Comparative Developmental Toxicity of Flavonoids Using an Integrative Zebrafish System

Abstract

Flavonoids are a large, structurally diverse class of bioactive naturally occurring chemicals commonly detected in breast milk, soy based infant formulas, amniotic fluid, and fetal cord blood. The potential for pervasive early life stage exposures raises concerns for perturbation of embryogenesis, though developmental toxicity and bioactivity information is limited for many flavonoids. Therefore, we evaluated a suite of 24 flavonoid and flavonoid-like chemicals using a zebrafish embryo-larval toxicity bioassay—an alternative model for investigating developmental toxicity of environmentally relevant chemicals. Embryos were exposed to 1–50 µM of each chemical from 6 to 120 h postfertilization (hpf), and assessed for 26 adverse developmental endpoints at 24, 72, and 120 hpf. Behavioral changes were evaluated in morphologically normal animals at 24 and 72 hpf, at 120 hpf using a larval photomotor response (LPR) assay. Gene expression was comparatively evaluated for all compounds for effects on biomarker transcripts indicative of AHR (cyp1a) and ER (cyp19a1b, esr1, lhb, vtg) pathway bioactivity. Overall, 15 of 24 flavonoids elicited adverse effects on one or more of the developmental or behavioral endpoints. Hierarchical clustering and principle component analyses compared toxicity profiles and identified 3 distinct groups of bioactive flavonoids. Despite robust induction of multiple estrogen-responsive biomarkers, co-exposure with ER and GPER antagonists did not ameliorate toxicity, suggesting ER-independence and alternative modes of action. Taken together, these studies demonstrate that development is sensitive to perturbation by bioactive flavonoids in zebrafish that are not related to traditional estrogen receptor mode of action pathways. This integrative zebrafish platform provides a useful framework for evaluating flavonoid developmental toxicity and hazard prioritization.

Flavonoids and structurally related compounds are a large and diverse class of highly bioactive phytochemicals, naturally abundant in fruits, vegetables, and legumes (reviewed by: Kumar and Pandey, 2013) (Figure 1). Owing to their ubiquity, exposures occur naturally through diet, and additionally, from a number of widely available over-the-counter dietary supplements. Consequently, flavonoids and flavonoid-like chemicals have been detected in human urine (Valentin-Blasini et al., 2003; Valentin-Blasini et al., 2005) and blood plasma (Peeters et al., 2007) in many populations with diets enriched with these chemicals. Flavonoids have also been detected in umbilical cord blood (Mustafa et al., 2007; Todaka et al., 2005) and amniotic fluid (Foster et al., 2002), which suggests a potential for in utero exposures, particularly given their potential to bioaccumulate in the developing fetus and cross the blood brain barrier (Adlercreutz et al., 1999; Todaka et al., 2005; Youdim et al., 2004). Soy-based baby formulas can contain an abundance of hormonally active flavonoids (i.e. phytoestrogens) (reviewed by: Chen and Rogan, 2004), as can human breast milk (Choi et al., 2002; Song et al., 2013), leading to early developmental exposures ex utero. Some soy based infant formulas contain up to 100 µM total isoflavones, whereas individual flavonoids measured in human breast milk are typically in the low to high nanomolar range (Choi et al., 2002; Song et al., 2013; Setchell et al., 1997; Setchell et al., 1998). Infants fed soy based formulas can reach low micromolar range levels in blood plasma (Setchell et al., 1997). The result of such in utero and ex utero exposures remains unclear, given that flavonoids are natural, but have high potential for exposure. For the majority of flavonoids, the bioactivity and the potential for adverse developmental effects in vivo are not fully understood.

FIG. 1.

Flavonoid and isoflavonoids backbones, and chemical structures of the 24 tested flavonoids and flavonoid-like compounds.

Despite structural similarities, flavonoids vary widely in pharmacological properties and can exhibit marked differences in toxicokinetics and structure–activity relationships (Spencer et al., 2004). Many well-known effects associated with flavonoids include hormone mimicry, neuroreceptor modulation, anti-cancer properties, anti-inflammatory mechanisms, and antioxidant activities. These have been reported for many normal and cancerous cell types in vitro, including bone (Kanno et al., 2004), breast (Harris et al., 2005), cervical (Vidya Priyadarsini et al., 2010), colon (Daskiewicz et al., 2005), leukemia (Plochmann et al., 2007), prostate (Kobayashi et al., 2002), skin (Iwashita et al., 2000), and thyroid (Vivacqua et al., 2006). Flavonoids are highly promiscuous, and are able to modulate the activity of important molecular receptors, including nuclear estrogen receptors (ER) (Han et al., 2002), membrane bound G-protein estrogen receptor (GPER) (Thomas and Dong, 2006), the aryl hydrocarbon receptor (AHR) (Zhang et al., 2003), glycine receptors (Huang and Dillon, 2000), GABA receptors (Hanrahan et al., 2011), nicotinic acetylcholine receptors (Lee et al., 2011), and others. As a result of the diverse bioactivities across numerous cell types, flavonoids are under investigation to discover desirable pharmacological activities that may be exploited for therapeutic applications (reviewed by: Nijveldt et al., 2001). However, it is these multifaceted bioactive properties that pose a potential risk to development. In higher vertebrates, reproductive developmental effects of hormonally active flavonoids (e.g. genistein) has been widely reported. Neonatal exposure to genistein in mice results in reproductive toxicity, including abnormal estrous cycles, altered ovarian function, early onset of reproductive senescence, and reduced fecundity and fertility (Jefferson et al., 2005). Similarly, postnatal exposure to high doses of genistein in rats results in permanent estrus and altered uterine weights (Lewis et al., 2003). Given the diversity of biological activity and effects, an examination of flavonoid bioactive properties in a developing organism can provide a useful framework in which to compare this chemical class.

A zebrafish embryo-larval bioassay can be used as an alternative screening platform to evaluate the developmental toxicity and bioactivity of flavonoids and flavonoid-like chemicals in vivo. The zebrafish model is a rapid-throughput model used to link unique adverse morphological phenotypes and behavioral effects with molecular endpoints to elucidate mechanisms of action for toxicants in vivo (reviewed by: Bugel et al., 2014; Hill et al., 2005). With over 70% genetic homology to the human, the zebrafish is a powerful translational alternative animal model for human health and disease (Howe et al., 2013). Zebrafish develop rapidly within 120 h including organogenesis of all organ systems except the reproductive system, and embryogenesis is well conserved and homologous to many vertebrates, including humans (Gilbert, 2003). To our knowledge, comprehensive suites of flavonoids have not been comparatively evaluated in zebrafish for developmental toxicity. There have been a few studies of individual flavonoid toxicity in zebrafish, which typically focused on acute lethality and secondary endpoints related to hormone modulating activity (Kim et al., 2009; Sassi-Messai et al., 2009) and antioxidant properties (Chen et al., 2012). As a result, there is limited information on many flavonoids. To overcome this gap in information, a multidimensional approach is necessary to thoroughly determine the bioactivity and toxicity of a broader set of chemicals in the zebrafish (Truong et al., 2014). The zebrafish embryo-larval toxicity bioassay allows for rapid characterization of diverse phenotyping outcomes, as well as the evaluation of sub-lethal adverse effects on behavior and biomarker gene expression. The larval photomotor response (LPR) is one such behavioral assay that allows for evaluation of adverse chemical effects on the light-to-dark phototransition response in phenotypically normal animals. This flexible and integrative model allows the use of a variety of developmental and gene expression endpoints to provide high content data for hazard prioritization and identifying bioactive flavonoids.

