Toxicogenomics of Bisphenol A and Neurodevelopmental Disorders

Bisphenol A (BPA) has been widely used in many industrial and consumer products and is known as an endocrine‐disrupting chemical. To find the underlying genetic basis and molecular mechanisms of BPA‐associated neurodevelopmental disorders (NDs), this chapter addressed the toxicogenomics of BPA with publicly accessed Comparative Toxicogenomics Database. The present results indicated that the key cellular components (CC) of the nervous system such as neuron, synapse, dendrite and axon are common in CC annotation; the commonly found molecular functions are neurotransmitter receptor or transducer binding or activity; and the main common biological processes include syn‐ aptic signalling, cognition, learning or memory, behaviour, the development of nervous system and brain. Neuroactive ligand‐receptor interaction, dopaminergic, glutamatergic and serotonergic synapses, monoamine transport and synaptic vesicle pathway were the common pathways. Simultaneously, the BPA‐disease may share the common pathways with drug addictions such as cocaine addiction. Unique pathways might also contribute to the BPA action in different NDs such as one carbon metabolism and detoxification of oxidative stress in Down syndrome. Although GO and pathway results indicate some common annotations, the predicted PPI molecular function clusters are quite different for each ND. In addition, some of the NDs share the same transcription factors (TFs) and miRNAs, which indicate these disorders have the similar expression profiles. Finally, chemicals having comparable interacting genes to BPA should be considered.


Introduction
The critical windows of vulnerability during human brain development are mainly from the third trimester to at least 2-3 years after birth [1]. Any developmental neurotoxicant exposure during this critical window has the possibility to cause various clinical neurodevelopmental disorders (NDs) in humans (e.g. autism, anxiety disorder, schizophrenia, dyslexia and epilepsy). Bisphenol A (BPA), as a proved endocrine-disrupting chemical, has been widely used as the plasticizer in many consumer products made of polycarbonate plastic such as baby bottles, tableware, food containers and water bottles. BPA can also be found in breast milk [2,3]. Infants and children were found to have the highest estimated daily intake of BPA per body weight [4,5]. A review by Healy et al. [6] affirmed that the potential for non-dietary sources make a substantial contribution to the total daily BPA exposure in young children and recommended risk-assessment models implement new frameworks, which specifically address exposure and hazard in early childhood. Pinson et al. [7] reviewed the human and rodent data on the neurodevelopmental alterations of BPA, and found that mostly reported effects were social and sexual behaviour and cognition that were unique to humans. The related mechanisms reported included the disruption of thyroid function, alterations of neurotransmitters levels, calcium signalling and neurotoxicity. Given the extensive BPA exposure during the critical windows of the brain development and the possible neurodevelopmental alterations, here we explore the possible genetic basis and the molecular mechanisms of BPA-associated NDs.

BPA-gene interactions and neurodevelopmental disorders
Comparative toxicogenomics database (CTD, http://ctdbase.org) is a robust, publicly available database that aims to advance understanding about how environmental exposures affect human health with manually curated information about chemical-gene/protein interactions, chemical-disease and gene-disease relationships, with functional and pathway data to aid in the development of hypotheses about the mechanisms underlying environmentally influenced diseases [8]. In this work, all BPA-gene/protein interactions were downloaded from CTD, in which BPA-gene/protein interactions associated to the following 17 NDs were selected as our targeted NDs for further analysis according to MESH ID used in CTD-anxiety disorders (AD), attention deficit and disruptive behaviour disorders (ADDBD), autism spectrum disorder (ASD), bipolar disorder (BD), developmental disabilities (DD), Down syndrome (DS), foetal alcohol spectrum disorders (FASD), intellectual disability (ID), language development disorders (LDD), learning disorders (LD), motor skills disorders (MSD), obsessive-compulsive disorder (OCD), pervasive child development disorders (PCDD), schizophrenia (Sch), speech disorders (SD), stereotypic movement disorder (SMD) and Tourette syndrome (TS). Thus, BPA-gene/protein interactions associated to these 17 NDs were collected for further analysis.

Microarray data and differently expressed gene screening
To explore the possible clinical application of the genes curated, we used Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) and ArrayExpress (http: //www. ebi.ac.uk / arrayexpress) to find the microarray data for peripheral blood, and special tissue gene expression profiling in NDs. The differently expressed genes (DEGs) in each ND samples were identified. If more than one microarray dataset is found, all the datasets are integrated to perform the meta-analysis for the DEGs. Based on the inference scores and the counts of inference BPA-interacted genes, we only selected ID, LD, Sch, ASD, AD and BD as our target diseases for the possible microarray data.

