Triphenyl phosphate induces coordinated DNA methylation and gene expression changes in aquatic embryonic cells

Abstract

Triphenyl phosphate (TPHP) is a high-production-volume flame retardant and plasticizer that is widely detected in the environment and in biomonitoring studies. TPHP exposure has been linked to endocrine disruption, metabolic disruption, genotoxicity, and neurodevelopmental effects in vitro and in vivo. The diverse toxicological outcomes across studies suggest disruption of fundamental regulatory processes, such as epigenetic control of gene expression. Here, we used an immortalized embryonic cell line derived from steelhead trout (STE-137) to investigate coordinated transcriptional and epigenetic responses to TPHP exposure. Cells were exposed to 0 or 80 μM TPHP for 24 h, followed by RNA sequencing (RNA-seq) or whole-genome bisulfite sequencing (WGBS). Differential expression analysis identified 1622 significant genes, with significant enrichment of DNA replication and repair, cell cycle regulation, and endocrine signaling pathways and prominent lipid metabolism genes. Weighted gene co-expression network analysis revealed modules highly correlated with exposure, including those enriched for protein and amino acid metabolism, ion transport, and genomic stability. WGBS methylome analysis detected 382 differentially methylated regions (DMRs), the majority hypermethylated and within gene bodies. Notable alterations included a DMR in the htr2cl1 gene, encoding a serotonin receptor, and the brca2 gene, a key DNA damage enzyme. Integration of RNA-seq and WGBS datasets identified nine genes with both expression and methylation changes, alongside altered gene expression of several key epigenetic regulators. Our results provide molecular evidence for early initiating events relevant to adverse outcome pathways and highlight the importance of epigenetic endpoints in developmental toxicity assessment.

Introduction

Triphenyl phosphate (TPHP; also called TPP) is an industrial chemical produced in high volumes. It is one of the most widely used flame retardants and plasticizers globally, with reported US production and import volumes ranging from 1 million to 10 million pounds annually over the last decade [1]. It is found in two common industrial products: polyvinyl chloride (PVC) plastics and the Firemaster® 550 flame retardant mixture [2]. Due to its high production and usage, TPHP and its predominant metabolite diphenyl phosphate (DPHP) are ubiquitously detected in both the environment, including rivers, drinking water, air, indoor dust [3, 4–6], and biotic samples worldwide [7]. TPHP has been detected in house dust at levels as high as 9810 ng/g in China [8, 9], and in surface water, as high as 13.2 ng/l in China [10]. TPHP can be absorbed via ingestion, inhalation, and dermal absorption of dust [11, 12], and is then metabolized quickly to diphenyl phosphate and hydroxylated metabolites in the liver [13, 14]. As a result, high levels of TPHP or DPHP are frequently detected in human samples, including blood serum, urine, breast milk, and even cerebrospinal fluid [15–19]. While TPHP is rapidly metabolized [20] and is not considered to be persistent nor bioaccumulative by several environmental regulatory bodies [21, 22], the consistent and globally ubiquitous use of this chemical likely causes life-long exposure to both humans and ecosystems at large. Environmental pollutant exposure at critical periods of development can result in later-in-life health effects. Therefore, the health effects of TPHP exposure during embryonic and fetal development are critically in need of further investigation.

There is growing evidence in the literature that TPHP causes a multiplex of toxicities. Firstly, TPHP has been shown to act as an endocrine disrupting chemical in estrogenic and thyroid pathways, as reviewed by Hu et al. (2023) [23]. There is also evidence that TPHP causes metabolic disruptions and obesogenic phenotypes through altered glucose and lipid balances, insulin signaling, and gut-microbiota dysbiosis across a variety of in vivo and in vitro models [24–28]. Some studies have shown TPHP-induced oxidative stress, mitochondrial membrane damage, and DNA damage effects, including in mouse spermatocyte and testicular Leydig cells [29, 30]. At levels as low as 1 μM, TPHP was shown to increase the proliferation and migration of a human colorectal cancer cell line [31]. Finally, neurotoxicity in the form of behavioral and brain morphological abnormalities as well as altered serotonin signaling has been observed [32–35]. A literature search of toxicity outcomes associated with TPHP exposure yields many interconnected and sometimes conflicting results, likely due to the different species models, metabolic capabilities, durations of exposure, concentrations, and even experimental design considerations. For example, regarding the estrogenic endocrine disrupting capacity of TPHP, several studies across in vitro and in vivo models have found TPHP to have pro‐estrogenic effects [36–38], while others have found anti‐estrogenic effects [36, 39]. Regardless of these differences, one thing remains clear; TPHP does not appear to be toxicologically inert across a range of concentrations and species. Furthermore, while there is mounting evidence for its toxicity, TPHP is not regulated as a toxic substance in Canada and remains under evaluation in the United States [40, 1]. The establishment of clear adverse outcome pathways (AOPs) for TPHP likely needs to be achieved for further regulatory action to be taken.

The diverse toxicological outcomes observed across TPHP studies suggest the involvement of fundamental regulatory mechanisms that can simultaneously affect multiple biological systems. Epigenetic modifications represent one such mechanism capable of coordinating widespread transcriptional changes during critical developmental windows. Studies have linked epigenetic alterations to the developmental origins of health and diseases (DOHaD), namely through the ability to regulate the transcriptome. DNA methylation is a type of epigenetic modification, commonly involving the covalent addition of a methyl group (−CH₃) to the 5′ carbon of cytosine residues at CpG sites. DNA methylation has been shown to result in transcriptional control of certain genes by preventing transcription factor binding to promoter or regulatory regions [41–43]. However, the impact of DNA methylation on the directionality of gene expression appears to be site, gene, and transcription factor-dependant [42, 44–46]. The epigenome and transcriptome are in a complex and dynamic interplay throughout an organism’s life course. However, as per the DOHaD hypothesis, embryonic development is an especially sensitive time for the methylome, and various environmental stressors can interfere with this process, which may lead to adverse phenotypes, such as metabolic disorders [47–49]. Therefore, it is important to identify how environmental toxicants may affect the epigenome, which may offer insights to the wide range of TPHP’s toxicological outcomes observed across many studies.

While teleost fish such as steelhead trout (Oncorhynch us mykiss) and zebrafish (Danio rerio) are established model organisms in the fields of ecotoxicology and developmental toxicology [50–54], the use of alternative in vitro models can contribute positively to the AOP framework. While whole-organism models capture integrated, physiological responses, in vitro models are useful for controlled and cost-effective investigation of early molecular initiating events that can be subsequently investigated in targeted in vivo experiments. Here, we rely upon an aquatic embryonic cell model to allow us to investigate novel genome-wide coordinated methylation [using whole-genome bisulfite sequencing (WGBS)] and transcriptional [RNA sequencing (RNA-seq)] changes following TPHP exposure.

