Using prep-UC to fractionate full, intermediate, and empty AAV capsids into enriched capsid populations

A single-stranded rAAV vector lot containing a heterogenous mixture of full (57.4%), intermediate (26.1%), and empty (16.5%) capsids (Fig. 1a) was subjected to prep-UC using a cesium chloride gradient to generate bands for each capsid population (Supplementary Fig. S2). Two rounds of separation were performed and the bands corresponding to full (OXBS1-F), intermediate (OXBS1-I), and empty (OXBS1-E) capsids were extracted individually. It should be noted that because both the OXBS1-I and OXBS1-F bands were so close together, to ensure the entire OXBS1-I band was extracted, it was inevitable that a subpopulation of OXBS1-F would also be present.

The first step in analytical characterization was to quantify capsid content purity for each extracted sample. Several accepted analytical methods can be used to determine the packaging profile of rAAV capsids, including using the ratio of the VG titer compared to the capsid titer, transmission electron microscopy, AUC, and CDMS [17]. AUC was chosen since this method provides sufficient resolution to distinguish between full, intermediate, and empty capsids and, therefore, allow quantitation of each species [15]. Figure 1 shows AUC size distribution profiles reflecting the relative size and abundance of the separated species in each sample. The OXBS1-F fraction was enriched for full capsids (91.1%) with a low level of intermediate capsids (8.6%) (Fig. 1b). The OXBS1-I fraction was a mixture of intermediate (55.8%) and full (43.8%) capsids (Fig. 1c). The OXBS1-E fraction was enriched for empty capsids (83.0%) with a low level of intermediate capsids (17.0%) (Fig. 1d).

CDMS as an orthogonal method for quantification of full, intermediate and empty AAV capsids and evaluation of capsid charge state

CDMS analyzes AAV capsid content through the differences in charge and mass between empty and full capsids. Figure 2 shows CDMS mass size distribution profiles reflecting the relative mass and abundance of the separated species in each sample. The genome mass (1.34 MDa) was calculated by subtracting the mass of empty capsid peak (3.71 MDa) from the full capsid peak (5.05 MDa) and aligned closely to the expected genome mass for this GOI (1.28 MDa). Narrow peak widths for OXBS1-F (Fig. 2a) and OXBS1-E (Fig. 2c) suggested homogenous packaging of full and empty capsids, respectively. In contrast, the peak width for intermediate capsids in OXBS1-I (Fig. 2b) was much wider (4.00–4.88 MDa) and suggested more heterogenous packaging. The genome mass of the major peak in OXBS1-I, calculated by subtracting the mass of intermediate capsid peak (4.39 MDa) from full capsid peak (5.07 MDa), was roughly half the genome length at 0.68 MDa. The OXBS1-E fraction consisted primarily of empty capsids at 3.71 MDa. However, a second peak of 20% relative abundance was identified with 7.41 MDa mass, which correlates with an empty capsid dimer. Interestingly, this peak was quantified as intermediate capsids by AUC, which highlights the importance of having orthogonal analytical methods to characterize quality attributes. The relative abundance of full, intermediate, and empty capsids in OXBS1-F (Fig. 2a), OXBS1-I (Fig. 2b) and OXBS1-E (Fig. 2c) fractions calculated by CDMS were highly comparable to that determined for AUC as shown in Fig. 1.

Fig. 2: Capsid content determination by CDMS.

CDMS mass distribution plots for rAAV samples post-preparative ultracentrifugation. Tables below the figure detail the identified peak, mass range (MDa) and % relative abundance, n = 1 for the assay. A OXBS1-F, B OXBS1-I, and C OXBS1-E.

CDMS can also determine charge state distribution and provide orthogonal information about the capsid structure. Despite the substantial mass increase from empty capsids (3.71 MDa) to intermediate capsids (4.39 MDa) to full capsids (5.05 MDa), the average charge increased only slightly (around 10 units) (Supplementary Table S2). This phenomenon was observed previously [18] and is thought to be consistent with a small expansion of the capsid with larger genome size and with all the genomes being fully internalized. The largest difference in charge was observed for empty dimer (7.41 MDa; 293.4 units) compared to empty monomer (3.71 MDa; 158.6 units) in the OXBS1-E fraction with the likelihood of charge (for the empty dimer) nearly doubling due to a doubling in overall size.

