Microfluidic-based patient-derived organoids recapitulate thyroid cancer heterogeneity and reveal NF-κB-driven maturation for precision therapy

Abstract

Background

Inter-tumor heterogeneity poses significant challenges for precision therapy in thyroid cancer (TC). The conventional organoid models are limited by inefficiency and poor physiological relevance.

Methods

We developed droplet-engineered organoids (DEOs) using microfluidic 3D bioprinting to rapidly generate patient-derived TC models. These DEOs were characterized via histology, whole-exome and RNA sequencing, and utilized for drug sensitivity testing and metastasis modeling.

Results

DEOs were generated within 10 days, exhibiting superior uniformity (CV: 2.54%) and a high success rate (76%). They faithfully recapitulated the histopathological architecture, genomic landscape (92% driver gene concordance), and native immune microenvironment (CD3+/CD56+/CD68+/α-SMA+) of parental tumors. Drug screening revealed patient-specific heterogeneity, accurately mirroring clinical responses, including cisplatin sensitivity and anti-PD-1 resistance. We established a novel TC and lung organoids co-culture model, which could be used to study the TC lung metastasis. Crucially, transcriptomics identified stage-specific maturation driven by NF-κB signaling. Pharmacological inhibition of NF-κB synergistically enhanced the efficacy of dasatinib, anti-PD-1, and paclitaxel, with combination index (CI) values of 0.58, 0.45, and 0.80, respectively.

Conclusions

Our microfluidic platform enables rapid, high-fidelity modeling of TC, offering a scalable and physiologically relevant tool for mechanistic studies, drug screening, and personalized therapy prediction, with highly promising translational potential.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-07882-z.

Introduction

Thyroid cancer is a malignant tumor originating from the follicular or parafollicular epithelium of the thyroid gland [1]. In recent years, the incidence of thyroid cancer has increased worldwide. According to the Global Cancer Statistics 2022, thyroid cancer is the fourth most common cancer among women, with approximately 820,000 new cases annually worldwide, and China accounting for about 460,000 of these new cases, representing an annual increase of 20% in the incidence rate [2, 3]. Differentiated thyroid cancer (DTC), which comprises over 90% of all thyroid cancers, is primarily managed through surgery [4]. Although the prognosis for most patients is favorable, the risk of distant metastasis ranges from 5% to 25% [5]. Treatment options remain limited for rarer forms such as undifferentiated carcinoma and high-grade papillary carcinoma, for which targeted therapies are still under development [6]. Over the past decade, the FDA has approved nine kinase inhibitors or combination therapies, including sorafenib, lenvatinib, and cabozantinib, mainly for progressive, radioiodine-refractory DTC and medullary thyroid cancer [7]. Nonetheless, therapeutic responses vary considerably among TC patients.

Thyroid cancer exhibits a high frequency of potentially druggable molecular alterations, including the highest frequency of oncogene-driven kinase fusions observed among solid tumor [8]. The BRAF V600E mutation is particularly common in thyroid cancer and is associated with tumor progression and immune escape [9], underscoring the need for personalized treatment strategies. The heterogeneity of treatment response may be attributable, in part, to the patient’s genetic background [10]. Furthermore, Myeloid cells contribute to the formation of the tumor immune microenvironment, influenced by alterations in monocytes even before tumor infiltration [11]. However, the advancement of precision medicine for thyroid cancer has been hampered by the lack of physiologically relevant in vitro models for evaluating therapeutic efficacy. Currently, in vitro models for thyroid cancer research include 2D monolayers, spheroids, and organoids. 2D monolayers, due to their non-physiological culture conditions, fail to recapitulate the natural follicular architecture of the thyroid, leading to discrepancies in cellular morphology, function, and drug response compared to the native tissue [12]. Spheroids, while offering a 3D structure, often lack sustained cell growth and do not fully mimic tumor-stroma interactions [13]. Organoids, self-organizing, multicellular structures derived from adult stem cells, embryonic stem cells, or induced pluripotent stem cells, preserve the heterogeneity of the original tissue and more accurately emulate tumor biology and drug responses, thereby improving predictive accuracy in drug testing [14, 15]. Patient-derived organoids (PDOs) from papillary thyroid cancer (PTC) can be established and are able to maintain consistent histological and genetic profiles [16]. Despite these advances, generating organoids with realistic and reproducible 3D structures remains challenging. Conventional methods often rely on manual fabrication of dome-like architectures, which are labor-intensive and yield results with high variability [17]. Microfluidic technology has emerged as a critical tool for automating organoid production, enabling high-throughput generation, enhancing nutrient and waste exchange, and supporting the formation of uniform organoids and spheroids in a cost-effective manner [18, 19]. The integration of 3D bioprinting with microfluidics thus offers a promising strategy for scalable, reproducible, and physiologically relevant thyroid cancer organoid models.

The NF-κB signaling pathway plays a crucial role in promoting thyroid cancer cell survival and proliferation [20]. Recent studies have demonstrated that RBM10 gene deletion leads to NF-κB pathway-dependent enhancement, which significantly promotes thyroid cancer cell proliferation and invasion via activation of RELA (p65) [21]. In the organoid culture systems, the NF-κB signaling pathway has been shown to be essential for maintaining cancer stem cell properties, particularly in terms of self-renewal capacity and differentiation potential [22]. Notably, pharmacological inhibition of the NF-κB signaling pathway results in a marked reduction of cancer stem cell populations, leading to significant suppression of tumor growth and metastatic potential. Collectively, these findings indicate that the NF-κB signaling pathway orchestrates multiple oncogenic processes in thyroid cancer, making it an attractive target for novel therapies.

