Multi-omics characterization of RNA binding proteins reveals disease comorbidities and potential drugs in COVID-19

The COVID-19 has led to a devastating global health crisis, which emphasizes the urgent need to deepen our understanding of the molecular mechanism and identifying potential antiviral drugs. Here, we comprehensively analyzed the transcriptomic and proteomic profiles of 178 COVID-19 patients, ranging from asymptomatic to critically ill. Our analyses found that the RNA binding proteins (RBPs) were likely to be perturbed in infection. Interactome analysis revealed that RBPs interact with virus proteins and the viral interacting RBPs were likely to locate in central regions of human protein-protein interaction network. Functional enrichment analysis revealed that the viral interacting RBPs were likely to be enriched in RNA transport, apoptosis and viral genome replication-related pathways. Based on network proximity analyses of 299 human complex-disease genes and COVID-19-related RBPs in the human interactome, we revealed the significant associations between complex diseases and COVID-19. Network analysis also implicated potential antiviral drugs for treatment of COVID-19. In summary, our integrative characterization of COVID-19 patients may thus help providing evidence regarding pathophysiology and potential therapeutic strategies for COVID-19.


Introduction
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has swept the world and attracted high international attention with millions of infection cases, which has become the largest pandemic called coronavirus disease 2019 (COVID-19) [1]. Some COVID-19 patients may have severe complications, including acute respiratory distress syndrome or even death [2]. The persistence and recovery of COVID-19 are largely affected by the health status and age of individual patients. In the same individual, when two diseases affect each other due to their common pathogenesis, they are called comorbidity. More and more epidemiological studies have indicated that COVID-19 interacts with other diseases. Of note, 25.2% of patients reported having at least one comorbidity, and the most common comorbidities of COVID-19 are hypertension, diabetes, coronary heart disease and malignancy [3]. In recent study, about 16.9% of patients with severe COVID-19 were reported to have hypertension, and patients with comorbidities are more susceptible to COVID-19 as compared to a healthy individual [4]. Clinical studies have previously shown that hypertension is also a risk factor for high mortality in SARS and MERS [5]. However, the molecular mechanisms of COVID-19 and the comorbidities are still largely unknown.
The majority of studies on COVID-19 have focused on the relationship between disease and clinical characteristics of patients [6]. To describe the response of host cells to SARS-CoV-2 infection, transcriptome and proteome data were obtained from patients infected with SARS-CoV-2, and the differences between different infected stages of SARS-CoV-2 were compared to predict potential biomarkers during SARS-CoV-2 infected stages [7][8][9]. Furthermore, to clarify the cascade reaction that occurs after the SARS-CoV-2 infection, the interaction network between the host protein and the interaction network between the host protein and the viral protein were constructed to identify the hub genes that play an important role in the infection process [10]. In addition, transcriptome analysis was performed for drug repositioning and providing potential drugs to treat COVID-19 [11]. However, multi-omics integrative analysis in COVID-19 is still needed to understand the potential molecular mechanisms and identify novel treatments.
RNA-binding proteins (RBPs) are proteins that bind to RNA through one or more globular RNA-binding domains (RBDs) and change the life course or functions of the binding RNA [12]. In the past few years, several studies have been performed to investigate the interaction between protein and RNA in infected cells, and found that many RBPs containing mature RBD are crucial to the infection of specific viruses [13]. As intracellular pathogens with small genomes, viruses rely on resources of host cell to complete its life cycle, and have evolved a complex mechanism to hijack the important composition of the host cell. RBP is even involved in almost every stage of virus RNA replication. Moreover, the RBP is essential to its anti-virus work because of its ability to identify abnormal products of virus replication [13]. Viruses typically target RBPs in host cells to disable them, or hijack RBPs involved in the processing of transcriptional RNA in order to invade the host [13]. It is still unclear to what extent the expressions or regulations of RBPs are perturbed in SARS-CoV-2 infection.
To address these issues, we comprehensively analyzed the transcriptomic and proteomic profiles of 178 COVID-19 patients, ranging from asymptomatic to critically ill (Fig. S1). We found that RBPs were likely to be perturbed in COVID-19 and interacted with viral proteins frequently. Moreover, we revealed the interaction patterns of RBPs, and network analysis helps our understanding the disease-disease associations and potential antiviral therapy.

