Introduction

Zebrafish (Danio rerio) is a well-known vertebrate model organism to study toxicology1,2. Advantages like cost-effective maintenance, high fecundity, external fertilization, short life cycle, fast growth rate, well-defined developmental stages, transparency of zebrafish embryos and, small size of the embryos as well as adults have made it ideal for toxicity testing3. Zebrafish also shows remarkable physiological and functional similarities with humans4. Approximately 70% of human genes have functional orthologs in zebrafish5. The use of zebrafish larvae also brings direct benefits of the 3R’s (Replace, Reduce, Refine) principle6,7.

Owing to rising critical health concerns upon toxicant exposure, this study focuses on the impact of such chemicals on gene expression networks during neurotoxicity and cardiotoxicity in zebrafish. Zebrafish is already an established animal model for evaluating cardiotoxicity and neurotoxicity as most of the signalling pathways and genes are conserved. This study sought to leverage the knowledge gained from zebrafish to model the potential gene networks in humans that may be affected during cardiotoxicity and neurotoxicity. We used meta-analysis to examine the alteration in genes during cardiotoxicity and neurotoxicity in zebrafish, caused by chemical exposure of any kind, using the database—ConsensusPathDB. ConsensusPathDB is a database that integrates information from multiple databases (31 public databases and constantly being updated) about physical entities and functional interactions to provide a more complete picture of biological interactions at the cellular level8,9. It achieves this through the application of a merging algorithm that identifies overlaps among physical entities/molecules (genes, proteins, enzymes, protein complexes and RNA) and interactions separately to identify the number of entities and interactions that are unique to each database. Upon input of genes that are differentially expressed in a particular phenotype, induced network module analysis of ConsensusPathDB can be used to generate interaction networks. The analysis maps the submitted genes to physical entities across all databases and then integrates information on interactions involving the entities from these databases. Multiple sub-networks are generated as modular hubs within the larger network of the studied phenotype, which may be concerned with different pathways. Through information on intermediate nodes and the types of entities and interactions that mediate a particular phenotype, the induced networks of ConsensusPathDB help generate novel insights that can be explored experimentally10.

Neurotoxicity is the ability of a chemical, biological, or physical agent to cause adverse functional or structural changes in the nervous system11. Advantages like easy penetration of chemicals through the external chorion membrane of zebrafish larvae by passive diffusion, development of the blood–brain barrier similar to mammals, and fast brain development have made zebrafish ideal for neurotoxicity screening12. Neurotoxicity endpoints include gene expression patterns, neural morphogenesis, and neuro-behavioral profiling13. We considered altered expression of genes related to neuronal function as neurotoxicity endpoint. Cardiotoxicity is defined as the toxicity that damages the heart muscle and other cardiac tissues and/or disrupts the electrophysiology of the heart14. Zebrafish are particularly suited to access cardiotoxicity since the heart rate and action potential are analogous to humans15. Along with this, optically transparent embryos allow the use of non-invasive techniques and whole animal imaging for cardiotoxicity evaluation. Cardiac function assessment can also be carried out by investigating teratogenic effects and through a variety of hemodynamic parameters, including heartbeat, cardiac output, fractional area change, fractional shortening, and vascular blood flow velocities and altered gene expression16. For the analysis in this study, we considered altered cardiac gene expression as cardiotoxicity endpoint.

In this study, we first manually curated articles published between 2019 and 2020 related to neurotoxicity and cardiotoxicity. We then subjected the genes altered during neurotoxicity and cardiotoxicity to interaction analysis using ConsensusPathDB. Pathophysiological interplays between the nervous and cardiovascular systems prompted us to investigate if there were any common target genes between neurotoxicity and cardiotoxicity and if known interactions existed between them.

Results

Interaction between genes upregulated exclusively during neurotoxicity in zebrafish

