Development and implementation of an LC-MS-based multi-attribute method for adeno-associated virus

Abstract

The multi-attribute method (MAM), a mass spectrometry technique for quantifying amino acid modifications at the peptide level, is becoming a prominent analytical tool in the development of biotherapeutics. The method has promise for adeno-associated virus (AAV) therapeutics, where capsid protein modifications have been directly linked to reduced transduction efficiency. Given this link, a robust and precise procedure to quantitate capsid modifications would be beneficial for implementation throughout biotherapeutic development. Herein, an AAV product was characterized, and capsid sequence liabilities were identified. A peptide map MAM method was developed to quantitate select sites of modifications and was validated according to ICH Q2(R2). Through this exercise, the method was demonstrated to be suitable to quantitate several sites of deamidation and the method was applied during stability, process development, and product comparability studies. Additionally, preliminary data demonstrated that the method was not limited to monitoring deamidation but also could be applied to other post-translational and chemical modifications.

Graphical abstract

Powers and colleagues have developed an AAV MS-based multi-attribute method (MAM) to quantitate deamidation, an important quality attribute for AAV therapeutics. The method was validated according to ICH Q2(R2) and implemented for routine testing of AAV for development and stability purposes.

Introduction

Recombinant adeno-associated virus (rAAV) therapeutics are an emerging class of biotherapeutics. The AAV vector consists of a small icosahedral capsid composed of 60 capsid proteins, VP1, VP2, and VP3, and a therapeutic single-stranded DNA transgene up to 4.7 kilobases.¹ The relative abundance of VP1:VP2:VP3 varies but is often reported as approximately 1:1:10, with the level of VP1 and VP2 dependent on factors such as the production system.^1,2 rAAV vectors utilize the same capsid components and structure as wild-type AAV; however, the wild-type protein coding sequences in the transgene are replaced with a therapeutic gene expression cassette.

In general, the use of mass spectrometry (MS) to characterize rAAV vectors has emerged as a useful tool to gain important information on rAAV quality attributes. MS has been used to characterize the assembled AAV capsid, capsid protein subunits, and proteolytic digested AAV capsids. Charge detection mass spectrometry (CDMS) and/or high-resolution mass spectrometry have been leveraged to determine the mass of the AAV capsid, the ratio of empty-to-full capsids, and obtain information about the packaged genome.^1,3,4,5,6,7 MS analysis of the individual capsid proteins has been used to support serotype identity testing, VP ratio, clips, alternative processing sites, and post-translational modifications.^8,9,10 Laboratories have performed proteolytic digestion followed by liquid chromatograph tandem MS (LC-MS/MS) to detect, identify, and quantify levels of host cell proteins.^11,12,13 Similar workflows have identified various capsid protein modifications, including N-terminal modifications, deamidation, phosphorylation, isomerization, and oxidation, some of which increase under stressed conditions.^14,15,16,17 These modifications, such as deamidation and phosphorylation, can impact AAV transduction efficiency.^17,18 Efficient monitoring of these modifications can enhance product and process development for rAAV therapeutics.

The multi-attribute method (MAM), formally introduced by Rogers et al. in 2015, is an MS-based approach to simultaneously monitor multiple product quality attributes (PQAs).¹⁹ At a high level, MAM enables divergent analytical procedures directed at monitoring multiple quality attributes to be consolidated into a single, MS-based analytical method. The initial version of MAM, a peptide map LC-MS method, has been optimized to enable the use of multiple enzymes for sequence-dependent digestions, the use of multiple columns and gradients to enable optimized chromatographic separations, automated digestion for higher throughput analysis, and online process monitoring for real-time process feedback.^{20,21,22,23,24,25,26} Although MAM has traditionally been applied for the analysis of peptides from monoclonal antibodies, the approach has been adapted to support broader applications, including antibody subunit analysis and intact mass analysis.^27,28,29 Furthermore, the approach has been applied to other biotherapeutic classes, including antibody drug conjugates (ADCs) and fusion proteins.^30,31,32,33

MAM has been implemented throughout the biotherapeutic life cycle, from development to product release in QC laboratories.³⁴ To support research and development, Dong et al. developed an automated protein A purification followed by LC-MS analysis to measure PQAs during the fermentation process.³⁵ Similarly, MAM has supported PQA understanding during cell culture process development, scale-up, and technology transfer.^36,37 In QC settings, MAM has been implemented to support release and stability testing, both for identity testing and PQA quantitation.^38,39 Method validation exercises have been detailed previously and include demonstration of specificity, accuracy, linearity, precision, limit of quantitation (QL), range, and robustness.^40,41

Given the utility of MAM across biotherapeutics, and the structural complexity of rAAV in general, the technology would be beneficial during rAAV therapeutic development. The complexity of the AAV capsid, due to the presence of multiple capsid proteins and various abundances, challenges traditional chromatographic separation methods for the quantitation of modifications. The MS component of the MAM procedure provides an additional degree of separation that enables quantitation of low abundant species in complex chromatographic profiles. Furthermore, given the relationship between certain modifications and transduction efficiency, MAM could be implemented to provide inferences on product potency, where traditional assays are complicated and time-consuming. Herein, we document the initial characterization of a rAAV product and identify molecular hotspots, residues prone to modification, pertinent for routine monitoring. An MAM method was developed to monitor these hotspots and validated according to ICH Q2(R2), demonstrating the method was precise and suitable for its intended purpose. Once developed, the method was implemented to support process development, ensure product stability, and demonstrate analytical comparability. While the method was validated for a subset of modifications, proof-of-concept data demonstrate that additional attributes could also be monitored for characterization purposes during development.