The objective of this study was to determine whether acute TPHP exposure induces coordinated transcriptomic and DNA methylation changes in embryonic cells and to identify molecular pathways that may represent early molecular initiating events within AOPs relevant to TPHP-induced developmental toxicity, as depicted in Fig. 1. To our knowledge, this is the first study to map coordinated transcriptional and methylation changes following TPHP exposure in an embryonic model. These findings provide mechanistic anchors for constructing AOPs relevant to TPHP developmental toxicity.

Figure 1.

Conceptual framework linking triphenyl phosphate exposure to developmental toxicity through epigenetic and transcriptomic mechanisms. Developmental exposure to the flame retardant triphenyl phosphate (TPHP) is hypothesized to induce coordinated changes in DNA methylation patterns and gene expression profiles in embryonic cells. These molecular alterations may represent initiating events in AOPs leading to diverse toxicological endpoints, including metabolic syndromes, endocrine disruption, neurotoxicity, oxidative stress, and increased cancer risk.

Results

Differentially expressed genes

To explore early molecular markers of disease risk, we analysed differentially expressed genes (DEGs) in embryonic cells derived from steelhead trout (STE-137) between those exposed to 80 μM of TPHP for 24 h compared to control (0 μM). PCA plot and volcano plot analysis revealed a total of 1622 significant DEGs were identified, with 40.9% (663) showing downregulation and 59.1% (959) showing upregulation in the exposed group (Fig. 2A and B). A heatmap of Z-scores for the top 50 most significant DEGs by sample demonstrated clear clustering of samples between control and exposed groups as well as between upregulated and downregulated DEGs (Fig. 3). The most significantly altered gene was soat2, which encodes for the sterol-O-acyltransferase 2 enzyme, with a 328-fold (8.37 log₂ fold change) increase in expression (adjusted P-value = 2.07 × 10⁻⁴²). Also notable was cidec, encoding for the cell death inducing DFFA like effector c enzyme, with a 50-fold (5.67 log₂ fold change) increase in expression (adjusted P-value = 2.07 × 10⁻⁴²) (Table 1). The top 10 most upregulated and downregulated DEGs can be found in Supplementary Tables 1 and 2.

Figure 2.

Transcriptomic response to TPHP exposure in the STE-137 steelhead trout embryonic cell line. (A) PCA of variance-stabilized transformed gene expression data following batch correction with ComBat-seq. STE-137 cells were exposed to either 0 μM (control, blue circles, n = 9) or 80 μM TPHP (exposed, red circles, n = 8) for 24 h. PC1 and PC2 show 22.65% and 9.95% of total variance. (B) Volcano plot displaying DGE analysis results from DESeq2 package analysis in R. Each point represents one gene, with log2 fold change on the x-axis and −log10 adjusted P-value on the y-axis. Red points indicate significantly DEGs (adjusted P-value < .05, Benjamini–Hochberg correction). A total of 1622 genes were significantly altered, with 959 upregulated and 663 downregulated in TPHP-exposed cells compared to controls.

Figure 3.

Heatmap of the top 50 most significantly DEGs following TPHP exposure. Each column represents an individual biological replicate, and each row represents a gene. Z-score normalized gene expression values are shown for each sample (columns) and gene (rows) following 24-h exposure to either vehicle control (0 μM TPHP, n = 9) or 80 μM TPHP (n = 8). Samples are annotated on the x-axis by experimental batch (top bar, cyan/magenta) and treatment condition (second bar, orange = control 0 μM, green = exposed 80 μM TPHP). Gene symbols are displayed where available, with NCBI gene identifiers shown for unannotated genes. Hierarchical clustering reveals distinct expression patterns, with clear separation between control and exposed groups, as well as between upregulated (upper cluster, blue tones) and downregulated genes (lower cluster, red/orange tones). Color scale represents standardized expression levels from low (blue, −2) to high (red, +2). Analysis performed using DESeq2 with Benjamini–Hochberg multiple testing correction (adjusted P < .05).

Table 1.

Top 10 most significantly DEGs in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP.

Gene name	Gene description	Log2 fold change	Adjusted P-value	Predicted protein function
soat2	Sterol O-acyltransferase 2	8.37	2.07E-42	Esterifies free cholesterol with long-chain fatty acyl-CoA
cidec	Cell death inducing DFFA like effector c	5.67	2.07E-42	Promotes lipid accumulation in adipocytes and suppresses lipolysis
ddit3	DNA-damage-inducible transcript 3	2.74	5.49E-27	Transcription factor implicated in adipogenesis and erythropoiesis, is activated by endoplasmic reticulum stress, and promotes apoptosis
mcm2	Minichromosome maintenance complex component 2	−1.86	1.02E-26	Involved in the initiation of eukaryotic genome replication
LOC110528856 (gtpbp2)	GTP-binding protein 2	2.92	2.00E-25	Binds GTP and enables G proteins to perform a wide range of biologic activities
mcm3	Minichromosome maintenance complex component 3	−2.11	1.80E-24	Involved in the initiation of eukaryotic genome replication
ywhag1	3-monooxygenase/tryptophan 5-monooxygenase activation protein, gamma polypeptide 1	2.56	3.17E-24	Involved in various signal transduction pathways and neuron development
cdca7a	Cell division cycle associated 7a	−2.567	6.67E-27	Responsive to c-Myc, involved with cell proliferation, tumour formation and stable inheritance of DNA methylation during cell division
LOC110504691 (ddit4l)	DNA damage-inducible transcript 4-like protein	2.43	1.54E-26	Inhibits cell growth by regulating TOR signaling pathway
LOC110486081 (hnrnpll)	Heterogeneous nuclear ribonucleoprotein L-like	4.11	9.90E-25	Master regulator of alternative splicing in T-cells

Gene names, descriptions, log2 fold changes, adjusted P-values, and predicted protein functions are shown for the 10 genes with the lowest adjusted P-values from differential expression analysis. Cells were exposed to either vehicle control (0 μM TPHP, n = 9 biological replicates) or 80 μM TPHP (n = 8 biological replicates) for 24 h. Differential expression analysis was performed using DESeq2 with Benjamini–Hochberg multiple testing correction; all genes shown have adjusted P-value < .05. Positive log2 fold change values indicate upregulation in TPHP-exposed cells; negative values indicate downregulation. Predicted protein functions are based on orthologous gene annotations from NCBI Gene, UniProt, and ENSEMBL databases.

