Novel a new prognostic model of thyroid carcinoma based on macrophage and cuproptosis-related lncRNA

Abstract

Background

Thyroid carcinoma is a type of malignant tumor with a relatively good prognosis, but some types have a poorer prognosis. Therefore, it is necessary to establish a new prognostic model to predict the survival outcomes and immune therapy responses of thyroid cancer patients.

Methods

ScRNA analysis was conducted to identify cuproptosis-related genes and macrophages-related genes. The Wilcoxon algorithm was employed to identify tumor-related genes. The overlapping genes were utilized to identify lncRNAs that are related to both macrophages and cuproptosis. Lasso was used to construct a prognostic model. Based on the model, we conducted survival analysis, mutational profile analysis, and drug sensitivity analysis.

Results

Our prognostic risk model has identified 11 macrophage and cuproptosis-related lncRNAs. It has been confirmed that our prognostic model showed favorable performance in predicting the survival outcomes of patients in the TCGA-THCA cohort, with an AUC value exceeding 0.8, supporting its potential value for prognostic assessment of thyroid carcinoma. In this study, we observed no significant disparity in gene mutation rates between the high-risk and low-risk groups. Patients categorized in the low-risk group exhibit heightened sensitivity to immunotherapy and demonstrate responsiveness to a variety of immunotherapeutic agents, such as dasatinib.

Conclusion

Our study highlights the potential of macrophage and cuproptosis-related lncRNAs as novel predictive biomarkers for thyroid carcinoma.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12935-026-04248-9.

1 Introduction

Thyroid cancer represents a significant global health burden with rising incidence worldwide, particularly among women. The disease is classified into several types, with differentiated thyroid cancer (DTC)—including papillary and follicular subtypes—accounting for the majority of cases. Papillary thyroid carcinoma (PTC), the most common subtype, is characterized by a favorable prognosis and extremely high survival rate, especially in younger patients [1–4]. However, papillary thyroid carcinoma (PTC) is prone to early lymph node metastasis, and lymph node metastasis remains one of the factors contributing to poor prognosis in thyroid carcinoma [5]. The global incidence of thyroid cancer has increased significantly, with estimates reaching up to 821,214 new cases in 2022 [6]. PTC remains the most commonly diagnosed type, comprising approximately 90% of cases. Recent studies report adult incidence rates ranging from 4.5 to 14.4 per 100,000 individuals, with notable geographic variations [7, 8]. Given these varying prognoses across thyroid cancer types; therefore, it is necessary to develop a new prognostic model.

Macrophages are key components of the immune system, originating from monocytes and exhibiting versatile functions that are crucial in various biological processes, including tumor development. In thyroid cancer, tumor-associated macrophages (TAMs) significantly influence tumor behavior and patient prognosis. Although most thyroid cancers are well-differentiated and have favorable outcomes, the presence of TAMs has been associated with poorer prognoses, particularly in anaplastic and medullary types [9]. TAMs, especially those of the M2 phenotype, promote tumor growth and metastasis by secreting immunosuppressive factors and supporting angiogenesis [10]. Recent findings indicate that these macrophages help create a tumor microenvironment conducive to cancer cell survival, thereby impairing immunotherapeutic responses [11]. Notably, targeting the polarization of macrophages from the M2 to the M1 phenotype has emerged as a potential therapeutic strategy to enhance anti-tumor immunity [12]. Thus, further investigation into the mechanisms by which TAMs affect thyroid cancer is essential for developing effective prognostic models and targeted therapies that can improve patient outcomes [13–16].

Cuproptosis is a novel concept in cancer biology, referring to a unique form of regulated cell death induced by excess copper. Long non-coding RNAs (lncRNAs) associated with cuproptosis have been identified as key regulators in various types of cancer, including thyroid cancer [17, 18]. Recent studies suggest that these lncRNAs can modulate tumor growth, metastasis, and therapeutic response [19, 20]. In thyroid cancer, specific cuproptosis-related lncRNAs have shown prognostic potential. For instance, one study identified a lncRNA signature capable of predicting patient outcomes, with high expression levels correlating with poorer survival rates [21]. Additionally, these lncRNAs impact the role of tumor-associated macrophages (TAMs) in the tumor microenvironment by regulating macrophage polarization and enhancing tumor-promoting activities [22, 23]. Thus, cuproptosis lncRNAs hold promise as both prognostic markers and therapeutic targets in managing thyroid cancer, highlighting their relevance in understanding tumor pathogenesis and improving clinical strategies [24].

