The powder form of fipronil (purity 98%) was purchased from ChemService (West Chester, PA) and was used to prepare a series of stock solutions. All stock solutions were prepared with HPLC grade methanol (Thermo Fisher Scientific, Waltham, MA). First, a 10 000000 μg L⁻¹ stock solution was made, and then diluted in a step-wise manner (1:10 serial dilutions).

Two sets of experimental solutions were prepared; the first set of solutions covers wide range of concentrations at 1:10 serial dilutions (0.1, 1.0, 10.0, 100, and 1 000 μg L⁻¹) and the second set was made to test concentrations between 200 to 1 000 μg L⁻¹ with an increment of 200 (200, 400, 600, 800 and 1 000 μg L⁻¹). The experimental solutions were prepared by spiking stock solutions into reconstituted water: pH 8.0, alkalinity 80 mg L⁻¹ CaCO₃, hardness 100 mg L⁻¹ CaCO₃ (Horning and Weber, 1985). The final concentration of the methanol in all the experimental solutions was 0.01% (v/v ratio) as suggested by American Society of Testing and Materials (ASTM E1241-92). A 0.01% methanol in reconstituted water (v/v ratio) was used as a vehicle control. A small portion of freshly prepared experimental solution was collected in amber glass vials prior to the initiation of the exposure and stored at −80 °C for measurement of fipronil concentration by enzyme-linked immunosorbent assay (ELISA) as described in the next section (Vasylieva et al., 2015).

All the buffers were prepared with ultrapure deionized (DI) water and the compositions are listed as follows: phosphate-buffered saline (PBS, 10 mM, pH 7.5), PBST (PBS containing 0.05% Tween 20), coating buffer (14 mM Na₂CO₃, 35 mM NaHCO₃, pH 9.8), blocking buffer (1% BSA in PBST), and substrate buffer (0.1 M sodium citrate/acetate buffer, pH 5.5). Substrate solution contained 0.2 mL of 0.6% 3,3′,5,5′-Tetramethylbenzidine (in dimethyl sulfoxide, w/v), 0.05 mL of 1% H₂O₂ in 12.5 mL of substrate buffer. Stop solution was 2M H₂SO₄.

Plates were coated with 1 μg mL⁻¹ antigen (1-CON, Vasylieva et al. 2015) diluted in the coating buffer (100 μL per well). After incubation for 1 h at room temperature (RT), the solution was replaced with the blocking buffer (200 μL per well) and plates were incubated over night at 4 °C or for 1–4 h at room temperature (RT). Plates were then washed with PBST 3 times prior to sample loading. The standards in the assay buffer were prepared in glass vials and loaded on the coated plate in triplicate wells (50 μL per well). The experimental solutions in low concentrations (0.1–10 μg L⁻¹) were directly added to the wells pre-loaded with the assay buffer while samples with high fipronil concentrations (100–1 000 μg L⁻¹) were diluted with the assay buffer before addition to the plate (Vasylieva et al. 2015). An equal volume of anti-fipronil serum (#2268 diluted in PBS, 50 μL per well) was added at 1:4 000 dilution, giving final 1: 8 000 dilution in the plate. The plates were incubated for 1 h at RT and then washed 5 times with PBST. Goat anti-rabbit IgG-HRP conjugate was added at 100 μL per well in a 1:20,000 dilution as instructed by manufacturer (Abcam, Cambridge, MA). The plates were incubated for 1 h at RT and washed 5 times with PBST. Substrate solution was added (100 μL per well) and was left to develop color for 10 min. The reaction was stopped by addition of stop solution (50 μL per well) and absorbance was read at 450 nm. SigmaPlot ver. 11.0 software was used for curve fitting and data analysis to find IC₅₀ values and the equation of the standard curve. The assay was performed four separate times.