First, we analyzed the interactions between genes that were exclusively upregulated during neurotoxicity, excluding genes that were upregulated during both neurotoxicity and cardiotoxicity. This list of genes was subjected to “induced network modules analysis” in the gene set analysis in ConsensusPathDB. The analysis yielded two clusters (Fig. 1). The first cluster comprised seed node IL10 (Interleukin-10) (annotated in black) which is a cytokine with potent immunosuppressive potential17. Intermediate nodes (annotated in pink) included transcription factors such as SMAD3/SMAD4/GATA3 and JUN/JUNB. This cluster also included gene regulatory interactions mediated by LTA (Lymphotoxin alpha), TP53 (tumor suppressor p53), PRKAB1 (Protein Kinase AMP-Activated Non-Catalytic Subunit Beta 1), ODC1 (ornithine decarboxylase), and CXCL8 (C-X-C Motif Chemokine Ligand 8 also known as Interleukin-8). A sub-cluster with seed node LTA and an intermediate node protein encoded by the PLAGL1 (Pleiomorphic Adenoma Gene-Like 1) gene connected hubs centered on IL-10 and TP53. LTA is known to mediate a large variety of inflammatory and immunostimulatory responses18. Another sub-cluster was formed by seed nodes like PRKAB1, TP53, ODC1, and intermediate node MYC/Max/Cbp/p300. Here, TP53 was central to the network and formed forward gene regulatory networks with both PRKAB1 and ODC1 genes and proteins, where additional intermediate node MYC/Max/Cbp/p300 was found to be involved in gene regulatory interactions specifically with ODC1. CXCL8 gene further displayed gene regulatory interactions with intermediate nodes HB-EGF/EGFR, ATF2/JUND/macroH2A, and Fra1/JUND. The ATF2/JUND/macroH2A gene regulatory interaction was interestingly shown to have an inhibitory effect on the CXCL8 gene and protein. CXCL8 is a major mediator of the inflammatory response19. p53 is also known to be involved in the regulation of several proinflammatory genes in human macrophages like IL-6, IL-8, and CXCL1 (C-X-C motif chemokine Ligand 1)20. This confirmed the importance of the seed nodes in this cluster as major inflammatory contributors to neurotoxicity.

Figure 1
figure 1

Interactions between genes that are exclusively upregulated in neurotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Line ending with an arrow represents stimulation whereas line ending with a small blunt line represents inhibition.

Full size image

The second cluster consisted of seed nodes NFE2L2 (Nuclear Factor, Erythroid 2 Like 2) and HMOX1 (Heme oxygenase) where the HMOX1 gene and protein were shown to be negatively regulated (inhibited) by the intermediate node Fra1/USF2. NFE2L2 is a transcriptional activator that binds to the ARE (antioxidant response elements) and protects against oxidative stress by upregulating other antioxidant proteins21. HMOX1 is involved in heme catabolism, it cleaves heme to form biliverdin. Biliverdin is subsequently converted to bilirubin and carbon monoxide (a gaseous neurotransmitter) by biliverdin reductase22,23. This gives it an anti-inflammatory property by upregulating IL10. It also exhibits cytoprotective effects since excess of free heme sensitizes cells to undergo apoptosis24. Overall, the interaction modules in this gene set indicated that most of the genes were linked with neuroinflammation.

Furthermore, a protein interaction was found between the TP53 gene and the CXCL8 gene. No further biochemical reaction or genetic interactions were observed between the nodes.

Interactions between genes upregulated in both neurotoxicity and cardiotoxicity in zebrafish

When we subjected the list comprising genes that were upregulated in both neurotoxicity and cardiotoxicity to analysis in ConsensusPathDB, one cluster comprising only gene regulatory interactions was obtained (Fig. 2). Seed node PTGS2 (Prostaglandin G/H synthase 2 precursor) gene and protein were found to be involved in gene regulatory interactions that also involved two intermediate nodes namely JUN/FOS/NFAT1-c-4 and JUN family. FOS and Jun are transcription factors also linked to inflammation25. No further biochemical reaction, genetic, or protein interactions were observed between the nodes.

Figure 2
figure 2

Interactions between genes that are upregulated both in neurotoxicity and cardiotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Grey square (JUN family) indicates that the entity whether gene/protein/compound etc. was not specified by the database.

Full size image

Interactions between genes exclusively downregulated in neurotoxicity in zebrafish

When the list of genes downregulated in neurotoxicity excluding the commonly downregulated genes in neurotoxicity and cardiotoxicity, was subjected to the induced network modules analysis, four distinct clusters were obtained, three of which were anchored by gene regulatory interactions (Fig. 3). The first cluster comprised of three gene regulatory sub-clusters anchored by the seed genes FOS, ESR1 (Estrogen receptor alpha), and SOCS3 (Suppressor of cytokine signaling 3). The SOCS3 sub-cluster presented a complex network with ESR1 through protein and biochemical level interactions mediated by seed nodes PIK3CA (Phosphatidylinositol 3-kinase catalytic subunit, alpha isoform), and mTOR (Mammalian target of rapamycin) proteins. The SOCS3 sub-cluster elucidated the presence of intermediate nodes Jak2/Leptin Receptor and MIR203 through gene regulatory interactions. MTOR protein networked with ESR1 protein through biochemical as well as protein level interactions. Seed node proteins ARRB2 (Beta-Arrestin 2, isoform 1), CDC42 (Cell Division Cycle 42), PIK3CA, EPAS1 (Endothelial PAS domain-containing protein 1), and ESR1 interacted at the protein level. Further, ATF2/JUN/ER alpha, ATF2(dimer)/ER alpha, and JUN/FOS/ER alpha intermediate node pathways were implicated in the gene regulatory hub anchored in the ESR1 seed node, whereas ERK1-2/ELK1 was the intermediate node implicated in the sub-cluster anchored by the seed node FOS.

