Drugs produce pharmaceutical and adverse effects that arise from the complex relationship between drug targets and signatures; by considering such relationships, we can begin to understand the cellular mechanisms of drugs. In this study, we selected 463 genes from the DSigDB database corresponding to targets and signatures for 382 FDA-approved drugs with both protein binding information for a drug-target score (KDTN, i.e., the degree to which the protein encoded by the gene binds to a number of drugs) and microarray signature information for a drug-sensitive score (KDSN, i.e., the degree to which gene expression is stimulated by the drug). Accordingly, we constructed two drug–gene bipartite network models, a drug-target network and drug-signature network, which were merged into a multidimensional model. Analysis revealed that the KDTN and KDSN were in mutually exclusive and reciprocal relationships in terms of their biological network structure and gene function. A symmetric balance between the KDTN and KDSN of genes facilitates the possibility of therapeutic drug effects in living organisms. These results provide new insights into the relationship between drugs and genes, specifically drug targets and drug signatures.
Research Article
Relationship Between Drug Targets and Drug-Signature Networks: A Network-Based Genome-Wide Landscape
https://doi.org/10.21203/rs.3.rs-967821/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this older preprint version
cellular mechanism
FDA
drug-sensitive score
KDTN and KDSN
drug signatures
Drugs produce pharmaceutical and adverse effects according to the complex relationship between drug targets and drug signatures (1). As gene analysis has become more prevalent, studies on the association between genes and drugs have also become more widespread in the field of drug research. As examples, Nagaraj et al. used a computational drug-repositioning approach to rapidly identify potent drug candidates for epithelial ovarian cancer treatment (2), Kim et al. assessed reversal gene expression profiles for gastric cancer using computational drug repositioning (3), and Grenie and Hu investigated drugs for inflammatory bowel disease using genetic information and computational methods (4). In such studies, genes are divided into two categories: drug-target genes and drug-signature genes. Drug-target genes (also known as “druggable genes”) code for proteins that physically bind with the drug compound (5, 6), whereas drug-signature genes (also known as “drug-sensitive” genes) are differentially expressed due to drug–protein binding following drug treatment (1). Studying the interactions between drug targets and drug signatures is important for drug discovery, drug repositioning, and identifying inference from potential adverse drug reactions (7, 8).
Ideally, the study of drug-related genes in both categories should extend to all human genes without any limitations. However, conventional studies have been limited to specific genes and specific drugs related to only a few diseases, and they have focused on only drug targets or drug signatures. Because such studies have investigated specific genes and drugs, the characteristics of all human genes in living cells are not typically taken into account. Additionally, for these reasons, conventional drug development studies may show effects on the phenotype of interest, and the presence of treated subjects is typically perceived as a cause of bias in genome-wide association studies (9). Therefore, in the present study, we focused on the relationship between drug targets and signatures based on their characteristics. Our aim was to identify the genetic landscape via a multidimensional network using genome-wide drug–gene binding data and gene expression data.
In general, drugs affect the activity of proteins that correspond to target genes. During drug treatment, biological networks are disturbed and the expression of many other genes is significantly changed by
unexpected responses to the drug. Thus, the pharmaceutical and adverse effects of the drug occur through complex relationships among drug targets and drug signatures (1). Changes in gene expression by drug treatment can imply a therapeutic effect at the cellular level. From another perspective, drug treatment can have a perturbative effect in cells via gene networks (10). Indeed, changes in gene expression in cells to maintain homeostasis arise due to perturbation (11).
To elucidate genome-wide inter-relationships between drug-target genes and signatures, we selected genes corresponding to targets and signatures for drugs that have both protein binding information for drug-target score (KDTN) and microarray signature information for drug-sensitive score (KDSN). KDTN represents the degree to which the protein corresponding to the gene binds to a large number of drugs, whereas KDSN represents the degree of the gene expression response following stimulation by drugs. Overall, we explored the network-based genome-wide landscape by comparing the cellular and functional characteristics of drug targets and drug signatures using the two variables KDTN and KDSN.
