Efficient approaches enabling the identification and prioritization of potential thyroid hormone-system-disrupting chemicals (THSDCs) acting through various molecular initiating events (MIEs), which are related to disrupted thyroid hormone homeostasis, are urgently needed (Haigis et al. 2023; Vergauwen et al. 2024). Inhibition of iodide uptake in thyrocytes mediated by NIS is one of the crucial identified MIEs, which can be affected by a diverse set of THSDCs (Friedman et al. 2016; Hallinger et al. 2017; Hornung et al. 2018; Olker et al. 2019; Titus et al. 2008; Wang et al. 2019, 2018). The effect of THSDCs on NIS activity has been connected with the decrease of TH production in the thyroid gland and significant health effects, namely during development (De Groef et al. 2006; Zimmermann 2011). Importantly, as NIS is often used as an important theragnostic gene in radionuclide uptake-based carcinoma treatment (Chung et al. 2018; Kitzberger et al. 2022; Spitzweg et al. 2021), NIS inhibition by xenobiotics could also adversely affect diagnostic and treatment methods and potentially jeopardize therapy outcomes since NIS activity in carcinoma cells is being inversely connected with the recurrence and metastatic potential of the thyroid carcinoma (Oh and Ahn 2021).

This study brings important information on NIS assay performance across different in vitro models and advances its in vivo relevance and applicability by developing biotransformation-competent assay set up for NIS-inhibition screening purposes by incorporating BTS capacities into the assay. The most crucial results are outlined and discussed in the following sections, with additional material given in the SI.

We examined SLC5A5 (NIS) gene expression in ten model cell lines. While human thyrocytes-derived cell line Nthy-ori 3–1, together with the commonly used human cell line HEK293T, were without detectable NIS expression, the transfected HEK293T cell lines, together with the rat FRTL-5 cell line, had significant NIS gene expression levels (Figure S3), as compared to commercially available human and rat total thyroid RNA samples. The highest gene expression was detected in the cell cultures HEK293T NIS 01 and 02 as well as with single-cell clone line HEK293T NIS C. Among these, HEK293T NIS C showed the highest NIS gene expression, followed by HEK293T NIS 02 and HEK293T NIS 01, while the rest of the transfected cell lines were removed from further evaluation due to low NIS gene expression and in some cases also poor growth characteristics (Figure S3).

Western blot analysis revealed protein bands corresponding to NIS in the prioritized transfected cell lines, confirming the presence of NIS protein. No NIS protein expression was observed in Nthy-ori 3–1 or non-transfected HEK293T cell lines (Figure S4). The three transfected human cell lines with good stable NIS protein expression also across higher cell culture passages (HEK293T NIS C, NIS 01, and NIS 02), along with the endogenously NIS-expressing rat FRTL-5 cell line, were selected for further characterization of NIS activity and their suitability for NIS-inhibition measurement.

The NIS gene and protein expression characterized via qPCR and Western Blot analysis (Figure S3 and S4) corresponded to NIS-mediated iodide uptake assessed by SK reaction. There was no iodide uptake due to the absence of NIS protein in the HEK293T and Nthy-ori 3–1 cell lines, where the recorded responses (absorbance at 415 nm) corresponded to the cell-free controls (Figure S5B). However, even though the NIS gene and protein expression patterns were similar between the HEK293T NIS 01 and NIS 02 cell lines (Figure S4), the former depicted lower iodide uptake in the pilot experiments (Figure S5B) and was excluded from further investigations.

NIS-expressing cell lines HEK293T NIS 02, HEK293T NIS C, and FRTL-5 with good iodide uptake, were finally selected for NIS-inhibition assessment by chemical exposure. The two HEK293T-derived models were employed for the detailed characterization to verify if they show the same performance characteristics as they share the same NIS-expression gene construct. The characterization was performed along with careful evaluation of cellular viability, which is crucial in the NIS-inhibition assay, as cytotoxic effects could be mistakenly interpreted as NIS inhibition, leading to false-positive results.