Figure 3
figure 3

Interactions between genes that are exclusively downregulated in neurotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Dotted lines represent at least one physical interaction. Line ending with an arrow represents stimulation whereas line ending with a small blunt line represents inhibition.

Full size image

The second cluster comprised of seed node proteins SOX9 (SRY-Box Transcription Factor 9) and MATN3 (Matrilin 3) interconnected by an intermediate node protein MATN1 (Matrilin 1) through gene regulatory and protein level interaction, respectively. The SOX9 protein has been associated with initiation as well as progression of solid tumors26. The third cluster consisted of seed node MMP9 (Matrix metalloproteinase-9) and intermediate nodes JUNB/FOSB, Fra1/FIAT, and FGFR-FGF2 complex/N-cadherin.

When biochemical reaction interactions were included in the analysis, all the clusters interacted. Seed nodes like MTOR and CDK6 showed a physical biochemical reaction interaction (dotted green lines in Fig. 3). ARRB2 and LHCGR (luteinizing hormone/choriogonadotropin receptor) also showed networking at the biochemical level in the first cluster. The protein encoded by the mTOR gene belongs to a family of phosphatidylinositol kinase-related kinases. These kinases mediate cellular responses to stress such as DNA damage and nutrient deprivation27. This protein acts as the target for cell-cycle arrest and immunosuppressive effects of the (FK506 Binding Protein 12) FKBP12-rapamycin complex28. The protein encoded by the CDK6 gene is a member of the cyclin-dependent protein kinase family and is an important regulator of cell cycle progression29,30. They are known to be over-expressed in leukemia and malignancies31. ARRB2 is a major regulator of GPCR (G-Protein Couple Receptor) signaling including GPCR desensitization and internalization32,33. It also links GPCRs to clathrin-coated pits, regulation of cytoskeletal rearrangement and cellular localization, translocation, and regulation of signaling elements in the GPCR cascade34. The fourth cluster was devoid of any gene regulatory interactions and only showed proteins CACNA1A, CACNA1D, CACNA1I, CACNA1F, and GRIA2 (Glutamate ionotropic receptor AMPA type subunit 2) interacting via physical biochemical interactions. One can note that this group is comprised of mainly membrane receptors such as GPCRs and voltage-gated calcium channels like CACNA1D, CACNA1F, CACNA1A, and CACNA1I which may be associated physically in membrane microdomains.

Several protein–protein interactions were also observed in the proteins encoded by genes that were downregulated during neurotoxicity (orange lines in Fig. 3). The protein level interactions in the biggest cluster involved seed node proteins ARRB2, CDC42, ESR1, EPAS1, FOS (Fos Proto-Oncogene, AP-1 Transcription Factor Subunit), MTOR, and PIK3CA. EPAS1 also known as Hypoxia-inducible factor 2-alpha is a transcription factor that is induced when oxygen levels fall35. The fourth cluster, the only one devoid of any gene regulatory interactions implicated GRIA2 and GRIA3 in protein–protein interaction. GRIA3 plays an important role in excitatory synaptic transmission and is involved in bipolar disorder and nonspecific X-linked mental retardation36.

Interactions between genes downregulated during neurotoxicity and cardiotoxicity in zebrafish

Only two genes FLT4 (Fms-related tyrosine kinase 4) and KDR (Kinase Insert Domain Receptor) were found to be downregulated with a fold change of more than 3 in both neurotoxicity and cardiotoxicity. When these two genes were subjected to gene regulatory interaction analysis, few intermediate nodes were observed (Fig. 4). Intermediate nodes comprised of HIF2A/ARNT, HHEX gene, HEY1, and ETS1. KDR or VEGFR2 (Vascular Endothelial Growth Factor Receptor 2), which is a tyrosine-protein kinase acts as a cell surface receptor for VEGFA, VEGFC and VEGFD. It is known as a critical player in angiogenesis, vascular development and permeability37.