In these studies, we use a zebrafish embryo-larval toxicity bioassay as an integrated approach to comparatively evaluate bioactivity and toxicity of 24 flavonoids and flavonoid-like chemicals (Figure 1). Using this platform, we evaluated early life exposure effects on morphological outcomes, behavior (photomotor response, thigmotaxis), and gene expression for a number of AHR (cyp1a) and ER (cyp19a1b, esr1, lhb, vtg1) biomarker transcripts (Figure 2).

FIG. 2.

Schematic of the exposure paradigm and assays used to characterize developmental effects of all test compounds on morphology, behavior, and gene expression. A, Embryos were developmentally exposed (static non-renewal) to each chemical individually from 6 to 120 h post fertilization (hpf) and observed at 24, 72, and 120 hpf for developmental toxicity and morphological defects. Example photomicrographs at 24 and 120 hpf show stage delay (SD), and effects on craniofacial endpoints (CF), pericardial edema (PE), yolk sac edema (YSE), axis and caudal fin. Targeted gene expression was evaluated for all compounds at 120 hpf for effects on AHR and ER biomarkers (cyp1a, cyp19a1b, esr1, lhb, vtg1). Behavioral effects were evaluated for all chemicals at 72 and 120 hpf. (B) The larval photomotor response (LPR) behavior assay at 120 hpf used a dark light stimulus to induce locomotor activity. Treatment effects on the photomotor response within each photoperiod were evaluated only for developmentally normal animals by binning the locomotor activity in the different Dark/Light min across trials 1–3. Data are reported as mean ± SD for the response at each minute. *Significance determined using 2-way repeated measures ANOVA, Tukey’s post hoc, P ≤ .05, N  ≥  6.

MATERIALS AND METHODS

Chemicals. The chemical structures for all tested chemicals are shown in Figure 1. Chemicals and purities purchased from Sigma-Aldrich (St. Louis, Missouri) were: 17β-estradiol (≥98%, CAS: 50-28-2), 3ʹ-hydroxyflavone (≥98%, CAS: 70460-18-3), 6-hydroxyflavone (≥98%, CAS: 6665-83-4), 7-hydroxyflavone (≥98%, CAS: 6665-86-7), α-naphthoflavone (≥98%, CAS: 604-59-1), β-naphthoflavone (≥98%, CAS: 6051-87-2), apigenin (≥95%, CAS: 520-36-5), biochanin A (≥99%, CAS: 491-80-5), chrysin (≥97%, CAS: 480-40-0), fisetin (≥98%, CAS: 345909-34-4), flavone (≥99%, CAS: 525-82-6), formononetin (≥99%, CAS: 485-72-3), galangin (≥95%, CAS: 548-83-4), luteolin (≥98%, CAS: 491-70-3), morin (≥85%, CAS: 654055-01-3), myricetin (≥96%, CAS: 529-44-2), naringenin (≥98%, CAS: 67604-48-2), puerarin (≥98%, CAS: 3681-99-0), quercetin 3-β-D glucoside (≥90%, CAS: 482-35-9), resveratrol (≥99%, CAS: 501-36-0). Chemicals and purities purchased from Cayman Chemical Company (Ann Arbor, Michigan) were: daidzein (≥95%, CAS: 486-66-8), (S)-equol (≥98%, CAS: 531-95-3), genistein (≥98%, CAS: 446-72-0), kaempferol (≥98%, CAS: 520-18-3), quercetin (≥95%, CAS: 6151-25-3). Tamoxifen (≥99%, CAS: 10540-29-1) was purchased from MP Biomedicals (Santa Ana, California). G-1 (GPER agonist, ≥98%, CAS: 881639-98-1) and G-15 (GPER antagonist, >99%, CAS: 1161002-05-6) was purchased from Tocris Bioscience (Minneapolis, Minnesota). Dimethyl sulfoxide (≥99.9%, DMSO, CAS: 67-68-5) was obtained from Avantor Performance Materials (Center Valley, Pennsylvania). All chemicals stocks were prepared in DMSO.

Zebrafish husbandry and chemical exposures. Adult wild-type 5D zebrafish (Danio rerio) were maintained on a 14:10 light/dark cycle at the Sinnhuber Aquatic Research Laboratory at Oregon State University (Corvallis, Oregon), in accordance with protocols approved by the Institutional Animal Care and Use Committee. Embryos were stage selected from group spawns to ensure all embryos were at the same developmental stage at the start of each experiment (Kimmel et al., 1995). Prior to exposures, chorions were enzymatically removed from embryos at 4–5 h post fertilization (hpf) with Pronase (Roche, Indianapolis, Indiana) following previously reported procedures (Usenko et al., 2007).

Animals were exposed throughout embryogenesis from 6 to 120 hpf, with phenotype observations, behavioral assessments, and gene expression analysis throughout development (Figure 2). Dechorionated embryos were individually transferred into clear BD 353075 Falcon 96-well plates (Corning, Corning, New York) containing 100 µl embryo medium (15 mM NaCl, 0.5 mM KCl, 1 mM CaCl₂, 0.15 mM KH₂PO₄, 0.05 mM Na₂HPO₄, 1 mM MgSO₄, 0.05 mM NaHCO₃). At 6 hpf, 100 µl of 2× treatments prepared in embryo medium were added to achieve the desired concentration in 0.1% DMSO. A standard concentration response (0, 1, 5, 10, 25 and 50 µM) was conducted on each plate with 16 embryos per concentration per plate, and 2 plates per chemical (N = 32 per treatment per chemical total). Concentrations were selected from preliminary studies and account for solubility limits (typically 50–100 µM). These nominal concentrations of flavonoids are also relatable to some high levels of exposure, which have been reported up to the low micromolar range in infant blood plasma, and 100 µM in soy based baby formulas (Cassidy et al., 1994; Setchell et al., 1997). Exposure plates were covered with parafilm, wrapped in aluminum foil to minimize chemical photo degradation, and incubated at 28.5 °C.

