We queried the AOP-wiki (https://aopwiki.org/) in April 2023, to identify AOPs relevant to THSD following the general extraction logic described by Haigis et al. (Haigis et al., 2023).

All AOPs involving thyroid hormone synthesis, metabolism, transport, signaling, or thyroid hormone–dependent biological processes were screened. AOPs were retained if their taxonomic applicability included mammals (human and/or rodents) given the objective of identifying effect biomarkers (EBMs) relevant for human epidemiology. For each AOP, we extracted the molecular initiating events (MIEs), key events (KEs), KE relationships (KERs), and adverse outcomes (AOs).

All identified thyroid-related AOPs and their species applicability are provided in Supplementary Table S1, and detailed MIE, KE, and AO information for mammalian-relevant AOPs is provided in Supplementary Table S2. These AOPs served as the basis for the network presented in Figure 1. Non-mammalian AOPs (e.g., amphibian metamorphosis, swim bladder inflation) were recorded but excluded from biomarker evaluation due to limited direct human relevance.

Mammalian-relevant AOPs were organized into a network to visualize overlapping pathway events and biological connectivity (Figure 1). Manual clustering was performed based on shared late KEs and AOs, consistent with Haigis et al. (Haigis et al., 2023). This approach allowed grouping of AOPs with common downstream pathways to facilitate the subsequent identification of EBMs aligned to specific neurodevelopmental, endocrine, and renal outcomes. The resulting clustered network (Figure 1) informed the selection of late KEs and AOs for targeted biomarker identification.

For each MIE and KE, all measurement methods described in AOP-Wiki were extracted, including molecular, cellular, tissue-level, and functional readouts. Each method was evaluated for its suitability for human epidemiological studies, considering:

biological matrix availability (blood, urine, saliva, cerebrospinal fluid, tissue)

Measures requiring invasive sampling (e.g., brain tissue histology) were noted as mechanistically informative but not feasible for routine epidemiological use.

Where AOP-Wiki did not identify feasible human measurement options, we performed targeted PubMed searches. Search terms combined biomarker-related keywords with KE-specific biological terms and applied filters for human or translational evidence, English language, and publication years 2018–2023.

The search workflow is illustrated in Figure 2, and full search strings are provided in Supplementary Table S3. Titles and abstracts were screened to identify studies reporting EBMs aligned to the late KEs and AOs.

Candidate EBMs were then evaluated based on their relevance and potential for use in human studies. Specifically, we considered: a) their mechanistic alignment with the corresponding AOP KEs and AOs, b) availability of measurable endpoints in human biological matrices, and c) feasibility for implementation in epidemiological settings, including scalability in population-based research. When human data was already available, we reviewed this evidence to assess applicability and maturity. Biomarkers supported mainly by experimental studies were noted as promising candidates that will require additional validation before widespread use in human populations. Markers that rely on highly invasive sampling, such as cerebrospinal fluid or post-mortem brain tissue, were not considered feasible for routine human biomonitoring and are presented separately in Supplementary Table S4, as they remain valuable from a mechanistic research perspective.

The results are organized according to the four mechanistic clusters identified in the AOP network (Figure 1). For each cluster, we describe the relevant biological context, the associated KEs and AOs, and the EBMs identified through our screening process. Where applicable, we highlight EBMs that are already measurable in human biological matrices and have been used in epidemiological or clinical studies, as well as those that appear promising but currently lack sufficient validation in human populations. We also note instances where no suitable biomarker could be identified, reflecting areas where further development and translational research are needed. Summary tables (Tables 1–4) present an overview of biomarker applicability across clusters, while detailed extraction sheets are provided in Supplementary Tables S1–S4.

Five AOPs (8, 42, 134, 152, 300) with similar KEs on alterations of the hippocampus were identified (Figure 1; Cluster 1). The hippocampal KEs were altered hippocampal gene expression (KE 756), altered hippocampal anatomy (KE 757), and altered hippocampal physiology (KE 758). The five AOPs had two common AOs: loss of cochlear function (AO 319) and decreased cognitive function (AO 402).