Medaka embryos were exposed to nine concentrations of fipronil over the entire embryonic period (the nominal concentrations: 0.1, 1.0, 10.0, 100, 200, 400, 600, 800, and 1 000 μg L⁻¹). A total of five fipronil exposures were run; three of the exposures occurred at concentrations of 0.1, 1.0, 10.0, 100, and 1 000 μg L⁻¹ while the other two were at 200, 400, 600, 800 and 1 000 μg L⁻¹. All the exposures were performed with a newly developed 96-well plate method (Truong et al. 2011). Traditional beaker or petri dish exposure methods generate a large volume of waste, accumulate costly disposal fees, require large and complex temperature control mechanisms (i.e. water baths), and expose embryos in large groups, making tracking of individual responses difficult. Exposure using the 96-well plate method eliminates all of these problems; temperature control is managed via a commonly-used environmental chamber, one well holds only 250 μL of experimental solution, and each developing embryo can be observed without disruption and tracked separately under a dissecting microscope. Medaka embryos (< 6 h post fertilization) were collected from the culture facility at the Aquatic Health Program, University of California, Davis. Immediately after collection, embryos were cleaned with DI water and sorted for health and viability, then placed in 15 mL of experimental solutions in 50 mL Pyrex beakers for batch exposure (50 eggs per beaker, one beaker per concentration). Medaka embryos were kept in the beakers until male and female embryos could be identified (~4 d post-fertilization). At 2 days post-exposure (dpe), a 50% (7.5 mL) water change was performed. At 4 dpe, embryos were separated by sex and moved into a 96-well plate (Costar 9016, Non-sterile, Polystyrene, Flat Bottom, Medium Binding). Embryos were not placed initially in 96-well plates both to allow time to determine sex of the embryos, and to remove unfertilized eggs. This also allowed us to ensure that equal numbers of each sex were exposed and to conclusively attribute hatching failures to experimental conditions. Each well contained 250 μL of fresh experimental solution and one embryo was placed in each well. A total of 32 or more eggs were used for the exposure per treatment. More than 50% of total well volume was changed every other day during the remainder of the experiment. The fish embryos in the 96-well plates were kept in an environmental chamber (Percival Scientific, Perry, IA, Model 136LLVL) at 25 °C on a 16:8 h light:dark cycle with light intensity of 808 Lux during the experiment. All embryos were observed daily, and signs of abnormal development, mortality, and hatching success were recorded. Hatched larvae were kept in the 96-well plates up to 24 h after hatch for assessing abnormal development as well as swimming ability, and then euthanized using tricaine methane sulfonate (Tricaine-S, Western Chemical, Inc., Ferndale, WA). Larval fish lying on their side at the bottom of the wells at 24 h post hatch were considered as “impaired swimming”. Hatchlings were not fed during the exposure. The fish embryos that failed to hatch were terminated at 14 dpe.

We analyzed the data for a sex specific effect of fipronil on hatching success, hatching time, and tail curvature (measured at hatch) using multi-model inference (Anderson et al., 2000; Burnham and Anderson, 2002). For statistical analysis, individual fish in each well were treated as replicates. For both the hatching success and tail curvature data, each model had binomial distributions of error because the response variable is a success/failure (i.e., the fish either hatched or it did not). Using the error distribution appropriate for the type of data (e.g., binomial distribution for success/failure data) provides accurate estimates of confidence intervals (Bolker et al. 2009). Models fit to the hatching success data included H~ (an intercept only model), H~C and H~C+S, where H is hatching success (1 is success, 0 is failure), C is the nominal concentration (μg L⁻¹), and S is a dummy variable for sex (1 is male, 0 is female). Models used to analyze the tail curvature data included TC~ (an intercept only model), TC~C and TC~C+S, where TC is a variable where ‘1’ is a curved tail and ‘0’ is no tail curvature, C is the nominal concentration of fipronil in μg L⁻¹, and S is a dummy variable for sex (1 is male, 0 is female). For the hatching time analysis we built models with Gaussian error distributions. Models included D~ (an intercept only model), D~C, D~C+S, D~C+S+C×S, where D is days to hatching, C is concentration (μg L⁻¹), and S is sex. We included the interaction model to determine whether fipronil influences hatching time differently for each sex, potentially indicating endocrine disrupting effects. Because exposures were run five times over the course of several months, we also tested for an effect of time for each of the responses by including a variable for ‘temporal block’ in the best model and comparing it to a model without the temporal block variable. In each case time was insignificant, so we did not include ‘time’ in the models and present the data from the five time points together. For the tail curvature analysis, more than half the larvae had the deformity at the highest concentration, allowing us to calculate an ED₅₀. We used the methodology described in Hammock et al. (2015) for the calculation.

In addition to the above analyses, we performed three one-way analysis of variances (ANOVAs) to determine the lowest-observed-effect concentration (LOEC) for hatching success, tail curvature, and hatching time. A logistic transformation was used on the hatching success and tail curvature data because the response variables are binary. Concentration was the only independent variable used for each ANOVA. Dunnett’s tests were conducted following all three ANOVAs to find the LOECs.