Pathway enrichment analysis on DEGs

Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis revealed that the 1622 DEGs were associated with specific biological processes, cellular components, and molecular functions (Fig. 4A and B). Top 12–15 most significant GO and KEGG terms are namely involved with DNA replication, fidelity and repair, cell cycle control, and endocrine signaling (oocyte meiosis and gonadotropin-releasing hormone signaling pathway) (Fig. 4A and B).

Figure 4.

Functional enrichment analysis of DEGs in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP. GO biological process and KEGG pathway enrichment analyses were performed on 1622 significantly DEGs (adjusted P < .05) using the clusterProfiler package in R with Benjamini–Hochberg correction. (A) Top 15 most significant enriched GO biological process terms, ranked by gene count. Bar color represents adjusted P-value. Major enriched processes include DNA replication, chromosome organization, DNA repair mechanisms, cell cycle regulation, and meiotic processes. (B) Top 12 most significant enriched KEGG pathways, ranked by gene count. Significantly affected pathways include cell cycle control, DNA replication, mismatch repair, oocyte meiosis, calcium signaling, endocrine signaling (GnRH pathway), and proteasome function. Count indicates the number of DEGs mapping to each term/pathway.

WGCNA analysis

To capture coordinated transcriptional responses beyond individual DEGs, we applied weighted gene co-expression network analysis (WGCNA). The WGCNA algorithm groups genes into modules based on correlated expression patterns across all samples. Genes within a module exhibit proportional changes relative to each other both within and between treatment groups, identifying coordinated biological responses independent of individual gene significance thresholds. By correlating these modules with TPHP exposure, we uncovered network-level responses, some involving genes with subtle expression changes. We found a total of 6 significant modules, with two showing very high correlation with TPHP exposure (Fig. 5A). Module 2 showed a positive correlation with TPHP exposure (r = 0.96), while Module 3 showed a negative correlation with exposure (r = −0.89), indicating these modules are disrupted in opposite directions. Module 2 contained 1134 significant genes (kME > 0.8 and P < .05), while Module 3 contained 1074. The top 10 hub genes within Modules 2 and 3 by module membership (kME) can be found in Supplementary Tables 3 and 4. GO and KEGG pathway enrichment analysis on these two most significant modules revealed several overlapping and several novel implicated pathways compared to the standard DEG analysis alone (Fig. 5B–E). Module 2 revealed protein and amino acid maturation, metabolism, and transport, as well as metal ion transport and sequestering as the main positively associated pathways by TPHP exposure (Fig. 5B and C). Module 3 showed cell cycle control, DNA replication, fidelity, and repair as the main pathways negatively implicated by TPHP exposure, along with oocyte meiosis and motor protein disturbances (Fig. 5D and E).

Figure 5.

WGCNA identified gene modules correlated with exposure in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP. (A) Module-trait correlation heatmap showing Pearson correlation coefficients between six identified gene modules (ME0-ME5) and TPHP exposure status. Each cell displays the correlation coefficient (top number) and P-value (bottom number, in parentheses). Module 2 (ME2) shows a strong positive correlation with TPHP exposure (r = 0.96, P = 7 × 10⁻¹¹), while Module 3 (ME3) shows a strong negative correlation (r = −0.89, P = 4 × 10⁻⁷). WGCNA was performed on the top 7000 most variable genes. (B) GO biological process enrichment for Module 2 genes. (C) KEGG pathway enrichment for Module 2 genes. (D) GO biological process enrichment for Module 3 genes. (E) KEGG pathway enrichment for Module 3 genes. All enrichment analyses performed with Benjamini–Hochberg correction (adjusted P < .05).

Global and targeted methylome changes

We investigated global and targeted epigenetic consequences to the methylome in STE-137 following exposure to 80 µM of TPHP in comparison to control (0 µM) using WGBS. Initial PCA plot of DMRs showed clear separation of control and exposed samples (Fig. 6A). Differentially methylated region (DMR) analysis revealed a total of 382 significant DMRs with a minimum methylation difference of 10%, with 250 (65.4%) being hypermethylated and 132 (34.6%) being hypomethylated (Fig. 6B). Genome-wide average CpG methylation, calculated as the mean percentage of methylated reads across all filtered CpG sites per sample, showed no significant difference between control and TPHP-exposed groups (Fig. 6C). Coverage analysis data for both CpG site and DMR read depth can be found in Supplementary Table 4A and B. A heatmap of Z-scores for the top 50 most significant DMRs by sample and annotated by the closest gene demonstrated clustering of samples between control and exposed groups as well as between increased and decreased methylation (Fig. 7).

Figure 6.

DNA methylation changes at the global and regional levels in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP. (A) PCA plot of 382 significantly DMRs showing clear separation between control (0 μM, blue circles, n = 7) and TPHP-exposed (80 μM, red circles, n = 7) samples. (B) Volcano plot of DMR analysis displaying methylation difference percentage (x-axis) versus −log10 q-value (y-axis). Red points indicate significant DMRs (q ≤ 0.01). Vertical dashed lines mark ± 10% methylation difference thresholds. Of 382 significant DMRs, 250 (65.4%) were hypermethylated and 132 (34.6%) were hypomethylated in TPHP-exposed cells. DMRs were identified using a sliding window approach (500 bp windows, 250 bp step size) with minimum 10 × coverage in ≥ 4/7 samples per group. (C) Genome-wide percentage of methylated CpG sites across all filtered CpG sites per sample. No significant difference between groups (unpaired t-test, P > .05; n = 7 per group, mean ± SD).

Figure 7.

Heatmap of the top 50 most significantly DMRs in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHPZ-score normalized methylation values are shown for each sample (columns) and DMR (rows). Samples are annotated by treatment condition (top bar: green = control 0 μM, orange = exposed 80 μM TPHP), genomic region type (second bar: green = gene body, blue = upstream, pink = downstream, purple = intergenic), closest associated gene symbol (where available), and average methylation difference percentage between control and exposed groups. Hierarchical clustering represents methylation patterns by treatment. Color scale represents standardized methylation levels. Significant DMRs were identified using methylKit (q ≤ 0.01, absolute methylation difference ≥ 10%, n = 7 per group).