To investigate the effects of cuproptosis-related genes on macrophages and thereby clarify their impacts on the immune microenvironment, we designed this study. Specifically, this study explores the influences of cuproptosis-associated lncRNAs on macrophages and the immune microenvironment of thyroid cancer, and establishes a novel prognostic model to predict the prognosis of thyroid cancer patients and identify potential therapeutic targets.

Matherials and methods

Single-cell data downloading and processing

The single-cell RNA sequencing dataset GSE191288 was downloaded from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov/gds/). This dataset included 7 samples: 6 tumor samples and 1 normal sample. We then processed the single-cell sequencing data using the “Seurat” package. Quality control was first performed by retaining cells with mitochondrial gene content less than 20% and ribosomal gene content less than 20%, with the number of expressed genes per cell between 200 and 10,000, and by keeping only genes expressed in at least 3 cells. Subsequently, we identified 2000 highly variable genes. Dimensionality reduction was performed using principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE). The “Harmony” package was used to eliminate batch effects among the 7 samples. Cell clusters were identified using the “FindNeighbors” and “FindClusters” functions and visualized via t-SNE. Lastly, we annotated the cell clusters using marker genes from previous studies (Supplementary Table S1). The “ssGSEA” method was used to calculate the activity of gene sets in each cell. The Wilcoxon test (adjusted p < 0.05) was used to identify differentially expressed genes (DEGs), with all other parameters set to default values. We then used the “AUCell” package to calculate the “AUCell Score” for each cell to classify cells into High and Low cuproptosis groups. Finally, the “DESeq2” package was used to identify differentially expressed genes (DEGs) among the 8 cell types, with all parameters set to default values.

TCGA data collection and processing

The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/) includes genomic sequencing data, proteomic data, methylation data, mRNA data, and miRNA data for 33 types of cancer. In this study, we downloaded transcriptome profiling (RNA-seq), single nucleotide data, and corresponding clinical information for TCGA-THCA, including 505 papillary thyroid cancer samples and 59 normal thyroid tissues. Patients with missing clinical data or 0 survival days were excluded. All data in this study were log2 transformed.

.Identification of tumor-related genes

The “Limma” package was used to identify Differentially Expressed Genes (DEGs) between tumor and normal thyroid tissue. We set the threshold at |logFC| > 1 and adjust p < 0.05 to identify the DEGs.

Identification of macrophage-related cuproptosis lncRNAs

Macrophage- and cuproptosis-related DEGs were identified by intersecting High_Cuproptosis_genes, TCGA-DEGs, and macrophage-related genes using the “VENN” package.Subsequently, we performed co-expression analysis of macrophage cuproptosis-related genes and lncRNAs, with the criteria of a correlation coefficient ≥ 0.3 and p < 0.05. Finally, we constructed a sankey diagram and a co-expression network to visualize the relationships between these genes and lncRNAs using the “ggplot2” and “ggalluvial” packages.

Construction of a prognostic model

TCGA-THCA samples were randomly divided into a training set and a internal validation set at a 7:3 ratio using the “caret” package in R. The training set data were used to screen prognostic long non-coding RNAs (lncRNAs) and construct the prognostic model, while the validation set data were used to verify the model’s accuracy. There were no significant differences in clinical characteristics (e.g., age, gender, TNM stage) between the two groups of patients (P > 0.05) (Table 1). Univariate analysis was performed on the training set to screen for prognostic lncRNAs (Supplementary Figure S1). Subsequently, a LASSO regression model was constructed using the “glmnet” package; specifically, the model was set as a Cox proportional hazards model (family = “cox”) to accommodate survival data, with the maximum number of iterations set to 1000 to ensure convergence. A 10-fold cross-validation was used to select ‘lambda.min’ (the lambda value corresponding to the minimum cross-validation error) as the optimal regularization parameter. Finally, patients in the TCGA-THCA cohort were stratified into high-risk and low-risk groups based on the median risk score, and the predictive performance of the model was evaluated using the internal validation set.

