MicroRNA Profiling as Tool for In Vitro Developmental Neurotoxicity Testing: The Case of Sodium Valproate

Studying chemical disturbances during neural differentiation of murine embryonic stem cells (mESCs) has been established as an alternative in vitro testing approach for the identification of developmental neurotoxicants. miRNAs represent a class of small non-coding RNA molecules involved in the regulation of neural development and ESC differentiation and specification. Thus, neural differentiation of mESCs in vitro allows investigating the role of miRNAs in chemical-mediated developmental toxicity. We analyzed changes in miRNome and transcriptome during neural differentiation of mESCs exposed to the developmental neurotoxicant sodium valproate (VPA). A total of 110 miRNAs and 377 mRNAs were identified differently expressed in neurally differentiating mESCs upon VPA treatment. Based on miRNA profiling we observed that VPA shifts the lineage specification from neural to myogenic differentiation (upregulation of muscle-abundant miRNAs, mir-206, mir-133a and mir-10a, and downregulation of neural-specific mir-124a, mir-128 and mir-137). These findings were confirmed on the mRNA level and via immunochemistry. Particularly, the expression of myogenic regulatory factors (MRFs) as well as muscle-specific genes (Actc1, calponin, myosin light chain, asporin, decorin) were found elevated, while genes involved in neurogenesis (e.g. Otx1, 2, and Zic3, 4, 5) were repressed. These results were specific for valproate treatment and―based on the following two observations―most likely due to the inhibition of histone deacetylase (HDAC) activity: (i) we did not observe any induction of muscle-specific miRNAs in neurally differentiating mESCs exposed to the unrelated developmental neurotoxicant sodium arsenite; and (ii) the expression of muscle-abundant mir-206 and mir-10a was similarly increased in cells exposed to the structurally different HDAC inhibitor trichostatin A (TSA). Based on our results we conclude that miRNA expression profiling is a suitable molecular endpoint for developmental neurotoxicity. The observed lineage shift into myogenesis, where miRNAs may play an important role, could be one of the developmental neurotoxic mechanisms of VPA.


Introduction
Exposures to xenobiotics during embryonic, fetal, and perinatal periods are of particular concern since scientific evidence suggest that the developing central nervous system (CNS) is much more vulnerable to chemicals than the adult CNS. Exposure to drugs and environmental chemicals during critical developmental stages is likely to contribute to the increasing incidence of neurodevelopmental disorders in children [1][2][3][4][5]. Today one out of six children is diagnosed with a developmental disorder [6] [7], which include, for instance, learning disabilities and delays, autism spectrum disorders (ASD), and the attention deficit and hyperactivity disorder (ADHD). There is a critical deficiency of knowledge when it comes to the developmental neurotoxicity of drugs and chemicals. So far, only very few compounds have been identified as developmental neurotoxicants [1], but this might not reflect the actual prevalence of neurotoxicants in the human environment, since only a minor portion of the more than 80,000 chemicals used worldwide have been tested to determine their potential to trigger developmental neurotoxicity (DNT) in vivo. An important restriction for routine assessment of chemical-mediated DNT is the high cost-approximately $1.4 million per substance-and the time consumption associated with the conduct of regulatory in vivo tests currently accepted at the level of international guidelines (OECD TG 426 and US EPA 712-C-98-239) [8] [9]. Hence, there is a critical need for the development of alternative non-animal, highthroughput methods for DNT assessment to ensure the safety of chemicals and drugs.
Several in vitro DNT approaches addressing different aspects of neurogenesis have been developed during the last two decades. These include studies on invertebrates such as nematodes (Caenorhabditis elegans) or flies (Drosophila melanogaster), vertebrate models such as zebrafish, as well as different cellular systems such as mouse (m) or human (h) embryonic stem cells (ESCs), neural progenitor cells (NPCs), or primary cell cultures (reviewed in [10]). We prioritized the use of pluripotent mESC-based approaches since differentiation of ESCs covers early stages of neurogenesis, which are not represented in NPCs or primary cells. The protocol for neural differentiation of mESCs has been previously developed and successfully adapted for chemical exposure [11]. This simple, robust, and reproducible method of neural differentiation of mESC is a suitable cellular model to study molecular perturbations induced by chemical exposure during neurogenesis that might provide an insight into the underlying molecular mechanisms and pathways mediating the toxicity of DNT substances.
As a toxicological molecular endpoint, microRNA (miRNA) expression profiling was chosen. MiRNAs are small noncoding RNA molecules (21-25 nt), which were discovered a decade ago as powerful post-transriptional regulators of embryonic development [12]. The expression of thousand of genes is regulated by hundreds of different miRNAs that bind to (imperfect) complementarity sites in the 39UTR regions of the respective target mRNAs, thereby interfering with the translation of the encoded proteins [13]. One miRNA may have several hundreds of targets and one target mRNA may be regulated by more than one miRNA. More then 60% of the protein coding genes are predicted to be regulated by miRNAs [14]. Increasing evidence points to the importance and significance of miRNA networks in coordination and fine-tuning of gene expression with high temporal and spatial specificity (reviewed in [15]). More than 50% of all identified miRNAs are expressed in the brain, where they play a particularly significant role in brain development by regulating developmental timing, cellular differentiation, proliferation, lineage determination, synaptogenesis, and brain morphogenesis (reviewed in [16] and [17]). In animal studies, depletion of the RNAse III Dicer triggers the loss of miRNA synthesis and results in severe defects in brain development and morphogenesis [18,19]. Taking into account both the significant role of miRNA in the development of the CNS, and the fact that mRNAs with miRNA binding sites are twice as likely to be sensitive to environmental chemicals exposure than those which lack miRNA binding sites [20], suggests that miRNA (with their ability to post-transcriptionally regulate thousands of genes) can rapidly respond to environmental disturbances. Moreover, many miRNA are phylogenetically conserved molecules, which makes them perfect endpoints in toxicological studies. Studying the perturbations of phylogenetically conserved miRNAs may help to overcome the problem of interspecies differences and allow the extrapolation of the results obtained in non-human species to humans. However, the role of miRNAs in toxicology, especially neurotoxicology, still remains in an exploratory phase. Several studies have addressed the question of the role and possible consequences of perturbations in miRNA expression under exposure to environmental toxicants or drugs (reviewed in [21][22][23]). Transcriptomics perturbations during neural differentiation of mESC induced by developmental neurotoxicants were extensively studied by Piersma's group [24,25]. They did not, however, address the question of miRNA profiling. Recently, miRNA profiling as a toll for DNT was assed by Bal-Price's laboratory with methyl mercury chloride as the example [26].
In this study we analyzed miRNA expression profiles (miR-Nome) after neural differentiation of mESCs under the exposure to a well known developmental neurotoxicant: sodium valproate (salt of valproic acid, VPA) [27] to establish miRNA profiling as a molecular tool for DNT testing. VPA is a broadly used antiepileptic drug, and is also applied for the treatment of bipolar disorder, cancer, and migraine [28][29][30]. The therapeutic effect of VPA is a combination of several effects on cellular signalling, including induction of GABAergic neurotransmission, promotion of neuronal remodelling, and neurogenesis [31][32][33]. The teratogenic effects of VPA that result in neural tube defects (NTD), heart malformations, and craniofacial and skeletal anomalies are well described. The mechanisms underlying these malformations, however, remain ambiguous [27]. VPA was shown to influence N-cadherin-mediated cell adhesion, to modulate Wnt-dependent gene expression, and to inhibit histone deacetylase (HDAC) activities [34], thereby facilitating the establishment of a pluripotent state [35][36][37]. Since miRNAs have been shown to represent a major target of HDAC inhibitors (reviewed in [38]), miRNA might also be important components of the molecular mechanisms mediating the neurotoxic effects of VPA. To analyze whether VPA effects on miRNA expression are compound-specific, we also exposed the cells to another epigenetically active developmental neurotoxicant, sodium arsenite. Arsenic is a very toxic, carcinogenic environmental contaminant associated with the induction of neurobehavioral dysfunctions [39,40]. The results of this study provided new molecular insight into the mechanisms mediating VPA toxicity: long term exposure to valproate but not arsenite during neural differentiation of mES cells promoted disbalance in lineage specification by induction of myogenesis in neuronally differentiating cultures. We conclude that miRNA expression profiling can be used as a novel molecular endpoint for the identification of DNT activities and substance specific effects.
Neural differentiation was induced according to the established protocol [11]. Briefly, prior to induction of neural differentiation (day 0), W4 cells were cultured at high density in routine culture media in the presence of mLIF for 24 h. On the next day (day 1), neural differentiation was induced by plating the cells at low density in DMEM-F12/Neurobasal Medium (1:1), supplemented with B27 (Gibco, 1:100), N2 (Gibco, 1:200), insulin (Sigma, 10 mg/ ml), bovine serum albumin, fraction V (Sigma, 150 mg/ml), bmercaptoethanol (Sigma, 0.1 mM), 50 U/ml penicillin, 50 mg/ml streptomycin, 0.2% FCS (from day one to five), 1 mg/ml laminine, from day one to day seven (Sigma), 10 ng/ml basic fibroblast growth factor, from day seven human (bFGF; Strathmann Biotec AG) on poly-L-ornithine(PLO)-coated dishes. The medium was changed on days 5, 7, 9, 12 and 14. Murine primary cortical neurons were prepared from C57Bl/6 mice pups on day 17 of embryonic development and kindly provided by Prof. Dr. S. Lehnardt. Dr Lehnardt's group has been routinely isolating murine cortical neurons for their experiments and provided us the surplus of the cultures as a gift. The experimental procedure of primary neuron isolation, ethical statement and respective permits are described elsewhere [42,43]. Neurons were plated at a density of 1.7610 5 /cm 2 on poly-D-lysine-coated culture flasks in Neurobasal medium (Gibco) supplemented with 2% B-27 (Gibco), 2 mM glutamine, 50 U/ml penicillin and 50 mg/ml streptomycin.