The current study applied a non-radioactive approach (SK reaction) to measure NIS-mediated iodide uptake inhibition by 23 model chemicals from diverse use categories (Table 1, Figs. 1 and 2). Four of them (PCL, TCS, BPA, DBP) have been previously tested using the same method on other in vitro models (Dong et al. 2019; Waltz et al. 2010; Wu et al. 2016; for more details see Table S2). Nevertheless, there are data for 16 of the chemicals from radioactivity-based method (RAIU) on several in vitro models (see Table S2), which enables the assessment of comparability of the results across the two different assay formats and the various in vitro models. RAIU detection was employed on hNIS-HEK293T-EPA cell model in the current version of ToxCast (US EPA 2024), papers building on data from the earlier ToxCast datasets (Wang et al. 2019, 2018), studies using the ToxCast cell model (Buckalew et al. 2020; Hallinger et al. 2017), FRTL-5 cell model (Buckalew et al. 2020; Waltz et al. 2010), or HeLa cells transfected with NIS genes from different vertebrates (Concilio et al. 2020). Table S2 in Supplementary Information provides a detailed overview of the results from NIS inhibition and cytotoxicity in the three in vitro models used in our study along with the relevant data across the studies in literature and the ToxCast database.

Detected effects of the first part of tested compounds on NIS inhibition (blue curve) and cell viability (red curve) in two hNIS transfected cell lines, namely HEK293T NIS02 (left), HEK293T NIS C (middle), and rat thyroid cell line FRTL-5 (right). Points represent the total mean from at least three independent experiments (n ≥ 3 experimental units, N ≥ 9 observational units/technical replicates) per tested concentration, with whiskers representing ± SD of all observations. The data are shown as FOC%, indicating the fraction of control. For NIS inhibition, the cell-free control (no iodide uptake) serves as a reference, with an increase in FOC% reflecting enhanced inhibition of iodide uptake. For cell viability, a decrease in FOC% indicates reduced viability relative to the NaI control (100% viable cells). The names and abbreviations of the tested chemicals and their respective IC50 NIS-inhibition values are provided in Table 1, the cytotoxicity IC20 values are in Table S2. The tested concentrations ranged from 0.78 to 100 µM, except for BPA and MET, which were tested up to 500 µM (Colour figure online)

Detected effects of the second part of tested compounds on NIS inhibition (blue curve) and cell viability (red curve) in two hNIS transfected cell lines, namely HEK293T NIS02 (left), HEK293T NIS C (middle), and rat thyroid cell line FRTL-5 (right). Points represent the total mean from at least three independent experiments (n ≥ 3 experimental units, N ≥ 9 observational units/technical replicates) per tested concentration, with whiskers representing ± SD of all observations. The data are shown as FOC%, indicating the fraction of control. For NIS inhibition, the cell-free control (no iodide uptake) serves as a reference, with an increase in FOC% reflecting enhanced inhibition of iodide uptake. For cell viability, a decrease in FOC% indicates reduced viability relative to the NaI control (100% viable cells). The names and abbreviations of the tested chemicals and their respective IC50 NIS-inhibition values are provided in Table 1, the cytotoxicity IC20 values are in Table S2. The tested concentrations ranged from 0.78 to 100 µM, except for RSC, whose maximal tested concentration was 10,000 µM (Colour figure online)

The results of our study, where all chemicals were assessed in parallel on three different in vitro models by the same SK reaction-based method, demonstrate some differences in the sensitivity of the used in vitro models, regarding both NIS inhibition and cytotoxicity (Table 1). To provide transparent results, we report not only the mean effective values but also their standard deviations across at least 3 independent experiments for both endpoints (Table S2), which are mostly not reported in previous literature. Nine of the tested chemicals were detected as active NIS inhibitors with six of them active at non-cytotoxic concentrations in at least two of the three in vitro models used in our study (HEK293T NIS02 and FRTL-5; Table 1, Figs. 1 and 2). For eight of these chemicals, our data are consistent with the literature (Table S2). In the case of RSC, we have detected NIS inhibition at concentrations beyond the range tested in ToxCast. The lower concentrations corresponding to those tested in ToxCast, where it was categorized as inactive, also did not cause significant inhibition in our assays. TPP was active in two of our in vitro models without significant cytotoxicity, allowing us to quantify its IC50 (Fig. 2, Table 1, S2). It was also recognized as active in ToxCast, but there was also significant cytotoxicity at the same concentrations (Table S2). For TBBPA, NIS-inhibition activity was very close to the cytotoxicity effect in both our study and ToxCast, so it cannot be classified as a specific NIS inhibitor.