Gene function enrichment analysis
The curated genes in CTD for each ND were uploaded to DAVID 6.8 Beta (https://david-d.ncifcrf.gov/tools.jsp). Homo sapiens were used as the background population. The gene ontology (GO) and pathway were analysed [11,12]. Simultaneously, WikiPathways and Reactome of EnrichR [13,14] were used for duplicated pathway prediction. STRING (http://string-db.org) is a database of known and predicted protein-protein interactions (PPI), and can be used to predict the functional associations between proteins [15]. Based on the information provided by the STRING database, all genes related to each neurodevelopmental disorder are used to construct a PPI network. The functional molecules in the PPI network are subsequently identified by the molecular complex detection (MCODE) plugin [16] of Cytoscape [17]. The MCODE is a well-known automated method to find highly interconnected subgraphs as molecular complexes or clusters in a PPI network. The proteins in each module will be transferred in Genemania app [18] to predict the possible target proteins or biomarkers.

Intellectual disability
Of the 119 interacted genes, 117 were bi-interacted. GO analysis with these 117 genes indicated that BPA bi-interacted genes are involved in the biological processes (BP) such as cognition, the development of nervous system, head, brain, forebrain, pallium, telencephalon and embryonic organ ( Table 2). Other BPs included learning or memory, behaviour (social, singleor multi-organism), intraspecies interaction between organisms, embryonic organ morphogenesis, regulation of synapse structure or activity and neuron apoptotic process. Of the BPs, nervous system development was also reported in a recent study on systematic phenomics analysis for the genes muted in ID by Kochinke et al. [19]. Some genes possibly involved in the cellular component (CC) of somatodendritic compartment and the molecular functions (MF) of chromatic binding. Pathway analysis only found MECP2 and associated Rett Syndrome in WikiPathways, in which six BPA bi-interacted genes were involved. This might suggest that ID and Rett syndrome possess the same pathway.
PPI analysis found four molecular modules (Figure 1). TSEN2, TSEN34, TSEN54 and VRK1 were involved in module 1. TSEN2, TSEN34 and TSEN54 are tRNA splicing endonuclease subunits and can interact through physical interaction or co-expression. VRK1 might interact with TSEN2, TSEN34 and TSEN54 through the same pathway, physical interaction or co-expression. CLP1 and TSEN15 can physically interact with TSEN2, TSEN34 and TSEN54 [20][21][22], therefore, BPA has the potential to interact with these two genes. BPA might also interact with WARS and WARS2 because of their predicted interaction with TSEN34 [23,24]. WARS and WARS2 are involved in the tryptophan metabolic pathway, which have been reported to appear to provide a unifying biochemical basis for ASDs [25]. Tryptophan is a precursor of important compounds, such as serotonin, quinolinic acid and kynurenic acid, which are involved in neurodevelopment and synaptogenesis. Decreased tryptophan metabolism may alter brain development, neuroimmune activity and mitochondrial function. In module 2, ASCL1, HDAC4, BDNF, KDM6B, TH, HIST3H3, PAX6 and SIN3A interacted through co-localization, co-expression and physical interactions. TH and KDM6B linked other six node proteins through CALCOCO1 [26]. In module 3, LARP7, GAMT and TRMT1 were interacted by physical interactions, co-expression and predicted, pathway or genetic interaction. In this module, CDK9 plays an import role to link the three genes mainly by physical interactions [27][28][29], although there was no direct evidence for BPA-CDK9 interaction. CDK9 might associate with ID by JUN binding [30] or by AFF family of RNA-binding proteins [31]. In module 4, PARP1, FASN and PTEN interact through physical interaction, co-expression, predicted, pathway, genetic interaction and shared protein domains.
Recent findings have proven that the mTOR pathway is altered in cells with defective DNA repair. PARP1 is related to the accumulation of irreparable DNA damage [32], while PTEN is a phosphatase to mediate switching off the PI3K/Akt/mTOR signalling pathway, which has been reportedly associated with ID [33][34][35]. FASN (expression of fatty acid synthase) is found negatively correlated with PTEN [36], but the in-between genes were not explored. FASN may co-express with MAST2 [26], PRKDC [26,37] or BMI1 to physically interact with PTEN.

