Identification of functional genetic components modulating toxicity response to PFOS using genome-wide CRISPR screens in HepG2/C3A cells

Abstract

Perfluorooctane sulfonate (PFOS) poses significant health and environmental risks due to its persistence and widespread use and has been linked to various adverse outcomes, such as liver toxicity. Although the molecular responses and toxicity effects of PFOS exposure have been extensively studied, considerable uncertainty remains regarding the causal mechanisms leading to PFOS-associated adverse effects. To help bridge this gap, we conducted CRISPR screens in HepG2/C3A human liver cells exposed to IC₂₅ (170 µM) of PFOS to identify genes and pathways influencing PFOS-induced cytotoxicity. Using a genome-wide CRISPR knockout library targeting 18,819 genes, we identified 340 candidate genes that modulate PFOS-induced cytotoxicity when genetically disrupted (189 gene disruptions increased sensitivity and 151 gene disruptions increased resistance). From these candidate genes, we individually disrupted two candidate genes, SLC6A9 which encodes the glycine transporter GlyT1, and CPSF2, and confirmed increased resistance to PFOS exposure. Further, molecular docking analysis predicts that PFOS directly binds to GlyT1 and functional inhibition of GlyT1 also increases resistance to PFOS exposure. Gene-Disease outcome association analysis using the Comparative Toxicogenomics Database (CTD) indicated an enrichment of candidate genes associated with cancer-related and liver disease phenotypes. KEGG and STRING enrichment analyses found over representation of several biological pathways including DNA damage response and cell cycle. Lastly, cross-species conservation analysis using the top two validated gene targets found that their pathways were highly conserved in several environmentally relevant species. These findings provide new mechanistic and functional insights into PFOS-induced cytotoxicity, highlight potential molecular targets for toxicity mitigation, and establish a foundation for cross-species toxicogenomic modeling of PFOS health effects.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00204-025-04294-8.

Introduction

The number of chemicals used in consumer and industrial products has increased at a pace faster than our ability to assess the potential risks they may pose to human and ecological health. Among these, a large group known as poly- and perfluoroalkyl substances (PFAS) has raised significant concerns due to widespread use and their resistance to degradation, making them ubiquitous in the environment. They have been linked to adverse human health and environmental effects (Wee and Aris 2023). Although its manufacture has been restricted across the globe, perfluorooctane sulfonate (PFOS), remains the most commonly detected PFAS globally and exposure represents an ongoing concern in human and ecological health (Göckener et al. 2020; Cheng et al. 2022; Wee and Aris 2023) as evidenced by a recent large-scale environmental survey done by US EPA (Environmental Protection Agency) detecting PFOS at levels higher than US guidelines in 31% of global groundwater samples (US EPA, Accessed June 27, 2025).

Several human epidemiological studies have indicated potential links between exposure to PFOS and human adverse outcomes, including reproductive and developmental defects, immunotoxicity, and, the focus of this study, liver toxicity (Cui et al. 2009; Salihovic et al. 2019; Fenton et al. 2021; India-Aldana et al. 2023). Previous research examining PFOS accumulation in human tissues revealed its presence in the brain, kidneys, liver, and lungs, with the highest concentration found in the liver (Pérez et al. 2013; Zeng et al. 2019). A systematic review and meta-analysis of human epidemiology studies found correlation between predicted or measured PFOS exposure and higher serum alanine aminotransferase (ALT) levels, a biomarker of liver injury (Borghese et al. 2022; Costello et al. 2022). Although most of the epidemiological studies suggest a potential association between PFOS exposure and liver toxicity, a causal link is still controversial. In rodent studies, PFOS treatment altered lipid utilization pathways, increasing hepatic lipid accumulation, in turn, hepatic steatosis (Marques et al. 2020; Pfohl et al. 2021; Ling et al. 2023). In addition, the combination of transcriptomics and proteomics revealed enrichment of hepatic steatosis-related pathways in mice exposed to PFOS in their diet, consistent with PFOS-induced liver damage (Zhang et al. 2023). An in vitro study that measured cellular triglyceride and cholesterol levels in human HepaRG cells exposed to PFOS supported PFOS-induced liver damage (Louisse et al. 2020). While current evidence suggests that liver toxicity can result from PFOS exposure, the molecular mechanisms and specific targets are not clearly understood.

Omics-based approaches have identified cellular pathways and putative molecular targets underlying PFOS effects (Rowan-Carroll et al. 2021; Beale et al. 2022). While these ‘omics’ analyses do identify molecular changes correlated with PFOS exposure, they do not provide direct causal evidence for specific targets and/or molecular mechanisms underlying cellular- and organismal-effects of PFOS. Genome-wide CRISPR screens can provide a complementary approach to help identify the key functional genetic components and pathways modulating PFAS toxicity. In this study, we carried out a genome-wide loss-of-function CRISPR screen in human HepG2/C3A liver cells to identify functional modulators of PFOS cellular toxicity. We independently validated the effect of targeted disruption of two selected top candidate genes. In addition, an inhibitor study and molecular docking prediction were performed with one of the validated genes based on structural similarity of PFOS and an inhibitor of the gene product (de Carvalho et al. 2024), which suggested mechanistic insights into PFOS-induced cellular toxicity. The Comparative Toxicogenomics Database (CTD) was used to identify potential linkages to adverse outcomes (diseases) associated with the candidate genes identified in our screens. KEGG and STRING enrichment analyses of candidate genes that increase or decrease sensitivity to PFOS-induced cytotoxicity, when genetically disrupted, identified cellular processes and pathways, consistent with gene-disease association identified in CTD analysis. The top two candidate genes that were experimentally validated were further analyzed for their evolutionary conservation and functional relevance across species to provide insights into shared mechanisms underlying the adverse environmental and ecological impacts of PFOS exposure, which could guide future one-health perspective studies.