Results

LC-MS/MS characterization of peptides from an rAAV9 product

The primary structure and post-translational modifications of an rAAV therapeutic were analyzed using reduced trypsin/Lys-C peptide mapping and LC-MS/MS (as described in the LC-MS/MS characterization method section). This analysis aimed to identify any modifications, particularly those changing on stability. The MS data corresponding to peaks in the UV chromatogram were thoroughly characterized and annotated for peak assignment (Figure 1A). Overall, the method used a combination of trypsin and trypsin/lysC, at an appreciable enzyme:substrate basis, to account for the fact that AAV has been characterized as resistant to proteolytic digestion.^13,42 Peptides are labeled according to their positions in the VP1, VP2, and VP3 sequences. “T” represents a tryptic cleavage, and the number represents the numerical order of the peptides from an in silico digest starting with the N terminus. For example, T1 is the des-methionine VP1 N-terminal protein representing amino acids 2–20 in this specific rAAV, while T64 denotes the final two amino acids, 735–736, in this rAAV sequence, forming the C-terminal tryptic peptide). After analysis, the detected proteolytic peptides accounted for 99.6%, 99.7%, and 99.6% sequence coverage of the respective VP1, VP2, and VP3 capsid proteins. Figure 1A demonstrates that very little overdigestion, or digestion following sites that were not lysine or arginine, occurred. In fact, even with the enzyme:substrate ratio, some underdigestion (or missed cleavage) was still observed.

Figure 1.

Annotated UV chromatogram (214 nm) for trypsin-digested AAV9 capsids

(A) The trypsin/Lys-C peptide map UV (214 nm) chromatogram of a rAAV product following capsid denaturation and digestion. (B) and (C) The peptide map UV (214 nm) chromatograms, performed on a different day than (A), of an rAAV product following thermal stress at 25°C for T = 0 w and T = 4 w. “T” represents tryptic peptides, “∗” represents system peaks including the trypsin auto-digestion products, “ˆ” represents N-terminal, acetylated, des-methionine peptide of VP1 and VP3, “^‡” represents des-threonine N-terminal peptide of VP2, “Δ” represents des-Thr, acetylated methionine N-terminal peptide of VP2, “†” represents des-methionine N-terminal peptide of VP3 without acetylation, “d” represents deamidation, “P” represents phosphorylation, “iso” represent isomerization, “$” represents a putatively disulfide linked peptide due to incomplete reduction, and “suc” denotes succinimide.

To investigate potential sequence liabilities and degradation pathways of the rAAV, the material was subjected to stress at 25°C for 4 weeks. Both the unstressed (control) and stressed materials were analyzed by LC-MS/MS. A comparison of the UV chromatograms (214 nm) is presented in Figures 1B and 1C. Although the thermally stressed and control materials appeared largely similar based on the UV data (Figures 1B and 1C), the LC-MS/MS identified several peptide modifications increased in abundance following thermal stress exposure (Table 1). The observed modifications that were present in the material or changed on stability included N-terminal processing on the capsid proteins, deamidation, succinimide formation, phosphorylation, isomerization, and oxidation (Table 1). Notably, many of these modifications were unique to the less abundant VP1 or VP2 capsid proteins.

Table 1.

Identification and quantitation of modifications on the rAAV product

Capsid protein	Peptide and residue	Relative abundance (%)
		T = 0	T = 4w 25C
VP1	Isomerization at T1ˆ (D4, VP1 N-term)	1.4	7.8
	Deamidation at T6 (N57)	10.7	48.2
	Deamidation at T10 (N94)	2.1	13.4
VP1, VP2	Phosphorylation at T20 (S149)	9.4	9.3
	Oxidation at T24 (M203)	2.4	10.9
VP1, VP2, VP3	Deamidation at T35 (N329)	3.7	8.9
	Oxidation T37 (M404)	0.9	1.3
	Deamidation at T40 (N452)	8.5	28.8
	Succinimide at T40 (N452)	1.4	1.4
	Deamidation at T45 (N515 and 519)	0.9	0.9
VP3 (204–238)	Isomerization at T24ˆ (D219, VP3 N-term)	2.2	5.9

Only select modifications that were associated with higher abundances and/or changes on stressed conditions were labeled in Figure 1. The relative abundance was computed by dividing the mass spectrometry peak area of the modified peptide by the total peak area of the peptide. Relative abundances for each modification were derived from a single injection.