In this study, we leveraged microfluidic technology to develop 3D patient-derived thyroid cancer droplet-engineered organoids (DEOs) that matured within 10 days, significantly decreasing the ripening time compared to conventional methods. We comprehensively characterized the molecular profiles of these DEOs and performed personalized drug screening using clinically relevant agents. Our models faithfully recapitulate inter- and intra-tumor heterogeneity and drug response variations. Furthermore, we established a novel model of thyroid cancer lung metastasis, providing a valuable platform for investigating metastatic mechanisms and evaluating new therapeutics. Finally, we identified stage-specific gene expression patterns in DEOs and revealed the critical role of NF-κB signaling in organoid maturation and growth, informing the design of synergistic combination therapies with enhanced anticancer efficacy.

Materials and methods

Human specimens

Thyroid tumor tissues were collected from patients undergoing surgical resection at Shenzhen People’s Hospital, China, between March and June 2024. For each patient, sex, age, tumor size, and clinical stage were documented where available (Supplementary Table 1). TNM staging was determined according to the eighth edition of the American Joint Committee on Cancer (AJCC) staging system. Paracancerous lung tissues were obtained from lung cancer patients at the same institution. This study was approved by the Medical Ethics Committee of Shenzhen People’s Hospital (Ethical Development No. 2024–08401). All patients provided written informed consent in accordance with the Declaration of Helsinki. Diagnoses were confirmed by two independent pathologists based on routine H&E-stained sections.

Tumor tissue processing

Tissue samples of thyroid cancer were obtained from Shenzhen People’s Hospital (Ethical Development No. 2024–08401). After the patients signed an informed consent form, the remaining cancer tissues from surgical resection were packed in tissue preservation solution and transported to the laboratory at 4℃. Tissues were washed and minced in PBS, then centrifuged to remove PBS. A digestive solution with collagenase (Yuanye Bio, Shanghai, China, S35952), DNA enzyme (Solarbio, Beijing, China, D8074), and dispersing enzyme (Yuanye Bio, Shanghai, China, S25046) in Advanced DMEM/F-12 medium (Gibco, USA, 12634010) with 1% BSA was prepared. Tissues were digested in this solution at 37 °C for 15 min, stopped with serum-containing medium, and centrifuged. The precipitate was resuspended in PBS, filtered through a 100 μm sieve, and centrifuged again. Erythrocyte lysate was added based on red blood cell count, incubated, and stopped with PBS. The mixture was centrifuged, and the supernatant was discarded.

Generation of DEOs

Droplet-engineered organoids (DEOs) were fabricated using the OrgFab automated bioprinter. The instrument was sterilized under UV light for 30 min prior to use. A pre-connected sample-loading chip was mounted on the cooling platform. A single-cell suspension was prepared through enzymatic digestion, centrifuged, and resuspended in thawed Matrigel (Corning, USA, #352434) at a density of 5 × 10⁴ cells/µL. The mixture was transferred to a chilled microcentrifuge tube. A 96-well culture plate was positioned on the printing stage, and PT tubing was placed in the solidification chamber followed by insertion of the PDMS print head. Printing parameters were set to Matrigel (Corning, USA, #352434), 100 µL dispensing volume, and 3 DEOs per well. The printing process was initiated under controlled cooling and solidification conditions to ensure cell viability.

Organoid culture and passage

The culture medium for thyroid cancer organoids consisted of Advanced DMEM/F-12 (Gibco, USA, #12634010) supplemented with 1% GlutaMAX (Gibco, USA, #35050-061), 1% penicillin-streptomycin (Biosharp, Anhui, China, #BL505A), 1% N2 supplement (Gibco, USA, #17502048), 10 µM SB202190 (Yeasen, Shanghai, China, #53005ES08), 5 ng/mL FGF-2 (Yeasen, Shanghai, China, #91330ES10), 100 ng/mL FGF-10 (Yeasen, Shanghai, China, #91306ES60), 5 ng/mL EGF (Yeasen, Shanghai, China, #92701ES10), 100 ng/mL Noggin (Yeasen, Shanghai, China, #92528ES60), 100 ng/mL R-Spondin-1 (Yeasen, Shanghai, China, #92278ES60), 10 mM nicotinamide (Sigma, USA, #N0636-100G), 500 nM A83-01 (MCE, USA, #HY-10432), 25 ng/mL HGF (Yeasen, Shanghai, China, #92055ES50), 10 µM forskolin (MCE, USA, #HY-15371), 5 µM Y-27,632 (MCE, USA, #HY-10071), and 1% B27 supplement (Gibco, USA, #17504044). DEOs were maintained at 37 °C under 5% CO₂, with medium changes every three days. For passaging, organoids were collected, treated with 0.05% trypsin/EDTA at 37 °C for 5 min, mechanically dissociated, washed with PBS, and centrifuged at 300×g for 5 min. The pellet was resuspended in Advanced DMEM/F-12 and subsequently in Matrigel. Organoids were passaged every twelve days.

