¹⁷⁷Lu-labeled DOTA-EB-RGD as a candidate for thyroid cancer therapy

Thyroid carcinoma (TC) is categorized by the World Health Organization into BRAF-like, RAS-like, and non-BRAF/RAS-like tumors based on its histology, differentiation, and molecular signature (1). The standard treatment for aggressive TC consists of the complete removal of thyroid gland and radioactive iodine (RAI) treatment (2). However, tumors become refractory to standard therapy in approximately 22% of patients (3, 4). Several studies have suggested that the survival rate of non–RAI-responsive TC patients is less than 3.5 years (5, 6). The use of tyrosine kinase inhibitors (TKIs), such as lenvatinib for treatment of advanced RAI refractory TC, is driven by its growth inhibitory effects elicited via multiple targets such as fibroblast growth factor receptors 1–4 (FGFR-1–4), vascular endothelial growth factor receptors 1–3 (VEGFR1–3), c-kit, RET, and platelet-derived growth factor receptor a (PDGFRa). However, therapy with TKIs leads only to transient responses (7, 8). Therefore, it is imperative to devise new therapeutic strategies to overcome these challenges.

Integrins are transmembrane receptors that can identify and bind to extracellular matrix proteins. The binding of integrins to their respective ligands leads to several structural, as well as signaling, changes in cells, including alterations in gene expression. Moreover, different levels of crosstalk exist between integrin-mediated signaling and signaling downstream of growth factors (9). A subtype of integrins (α_Vβ₃) recognizes Arg-Gly-Asp (RGD) peptide motifs, which have a role in neovascularization and development of different cancers (10).

Radiolabeled RGD peptides are utilized as radiotracers for targeted tumor imaging and therapy. In fact, there are preliminary data revealing that the radiolabeled α_Vβ₃ agonist ⁶⁸Ga-DOTA-RGD₂ could be successfully used for positron emission tomography/computed tomography (PET/CT) imaging in patients with radioiodine-refractory TC. These findings suggest that targeting α_Vβ₃ with RGD-based ligands radiolabeled with β emitters such as lutetium-177 (¹⁷⁷Lu) may have therapeutic potential in TC. However, given the relatively short half-life of DOTA-RGD (11, 12), we believe that the therapeutic efficacy of ¹⁷⁷Lu-labeled RGD peptides can be improved by conjugating the Evans Blue (EB) moiety to the cyclic form of RGD peptide that extends its half-life and therapeutic efficacy (13). Therefore, the goal of this study was to assess the expression of α_Vβ₃ in different histological and molecular subtypes of TC and to test the therapeutic efficacy of the radiolabeled integrin α_Vβ₃ targeting agent, ¹⁷⁷Lu-DOTA-EB-cRGDfK, in the therapy of TC in mouse xenografts.

Expression of α_Vβ₃ integrins in BRAF-like and RAS-like tumors. We utilized the publicly available database, The Cancer Genome Atlas (TCGA), consisting of mRNA expression profiles of 496 TC tissues to evaluate the association between BRAF-RAS scoring (BRS) and the expression of integrins α_V and β₃. We observed that there was a negative correlation between mRNA expression of α_V and BRS (r = –0.5, P < 0.001) (Figure 1A). In addition, we found a low negative correlation between mRNA expression of β₃ and BRS (r = –0.27, P < 0.001) (Figure 1B), suggesting that expression of α_V and β₃ integrins might be higher in BRAF-like tumors. Consistently with this result, RNA-seq data in selected samples harboring the BRAFV600E mutation (n = 143) and HRAS/NRAS/KRAS pathogenic variants (n = 49) showed that the expression of α_V was significantly higher (P = 0.001) in BRAFV600E-driven TC as compared with RAS-driven tumors (Figure 1C), but there was no statistically significant difference in β3 expression between BRAF-like and RAS-like tumors (P = 0.07) (Figure 1D). To summarize, mRNA expression of α_V is higher in BRAF-like than RAS-driven tumors, but its dimer’s partner β₃ is not significantly associated with BRAF/RAS transcriptome profile. Therefore, we next focused on the analysis of α_Vβ₃ protein dimers in TC.

Figure 1

Expression of α_Vβ₃ integrins. (A) A moderate negative correlation (Spearman, 2-sided) between mRNA expression of α_V and BRS (r = –0.5, P < 0.001) was observed. (B) A weak negative correlation (Spearman, 2-sided) between mRNA expression of β₃ and BRS (r = –0.27, P < 0.001) (n = 496). (C) The Cancer Genome Atlas (TCGA) was utilized to evaluate the RNA-seq data in selected samples with BRAFV600E mutation (n = 143) and HRAS/NRAS/KRAS pathogenic variants (n = 49) and observed that α_V expression was significantly higher (1-tailed non-parametric t test, Mann-Whitney, P = 0.001) in BRAFV600E-driven TC in comparison to RAS-driven tumors. (D) Non-significant (1-tailed non-parametric t test, Mann-Whitney) difference in β₃ mRNA expression in BRAFV600E (n = 143) vs. RAS-mutant tumors (n = 49). (E) Representative immunostaining revealing a higher α_Vβ₃ dimers expression in papillary TC (PTC) compared with follicular TC (FTC) (P = 0.03), and normal thyroid (P < 0.001). Original magnification, ×100. Scale bars: 100 μm. (F) Quantitative analysis of the immunostaining of α_Vβ₃ dimers expression in different thyroid tissues (1-way ANOVA, Kruskal-Wallis). Normal thyroid (NT) n = 10, PTC n = 44, FTC n = 20, PDTC n = 6.

Papillary TC has higher α_Vβ₃ expression compared with follicular TC and normal thyroid. Human tissue microarrays including 70 TC (papillary thyroid carcinoma [PTC] = 44, follicular thyroid carcinoma [FTC] = 20, poorly differentiated thyroid carcinoma [PDTC] = 6) and 10 normal thyroid samples were employed for immunohistochemistry. Higher α_Vβ₃ expression was found in PTC compared with FTC (P = 0.03), and between PTC and normal thyroid (P < 0.001) (Figure 1, E and F). We next modeled the observed variable expression of α_Vβ₃ in vitro in TC cell lines.

Differential expression of α_Vβ₃ in TC cell lines. We analyzed the mRNA expression of 24 genes, including integrins α_Vβ₃ and thyroid-specific genes (thyrotropin receptor [TSHR], thyroglobulin [TG], thyroid peroxidase [TPO], thyroid hormone receptor α [THRA], and THRB) in 13 TC cell lines and the normal thyroid cell line (HTori-3) (Figure 2A). As expected (14), TC cell lines were characterized by a significantly reduced expression of thyroid-specific genes compared with normal thyrocytes. We found a differential mRNA expression of α_Vβ₃ integrin in TC cell lines (Figure 2A). While the mRNA level of α_V (ITGAV) expression was relatively comparable between the cell lines and normal thyroid, the highest β₃ (ITGB3) integrin expression was observed in TC cell lines driven by the BRAFV600E oncogenic mutation (K1, OCUT2, SW1736), MKRN1-BRAF fusion (THJ16T), or CCDC6-RET fusion (TPC1).