We found human relevant measurement methods being described in AOP-wiki, however, the methods were generally not applicable for epidemiological studies as they required brain tissue, e.g., to measure altered gene expression in the hippocampus (Zhang et al., 2002). Some non-invasive, but demanding methods were described in AOP-wiki, such as electroencephalography and structural and functional magnetic resonance imaging. These techniques have been used in several human studies to assess associations between thyroid hormone levels and hippocampal volume/function and cognitive function (Smith and Jonides, 1997; Willoughby et al., 2014; Wheeler et al., 2015; Gilbert et al., 1995). Even though the techniques are well accepted, they are not feasible for larger epidemiological studies due to the requirement of extensive equipment and procedures.

Thus, we searched PubMed for potential EBMs for hippocampal alterations. Of the 56 studies, 8 studies (Morrison et al., 2023; Ginsberg et al., 2019; Liu et al., 2021; Han et al., 2021; Reemst et al., 2022; Fusilier et al., 2021; Islam et al., 2021; Gasparini et al., 2020) were published in 2018 or later and included relevant information on possible EBMs (Table 1). Even though two AOs (decrease cognitive function (AO 402) and loss of cochlear function (AO 319)) were included in the 5 AOPs relevant to this search (Figure 1), all 8 included studies focused on decreased cognitive function and especially Alzheimer’s disease, reflecting the available human evidence base for hippocampal dysfunction.

One study (Morrison et al., 2023) measured the hippocampal grading in humans with magnetic resonance imaging using scoring by Nonlocal Image Patch Estimator (SNIPE) (Table 1). They were able to accurately classify people with normal cognitive function from those with cognitive impairment and Alzheimer’s disease. Even though the tool could be promising for detecting early/mild cognitive impairment in a clinical setting, using magnetic resonance imaging in larger epidemiological studies may be too inconvenient and logistically challenging (Morrison et al., 2023).

The remaining studies measured gene expression levels in postmortem brain/hippocampus tissue, in either humans (Ginsberg et al., 2019; Liu et al., 2021; Han et al., 2021; Reemst et al., 2022) or mice (Fusilier et al., 2021; Islam et al., 2021; Gasparini et al., 2020) (Table 1). Before the possible EBMs can be used in epidemiological studies, they need to be validated in non-/low invasive human matrices such as blood, urine, or saliva.

One of the most promising EBMs identified is the brain-derived neurotrophic factor (BDNF), which has been identified previously as an EBM for neurological outcomes/behavioural function (Rodríguez-Carrillo et al., 2023a; Mustieles et al., 2022). It has already been verified in several epidemiological studies showing that BDNF can be measured in serum, plasma, and urine. High serum BDNF level has also recently been associated with decreased risk of poststroke cognitive impairment after 3 months (Chang et al., 2024). However, to our knowledge, there are no studies in human investigating if the BDNF level in blood and urine is correlated with the BDNF level in the hippocampus, however it is likely, as studies have shown positive correlations between blood and hippocampal BDNF levels in animal models (Klein et al., 2011).

Many of the included studies utilized samples from Alzheimer’s patients (Morrison et al., 2023; Ginsberg et al., 2019; Liu et al., 2021; Han et al., 2021) or from mouse models for Alzheimer’s disease (Fusilier et al., 2021) (Table 1). Alzheimer’s is known for its impact on the hippocampus. It's important to note that some of the EBMs might only emerge after Alzheimer’s disease has become evident. Therefore, these biomarkers may not be effective in early detection of cognitive impairment in epidemiological studies. Some thyroid hormones (free triiodthyronin (fT3) and total triiodthyronin (T3)) have been associated with Alzheimer’s, whereas no associations were seen for thyroid-stimulating hormone (TSH) or thyroxine (T4) in a meta-analysis (Dolatshahi et al., 2023).

One AOP (54) on impaired learning and memory was identified (Figure 1; Cluster 2). The AOP includes 1 MIE and 9 KEs, and is reviewed and endorsed. We have focused our search on EBMs on the four late KEs: 1. reduced levels of BDNF (KE 381), 2. decrease of GABAergic interneurons (KE 851), 3. decrease of synaptogenesis (KE 385), and 4. decrease of neuronal network function (KE 386).

The KEs are generally well described in AOP-wiki, and the description also includes information on measurement methods. AOP-wiki includes measurement methods for BDNF applicable for epidemiological studies, and BDNF can be measured by available commercial ELISA kits in whole blood, serum, plasma, platelets, and urine (Trajkovska et al., 2007). Furthermore, the EBM has been validated as a marker for neurological outcomes/behavioural function in epidemiological studies (Mustieles et al., 2022; Rodríguez-Carrillo et al., 2023b).