Materials and methods

Cell cultures and PFOS cytotoxicity

HEK293T and HepG2/C3A cells were purchased from the American Type Culture Collection (ATCC, Manassas, Virginia). Cells were maintained and passaged following ATCC’s recommended protocol (for the detailed condition, see Supplementary Information). Cytostasis/Cytotoxicity of PFOS exposure in HepG2/C3A cells exposed to a range of nominal concentrations (0–300 µM) in triplicate for 6 days was assessed by measuring ATP levels using the CellTiter-Glo2.0 cell viability assay kit (Promega, Madison, WI) following the manufacturer’s instruction. Of note, we chose the duration of exposure as 6 days for the IC₂₅ calculation since PFOS exposure didn’t show detectable cytotoxicity in HepG2/C3A cells after 3 days of exposure, which is the exposure period that we routinely use to estimate sublethal concentrations for CRISPR screens. Additionally, we performed a pre-screen time-course treatment with the initial benchmark IC₂₅ (6 days) dose over 18 days by passaging the treated cells every 6 days to refine the exposure conditions for the screen and to assess longer-term cellular responses (including cumulative cytostasis and mild cytotoxicity) by decreasing IC₂₅ (6 days) down to 170 µM as higher cytotoxicity was observed with IC₂₅ (6 days) for 18 days.

Genome-wide CRISPR loss-of-function screening

For the genome-wide loss-of-function genetic screens, we used a customized minimal genome-wide sgRNA library (Table S1) termed MinLibCas9 based on a published study which selected two sgRNAs with maximal on-target activity for each gene based on computational analysis of the screening results of several genome-wide screens (Gonçalves et al. 2021). Lentivirus production was carried out as previously described (Sobh et al. 2019). A pool of the MinLibCas9 lentiviral library was transduced into HepG2/C3A cells at a0.3 multiplicity of infection (MOI) to generate a mutant pool as in previous studies (Russo et al. 2020; Zhao et al. 2021) (detailed methods available in Supplementary Information). The mutant pool was then expanded for the PFOS exposure experiments to obtain sufficient number of cells for replicates and controls. Control and PFOS exposure samples of 16 × 10⁶ cells in T225 flasks (three replicates) were used for screening. Exposure samples were treated with PFOS at inhibitory concentration (IC₂₅) for 18 days (10 doubling times). Representation of the library was maintained throughout the screening experiment at approximately 400X the MinLibCas9 library size (approximately 40,000, so 400X = 16,000,000 cells) for each replicate.

DNA extraction, library preparation, and next-generation sequencing (NGS)

Genomic DNA was extracted from 16 × 10⁶ cells of each sample using the Quick-DNA Midiprep Plus Kit (ZYMO Research) following the manufacturer’s protocol. Amplicons for NGS Illumina sequencing were generated using the pairs of universal CRISPR-FOR1 forward primer and CRISPR-REV# reverse primers (#: 1 to 48) specific for each sample (see Table S2). Amplicons amplifying the sgRNA region in each sample were then pooled and gel purified using the QIAquick Gel Extraction Kit (Qiagen) and quantified using the Qubit HS dsDNA assay (Thermo Scientific). Equimolar amounts of each amplicon library were multiplexed in one pool. The NGS was carried out at the Interdisciplinary Center for Biotechnology Research (ICBR), University of Florida at Gainesville, using the NovaSeqX paired 150 bp high-throughput platform (Illumina).

Data processing and bioinformatics

After data acquisition (raw fastq.gz files), read quality was checked using FASTQC tools. A Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout (MaGeCK) (Li et al. 2014) was then used to demultiplex raw fastq data files, which were further processed to generate reads containing only the unique 20 bp guide sequences. The resulting read counts mapped to sgRNAs from control samples (rep.1 = 24,456,189, rep.2 = 35,209,481, rep.3 = 41,602,626) and PFOS-exposed samples (rep.1 = 8,177,206, rep.2 = 38,374,147, rep.3 = 13,033,381) corresponded to an average coverage of 892X for control and 524X for PFOS samples, both of which are above the typical coverage used in genome-wide CRISPR screens. The counts were then normalized to adjust for the effect of library sizes and read count distributions. The sgRNAs were then ranked by a modified robust ranking aggregation (α-RRA) algorithm to generate a ranked gene list. Candidate genes were then defined as genes with p-value < 0.01 and Log₂FC < -0.6 (sensitive) or Log₂FC > 0.6 (resistant) for further analysis (e.g., KEGG and CTD analyses). We define sensitive genes as those that when disrupted by CRISPR targeting, show increased cellular sensitivity (less abundant than control) to PFOS exposure whereas resistant genes show increased resistance (more abundant than control) to PFOS exposure. We compared the MaGeCK-produced results with the results obtained using Platform-independent Analysis of Pooled Screens using Python (PinAPL-Py) (Spahn et al. 2017).

Single gene knockout validations

We performed individual disruptions of the top two candidate genes that increase resistance to PFOS when genetically disrupted. The sgRNAs targeting the SLC6A9 and CPSF2 genes were selected from the MinLibCas9 sgRNA library (Table S3). Each sgRNA was cloned into the LentiCRISPRv2 using the golden gate assembly strategy as described elsewhere in detail (Joung et al. 2017). After cloning of each sgRNA, production and transduction of the corresponding lentivirus were conducted, generating knockout cell pools (for details, see Supplementary Information). The cellular pools for each targeted gene were used for cytotoxicity assays to evaluate their resistance to PFOS exposure.

