Human/SARS-CoV-2 genome-scale metabolic modeling to discover potential antiviral targets for COVID-19

Abstract

Background: Coronavirus disease 2019 (COVID-19) has caused a substantial increase in mortality and economic and social disruption. The absence of US Food and Drug Administration-approved drugs for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) highlights the need for new therapeutic drugs to combat COVID-19.

Methods: The present study proposed a fuzzy hierarchical optimization framework for identifying potential antiviral targets for COVID-19. The objectives in the decision-making problem were not only to evaluate the elimination of the virus growth, but also to minimize side effects causing treatment. The identified candidate targets could promote processes of drug discovery and development.

Significant findings: Our gene-centric method revealed that dihydroorotate dehydrogenase (DHODH) inhibition could reduce viral biomass growth and metabolic deviation by 99.4% and 65.6%, respectively, and increase cell viability by 70.4%. We also identified two-target combinations that could completely block viral biomass growth and more effectively prevent metabolic deviation. We also discovered that the inhibition of two antiviral metabolites, cytidine triphosphate (CTP) and uridine-5'-triphosphate (UTP), exhibits effects similar to those of molnupiravir, which is undergoing phase III clinical trials. Our predictions also indicate that CTP and UTP inhibition blocks viral RNA replication through a similar mechanism to that of molnupiravir.

Keywords: Bioprocess systems engineering; Computer-aided drug discovery; Constraint-based modeling; Evolutionary optimization; Flux balance analysis; Fuzzy optimization.

Graphical abstract

Fig. 1

Flowchart of computer-aided screening for antiviral targets to combat COVID-19. (A) Gene and protein sequences of SARS-CoV-2 were downloaded from NCBI database. (B) A pseudo-reaction was constructed as a viral biomass objective function. (C) The human genome-scale metabolic network Recon3D was downloaded from Virtual Metabolic Human (https://www.vmh.life/). (D) The genome-scale metabolic model of the host-virus cells was created. (E) Flux distribution patterns for host cells were obtained from clinical data if available; otherwise, the template values were computed with flux balance analysis (FBA) and uniform flux distribution (UFD) problem without perturbation. (F) A set of antiviral targets was identified using the nest hybrid differential algorithm (NHDE), and used to compute the flux distributions of the host-virus cells during treatment. (G) The same targets were used to compute the flux distributions of perturbed host cells during treatment. (H) From fuzzy set theory, the flux distributions of the template, treated and perturbed cells are determined to evaluate the fitness of the targets. (I) If the fitness was unsatisfactory, the next antiviral targets were generated using the NHDE algorithm, and the procedure was repeated. (J) If the fitness was satisfactory, the target was identified as potential candidate.

Fig. 2

Illustrations of gene- and metabolite-centric approaches. (A) In the gene-centric approach, the reaction r₁ is catalyzed by isozymes, E₁ and E₂, and the reaction r₂ is regulated by E₁. The isozymes, E₁ and E₂, are knocked out to inhibit r₁, and r₂ is also blocked by E₁. (B) In the metabolite-centric approach, the metabolite, M₁, is inhibited, and the synthesis reactions, r₂, r₃ and r₄ are there by also inhibited.

Fig. 3

Stoichiometric analysis of amino acids and nucleotides in the viral biomass reaction. (A) Log2 fold changes of stoichiometric coefficients in amino acids and nucleotides of Delta and Alpha variants versus to host cells. (B) Stoichiometric coefficients of amino acids and nucleotides in the biomass reaction of Delta variant and host cells.

Fig. 4

Integration of a concise metabolic network regulated by the enzyme DHODH with viral replication. DHODH inhibition downregulates the conversion of Dhor-S to Orot in the host–virus cells. The numerical values in the box $\frac{\text{TR}|\text{PB}}{\text{HV}|\text{HT}}$ indicate the metabolite flow rates (mmol/gDW/h) for treated cells (TR), perturbed cells (PB), host–virus cells (HV), and host cells (HT).

Fig. 5

Membership grades for viral biomass objective function (VBOF), cell viability (CV), and metabolic deviation (MD) under various multi-target combination treatments.

Fig. 6

Membership grades for viral biomass objective function (VBOF), cell viability (CV), and metabolic deviation (MD) for treatments targeting each of 20 amino acids and four nucleotides.

Fig. 7

Membership grades for two-target combinations. The NHDE algorithm was applied to discover the most favorable two-target combinations, each of which consisted of one of 20 amino acids or one of four nucleotides with another metabolite in the model.

Fig. 8

Membership grades for single-target antimetabolite and two-target combination treatments. These antimetabolites were determined by the NHDE algorithm and excluded the viral biomass building blocks—comprising 20 amino acids and four nucleotides—as candidates.