Differentiated thyroid cancer (DTC), accounting for 90–95% of thyroid malignancies [1], typically carries a favorable prognosis following standard therapies, including surgery, radioactive iodine (RAI), and thyroid-stimulating hormone suppression [2]. However, 5–20% of patients experience recurrence or metastasis within 20–25 years, and 10–15% develop distant metastases [3, 4]. For these metastatic cases, RAI remains a cornerstone treatment. Yet its efficacy is hampered by two intertwined challenges: dose-limiting toxicity and RAI resistance, particularly in metastases that retain iodine uptake capacity (e.g., bone metastases) but evade apoptosis, a phenomenon termed RAI-refractory (RAI-R) DTC [3]. RAIR-DTC patients face dismal outcomes, with a median survival of 3–5 years [5], underscoring the urgent need to unravel resistance mechanisms. The 2016 ATA guidelines 3 define four RAI-R categories, from lesions with no uptake to those that retain uptake but do not respond [3]. In this study we focus specifically on the ATA category characterized by measurable ¹³¹I uptake but failure to achieve a clinical response (i.e., lesions that take up radioiodine yet remain refractory), because the therapeutic strategy of impairing DNA repair to potentiate ¹³¹I cytotoxicity is most relevant to tumors that receive radioactive iodide but fail to undergo cell death.

RAI exerts cytotoxicity primarily through ionizing radiation–induced DNA damage, classically attributed to DNA double-strand breaks (DSBs) [6]. Unlike external beam radiotherapy, RAI efficacy relies on functional sodium/iodide symporter (NIS)–mediated uptake, a process frequently impaired in advanced disease [7]. Nevertheless, approximately 47% of metastatic DTCs retain NIS expression but remain RAI-resistant [4], suggesting that resistance involves factors beyond iodine transport.

One critical, yet underexplored, contributor is the DNA damage response (DDR). RAI-induced DSBs activate the apical DDR kinase ataxia-telangiectasia mutated (ATM), which regulates cell cycle arrest, DNA repair, and apoptosis [8–10]. Paradoxically, excessive DDR activation may promote tumor survival by enabling DSB repair and disrupting apoptosis [11]. This duality is mirrored in clinical observations: cells from Ataxia Telangiectasia (A-T) patients (lacking functional ATM) exhibit radiosensitivity due to defective DSB repair [12], and ATM inhibition in thyroid cancer cells has been shown to restore NIS function and enhance RAI uptake and cytotoxicity [13].

While NIS downregulation remains the primary focus in RAI-R research, the role of DDR, particularly ATM signaling, is less understood. Notably, ATM not only orchestrates DSB repair but also influences NIS expression [13], linking DDR directly to RAI responsiveness. Although ATM inhibitors have been evaluated in combination with external radiation across various tumor types [14, 15], their interaction with internal radiation modalities like RAI has not been systematically studied.

Here, we investigate the role of ATM in DTC progression and RAI resistance. We assess ATM expression across thyroid cancer stages and evaluate whether ATM inhibition can sensitize DTC cells to RAI both in vitro and in vivo. Our study aims to provide mechanistic insights into how ATM activity may promote resistance to RAI and to explore its potential as a therapeutic target in RAIR-DTC.

An 89-sample tissue microarray from Shanghai Tenth People's Hospital, comprising normal, cancerous, and metastatic thyroid tissues, was used. All samples were obtained with patient consent.

Human samples (normal thyroid, TC, paraneoplastic, and metastatic tissues specimens) were obtained with the patient's consent through the Shanghai Tenth People's Hospital. This study was approved by the Institutional Review Board of Shanghai Tenth People's Hospital (approval number: 24KN265). To identify the expression of ATM in TCs, a tissue microarray (TMA) was constructed, including 89 samples, including normal thyroid tissues, thyroid cancer (TC) tissues, paraneoplastic tissues, and metastatic lymph nodes (2 mm tissue core per sample). All tissue samples were fixed in formalin and embedded in paraffin for IHC. Samples were dewaxed with xylene and then rehydrated in anhydrous ethanol solution. Sections were then incubated overnight at 4°C with an anti-ATM antibody (1:100; #2873T, CST) and followed by incubation with the secondary antibody. Two independent pathologists assessed the positivity and staining intensity of the tissue sections blindly.

