A novel method for quantitation of AAV genome integrity and residual DNAs using duplex digital PCR

Lauren Tereshko, Xiaohui Zhao, Jake Gagnon, Tin-Chi Lin, Trevor Ewald, Marina Feschenko, Cullen L. SchmidbioRxiv 2023 · DOI

TL;DR

This paper introduces a novel statistical model, the Poisson-multinomial distribution, to accurately quantify the integrity of adeno-associated virus (AAV) genomes using duplex droplet digital PCR (ddPCR). This new model significantly improves accuracy and expands the quantifiable range compared to existing methods, addressing a critical need for robust AAV characterization in gene therapy development.

What Problem Does This Paper Address?

Gene therapies utilizing recombinant adeno-associated virus (rAAV) vectors are a promising frontier, but their efficacy and safety depend heavily on the quality of the AAV capsid content. AAV production processes can lead to heterogeneous capsid populations, including empty, partial, or over-packaged capsids, as well as unwanted residual DNA from the production process. Accurately assessing AAV genome integrity is crucial for ensuring product quality. While droplet digital PCR (ddPCR) is a powerful tool for this, existing methods for analyzing multiplexed ddPCR data, particularly for duplex assays (which simultaneously detect two targets), are limited in their accuracy and dynamic range. This gap hinders robust process development and quality control for AAV-based gene therapies.

Methods

The core of this paper is the development and validation of a novel Poisson-multinomial mixture distribution model for analyzing duplex ddPCR data. This model goes beyond simpler Poisson-based approaches by accounting for the three possible outcomes in a duplex reaction: detection of only the 5' target, only the 3' target, or both targets (indicating an intact genome).

The authors compared their new model against two existing "linkage" models (Formula 3 and Formula 4) that rely on calculating the percentage of linked targets. These comparisons were performed using:

  • Plasmid simulated genomes: These were created by digesting viral vector plasmids to simulate varying degrees of genome integrity (0-100%).
  • rAAV simulated genomes: These were generated by mixing two rAAV samples with a common promoter but different poly(A) sequences, allowing for controlled integrity levels.
  • Heat-fragmented rAAV genomes: This involved subjecting rAAV samples to thermal stress to induce DNA degradation, mimicking real-world degradation scenarios.

The performance of the models was evaluated based on accuracy, precision (RSD), recovery, and linearity.

Datasets

The study utilized simulated datasets for evaluation:

  • Plasmid simulated genomes: Created from digested viral vector plasmids, with expected genome integrity levels ranging from 0% to 100%.
  • rAAV simulated genomes: Prepared by mixing two rAAV samples with varying concentration ratios to achieve expected integrity levels of 0%, 5%, 10%, 21%, 42%, 63%, and 84%.
  • Heat-fragmented rAAV genomes: These samples had an expected integrity of 48% as determined by long-read NGS, and were further fragmented by incubation at 95°C for varying durations (0-30 minutes).

No new benchmark datasets were introduced; rather, the existing simulated and prepared samples served as the evaluation material.

Key Insights and Ideas

  • The limitations of simple models: Traditional percentage-based calculations and even basic Poisson models struggle with the complexity of duplex ddPCR data, leading to inaccurate estimations, especially at lower concentrations or when dealing with fragmented genomes.
  • The power of the Poisson-multinomial model: This new model's ability to consider multiple droplet species (5' only, 3' only, both) provides a more nuanced and accurate interpretation of duplex ddPCR results.
  • Linkage models overestimate integrity: A critical finding is that existing linkage models consistently overestimate genome integrity, particularly for fragmented samples. This is hypothesized to be due to their focus on the concentration of linked targets rather than the total DNA present.
  • Expanding the dynamic range: The Poisson-multinomial model significantly broadens the quantifiable range of AAV genome integrity assays, making it suitable for a wider spectrum of sample concentrations and integrity levels.
  • Robustness to fragmentation: The new model demonstrates superior performance in accurately assessing integrity even when genomes are fragmented, a common challenge in AAV preparations.

Key Results

  • The Poisson-multinomial model achieved overall accuracy of 96.3% for plasmid samples and 98.1% for rAAV samples, with intermediate precision below 1%.
  • Compared to linkage models, the Poisson-multinomial model showed greater overall accuracy (149.6% vs. 98.1%) and intermediate precision (21.8% vs. 6.5%) for rAAV samples.
  • Linkage models consistently overestimated genome integrity for both simulated and heat-fragmented rAAV samples, with overestimation worsening as expected integrity decreased.
  • The Poisson-multinomial model demonstrated better linearity with a slope of 1.01 and pseudo R² of 0.996, indicating more reliable results across the expected integrity range.
  • The model's dynamic range was expanded to cover at least four orders of magnitude in template concentrations.

Limitations and Future Work

  • Limitations of Linkage Models: The paper strongly suggests that linkage models are less accurate due to their assumption that both 5' and 3' regions are present even in fragmented genomes, and their focus on average or maximum target concentrations rather than total DNA.
  • Future Work: The authors highlight that the developed Poisson-multinomial model has been coded into a Shiny application, making it publicly available and easy to use. They suggest that this model can be applied to other duplexed ddPCR assays, including those for monitoring residual DNA size and identity. Furthermore, they propose expanding multiplexed ddPCR to simultaneously measure quantity and integrity of specific residual DNAs from plasmids or host cells, potentially by controlling decapsidation timing. The expanded dynamic range of the Poisson-multinomial model is seen as advantageous for such analyses.

Summary updated

Abstract

Abstract Recombinant adeno-associated virus (rAAV) vectors have become a reliable strategy for delivering gene therapies. As rAAV capsid content is known to be heterogeneous, assays for rAAV characterization are critical for assessing the efficacy and safety of drug products. Multiplex droplet digital PCR (ddPCR) has emerged as a popular molecular approach for characterizing capsid content due to its high level of throughput, accuracy, and replicability. Despite growing popularity, tools to accurately analyze multiplexed data are scarce. Here, we introduce a novel model to estimate genome integrity from duplex ddPCR assays. This work demonstrates that use of a Poisson-multinomial mixture distribution significantly improves the accuracy and quantifiable range of duplex ddPCR assays over currently available models.
BibTeX
@misc{tereshko2023novel,
  author = {Lauren Tereshko and Xiaohui Zhao and Jake Gagnon and Tin-Chi Lin and Trevor Ewald and Marina Feschenko and Cullen L. Schmid},
  title = {{A novel method for quantitation of AAV genome integrity and residual DNAs using duplex digital PCR}},
  year = {2023},
  doi = {10.1101/2023.07.19.546797},
  howpublished = {bioRxiv},
}
Figure