Data
The drug and genome database DSigDB (http://dsigdb.tanlab.org/DSigDBv1.0/) (12) is an open-source database that currently includes 22,527 gene sets and consists of 17,389 unique compounds covering 19,531 genes. Gene sets provide seamless integration by which to link gene expression with drugs/compounds. DSigDB organizes drugs and small molecule-related gene sets into four domains based on data for drug-induced quantitative inhibition and/or changes in gene expression. The data from DSigDB contains four domains (D1–D4) that collect drug and genome data for four purposes, as shown in Table 1. The D1 and D3 domain datasets were used to construct each drug-target network (DTN) and drug-sensitive network (DSN) in this study.
Construction of the network
FDA-approved drugs for which both protein binding data and microarray experiment data were available were used from the DSigDB database. Following these principles, the D1 and D3 datasets were used and matched to their drug and gene ID terms.
In the process of matching terms from each domain, the Pubchem compound term (https://pubchem.ncbi.nlm.nih.gov/) (16) and Entrez term (https://www.ncbi.nlm.nih.gov/Web/Search/entrezfs.html) (17) were used as the drug and gene ID. Consequently, 382 drug compounds and 9,490 genes intersecting with D1 and D3 were extracted. Using these data, the DTN and DSN were constructed from calculated gene scores, i.e., KDTN and KDSN, based on the relationship between drugs and genes as follows:
where the range of KDTN and KDSN was from 1 to 186 and from 1 to 104, respectively, and the number of genes with KDTN ≥ 1 and KDSN ≥ 1 was 720 and 9,233, respectively. Among these genes, 463 simultaneously had both KDTN ≥ 1 and KDSN ≥ 1. From the relationship between the KDTN and KDSN of genes, the genes were divided into three groups as shown in Figure 1A: the DTG set (drug-target genes: 257 genes), DSG set (drug-sensitive genes: 8,770 genes), and the DTSG set (drug-target and -sensitive genes: 463 genes).
Figure 1B shows the process by which a drug–gene network was constructed. Two kinds of bipartite network (DTN and DSN) were constructed, which were then merged into a multidimensional network. For example, Gene A binds to three drugs among nine drugs in the DTN (drug 2, 3, and 4) but its expression was changed by only one drug in the DSN (drug 7). Using such analyses, the target score and drug-sensitive score of each gene from the binary network model were calculated.
Analysis
Gene enrichment analysis and network analysis were conducted in three groups of genes (DTG, DTSG, and DSG) using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses via DAVID (https://david.ncifcrf.gov/) (18). The top 10 GO terms and top 8 KEGG pathway terms for each gene group, i.e., those terms that were most enriched, were determined and are shown in Figure 5. The distributions among KDTN of D1, KDSN of D3, and KDTN and KDSN of DTSG were compared using the ggplot package in R. We then constructed a drug-target bipartite network for the drugs and genes involved in the DTSG set using Cytoscape. For network visualization, 19 drugs and 170 genes were used; these 19 drugs represented about 5% of the 371 drugs included in the DTSG set. Furthermore, network analysis was conducted using the concept of m-core decomposition to analyze the central function according to KDTN and KDSN. A “peeling algorithm” aims to characterize a network hub and elucidate the relationships between nodes based on network connectivity (19). In a multidimensional network, the nodes represent the drugs and genes of the DTG and DSG sets, respectively, and the edges represent the relationships between drug-target genes or drug signatures. In the present study, we applied a peeling algorithm represented by m-core (19). Specifically, m-coreDSN and m-coreDTN are defined as the maximal connected subgraph of the DSN and DTN, respectively, in which all genes have a degree of KDSN and KDTN greater than the m value.