Figure 4
figure 4

Interactions between genes that are downregulated in both neurotoxicity and cardiotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Dotted line represents at least one physical interaction.

Full size image

Seed nodes FLT4 and KDR interacted via biochemical reaction interactions (Fig. 4). FLT4 is a receptor tyrosine kinase for VEGFC and VEGFC. It is involved in adult lymph angiogenesis besides development of the cardiovascular system and vascular network during embryonic development38.

Interactions between genes upregulated exclusively during cardiotoxicity in zebrafish

When genes that were upregulated during cardiotoxicity were subjected to gene regulatory interactions, three interaction clusters were observed (Fig. 5). Each of the three seed nodes ESR1, NR4A1 (Nuclear Receptor 4A subunit 1), and IL1β (Inerleukin-1 beta) formed a separate cluster. The protein encoded by NR4A1 acts as a nuclear transcription factor. This is an orphan nuclear receptor, the loss of which has been shown to lead to an inflammatory phenotype and increased atherosclerosis39. The protein encoded by the IL1β gene is a member of the interleukin 1 cytokine family, the major group of ligands and receptors associated with acute and chronic inflammation that play a central role in many cardiovascular diseases40. Suppression of IL1β has been used for treating severe symptoms in these conditions40. Intermediate nodes like JUN/FOS/ER alpha, ATF2 (dimer)/ER alpha, and ATF2/JUN/ER alpha were included in the first cluster. Intermediate nodes like MEF2D/NFAT1/Cbp/p300 were included in the second cluster. Intermediate nodes like STAT1-3 (Signal Transducers and Activators of Transcription) dimer and p300/CBP/RELA/p50 were included in the third cluster.

Figure 5
figure 5

Interactions between genes that are upregulated exclusively in cardiotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Grey square (STAT1-3 dimer) indicates that the entity whether gene/protein/compound etc. was not specified by the database.

Full size image

Interactions between genes exclusively downregulated in cardiotoxicity in zebrafish

Three clusters of gene regulatory interactions were observed between the genes downregulated during cardiotoxicity (Fig. 6). Seed node VEGFA interacted with intermediate nodes PAS, HIF1A/ARNT/Cbp/p300/HDAC7, and SEMA4A: PLXND1 (Fig. 6). VEGFA (Vascular Endothelial Growth Factor A) is a growth factor that induces proliferation and migration of vascular endothelial cells41. In the second cluster, seed node ITGB1 (Integrin Subunit Beta 1) interacted with the intermediate node SRF (dimer)/MAL (dimer)/SCAI. ITGB1 are membrane receptors involved in cell adhesion, embryogenesis, homeostasis, tissue repair, immune response, and metastatic diffusion of tumor cells42. It has also been shown to increase stem cell survival and cardiac function after myocardial infarction43. In the third cluster, seed node MEF2C (Myocyte Enhancer Factor 2C) interacted with intermediate nodes SMAD2/SMAD2/SMAD4/FOXH1/NKX2-5 and CSX/GATA4. The protein encoded by MEF2C is a cardiac transcription factor that has been well-studied as a differentiation factor known to regulate a variety of cellular processes44.

Figure 6
figure 6

Interactions between genes that are downregulated exclusively in cardiotoxicity. Black labels denote seed nodes and pink labels denote intermediate nodes. Each line represents an interaction. Dotted line represents at least one physical interaction.

Full size image

Two additional clusters were observed with biochemical interactions. (Fig. 6). The first cluster consisted of seed nodes NOTCH4 and NOTCH1 (neurogenic locus notch homolog protein 4 and 1) and intermediate node FBXW7:SKP1:CUL1:RBX1. Mutations in the NOTCH1 gene have been implicated in multiple conditions including aortic valve disease, Adams-Oliver syndrome and leukemia, while mutations in the NOTCH4 gene have shown association with schizophrenia45. The second cluster consisted of seed nodes RHO (Rhodopsin), OPN1MW (Opsin 1, Medium Wave Sensitive, and OPN1SW (Opsin 1, Short Wave Sensitive) all of which interacted with a compound 11cRAL (11-cis-retinyl-Green-sensitive opsin, Opsin-1) that emerged as an intermediate node (Fig. 6 dark yellow square). The RHO gene encodes a protein found in rod cells; mutations in this gene are a cause of congenital blindness46. OPN1MW encodes opsins that are sensitive to medium-wavelengths of light while OPN1SW encodes for the blue cone pigment protein. Deuteranopic color blindness has been attributed to defects in OPN1MW gene, while tritanopia is attribute to mutations in OPN1SW47,48.