Based on the MS data, modifications that increased include Asp4 (D4) isomerization, Asn57 (N57) deamidation, Asn94 (N94) deamidation, Asp219 (D219) isomerization, Met203 (M203) oxidation, and Asn452 (N452) deamidation. Among all modifications, deamidation increased the most in terms of total precent relative abundance, where three sites had increases of more than 10% deamidation.

Generation of an MAM method for select PQA monitoring

The product-specific peptide map characterization results showed degradation through asparagine 57, 94, and 452 deamidation (denoted as N57, N94, and N452 deamidation throughout). These sites were particularly of interest based on both the characterization data and prior literature. Giles et al. demonstrated a significant reduction in relative potency as a result of N57 and N94 deamidation in AAV8 capsids.¹⁷ Similarly, He et al. demonstrated a correlation of N57 deamidation and decrease in both expression and activity for an alternate serotype.⁴³ Given this relationship, an MAM method was developed for the automated calculation of asparagine deamidation on these specific amino acid residues. This MAM method (as described in the sample preparation and analysis for the multi-attribute method section) was used for all subsequent MS data presented herein. The MAM procedure used a higher enzyme/substrate ratio compared with the characterization method to minimize missed cleavages observed on the N57, N94, and N452 peptides.

MS peak areas were computed at the peptide level and the relative abundance was computed by dividing the MS peak area of the peptide containing the deamidated residue by the total peak area of the peptide. The total peak area included signal from the deamidated, unmodified, and succinimide modified forms, as applicable. The processing method is described in Table S1 and examples of integrations are shown in Figure S1. In general, the succinimide species, when included, contributed only a small amount of signal and are not shown in Figure S1.

Robustness experiments were initially performed to assess the stability of the digested sample prior to analysis. Digested samples were kept at 2–8°C or −60°C for a defined time to determine allowable storage times at those conditions. At 2–8°C over the course of 46 h, N94 and N452 deamidation showed a slight upward trend, 2.0%–3.7% respectively, in the relative abundance of deamidation, suggesting sample storage at this temperature should be limited (Figure S2A). These data suggest the time a prepared sample should be stored in the autosampler (2–8°C) should not exceed 24 h. Although some change was observed during this duration, this time frame was selected as the magnitude of the change was acceptable and within expected method variability. When the bulk prepared sample was stored at −60°C, no trend of increased deamidation was observed (Figure S2B). These data suggest that the sample can be stored frozen at −60°C if prolonged time is needed between preparation and analysis. No prepared samples were stored longer than 7 days at −60°C, but it is possible that longer storage times at this condition would be feasible.

In addition to sample stability, experiments were performed to understand the procedure robustness across various protein injection loads and MS intensities. Toward this aim, a single sample was injected from 10% to 200% of the target injection level (Figure 2). These data demonstrated that there is no difference in the attribute relative abundance as the sample load changes. Furthermore, the TIC signal of the injection (Figure 2B) and MS area (Figure 2C) from selected components showed linearity with respect to the injection load. Collectively, these data illustrate that the method is robust for the use of calculating attribute relative abundances within the normal operating range of the procedure.

Figure 2.

Assessment of method characteristics across different injection levels

(A) The % deamidation value obtained for N57, N94, and N452 from 10% to 200% of the target injection load (1 μg). (B) The TIC signal obtained from the 10% to 200% of the target injection loads. (C) The component peak area obtained from the 10% to 200% of the target injection loads for the unmodified N57 peptide (diamonds) and the unmodified N452 peptide (circles). All datapoints represent single injections.

Based on these data, system suitability (SST) and assay acceptance criteria (AC) were established for a product-specific control sample to ensure the LC-MS system was operating as intended and the sample preparation for the control was in line with historical experience (Table S2). As all injection loads from 10% to 200% of the target sample load enabled consistent quantitation of the targeted sites of deamidation, TIC signal and component areas from the lowest injection level were set as the minimum values for these parameters. Additional criteria were established for mass accuracy, component retention time, proper cleavage rate, and deamidation levels. Mass accuracy and allowable retention times were established based on the processing method design. The proper cleavage level and abundance of modifications were based on historical control sample data. Sample suitability (SS) criteria were designed to make sure the digestion and analysis of the individual samples was in line with method expectations.

Validation of the MAM procedure

The analytical procedure was validated in accordance with International Council for Harmonization (ICH) Q2(R2) guidelines for precision, accuracy, linearity, specificity, QL, and range. A summary of the validation results is presented in Table 2. Sample A represents unstressed rAAV material. To generate variable attribute relative abundance levels, material was stressed (Sample E) and mixed with unstressed material at various levels (Samples B–D), as described in the materials and methods section. Lower injection volumes (Samples F and G) were performed to assess performance for modification levels below the unstressed material, as described in the materials and methods section.