Table 1.

clinical characteristics of training and validation cohort

Characteristics	Training cohort	Testing cohort	Total	P-value
Age (years)
≥ 65	49(13.76%)	27(18.24%)	76(15.08%)	0.238
<65	307(86.24%)	121(81.76%)	428(84.92%)
Gender
Female	266(74.71%)	102(68.91%)	368(73.02%)	0.139
Male	90(25.29%)	46(31.09%)	136(26.98%)
Stage
I- II	237(66.57%)	99(66.89%)	336(66.67%)	0.916
II- III	117(32.87%)	49(33.11%)	166(32.94%)
Unknow	2(0.56%)	0(0%)	2(0.39%)
T
1–2	221(62.08%)	87(58.78%)	308(61.11%)
3–4	134(37.64%)	60(40.54%)	194(38.49%)	0.653
X	1(0.28%)	1(0.68%)	2(0.40%)
N
0	171(48.03%)	59(39.86%)	230(45.63%)	0.051
1	155(43.54%)	69(46.62%)	224(44.44%)
x	30(8.43%)	20(13.52%)	50(9.93%)
M
0	195(54.78%)	87(58.78%)	282(55.95%)	0.481
1	6(1.69%)	3(2.03%)	9(1.79%)
X	154(43.26%)	58(39.19%)	212(42.06%)
Unknow	1(0.27%)	0(0%)	1(0.20%)

Assessment of the independence and validity of the prognostic model

To investigate the impact of clinical characteristics on patient prognosis, we performed univariate and multivariate regression analyses of clinical features including risk score, gender, age, and pathological stage. By calculating Overall Survival (OS) probabilities at 1, 3, and 5 years, we constructed a nomogram integrating risk score, age, gender, pathological stage, and other clinical parameters as independent prognostic factors. To evaluate the nomogram’s accuracy, we plotted calibration and ROC curves. To assess the prognostic significance of risk score across different subgroups, we performed stratified analysis according to age, gender, tumor grade, clinical stage, and TNM stage. To explore the relationship between risk score and various clinical characteristics, we conducted comprehensive analyses using the “survival” R package. Specifically, we examined the associations of age, gender, clinical stage, T (tumor), N (node), and M (metastasis) stages with risk score. We performed Kaplan-Meier analysis to identify patterns and differences in survival outcomes among these variables and visualized the results to enhance interpretability and understanding of the potential prognostic value of these clinical factors.

Mutation landscape and survival analysis

The “maftools” package was used to generate genetic mutation features for THCA patients from the TCGA database. Subsequently, we combined the mutation feature files with risk scores. We then used a waterfall plot to visualize the top 30 genes with the highest mutation frequencies. To explore the impact of mutation frequency on survival outcomes in high- and low-risk THCA patient groups, we conducted Kaplan-Meier analysis using the “survival” package.

Differential gene expression analysis between two risk groups

The “limma” package was used to explore the differentially expressed genes (DEGs) between high- and low-risk patients, aiming to investigate new therapeutic targets. Eventually, we visualized the DEGs and their expression levels between the two risk groups using the “ggplot2” and “pheatmap” packages.

KEGG and GO analysis

In order to explore the potential biological pathways between two risk groups, we performed Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) Enrichment analysis. Based on the “org.Hs.eg.db” package online, we identified the Molecular Function (MF), Cellular Component (CC), and Biological Process (BP) between the two groups.

Gene set enrichment analysis and tumor immune microenvironment

Gene Set Enrichment Analysis (GSEA) was performed using the “c2.cp.kegg.Hs.symbols.gmt” gene set from MsigDB to identify potential pathways and molecular mechanisms differentiating the two risk groups. Subsequently, we conducted single-sample GSEA (ssGSEA) on the high- and low-risk groups to characterize immune cells and immune functions in the tumor microenvironment using the GSVA package and the immune.gmt gene set (Supplementary Table S7–S8) obtained from previous studies. To further identify potential immunotherapy targets, we performed an immune checkpoint analysis on both risk groups using known immune checkpoint genes (Supplementary Tables, Table S9) with the limma package. Next, the ESTIMATE package was used to assess immune cell infiltration levels, stromal cell infiltration extent, and tumor purity in the two risk groups. We performed Kaplan-Meier analysis on the ESTIMATE results and visualized the findings. Finally, we employed multiple immune infiltration algorithms and corresponding data from previous studies to evaluate the correlations between immune cells and risk scores.

Tumor immunotherapy response and drug sensitivity analysis

Based on the riskscores of the TCGA cohort, the online website TIDE (http://tide.dfci.harvard.edu/) was used to calculate tumor immune evasion risk scores for the high- and low-risk groups. Meanwhile, the “oncoPredict” package was utilized to detect the IC50 of chemotherapy agents in patients from different risk groups in order to evaluate their sensitivity to various chemotherapy drugs.

Results

The flowchart of this study was illustrated in Fig. 1.

Fig. 1.

