Comparison of transcriptomic points of departure with apical endpoints in developing zebrafish exposed to PFOS, PFOA, or PFHxS

Abstract

Zebrafish are widely leveraged for investigating per- and polyfluoroalkyl substance (PFAS) toxicity; however, there exists a data gap for employing zebrafish transcriptomic data for PFAS quantitative risk assessment. The present study exposed developing zebrafish at different start times and durations to 0 to 100 μM of perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorohexanesulfonic acid (PFHxS) to examine chemical-specific effects on transcriptomics-based point of departure estimates (tPODs). Study goals were to investigate if the tPODs were robust to differences in exposure start and duration, were protective of zebrafish apical outcomes, and were similar to recently published PFAS tPODs in other model organisms. Zebrafish tPODs were within 1.5 orders of magnitude across the 3 chemicals, exposure durations, and start times, except for the shortest and earliest exposure (6 to 24 h post fertilization) for PFOS/PFOA. Although corresponding apical endpoints were influenced by exposure scenario, tPODs were generally lower, and thus protective of zebrafish mortality and sublethal endpoints. The relationship between our tPODs and literature zebrafish toxicity effects was chemical-specific; only the PFOA tPODs were protective of all apical effects together, and PFOS/PFHxS tPODs were generally within the interquartile range of the effects concentrations. The zebrafish PFAS tPODs were also similar to previously published fathead minnow tPODs, whereas only the PFOS zebrafish tPODs were within one order of magnitude of the Daphnia magna PFOS tPOD. Our study emphasizes that the high-throughput-compatible, in vivo developmental zebrafish platform is a feasible tool for detecting dose-dependent gene expression responses that can be leveraged for higher-tier assessments.

There are thousands of substances in commerce with many, such as, pesticides and pharmaceuticals, known to negatively impact human and ecosystem health. The growing burden of new and existing environmental pollutants has driven public and environmental health agencies to invest in novel approaches for rapidly assessing chemical risk. Increasing awareness of animal welfare considerations is also refining and replacing traditional animal-based toxicity studies with shorter methods that are more reliant on mechanistic endpoints (Brooks et al. 2020). Thus, 21st century toxicology integrates advances in biological methods such as omics technologies, high-throughput screening, and in vitro and in silico approaches, to enhance the chemical risk assessment paradigm (Villeneuve and Garcia-Reyero 2011). Incorporation of molecular and biochemical data in toxicity studies is beginning to establish a mechanistic basis for read-across and a link between phenotypic and genotypic data, improving chemical classification and hazard assessment (Settivari et al. 2015).

The zebrafish (Danio rerio) is a widely used vertebrate model for toxicity testing. Well-established for studying embryonic development, zebrafish have been incorporated into routine developmental toxicity assessments in both human and ecotoxicology fields (Scholz et al. 2008; Garcia et al. 2016). Their optically transparent embryos and rapid external development (Kimmel et al. 1995) enable early-stage morphological, behavioral, and molecular-based toxicity studies within the first 5 d of development (Nishimura et al. 2016). The rapid, sensitive, and dynamic nature of zebrafish development makes them a valuable model for identifying teratogens, understanding chemical toxicity mechanisms, and prioritizing chemical hazards through whole-animal phenotypic and molecular evaluations (Tal et al. 2020; Truong et al. 2020).

The developing zebrafish assay has recently been applied to screen the bioactivity of various per- and polyfluoroalkyl substances (PFAS) using apical endpoints (Dasgupta et al. 2020; Han et al. 2021; Truong et al. 2022). PFAS are a diverse group of persistent anthropogenic contaminants that have attained worldwide prominence in public media and in scientific and regulatory communities, due to their ubiquitous nature and environmental stability of the carbon-fluorine bond. PFAS have been detected throughout the environment, including in wild animals and humans (De Silva et al. 2021). PFAS exposure has been associated with several adverse health outcomes, including negative reproductive and neurobehavioral effects, liver and kidney diseases, thyroid and immune dysfunction, as well as developmental malformations, sensory deficiencies, and altered transcriptomic effects (Ankley et al. 2021; Rericha et al. 2023).

Although zebrafish transcriptomics have helped with elucidation of mechanisms of toxicity of PFAS and other chemicals, application of transcriptomic data for quantitative risk assessment has, until recently, remained limited due to several challenges (Sauer et al. 2017; Ursula et al. 2017; Pain et al. 2020). However, decreasing costs and increasing throughput now support omics-based studies, and recent proof of concept publications suggest that benchmark dose response analysis of transcriptomic data from shorter exposures can identify points of departure (PODs) consistent with chronic PODs from traditional long-term animal studies (Thomas et al. 2013; National Toxicology Program 2018). Transcriptomic PODs (tPOD) represent chemical exposure concentrations below which concerted molecular changes are not expected to occur and are thus hypothesized to be protective against chemical-induced adverse effects (Johnson et al. 2022). Importantly, tPODs can be identified without need to understand the biological processes and links to specific hazards. Recent studies have demonstrated that tPODs can be protective of apical outcomes across various vertebrate models: e.g., short-term chemical exposures in fathead minnows and developing zebrafish, as well as chronic exposures in adult zebrafish and rainbow trout (Pagé-Larivière et al. 2019; Alcaraz et al. 2022; Villeneuve et al. 2023; Mahoney et al. 2024). These findings support the use of transcriptomic studies in chemical risk assessment. However, questions remain as to whether PFAS exposure during zebrafish development consistently produces tPODs that are protective of (i.e. lower than) phenotypic responses, and whether developmental window of exposure influences the relationship between phenotypic and transcriptomic responses.

This study leveraged the developmental zebrafish model to assess if zebrafish tPOD estimates for well-studied PFAS, perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorohexanesulfonic acid (PFHxS), would vary significantly depending on the duration of exposure and/or developmental window over which the exposure took place, and whether those tPODs would be lower than apical effect concentrations. To investigate this, zebrafish were exposed to 8 concentrations of each PFAS with exposures starting at 6, 24, 48, or 96 hpf and extending to 24, 48, and 120 hpf. We hypothesized that, (i) tPODs would be more sensitive than apical endpoints regardless of exposure window, (ii) chemical-specific tPODs would remain consistent across exposure windows, with up to one order of magnitude difference between the individual tPODs, and (iii) zebrafish tPODs for PFOS, PFOA, and PFHxS would be within an order of magnitude of those from fathead minnows exposed to the same chemicals in a previous study (Villeneuve et al. 2024b).