Thyroid cancer lung metastasis model

Lung and thyroid cancer DEOs were printed using OrgFab at a density of 2 × 10⁴ cells/µL. Thyroid cancer DEOs were stained with DiI perchlorate (Yeasen, Shanghai, China, #40726ES10) for 10 min and washed twice with PBS. One stained thyroid DEO and one lung DEO were transferred to a centrifuge tube, supernatant was aspirated, and 5 µL Matrigel was added to embed the organoids. The gel was pipetted into a 48-well plate. After 10 days of culture, migration was assessed using fluorescence microscopy. The distance between thyroid and lung DEOs was quantified with ImageJ software. The migration distance between thyroid cancer and lung organoids after co-culture was measured using ImageJ software. Statistical significance was determined using a paired two-sided Student’s t-test. A p-value < 0.05 was considered significant.

Histology and immunostaining

Tissues and organoids were fixed in 4% PFA for 24 h at room temperature, embedded in paraffin, and sectioned at 5 μm thickness for H&E staining and immunofluorescence. Sections were incubated with primary antibodies at 4 °C overnight. The following antibodies were used: TTF1 (HuaBio, Hangzhou, China, #RT1633, 1:500), calcitonin (HuaBio, Hangzhou, China, #ER1905-32, 1:300), thyroglobulin (ABclonal, Wuhan, China, #A3407, 1:200), CEACAM6 (HuaBio, Hangzhou, China, #EM1901-68, 1:100), CD3 (ProteinTech, Wuhan, China, #17617-1-AP, 1:300), CD56 (HuaBio, #ET1702-43, 1:100), CD68 (ABclonal, Wuhan, China, #A13286, 1:200), α-SMA (ABclonal, Wuhan, China, #A17910, 1:200), and SFPTC (ABclonal, Wuhan, China, #A1835, 1:100). After PBS washes, samples were incubated with appropriate secondary antibodies and counterstained with DAPI (Thermo Fisher Scientific, USA, #D1306). Imaging was performed using a Leica fluorescence microscope.

Cell viability and imaging

Viability of TC DEOs was assessed on day one using the LIVE/DEAD Viability/Cytotoxicity Kit (Yeasen, Shanghai, China, #40747ES76). Organoids were rinsed with PBS and incubated in PBS containing 1 µL Calcein AM (2 mM) and 3 µL propidium iodide (1.5 mM) for 1 h at room temperature. Unbound dye was removed by washing with PBS. Stained organoids were visualized under a fluorescence microscope (Nikon Eclipse Ts2r).

Whole-exome sequencing

Whole-exome sequencing (WES) was conducted by ChiBiotech Co., Ltd. (Shenzhen, China). Genomic DNA was extracted from patient tumors and organoids using the DNeasy Blood & Tissue Kit (QIAGEN, Germany, #69504) and Maxwell 16 Tissue DNA Purification Kit (Promega, USA, #AS1030). Libraries were constructed through shearing, end repair, A-tailing, and adapter ligation. Exome capture was performed using the HaloPlex Exome Kit (Agilent, USA, #G9906A). Sequencing was carried out on an Illumina HiSeq 2500 platform (2 × 100 bp). Reads were aligned to the GRCh37/hg19 reference genome using the Burrows-Wheeler Aligner and processed via the IPM-Exome pipeline (v0.9).

RNA-sequencing

Total RNA was extracted from tissues and organoids at early and late stages using a commercial kit (Tiangen Biotech, Beijing, China, #DP424). RNA quality and quantity were assessed prior to library preparation. Sequencing libraries were constructed from 2 µg total RNA and sequenced on an Illumina NovaSeq 6000 platform (ChiBiotech Co., Ltd.). Differential expression analysis was performed using the edgeR package (v3.40.2) in R. Genes with an absolute fold change > 2 and a false discovery rate (FDR) adjusted p-value (q-value) < 0.05 were defined as differentially expressed genes (DEGs).

Drug screening

Drug sensitivity assays were performed using 5-point, 5-fold dilution series of anlotinib (Selleck, USA, #AL3818,), dasatinib (Aladdin, Shanghai, China, #D1371597), and anti-PD-1 (Aladdin, Shanghai, China, #C412010), with concentrations ranging from 0.4 µM to 250 µM; and 4-point, 5-fold dilutions of cisplatin (Aladdin, Shanghai, China, #D109812) and paclitaxel (Aladdin, Shanghai, China, #P106868), ranging from 8 µM to 1000 µM. For combination treatments, DEOs were co-treated with 1 µM BAY 11-7082 (MCE) and dasatinib (0.4, 2, or 10 µM), anti-PD-1 (20–100 µg/mL), or paclitaxel (4, 10, or 20 µM). After six days, viability was assessed using the CellTiter-Glo 3D Assay (Promega, USA, #G9681), and luminescence was measured on a SpectraMax iD3 plate reader (Molecular Devices, USA). Dose-response curves and IC₅₀ values were derived through nonlinear regression analysis.

The nature of drug interactions (synergy, additivity, or antagonism) was quantitatively assessed using the Chou-Talalay method for combination index (CI) calculation. Dose-response data for each drug alone and in combination (at a fixed molar ratio based on their individual IC₅₀ values) were analyzed. CI values were calculated using the median-effect equation: CI = (D₁)/(Dx)₁ + (D₂)/(Dx)₂, where (D₁) and (D₂) are the concentrations of drug 1 and drug 2 in combination that produce a given effect level (Fa), and (Dx)₁ and (Dx)₂ are the concentrations of each drug alone required to produce the same effect. A CI < 1, = 1, and > 1 indicates synergy, additivity, and antagonism, respectively. The statistical significance of synergy (CI < 1) was determined by a one-sample t-test comparing the mean CI values at different effect levels (Fa) against the theoretical value of 1.