Learning disorders
A total of 29 BPA bi-interacted genes related to LD were found. Some of these genes are involved in the BPs such as behaviour, learning or memory, cognition, synaptic-signalling related, cell-cell signalling related, secretion, neuron death or apoptotic-related processes ( Table 3). Simultaneously, some of these genes may participate in the CCs such as synapse, presynapse, dendrite, somatodendritic compartment, axon or secretory vesicle. Five KEGG pathways might be influenced by BPA ( Table 4). Retrograde endocannabinoid signalling, serotonergic synapse and glutamatergic synapse pathways may be influenced by BPA and involve in LD. Nicotine addiction and Alzheimer's disease may share the same pathway with LD. WikiPathways indicated monoamine transport, synaptic vesicle pathway, MECP2 and associated Rett syndrome pathway might also be influenced. Interestingly, some genes involved in LD are also found in sudden infant death syndrome (SIDS) susceptibility pathways. Reactome pathway confirmed the serotonergic synapse pathway found in KEGG, and additionally suggested that organic anion transporters pathway may be influenced by BPA.  Only one module was found for LD (Figure 2: LD). In this module, HTR1A, MAPT, TH, PARK2 and TRH were interacted by physical interactions, co-expression, predicted, pathway, co-localization, genetic interactions and shared protein domains. PARK2 is known to play a role in neurological development or function and when disturbed can account for LD [38,39]. TH links PARK2 by sharing the same pathway [40], in which HOXA4 could co-express with TRHR [41], which could co-express with MAPT [42]. Serotonin gene (HTR1A) is also involved in this module, which is consistent with the serotonergic synapse and serotonin receptors pathway in KEGG and Reactome, respectively.

Schizophrenia
A total of 149 genes related to Sch were found to be BPA bi-interacted. Significant BPs included those found in LD such as behaviour, learning or memory, cognition, synaptic signalling related and cell-cell signalling related. Synaptic transmission-related, cell communicationrelated and phosphorus metabolic processes were also found ( Table 5). Like LD, the genes participate in synapse, presynapse, dendrite, somatodendritic compartment and axon cellular components. Other CCs include cell body, intrinsic or integral component of plasma membrane, synaptic membrane and plasma membrane region. Unlike LD with little significant MFs found, Sch showed many significant MFs including the activity of signal transducer, neurotransmitter receptor, transmembrane-signalling receptor, molecular transducer, glutamate receptor and dopamine neurotransmitter receptor, dopamine or catecholamine binding. The same as LD, WikiPathways found that BPA could be linked to Sch through Alzheimer's disease, monoamine transport and SIDS susceptibility pathways ( Table 6). The KEGG pathways such as neuroactive ligand-receptor interaction, cocaine addiction, dopaminergic synapse, cAMP signalling pathway and calcium-signalling pathway were also found to be significant. Reactome pathways included transmission across chemical synapse, amine ligand-binding receptors, neuronal system and signalling by GPCR, PDGF, FGFR4, FGFR3, FGFR1 or EGFR.

Autism spectrum disorders
We found 51 ASD genes that could biinteract with BPA. These genes are partly involved in the BPs such as organism development, system development, synapse organization, behaviour, learning or memory, regulation of synapse structure or activity, multicellular organismal process, nervous system development, membrane potential, cellular process and cell-cell signalling-related processes. Involved CC was synaptic membrane, and MFs included neuroligin family protein binding and chromatin binding. The KEGG pathways such as neuroactive ligand-receptor interaction and cocaine addiction were found as Sch. Like ID and LD, MECP2 and associated Rett syndrome (WikiPathways) was also found in ASD. PRC2 methylates histones and DNA, non-integrin membrane-ECM interactions, and gastrin-CREB-signalling pathway via PKC and MAPK were the main pathways found in Reactome.
Two molecular modules were found for ASD (Figure 2: ASD-1, ASD-2). In module 1, TET3, SIN3A, DNMT3A and DNMT3B were connected to each other by physical interactions, co-expression, predicted, pathway, co-localization, genetic interactions and shared protein domains. HDAC2 and MORF4L1 might be the candidate genes interacted with BPA because of their predicted interaction with SIN3A [24]. DNMT3L and MYB are another two main genes in this module because of their direct or indirect interactions with other genes. AVPR1A, OXTR and NTSR1 composite the module 2 through physical interactions, co-expression, pathway, co-localization and shared protein domains. OXT and NTS are another two main genes in this module; they share the same pathway with OXTR, NTSR1, AVPR1A and some other genes [40] and, thus, might be influenced by BPA.