Figure 2

Comparison of α_Vβ₃ expression. (A) Relative mRNA expression analysis of thyroid-specific genes, thyroid hormone receptor genes, and integrins in 13 TC cell lines in comparison to a normal thyrocyte cell line, HTori-3 (n = 3). TC cell lines are characterized by variable expression of ITGAV (α_V) and ITGB3 (β₃). (B) Immunoblot depicting the expression of α_Vβ₃ dimers in different TC cell lines. The highest expression was found in TPC1 (CCDC6-RET fusion), followed by oncocytic TC cell line XTC1, OCUT2 (BRAF-like TC), and FTC133 (RAS-like TC). C643 (with HRAS p.G13R mutation) and THJ16T (with MKRN1-BRAF fusion) cell lines had very minimal expression of α_Vβ₃ protein. (C) Quantification of the Western blot demonstrating α_Vβ₃ dimer protein expression (n = 3). (D) Low non-significant correlation (Pearson, 2-tailed) between α_Vβ₃ protein expression normalized to GAPDH (x-axis) and BRAF/RAS transcriptome profile quantified as BRAF-RAS score (BRS, y-axis): r = –0.34 (n = 3).

Relative mRNA expression patterns were confirmed at the protein level by immunoblotting and cell surface immunofluorescent staining, and 6 TC cell lines with variable α_Vβ₃ expression were selected for further analyses. There was no statistically significant correlation between α_Vβ₃ dimers and BRAF/RAS status of TC cell lines (r = –0.34, P = 0.51, Figure 2, B–D), despite a moderate-to-high correlation of individual monomers α_V (r = –0.7, P = 0.12, Supplemental Figure 1, A–C; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.200594DS1), and β₃ (r = –0.61, P = 0.19, Supplemental Figure 1, D–F).

Taken together, these data suggest no association between α_Vβ₃ dimer expression and BRAF-like or RAS-like transcriptome profile. However, we have established a preclinical model of differential α_Vβ₃ expression in human TC.

Cell surface expression of α_Vβ₃ integrin receptors. To analyze cell surface expression of α_Vβ₃, we performed flow cytometry on cell lines characterized by high (OCUT2) and low (FTC133) overall α_Vβ₃ expression. The dot plots (Figure 3A) show a higher percentage of α_Vβ₃-positive cells in OCUT2 (99.7%) than in FTC133 (63.2%), as is corroborated by the staggered overlay of the univariate histogram (Figure 3B) for the detector channel. Moreover, comparison of the mean fluorescence intensity (MFI) showed that OCUT2 is characterized by a 2.6-fold higher (P < 0.001) expression of α_Vβ₃ on the cell surface as compared with FTC133 (Figure 3C). When ITGB3 was knocked out in OCUT2 (OCUT-β3KO), the cell surface expression of α_Vβ₃ dimer was not detected (Figure 3D). The corresponding scatter dot plot for the MFI supported the observation, showing a 10-fold decrease to a near-negligible level of α_Vβ₃ in OCUT-β3KO cells (P < 0.0001). On the other hand, with adenoviral overexpression of ITGAV and ITGB3 in the FTC133 cell line (FTC133-α_Vβ₃⁺, Supplemental Figure 2), the α_Vβ₃ cell surface expression increased, as shown in the staggered histogram (Figure 3E) and in the adjoining dot plot, with a 4-fold increase compared with parental FTC133 (P < 0.0001).

Figure 3

Evaluation of α_Vβ₃ integrin receptor on TC cell surface. (A) Representative flow cytometry dot plots (10,000 events acquired; N = 3, n = 3) showing cell-surface expression of integrin α_Vβ₃ in OCUT2 (99% positive, upper panel) and FTC133 (63% positive, lower panel) cells. Red gates indicate α_Vβ₃-positive cells detected by AF633-conjugated secondary antibody fluorescence (APC channel, x-axis) versus side scatter (y-axis). Plots are shown for DAPI, secondary antibody control, and anti-α_Vβ₃–stained cells. (B) Overlay histograms of α_Vβ₃ staining show a greater rightward fluorescence shift in OCUT2 cells compared with FTC133 cells, indicating higher α_Vβ₃ expression. (C) The mean fluorescence intensity (MFI) analysis revealed higher α_Vβ₃ cell surface expression in OCUT2 than in FTC133. The results show an average of N = 4 biological replicate experiments, each done in triplicate (n = 3) and 2-tailed Welch’s t test was performed. (D) The staggered overlay for the univariate histograms of α_Vβ₃ staining for OCUT2-β3KO (blue) over OCUT2 parental cells with secondary (green) and anti-α_Vβ₃ (red) antibodies shows no expression of α_Vβ₃ receptor on the knockout cell surface. The adjacent dot plot quantitatively reveals the significant downregulation in MFI (N = 4, n = 2 or 3, 2-tailed Welch’s t test). (E) Adenoviral overexpression of α_Vβ₃ on FTC133 (red) resulted in a substantial shift of the staggered histogram compared with parental FTC133 (blue) and parental cells with secondary antibody (green). The corresponding MFI analysis showed a significant increase in α_Vβ₃ cell surface expression in the overexpressing cells compared with the parental cells (N = 4, n = 3, 2-tailed Welch’s t test). ****P < 0.0001.

Saturation radioligand binding to α_Vβ₃ integrin receptors. Consistently, radioligand binding assays to the α_Vβ₃ receptors, showed a significantly higher maximum binding sites (B_max) in OCUT2 cells as compared with FTC133 cells (mean ± SEM of 422.5 ± 49.4 and 88.1 ±13.9 fmol, respectively, P = 0.016; Figure 4, A and B), with the former having a significantly higher binding affinity (K_d) value (OCUT2 344.1 ± 57.6 nM vs. FTC133 97.4 ± 10.0 nM, P = 0.047). Accordingly, we have identified TC cell line models with high and low α_Vβ₃ cell surface expression for testing in vivo. In order to identify the specificity of the RGD peptide to the cell surface dimer α_Vβ₃ receptor, saturation binding of ⁶⁸Ga-DOTA-RGD₂ in OCUT-β3KO cells shows (Figure 4A) near-negligible affinity for the ligand (B_max 11.3 ± 1.7 fmol) compared with the parental OCUT2 (extra-sum-of-squares F test, F(2,176) = 907.8, P < 0.0001). Saturation binding curves differed significantly between FTC133 and FTC133-α_Vβ₃⁺ cells (extra-sum-of-squares F test, F(2,116)=206.6, P < 0.0001). Nonlinear regression revealed a marked increase in B_max in FTC133-α_Vβ₃⁺ cells (206.6 ± 24.5 vs. 88.1 ±13.9 fmol/mg, P < 0.0001), whereas K_d values were similar (107.1 ± 11.8 vs. 97.4 ± 10 nM, P = 0.58), as shown in Figure 4B.

Figure 4

⁶⁸Ga-DOTA-RGD₂ saturation binding assay. The nonlinear regression curves showed saturation binding of the radioligand ⁶⁸Ga-DOTA-RGD₂ on (A) OCUT2 and OCUT-β3KO. OCUT-β3 KO display negligible binding compared with parental cells (mean ± SEM 11.3 ± 1.7, extra-sum-of-squares F test, F(2,176) = 907.8, P < 0.0001, N = 3, n = 3) (B) FTC133 and FTC133 α_Vβ₃-overexpressing cells (named FTC133-α_Vβ₃⁺). FTC133-α_Vβ₃⁺ have markedly high binding compared with parental FTC133 (mean ± SEM 206.6 ± 24.5, extra-sum-of-squares F test, F(2,116) = 206.6, P < 0.0001, N = 2, n = 3).