Gene set analysis
Distribution of KDTN and KDSN in the DTSG set. First, the distributions of KDT and KDS for the DTSG set were analyzed. Each distribution and the three-dimensional distribution for DTSG set were identified to elucidate the relationships between KDT and KDS. As shown in Figure 2, KDT and KDS had scale-free and power-law distributions.
Construction of the drug–gene network. The distributions of KDT and KDS for the DTSG set were compared with the D1 and D3 sets to determine whether the DTSG set had representativeness for the D1 and D3 set and to show the tendency of the DTSG set before a drug–gene network was constructed. Results showed that the distributions of KDT and KDS for the DTSG set presented representativeness for D1 and D3 (Fig. 3). Once representativeness was confirmed, an integrative drug–gene interaction network for the DTSG set and the drugs was visualized. However, visualizing all the target genes and sensitive genes made it difficult to intuitively observe the characteristics of the network. Therefore, only the target genes and sensitive genes for about 5% of the 371 drugs included in the DTSG set were analyzed in alphabetical order.
To construct a drug-target subnetwork, 16 drugs binding to 84 genes were extracted. To construct a drug-signature subnetwork, 13 drugs that affect the expression level of 103 genes were extracted. The drug-target subnetwork and the drug-signature subnetwork were merged as a drug–gene subnetwork containing 189 nodes (19 drug and 170 genes) and 266 edges. As shown in Figure 4 (in which the black-colored nodes represent drugs, the orange-colored nodes represent sensitive genes, and the green-colored nodes represent target genes), PTGER2 (purple-colored node) was the only gene included as both a target gene and a sensitive gene for 19 drugs in the subnetwork. Except for PTGER2, the relationships between the target genes and sensitive genes were exclusive and independent (Fig. 4).
Enrichment analysis
GO/KEGG pathway analysis for the DSG, DTSG, and DTG sets. Through GO analysis, the cellular components, biological processes, and molecular functions associated with each gene set were investigated. In the DTG set, 257 genes were associated with 13 cellular components, 28 biological processes, 20 molecular functions, and 11 KEGG pathways (FDR-adjusted p-value <0.05). In the DSG set, 8,770 genes were associated with 41 cellular components, 63 biological processes, 21 molecular functions, and 8 KEGG pathways (FDR-adjusted p-value <0.05). In the DTSG set, 463 genes were associated with 24 cellular components, 95 biological processes, 35 molecular functions, and 28 KEGG pathways (FDR-adjusted p-value <0.05).
GO analysis revealed that most proteins synthesized by drug-sensitive genes were located in inner cellular zones such as the nuclear chromosome, nuclear pore, and nucleosome rather than in outer cellar zones such as the cell wall. The proteins synthesized by drug-sensitive genes were shown to be involved in gene transcription, gene expression regulation, and DNA replication, and to function in DNA, RNA, and protein binding. In contrast, GO analysis showed that most proteins synthesized by drug-target genes were located in outer cellular zones and played roles, for example, in receptor complexes, voltage-gated channel complexes, synapses, and cell junctions. Most proteins synthesized by drug-target genes were found to be involved in the catabolic process of cGMP and cAMP and in transmission and transport processes; they played roles in ion channels and enzyme activity.
Through KEGG pathway analysis (Fig. 5), it was revealed that drug-sensitive genes were involved in central dogma-related pathways such as the spliceosome, transcriptional regulation, and protein-processing progress. However, drug-target genes were involved in neural signaling pathways such as addiction to nitrogen, nicotine, and morphine, serotonergic synapses, and retrograde endocannabinoid signaling.
As shown in Figure 5, only 6 terms (GO and KEGG) from 114 terms associated with the three gene sets (DTG, DSG, and DTSG) overlapped. Two cellular component terms overlapped in the DTG and DTSG sets: postsynaptic membrane and voltage-gated calcium channel complex. Of the molecular function terms, 3′,5′-cyclic-nucleotide phosphodiesterase activity and 3′,5′-cyclic-AMP phosphodiesterase activity overlapped in the DTG and DTSG sets. From the biological processes terms, only one term overlapped between the DTG and DTSG sets: cAMP catabolic process. Similarly, one KEGG pathway term, morphine addiction pathway, overlapped in the DTSG and DTG sets. These results suggest that drug-target genes and drug-sensitive genes are exclusive and independent in terms of their cellular locations, genetic functions, processes, and pathways.