An additional exposure was performed to characterize gene expression patterns of embryos at the lowest concentration of fipronil. For RNA-seq, the lowest-observed effect concentration at which development of the tail curvature was observed (200 μg L⁻¹) was used since non-specific responses due to overdose fipronil exposure (e.g. cell necrosis) were our concern. The exposure was conducted as follows: Medaka embryos were collected as described above (2.3. Embryo exposures), followed by batch exposure in glass beakers (210 embryos in 50 mL of fipronil solution, 200 μg L⁻¹). The same number of embryos were also exposed to 0.01% methanol as a vehicle control. On 4 dpe, the embryos were separated by sex and then placed into a 96-well plate as described above (one embryo per well). The experimental solution (> 50%) was changed every other day. At the end of the exposure (7 dpe, Stage 37: Pericardial cavity formation stage, Kinoshita et al. 2009), all the embryos in each treatment for each sex were pooled together and then randomly divided into 3 groups (n = 10 per group) to obtain a sufficient amount of total RNA for library preparation and to minimize bias from individual variation. The sampling time was chosen to ensure that 1) all the fish were still at embryonic stage and 2) embryos were exposed to fipronil long enough to induce changes in gene expression patterns.

Only male embryos were used for the gene expression analysis in order to standardize test results. Total RNA was isolated using TRIzol Reagent by following the manufacturer’s instructions (Thermo Fisher Scientific). The quality of the total RNA was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA) available at the DNA Technologies Core Facility at UC Davis (http://dnatech.genomecenter.ucdavis.edu) and ensured that the quality and integrity of the total RNA samples were above the required criteria for library preparation (RNA Integrity Number > 8.0). No genomic DNA contamination was detected.

The library preparation and sequencing reaction for RNA-seq was performed by the DNA Technologies Core Facility at University of California, Davis. Transcripts with poly(A) tail were enriched from the total RNA samples with Bio-Poly(A) beads (Bioo Scientific, Austin, Texas). Strand-specific RNA-seq libraries were generated with barcoded adapters using the NEXTflex Rapid Directional kit (Bioo Scientific) according to the instructions of the manufacturer. Uniform fragment size distribution for all libraries was verified with a Bioanalyzer. The concentration of the libraries were quantified with Qubit Fluorometer (Thermo Fisher Scientific), pooled equimolarly, and sequenced on a HiSeq 2500 instrument (Illumina, San Diego, CA) with single-end 50 bp reads.

The sequencing data were processed using a series of bioinformatics programs. Prior to the data analysis, the sequence data in FASTQ format were subjected to quality check as well as trimming poor bases using programs in the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/index.html). The Medaka reference genome sequence was obtained from the Ensembl database (downloaded on December 18^th 2014, http://www.ensembl.org/index.html) and indexed using Bowtie2-build available within the package of Bowtie2 ver. 2.2.4 (Langmead and Salzberg, 2012). The FASTQ sequence data were used for mapping to the reference genome sequence using TopHat ver. 2.0.13 (Kim et al., 2013). The output files generated for each library were then used for transcript assembly, merging, and statistical analysis using Cufflinks, Cuffmerge, and Cuffdiff within Cufflinks ver. 2.2.1 package, respectively (Trapnell et al., 2010, 2012). All the analyses were performed with the default settings. The gene expression level was expressed as “mapped fragments per kilobase of exon model per million mapped reads (FPKM)” and genes with a false discovery rate (FDR) lower than 0.05 were considered as differentially expressed by the fipronil exposure. A two-tailed t-test was used to test for significance against a null hypothesis that there was no change while a one–tailed t-test using the estimated posterior distribution for that gene was used in cases where there were zero fragments mapped in one condition (Trapnell et al., 2013). The genes for which expression was significantly altered by fipronil exposure were subsequently annotated by BLASTX against protein database (nr) obtained from the NCBI website (downloaded on April 8^th 2015, approximately 64 million sequences were available, http://www.ncbi.nlm.nih.gov/). All the data processing with bioinformatics programs was performed using custom workstation built with 2X Xeon E5-2630 6 core CPU with 128GB ECC RAM, 4× HDD in RAID 10 configuration for Data Storage, 64bit Linux system.

Data interpretation of RNA-seq datasets based on a limited number of selected individual genes has a risk of misinterpretation and likely produces inconsistent results when analyzed by researchers with different research backgrounds and expertise. To overcome this issue, a Gene Set Enrichment Analysis (GSEA) was performed to investigate biological functions altered by fipronil exposure using gene ontology terms rather than individual genes. A GSEA was performed using GOstat, a Bioconductor package written in R (Falcon and Gentleman, 2006; R Core Team, 2016). A setting for Cuffdiff (FDR < 0.075) was used to obtain a larger number of differentially expressed genes to perform the analysis, and a total of 2,021 genes were identified as differentially expressed by the fipronil exposure with the setting. The DNA sequences for the differentially expressed Medaka genes by fipronil exposure were retrieved using a custom shell script and further used to obtain Gene Product IDs by running a BLASTX search against the annotated Medaka protein sequences downloaded from the QuickGO database available at the EBI website (http://www.ebi.ac.uk/QuickGO/GAnnotation). For BLASTX, a cutoff value of 95% identity was used to remove sequences showing low similarities, and only non-redundant Gene Product IDs were used for the GSEA. All the Medaka genes expressed in the samples as well as successfully annotated by BLASTX were utilized as “universe” or “background” (Falcon and Gentleman, 2006). The hyperGTest function, which implements the hypergeometric calculation, was used for the analysis (P value cutoff = 0.05).

