Introduction

Follicular cell-derived TCs mostly develop through the aberrant growing of endodermal layer-derived follicular epithelial cells and comprise the main histological differentiated and undifferentiated subtypes1,2. Differentiated TCs include the most common papillary (PTC, 80-85%) and follicular (FTC, 10-15%) TCs. Although most differentiated TCs are indolent tumors with a favorable outcome, 5–20% of them have a high risk of relapse and death3. Anaplastic TC (ATC) represents 1-2% of all TC cases and behaves very aggressively by invading adjacent tissues and metastasizing distant organs4. ATC can be derived either from the follicular thyroid epithelium gland by losing the biological features, including the iodine uptake (de novo), or more frequently from a dedifferentiation process of a pre-existing differentiated TC, following the accumulation of multiple mutations5. Despite the intense efforts to improve therapeutic regimens, there are no effective therapies available for metastatic TC patients. The majority of TC metastases do not express the sodium/iodine symporter (NIS) and are not responsive to radioactive iodine, the most effective non-surgical therapy in TC6,7,8. Thus, there is an urgent need to develop experimental models to study the fundamental mechanisms underlying tumorigenesis and metastasis formation, particularly in radioiodine-resistant patients.

Despite major advances in the know-how of TC’s hallmarks and biological behavior, the cell subpopulation in the lineage hierarchy that serves as cell of origin for the different TC histotypes, following the acquisition of somatic mutations, remains unknown. A critical issue argues whether the different TC histotypes mirror a distinct transformed cell subpopulation or the same target cell harboring different genetic mutations. The existence of different follicular histotypes of cells-derived TCs endowed with distinctive biological behavior led to formulate several carcinogenesis models that describe the cell of origin1,2. However, these models do not reflect the phenotypic and genetic intertumoral heterogeneity of follicular cell-derived TCs. In adults mammals, tissue homeostasis and repair of injured organs depends on small reservoirs of tissue-specific stem cells able to self-renew and differentiate. Indeed, the hierarchical organization of adult tissue has been accomplished to slow aging, help the organism recover from injury, and finally protect cells from accumulating damage that would ultimately lead to cancer. Although the progenitor cells with limited lifespan escape the risk of accumulating mutations, adult stem cells and their immediate offsprings could potentially be transformed either by acquiring mutations inducing self- renewal, or by inheriting existing mutations from stem cells9. This model introduced the concept of tumorigenesis assuming stem cells as source of tumor formation in light of their longevity and innate self-renewal capacity, essential for the accumulation of genetic or epigenetic alterations10,11. To explain the different tumor histopathology and behavior, a genetic mutation model, in which a variety of genetic mutations occurs within the same “target cell”, was conceptualized2. For most cancers, the “target cell” of transforming mutations is unknown. Nevertheless, there are considerable evidence that certain types of leukemia arise from mutations which accumulate in hematopoietic progenitor cells12,13,14. This observation mostly reflects the need to better understand the oncogenic events which occur in tissue-resident stem cells in order to explain the transition from non-malignant hyper-proliferative lesions to well-established cancers endowed with different biological behaviors. Although we and other research groups have identified a subpopulation of cancer stem cells (CSCs)/progenitor cells that is crucially involved in the TC initiation, the transformation stage in the hierarchy lineage remains elusive15. Here, we show the ability of human embryonic stem cells (hESCs) to recapitulate the cellular hierarchy of thyroid gland and reproduce the oncogenic process upon CRISPR-based genetic editing.

Results

hESCs engineered with the most common TC genetic alterations recapitulate TC histotypes

To establish a model that mirrors the follicular cell-derived TC histotypes, hESCs-derived thyroid progenitor cells (TPCs) and their differentiation lineages were obtained following Longmire’s established protocol16 and genetically modified using a CRISPR/Cas9 technology by introducing the most common TC genetic alterations including BRAF, NRAS and TP53 (Fig. 1A and Supplementary Fig. 1a). According to literature, cells at day 6, upon activation of BMP-FGF2 signaling, promoted thyroid lineage specification gradually increasing, up to day 30, the expression levels of thyroid markers PAX8, TTF1, TSH-R, TPO, Tg and NIS (Supplementary Figs. 1b–e)16,17,18. D22 cells triggered the expression of specific thyroid lineage markers, PAX8 and TTF1, which are retained and concomitantly complemented, at D26 and D30, by the activation of thyroid differentiation markers, TSH-R, TPO, Tg, and NIS (Supplementary Figs. 1c–e). Besides, D22 cells expressed high levels of other putative thyroid progenitor cell hallmarks such as CD133, ABCG2, Nestin and HNF-4α19, along with the thyroid transcription factors HHEX and FOXE120 (Supplementary Figs. 1f–j). All together these data indicate that D22 thyroid lineage is enriched in thyroid progenitor cells (TPCs). Mature lung markers SCGB1A1, AQP5, SFTPB and SFTPC were barely detectable in the thyroid lineage from days 22 to 30 (Supplementary Fig. 1k). While throughout hESC differentiation, cells at days 0, 6 and 22 were progressively enriched in G0-G1 and depleted in G2-M/S phase fractions, most of the differentiated thyroid precursors at days 26 and 30 displayed a slight decrease of G0-G1 and increase of G2-M phases (Fig. 1B).