Mortality and morphological evaluations. Zebrafish were observed at 24, 72, and 120 hpf for mortality, developmental malformations, or abnormal involuntary movement. Observations at 24 and 120 hpf have been previously described (Noyes et al., 2015; Truong et al., 2011; Truong et al., 2014). The developmental toxicity assessments were performed blind to the identity of each chemical, though not to concentration because a standard plate format was used. Sublethal morphological anomalies at 24 hpf included developmental delay, notochord malformations, spontaneous motion. Larvae were observed at 72 hpf for spastic movement or tremors of the head or tail, and pectoral or caudal fins. At 120 hpf endpoints noted were malformations of the axis, brain, circulation, eyes, fins (caudal/pectoral), jaw, otic vesicle, pigmentation, snout, somites, swim bladder, trunk, edema around the heart (pericardial edema), or yolk sac, and response to tactile probing touch response (Figure 2A). For each animal, binary responses were recorded for each endpoint as either absent (0) or present (1). Percentage for mortality at 24, 72, and 120 hpf was calculated out of 32 animals. Percentage of mortality from 24 to 120 hpf was calculated out of survivors at 24 hpf. Percentage of animals with each morphological endpoints at each time point was calculated out of survivors at the respective time point. Percentage for “Any defect” encompasses any morphological endpoints observed at 24, 72, and 120, while “Any defect + mortality” includes morphological endpoints and mortality at any time point. Percentage for all 24 and 72 hpf endpoints was calculated out of 32 animals. Biomarker gene expression was evaluated for all chemicals at 120 hpf using 10 µM treatment, though 3ʹ-hydroxyflavone and biochanin A was evaluated at 5 µM because of mortality at 10 µM.

Larval photomotor response (LPR) behavioral analysis. Zebrafish larvae were subjected to a light-dark transition photomotor response behavior assay at 120 hpf using a ViewPoint Zebrabox system and video tracking software (ViewPoint Life Sciences, Lyon, France). The LPR behavioral assay was used to evaluate sub-lethal adverse effects on locomotor behavioral responses to a phototransition in phenotypically normal animals, and is similar to those previously described (Noyes et al., 2015; Truong et al., 2014). The light-dark photoperiod cycling consisted of: 4-min light acclimation, following by 4 repeat trials of a 2-min dark stimulation and 2-min light resting periods (Figure 2B). The first conditioning trial was treated as a stimulus training period and excluded from data analysis. For each animal, the mean activity was calculated for each min of each photoperiod (i.e. “Dark1” is the mean of the first min of the dark periods in trials 1–3, min 9, 13, and 17). Multiple trials were used to obtain an average response for each animal for each minute to reduce variability within the assay and increase confidence in each animal’s measured response. Data from animals with defects were removed from the analysis post-hoc so that behavior was analyzed only for phenotypically normal animals that did not exhibit any morphological malformations.

Analysis of mRNA expression by quantitative real-time polymerase chain reaction (qRT-PCR).For all 24 tested flavonoids and flavonoid-like compounds, messenger RNA expression was comparatively evaluated for 5 biomarkers indicative of AHR (cyp1a) and ER (cyp19a1b, esr1, lhb, vtg1) activity (Supplementary Table 1). Transcripts were measured using qRT-PCR methods previously described (Bugel et al., 2016). Briefly, total RNA was isolated using RNAzol^® RT (Molecular Research Center, Inc., Cincinnati, Ohio) and converted to cDNA using the Applied Biosystems High-Capacity cDNA Reverse Transcription kit (Life Technologies, Carlsbad, California). qRT-PCR was performed using Power SYBR^® Green PCR Master Mix with a StepOnePlus™ Real-Time PCR System (Applied Biosystems, Foster City, California). Gene expression was evaluated for 22 compounds at 10 µM, although 5 µM group was used for 3ʹ-hydroxyflavone and biochanin A because of mortality at 10 µM. For each chemical, 3 biological replicates were sampled, and each was a pool of 8–12 whole larval animals collected at 120 hpf. β-actin was used as a housekeeping transcript for normalization, and relative gene expression was quantified using the ΔΔCt method (Pfaffl, 2001).

Data and statistical analyses. Statistical tests were performed using SigmaPlot™ (v. 13.0), Prism (v. 6.01), XLSTAT (2016), and R (v. 3.2.2). A P ≤ .05 was universally regarded as significantly different for all studies. For discrete morphological observation data, a Fisher’s exact test was used to compare control versus each individual treatment. To evaluate LPR behavior data, a 2-way repeated measures ANOVA (Tukey’s post-hoc) was used. Gene expression data was evaluated using a 1-way ANOVA (Dunnett’s post-hoc) with 5% false discovery rate (FDR) and a 1.5 fold change cutoff threshold (Benjamini and Hochberg, 1995). GENE-E (v. 3.0.204) was used to generate heat maps of significant findings and for agglomerative hierarchical clustering analysis (HCA). HCA for the lowest observable adverse effect level (LOAEL) heat map was performed using a Euclidian distance based dissimilarity metric. Principle component analysis (PCA) of LOAEL values was performed using a Pearson (n) metric. A bootstrapped k-means clustering algorithm was applied to the PCA to determine cluster centers (Trace(W) criterion, 1000 iterations, 0.00001 convergence, 100 repetitions with random initial partition). All raw and summary data are provided as supplemental materials.

RESULTS

Developmental Toxicity Evaluations and Phenotype Characterizations

A zebrafish embryo-larval toxicity bioassay was used to assess the developmental toxicity for a suite of structurally diverse flavonoid and flavonoid-like chemicals. A complete summary of the mortality and developmental malformations observed at 24, 72, and 120 hpf is presented in Figure 3 and Supplementary Data Table 2A. Out of the 24 chemicals evaluated, 14 elicited significant toxicity within the tested concentration range (1–50 µM). For many of the chemicals which induced toxicity, morphological effects were typically concentration dependent. Ten chemicals significantly delayed development (stage delay) at 24 hpf, and all chemicals that affected stage progression caused a significant increase in mortality by 120 hpf at the same concentrations. Biochanin A, galangin, and α-naphthoflavone induced mortality as early as 24 hpf at the highest concentration (50 µM). However, α-naphthoflavone induced only intermediate toxicity at the 120 hpf time point. In contrast, biochanin A, galangin, and 3ʹ-hydroxyflavone, were the most lethal, with 100% mortality at 25 and 50 µM as early as 72 hpf.

FIG. 3.

A concentration-response heat map for developmental toxicity morphological evaluations at 24, 72, and 120 h post fertilization (hpf) for bioactive phytoestrogens, flavonoids, and flavonoid-like chemicals. All 24 chemicals in Fig. 1 were tested, and those not found to elicit adverse effects on these developmental toxicity endpoints are not shown (daidzein, fisetin, luteolin, morin, myricetin, naringenin, puerarin, quercetin, quercetin 3-β-D glucoside, and resveratrol). Embryos were developmentally exposed (static non-renewal) to each chemical individually from 6 to 120 hpf. Shaded cells indicate the prevalence of animals exhibiting each endpoint only when significantly elevated relative to control groups. Gray shaded cells indicate data not available due to 100% mortality. Significance determined using Fisher’s exact test, P ≤ .05, N = 32 animals per concentration. Developmental toxicity evaluation data for all 24 tested chemicals is provided in Supplementary Table 2A.