Molecular docking

A Protein Data Bank (PDB) format file for the human and zebrafish GlyT1 and glycine transporter 2 (GlyT2) structures were obtained from AlphaFold Protein Structure Database (human GlyT1: XP_047284695.1; human GlyT2: XP_016874033.1; zebrafish GlyT1: XP_005163551.1; zebrafish GlyT2: XP_009301240.1) (Jumper et al. 2021; Varadi et al. 2024). Molecular docking to predict the binding of PFOS to these transporters was performed using AutoDock Vina (AutoDock Vina). The docking search grid was defined by a box measuring 84 × 48 × 78 Å. The highest scoring poses, indicated by the lowest ΔG values (in kcal/mol), were identified in the central region of each transporter. All structural figures were created using PyMOL (The PyMOL Molecular Graphics System, Version 3.0, Schrödinger, LLC.; PyMOL| pymol.org).

GlyT1 inhibitor assay

ALX5407, a selective human GlyT1 inhibitor (Atkinson et al. 2001), was used for an inhibitor assay to functionally disrupt GlyT1 activity (i.e., PFOS-GlyT1 interaction). A dose-response pilot study with ALX5407 in DMSO found 10 nM to be the maximal concentration not affecting HepG2/CA3 cell viability after 6 days consistent with a previous study which reported 3 nM as the EC50 of the inhibitor (Atkinson et al. 2001). To assess the effect of GlyT1 inhibition on PFOS cytotoxicity, HepG2/C3A cells were plated into a 96-well plate at a density of 12,000 cells per well a day before exposure. ALX5407 (10 nM) was added to triplicate sets of wells with corresponding DMSO controls and different concentrations of PFOS (0, 75, 150, and 300 µM) at the same time for 6 days. ATP concentration in each well was determined by CellTiter Glo2.0 assay as in 2.1 above. Statistical significance evaluation was done by t-test using GraphPad Prism (version 10.1.0).

Functional enrichment analysis

KEGG gene set enrichment analysis (Log₂FC values were included) was conducted with sensitive and resistant candidate genes to PFOS via clusterProfiler and the corresponding enrichment plots were generated by SRPLOT software (Tang et al. 2023). STRING database (Szklarczyk et al. 2023) was used to cluster protein products of candidate genes that confer sensitivity or resistance to PFOS using default settings (accessed on Mar 13th, 2025). We annotated each cluster of two or more gene products based on Gene Ontology-Biological pathway and KEGG pathway as available within STRING.

Gene-disease outcome association analysis

Statistically significant candidate genes (189 PFOS-sensitive genes and 151 PFOS-resistant genes) were used to perform over- and under-representation analyses using the CTD (Davis et al. 2025) accessed on Jan 30th, 2025. The proportion of PFOS-sensitive and PFOS-resistant genes linked to each Disease was compared to that of equally sized random sets of genes. Disease outcomes with a proportion above 0.001 in our significant candidate gene sets were considered for further analysis. Z-scores were calculated by subtracting our observed value with the mean proportion for each Disease outcomes across 1,000 random sets and divided by the standard deviation of the random sets. p-values were obtained using a standard normal distribution, and 0.05 was used as the significance threshold.

G2P-SCAN analysis: conserved reactome pathways analysis among other model species

The Genes-to-Pathways Species Conservation Analysis (G2P-SCAN) (Rivetti et al. 2023) R package was run using the latest source code for the G2P-SCAN library that is available on GitHub (https://github.com/seacunilever/G2P-SCAN). Based on the gene input, the G2P-SCAN outputs the overall analysis of orthology and functional families to substantiate the identification of conservation and susceptibility at the Reactome pathway level.

Results

CRISPR screens identified functional genetic components modulating PFOS-induced cytotoxicity

The overall design of this study is illustrated in Fig. 1A. We first identified a dose of PFOS which partially inhibited cellular proliferation of HepG2/C3A over the short-term period of exposure. We determined the IC₂₅ for 6 days exposure to be 192 µM for PFOS (Fig. S1), which was further refined to be 170 µM after a pre-screen time-course treatment for the subsequent CRISPR screens. After quality control sequencing of the MinLibCas9 library (Fig. S2), we used the library to generate a mutant pool of knockouts in HepG2/C3A at ~ 400X coverage of each sgRNA. After expansion of the mutant pool, the pool was either maintained (Control) or exposed to PFOS (Exposed) at the IC₂₅ concentrations for 18 days (approximately 10 doublings). We utilized the IC₂₅ of PFOS to identify genes which increase sensitivity to PFOS when disrupted (corresponding sgRNAs depleted in the exposed mutant pool) as well as genes which increase resistance to PFOS when disrupted (corresponding sgRNAs enriched in the exposed mutant pool). We used MaGeCK to rank sgRNAs and corresponding genes (gene hits) whose disruption either increased sensitivity or resistance to PFOS exposure (Table S4). We then filtered those hits with an additional threshold of p-value < 0.01 and log₂FC < − 0.6 or > 0.6 (Fig. 1B) to identify candidate genes for downstream pathway enrichment and CTD analyses, thereby ensuring selection of biologically meaningful hits. At this threshold, we identified a total of 340 candidate genes (189 genes increased sensitivity when disrupted; 151 genes increased resistance when disrupted) to PFOS exposure (a list of the candidate genes in Table S4) (Fig. 1B).

Fig. 1.