Table 2.

Analytical results from validation of the MAM procedure

Validation characteristic	Experimental design	N57 deamidation	N94 deamidation	N452 deamidation
Repeatability (RSD)	Tested at three levels in a single test instance with three independent sample preparations for each level.	Sample A = 1.3%	Sample A = 4.3%	Sample A = 0.7%
		Sample C = 1.5%	Sample C = 1.8%	Sample C = 0.8%
		Sample E = 1.0%	Sample E = 1.6%	Sample E = 1.0%
Intermediate precision (RSD)	Six independent assay instances of testing of seven sample-level preparations over the range of the assay. Testing was performed by two analysts using two instruments and two analytical columns in one laboratory.	Sample A = 2.4%	Sample A = 5.4%	Sample A = 3.8%
		Sample B = 2.0%	Sample B = 7.8%	Sample B = 1.3%
		Sample C = 2.0%	Sample C = 5.5%	Sample C = 2.4%
		Sample D = 1.4%	Sample D = 6.5%	Sample D = 1.3%
		Sample E = 1.1%	Sample E = 5.5%	Sample E = 1.4%
		Sample F = 2.9%	Sample F = 12.8%	Sample F = 7.3%
		Sample G = 1.9%	Sample G = 7.2%	Sample G = 4.9%
Accuracy (recovery)	Accuracy was evaluated using data generated from the assessment of seven sample levels in the intermediate precision testing. The accuracy was computed by comparing the numerical result observed to the theoretical deamidation level.	Sample A = 106.4%	Sample A = 109.4%	Sample A = 95.7%
		Sample B = 104.5%	Sample B = 100.0%	Sample B = 96.6%
		Sample C = 100.9%	Sample C = 93.7%	Sample C = 97.0%
		Sample D = 102.1%	Sample D = 94.7%	Sample D = 97.9%
		Sample E = 102.7%	Sample E = 97.5%	Sample E = 98.9%
		Sample F = 109.7%	Sample F = 116.7%	Sample F = 96.9%
		Sample G = 107.7%	Sample G = 100.0%	Sample G = 90.1%
Linearity	Linearity was evaluated using data generated from the assessment of seven sample levels in the intermediate precision testing. The theoretical deamidation level (x axis) of each sample level against the mean measured deamidation level (y axis) is plotted to evaluate linearity.	1. Visual inspection of plot appears linear. and 2. R2 = 1.00	1. Visual inspection of plot appears linear. and 2. R2 = 1.00	1. Visual inspection of plot appears linear. and 2. R2 = 1.00
Quantitation limit and detection limit (attribute abundance)	The quantitation limit (QL) was set using the deamidation level from the lowest injection volume sample that met the precision criteria.	QL = 3.1%	QL = 0.6%	QL = 3.2%
Range (attribute abundance)	The range was established using the data from intermediate precision, accuracy, and linearity assessments. The range was established for each deamidation site based on the lowest and highest deamidation value (and all points in-between) that met the precision, accuracy, and linearity criteria.	3.1%–48.7%	0.6%–15.8%	3.2%–37.2%
Specificity	Specificity was evaluated using formulation buffer. The component areas from the formulation buffer were compared with the mean Sample A component areas from the intermediate precision testing.	Sample H (FB) Component Area = 0.0% of Sample A Component Area	Sample H (FB) Component Area = 0.0% of Sample A Component Area	Sample H (FB) Component Area = 0.0% of Sample A Component Area
Prepared sample stability	Prepared sample stability was assessed by assessing one sample level post-digestion and stored at ≤ −60°C for 1, 3, and 7 days and at 2°C–8°C for 24 h.	2°C–8°C: NLT 24 h	2°C–8°C: NLT 24 h	2°C–8°C: NLT 24 h
		≤ −60°C: NLT 168 h (7 d)	≤ −60°C: NLT 168 h (7 d)	≤ −60°C: NLT 168 h (7 d)

Sample Description: Sample A represents unstressed rAAV material and sample E represents thermally stressed AAV material. Samples B-D represent various mixtures of Samples A and E. Samples F and G are lower injection levels of Sample A. These specific mixtures are described in more detail in the materials and methods section.

The method validation demonstrated the procedure was specific and robust. Robustness and stability data are presented in Figures S2 and 2, and show the digested material is generally stable, and the method is reproducible across a range of injected materials. Specificity was demonstrated by comparing Sample A with a formulation buffer injection, where the formulation buffer had ≤0.1% of the component area of the Sample A result for each site of deamidation. The QL was established for each deamidation site based on the lowest deamidation value that had suitable precision and accuracy (Table 2). The QL and range were established for each deamidation site based on the lowest and highest deamidation value (and all points in-between) that met the precision, accuracy, and linearity criteria.