Our data also align with the literature for the five inactive chemicals for which any previous data were available. Only in the case of PFOA, recognized as inactive in all three cell models in our study, the literature data were ambiguous. It was also recognized as inactive in ToxCast (US EPA 2024) and Wang et al. (2018), which was based on an older version of ToxCast dataset, but it was active at relatively high concentrations in RAIU in FRTL-5 cells and the hNIS-HEK293T-EPA cell model in Buckalew et al. (2020). Since we used the same rat cell model and did not detect NIS inhibition, it might be possible that the RAIU assay was more sensitive than the SK assay. However, this is not supported by the ToxCast data (US EPA 2024), which do not show significant NIS inhibition with the same detection method and human cell model as in Buckalew et al. (2020). Thus, considering this is the only discrepancy in categorization of chemicals between our study and the literature, the SK assay appears to have similar sensitivity to the more complex RAIU assay. The IC50 levels detected using HEK293T NIS02 and FRTL-5 cell models were also very similar to those detected with the RAIU method in the literature (Table S2).

There are differences in cytotoxic concentrations of BPA, TCS, and PFOS across various in vitro models across studies. In our study and literature, BPA showed NIS inhibitory effects at relatively high µM IC50 concentrations near cytotoxicity limits. PFOS and TCS demonstrated NIS inhibition with lower IC50 values, and cytotoxicity findings varying across studies (Table S2). For an independent orthogonal cytotoxicity assessment, we have evaluated their cytotoxicity also using the CTG assay, which confirmed identical cytotoxicity cutoff values in both the NR and CTG assays (Figure S6). We attribute discrepancies primarily to variations in cell sensitivity across the studies but differences in definition and calculation of cytotoxicity cutoff values can play a role here. These findings emphasize the need to assess cytotoxicity alongside NIS inhibition, ensuring that identified NIS inhibitors are genuine and not merely artifacts of cell death, which can obscure inhibition effects. This has been also highlighted throughout the application of the Radioactive Iodine Uptake assay (RAIU) in a study using hNIS-HEK293T-EPA test model, which emphasized the importance of distinguishing between true NIS inhibition and nonspecific reductions in iodide uptake caused by cell death or compromised cellular health (Buckalew et al. 2020).

In our study, the HEK293T NIS 02 model showed the highest sensitivity to NIS inhibition among the three tested models, while NIS C was more prone to cytotoxic effects and exhibited lower sensitivity to NIS inhibition, despite having higher NIS-expression levels (Figure S3, S4; Table 1). The increased cytotoxicity in the NIS C model may result from the specific genomic integration site of the NIS-expression construct, potentially affecting genes vital for cellular metabolism. Alternatively, high NIS activity in this model might disrupt cellular homeostasis. Comparisons among HEK293T NIS 01, NIS 02, and NIS C clones—all transfected with the same vector—highlight that assay performance depends on factors beyond the vector itself. The number and genomic location of integrated gene constructs may influence the expression of other essential genes and possibly also the translocation of NIS protein to the membrane (Oh and Ahn 2021). These factors could indirectly impact cell model performance. This should be taken into account when comparing the data from different in vitro models overexpressing NIS. This can at least partly account for the differences in IC50 values across various human NIS-based in vitro models.

The rat FRTL-5 model, with intrinsic NIS expression, produced NIS-inhibition data generally consistent with the NIS 02 cell model (Figs. 1, 2, and Table 1), identifying the same compounds as active or inactive with comparable or slightly higher IC50 values. This consistency supports the relevance of the newly established human NIS 02 cell model, which makes it suitable for NIS-inhibition studies. While the FRTL-5 cell line showed comparable results as HEK293T NIS02, it was not as suitable model for efficient testing of chemicals due to its slow growth rate and expensive growth medium. In addition, NIS C with its lower sensitivity to NIS inhibition and higher sensitivity to cytotoxicity is less suitable for this purpose.

Although differences between our results and published data were generally minor, our results show there can be sensitivity differences even among similar in vitro models expressing the same NIS gene, as seen between HEK293T NIS02 and NIS C cell lines, which share the same transfection vector. While various in vitro models can effectively identify stronger NIS inhibitors, their ability to identify weak inhibitors may be compromised if they are more sensitive to general cytotoxic effects. IC50 values for NIS inhibition should be interpreted cautiously in parallel with cytotoxic effects and ideally validated with models of greater in vivo relevance. A key difference from in vivo systems remains the absence of biotransformation mechanisms in these in vitro models.

To better mimic the in vivo situation, the established NIS-inhibition assay was retrofitted with an external biotransformation system (BTS). Based on its demonstrated sensitivity and applicability as well as human relevance, the HEK293T NIS 02 cell line was selected as the test model for investigating the effect of BTS on the overall assay outcomes. Biotransformation was addressed by incorporating a pre-exposure phase, which included the incubation of the test chemicals with S9 and respective cofactors. A set of compounds previously identified (Table 1, Figs. 1 and 2) as effective NIS inhibitors in HEK293T NIS 02 cells at non-cytotoxic concentrations, namely BPA, DBP, ETX, MET, PCL, PFOS, RSC, TCS, and TPP, was chosen for these follow-up experiments. Both NIS inhibition and cell viability were assessed concurrently (Figs. 3, 4, S7).