Most DMRs (249, 65.2%) overlapped with a known gene body (Fig. 8A). Further stratifying this group, 147 gene body DMRs were located within introns, while 102 were located within exons (Fig. 8B). Additionally, 29 DMRs were found to be located within upstream promoter regions (≤5 kb from transcriptional start site) (Fig. 8C). The most significantly altered DMR was found to overlap with the htr2cl1 gene, with an 88.5% decrease in methylation in the exposed group compared to control (Table 2). The htr2cl1 gene encodes for the 5-hydroxytryptamine receptor 2C. Notably, several of the most significant gene body DMRs in Table 2 are predicted to be associated with genes involved in central nervous system function, including as synaptic scaffolding proteins (magi2) [55], serotonin signaling (htr2cl1) [56] and glutamate receptor regulation (cnih3) [57] and blood-brain barrier integrity (tjp3) [58], suggesting a coordinated disruption of neurodevelopmental pathways. The top 10 most significant DMRs regardless of genomic location, most hypermethylated and most hypomethylated DMRs can be found in Supplementary Tables 5–7, respectively. Pathway enrichment analysis using clusterProfiler [59] on DMR-associated genes yielded no significant results. The number of DMR-associated genes was limited, which may have reduced statistical power for enrichment analysis.

Figure 8.

Genomic distribution of DMRs in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP. (A) Broad genomic classification of all 382 significant DMRs. The majority (249 DMRs, 65.2%) overlap with gene bodies, 29 (7.6%) are in promoter regions (≤5 kb upstream of transcriptional start site), 13 (3.4%) are downstream (≤5 kb past transcriptional end site), and 91 (23.8%) are intergenic. (B) Subcategory distribution of the 29 upstream promoter-associated DMRs by distance from transcriptional start site: 0–100 bp (1 DMR), 100–500 bp (4 DMRs), 500–1 kb (9 DMRs), 1–5 kb (15 DMRs). (C) Subcategory distribution of the 249 gene body DMRs by genomic feature: 147 (59.0%) in introns and 102 (41.0%) in exons. DMR annotations and distance to nearest genes were identified using the USDA_OmykA_1.1 steelhead trout genome assembly (GCF_013265735.2) with AnnotationHub and biomaRt packages, supplemented by manual curation using NCBI Gene database.

Table 2.

Top 10 most significantly DMRs overlapping with gene bodies in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM triphenyl phosphate.

Gene name	Gene description	Meth Diff %	Adjusted P-value	Predicted protein function
LOC110531418 (htr2cl1)	5-hydroxytryptamine receptor 2C	−88.489	6.56E-37	Encodes a receptor responding to the neurotransmitter serotonin
LOC110503715 (ccdc9)	Coiled-coil domain-containing protein 9	98.46	2.75E-30	Enables RNA binding activity
LOC110490483 (magi2)	Membrane-associated guanylate kinase, WW and PDZ domain-containing protein 2	46.51	5.07E-11	Interacts with atrophin-1 protein; function unknown
tjp3	Tight junction protein 3	−40.81	3.05E-08	Scaffold proteins that plays a role in cell cycle progression
LOC110490075 (ap2m1b)	AP-2 complex subunit mu	−43.20	3.58E-06	Involved in membrane trafficking pathways, including clathrin-dependent endocytosis
LOC110500009	Putative ferric-chelate reductase 1	−33.82	3.84E-06	Involved in reducing extracellular copper iron prior to import
LOC110529662 (cnih3)	Protein cornichon homolog 3	22.63	6.61E-06	Regulates activation/properties of AMPA-selective glutamate receptors
LOC110527506 (nlgn4x)	Neuroligin-4, X-linked	−23.12	7.12E-06	Involved in cell-cell interactions and may be involved in formation of central nervous system
txlnba	Taxilin beta a	26.27	9.28E-06	Predicted to enable syntaxin binding activity in membrane trafficking
LOC110498867 (znf420)	Zinc finger protein 420	−19.11	1.69E-05	Az inc finger protein that negatively-regulates p53-mediated apoptosis

Gene names, descriptions, methylation difference percentages, adjusted P-values (q-values), and predicted protein functions are shown for the 10 gene body-associated DMRs with the lowest q-values. Cells were exposed to either vehicle control (0 μM TPHP, n = 7) or 80 μM TPHP (n = 7) for 24 h. Significance criteria was a q-value ≤ 0.01 and absolute methylation difference ≥ 10%. Predicted protein functions are based on orthologous gene annotations from NCBI Gene, UniProt, and ENSEMBL databases.

Multi-omic integration of differential methylation and expression

We identified genes with coordinated methylation and expression changes. In total, there were nine genes that were identified as both DEGs and have DMRs overlapping with their gene body (Fig. 9); including cspp1 (LOC110508768), ap2m1 (LOC110490075), sgsm2 (LOC110519585), add3 (LOC110529161), utp3, igfbp6b, uqcrc1 (LOC110527932), brf1a, and slka. The brf1a gene, encoding the transcription factor IIIB 90 kDa subunit, was found to have two separate DMRs overlapping with the gene body. Of these nine genes, five showed hypomethylation with increased expression, three showed hypermethylation with increased expression, and one showed hypomethylation with decreased expression relative to controls.

Figure 9.

Multi-omic integration identifies genes with coordinated transcriptomic and epigenetic changes in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM TPHP. Scatter plot displaying the nine genes that show both significant differential expression (y-axis: log₂ fold change from RNA-seq) and overlapping gene body DMRs after significance filtering (x-axis: methylation difference percentage from WGBS). Gene names with LOC identifiers are shown for manually annotated loci. The brf1a gene appears twice due to two distinct DMRs within its gene body.

Discussion

Transcriptomic disruption of metabolic and cellular pathways

Exposure to the environmental pollutant TPHP for 24 h (80 μM) markedly altered the transcriptome of STE-137 embryonic cells, with 1622 genes showing significant differential expression. While this concentration exceeds typical environmental exposures, it was selected to induce sub-lethal effects that reveal molecular pathways of toxicity and falls within the range used in several other mechanistic studies examining TPHP’s mode of action. Enrichment analyses highlighted lipid metabolism, DNA replication and repair, cell cycle control, endocrine signaling, and cellular senescence as major affected pathways (Figs 4–5). These functional categories are highly relevant to embryogenesis, where coordinated metabolic and proliferative programming is required to support rapid growth and tissue differentiation.