Transcription factor (TF) analysis
In gene set analysis, it is important not only to characterize the gene set but also to identify the number and type of TFs as this can help to improve understanding of gene regulatory networks. Thus, we examined whether there were differences in the number of TFs involved in each gene set. We used X2Kweb (20) as a TF analysis tool to examine the binding frequency and types of TFs for each gene set. Results showed that TFs bound on DNA strands on average six times per gene in the DSG set, which was three-fold greater than the TF binding in the DTG and DTSG sets (both two times per gene on average, Fig.6a). In total, 737 TFs were associated with the three gene sets. Of these, 30 TFs overlapped between two or more gene sets as shown in Figure 6b. Therefore, the TFs involved in each gene set differed. Of the 30 overlapping TFs, 9 were derived from essential genes in humans (<10% of all human genes are considered essential) (21).
Core gene analysis of DTN and DSN
Characterization of the core genes in KDT and KDS for the DTG and DSG sets was examined by applying a peeling algorithm. Each network that included >50 genes was analyzed according to m-core. Thus, m-coreDSN had 1 to 36 core gene groups whereas m-coreDTN had 1–17 core gene groups. Figure 7 indicates the gene ontological characterization in each network according to m-core. In cellular component analysis, the core genes of each DSN and DTN showed exclusive distributions. Proteins synthesized by core genes of the DSN were located in the cytosol, cytoplasm, nuclear chromosome, and nucleosome. Conversely, proteins synthesized by the core genes of the DTN were located in the synapses, dendrites, plasma membrane, and axon terminus. In molecular function analysis, the core genes of each DTN and DSN were also exclusively distributed. Proteins synthesized by the core genes of the DSN functioned during cell–cell adhesion and in protein heterodimerization activity by binding proteins and cadherin. The proteins synthesized by the core genes of the DTN functioned in ion binding, hormone binding, chemical receptor activity, and enzyme activity functions (Supplementary Fig. 1A). In biological processes analysis, the core genes of the DSN and DTN networks also showed exclusive distributions. Proteins synthesized by the core genes of the DSN were involved in the PERK-mediated unfolded protein response, response to hypoxia, positive regulation of angiogenesis, and regulation of cell death. In contrast, proteins synthesized by the core genes of the DTN were involved in the response to drugs, dopamine transport, receptor signaling pathways, and monoterpenoid metabolic processes (Supplementary Fig. 1B).
In summary, characterizations of genes in the DSG, DTG, and DTSG sets in terms of GO and KEGG pathways could clearly be distinguished. In addition, the numbers and types of TFs differed among the DSG and DTG sets with different binding frequencies of the TFs on DNA strands. Finally, m-core analysis of the core genes in each DSN and DTN exhibited reciprocal balanced characteristics.
Here, we investigated the relationships between two major gene groups, i.e., target and signature genes, according to various perspectives related to drug discovery and development. Our purpose was not only to investigate the relationship between DTG and DSG sets but also to demonstrate the genome-wide landscape of drug–gene multidimensional networks based on relationships between DTNs and DSNs. We classified genes into three sets based on their operational roles such as their physical binding and/or response to drugs. First, we constructed DTNs and DSNs from experimental data such as drug-target and drug-signature data. Subnetworks were constructed using experimental data from drug targets and drug signatures. In a multidimensional network model constructed by merging each subnetwork, the KDTN and KDSN of the DTG, DSG, and DTSG sets were calculated; thus, the mean network interaction degree of each subnetwork was calculated. The three gene sets, namely DTG, DSG, and DTSG, were compared in four ways: 1) the distribution between KDTN and KDSN, 2) GO and KEGG pathway analysis, 3) the number and type of TFs, and 4) GO analysis with m-coreDSN and m-coreDTN according to KDSN and KDTN.