Flow cytometry
The efficiency of neural differentiation was estimated by flow cytometry as described in [44]. Briefly, cells were harvested either 12 or 16 days after induction of neural differentiation by trypsin/ EDTA treatment, fixed with 2% paraformaldehyde (PFA), permeabilized with 0,15% saponin (Sigma) and stained with anti-bIII-tubulin (1:3200, Sigma) and anti-MAP2 (1:800, BD Biosciences) followed by incubation with R-phycoerythrin-conjugated goat antiserum (1:60, DakoCytomation). The cells were analyzed in a BD Calibur flow cytometer (Fig. S1C). Cells lacking primary antibody were used as a control. Expression of stem cell markers Oct3/4 and Nanog (Fig S1A) was analyzed using Human Pluripotent Stem Cells Transcription Factor Analysis Kit according to the manufacturer's instructions (BD).
Substance treatment and viability assay mESCs were exposed from day one of neural differentiation onward to sub-toxic concentrations of the sodium valproate (VPA, Sigma Aldrich), sodium arsenite (As-III, Sigma Aldrich) or trichostatin (TSA, Sigma Aldrich). VPA stock was prepared in PBS, while arsenite and TSA stocks were prepared in water, thus PBS/water controls were included as solvent/untreated controls. The cell culture medium, containing appropriate concentrations of test substances, was exchanged every second day. Cells were harvest on day 0, 5, 7, 9, 12, and 16 of neural differentiation. To estimate the sub-toxic concentrations (IC 10 -C 20 ) of test chemicals (VPA, arsenite and TSA) applicable for further gene expression analyses, cell viability was measured using the resazurin dye reduction assay (CellTiterBlue assay, Promega, Fig. 1A and Fig.  S4A and B). The effective concentrations that interfere with neural differentiation were estimated using flow cytometry with the neuron-specific marker bIII-tubulin. For microarray analyses cells were treated with 300 mM VPA (IC 10 ) or 0.75 mM (IC 10 ) arsenite, respectively. For RT-PCR analyses cells were exposed to 200, 300 or 400 mM VPA and to 0.75, 1 or 2 mM TSA, respectively.
Mouse primary embryonic cortical neurons were exposed to higher VPA concentrations than differentiating cultures, since they were found to be less sensitive to the toxicant (observation based on cell viability (resazurin dye reduction) assay, Fig. S4C). Four hours after isolation cells were treated with 0.8 mM, 1 mM and 2 mM VPA for 10 days in vitro.

