Structural and dynamical analysis of integrated human/SARS-CoV-2 metabolic models present novel treatment strategies against COVID-19

The coronavirus disease 2019 (COVID-19) pandemic caused by the new coronavirus (SARS-CoV-2) is currently responsible for over 500 thousand deaths in 216 countries across the world and is affecting over 10 million people. The absence of FDA approved drugs against the new SARS-CoV-2 virus has highlighted an urgent need to design new drugs. We developed an integrated model of the human cell and the SARS-CoV-2 virus to provide insight into the pathogenetic mechanism of the virus and to support current therapeutic strategies. We show the biochemical reactions required for the growth and general maintenance of the human cell, rst of all, in its healthy state. We then demonstrate how the entry of the SARS-CoV-2 virus into the human cell causes biochemical and structural changes, leading to a change of cell functions or cell death. We have completed a comparative analysis of our model and other previously generated cell type models and highlight 48 pathways and over 800 reactions hijacked by the virus for its replication and survival. We designed a new tool which predicts 15 unique reactions as drug targets from our models (the integrated human macrophage, human airway epithelial cells and the SARS-CoV2 virus) and provide a platform for future studies on viral entry inhibition and drug optimisation strategies.


Introduction
SARS-COV-2, the causative agent of the COVID-29 disease, belongs to a group of viruses commonly known as β-coronavirus. This class of viruses is responsible for mild-to-fatal respiratory tract infections in animals and birds. Whilst the common cold is more commonly associated with the mild forms of the disease, the previous MERS and SARS-2002 infections and the current COVID-19 disease belong to the group of fatal diseases. The genome of the virus responsible for the ongoing COVID-19 disease, SARS-CoV-2, has ~80% sequence identity to SARS-CoV and 96% identical at the whole-genome level to a bat coronavirus . The SARS-CoV-2 virus affects the lower respiratory tract cells and the upper cells in the pharyngeal region (Huang et al., 2020;Chen et al., 2020); and the range of viral infections ranges from asymptomatic, mild, moderate and severe cases. Previous studies in China show that 86% of cases of infection and the contagiousness of the virus were undocumented before travel restrictions were imposed . Therefore, there are still many unknown factors regarding the stages of infection and transmissibility patterns of the virus. Studies in France demonstrate the transmission potential of asymptomatic persons and suggest varying dynamics of transmission in children (Danis et al., 2020). The human angiotensin-converting enzyme 2 (human-ACE-2 protein) has been identi ed as the cell receptor for both the SARS-2002 virus and the SARS-CoV-2 virus. The ACE-2 enzyme, which has a primary function of controlling blood pressure, is usually found in the epithelial cells of the heart, lungs, kidneys and intestine (Hamming et al., 2014;Donoghue et al., 2000).
The mechanism of replication of the SARS-CoV-2 virus in the human cell involves an initial binding and attachment of the spike (S) glycoprotein to the ACE2 receptor of its host. During endocytosis, the genetic material of the virus is injected into the host cell, where it loses its protective envelope. The virus, now ready for replication, is released into the nucleus of the human cell (Fig 1). Subsequent assembly and maturation of viral proteins lead to cell death and a proliferation of the virus within the human body.

Page 3/20
The lack of FDA approved drugs against Covid-19, coupled with the di culties encountered globally in containing the virus, prompted the WHO to declare the outbreak a pandemic in March 2020. The has led to intensi ed efforts around the world to ght this disease. Previous studies in drug target identi cation against viral diseases such as Zika, Chikungunya and Dengue by Aller et al., 2018 introduced a system of integrating the host's macrophage and viral metabolic networks to predict a set of host reactions which, when constrained, can inhibit viral production. A recent study by Renz et al., 2020 demonstrates a similar approach and predicts drug targets against the SARS-CoV2 virus. Targets of known antiviral drugs predicted from both studies demonstrate the applicability of the integrated human/virus metabolic modelling in drug target identi cation.
We have built on these approaches by developing an integrated epithelial cell / SARS-CoV-2 virus metabolic model and employed a combination of structural and dynamical analyses to assess the model and make predictions. We have designed and developed a new tool ( ndCPcli) to carry out such analyses and to predict drug targets. We have also performed a comparative analysis of our model and another previously generated cell-type models.