The flowchart of this study

Single-cell sequencing analysis

7 Thyriod Cancer (THCA) samples were included in our study. We removed cells with mitochondrial genes and ribosomal genes exceeding 20%, and recalculated the correlation between these two types of genes and total RNA (Fig. 2A-F). Then, we preserved cells with high-variable genes ranging from 200 to 7000. Subsequently, we performed dimensionality reduction on the data after quality control (Fig. 2G). Harmony was used to eliminate batch effects between the 7 samples (Fig. 2H). We set the resolution to 1 and divided all the cells into 29 subclusters (Fig. 2I). Based on the marker genes, all cells were annotated into 8 cell types, including Thyroid follicular cells, T cells, Epithelial cells, Fibroblasts, Endothelial cells, Plasma cells, Macrophages, and B cells (Fig. 2J-K). As shown in Fig. 2L-M, we calculated the Cuproptosis gene(Supplementary Table S2) score in different cell types. By setting the standard at a p-value < 0.05 and min.pct > 0.25, we identified 3584 High_Cuproptosis_Score genes (Supplementary Table S3). Finally, 4789 Macrophage-related genes were identified with the criterion |log2FC| > 0.25 and min.pct > 0.25 (Fig. 2N, Supplementary Table S5).

Fig. 2.

Single-cell RNA sequencing analysis. (A-C) single-cell sequencing data quality control. (D-F) The percentage of nCount RNA, mitochondria RNA and ribosome RNA after quality control.(G) Dimensionality reduction of scRNA data.(H) Dimensionality of the batch effects by Harmony.(I) The result of cell clustering.(J-K) The result of cell annotation. (L-M) The result of Cuproptosis score between different celltypes. (N) The differential expressed genes(DEGs) between celltypes

Construction of the prognostic model associated with Macrophage and Curproptosis-related lncRNA

Wilcoxon rank-sum test was performed to identify differentially expressed genes (DEGs). We identified 1,255 DEGs using the criteria |log2FC| > 1.5 and p-value < 0.05 (Fig. 3A-B, Supplementary Table S4). Macrophage and cuproptosis-related genes were identified by intersecting DEGs, High_Cuproptosis_Score genes, and Macrophage-related genes (Fig. 3C). Using co-expression analysis, we identified 381 lncRNAs co-expressed with these macrophage and cuproptosis-related genes (Fig. 3D-E). The TCGA-THCA cohort was subsequently divided into a training set and an internal validation set in a ratio of 7:3. Univariate Cox analysis was then performed on these lncRNAs in the training set, ultimately identifying 11 prognosis-associated lncRNAs (Supplementary Figure S1). To further filter prognosis-related lncRNAs, Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was performed. Finally, we identified 11 lncRNAs to construct a prognostic model (Fig. 3F-G). The prognostic model can be represented by the formula below:

Fig. 3.

Constructing a prognostic model. (A) Show the volcano plot of DEGs. (B) Show the heatmap of the top 50 DEGs. (C) Intersecting the DEGs, High_Cuproptosis_Score gene and Macrophage genes. (D-E) The co-expression network of Macrophage and cuproptosis-related genes with lncRNAs. (F-G) Lasso regression was performed to construct a prognostic model

Risk score = ∑(Coefi*Expi)= (-0.3970)* AC107214.1 + (6.1614)* AC135050.1 + (2.9477)* AC026355.3 + (1.9100)* FOXP1-AS1 + (-0.5844)* AC123768.1 + (2.5894)* MIR3945HG + (5.1806)* AC022819.1 + (1.5998)* AC009716.1 + (-7.7286)* AC120498.9 + (1.0665)* AL133444.1 + (1.6494)* AL132709.8.

Validation of the prognostic risk model

Based on the median risk score of the prognostic model, the TCGA training set, internal validation set, and total set were stratified into high- and low-risk groups. Kaplan-Meier survival analysis, ROC curve analysis, and risk stratification analysis were performed on these groups. Survival analysis of each set revealed that the high-risk group exhibited poorer prognosis than the low-risk group (Fig. 4A-C). ROC curve analysis of each set showed that the 1-, 3-, and 5-year AUC values in the training set approached 1.0, while those in both the internal validation and total sets exceeded 0.75, indicating that the model reliably distinguished survival outcomes between the risk groups (Fig. 4D-F). Risk curve analysis of the training, validation, and total sets demonstrated that mortality increased significantly as risk scores increased (Fig. 4G-L). Additionally, the heatmap revealed that AC026355.3, AC022819.1, AL133444.1, MIR3945HG, AL132709.8, FOXP1-AS1, AC135050.1, and AC009716.1 were risk lncRNAs, whereas AC120498.9, AC107214.1, and AC123768.1 were protective lncRNAs (Fig. 4M-O).