Fig. 1: hESCs engineered with the most common TC genetic alterations recapitulate thyroid cancer histotypes.
figure 1

A Model of directed differentiation of human embryonic stem cell (hESC) into thyroid lineage. To promote thyroid lineage specification, hESCs were exposed to the indicated stimuli. Treatment with FGF10, KGF, BMP4 and EGF is reported as FKBE. B Cell cycle analysis in non-targeting control (NTC) hESC-derived cells at the indicated stage of thyroid differentiation lineage. The data show percentage of cell number in G0/G1, S, and G2/M cell cycle phases. Data are expressed as mean ± SD of three independent experiments. C Cell proliferation in hESC-derived cells, engineered with the indicated mutations, at the indicated stage of thyroid differentiation lineage, at 7 days. D Invasion analysis in cells engineered as in (C) at 72 h. E Clonogenic assay in D22 thyroid progenitor cells (TPCs) engineered as in (C) at 21 days. For (C–E) statistical significance was calculated using the two-tailed unpaired t test and data are mean ± standard error of three independent experiments. F (upper panel) Growth kinetics of xenograft tumors generated by subcutaneous injection of D22 TPCs. (lower panel) Frequency of teratocarcinoma (TerC) or TCs obtained by the injection of hESC-derived cells harboring different mutational background, at different stages of thyroid differentiation lineage. Data are shown as mean ± SD. n = 12 mice per group. G H&E staining and immunohistochemistry analysis of thyroglobulin (Tg), cytokeratin 19 (CK19), β-catenin and NIS on xenograft tumors obtained following injection of D22 TPCs engineered with different mutational background and compared with patient-derived PTC, FTC and ATC. Number of tissues analyzed n = 5. Mutational status of human tissues: PTC BRAF mutated ID#6, FTC NRAS mutated, and ATC BRAF/TP53 mutated ID#96. Scale bars, 100 µm. H Correlation analysis in Gene Expression Omnibus (GEO) (GSE33630) of CTNNB1 mRNA expression levels in normal thyroid tissue, PTC and ATC. Boxes represent the interquartile range (IQR) and midline represents the median. Statistical significance was calculated using Kruskal-Wallis test. i Immunoblot analysis of β-catenin in D22 TPCs engineered as in (c). β-actin was used as loading control. Source data are provided as a Source Data file.

Full size image

D22 TPCs showed a similar cell cycle profile upon BRAF, NRAS and TP53 mutational background introduction. Sanger and Next-generation sequencing analyses exhibited a high-efficiency percentage of the introduced exogeneous genetic mutations that was consistent with a correlation coefficient of 0.85% (Supplementary Figs. 2a–c and Supplementary Table 1). hESC-derived anterior foregut endoderm at day 6 and D22 TPCs, harboring TP53R248Q in combination with BRAFV600E or NRASQ61R, showed more pronounced proliferative, invasive capabilities and clonogenic activity, compared to NTC control (Fig. 1C–E). Upon the acquisition of specific genetic alterations, except the inactivating point TP53R248Q mutation alone, exclusively cells at days 6 and 22 retained tumorigenic capacity (Fig. 1F). Subcutaneous injection of the genetically engineered D22 TPCs generated avatars that recapitulate the recognizable phenotypic and morphological features of follicular cell-derived TC histotypes, whereas D6 cells-derived xenograft masses resembled a teratocarcinoma phenotype expressing S-100, CDX2, Oct3/4 and p40 (Fig. 1G and Supplementary Fig. 2d).

Specifically, BRAFV600E and NRASQ61R-engineered D22 cells led to the formation of tumors that reflect the cellular heterogeneity of patient-derived PTC and FTC phenotype, respectively, mainly characterized by different degrees of nuclear atypia and variable mitotic activity (Fig. 1G).

