Multi-omics data integration reveals the complexity and diversity of host factors associated with influenza virus infection

Data collection

Five kinds of genes or proteins including differentially expressed genes (DEGs), differentially expressed proteins (DEPs), proteins with differential post-translational modifications (DPMs), proteins that participate in human-virus protein–protein interactions (P-PPIs) and human susceptibility factors (SHFs) associated with infection of human or human cells were obtained for 14 subtypes of influenza A viruses, as well as influenza B and C viruses, from publicly available databases and the literature.

Specifically, DEGs were extracted from the HVIDB (Version 1.0) (Yang et al., 2021) database. The DEPs were manually collected from the Influenza Research Database (IRD) (https://www.fludb.org/) (Zhang et al., 2017) and from literatures in the PubMed database by searching “(influenza[TIAB] OR flu[TIAB]) AND ((proteome[TIAB]) OR (proteomics[TIAB]))” on March 1st, 2023. The DPMs were manually collected from literatures in the PubMed database. The PPIs between influenza viruses and humans were obtained from public databases including Viruses.STRING (Version 10.5)  (Cook et al., 2018), IntAct (Version 4.2.3.2) (Orchard et al., 2014), BioGrid (Version 4.4.208) (Stark et al., 2006), Virhostnet(Version 3.0) (Guirimand, Delmotte & Navratil, 2015), HVIDB (Version 1.0), VirusMINT (Version 1.0) (Chatr-aryamontri et al., 2009), VirusMentha (Version 1.0) (Calderone, Licata & Cesareni, 2015) and HVPPI (Version 1.0) (Li et al., 2022). To ensure the reliability of the data, only experimentally validated PPIs from these databases were selected. The PPIs were organized by virus type or subtypes.

To obtain the SHFs of the influenza virus, we firstly collected 1254 abstracts from the PubMed database (Canese & Weis, 2013) by searching “(influenza[TIAB] OR flu[TIAB]) AND human[TIAB] AND((susceptible[TIAB]) OR (susceptibility[TIAB]) OR (sensibility[TIAB]) OR (sensitiveness[TIAB]) OR (susceptivenes[TIAB]))” on April 13th, 2023. Then, each abstract was manually screened based on whether it contained SHFs, which resulted in 48 abstracts. Finally, the full texts of these abstracts were read carefully and 57 SHFs were compiled from these papers.

Functional enrichment analysis

The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of human genes were conducted with functions of enrichGO() and enrichKEGG() in the package “clusterProfiler” (version 4.8.2) (Wu et al., 2021) and org.Hs.eg.db(version 3.17.0) (Carlson et al., 2019) in R (version 4.0.3) (R Core Team, 2013).

Statistical analysis

All statistical analyses were conducted in R (version 4.0.3) (R Core Team, 2013). The Wilcoxon rank-sum test was conducted using the wilcox.test() function; the linear regression fitting was performed using the lm() function; the correlation coefficient was calculated using the cor.test () function.

Data summary

As shown in Fig. 1A and Table S1, two human influenza A viruses including A(H1N1) and A(H3N2) had a large number of genes or proteins associated with human infection (for clarity, they were defined as Virus-Infection-associated Human Factors (VIHFs)). The former had a total of 12,556 VIHFs which included 10,068 DEGs, 1,435 DEPs, 104 DPMs, 2,654 P-PPIs and 48 SHFs; the latter had 15,857 VIHFs which included 15,250 DEGs, 41 DEPs, 342 DPMs, 534 P-PPIs and seven SHFs. Besides human influenza A viruses, two avian influenza viruses, i.e., A(H5N1) and A(H7N9), also had a large number of VIHFs, with the former having 24,683 VIHFs and the latter having 6,244 VIHFs. The remaining viruses, including H3N8, H4N6, H5N2, H5N3, H5N6, H6N8, H7N1, H7N4, H7N7, H9N2, Influenza B virus and Influenza C virus possessed one, one 286, 83, one, one, one, one 5,093, 128, 21 and six VIHFs, respectively.

When comparing VIHFs between influenza types or subtypes, as shown in Fig. 1B, we observed a very small proportion of shared VIHFs. Two human influenza A viruses, i.e., A(H1N1) and A(H3N2), had a high ratio of shared VIHFs. Subtypes with the same HA subtype seemed to have a relatively high ratio of shared VIHFs, such as subtypes of H5N1, H5N2 and H5N3, and subtypes of H7N9 and H7N7.