For the remaining KEs, the measurement methods in AOP-wiki were less applicable for epidemiological studies, including for instance immunohistochemical staining on brain tissue and different types of electroencephalography measurements. Therefore, we searched PubMed on relevant novel EBMs for the remaining three late KEs. Twenty articles (Woodworth et al., 2018; Kivimäki et al., 2021; Mullins et al., 2021; Rübsamen et al., 2022; Cortese et al., 2020; Kantrowitz et al., 2021; McClure et al., 2022; Dyrba et al., 2018; Richard et al., 2023; Dong et al., 2023; Kaar et al., 2019; López-Cerdán et al., 2022; Christiansen et al., 2018; Logan et al., 2020; Fagan et al., 2021; Ng et al., 2021; Blennow et al., 2019; Fabbri et al., 2018; Morphett et al., 2024; Gillespie et al., 2024) published in 2018 or later included relevant information on possible EBMs (Table 2). Of the evaluated studies, 8 studies (Woodworth et al., 2018; Kivimäki et al., 2021; Mullins et al., 2021; Rübsamen et al., 2022; Cortese et al., 2020; Kantrowitz et al., 2021; McClure et al., 2022; Dyrba et al., 2018) included potential future EBMs that could be used for human epidemiological studies.

One study (Woodworth et al., 2018) measured urinary protein levels of metalloproteinase-9 (MMP9) and MMP9/Neutrophil gelatinase-associated lipocalin (NGAL) complex and found that higher levels may enhance neuronal damage in the brainstem nuclei and enhance plasticity of sensorimotor regions in human patients with urological chronic pelvic pain syndrome (Table 2). Further validation of the associations with the KEs and the AOs in healthy individuals are, however, needed before implementation of the EBMs in epidemiological studies.

Four studies used blood as a matrix and measured protein levels (Kivimäki et al., 2021; Rübsamen et al., 2022), gene expressions (Mullins et al., 2021), and vitamin D (Cortese et al., 2020) (Table 2). The studies were conducted in people with diseases (multiple sclerosis, bipolar disorder, Alzheimer’s disease, sporadic Creutzfeldt-Jakob disease) (Kivimäki et al., 2021; Mullins et al., 2021; Rübsamen et al., 2022; Cortese et al., 2020). Thus, in all cases further validation of the associations with the KEs and the AOs in healthy human individuals are needed before implementation of the EBMs in epidemiological studies.

Three studies (Kantrowitz et al., 2021; McClure et al., 2022; Dyrba et al., 2018), with potential future EBMs, used brain imaging of rodent and human brains to evaluate the neuron function (Table 2). These methods can be used in human studies, however, for larger epidemiological studies the inconvenience and logistic challenges may be difficult to overcome.

The remaining 12 studies (Richard et al., 2023; Dong et al., 2023; Kaar et al., 2019; López-Cerdán et al., 2022; Christiansen et al., 2018; Logan et al., 2020; Fagan et al., 2021; Ng et al., 2021; Blennow et al., 2019; Fabbri et al., 2018; Morphett et al., 2024; Gillespie et al., 2024) without potential EBMs recommended for human epidemiological studies mainly used invasive samples (e.g., brain tissue and cerebrospinal fluid) as a matrix. These studies may increase the understanding of the mechanisms involved in the pathways towards AOs and disease (Supplementary Table S4).

Three AOPs (110, 119, 162) with increase in adenomas/carcinomas in thyroid follicular cells as AO (AO 741) were identified, and included several identical KEs (Figure 1; Cluster 3). We have focused our search on EBMs for the two late KEs: increase in hypertrophy and proliferation (KE 739) and increase in hyperplasia (KE 740) in follicular cells.

The description of the two KEs in AOP-wiki is very limited and does not include any information on measurement methods. Therefore, we searched PubMed, and of 115 hits, 13 articles were published in 2018 or later and included relevant information on possible EBMs for increase in hypertrophy, proliferation, and hyperplasia thyroid and/or adenomas/carcinomas in thyroid follicular cells (Zhang P. et al., 2020; Ameziane et al., 2019; Halsall and Oddy, 2021; Yan et al., 2021; Wang et al., 2020; Bertoni et al., 2021; Pang and Yang, 2021; Wang et al., 2018; Zhou et al., 2018; Hong et al., 2023; Bertoni et al., 2019; Mostoufi-Moab et al., 2018; Qu et al., 2020).