Although the inactivation of p53 alone was not required for the tumor initiation, D22 TPCs harboring TP53R248Q in combination with BRAFV600E or NRASQ61R generated undifferentiated tumors with a pleomorphic and undifferentiated morphology recapitulating the heterogeneity and the phenotypic characteristics of ATC, expressing low levels of Tg, CK19 and NIS as compared with the well-differentiated TC histotypes (Fig. 1F, G and Supplementary Fig. 2e). In line with current knowledge, these data suggest that TP53 mutations are required for tumor evolution and a higher disease stage exclusively in cooperation with other oncogenic hits21. Accordingly, TP53 mutation was more frequently positively associated with undifferentiated tumors showing aggressive features and distant metastases22,23,24 (Supplementary Fig. 2f). This phenomenon is in line with the analysis of TP53 germline mutation distribution in a large cohort of patients affected by Li-Fraumeni syndrome, which highlighted a rare positive association with inherited risk of develo** TCs (Supplementary Fig. 2g). Moreover, in accordance with the positive association between CTNNB1 mRNA levels and aggressive histotypes of TC25, D22 TPCs engineered with TP53R248Q in combination with BRAFV600E or NRASQ61R mutations displayed β-catenin accumulation and nuclear localization (Fig. 1G–I and Supplementary Fig. 2h). In D22 cells, BRAFV600E and NRASQ61R alone or in combination with TP53R248Q mutation promoted the expression of CSC-, EMT-, and metastasis-related genes (Fig. 2A). These findings suggest that thyroid carcinogenesis follows a genetic mutation model in which the TPCs are the elective target of genetic alterations.

Fig. 2: The ternary complex TIMP1-MMP9-CD44 sustains the tumor initiation capability of engineered D22 TPCs.
figure 2

A Heatmap of Wnt-, stemness-, metastasis- and EMT- related genes in D22 TPCs and in D30 mature thyrocytes engineered with the indicated mutations. Data are presented as 2−ΔCt normalized values of three independent experiments. B Table and Venn diagrams of common (black color) and exclusive upregulated genes (red color) with logFC > 1.5 in D22 TPCs vs thyrocytes (D30). C Immunoblot analysis of TIMP1 and CD44 in hESC-derived cells harboring different mutational background, at D22 and D30. β-actin was used as loading control. One representative of three independent experiments is shown. Source data are provided as a Source Data file. D MMP9 production in cells as in (C) at 48 h. Statistical significance was calculated using the unpaired two-tailed t test. Data are expressed as mean ± SD of three independent experiments. E Growth kinetics of xenograft tumors generated by subcutaneous injection of the indicated engineered D22 TPCs and D30 cells overexpressing TIMP1, MMP9 and CD44 alone and in combination. Data are shown as mean ± SD. n = 10 mice per group. F Relative mRNA expression levels of TIMP1, MMP9 and CD44 in normal (n = 59) and tumor (n = 509) thyroid tissue from TGCA database, thyroid carcinoma (THCA) branch. Boxes represent the IQR and midline represents the median. Statistical significance was calculated using the Wilcoxon test. G (left panel) Immunohistochemical analysis of TIMP1, CD63, MMP9 and CD44v6 on xenografts obtained by injecting engineered D22 TPCs and compared with normal thyroid tissue, patient-derived PTC, FTC and ATC. Ctrl= negative control. The number of tissues analyzed n = 5. Mutational status of human tissues: PTC BRAF mutated ID#6, FTC NRAS mutated, and ATC BRAF/TP53 mutated ID#96. Scale bars, 100 µm. (right panel) Histograms represent the percentage of TIMP1, CD63, MMP9 and CD44v6 positive cells. Data are expressed as mean ± SD of three independent experiments. H CD44v6 flow cytometry analysis (orange histograms) and corresponding isotype-matched control (grey histograms), in engineered D22 TPCs and compared with isolated normal thyroid cells and patient-derived PTC, FTC and ATC cells.

Full size image

TIMP1-MMP9-CD44 complex sustains tumor initiation capability of engineered D22 TPCs

To explore the signaling pathway implicated in the tumorigenesis of engineered TPCs, we sought to analyze the reprogramming of the transcriptomic profile sustained by specific oncogenic alterations.

From transcriptomic analysis of engineered D22 TPCs with BRAFV600E or NRASQ61R alone and in combination with TP53R248Q mutation emerged a common higher expression of HGF, TIMP1, MMP9 and CD44 compared with D30 mature thyroid cells (Fig. 2A, B and Supplementary Fig. 3a). As the previously reported activation of HGF/MET signaling pathway in ATCs26, HGF was found among the top differentially upregulated genes in D22 compared with D30 (Fig. 2A, B).

These findings are in line with TIMP1 and CD44 protein expression patterns and MMP9 activity in genetically edited cells at days 22 and 30 stage of thyroid differentiation lineage (Fig. 2C, D, and Supplementary Fig. 3b). mRNA and protein analysis of hESCs at different stages of differentiation revealed that expression levels of TIMP1, MMP9 and CD44, in a time-dependent manner, increase from D0 to D22, and then decrease to D30 (Supplementary Figs. 3C–E). In accordance with the lack of tumorigenic activity, TP53R248Q mutation alone did not endorse the expression of these molecules (Fig. 2A–D and Supplementary Figs. 3a, b). Although TIMP1 is a natural inhibitor of MMPs involved in the matrix remodeling during carcinogenesis, it converges toward cancer cell survival by binding MMP9/CD44 complex and mediating PI3K/AKT pathway activation27,28. TIMP1, together with its binding protein CD63, has emerged as a predictive biomarker with regard to tumor therapeutic response and spreading29. Differentiated cells at days 26 and 30, characterized by negligible expression levels of TIMP1, MMP9 and CD44, were unable to induce tumor growth (Supplementary Figs. 3f, g). Interestingly, the overexpression of TIMP1 together with CD44 and MMP9, is able to significantly increase the proliferation rate and in lesser extent the invasive capacity of D30 cells harboring NRASQ61R/TP53R248Q mutations, whose cells have acquired a weak tumorigenic activity, tardily generating small tumors. This phenomenon could be attributed to the concurrent epigenetic alterations, induced by the ternary complex, along the differentiation process (Fig. 2E and Supplementary Figs. 4a–d).