Transcriptome and proteome of COVID-19
We performed multi-omics analysis of patients infected with SARS-CoV-2, including transcriptome sequencing data and liquid chromatography-mass spectrometry (LC-MS) to capture proteomic data of patients' serum. All these data and clinical information of patients were obtained from one of our recent studies [14]. All COVID-19 patients (n = 1432) enrolled in this study were screened to exclude those with selected complications, surgical history, or aged younger than 19 or older than 70. The complications excluded in this study included hypertension, coronary heart disease, diabetes, chronic obstructive pulmonary disease, malignancy, chronic kidney disease, cerebrovascular disease, immunodeficiency disease, chronic hepatitis, and tuberculosis. After filtering these patients, 231 patients were enrolled and grouped as asymptomatic (n = 64), mild (n = 90), severe (n = 55), or critical (n = 22) and the male to female ratio was 1.12:1. The mean age of patients was 46.7 years old (standard deviation (SD) = 13.5). The multi-omics profiling was conducted and included transcriptome sequencing (178 samples) and proteomic sequencing (161 samples). In total, transcriptome data from 178 patients with symptoms ranging from asymptomatic to critically ill were included in our analysis, including 64 asymptomatic patients, 64 mild patients, 34 severe patients and 16 critical patients, respectively. Proteomic data included 161 patients, 53 asymptomatic, 54 mild, 33 severe, and 21 critical patients.
We constructed sequencing library and generated transcriptome RNA data [14]. The RNase H method was used to remove rRNA, and globin RNA was moved through a QAIseq FastSelect RNA Removal Kit. After combining purified fragmented cDNA with End Repair Mix and A-Tailing Mix, they were mixed well by pipetting and incubated. Next, PCR amplification, library quality control and pooling cyclization was performed. The RNA library was sequenced using the MGI2000 PE100 platform with 100-bp paired-end reads. Transcriptome of COVID-19 patients were performed by RNA-Seq. To remove the reads with low quality, base ratios >20% and unknown base ('N' base) ratios> 5%, we used SOAPnuke to filter the raw reads [15]. HISAT2 was used to map the filtered reads to the reference genome [16]. Next, Bowtie2 (v2.2.5) was used to align the filtered reads to the transcriptome. Gene expression level was measured by Fragments Per Kilobase of exon model per Million mapped fragments (FPKM) based on the RSEM tool [17]. Protein-coding genes with FPKM >0.1 in at least one sample were reserved for subsequent analysis. In addition, genes with low expression levels (with expression level = 0 in more than 50% patients) were exclude. In total, 14,037 genes including 354 RBPs were analyzed in this study.
Proteomic data was independently collected using Q Exactive HF mass spectrometer (Thermo Scientific, San Jose, USA) and UltiMate 3000 UHPLC liquid chromatography (Thermo Scientific, San Jose, USA). The raw data was analyzed using Spectronaut software (12.0.20491.14.21367) to achieve deeper proteome quantification. The FDR cutoff value for protein level was set to 1% [14]. Proteins with low expression levels (with expression level = 0 in more than 50% patients) were exclude. In total, 634 proteins including 5 RBPs were included for further analysis.

Data processing
The pre-processing processes of the data included filtering, normalization and data transformation. Data filtering concentrates on the removal of genes with low expression levels. In transcriptome data, protein-coding genes with FPKM >0.1 in at least one sample were reserved for subsequent analysis. In addition, genes or proteins with low expression levels (with expression level = 0 in more than 50% patients) were exclude in transcriptome and proteome data. After filtering, 14,037 genes including 354 RBPs were analyzed in this study, and 634 proteins including 5 RBPs were included for further analysis. Data normalization and data transformation were used to adjust data distributions, thus variations in data could be more easily displayed. Log2 transformation was performed to the expression profiles of transcriptome and proteome.

Collection of RBPs
We collected 356 RBP genes from one previous study [18], which provided the function, localization, domain structure and experiments of RBPs (Table S1). Briefly, these RBPs were classified into different subclasses, including transcriptome-wide RNA-binding sites of RBPs, RBP-responsive genes and alternative splicing events, in vitro RBP binding motifs, subcellular localization of RBPs and association of RBPs with chromatin. Functionally, these RBPs most often contribute to the regulation of RNA splicing processes (98 RBPs, 23%), of which 162 RBPs (46%) have more than one function, and 57% of the RBPs surveyed contain well-defined RNA-binding domains.