Validation testing also demonstrated the method to be precise. Repeatability testing was performed to demonstrate instrument precision, as the same sample was injected three times on the same instrument on the same day at three levels. During repeatability, the quantitation of deamidation at all sites had a relative standard deviation (RSD) of ≤5%, with one instance being 4.3% and all other results were less than 2.0% RSD (Table 2). Intermediate precision testing was performed to monitor the overall assay precision when performed by multiple analysts on multiple systems, mimicking how the assay would be performed over time to support the development of an AAV therapeutic. For N57 and N452 deamidation, all intermediate precision RSD values were less than 10.0%, with nearly all values being well below 5.0%. N94 deamidation had a higher RSD, ranging from 5.4% to 12.8% RSD. This was not surprising given the lower relative abundance of N94 deamidation compared with the other two deamidation sites (Figure 3) and the known impact of autosampler stability on this site of deamidation. All individual values that went into the intermediate precision calculation are shown in Figures 3D–3F.

Figure 3.

Demonstration of the MAM method linearity and intermediate precision

Linearity data from N57 (A), N94 (B), and N452 (C) deamidation relative abundances are provided across samples A–G. Samples A–G represent samples with different deamidation levels, as described in the materials and methods section - samples and validation. Each linearity datapoint represents the average across six independent datapoints. To understand the individual measurements within the linearity datapoints, each individual measurement (from the intermediate precision set of experiments) is in (D)–(F), respectively.

Accuracy and linearity were demonstrated by using the intermediate precision testing values. The test method was shown to be accurate, with all levels across the three sites of deamidation demonstrating between 90% and 120% accuracy (Table 2). Sample F for N94 deamidation was the only sample that had an accuracy outside of 90%–110% accuracy. The accuracy was computed by comparing the numerical result observed with the theoretical deamidation level. As with accuracy, linearity was demonstrated for each site of deamidation, with all sites of deamidation having an R-square of 1.00 (Figures 3A–3C).

Implementation of the MAM procedure on forced degradation and stability studies

The MAM method was tested on stability samples to understand deamidation level at the long-term stability condition (−70°C), accelerated stability condition (2–8°C), and forced degraded conditions (25°C). At both long-term and accelerated conditions, the deamidation values remained largely consistent at all time points tested (Figures 4A and 4B). In contrast, when applied to forced degraded materials, all three sites of deamidation increased (Figure 4C). In addition to MAM, relative potency results were assessed for all time points and conditions. The relative potency value is obtained by dividing the potency measurement of a given sample by the measurement of a reference sample. Thus, as the sample in Figure 4C goes from approximately 100%–40% relative potency over 4 weeks at 25°C, and exhibits a consistent trend of reduced relative potency across time points, the sample is markedly less potent because of the thermal degradation. Given the mechanism of action of the rAAV therapeutic, the potency procedure is subject to larger levels of analytical variability. As such, it is not uncommon to see oscillations in values like those observed at the 18- and 30-month time points in Figure 4A and the 1-month time point in Figure 4B. That said, prior and subsequent time points each revealed similar levels of relative potency to the initial measured value, demonstrating that the long-term and accelerated conditions had no impactful change in relative potency, beyond standard assay variability (Figures 4A and 4B). In contrast, the forced degraded samples did exhibit a decrease in relative potency at each time point which inversely correlated with the increase in deamidation (Figure 4C). These data demonstrate that the MAM procedure is stability indicating, can be applied consistently over time to support stability studies, and results can be correlated with the potency procedure. This is important, as the AAV potency procedures can be lengthy, complex assays and subject to appreciable analytical variability, depending on the format of the assay and mechanism of action of the therapeutic AAV. Given the correlation between asparagine deamidation and relative potency, the MAM procedure represents a valuable tool to implement during stability studies.^17,43

Figure 4.

Analysis and correlation of deamidation and relative potency on stability and forced degradation conditions

(A–C) The deamidation level observed for N57 (black), N94 (gray), and N452 (patterned) under long-term stability conditions at −70°C (A), accelerated stability conditions at 2–8°C (B), and forced degraded conditions at 25°C (C). Relative potency results are graphed with red lines. The relative potency value is obtained by dividing the potency measurement of a given sample by the measurement of a reference sample. MAM data were not collected for the 36-month sample in (A), denoted by ∗. All datapoints represent single measurements for both MAM and relative potency results.

Implementation of the MAM procedure for process development and comparability studies

As deamidation has been inversely linked to potency, it was beneficial to understand the manufacturing process and how changes to the process or allowable manufacturing hold times impacted the level of deamidation. To assess this, the hold times at manufacturing steps were evaluated for three different rAAV lots of material. N57, N94, and N452 are graphed separately in Figures 5A–5C, respectively. Each graph shows the deamidation change between the terminal and initial time point. The data were evaluated to look for trends across deamidation sites and lots for a given manufacturing/purification step. For most manufacturing steps (labeled Affinity Eluate, AEX Pool, VRF Filtrate, UF/DF Concentrate, and DS Sublot), there was not an appreciable increase in deamidation as a result of the hold time. There were some instances where increases were observed (Figure 5B, UF/DF Concentrate) but a firm trend could not be established across lots and all deamidation sites. Likewise, some trends were observed suggesting higher levels of deamidation at each time point for each site of deamidation (Figure 5, VRF Filtrate), but the increase in deamidation was small and not practically meaningful. In contrast, the AEX Diluted Load step had a pronounced effect on deamidation, with all deamidation sites and material lots showing a consistent and large increase. These data suggest that hold times at the AEX Diluted Load step should be minimized if possible, and controlled in the manufacturing process, to yield material that is consistent in the level of deamidation. The increase in observed deamidation at this step was likely due to the temperature or pH at that step.