In the light of these data, we investigated whether the expression of each member of the ternary complex TIMP1-MMP9-CD44 correlated with thyroid tumor. Analysis of a large cohort of primary TCs showed a significant positive association with the expression of TIMP1, MMP9 and CD44 (Fig. 2F and Supplementary Fig. 4e)30. Following the ternary complex formation, TIMP1 leads to the activation of MMP9, which in turn cleaves the CD44 extra-cellular domain hence releasing the intra-cellular fragment into the cytoplasm31. The CD44 fragment translocates into the nucleus, where it acts as a transcription factor, by regulating the expression of several target genes including the variant isoforms of CD44 (CD44v)32.

Some of CD44v have been considered as functional markers of cancer progression33. We have recently reported that constitutive and reprogrammed metastatic cancer cells are identifiable by the expression of CD44v6, which is positively associated with PI3K/AKT pathway activation34.

We next evaluated whether TIMP1, interacting with CD63, is required for cell survival and migration, cooperating with CD44v6. D22 cell-derived xenograft tumors, harboring BRAFV600E and NRASQ61R alone or in combination with TP53R248Q mutation expressed comparable levels of TIMP1, CD63, MMP9 and CD44v6 with those found in the matching PTC, FTC and ATC histotypes (Fig. 2G, H, and Supplementary Figs. 4f, g). In accordance with the inability to promote tumor initiation, TP53R248Q mutation alone lacked the capability to induce CD44v6 expression in D22 TPCs (Supplementary Fig. 4h). TIMP1 targeting significantly impaired the proliferative and invasive capacity of the engineered D22 TPCs, regardless the introduced genetic alteration, by impeding the activation of PI3K/AKT pathway, even in presence of TIMP1 exogenous cytokine, driven by a reduced expression level of CD44v634 (Fig. 3A–D, and Supplementary Figs. 5a–d).

Fig. 3: TIMP1 blockade and CD44v6 silencing significantly impair survival and invasive capacity of D22 TPCs.
figure 3

A Cell proliferation in D22 TPCs engineered with the indicated mutations treated with vehicle, TIMP1, TIMP1 inhibitor (TIMP1i) alone or in combination up to 3 days. B Invasion analysis in D22 TPCs harboring the indicated mutations treated as in (A) up to 4 days. For (A and B) statistical significance was calculated using the two-tailed unpaired t test and data are mean ± standard error of three independent experiments. C Immunoblot of pAKT and AKT in engineered D22 TPCs treated as in (A) for 48 h. β-actin was used as loading control. One representative of three independent experiments is shown. Source data are provided as a Source Data file. D Relative mRNA expression levels of CD44v6 in D22 TPCs engineered with the indicated mutations and treated with TIMP1 inhibitor (TIMP1i) for 24 h. Data are presented as fold change over vehicle ± SD of three independent experiments. E Immunoblot analysis of pAKT and AKT levels in D22 TPCs engineered with the indicated mutations and transduced with control shRNA (scramble, scr) or CD44v6 shRNA (shCD44v6). β-actin was used as loading control. One representative of three independent experiments is shown. Source data are provided as a Source Data file. f Cell proliferation in D22 TPCs engineered with the indicated mutations transduced with control shRNA (scramble, scr) or CD44v6 shRNA (shCD44v6), up to 72 h. G Invasion analysis in D22 TPCs harboring the indicated mutations transduced with control scramble (scr) or shCD44v6, up to 4 days. For (F and G) statistical significance was calculated using the two-tailed unpaired t test and data are mean ± standard error of three independent experiments. H Immunoblot analysis of pAKT and AKT in the indicated engineered D22 TPCs overexpressing CD44v6, untreated and treated with TIMP1i for 48 h. β-actin was used as loading control. One representative of three independent experiments is shown. Source data are provided as a Source Data file. I Heatmap of Wnt-, EMT- stemness and metastasis-related genes in D22 TPCs engineered with the indicated mutations and treated with vehicle or TIMP1i for 24 h. Data are presented as normalized 2−ΔCt values of three independent experiments.