The expression levels were assessed based on the staining intensity (0 for no staining, 1 for weak staining, 2 for moderate staining, and 3 for strong staining) and the positive cell ratio (0 for < 10%, 1 for 10 to < 50%, and 2 for ≥ 50% cell). The histochemistry score (H-SCORE) was calculated as follows: H-SCORE = (percentage of cells with weak staining × 1) + (percentage of cells with medium staining × 2) + (percentage of cells with strong staining × 3).

To enhance the understanding of the role of ATM in thyroid cancer, the Cancer exploration module from the TIMER2.0 database was utilized to investigate the correlation between ATM expression and the expression levels of genes associated with thyroid cancer dedifferentiation. The Gene_Corr module analyzes the relationship between ATM expression and the BRAF, KRAS, MTOR, and PIK3CA gene expression. A correlation coefficient of 0.1 ≤ ρ ≤ 1.0 indicates a significant correlation between ATM and thyroid cancer dedifferentiation genes, while a p < 0.05 denotes statistical significance.

The K1 (RRID: CVCL_2537), BCPAP (RRID: CVCL_0153), and TPC-1 (RRID༚CVCL_6298) cell lines were procured from the American Type Culture Collection (ATCC, Manassas, VA, USA). K1 cells were incubated in medium DMEM (Gibco) supplemented with 1% Penicillin-Streptomycin solution (100x Gibco) and 10% FBS (Avantor). The cells were maintained at 37°C in 5% CO2. BCPAP and TPC-1 cells were incubated in medium RPMI-1640 (Gibco) supplemented with 1% Penicillin-Streptomycin solution (100x Gibco), and 10% FBS (Avantor). The cells were maintained at 37°C in 5% CO2. Treatments included AZD1390 (5–50 nM) and ¹³¹I (0.74–4.6 MBq/mL), alone or combined.

SPF nude BALB/c female mice (4–6 weeks old, weighing approximately 20 g) were purchased from Shanghai SLAC Laboratory Animal Co., Ltd., China, and placed in the SPF facility of the Tenth People's Hospital, Tongji University School of Medicine. To improve RAI uptake in mouse tumors, thyroid tissues in mice were removed by electroacupuncture before tumor cell injection. Postoperatively, they were given a normal diet of 0.1% calcium gluconate (to prevent calcium deficiency due to intra-operative parathyroid damage) to feed. Levothyroxine was also supplemented, and L-T4 at a dose of 14ug/Kg BW was given daily by gavage to prevent hypothyroidism until 7 days before treatment with RAI. Shanghai Tenth People Hospital's Animal Ethics Committee approved these experiments according to the Guidelines for the Care and Use of Laboratory Animals (approval number: SHDSYY-2024-4456).

Xenograft tumors in nude mice: K1 cells (1×10⁶) were implanted into mice through hypodermic injection, the mice were randomized into four groups (n = 7): vehicle (intraperitoneal injection of 100uL saline), RAI (a single intraperitoneal injection of 37 MBq of ¹³¹I solution per mice once), iATM (per mice was injected intraperitoneally with 5mg/kg AZD1390 QD for 12 days), RAI + iATM (a single intraperitoneal injection of 37 MBq of ¹³¹I solution once and intraperitoneal 5mg/Kg AZD1390 QD per mice for 12 days). These mice were housed in a clean barrier system at 20–25°C and 50–60% humidity, with free access to food and water. Tumor growth was monitored by digital calipers every 3 days, and tumor volumes were recorded using the following formula: Volume = (longer diameter × shorter diameter²)/2. All mice were executed on day 12.

Histological analyses of the mice's tumors: Tissues were fixed in 10% formalin at room temperature, treated with anhydrous ethanol and xylene, and embedded in paraffin. The thickness of paraffin-embedded tissue sections was 5 µm and was stained with H&E, Ki-67, and TUNEL, respectively. Immunohistochemical staining was performed using anti-γH2AX (1:200, #NB100-638, Novus Biologicals), and the slides were scanned with a digital pathology slide scanner (KFBIO). The scanning results were statistically positive ratio using ImageJ software.

CCK-8: Cell viability was examined using the Cell Counting Kit-8 (CCK-8) (#GK10001, GLPBIO) assay. The cells were plated in 96-well plates (about 5×10³ cells/well in 100 µL culture medium) and incubated for 48h. Then we replaced the medium with 0-1500uCi/mL concentration of ¹³¹I (adding 5nM KI per well) or 0-300uM AZD1390 2.3 MBq/mL ¹³¹I + 0-300nM AZD1390 (adding 5nM KI per well). After maintaining 48h or 96h, the CCK-8 working solutions were added. Then the cells were incubated continuously for 2h at 37℃. Next, a microplate reader (Tecan, Infinite 200Pro) was used to detect each group's light absorption values at 450nm. Cell survival could be displayed by the ratio of absorption values, with the calculation formula of: (dosing-blank) / (control-blank).