Determination of VP purity and VP1:VP2:VP3 ratio by CE-SDS

The AAV capsid is composed of three viral proteins (VP1, VP2, and VP3) assembled into an icosahedron at a molar ratio of approximately 1:1:10 of VP1:VP2:VP3 [3]. The relative expression levels of VP1, VP2 and VP3, as well as PTMs of the amino acid side chains, can be affected by the production method used and deviation from the 1:1:10 ratio may result in alterations in the amount of PTMs, potentially impacting AAV in vivo activity [19, 20]. To determine whether VP heterogeneity is observed at the capsid level, OXBS1-F, OXBS1-I and OXBS1-E samples were characterized by CE-SDS. OXBS1-F and OXBS1-I samples were comparable in VP purity (100%) and VP1:VP2:VP3 ratio (1:1:10) (Fig. 3a). However, the OXBS1-E samples had a slightly lower amount of VP1 and a slightly higher amount of VP2, resulting in a 1:2:12 ratio.

Fig. 3: Characterization of VP purity, VP1:VP2:VP3 ratio, and post-translational modifications of full, intermediate, and empty capsids by CE-SDS and LC-MS/MS.

A VP purity and VP ratio of OXBS1-F, OXBS1-I, and OXBS1-E were determined by CE-SDS. B, C Peptide mapping of the VP proteins present in OXBS1-F, OXBS1-I, and OXBS1-E pools was determined by LC-MS/MS. B Sequence coverage was confirmed according to the primary amino acid sequence using BioPharma Finder™ software (Thermo Fisher Scientific) and C post-translational modifications were identified using both BioPharma Finder™ and Skyline. N = 1 for the assay. VP Viral Protein, I aspartic acid isomerization, D deamidation, S succinimide, P phosphorylation, M methylation.

Identification and quantification of PTMs in full, intermediate, and empty AAV capsids by LC-MS/MS

VP integrity and PTMs of the enriched capsid samples were characterized using LC-MS/MS. The determined amino acid sequence matched the expected amino acid sequence for AAVHSC-15 with ≥98% coverage of VP1, VP2 and VP3 for OXBS1-F and OXBS1-I samples (Fig. 3b). The amino acid sequence coverage for VP1, VP2 and VP3 for OXBS1-E was 72–80%. Capsid titers were similar for all three enriched samples (3E + 13 capsids/mL) (Fig. 4a).

Fig. 4: Measurement of VG titer, capsid titer, process-related impurities, and residual sequence expression in HeLa cells.

A VG titer and capsid titer for OXBS1-F, OXBS1-I, and OXBS1-E pools were determined by ddPCR and ELISA, respectively. B Levels of plasmid-derived DNA impurities (Rep/Cap, Helper plasmid, KanR; left y-axis) and host cell DNA (right y-axis) were determined by ddPCR and qPCR, respectively. Results for each target were normalized to 1E + 13 capsids to allow direct comparison between samples. C mRNA expression of GOI and KanR was measured across a dose range of MOIs for OXBS1-F and OXBS1-I in HeLa cells by RT-qPCR. Each point represents the average of n = 2 wells. A no RT control (dotted lines) was run for each to confirm the assay is specifically measuring gene expression.

The percent relative abundance for PTMs is shown in Fig. 3c; identified modifications include isoaspartic acid isomerization, asparagine/glutamine succinimide, asparagine/glutamine deamidation, serine phosphorylation and arginine methylation. The abundance of most AAV VP PTMs were comparable across the OXBS1-F, OXBS1-I, and OXBS1-E samples. The most common spontaneous PTM, asparagine deamidation, can result in an aspartic acid or isoaspartic acid modification. Deamidation was observed at comparably low levels across full, intermediate, and empty capsid populations, except for VP1 residue N57, which was 47% deamidated in the OXBS1-E fraction compared to 1% and 2% N57 deamidation for OXBS1-F and OXBS1-I, respectively. In addition, phosphorylation at S149 was approximately 3-fold higher for OXBS1-E (23%), when compared to OXBS1-F (7%) and OXBS1-I (7%) fractions.

Taken together, the data suggests that full and intermediate AAV capsids have a highly similar profile of PTMs. Empty capsids, on the other hand, had higher deamidation and phosphorylation, specifically at N57 and S149, respectively. The presence of S149 phosphorylation, which is present on both the VP1 and VP2 regions, may play a role in cell signaling and/or trafficking while N57, unique to VP1, is located on a domain that has been shown to play a role in AAV in vivo and in vitro transduction [20]. The aforementioned PTMs are less abundant in intermediate and full capsids and may influence the packaging profile – or lack thereof – for empty capsids. These characteristics provide insight into the biophysical properties of both full and intermediate capsids and provide further insight into why separation and purification of these species remains a current challenge.

