Viral vectors are the next therapeutic revolution, with an immense ability to cure otherwise untouchable diseases, from congenital to communicable maladies. However, the vectors themselves are incredibly complex entities, and their relative novelty means the analysis, technology and understanding is immature compared to other biopharmaceutical classes. The properties of viral vectors – their size, lability and potentially hazardous nature – imposes restrictions on experimental throughput and data quality and turn viral purification process development into a highly complex affair. To realise their full potential, development must be ready to evaluate and implement new methodologies and techniques, from novel consumables and equipment, such as new adsorbers, HTPD devices, automation systems and analytical tools. Statistical analysis is likely to remain an important facet for development, though for chromatography, fortunately, the nascence of viral vectors is coincident with that of another powerful tool in the downstream process development arsenal: Mechanistic modeling. Such models allow one to understand, predict and therefore develop and validate a process almost entirely in silico through employing the fundamental principles of fluid dynamics and thermodynamics. An in silico approach to purification process development of viral vectors enables fundamental benefits along the entire value creation chain and has proven successful in various showcases.
The world has changed drastically over the past few months, with entire countries being quarantined, billions of people living under lockdown, millions of cases and hundreds of thousands of deaths worldwide (WHO, 2020). The risk viruses present to global 21st century life has never been clearer than with the advent of COVID-19. But viruses are likely to be also significant in the fight against COVID-19: Virus-based vaccines have been established for over 200 years (Riedel, 2005) and are responsible for the eradication of many diseases before. Inactivated viruses may be ultimately used in the fight against COVID, with three of the five treatments entering Phase II or III development of this type (Milken Institute, 2020).
The ability to manipulate viruses further and exploit their ability to transduce human cells as a vector, rather than just their immunogenicity, is a very powerful tool to further employ viruses for treating, rather than causing, disease. The recent renaissance of viral vectors for gene therapy has been providing positive patient outcomes through fine control of biological processes – which conventional therapies cannot. At the moment, one of the leading COVID candidates in Phase 2 is of this class, with dozens more in earlier development (Milken Institute, 2020). The viral vector industry is growing at a rapid pace and by 2025 the market size is expected to surpass $10 billion (BIS Research, 2019) and the FDA is expecting to approve 10 to 20 cell and gene therapy products annually (Gottlieb, 2019).
The list of viral vector therapies already granted marketing approval by one or more regulatory body covers a myriad of maladies. Candidates in development demonstrate an even greater therapeutic potential, from the first viral vector gene therapy candidate trialled in 1989, there are now over 3,700 registered clinical trials for gene therapies across 204 countries, of which viral vectors represent most of these candidates (Yin et al., 2014). They include treatments for such blockbuster indications as cystic fibrosis, Parkinson’s and sickle cell disease, with many of these candidates in Phase III. They represent a veritable viral menagerie, with the most popular groups of viruses including many pseudotypes and serotypes of Adenovirus (Ad), Adeno-associated Virus (AAV), Retrovirus (RV) and herpes simplex virus (HSV).
|Luxturna™||Leber's congenital amaurosis||AAV||2017|
These viruses possess a wide array of morphologies, genomes types, structures, mechanism of action and indications, and cannot be adequately described in a few paragraphs. In short, even the few viral vector types given regulatory approval represents ranges in size from 20 nm to 240 nm, includes ssRNA to dsDNA genomes from 4 kb to over 30 kb in capacity, and from relatively simple particles with a small capsid, to enveloped structures with a payload of enzymes. There is also significant variation between subtypes and serotypes, though there are also a few similarities which this class of biopharmaceutical share from a processing standpoint, largely complexity and the penalty of novelty.
Being a nascent class of therapeutics, all aspects of a viral vector bioprocess are typically under rapid development, aiming to improve the production, purification and analysis. The common factors across all activities is the immense complexity of viral vectors, poor stability, the very high cost of goods of manufacture, often $10k – $100k/dose (Cameau et al., 2020), translating to prices of hundreds of thousands, or millions of dollars, per patient (Spink & Steinsapir, 2018). In light of viral vector therapies being withdrawn from the market, or not granted reimbursement due to the cost (Mullin, 2017), this is of particular concern. Upstream and analytical uncertainty all impact downstream processing , in which material is variable, typically low concentration, contaminated by a myriad of process and product related impurities and poorly characterised. The inherent safety risks of handling, the variability between viral types, performance and ease of analysis and the lack of the decades of experience other biopharmaceutical classes can draw on affects every stage of bioprocess development. On top of these, the typical drivers impacting most bioprocess development are also present; bigger pipelines feeding into early stage development, ever shrinking timelines, greater production scales, leaner process economics and push towards process intensification, whilst in the face of regulatory pressure to establish appropriate product and process understanding.
Development of viral vector processes therefore must be rapid, agile, effective and robust, with a reduced experimental throughput still providing good process understanding and efficiency. To achieve this may well require new techniques, technologies and paradigms; fortunately, there are several developments in the field of downstream process development which are especially promising when applied to viral vector processes.