The information on possible EBMs in the 13 included studies are listed in Table 3.

A potential EBM in blood is interleukin-34 (IL-34), which is increased in both tumour tissue and serum of papillary thyroid cancer patients compared with age-matched controls (Zhang P. et al., 2020) (Table 3). The expression of serum IL-34 was also significantly associated with tumour size, tumour stage, and lymph node metastasis (Zhang P. et al., 2020). IL-34 has also been found to play a role in many other cancer types, such as hepatocarcinoma, osteosarcoma, multiple myeloma, colon cancer, and lung cancer (Franzè et al., 2020). Furthermore, IL-34 has been linked to several other disease types including autoimmune disorders, infections, neurological disorders and metabolic diseases, indicating that it is not very specific as EBM (Baghdadi et al., 2018).

Two other EBMs measurable in serum, oxidative stress (Ameziane et al., 2019) and reverse triiodothyronine (rT3) (Halsall and Oddy, 2021), have been identified in the literature (Table 3). They are, however, too unspecific to be EBMs alone on increase in hypertrophy, proliferation, hyperplasia and/or adenomas/carcinomas in thyroid follicular cells but might serve as EBMs in combination with other markers.

Eight studies (Yan et al., 2021; Wang et al., 2020; Bertoni et al., 2021; Pang and Yang, 2021; Wang et al., 2018; Zhou et al., 2018; Hong et al., 2023; Bertoni et al., 2019) have identified differences in expression of genes, microRNA and circularRNA, as well as slicing variants and adenosine monophosphatase (AMPase) activity between tumour tissue and matched adjacent or normal non-cancerous tissue (Table 3). These markers may serve as future EBMs, if the differences can be detected in blood samples or other non-/low invasive matrixes which can be collected in larger epidemiological studies. Development and validation of these suggested EBMs need further studies.

Two studies (Mostoufi-Moab et al., 2018; Qu et al., 2020) found somatic and germline DNA variations present in the thyroid tumours compared to non-cancerous tissues (Table 3). The somatic variation might serve as future EBMs if detectable in blood samples. Both studies found somatic variations related to the receptor tyrosine kinase (RTK)/RAS pathway (Mostoufi-Moab et al., 2018; Qu et al., 2020), which could be relevant to investigate more in relation to EBMs, as well as the hippo pathway identified by Qu et al. (Qu et al., 2020).

One AOP (128) on kidney toxicity was identified as a thyroid-related AOP (Figure 1; Cluster 4). The AOP includes two events related to thyroid hormone balance (MIE 277 and KE 281), increased serum creatinine (KE 813) and seven late KEs related to kidney function including cystic dilatation (KE 823), cytoplasmic vacuolization (KE 818 and KE 824), decrease renal plasma flow (KE 820), sodium reabsorption (KE 821), ability to dilute urine (KE 825), and glomerular filtration (KE 819).

The descriptions of the AOP and KEs in AOP-wiki are very limited and do not include any information on measurement methods. However, several EBMs of kidney function have already been identified and reviewed by others (Zare et al., 2021; Rodríguez-Carrillo et al., 2023a; Lopez-Giacoman and Madero, 2015; Fassett et al., 2011; Wasung et al., 2015; Kidney Disease: Improving Global Outcomes KDIGO, 2013; Younes-Ibrahim and Younes-Ibrahim, 2022). Therefore, we have not conducted additional systematic literature searches, but Table 4 comprises the most relevant information on the already identified EBMs. In Table 4, we have also indicated the potential relationship between the EBMs and the KEs. However, establishing definitive connections is challenging due to the limited information available in AOP-wiki regarding the AOP and KEs.

Utilizing a panel of EBMs can provide a more comprehensive assessment of kidney damage and functional changes and help to identify early signs of kidney injury that may not be detectable with traditional markers alone. The U.S. Food and Drug Administration (FDA) has in 2018 assessed a panel of six markers for kidney tubular injury in clinical Phase 1 trials for drugs (U.S. Food and Drug Administration, 2020): clusterin (CLU), cystatin-C (CysC), kidney injury molecule-1 (KIM-1), N-acetyl-beta-d-glucosaminidase (NAG), neutrophil gelatinase-associated lipocalin (NGAL), and osteopontin (OPN). The U.S. FDA determined that the Composite Measure composed of these six urinary biomarkers was qualified as a “safety composite biomarker panel to be used in conjunction with traditional measures to aid in the detection of kidney tubular injury in phase 1 trials in healthy volunteers when there is an a priori concern that a drug may cause renal tubular injury in humans” (U.S. Food and Drug Administration, 2020).