Taken together, these data confirm the specificity of ⁶⁸Ga-DOTA-RGD₂ binding to α_Vβ₃ integrin.

¹⁷⁷Lu-DOTA-EB-RGD reduces tumor growth in TC tumor models with high α_Vβ₃ integrin expression. We compared the therapeutic potential of radiolabeled RGD peptides with lenvatinib, the existing standard of care, in vivo using our established mouse models (15, 16). We selected subcutaneous and/or metastatic models with variable α_Vβ₃ expression and binding capacity; OCUT2 served as a model with high expression and high binding capacity and FTC133 as a model with low α_Vβ₃ expression and low B_max. Mixed-effects model-based analysis of tumor growth velocity revealed that treatment with ¹⁷⁷Lu-DOTA-EB-RGD (¹⁷⁷Lu) led to a significant reduction in the tumor growth over time as compared with placebo (P < 0.001) and lenvatinib alone (P < 0.001, Figure 5A). The combination therapy (¹⁷⁷Lu+L) resulted in decreased tumor growth compared with monotherapy with lenvatinib (P < 0.001) but had a similar effect to ¹⁷⁷Lu monotherapy (P = 0.79) (Figure 5, A and C). In addition, PET scanning revealed that the uptake of ⁶⁸Ga-DOTA-RGD₂ was significantly reduced due to a smaller tumor volume in mice exposed to combination therapy (¹⁷⁷Lu+L, maximum standard uptake value [SUV_max] 0.75 ± 0.11) as compared with treatment with placebo (SUV_max 1.12 ± 0.11; P = 0.001) as well as lenvatinib (0.98 ± 0.21; P = 0.045) (Figure 5B).

Figure 5

¹⁷⁷Lu-DOTA-EB-RGD treatment in subcutaneous TC tumor models. (A) Tumor growth velocity analysis by 2-way mixed-effects model ANOVA revealed that treatment with ¹⁷⁷Lu-DOTA-EB-RGD (¹⁷⁷Lu) led to significant tumor growth reduction over time compared with placebo (P < 0.001) and lenvatinib alone (P < 0.001). Combination therapy (¹⁷⁷Lu+L) decreased tumor growth compared with lenvatinib monotherapy (P < 0.001), although having similar effect as ¹⁷⁷Lu monotherapy (P = 0.79). (B) Representative PET scan images depicting uptake of ⁶⁸Ga-DOTA-RGD₂ in tumors showing significantly lower uptake in mice exposed to combination therapy (¹⁷⁷Lu+L, SUV_max 0.75 ± 0.11) in comparison with the placebo group (SUV_max 1.12 ± 0.11; P = 0.001) as well as lenvatinib (0.98 ± 0.21; P = 0.045) (1-way ANOVA with Holm-Šídák multiple-comparison test). (C) Resected tumors collected from each group show that treatment with ¹⁷⁷Lu in monotherapy or combination therapy led to a significant reduction of tumor volume (median [IQR], 2-way mixed-effects model ANOVA) with respect to placebo [P] (¹⁷⁷Lu 0.350 cm³ [0.172–0.508] vs. P 0.928 cm³ [0.196–1.674], P = 0.0001; and ¹⁷⁷Lu+L 0.310 cm³ [0.214–0.335] vs. P 0.928 cm³ [0.196–1.674], P < 0.0001). The combination therapy (¹⁷⁷Lu+L, 0.337 cm³) resulted in decreased tumor volume as compared with monotherapy with lenvatinib (0.855 cm³, P = 0.05) but had a similar effect as compared with ¹⁷⁷Lu monotherapy (P = 0.99). (D) No treatment-associated difference in ⁶⁸Ga-DOTA-RGD distribution at the conclusion of the experiment in normal organs other than from higher variability in heart exposure in ¹⁷⁷Lu monotherapy as compared with placebo (2-way ANOVA). (E) Ex vivo biodistribution studies in ¹⁷⁷Lu monotherapy and ¹⁷⁷Lu +L revealed that the kidneys had a similar radiation exposure to the tumors, while exposure to the heart and liver was significantly lower (2-way ANOVA). The number of mice (n) studied for each group was: placebo, n = 8; lenvatinib, L, n = 7; ¹⁷⁷Lu, n = 8; ¹⁷⁷Lu+L, n = 6. **P < 0.01; ***P < 0.001; ****P < 0.0001.

There was no treatment-associated difference in ⁶⁸Ga-DOTA-RGD₂ distribution in normal organs apart from higher variability in heart exposure in ¹⁷⁷Lu monotherapy as compared with placebo (Figure 5D).

Ex vivo biodistribution studies in ¹⁷⁷Lu monotherapy and ¹⁷⁷Lu+L revealed that the kidneys had a similar radiation exposure to the tumors, while exposure to the heart (P < 0.0001) and liver (P < 0.001) was significantly lower (Figure 5E). These data suggest that kidneys might be the dose-limiting organ for bench-to-bedside translation.

RNA-seq analysis of resected tumors demonstrated that ¹⁷⁷Lu either in monotherapy or combination therapy induced a type I IFN response, characterized by upregulation of IFN-α–inducible protein 6 (IFI6) that was the only transcript passing the false discovery rate (FDR) cutoff of q less than 0.05 (log₂FC 3.5, P < 0.001, q = 0.01). Therefore, given the exploratory nature of the analysis and the limited sample size, genes with unadjusted P of less than 0.05 and |log₂FC| of 1 or greater were considered candidates for further validation. Consistently, other IFN response element genes were also significantly upregulated IFN-induced protein with tetratricopeptide repeats 1 (IFIT1), IFIT2, and MX dynamin-like GTPase 1 (MX1), compared with both placebo- and lenvatinib-treated tumors (Supplemental Figure 3, A, C, and D, Supplemental Figure 4, and Supporting Data Values; unadjusted P < 0.05, and |log₂FC| ≥ 1). Treatment with ¹⁷⁷Lu was also associated with reduced expression of the proangiogenic factor vascular endothelial growth factor β (VEGFB) (Supplemental Figure 3A and Supporting Data Values), whereas lenvatinib predominantly suppressed epithelial-mesenchymal transition–associated (EMT-associated) and invasion-associated genes, including fibronectin 1 (FN1), cluster of differentiation 44 (CD44), MET protooncogene, catenin β1 (CTNNB1), and matrix metallopeptidase 2 (MMP2) (Supplemental Figure 3B and Supporting Data Values). No significant changes in α_V and β₃ integrin mRNA expression were observed in mice treated with ¹⁷⁷Lu monotherapy, whereas lenvatinib monotherapy was associated with reduced expression of integrin-associated adhesion and invasion genes, including ITGAV, ITGB5, FN1, CD44, and MMP2 (Supplemental Figure 3B and Supporting Data Values). However, downstream focal adhesion signaling components, including focal adhesion kinase (FAK, encoded by PTK2), SRC protooncogene, paxillin (PXN), and vinculin (VCL), did not exhibit a significant differential expression across treatment groups. Consistently, the protein expression of α_Vβ₃ dimers was significantly reduced in lenvatinib (3.0 ± 0.82 vs. 8.86 ± 2.27, P = 0.001) and L+¹⁷⁷Lu combination therapy (3.5 ± 01.52 vs. 8.86 ± 2.27, P = 0.01), as compared with placebo (Figure 6). However, its decreased expression was also observed in ¹⁷⁷Lu monotherapy (3.33 ± 0.52 vs. 8.86 ± 2.27, P = 0.01), despite no detectable changes in integrin transcript levels. This finding suggests that ¹⁷⁷Lu treatment may have preferentially eliminated clones with high integrin expression. Furthermore, immunostaining revealed uniform FAK expression across treatment groups (Supplemental Figure 5, A and B), indicating that cell-matrix anchorage signaling was not differentially affected. Additional downstream targets of integrin signaling, pan-AKT, p-ERK1/2, and p-MEK, were also not significantly affected in any of the treatment groups. Similarly, no differential expression was found for tumor-suppressing PTEN in any of the studied groups (Supplemental Figure 5, A and B). Collectively, these findings suggest that the therapeutic activity of ¹⁷⁷Lu-DOTA-EB-RGD is not mediated by major transcriptional downregulation of α_Vβ₃ integrin.