Full size image

Transduction of engineered D22 TPCs with a lentiviral vector encoding for CD44v6 short hairpin RNA (shRNA) sequences, significantly decreased AKT phosphorylation, proliferative and invasive potential (Fig. 3E–G and Supplementary Fig. 5e–g). Conversely, CD44v6 overexpression led to an upregulation of pAKT, whose protein expression levels were restored by the targeting of TIMP1, to lower levels expressed by untreated TPCs (Fig. 3H and Supplementary Figs. 5h–j). These findings suggest that TIMP1 through CD44v6 is required for the activation of PI3K/AKT pathway.

Likewise, in the TPCs harboring the different mutation profiles, TIMP1 blockade downregulated the expression of Wnt, EMT- and stemness-related genes, as also confirmed by enrichment analysis computed with Reactome Database (Fig. 3I and Supplementary Figs. 6a–d). Specifically, targeting of TIMP1 hampered the mRNA levels of ß-catenin targets, such as AXIN2, MYC and CCND1 and the cell growth rate, which were both restored in presence of Wnt pathway agonists, Wnt3A and R-spondin1 (Supplementary Figs. 6e–h). In line with these findings, engineered D22 TPCs and their derived mouse avatars, showed similar transcriptomic profiles and an enrichment of pathways related to EMT- and stemness-associated signature, as well as PI3K/AKT and MAPK signaling, comparable to that observed in primary PTC- and ATC tissues (Fig. 4A–D and Supplementary Figs. 7a, b). Altogether, these data indicate that specific mutations regulate the expression of TIMP1/CD44v6 complex, which sustain the tumor growth of the different TC histotypes through the activation of PI3K/AKT pathway.

Fig. 4: MAPK and PI3K/AKT pathways, along with EMT-associated signature, are enriched in D22 TPCs, mouse avatars and human PTC and ATC.
figure 4

A (upper panels) Heatmaps showing the top 100 differentially expressed genes (DEGs) with logFC > 2 obtained by total transcriptome analysis (mRNAseq), in engineered D22 TPCs harboring the pathogenetic TC mutations versus their derived-xenograft tumors. (lower panels) Barplots indicate the significantly enriched pathways (p value < 0.05) performed using a Gene set enrichment analysis (GSEA) on Hallmark category from MSigDB database of differentially expressed genes (DEGs) obtained by RNAseq. B Venn diagram of the common enriched pathways in xenograft tumors versus D22 TPCs. List of the common 13 differentially regulated genes, associated with EMT-pathway. C Gene set enrichment analysis (GSEA) of EMT pathway in xeno NRAS/TP53 versus D22 NRAS/TP53 (ES = 0.549, NES = 2.378, p = 2.31E-07). Statistical significance was calculated using Benjamini & Hochberg test. D Heatmap showing normalized mRNA expression values of genes involved in PI3K/AKT-, MAPK-, EMT- and stemness-related pathways in the indicated engineered D22 TPCs, D22-derived xenograft tumors and human PTC and ATC. Mutational status of human tissues: PTC BRAF mutated ID#6, and ATC BRAF/TP53 mutated ID#96. Values are expressed as Log2 upper quartile (UQ).

Full size image

KISS1R is a prognostic factor and a potential therapeutic target in advanced TCs

Given that, BRAFV600E or NRASQ61R in combination with TP53R248Q mutations gave rise to tumors that resemble the ATC phenotype, we evaluated whether these genetic alterations are required for the metastatic potential of TPCs. Cells were orthotopically injected and their xenograft-derived lesions analyzed for the gene expression related to the metastatic signaling profile.

TPCs engineered for BRAFV600E/TP53R248Q and NRASQ61R/TP53R248Q, robustly generated orthotopic and metastatic tumors (Fig. 5A). Immunohistochemical analysis showed that TIMP1/MMP9/CD44 ternary complex expression in primary tumor avatars paralleled an increase of the expression levels of TWIST, SNAIL in the xenograft metastasis, resembling the phenotypic landscape of PTC and ATC patient-derived primary and metastatic lesions (Fig. 5A, B).

Fig. 5: KISS1R is a prognostic factor and a potential therapeutic target in advanced TCs.
figure 5