Materials and methods

Zebrafish husbandry

All procedures involving zebrafish (D. rerio) were performed in agreement with the Interagency Research Animal Committee guidelines and under an in-house, approved Animal Care and Use Committee protocol (ACUP), “The culture, rearing and holding of zebrafish (D. rerio)” (ACUP#: Eco25-09-001). Adult zebrafish (AB wild type strain) were housed in an onsite culture facility at the US Environmental Protection Agency (USEPA) laboratory in Duluth, Minnesota. The fish were raised in a flow-through system with control UV-treated, filtered Lake Superior water (LSW) at a flow rate of 250 ml/min. Fish were held at 25 ± 1 °C under a 16/8-h light/dark photoperiod cycle. Fish were housed in 20.8-l tanks with <100 zebrafish per tank. Adult fish were fed live brine shrimp (Artemia) nauplii twice daily.

The night prior, sexually mature zebrafish in groups of 4 males and 6 females were separated into breeding tanks containing a plastic screened spawning basket. The next morning, approximately 2 h after the lights turned on, embryos were collected and maintained in filtered LSW in an incubator set at 28.5 °C. At approximately 4 hpf, zebrafish embryos were screened under a stereo microscope, and only similarly-staged embryos were sorted out for chemical exposure (Kimmel et al. 1995).

Chemicals

All solvents were of HPLC grade or higher from Fisher Scientific. DMSO and chlorpyrifos (>98% purity) were purchased from Sigma Aldrich. PFOS (CASRN 1763-23-1; DTXSID3031864), PFOA (CASRN 335-67-1; DTXSID8031865), and PFHxS (CASRN 355-46-4; DTXSID7040150) were procured solubilized in DMSO from Evotec (Branford CT, USA; Contract #EP-D-12-034). Mass labeled ¹³C8-PFOS, ¹³C8-PFOA, and ¹³C3-PFHxS were purchased from Wellington Laboratories Inc. for use as internal standards.

Waterborne PFAS chemical exposure

This study investigated 4 different exposure start times, 6, 24, 48, and 96 hpf (Fig. 1). Exposures ended at 24, 48, 96, 120, or 144 hpf. The longest exposure window was from 6 to 144 hpf (~6 d), and the shortest windows were 6–24, 24–48, and 96–120 hpf (~1 d). For the windows that began after 6 hpf (24, 48, 96 hpf), zebrafish embryos were plated in 6-well polystyrene plates (Costar 3736; Corning, Kennebunk, Maine) with 12 embryos/well in approximately 5 ml of LSW until the start of the chemical exposure. Plates were sealed with parafilm and placed in the 28.5 °C incubator, and 50% of the LSW was renewed daily, removing any dead or malformed embryos. For the windows that began at 6 hpf, embryos were placed directly into the chemical exposure plates after sorting.

Fig. 1.

Fourteen windows of exposure beginning at 4 exposure start times (6, 24, 48, 96 h post fertilization [hpf]) with the developmental toxicity endpoints that were measured. Yellow stars indicate the timepoints when RNA samples were collected for transcriptomics and tPOD estimations.

At the start of exposure for all windows, zebrafish were placed one per well in round bottom 96-well polystyrene plates (Sigma-Aldrich) in 20 μl LSW. Chemical solutions (180 μl/well) were manually added to their respective wells to reach the final exposure concentrations. Zebrafish were exposed to 0.33% DMSO (vehicle control), 1 μM chlorpyrifos (behavior positive control), or a half-log dilution concentration range of PFOS, PFOA, or PFHxS: 100, 31.6, 10.0, 3.2, 1.0, 0.32, 0.10, and 0.032 μM (Table 1). Each PFAS concentration was included on two of the three 96-well plates (n = 5 individuals/plate), and DMSO controls (n = 8 individuals/plate) and chlorpyrifos (n = 6 individuals/plate) were included on all 3 plates. The 96-well plates were sealed using a silicone mat (catalog #506065, NEST Scientific USA, New Jersey) and placed in the dark at 28.5 °C overnight. Every 24 h, after developmental toxicity endpoints were evaluated (see below), 50% renewal of each exposure solution was conducted.

Table 1.

Nominal and measured exposure concentrations in μM and mg/l for each PFAS.

Chemical	Nominal (μM)	Nominal (mg/l)	Measured (mg/l)	SD (mg/l)
PFOS	0.032	0.0171	0.007	0.004
	0.1	0.0538	0.033	0.009
	0.317	0.1705	0.113	0.030
	1.003	0.5382	0.506	0.139
	3.169	1.7045	1.639	0.670
	10.014	5.3822	6.386	1.220
	31.64	17.0454	20.825	4.064
	100	53.822	74.407	21.279
PFOA	0.032	0.0131	0.016	0.006
	0.1	0.0414	0.052	0.014
	0.317	0.1311	0.160	0.061
	1.003	0.4141	0.571	0.196
	3.169	1.3114	1.687	0.470
	10.014	4.1407	6.461	2.076
	31.64	13.1136	15.875	7.146
	100	41.407	57.325	23.304
PFHxS	0.032	0.0139	0.015	0.005
	0.1	0.0438	0.049	0.014
	0.317	0.1388	0.191	0.052
	1.003	0.4382	0.578	0.165
	3.169	1.3878	1.790	0.436
	10.014	4.382	7.061	2.268
	31.64	13.8778	23.251	11.937
	100	43.82	65.395	21.619

Average measured concentrations across all windows of exposure that were used for analyses, along with the corresponding standard deviations (SD).

Exposure analytical verification

At the end of each zebrafish exposure window, 100 μl of the 200 μl exposure solution was collected for analytical verification using liquid chromatography–tandem mass spectrometry (LC–MS/MS; Table 1, Table S1). The sample was diluted with acetonitrile including internal standards and directly injected into an Agilent 1260 system (Agilent Technologies) coupled to an Agilent 6410 mass spectrometer.