Viral vector analytics can be expensive, variable, low throughput and sparse, typically relying on biological assays, analytical ultracentrifugation, electron microscopy, ELISA and qPCR for titration, each with their own caveats. Many novel analytical tools have also been fruitful in developing viral vector processes; from droplet digital PCR (ddPCR) enabling one to forego standard curves whilst reducing plate-to-plate variability, the use of dynamic light scattering (DLS) to immediately probe the size distribution of particles in a sample, as well as the advent of benchtop transmission electron microscopes (TEM) to view, and quantify vector morphology and genome enrichment. With each passing month, new techniques are developed and ways of employing established techniques to viral vector processes are discovered.
Several novel ligands and adsorber types have demonstrated great potential in purification of viral vectors. Affinity ligands derived from camelid antibodies, such as Cytiva’s AVB™ and Thermofisher’s AAVX™ ligands are widely used in AAV processes for their wide specificity across a range of AAV serotypes, affording the same benefit Protein A has provided mAb processes; a single step providing high purity and concentration from crude feedstocks. Unlike the available state-of-the-art variants of recombinant Protein A, there is poor stability with respect to high pH’s, precluding the standard sodium hydroxide cleaning regimes whilst maintaining the high costs associated with such complex ligands, though the innovations seen with protein A in the past decades will hopefully be replicated here with respect to cost and stability. Affinity ligands have been trialled for adeno and influenza viruses. There is further active development towards a lentiviral affinity ligand, though the development of conditions for elution suitable for enveloped virus recovery will be a difficult endeavour. Mixed-mode resins for viral purification are also increasingly employed. Cytiva’s CaptoCore™ range, which relies on steric exclusion of large particles from the shell of a bead, but pore-penetration and therefore clearance of smaller impurities, enabling one to forgo the elution issue by operating in a flowthrough mode.
Viral vector chromatographic processes typically do not rely on high surface-area, porous bead adsorbers for bind-elute, but instead typically employ more novel formats such as membranes, monoliths and fibres, enabling one to rapidly process material in a low shear environment through relying on convection for mass transport. This focus on reducing shear has also been applied to TFF processes, in which hollow-fibre formats dominate compared to flat-sheet cassettes.
As with the viral vectors themselves, these adsorber formats and types are relatively novel, and are therefore not as well understood as more established operations. Binding capacities, reusability and cleaning regimes have yet to be defined and standardised across the industry, but their current utility cannot be understated and will only improve with time and greater understanding of both product and process. Other novel tools, such as high throughput process development (HTPD) technologies and mechanistic modeling will help bridge this knowledge gap in an efficient manner
The general theme of the pressures on bioprocess development, especially with viral vectors, in the need to obtain more information with less time and material. In this regard, high throughput technologies are incredibly powerful, especially when combined with robotic liquid handlers, enabling the scale-down, automation and parallelization of experiments. Upstream tools include microscale, parallel bioreactors and microwell formats, whereas DSP uses formats such as miniaturized columns, resin slurry plates, miniature TFF cartridges and filter plates enabling one to evaluate between 8 and 96 experiments simultaneously. A common issue, however, is interpreting and validating a scale down model – fluid flow between a microscale and production scale column or bioreactor is very different, and this often impacts performance. One often expects more variation in a high-throughput system in which small errors are magnified, and with generally scarcer data for each experimental condition. From a chromatography perspective, replacing a column-ÄKTA™ system, with its online UV, pressure, pH and conductivity measurements and continuous, finely controlled flow instead with a microscale column-LHS system, which has none of those benefits but only offline analysis, may well impact the data quality and what process understanding may be gleaned. Many have used a model framework to better use this valuable data to help alleviate the issues arising from scale, either statistical models to handle the array of data, or mechanistic models to better comprehend the data.
Multivariate data analysis (MVDA) is a mainstay in modern process development, enabling one to investigate the interaction between variables and determine correlations otherwise ignored. This approach is widely used across all the bioprocess industries at every stage, from cell-line development to formulation, and is a very powerful tool for understanding how the parameters in a bioprocess correlate, as well as being able to screen a combination of these parameters to move towards a process optimum and establish a design space. However, all a data-driven method such as Design of Experiments (DoE) or multivariate regression can do is correlate parameters empirically, but not prove causation. Additionally, the assumption that responses are linearly or polynomially, depending on design, related to a combination of inputs is a patently weak one – most interactions in bioprocesses are non-linear, at least at any meaningful range. Additionally, these techniques, whilst a reduction compared to the one-factor-at-a-time approach of yore, require a lot of experimentation, again, depending on design. Therefore DoE/MVDA is an especially powerful tool for USP development, where the systems are very complex and poorly understood, and is typically used in tandem with the high-throughput techniques. This, however, is not the case for chromatography, in which one may employ established principles of thermodynamics, hydrodynamics and kinetics to better interpret the valuable data generated at every scale and mode of operation through establishing a mechanistic model of the process.