Figure 6

Immunohistochemistry of tumor tissues. (A) Representative images of immunostaining for α_Vβ₃ integrin, proliferation marker PCNA, and survival marker survivin (placebo n = 7, lenvatinib n = 7, ¹⁷⁷Lu-DOTA-EB-RGD n = 6, ¹⁷⁷Lu-DOTA-EB-RGD + lenvatinib n = 6). Original magnification, ×200. Scale bars: 100 μm. (B) Quantitative analysis of immunostaining of different groups with respect to control showing that compared to the tumors treated with placebo, the expression of integrin α_Vβ₃ dimers (1-way ANOVA, Kruskal-Wallis) was significantly reduced in lenvatinib (P = 0.001), ¹⁷⁷Lu monotherapy group (P = 0.01), and ¹⁷⁷Lu+L combination therapy (P = 0.01). Proliferation marker PCNA (1-way ANOVA, Kruskal-Wallis) was significantly reduced in the tumors exposed to combination therapy with ¹⁷⁷Lu+L (P = 0.004) and monotherapy with lenvatinib with respect to placebo (P < 0.001). Also, in comparison with the tumors treated with placebo (Welch’s t test), survivin (cancer cells’ survival marker) was found to be significantly reduced in ¹⁷⁷Lu+L combination therapy (P = 0.03) and ¹⁷⁷Lu monotherapy (P = 0.006).

Immunostaining for proliferation and anti-apoptotic markers (Figure 6A) revealed that compared with placebo, proliferation marker PCNA was significantly reduced in the tumors subjected to combination therapy with ¹⁷⁷Lu+L (median score 2.5 [IQR 2–3.75] vs. 9 [IQR 6–12], P = 0.001) and monotherapy with lenvatinib (median score 2 [IQR 2–4] vs. 9 [IQR 6–12], P < 0.001), while there was a slight, statistically nonsignificant decrease in the proliferation marker with ¹⁷⁷Lu alone as compared with placebo (median score 6 [IQR 5.25–6] vs. 9 [IQR 6–12], P = 0.47). Compared with tumors exposed to placebo, the cancer cells’ survival marker survivin was significantly reduced in ¹⁷⁷Lu combination therapy (score 3.83 ± 2.4 vs. 7.71 ± 3.14, P = 0.03) and ¹⁷⁷Lu monotherapy (score 2.83 ± 1.94 vs. 7.71 ± 3.14, P = 0.01) (Figure 6B). However, expression of key apoptotic regulators BAX and BCL2 (Supplemental Figure 6, A and B) maintained steady, localized cytoplasmic positivity across all cohorts, indicating that neither mono nor combination treatments significantly altered baseline proapoptotic signals compared with placebo control.

In contrast, the luciferase-tagged FTC133 induced lung metastatic mouse model, characterized by a low α_vβ₃ expression, did not identify metastatic lesions with ⁶⁸Ga-DOTA-RGD₂ PET imaging despite their presence being documented by IVIS imaging (Supplemental Figure 7A), consistent with low to no α_Vβ₃ expression in metastatic lesions. This was associated with no reduction in tumor growth in response to ¹⁷⁷Lu, since no significant difference in tumor burden via bioluminescent counts from the IVIS imaging studies was observed across all treatment groups compared to placebo (Supplemental Figure 7, A and B).

Taken together, these in vivo data indicate that treatment with ¹⁷⁷Lu, either as monotherapy or in combination with lenvatinib, is effective in reducing the growth of TC characterized by high α_vβ₃ expression, and its effects are associated with eliciting a type I IFN signaling, increased prosurvival signaling, and reduced proliferation in vivo.

In this study, we demonstrated that integrin α_Vβ₃ is upregulated in a subset of TC, particularly in the most common histological type, PTC, and that ⁶⁸Ga-labeled RGD can be used for PET imaging of tumors characterized by a high α_Vβ₃ expression. While PTC is characterized by the highest α_Vβ₃ expression, there is no association between its expression and BRAF-like or RAS-like status of TC. We next demonstrated that the radiolabeled α_Vβ₃ analog ¹⁷⁷Lu-DOTA-EB-RGD targets α_Vβ₃ in TC, leading to tumor growth inhibition in models characterized by high α_Vβ₃ expression, but not in models characterized by low α_Vβ₃ expression. The therapeutic efficacy was associated with eliciting a type I IFN signaling, inhibition of proliferation and decreased TC cell survival, as well as with decreased α_Vβ₃ expression in treated tumors. Decreased proliferation and diminished prosurvival markers are consistent with the growth inhibitory potency of therapy with ¹⁷⁷Lu-DOTA-EB-RGD. Because no changes in α_Vβ₃ were observed at the transcriptional level, and no changes in activation of its downstream targets have been observed on a protein level, the lower α_Vβ₃ protein expression in treated tumors may reflect selective elimination of α_Vβ₃-high tumor cells, resulting in enrichment of residual cells with lower receptor expression. In addition, reduced tumor cellularity, stromal remodeling, may also contribute to the observed decrease in protein expression.

Our preliminary transcriptomic data showing induction of IFN-stimulated genes after ¹⁷⁷Lu-DOTA-EB-RGD are consistent with emerging evidence that ¹⁷⁷Lu-based radiopharmaceutical therapies can activate a STING-dependent type I IFN program in tumor cells associated with an increased intratumoral CD8⁺/Treg ratio (17). Furthermore, ¹⁷⁷Lu-labeled ligands targeting different protein receptors, including ¹⁷⁷Lu-iPD-L1, have been shown to remodel the tumor immune microenvironment, enhance immune effector functions, and augment antitumor immunity (18). However, activation of IFN-stimulated genes in our model as potential mechanism of therapeutic efficacy contributing to treatment-associated immune modulation needs to be tested in an immunocompetent model.