Separations were performed using a Poroshell 120 EC-C18 column (Agilent, Santa Clara, California) (2.1 × 50 mm, 2.7 μm) with a 5 min run time and gradient mobile phase flowing at 0.4 ml/min. Composition of mobile phase was: Mobile phase A—5 mM ammonium acetate in water and mobile phase B—5 mM ammonium acetate in methanol. Mobile phase programming was started from 50% B for 0.1 min and then gradually increased to 100% B at 1.50 min. 100% B was held until 3.50 min and was dropped back to 50% B at 3.51 min. Concentrations of analytes were determined by internal standard quantification method using Agilent MassHunter software.

MS parameters were as follows: Electron spray ionization (ESI) was in negative mode at 2500 V, gas temperature was at 350 °C, and the nebulizing gas flow was at a rate of 13 l/min at 60 psi. PFOS precursor Ion was 498.9 m/z, product ion 99 m/z, and 80 m/z, with a collision energy of 56 V for both. PFOA precursor Ion was 413 m/z, product ion 368.9 m/z and 169 m/z, with a collision energy of 6 and 18 V. PFHxS precursor Ion was 399 m/z, product ion 99 m/z and 80 m/z, with a collision energy of 44 and 48 V.

Developmental toxicity (Apical endpoints)

Morphology

Mortality, hatching, and posterior swim bladder (SB) inflation were recorded as a binary presence or absence of each endpoint (Fig. 1). Mortality was evaluated at all time points of each exposure window (Fig. S1). Hatching was evaluated at 24, 48, 72, 96, and 120 hpf (no PFAS-related effects, hence data are not shown), and SB inflation was evaluated at 120 and 144 hpf. At 28.5 °C, hatching occurs at 48 hpf (Kimmel et al. 1995), and whereas SB inflation is expected to begin at 72 hpf (Kimmel et al. 1995), we noted that the SB of control animals began to inflate at approximately 120 hpf in our 96-well plate format (Fig. S2). At 24 hpf, control animal tail detachment and spontaneous movement (Kimmel et al. 1995) were evaluated to confirm that zebrafish were developing normally (data not shown). Any other lethal abnormalities, including edema (pericardial and yolk sac), and axis malformations, were also noted if observed (data not shown). These animals were considered “abnormal” and were removed from the behavior analyses (see below).

Length

At 96, 120, and 144 hpf of all exposure windows that included these time points, images of all zebrafish laying on their sides were captured 4 wells at a time, using a Nikon DS-fi3 high-definition color microscope camera mounted on a Nikon SMZ1270 stereomicroscope. If animals were upright and swimming, they were gently nudged with a pipette tip so they would lay on their sides. Total lengths of individual animals were measured from the tip of the nose to the end of the caudal fin using ImageJ (Schneider et al. 2012). Measurements in pixels were normalized to the known 96-well plate diameter (6.858 mm). Individual animals that were identified as dead at the timepoint of the length measurement were removed from further analyses. Data for length over time is presented in Fig. S3.

Behavior

For the exposure windows that ended at 120 or 144 hpf, larval behavior at 120 hpf was evaluated with a larval photomotor response assay using a DanioVision instrument (Noldus, Wageningen, The Netherlands). The 96-well plates were taken out of the dark incubator and immediately transferred to the DanioVision Observation Chamber. Zebrafish were then allowed to acclimate for 10 min in the dark, after which the assay was immediately performed. The assay consisted of 2 cycles of 2 min in the dark (infrared light) and 2 min in visible light, with 120-hpf fish generally having greater movement in the dark compared to in the light (Kristofco et al. 2016). High-definition video was captured to track each individual fish’s movement, and the total distance moved (mm) was integrated over 20 s time bins by the manufacturer’s software (Ethovision XT Version 15). An R script (version 4.3.0) was written to further analyze data using RStudio (version 2024.04.2) (R Core Development Team 2010). Both control and PFAS-exposed fish with uninflated swim bladders were included in the analysis since majority of both the control and PFAS-exposed fish only had their swim bladders fully inflated by 144 hpf. The caveat was that the uninflated swim bladder could affect the presented swimming behavior results as seen in previous studies (Hagenaars et al. 2014). Although initial sample sizes were up to n = 20 animals/PFAS concentration/exposure start time, n = 48 for DMSO, and n = 36 for Chlorpyrifos, fish exhibiting lethal malformations or complete immobility throughout the assay were excluded from the behavioral analysis. The final number of individuals analyzed for each exposure start time for all PFAS treatments ranged as follows: 6 hpf: 7–14, 24 hpf: 1–18, 48 hpf: 2–20, and 96 hpf: 12–19. Final sample sizes for DMSO ranged from 18 to 41, and for Chlorpyrifos ranged from 13 to 29 across all exposure start times. Data for resulting concentration-dependent behavior responses are presented in Fig. S4.