Genome-wide CRIPSR screens to identify functional genetic modulators of PFOS-induced cellular toxicity. The CRISPR screening workflow employed in the study is shown in panel (A), while panel B illustrates the bioinformatics analysis and specific criteria used to identify statistically significant candidate genes from the screens. The volcano plot C displays candidate genes that modulate PFOS cellular toxicity with different colors. Blue dots represent PFOS-sensitive genes and red dots represent PFOS-resistant genes. The top 10 sensitive and resistant genes are also listed next to the plot. These top candidate genes reflected the most significant ranks from MaGeCK analysis (α-RRA) based on the criteria (B) applied. The x-axis represents Log₂ FC (fold change), and the y-axis shows -Log10 (p-value). Panel D shows the representative individual sgRNA abundance results of the top 2 resistant candidate genes (sgRNA#1 for SLC6A9 and CPSF2) from the screens

The Log₂FC, p-value, and known or predicted functions (abbreviated STRING annotation for corresponding gene product) of the top ranked gene hits identified in our CRISPR screens are shown in Table S4. The top ten gene hits that increased resistance to PFOS when disrupted are SLC6A9, CPSF2, CTNNB1, ZNF860, ZNF486, WDR70, LRRC56, SMC2, PHLPP1, and SLC46A3 (Fig. 1C). Similarly, the top ten gene hits that increased sensitivity to PFOS exposure when disrupted were TMEM254, YRDC, CUBN, CFAP299, UNC45A, MMP20, SCNN1G, ABTB1, URI1, and ASTL. Analysis using a different algorithm, PinAPL-Py (Rockett 2002), identified a similar set of candidates particularly among the significantly ranked genes (all top sensitive and resistant genes were identified as significant except for ASTL gene) (Fig. S3). We selected the two most significant candidate gene hits identified in both analysis methods as top-ranked genes that increase resistance to PFOS exposure, SLC6A9 and CPSF2 for further investigation. Individual sgRNA abundance of the selected genes in our screening data (normalized sgRNA counts for each gene in our CRISPR screens) showed an increase in abundance upon PFOS exposure (Fig. 1D).

3.2 Functional validation of selected candidate genes supports a role for SLC6A9 and CPSF2 during PFOS exposure.

To validate the selected candidate genes identified as being resistant to PFOS in our screens, we performed single gene CRISPR/Cas9 targeting of SLC6A9 and CPSF2. Consistent with the results of the CRISPR screens, the cell pools with single gene knockouts (KOs) validated at the gene and mRNA levels (Figs. S4 and 2A) exhibited dose-response cellular resistance to PFOS exposure for 6 days with improved cell viability compared to wild-type HepG2/C3A cells (Fig. 2B and C). Disruption of the SLC6A9 in HepG2/C3A cells resulted in significantly increased resistance to PFOS exposure at 150 µM (p = 0.00013) and 300 µM (p = 0.00096) (Fig. 2B) and disruption of CPSF2 similarly showed significantly enhanced cell viability upon PFOS exposure at 150 µM (p = 0.0002) and 300 µM (p = 0.00149) (Fig. 2C). Partly due to the lower average KO efficiency of CPSF2 detected at the mRNA level, the cell viability assay may show weaker overall cellular resistance to PFOS exposure in the CPSF2 KO cell pool (heterogenous mutant cell population) compared with the SLC6A9 KO cell pool (Fig. 2). The results support a functional role for the corresponding gene products in modulating PFOS-induced cytotoxicity.

Fig. 2.

Targeted validation of candidate genes identified from the PFOS CRISPR screens. For the top 2 candidate genes resistant to PFOS exposure, SLC6A9 and CPSF2, individual knockouts (KOs) of each gene are generated in HepG2/C3A cells by single gene CRISPR/Cas9 with the effect of each sgRNA#1, targeting each gene, on corresponding mRNA levels is shown, with error bars representing the mean ± SD from two samples of each KO cell pool (A). Using these KO cells, cell viability assays were performed and compared to wild-type HepG2/C3A cells for validation. Cell viability for 6-day exposure to PFOS was used to measure the relative chemical susceptibility (B and C). Three replicates were used for cell viability assays, and the error bars represent the mean ± SD from three independent experiments

Inhibitor assay and molecular Docking study suggest PFOS interacts with GlyT1 encoded by SLC6A9

Our screening and validation results showed that the functional disruption of SLC6A9, which encodes Glycine Transporter 1 (GlyT1), increased cellular resistance to PFOS exposure. Disruption of a cellular transporter resulting in increased resistance to a xenobiotic is consistent with a potential role in transport of the xenobiotic (Klaassen 2002; George et al. 2017). We next tested whether ALX5407, a selective irreversible inhibitor of human GlyT1 (Atkinson et al. 2001), induces cellular resistance to PFOS exposure (Fig. 3A). Indeed, HepG2/C3A cells treated with ALX5407 showed significantly increased resistance (p = 0.025) to PFOS exposure at 300 µM for 6 days in the presence of 10 nM ALX5407 (Fig. 3A), consistent with a functional role for GlyT1 in modulating PFOS toxicity. No interaction with or role of GlyT1 in PFOS transport has been previously reported to our knowledge, although other solute carrier transporters (SLCs), including sodium/taurocholate co-transporting polypeptide (NTCP) and organic anion transporter 4 (OAT4), have been reported to transport PFOS in several cell types (Vujic et al. 2024). We noted that Bitopertin, a known inhibitor of GlyT1, showed possible structural similarity to PFOS with a sulfonyl group (SO₂CH₃) and multiple trifluoromethyl groups (CF₃) (Fig. 3B). Previous studies had carried out molecular docking studies between Bitopertin and GlyT1 and GylT2 (encoded by SLC6A5) (de Carvalho et al. 2024), and thus we employed molecular docking to predict potential intermolecular contacts between PFOS and the GlyT1 transporter as well as the closely related GlyT2. We hypothesized that PFOS might interact with amino acid side chains in the central regions of the GlyT1 and GlyT2 transporter similar to as seen with Bitopertin (de Carvalho et al. 2024). Indeed, docking simulations using AutoDock Vina predicted PFOS binding to a conserved site within the GlyT1 and GlyT2 transporters. In GlyT1, PFOS was predicted to form hydrogen bonds with residues Y196, Y370, S371, and T472, (ΔG value − 9.5 kcal/mol) and in GlyT2, predicted hydrogen bonding occurs with residues at similar corresponding positions, Y289, F478, S481, and T580 (Fig. 3C). ΔG scores were − 9.5 kcal/mol and − 9.4 kcal/mol for PFOS interaction with GlyT1 and GlyT2, respectively, as compared to predicted ΔG of -3.89 for glycine interaction with GlyT1 at this site and ΔG of -4.3 for glycine interaction with GlyT2 at this site (Fig. 3C). Based on the potential interaction of PFOS with GlyT1 predicted by molecular docking simulation (potential substrate or inhibitor), inhibition assay (enzymatic inhibition), and KO validation (genetic disruption), we suggest that GlyT1 (SLC6A9) modulates PFOS cellular toxicity through a direct interaction with PFOS, but we cannot differentiate between a role for PFOS as an GlyT1 inhibitor or as a substrate.