Identification of differentially expressed RBPs in COVID-19
Differential expression analysis was performed for patients with different infection stages. Several methods employed in differential expression analysis to identify important signatures in in disease progression [19,20]. First, the transcriptome profiles were log2 transformation. Therefore, the Shapiro-Wilk normality test was conducted for the expression profiles to assess the data normality. The results demonstrated that the expression profiles of the transcriptome and proteome did not adhere to the normal distribution (p value < 0.05). Consequently, Wilcoxon's rank sum test was used to analyze the differences between the two stages of COVID-19 patients. In addition, the FC was used to evaluate the expression levels of genes in diseases, which are the value of dividing the average gene expression in diseased tissues by the average gene expression in normal tissues. When FC was greater than 1, it means up regulation of gene in disease tissues, while less than 1 means down regulation. Genes with adjusted p-value ≤0.05 was considered as differentially expressed in COVID-19.

RBP-gene interaction network
We obtained the protein-protein interactions among human proteins or between SARS-CoV-2 proteins and human proteins from one of our recent studies [21]. There were 2,724,723 interactions among 19,779 human proteins, and 389 interactions between SARS-CoV-2 proteins and human proteins, which involved in 28 SARS-CoV-2 proteins and 384 human proteins (Table S2). In addition, additional human proteins interacting with SARS-CoV-2 proteins were collected from a recent study (Table S2) [22].
In addition, we also downloaded human PPI data from HumanNet V3 [23], which contained the interactions among approximate 99.8% of human protein-coding genes. In our analysis, we used three models, Based on the interaction information between viral proteins and human proteins and differential RBPs in COVID-19, we obtain the important subsets of RBP that are differentially expressed and interact with viral proteins during SARS-CoV-2 infection. Combining the interactions among human proteins, we constructed the interaction network among viral protein, RBPs and human proteins (Table S3).

Functional enrichment analysis of RBPs
To identify the functions of virus infection-related RBPs, we used the R package 'clusterProfiler' for functional enrichment analysis [24]. In this analysis, we only considered biological processes (BPs) with 10-500 overlapping genes, and considered the Gene Ontology (GO terms) with p values less than 0.05 and adjusted-p values less than 0.1 as thresholds for screening significant functions. Next, all significant GO terms were calculated and clustered using the GO_similarity() function in the 'simplifyEnrichment' R package, and the simplifyGO() function was used to display the GO clusters [25].

Dissection of disease comorbidities of COVID-19
The 299 disease-related proteins were collected from a recent article to evaluate the associations between SARS-CoV-2 infection and other 299 diseases [26] (Table S4). First, using the protein-protein interaction network G, we assessed the shortest distance d bb between proteins within the disease-related datasets b and the shortest distance d vv between proteins within COVID-19-related datasets v.
We next calculated the average shortest path length between diseaserelated proteins b and COVID-19-related RBPs v.
where d(i, j) is the shortest distance between protein i (COVID-19-related RBPs) and j (disease-related proteins) in the human protein interaction networks G. Finally, the index S vb was calculated, where S vb <0 suggests a network-based overlap between the viral-related RBPs v and proteins associated with complex disease b [21,27]. S vb was calculated as follows: S vb compared the shortest distances between proteins within viral interacting proteins d vv or diseases-related proteins d bb , to the shortest distances d vb between virus interacting proteins and disease-associated proteins.

Identification of potential drugs for COVID-19
The human protein interaction networks have provided a reasonable method for predicting potential drugs for treatment of SARS-CoV-2 infections. Thus, we first downloaded the Food and Drug Administration (FDA) approved drug target interaction data from DrugBank (https: //www.drugbank.com/), including 2710 drugs and 2694 targets. In addition, the drug-target interaction data was downloaded from Therapeutic Target Database (TTD, https://db.idrblab.net/ttd/) [28] to further verify the results, including 1650 FDA-approved drugs and 561 targets.
We prioritized the candidate drugs or small molecules based on the network-based method [21,26]. Based on the protein V encoded by the host gene and interacted with SARS-CoV-2, and the drug targets T, the network proximity between them of each candidate drug was calculated as follows: where d(i, j) is the shortest distance between protein i (viral interacting proteins) and j (drug target proteins) in the human protein interaction networks.
Next, the results of network proximity were further transformed into Z-score based on permutation tests: d r and σ r are the average network proximity and standard deviations of 1000 times permutation, respectively. In addition, gene-encoded proteins were randomly selected from the entire human proteome, which had degree distributions similar to SARS-CoV-2-associated protein V and drug-target protein T. The p value was calculated as the proportion of random conditions in which d VT was lower than observed value, and drugs with Z-score < − 1.5 and p value < 0.001 were identified as potential antiviral drugs.