Concentration response modeling of developmental toxicity data

Benchmark concentrations (BMCs) were estimated for mortality, SB inflation, growth, and behavior. These analyses involved fitting concentration-response curves (Sheffield 2021), so concentration was treated as continuous. Multiple mathematical models were tested against the data using the R package, tcplfit2 (Sheffield et al. 2022), and the best fitting concentration-response curves with the minimum Akaike Information Criterion (AIC) value were generated for each exposure start time (6, 24, 48, or 96 hpf). We then calculated the BMC (modeled concentration that corresponds to a benchmark response of 1.349 standard deviation over the modeled control response), and associated upper (BMCU) and lower (BMCL) 95% confidence limits around the BMC. We designated the BMCL as the apical POD (aPOD), acknowledging that there are several different ways of calculating aPODs, for example, the lowest observed effect concentration in Pagé-Larivière et al. (2019). A quality filter was applied such that if the BMCL was greater than the highest concentration tested, we instead reported the highest tested concentration for comparison of tPOD data within this study. For mortality, we examined the final timepoint observed, and normalized responses as percent of control (DMSO). We analyzed SB inflation at 120 hpf, at which point only some individuals are expected to have inflated swim bladders. For length, we analyzed growth from 120 to 144 hpf, and from 96 to 120 hpf; but because all estimated BMCLs were greater than the highest tested concentration for each PFAS (Fig. S7A and B), data were not presented for comparisons with other aPODs and the tPODs determined in this study. To meet the requirement of centering the data at zero for SB inflation, we again normalized as percent of control (DMSO), but then subtracted from each observation the median response exhibited by the 2 lowest non-zero concentrations. The latter step was done to try to improve model fit because the data did not show an obvious dose-response relationship relative to DMSO (Sheffield et al. 2022). For behavior, we followed the approach of Thunga et al. (2021) and examined total distance moved separately for each phase. We used here a normalization procedure that also included the positive control: (total distance moved – DMSO control)/Chlorpyrifos positive control ×100 (Sheffield et al. 2022). For all endpoints, to get one observation per concentration, we collapsed replicates by taking the median across all wells. To determine the biological realism of the model chosen for BMC derivation of each endpoint, we considered the continuous Hitcall, which takes into account the AIC of the winning model, as well as the concentrations at which the median response exceeds the cutoff (Sheffield et al. 2022). Hitcall values <0.5 should be interpreted with caution (Harrill et al. 2021) and values closer to 1 have high probability of representing realistic biology (Sheffield et al. 2022).

Transcriptomics

Sample collection, RNA isolation, library preparation, and sequencing

For each PFAS, 5 individual animals from each of the 8 exposure concentrations (0.032 to 100 μM) were collected for transcriptomics at the end of 7 (of the 14) exposure windows: 6–24, 6–48, 24–48, 6–120, 24–120, 48–120, and 96–120 hpf. Thus, we collected 5 individuals ×8 concentrations = 40 samples per PFAS and exposure window combination. The only exceptions were the 2 highest PFOS concentrations from the 6 to 120 hpf window where we observed mortality by the end of the exposure, and thus animals were not sequenced. Additionally, individual animals exposed to DMSO (n = 8/exposure plate) were collected from each exposure window. At the end of the exposure, exposure solutions were removed and replaced immediately with homogenization buffer RLT (Qiagen) and beta-mercaptoethanol (10 μl/1 ml buffer). Animals were homogenized using a bead mixer mill with a 2.3 mm stainless steel bead, at 30 Hz for 2 min, and frozen immediately at −80 °C.

Detailed methods for RNA isolation, library preparation, and sequencing are provided in Information SM.1. Briefly, total RNA was isolated with the Ambion MagMAX-96 Total RNA Isolation kit (ThermoFisher Scientific AM1830) following the manufacturer’s user protocol using the epMotion 5075 liquid handling system. RNA quantification was done using a Synergy HTX multimode plate reader, and Take3 plate (BioTek). The 260/280 ratio was 2.0 ± 0.5 for majority of the samples. RNA quality was checked by TapeStation 4200 RNA assay (Agilent), and RNA integrity (RIN) scores ranged from 3.4 to 10, with a median RIN of 9.0. Samples with a 260/280 ratio outside of 2.0 ± 0.5 had a RIN > 7.0, therefore, were considered of sufficient quality for sequencing. 60 ng (10 μl) per sample of total RNA was used as input for the Lexogen Poly(A) Selection component of library generation kits. RNA-Seq libraries were generated using CORALL RNA-Seq Library Prep Kits with UDIs (version 1) according to manufacturer’s protocol. Pooled libraries were shipped to the Michigan State University RTSF Genomics Core for sequencing.

Each pool was loaded onto one lane of a NovaSeq S4 flow cell and sequenced in a 2 × 150 bp paired end format using a NovaSeq 6000 v1.5 300 cycle reagent kit. Base calling was done by Illumina Real Time Analysis (RTA) v3.4.4, and output of RTA was demultiplexed and converted to FASTQ format with Illumina Bcl2fastq v2.20.0. Samples had an average sequencing depth 21 M reads. Trimmed reads were aligned to the zebrafish reference genome (GRCz11) with STAR (v2.7.10a), and FastQC (v0.11.9) and MultiQC (v1.14) were used to assess the quality control metrics of the processed reads. All additional sequencing analyses details are provided in Information SM.1.

Raw data were deposited in the National Center for Biotechnology Information’s (NCBI) Sequence Read Archive (SRA; Bioproject ID PRJNA1011844). Processed data and associated sample meta-data were deposited in the NCBI Gene Expression Omnibus (GSE242402).

Concentration response modeling of transcriptomic data

Concentration-response modeling was conducted using BMDExpress3 (version 3.0; Yang et al. 2007) following the general approach detailed by the National Toxicology Program (2018). After statistically pre-filtering the concentration response data, features with concentration-dependent, monotonic trends in expression were fit to 6 different regression models (Hill, Linear, Power, Polynomial 2, and Exponential 3, 5). The best fit model (i.e. lowest Akaike information criterion [AIC]) was used to calculate a BMC and associated upper (BMCU) and lower (BMCL) 95% confidence range using 1 standard deviation above the control mean as a benchmark response. Quality filters were applied such that BMCs more than 10-fold lower than the lowest concentration tested, greater than the highest concentration tested, with a best fit P-value > 0.1, or that yielded a BMCU/BMCL greater than 40 were removed. The median of the BMC distribution was calculated, in addition to the median of the BMCU and BMCL distributions. The tPOD was defined as the 10th percentile of the distribution of the BMCs determined for genes whose concentration-response curves passed the applied quality filters for each chemical, similar to our previously published studies in fathead minnow and Daphnia magna (Flynn et al. 2024; Villeneuve et al. 2024a, 2024b) for easier comparison with the published tPODs. Error estimates around the tPOD were calculated as the 10th percentile of the distribution of BMCUs and BMCLs.