Determination of titer for vector genome and residual impurities

Following purification, a portion of AAV capsids containing fragments of nucleic acids other than the fully intended VG may be present as product-related impurities (Supplementary Fig. S1). These impurities, which can include fragments of the VG, host cell DNA and production plasmid DNA, have the potential to impact the efficacy and safety of the product. A comprehensive characterization of these impurities was performed to assess the potential risks of product-related impurities to patients.

The VG titer of the non-fractionated AAV-OXBS1 sample and the enriched fractions (OXBS1-F, OXBS1-I, and OXBS1-E) was determined by ddPCR. The VG titers of OXBS1-F, OXBS1-I and OXBS1-E samples were 1.45E + 13, 1.01E + 13 and 3.26E + 10, respectively (Fig. 4a). All samples were formulated at 3E + 13 capsids/mL. The VG titer for OXBS1-I was ~70% less than OXBS1-F. When considering the percent full capsids by AUC for OXBS1-I (43.8%) and OXBS1-F (91.1%), the percent full for OXBS1-I was 48% less than OXBS1-F. Overall, these findings suggest that the intermediate genomes in the OXBS1-I sample may also be contributing to VG titer.

The amounts of total residual nucleic acid process-related impurities are shown in Fig. 4b. Sample results were normalized to capsid titer to generate residuals per 1E + 13 capsids. The plasmid-derived DNA impurities (Rep/Cap, Helper plasmid and KanR) were present at comparable levels in the OXBS1-F and OXBS1-I pools at 2–4 logs lower than vector genome. Adenovirus E1A levels were below the limit of quantitation (<1E + 07 copies/mL). Residual host cell DNA impurities are fragments of human genomic DNA unintentionally packaged within AAV capsids [4]. In a previous study, AAV vectors produced by transient transfection contained up to 0.30% host cell DNA [21]. Our data shows that residual host cell DNA is enriched in the OXBS1-I fraction, indicating that the host cell fragments are smaller than the optimal 4.7 kB packaging size. All DNA impurities were low in the OXBS1-E empty capsid population (<7E + 09 copies/1E + 13 capsids).

Evaluation of residual impurity sequence expression in vitro

To determine whether the encapsidated DNA impurities were capable of being expressed upon transduction into cells, a residual gene expression assay was performed following transduction of HeLa cells with a dose response curve of OXBS1-F and OXBS1-I samples. Dose-dependent expression of the GOI was observed across the fixed MOI range for OXBS1-F and OXBS1-I (Fig. 4c). No signal was detected for E1A, E2A, E4 or Rep/Cap residual gene expression in both OXBS1-F and OXBS1-I samples (qPCR Ct value was undetermined; data not shown). Dose-dependent expression of KanR mRNA was detected in the reverse transcriptase (RT) group but not the in the absence of RT (no RT group), indicating that this was specifically detecting KanR mRNA expression (Fig. 4d). However, KanR mRNA expression was only observed below baseline 40 Ct level at very high VG/cell (above 5.6E + 04) and up to 15 Ct higher when compared to expression of the GOI. It is uncertain whether such expression would be triggered in vivo with much lower expected VG/cell, although further studies are warranted to confirm this.

Determining the proportion of VG and impurities by next generation sequencing (NGS)

DNA impurity analysis above shows the heterogenous nature of DNA packaging for this AAV preparation. To characterize encapsidated DNA further, DNA was extracted from OXBS1, OXBS1-F, and OXBS1-I samples and prepared for long-read next generation sequencing using the PacBio platform. Read size distributions corresponded with AUC fractionation (Supplementary Fig. S3). Sequencing reads were compared to the packaging plasmids and sorted in bulk into bins based on sequence identity: VG (Full or intermediate); Reverse Packaging; ITR only; Plasmid Backbone; Plasmid Backbone and VG; RepCap; pHelper; and Chimera (which contained reads that had subsequences from multiple plasmids) (Table 1). When compared to unfractionated OXBS1, the OXBS1-F fraction was enriched for VG sequences, whereas the OXBS1-I fraction was enriched for non-VG sequences. As was observed for the ddPCR data above, absolute read counts for RepCap and pHelper containing reads were 2–4 orders of magnitude lower than the vector genome containing reads in all three samples.