Results
Comparative analysis of integrated models of infected human epithelial cell and the macrophage cell with the SARS-CoV-2 virus

Host dependent metabolic pathways
We initially demonstrated the biochemical requirements for the growth and maintenance of the human airway epithelial and macrophage cells and used the integrated models to show the essential host reactions needed for the survival and viability of the SARS-CoV-2 virus within the host's cell compartments. We have validated our models by mapping the experimentally characterised human/SARS-CoV-2 virus protein-protein interaction data from Gordon et al., 2020 on the in-silico virusintegrated human macrophage and epithelial cells. We identi ed 48 metabolic pathways from 334 metabolic pathways in the human metabolic network, including the biosynthesis and degradation pathways of amino acids, fatty acids, carbohydrates, amines, cofactors as well as core components of the central mRNA metabolism (Fig 2).
The 48 metabolic pathways that were mapped to the protein-protein interaction network produced by Gordon et al., 2020 are referred to as PPi-Pathway Intersection nodes in this manuscript (Fig 3). These include cysteine, methionine and selenocysteine amino acid biosynthetic pathways, C20 prostanoid hormone biosynthetic pathways, Vitamin D3 and Vitamin K epoxide cycle. The degradation pathways identi ed include the lysine, tryptophan, methionine, fatty acid degradation, ceramide and sphingolipid recycling pathways and phospholipases degradation; amine and heme degradation (Fig 3).  Tables 1 in S1 Table and 2 in S2 Table). It was also demonstrated that: a) the maximum growth rate of the macrophage cell in the absence of virus was 0.0269 h-1 ( Table 1 in S1 Table); and 0.012 for the human airway epithelial cell ( Table 2 in S2 Table); b) the maximum growth rate of the virus in the macrophage cell was 0.0144 h-1 and 0.0181 in the human airway epithelial cell. These numerical results mean that 0.0144 h-1 is the theoretical maximum of the growth rate of the virus in the human macrophage cell. If this ux is assigned to the viral growth reaction, then Flux Variability Analysis (FVA) (Orth et al., 2010) can be used to calculate the ranges of uxes allowed for the remaining reactions in the cell while the virus is being replicated at its optimum condition. The execution of FVA under such conditions produced a zero growth of the host cell, i.e. both the lower and upper ux bounds of the reaction indicate that the growth is zero. This means that if the virus is replicating at its maximum rate then the cell cannot reproduce.

Bottleneck reactions and the prioritization of potential drug targets
The bottleneck reactions identi ed by the ndCPcli tool are unique reactions in a metabolic network required for the growth and survival of the organism and, like chokepoint reactions, are potential drug targets ( Yeh et al., 2004;Oarga et al., 2020). Whilst the classical chokepoint reactions identify unique reactions from a stoichiometric model, we improve on this approach by using the structural and dynamical information of the integrated Human/SARS-CoV-2 metabolic model within the airway epithelial cell and the macrophage cell to predict potential drug targets against the SARS-CoV-2 virus.
We initially identi ed 1595 bottleneck reactions required for the virus' maintenance and replication in the human macrophage cell; these include pathways in lipid metabolism, coenzyme transport and metabolism, energy production and conversion, amino acid and nucleotide transport and metabolism ( Table 1 in S1 Table). In the human airway epithelial cell, 1819 bottleneck reactions were initially identi ed; these include the biosynthesis and degradation pathways of amino acids, fatty acids, carbohydrates, amines, cofactors as well as some components of the central mRNA metabolism ( Table 2 in S2 Table).
To validate/account for the results, and because each bottleneck reaction should be balanced by at least one other reaction that produces or consumes that metabolite, we have excluded reactions in the model with dead-end metabolites. The bottleneck reactions are further prioritised by interrogating the dynamical information in the model using the ux variability analysis, which determines if a reaction is reversible.
The bottleneck reactions are potential drug targets as they are indispensable for the maintenance and replication of the virus within the host. In order to rank the potential drug targets identi ed, we prioritised enzymes for unique reactions that occur at the nodes of intersection between the bottleneck and essential reactions and the experimental results from the human/virus protein-protein interaction network (Gordon et al., 2020) (Fig 3). We refer to these as PPi-Pathway intersection nodes.
The PPi-Pathway intersection (PPi) nodes identi ed are present in biosynthesis pathways such as the cysteine and S-adenosyl-L-methionine biosynthetic pathways. In both pathways, the enzyme Sadenosylmethionine synthase (Mat2b), catalyses the phosphorylation reaction of methionine to Sadenosyl-L-methionine. During infection, the viral protein Nsp9 is seen to react with MAT2B (Gordon et al., 2020) (Fig 4a/b). Another viral protein, Nsp8 also interacts with the enzyme O-phosphoseryl-tRNA(Sec) selenium transferase (SEPSECS), which catalyses the last step of the L-selenocysteine biosynthesis pathway ( Fig 5).
PPi nodes also occur in a network of various fatty acid and stearate biosynthetic pathways with Nsp2 interacting with the very long-chain acyl-CoA synthetase (SLC27A2) (Fig 6a). In other fatty acid biosynthetic pathways, γ-linolenate biosynthesis, Nsp7 interacts with ACSL3 (Fig 6a)The viral protein, Nsp2, also interacts with POR in other pathways including vitamin D3 biosynthesis, L-tryptophan degradation, ceramide and sphingolipid recycling (Fig 6b).
In carbohydrate metabolism, a PPi node is identi ed at the glycan & oligosaccharide biosynthetic pathways, and speci cally where two mannose residues are added in α(1→2) linkages to the nascent oligosaccharide and catalysed by the enzyme ALG11. The viral protein Nsp4 interacts with ALG11 during the infection of the SARS-CoV-2 virus (Fig 7). Other viral proteins, Nsp7 (Fig 8) reacts with ACSL3 and ORF8a interacts with HS2ST1 (Fig 8b), a key enzyme involved in the heparan sulfate biosynthesis pathway. The rst enzyme of the N-linked oligosaccharide processing pathway, mannosyloligosaccharide α-1,2-glucosidase (MOGS), also interacts with Nsp7 and ORF8a (Fig 8c).
PPi nodes speci c to the human macrophage cell include the O-phosphoseryl-tRNA(Sec) selenium transferase in the L-selenocysteine biosynthetic pathway, which interacts with the viral protein Nsp8. The alkylglycerone-phosphate synthase/Nps7 PPi node, which is present in the Phospholipid/Plasmalogen biosynthetic pathway is also speci c to the macrophage cell. Alternatively, PP-pathway intersection nodes common to both human airway epithelial cell and the macrophage cell are the MAT2B/Nsp9 intersection pathways present in the cysteine metabolism and L-methionine degradation. We did not identify PPi nodes speci c to the human epithelial cell. The dynamical changes of ux metabolism in our in-silico virus optimal models show signi cant increase in viral infection. Four reactions involved in the biosynthesis of fatty acids with predicted non-zero uxes in the host model exhibit an average increase of 190% in their maximum uxes in the viral model (the maximum increase is 298%). The average increase of 32 reactions in lipid metabolism with non-zero uxes in the host model is 277% (the maximum increase is 498%). With respect to sphingolipid metabolism, 14 out of 15 reactions with non-zero uxes in the host model exhibit an average increase of 228% (the maximum increase is 298%) and similar increases in phospholipases and palmitic acid biosynthesis (Fig 6). We show an average increase of 190% in cholesterol and fatty acid metabolism during viral infection and demonstrate the essentiality of these pathways to SARS-CoV-2 virus. Previous studies have shown that cholesterol and fatty acids are main components of the viral membranes and needed for viral replication (Heaton and Randall, 2011); therefore, drugs inhibiting these pathways such as AM580, Statins, Fibrate (Fiévet et al., 2009) will be essential for both early and late stages of the Covid-19 disease.