Widespread transcriptome alterations of RBPs in COVID-19
We first interrogated if and to what extent the human transcriptome is altered in COVID-19 at the gene expression level. We identified differentially expressed genes across different stages of SARS-CoV-2 infections. We found that the human transcriptomes were greatly perturbed by virus infections and a large number of genes exhibited differential expressions at the RNA level (Fig. S2A). In addition, we identified 332 RBPs which exhibited dynamic expression patterns during SARS-CoV-2 infections (Fig. 1A). We also collected another 1616 RBPs from literature and identified 1398 RBPs that exhibited abnormal expression in at least one infection stage of COVID-19 (Fig. S3A). For example, LARP7, SRSF2, HNRNPK and PABPC1 exhibited high expression in asymptomatic patients and decreased expression in critical patients (Fig. 1A). In contrast, DHX29, PSPC1 and LARP1 exhibited increased expression during the process of infection (Fig. 1A).
Next, we calculated the proportions of protein-coding genes and RBPs that exhibited differential expressions between different stages of SARS-CoV-2 infections. We found that there were higher proportions of RBPs perturbed when compared to other coding genes in comparisons between "Critical-Severe", "Critical-Mild", "Critical-Asymptomatic" and "Mild-Asymptomatic" (Fig. 1B). In addition, we performed Fisher's exact test to evaluate the significance and found that the Odd Ratios (ORs) of RBPs vs. other genes were significantly higher than 1.0 in these four groups of comparisons (Fig. 1C). These results indicated that RBP genes were more likely to be perturbed during infection, suggesting their important roles in the process of SARS-CoV-2 infection.

Perturbations of RBP translations in COVID-19
Genome-wide proteome data during SARS-CoV-2 infection allow us to systematically investigate the alterations of proteins. We identified 98-273 proteins were up-regulated and 30-116 proteins were downregulated in the comparison of patients in different infection stages (Fig. S2B). In total, we identified 506 proteins exhibited differential expression in at least one stage ( Fig. 2A). In addition, we obtained the similar results using another RBPs (Fig. S3B). These results suggest that the protein translations were perturbed in COVID-19.
In particular, we identified three RBPs that showed expression perturbations in both RNA and protein level, including CORO1A, DHX29 and RPS5. CORO1A exhibited significantly higher expressions of RNA and protein in asymptomatic patients (Fig. 2B-C). It has been demonstrated that the inborn mutations in CORO1A affect the development, differentiation, and function of key factors involved in the immunity against epstein-barr virus (EBV) [29]. Coronin 1A (Coro1A) also plays important roles in host against bacterial infection, which is an important immunity related gene that helps to inhibit NF-kB activation [30]. In addition, DHX29 exhibited lower expression at RNA and protein levels in asymptomatic patients but higher expression in mild and severe patients ( Fig. 2B-C). DHX29 has been demonstrated to function as a new canonical Wnt signaling tumor suppressor [31]. DHX29 was a candidate biomarker for the diagnosis of tuberculosis infection [32] and functions as an RNA co-sensor for antiviral immunity [33]. RPS5 exhibited high expression in severe patients at both RNA and protein levels ( Fig. 2B-C). It has been demonstrated that RPS5 interacted with the Rabbit hemorrhagic disease virus (RHDV) and played a role in virus replication [34]. Together, these results suggest that the transcription and translations of several RBPs were perturbed in SARS-CoV-2 infection.