Integrins have been targeted with radiolabeled ligands in various metastatic cancers such as gliomas and non–small cell lung cancer (NSCLC) (19–23). There are several clinical trials successfully utilizing PET/CT imaging for evaluation of α_Vβ₃-positive cells in lymphomas, prostate, lung, breast, or head and neck cancers, as well as clinical trials focused on the selection of patients that may benefit from integrin targeting peptide receptor radionuclide therapy (PRRT) based on PET/CT imaging (e.g., ClinicalTrials.gov NCT05549024, NCT06638034, and NCT05013086) (24, 25). There are also preclinical data in breast, prostate, and pancreatic cancers utilizing α_Vβ₃ ligands radiolabeled with α emitters (²²⁵Ac) showing a significant reduction in tumor burden (26–28). However, the application of this management strategy has not yet reached the clinical trial stage. There are also efforts to leverage integrins’ role in tumor neovascularization and combine RGD-analog-based therapy with antiangiogenic agents (24).

Interestingly, patients with metastatic TC have also been enrolled in small pilot studies focused on screening for α_Vβ₃ integrins using PET/CT (e.g., ClinicalTrials.gov NCT05549024 and NCT06638034). Parihar et al. evaluated the diagnostic accuracy of ⁶⁸Ga-DOTA-RGD₂ PET/CT in patients with radioiodine-refractory TC (RAIR-TC), validating the utilization of this radiotracer in the human population (12). The authors observed a high specific uptake of radiolabeled RGDs in metastatic lesions, inferring an optimal target for ¹⁷⁷Lu-labeled RGDs for therapy. In a follow-up study, the same group presented a case report of a safe use of ¹⁷⁷Lu-DOTA-RGD₂ in a patient with metastatic RAIR-TC that resulted in improved pain control due to a reduction in size of one of the bone lesions located in the sternum at 4 months after therapy (29). However, the duration of such a response is unknown, as there are no follow-up data available. To optimize this therapeutic approach, patient selection may need to be guided by PET/CT imaging using ⁶⁸Ga- or ⁶⁴Cu-labeled α_Vβ₃ ligands to identify candidates for PRRT based on high tumor uptake. Given that our preclinical biodistribution studies identified the kidneys as the dose-limiting organ, PRRT may need to be accompanied by renoprotective amino acid infusions, similar to the standard practice used in PRRT targeting somatostatin receptors (30).

One of the limitations of RGD-based treatment is its limited stability, short half-life (24), and rapid renal clearance. Thus, it is crucial to improve its pharmacokinetics and pharmacodynamics. One strategy to overcome the drawback of low solution stability of RGD peptides is to cyclize the linear form of the RGD peptide (31, 32). Stability is enhanced by reducing conformational flexibility, leading to a lower susceptibility to proteolytic damage. Therefore, the cyclic form of RGD was used for therapy purposes, and RGD dimers for PET imaging. However, even the cyclic dimeric form has been characterized by a high K_d in our model. Therefore, in addition, in our study, cyclic RGD was chemically conjugated with an albumin-binding moiety-truncated Evans Blue (denoted as EB-RGD), leading to an increased half-life and elevated tumor uptake and tumor residence time (13, 33, 34). Zhao et al. utilized ¹⁷⁷Lu-conjugated DOTA-EB-RGD in a preclinical model using patient-derived xenografts from subjects with advanced NSCLC (13), showing superior theranostic properties compared with non–EB-bound RGD. Another preliminary clinical study found that utilizing ¹⁷⁷Lu-labeled dimeric RGD peptide to treat patients with TC resulted in improved biodistribution and prolonged retention in locoregional/metastatic lesions (35). Another avenue to extend the biological half-life of RGD analogs is to modify them using lipid particles. Addition of lipids like polyethylene glycol (PEG) into RGD peptides results in lower immunogenicity, as the peptide is protected from the rapid clearance by the immune system (36–38). In addition, there are efforts to embed RGD analogs into the liposomes. In order to decrease the risk of severe allergic response to liposomes, some studies have suggested that the introduction of an SGS (serine-glycine-serine) amino acid sequence in c(RGDyK)-modified liposomes may lead to reduced anaphylaxis events and improve the safety profile of drug delivery (39).

One of the strengths of our study is the utilization of the robust clinical data to evaluate the expression of α_Vβ₃ integrins in different types of TC. Moreover, our in vivo studies used new advances in RGD analog synthesis using stable cyclic analogs either in dimeric or monomeric forms and extending their biological stability by adding the EB moiety.

One of the limitations of our study is that the mechanisms of α_Vβ₃ expression, dimerization, and cell membrane trafficking in TC have not been investigated. However, we analyzed transcriptome and protein expression of the resected tumors and demonstrated that ¹⁷⁷Lu-DOTA-EB-RGD induces a type I interferon response program that needs to be validated in immunocompetent model. In conclusion, the radiolabeled α_Vβ₃ analog ¹⁷⁷Lu-DOTA-EB-RGD has potent growth inhibitory effects in TC characterized by a high α_Vβ₃ integrin expression that has been observed mostly in PTC. The α_Vβ₃ integrin could potentially serve as a molecular target for therapy with radiolabeled RGD analogs in a subset of TC that are characterized by a high ⁶⁸Ga-DOTA-RGD₂ uptake. Because kidneys appear to be dose-limiting organ, clinical translation should be preceded by pharmacokinetic studies that evaluate the addition of nephroprotective agents to PRRT.

Sex as a biological variable. Our study examined male and female mice, and similar findings are reported for both sexes.

Expression profiling of integrin α_Vβ₃. Integrin α_Vβ₃ mRNA and/or protein expression was evaluated in 496 TCs included in TCGA, commercially available Human TC Tissue Microarray (TMA) slides (US Biomax Inc.), including 70 TC and 10 normal thyroid samples, and 14 TC cell lines.

BRS. BRAF-like or RAS-like expression profiles were determined through standard BRS scores ranging from –1 to 0 for BRAF-like and 0 to 1 for RAS-like TC (40). The association between BRS score and α_Vβ₃ expression was tested by Spearman’s correlation (r).

Immunohistochemistry. Immunostaining was performed on commercially available human TC tissue microarray slides (US Biomax Inc.). Sections were dewaxed, soaked in alcohol (descending alcohol series 100%, 96%, 70%), washed in deionized water, and then microwave-treated in antigen unmasking solution (Vector Laboratories) for 10 minutes. Endogenous peroxidase activity was quenched by incubation in BLOXALL Blocking Solution (Vector Laboratories) for 10 minutes. Sections were then incubated for 20 minutes in 2.5% horse serum and then at 4°C overnight with anti–integrin α_Vβ₃ antibody (clone EM22703, ZooMAb Rabbit Monoclonal, Sigma-Aldrich; diluted 1:100).

Immunostaining was performed using ImmPRESS Horse Anti-Rabbit IgG PLUS Polymer Kit (Vector Laboratories) as per the manufacturer’s instructions. Peroxidase staining was revealed with the ImmPact DAB peroxidase substrate kit (Vector Laboratories). Sections were counterstained with hematoxylin and mounted. The antiserum was omitted in the negative control.