Figure 5.

Evaluation the impact of manufacturing hold times on deamidation

Three product lots (black = lot 1, gray = lot 2, and patterned = lot 3) were tested at various manufacturing steps and associated hold times (defined on the x axis with the number in parentheses representing the max hold time). The change in deamidation from T0 to the specified step is graphed. Results from N57 (A), N94 (B), and N452 (C) are shown. All datapoints represent single measurements.

In addition to improving hold time understanding, MAM testing was implemented to support an assessment of deamidation across manufactured lots. Herein, MAM was used as a part of an analytical comparability study, with the goal of demonstrating similarity in material manufactured before and after a process change. This was especially important as the pre-change material had been utilized in the clinic, and it was therefore important to demonstrate consistency in product quality attributes of the post-change material. Seven lots of material, four pre-change and three post-change, were analyzed by MAM to assess deamidation (Figure 6). Overall, deamidation in post-change lots was highly consistent to pre-change lots, with better precision in the measure of deamidation observed after the process change.

Figure 6.

MAM implementation for analytical comparability studies

Seven product lots were tested by MAM to demonstrate comparability following a process change. Solid bars represent lots manufactured with the pre-change process and patterned bars represent lots manufactured with the post-change process. All datapoints represent single measurements.

Assessment of the MAM procedure for additional attributes and sites of modification

The decision to validate the method for the quantitation of select deamidation sites was based on internal development experience and the link between deamidation and transduction efficiency.^17,43 While deamidation is an important attribute to monitor for AAV gene therapy products, other modifications may also be of interest for routine monitoring during development. The MAM processing method was adjusted to include additional sites of deamidation, methionine oxidation, isomerization, phosphorylation, and succinimide intermediates and applied to select validation data from Samples A–E (Figure S3).

While additional sites of deamidation were monitored (Figure S3A), the three sites most susceptible to deamidation were the sites that were in scope for the method validation. Deamidation at other sites remained at low levels and did not increase in magnitude as much as the three sites in scope of the validation. Several sites of oxidation did demonstrate a trend of increased oxidation in the validation samples. However, the most pronounced oxidation site exhibited only a small increase, from approximately 3% to 5% (Figure S3B). Isomerization was observed to increase at two sites, both of which were at the N terminus of the capsid proteins for VP1 and VP3 (Figure S3C). While phosphorylation and succinimide were observed, there was no increase observed under the conditions assessed (Figure S3D).

While the MAM method was validated for only three sites of deamidation, these data demonstrate that the method could be used to assess numerous modifications. Monitoring additional attributes can aid in process development and may be required for alternative stress or manufacturing conditions.

Discussion

The ability to routinely monitor AAV capsid deamidation is important for AAV development given the link between capsid protein deamidation and reduced transfection efficiency. Similar principles were applied herein for monitoring AAV deamidation as have been applied across the pharmaceutical industry. One component of a traditional MAM method that was not evaluated in the study herein is new peak detection. New peak detection is often implemented to identify peaks that are new, missing, or have a change in relative abundance to allow the procedure to serve as a purity test.⁴⁴ While new peak detection could have been implemented for the data provided herein, MAM was implemented explicitly to report deamidation of select capsid proteins, and was not implemented with the intent of removing the traditional capsid purity method. Furthermore, implementation of new peak detection can be complex and is still being optimized for biotherapeutics. The challenges of new peak detection may be further exacerbated for AAV, where the lower relative abundances of VP1 and VP2 would make the assessment of new peaks on these capsid proteins challenging without lowering conventional signal thresholds.

While new peak detection was not evaluated, the method was validated to quantitate three individual sites of capsid deamidation per ICH Q2(R2) and was demonstrated to be precise, accurate, linear, suitable for QL and range, robust, and stability indicating (Table 2). To ensure quality analytical data, SST, AC, and SS criteria were established. While the procedure was validated for deamidation at select sites, preliminary data show the method is precise and linear for other attributes (Figure S3). While the method demonstrates strong robustness and precision, it is acknowledged that appropriate controls must be maintained during analysis. For example, the incubation step at 95°C for 5 min was controlled during the digestion procedure, as elevated temperatures can promote deamidation. Although some of the deamidation observed in the manuscript may be attributed to method, the impact was manageable. This is supported by the intermediate precision data, which confirmed the method’s ability to detect deamidation changes reliably when applied to stability and process support studies.