Results from the biotransformation (BTS)-augmented NIS-inhibition assay using HEK293T NIS02 cells under four experimental conditions (detailed below), each assessed in 3 independent experiments. The Y-axis shows FOC% (fraction of no uptake control), representing NIS inhibition, where 100% FOC indicates complete inhibition (based on cell-free controls). The names and abbreviations of the tested chemicals are provided in Table 1. Dots represent the mean normalized NIS activity per exposure concentration, whereas the whiskers represent the standard deviation across replicates. The statistically significant differences among BTS setups are indicated by different letters (p < 0.05). The BTS setups were as follows: W/O S9 and W/cof (green curve), representing chemicals treated without the S9 fraction but with all cofactors added; D S9, W/cof (orange curve), where chemicals were treated with denatured S9 and all cofactors added; W/S9, W/cof (purple curve), indicating treatment with both S9 fraction and cofactors; and SM (blue curve), the standard NIS-inhibition assay method (Colour figure online)

Results from the biotransformation (BTS)-augmented NIS-inhibition assay using HEK293T NIS02 cells under four experimental conditions (detailed below), each assessed in 3 independent experiments. The Y-axis shows FOC% (fraction of control), representing NIS inhibition, where 100% FOC indicates complete inhibition (based on cell-free/no uptake controls). Chemical abbreviations are explained in Table 1 and Table S1. Dots represent the mean normalized NIS activity per exposure concentration, whereas the whiskers represent the standard deviation across replicates. The BTS setups were as follows: W/O S9 and W/cof (green curve), representing chemicals treated without the S9 fraction but with all cofactors added; D S9, W/cof (orange curve), where chemicals were treated with denatured S9 and all cofactors added; W/S9, W/cof (purple curve), indicating treatment with both S9 fraction and cofactors; and SM (blue curve), the standard NIS-inhibition assay method. The statistically significant differences among BTS setups are indicated by different letters (p < 0.05), “n.s.” indicates no significant differences (Colour figure online)

Incorporating BTS into the NIS assays impacted the effects of the tested compounds in three different ways, as compared to the standard NIS assay. First, we recorded a statistically significant reduction in NIS inhibition for DBP, TCS, and BPA (Fig. 3) in the subgroups utilizing a fully active BTS (“W/S9, W/cof”). Contrarily, the variants consisting of denatured S9 (“D S9, W/ cof”) and cofactors-only (“W/O S9, W/cof”) did not lead to a statistically significant impact on NIS inhibition, also when compared to the NIS method without BTS (“SM” – standard method) that was run in parallel.

We derived IC20 values and their ratios for all NIS inhibitors, that depicted statistically significant differences between fully active BTS systems and respective controls for a facilitated quantitative comparison by dividing the IC20 of the fully active BTS setup (“W/S9, W/cof”) by the IC20 of the denatured BTS setup (“D S9, W/cof”) (see Table S3 for the summary). IC20 values were prioritized for this comparison as the NIS inhibition often did not reach up to 50% effect in the fully active BTS setup (Fig. 3). The preincubation in fully active BTS setup increased IC20 values by approximately 5 (TCS), 6.3 (BPA), and more than ninefold (DBP) as compared to the denatured BTS setup (Table S3). According to those results, these compounds were metabolized by the added BTS and subsequently caused less NIS inhibition. The impact of compound bioavailability can be ruled out as the control setups showed no statistically significant differences in IC20 values among the standard method and the different BTS control variants.

Second, for another group of investigated compounds (PCL, PFOS, and RSC; Fig. 4A, B, C), no significant differences were observed among the effects of the variants with the active BTS setup and BTS-related control setups. As the concentration–response patterns were almost identical in between test setups, we deduced that these respective compounds are not impacted by biotransformation, at least in their NIS-inhibition ability within this specific in vitro assay.