Fig. 3.

Further analysis of the top resistant candidate gene for PFOS toxicity, SLC6A9, and predicted binding between GlyT1 (encoded by SLC6A9) and PFOS. A Enzymatic activity inhibition of GlyT1 by treatment of ALX5407 and its functional effect on PFOS-induced cellular toxicity. Cell viability of HepG2/C3A cells exposed to PFOS (0, 75, 150, 300 µM) when treated with ALX5407 (10 nM), a selective GlyT1 inhibitor for 6 days. Three replicates were used for cell viability assays. Statistical significance evaluation was done by t-test using GraphPad Prism (version 10.1.0). B Structural similarity between another GlyT1 inhibitor, Bitopertin, and PFOS. Putative structural similarity of Bitopertin to PFOS was depicted with a sulfonyl group (SO₂CH₃) and multiple trifluoromethyl groups (CF₃). C Molecular docking results present predicted intermolecular contact between GlyT1 (encoded by SLC6A9), GlyT2 (encoded by SLC6A5) and PFOS. PFOS is shown as sticks, with yellow for carbon, orange for sulfur, red for oxygen, and light blue for hydrogen. In the upper part, GlyT1 was depicted as a copper-colored ribbon diagram. In the lower part, GlyT2 was depicted as a green ribbon diagram. Residues involved in hydrogen bonding with PFOS were shown as stick models, with violet for carbon, red for oxygen, and blue for nitrogen. Residues that were predicted to form hydrogen bonds with PFOS are shown as stick models, with cyan representing carbon, red representing oxygen, and blue representing nitrogen

3.3 Gene-Disease outcome association analysis using the Comparative Toxicogenomic Database indicated cancer (neoplasm, carcinoma) and liver disease as potential adverse outcomes of PFOS toxicity.

To survey the potential association with adverse outcomes (disease phenotypes) of our significant candidate genes, we performed a custom Gene-Disease outcome association analysis using the CTD (Fig. 4A). We identified disease terms over-represented in the PFOS-sensitive and PFOS-resistant genes. The top 15 over-represented diseases for the PFOS-sensitive gene set as compared to randomly sampled gene sets of the same size are shown in Fig. 4B. 8 of the 15 top over-represented diseases in PFOS-sensitive genes included terms related to cancer such as Pancreatic Neoplasms, Adenocarcinoma, Prostatic Neoplasms, and Liver Neoplasms (Fig. 4B). In contrast, the analysis of the PFOS-resistant genes found 6 out of the top 15 associated phenotypes are related to liver disease, including liver failure, cholestasis, and liver cirrhosis (Fig. 5A).

Fig. 4.

Functional enrichment analyses of PFOS-sensitive genes. A Schematic overview of the Gene-Disease outcome analysis workflow. B Results of the top 15 significant disease outcomes over-represented in PFOS-sensitive genes are presented. In the histograms, the x-axis indicates the proportion of Disease hits with the dashed black line representing the mean of the random sets shown in gray, and our observed proportion represented by the solid blue line. The y-axis indicates the frequency of the proportion in the randomly sampled sets. The disease categories are displayed at the top of each plot. C KEGG pathway enrichment result of PFOS-sensitive genes. D STRING clusters of the functional protein interactions enriched in PFOS-sensitive genes

Fig. 5.

Functional enrichment analyses of PFOS-resistant genes. A Results of the top 15 significant disease outcomes over-represented in PFOS-resistant genes. In the histograms, the x-axis indicates the proportion of Disease hits with the dashed black line representing the mean of the random sets shown in gray, and our observed proportion represented by the solid blue line. The y-axis indicates the frequency of the proportion in the randomly sampled sets. The disease categories are displayed at the top of each plot. B KEGG pathway enrichment result of PFOS-resistant genes. C STRING clusters of the functional protein interactions enriched in PFOS-resistant genes

Enrichment analyses identified potential cellular pathways modulating PFOS toxicity

Next, we carried out KEGG and STRING analysis on candidate genes (sensitive and resistant genes) identified in the screens to gain insight into potential involved pathways and to explore functional interactions modulating PFOS-induced cytotoxicity (Figs. 4C, D and 5B, C). KEGG gene set enrichment analysis revealed that ATP-dependent chromatin remodeling and chemical carcinogenesis-DNA adducts related pathways are among the most significantly enriched pathways in PFOS-sensitive genes (Fig. 4C), consistent with the results of CTD Gene-Disease outcome association analysis which found cancer-related phenotypes being significantly over-represented (Fig. 4B). Functionally annotated STRING clusters of the PFOS-sensitive genes included Chemokine and Cytokine-mediated signaling pathway (CXCR5, CXCL5, CXCR6, AICDA, TNFRSF13C), the Hedgehog signaling pathway (SUFU, IHH, SAP18, EHD1), and cell cycle-related pathway such as Apoptotic process, Replicative senescence, and DNA damage response (TP53, NR2E1, CHEK2, BRPF1) (Fig. 4D) as well as Purine biosynthesis, Nuclear pore complex disassembly, Synaptic membrane adhesion, and others (Fig. S5A). KEGG analysis of PFOS-resistant genes identified Systemic lupus erythematosus, Cell cycle, and others as significantly enriched pathways (Fig. 5B). Functionally annotated clusters identified by STRING analysis of the PFOS-resistant genes included Chromosome organization (H3C12, TBX2, ZMYM3, H2BC13, H2AC21, SMC2, PDS5A, RAD21, PAXIP1, H2BU1), mRNA 3-end processing (SNRPG, CPSF2, WDR70, ZNF473), the Wnt signaling pathway (CTNNB1, AXIN1, TJP3, RET, ETV3, NDP) (Fig. 5C) as well as several others including Ionotropic glutamate receptor complex and Regulation of oxidative phosphorylation (Fig. S5B).