A (left panel) H&E analysis of primary tumor (P), lymphnodes (Ly) and lung (Lu) metastatic lesions, generated by orthotopic injection of D22 TPCs engineered with the indicated mutations. Scale bars, 100 µm. (right panel) Immunohistochemical analysis of TIMP1, MMP9, CD44, β-catenin, TWIST and SNAIL on primary and metastasis mouse avatars. Ctrl= negative control. Scale bars, 100 µm. B Immunohistochemical analysis of TIMP1, MMP9, CD44, β-catenin, TWIST and SNAIL on primary and metastatic PTC (ID#61) and ATC (ID#96) patient-derived tumors harboring BRAF/TP53 mutations. Ctrl= negative control. Scale bars, 100 µm. For (A and B) n = 5 tissues analyzed. c Heatmap of Wnt-, EMT- stemness and metastasis-related genes (2−ΔCt expression values) in primary (P), metastatic (M) tumor xenografts and primary (P) and metastatic (M) PTC- and ATC-patient derived tumors, harboring BRAF/TP53 mutations. Data are presented as normalized mRNA expression values of three independent experiments. D Venn diagram showing common upregulated genes with logFC > 3.5 in metastasis (M) versus primary (P) tumor xenografts, in PTC-derived metastasis (PTC M) versus primary tumors (PTC P) and in ATC primary versus PTC primary tumors. E and F, (left panels) Immunohistochemical analysis of KISS1 (E) and KISS1R (F) on primary and metastatic PTC (ID#61) and ATC (ID#96) patient-derived tumors harboring BRAF/TP53 mutations. n = 5 tissues analyzed. Scale bars, 100 µm. (right panels) R2 database analysis of KISS1 (E) and KISS1R (F) mRNA expression levels in normal thyroid (nt, n = 6), PTC (n = 6), FTC (n = 6) and ATC (n = 10) patient-derived tissues (R2 database Tumor Thyroid Carcinoma all - Huettelmaier − 28 - tmm - jbseqrnanb1). Boxes represent the IQR and midline represents the median. Statistical significance was calculated using ANOVA test. G Immunohistochemistry score analysis of KISS1R expression in normal thyroid tissue (n = 56), primary PTC tumors (PTC P, n = 73), PTC-derived loco-regional lymphnode metastasis (PTC M, n = 19) and primary ATC tumors (n = 5). Mutational status of 97 primary and metastatic human tissues is reported in Supplementary Data 2. Boxes represent the IQR and midline represents the median. Statistical significance was calculated using Kruskal-Wallis test.

Full size image

From the transcriptomic analysis emerged a high expression of KISS1 and KISS1R in TC metastatic lesions and in xenografts derived from TPCs harboring the mutation of BRAF or NRAS in combination with TP53, likely induced by the functional inactivation of p53, along with the activation of Wnt signaling and EMT-related pathways (Fig. 5C, D and Supplementary Figs. 8a–d). Although KISS1 is known as a metastasis suppressor gene, recent evidence highlighted its role in triggering the molecular events of the metastatic cascade35,36,37,38. Consistently, by binding its cognate receptor, KISS1 promotes invasion, intravasation and secondary lesion engraftments37,39.

Immunohistochemistry and transcriptomic analysis of a cohort of TC patients (Supplementary Table 2 and Supplementary Fig. 8e) revealed that KISS1 and KISS1R expression levels are positively associated with the aggressive phenotype including PTC-derived metastatic lesions and ATCs, in line with a RNAseq and DFS probability analysis of a large cohort of thyroid carcinoma, showing a significant correlation between high levels of KISS1R and the risk of develo** TC metastatic disease (Fig. 5E–G and Supplementary Figs. 8f–i).

KISS1R targeting reduced both proliferative and invasive activity and also inhibited the activation of the PI3K/AKT and ERK pathways (Supplementary Figs. 9a–f). Indeed, several clinical trials have been designed to target concomitantly PI3K/AKT and MAPK pathway (NCT01830504). Although surgical resection in combination with radioactive iodine and l-thyroxine treatment is known to be effective in differentiated TCs3, patients who developed distant metastases or affected by ATCs are refractory to radioactive iodine7,8. Indeed, ATC and metastatic disease are still incurable due to their systemic nature and resistance to the currently accessible therapeutic approaches. In accordance with targeting of molecules regulating NIS trafficking, tyrosine kinases or HDAC and overexpression of PAX8 gene40,41,42,43, we intriguingly observed that inhibition of KISS1R together with TIMP1 induced the expression of thyroid differentiation markers such as PAX8, TG, TSHR, TPO, TTF1 and NIS, which sustained the increase of functional radioiodine uptake in BRAF/TP53 and NRAS/TP53 mutant D22 TPCs, whose 125Iodine retention levels were similar to those of untreated PTC established cell lines (Fig. 6A–G and Supplementary Fig. 9g). In xenograft thyroid tumors derived from the subcutaneous injection of mutated D22 TPCs, the intra-tumoral administration of TIMP1 and KISS1R inhibitors44,45,46 lead to a significant delay in tumor growth, characterized by a marked expression of NIS, and a prolonged percentage of mice survival (Fig. 6H, I).