Results

Transcriptomic points of departure

RNA sequencing conducted on individual fish (n = 5/treatment) at the end of 6 exposure windows: 6–24, 24–48, 6–120, 24–120, 48–120, and 96–120 hpf revealed significant transcriptomic alterations. The total number of annotated transcripts across all exposure windows for the 3 PFAS ranged from 19,545 to 22,538 (Table 2). Of these, the number of concentration responsive genes (CRGs) (evaluated using ANOVA; P-value < 0.05, minimum absolute fold change ≥2, and a concentration-dependent trend based on the Williams Trend test in BMDExpress3) to which curve fitting was applied, ranged from 27 (6 to 48 hpf) to 295 (6 to 24 hpf) for PFOS, 29 (24 to 48 hpf) to 341 (6 to 24 hpf) for PFOA, and 23 (24 to 48 hpf) to 138 (24 to 120 hpf) for PFHxS. For both PFOS and PFOA, the highest number of CRGs for curve fitting was identified in the 6 to 24 hpf exposure window. The total number of CRGs ranged from 3 (PFHxS; 6 to 48 hpf) to 116 (PFOA; 6 to 24 hpf), an average decrease of ~70% compared to the total number of CRGs for which curve fitting was applied. This percent decrease ranged from 51.5% in the 48 to 120 hpf window for PFOA to 92.9% in the 6 to 48 hpf window for PFHxS.

Table 2.

Summary of BMDExpress3 results for developing zebrafish exposed to 8 concentrations (n = 5) each of PFOS, PFOA, or PFHxS for 7 exposure windows.

Chemical (0.1–100 μM)	Lowest conc. (0.032 μM) (mg/l)	Highest conc. (100 μM) (mg/l)	Exposure window (hpf)	Input transc-ripts	CRGs pre-filter	CRGs post-filter	% decrease from pre to post filter	Median BMC (mg/l)	Median BMCL (mg/l)	Median BMCU (mg/l)	10th Percentile tPOD (mg/l)
PFOS	0.007	74.41	6–24	19,545	295	69	76.61	0.04	0.02	0.15	0.01
			6–48	20,042	27	7	74.07	12.01	7.46	28.49	0.92
			24–48	20,291	72	33	54.17	12.81	7.73	39.33	9.58
	0.007	20.83	6–120	21,081	234	76	67.52	8.92	6.51	14.89	2.47
			24–120	21,052	143	56	60.84	11.15	7.72	21.17	1.95
			48–120	20,950	163	62	61.96	11.11	7.84	18.98	1.96
	0.007	74.41	96–120	20,959	71	29	59.15	31.15	22.96	64.85	5.38
PFOA	0.160	57.33	6–24	20,650	341	116	65.98	0.02	0.01	0.06	0.01
			6–48	21,230	66	13	80.30	24.17	17.86	50.15	8.48
			24–48	21,181	29	4	86.21	36.62	24.05	78.07	15.96
			6–120	22,480	147	46	68.71	8.84	5.55	14.16	0.93
			24–120	22,522	68	18	73.53	26.31	18.97	44.49	4.24
			48–120	22,538	134	65	51.49	40.43	26.48	76.42	6.01
			96–120	22,530	64	24	62.50	45.76	27.97	99.66	8.01
PFHxS	0.015	65.40	6–24	19,911	76	16	78.95	54.59	33.02	110.22	28.61
			6–48	20,426	42	3	92.86	53.09	32.63	65.28	21.98
			24–48	20,891	23	7	69.57	18.02	8.77	51.76	7.19
			6–120	22,363	132	53	59.85	7.27	3.25	12.40	1.19
			24–120	21,947	138	47	65.94	47.45	30.19	65.69	6.84
			48–120	22,215	61	15	75.41	44.93	29.13	96.69	7.56
			96–120	22,113	105	30	71.43	33.44	24.00	74.48	1.69

Table shows the concentration responsive genes (CRGs) pre- and post-filters, as well as the median benchmark concentration (BMC), lower bound of the 95% confidence interval of the BMC (BMCL), upper bound of the 95% confidence interval of the BMC (BMCU), and the transcriptomic point of departure (tPOD), calculated as the 10th percentile of the BMC. The first (black) and second (gray) most potent exposure windows by median BMCs and tPODs are highlighted.

Based on the tPOD estimates, the zebrafish transcriptome response was most sensitive when exposed during the 6 to 24 hpf window for PFOS and PFOA, and the 6 to 120 hpf window for PFHxS (Fig. 2). Considering the tPODs calculated for different exposure windows ending at 120 hpf, there was a general trend for the longer exposure durations to have lower tPOD estimates. This was most apparent for PFOA, whereas the PFOS 6 to 120 hpf and the PFHxS 96 to 120 hpf exposures were notable exceptions to that overall trend (Fig. 2). However, even within the set of exposures ending at 120 h, the overall variability in the tPOD estimates was less than 1 order of magnitude (Table 2; 3.5-fold for PFOS; 5.2-fold for PFOA; 6.7-fold for PFHxS). With the exception of the tPODs calculated for PFOS and PFOA for the exposure window 6 to 24 hpf, all zebrafish tPOD estimates differed by less than 2 orders of magnitude from each other, both within individual chemicals and across all 3 PFAS.

Fig. 2.

Transcriptomic points of departure (tPOD) for zebrafish exposed to PFOS, PFOA, or PFHxS for 7 developmental windows of exposure. Circles are the 10th percentile of the distribution of the benchmark concentrations determined for genes whose concentration-response curves passed our quality filters. Error bars represent the 10th percentile of the upper and lower confidence limits of the benchmark concentrations (BMCU and BMCL, respectively). In addition, published tPODs for fathead minnow exposed from 120 to 144 h post fertilization (hpf) (5 to 6 d post fertilization) (Villeneuve et al. 2024b), and Daphnia magna exposed from 72 to 96 hpf to each PFAS (Villeneuve et al. 2024a) are included as blue and red horizontal lines, respectively. The dashed blue and red lines for fathead minnow and D. magna are the corresponding BMCUs and BMCLs for each species.