RNA isolation and reverse transcription
Total RNA was purified from cells using TRIZOL reagent (Invitrogen) from three to four independent differentiation experiments. Isolated RNA was treated with RNAse-free DNAse I (Roche) prior to reverse transcription. For reverse transcription of total RNA, 1.5 mg RNA was reverse-transcribed into singlestranded cDNA using random primers and high capacity cDNA reverse transcription kit (Applied Biosystems) according to manufacturer's instructions. Short stem-loop cDNA libraries from individual miRNAs were generated using TaqMan microRNA assay and TaqMan miRNA reverse transcription kits (Applied Biosystems). For simultaneous detection of multiple miRNAs up to six miRNA-specific stem-loop RT-primers were multiplexed in one reverse transcription reaction. Briefly, In 30 ml reaction 30 ng of total RNA were mixed with 2 ml of each miRNA-specific primer, 3 ml 106 RT Puffer, 0.6 ml dNTP mix (100 mM total), 0.38 ml RNAse inhibitor (20 U/ml) and 2 ml Multiscribe reverse transcriptase (50 U/ml). One ml of cDNA was used for further realtime PCR. cDNA synthesis was performed at 42uC for 30 min.

miRNA expression microarray profiling
Sample preparation, microarray hybridization and scanning were performed at MFT Services (Faculty of Medicine, University of Tübingen). Briefly, one microgram of RNA, isolated on day 16 of neural differentiation from cells treated either with solvent, VPA or arsenite, was labeled with FlashTag Biotin RNA labeling kit (Genisphere, Hatfield, PA, USA) for Affymetrix GeneChip miRNA 2.0 arrays (Affymetrix, Santa Clara, CA, USA) according to the manufacturer's recommendations. A simple colorimetric Enzyme Linked Oligosorbent Assay (ELOSA) was used to confirm successful biotin labeling. After labeling, the samples were hybridized on Affymetrix GeneChip miRNA 2.0 arrays, washed, stained, and scanned according to manufacturer's instructions (Affymetrix). The data files (.CEL) were imported to Partek 6.5 software (www.partek.com). The RMA method was used to perform background adjustment, quantile normalization and summarization of the log-expression values for each gene on each array. We performed a differential expression analysis using tstatistics methods implemented in the Partek software. Ingenutity (IPA) software (Ingenuity Systems, www.ingenuity.com) was used to perform miRNA target genes filtering and to identify potential gene networks, pathways and biological functions in which VPA sensitive miRNAs could be involved.

Whole-genome expression microarray profiling
Total RNA was purified from either VPA-treated or untreated mESCs on day 16 of neural differentiation. Their expression profiles were analyzed with the GeneChip Mouse Gene 1.0 ST Array (Affymetrix). The amount and integrity of purified RNA were estimated by microcapillary electrophoresis (2100 Bioanalyzer, Agilent Technologies), whereas purity was assessed by the A260/A280 ratio.
Amplified and biotinylated sense strand DNA was generated from total RNA with the Affymetrix whole transcript sense target labeling assay. Labeled, single-stranded DNA (5.5 mg) was combined with hybridization and spike controls and hybridized with the pre-equilibrated Affymetrix GeneChip Mouse Gene 1.0 ST Array for 16-18 h. Following hybridization, arrays were washed and stained with a streptavidin phycoerythrin conjugate using an automated GeneChip Fluidics Station 450. They were then scanned with a GeneChip Scanner 3000 using a 570-nm excitation wavelength laser. Sample preparation, microarray hybridization and scanning were performed at MFT Services (Faculty of Medicine, University of Tübingen). Microarray quality control assessment and data acquisition were performed with the GeneChip Operating Software (Affymetrix). Normalization among the different microarray data files was performed by robust multi-array analysis (RMA) inside the Partek 6.5 software. Unsupervised hierarchical clustering analysis and a principal component analysis were performed with the Partek software to establish non-forced groups of samples. Differential expression was determined by using t-statistics. False discovery rates were determined following the Benjamini-Hochberg procedure [45]. Data were expressed as mean 6 S.D. Student's t-test was done to determine the statistical significance of the pairwise comparisons.

Gene network analysis
Upregulated and downregulated miRNA together with their reciprocal expressed mRNA targets were further analyzed using IPA software (Ingenuity Systems, www.ingenuity.com). A data set containing gene identifiers and corresponding expression values was uploaded into the application. Each identifier was mapped to its corresponding object in the Ingenuity Knowledge Base. A $2fold cutoff was set to identify molecules whose expression was significantly differentially regulated. These molecules, called network eligible molecules, were overlaid onto a global molecular network developed from information contained in the Ingenuity Knowledge Base. Molecular networks were then algorithmically generated based on their connectivity. The functional analysis identified the biological functions and/or diseases that were most significant affected by the VPA treatment. Right tailed Fisher's exact test was used to calculate a p-value determining the probability that each biological function and/or disease assigned to that data set is due to chance alone. Canonical pathways analysis identified the pathways from the IPA library of canonical pathways that were most significant altered by the VPA treatment.
The significance of the association between the data set and the canonical pathway was measured in 2 ways: 1) A ratio of the number of molecules from the data set that map to the pathway divided by the total number of molecules that map to the canonical pathway is displayed; 2) Fisher's exact test was used to calculate a p-value determining the probability that the association between the genes in the dataset and the canonical pathway is explained by chance alone.

Quantitative RT-PCR
Quantitative real-time PCR (RT-PCR) on miRNAs was performed using the TaqMan microRNA assay kit in combination with TaqMan universal PCR master mix, no AmpErase UNG (Applied Biosystems) according manufacturer's instructions. Expression of individual miRNAs was normalized to sno-202 expression. Expression of primary miRNA transcripts or mRNAs was quantified using SYBR Green reagents (Applied Biosystems). Primers for SYBR Green probes were designed using the NCBI Primer-BLAST software (www.ncbi.nlm.nih.gov/tools/primerblast/) and are listed in Table S1. 18S ribosomal RNA was used as endogenous control gene. All RT-PCRs were performed in triplicates on Applied Biosystems 7500 Real-Time PCR System with following thermal cycling parameters: miRNA RT-PCR (95uC for 2 min, followed by 40 cycles of 15 s at 95uC and 1 min 60uC); SYBR Green RT-PCR (50uC for 2 min, 95uC for 2 min, followed by 40 cycles of 15 s at 95uC and 1 min 60uC); Melt curve stage was included in SYBR Green reaction (95uC for 15 s, 60uC for 1 min, 95uC for 30 s and 60uC for 15 s). The relative levels of RNA expression in treated samples in comparison to untreated controls were quantified using the comparative CT (2 2DDCT ) method [46]. Data collected from three or four independent differentiation experiments are reported as average log 2 -fold change of independent biological replicates 6 SEM. Differences in treated and untreated samples were analyzed for statistical significance using Student's t-test. P-Value,0.05 was denoted on the graphs by *, p,0.01 by **, and p,0.001 by ***.