(S)-equol had the steepest concentration response curve, with 100% mortality by 120 hpf and few sublethal effects at 25 µM: only pericardial and yolk sac edema and decreased pigmentation. (S)-equol and kaempferol were also 2 of 7 compounds to induce significant spasms in the head, pectoral or caudal fins (trunk) at 72 hpf (Supplementary Video 1). Others included 3ʹ-, 6-, and 7-hydroxyflavone, β-naphthoflavone, and biochanin A. Changes in normal embryo and larval movement were already apparent at 24 hpf, however, with a decrease in spontaneous motion caused by exposure to flavone (50 µM), biochanin A (25 and 50 µM), and galangin (10-50 µM). With the exception of (S)-equol, 13 of the 14 toxic chemicals reduced the larval touch response (thigmotaxis) at least at 50 µM, with 3ʹ-hydroxyflavone, 6-hydroxyflavone, and chrysin affecting touch response significantly as low as 10 µM.

Overall, 3ʹ-hydroxyflavone was the most toxic of the chemicals tested, as it caused a significant increase in the occurrence of 13 out of the 26 sublethal endpoints at 5 µM. Biochanin A and galangin were comparably toxic in terms of lethality and deleterious effects at early time points, but had steeper response curves than 3ʹ-hydroxyflavone, and appeared less toxic based on overall lowest observed adverse effect levels. Pericardial and yolk sac edemas were the most commonly observed sublethal malformation, occurring in all 14 of the flavonoid and flavonoid-like chemicals within this concentration range. These data demonstrate that small changes in structure within the class of flavonoids can result in vastly different phenotypic outcomes in the developing organism. In addition to the effects on the cardiovascular/lymphatic systems and craniofacial and trunk morphologies, significant effects on embryo and larval movement were observed.

Evaluation of Larval Photomotor Responses at 120 Hpf

A larval photomotor response (LPR) behavioral assay was used to evaluate adverse chemical effects at 120 hpf for animals developmentally exposed to each individual test chemical. The study sought to determine if there were effects on the light to dark phototransition in animals with no observable adverse morphological effects. Effects on the LPR assay at 120 hpf are summarized in Figure 4 and Supplementary Table 2B. The number of chemicals that altered the LPR assay was fewer than those which induced developmental toxicity, and all chemicals that significantly altered the LPR assay elicited a hypoactive phenotype with reduced activity relative to controls. The 5 developmental toxic flavonoids that did not detrimentally impact LPR were β-naphthoflavone, apigenin, galangin, genistein, and kaempferol. Only one chemical, fisetin, was found to significantly impact the LPR assay in the absence of other developmental toxicity endpoints, and the effect was not concentration dependent, although this effect was repeated in 2 additional studies to confirm and validate the finding (data not shown). Not all altered responses in the dark phases were concordant with altered responses in the light phase. Three chemicals had altered responses in one of the photoperiod, and not the other (6-hydroxyflavone, 7-hydroxyflavone, biochanin A, chrysin, flavone, and formononetin). When comparing the 2 individual minutes within each photoperiod, several chemicals induced changes in only one of the 2 min (e.g. biochanin A in Dark2 but not Dark1).

FIG. 4.

Larval photomotor response (LPR) behavior assay concentration-response heat map for bioactive flavonoids and flavonoid-like chemicals. All 24 chemicals in Figure 1 were tested, and those not found to elicit adverse effects in the LPR assay are not shown (β-naphthoflavone, apigenin, daidzein, galangin, genistein, kaempferol, luteolin, morin, myricetin, naringenin, puerarin, quercetin, quercetin 3-β-D glucoside, and resveratrol). Embryos were developmentally exposed (static non-renewal) to each chemical individually from 6 to 120 h post fertilization (hpf). Only morphologically normal animals at 120 hpf were included in the LPR analysis. Shaded cells indicate the percent decrease in activity for each treatment and photoperiod only when significantly different than control activity levels. Gray shaded cells indicate data not available due to >80% mortality or defects. Significance determined using 2-way repeated measures ANOVA, Tukey’s post hoc, P ≤ .05, N ≥ 6 animals per concentration. LPR data for all 24 tested chemicals is provided in Supplementary Table 2B.

Hierarchical Clustering and Principle Component Analyses for Ranking Developmental Toxicity

To compare bioactivity for the chemicals exhibiting developmental toxicity, hierarchical clustering analysis (HCA) and principle component analysis (PCA) were performed using LOAELs for chemicals with at least one significant developmental endpoint (phenotypic evaluations and behavior). The multidimensional analysis grouped chemicals based on potency and toxicity profile for all endpoints measured. HCA using LOAEL values resulted in 3 unique groups, though without a distinction between chemical classes (Figure 5). The most potent HCA-grouped chemicals, with LOAELs between 5 and 10 µM, included 7 compounds: 3ʹ-hydroxyflavone, flavone, galangin, 6-hydroxyflavone, formononetin, chrysin, and 7-hydroxyflavone. The second group included 5 compounds which clustered based on similar moderate LOAELs in the 10–25 µM range, including biochanin A, α-naphthoflavone, β-naphthoflavone, and kaempferol. The third and final group included low potency chemicals with few observed effects at generally higher concentrations (25–50 µM), including (S)-equol, genistein, apigenin, and fisetin.

FIG. 5.

Heat map representing the lowest observed adverse effect levels (LOAEL) for all morphological and behavioral endpoints for chemicals with at least 1 significant endpoint. White colored cells indicate no significant observable adverse effect for the respective endpoint. Chemicals that did not elicit adverse developmental effects on any endpoint were not included in analysis (daidzein, luteolin, morin, myricetin, naringenin, puerarin, quercetin, quercetin 3-β-D glucoside, and resveratrol). A Euclidian distance metric with complete linkage was used for hierarchical clustering analysis (HCA).

To further compare chemical toxicity, PCA was used as an alternative dimensional reduction tool to determine relationships based on the principle components using LOAELs (Figure 6). This grouping method led to a separation of the chemicals very similar to the HCA method, with minor differences. The first group was comprised of 6 of the 7 compounds from the most potent group (3ʹ-hydroxyflavone, flavone, galangin, 6-hydroxyflavone, formononetin, and chrysin). The second group was comprised of 5 moderately potent chemicals (7-hydroxyflavone, biochanin A, α-naphthoflavone, β-naphthoflavone, and kaempferol). The third and smallest group, was nearly identical to the lowest potency group observed in the HCA, and included the chemicals with fewest effects ((S)-equol, genistein, and apigenin), though fisetin was excluded from this group.

FIG. 6.

Two-dimensional principle component analysis (PCA) for identifying clustering patterns and determining relationships between bioactive flavonoids based on lowest observed adverse effect levels (LOAEL) values for developmental toxicity endpoints including morphological observations at 24 and 120 h post fertilization (hpf), and behavioral effects at 72 and 120 hpf. A Pearson (n) metric with bootstrapped k-means clustering algorithm was used for PCA.