In addition to comparing zebrafish tPODs across the different exposure windows, we also compared the zebrafish tPOD estimates to recently published fathead minnow (Villeneuve et al. 2024b) and D. magna (Villeneuve et al. 2024a) tPOD data for the same PFAS. Fathead minnows were exposed for 24 h to a concentration range of each of the 3 chemicals from 5 to 6 d post fertilization (~120 to 144 hpf), and the identified tPODs were similar to those estimated for the zebrafish for most windows of exposure. Any exceptions to this trend were exhibited from zebrafish exposures that started at 6 hpf (Fig. 2). The zebrafish tPOD estimates for PFOS and PFOA for the 6 to 24 hpf window were 3 orders of magnitude lower than the fathead minnow estimates, and the PFOS (6 to 48 hpf) and PFOA (6 to 120 hpf) estimates were an order of magnitude lower, and thus more conservative than the fathead minnow estimates. The Daphnia tPOD for PFOS was similar to that of fathead minnow; however, the Daphnia estimates for PFOA and PFHxS were 2 to 3 orders of magnitude lower than the 2 fish species, which was consistent with the lower sensitivity of the fathead minnow transcriptome to PFOA and PFHxS exposure compared to Daphnia (Villeneuve et al. 2024a).

Apical points of departure

To facilitate comparison with the tPODs, concentration-response modeling was successfully applied to mortality, swim bladder inflation, and the 4 phases of the behavior assay conducted at 120 hpf using the average measured concentrations to derive benchmark concentrations and corresponding aPODs (Figs S5, S6, and S8A–D). The relationship between endpoint-specific aPODs and exposure start was PFAS-specific.

For PFOS, in general, the earlier exposure start times had lower aPOD values compared to the later start times (Fig. 3). Some exceptions included SB inflation when exposure started at 24 hpf, and the first Dark phase when an aPOD could not be calculated at the 6 hpf exposure start time. Both light phases of the behavior assay and the second dark phase (light 1, light 2, and dark 2; Fig. 3) had significant concentration-response effects at the earliest 2 start times, 6 and 24 hpf, but not when exposure started later at 48 or 96 hpf. For PFOA and PFHxS, we noted that only 3 and 2 aPODs, respectively, were determined, exclusively for mortality and SB inflation (Fig. 3). This contrasted with PFOS’ 13 aPODs, suggesting that PFOA and PFHxS were less bioactive compared to PFOS in our study, as has been observed in other studies (Ding et al. 2013; Gaballah et al. 2020).

Fig. 3.

Summary of the apical points of departure (aPODs) for each developmental toxicity endpoint (mortality, posterior swim bladder [SB] inflation, and 4 phases of behavior assay). The light blue circles are the calculated BMCLs (aPODs) for each endpoint, and if a BMC could not be calculated, the highest exposure concentration (dark blue circles) is included for comparison. The smaller the circle size, the lower the aPOD. Hitcall > 0.5 is indicated with “þ”, and Hitcalls < 0.5 should be interpreted with caution (Harrill et al. 2021).

When focusing on specific endpoints and exposure start times, we found that fish exposed to PFOS starting at 6 hpf were approximately 5 times more sensitive to effects on survival than the fish that were not exposed until 48 hpf. When considering total distance moved in the light phases of the behavioral assay, fish exposed to PFOS from 6 hpf were around 200 and 500 times (PFOS light 1 and light 2, Fig. 3) more sensitive than fish exposed from 48 or 96 hpf. For dark 2, PFOS-exposed fish were less than three times more sensitive when exposed from 6 hpf compared to later exposure start times. For SB inflation for both PFOS and PFOA, fish exposed from 48 hpf were at least 15 times more sensitive than fish exposed from 24 hpf, and for PFHxS, fish exposed from 96 hpf were at least 400 times more sensitive than fish exposed from the other 3 exposure start times. Although the SB inflation results could be an artifact of exposure design differences between the exposures beginning at different start times, we also acknowledge that with the current exposure design, we cannot know if fish exposed from later start times would have been equally as sensitive as the earlier start times had their exposure duration been extended to a total of > 96 h. Nevertheless, in this study, the variability in effect concentration estimates one might expect for different chemical exposure windows was dependent on not just the type of PFAS but also on the endpoint of interest. In addition, aPODs varied by several orders of magnitude across exposure start times particularly for the mortality and behavior endpoints for PFOS.

Comparison between tPODs and aPODs

Investigating both morphological and behavioral endpoints, as well tPODs allowed us to compare aPODs with tPODs (Fig. 4) to test our hypothesis that the tPOD estimations would be protective of (lower than) any of the corresponding morphological or behavioral points of departures (aPODs) irrespective of when the exposure commenced or the exposure duration.

Fig. 4.

Comparison of the zebrafish tPOD estimates for each window of exposure with phenotypic PODs (aPODs) for mortality (A), posterior swim bladder [SB] inflation (B), and PFOS behavior (C) presented for each exposure start time. If an aPOD (light blue square) could not be estimated, the highest exposure concentration (dark blue square) is presented. aPODs could not be estimated for any of the PFOA and PFHxS behavior data and hence are not present in the figure.

For PFOS, tPODs were generally protective of mortality, swim bladder inflation, and total distance moved in each of the phases of the behavior assay (Fig. 4A–C). Exceptions included the 24 to 48 hpf tPOD when considering mortality when exposure commenced at 24 hpf, and the 6 hpf exposure start time in the 2 light phases of the behavior assay.

For PFOA and PFHxS, because low levels of mortality were detected in most exposure windows, the tPOD was the more sensitive endpoint. It was possible to estimate a BMC for PFHxS mortality when exposure begun at 6 hpf, and the corresponding tPOD estimates were lower than the calculated aPOD (Fig. 4A). When focusing on the effect of PFOA exposure on swim bladder inflation, tPODs were lower than the aPODs when exposure started at 24 and 96 hpf, but not 48 hpf (Fig. 4B). For PFHxS, the tPOD was greater than the SB inflation-aPOD when exposure commenced at 96 hpf.