Among the most strongly upregulated genes were soat2 (∼329-fold) and cidec (∼50-fold), both central to lipid storage and metabolic regulation. The soat2 gene encodes sterol O-acyltransferase 2, which esterifies free cholesterol with long-chain fatty acyl-CoA, thereby regulating cholesterol storage and homeostasis [60]. Cholesterol serves as a fundamental component of cell membranes, a precursor for steroid hormone synthesis, and is essential for neuronal synaptogenesis and sonic hedgehog signaling [61]. During embryogenesis, cholesterol alterations can lead to multiple developmental abnormalities. The dramatic upregulation observed in our study may perturb membrane stability, steroid hormone synthesis, and critical embryogenic signaling pathways, consistent with associations between soat2 dysregulation and both tumorigenic processes and metabolic alterations in mammalian models [62, 63]. The gene cidec encodes the cell death-inducing DFFA-like effector C protein, which promotes lipid accumulation in adipocytes and suppresses lipolysis [64]. CIDEC may also function as a tumor suppressor in human non-small cell carcinoma through modulation of lipid metabolism [65]. In fish (Ctenopharyngodon idella) adipocytes, cidec expression is regulated by cAMP-response element binding protein (CREB), a transcription factor controlling metabolic gene expression [66]. The dramatic upregulation of cidec in our study, combined with concurrent significant upregulation of CREB family members (creb3l3l and creb3l2) (Supplementary File 1), suggests that TPHP may activate cidec-associated metabolic pathways, potentially promoting lipid storage and adipocyte-like metabolic reprogramming. These findings provide molecular support for TPHP’s metabolic disrupting potential and align with studies linking TPHP exposure to obesogenic and diabetic phenotypes [24, 67].

Beyond metabolic disruption, TPHP altered the expression of numerous epigenetic regulators, strengthening evidence for an epigenetic basis of TPHP-induced toxicity. This includes a maintenance DNA methyltransferase (dnmt1), a DNA demethylase (tet1), histone chaperones (asf1bb), histone-binding proteins (rbbp), histone lysine methyltransferases (nsd2, mecom, suv39h1a), histone deacetylase complex subunits (sap18, dpy30), and a methyl-donor enzyme (mat2b) (Supplementary File 1). The mixed directionality of these changes suggests a broad shift in the epigenetic regulatory landscape. We and others have previously shown that global DNA methylation and several global histone modifications are altered following TPHP exposure in vitro and in vivo [68–70]. Taken together, these epigenome changes provide a mechanistic link between the observed transcriptomic reprogramming and the methylome alterations also explored in this study.

WGCNA identified two modules highly correlated with TPHP exposure: one positively correlated module enriched for protein and amino acid metabolism and metal ion transport/sequestration, and one negatively correlated module enriched for DNA replication, fidelity, and repair, as well as oocyte meiosis and motor protein function (Fig. 5). Collectively, the transcriptomic profile indicates that TPHP simultaneously disrupts lipid metabolic regulation, cell cycle control, and genomic stability, potentially through combined endocrine and metabolic mechanisms. These pathways are essential for normal embryogenesis, and their disruption by TPHP, including upregulation of lipid metabolism genes, alteration of DNA repair and cell cycle networks, and disturbance of reproductive endocrine signaling, aligns with reported metabolic, genotoxic, and endocrine effects in other studies.

Methylome changes in response to TPHP exposure

Given previous findings of altered DNA methylation profiles in three studies across three species following TPHP exposure (Germain & Winn, 2024; Negi et al., 2024; Shafique et al., 2023), we sought to further investigate the involvement of the methylome in TPHP toxicity. We identified 382 significant DMRs following TPHP exposure, with a predominance of hypermethylation (65.4%) over hypomethylation (34.6%). The majority of DMRs were located within gene bodies, particularly intronic regions, with fewer in promoter or downstream regions. Methylation within gene bodies can have variable effects on gene expression [71]. The most significantly altered DMR was found to overlap with the htr2cl1 gene, with an 88.5% decrease in methylation in the exposed group compared to control. This finding is particularly noteworthy as htr2cl1 encodes the 5-hydroxytryptamine receptor 2C, and the dramatic hypomethylation observed may contribute to the neurotoxic effects associated with TPHP exposure through altered serotonergic signaling pathways [32–35]. Additionally, we identified a notable gene body hypermethylated DMR in brca2 (+29.6% methylation difference) (Supplementary File 2). BRCA2 in humans is essential for homologous recombination–mediated repair of DNA double-strand breaks and maintenance of replication fork stability [72]. While there was a lack of significant pathway enrichment for DMR-associated genes, this was likely due to the low number of identified DMR-associated genes rather than having no biological relevance and reflects the difficulty of methylome analyses due to the inherent variability and dynamic nature of the epigenome.

The absence of changes to genome-wide CpG-site methylation observed in this WGBS analysis contrasts with our previous findings of decreased global 5-methylcytosine levels using ELISA-based quantification [68]. This discrepancy likely reflects methodological differences between the two approaches. The ELISA-based global methylation assay previously used specifically quantified 5-methylcytosine (5-mC) content across the entire genome, whereas WGBS detects both 5-mC and 5-hydroxymethylcytosine (5-hmC). As demethylation proceeds through the stepwise oxidation of 5-mC to 5-hmC via TET enzymes [73, 74], the global balance may have shifted toward the 5-hmC state at 24 h of exposure, maintaining apparent total methylation levels in WGBS while reducing true 5-mC content detected by ELISA. Nonetheless, the targeted DMR approach used here captures localized methylation changes that may be more functionally relevant than global shifts. The genomic distribution of DMRs primarily within gene bodies suggests targeted rather than global methylation alterations. These methylome changes occur alongside transcriptional modulation of key DNA methylation regulators, including downregulation of dnmt1 and upregulation of tet1. Decreased dnmt1 expression may impair maintenance methylation during replication, whereas increased tet1 could enhance demethylation via oxidation of 5-methylcytosine. Such opposing shifts in methylation machinery may underlie the mixture of hyper- and hypomethylated regions observed. While this study examined acute molecular responses at 24 h, the observed methylation changes may represent early markers of more persistent epigenetic reprogramming. The altered expression of key methylation regulators additionally suggests that TPHP exposure initiates processes that could maintain aberrant methylation patterns through subsequent cell divisions, potentially explaining how transient developmental exposures can result in lasting phenotypic changes.