Discussion
Sphingolipids are composed of both hydrophobic and hydrophilic units and play a large role in the endocytic or exocytic viral entry processes into the cell (Dimitrov et al., 2004). The pH-dependent endocytic process is further enhanced by the presence of clathrin, a protein present in the plasma membrane, Golgi apparatus and in the cytoplasm, whilst the exocytic route involves viral crossing through the plasma membrane at neutral pH. Our results show a 3-fold increase in sphingolipid metabolism during viral infection (Fig 6); we hypothesise that drugs inhibiting sphingolipid metabolism and/or the endocytosis process will inhibit infection of SARs-CoV-2 virus. Indeed, a sphingosine kinase-2 (SphK2) inhibitor, Opaganib, which has proved bene cial in the treatment of Covid-19 is currently in  (table 2) and in our PPi-pathway intersection nodes (Fig 3).
Our lists of bottleneck, essential reactions and PPi-pathway intersection nodes also include critical points in the biosynthesis of phospholipids. We show that these reactions are essential for viral infection and replication (Fig 3) and propose that targeting the phospholipase enzyme or the interacting Nsp2 protein could inhibit viral replication. Our results support previous studies from Muller et al., 2018, that targeting the phospholipase enzyme could inhibit the early stage of Covid-19 disease.

Redox homeostasis and antioxidant therapy
Redox homeostasis refers to the ability of the cell to maintain its balance amidst infections and other unstable cellular environmental factors. Delgardo-Roche and Mesta, 2020 have described oxidative stress as a key player in Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV) Infection with cytokine production. Foyer and Noctor 2005 have previously shown that antioxidants, such as glutathione and ascorbate, are important metabolites for the cellular redox state. Our studies have identi ed key target enzymes involved in the metabolism of glutathione and ascorbic acid as bottleneck and essential reactions; including glutathione synthase, glutathione peroxidase and ascorbic acid oxidase (Table 2). We also demonstrate an increase in ux of these enzymatic reactions on infection of the virus. In a recent study, Horowitz et al., 2020 previously demonstrated how the use of high dose oral and/or IV glutathione on severe outcomes of SARS-CoV-2 virus led to favourable treatment outcomes. Other studies have shown that steroids such as dexamethasone and Methylpredisone to treat severe cases of Covid-19. Due to possible side effects of steroid treatment, we propose the use glutathione as therapy for severe cases of Covid-19 in the aged population and other severe cases with cytokine storm syndrome.