Table 1 Summary of bulk NGS sequence attribution.

Reads classified, or determined to be, as VG in Table 1 were further analyzed to determine the proportion of full, snapback VG [22], and truncated VG subpopulations in each sample (Table 2). When compared to unfractionated OXBS1, the OXBS1-F fraction was enriched for full VG, whereas the OXBS1-I sample was enriched for snapback partial VGs (Table 2). Truncated partial VGs appeared at similar rates across all three samples. Unexpectedly, most of the classified intermediate VG species captured were snapback partials, not truncated partials. Both truncated VGs and snapbacks contain partial vector genome sequences and are typically unable to produce a full recombinant product in the case of recombinant gene therapies; therefore, both types of subgenomes can be considered contaminants [7]. However, unlike truncated VGs, snapbacks have a hairpin-like genome and contain ITRs on both the 5’ and 3’ genome ends, which are important for episome formation [23, 24]. Further studies are needed to explore the nuclear durability of both truncated and snapbacks VGs and assess if high proportions of snapbacks within the intermediate genome subpopulation is a general phenomenon.

Table 2 Summary of vector sequence attribution for vector reads by NGS sequencing.

To characterize the snapback genomes further, the 5’ payload aligning segment was mapped for each read to determine the nucleotide at which alignment to the full expected VG ended. The positions where these alignments ended for all snapback reads were enumerated individually for the plus and minus strands (Supplementary Fig. S4). Surprisingly, the location of the terminal nucleotide was extremely discrete. In the OXBS1-I sample, three nucleotides in the reference accounted for over 25% of the snapback nucleotides in + strand reads, suggesting that there is a specific phenomenon responsible for formation of the snapback structure. Local folding structure was calculated across the vector genome with a sliding window approach. Despite observing significant predicted structure in some snapback nucleotides, several high-count snapback nucleotides did not have a high degree of predicted local folding structure (Supplementary Fig. S4). While the occurrence of snapback structure appears to be influenced by local folding structure, it does not explain the existence of all major breakpoints. Both C and G nucleotides were significantly enriched at breakpoints on the plus and minus points for all vectors (X2p-value < 2.2E−16). In aggregate, these data indicate that there is a characteristic process to snapback genome formation [7]. However, a more diverse panel of AAV vector genomes would be needed to clarify the mechanism.

Evaluating the impact of intermediate capsids on in vitro and in vivo potency

AAV potency consists of three parts: 1) transduction of the AAV capsid into target cells; 2) expression of the transgene within the cell; and 3) biological activity of the expressed protein. A full matrix of in vitro potency assays (infectivity, transgene expression and biological activity) was developed for this rAAV product to evaluate the impact of intermediate and empty capsids on potency. The infectivity assay determines the infectious titer of the sample and is reported relative to the VG titer (i.e., VG/IU). The in vitro transgene expression and biological activity assays determine the percent relative transgene expression (%RGE) and percent relative potency (%RP), respectively, when compared to a product-specific reference standard.

In vitro potency results for OXBS1-F and OXBS1-I samples are shown in Fig. 5a. The sample VG titer was used for each assay to allow determination of potency as a function of vector genome. Infectivity results (VG/IU) were comparable for OXBS1-F and OXBS1-I samples and within the expected variability for this method, indicating that the samples enriched for full or intermediate capsids were equally infectious. However, sample divergence was observed for transgene expression (%RGE) and biological activity (%RP) where the OXBS1-I fraction was 1.7-fold and 2-fold less potent than the OXBS1-F sample, respectively. The fact that transgene expression and biological activity was even observed in the OXBS1-I sample was likely due to the presence of 43.8% full capsids. To test this theory, OXBS1-F, OXBS1-I and OXBS1-E samples were next tested in the transgene expression assay by running all samples using the OXBS1-F vg titer (1.45E + 13 vg/mL). Since all three samples had an identical capsid titer of 3E + 13 capsids/mL, this resulted in equal loading of capsids per cell and allowed for determination of transgene expression as a function of capsid content. As shown in Fig. 5b, the %RGE was highest for the OXBS1-F sample (96%) and correlated well with % full capsids by AUC (91.1%). The OXBS1-I sample was 45% RGE, which correlated well with AUC results for % full in this sample (43.8%). The OXBS1-E sample, which had no full capsids, was 0% RGE.