Statistical analysis

All data are expressed as mean ± SEM unless otherwise specified. Statistical analyses were determined by Prism GraphPad 7.0 (GraphPad Software). Differences among groups were determined by one-way analysis of variance followed by Tukey multiple comparison test using Prism GraphPad 7.0 (GraphPad Software). Significance was set at P less than 0.05.

Results

Construction of engineered patient-derived thyroid cancer organoids

Traditional approaches to establishing DEOs typically involve manual pipetting [23, 24], which is inefficient and has a low success rate. In this study, we employed OrgFab, an integrated bioprinter, to generate DEOs by forming cell-matrix gel droplets [25]. The collected patient-derived thyroid cancer tissues were divided into three parts, which were used for tissue sequencing, histopathological testing, and organoid construction, respectively (Fig. 1a). Tissue samples were minced, enzymatically digested to isolate single cells, and mixed with matrix gel. The resulting cell-matrix suspension was then loaded into OrgFab device to print uniform droplets, which were then gelled at 37 °C to form DEO precursors (Fig. 1a). To determine whether the traditional manual pipetting method could stably produce sizeable organoids within 10 days, we constructed non-engineered organoids (NEOs) from the same tumor samples at a cell concentration of 5 × 10⁷ cells/µL. On day 10, NEOs manifested as small clusters of organoids, with the majority of cells dispersed throughout the Matrigel, lacking any distinct architectural form. In contrast, the DEO precursors showed increasing cell density with apparent vesicular structures by day 4, and these structures grew further with more tightly packed cells by day 10 (Fig. 1b). We performed H&E staining on both primary thyroid cancer tissues and organoids (DEO and NEO) to evaluate the histological features. On day 10, NEO exhibited only a small number of cells, and no discernible structure was observed (Fig. 1c). The results revealed enlarged nuclei and characteristic clear cytoplasm (indicated by black arrows) in both native cancer cells and DEOs (Fig. 1c). The parental tissue and DEOs all formed cyst-like structures with irregular morphology. Given that thyroid cancer tissues primarily originate from the thyroid follicular epithelium, these vesicular structures are likely formed by these epithelial cells [26]. DEOs from the same batch exhibited uniform size, with a coefficient of variation (CV) of 2.54%, proving that our platform significantly reduced variability among the fabricated organoids. Due to manual handling, NEO diameter showed significant variation, with a CV of 7.56%, higher than that of DEOs (Fig. 1d). We collected 50 samples from TC patients, and for each sample, both DEO and NEO were generated simultaneously (Supplementary Table 1). The success rate for DEOs was 76%, higher than that for NEOs (42%) (Fig. 1e, Supplementary Table 2). To characterize how signatures of tumor tissues affect the success rate of establishing DEO cultures, we compared clinical features and cell viability of tumor tissues that failed to establish DEO cultures to those that were successful (Fig. 1f). We found that the success rate of DEO increased when tissue cell viability exceeded 40%. However, NEO demanded the maintenance of tissue cell viability above 60% to successfully establish organoids (Fig. 1g). The patient age, gender, and TNM stage did not show significant correlation with the successful generation of TC DEO (Supplementary Table 2). Together, these findings demonstrate that DEO exhibits greater dimensional uniformity and reproduces the key histopathological features of primary thyroid carcinoma more rapidly and accurately than NEOs.

Fig. 1.

Construction of engineered patient-derived thyroid cancer organoids. a Experimental design. Fresh TC samples were obtained from patients and then used to construct DEO by using OrgFab device. b The bright-field image of DEOs and NEOs. Scale bar, 100 μm. c The H&E staining of parental tissue and DEOs and NEOs on day 10. Black arrows show clear cytoplasm in tumor tissue and DEO. Scale bars, 100 μm. d. The violin plots of diameters of DEOs and NEOs. e. Comparison of DEO and NEO construction success rates. f. Clinical features and cell viability of successful DEO culture tissues. The features are indicated with a color code. g. Comparison of tissue cell viability in successfully constructed DEO and NEO

DEOs recapitulate the histopathological characteristics of the originating tumor

To verify whether DEOs restore the cell components of the parental tissue, we used immunofluorescence (IF) staining to compare the expression of thyroid tumor markers in originating tumor tissue and DEOs. Thyroid transcription factor-1 (TTF1) is expressed in thyroid follicular and parafollicular cells and is a marker of differentiated thyroid carcinoma. TTF1-positive cells were scattered throughout the DEOs (Fig. 2a). CEACAM6 is a cell adhesion protein that is overexpressed in thyroid cancer and was found to be expressed in both the parental tumor tissue and DEOs (Fig. 2a). Additionally, calcitonin (CT) and thyroglobulin (TG) are secreted by follicular cells of thyroid cancer [27]. The DEOs retained expression levels of CT and TG comparable to that of the parental tissue (Fig. 2a). Overall, the DEOs recapitulated the immunohistochemical characteristics of their corresponding primary cancers. In addition to the TC cells, the tumor immune microenvironment also plays an important role in tumor growth. Infiltrating immune cells in the PTC microenvironment may be closely related to tumor progression and serve as an important indicator for assessing the prognosis of patients [28]. We examined the expression of immune cell markers in tissues and DEOs to elucidate whether organoids retained an immune cell population mimicking that of the native tumor. IF analysis revealed consistent distribution and expression intensity of CD3, a T-cell marker, between organoids and their original tumor tissue (Fig. 2b, c). The same was true for three additional biomarkers, including CD56 (NK-cell marker), CD68 (macrophage marker) and α-SMA (tumor-associated fibroblast marker) (Fig. 2b, c). Immunofluorescence showed that expression of the above markers was not significantly different from the parental tissue. Collectively, these results suggest that DEOs retain the structural and genetic characteristics of their originating tumor tissue.