The synergistic effect of RAI and ATM inhibitor was demonstrated through isohologram analysis. The CompuSyn software generates a CI value to quantify the interactions between drugs [17]. This CI value was calculated following the methodology outlined by Chou et al. [18, 19]. A CI value approaching 1 suggests additive effects of the drug combinations, while values less than 1 indicate synergy, and values greater than 1 indicate antagonism.

Immunofluorescence staining of dsDNA and ssDNA: Cells were collected at the indicated times to detect cytosolic DNA in K1 cells after RAI irradiation. K1 cells were fixed with 4% PFA and permeated with 5% Triton X-100. Then blocked with 1% BSA in PBS, and incubated with anti-dsDNA (1:400; #MAB1293, AE2, Millipore) or anti-ssDNA (1:400, # MAB3299, F7-26, Millipore) primary antibody overnight at 4℃. After 3 washes with PBS, the cells were incubated with donkey anti-mouse-Alexa Fluor 555 (#A-31570, Thermo Fisher) for 1h at room temperature. Subsequently, DAPI staining was performed at room temperature for 10 min. Confocal microscopy was conducted using a Zeiss LSM900 microscope with a 60x oil-immersion lens following standard protocol, and images were analyzed using Image J software.

Single-Cell Analysis Indicates ATM as a Driver of Thyroid Cancer Progression.

Single-cell RNA sequencing (scRNA-seq) of 28 thyroid tissue samples, including 15 anaplastic thyroid cancers (ATC), 7 papillary thyroid cancers (PTC), and 6 normal thyroid follicular epithelial cell samples delineated the trajectory of malignant progression [26]. Uniform Manifold Approximation and Projection (UMAP) revealed eight distinct cell populations [26], including epithelial, immune subsets, fibroblasts, endothelial cells (Fig. 1A, S1A-B), validated by marker gene profiling (KRT18 for epithelium, PECAM1 for endothelium) (Fig. 1B).

To visualize the dynamic transcriptional evolution of epithelial cells, we reconstructed a pseudotime developmental trajectory using Monocle3 (Fig. 1C & Fig. S1C). This trajectory clearly revealed a gradual transition from normal thyroid follicular cells (NORM) through PTC to ATC, indicating a continuous dedifferentiation process during tumor evolution. KEGG pathway enrichment along this trajectory revealed a stepwise activation of metabolic (e.g., ribosome biogenesis, oxidative phosphorylation) and oncogenic signaling pathways (e.g., PI3K-Akt, relaxin) (Fig. S1D). A comparative analysis highlighted the ATC-specific upregulation of cell cycle (adjusted P < 0.05) and chromosomal instability genes (BUB1, MCM2) compared to NORM (Fig. 1D-E). Critically, among cell cycle-associated genes, ATM was identified as progressively upregulated during dedifferentiation (Fig. 1F), implicating it as a potential driver of genomic instability and aggressive tumor progression.

Clinical Validation of ATM Overexpression in Advanced and RAI-Resistant Thyroid Cancer.

We validated ATM's clinical relevance via immunohistochemistry on 89 thyroid samples, including normal thyroid tissues, primary tumors, adjacent tissues, and metastatic lymph nodes (Fig. S2A-B and Supplemental Table 1). Consistent with scRNA-seq data, ATM protein levels were significantly elevated in thyroid cancer tissues compared to adjacent normal tissues (p < 0.05) (Fig. 2A). ATM expression was highest in ATC tissues, significantly exceeding levels in PTC and follicular thyroid carcinoma (FTC) samples (P < 0.0001) (Fig. 2B).