G2P-SCAN analysis provides cross-species applicability of adverse effects using gene targets of PFOS-induced toxicity

To provide insight into potential cross-species PFOS toxicity based on our human cell-based functional data, we conducted an in silico cross-species pathway prediction using G2P-SCAN. The G2P-SCAN analyzes data from various databases, integrating gene orthologs, protein families, entities, and reactions linked to human genes and respective pathways across six relevant model species including rat (Mus musculus), mouse (Rattus norvegicus), zebrafish (Danio rerio), fruit fly (Drosophila melanogaster), nematode (Caenorhabditis elegans), and budding yeast (Saccharomyces cerevisiae). We focused on top significant candidate genes including SLC6A9 and CPSF2, the two experimentally validated genes. SLC6A9 gene was mapped to only a single pathway (i.e., Na+/Cl- dependent neurotransmitter transporters, Fig. 6A) compared to result of CPSF2 (Fig. 6B) which showed mapping to multiple pathways, including RNA polymerase II transcription termination, Processing of intronless pre-mRNAs, and mRNA 3’-end processing. While SLC6A9 showed 100% pathway conservation in mammals, zebrafish and other three non-vertebrate species showed much lower conservation (zebrafish: 74%, nematode: 21%, fruit fly: 32%, Budding yeast: 0%) of this pathway. CPSF2 was highly conserved across vertebrates in the related pathways (83% for Processing of intronless pre-mRNAs of zebrafish to 100% for the same pathway of Mouse) whereas we observed lower conservation in other non-vertebrate species (39% for mRNA 3’-end processing of yeast to 78% for Processing of intronless pre-mRNAs of nematode and fruit fly) with notable exception of fruit fly (93%) in the pathway of ‘transport of mature mRNA derived from an intronless transcript’. Based on the G2P-SCAN result showing conservation in piscine species, we used molecular docking to assess if PFOS is predicted to interact with the GlyT1 and GlyT2 homologs in zebrafish (Danio rerio). Molecular docking of zebrafish GlyT1 (zGlyT1) and closely related GlyT2 (zGlyT2) also showed strong interaction (Fig. 6C). ΔG scores were − 6.4 kcal/mol and − 5.6 kcal/mol for PFOS interaction with zGlyT1 and zGlyT2, respectively, as compared to predicted ΔG of -6.4 for glycine interaction with zGlyT1 at this site and ΔG of -6.0 for glycine interaction with zGlyT2 at this site, indicating substrate-level interaction. Together, these results showcased that functionally identified gene targets responsible for a certain chemical-induced toxicity can be used to predict a potential role in other species under the same environmental exposure and that conservation levels of gene targets and related pathways may be different across the species.

Fig. 6.

The predicted conservation percentage for each species compared to the human ortholog based on the Genes-to-Pathways Species Conservation Analysis (G2P-SCAN) output, using GlyT1 (SLC6A9) (A) and CPSF2 (CPSF2) (B) as inputs, is shown. The individual conservation level related to each human gene input was analyzed using G2P-SCAN to determine the level of conservation across six model species presented by Reactome pathways. The size of circle is proportional to the percentage of conservation. The color of circle indicates different species, including Rat, Mouse, Zebrafish, Nematode, Fruit fly, and Yeast. C Molecular docking results present predicted intermolecular contact between zebrafish GlyT1 (zGlyT1) and zebrafish GlyT2 (zGlyT2) and PFOS

Discussion

Extensive epidemiological studies as well as controlled exposure studies in multiple species provide support for PFOS exposure being associated with adverse outcomes on liver function (Fenton et al. 2021; Jane et al. 2022) while the mechanisms underlying adverse outcomes remain controversial. Several potential mechanisms may play a role such as Peroxisome proliferator-activated receptor alpha (PPARα) activation (Azhagiya Singam et al. 2024), but definitive causal connections remain elusive. To provide insights into the functional effects of PFOS on the human liver, we carried out a genome-wide CRISPR screen with PFOS in this study. We identified multiple candidate genes that when disrupted modulate the cellular toxicity of PFOS and among these candidates, two selected genes (SLC6A9 and CPSF2) were experimentally validated, recapitulating the screening results. We further identified potential associated phenotypic outcomes (i.e., diseases), cellular pathways, and molecular networks potentially involved in PFOS toxicity using the PFOS candidate genes, which together suggest previously unrecognized potential mechanisms of PFOS-induced adverse effects.