Fig. 2.

TC-derived DEOs recapitulate histopathological characteristics of parental tumors. a Representative immunofluorescence images of paired TC DEOs and tumor tissues for TTF-1, CEACAM6, CT and TG. Nuclei were stained with DAPI (blue). Scale bar, 100 μm. b Representative immunofluorescence images of paired TC DEOs and tumor tissues for CD3, CD56, CD68 and α-SMA. Nuclei were stained with DAPI (blue). Scale bar, 100 μm. c Mean gray value statistics for tissue and DEOs immunofluorescence images (n = 3). ns, no significant

Genetic concordance between advanced-stage thyroid cancer organoids and parental tissues

TC DEOs exhibit distinct morphological and structural alterations during development. On the early-stage (Day 1-Day 5), the cells in organoids are distributed uniformly, there is no obvious lumen structure, and the diameter of the organoids is approximately 500 μm. By the late stage (Day 6-Day 10), the cell growth rate accelerates, presenting a clustered growth pattern. Within the organoids, mature lumen structures are formed. At the edges of the organoids, budding structures are present. The diameter of the organoids further shrinks to about 300 μm (Fig. 3a). To evaluate whether mature DEOs maintained the genetic mutation profile of the original tumor tissue, we performed whole-exome sequencing (WES) of DEOs and parental tumor samples. The distribution of somatic insertion and deletion (InDel) mutations was highly similar between primary TC tissues and their corresponding DEOs (Fig. 3b). C > T/G > A transitions were the most prevalent substitution type in both tumor samples and DEOs, while T > A/A > T transitions were the least frequent (Fig. 3c). Copy number variation (CNV) analysis revealed comparable patterns of genomic gains and losses in DEO lines and their corresponding primary tumors (Fig. 3d, e). By comparing somatic mutations with known cancer driver genes in public databases, we identified a 92% concordance in driver gene mutations between DEOs and original tissues (Fig. 3f). Notably, mutations were detected in genes such as NCOA4 [28], AHNAK2 [29], and PRDM16 [30], which have been previously implicated in papillary thyroid carcinoma pathogenesis. To further assess genetic stability during organoid culture, we compared WES data from early-stage (day 5) and late-stage (day 10) DEOs alongside their respective parental tissues. Late-stage DEOs exhibited CNV profiles that more closely resembled the original tumor than early-stage cultures (Fig. 3g, h). Transcriptomic analysis via RNA-seq showed that late-stage DEOs and primary tissue shared similar expression levels of genes related to cell proliferation and migration (Fig. 3i). Moreover, late-stage DEOs had significantly fewer differentially expressed genes relative to parental tissue compared to early-stage cultures (Fig. 3j). Together, these results indicate that late-stage DEOs faithfully preserve the genetic alterations and global gene expression features of the original thyroid cancer tissue.

Fig. 3.

Genomic features of DEOs at different stages and matched original tumor tissues. a The bright-filed image of DEOs in early-stage and late-stage, respectively. Scale bar, 100 μm. b Proportions of somatic InDel in DEOs and parental tumor tissue. c Percentage of six types of base substitutions in DEOs and parental tumor tissue. d, e Copy number variation plot (d) and CNV statistical pie chart (e) of DEOs and parental tumor tissue. The red and blue dots in c represent copy number increases and decreases, respectively. f Somatic mutation comparison of known driver gene mutation types. Displayed are the top 20 mutated genes observed in the data. The sources of driver genes compared from Cancer Gene Census, BertVogelstein125, SMG127, Comprehensive435, intOGen. g, h Copy number variation plot (g) and CNV statistical pie chart (h) of early-stage organoids (DEO_Early), late-stage organoids (DEO_Late) and parental tumors (Tissue). The red and blue dots in h represent copy number increases and decreases, respectively. i Heatmap of gene expression of early-stage organoids (DEO_Early), late-stage organoids (DEO_Late) and parental tumors (Tissue) by RNA-seq. j A bar chart comparing differentially gene expression (DEG) between late-stage organoids and parental tissue versus those between early-stage organoids and parental tissue by RNA-seq. Bar height represents the number of DEGs (FDR < 0.05), showing significantly fewer DEGs between DEO_Late and Tissue compared to DEO_Early and Tissue

Clinical drug screening using thyroid cancer organoids

To assess the potential of DEOs as a preclinical drug screening platform, we treated DEOs with a panel of clinically relevant agents, including the chemotherapeutic agents cisplatin and paclitaxel (PTX), the targeted drugs dasatinib and anlotinib, and the immune checkpoint inhibitor anti-PD-1. Drug selection was based on their current application in thyroid cancer therapy or investigational status in clinical trials. In vitro drug sensitivity assays revealed considerable interpatient heterogeneity in DEO responses. Most DEOs exhibited greater sensitivity to cisplatin, with area under the curve (AUC) values below 30% (Fig. 4a), whereas limited responsiveness was observed to anti-PD-1 treatment, reflected by AUC values exceeding 70% (Fig. 4d), which was consistent with clinical profiles. Responses to dasatinib and anlotinib varied across DEOs, ranging from resistant to partially sensitive or sensitive (Fig. 4b, c). Moreover, DEOs derived from three different patients showed markedly divergent reactions to the same drug regimen, highlighting not only interpatient but also intratumor heterogeneity in drug susceptibility (Fig. 4e-h). These results strongly support the value of DEOs in replicating patient-specific drug responses and their potential for guiding personalized therapeutic strategies.