Evaluation of Biomarker Gene Expression

Whole animal gene expression was comparatively evaluated at 120 hpf following developmental exposure to each of the 24 tested flavonoid and flavonoid-like compounds using 5 biomarker transcripts as indicators of AHR (cyp1a) and ER (cyp19a1b, esr1, lhb, vtg1) bioactivity. These transcripts are highly specific and responsive indicators of bioactivity for their respective pathway. Expression was analyzed at 120 hpf in whole animals exposed to 10 µM for all compounds, though 5 µM had to be used for 3ʹ-hydroxyflavone and biochanin A because of mortality associated with 10 µM exposure.

Of the 24 tested compounds tested at 10 µM, 4 significantly affected expression of the AHR biomarker, cyp1a (Figure 7 and Supplementary Table 2C). The 3 compounds that significantly induced cyp1a expression in order from lowest to highest fold change were: α-naphthoflavone (3 fold) < flavone (15 fold) < β-naphthoflavone (46 fold). Genistein was the only compound found to significantly reduce expression of cyp1a, which was a modest ∼2-fold decrease in expression levels.

FIG. 7.

Heat map of biomarker gene expression responses for bioactive flavonoid and flavonoid-like chemicals. Gene expression was evaluated for all compounds at 120 h post fertilization (hpf) following developmental exposure to 10 µM, though 5 µM was used for 3ʹ-hydroxyflavone and biochanin A because of mortality with 10 µM. All 24 chemicals shown in Fig. 1 were evaluated, and those not found to elicit effects on these biomarker endpoints are not shown (3ʹ-hydroxyflavone, 6-hydroxyflavone, 7-hydroxyflavone, apigenin, chrysin, daidzein, fisetin, kaempferol, luteolin, morin, myricetin, naringenin, puerarin, quercetin, quercetin 3-β-D glucoside, and resveratrol). Data represents fold changes (treatment to control ratio) calculated using the ΔΔCt method only for significant findings. Significance determined using 1-way ANOVA, Dunnett’s post-hoc with 5% FDR (Benjamini–Hochberg) and 1.5-fold change cutoff threshold, P ≤ .05, N = 3 replicates with 8–12 animals per pool. Gene expression data for all 24 chemicals is provided in Supplementary Table 2C. The scale bar indicates the log10 transformed fold change.

Robust changes in gene expression were observed for the 4 estrogen-responsive biomarkers (cyp19a1b, esr1, lhb, vtg1) by 5 of the 24 tested compounds tested at 10 µM, including (S)-equol, formononetin, biochanin A, galangin, and genistein (Figure 7 and Supplementary Table 2C). For vtg1, a highly specific hepatic biomarker, varied induction levels were indicative of differential estrogenic potencies. From lowest to highest vtg1 fold induction: galangin (4-fold) < formononetin (46 fold) < biochanin A (312 fold) < genistein (2658 fold) < (S)-equol (22425 fold). A similar trend was observed for highly tissue specific brain aromatase, cyp19a1b: < biochanin A (2 fold) < formononetin (3 fold) < genistein (8 fold) < (S)-equol (31 fold). For lhb, the brain specific luteinizing hormone, expression patterns were: biochanin A (4 fold) < genistein (7 fold) < (S)-equol (9 fold). Only 1 chemical, galangin, reduced expression of cyp19a1b (-3 fold) and lhb (-7 fold). (S)-equol was the only chemical to elevate esr1 expression (∼2 fold), which in contrast to the other endocrine biomarker genes, is generally not tissue specific. In a separate study, gene expression was evaluated to determine whether one of the non-estrogenic flavonoids (i.e. kaempferol) may elicit estrogenic effects at higher concentrations. Despite no estrogenic response to 10 µM kaempferol, vtg1 was significantly induced 11.6 ± 6.9-fold (mean ± SD) by 50 µM kaempferol relative to controls (1.0 ± 0.3-fold) at 120 hpf (1-way ANOVA, Dunnett’s post-hoc with 5% FDR (Benjamini-Hochberg), P ≤ .05, N = 4 replicates with 8–12 animals per pool).

Investigation of the Potential Role for ERs as Mediators of Phytoestrogen Developmental Toxicity

Chemical inhibitors were used to test the hypothesis that developmental toxicity of flavonoids is ER dependent. Antagonists for nuclear and membrane ERs were used in an attempt to ameliorate developmental toxicity at 120 hpf . For these studies, embryos were developmentally exposed to 5 phytoestrogens individually at an EC85 concentration or greater, and co-treated with each ER inhibitor individually (Figure 8A and Figure 8A; Supplementary Table 2D). The 5 flavonoids evaluated for ER dependency were: biochanin A (15 µM), (S)-equol (40 µM), galangin (7.5 µM), genistein (15 µM), and kaempferol (50 µM). These 5 flavonoids were found to elicit differential estrogenic potencies in the biomarker analysis of estrogen-responsive transcripts. Kaempferol was included as a compound that elicited significant levels of toxicity with a relatively weak estrogenic response only observed at high concentrations. Chemical inhibitors used were either tamoxifen, a nuclear ER antagonist, or G-15, a membrane G-protein estrogen receptor (GPER) antagonist. For each treatment group, mortality and developmental toxicity at 120 hpf was evaluated. Neither tamoxifen nor G-15 significantly ameliorated the prevalence of mortality or developmental defects induced by any of the tested flavonoids (Figure 8A; Supplementary Table 2D).

FIG. 8.

A, Example photomicrographs of 120 h post fertilization (hpf) animals individually treated 6–120 hpf (static non-renewal) with 5 select phytoestrogens and co-treated with either 5 µM tamoxifen (ER antagonist) or 25 µM G-15 (G-protein estrogen receptor antagonist). Neither chemical inhibitor using effective and potent concentrations ameliorated developmental toxicity, suggestive that developmental toxicity was ER independent. Concentrations of phytoestrogens were selected to be an EC85 or greater for “Any defect + mortality” at 120 hpf and were: 15 µM for biochanin A, 40 µM for (S)-equol, 7.5 µM for galangin, 15 µM for genistein, and 50 µM for kaempferol. Animals were only evaluated at 120 hpf. No significant differences were observed between treatments or antagonist co-treatment groups for the prevalence of any individual or cumulative defect, mortality, or “Any defect + mortality”. Significance determined using Fisher’s exact test, P ≤ .05, N = 32 animals per concentration. All developmental toxicity evaluation data for treatments and co-treatments are provided in Supplementary Table 2D. B, Induction of estrogen-responsive biomarkers (cyp19a1b, vtg1) by E2, biochanin A, and (S)-equol were significantly inhibited by co-treatment with 5 µM tamoxifen. Embryos were developmentally exposed (static non-renewal) to each chemical individually or in combination with 5 µM tamoxifen from 6 to 120 hpf. Data are reported as mean ± SD. Significance was determined using 2-way ANOVA, Tukey’s post hoc, P ≤ .05, N = 4 replicates per treatments with 12 animals per replicate. *Indicates significant difference between treatment and respective control. †Indicates significant difference with tamoxifen co-treatment within respective treatment group. Percent inhibition is indicated for tamoxifen co-treatment groups with significant reductions. Inhibition was calculated where 1 fold represents 100% inhibition and the induction level with the test chemical represents 0% inhibition. C, Developmental toxicity of G-1 (GPER agonist) was completely ameliorated by co-treatment with 25 µM G-15 (G-protein estrogen receptor antagonist). G-1 (2.5 µM) elicited axial defects in 100% of animals, and co-treatment with G-15 completely ameliorated the phenotype. Significance determined using Fisher’s exact test, P ≤ .05, N = 32 animals per concentration. Evaluation data for G-1 and G-15 treatments and co-treatments are provided in Supplementary Table 2D.