Discussion

This study aimed at using gene expression data of cardiotoxicity and neurotoxicity phenotypes from the zebrafish animal model to understand similar gene networks in humans. Induced network modules analysis allowed to find major pathways involved in the phenotypes and to identify hidden molecular players that could be explored experimentally as useful biomarkers. Intermediate nodes thus obtained provided an opportunity to observe converging pathways and identify molecular players that may be affected, though not transcriptionally. These nodes may encode proteins that signal together in a microdomain complex and thus may be relevant for the particular phenotype9.

The interaction networks obtained in our study predominantly involved gene-regulatory interactions. The analysis was successful in identifying major pathways anchored by seed nodes (input genes) responsible for cardiotoxicity or neurotoxicity phenotype. Table 1 describes these anchoring seeds at the gene regulatory, protein as well as biochemical reaction levels. We propose that these genes and proteins possess the potential to be utilized as biomarkers for investigatingneurotoxicity and cardiotoxicity phenotypes in toxicity studies and for the assessment of health and diseased states in the zebrafish model organism and possibly in humans.

Table 1 Specific genes that can serve as biomarkers for studying toxicity in zebrafish (Danio rerio).
Full size table

Estrogen receptor alpha (ESR1), a gene downregulated in neurotoxicity but upregulated in cardiotoxicity, revealed itself as a unique gene implicated in both the pathologies at both the gene-regulatory and protein level interaction. The ESR1 gene is generally implicated in estrogen resistance in breast cancer, it encodes for a nuclear hormone receptor and regulates cellular proliferation, and it also associates with DNA-binding transcription factors such as ATF-2, FOS, and JUN to mediate signaling independent of the estrogen response element49,50. These are also the intermediate nodes observed in our analysis. ESR1 has been shown to mediate neurodevelopmental toxicity in zebrafish upon exposure to aromatase and phthalates through impaired neurogenesis51. It also accelerates heart regeneration via inflammatory signaling52. These studies assert the results of the interaction modules analysis in our study. These cross-observations also indicate the potential of ESR1 to be employed as a biomarker for both phenotypes through pathways implicated in regeneration and inflammation. PTGS2, a key enzyme in prostaglandin biosynthesis, implicated in epilepsy53, cardiomyopathy and regeneration54 was also observed to be upregulated in both cardiotoxicity and neurotoxicity in our studies and through the interaction networks it associated with JUN family of transcriptions factors underscoring the overlapping pathways in toxicity and cancer. Similarly, the Kinase insert domain (KDR) gene, encoding for the VEGF receptor has been implicated in angiogenesis and glioma55,56,57 and was found to be downregulated in both neurotoxicity and cardiotoxicity phenotypes. These observations highlight a major use case of molecular players like ESR1, PTGS2, and KDR as biomarkers of the cardiotoxicity and neurotoxicity phenotypes anchored in regenerative and inflammatory signaling pathways. A unique observation of our study is the substantial overlap of genes involved in tumor manifestation/association with the genes involved in cardiotoxicity and neurotoxicity such as ESR1, PTGS2, LTA and KDR. Many of these genes are already proposed as cancer biomarkers58,59,60,61. It would be pertinent to determine if cardiotoxicity and neurotoxicity would lead to carcinogenesis though some epidemiological evidence exits. It is worth to mention that though there appears a convergence of toxicity pathways with cancer pathways, the intermediate nodes that group these networks into sub-networks underscore the absence of bias in our analysis and may represent a true overlap of toxicity pathways with cancer pathways.

The induced networks aim to connect as many input seed nodes as possible, however, certain seed nodes that displayed high differential expression in experiments, did not appear in the networks generated through ConsensuPathDB in our study. This absence however does not diminish the significance of genes such as cyp1a1, gstp1, cyp19A1, and Rgr. Some of these genes such as cyp1a1 and gstp1 are already associated with toxicity and may act as biomarkers for toxicity62,63,64. It is possible that they are involved in the toxicity manifestation through other functional interactions in differently connected pathways, and therefore are not depicted in the pathways in our figures.

In our study, input genes were not separated based on the differential expression at a particular developmental or adult stage. Most of the studies that are referenced, used embryonic/larval stages of zebrafish for toxicity analysis (Supplementary Table S9). Hence, our study may provide a limited opportunity to understand the concerned phenotypes from the perspective of adult zebrafish stages. However, all the genes used in our analysis are expressed in humans and therefore our results may be extrapolated to human toxicity.