Immune regulation
The SARS-CoV-2 virus is able to proliferate unhindered in infected cells, due to the lack of immunity in humans (Felsenstein et al., 2020). The result is cell death, a release of viral particles to the extracellular environment and a general hyperactivity of the immune system in some patients with severe Covid-19 disease and subsequent, lung in ammation and cytokine syndrome. Immunocompromised patients or those with underlying symptoms such as diabetes hypertension and transplantation are most affected (Zhong et al., 2020). Whilst clinical trials are ongoing worldwide with various antivirals and immune modulating treatments, there is currently limited knowledge on the host dependency factors responsible for the individual outcomes of the disease. Our results provide insight into the immune evasion strategies of SARS-CoV2; we demonstrate changes in the ux metabolism of vitamin D and tryptophan metabolism during viral infection. Vitamin D is important for bone growth and turnover and a low vitamin D status is associated with an increased susceptibility to upper respiratory tract infections (Mitchell F, 2020).
Previous studies have shown that a supplementation of vitamin D prevents acute respiratory tract infections (Martineau et al., 2016). Our results highlight vitamin D as an essential reaction in the PPipathway intersection nodes and we show the viral protein Nsp2 interaction with key enzymes in the vitamins D and C metabolism pathways (Fig 3 and 6). SARS-CoV-2 viral infection causes metabolic perturbations of vitamin D metabolism in the host resulting to disruptions in cellular homeostasis. We propose support therapy management strategies where vitamin D supplements are provided to all Covid-19 patients. Our results also show that tryptophan, melatonin and prostaglandins, important compounds for immunity and homeostasis (Platten et al., 2019;Gitto et al., 2010), are affected by the infection of SARS-CoV-2 virus and we provide insight into the viral mechanism of action within the human body.
In summary, we have provided a platform for drug target prediction against Covid-19, and for future studies on viral entry inhibition, antioxidant therapy and immune regulation.

Methods
We manually curated the human airway epithelial cell initially constructed by Wang et al., 2017 with gene expression datasets of the human airway epithelial cell (Deprez et al., 2020;Braga et al., 2020) and the humancyc database (Romero et al., 2014) to produce a new GEM, (iBBEC4660); i for in-silico, BB for the rst author's name, EC for airway epithelial cell and 4660 for the number of open reading frames. In order to assess and predict the performance of the models, we made use of Flux Balance Analysis (FBA) and Flux Variability Analysis (FVA) (Orth et al., 2010). FBA is a computational method that can be applied e ciently to genome scale models to estimate the uxes of reactions at steady state. It is based on the solution of a linear programming problem that maximizes an objective function of interest subject to a set of constraints on the uxes of the reactions. The linear programming problem associated with FBA can be expressed as: where v is the vector of uxes, c represents the objective, S is the stoichiometry matrix, and L and U are lower and upper bounds on the uxes. Thus, · is the objective function, which usually refers to the growth rate of the organism, and · = 0 represents the balance of uxes at steady state.
FVA is also based on the solution of linear programming problems, and its main use is the computation of ranges of uxes that are compatible with given ux constraints. For instance, if the growth rate predicted by FBA is µ max , then the range of uxes of a given reaction i that are compatible with such growth rate can be obtained by minimizing and maximizing the following programming problem: where v growth is the ux of the reaction associated with growth. FBA and FVA were applied on the metabolic network of the host, both with and without reaction modelling the production of the virus, by using the Python toolbox COBRApy (Ebrahim et al., 2013).
Bottleneck reactions, like chokepoint reactions, are required for the reaction synthesis and the removal of these reactions will cause an accumulation or depletion of the metabolites; they represent potential drug targets. The tool ndCPcli is a command line application intended for the computation of bottleneck reactions on genome scale models. The source code of the tool is available at github.com/ ndCP/ ndCPcli. This tool is distributed as a Python package and requires an installation of a Python language interpreter with a version 3.5 or higher and the application can be installed with the pip package management tool.
The pathway maps were created with the pathway collage software (Paley et al., 2016) and the model was deposited in the BioModels database (Chelliah et al., 2015) and assigned the identi er "MODEL2007210001".
Declarations Figure 1 The mechanism of replication of the SARS-CoV-2 virus in the human cell PPi-pathway intersection nodes PPi-Pathway intersection node -Nsp9 PPi-Pathway intersection node -Nsp8

Figure 6
Page 19/20 PPi-Pathway intersection node -Nsp2 as denoted by the red square blocks. The red lines indicate an increase in ux of the highlighted reactions when the virus hijacks the system Figure 7 PPi-Pathway intersection node -Nsp4 Supplementary Files This is a list of supplementary les associated with this preprint. Click to download. Supplementary.zip