Multi-omics integration

The intersection of DMRs and DEGs identified nine genes altered at both the methylation and transcriptional levels by TPHP exposure, each containing at least one DMR within its gene body. Most of these genes showed increased transcriptional expression with a mixture of hypo- and hypermethylated DMRs, supporting the concept that DNA methylation does not uniformly lead to gene silencing [75–78]. These genes have a variety of roles, as shown in Table 3. The relatively small overlap between DEGs and DMR-associated genes reflects the complex relationship between DNA methylation and gene expression, where methylation changes may not immediately translate to expression differences within the 24-h exposure window, and expression changes may occur through methylation-independent mechanisms. While direct causation cannot be determined from this experimental design, the co-occurrence of transcriptomic and methylome changes at several gene loci, combined with expression changes in epigenetic modifying enzymes, suggests a DNA methylation component to TPHP-induced toxicity. Importantly, the epigenetic disruption documented in this study may represent a key mechanism underlying TPHP’s developmental toxicity. Changes to DNA methylation profiles during embryonic development can lead to lasting alterations in the epigenome and transcriptome, potentially contributing to later-life health and disease outcomes. Embryonic development is characterized by dynamic epigenetic remodeling essential for proper cell fate specification and tissue differentiation [47–49]. Environmental interference with this process, as demonstrated by TPHP exposure, could establish aberrant epigenetic marks that persist through subsequent developmental stages and potentially into adulthood. The multi-omics approach employed in this study provides new mechanistic targets into TPHP toxicity that may explain the diverse adverse outcomes reported across different experimental systems. The coordinated disruption of metabolic regulation, cell cycle control, DNA repair mechanisms, and epigenetic programming suggests that TPHP acts through multiple interconnected pathways rather than a single mode of action.

Table 3.

Genes with coordinated transcriptomic and methylomic changes following TPHP exposure in STE-137 steelhead trout embryonic cells following 24-h exposure to 80 μM triphenyl phosphate.

Gene name	Gene description	Log2 fold change	Meth Diff %	Predicted protein function
LOC110508768 (cspp1)	Centrosome and spindle pole-associated protein 1	2.07	−67.44	Involved in cell-cycle-dependent microtubule organization
LOC110490075 (ap2m1)	AP-2 complex subunit mu	1.90	−43.20	Involved in membrane trafficking pathways, including clathrin-dependent endocytosis
LOC110519585 (sgsm2)	Small G protein signaling modulator 2	3.73	−21.08	Activates regulators of membrane trafficking
LOC110529161 (add3)	Gamma-adducin	1.32	−19.17	Cytoskeleton protein involved in cell–cell contact and spectrin-actin network
utp3	UTP3 small subunit processome component	0.55	−12.42	Involved in ribosomal biogenesis and enables RNA binding
igfbp6b	Insulin-like growth factor binding protein 6b	−1.47	−12.69	Part of the insulin-like growth factor signaling pathway
brf1a	BRF1 general transcription factor IIIB subunit a	1.28	22.77	Predicted to be part of transcription factor TFIIIB complex
brf1a	BRF1 general transcription factor IIIB subunit a	1.28	13.71	Predicted to be part of transcription factor TFIIIB complex
LOC110527932 (uqcrc1)	Cytochrome b-c1 complex subunit 1, mitochondrial	0.48	12.20	Part of the mitochondrial electron transport chain
slka	STE20-like kinase a	0.59	39.62	Predicted to enable protein serine/threonine kinase activity involved with apoptosis and cell migration

STE-137 cells were exposed to either vehicle control (0 μM TPHP) or 80 μM TPHP for 24 h. Nine genes showing both significant differential expression and significant gene body DMRs are listed with their descriptions, log2 fold changes from RNA-seq, methylation difference percentages from WGBS, and predicted protein functions. RNA-seq analysis included n = 8–9 biological replicates per group; WGBS analysis included n = 7 biological replicates per group, with separate experiments conducted for transcriptomic and methylomic analyses. Differential expression was determined using DESeq2 (adjusted P < .05, Benjamini–Hochberg correction). DMRs were identified using methylKit (q ≤ 0.01, absolute methylation difference ≥ 10%). Predicted protein functions are based on orthologous gene annotations from NCBI Gene, UniProt, and ENSEMBL databases.

Limitations and future directions

This study is limited to in vitro analysis using a single embryonic cell line and exposure condition, which provides mechanistic insights but cannot fully recapitulate the complexity, or inter-individual variability of in vivo developmental processes. However, embryonic cells represent a highly plastic developmental context in which epigenetic regulation is particularly dynamic, making this model well suited for detecting subtle, early molecular perturbations. Consistent with this, we previously observed no detectable epigenetic alterations following TPhP exposure in a differentiated gill epithelial cell line (RTgill-W1) [68], indicating the sensitivity of this cell line model. We expect that the transcriptional and epigenetic changes observed here reflect conserved molecular initiating events. Scaling these findings to the whole organism will require validation across tissue types and developmental stages, where compensatory mechanisms, metabolism, and systemic endocrine regulation may modulate the magnitude and persistence of these responses. Additionally, the single time point analysis (24 h) captures immediate coordinated molecular responses but does not address the complex temporal dynamics of epigenetic and transcriptomic changes.

The relatively low WGBS alignment rate (∼19.7%) is consistent with the complex salmonid genome and loss of complexity from bisulfite conversion. Because this study was conducted in an immortalized embryonic-derived cell line, divergences between cell-line genomic structure and whole-organism reference assemblies may further contribute to alignment challenges, reinforcing our use of conservative alignment and downstream filtering strategies. While sufficient coverage was achieved for DMR calling, future studies could aim to achieve greater individual CpG site read depth to allow site-specific analysis. Note that the WGCNA analysis was not performed on the WGBS dataset as the minimum number of cumulative samples across all groups recommended for this analysis is 15 [79, 80].

Additionally, the bioinformatics approach identifies candidate pathways and mechanisms that will require further validation through targeted functional studies. While this study focused on transcriptomic and epigenomic endpoints, protein-level validation and functional assays represent important next steps to confirm the biological consequences of the molecular changes observed. Validation of key pathways related to metabolic and endocrine signaling, cell cycle regulation, and DNA damage would strengthen causal interpretation of these early molecular markers and help link coordinated gene expression and methylation changes to functional outcomes. The establishment of AOPs linking the molecular initiating events identified in this study to adverse developmental outcomes observed in vivo represents an important research priority. Such mechanistic frameworks are essential for regulatory decision-making and risk assessment of TPHP and related chemicals.

