to an Orbitrap Exploris 480 or Fusion Lumos or Exploris 480 mass spectrometer. For data-dependent acquisition (DDA), precursor m/z range was set from 400 to 1600. Ion source parameters were as follows: spray voltage, 3500 V; ion source temperature, 350 °C; Orbitrap resolution, 120,000 for precursor and Full MS scans and 30,000 for MS/MS scans.

Simultaneous MS1 and data-independent acquisition (DIA) were performed using the same MS1 m/z range and an Orbitrap resolution of 360,000. For DIA, overlapping 5 m/z isolation windows were used with a precursor range of 400–1200 m/z and 100 m/z isolation window width.

For PTM evaluation, PTM libraries were prepared using BioPharmaFinder (ver. 5.1, ThermoFisher) to identify mAb-derived peptides. Peptide information associated with PTMs was then transferred to Chromeleon (ver. 7.3.1, ThermoFisher) to create a PTM library for quantitation. Modification ratio was calculated using the following formula. Since multiple PTMs were identified within peptides of the same sequence length, the sum of the extracted ion chromatograms was used as denominator.

Modification ratio (%) = (XIC area of specific modified peptide)/(Sum of XIC area of peptides with same length × 100). For HCP, Proteome Discoverer (ver. 2.4.1.15, ThermoFisher) was utilized to identify the results of DDA analysis of UPBH samples. Protein data annotated with the taxonomy Cricetulus griseus from UniProt were used to construct the HCP library. HCP quantitation was performed with Skyline (ver. 22.2, MacCoss Lab).

All statistical analyses were carried out using JMP (ver. 18.0.1, JMP Statistical Discovery LLC).

4.9. HCP Linearity Evaluation

To confirm the linearity of DIA-based HCP evaluation, all UPBH samples from the Ambr 15 system were pooled and processed according to the sample preparation protocol described above. After tryptic digestion, the samples were concentrated using centrifugal evaporation to increase peptide concentration. The dried samples were then reconstituted in water. Column loading was adjusted based on the mAb concentration, ranging from 10% to 300% of the nominal load.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c11877.

• Table S1: cultivation conditions using an automated cultivation system; Table S2: PTM comparison between UPBH samples and protein A-purified UPBH samples (oxidation, deamidation, succinimide (NH₃ loss), glycation); Table S3: PTM comparison between UPBH samples and protein A-purified UPBH samples (glycosylation); Table S4: gradient programs for protein A purification; Table S5: correlation analysis among HCPs, cultivation performances and PTMs; Figure S1: correlation of PTM values between purified/unpurified (harvest) UPBH sample; Figure S2: PTM levels (%) in cultivation samples on days 10, 14, and 17; Figure S3: PTM levels (%) in cultivation samples at target pH 6.95 and 7.05 (purified); Figure S4: PTM levels (%) in cultivation samples at target dissolved oxygen 15% and 40% (Purified); Figure S5: linearity assessment of detected HCPs (PDF)

This study was funded by Astellas Pharma Inc.

The authors declare no competing financial interest.