Fig. 4.

Validation of the Risk Model. (A-C) Kaplan-Meier Curve of the TCGA set. (D-F) ROC Curve of the TCGA set. (G-I) Risk curve of the TCGA set. (J-L) Survival Status Curve of the TCGA set. (M-O) Risk Score Heatmap of the TCGA set

Subsequently, the accuracy of the model’s predictions was validated through Kaplan-Meier analysis of various clinical subgroups, including T/N/M, age (≥ 65 vs. <65 years), sex (female vs. male), and stage. We found that low-risk patients had better survival outcomes across all these subgroups (Fig. 5A–F).

Fig. 5.

K-M curve of diferent clinical characteristic in TCGA-THCA. (A) Kaplan-Meier Curve of T(1–2) and T(3–4). (B) Kaplan-Meier Curve of N0 and N1. (C) Kaplan-Meier Curve of M0 and M1. (D) Kaplan-Meier Curve of Age( > = 65) and Age(< 65). (E) Kaplan-Meier Curve of Male and Female. (F) Kaplan-Meier Curve of Stage(I-II) and Stage(III-IV)

Construction of the prognosis-related nomogram

Based on the risk score and clinical characteristics of the TCGA total sample, we conducted univariate and multivariate analyses and found that age, Metastasis Stage (M), and risk score are independent risk factors for THCA patients (Fig. 6A-B). Through the Nomogram, we found that risk scores, age, gender, clinical staging, T staging, N staging, and M staging can serve as independent factors affecting patient prognosis, and all these factors are risk factors for the prognosis of thyroid carcinoma patients. The survival rates of patients, as obtained from the aforementioned line charts for 1, 3, and 5 years, are all 0.99 (Fig. 6C). Finally, through the calibration curve, it can be observed that the predicted values of this model are close to the actual values, indicating the model’s accuracy (Fig. 6D). In Decision Curve Analysis (DCA), the nomogram model demonstrated greater net benefit than the ‘risk model’ and ‘clinical features’ across the range of threshold probabilities from 0.015 to 1 (Fig. 6E) .We conducted ROC curve analysis on the Nomogram risk scores, model risk scores, and patient clinical characteristics and found that in all three groups of patients, the AUC values for both the Nomogram scores and age were over 0.9, suggesting that the Nomogram risk scores and patient age have good accuracy (Fig. 6F-G).

Fig. 6.

Constructing a Nomogram. (A-B) Univariate and multivariate analysis based on the clinical characterstic. (C) The Nomogram. (D) Calibration curve of the Nomogram. (E) Decision curve analysis for all characteristic. (F-G) The ROC curve of 3 groups based on the clinical characteristic

Tumor mutational burden analysis

Based on the risk score, we conducted gene mutation analysis in the high- and low-risk groups and found that the top five genes with the highest mutation frequencies in both groups were BRAF, NRAS, TG, HRAS, and TTN (Fig. 7A, B). There was no statistically significant difference in the mutation rates of these genes between the two groups, and no obvious correlation between gene mutations and risk status (p > 0.05, Fig. 7C, D). Subsequently, Kaplan-Meier analysis of the high- and low-mutation groups revealed that the high-mutation group had poorer survival outcomes than the low-mutation group (p < 0.001, Fig. 7E). In the low-risk group, there was no significant difference in survival outcomes between patients with high and low mutation rates. In the high-risk group, a high mutation rate was associated with poorer survival outcomes (p < 0.001, Fig. 7F).

Fig. 7.

Tumor Burden Mutational Analysis. (A) The mutant landscape of the low-risk group. (B) The mutant landscape of the high-risk group. (C) A violin plot showing the difference in mutant rate between high- and low-risk group. (D) A corrlation map show the relationship between riskScore and TMB. (E) Kaplan-Meier curve shows the diference OS between the high- and low-TMB groups. (E) Kaplan-Meier curve shows the diference OS after combining the TMB with risk score

Diferential analyis between two risk groups

Furthermore, differential analysis was performed on the two risk groups. We identified 7 down-regulated and 237 up-regulated genes in the high-risk group when we set the standard |log2FC| > 1 and p-value < 0.05 (Supplementary Figure S2A-B). Through GO and KEGG enrichment analysis of the high-risk and low-risk groups, we found that the 244 DEGs were mainly enriched in some potential biological pathways, including extracellular matrix organization, extracellular structure organization, external encapsulating structure organization, Cytoskeleton in muscle cells, Cytokine-cytokine receptor interaction, and the PI3K-Akt signaling pathway (p < 0.05, Supplementary FigureS3A-F).