Concentrations used for tamoxifen (5 µM) and G-15 (25 µM) were determined to effectively inhibit the 120 hpf effects of 17β-estradiol and G-1 (GPER agonist) in zebrafish (Figure 8B and C). This concentration of tamoxifen inhibits the 17β-estradiol induction of various estrogen biomarkers at 120 hpf by >95%, and is the highest concentration possible without eliciting adverse effects (Bugel et al., 2016). Expression levels of various estrogen-responsive transcripts were measured to demonstrate the antiestrogenic efficacy of tamoxifen co-treatment (Figure 8B). Tissue-specific brain aromatase (cyp19a1b) and hepatic vitellogenin (vtg1) were chosen as biomarkers to demonstrate tamoxifen’s effect on induction levels of 17β-estradiol (E2) and 2 potent phytoestrogens (biochanin A and (S)-equol). Co-treatment with 5 µM tamoxifen significantly inhibited the robust induction levels of both transcripts by E2 (100 nM), biochanin A (5 µM), and (S)-equol (10 µM). Induction of cyp19a1b by E2, biochanin A, and (S)-equol was inhibited by tamoxifen co-treatment 96, 84, and 90%, respectively. Similarly, induction levels of vtg1 expression by E2, biochanin A, and (S)-equol were inhibited by >98% by tamoxifen. To test for GPER involvement, the concentration of G-15 used inhibits the developmental toxicity of the GPER agonist, G-1, at 120 hpf in our lab, and others (Jayasinghe and Volz, 2012). Developing zebrafish exposed to 2.5 µM G-1 exhibited an axial defect in 100% of animals (31/31 animals total), whereas co-treatment with the GPER antagonist G-15 (25 µM) completely blocked this phenotype from developing (0/32 animals total) (Figure 8C; Supplementary Table 2D).

DISCUSSION

The potential for flavonoid and related phytochemicals to induce neurobehavioral and endocrine disrupting effects throughout development is a relevant toxicological question given the ubiquitous nature of these compounds and possible high levels of exposure during critical windows of development. In the present studies, we used a rapid-throughput zebrafish embryo-larval bioassay as an alternative animal model and flexible platform to evaluate developmental toxicity of flavonoids and flavonoid-like chemicals in vivo. Developmental toxicity for many of the tested flavonoids is currently limited or unavailable, and our findings underscore the possibility for impacts at critical developmental windows. Overall, 15 of 24 tested flavonoids and flavonoid-like chemicals were developmentally bioactive at the concentrations tested, with an effect on 1 or more adverse morphological endpoint or in the LPR behavioral assay (Figure 5). The distinct flavonoid toxicity profiles for the morphological, behavioral, and gene expression endpoints demonstrated chemical- and gene-specific effects. Our studies provide evidence for complex modes of action and emphasize the need for broad, unbiased global approaches to further evaluate this developmentally bioactive class of chemicals.

The evaluation of developmental toxicity endpoints in embryonic and larval zebrafish has been used to comparatively evaluate large sets of structurally similar and dissimilar chemicals (Noyes et al., 2015; Padilla et al., 2012; Truong et al., 2014). Multiple studies have evaluated the developmental toxicity of individual flavonoids in zebrafish, such as genistein (Kim et al., 2009; Sassi-Messai et al., 2009), nobiletin (Lam et al., 2011), quercetin (Zhao et al., 2014), and zearalenone (Bakos et al., 2013). The present study provides the largest comparative evaluation of flavonoid developmental toxicity to date, including 24 flavonoid and flavonoid-like chemicals, many of which lack developmental toxicity data (Figure 3). We identified 15 flavonoids that were developmentally bioactive, resulting in adverse developmental effects. A previous study evaluated acute lethality for a subset of 15 flavonoids in larval zebrafish, including 7 used in our studies (Chen et al., 2012). These 7 typically had LC50s ≥150 µM, which included 3 flavonoids not found to be bioactive in our study (morin, myricetin, quercetin). This suggests that LC50s are not reflective of developmental effects at lower concentrations, and highlights the need to use a comprehensive list of apical endpoints to rank toxicity and prioritize chemicals for further study based on toxicity profiles.

In our present study, we used HCA and PCA as a multidimensional data reduction tool to compare and rank developmental toxicity profiles for all endpoints simultaneously. The basis for this approach is that adverse developmental morphological effects are apical outcomes resulting from perturbation of various pathways related to chemical bioactivity. However, a limitation of this approach is that this analysis is based on nominal LOAELs, and does not account for differences in toxicokinetics. Differences in bioavailability may also account for some of the flavonoids exhibiting no toxicity in our studies, and the lack of toxicity should be interpreted accordingly. Nevertheless, this has been a useful for evaluating flame retardant developmental toxicity (Noyes et al., 2015). Using HCA and PCA, we identified 3 major groups of bioactive flavonoids ranked by potency and toxicity profiles, that did not, as might be expected, cluster according to flavone and isoflavone sub-classes (Figures 4 and 5). Our study demonstrates that minor structural changes alter the apparent potency and toxicity profile for flavonoids. For example, when comparing flavone to the hydroxyflavones, the addition of a single hydroxyl group at the 3ʹ-, 6-, or 7-positions resulted in spasms at 72 hpf, whereas flavone had no effects on the spasm endpoint (Figure 5). The flavones and hydroxyflavones did show similar toxicity profiles based on the HCA and PCA, with minor differences throughout the LPR assay (Figure 5). Similarly, when comparing the isoflavones genistein and biochanin A, the O-methylation of the 4ʹ-hydroxyl group resulted in lower more potent LOAELS for many endpoints, and an altered toxicity profile that led to separate classifications based on HCA and PCA (Figures 5 and 6). All flavonoids with an unsubstituted 2-position benzenoid rings displayed toxicity (e.g. galangin), whereas hydroxylated flavones and isoflavones had either varying levels of toxicity, or no effects. These findings suggest that metabolism of parent flavonoids may result in active metabolites that could have very different bioactivities than parent compounds (Figure 5). Daidzein, for example, is preferentially metabolized to the (S)-equol enantiomer by intestinal microflora (Setchell et al., 2005), resulting in a bioactive compound in our toxicity assays whereas the parent was nontoxic. In contrast, daidzein’s glucoside conjugate, puerarin, was non-responsive in the toxicity assays, similar to daidzein. In addition to spasms and changes in the LPR assay, (S)-equol also resulted in hypopigmentation, together with several others (3ʹ-hydroxyflavone, flavone, galangin, 6-hydroxyflavone, and formononetin) to a lesser extent. (S)-equol and daidzein are antimelanogenic compounds in mice via inhibition of the melanocortin 1 receptor (Chang and Tsai, 2016). Our studies demonstrate that other flavonoids may also elicit this activity. Collectively, these studies indicate that small changes in substituent functional groups could significantly change toxicity profiles and potency, and future studies should include additional flavonoids or endpoints to further elucidate structure–activity relationships.