Finally, a set of compounds (ETX, MET and TPP, Fig. 4D, E, F) demonstrated varying, setup-specific concentration–response patterns, which can most likely be attributed to kinetic compartmentalization effects, impacting bioavailability. Importantly, no biotransformation-related effect changes were detected for these compounds, as there were no statistically significant differences between their effects in fully active BTS setups compared to setups containing denatured S9. As mentioned previously, ETX and MET were specifically added to the set of test compounds due to their hydrophobicity (log Kow > 5). Indeed, differences in effect measures are evident for ETX (log Kow = 5.6) between the standard method (“SM”) and all BTS setups that include a 5-h preincubation step (Fig. 4D and Table S3). Although glass vessels were utilized, preincubation likely diminished free available ETX concentrations by binding to the vessel or losses of the chemical by evaporation, although the incubation vials were tightly sealed and with minimal head space. Interestingly, for MET (log Kow = 5.1), effect measures differed similarly between the standard method and all BTS-related setups; however, a statistical difference was also computed between S9-bearing setups (“W/S9, W/cof” and “D S9, W/cof”) and the “W/O S9, W/cof” control (Fig. 4E, Table S3).

This difference could be attributed to “serum-mediated passive dosing” (SMPD Fischer et al. 2018; Lungu-Mitea et al. 2021; Proença et al. 2021). SMPD describes the characteristics of hydrophobic chemicals binding to structural lipids or proteins that are present in certain in vitro test systems, such as serum albumin. Therefore, structural lipids or proteins can act as a sink for hydrophobic compounds and will impact the steady-state equilibrium of freely available concentrations. In our example, we do not employ serum albumin in the BTS preincubation step, but the structural protein within the active or denatured S9 can act as a sink. Within the “W/O S9, W/cof” setups, available exposure concentrations of MET and TPP had been partly lost, and we saw no or reduced effect on NIS inhibition, whereas all S9-containing setups retained some amount of available compound—bound to protein—that was still causing NIS-inhibition effect. Hence, we can also explain the significant differences in effect measures for TPP (log Kow = 4.6; Fig. 4F and Table S1 and S3) via SMPD. Due to its lower hydrophobicity compared to ETX and MET, less available concentration is lost during preincubation to surface binding or evaporation (compare “SM” with “W/O S9, W/cof” for TPP and MET). Still, this effect is overcompensated by the presence of S9 protein (setups “W/S9, W/cof” and “D S9, W/cof”). By binding to the available S9 in the specific BTS setups during the preincubation step, more S9-bound TPP is available post-incubation via SMPD-related partitioning. Consequently, we observed a statistically significant increase in the overall NIS-inhibition effect for the BTS setups that included active and denatured S9 compared to the setup w/cof without BTS (Fig. 4F). Notably, in TPP’s specific case, bioavailability-related effects are rather minuscule compared to MET (Fig. 4E). We conclude that the biotransformation-augmented NIS assay is suitable for chemicals with a log Kow up to approximately 5. However, further testing of compounds within that hydrophobicity margin is necessary to substantiate our results.

To contextualize our experimental results, we employed the in silico BioTransformer 3.0 tool (Wishart et al. 2022); a rule-based, machine-learning algorithm capable of predicting biotransformation metabolites from SMILES inputs and naming the potentially involved enzymatic actors and reactions of detoxification. The computed results (Table S4) corroborate our experimental data. The algorithm predicted the occurrence of phase I and II biotransformation for BPA, DBP, and TCS, involving Cytochrome P450 1A2 (CYP1A2)-mediated hydrolysis reactions, Uridine diphosphate glucuronosyl transferase (UGT)-mediated glucuronidations, Sulfotransferase (SULT)-mediated sulfonations, and Glutathione S-transferase (GST)-mediated conjugations. This prediction corresponds to less NIS inhibition by these compounds after BTS treatment (Fig. 3). Per se, all predicted phase I and II reaction types align with the technical capabilities of the employed BTS, as all necessary cofactors were included in the used BTS (see method section on BTS).

The BioTransfomer algorithm also predicted the metabolization of TPP via CYP2B6-mediated (Cytochrome P450 2B6) dehalogenative photocyclization, which was not mirrored by any decrease of NIS-inhibition activity in our in vitro results (Fig. 4F). This might be caused by comparable NIS-inhibition potential of its metabolites or insufficient TPP metabolization. Notably, human CYP2B6, as BioTransformer outputs relate to human nomenclature, corresponds to CYP2B1 in rats (Rattus norvegicus) (Vanoye-Carlo et al. 2015). Rat CYP2B1 expression has been primarily located within the liver, lungs, and brain (Miksys et al. 2000) and has been shown to be inducible via phenobarbital (PB) treatment. However, PB-mediated CYP2B1 induction in rat liver was reportedly 300-fold lower as compared to CYP1A1 induction for identically treated female Wistar rats (Chanyshev et al.