Network analysis

For this first analysis we used the Graph Tools of the RING database (https://precious.polito.it/theringdb/login) [4]. This tool integrates more than 30 publicly available data repositories and allows to reconstruct interaction networks between genes, Transcription Factors, miRNAs, Drugs, Diseases, and SNPs.

As a first step we performed a network reconstruction looking only for interactions among the proteins of the HCoV-host interactome as reported in S3 Table in the Supplemental material of [1]. In order to build the minimum set of most reliable interactions, we used the most conservative settings, where all interactions are validated, manually curated, and limited to signaling or regulatory actions like inhibition/activation (data sources: TRRUST [8] and SIGNOR [9]). This phase allowed to divide the original HCoV-host interactome proteins in two sets: one that shows no obvious regulatory interactions, and one made of 20 genes (see Table 1) out of the original 135 proteins, that forms a well-defined regulatory interactome. For all interactions we report (in S3 Table of the supplemental material), the Pubmed references of the supporting papers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. List of highly-likely interacting genes within the HCoV-host interactome.

https://doi.org/10.1371/journal.pcbi.1008238.t001

The HCoV-host Network has been then enhanced looking for co-expressed microRNAs (in this case we selected the MIRIAD data source [10], while relaxing the Validated and Manually curated filters). This operation highlighted that 3 miRNAs are co-expressed by the identified cluster: hsa-miR 3912-3p and hsa-miR 3912-5p hosted by the NPM1 gene, and hsa-miR 4751, hosted by ATF5 the gene. We decided to highlight miRna possibly involved in the Covid interactome because miRNAs are known to mediate several regulatory mechanisms, but also to be powerful biomarkers for several diseases. Finally, we added the IL-6 gene to see if it presented any interaction with the identified cluster.

The resulting network is presented in Fig 1. Yellow edges represent multiple interactions. For each node of the network, the S1 Table reports the node name, type, UniprotId, and possible aliases. For each interaction, the S2 Table reports the interaction type (to match the symbol with its meaning refer to [4]), the database of origin, and, where available, the PubMed Ids of the related papers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. HCoV-host and IL-6 interactome cluster.

Nodes are represented as follow: genes in violet, transcription factors in blue, and microRNAs in green. Edges are marked in: red for inhibition/silencing effects, green for activation/silencing, black for undirected/unknown and yellow when multiple regulations are available in literature (see the list of all the interactions reported in Supplemental Material in S2 Table).

https://doi.org/10.1371/journal.pcbi.1008238.g001

As a third step, we further enhanced the network looking for drug interactions that had at least two targets in the identified cluster (data sources: DGIdB [11] and DRUGBANK[12]). In this way we limited the number of drugs to only the ones that possibly have a stronger effect on the HCoV-Cluster of proteins, by presenting multiple targets. The resulting network is presented in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. HCoV-host and IL-6 interactome cluster enriched with drugs interaction data.

Nodes are represented as follow: genes in violet, transcription factors in blue, microRNAs in green, and drugs in cyan. Edges are marked in: red for inhibition/silencing effects, green for activation/silencing, black for undirected/unknown, fuchsia for drug-target association, and yellow when multiple regulations are available in literature (see the list of all the interactions reported.

https://doi.org/10.1371/journal.pcbi.1008238.g002

For the sake of clarity and readability we kept the proposed HCoV interacting network as small as possible and we avoided including further possible regulations with weaker reliability. Custom enhancements are obviously possible for interested scientists, and instructions are reported in the S1 Instruction file.

The list of drugs that have at least two interactions with the cluster is reported in Table 2. Although, the list of drugs reported only shares Paroxetine with of [1], the others could still be of interest as possible repurposing candidates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. List of possible drug interactions with the identified HCoV-host cluster.

https://doi.org/10.1371/journal.pcbi.1008238.t002

The proposed HCoV interacting network may be helpful to all scientists currently involved in SARS-CoV-2 research, since it provides a compact but comprehensive glance on some of the current state-of-the-art regulations that take place among the HCoV-host interactome.

Enrichment analysis

For this second analysis we resorted to an R [13] pipeline, taking advantage of clusterProfiler package [14], and we fed it with both the full-set of 135 HCoV-host proteins, and HCoV cluster of 20 proteins identified in the previous analysis.