While AAV therapeutics have emerged as a promising class of biotherapeutics, a claim supported by a recent surge in licensure approvals, the field continues to develop. Advances in rAAV production processes are being implemented to enhance product quality, reduce manufacturing costs, and to increase manufacturing yields to meet production demands.^45,46,47 As manufacturing processes evolve and new products continue to emerge, it is crucial to have advanced analytical tools that can quickly and accurately monitor key attributes. MAM is one such promising tool.

Materials and methods

AAV material

AAV materials used in this study are as follows. All materials used in the study are recombinant AAV (rAAV) and are referred to as AAV for simplicity. All therapeutic genomes are flanked by AAV2 inverted terminal repeats. An AAV9 AAV encoding a proprietary protein sequence was produced through triple transfection of HEK293 cells. Purification involved viral inactivation, affinity chromatography and anion exchange (AEX) chromatography, a viral filtration step (VRF), an ultrafiltration/diafiltration (UFDF) step, and formulation and Drug Substance (DS) sublot filtration steps.

LC-MS/MS characterization method

AAV samples were first normalized to a concentration of 1.0E13 vp/mL (approximately 0.1 mg/mL). Vp/mL concentrations were obtained using an SE-HPLC method, as described previously.⁶ Samples were then then denatured with 2% (w/v) RapiGest SF Surfactant (Waters Company, Part Number 186001861) and incubated at 95°C for 5 min. Samples were then enzymatically digested with approximately a 1:15 Protein:Trypsin ratio (Promega Corporation, Part Number V5111) and a 1:30 trypsin/LysC ratio (Promega Corporation, Part Number V5073) and incubated at 37°C for 90 min. The addition of trypsin/LysC ratio was included to reduce the level of missed cleavage. Following enzymatic digestion, the samples were treated with DNase to digest DNA and, further, reduced with TCEP (tris(2-carboxyethyl)phosphine). The reaction was quenched with neat trifluoroacetic acid.

The resulting peptide mixture was separated on a Waters XSelect XP CSH C18 column (Waters Company, 2.5 μm, 2.1 × 150 mm, Part Number 186006727) on an Agilent 1260 UPLC system. Mobile phase A was 0.1% trifluoroacetic acid (TFA) in water and mobile phase B was 0.1% TFA in 95% acetonitrile with 5% water. Gradient conditions were as follows: initial, 0% B; 5 min, 0% B; 140 min, 40% B; 155 min, 90% B; 160 min, 90% B. The flow rate was 0.2 mL/min, and column temperature was set to 60°C. UV absorbance data was acquired at 214 nm and 280 nm.

Mass spectrometric detection was performed using a Thermo Scientific Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific, Bremen, DE). Key source parameters include an H-ESI voltage of +3.5 kV, ion transfer tube temperature of 300°C, vaporizer temperature of 200°C, and RF lens % of 40. MS detection was set as Full MS1 scan range from m/z 180–2,000 with a resolution setting of 480,000 at m/z 200. MS2 was collected with a resolution setting of 30,000 with stepped higher-energy collisional dissociation (HCD) energies of 28, 30, 32. The resulting data were analyzed using Thermo Scientific’s BioPharma Finder (Thermo Scientific, Waltham, MA). The relative abundance of each targeted product quality attribute was quantified using the sum of the extracted ion chromatogram (EIC) peak area for the modified version(s) of the peptide divided by the sum of the EIC peak area of the modified and unmodified versions of the peptide.

Sample preparation and analysis for the multi-attribute method

AAV samples were denatured with 2% (w/v) RapiGest SF Surfactant (Waters #186001861) in 200 mM Tris-HCl, 20 mM L-Methionine, pH 8.0, and incubated at 95°C for 5 min. The samples were then allowed to cool to room temperature prior to enzymatic digestion. Digestion was performed by adding an approximately 1:1 ratio of Protein:Trypsin/Lys-C Mix (Promega, catalog #V5071) and 2:1 ratio of Protein:Trypsin (Promega, catalog #V5111) to each sample, followed by incubation at 37°C for 3.5 h. DNase digestion was carried out by adding 1.3 μL of DNase I (BioLabs #M0303L) and incubating at 37°C for 30 min. The digestion was quenched with neat trifluoroacetic acid (Thermo #28904).

The resulting peptide mixture was separated on an Agilent ZORBAX 300 SB-C18 column (Agilent, Rapid Resolution HD, 2.1 mm × 150 mm, 1.8 μm) using a Vanquish UHPLC+ system (Thermo Scientific) with a binary pump, autosampler, UV detector, and column compartment. Samples were injected at approximately 1 μg per injection, unless the sample injection volume was deliberately changed and specified in the manuscript. Mobile phase A was 0.1% formic acid in water, and mobile phase B was 0.1% formic acid in acetonitrile. The gradient ranged from 0% mobile phase B to 90% mobile phase A over 75 min. The flow rate was 0.25 mL/min, and the column temperature was set to 50°C.