Fig. 6: KISS1R and TIMP1 targeting restore the functional iodine uptake by increasing NIS expression.
figure 6

A Relative mRNA expression levels of PAX8, TG, TSHR, TPO and TTF1 in the indicated engineered D22 treated with KISS1R inhibitor (KISS1Ri) alone or in combination with TIMP1 inhibitor (TIMP1i) for 24 h. Data are presented as fold change over vehicle ± SD of three independent experiments. B and c Immunofluorescence analysis of NIS and cell positivity in D22 TPCs engineered with the indicated mutations and treated as in (A) for 72 h. Nuclei were counterstained by Toto-3. Statistical significance was calculated using the unpaired two-tailed t test. Scale bars, 20 µm. Data are mean ± SD of 3 independent experiments. n = 12 wells analyzed. D Relative mRNA expression levels of NIS in cells engineered and treated as in (A) for 24 h. Data are presented as fold change over vehicle ± SD of three independent experiments. E and F Immunoblot of NIS (E) and relative optical density ratio (F) in cells as in (A) for 48 h. β-actin was used as loading control. One representative of three independent experiments is shown. Source data are provided as a Source Data file. G Radioiodine uptake in the indicated established thyroid cancer cell lines and engineered D22 TPCs treated as in (A) for 48 h. For (F and G) statistical significance was calculated using the two-tailed unpaired t test and data are mean ± standard error of three independent experiments. H (left panel) Growth kinetics of xenograft tumors generated by subcutaneous injection of D22 TPCs engineered for NRASQ61R/TP53R248Q treated with TIMP1 inhibitor (TIMP1i) and KISS1R inhibitor (KISS1Ri) alone and in combination. Statistical significance was calculated using the unpaired two-tailed t test. Data are shown as mean ± SD. n = 6 mice per group. (right panels) Immunohistochemical analysis of NIS on xenograft tumors obtained following injection of D22 TPCs engineered for NRASQ61R/TP53R248Q treated as indicated, left panel. Scale bars, 100 µm. I Kaplan-Meier graph showing the murine survival of D22 TPCs engineered with NRASQ61R/TP53R248Q treated as in (H). The statistical significance between groups was evaluated using a log rank Mantel-Cox test.

Full size image

Thus, these data indicate KISS1R and TIMP1 blockade as adjuvant therapy to sensitize undifferentiated TCs to radioiodine-based therapeutic regimen, reprogramming thyroid cancer cells behaving aggressively toward a more differentiated phenotype.

Our findings suggest that TC histopathology and behavior are determined by distinct genetic alterations and that while the ternary complex is crucial for the promotion of TC, KISS1R sustains the metastatic growth (Fig. 7A). Triggering of KISS1R leads to the acquisition of an invasive phenotype through the activation of both PI3K and ERK-mediated MMP9 signaling pathway47, which positively regulated the expression of EMT-associated genes (Fig. 7B and Supplementary Fig. 9h).

Fig. 7: Model of TC tumorigenesis and progression.
figure 7

A Schematic model illustrating that the most common TC genetic alterations (BRAFV600E, NRASQ61R, BRAFV600E/TP53R248Q and NRASQ61R/TP53R248Q) in thyroid progenitor cell (TPC) recapitulate the different TC histotypes (FTC, PTC and ATC). Of note TP53R248Q alone is not required for TC initiation. B Model of ternary complex (TIMP1/MMP9/CD44) and KISS1R driven pathways in engineered D22 TPCs. TIMP1 and pro-MMP9 complex formation activates MMP9 and consequently leads to the cleavage of CD44. The CD44 intracytoplasmic domain (CD44icd) translocates into the nucleus where it induces CD44v6 transcription. CD44v6 promotes TPC proliferation through PI3K/AKT pathway. The binding of kisspeptins (KP) to KISS1R activates ERK and cooperates with MMP9 to promote the transcription of EMT-related genes, including TWIST and SNAIL, driving metastatic engraftment.

Full size image

Discussion

In this study, we reproduced the oncogenic events that occur in tissue-resident stem cells/progenitor cells during TC transformation. We demonstrated that genetic mutations occurring in tissue-resident progenitor cells, characterized by a long-lasting cell lifespan, govern the evolutionary dynamics of thyroid tumorigenesis. Thus, this phenomenon could explain the TC incidence rates dictated by the rare (1/1000) presence of stem/progenitor cells in the thyroid gland48, which is renewed few times during the whole life49. We recapitulated all the different TC histotypes based on the induction of specific genomic alteration delivered by CRISPR-Cas9 in human ESC-derived TPCs at day 22. Thus, this is a genetic edited model of tumor transformation derived from human ESCs-derived cells. Scientists have either generated in vitro a normal functional thyroid gland starting from murine ESCs16 or they have introduced oncogenic driver mutations in breast, colon, or pancreas organoids by using CRISPR/Cas9 technology to induce tumor transformation50,51,52. Nowadays, the most common genetic and epigenetic alterations reported in TC, lead to the constitutive activation of two main signaling pathways: i) the MAPK and ii) PI3K/AKT/mTOR signaling pathways, which promote uncontrolled cell division, cell migration and longevity53. The spectrum of altered genes, including point somatic mutations, which differ on TC histotypes, endorse tumor initiation and progression. BRAF is mutated in 35-40% of PTCs and in 25-35% of ATCs, while it is not mutated in FTCs. RAS point mutations are reported in 18-52% of FTCs and in 6-55% of ATCs54.