With both the sets we performed statistically valid enrichment analysis on:

• the three Gene Ontologies (GO [5]) (i.e., Molecular Function, Cellular Component and Biological Process);
• two of the most widely used pathway repositories KEGG [6] and WikiPathways[7]);

Finally, we also added a one non-statistical GO classification.

The result is provided as an interactive website (available at https://precious.polito.it/covid-19) that collects and presents all results both in graphical and tabular form; all the tables are immediately available to download to, once again, provide scientists with immediately usable result for further analysis. All results are also available in the S1 Data file.

Each enrichment analysis result is provided with its entity (i.e., a GO or Pathway) defined with its ID and Description, the gene ratio coverage, a list of proteins involved in that pathway or GO, and its enrichment p-value adjusted by Benjamini & Hochberg (BH) method [14]. This value has been computed to better control the expected proportion of false discoveries amongst the rejected hypotheses (i.e., false discovery rate, FDR), thus allowing for a less stringent condition on false discoveries, and allowing more candidate options to be included as results. Furthermore, we also computed, only for GOs, a non-statistical classification, in this case, possibly helpful for preliminary evaluation or hypothesis definition. In this case we simply showcase the frequency of recurring GOs among the entities, with no statistical consideration. Because of that, data tables related to GO classification do not report any p-value.

Finally, a cluster analysis has been computed in order to selectively elucidate possible inner differences between the HCoV-host protein-set and the HCoV-host proteins cluster-set in all the computations performed in the previous steps (i.e., GO Classification GO Enrichment and Pathway Enrichment). The same statistical assumption and limits, previously discussed for enrichment analysis, do apply to cluster analysis.

As a proof of concept, we hereby show in Table 3, the top five Wiki Pathway enriched pathways.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. List of the top 5 enriched Wiki pathways.

To see the patway full content in the original wikipathway website use the following link https://www.wikipathways.org/index.php/Pathway:<PathwayID> by replacing <PathwayID> with the id reported in the table.

https://doi.org/10.1371/journal.pcbi.1008238.t003

WP3872 describes integrin mediated cell survival regulation induced by parathyroid hormone-related protein. Although the pathway may seem not of immediate applicability for SARS-CoV-2, it is interesting to highlight the involvement of the PI3-K/Akt pathway by increasing levels of integrin A6B4, which further modulate the pro/anti-apoptosis members in the Bcl-2 family. Bcl-2 family of genes have been shown to play an important role in the IL-6–mediated protective response to oxidative stress. Authors in [15] showed that IL-6 induced Bcl-2 expression, both in vivo and in vitro, disrupted interactions between proapoptotic and antiapoptotic factors, and suppressed H2O2-induced loss of mitochondrial membrane potential in vitro. Concluding that IL-6 induces Bcl-2 expression to perform cytoprotective functions in response to oxygen toxicity, and conclude that IL-6 induces Bcl-2 expression to perform cytoprotective functions in response to oxygen toxicity, and that this effect is mediated by alterations in the interactions between BAK and MFNS.

WP4298 refers to Viral Myocarditis (VM) pathway. VM is a rare cardiac disease associated with the inflammation and injury of the myocardium result of cooperation between viral processes and the adaptive as innate host's immune response (see [16–18]). Recent papers address SARS-CoV-2 as responsible for acute myocarditis or fulminant myocarditis nevertheless author state that the mechanism of cardiac pathology caused by SARS-CoV-2 needs further study [19, 20].

WP127 and WP286 are related to interleukine-mediated (IL-5, IL-3) inflammatory response.

Acute Respiratory Distress Syndrome (ARDS) induced by SARS-CoV-2, has been recently highlighted as mediated by high level of cytokine IL-6 [2, 21] that leads to excessive inflammatory response, which is further related to bad prognosis. While IL-6 may be considered as a therapeutic target on his own, common ILs regulatory traits (shared with IL-3 and IL-5, may be helpful to highlight more detailed mechanisms of action in the inflammatory response.

WP3646 pathway reports main hub genes and their related miRNAs. The pathways has been built on a set of differentially expressed genes in both chronic HCV (hepatitis C virus) and HCC (hepatocellular carcinoma) to highlight how Hepatitis C Virus leads to hepatocellular carcinoma [22]. This pathway suggests a possible similar behavior for corona family viruses and the liver involvement during infection has been recently highlighted, but still largely uncovered [23].