The zebrafish has many readily assessable behavioral endpoints that are amenable to large-scale chemical screens for discovering the pharmacological effects of drugs and neurotoxicants on neurodevelopment (Drapeau et al., 2002; Ellis et al., 2012; Kokel and Peterson, 2008). In this study, we evaluated the potential for behavioral effects of flavonoids throughout development with 2 endpoints, abnormal spastic behaviors at 72 hpf, and changes in the LPR assay at 120 hpf. Behavioral effects associated with flavonoids are widely reported in rodents for multiple endpoints, including: anxiogenic and anxiolytic effects (Lund and Lephart, 2001), hyperactivity and fear (Garey et al., 2001), spatial learning and memory (Ball et al., 2010; Kohara et al., 2014), and proconvulsant and anticonvulsant activity (Avallone et al., 2000; Medina et al., 1990). The relationship between neuroendocrine and hormonal actions of estrogens are well known, and may explain the behavioral effects of phytoestrogens (Lephart et al., 2002). However, flavonoids can directly interact with and modulate synaptic neuroreceptor activity and ion currents in vitro for glycine (Huang and Dillon, 2000), GABA (Goutman et al., 2003; Hanrahan et al., 2011), and nicotinic acetylcholine receptors (Lee et al., 2011). In our present study, preliminary observations identified seizure-like behaviors throughout early development, manifesting as spastic pectoral fin movements and body twitching at 72 hpf (Supplementary Video 1). We identified 7 flavonoids that induced the seizure-like spasms at 72 hpf, at concentrations that typically preceded overt toxicity and lethality at 120 hpf (Figure 3). However, for some chemicals, the spasm responses at 72 hpf were atypical and higher concentrations less efficacious because of onset of tetany or morbidity (e.g. 3ʹ-hydroxyflavone, (S)-equol). Seizure-like activity is phenotypic of neurostimulation, which may implicate a role for neuroreceptor modulation in mediating these endpoints. Zebrafish larvae exposed to the proconvulsant pentylenetetrazole, exhibit a similar phenotypic clonus-like convulsion indicative of seizure like activity (Baraban et al., 2005). Similarly, chemical-induced convulsions have been previously reported in zebrafish larvae for a broad range of pharmacological compounds with diverse proconvulsive mechanisms (Winter et al., 2008). To further evaluate the behavioral effects of developmental exposure to flavonoids, we employed the LPR behavioral assay (Figure 4). This assay used a light-to-dark phototransition to evaluate changes in locomotor activity of phenotypically normal animals as a surrogate measure of chemical effects on behavior. The LPR behavioral assay has also been used previously to characterize the behavioral effects of flame retardants in zebrafish (Noyes et al., 2015). Overall, 10 of the 24 tested flavonoids elicited hypoactive phenotypes in the LPR behavioral assay. Only fisetin and formononetin affected the LPR assay at concentrations lacking other phenotypic effects. There was high concordance between chemicals that induced the spastic phenotype at 72 hpf, and those that induced changes in the LPR behavioral assay at 120 hpf. However, it is important to note that the hyperactive spastic phenotype is contrary to the hypoactive responses in the LPR assay, and may indicate different modes of action for these different types of effects. Altered touch responses (thigmotaxis) at 120 hpf were also highly concordant with the other behavioral assays. However, effects on spontaneous motion at 24 hpf were typically not predictive of these behavioral effects at 72 and 120 hpf. Taken together, flavonoids induce complex effects on neurodevelopmental behavior, and further studies are necessary to explore the acute and long-term behavioral effects of early life stage low level developmental exposures to flavonoids.

Targeted gene expression was evaluated for several biomarker transcripts with the goal of comparatively evaluating chemical- and gene-specific modes of action focused on the AHR and ER pathways (Figure 7). The AHR-dependent biomarker cyp1a was included because of evidence that some flavonoids behave as selective AHR modulators, depending on cell type and chemical (Zhang et al., 2003). In Hepa-1 cells in vitro, several flavonoids (apigenin, chrysin, daidzein, galangin, genistein) behave as AHR ligands, whereas others are inactive (kaempferol, luteolin, myricetin, naringenin, quercetin) (Zhang et al., 2003). Other relevant flavonoids have been predicted to be AHR agonists using an in vitro transformation assay, including chrysin, daidzein, flavone, chrysin, quercetin (Ashida et al., 2000). However, in the present study of 24 flavonoids, only β-naphthoflavone, flavone, and α-naphthoflavone upregulated cyp1a, which may suggest that most flavonoids are not AHR active at 10 µM in vivo.