Conclusions

Embryonic development is characterized by dynamic epigenetic remodeling essential for proper cell fate specification, growth, and tissue differentiation. Environmental interference with this process, as demonstrated by TPHP exposure, may establish aberrant epigenetic marks that persist through subsequent developmental stages and potentially into adulthood. This multi-omics analysis demonstrates that TPHP exposure induces coordinated disruption of several transcriptomic and epigenetic regulatory pathways in embryonic cells. The simultaneous alteration of genes involved with metabolic processes, genomic stability mechanisms, and epigenetic machinery provides molecular evidence for TPHP’s capacity to disrupt normal developmental processes through multiple interconnected pathways. These findings support the need for epigenetic assessment in developmental toxicity testing and provide nine genes as specific molecular targets for further investigation.

Materials and methods

Cell culture and exposure model

This study used the immortalized cell line STE‐137 (Cellosaurus STE-137, CVCL_4308), derived from pooled steelhead trout (Oncorhynchus mykiss irideus) embryonic tissue. This cell line was kindly provided to us from the United States Geological Survey Western Fisheries Research Center (Seattle, WA, USA) and has been previously described by Lannan et al. (1984) [81] and reviewed by Bols et al. (2017) [82]. Cells were maintained in supplemented growth medium consisting of 88% Leibovitz’s L-15 medium with L-glutamine (Sigma-Aldrich, St. Louis, MO, USA), 10% fetal bovine serum (Gibco, Grand Island, NY, USA), and 1% penicillin-streptomycin solution (Wisent Bioproducts, Saint-Jean-Baptiste, QC, Canada). Cultures were grown in T25 flasks with vented caps at 18°C until reaching 80%–90% confluency (∼250 000 cells/ml) at passage numbers 20–30.

Preparation of test chemical solution

For experimental exposures, TPHP (CAS No. 115-86-6, Sigma-Aldrich) was dissolved in dimethyl sulfoxide (DMSO, Sigma-Aldrich) to create a 100 mM stock solution, stored in darkness at room temperature for maximum 6 months. On the day of exposure, a 1 mM working stock was generated by diluting the storage stock 1:100 in fresh supplemented growth medium, which was then added to cell culture flasks to achieve the final exposure concentration of 80 or 0 μM (control) for 24 h. All exposure groups contained 0.08% (v/v) DMSO as the final vehicle concentration. The 80 μM concentration and 24 h exposure period were selected based on prior proliferation and viability experiments demonstrating that this exposure is sub-lethal and does not significantly reduce cell proliferation, while being sufficient to induce changes in global DNA methylation, histone modifications, and gene expression in this model [50, 68]. While estimates vary on detected levels in water systems, this exposure regimen is above environmentally relevant concentrations as reported in the literature [4, 10, 83–85]. It was selected to enable identification of early molecular initiating events relevant to ecological hazard assessment and AOP development. Each biological replicate represented an independent cell population in a separate culture flask exposed on different experimental days. Two separate experiments were conducted for transcriptomic and methylomic analyses using different cell aliquots to ensure independence.

RNA sequencing

Total RNA was extracted using the Aurum™ Total RNA Mini Kit (Bio-Rad, Hercules, CA, USA) with on-column DNase I digestion. RNA quality and quantity were assessed using a NanoDrop 2000 spectrophotometer, with samples meeting A260/280 ratios > 2.0 proceeding to library preparation. Libraries were prepared using the QuantSeq 3’ mRNA-Seq Library Prep Kit V2 (Forward) with Unique Dual Indices (Lexogen, Vienna, Austria). Each library was generated from 250 ng total RNA, followed by second-strand synthesis and PCR amplification. RNA sequencing was performed on 18 samples, representing two exposure groups (0 and 80 µM TPHP) with 9 initial biological replicates per group. Sequencing and quality control were performed at the Queen’s Cardio Pulmonary Unit (Kingston, ON, Canada). Final libraries were pooled and sequenced as 75 bp single-end reads on an Illumina NextSeq 550 platform. Sequencing reads were evaluated using FastQC 0.11.9 [86], summarized with MultiQC 1.13 [87], and aligned to a reference genome. Adapter sequences and low-quality bases were trimmed using bbmap with default settings [88]. Reads were aligned to the Oncorhynchus mykiss USDA_OmykA_1.1 reference genome using STAR 2.7.9a [89]. Gene-level quantification was performed with HTSeq-count [90]. One sample was excluded post-quality control due to technical failure, yielding n = 8–9 per group for final analysis. Prior to differential expression analysis, raw count data were batch-corrected using the ComBat-seq sva package 3.46.0 [91] to address technical variation between cell line aliquot batches (Supplementary Figs. 1 and 2). Normalization and differential expression analysis was performed in R Studio 4.4.2 using the DESeq2 package 1.46.0 [92]. Condition was specified as the primary experimental design factor. Count data was normalized using the median-of-ratios method, and dispersion estimates were fitted using the default parametric method. Principal component analysis (PCA) was performed on variance-stabilizing transformed (VST) data to assess sample clustering and verify batch correction effectiveness. Statistical testing for differential expression was performed using Wald tests with Benjamini–Hochberg correction for multiple testing. Gene annotation was performed using AnnotationHub [93] and biomaRt [94] packages, with manual curation using the NCBI Gene database when automated mapping was unavailable. Plots were generated using the ggplot2 [95] and pheatmap [96] packages, and base R. In addition to differential gene expression (DEG) analysis, we performed WGCNA, an unsupervised clustering method that groups genes into modules based on shared expression patterns across samples [80]. This network-based approach enables the identification of co-regulated gene modules and allows for correlation of these modules with external traits, such as TPHP exposure. Batch-corrected and VST-transformed counts from DESeq2 were used for WGCNA. The top 7000 most variable genes (by median absolute deviation) were retained for analysis, and a signed network construction utilized a soft-thresholding power of β=8 selected based on scale-free topology criterion (R² > 0.8) (Supplementary Fig. 3A and B). Signed network construction was performed with a minimum module size of 75 genes, a module merging threshold of 0.25, and the deepSplit parameter set to 0 to prevent over-splitting. A topological overlap matrix was generated and used for hierarchical clustering in signed mode. Module detection was performed using the dynamic tree cut method. Module eigengenes were correlated with TPHP exposure status using Pearson correlation, and significant modules were defined as those with P < .05. Genes within top two most significant modules (ME2 and ME3) were extracted, and the top 10 hub genes by module membership strength (kME) were identified. Gene annotation was performed as described above.