RBPs frequently interact with viral proteins
The protein-protein interactions (PPIs) between human and viruses play important roles in viral infection and host immune responses. We thus investigated the interactions among human RBPs and viral proteins. We identified 51 RBPs of 356 RBPs exhibited perturbed expressions in SARS-CoV-2 infection and also interacting with SARS-CoV-2 proteins, and using other potential RBPs collected from literatures, we got the similar results (Table S3). In particular, we constructed the human-virus interaction network, which involving 19 human RBPs and 7 SARS-CoV-2 proteins (Fig. 3A). The majority of RBPs interacted with nucleocapsid (N) and nsp8 proteins, such as G3BP1, RBM41 and EXOSC5 (Fig. 3A). The N protein is one of the most crucial structural components of SARS-CoV-2 [35], which packages the viral RNA into helical ribonucleocapsid and interacts with the other structural proteins during virions' assembly [36]. Nsp8 is an enzyme in viruses that contributes to the activation and sustained production of viral RNA. It has been demonstrated that nsp8 may play a regulatory role in the initiation of viral replication [37]. In addition, we found that RBPs can potentially interact with numerous human proteins in the protein-protein network (Fig. 3A).
Next, we investigated the expression patterns of RBPs and corresponding targets during SARS-CoV-2 infection. We found that the majority of target genes exhibited higher expressions in asymptomatic patients and lower expressions in severe and critical patients (Fig. 3B). For instance, the expressions of PTBP1, EIF4A1 and HSP90AB1 decreased gradually during the SARS-CoV-2 infection stages (Fig. 3C-E). It has been shown that SARS-CoV-2 infection perturbed the expression of three genes crucial for neuronal survival in its host cells, including PTBP1, YWHAZ and YWHAE [38]. Emerging evidence suggested that the SARS-CoV-2 virus regulates the host processes involved in protein synthesis, such as the control of translation factors EIF4A [39]. EIF4A1 is part of the cellular EIF4F translation initiation complex, which plays important roles in mRNA binding to the ribosome [39]. Moreover, HSP90AB1 exhibited significantly decreased expression during infection (Fig. 3E), which is consistent with previous study [40].
In addition, we also observed that several RBPs exhibited increased expressions during infections, such as LARP1 and PA2G4 (Fig. S4). It has been demonstrated by genetic perturbation that LARP1 can restrict SARS-CoV-2 replication in infected cells [41]. The inconsistence between RNA and protein levels of LARP1 suggested that we need more experiments to investigate its roles in infection. PA2G4 also exhibited increased expression in severe and critical patients (Fig. S4), which plays important roles in virus infections [42]. Together, these data demonstrate that our RBP regulatory network analysis recovered known

SARS-CoV-2 virus is likely to target PPI network center
The PPIs among human and virus mediate viral infection and host immune response [21,43]. The function importance of proteins can be represented by their locations in human PPI or regulatory networks [44][45][46]. Thus, we next investigated the location of SARS-CoV-2 interacting RBPs that were with expression perturbations during infection in the context of human PPIs. RBPs were firstly divided into two classes: one was those interacting with SARS-CoV-2 proteins and exhibited expression perturbations during infection; the other RBPs were classified into another group. We calculated the topological features of RBP in human PPI, including the degree, closeness and betweenness of proteins. We found that virus-related RBPs have significantly higher degrees in networks than other RBPs in the PCNet (Fig. 4A, p = 6.3e-08). We next compared the closeness between SARS-CoV-2 virus-related RBPs and other RBPs, and found that SARS-CoV-2 virus-related RBPs have higher closeness coefficient (Fig. 4B, p = 2.1e-08). When comparing the betweenness of SARS-CoV-2 virus-related RBPs with other RBPs, we found that the betweenness of virus-related RBPs are significantly higher than other RBPs (Fig. 4C, p = 7.6e-05). In addition, we performed the same analyses using three human protein interaction networks from HumanNet [47] and we observed the same results (Fig. S5). All these results suggested that SARS-CoV-2 virus was likely to interact with RBPs that are located in the center of the network or with those that play important roles in information spread.
To determine the functions of these virus-related RBPs in SARS-CoV-2 infection, we performed function enrichment analysis based on biological process (BP). The results showed that SARS-CoV-2 virus-related RBPs were significantly enriched in the biological functions of regulating chromosome and RNA transport, apoptosis, stress signal pathway and viral genome replication (Fig. S6). In previous studies, the related process of virus infection in human body may be related to RNA replication and transport [48], and inducing inflammatory response and apoptosis of infected cells is an effective method to inhibit SARS-CoV-2 infection [49].