Mechanistic models are deceptively simple; the full complexity of protein interactions is something incredibly difficult to predict and model, considering the shear number, nature and convolution of interactions. Fortunately, a mechanistic model may take all these complex physicochemical phenomena and reduce them to an adsorption model.
Such models typically comprise only a handful of parameters describing how much material adsorbs/desorbs, how material interacts with other components when doing so, and how quickly these events happen. These parameters may be estimated in several ways, from batch adsorption to quantitative structure activity relationships. Using GoSilico’s standard approach only few data is needed, that is often already at hand: simple gradient elutions, a standard experiment in early process development. The remaining characteristics of the process; the flow non-idealities, the structure of columns, beads, fibers, membranes or monoliths and mass-transfer resistances are represented by a series of constants in differential equations. This approach is particularly well suited when the experimental optimization is difficult, whilst chromatograms present a lot of information which is typically not used to its full potential, such as during polishing steps and empty/full vector separation.
Mechanistic models simulate the innate complexity of the system. The simulation is however not based on providing enough data of the problem to establish complex empirical correlations, but through implementing base scientific principles to determine causation.
Nick Whitelock, Business Development Manager at GoSilico
Mechanistic models provide a very powerful avenue for chromatographic development of viral vector purification processes and provide unparalleled insights into the dynamics of the system. This is in particular true for applications, with non-linear variations caused by different flowrates, intermittent fluid handling and poor data completeness of microscale formats, where DoEs would not be able to model these complex behaviours. A difference in upstream material quality, buffer tonicity or pH, feed titre or differences in column packing is reduced to a series of model parameters, and the solution framework can thereafter predict the impact on performance. Of particular benefit is that these model parameters have real physical significance, rather than being incomprehensible fitting parameters, they directly describe measurable properties such as diffusion rates, HETP, system volumes and column void volumes. With this method, the expense and safety risks typical of working with viral vectors, the poor experimental and analytical throughput is then mitigated. The experiments performed to calibrate the model are designed in such a way to get as much information as possible with the smallest amount of experimentation, in stark contrast to the guessing game a DoE may become. This is largely because this approach is also valid for extrapolation outside a calibration range and better describing behaviours within it, by employing these base principles.
This enables the majority of standard development activities – process optimization, characterization and validation, to be performed in silico, evaluating a far greater range of process conditions than what’s practicable experimentally. Additionally, by being able to describe a complex system using causal mathematical relationships alone, based upon established scientific principles, demonstrates the pinnacle of process understanding. Rather than merely understanding correlations between process parameters and performance, one can describe the influence of one on the other, enabling significantly more avenues for process evaluation, extrapolation of process conditions and a mechanistic understanding of the process, compliant with the principles of Quality by Design (QbD).
To help realize the potential of viral vectors, the industry must adapt to new technologies and methodologies. Several approaches have promise to help develop effective downstream processes: high throughput technologies, novel adsorber formats, and of course mechanistic modeling. This enables one to establish a mathematical understanding of the processes, allowing mitigation of the variability plaguing traditional approaches, the increase in cost and the reduction in experimental throughput, by enabling a greater level of understanding with fewer experiments.
If you would like to discuss how ChromX has been used to improve development for a range of viral vectors, vaccines and other biopharmaceutical processes across multiples of chromatographic scales, formats, absorbers and modes, and its applications from early-stage screening to process validation, please reach out to our business development team.
BIS Research (2019) “Global Cell and Gene Therapy Market: Focus on Products, Applications, Regions and Competitive Landscape – Analysis and Forecast, 2019-2025” https://bisresearch.com/industry-report/cell-gene-therapy-market.html
Gottlieb, S (2019) https://www.fda.gov/news-events/press-announcements/statement-fda-commissioner-scott-gottlieb-md-and-peter-marks-md-phd-director-center-biologics – accessed 15JUN20
Milken Institute (2020) COVID-19 Treatment and Vaccine Tracker Accessed 16JUN20
Mullin, E (2017), Tracking the Cost of Gene Therapy, MIT Technical Review,
Riedel, S. (2005). Edward Jenner and the History of Smallpox and Vaccination. Baylor University Medical Center Proceedings, 18(1), 21–25.
Spink, K., & Steinsapir, A. (2018). The long road to affordability: a cost of goods analysis for an autologous CAR-T process. Cell and Gene Therapy Insights, 4(11), 1105–1116.
Yin H, Kanasty RL, Eltoukhy AA, Vegas AJ, Dorkin JR, Anderson DG. Non-viral vectors for gene-based therapy. Nat Rev Genet. (2014) 15:541–55.
WHO (2020) World Health Organisation, Coronavirus Disease 2019. WHO Report 127 (26 May 2020)