Integrating AOPs into epidemiological research provides a mechanistic framework for linking chemical exposures (environmental pollutants or drugs) to health outcomes. Many widely used chemicals engage the molecular initiating events (MIEs) within the presented AOP network (Figure 1). Here, we summarize some of the broadly prevalent exposures that can act through these events, with emphasis on three MIEs present in multiple AOPs: decreased thyroid hormone synthesis, inhibition of thyroid peroxidase (TPO), and inhibition of the sodium/iodide symporter (NIS).

Decreased thyroid hormone synthesis (MIE/KE 277) is part of all four identified AOP clusters and included in six of the AOPs (Figure 1; Supplementary Tables S2a, b). It is one of the most essential events for THSD, and generally included as a KE, but for AOP 128 it is given as the MIE. Deceased thyroid hormone synthesis cannot be measured directly but is often measured indirectly by TPO and NIS inhibition or by T4 serum levels.

Inhibition of TPO (MIE 279) is the MIE of two included AOPs (42 and 119) leading to decreased cognitive function and follicular cell adenomas and carcinomas, respectively. TPO is the enzyme responsible for the de novo synthesis of thyroid hormones, and its inhibition can disrupt systemic thyroid hormone homeostasis. High-throughput screening of over 1000 chemicals from the ToxCast library identified more than 300 putative TPO inhibitors, spanning pharmaceuticals, pesticides, industrial compounds, and consumer product ingredients (Paul et al., 2016). Among these, bisphenols and phthalates stand out due to their widespread use in plastics, food packaging, and personal care products, resulting in ubiquitous exposure across the general population (Gerofke et al., 2024; Govarts et al., 2023). Observational evidence suggests associations between exposure to these compounds and altered circulating thyroid hormone levels in humans (Milczarek-Banach et al., 2020; Kim et al., 2019). Given the prevalence of bisphenols and phthalates in everyday environments and their demonstrated potential to interfere with thyroid hormone synthesis via TPO inhibition, further epidemiological studies to validate the effects in humans are warranted. Focusing on the AOs of the included AOPs (decreased cognitive function and follicular cell adenomas/carcinomas) and aim to validate the identified EBMs.

Inhibition of NIS (MIE 424) leads to decreased cognitive function, impaired learning and memory, and follicular cell adenomas and carcinomas in the included AOP (54, 110, and 134). NIS is essential for iodide uptake into thyroid follicular cells, the initial step for thyroid hormone synthesis. Recent high-throughput screening of over 1000 environmental chemicals identified 273 compounds with significant inhibition of NIS-mediated iodide uptake, highlighting the broad chemical space with potential to impair thyroid hormone synthesis (Wang et al., 2019). Several of the identified NIS inhibitors are common exposures in the general population, such as perchlorate, thiocyanate and nitrate that occur in drinking water, foods and tobacco smoke and have been associated with altered circulating thyroid hormone levels in population studies, with stronger effects in women of lower iodine status (Blount et al., 2006; Blount et al., 2007; Bruce et al., 2013; King et al., 2023). Several per- and polyfluoroalkyl substances (PFAS) have also been identified as potential NIS inhibitors in humans (Stoker et al., 2023). Population exposure to PFAS is well documented worldwide, including Europe, the United States, Asia and the Arctic (Richterová et al., 2023; Botelho et al., 2025; Kookana et al., 2025; Bonefeld-Jørgensen et al., 2023), underscoring their epidemiologic relevance. Accordingly, epidemiological studies should evaluate associations between NIS inhibitors, such as perchlorate, thiocyanate, nitrate and PFAS, and health effects via AOP-aligned EBMs.

The 11 AOPs included in this study are at various stages of development, highlighting the evolving nature of thyroid hormone-related research within the AOP framework. Two AOPs (42 (cluster 1) and 54 (cluster 2)) have received endorsement (Supplementary Table S1), indicating a higher level of scientific consensus and validation, and provide a strong foundation for regulatory applications. The 9 AOPs (within cluster 1, 3, and 4 and AOP 402 without any events) are under development with several requiring additional empirical evidence to strengthen key event relationships (KERs) and improve predictive utility for epidemiological studies (Supplementary Table S1).