Insights regarding the role and networking of these genes can serve to establish phenotypes and test hypotheses involving the novel intermediate genes revealed as a result of the induced network modules analysis. This study makes a case for utilizing gene expression data sets and induced network modules analysis to lead researchers to novel biomarkers that can be used to investigate and establish phenotypes.

Methods

We searched the literature in academic databases PubMed and Google Scholar for articles that reported cardiotoxicity and neurotoxicity in zebrafish in the year 2019 and 2020. The study was conceptualised in early 2021 and hence literature from the previous two years was reviewed. Also, there was a considerable increase in the publications in this area from previous years and therefore it was not possible to sample all the publications. Therefore, we chose 2 years which had around 2678 publications with the search terms used in our study. The search terms used were—“Neurotoxicity in zebrafish”, “Cardiotoxicity in zebrafish”, “acute toxicity in zebrafish”, “Chronic toxicity in zebrafish”. 592, 218, 938 and 930 hits were received for each string, respectively. Then, the articles were manually screened for gene expression studies and data such as gene name, gene alteration (upregulated or downregulated), primer sequences, study design including the chemical used, time of exposure, the time point at which qPCR was done, strain, and the developmental stage of the zebrafish. Toxicity articles that did not report gene expression data were excluded from our analysis.

A total of 76 articles for neurotoxicity and 34 articles for cardiotoxicity were found relevant and were included in the analysis presented here (Titles of these articles are provided in the Supplementary Tables S1 and S2 respectively). From the data collected from the relevant articles, the gene lists were created and divided based on the observed alteration. Six gene lists were created: (1) Exclusively upregulated genes in neurotoxicity (Supplementary Table S3), (2) Common upregulated genes in neurotoxicity and cardiotoxicity (Supplementary Table S4), (3) Exclusively downregulated genes in neurotoxicity (Supplementary Table S5), (4) Common downregulated genes in neurotoxicity and cardiotoxicity (Supplementary Table S6), 5) Exclusively upregulated genes in cardiotoxicity (Supplementary Table S7), and (6) Exclusively downregulated genes in cardiotoxicity (Supplementary Table S8). Table 2 shows the number of genes in each list. Table 3 shows gene names and their abbreviations. From these genes, only those genes having fold change greater than 3 were considered for induced network modules analysis to introduce stringency in the interaction analysis (Supplementary Tables S3–S8). In the induced network modules of ConsensusPathDB (Humans), “gene set analysis” was used to analyze the six sets of genes. Prior to submission, each gene list was corrected to only include those paralogs of zebrafish that share greater homology to the human ortholog. We employed this approach to find putative gene biomarkers of toxicity that can be extrapolated to humans since majority of pathways are conserved in humans and zebrafish and zebrafish is already an established model of cardiotoxicity and neurotoxicity. Each input gene served as a “seed node” for which we determined, gene regulatory interactions, protein interactions and biochemical interactions in the form of an induced network module. Additionally, the generated networks included genes/proteins etc. called as “intermediate nodes” that associate two or more seed genes with each other and have many significant connections within the entire induced network module. Though these intermediate nodes are associated with the seed nodes, however they did not appear in the input gene list. So, one may speculate that they are not regulated at the transcriptional level in the phenotype under study, but have been otherwise shown in the literature to have some association with the seed nodes. Table 4 shows intermediate node complex genes obtained in our study and their abbreviations.

Table 2 Numbers of genes in each list.
Full size table
Table 3 Gene names and their abbreviations.
Full size table
Table 4 Intermediate node complex genes and their abbreviations.
Full size table

In our analysis, these interactions were determined at the high-confidence option to obtain intermediate nodes. These intermediate nodes were further ranked based on the significance of their association with our supplied gene list given their overall connectivity in the background network. This rank was denoted by a z-score, which is computed using a binomial proportions test and which can be dynamically controlled by the user to create sub-networks with intended stringency9. For determining a suitable z-score for our analysis, induced network modules were generated at a z-score threshold of 0, 25, 50, 75, and 100. While the networks generated at z = 0 were too dense to comprehend, networks at z = 100 were too sparse and barely gave any information about gene networks. Thus, z = 75 was chosen as a suitable threshold for generating networks with reliable stringency and sufficient information for identifying important sub-networks regulating the particular phenotype in our study.