In RAI-refractory (RAI-R) PTC, ATM expression was substantially elevated, with a median score of 21.82 (95% CI: 12.59–29.20), compared to radioiodine-avid (RIA) PTC cases (median score 4.85, 95% CI: 3.17–12.99) and normal paraneoplastic tissues (median score 2.89, 95% CI: 1.31–8.44) (Fig. 2C). Metastatic lymph nodes from RAI-R PTC cases exhibited higher ATM scores (median 7.20, 95% CI: 5.60–16.60) compared to those from RIA cases (median 2.33, 95% CI: 0.73–13.84) (Fig. 2D). Correlation analysis using the TIMER2.0 database revealed positive associations between ATM and oncogenic drivers including BRAF (ρ = 0.546), KRAS (ρ = 0.636), MTOR (ρ = 0.546), and PIK3CA (ρ = 0.704) (Fig. 2E). Figure 2F presents representative images of ATM staining for different tissue types.

Targeting ATM Inhibition Synergizes with RAI to Suppress Tumor Growth In Vivo.

To evaluate ATM inhibition in vivo, we selected K1 cells for xenograft studies due to their selective ¹³¹I uptake (Fig. S3A), as baseline ATM expression did not differ significantly across tested cell lines (Fig. S3B). Having established this model, we next ensured that the ATM inhibitor AZD1390 itself did not interfere with the fundamental mechanism of RAI action. A radioactive uptake assay confirmed that AZD1390 treatment did not alter ¹³¹I accumulation in K1 cells compared to the control (Fig. S3C).

Tumor-bearing nude mice were treated with RAI (37 MBq, single dose), the ATM inhibitor AZD1390 (5 mg/kg, daily), or the combination for 12 days (Fig. 3A, B). Tumor volumes were significantly reduced in the combination group compared to monotherapies (P < 0.01, two-way ANOVA) (Fig. 3C). Final tumor weights were also significantly lower in the combination group.

Immunohistochemistry analysis of excised tumors showed decreased Ki-67 staining and increased TUNEL staining in the combination group (P < 0.001) (Fig. 3D). While ¹³¹I alone activated the DDR, evidenced by γH2AX foci accumulation, co-treatment with AZD1390 suppressed γH2AX levels, indicating compromised DDR signaling (Fig. 3D). These findings suggest that ATM inhibition potentiates RAI efficacy by impairing DNA repair and promoting apoptosis.

ATM Blockade Potentiates RAI-Induced Cytotoxicity and Suppresses Migration In Vitro.

In vitro, ATM inhibition radiosensitized K1 cells treated with ¹³¹I (0–55.5 MBq/mL) or AZD1390 (0–1200 nM) for 48–92 hours. CCK-8 assays revealed time- and dose-dependent inhibition of cell proliferation by ¹³¹I, while AZD1390 alone showed minimal cytotoxicity (Fig. 4A-B). Co-treatment with ¹³¹I (2.3 MBq/mL) and AZD1390 (1.2–75 nM) significantly reduced cell viability compared to monotherapies (Fig. 4C). CompuSyn analysis confirmed synergistic inhibition (combination index < 1) (Fig. 4D) [27, 28].Colony formation assays demonstrated reduced clonogenic survival with the combination therapy (P < 0.001), with fewer colonies observed at higher AZD1390 doses (Fig. 4E-F). Annexin V/PI staining showed activity-dependent increases in apoptosis upon ¹³¹I treatment (P < 0.001), which was significantly enhanced by combination therapy (P < 0.01) (Fig. 4G–J). We established stable ATM-knockdown K1 cells (shATM) to genetically validate on-target effects (Fig. 4K). We first assessed the impact of ATM depletion on cell migration. Wound-healing assays revealed that ATM knockdown significantly impaired the migratory capacity of K1 cells compared to the non-targeting control (shNC) (p < 0.05) (Fig. 4L-M). We next asked whether genetic ATM knockdown could recapitulate the chemosensitization effect observed with AZD1390. Flow cytometric analysis of Annexin V/PI staining demonstrated that ATM knockdown alone induced an increase in apoptosis (p < 0.01). Crucially, the combination of ATM knockdown and RAI treatment resulted in a synergistic enhancement of apoptotic cell death, significantly exceeding the effects of either single treatment (p < 0.0001) (Fig. S4A-B). This genetic evidence unequivocally confirms that the sensitization to RAI is an on-target effect of ATM ablation.

RAI Induces Genotoxic Stress with Limited Detectable Double-Strand Breaks in RAI-R DTC cells.

RNA-seq of RAI-treated K1 cells identified 1,144 differentially expressed genes (|log2FC| >1.3, FDR < 0.05) (Fig. S5A) [29–34]. GO and KEGG analyses revealed enrichment of DNA repair, cell cycle regulation, and G1/S transition pathways (Fig. 5A, Fig. S5C). GSEA demonstrated downregulation of mitosis- and G2/M checkpoint-related genes, alongside activation of DNA repair pathways (FDR < 0.05) (Fig. 5B).