When comparing VIHFs of different types, we found a large difference between them. Although DEGs accounted for most VIHFs, each kind of VIHFs had unique genes. As shown in Fig. 1C, DEPs, DPMs, P-PPIs and SHFs had 180, 257, 508 and five unique genes, respectively. Interestingly, there were some genes shared among multiple kinds of VIHFs. For example, 19 genes were shared between DEGs, DEPs, DPMs and P-PPIs, suggesting that they may play important roles in influenza virus infections. As shown in Fig. 1D, over 90% of SHFs overlapped with other kinds of VIHFs, indicating that most susceptible genes were disrupted after influenza virus infection.

The meta information of VIHFs including the relavant viral strains, cell type and infection time points were also provided (Fig. 1E). Transcriptomic data from 14 influenza virus strains infecting 8 cells at 16 different time points, and proteomic data from seven influenza virus strains infecting 3 cell lines at 8 time points, were compiled separately.

Dynamic response of human cells to influenza virus infection

The dynamic response of humans to influenza virus infection was analyzed since there were a large number of DEGs. The number of DEGs identified at different time points post infection was analyzed for 14 viral strains which infected 8 different cell types (Fig. 2A). It was found that there was a small number of DEGs within 10 h post infection(hpi); then, the number of DEGs increased rapidly from 10 to 20 hpi, reaching a peak around 20 hpi; then, it began to decrease and kept stable after 30 hpi. When comparing the shared DEGs between different time points, we found that there was a negative correlation (Pearson Correlation Coefficient = −0.30) between the shared ratio of DEGs and the size of time intervals (Fig. 2B). DEGs in different time points had a small ratio of overlaps except those after 24 hpi (Fig. 2C), indicating distinct DEGs at different stages of virus infections.

The cell specificity contributed larger to the diversity of DEGs than the strain specificity

We further analyzed the strain and cell type specificity for DEGs. As shown in Fig. 3A, different strains of influenza viruses had varying numbers of DEGs with the median number of DEGs ranging from 22 to 4,798. Even within the same subtype such as A(H1N1), DEGs in different strains also differed much. Interestingly, two strains of highly pathogenic avian influenza H5N1 viruses which led to human infection and death, i.e., A/Vietnam/1203/2004 and A/Vietnam/UT3028II/03, had the largest number of DEGs, while avian influenza viruses such as A/duck/Malaysia/F118/08/2004 and A/duck/Malaysia/F119/3/1997 generally had a small number of DEGs. To remove the influence of cell type, we also compared the number of DEGs between virus strains which infected the same cell type. As shown in Fig. S1, large differences were still observed between the number of DEGs of viral strains infecting the same cell type, even for the strains of the same subtype such as four strains of A(H1N1) infecting the A549 cell.

Similarly, we observed large differences between DEGs in different human cell types after influenza virus infections (Fig. 3B). The median number of DEGs ranged from 55 to 6,710 across eight human cell types after influenza virus infections. Overall, when comparing the shared ratios of DEGs between different strains and different cell types, we found that the latter was smaller than the former, suggesting the cell specificity was greater than strain specificity in DEGs, and both of them contributed significantly to the diversity of DEGs in influenza virus infections.

The heterogeneity of P-PPIs in influenza viruses

The P-PPIs were the second largest type of VIHFs. Thus, the heterogeneity of P-PPIs in different types or subtypes of influenza viruses was analyzed. As shown in Fig. 4A, the shared ratios of P-PPIs between influenza virus types or subtypes were generally smaller than 0.1, suggesting different influenza viruses had large differences between P-PPIs. Further analysis of the P-PPIs by proteins of the influenza virus still showed large differences between different viruses (Fig. 4B).

Overview of the H2Flu

A database named H2Flu was created to store and organize different types of VIHFs of influenza viruses. The database is publicly available at http://computationalbiology.cn/H2Flu. It mainly includes pages of Homepage, Browse, Search, Statistic, Download and Contact us(Fig. 5).

An application case of H2Flu

We illustrated the potential usage of H2Flu in integration of multi-omic data in influenza virus infection by comparing the DEGs and DEPs during the infection of Calu-3 cells at 18 hpi by a viral strain of H5N1 virus (A/Vietnam/1203/2004). As shown in Fig. 6A, there were only two genes that were both up-regulated at the transcription and proteome level, suggesting the large difference of upregulated genes in host response to viral infection at different levels. However, more than one-third of downregulated DEPs were also found to be downregulated at the transcriptional level, suggesting the consistent role of these genes in the viral infection. Functional analysis of these genes showed that they were enriched in biological processes related to protein folding, localization, regulation of protein stability, and in KEGG pathways related to metabolisms of carbon and amino acid, actin cytoskeleton, focal adhesion, and so on.