Accordingly, we show that BRAFV600E or NRASQ61R recapitulates PTC and FTC histotypes, respectively, at transcriptomic, morphological and phenotypic levels. The mutation of tumor suppressor gene TP53 rarely occurs in differentiated PTC and FTC while it is frequently reported in ATC (70%)54,https://hgserver1.amc.nl/cgi-bin/r2/main.cgi). Analysis of enriched pathways from the Reactome database was performed through EnrichR R library v3.0.

Total RNA from FFPE thyroid tumor tissue specimens was purified with RNeasy FFPE Kit (Qiagen). 200 ng of total RNA were retrotranscribed with the high-capacity c-DNA reverse Transcription kit (Applied Biosystem).

Droplet digital PCR (ddPCR) was performed using ddPCR supermix for probes no-dUTP (Bio-Rad), and KISS1R (FAM) and GAPDH (HEX) ddPCR GEX assays (Bio-Rad). Following droplets generation and RT-PCR amplification, samples were quantified using QX200 Digital Droplet Reader (Bio-Rad). The obtained data were analysed with QuantaSoft Software (Bio-Rad) and KISS1R expression levels were calculated using the KISS1R (copies/μl)/GAPDH (copies/μl) ratio. Total RNA-seq analysis was carried out on 28 samples, including engineered D22 TPCs, D22-derived xenograft tumor tissues and human tissues from primary tumors and metastatic lesions of PTC and ATC patients. Read alignment and gene expression quantification were performed using STAR and Stringtie, respectively. Differentially expressed genes (DEGs) analysis was conducted by R with edgeR library based on the following criteria: (1) adjusted P < 0.05 and (2) Log 2 FC ≥ 2. The top 100 DEGs were selected. Gene set enrichment analysis (GSEA) was performed by using R library msigdbr and querying the hallmark (H) category, which includes 50 pathways. Pathways statistically significant with p value <0.05 were selected. Genes involved in PI3K/AKT-, MAPK-, EMT-related pathways were extracted from REACTOME signatures from MSigDB. Heatmaps were plotted using R library ggplot2 and ComplexHeatmap.

Nonradioactive assay

Iodine uptake of D22 TPCs, D30 and BCPAP cells has been measured by using Nonradioactive Iodine Assay kit (#D05076.96 wells; Bertin Pharma, Cayman Chemical), according to the manufacturer’s specifications. D22 TPCs, D30 and B-CPAP cells were plated (500000 cells in 24-well plates) and treated with KISS1R inhibitor (10 nM; Tocris) and TIMP1 inhibitor (1 µg/ml;R&D) alone or in combination for 48 h. Then, cell supernatant was analysed using Nonradioactive Iodine Assay kit (#D05076.96 wells; Bertin Pharma, Cayman Chemical).

Radioactive assay

D22 TPCs, B-CPAP, TT2609co2 and Cal62 cells were plated (50,000 cells in 24-well plates) and treated with KISS1R inhibitor (10 nM; Tocris) and TIMP1 inhibitor (1 µg/ml;R&D) alone or in combination for 72 h. Then, cells were incubated with iodine-125 (125I) (0.05 mCi; PerkinElmer). After 1 h, cells were washed twice in HBSS (Sigma-Aldrich) and lysed in 100 ml 2% SDS. Radioiodine uptake was determined using PerkinElmer Wallac 1470 Wizard instrument. Iodine 125 uptake was calculated normalized the counts per minute on the relative protein concentration revealed using Protein assay dye reagent concentrate (BIO-RAD).

Statistical analysis

Kruskal-Wallis test and pairwise Wilcoxon tests were used to compare the expression levels of CTNNB1 in normal thyrocytes, papillary and anaplastic thyroid carcinomas from the GSE33630 series in Gene Expression Omnibus (GEO). The Wilcoxon test was used to compare the expression levels of TIMP1, MMP9 and CD44 in normal and tumorigenic samples from the The Cancer Genome Atlas (TCGA) thyroid carcinoma (THCA) branch. TNMplot database was used for the analysis of thyroid cancer RNAseq-based data, for TIMP1/MMP9/CD44 and KISS1R gene expression analysis, respectively, using Kruskal-Wallis test.

Survival analyses were performed on data obtained from Gene Expression Profiling Interactive Analysis (GEPIA http://gepia.cancer-pku.cn/), showing the hazards ratio based on Cox PH Model, and expressed as disease-free survival (DFS) curves.

Following Kolmogorov-Smirnov test to assess the samples distribution, statistical significance was estimated by unpaired two-tailed t test, or by two-tailed Mann–Whitney test.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.