Transcriptomic analysis revealed differential expression of numerous DNA repair genes in response to RAI. These included a modest upregulation of key components of the base excision repair (BER) pathway, such as POLB (Fig. 5C), which is consistent with a cellular response to RAI-induced oxidative base damage, a known source of apurinic/apyrimidinic (AP) sites. EdU staining confirmed a time-dependent decrease in proliferation after RAI (Fig. 5D, K), but TUNEL and dsDNA staining did not reveal a sustained increase in detectable double-strand breaks (Fig. 5F–I, K). Comet assays showed limited DNA fragmentation under these experimental conditions in ¹³¹I-treated cells compared to DNase I controls (Fig. 5G, K). Notably, RAI induced modest increases in single-stranded DNA (ssDNA) accumulation (P < 0.01) (Fig. 5H–J, K). These results suggest that in K1 cells with high ATM activity, RAI-induced double-strand breaks, if present, are rapidly resolved, whereas non-DSB lesions such as AP sites may persist and constitute a dominant source of unresolved genotoxic stress.

ATM Inhibition Exacerbates Genotoxic Stress and Induces Mitotic Catastrophe.

Flow cytometry revealed that ¹³¹I (2.3–4.6 MBq/mL) induced G2/M arrest in K1 cells, with dose-dependent enrichment (Fig. 6A–B, D). While ¹³¹I alone caused S-phase accumulation, co-treatment with AZD1390 (50 nM) further augmented G2/M arrest (Fig. 6C–E). Western blotting confirmed that ATM inhibition attenuated phosphorylation of checkpoint kinases (pChk2, pCDC2), suggesting dysregulated G2/M checkpoint control and premature mitotic entry. (Fig. 6F–G, Fig. S6A).

To assess DNA damage persistence, cells were irradiated with ¹³¹I (4.6 MBq/mL) for 24 hours, washed to remove residual RAI, and then incubated with or without AZD1390. Co-treatment of ¹³¹I-AZD1390 resulted in a time-dependent accumulation of AP sites compared with RAI alone. At the same time, cell survival was reduced, emphasizing the cytotoxic consequences of impaired DNA repair (Fig. 6H). These results support that ATM inhibition disables DDR checkpoints, driving cells with unresolved damage into mitosis, ultimately triggering mitotic catastrophe and apoptosis.

RAI remains a cornerstone therapy for DTC [6], yet its efficacy is often limited by intrinsic or acquired resistance. Our study identifies ATM as a molecular determinant associated with disease progression and therapeutic resistance in thyroid cancer. By integrating single-cell transcriptomics, clinical TMAs, functional preclinical models and in vitro mechanistic explorations, we delineate a central role for ATM in promoting dedifferentiation, genomic instability, and escape from RAI-induced cytotoxicity.

Single-cell RNA-sequencing revealed a continuum from normal thyroid follicular cells through papillary thyroid carcinoma (PTC) to anaplastic thyroid carcinoma (ATC), with a stepwise upregulation of ATM along this dedifferentiation trajectory. Pseudotemporal analysis showed progressive ATM activation during tumor evolution, coinciding with cell cycle, DNA replication, and chromosomal instability programs. These findings were validated in clinical specimens, where RAI-R tumors exhibited significantly higher ATM levels than RAI-sensitive counterparts. ATM expression positively correlated with oncogenic drivers (BRAF, KRAS, PIK3CA [35]), suggesting its integration into pro-tumorigenic signaling networks. While these data strongly associate ATM upregulation with the dedifferentiated state, future studies manipulating ATM expression (e.g., knockdown/overexpression) in thyroid models are needed to definitively establish its causal role in driving dedifferentiation. These findings align with studies in breast and gastric cancers [36, 37], where ATM overexpression is linked to both oncogenic transformation and therapy resistance, underscoring its context-dependent roles in cancer biology [38]. Notably, in other cancer types, ATM has been implicated in epithelial-to-mesenchymal transition and loss of differentiation markers, supporting the concept that it may play a similar role in thyroid carcinogenesis [39].