Results
Valproate effects on cell viability and neural differentiation of mESC mESC line W4 were differentiated into neurons according to an established in-house protocol [11]. The undifferentiated W4 cells showed a typical ESC morphology and expressed alkaline phosphatase as well as Oct3/4 and Nanog (Fig. S1A). Neural differentiation efficiency was monitored using neural-specific markers by western blot (Fig. S1B), flow cytometry (Fig. S1C) and immunocytochemistry (Fig. S1D). Furthermore, expression of neural-specific/enriched miRNAs, neuronal marker (b-III-tubulin) and neuro-progenitor marker (nestin) during differentiation was analyzed by qRT-PCR. All neural miRNAs as well as neuronal marker b-III-tubulin were strongly induced upon induction of differentiation ( Fig. S1E and F). The VPA-induced alterations in the neural differentiation process were monitored after exposure to a range of different valproate concentrations for 16 days. To estimate the sub-toxic concentrations of this compound for subsequent gene expression analyses, we determined the effects of VPA on cell viability and neural differentiation using the resazurin reduction assay (CellTiterBlue) and flow cytometry applying the neuron-specific marker b-III-tubulin, respectively. VPA affected cell viability in a concentration-dependent manner (Fig. 1A). The substance concentration that reduced cell viability to 90% (effective concentration 10%, EC 10 ) was 314 mM. Expression of bIII-tubulin was reduced up to 65% of control by VPA (Fig. 1B).

VPA effects on miRNome of neural-differentiated ESC
As a next step, neurally differentiating mESCs were used to determine the effects of valproate on miRNA expression. In these experiments cells were also treated with a structurally and mechanistically unrelated developmental neurotoxicant and epigenetically active substance, sodium arsenite, in order to investigate substance-specific effects on the miRNA expression profile. To monitor effects of VPA and arsenite on the miRNome, mESCs were differentiated into neurons for 16 days under exposure to sub-toxic concentrations of either sodium valproate (300 mM) or sodium arsenite (0.75 mM). The 16-day period was chosen based on analysis of the expression patterns of neuralspecific miRNAs during neural differentiation of mESCs. mir-9 was strongly induced from day 9 of differentiation onwards, while mir-124 and mir-128 were induced later on reaching their maximum expression levels both on day 16 ( [47] and Fig. S1F). Brain enriched miRNAs such as mir-125b, let-7a,c, mir-138 and mir-181a were also strongly induced after 16 days of differentiation ( [47] and Fig. S1F). We analyzed the global changes in miRNA profiles in VPA-treated cells using an Affymetrix microarray. The normalized data of miRNA and whole genome microarrays are deposited in the GEOarchive (see: http://www.ncbi.nlm.nih.gov/ geo/query/acc.cgi?acc = GSE50215, http://www.ncbi.nlm.nih. gov/geo/query/acc.cgi?acc = GSE50216).
Differently expressed miRNAs are summarized in Fig. 2 and Table S2. The exposure to valproate during neural differentiation of mESCs led to significant alterations (fold change (FC)$|2|, p# 0.05) in the expression of 110 miRNAs ( Fig. 2A).
From 110 miRNAs that responded to VPA exposure, 74 were upregulated up to 100-fold, while 36 were 2-to 4-fold downregulated (Table S2). Six miRNAs were regulated on the level of primary transcripts (pri-mir s ) but not on the level of the mature 22-nt forms, as determined via the whole genome array. In contrast, arsenite significantly altered the expression only of 27 miRNAs (Fig. 2B and Table S2). Ten miRNAs were in intersection for both substances, but most of them were regulated in opposite direction (Table S3). From 27 arsenite-regulated miRNAs, expression of 15 miRNAs was significantly increased (up to 4-fold), and the expression of 12 miRNAs was decreased (up to 3-fold) (Table S2). Volcano plots presented in Figure 3 demonstrate substance-specific effects on miRNA expression. The plots represent log 2 ratios in miRNA expression in substance-treated cells in comparison to control (solvent-treated) cells plotted against 2log 10 of the p-value. Comparing to the solvent control, in cells treated with VPA we observed a strong upregulation of myogenic miRNAs (myo-mirs: mir-206, mir-133a,b), or miRNAs shown to be involved in muscle differentiation and specification (mir-10a, mir-143/mir-145 cluster, mir-214, mir-322, mir-199a). By contrast, we could not observe any induction of muscle-specific miRNAs upon exposure of cells to arsenite. Moreover, mir-214, mir-145 and mir-199a were significantly downregulated in these cells. The myogenic mir-206 was the most strongly induced miRNA in cells treated with VPA (up to 100-fold increase in expression). However, mir-133b that clusters with mir-206 in a bicistronic loci, was induced by VPA only up to 2.2-fold arguing that the induction of mir-206 was at least not solely due to a change in chromatin structure.
miRNAs involved in embryonic and adult neurogenesis such as mir-137, mir-128, mir-124a, mir-326, or mir-7 were found significantly downregulated by VPA. No effects of arsenite were observed on the expression of these miRNAs. All of these repressed neural miRNAs were shown to inhibit neural stem cell and neural progenitor cell proliferation and stimulate neuronal differentiation and specification (cf. section Discussion [16]).
Known tumor-suppressor miRNAs, such as mir-34a and c, were strongly upregulated by arsenite and only slightly downregulated in cells treated with VPA. Further, mir-491, a miRNA which is involved in neurosteroidogenesis and pathogenesis of multiple sclerosis [48], was found strongly upregulated by arsenite and downregulated by VPA, and mir-383, a miRNA expressed in the reproductive system and a negative regulator of proliferation, was highly expressed in control samples but significantly downregulated by both substances (4.2-and 2-fold by VPA and arsenite, respectively).
Altogether our data demonstrate substance-specific effects on miRNA expression profiles. In particular, long term exposure to valproate, but not arsenite, during neural differentiation of mESCs led to the repression of neural-and the induction of muscle-specific miRNAs.

