Mengran Wang, Johanna B Withers, Piero Ricchiuto, Ivan Voitalov, Michael McAnally, Helia N Sanchez, Alif Saleh, Viatcheslav R Akmaev, Susan Dina Ghiassian
Abstract:
Introduction: Viruses are obligate intracellular parasites that can only replicate by entering into a host cell and hijacking host cell machinery to produce and assemble new progeny virions. The arsenal to treat viral infections focuses on viral and cellular components that are essential for the viral life cycle. There are two main categories of antiviral therapies: (1) host-targeted antivirals that modulate host proteins that interact with or are influenced by the virus and (2) virus-targeted antivirals that directly bind to and modulate the activity of viral proteins (1). Drugs that target viral proteins may directly limit viral replication and propagation (1).
The medical and scientific communities do not have the luxury of time to develop new compounds that target highly contagious life-threatening viruses with the rapidity required to combat an ongoing pandemic (2). To facilitate the advancement of drugs into clinical trials, identification of drugs that were previously approved for other indications and exhibit a reasonable safety profile represent a resource for potential antiviral therapeutics (3, 4, 5 Preprint).
The coronavirus disease 2019 (COVID-19) pandemic, caused by the spread of SARS-CoV-2, has highlighted the need for tools to rapidly identify effective therapies against emerging pathogens. Numerous drugs have been identified that are predicted to alleviate COVID-19 symptoms and are in clinical trials to assess their safety and efficacy (6, 7). Identification of multiple treatment options is important to control the spread of the disease: patients respond differently to the same treatments because of their genetics (8), sequence variations in the viral genome may influence drug efficacy (9), and bottlenecks in drug availability may occur once an effective treatment is identified (10).
One consideration when evaluating drug repurposing opportunities is the ability of a drug to bind to multiple protein targets, called drug promiscuity. It has been previously shown that a drug’s promiscuity is correlated with structural similarity and binding site similarity between the intended and unintended protein targets (11). In this study, network-based tools and bioinformatic approaches identified drug repurposing opportunities that were predicted to directly target viral proteins. The two antiviral drug discovery methods developed in this study were designed to first identify existing protein targets that are structurally similar to viral proteins then, subsequently, predict and rank the interactions likely to occur between existing compounds and viral proteins. These methods were validated using the human immunodeficiency virus 1 (HIV), hepatitis B virus (HBV), hepatitis C virus (HCV), and human papillomavirus (HPV) then used to predict candidate antiviral drug repurposing opportunities for SARS-CoV-2.
Materials and Methods: The Human Interactome was consolidated as previously described (62, 63) from 21 public databases containing different types of experimentally derived PPIs data:
- Binary PPIs, derived from high-throughput yeast-two hybrid (Y2H) experiments (HI-Union 2016), 3D protein structures (Interactome3D (65), Instruct (66), Insider (67)) or literature curation (PINA 2014, MINT 2019, LitBM17 2013, Interactome3D, Instruct, Insider, BioGrid 2019, HINT 2019, HIPPIE (72), APID (73), InWeb (74), IntAct (75))
- PPIs identified by affinity purification followed by mass spectrometry present in BioPlex2 (2017) (76), QUBIC (77), CoFrac (78), HINT, HIPPIE, APID, LitBM17, and InWeb
- Kinase-substrate interactions from KinomeNetworkX (79) and PhosphoSitePlus (80)
- Signaling interactions from SignaLink (2019) (81) and InnateDB (2019) (82); and
- Regulatory interactions derived by the ENCODE consortium (2012). The curated list of molecular interactions provided by Alonso-López et al (73) was used for differentiating binary interactions among the experimental methods present in the literature curation databases. For InWeb, interactions with a curation score <0.175 (75th percentile) were not considered. All proteins were mapped to their corresponding Entrez ID (NCBI), with unmapped proteins removed. The resulting interactome includes 18,505 proteins and 327,924 interactions. The network analyses were limited to the largest connected component, containing 18,446 proteins and 322,159 interactions.
Results: In this study, two network- and sequence-based methods were developed to identify candidate drug repurposing opportunities that directly target viral proteins. An underlying assumption behind many drug–target interaction predictions is that structurally similar proteins are more likely to be targeted by similar drugs (11, 12, 13). Using complementary approaches, the methods described herein identify drugs by virtue of homology between viral proteins and proteins that are the known target of therapeutics developed for other indications. Both methods were designed to find proteins that are structurally similar to viral proteins; method 1 inferred structural similarity through common protein–protein interaction (PPI) patterns and method 2 derived structural similarity through sequence homology (Fig 1). When used in combination, these complementary methods derive high-confidence predictions of drug repurposing candidates.