Fig. 4.

Differential drug responses in patient-derived DEOs under drug treatment. a-d Summary of chemosensitivity responses for DEOs to cisplatin (a), dasatinib (b), anlotinib (c) and anti-PD-1(d). Dose-response curves (n = 3) are shown, with the AUC calculated from raw data and displayed as a violin plot. e-f Heterogeneous drug responses of DEOs derived from patient P5 (e), patient P6 (g) and patient P7 (f). h The heatmap illustrates the chemosensitivity responses of organoids to cisplatin, PTX, and dasatinib (n = 3), quantified using AUC Z-scores. Color intensity ranges from purple (low scores, indicating sensitivity) to yellow (high scores, indicating resistance)

Construction of an organoid model for thyroid cancer lung metastasis

The lung is the most common site of distant metastasis in patients with differentiated thyroid cancer [31], and the development of lung metastases is associated with poor prognosis, significantly affecting quality of life and overall survival [32]. To model this process, we generated lung organoids using paracancerous tissue from human lung cancer samples. By day 10 of culture, the organoids developed large cystic structures morphologically resembling alveolar compartments of the native lung tissue (Fig. 5a, b). Immunofluorescence analysis confirmed expression of the lung epithelial marker TTF1 and type II alveolar epithelial cell markers in these organoids, consistent with the expression profile observed in the original tissue (Fig. 5c). To investigate the metastasis potential of thyroid cancer cells to the lung microenvironment, we established a paracrine-signaling model by co-culturing thyroid cancer DEOs with lung-derived organoids. This setup aimed to mimic the lung metastatic niche in vitro. We observed pronounced morphological changes in thyroid cancer DEOs following co-culture. By day 10, the DEOs adopted a dispersed architecture, forming multiple smaller structures and exhibiting marked migratory behavior toward the lung organoids (Fig. 5d, e). These results suggest that the lung microenvironment promotes migratory responses and may facilitate the metastatic dissemination of thyroid cancer cells.

Fig. 5.

Construction of an organoid model for thyroid cancer lung metastasis. a DEOs derived from human lung cancer paracancerous tissue. Scale bar, 100 μm. b H&E staining of parental lung tissue and DEOs. Scale bar, 100 μm. c Representative immunofluorescence images of TTF-1 and SFPTC in parental lung tissue and DEOs. Nuclei were stained with DAPI (blue). Scale bar, 100 μm. d Representative bright-field and confocal images of co-culture of TC DEOs and lung DEOs. TC DEOs were dyed with Dil perchlorate (red). Scale bar: 100 μm. e Quantification of the distance between two types of organoids. Each data point represents an independent co-culture unit. Data are presented as mean ± SD; *p < 0.05 (paired two-sided Student’s t-test)

NF-κB signaling promotes maturation and enhances therapeutic sensitivity in thyroid cancer organoids

To investigate the molecular mechanisms underlying DEO maturation, we compared the transcriptomes of early-stage (Day 5) and late-stage (Day 10) DEOs. Late-stage DEOs showed elevated expression of pro-tumorigenic genes such as MCAM and HKDC1, the thyroid hormone synthesis gene DUOXA2, and the developmental genes WNT8B and RET. In contrast, early-stage DEOs exhibited upregulation of apoptosis-related genes including LMNB1, RRM2, and TNF (Fig. 6a). Gene Ontology (GO) analysis indicated that upregulated differentially expressed genes (DEGs) were enriched in terms related to positive regulation of cell activation, plasma membrane components, and signaling receptor binding (Fig. 6b). KEGG pathway analysis further revealed significant activation of the NF-κB signaling pathway in late-stage DEOs (Fig. 6c), suggesting its functional role in promoting proliferation and maturation of organoids. Consistent with this, gene set enrichment analysis (GSEA) also showed upregulation of NF-κB-related gene sets in late-stage compared to early-stage DEOs (Fig. 6d), indicating that NF-κB inhibition may represent a promising therapeutic strategy to suppress tumor growth.

Fig. 6.

NF-κB signaling promotes maturation and enhances therapeutic sensitivity in thyroid cancer organoids. a Heatmap shows the expression levels of DEGs between early-stage organoids (DEO_Early), late-stage organoids (DEO_Late). b Gene ontology analysis of DEGs in DEOs at day 5 versus day 10 (p-value < 0.05). Bar chart showing the top 5 GO terms for biological process, cellular component and molecular function, ranked by -Log10 p-value. The size of the bubble indicates the number of genes enriched in each GO item. c Scatterplot of KEGG pathway enrichment analysis for DEGs in DEOs at day 5 versus day 10. The x-axis represents the rich factor (ratio of DEGs to total genes in a pathway), and point colors indicate p-values. d GSEA using DEGs in RNA-seq data of early-stage organoids and late-stage organoids. e DEOs treated with anti-tumor drugs (dasatinib, anti-PD-1, PTX), BAY 11-7082, or their combination for 6 days. The combination index (CI) for each pair is indicated in the graph, demonstrating synergy. f-h Statistical analysis of DEOs cell viability and dead cell fluorescence staining following anti-tumor drugs: 0.4 µM dasatinib (f), 100 µg/mL anti-PD-1 (g), 4 µM PTX (h), 1 µM BAY 11-7082, or their combination. Scale bar: 100 μm. Data are representative of at least three independent experiments. Error bars represent means ± SD. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001

