Novel Approaches

Major challenges in antiviral drug development are connected to the huge size of data sets and the limitations of conventional methods. The PANVIPREP drug development workflow will incorporate innovative approaches to increase speed and success rate. Important elements are highlighted below.

State-of-the-art antiviral toolboxes for screening and antivirome studies

Virological toolboxes will be expanded by PANVIPREP’s molecular virology expert groups, to accelerate hit identification and mode of action studies further. Engineered or attenuated viruses will be developed to be handled at lower biosafety levels to facilitate work with pathogenic viruses. We will engineer biosafe, single-round viruses (only able to replicate in an engineered complementing host cell line), and use genome-wide recoding to generate attenuated viruses that can be used at the BSL-2 level. Viruses expressing fluorescent and/or luminescent reporter proteins will be designed for high-content imaging assays to support phenotypic screening. It enables the simultaneous evaluation of broad-spectrum activity against multiple viruses, each expressing a different reporter protein. The antiviral screening will be refined using physiologically relevant models based on primary human epithelium cells to study respiratory virus pathogenesis and evaluate antiviral drug efficacy. Partner Epithelix has developed standardized air-liquid interface 3D human airway epithelial cultures (nasal, bronchial, small-airway and alveolar) that closely mimic morphology and function of native tissues (e.g., cilia formation and mucus production). Screening data will feed back into the project’s pipeline to optimize further these innovative primary cell culture models, which can also be used to evaluate airborne drug delivery methods (nasal spray or nebulizers). The models constitute an additional screening layer to reduce the number of animal experiments for compound evaluation, in line with the 3R principle.


Proteolysis targeting chimeras (PROTACs)

As an alternative broad-spectrum approach to target viral proteins, we will explore PROTACs as an innovative antiviral format constituting a recent paradigm shift in drug discovery. So far, PROTACs have received attention in other areas, particularly oncology, but have hardly been considered for antiviral use. PROTACs rely on a catalytic (‘event-driven’) instead of occupancy-driven MoA, as they deploy the host’s proteasome system to degrade their viral target. This fast and reversible chemical knockdown strategy allows for lower intracellular concentrations than conventional non-catalytic inhibitors. Due to much lower target affinities, PROTACs are potentially less susceptible to resistance development. Moreover, they allow access to previously “undruggable” protein targets as they can bind at any site of the protein. Partners HZI and UzL have set up a PROTAC toolbox that will be adapted to binders provided by other partners. We will explore the PROTAC approach to combat viral infections using selected hits from the WP4 mini-projects by designing PROTACs with different linkers and E3-ligase binders to boost activity.


Innovative viral drug targets

For many virus groups, worldwide screening efforts, studies of evolutionarily conserved viral enzymes and other functions, and virus variant analysis have identified potential drug targets in unexpected viral genes or sequences. The latter will powerfully complement antiviral drug design efforts on more conventional targets such as viral polymerases and proteases, thus expanding possibilities for developing synergistic drug combinations. PANVIPREP will explore these potentially important novel viral targets, specifically conserved RNA sequences/structures, membrane-associated proteins that orchestrate viral replication organelles and RNA methyltransferases, which act at the crossroad of genome expression and stability, but also counteract innate immune responses of the host cell.


Innovative drug delivery

Some of PANVIPREP’s selected virus groups cause brain infections, and newly emerging viruses with epidemic/pandemic potential may be neurotropic. Thus, it will be essential to have methods to deliver antiviral drugs to the brain efficiently and in a patient-friendly manner. The well-vascularised nasal cavity is directly connected to the brain through the olfactory and trigeminal nerves. It can thus be explored for direct nose-to-brain (NtB) delivery of antiviral molecules packaged as nanoparticles. This approach also allows the masking of physicochemical properties that, when administered as an unformulated drug, may impede efficient passage through the blood-brain barrier. Protamine and polyarginines are cell-penetrating peptides that, when used in nanoparticle composition, can promote the delivery of molecules via the NtB route. Partner CSIC will prepare nanoparticles by functionalising the surface with lectin-binding sugars, folic acid and specific antibodies to increase affinity for specific cells of the CNS and enhance the correct delivery of the associated payload. The efficacy of brain delivery of antiviral drugs for treating CNS infections, will be evaluated in animal models in collaboration with other PANVIPREP Partners.


Computational tools advancing hit identification and hit-to-lead development

Ligand-based machine learning (ML) models will be used to advance both the hit identification (ID) phase and the hit-to-lead (H2L) phase. Models will be trained using chemical information and antiviral screening data. During the hit ID phase, compound activity is used to train a classifier ML model that predicts the activity of untested compounds. This model virtually screens billions of compounds and selects novel chemical space for in vitro testing. In the H2L phase, biological activity will be encoded as a pIC50 value describing potency. The chemical representations and potency values from compound-ligand series are used to train regression ML models that predict the potency of untested compounds that underwent chemical modifications. For both hit ID and H2L phases, we will explore:

  • classical ML models, like random forest using chemical fingerprints;
  • more advanced neural network models, such as graph convolutional and transformer architectures;
  • few-shot learning approaches to maximally leverage the datasets.

Only the compounds with the highest probability of being active will be selected to be synthesized and tested in vitro. ML can help chemists select chemical modifications that are more likely to result in improvements during H2L, thus accelerating potency and solubility.

Structure-based drug discovery will also predict which compounds will bind to a target, particularly if knowledge of the protein’s structure (crystal/cryo-EM structure or homology model) and the active site is available. Such docking studies can screen compounds against a target of interest and predict which chemical modifications result in improved binding modes. 

Source data: Access to various chemical structures to run predictions on is highly desirable. Our sources include the “physical” screening decks from partners (e.g. CISTIM and Janssen) and databases of “off the shelf” compounds from major suppliers. In complement, pre-compiled chemical space of over 7 trillion compounds can be searched through BiosolveIT’s Infinisee and finally, generative chemistry methodology will be applied.

CISTIM, UCTP and Janssen have the experience and facilities to perform ligand-based virtual screenings, ML-driven hit identification/expansion, and ligand optimisation. Multiple methods are available through the Schrödinger suite. Methodologies are also in place for generating further QSAR/QSPR models or AI applications, ranging from “simple” ML models to Deep Learning models. All three partners can perform structure-based studies, now applicable to increasing targets due to recent advances in structure prediction. Structure-based methods such as docking generally require greater computational resources than ligand-based methods, for which the PANVIPREP consortium has access to high-performance computing resources.

in silico PK modelling: During hit-to-lead development, it is critical to assess parameters for (pre-) clinical development. HZI will apply and develop in silico PK modelling methods (rather simplistic compartmental or more sophisticated PBPK structural constructs) to guide in vivo studies. As a measure of lipophilicity, LogP will serve as a key parameter for distribution towards target tissues in PBPK modelling. It will be crucial for PK and, in particular, efficacy studies to ensure that compounds reach the tissues infected by a particular virus. Information on metabolites gained from in vitro and in vivo experiments using HR-MS2 combined with a metabolite-finder pathway algorithm and information about CYP inhibition, induction, and permeability from Caco-2 or PAMPA assays will render predictions more accurate. Validations will be performed using different in vivo data sets. Compounds will be ranked based on antiviral, ADME and modelling data to select candidates for in vivo studies. PBPK modelling will also offer cross-species extrapolation of PK behavior towards human physiology and enable to take factors like age, gender or ethnicity into account.

Taken together, the computational approaches outlined above will offer novel ways to advance drug discovery efforts during both the hit ID and hit-to-lead stages and can help to develop drugs with more desirable profiles.