The results of staining were interpreted and scored independently by 2 investigators. The evaluation of immunoreactivity scores (IHS) was carried out using ×200 magnification. Staining intensity (SI) was assessed according to a categorical scale: 0, no staining; 1, faint staining; 2, slight staining; 3, moderate staining; and 4, strong staining. The percentage of positively stained cells (PP) was assessed as the following: 0, no positive cells; 1, 0%–25% positive cells; 2, 26%–50% positive cells; 3, 51%–75% positive cells; and 4, 76%–100% positive cells. An overall IHS score was calculated by multiplying SI by PP (range of possible scores 0–16). The method for immunostaining TC tumor tissue microarrays derived from mouse xenografts used in this study is described in the Supplemental Methods.

Cell culture. Fourteen TC cell lines and 1 follicular thyroid cell line (HTori-3) were used for our study. These cell lines were either gifted by Electron Kebebew (Stanford University) or purchased from the University of Colorado Cancer Center, Sigma-Aldrich, or ATCC: FTC (FTC133); PTC (K1, BCPAP, and TPC1); medullary thyroid cancer (TT); anaplastic thyroid cancer (THJ16T, THJ29T, 8505C, OCUT2, KAT18, SW1736, and C643); and oncocytic thyroid cancer (XTC1). The cell lines were grown in the following media, based on recommended optimal conditions — HTori-3, FTC133, and K1: DMEM/F-12 (Gibco); XTC1, 8505C, OCUT2, and OCUT2-β3KO: DMEM (Gibco); TT: F12K (Gibco); TPC1, THJ16T, THJ29T, KAT-18, SW1736, BCPAP, and C643: RPMI1640 (Gibco); per the manufacturers’ instructions. All media were supplemented with 10% fetal bovine serum (R&D Systems), 10 U/mL penicillin-streptomycin (Gibco), and 0.25 μg/mL amphotericin B (Gibco). Luciferase-tagged FTC133 cell lines, designated as FTC133-Luc, were also maintained in respective media supplemented with G-418 (Geneticin) to preserve antibiotic selection of the luciferase lentivirus. The cells were authenticated by short tandem repeat analysis, and contamination was excluded via IMPACT II PCR analysis (Supporting Data Values).

RT-PCR array. For gene expression analysis, we used customized RT² Profiler PCR Array plates manufactured by QIAGEN. The plate contained primers for 4 thyroid-specific genes: thyroid-stimulating hormone receptor (TSHR), sodium-iodine symporter (NIS), thyroglobulin (TG), thyroid peroxidase (TPO); 2 thyroid hormone receptor genes: thyroid hormone receptor α (THRA) and THRB; 18 integrin genes: integrin β1–β7 (ITGB1–ITGB7), integrin α1–α6 (ITGA1–ITGA6), integrin α_V (ITGAV), integrin α2B (ITGA2B), integrin αL (ITGAL), integrin αM (ITGAM), integrin αX (ITGAX); 5 housekeeping genes: β-actin (ACTB), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), β-2-microglobulin (B2M), hypoxanthine phosphoribosyltransferase 1 (HPRT1), ribosomal protein lateral stalk subunit P0 (RPLP0); 1 human genomic DNA contamination control (HGDC); 1 positive PCR control (PPC); and 1 reverse transcriptase control (RTC). The gene expression within the cDNA samples was analyzed using iQ SYBR Green Supermix (Bio-Rad) on a StepOnePlus Real-Time PCR Detection System (Applied Biosystems). The relative gene expression was normalized to the housekeeping genes and calculated using the 2^–ΔΔCt method.

Immunoblotting. TC cells were lysed in M-PER (Mammalian Protein Extraction Reagent, Thermo Fisher Scientific) according to the manufacturer’s instructions. The protein concentration was calculated through the Bradford assay (DC protein assay Kit II, 5000112, Bio-Rad). The protein samples were loaded on 4%–12% SDS-PAGE gels (Novex; 40 μg/lane) and transferred to polyvinylidene difluoride (PVDF) membranes through a Trans-blot Turbo system (from Bio-Rad). The membranes were then blocked with 3% BSA diluted in PBST at room temperature for 1 hour. Next, the membranes were incubated at 4°C overnight with primary antibodies against α_Vβ₃ dimers (1:1000; Novus Biologics, SC56-07), α_V monomer (1:1000; Abcam, ab179475), β₃ monomer (1:1000; Abcam, ab119992), or GAPDH (1:1,000; Invitrogen, MA5-15738) as housekeeping protein. The membranes were washed thrice with PBST and incubated further with either an anti-rabbit (1:1,000; sc-2357) or anti-mouse (1:1,000; sc-516102) secondary antibody (Santa Cruz Biotechnology) at room temperature for 1 hour. The membranes were then washed with PBST thrice and visualized through an Imaging System (Bio-Rad) using ECL (obtained from Azure Biosystems).

Generation of integrin β3–knockout cell lines (ITGB3^–/^–) using CRISPR/Cas9 genome editing. Guide RNAs (gRNA) targeting exon 3 (3′-GCGAGGUGAGCCCAGAGGCA-5′, 3′-UCACAGCGAGGUGAGCCCAG-5′, and 3′-UUCUCCUUCAGGUCACAGCG-5′) of the integrin β3 gene (ITGB3) were designed and synthesized by Synthego. Cas9 ribonucleoprotein (RNP) was synthesized by the UC Berkeley QB3 MacroLab. The Lipofectamine CRISPRMAX Cas9 Transfection Reagent (Thermo Fisher Scientific) was used to transfect the Cas9/gRNA complex into cells, following the manufacturer’s protocol. A 2:1 molar ratio of RNA/Cas9 was determined to be optimal. Cells were seeded 24 hours prior and transfected at 50% confluence in 24-well plates. Single cells were isolated using limiting dilutions and propagated to obtain single-cell–derived colonies. Mutations in the targeted region were determined by PCR amplification and capillary DNA sequencing. Knockout of the ITGB3 gene was confirmed at the protein level by flow cytometry.

For capillary DNA sequencing, DNA was isolated from cells using the QIAGEN DNeasy Blood and Tissue Kit, and the target region amplified by PCR with the following primer pairs: 5′-GCTGGTCTACCAGGCATCTT-3′ and 3′-CCATTCTCTTGCCCACACCT-5′; and 5′-TCAGCTTCGGTTCCAAGGAC-3′ and 3′TGGATGACGGTGTCCAGTTG-5′, using the PCR Master Mix (Thermo Fisher Scientific) following the manufacturer’s instructions. The PCR product was cleaned using the Zymo Research DNA Clean & Concentrator kit and sequenced by capillary DNA sequencing. Raw sequences were analyzed using Chromas DNA sequence analysis software (Technelysium, https://technelysium.com.au/) version 2.6.6. The alignment programs EMBOSS WATER Pairwise Sequence Alignment (EMBOSS WATER < Job Dispatcher < EMBL-EBI) and Clustal Omega Multiple Sequence Alignment (Clustal Omega < Job Dispatcher < EMBL-EBI) were employed for pairwise and multiple sequence alignments respectively.