To functionally validate these findings, we treated Day 10 DEOs with the NF-κB inhibitor BAY 11-7082, either alone or in combination with anticancer agents (dasatinib, anti-PD-1, or PTX). Inhibition of NF-κB significantly sensitized DEOs to these drugs, resulting in a pronounced decrease in cell viability compared to monotherapy (Fig. 6e-h). While neither BAY 11-7082 nor single-agent treatments alone markedly affected viability, their combination led to a synergistic reduction in cell survival. Moreover, dose-dependent decreases in viability were observed with increasing concentrations of dasatinib or PTX, either alone or in combination with BAY 11-7082. The synergistic effect of BAY 11-7082 combination therapy was further quantified. For each drug combination, the combination index (CI) was calculated using the Chou-Talalay method, where CI < 1, = 1, and > 1 indicate synergy, additive effect, and antagonism, respectively. All combinations showed CI values significantly below 1 (p < 0.01, one-sample t-test against a hypothetical mean of 1), confirming synergistic interactions. Together, these results underscore the potential of targeting NF-κB signaling to increase the effectiveness of conventional anticancer therapies in TC.

Discussion

The traditional models for TC study have largely relied on 2D monolayers cell cultures, patient-derived xenograft (PDX) models, and transgenic mice, which are limited in recapitulating the physiological complexity of human tumors. Droplet-engineered organoids (DEOs) overcome many of these limitations by closely mimicking the human tumor microenvironment, enabling more accurate modeling of disease biology and behavior, as well as improved prediction of drug responses and disease progression. While patient-derived DEOs from papillary thyroid carcinoma (PTC) have previously been established using conventional methods [33], our study introduces an advanced approach involving the encapsulation of patient-derived thyroid cancer cells in Matrigel to generate uniform droplet-based organoid precursors. These precursors developed into mature organoid structures within 10 days, substantially reducing the time and variability associated with traditional organoid culture. Importantly, the resulting organoids retained critical characteristics of the original tumors, including histopathological features, genomic landscapes, and immune microenvironments.

The immune microenvironment plays a critical role in thyroid cancer, influencing tumor progression, therapeutic response, patient outcomes, and male reproductive health [34, 35].Its composition-including T cells, macrophages, dendritic cells, and NK cells-exhibits considerable heterogeneity across histological and molecular subtypes, contributing to varied clinical behaviors and differential responses to immunotherapy and targeted treatments [36, 37]. This heterogeneity underscores the necessity of personalized therapeutic approaches. Recapitulating the tumor immune microenvironment (TIME) has become a crucial criterion for evaluating the success of cancer organoid models. Recent advances, such as those reported by Neal et al. [38] and Dijkstra et al. [39], demonstrate that patient-derived organoids can incorporate autologous immune components when cultured under optimized conditions that preserve immune cell viability and function, such as using minimally digested tumor fragments or co-culture systems. These immune-competent organoids maintain key features of the original TIME, including immune cell diversity, spatial organization, and cytokine signaling. Our thyroid cancer organoid model offers significant advantages in reconstructing the TIME: it retains patient-specific immune profiles, including functional T cells, macrophages, tumor-associated fibroblast, and NK cells. This makes our model particularly valuable for assessing interpatient heterogeneity and predicting individualized treatment responses.

Organoids have demonstrated clinically related drug responses with a certain degree of fidelity, usually as good as and sometimes even more preferable than animal models, with an overall consistency of 83.33% between drug sensitivity and clinical responses [40]. Engineered organoids provide significant advantages over conventional cell lines in high-throughput drug screening due to their physiological relevance and structural and functional resemblance to human tissues [41, 42]. Moreover, DEOs offer precise control of the microenvironment of living cells, mechanical cues, biochemical signals, tissue interface, and cell-cell/matrix communication, which is not possible with traditional cell lines. These capabilities highlight the potential of engineered organoids to transform high-throughput drug screening by enabling more accurate and personalized prediction of drug efficacy and toxicity. In this study, we treated DEOs to a broad panel of anticancer agents, including chemotherapeutic, targeted, and immunotherapeutic compounds, and successfully established the feasibility of high-throughput screening using this platform. The assays revealed substantial heterogeneity in drug responses across different thyroid cancer organoid lines, reflecting patient-specific variability.

The lung represents the most common site of distant metastasis in thyroid cancer, contributing significantly to disease morbidity and mortality. Therefore, understanding the underlying mechanisms of TC lung metastases and developing effective therapeutic interventions are critically important [43, 44]. At present, there is a lack of a proper model for studying lung metastasis of thyroid cancer. In this study, we established a co-culture of patient-derived lung organoids and TC organoids to simulate TC lung metastasis. The TC organoids faithfully preserve the original tumor’s genetic and phenotypic profile, incorporating human-specific stromal and immune elements, and co-culturing with lung organoids enable reproducible modeling of the metastatic niche under controlled conditions. This system allows for real-time analysis of tumor-stroma interactions, high-throughput drug screening, and personalized prediction of metastatic behavior, providing a more physiologically relevant and scalable platform for investigating thyroid cancer lung metastasis and evaluating potential therapeutics.