In preclinical models, both pharmacologic and genetic ATM suppression synergized with RAI to inhibit tumor growth, induce apoptosis, and impair migration. This consistency confirms ATM's role in RAI resistance. Mechanistically, under the experimental conditions used, RAI treatment was associated with prominent accumulation of AP sites arising from oxidative base damage, while persistent or readily detectable double-strand breaks were limited. While γH2AX was observed post-RAI, comet assays confirmed no significant DSB induction. This observation aligns with the physical properties of ¹³¹I, which is primarily a low-LET β particle emitter. Unlike high-LET radiation, the sparse ionization events produced by electrons and β particles predominantly cause localized oxidative base damage and single-strand breaks, which are significantly more frequent than immediate DSBs [40]. These lesions often manifest as non-DSB clustered DNA damages, including a high density of AP sites [41]. Our findings suggest that in the context of high ATM expression, these pervasive sublethal lesions are the primary genotoxic stress managed by the cell's repair machinery. Instead, γH2AX likely reflects replication stress from unresolved AP sites, as persistent AP lesions stall replication forks, activating ATR and downstream DDR signaling [42]. ATM inhibition exacerbated this stress by abrogating G2/M checkpoint control (pChk and pCDC2 downregulation), forcing cells with unrepaired AP sites into mitosis, where replication errors culminated in mitotic catastrophe. While AP sites are canonically repaired via base excision repair (BER), the role of ATM in this context is uncharted. Intriguingly, ATM inhibition exacerbated AP site retention, suggesting its potential involvement in BER coordination. This corroborates earlier findings indicating that ATM or it’s signaling intermediates govern oxidative DNA damage responses [43, 44]. And researchers have shown that ATM directly participates in BER through phosphorylation of TDP1 at Ser81, a modification that enables XRCC1-dependent recruitment of TDP1 to sites of base lesions [45]. The limited detection of SSB intermediates suggests they are either transiently repaired or AP sites are processed via long-patch BER. This aligns with studies showing that oxidative agents, such as ionizing radiation, can directly generate AP sites without requiring SSB intermediates [46, 47]. Beyond its well-established role in DSB repair, our data suggest that ATM also contributes to the cellular response to RAI-induced AP site–associated genotoxic stress, potentially through coordination of BER and cell cycle checkpoints. Critically, ATM inhibition prolonged AP site retention, suggesting that ATM facilitates AP site resolution potentially through direct BER coordination or indirect checkpoint-mediated repair windows. This mechanism represents a distinct paradigm of synthetic lethality, diverging from PARP inhibitor strategies in BRCA-mutant cancers; here, it is the massive accumulation of 'hidden' sublethal AP sites, rather than a canonical deficiency in DSB repair, that drives genomic instability and cell death.

In addition to DNA-damage–related pathways, alterations in membrane transporters and ion-channel regulators have also been implicated in RAI refractoriness. A recent transcriptomic analysis identified ANO1 as a potential biomarker distinguishing RAI-avid from non-avid thyroid tumors [48]. Although ANO1 was not examined in the present study, its reported association underscores that multiple molecular mechanisms, including DNA damage response, oxidative stress, and altered ion transport, may collectively determine RAI uptake and treatment sensitivity.

The therapeutic implication is twofold: (1) ATM upregulation serves as a biomarker of aggressive, RAI-resistant thyroid cancer; and (2) pharmacologic inhibition of ATM may re-sensitize these tumors to RAI by dismantling DDR checkpoints and enhancing genotoxic burden. Importantly, our findings parallel similar observations in other solid tumors, where ATM deficiency or inhibition exacerbates therapeutic DNA damage, underscoring a broader applicability of ATM-targeted radiosensitization strategies.

In summary, this study establishes ATM as a key driver of thyroid cancer progression and a therapeutic vulnerability in RAI resistance. While its direct causal role in dedifferentiation requires further validation, our data position ATM as a critical contributor to the RAI-resistant phenotype. Our findings are most applicable to tumors that retain measurable radioiodine uptake but are clinically refractory (ATA category 4), and do not imply that ATM inhibition will restore radioiodine responsiveness in tumors that lack iodide uptake. These insights provide a strong mechanistic rationale for incorporating ATM inhibitors into treatment regimens for advanced or refractory thyroid cancers. Future clinical trials of ATM inhibition combined with RAI may offer a novel strategy to overcome radioiodine resistance and improve patient outcomes, and future work extending these findings to ATC and other dedifferentiated thyroid cancer models will further clarify the role of ATM in tumor progression and therapeutic resistance.