Anxiety disorders
A total of 34 genes associated with AD were found bi-interacted with BPA. GO analysis for these genes indicated that behaviour, learning or memory, cognition, monoamine transport, chemical synaptic transmission, cell-cell signalling, anterograde trans-synaptic signalling and neurological system process were all significant BPs such as LD, Sch and ASD. Interestingly, blood circulation, circulatory system process and regulation of blood pressure were found significant for AD. Like ID, LD, Sch and ASD, significant CCs included neuron part, neuron projection, somatodendritic compartment, synapse part, axon, synapse, dendrite, cell body, presynapse and neuronal cell body. Significant MFs included receptor binding, neuropeptide hormone activity and G-protein-coupled receptor binding. The significant KEGG pathways were as those found in LD. Neuroactive ligand-receptor interaction, alcoholism, cAMP signalling pathway, serotonergic and dopaminergic synapse, Rap1 signalling pathway and retrograde endocannabinoid signalling are the potential KEGG pathways that might be influenced by BPA in AD. For WikiPathways, monoamine transport was again found related to BPA biinteracted genes in AD. Other significant pathways included circadian rhythm-related genes, nicotine activity on dopaminergic neurons, corticotropin-releasing, GPCRs, cytosine methylation, myometrial relaxation and contraction pathways and estrogen-signalling pathway. In Reactome pathways, GPCR related, Class A/1, Class B/2, G alpha-related signalling events and peptide ligand-binding receptors were found possibly involved in the BPA-AD interactions.
Only one molecular module was found for AD, in which UCN, ADORA2A, CRH, NPS, CRHR2, NPY, NPY1R, APP, HTR7, CNR1, GRM8, SLC6A3, DRD2 and CARTPT were involved (Figure 2: AD). The interaction modes for these genes included predicted, physical interactions, shared protein domains, co-expression and co-localization. The genes in this module are all involved in neuroactive ligand-receptor interaction (KEGG pathway) which is consistent with the Reactome pathways of GPCR signalling and G alpha signalling events.

Bipolar disorder
A total of 39 genes were found bi-interacted with BPA for BD. The BPs, CCs and MFs were quite the same as AD. Neuroactive ligand-receptor interaction, dopaminergic synapse, calcium-signalling pathway, neurotrophin-signalling pathway, synaptic vesicle cycle, insulin secretion, morphine signalling pathway, MAPK signalling pathway, glutamatergic synapse and serotonergic synapse were found in KEGG pathways. Like LD and Sch, SIDS susceptibility pathways was also found significant for BD. GPCR-related pathways such as monoamine GPCRs in WikiPathways and GPCR ligand binding in Reactome were found to be involved in BPA-BD as in BPA-AD.
One molecular module was found for BD (Figure 2: BD), in which D1, NTRK1, DRD5, PVALB, NTRK2, HTR2A, COMT and INS were involved. Shared protein domains, co-localization and co-expression were the main interactions in this module. COMTD1 and LRTOMT might also be influenced by BPA because of their shared protein domains with COMT.

Other neurodevelopmental disorders
A total of 14 genes were found for bi-interacted BPA in DS. GO analysis indicated cellular oxidant detoxification-related BPs significant for these genes, and the MF of antioxidant activity was found significant consistently. The pathway analysis showed that KEGG pathway like one carbon pool by folate, and some pathways related to folate, one carbon or water-soluble vitamins metabolism in WikiPathways or Reactome pathways. Detoxification of reactive oxygen species and cellular responses to stress were also found significant in Reactome pathways. Consistent with the results of GO and pathway analyses, PPI interaction showed two different molecular modules, one with SLC19A1, MTR and MTHFR, and the other with SOD1, PRDX2 and PRDX6. It is clear that the module 1 is related to the clustering function of folate and other water-soluble vitamins metabolism, and the module 2 is for the detoxification of reactive oxygen species. Folate pathway has been regarded as involved in the pathogenesis of DS. Simultaneously, BPA exposure has the potential effects on the human phenotypes and altering DNA methylation [49,50], which could be counteracted by the supplementation of methyl donors such as folate, choline, betaine and vitamin B12 [50]. Detoxification of reactive oxygen species and cellular responses to stress are important to maintain the mitochondrial function, which has been associated with the aetiology of early-onset dementia in patients with DS [51,52].
For other NDs, less reference count or low inference score was found. But the limited results of GO and pathway analyses showed similar BPs, CCs, MFs and pathways with the above mentioned NDs in some extent.