Sorting gene processes using D1 and D3 showed that drug targets are not usually affected by the drug as reported in a previous conventional study (22). Therefore, only 463 genes intersected between 720 DTG and 9,233 DSG sets. The degree distributions of KDTN, KDSN, and subnetwork visualization showed that the relationship between the DTG and DSG sets was exclusive. Each degree of KDTN and KDSN showed a power-law distribution and their relationship was reciprocal. This is indirect evidence that the response of cells to drugs is structured and organized systematically (23, 24).
The relationships of GO terms (cellular component, molecular functions, and biological process) and KEGG pathways between the DTG, DSG, and DTSG sets was also reciprocal. This shows that the DTG and DSG sets have distinct functional differences in cells. Thus, studies to investigate target genes, not signature genes, should be conducted according to aspects of GO and KEGG pathways, as shown in the current study.
TF analysis of each gene set showed that the average binding frequency of TFs involved in the DSG set was six times per gene. The TFs involved in each gene set were also exclusive and different. Thus, genes acting as drug signatures seem to be regulated with binding frequencies three-fold greater than those of genes acting as drug targets. Consequently, genes regulated with 1–3 TFs would be good candidates for drug-target genes.
GO analysis using the m-core of each network showed that the functional and spatial characteristics of the target gene core and the signature gene core differ. Mutually exclusive characteristics were also exhibited.
In conclusion, the expression of target genes was barely affected by drug treatments. Therefore, the pharmaceutical effect of drugs was due to the DSGs for which expression levels were significantly changed by drug treatment rather than the direct action of DTGs. These complex drug–gene relationships can produce drug side effects as well as therapeutic effects. This study provides a potential new approach to discovering drugs. However, further studies are needed to identify the therapeutic effects and adverse drug reactions associated with the relationship between the DTN and DSN.
Ethics approval and consent to participate
Not applicable
Consent for publication
Not applicable
Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Competing interests
The authors declare that they have no competing interests.
Funding
This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning [grant number NRF-2020R1F1A1068423], the Ministry of Science and ICT [grant number NRF-2019M3C7A1032262] and the Technology Innovation Program (or Industrial Strategic Technology Development Program (20015086, Development of AI-based drug adverse events monitoring service.) funded By the Ministry of Trade, Industry & Energy(MOTIE, Korea).
Author’s contributions
H.W.H. conceived the project; H.W.H and C.W.L. designed the research; C.W.L. performed the research; C.W.L. analyzed the data; and H.W.H., C.W.L., S.M.K., S.O.S., and S.M.N. wrote the paper.
Acknowledgements
Not applicable
- Huang Y, Furuno M, Arakawa T, Takizawa S, de Hoon M, Suzuki H, et al. A framework for identification of on- and off-target transcriptional responses to drug treatment. Sci Rep. 2019;9(1):17603.
- Nagaraj AB, Wang QQ, Joseph P, Zheng C, Chen Y, Kovalenko O, et al. Using a novel computational drug-repositioning approach (DrugPredict) to rapidly identify potent drug candidates for cancer treatment. Oncogene. 2018;37(3):403-14.
- Kim IW, Jang H, Kim JH, Kim MG, Kim S, Oh JM. Computational Drug Repositioning for Gastric Cancer using Reversal Gene Expression Profiles. Sci Rep. 2019;9(1):2660.
- Grenier L, Hu P. Computational drug repurposing for inflammatory bowel disease using genetic information. Comput Struct Biotechnol J. 2019;17:127-35.
- Finan C, Gaulton A, Kruger FA, Lumbers RT, Shah T, Engmann J, et al. The druggable genome and support for target identification and validation in drug development. Sci Transl Med. 2017;9(383).