Adenovirus for hITGAV and hITGB3 and α_Vβ₃ overexpression. Adenoviruses expressing human ITGAV (NCBI: NM_002210.5) and ITGB3 (NCBI: NM_000212.3), genes corresponding to α_V and β₃ proteins, were custom-designed (Supplemental Figure 2) and procured from VectorBuilder Inc. After titration to determine transfection efficiency, 10 MOI/cell for each adenovirus was used throughout the study. Cells (2 × 10⁵) were seeded overnight and treated with desired concentration of adenovirus for 48 hours in fresh media, then harvested to isolate total RNA according to the RNeasy Mini Kit (QIAGEN, 74106). cDNA was prepared using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4368813), and real-time qPCR was performed to determine overexpressed genes using the TaqMan Gene Expression Assay for hITGAV (Hs00233808_m1) and hITGB3 (Hs01001469_m1). After confirming satisfactory overexpression efficiency and time point, α_Vβ₃ dimer protein levels were analyzed.

Cell surface staining of α_Vβ₃. FTC133 and OCUT2 cells were analyzed for live-cell surface expression of α_Vβ₃ using flow cytometry. Optimally maintained cells were detached with Accutase cell detachment solution (Innovative Cell Technologies, AT-104), washed with PBS followed by FACS buffer (PBS, 5% FBS, 0.1% NaN₃), and 2 × 10⁵ cells were used in each sample for staining. Rabbit anti-α_Vβ₃ monoclonal antibody (1:250; Novus Biologicals, NBP2-67557) was added to designated samples and incubated for 90 minutes. Alexa Fluor 633–conjugated (AF633-conjugated) F(ab′)₂-Goat anti-Rabbit IgG (H+L) (Thermo Fisher Scientific, A-21072) secondary antibody was added at a dilution of 1:1000 after the primary antibody incubation was complete. Finally, DAPI was added as a viability marker, and flow cytometric analysis was performed with 10,000 events counted for each sample. The blank control contained only secondary antibody and DAPI. The flow cytometry results were analyzed using FlowJo v10 software (https://www.flowjo.com/flowjo10/overview), and the built-in geometric mean analysis was employed to calculate the MFI. MFI measures the fluorescence intensity of a specific cell subpopulation, normalized to cell number, to quantify that subpopulation’s contribution to the overall marker expression in the entire cell population (41).

For the comparison of the cell surface expression of α_Vβ₃ receptor in parental OCUT2 and CRISPR/Cas9 ITGB3-knockout OCUT2 (named OCUT2-β3KO) cell lines, cells were prepared as described above and stained with mouse anti–integrin α_Vβ₃ monoclonal antibody (1:200; Abcam, ab7166), followed by AF647-conjugated Goat anti-Mouse IgG (H+L) Cross-Adsorbed secondary antibody (Thermo Fisher Scientific, A-21235) added at a dilution of 1:1000. For the parental FTC133 cells and FTC133 cells with adenoviruses overexpressing ITGAV and ITGB3 (named FTC133-α_Vβ₃⁺), cell surface α_Vβ₃ dimer staining was done in the same way after adenoviral transfection was completed.

Synthesis of radiolabeled peptides. The radiolabeled peptides ⁶⁸Ga-DOTA-c(RGDfK)₂, a cyclic RGD dimer labeled with ⁶⁸Ga via DOTA as a linker (named ⁶⁸Ga-DOTA-RGD₂), and ¹⁷⁷Lu-DOTA-EB-cRGDfK, a cyclic RGD labeled with ¹⁷⁷Lu via DOTA as a linker and attached to Evans Blue (EB) moiety (alias ¹⁷⁷Lu-DOTA-EB-RGD), were synthesized following a previously described method (13, 15). Briefly, 500 μL of ⁶⁸Ga elution fraction (obtained from the iThemba, South Africa, generator) in 0.6 M HCl was added to a solution of 100 μL 5 M NaOH and 500 μL 1 M HEPES (pH 7.2). This solution was then added to 20 μg (1 mg/mL in water) of in-house synthesized DOTA-c(RGDfK)₂. ⁶⁸Ga-DOTA-c(RGDfK)₂ was obtained by incubating the reaction mixture at 85°C for 10 minutes. The reaction was diluted into 5 mL of water and passed through BondElute (C-18 100 mg, Agilent). The cartridge was washed with 10 mL of water, and the product was eluted twice with 150 μL 80% ethanol containing 1 mM HCl. The column-purified radiotracer was diluted in PBS for administration in mice. Radiochemical purity was greater than 95 %, as assayed by radio HPLC.

The 30 μL stock solution of 1.94 GBq ¹⁷⁷LuCl₃ was diluted with 0.25 mL 0.4 M sodium acetate (pH 5) and mixed with 50 μg (1 mg/mL) in-house synthesized DOTA-EB-cRGDfK (33). After a 20-minute incubation at 95°C on an orbital shaker, the ¹⁷⁷Lu-DOTA-EB-cRGDfK reaction solution was assayed by rapid TLC on iTLC (silica on fiber glass) with 0.1 M citric acid, pH 5. The iTLC plate was assayed with a Bioscan TLC plate reader. Radiochemical purity was greater than 90%, and the reaction solution was used directly following dilution with PBS prior to injection.

Determination of maximum binding sites (B_max) and binding affinity (K_d). Saturation binding analysis was performed to determine the α_Vβ₃ receptor densities and radiopeptide binding affinity of cell lines (42). Parental OCUT2, OCUT2-β3KO, FTC133, and FTC133-α_Vβ₃⁺ cell lines were cultured in their respective media and maintained at 60%–70% confluence. Cells were individually seeded at a density of 4–5 × 10⁴ per well in 48-well flat-bottom microplates and recuperated overnight at 37°C. The wells were designated as non-specific binding (NSB) and total binding (TB). For the FTC133-α_Vβ₃⁺ cell line, cells were counted after adenoviral transfection was complete and before NSB or TB designation. Increasing concentrations of ⁶⁸Ga-labeled peptide, ⁶⁸Ga-DOTA-RGD₂ (10–3000 nM), were added, and cells were incubated at 37°C for 60 minutes. Cold (⁶⁸Ga-unlabeled) peptide at a 5× concentration relative to the highest radiolabeled peptide concentration (i.e., 15 μM) was used as a blocking solution to pretreat cells in the NSB wells. Cells were washed, the bound fractions were collected using 1.5 M NaOH, and the radioactivity was counted with a well-type automatic γ counter (2480 Wizard2, PerkinElmer). Specific binding counts were obtained by subtracting NSB counts from TB counts and converted to specifically bound radioligand (in fmol). Nonlinear regression analysis with extra-sum-of-squares F test was used to analyze data from at least 3 different experiments and calculate K_d and B_max values using the 1-site specific binding function (Y = B_max × X/[K_d + X], where X = radioligand concentration and Y = specific binding) in GraphPad Prism (43), and Welch’s t test was used to compare the B_max and K_d values.