Gene regulation
Transcription factors (TFs) and microRNAs (miRNAs), the largest families of transacting, share a common regulatory logic and represent the most numerous gene regulatory factors in multicellular genomes [53,54]. The library of ENCODE and ChEA Consensus TFs from ChIP-X in EnrichR (http://amp.pharm.mssm.edu/Enrichr/ [13,14]) were used for the possible TFs and related networks. The TargetScan library in EnrichR was used for the possible miRNA interaction. Here we only analysed the genes of ID, LD, Sch, ASD, LD, SMD, BD and SD whose inference score all over 10.
For the TFs, it was only ID, ASD, AD and BD that were found significant TFs (Table 7). USF2, MAX, SPI1, SMAD4, POU5F1, PPARD, MYC and RUNX1 were found significant for ID. The regulated genes for each of these TFs are shown in Table 7. The direct evidences for the USF2 linked to ID were the regulating role of USF2 on FMR1 of Fragile X mental retardation [55,56]. SUZ12 was found common in ASD, AD and BD, and REST was found in both ASD and BD. SUZ12, as a component of the polycomb repressive complex, was shown to interact with some of long non-coding RNAs like AK055040 to involve in neural development and brain function [57]. REST is a key TF that represses expression of genes involved in neurogenesis and neuronal function in non-neural and immature neural cell types [58].
Some miRNAs were found in Sch, ASD, AD and BD ( Table 8). MIR-218 and MIR-485-3p were significant in both Sch and AD. It has been reported that miR-218 is involved in Sch [59], and miR-485-3p is associated with obsessive-compulsive disorder, a type of AD [60]. MIR-380-3p was found significant in both ASD and BD, but no direct evidence in human studies.

Comparable chemicals
The CTD provides a way to group chemicals based upon their biological effects, instead of their physical or structural properties, which provides a novel way to view and classify genes and chemicals and will help advance testable hypotheses about environmental chemical-gene disease networks [61]. Comparable chemicals were curated for the possible sharing with many of the networks common to BPA in neurodevelopmental disorders (

Future trends and conclusion
With the existed data libraries (mainly CTD, GO, pathway, TFs and miRNA relate databases), bioinformatics softwares (Cytoscape, MCODE and Genemania) or web-based tools (STRING, GEO, ArrayExpress, David and EnrichR), BPs, CCs, MFs, signal pathways and gene regulation in the BPA-gene-disease networks were presented. These data integration and curation CIGs: common interacting genes; NDs: neurodevelopmental disorders. yielded insight into the actions of BPA and provide a basis for developing hypotheses about the molecular mechanisms underlying the aetiology of the neurodevelopmental disorder ID, LD, Sch, ASD, AD and BD, although most of the other neurodevelopmental disorders showed no enough information to make a conclusion. The nervous system-related CCs such as neuron related, synapse related, dendrite and axon related are common in CC annotation; the commonly found MFs are neurotransmitter receptor binding or activity, signal transducer or receptor binding or activity; and the main commonly involved BPs include synaptic signalling, cognition, learning or memory, behaviour, the development of nervous system and brain, and the regulation of the related BPs. Neuroactive ligand-receptor interaction, dopaminergic, glutamatergic and serotonergic synapse, monoamine transport, synaptic vesicle pathway may involve in the action of BPA in the neurodevelopmental disorders. Simultaneously, the BPA disease may share the common pathways with drug addictions (cocaine addiction, nicotine addiction and alcoholism), or other types of neurological diseases (Alzheimer's disease, Rett syndrome and sudden infant death syndrome). Unique pathways might also contribute to the BPA action in different NDs like one carbon metabolism and detoxification of oxidative stress-related pathways in Down syndrome. Although GO and pathway results indicate some common characteristics, the predicted PPI molecular function clusters are quite different for each ND. In addition, some of the NDs share the same TFs and miRNAs, which indicate these disorders have the similar expression profiles. What needs to be emphasized that the BPA-gene-disease networks might be influenced by some of the comparable chemicals such as environmental pollutants, drugs, dietary pollutants or occupational exposure, which share the same interacted genes with BPA.
The integrated and curated biological processes and pathways shall shed light on the future studies to find the possible BPA interacted or influenced genes. This will contribute to complete the BPA-disease networks, which surely help to screen the potential biomarker of BPA-induced neurodevelopmental diseases. However, it should be noted that most of the evidences were from curation of the cell or animal experiments. Simultaneously, the biinteraction mode for BPA-gene interaction was adopted for the precise network. Therefore, the future study design should consider the human subjects. Given the sample shortage, the peripheral blood instead of the brain tissue should be preferred in the future. This will contribute to the clinical diagnosis or intervention. Finally, our results should be carefully interpreted because the results might be changed with the increasing abundance of the enrichment of BPA bi-interacted genes.