To evaluate if the zebrafish tPOD estimates from this study are protective of zebrafish toxicity data for PFOS, PFOA, and PFHxS from other studies, we compared the tPODs calculated here with the zebrafish effect concentrations compiled in the ECOTOX Knowledgebase (Olker et al. 2022). Methods for searching ECOTOX closely followed 2 previous studies in fathead minnow (Villeneuve et al. 2023, 2024b) (Information SM.2) and included effects coded as mortality, reproduction, behavior, population, growth, development, morphology, and intoxication in the ECOTOX Knowledgebase [“Tier 1” in (Villeneuve et al. 2023)]. After filtering of results (Information SM.2), the search for effects from PFOS yielded 601 results from 50 references; PFOA had 388 results from 48 references; and PFHxS had 34 results from 5 references. tPODs from all windows of exposure for PFOS and PFOA, except for the tPOD from 6 to 24 hpf, were within the interquartile range of the effect concentrations generated from ECOTOX for each chemical (Fig. 5), except for the PFOS population and reproduction effects. The PFOA tPODs in particular were generally lower than the median of the ECOTOX effect concentrations, when effects concentrations from all effects are considered, and also when focusing on only development, morphology, and mortality (Fig. 5). We also note that the median effects concentrations for only behavior, population, and reproduction effects for PFOS, and behavior, growth, and reproduction effects for PFOA were lower (more sensitive) than majority of the tPODs for each chemical. For PFHxS, only the 6 to 24 and 6 to 48 hpf tPODs were both higher than the median and outside the interquartile range of the ECOTOX effect concentrations, for all effects, and also for behavior, development, and mortality, wheeas the 96 to 120 and 6 to 120 hpf tPODs were lower than the “all effects” effects concentrations. Of note, effect concentrations reported in ECOTOX were not points of departure based on benchmark responses, which was the case with the majority of the aPODs calculated in this study. Rather, the ECOTOX values constituted a mixture of LCx, ECx, LOECs, etc., and so all comparisons of results should be interpreted with caution.

Fig. 5.

tPOD estimates (dashed lines) for each zebrafish exposure window for PFOS, PFOA, and PFHxS, along with zebrafish ECOTOX records for each PFAS (box and whisker plot) for all Tier 1 effects combined, as well as individual effects. The center line of each box and whisker plot = median; box = interquartile range; whiskers = 1.5× interquartile range; black and gray points = values outside 1.5× the interquartile range.

Discussion

Overall study goals were to evaluate zebrafish tPOD robustness across exposure start times and durations (from 6 to 120 hpf), assess tPODs protectiveness of zebrafish apical outcomes within this study and from the literature, and compare the newly derived zebrafish tPODs with recently published PFAS tPODs in fathead minnow and D. magna. We chose 120 hpf because most of major organ systems are well-developed by then (Kimmel et al. 1995), and many phenotypic endpoints, including behavior, are typically measured at or after 120 hpf (Gaballah et al. 2020). Early time points were included to determine if tPOD estimates from early and short windows of exposure would be protective of tPOD estimates for later developmental outcomes.

Comparison between zebrafish tPODs and aPODs

The relationship between tPODs and aPODs was dependent on chemical, exposure start, and endpoint, and specifically for PFOS, tPOD estimates were generally lower, and thus protective, compared to aPODs. When lower, tPODs were around 1 to 2 orders of magnitude lower (Fig. 4). Our results are comparable to recent fathead minnow studies comparing whole-animal tPODs derived from short-term exposures (up to 7 d of several chemicals to apical outcomes either evaluated in the study (Alcaraz et al. 2022) or from previous literature (Alcaraz et al. 2021; Martin et al. 2023; Villeneuve et al. 2023; Flynn et al. 2024). Similar relationships have been described in rainbow trout cell lines (Mittal et al. 2022), and in specific tissues (Pagé-Larivière et al. 2019); both studies reported tPODs approximately an order of magnitude lower than apical endpoints. Overall, our zebrafish developmental toxicity study with PFOS, PFOA, and PFHxS provides strong support for using zebrafish PFAS transcriptomic data for quantitative risk assessment. Although mortality and neurobehavior, especially for the more bioactive PFOS, were reliable endpoints for aPODs, it is possible that the chemical exposure lasting only 24 h at a later developmental life stage did not manifest significant malformations relative to other exposure windows. Previous studies have demonstrated that malformation severity during early zebrafish development varies with life stage and exposure duration (Chen et al. 2016; Dasgupta et al. 2020; Robinson et al. 2020; Curcio et al. 2022). It is possible that our low sample size (10 individuals/exposure window/PFAS concentration), compared to 6 to 102 individuals/chemical concentration in Gaballah et al., for example, wasn’t enough to detect effects for the later exposure windows for PFOS neurobehavior, and majority of the endpoints for PFOA and PFHxS. Including multiple concentrations (8/exposure window) improved our tPOD estimates, but resulted in fewer fish than ideal for high-throughput developmental toxicity characterization.

Although comparisons within a study are less confounded and can provide for more certain conclusions, it is informative to understand how broadly our tPODs compare to toxicity results from other exposure scenarios. In this regard, our PFAS tPODs were generally within the interquartile range of the effects concentrations from other zebrafish studies of varying exposure start times and durations; Fig. 5. Although all PFOA tPODs were protective of the ECOTOX effects when all effects were considered, exposure window was important for determining that relationship for PFOS and PFHxS, and for individual effect categories. Similarly, our zebrafish tPODs for the 3 PFAS were generally within a factor of 5 of the median of effects concentrations from all fish species for both mortality and sublethal effects compiled from ECOTOX and presented previously (Villeneuve et al. 2024b). Overall, the comparisons between the tPODs generated in this study with mortality and sublethal apical endpoints for both zebrafish and other fish species from the literature emphasizes that tPODs derived from developmental zebrafish are not overly conservative but are consistent with the central tendency of effect concentrations observed over a range of studies. Additionally, tPODs tend to be more sensitive than PODs derived from traditional endpoints, despite the variability in apical endpoints considered, variety of chemicals investigated, and the various exposure scenarios, species, and life stages examined (Villeneuve et al. 2024a, 2024b). Sometimes tPODs are greater than aPODs; for example, PFOS behavior, population, and reproductive effects in this study (Fig. 5), and few sublethal effect concentrations from ECOTOX (Villeneuve et al. 2024b), suggesting the importance of experimental design in relationships between apical endpoints and tPOD estimations.