Animal studies. Mice were housed up to 5 per cage, provided ad libitum access to water and standard chow (NIH-31 formula; 5L7X, Animal Specialties and Provisions). The animal room was maintained at 22°C with 50% humidity and a 12:12-hour light/dark cycle with lights off at 6 pm. Experiments were performed with equal numbers of male and female mice ranging from 8 to 13 weeks of age, with a starting weight of 20 g or more. To generate cell-derived xenografts (CDXs), athymic nude mice (NU/J, strain 002019, The Jackson Laboratory) were injected subcutaneously with 5 × 10⁶ OCUT2 cells. Tumor growth and mouse weights were measured at regular intervals, and treatment was started when the average tumor size reached approximately 200 mm³. Along with the consideration of similar average tumor size within each group, mice were randomly assigned to maintain an equal sex ratio and body weight across groups prior to treatment initiation. The treatment groups were as follows: (a) placebo group n = 8 mice: PBS via oral gavage daily and via tail vein injection once; (b) lenvatinib group n = 8: lenvatinib 10 mg/kg via daily oral gavage and PBS via tail vein injection once; (c) ¹⁷⁷Lu group n = 8: 0.5 mCi of ¹⁷⁷Lu-DOTA-EB-RGD via tail vein injection and PBS via oral gavage daily; (d) ¹⁷⁷Lu+L group n = 8: 0.5 mCi of ¹⁷⁷Lu-DOTA-EB-RGD via tail vein injection and lenvatinib 10 mg/kg via daily oral gavage. 200 μL was the delivered volume for all treatments.

Static PET imaging for ⁶⁸Ga-DOTA-RGD₂ (100 μCi) was conducted 60 minutes after injection. An Inveon scanner (Siemens) was utilized to capture PET images. Thereafter, the images were analyzed using ASIPro software (Siemens) to determine regions of interest and quantify SUVs. The SUVs were further corrected with respect to the injected radioactivity dose, decay values, branching, and respective mouse weight. After the last scan, mice were euthanized and tumors, heart, liver, and kidneys were collected to perform biodistribution studies. Ex vivo quantification of relative γ radioactivity in tumors and normal organs (liver, heart, and kidneys) was performed by direct measurement using a γ counter (2480 Wizard2, PerkinElmer) 1 hour after the conclusion of the experiment. The counts per minute (CPM) readings were normalized by the weight of the tissues exposed to ¹⁷⁷Lu that was received by 2 groups of animals (¹⁷⁷Lu and ¹⁷⁷Lu+L) 16 days before the conclusion of the experiment and are represented as relative values for each group.

For the generation of metastatic nodules of TC in the lungs, extremely immunodeficient mice, NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, commonly referred to as NSG mice, were procured from The Jackson Laboratory (strain 005557). FTC133-Luc cells (5 × 10⁵) were injected i.v. through the tail vein to generate metastatic nodules in the lungs. The presence of lung and/or bone metastases was monitored weekly by bioluminescence quantification via IVIS imaging (IVIS Lumina Series III, Revvity). Mice were intraperitoneally injected with a freshly prepared solution of D-luciferin (potassium salt, Regis Technologies, 1-360222-200, 150 mg/kg body weight) in sterile PBS and anesthetized by isoflurane inhalation before IVIS imaging was performed with automatic acquisition and luminescent exposure of 60 seconds. Regions of interest (ROIs) of the same area were marked in the captured images of the animals, and the built-in analysis function was used to determine the levels of bioluminescence in terms of total counts. IVIS imaging was performed on the days of inoculation, start of treatment, and euthanasia. Sex-unbiased randomly divided mice with similar average tumor size and/or metastatic tumor burden quantified by IVIS bioluminescence with a threshold of >10⁴ counts, were categorized into 4 groups (n ≥ 6): (a) placebo, designated P, received vehicle only; (b) lenvatinib alone (L) received 10 mg/kg body weight lenvatinib via oral gavage; (c) ¹⁷⁷Lu-DOTA-EB-RGD (¹⁷⁷Lu) subjected to monotherapy with 0.5 mCi radiotherapy tracer DOTA-EB-RGD conjugated to ¹⁷⁷Lu (t_½ 6.6 days) via tail vein injection, or (d) in combination with 10 mg/kg body weight lenvatinib (¹⁷⁷Lu+L). All the treatment groups received placebo doses complementary to other groups via i.v. or p.o. at the same rate to normalize the levels of stress in animals among groups.

RNA-seq analysis of resected tumors. Total RNA from each tumor was quantified through a NanoDrop ND-1000. The library preparation was done using 1–2 g of the total RNA. Thereafter, an Agilent 2100 Bioanalyzer was employed to verify the completed libraries and quantified through the method of absolute quantification qPCR. Next, the barcoded libraries were mixed and further denatured in NaOH to single-stranded DNA, recorded on an Illumina flow cell, then amplified in situ, followed by sequencing with an Illumina NovaSeq 6000 instrument for 150 cycles.

The quality of the sequence was assessed using the FastQC software version 0.11.7 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The trimmed reads (trimmed 5′, 3′-adaptor bases utilizing cutadapt 1.17 (https://cutadapt.readthedocs.io/en/stable/) were then aligned to the reference genome with the Hisat2 software version 2.1.0 (https://cutadapt.readthedocs.io/en/stable/). StringTie 1.3.3 (https://github.com/gpertea/stringtie) was used to analyze the transcript abundances in each sample and the R package Ballgown 2.10.0 (https://bioconductor.org/packages/3.22/bioc/html/ballgown.html) was used to calculate the FPKM (fragments per kilobase of exon per million mapped reads) value for gene and transcript level. Thereafter, the R package Ballgown was used in order to filter the differentially expressed transcripts and genes. Next, the assembled results were employed to predict the new genes and transcripts through comparison with the reference annotation (by StringTie and Ballgown). Furthermore, detection of alternative splicing plot and events was performed through Replicate Multivariate Analysis of Transcript Splicing (rMATS). Correlation analysis and Principal Component Analysis (PCA) were based on FPKM. Gene Ontology, Hierarchical Clustering, gene expression level, Pathway analysis, volcano plots and scatter plots were conducted using the differentially expressed genes in Python 2.7 (https://www.python.org/), R or shell environments for the graphics as well as statistical computing. The RNA-seq was performed by Arraystar Inc.

Statistics. The normal distribution of continuous data is represented as mean with standard deviation (mean ± SD) or standard error of mean (mean ± SEM), and parametric tests were used for intergroup comparisons (t test or ANOVA for more than 2 groups) as appropriate. For non-normally distributed data, non-parametric tests such as the Mann-Whitney U test or the Wilcoxon rank-sum test were used. Mixed-effects model was used to assess the differences in the tumor growth rates in the 4 groups of mice subjected to different treatment modalities, and non-linear regression modeling was used for binding assays. GraphPad Prism 10.2.3 software was utilized to analyze the data, and a P value of 0.05 or less was considered to be statistically significant.

Study approval. All animals were maintained at NIH facilities in accordance with the guidelines established by the Association for Assessment and Accreditation of Laboratory Animal Care (AALAC), and all protocols were approved by the Institutional Animal Care and Use Committee (protocol: K122-MDB-24) at the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK).

Data availability. Raw data for all the experiments performed in this study as well as next-generation sequencing data are stored in the publicly available NCBI Gene Expression Omnibus (accession number GSE343650). The analytical codes are submitted in Github (commit URL: https://github.com/sonamkumari2025/RNAseq-Volcano-Plot.git; and the ID: 8a88f17ae69d2c24031898a21fba92b8fcde78d7). All the raw data from which the plots are generated in the main and supplemental figures are submitted as a Supporting Data Values file.

Statement on use of AI. On June 3, 2026, ChatGPT v.5.5, US Department of Health Enterprise was used to conceptualize the graphical abstract and was created using BioRender. Prompt used: “Using the results summary create graphical abstract. Do not recapitulate the data, rather put schematic of take-home message.”