Identification of the GlyT1 as a candidate gene functionally modulating PFOS-induced cellular toxicity, which transports glycine into multiple cells and tissues (including the hepatocyte in the liver (Howard et al. 2010; Wei et al. 2024), suggests a previously unknown connection between this amino acid transporter and PFOS. Several previous studies have noted increased mRNA expression level of SLC6A9 associated with PFOS exposure in mice (Pfohl et al. 2021; Deng et al. 2022) and in vitro (Louisse et al. 2023) but the effect of genetic disruption of the gene on PFOS toxicity hasn’t been studied. We considered the hypothesis that PFOS may directly interact with GlyT1 based on a previous study which carried out molecular docking of a GlyT1 inhibitor, Bitopertin, with structural features reminiscent of PFOS (Fig. 3B). Our molecular docking studies of PFOS with structural models of GlyT1 and the closely related GlyT2 unexpectedly predicted considerable affinity of PFOS for both GlyT1 and GlyT2 in the same region with the known inhibitor, with a predicted ΔG values exceeding the interaction with glycine alone (Fig. 3C). This potential interaction was further supported by chemical disruption of GlyT1 enzymatic activity with another selective GlyT1 inhibitor, ALX5407, which also conferred resistance to PFOS exposure (Fig. 3A). Together, molecular docking and enzymatic inhibition studies with SLC6A9 KO validation suggest that functional disruption of GlyT1 confers increased cellular resistance to PFOS exposure possibly via disrupted or reduced interaction between GlyT1 and PFOS. However, further follow-up studies such as PFOS and glycine (natural substrate) competitive uptake studies would be needed to confirm direct interaction. A recent review for the role of xenobiotic transporters expressed in the liver in PFAS transport noted that most of functional studies have relied on heterologous expression of transporters in cell lines (Vujic et al. 2024). Indeed, the Na⁺-taurocholate co-transporting polypeptide (NTCP) and several members of the organic anion transporting polypeptides (OATP) have been identified to play roles in PFOS uptake and reabsorption (Louisse et al. 2024). However, there remains uncertainty about the physiological role of the NTCP and OATP transporters, particularly in the liver. We did not find any of these previously identified transporters in our genetic screens, which indicates that they may not play primary roles in HepG2/C3A, or alternatively, functional redundancy may exist (e.g., multiple transporters could be involved and loss of single one may not alter the cellular response). In sum, our results suggest a possible role of GlyT1 in modulating PFOS cytotoxicity which has not been previously documented to our knowledge. While our studies primarily suggest a direct interaction of PFOS with GlyT1 underlies its role in PFOS-induced toxicity, further study will be needed to determine if it represents a substrate of GlyT1. Disruption of CPSF2 resulted in increased resistance to PFOS toxicity as well (Fig. 2A and C). CPSF2 encodes the CPSF2 protein which mediates 3’ end pre-mRNA processing, including poly-adenylation and histone mRNA 3’ end processing (Dominski and Tong 2021). The exact role of CPSF2 remains controversial (Liu and Manley 2024) and evidence for involvement of CPSF2 in mediating PFOS toxicity is limited. While several studies have identified histone epigenetic modifications associated with PFOS exposure (Kim et al. 2021; Li et al. 2024), direct effects on histone mRNA processing have not been noted to our knowledge. Interestingly, a genome-wide association study (GWAS) of perfluorooctanoic acid (PFOA) in vitro toxicity identified the CPSF2 locus as a candidate modulator of cytotoxicity (O’Shea et al. 2011). Transcriptional studies in zebrafish done by Dasgupta et al. and Rericha et al. with perfluorooctane sulfonamide (PFOSA), a precursor of PFOS, reported conflicting results with increased⁷⁰ or decreased⁷¹ CPSF2 mRNA expression, respectively. CPSF2 was identified as hub gene involved in lipid metabolism from weighted gene co-expression network analysis on RNA-seq data of bovine samples (Wang et al. 2022) potentially in line with lipid metabolism disruption which is suggested as one of molecular mechanisms of PFAS toxicity (Evans et al. 2022).

We used the CTD (Davis et al. 2025) to identify potential correlations between the candidate genes identified in our screens and associated disease phenotypes curated in the CTD. Surprisingly, the PFOS-sensitive genes showed over-representation in cancer-related terms, including Pancreatic neoplasms, Adenocarcinoma, Prostatic neoplasms, and Liver neoplasms among the top over-represented diseases. The adverse outcomes of PFAS exposure in general as well as PFOS specifically in the risk of developing different forms of cancer, including of the liver, kidney, thyroid, and blood remains uncertain (Steenland and Winquist 2021; Gerwen et al. 2023; Rhee et al. 2023; Winquist et al. 2023). The International Agency for Research on Cancer (IARC) classified PFOS as a Group 2B carcinogen (possibly carcinogenic) in Dec 2023 (Drury et al. 2024; Zahm et al. 2024). Several studies of PFOS exposure in rodents find evidence for hepatotoxicity, liver dysfunction (Bjork et al. 2011; Costello et al. 2022), and an early study in a rat model chronically exposed to PFOS observed significant increases in hepatocellular adenoma (Butenhoff et al. 2012). Similar to the CTD analysis, KEGG and STRING enrichment analyses identified several potentially cancer-related pathways and protein networks, respectively, over-represented in the set of PFOS sensitive mutants. Both DNA damage and non-DNA damage related tumorigenic pathways are suggested by the enrichment analysis as potentially involved in PFOS associated toxicity. For example, we noted enrichment of ATP-dependent chromatin remodeling (Fig. 4C) pathways which can be a key driver of cancer in some contexts, including by promoting the abnormal gene expression that maintains cancer status (Mayes et al. 2014) and/or by altering the accessibility of DNA in the chromatin to DNA-binding factors, which regulates DNA replication and DNA repair (Zhang et al. 2011). The enrichment of the chemical carcinogenesis-DNA adducts (Fig. 4C) KEGG pathway and the STRING network analysis which identified apoptosis, DNA replication, and the DNA damage response genes clusters supports a potential role for these processes in response to PFOS toxicity. Consistent with these findings, in marine vertebrates and bacteria (Liu et al. 2016; Otero-Sabio et al. 2022), cell cycle alterations and DNA damage responses have been postulated as potential mechanisms of PFOS-induced toxicity. And more recently, Li et al. identified TP53, a key player of DNA damage checkpoint (Eldridge et al. 2020), as a central node of PFOS-induced toxicity network (Li et al. 2025).

Together, our findings provide some support for a tumorigenic potential for PFOS but we recognize that the enrichment of cancer-related pathways might be related to intrinsic susceptibilities in HepG2/C3A due to pre-existing germline or somatic mutations (Sondka et al. 2024), which may mask or potentiate PFOS-specific toxicological effects. Thus, the observed enrichment findings should be viewed with caution and further validation of oncogenic relevance is essential.