The NF-κB signaling pathway plays a critical role in thyroid cancer invasion and metastasis [45]. Mechanistically, HIF-1α-dependent activation of NF-κB induces epithelial-mesenchymal transition (EMT), promotes the translocation of adhesion molecules, and upregulates Vimentin expression, collectively enhancing the invasiveness and migratory capacity of thyroid cancer cells [46]. The inhibition of NF-κB has emerged as a promising therapeutic strategy for advanced and refractory thyroid cancer. Preclinical studies suggest that using NF-κB inhibitors, such as small-molecule IKK inhibitors, or genetically suppressing its activity, can effectively induce apoptosis in thyroid cancer cells. This process also suppresses proliferation and invasion and significantly enhances the efficacy of conventional radiotherapy or chemotherapy [47, 48]. Our results showed that inhibition of NF-κB significantly sensitized DEOs to dasatinib, anti–PD-1, and PTX, resulting in a synergistic therapy in TC. Further research is needed to clarify these mechanisms and identify therapeutic targets within the NF-κB pathway.

Limitations of study

The limitation of this study is the geographical and histologic focus on papillary thyroid cancers (PTC) from a Shenzhen cohort, which may not fully represent the biologic spectrum of thyroid cancer, particularly rare subtypes like anaplastic carcinoma. Furthermore, analysis of culture failures revealed that successful organoid establishment was strongly dependent on initial cell viability, with a clear threshold above 40%. The primary reasons for low viability were pre-analytical factors intrinsic to the specimens, such as extensive fibrosis (common in PTC), necrosis, and ischemia time post-resection, rather than tumor grade alone. These findings underscore the practical challenges in biobanking and highlight the necessity for optimized, rapid specimen handling protocols in future multi-center studies aimed at modeling all thyroid cancer subtypes.

Several additional technical considerations warrant mention. First, while our RNA-seq analysis provided insightful mechanisms, the sample size for sequencing was limited. Future studies with expanded cohorts are needed to validate the stage-specific gene signatures and NF-κB pathway activity we identified. Second, regarding long-term culture, our microfluidic 3D bioprinting method generates DEOs with high initial cell density, which facilitates rapid maturation (within ~ 10 days). However, this compact structure also imposes physical constraints on growth. Prolonged culture beyond 15–20 days often leads to necrotic cores due to diffusion limitations, a common challenge in organoid systems. Therefore, to maintain optimal cell viability and proliferation, we routinely passaged DEOs at the 12-day point when mature structures were evident. This protocol ensures the retention of healthy, proliferative organoids for downstream assays but limits observations of very long-term (e.g., > 1 month) stability in a single culture cycle. Future optimizations of the matrix composition or incorporation of perfusable microfluidic systems may help overcome this spatial limitation.

Conclusion

Our study establishes a robust and efficient microfluidic-based platform for generating patient-derived engineered organoids (DEOs) that faithfully recapitulate the histopathological, genomic, and immunological features of thyroid cancer. These DEOs not only mirror inter- and intra-tumor heterogeneity and drug response profiles, but also serve as a scalable system for high-throughput drug screening and personalized therapy testing. Furthermore, we developed a novel model for thyroid cancer lung metastasis, providing a valuable tool for investigating metastatic mechanisms. Significantly, we identified NF-κB signaling as a key regulator of organoid maturation and demonstrated that its inhibition synergizes with conventional anticancer agents to enhance efficacy. This work underscores the potential of engineered organoids in advancing precision oncology and offers a translatable framework for mechanistic exploration and therapeutic discovery in thyroid cancer.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

H.G.: Writing – original draft, Methodology, Investigation, Formal Analysis, Conceptualization. J.L.: Methodology, Investigation, Formal Analysis, Data curation, Visualization. X.C.: Resources, Methodology, Investigation. Y.S.: Methodology, Investigation. Y.H.: Methodology, Investigation. Y.Z.: Resources, Data curation, Visualization. J.Z.: Methodology, Investigation. N.X.: Methodology, Investigation, Funding acquisition. X.D.: Writing – review and editing, Methodology, Investigation, Formal Analysis, Data Curation, Conceptualization, Visualization, Supervision, Project administration, Funding acquisition.

Funding

This work was supported by Science and Technology Projects in Guangzhou (No. 2025A03J3474), the Fundamental Research Funds for the Central Universities (No. 21625311), and the Shenzhen Fundamental Research Program (No. JCYJ20240813112004006).

Data availability

The RNA-seq datasets have been deposited in the Gene Expression Omnibus with the accession number GSE296099. The raw FASTQ files in the WES study will be provided for scientific research upon request due to human patient privacy concerns. The datasets generated and/or analyzed during this study are available upon request from the corresponding author.

Declarations

Ethics approval and consent to participate

The Ethics approval of this study was derived from the project of “Mechanism of PUS10-mediated pseudouridylation of pre-mir-152 in papillary thyroid carcinogenesis and progression via the PRR15-PI3K/Akt axis”, which was approved by the Scientific Research Ethics Committee of the First Affiliated Hospital of Jinan University (Ethical Development No. KY-2024-308).

Consent for publication

All authors agreed on the manuscript.

Conflict of interest

Authors declare no conflict of interest.