- Yildirim MA, Goh KI, Cusick ME, Barabasi AL, Vidal M. Drug-target network. Nat Biotechnol. 2007;25(10):1119-26.
- Hughes JP, Rees S, Kalindjian SB, Philpott KL. Principles of early drug discovery. Br J Pharmacol. 2011;162(6):1239-49.
- Luo Y, Zhao X, Zhou J, Yang J, Zhang Y, Kuang W, et al. A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information. Nat Commun. 2017;8(1):573.
- Schmidt AF, Heerspink HJL, Denig P, Finan C, Groenwold RHH. When drug treatments bias genetic studies: Mediation and interaction. PLoS One. 2019;14(8):e0221209.
- Santolini M, Barabasi AL. Predicting perturbation patterns from the topology of biological networks. Proc Natl Acad Sci U S A. 2018;115(27):E6375-E83.
- Moschovi M, Critselis E, Cen O, Adamaki M, Lambrou GI, Chrousos GP, et al. Drugs acting on homeostasis: challenging cancer cell adaptation. Expert Rev Anticancer Ther. 2015;15(12):1405-17.
- Yoo M, Shin J, Kim J, Ryall KA, Lee K, Lee S, et al. DSigDB: drug signatures database for gene set analysis. Bioinformatics. 2015;31(18):3069-71.
- Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science. 2006;313(5795):1929-35.
- Qin C, Zhang C, Zhu F, Xu F, Chen SY, Zhang P, et al. Therapeutic target database update 2014: a resource for targeted therapeutics. Nucleic Acids Res. 2014;42(Database issue):D1118-23.
- Davis AP, Murphy CG, Johnson R, Lay JM, Lennon-Hopkins K, Saraceni-Richards C, et al. The Comparative Toxicogenomics Database: update 2013. Nucleic Acids Res. 2013;41(Database issue):D1104-14
- Kim S, Thiessen PA, Bolton EE, Chen J, Fu G, Gindulyte A, et al. PubChem Substance and Compound databases. Nucleic Acids Res. 2016;44(D1):D1202-13.
- Maglott D, Ostell J, Pruitt KD, Tatusova T. Entrez Gene: gene-centered information at NCBI. Nucleic Acids Res. 2011;39(Database issue):D52-7.
- Huang DW, Sherman BT, Tan Q, Kir J, Liu D, Bryant D, et al. DAVID Bioinformatics Resources: expanded annotation database and novel algorithms to better extract biology from large gene lists. Nucleic Acids Res. 2007;35(Web Server issue):W169-75.
- Han HW, Ohn JH, Moon J, Kim JH. Yin and Yang of disease genes and death genes between reciprocally scale-free biological networks. Nucleic Acids Res. 2013;41(20):9209-17.
- Clarke DJB, Kuleshov MV, Schilder BM, Torre D, Duffy ME, Keenan AB, et al. eXpression2Kinases (X2K) Web: linking expression signatures to upstream cell signaling networks. Nucleic Acids Res. 2018;46(W1):W171-W9.
- Chen H, Zhang Z, Jiang S, Li R, Li W, Zhao C, et al. New insights on human essential genes based on integrated analysis and the construction of the HEGIAP web-based platform. Brief Bioinform. 2020;21(4):1397-410.
- Isik Z, Baldow C, Cannistraci CV, Schroeder M. Drug target prioritization by perturbed gene expression and network information. Sci Rep. 2015;5:17417.
- Park NI, Rogan PK, Tarnowski HE, Knoll JH. Structural and genic characterization of stable genomic regions in breast cancer: relevance to chemotherapy. Mol Oncol. 2012;6(3):347-59.
- DeLorenzo DM, Moon TS. Selection of stable reference genes for RT-qPCR in Rhodococcus opacus PD630. Sci Rep. 2018;8(1):6019.
Due to technical limitations, table 1 is only available as a download in the Supplemental Files section.
No competing interests reported.
- Table1.xlsx
Table 1. DB table from DSigDB.
- Supplementaryfigure1.gif