Disease comorbidities of COVID-19
Emerging evidence has revealed complications and comorbidities of COVID-19, such as neuroinflammatory reactions and kidney damage [50][51][52]. We thus systematically evaluated the disease associations with COVID-19 based on the state-of-art network proximity measure S vb (Fig. 5A and Table S4, see details in methods). Based on the gene sets of 299 human complex diseases and SARS-CoV-2 related RBPs, we compared the shortest distance d vv between virus-related RBPs and the shortest distance d bb between disease-related proteins with the shortest distance d vb between virus-related RBPs and disease-related gene sets (Fig. 5A). Network analysis revealed that SARS-CoV-2 infection was significantly correlated with metabolic diseases, kidney diseases and congenital diseases (Fig. 5B).
Recent reports suggested that hospitalized SARS-CoV-2 patients with acute kidney injury (AKI) often have poor survival rates. In previous studies, SARS-CoV-2 nucleocapsid protein staining was performed on 62 kidney samples from COVID-19 patients. SARS-CoV-2 nucleocapsid protein was detected positive in all 62 samples, mainly detected in proximal renal tubular cells [53,54]. These observations suggested that SARS-CoV-2 may be able to directly infect the kidney. In our study, we found that five SARS-CoV-2 proteins (M, N, nsp8, nsp12 and nsp13), interact with the kidney-related gene sets (Fig. 5C), which can antagonize the immune response of the host and carry out viral replication and transcription [37]. In addition, we found that SARS-CoV-2 directly interact with disease-related proteins, NONO and SFPQ (Fig. 5C). SFPQ has been shown to promote viral RNA amplification by interacting with SARS-CoV-2 genome [55].
On the other hand, SARS-CoV-2 infection may cause esophagealrelated diseases by interacting with RBPs (Fig. 5D). Emerging evidences have showed that SARS-CoV-2 infection often causes esophageal ulcers [56,57]. We found that SARS-CoV-2 directly interact with several proteins that play important roles in signaling pathways, such as TP53, MYCN, SOX2 and RUNX1 (Fig. 5D). These results suggest that SARS-CoV-2 may affect esophageal and even intestinal motility. In addition, we found that SARS-CoV-2 infection was correlated with the urologic neoplasms by interacting with NONO, CCND1 and SFPQ (Fig. S7A). In the analysis of leukemia-related diseases, we found that leukemia mainly affected RPLP1, ATM, PRKD2 and MYC, which were closely related to cell cycle, DNA damage and cell proliferation (Fig. S7B). All these results suggested that the comprehensive RBP interactome map helps to understand the association between COVID-19 and other human complex diseases.

Potential drugs for treatment of COVID-19
Understanding the complex interactions between virus-interacting proteins and proteins-associated with human diseases suggest that we  (Table S5). All the drugs were ranked based on Z dVT and top 31 drugs were visualized as a network, which showing the interactions among drug targets and SARS-CoV-2-related RBPs (Fig. 6A). We found that the majority of these candidate drugs target TOP2A and TOP1, which interacting with several SARS-CoV-2-related RBPs (Fig. 6A).
Although Doxorubicin, the preferred drug, has not been shown in literature to be effective in treating SARS-CoV-2, the molecular docking and kinetic simulation experiments with SARS-CoV-2 spike protein (S) and main protein (M) found that they have strong binding energy. These results indicated that Doxorubicin may inhibit SARS-CoV-2 S and M proteins and prevent SARS-CoV-2 from entering host cells [58,59]. We found that Doxorubicin mainly targets DNA topoisomerase ii (TOP2A) and nucleolar protein (NOLC1), and TOP2A interacts with most SARS-CoV-2-related RBPs (33/51, 65%) (Fig. 6B). It also has been demonstrated that TOP2A and NOLC1 may play an important role in virus proliferation and clonal invasion [60,61].
The Topotecan, a water-soluble semi-synthetic derivative of camptothecin, has shown antitumor activity in a variety of cell culture and xenotransplantation systems and inhibiting rapid cell division [62]. In the experiment of treating ACE2 cells infected with SARS-CoV-2 with Topotecan, Topotecan can inhibit the expression of IL-6, CXCL2 and other genes. However, Topotecan does not inhibit virus replication, but achieves the purpose of anti-infection by inhibiting the expression of host genes [63]. We found that Topotecan mainly targets DNA topoisomerase i (TOP1), may affect the expression of TOP1 and then inhibits other host genes to prevent viral infection (Fig. 6C).
In addition, we performed the same analysis using drug-target interaction data from Therapeutic Target Database (TTD) [28]. In total, 1650 drugs and 561 targets were included in this analysis. We identified 166 SARS-CoV-2 drug candidates ranked by Z dVT , and top 50 drugs were visualized (Fig. S8A and Table S5). Based on drug-target The dots represent diseases whose radius reflects the number of associated diseases genes. The diseases closest to the center, whose names are marked, are expected to have higher comorbidity with viral infection. C-D, Network visualization showing the protein-protein interactions among SARS-CoV-2 viral-interacting proteins and diseases-associated proteins. C for kidney neoplasms and D for esophageal diseases. Red represents SARS-CoV-2 proteins, and green represents proteins associated with SARS-CoV-2. Blue indicates diseasesassociated proteins, and purple means the intersection of virus-related proteins and disease-related proteins. Line indicates the connection between them.
interaction data of TTD, we observed similar results that TOP1 and TOP2 (TOP2A and TOP2B) were targeted by more drugs, and interacted with more RBPs. Moreover, Doxorubicin was also identified as candidates, which was FDA-approved drugs and considered as an anticancer agent, and targeted TOP2A, TOP2B and TERT, which were classified as successful targets with at least one approved drug (Figs. S8A-B).
In addition, Mycophenolate mofetil, widely and safely used as an immunosuppressive agent to treat autoimmune diseases, have been proved to block SARS-CoV-2 infection [64,65]. In our analysis, inosine-5 ′ -monophosphate dehydrogenase 2 (IMPDH2) interacting with SARS-CoV-2 nonstructural protein 14 (nsp14) was targeted by Mycophenolate mofetil, and interacted with many RBPs related to viral protein (Fig. S8C). A recent study predicted that Mycophenolate mofetil modulates the interaction between IMPDH2 and viral protein nsp14 [66]. It is crucial to target these host proteins that interact with viral proteins, and it will provide more targeted drug design to inhibit SARS-CoV-2 infection.