Whole genome bisulfite sequencing

WGBS was performed on 14 samples, representing two exposure groups (0 and 80 µM TPHP) with seven biological replicates per group. Following the 24 h exposure period, cells were harvested, and genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen) as per the manufacturer’s instructions. DNA concentration and purity were assessed using a Qubit fluorometer (Invitrogen) at the McGill Genome Centre (Montreal, Canada). Library preparation and bisulfite conversion were performed by the McGill Genome Centre, using the NxSeq® AmpFREE™ Low DNA Library Kit (Lucigen), following the manufacturer’s instructions as adapted by the McGill Genome Centre. Briefly, 1 µg of genomic DNA was sheared to approximately 350 bp using a Covaris ultrasonicator and purified with AMPure XP beads (Beckman Coulter). Following end-repair and A-tailing, methylated index adaptors were ligated, and the DNA was purified and size-selected using AMPure XP beads. Bisulfite conversion was conducted using the EZ DNA Methylation-Gold™ MagPrep Kit (Zymo Research) with the BS-ALT2 thermal cycling protocol (98°C for 10 min, 53°C for 4 h). As an internal quality control for bisulfite conversion efficiency, 1 ng of unmethylated Enterobacteria lambda phage genomic DNA (Roche Capture & Bisulfite-Conversion Control, NC_001416.1) was spiked into each sample prior to bisulfite treatment. This spike-in control allowed independent estimation of conversion efficiency by measuring the extent of cytosine-to-thymine conversion in a defined, non-methylated region (coordinates 4500–6500 bp of the lambda genome). Converted libraries were eluted, amplified using 6 cycles of LM-PCR, and cleaned again with AMPure XP beads. Final libraries were quantified using the sparQ Universal Library Quant Kit (Quantabio). WGBS, quality control, and genome alignment were performed at the McGill Genome Centre (Montreal, QC, Canada). Libraries were normalized, pooled, denatured with 0.05N NaOH, neutralized with HT1 buffer, and loaded at 150 pM on a 25B flow cell of the Illumina NovaSeq X platform. Sequencing was performed in paired-end mode (2 × 150 bp) with a 5% PhiX control. Base calling was carried out using the Illumina Real-Time Analysis Software (RTA 3.4.4). Raw paired-end reads were assessed for quality using FastQC [86] and summarized with MultiQC [87]. Adapter and quality trimming were performed using Trim Galore [97] in paired-end mode. Trimmed reads were aligned to the bisulfite-converted Oncorhynchus mykiss USDA_OmykA_1.1 reference genome using Bismark 0.24.0 [98] with Bowtie2 [99] in paired-end mode. Duplicate reads were removed, and methylation extraction was performed in Bismark. Cytosine methylation calls were summarized into coverage files for downstream differential analysis. Methylation call data were processed in R using the methylKit package [100]. CpG sites with a minimum coverage of 10× were retained, and the top 0.1% most highly covered sites were excluded to reduce PCR amplification bias. Coverage was normalized across samples using the median. To assess genome-wide methylation patterns, the percentage of CpG-site methylation was calculated for each sample from the filtered, normalized methylKit object as the percentage of CpG sites classified as methylated across all sites meeting quality criteria (≥10 × coverage, top 0.1% excluded, present in ≥4/7 samples per group), where methylation at a site was defined as ≥50% of sequencing reads exhibiting cytosine retention. Statistical comparison of CpG-site methylation percentage for each sample between the treatment and control groups was performed using unpaired t-test in GraphPad Prism 9 (P < .05). DMRs were defined in this study using a sliding window approach (tileMethylCounts(); 500 bp window, 250 bp step size) as described by Akalin et al. (2012) [100], requiring ≥ 4 out of 7 samples per group to meet the 10 × coverage threshold to be included. Similar sliding window approaches have been used by others [101–107]. Logistic regression was used to test for differential methylation between control and TPHP-exposed groups. Regions with a q-value ≤ 0.05 and an absolute methylation difference ≥ 10% were classified as significant DMRs. DMR annotation utilized the USDA_OmykA_1.1 genome assembly (GCF_013265735.2) with classification as gene body [exonic and inferred intronic (non-exonic regions within annotated gene boundaries)], promoter (≤5 kb upstream of transcriptional start site), downstream (≤5 kb past transcription end site), or intergenic regions. For gene mapping, we used AnnotationHub [93] and biomaRt [94], with manual annotation using the NCBI Gene database when automated methods were unavailable. DMR-to-gene distances were calculated and binned into variable base pair intervals for visualization. Annotated DMRs were visualized using ggplot2 [95] and pheatmap [96].

Multi-omic integration

Genes associated with significant DMRs were compared to DEGs to identify coordinated transcriptomic and epigenetic changes. For genes present in both the DEG and DMR datasets, directionality of change was examined to identify concordant regulatory patterns. A simple overlap analysis between the two datasets was performed based on gene symbols and the broad genomic location indicated. These overlapping genes were visualized to highlight coordinated epigenetic and transcriptomic responses to TPHP exposure.

Pathway enrichment analysis

GO and KEGG pathway enrichment analyses were performed using clusterProfiler [59] for Biological Process, Molecular Function, and Cellular Component ontologies using the Oncorhynchus mykiss OrgDb annotation object with Benjamini–Hochberg correction and significance threshold P < .05. Separate analyses were conducted for DEGs, DMR-associated, and WGCNA genes. All significant enrichment results were visualized using the ggplot2 [95] and enrichplot [108] packages in R.

Conflicts of interest

The authors declare no competing financial interests.

Author contributions

Logan S. Germain (Conceptualization [equal], Data curation [equal], Formal Analysis [equal], Investigation [equal], Methodology [equal], Writing – original draft [equal]), Benjamin P. Ott (Formal Analysis [supporting], Investigation [supporting], Methodology [supporting], Validation [supporting], Writing – review & editing [supporting]), Charles C. Hindmarch (Data curation [supporting], Investigation [supporting], Methodology [supporting], Software [supporting], Validation [supporting], Writing – review & editing [supporting]), Louise M. Winn (Conceptualization [lead], Funding acquisition [lead], Project administration [lead], Resources [lead], Supervision [lead], Writing – review & editing [equal])

Funding

This research was supported by a grant from the Natural Sciences and Engineering Council of Canada (NSERC) awarded to L.M. Winn (RGPIN-2019-05638).

Data availability

Raw RNA-seq and WGBS data have been deposited in the European Nucleotide Archive (ENA) under study accession ERP183977. The full list of processed DEGs can be found in file titled “Supplementary_F1_Annotated_DEGs,” and the full list of processed DMRs can be found in file titled “Supplementary_F2_Annotated_DMRs.”