VPA effects on protein coding gene expression in neuraldifferentiated mESCs
To support the miRNomics data in cells treated with VPA we performed whole genome microarray analysis using the same RNA samples as for miRNA profiling. Differently expressed genes in cells upon exposure to VPA are summarized in Fig. 4 and Table  S4. The expression of 377 genes was significantly affected by VPA (FC.2.0, p,0.05). 267 genes were upregulated up to 7-fold, while 110 genes were 2-to 5-fold downregulated. At the mRNA level, we also could confirm the VPA-mediated induction of myogenesis in cells through the detection of upregulation of the respective genes Actc1 (most strongly induced by 7.6-fold), calponin, myosin light chain, asporin, and decorin. In addition, a number of Hox genes, including HoxB2, 3,4,5,7,8,9; HoxA2,3; HoxC4,5,6,8 and HoxD4,8, which are known to be involved in cell fate determination during embryonic development, the regulation of the anterior-posterior polarity, in particular, were strongly upregulated by VPA treatment. Finally, Twist1, a transcription factor involved in cell lineage determination and specification, skeletal and muscle system morphogenesis, and strongly expressed in mesodermal derived tissues, was 2.1-fold upregulated by VPA. Twist1 is also a predicted target of mir-137 and mir-363, both of which were repressed in their expression by VPA (cf. above).
In contrast, neural factors like Otx1 and Otx2, known regulators of forebrain development [49] as well as members of ZIC protein family (Zic3, 4, and 5), were downregulated in cells upon exposure to VPA. Zic family is also involved in neural differentiation, forebrain development, and neural tube closure [50]. Otx2 and Zic4 are predicted targets of mir-206, while Otx1 and Zic3 are predicted targets of mir-199a and the muscle-specific mir-133a, respectively. Thus, the expression of protein encoding genes strongly correlated with the obtained miRNA results, confirming the induction of mesodermal and the repression of neural marker genes as well as a reciprocal expression of mRNAs in respect to the corresponding inhibitory miRNAs.
Pathway and functional analysis of miRNA targets regulated by VPA We performed a network and functional analysis of differently expressed genes using Ingenuity (IPA) software. As a first step of data analysis, we comprised a date-set by applying an IPA miRNA target filter. From 377 VPA-sensitive genes, 186 were found to be predicted or experimentally validated targets of 48 VPA-regulated    (Table S5). As an increasing number of studies demonstrate not only a post-transcriptional regulation by miRNAs but also the reduction in the expression of target mRNAs, for subsequent biofunctional analysis using IPA we only included miRNA targets, which exert reciprocal expression patterns to corresponding miRNAs (cf. material and methods). The functional and network analysis of VPA-sensitive miRNA and their targets demonstrated that they are involved in crucial cell signaling pathways and bioprocesses. The most relevant biological and disease processes as well as cell signaling and metabolic pathways are summarized in Table 1. Canonical pathway analysis identified the pathways from the IPA library that were most significant to VPA-sensitive miRNA and target genes.
Top molecular interactions of VPA regulated miRNAs and their targets are presented in Fig. S2. Molecules connected by the first network (A) were shown to be involved in cancer, cardiovascular, and reproductive diseases. The second network (B) combined molecules, which are involved in skeletal and muscle system development and functions, as well as embryonic and organism development that correlate to the upregulation of muscle specific miRNAs and genes. Further analysis of networks between miRNAs and their reciprocally expressed targets and the experimental validation of miRNA-mRNA regulations may further contribute to the understanding of pathways of VPA toxicity.

Real-time RT-PCR validation of VPA-responsive miRNAs
We validated the microarray data using real-time RT-PCR. Taqman microRNA assay (ABI) was used to analyze the expression of a set of muscle-and neuro-specific miRNAs 16 days after induction of neural differentiation under VPA exposure. In four independent differentiation procedures we could confirm the microarray data (Fig. 5A)-that is, a strong concentrationdependent induction of muscle-specific/abundant miRNA (mir-206, mir-10a, mir-214, mir-145, mir-143, mir-199a) and a significant downregulation of the expression of neuro-specific miRNAs (mir-124, mir-128, mir-137, mir-491, mir-383) in comparison to the solvent control. We did not observe any changes in the expression of neuro-specific mir-9 in our microarray data, so this miRNA was included in RT-PCR analysis as a negative control. In addition, we analyzed the expression of the same pool of miRNAs in primary cortical neurons, isolated on day 17 of embryonic development from mouse fetus and cultivated for further 10 days under VPA exposure. Only mir-10a was significantly affected by VPA in primary cortical neurons, suggesting that this miRNA could be a primary target of VPA (Fig. S3). We could not observe any induction of mir-206, or a significant reduction of neuronspecific miRNAs. Taking into consideration that miRNA expression can be regulated on the level of processing [53], we analyzed the expression of primary transcripts of selected miRNAs as well (Fig. 5B). Both, pri-mir-10 and pri-mir-206 were significantly upregulated upon VPA treatment, suggesting that VPA affects the transcription and not the processing of these two miRNAs. Taking into account the known mechanism of VPA as a HDAC inhibitor we can suggest possible activation of mir-10a and mir-206 genes by VPA through the inhibition of HDAC and, as a result, acetylation/activation of histones in the miRNA promoter regions. This would be in line with the induction of mir-10a expression in primary neurons (Fig. S3). From two primary mir-128 transcripts only pri-mir128-1 was detectable and slightly downregulated in our cell system. From three known primary mir-124 transcripts pri-mir124-1 was most strongly upregulated 16 days after induction of neural differentiation and was significantly downregulated by VPA. The reduction in the expression of pri-mir-124-2 and 3 was not statistically significant. As expected, the substance did not affect the expression of pri-mir-9.