RNA-seq results for the selected number of genes were validated by RT-qPCR as described in our previous publications (Rochman et al. 2014; Ramírez-Duarte et al. 2016). Briefly, total RNA (1 μg) used for RNA-seq was subjected to DNase treatment (AMPD1, Sigma-Aldroch), followed by cDNA synthesis using Superscript II Reverse Transcriptase following to the manufacturers’ instructions (Life Technologies). The primers for RT-qPCR were designed by PrimerExpress ver. 3.0 (Thermo Fisher Scientific, Waltham, MA, USA) and targets and corresponding primer sequences are listed in Table S1 (Supplementary Data). The reaction cocktail was prepared with Maxima SYBR Green qPCR Master Mix (Thermo Fisher Scientific) and the reactions were performed using an AB7900 HT FAST Thermocycler available at UC Davis Real-time PCR Research and Diagnostics Core Facility (http://www.vetmed.ucdavis.edu/vme/taqmanservice/). Each sample was analyzed in triplicate. The cycle threshold value (Ct) for each target gene was determined and relative expression levels in fold change was calculated by the 29^−ΔΔCt method (Schmittgen and Livak, 2008). The geometric means of three house keeping genes, glyceraldehyde-3-phosphate dehydrogenase (GADPH), ribosomal protein L7 (RPL7), and 18S ribosomal RNA were used for data normalization (Zhang and Hu, 2007). A Pearson’s correlation coefficient test was performed to assess the linear relationship of the gene expression data from RNA-seq and RT-qPCR using the package “Hmisc” written in R (R Core Team, 2016). The gene expression data were log2 transformed prior to the statistical test.

The measured fipronil concentrations for the selected experimental solutions are listed in Table S2 (Supplementary Data). The 0.1 μg L⁻¹ concentration was below the limit of detection (LOD) of the ELISA method. Assay #1 found no results for the 1.0 μg L⁻¹ sample (Table S2, Supplementary Data), therefore, only three replications were utilized for measuring the concentration for the 1.0 μg L⁻¹ sample, which produced a mean (± standard deviation) of 3.0 ± 1.14 μg L⁻¹. Four replicate measurements for the 10.0 μg L⁻¹ sample had a mean of 6.3 ± 0.94 μg L⁻¹. The 100 μg L⁻¹ sample had a mean of 86 ± 8.20 μg L⁻¹. A mean of 910 ± 113.52 μg L⁻¹ was found for the 1 000 μg L⁻¹ sample. The experimental solutions collected for ELISA (200, 400, 600, and 800 μg L⁻¹) were accidentally lost during the analysis. Throughout the rest of the paper, we report measured concentrations for the experimental solutions for which the ELISA data are available, otherwise nominal concentrations are reported (0.1, 3.0, 6.3, 86, 200, 400, 600, 800, and 910 μg L⁻¹).

The sequencing reaction generated approximately 61 million (M) reads for fipronil exposed groups and 58 M reads for the control groups (Table 3). Of them, 83 to 86% of the reads were successfully aligned and mapped to the Medaka reference genome sequence, providing over 15 M reads per sample for the analysis (Table 3). In total, 36,102 unique genes, including protein coding and non-coding RNA, were detected in at least one of the six groups. This includes novel genes, which were not predicted in a previous publication (Kasahara et al., 2007). With a false discovery rate (FDR) of 0.05, the expression of 174 genes was altered by the fipronil exposure with 69 and 105 genes significantly up- and down-regulated, respectively. This includes genes belonging to various categories, such as muscle growth and cardiac function (Supplemental Data Table S3).

The fipronil exposure altered a number of molecular functions as shown in Table 4. This includes biological functions involved in homeostasis of nerve cells in the central nervous system, such as voltage-gated calcium channel activity, calcium channel activity, and calcium ion transmembrane transporter activity. In addition, molecular functions of actin binding and cytoskeletal protein binding were also affected by the fipronil exposure (Table 4). However, alternation of molecular functions that are involved in endocrine system (e.g. thyroid hormone signaling pathway) were not detected.

Pearson’s correlation coefficient test indicated that there was a significant positive correlation between the RNA-seq and RT-qPCR results (r = 0.61, n = 15, P = 0.025, Fig. 5).