Discussion
It has been proved that RBPs are crucial to the infection of specific viruses, and even involved in almost every stage of virus RNA replication. However, it is unclear to what extent the expression or regulation of RBPs is interfered in COVID-19 infection. In this study, we systematically analyzed the transcriptome and proteome perturbations of RBP in COVID-19. We found that 93.26% of RBPs exhibited expression perturbations in at least one stage of COVID-19, which were significantly higher than that of other genes. These results suggested that RBPs were likely to be perturbed in COVID-19. Several important RBPs were identified, such as CORO1A, DHX29 and RPS5, which have been demonstrated to play important roles in virus infections [67,68].
Based on interaction between SARS-CoV-2 proteins and human proteins, 51 virus-related RBPs were extracted, and 19 of them interacted with known SARS-CoV-2 proteins. Combined with human PPI network, the interaction network among SARS-CoV-2 protein, RBPs and human proteins was constructed. Most of the SARS-CoV-2-related RBPs interacted with SARS-CoV-2 protein N and nsp8. The topological properties of RBPs were calculated, and the virus-related RBPs group was compared with other RBPs group. The results showed that virus-related RBPs were closer to the network center and played more important functions. These results were consistent with one of our recent studies [21], suggesting that SARS-CoV-2 was likely to perturb the network center thus to disturb the biological pathway in a more efficient way.
Emerging evidence indicated the connection between COVID-19 and a variety of diseases, leading to the aggravation of these diseases. However, the mechanism between them is still unclear. In the analysis of the correlation between COVID-19 and other complex diseases, we found that COVID-19 was mainly significantly associated with metabolic diseases, congenital diseases and kidney diseases. NONO and SFPQ genes are both kidney disease-related genes and SARS-CoV-2-related RBPs, which may be the key genes causing both diseases. In addition, genes associated with kidney have indirect effect mainly by M, N, nsp8, nsp12 and nsp13. The underlying mechanisms of kidney damage and renal failure caused by SARS-CoV-2 infection is unclear [51], while our results suggested that SARS-CoV-2 infecting kidneys may invade the kidney cells through the above viral proteins and key genes and then cause kidney complications.
Based on the network analysis, we also identified 157 potential drug candidates for the treatment of COVID-19, of which 15 (50%) of the top 30 preferred drugs targeted DNA topoisomerase ii alpha (TOP2A). In previous experiments, the molecular docking and kinetic simulation results between the preferred drug Doxorubicin and SARS-CoV-2 protein showed strong binding energy, and Doxorubicin may be a candidate drug for SARS-CoV-2 protein [58]. The prioritized drug Topotecan can reduce the pathological features of lung injury in infected animals and reverse the gene expression response induced by COVID-19. Therefore, Topotecan may be a candidate drug for treatment of COVID-19.
In summary, our integrative characterization revealed that RBPs were likely to be perturbed in COVID-19. Network analysis revealed the regulatory patterns of RBPs and their potential functions. Our results may thus help providing evidence regarding pathophysiology and potential therapeutic strategies for COVID-19.