Real-time RT-PCR validation of VPA-responsive mRNAs: Time kinetics of myogenesis and miRNA involved in myogenesis
In order to verify whole genome microarray data by real-time RT-PCR, we selected genes known to be involved in the development and neural tube closure (Hox genes, Twist1, Hamga2, Pax6, Otx1, Otx2, Zic4 and Zic5, Fzd4, Dkk2, Vcl), as well as musclespecific Actc1. We could confirm the microarray data by RT-PCR for all selected genes (Fig. 6A). The most upregulated mRNA in microarray experiments was Actc1 that encodes a-actin, a major component of muscle tissue (Fig. 6B).
Next we addressed the effect of VPA on several myogenic regulatory factors (MRF) known to control muscle cell differentiation, in particular Mef2A, Myf5, MoyD, myogenin as well as the paired box (Pax) transcription factors Pax3 and Pax7 that are known to trigger myogenesis through interplay with each other and with myo-mirs (reviewed in [54,55]). In the first step we were able to demonstrate a concentration-dependent induction in the expression of Myf5, myogenin, and MyoD, but not Mef2A and Pax3 in VPA-treated samples on day 16 of neural differentiation. Pax7 expression was significantly reduced (Fig. 6B). The expression of Hdac4, a validated mir-206 target, remained unchanged, suggesting that the regulation of this transcript by mir-206 can probably occur at the post-transcriptional level and does not affect mRNA expression. To understand which transcription factor or miRNA expression was affected first by VPA treatment, and thus possibly contributing to the stimulation of myogenic differentiation, we analyzed the time kinetics of the expression of myo-genes together with crucial miRNAs (mir-206, mir-10a) and marker for differentiated muscle tissue (Actc1) in the course of differentiation. The expression of these genes was analyzed in undifferentiated mESCs as well as on days 5, 7 9, 12 and 16 after induction of neural differentiation under exposure to 300 mM VPA. The gene expression at each time point was normalized to the expression in undifferentiated cells (Fig. 7). The expression of three-out-offour tested MRFs (Myf5, MyoD, myogenin) was affected by VPA, although their expression during differentiation was only slightly above the RT-PCR detection limit. Myogenin was induced up to 5 times on day 9 in both untreated and VPA-treated cells, then remained unchanged in untreated cells but was further and significantly (p,0.05) induced up to 65-fold in VPA-treated samples. Myf5 was not induced in untreated cells and significantly increased (p,0.05) from the beginning of day 12 of differentiation in VPA-treated cells, similar to Actc1 kinetics. MyoD has shown a biphasic expression profile during differentiation under VPA exposure. It was slightly induced on day 5 of differentiation (5-fold in VPA-treated samples and 2-fold in untreated cells), then downregulated on days 7 and 9, and then significantly upregulated (p,0.05) again on days 12 and 16, but only in VPA-treated samples. The induction of MyoD in early stages of differentiation by VPA could be explained based on the finding that proliferating myoblasts express MyoD [54]. The expression of Myf2A was induced in the course of neural differentiation beginning from day 7 in both untreated and VPA-treated samples in comparison to undifferentiated mESCs. Altogether, MRFs were significantly affected by VPA as early as day 12 of neural differentiation (Fig. 7A). Transcription factors Pax 3 and Pax 7, which are known to be regulated during early stages of neural and muscle differentiation (including neural tube closure) were strongly induced in the course of neural differentiation. VPA treatment did not affect expression of Pax 3 but significantly reduced (p, 0.01) the expression of Pax 7 as early as on day 16 of differentiation. Neither mir-206 nor mir-10a could be detected in undifferentiated mESCs. As expected, we observed a significant induction (p,0.05) of mir-206 expression in VPA-treated cells as early as on day 12 of differentiation, while mir-10a was strongly induced in VPA-treated but not in control samples from day 5 of differentiation, suggesting that this miRNA could be the first signal stimulating myogenesis under VPA exposure during neural differentiation of mESCs (Fig. 7B).

Morphological confirmation of the induction of myogenesis by VPA during neural differentiation of mESCs
We performed immunocytochemistry staining with an antibody against the muscle-specific marker a-sarcomeric actinin to confirm the gene expression data morphologically. We observed a 10-fold increase of a-actinin-positive cells in VPA-treated samples in comparison to the negative control (n = 3, Fig. 8, Fig. S5). Several clusters containing hundreds of a-actinin-positive cells per slide were observed in VPA treated samples, while only few sporadically distributed single a-actinin-positive cells were observed in PBS treated samples (40 cells per slide in average). No increases in aactinin-positive cells were detected in arsenite-treated samples in comparison to solvent control (data not shown). The staining with the neuron-specific marker bIII-tubulin was performed in parallel.

Induction of myogenic miRNAs expression during neural differentiation of mESCs is a common phenomenon for HDAC inhibitors
To examine whether the effects observed in VPA-treated samples are common for HDAC inhibitors, we induced the neural differentiation of mESCs under exposure to sub-toxic concentrations (0.75, 1 and 2 mM, Fig. S4B) of trichostatin A (TSA), a structurally unrelated HDAC inhibitor.
As expected, we observed the induction of mir-206 and mir-10a expression upon TSA treatment. Mir-214, mir-199a and mir-145 were also induced but not to such an extent as seen for VPA. Mir-128 and mir-137 were downregulated, while mir-124 was not. A slight induction of mir-9 was observed at the highest concentrations of TSA (Fig. 9). The results suggest that the induction of myogenic miRNA expression by VPA might be at least partially mediated by HDAC inhibition.