In anticipation of complex hormonal and endocrine disrupting effects, we examined gene expression for several estrogenic biomarker transcripts (cyp19a1b, esr1, lhb, vtg1) and evaluated the ER-dependence of developmental toxicity for 5 select flavonoids (Figs. 7 and 8). Many flavonoids modulate hormonal signaling directly through agonistic binding to both ER subtypes (α < β), though some interact antagonistically and are thus classified as selective estrogen receptor modulators (SERMs) (reviewed by: Oseni et al., 2008). As a result, numerous flavonoids are weak to highly estrogenic in vitro, including many used in the present study: 7-hydroxyflavone, apigenin, chrysin, equol, genistein, kaempferol, luteolin, naringenin, quercetin, and resveratrol (Collins et al., 1997; Kuiper et al., 1998; Le Bail et al., 1998; Mueller et al., 2004). However, many of these flavonoids were not found to alter biomarker transcript expression in the present study, suggesting that many of these flavonoids are not bioactive at 10 µM in vivo. However, for the 5 flavonoids found to be bioactive for ER activity of the 24 flavonoids tested, some of the effects on the biomarkers were suggestive of complex SERM bioactivities that are chemical-, gene-, and tissue-dependent (Fig. 7). The biomarkers we used are highly tissue-specific and sensitive to xenoestrogens, including the liver specific vtg1, and 2 brain specific transcripts, cyp19a1b and lhb (Brion et al., 2012; Chen and Ge, 2012; Wang et al., 2005). Fold induction of vtg1 by 5 flavonoids (biochanin A, (S)-equol, formononetin, galangin and genistein) indicated differential ER activation with a large range in potency (based on fold induction). However, responses with the brain specific biomarkers were not entirely in agreement. Four isoflavones (biochanin A, (S)-equol, and genistein) induced cyp19a1b and lhb, indicative of ER activity in the brain, and formononetin only induced cyp19a1b but not lhb. In contrast, galangin had repressive activity on cyp19a1b and lhb expression, suggestive of an antiestrogenic effect in the brain. (S)-equol, the most potent inducer of vtg1 (22425 fold change), was also the only flavonoid to induce esr1, which may be due to gene specific differences in estrogen sensitivity. Collins et al. (1997) reported potent antiestrogenic effects for several flavonoids (e.g. biochanin A, chrysin, flavone, and naringenin), though none of these flavonoids was found to elicit a repressive effect on the biomarkers in our study. Collins et al. (1997) determined this antiestrogenic activity to be due to disruption of ER dimerization, similar to the tissue-selective SERM effects of tamoxifen, which are determined by highly cell-specific differences in ER binding and recruitment of co-factors. Considering the estrogenic activity for 5 of the test compounds, we sought to test the hypothesis that developmental toxicity for these compounds is dependent on ER activation. However, co-treatment with tamoxifen (ER α/β antagonist) and G-15 (GPER antagonist) did not ameliorate toxicity or provide any beneficial effect (Figure 8A). Tamoxifen however was effective in blocking the estrogenic effects of biochanin A and (S)-equol on hepatic and neural biomarkers, vtg1 and cyp19a1b, respectively (Figure 8B). Sassi-Messai et al. (2009) reported similar findings in zebrafish, and using ER antagonist ICI 182,780, demonstrated that neural apoptosis induced by genistein during development is not dependent on ER activity, despite blocking ectopic induction of cyp19a1b in the anterior brain. Taken together, the overt developmental toxicity of the tested flavonoids is likely ER- and GPER-independent. Collectively, these studies emphasize the need to integrate global transcriptomic approaches to more fully define the endocrine disrupting effects of phytoestrogens, and the ER-independent modes of action responsible for toxicity.

The nominal concentrations of flavonoids used in our studies are relatable to exposure levels measured in various biological fluids. Generally, inter-individual exposure levels vary depending on the population, flavonoid, and matrix. Several flavonoids commonly detected in biological fluids were found to be developmentally bioactive in our study (e.g. genistein, equol, kaempferol, biochanin A). In a controlled study of infants fed soy based baby formula, low-micromolar concentrations of daidzein (1.2 ± 0.2 µM) and genistein (2.5 ± 1.6 µM) were measured in blood plasma (Setchell et al., 1997). These blood plasma levels were nearly 1000 fold higher than in infants fed cow or human breast milk (5–12 nM), as well as those measured in adults fed soy-rich diets (Setchell et al., 1997). The dosages from the soy based infant formulas that resulted in these low-micromolar levels were also estimated to be 6–11-fold higher than those that elicit hormonal effects in adults (Cassidy et al., 1994). Some soy based infant formulas contain 75–111 µM (32–47 µg/ml) total isoflavones (Setchell et al., 1997, 1998). These levels are much higher than ranges for single flavonoids typically measured in human breast milk, such as daidzein (13–177 nM), genistein (23–106 nM), kaempferol (8–71 nM), naringenin (64–722 nM), and quercetin (33–109 nM) (Choi et al., 2002; Song et al., 2013). Soy based baby formulas are therefore a much more significant source of flavonoids ex utero than human breast milk (reviewed by: Chen and Rogan, 2004). However, several studies have suggested that some flavonoids bioaccumulate in the developing fetus from maternal transfer to the fetal compartment (Adlercreutz et al., 1999; Todaka et al., 2005). Levels in cord blood plasma are generally higher than those in maternal blood plasma or amniotic fluid, likely due to differences in toxicokinetics leading to prolonged retention and ultimately bioaccumulation in the developing fetus. Maternal blood plasma generally has low- to mid-nanomolar levels of total isoflavones (19–744 nM), daidzein (2–243 nM), equol (1–401 nM), and genistein (9–303 nM) (Adlercreutz et al., 1999). Amniotic fluids contain similar levels of total isoflavones (52–779 nM), daidzein (16–156 nM), equol (1–397 nM), and genistein (11–212 nM) (Adlercreutz et al., 1999). However, fetal cord blood plasma has slightly higher levels of total isoflavones (58–831 nM), daidzein (10–137 nM), equol (0–267 nM), and genistein (40–417 nM) (Adlercreutz et al., 1999). Other studies have recapitulated these findings and reported similar low- to mid-nanomolar levels in amniotic fluids (Foster et al., 2002), cord blood (Mustafa et al., 2007), and adult plasma (Choi et al., 2002; Todaka et al., 2005; Valentin-Blasini et al., 2003). Taken together, humans are developmentally exposed to flavonoids in utero and ex utero as a result of complex exposures from a variety of mediums resulting in sub- to low-micromolar concentrations of total flavonoids. A limitation of our present studies is that we did not measure toxicokinetics, though the nominal low-micromolar concentrations used to elicit developmental effects in the exposure medium (1–50 µM range) may be relevant and relatable to those measured in various biological fluids. Furthermore, exposures in humans likely involve complex mixture, whereas our study focused on single chemical exposures. The morphological effects largely represent adverse phenotypes resulting from overdose exposures, and it is possible that changes in gene expression and behavior may be found at much lower and biologically relevant no observable adverse effect levels. We therefore recommend that future studies be focused on potentially long term behavioral and neuroendocrine effects from early life stage exposure to low/no adverse effect concentrations.

CONCLUSIONS

Using an integrative approach to comparatively evaluate developmental toxicity for a suite of flavonoids and related compounds, our studies indicate that vertebrate development is sensitive to perturbation by many flavonoids. Of the 24 flavonoids tested, 15 were developmentally bioactive and elicited adverse effects on morphology and/or behavior. Adverse sub-lethal effects on behavior were also observed in phenotypically normal animals. Evaluation of biomarker gene expression indicated differential effects that were chemical-, and gene-specific effects, particularly for the hormonal SERM activity in the brain, resulting in complex estrogenic and antiestrogenic effects. Our studies suggested that the developmental toxicity was not ER- or GPER-dependent, which also indicated alternative modes of action were involved. Given the scarcity of existing developmental toxicity data for the broad class of flavonoids and related compounds, and the growing evidence for bioactivity within this class, further studies are warranted to elucidate modes of action responsible for perturbation of neuroendocrine development, particularly at no adverse effect levels of exposure. Overall, our study underscores the utility of zebrafish as an integrative, multidimensional and translational platform for examining the developmental toxicity of environmental chemicals.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.oxfordjournals.org/.

FUNDING

This work was supported by the U.S. National Institute of Environmental Health Sciences (NIEHS) with an Environmental Health Sciences Core Center grant [P30 ES000210], an NIEHS Training grant [T32 ES007060], an NIEHS Superfund Basic Research Program grant [P42 ES016465], an NIEHS Pathway to Independence award [K99 ES025280], and an EPA STAR Grant [R835168].