Effects of timing and duration of chemical exposure on zebrafish tPODs

In general, our tPOD estimates were fairly consistent regardless of duration or window of exposure (Fig. 2), unlike the aPODs (Fig. 3). A notable exception was the shortest and earliest window (6 to 24 hpf) for PFOS and PFOA; its strong transcriptional potency (2 to 3 orders of magnitude lower than other tPODs) was corroborated by the corresponding high transcriptional activity, where 69 and 116 genes, respectively, passed our filters and thus demonstrated a modeled concentration response (Table 2). One possible explanation is that early development includes an intense period of cell division and differentiation, with blastulation, gastrulation, and somitogenesis all occurring by 24 hpf (Kimmel et al. 1995). Although other chemical classes have also been shown to cause toxicity during this sensitive period (Goodale et al. 2013; Chen et al. 2016), PFHxS is a possible outlier likely due to toxicokinetic differences. One study featuring PFDMO2OA, a proposed PFOA substitute, identified one differentially expressed gene (DEG) at 24 hpf, and 3268 at 120 hpf (Chen et al. 2024). The finding suggests significant transcriptional alterations in early zebrafish development may not be completely chemical agnostic, and similar to our results for PFOS and PFHxS, the highest transcriptional activity may occur at the end of the longest exposure (6 to 120 hpf) (Table 2). The 6 to 120 hpf tPODs for all 3 PFAS were around one order of magnitude lower than the tPODs from other exposure windows ending at 120 hpf (24–120, 48–120, and 96–120 hpf), suggesting a combination of chemical bioaccumulation and exposure timeline can impact transcriptional responses, along with changes in chorion physiology and hatching dynamics (Vogs et al. 2019; Mylroie et al. 2021). Although we did not measure PFAS uptake, exploring its correlation with tPODs would be valuable. Nevertheless, our findings suggest tPODs from early, short exposure windows (e.g. 6 to 24 hpf) may be too conservative and variable for PFAS risk assessment.

More intriguing is the overall stability of the remaining PFAS tPODs across the different exposure timelines. One zebrafish study reported similar tPODs for methyl-, ethyl-, and propyl-parabens despite vastly different sensitivities for apical outcomes (Tran et al. 2023) emphasizing the inherent nature of tPOD estimations as bioactivity thresholds rather than an indication of specific biological mechanisms. To our knowledge, this is the first study evaluating exposure design impacts on tPODs, but one study concluded that of the different analyses methods, utilization of the most sensitive genes resulted in the most stable tPOD estimates (Pagé-Larivière et al. 2019). It is noteworthy to compare tPODs from our study with recent zebrafish studies. Mahoney et al., used the 20th gene (0.003 mg/l), gene mode (0.025 mg/l), and 10th percentile of all genes (0.002 mg/l) methods after PFOS exposure from 24 to 96 hpf (Mahoney et al. 2024). Another study determined the tPOD using the lower confidence limit of the benchmark dose to be 0.011 mg/l after an exposure from 48 to 120 hpf (Martínez et al. 2019). Only our 6 to 24 hpf tPOD (0.01 mg/l) was similar to the published tPODs, with our next highest tPOD (6 to 48 hpf; 0.92 mg/l) being at least an order of magnitude greater. Despite discrepancies in exposure design, such as, exposure concentrations and analysis methods, our tPODs are on par with previously published values. Future studies may clarify factors influencing tPOD variability, but our data suggest that exposure window has minimal impact on PFAS tPOD estimates, especially after the first 24 h of zebrafish development.

Comparison of zebrafish tPODs with previously published fathead minnow and Daphnia magna tPODs

Despite the differences in experimental designs (e.g. the fathead minnow study leveraged a reduced transcriptome of 1832 genes rather than full transcriptome and exposed fish for 24 h from 5 to 6 d post fertilization (Villeneuve et al. 2024b) whereas the current study measured a whole-transcriptome response in fish exposed up to 5 d post fertilization [120 hpf]), the tPODs were similar between the 2 species for all 3 PFAS (Fig. 2). In contrast, the D. magna tPODs (Villeneuve et al. 2024a) for PFOA and PFHxS were 2 to 3 orders of magnitude lower than those derived for zebrafish and fathead minnows. The fact that zebrafish tPODs, based on full transcriptome RNAseq, were similar to the fathead minnow tPODs suggests that the greater sensitivity of D. magna was physiologically based, rather than an artifact of the different number of features considered in the analyses. Overall, identification of the similarities between tPODs from different fish species may be the first step towards being able to leverage tPODs for cross-species extrapolations and chemical susceptibility predictions for individual PFAS and beyond for other chemicals.

Conclusions

Although unanswered questions remain, our analysis focused on gene-agnostic tPOD estimates without examining specific CRGs or DEGs, or conducting gene set-based methods grouping CRGs into pathways (National Toxicology Program 2018). Raw data are publicly available for additional analyses. Nonetheless, we addressed several key research questions related to the use of tPODs for hazard evaluations. First, we showed that tPODs generated from developmental zebrafish PFAS exposures were relatively stable across varying exposure scenarios, compared with corresponding apical endpoints which were significantly influenced by exposure start and duration. Second, although the tPODs were protective, but not overly conservative of mortality and sublethal endpoints identified in this study, they were within range of the effect concentrations from other zebrafish studies. Finally, our zebrafish tPODs were similar to previously published fathead minnow tPODs, emphasizing that the high throughput-compatible, short-term, in vivo developmental zebrafish platform is a feasible tool for detecting dose-response gene expression responses that can be leveraged for screening level ecological assessments.

Supplementary Material

Supplementary material is available at Toxicological Sciences online.

Impact Statement.

The current study shows that PFAS tPODs are generally protective of apical outcomes from short-term developmental zebrafish exposures, in addition to the tPODs being robust to differences in exposure start time and duration. Thus, the high-throughput zebrafish model is a feasible tool for detecting dose-dependent gene expression responses that can be leveraged for higher-tier assessments.

Data availability

All transcriptomic raw data are available in the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA; Bioproject ID PRJNA1011844), and all morphology and behavior raw data can be accessed through Data.gov (DOI: 10.23719/1531975).