Discussion
In this study we analyzed perturbations in the miRNome in neural-differentiated mESCs exposed to the developmental neurotoxicant VPA to establish this endpoint as a suitable tool for the prediction of developmental neurotoxicity in vitro and to elucidate the molecular mechanisms of VPA-dependent developmental neurotoxicity. Using microarray technology we could demonstrate substance-specific effects on miRNA expression. VPA, but not arsenite, induced the expression of muscle-specific miRNAs during neural differentiation of mESCs. By contrast, the expression of a panel of neural-specific/abundant miRNAs was found repressed by VPA (Fig. 3, 5). VPA-specific miRNA expression patterns were further confirmed at the level of mRNA by demonstrating the induction of muscle-specific genes and the accompanied inhibition of genes involved in neurogenesis (Fig. 4,  6, 7), and phenotypically by immunofluorescent staining (Fig. 8). Supporting functional analysis of VPA-specific miRNA targets performed by IPA pathway analysis revealed that VPA might affect muscle and skeletal system development, cellular and tissue development, cellular growth, proliferation, and apoptosis (Table 1). In addition, we observed similar effects on musclespecific miRNAs with the structurally unrelated but well characterized HDAC inhibitor TSA, suggesting that the VPA effect is at least partially dependent on its known HDAC inhibitor activity (Fig. 9). In line with these observations, Gurvich and coworkers have analyzed alterations in the gene expression pattern of zebrafish and Xenopus embryos upon treatment with VPA or TSA and have shown that both HDAC inhibitors exert common effects at the gene expression level in these two different vertebrate model systems [56]. One of the greatest advantages of using miRNA profiling as a tool for toxicity testing is their conservation throughout the animal kingdom/the phyla. For instance, the first discovered C. elegans miRNAs, let-7 and lin-4 (mir-125 for mammals), their targets and functions, have been proven to be conserved in regulating developmental timing, stem cell differentiation and neural development in worms, flies and mammals [57][58][59][60]. Most regulated miRNAs shown in our study are highly conserved between mice and humans (e.g. mir-206, mir-214, mir-10a, mir-124, mir-137, mir-128, mir-9) [61,62]. Myogenesis regulating mir-206 is highly expressed in skeletal muscles in both species [62,63] as is mir-124, mir-9, mir-128 and mir-137 in mouse and human brain where they are responsible for fine-tuning of neurogenesis [62] [64]. Importantly, clustering of some miRNA genes within the genome is also conserved between human and mice (e.g. mir-206/mir-133b, mir-214/mir199a, mir-10a/HoxB4, mir-145/143 clusters, all of which have been studied here) (www. miRbase.org [61]). Hence, miRNA profiling may help to overcome interspecies differences what is certainly one of the obstacles in using non-human cell lines to predict human toxicity. Nevertheless, similar experiments using human embryonic stem cells or induced pluripotent stem cells (iPSC) will answer the question whether the effects will be applicable for human cells and human neural development in general.
The mir-206, a skeletal muscle-specific miRNA, was the most strongly upregulated miRNA in our experimental system (up to 100-fold after exposure, Fig. 3, 5 and Table S2). Expression of mir-206 was also found strongly induced during differentiation of C2C12 myoblasts [65,66]. This miRNA promotes skeletal myogenesis by translational but not transcriptional repression of multiple targets including Hdac4, a key regulator of myogenesis [66]. In agreement with these observations, we did not observe any VPA-induced changes in Hdac4 transcription levels during the differentiation of mESCs. In addition, mir-206 regulates the developmental timing of muscle differentiation by targeting early markers of myogenesis, such as Pax3 and Pax7 [67,68]. Pax7 is highly expressed in proliferating myoblasts, where it represses the expression of myogenic factors like MyoD. Thus, mir-206 can stimulate the expression of MyoD by targeting and translational repression of Pax7. On the other hand, mir-206 itself is transcriptionally regulated by MyoD, meaning that both molecules reciprocally regulate each other while driving myogenesis (reviewed in [55]). These changes in the differentiation process of muscle tissue also correlate with the observed transcriptional downregulation of Pax7 and the transcriptional induction of MyoD by VPA (Fig. 6B). Interestingly, mir-206 is also known as key regulator of the signaling between motor neurons and skeletal muscles at neuromuscular junctions and is required for proper formation and/or regeneration of neuromuscular synapses after injury by targeting Hdac4 and Connexin43 [69,70].
Mir-10a and 10b are both highly evolutionarily conserved and localized within Hox gene clusters. Hox genes are conserved transcription factors, regulating the anterior-posterior pattern formation during development and have been implicated in neural tube closure [71]. Mir-10a/b as well as a number of Hox genes (Table S4) were strongly induced during neural differentiation under VPA exposure in our mESC system (23.9/4.3-fold, Fig. 3, 5 and Table S2). Both miRNAs are upregulated in response to retinoic acid, a common inducer of cellular differentiation in different cell types and mir-10a/b were shown to contribute to retinoic acid-induced differentiation of neuroblastoma cells [72,73]. Huang and co-authors were able to demonstrate that mir-10a expression is strongly increased upon retinoic acid-induced muscle differentiation of ESCs, where it determines the smooth muscle lineage by targeting Hdac4 [72]. The finding that mir-10a is the earliest upregulated gene, which, in contrast to myogenic factors or mir-206, already displayed a significant increase in the expression level on day 5 of differentiation in VPA-treated mESCs, suggests that it plays a central role in the stimulation of myogenesis during VPA-mediated disturbance of neural differentiation. In addition, only mir-10a was found upregulated after VPA treatment of primary cortical neurons (Fig. S3), thus suggesting that the expression of mir-10a might be directly affected by VPA even at later stages of neural development. On the contrary, mir-206 was not upregulated in primary cultures, suggesting that the strong upregulation observed during neural differentiation of mESCs is (i) a result of myogenic lineage specification triggered by exposure to VPA at earlier stages of differentiation process, and (ii) that VPA is unable to directly stimulate the expression of mir-206 in already committed cells. For the same reason we did not observe any differences in the expression of neuron-specific miRNA in primary cultures (mir-124, mir-128, Fig. S3), which are known to play a significant role in earlier, but not later stages, of neurogenesis. In addition, the fact that about 40% of all known miRNAs could be affected by HDAC inhibitors while the percentage of protein encoding genes rather remains low [38] reinforces our hypothesis that mir-10a could be a first and direct target of VPA. In the future studies chromatin immunoprecipitation (CHIP) analysis of the mir-10 promoter region may be helpful to clarify whether VPA induces the acetylation/activation of histone complexes in the regulatory regions of the mir-10a locus.
We also observed a VPA-dependent induction of myogenic regulatory factors (MyoD, Myf5, and myogenin, Fig. 6 and 7) as well as an increased number of a-actinin-positive myocytes in neuraldifferentiated mESCs (Fig. 8). These results further support the findings on the miRNA level. It was previously shown that HDAC inhibitors, including VPA, enhance differentiation of myoblasts by regulation of MRFs [74,75]. This, together with the observation that ectopic expression of myogenic factors (such as Pax3, MoyD, Myf5 and myogenin) in neural tubes of chicken embryos activates muscle-specific gene expression and myogenesis in neural tissue [76] might explain the upregulation of myo-mirs and MRFs in our system and, as consequence, an activation of myogenesis by VPA during neural differentiation of mESCs.
Mir-214/mir-214* and mir-199a-5p/mir199a-3p are clustered on pri-mir-199a-2 within the Dnm3os gene in mouse as well as in human genome and were strongly upregulated by VPA (8.7/20.6and 9.8/11.1-fold, respectively), while exposure to arsenite significantly reduced the expression of these miRNAs (22.4/2 1.97-and 22.4/22.9-fold, respectively). The mir-214/mir-199a cluster is regulated during embryonic development by transcription factor Twist1 [77,78], the expression of which was significantly induced by VPA in our cell system (2.2-fold). Twist1 is implicated in the differentiation of multiple cell lineages including muscle formation, and regulates maturation and differentiation of human ovarian cancer stem cells through induction of the aforementioned miRNAs and subsequent inhibition of Pten and Ikkb by those miRNAs that lead to inhibition of NF-kB and activation of Akt pathways [77]. In addition, Twist1 plays a significant role in mesoderm formation and epithelial mesenchymal transition, which is a hallmark of gastrulation, and particularly in neural crest specification and migration [79,80]. Mir-214 was shown to specify muscle cell fate in the embryonic development of zebrafish by modulation of Hedgehog signaling during early segmentation stages [81]. Mir-214 is expressed in C2C12 myoblasts and strongly induced in the course of differentiation where it regulates both proliferation and differentiation depending on the culturing conditions, probably by targeting negative regulators of MRFs [82]. Furthermore, the histone methyltransferase Ezh2, a polycomb group protein, as well as N-Ras, both of which known repressors of skeletal myogenesis, were shown to be negatively regulated by mir-214 during muscle differentiation [83,84]. In situ hybridization of regulated miRNA may contribute to the understanding of the role of these miRNA in differentiated cells and clarify whether the expression of mir-214 is restricted to muscle cells or induced in neural cells as well.
VPA treatment of mESCs during neural differentiation led to the downregulation of some crucial neuronal miRNAs. The most VPA-sensitive miRNA was mir-137 (3.8-fold downregulated in comparison to control). Mir-137 is required for neural differentiation of ESCs [85] and modulates differentiation of adult mouse neural stem cells as well as brain tumor stem cells by crosstalk with methylation agents [86,87].
Mir-9 and mir-124, the two most abundant miRNAs in the CNS, showed different responses to VPA treatment: while mir-124 was downregulated, no effect on mir-9 expression could be observed after VPA exposure. Both were shown to promote neuronal lineage commitment by targeting multiple anti-neuronal factors in neuronal cells (reviewed in [16]). These results suggest that VPA affects neural differentiation processes and pathways, which are specific for mir-124 but not mir-9. Exposure of differentiating mESCs to VPA also reduced expression of mir-128, which is an enhancer of neural differentiation [88]. Interestingly, a negative regulator of dendrite outgrowth and maintenance, mir-375 [89], was significantly upregulated (2-fold) by VPA in neural-differentiated mESCs.
The effect of VPA on transcriptome during neural differentiation was extensively studied in hESC ( [90][91][92][93]), mESC systems [24,94] and P19 mouse embryonic carcinoma cells ( [51,52]). However, these studies did not address the question whether VPA may perturb miRNA expression during neural differentiation. In agreement with aforementioned studies we observed perturbations in expression of the genes associated with neural tube defects, HDAC inhibition pathways, genes involved in embryonic development especially neural development. Neither of these studies reported induction of myogenesis or lineage shift that can be explained by differences in exposure time: our cell system was exposed throughout the whole differentiation process, beginning from the first day of differentiation, while Theunissen et al. and Jergil et al. performed transcriptomics analysis after 24 hours of exposure (at day 11 of differentiation or in undifferentiated embryonic stem cells, respectively). Leist and colleagues analyzed perturbations in gene expression by VPA at earlier stages of differentiation (stage of neural ectodermal progenitor cells, four and six days after induction of differentiation). Overall, performing short-term exposure allows for identification of genes involved in the initial cellular response to the toxicant, while changes in the differentiation process such as lineage shifts rather will be measured upon long-term exposures covering the entire differentiation process.