Fig. 5: Evaluating the impact of intermediate capsids on in vitro and in vivo potency.

A Infectivity (VG/IU; left y-axis), transgene expression (%RGE; right y-axis) and biological activity (%RP; right y-axis)= were determined for OXBS1-F (black bars) and OXBS1-I (blue bars) samples using the nominal VG titer to determine potency as a function of vector genome. N = 1 for each sample. B Transgene expression was determined in vitro (%RGE; black bars) and in vivo (%RGE; blue bars) for OXBS1-F, OXBS1-I, and OXBS1-E samples using the nominal capsid titer to determine potency as a function of capsid content. % Full capsids were determined by AUC (purple bars). N = 1 for all assays. C Transgene expression was determined in liver tissue 5 weeks-post infusion of mice with 9E13 capsids/kg of OXBS1-F, OXBS1-I, or OXBS1-E samples. Formulation buffer was used as a vehicle control. GOI copies per ng RNA was determined by ddPCR. One-way analysis of variance (ANOVA) was used to determine any statistically significant differences between the means of the capsid groups. Error bars are SEM. N = 4 mice per cohort.

Potency as a function of capsid content was also determined in vivo. Mice were infused with an equivalent dose of 9E + 13 capsid/kg of either OXBS1-F, OXBS1-I or OXBS1-E. This dose was expected to result in robust transgene expression in the liver. VG transcript levels in liver samples were determined in a ddPCR transgene expression assay using the same primer/probe set used for VG titer, infectivity, and in vitro transgene expression assays. The GOI copies per ng RNA were calculated and plotted for each group in Fig. 5c. When dosed at equivalent capsids/kg, the OXBS1-F sample was significantly more potent than the OXBS1-I or OXBS1-E samples. This is not entirely surprising since the OXBS1-F cohort had the highest vg/kg of all the groups (OXBS1-F, 4.1E + 13 vg/kg; OXBS1-I, 3.0E + 13 vg/kg; OXBS1-E, 9.6E + 10 vg/kg). As shown in Fig. 5b, the %RGE for OXBS1-F (100%), OXBS1-I (44%) and OXBS1-E (0%) samples correlated well with % full capsids by AUC (91.1%, 43.8% and 0%, respectively).

These results together suggest that while intermediate AAV capsids (containing mostly snapback partial VGs) are capable of infecting cells, they are not transcribed into mRNA to generate a functional protein. It is therefore imperative when analyzing the function of partial VG sequences to leverage a full matrix of assays that evaluate every step of the potency lifecycle i.e., transduction, expression, and biological activity.

Comparability of AUC, CDMS, NGS and potency results

We next compared the analytical results obtained for the AUC, CDMS, PacBio sequencing, and potency methods. For this analysis, we took the PacBio total read counts of full-sized VG (i.e., ITR through ITR), partial-sized VG (i.e., snapback, truncated or sequences categorized as other that did not contain a full payload) or non-VG (i.e., Rep/Cap, pHelper, Plasmid Backbone, Reverse Packaging, Chimeric sequences, etc.) and normalized each to the total classifiable reads to generate a % full VG, % partial VG and % non-VG value, respectively, by NGS (Supplementary Table S3). A % intermediate genomes were calculated by combining % partial VG and % non-VG groups.

The % full VG results determined by NGS for each sample were remarkably similar to the % full capsids measured by AUC and CDMS, as well as the %RGE measured by in vitro and in vivo transgene expression methods (Table 3), strongly supporting that the full peak quantified by AUC and CDMS contains predominately functional full-length ITR-through-ITR vector. The most striking comparison was observed for the OXBS1-I sample, where the % full (44% by AUC, 45% by CDMS, 44% by NGS; Table 3) and % intermediate (56% by AUC, 54% by CDMS, 56% by NGS; Figs. 1,  2, and Supplementary Table S3) were identical between methods. NGS sequencing was able to determine that the intermediate peak of the OXBS1-I fraction is comprised of both intermediate VGs (predominately snapback partials) and non-VG sequences. Thus, together, AUC, CDMS and NGS analysis provide a comprehensive understanding of the vector profile of the product. Furthermore, these results strongly support that the potency observed in the OXBS1-I sample is attributable to the presence of full VGs in this sample and that the intermediate VGs themselves are not functional.

Table 3 Comparison of results for AUC (% full capsids), NGS (% full VG), in vitro transgene expression (%RGE) and in vivo transgene expression (%RGE).