Mass spectrometric detection was performed using a Thermo Scientific Exploris-240, Exploris MX, or Exactive Plus (Thermo Fisher Scientific) mass spectrometers with Chromeleon software version 7.2. Calibration was conducted using Pierce FlexMix Calibration solution (Pierce #A39239). Key source parameters included a spray voltage of +3.00 kV, a capillary temperature of 225°C, and an auxiliary gas heater temperature of 200°C. MS detection was set as Full MS1 scan range from m/z 250–1800 with a resolution setting of 120,000 for the Exploris-240 and Exploris MX and a resolution setting of 140,000 for the Exactive Plus. The relative abundance of each targeted product quality attribute was quantified using the sum of the EIC peak area for the modified version(s) of the peptide divided by the sum of the EIC peak area of the modified and unmodified versions of the peptide.

Samples and validation

A summary of the validation samples is provided in Table S3. Validation samples were prepared as follows. A bulk volume of a single sample was divided. A portion of that material was directly aliquoted into Sample A or left unstressed and not aliquoted for mixing. The remainder was stressed at 25°C for approximately 1 month. A portion of the stressed material was aliquoted into Sample E and the remainder was saved for mixing. Sample A and Sample E were then mixed at defined ratios to make Sample B, Sample C, and Sample D. All materials were measured for the vp/mL concentration prior to mixing, to ensure the samples had consistent concentrations.

In general, the calculation of relative percent abundance of a given modification was computed by dividing the peak area of the modified peptide by the total peak area of the peptide (including both the modified and unmodified forms).

Samples F and G were used to compute the QL by lowering the injection volume of the unstressed Sample A. The calculation for Samples F and G was performed as follows, including the computation of the theoretical values.

The injection load ratio (defined as LR) was calculated between the low injection volume sample (Sample F or Sample G, denoted as Sample X in the equation below) and Sample A.

$$\text{LR}=\text{Injection}\text{volume}\text{of}\text{Sample}\mathrm{X}/\text{Injection}\text{volume}\text{of}\text{Sample}\mathrm{A}$$ (Equation 1)

To adjust between signal differences across injections, a correction factor (defined as CF) was calculated between Sample X and Sample A. Area_U represents the peak area of the unmodified peptide.

$$\text{CF}={\text{Area}}_{\mathrm{U}}\left[\text{Sample}\mathrm{X}\right]/\left({\text{Area}}_{\mathrm{U}}\left[\text{Sample}\mathrm{A}\right]\times \text{LR}\right)$$ (Equation 2)

The computation of the relative abundance of the low injection volume samples used Equation 3. Area_M denotes the summed peak area of the modified peptide of interest. Area_T denotes the total area of the peptide, including all modified and unmodified forms.

$$\text{Relative}\text{Abundance}=({\text{Area}}_{\mathrm{M}}\left[\text{Sample}\mathrm{X}\right]/\text{CF})/\left({\text{Area}}_{\mathrm{T}}\left[\text{Sample}\mathrm{A}\right]\right)\times 100$$ (Equation 3)

Quantitative relative potency (expression) assay

The potency of AAV9 test samples was determined by quantitating expression of the mini-dystrophin protein using Protein Simple capillary immunoassay technology. IMMORTO-mdx myoblast cells⁴⁸ were seeded on fibronectin-coated 96-well cell culture plates, and subsequently differentiated into myotubes using DMEM with 1.5% horse serum at 37°C, 5% CO₂. Approximately 48 hours post-differentiation, the cells were transduced with a serial dilution of AAV9 reference material and test samples, and incubated at 37°C, 5% CO₂ for an additional 6 days. After incubation, AAV9-transduced myotubes were lysed, and the expression of mini-dystrophin protein (Leica Biosystem, NCL-DYSB Clone 34C5) was quantitated and normalized with a myotube differentiation marker, alpha (α)-sarcomeric actinin (Sigma, A7811, Clone EA-53), using Protein Simple capillary immunoassay technology. The assay response from three assay plates was averaged, and the relative potency (expression) of the test sample was determined by the slope ratio of the normalized AAV9 test sample dose response slope divided by the normalized reference material dose response slope. The potency (expression) of the AAV9 test sample was expressed as percent relative potency (%RP).

Data availability

All data necessary to interpret, verify, and extend the research in the article were made available.

Acknowledgments

Funding for this study was provided entirely by Pfizer Inc. The graphical abstract was created in BioRender. Castaneda, C. (2025) https://BioRender.com/d02m391.

Author contributions

T.W.P., S.M., H.N., D.R., and S.S. generated materials, authored protocols, performed experiments, and analyzed data. T.W.P. is the lead author of the manuscript. T.W.P., S.M., H.N., D.R., and S.S. generated figures and tables for the manuscript. All authors were involved in the planning, review, editing, and approval of the article.

Declaration of interests

The authors declare no competing interests.