Conclusions
The findings presented here may contribute to a better understanding of neural tube defects induced by VPA. As the data shows, one of the main advantages of miRNome against whole transcriptome profiling is the amount of data: there are 2578 mature human and 1908 mouse miRNAs identified so far (http://www.mirbase.org) [95]). Therefore, miRNA profiling provides a more compact data set that might facilitate the analysis of a specific substance's effects to greater extent. In addition, miRNA profiling may have stronger predictively than whole genome profiling, as miRNAs were shown to be crucial development regulators, influencing developmental timing as well as cell specification (reviewed in [15]). An explosive number of studies analyzing miRNA functions in numerous cellular processes, especially neural development, open a new dimension for analysis of pathways of toxicity, where miRNAs might be involved. Thus, miRNA profiling may serve as an accomplishment to the transcriptomics and as a good molecular marker for developmental neurotoxicity testing and for discovery of new pathways of toxicity. Based on miRNA profiling we could demonstrate an unexpected and previously unknown adverse activity of VPA on neural differentiation of mESCs inducing a shift in lineage determination from neurogenesis into myogenesis. The mechanisms underlying these observations are unclear. However, it is tempting to speculate that some of the aforementioned miRNAs could be primary targets of HDAC inhibition that at least partially may trigger myogenesis in our system. In general, analysis of miRNA expression profiles after VPA exposure provided the first glance of certain miRNAs targets that might be involved in mediating the developmental neurotoxicity of this substance. Gates (x-axis) were gated based on negative controls lacking first antibody. D. The neuronal morphology was visualized on day 9, 12, and 16 of neural differentiation by immunostaining with Nestin, MAP2 and b-III Tubulin. Presence of astroglia cells was demonstrated by the staining with antibody against GFAP. E. Expression of neural specific or enriched miRNAs was strongly induced on day 16 of neural differentiation. The graph represents mean of absolute miRNA expression levels normalized to undifferentiated ES cells measured in four independent RT-PCR experiments. b-III-Tubulin expression was analyzed in parallel as a positive control of neural differentiation. F. Expression of neuronal specific markers mir-9, mir-124 and b-III-tubulin and neuro-progenitor specific marker nestin at different time points of neural differentiation. The expression of genes at each time point of differentiation is normalized to the expression in undifferentiated mES cells. The graph demonstrates mean of log 2 fold change in three independent biological replicates (differentiated vs. undifferentiated) 6 SEM. (TIF) Figure S2 Molecular networks of VPA-sensitive miRNA and their reciprocal expressed target mRNA. The IPAgenerated network is a graphical representation of the molecular relationships between molecules. Molecules are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). All edges are supported by at least one reference from the literature or from canonical information stored in the Ingenuity Knowledge Base. The intensity of the node color indicates the degree of up-(red) or down-(green) regulation. Nodes are displayed using various shapes that represent the functional class of the gene product. (TIF) Figure S3 miRNA expression in mouse primary cortical neurons exposed to VPA for 10 days in vitro. No changes could be observed in expression of neural specific miRNAs between VPA and solvent control. mir-206 was not detected in primary cultures, while mir-10a was strongly induced upon VPA treatment. The graph demonstrates mean of log 2 fold change (VPA vs. solvent control) in two independent biological replicates 6 SEM. (TIF) Figure S4 Toxicant (VPA, arsenite and TSA) effects on cell viability. A. and B. The W4 ES cells were induced to differentiate into neurons for 16 days under continuous substance (arsenite (A) and TSA (B)) exposure. C. Primary neurons were exposed to VPA from day 1 until day 10 in vitro. Cell viability was estimated using CellTiterBlue assay and is shown as a percentage of solvent control.