Another protein interaction cluster comprised of CXCR5, CXCL5, CXCR6, AICDA, TNFRSF13C involved Chemokine and Cytokine-mediated signaling pathways. Chemokines are known to play a crucial role in maintaining the innate immune system (Mackay 2001), and epidemiological studies have linked immune dysfunction with PFOS exposure (Lee et al. 2018; Ehrlich et al. 2023). A 3D in vitro study to examine PFOS effects on cardiac development reported decreased expression of Hedgehog pathway genes, including HHIP, HHATL, and LQCE, as well as an increased expression of ENPP1, a Hedgehog repressor, suggesting PFOS exposure impairs the Hedgehog signaling pathway (Davidsen et al. 2022).

In contrast, we identified over-representation of liver disease-related terms in our PFOS-resistant genes. In people, several epidemiological studies linked PFOS exposure to elevated levels of alanine aminotransferase (ALT) in serum, a widely used marker of liver damage (Addicks et al. 2023; Zhang et al. 2023). A rodent study demonstrated that PFOS exposure exacerbates alcohol-induced liver injury, suggesting its potential as a significant risk factor in this phenotype (Ekuban et al. 2025). KEGG analysis identified Systemic lupus erythematosus (SLE) and cell cycle as significant pathways enriched for PFOS-resistant genes (Fig. 5B). Interestingly, liver abnormalities defined as higher levels of liver damage markers have been found among newly diagnosed SLE patients (Bessone et al. 2014). Cell cycle was also identified as enriched in KEGG (Fig. 5B) analysis and PFOS exposure in vitro was reported to promote hepatic cell proliferation, as evidenced by biomarkers such as nuclear protein Ki67 (Ki67) and topoisomerase 2 alpha (Top2α) (Cui et al. 2015). As the cell cycle represents one of the most significantly enriched KEGG pathways, this process may be critically relevant to the phenotypic association results. In line with this, the ‘liver neoplasms’ CTD over-representation (Fig. 5A) could be attributed to cell cycle-related pathway enrichment and corresponding genes (cell cycle: PDS5A, ORC6, RAD21, ZBTB17; Thyroid cancer: CTNNB1, RET; see Table S5) identified from our CRIPSR screens. STRING clusters of PFOS-resistance gene products included protein networks involved in chromosome organization, mRNA 3’ end processing, and the Wnt signaling pathway (Fig. 5C). Among these, the Wnt/b-catenin signaling pathway was linked to PFOS-induced liver damage in in vitro study where PFOS causes oxidative stress attributed to activation of the Wnt/b-catenin signaling pathway (Jiajing et al. 2024). Moreover, exposure to low concentration of PFOS (10 µM) was reported to alter Wnt signaling in mouse bone cells, suggesting the importance of Wnt/β-catenin signaling pathway in PFOS toxicity mechanisms (Xue et al. 2025). Overall, the gene-disease outcome association analysis highlights potential connection between the functional genetic components modulating PFOS toxicity and cancer and liver disease phenotypes.

The One Health approach (Buttke 2011; Sauvé 2024) can take advantage of available data in diverse species to extrapolate the effects of a chemical on molecular targets, affected biological pathways, and signaling cascades in one species to molecular targets conserved in other exposure species. We used G2P-SCAN analysis to assess the relevance of our studies in human cells to other species and identify candidate molecular targets conserved in those species (Schumann et al. 2024). Beyond demonstrating the general utility of G2P-SCAN, our results provide specific mechanistic insights into the broader potential toxicity effects of PFOS across species. The cross-species comparison of human gene targets linked to PFOS-mediated toxicity, such as SLC6A9 and CPSF2, enabled evaluation of the relevance of these targets in other species. In particular, based on G2P-SCAN conservation result of GlyT1 in zebrafish compared to human, we simulated molecular interaction between PFOS and zGlyT1 (and closely related zGlyT2), suggesting strong interaction (similar to interaction score with original substrate, glycine). This at least partly demonstrated that our G2P-SCAN based cross-species approach is valid to predict potential toxicity of other species using human relevant findings. Furthermore, the finding related to CPSF2 suggests that basal mRNA regulatory pathways (multiple conserved pathways) may play a critical role in PFOS toxicity across taxa. Although only two genes were included in the G2P-SCAN analysis based on their experimental validation in this study, additional candidate genes (e.g., the top 10 hits) could be further examined to generate pathway-level conservation data useful for elucidating conserved mechanisms of PFOS toxicity across species (Owen et al. 2025). While the applicability of such target-based inference of adverse effects depends on the conservation of molecular targets and pathways across species, this cross-species prediction tool can help identify priority species that may be at higher risk through similar mechanisms to an environmental toxicant to which humans are being exposed (Rivetti et al. 2023; Schumann et al. 2024).

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chanhee Kim: Conceptualization, Investigation, Methodology, Formal analysis, Visualization, Writing- original draft preparation, Abderrahmane Tagmount: Investigation, Methodology, Formal analysis, Writing- review and editing, Zhaohan Zhu: Methodology, Data analysis, Writing- review and editing, Frances Wilson: Investigation, Methodology, Danmeng Li: Methodology, Formal analysis, Writing- method, review, and editing, David A. Ostrov: Writing- review and editing, W. Brad Barbazuk: Writing- review and editing, Rhonda L. Bacher: Investigation, Methodology, Data analysis, Writing- method, review, and editing. Chris D. Vulpe: Conceptualization, Supervision, Writing- review and editing, funding acquisition.

Data availability

The raw and processed CRISPR screen data with the corresponding metadata in this study were deposited into Gene Expression Omnibus (GEO) database repository and available with an accession number of GSE291529.

Declarations

Conflict of interest

The authors declare that they have no conflicts of interest.