The developmental stage of the AOP and associated events directly affect the reliability of the identified EBMs. EBMs linked to well-developed AOPs and validated KEs are more reliable and applicable for human studies, while those associated with less-developed pathways require further validation before they can be widely used in population-based research. The variability in the developmental status of the AOPs highlights the need for continued research to enhance their robustness and regulatory acceptance. Future efforts should prioritize the validation of key events and measurement methods, ensuring their applicability for human biomonitoring and risk assessment purposes.

All AOPs with a mammal (human and rodent) applicability domain were included in the search of EBMs for epidemiological studies. Five AOPs (42, 54, 128, 134, 300) explicitly state human relevance, while six (8, 110, 119, 152, 162, 402) had rat and/or mouse as the taxonomic applicability domain (Supplementary Table S1). This inclusive approach ensures broader coverage, capturing a wide range of potential EBMs. Although animal models provide valuable mechanistic insights, human relevance must be validated, especially for AOPs such as those describing follicular cell adenomas and carcinomas, which may not be directly applicable to humans.

Thyroid disruption can profoundly affect human health, particularly neurodevelopment during critical windows such as the foetal and early postnatal periods. Despite the comprehensive nature of current AOPs for THSD, there may still be endpoints that are not fully covered (Noyes et al., 2019). Current AOPs primarily focus on well-characterized MIEs and KEs, but there are gaps in understanding the full spectrum of thyroid disruption effects (Noyes et al., 2019).

The use of EBMs in epidemiological studies offers significant advantages in THSD. EBMs can provide early indications of biological changes, facilitate timely intervention and improve risk assessment. They can also enhance the specificity and sensitivity of epidemiological studies, allowing for more accurate detection of thyroid-related health effects. Specific EBM for the thyroid hormone system, such as serum TSH and thyroxine (T4), are already well-established in clinical practice for monitoring thyroid function (D'Aurizio et al., 2023; Van Uytfanghe et al., 2023). However, novel EBMs, including the neurotoxicity marker (BDNF) and kidney markers (e.g., NGAL and KIM-1), have shown promise in linking thyroid disruption to more specific health outcomes. When individual EBMs exhibit limited specificity, their interpretative value in epidemiological studies can be strengthened through the combined use of multiple EBMs capturing complementary biological processes within an AOP. Such panel-based approaches can improve biological plausibility and robustness, particularly for complex outcomes where multiple mechanistically linked pathways contribute to the adverse outcome.

By integrating EBMs into the AOP framework, we can bridge the gap between mechanistic toxicology (in-vitro and animal studies) and epidemiological research, ultimately leading to better understanding and management of THSD’s impact on human health. This approach not only enhances the predictive power of AOPs but also supports the development of targeted interventions and regulatory policies to mitigate the adverse effects of thyroid disruptors. However, validation of the EBMs is needed to ensure the validity of the results. Thus, while the use of BDNF and kidney markers seem to be implemented and validated in several epidemiological studies, further work is needed to validate other novel EBMs (Tables 1–4).

To support practical implementation of AOP-aligned EBMs in human studies, we prioritized the identified EBMs according to their biological specificity, feasibility in commonly used human biological matrices, strength of human evidence, and remaining validation needs (Table 5). This provides a translational link between the mechanistic evidence captured in the AOPs and the requirements of population-based research and regulation.

Brain-derived neurotrophic factor (BDNF) and the kidney injury biomarkers (clusterin (CLU), cystatin-C (CysC), kidney injury molecule-1 (KIM-1), N-acetyl-beta-d-glucosaminidase (NAG), neutrophil gelatinase-associated lipocalin (NGAL), osteopontin (OPN)) emerged as high-priority candidates, as they are biologically relevant, measurable in minimally invasive matrices, and increasingly supported by human data. Calcitonin is promising for C-cell neoplasia but has limited epidemiological applicability due to the rarity of this outcome. Interleukin-34 (IL-34) and oxidative stress markers are mechanistically relevant but require further specificity and validation before broad use.

Overall, this prioritization highlights a set of EBMs that can be deployed in epidemiological research now, as well as those requiring targeted development to enable regulatory uptake and refinement of thyroid-related AOPs.