The landscape of tumor microenvironment

We performed Gene Set Enrichment Analysis (GSEA) on the high- and low-risk groups to identify potential pathways. In both risk groups, GSEA was performed using several diagrams based on the following screening criteria: FDR > 0 and P < 0.05 (Fig. 8A-B). The five most significant functions associated with the high-risk group include “KEGG_CHEMOKINE_SIGNALING_PATHWAY”, “KEGG_CYTOKINE_CYTOKINE_RECEPTOR_INTERACTION”, “KEGG_ECM_RECEPTOR_INTERACTION”, “KEGG_FOCAL_ADHESION”, and “KEGG_HEMATOPOIETIC_CELL_LINEAGE”. In addition, the five most significant functions associated with the low-risk group include “KEGG_HUNTINGTONS_DISEASE”, “KEGG_OXIDATIVE_PHOSPHORYLATION”, “KEGG_PARKINSONS_DISEASE”, “KEGG_RIBOSOME”, and “KEGG_TYPE_II_DIABETES_MELLITUS”.

Fig. 8.

The landscape of tumor. (A) (B) The GSEA result of high- and low-risk groups. (C) The result of ssGSEA. (D) The result of ssGSEA immune function. (E) Immune checkpoint analysis. (F) The result of Tumor Microenvironment. (G)-(I) The result of Kaplan-Meier analysis based on the result of TME. (J) Immune infiltration analysis based on multiple algorithms. (K)-(M) Correlation between Macrophages and risk score

Thereafter, we applied the ssGSEA method to explore the infiltration level of 24 kinds of immune cells and retrieved the scores of 13 immune functions to assess the relationship between the risk model and immune infiltration. The high-risk group had increased macrophages and enhanced all immunological functions (p < 0.05, Fig. 8C-D). Subsequently, we applied immune checkpoint analysis based on 47 immune genes. We found that the high-risk group had higher expression levels of those genes than the low-risk group (p < 0.05, Fig. 8E).

The results of the Tumor Microenvironment (TME) analysis showed that the high-risk group had higher stromal scores, immune scores, and estimation scores, which means that the TME of the high-risk group was more complex and possibly associated with invasiveness, resistance to treatment, and a poorer prognosis (P < 0.05, Fig. 8F). We also found that higher TME scores were associated with worse survival outcomes (P < 0.001, Fig. 8G-I).

Based on the TME analysis findings, we further investigated the tumor immune microenvironment. Using seven types of software, we found that various immune cells had a positive effect on tumor development, among which M0, M1, and M2 macrophages were associated with higher risk scores (p < 0.05, Fig. 8J-M).

The landscape of tumor microenvironment and drug sensitive

To further explore the differences in the immune microenvironment between high- and low-risk groups, we analyzed a variety of common immune markers. Figures 9A-B show that the high-risk group has high TIDE and CAF scores (p < 0.001), suggesting that the high-risk group may exhibit enhanced immune escape mechanisms and reduced efficacy of immunotherapy. However, there was no significant difference in CD8 and CD274 scores between the high- and low-risk groups (Fig. 9C-D, p > 0.05). Concurrently, the high-risk group exhibited higher Dysfunction, Exclusion IFNG, Merck18, and TAM.M2 scores (Fig. 9E-G, I and K, p < 0.05), while the MDSC scores showed no significant difference (Fig. 9H, p > 0.05), suggesting potential immune evasion and reduced effectiveness of immunotherapy.

Fig. 9.

Tumor Burden Mutational Analysis. (A) TIDE (B) CAF (C) CD8 (D) CD274 (E) Dysfunction (F) Exclusion (G) IFNG (H) MDSC (I)Merck18 (J) MSI (K)TAM.M2

IC50 is an indicator that illustrates the tolerance of tumor cells to drugs. We conducted a differential sensitivity analysis of 198 drugs between high and low-risk groups. The results showed that Low-risk group THCA patients are highly sensitive to targeted drugs such as Afuresertib, Elephantin, Leflunomide, Linsitinib, Nilotinib, and Sabutoclax; whereas High-risk group THCA patients are highly sensitive to targeted drugs such as Cediranib, Dasatinib, Entospletinib, Olaparib, Osimertinib, and Saptinib (Fig. 10A-L).

Fig. 10.

The chemotherapeutic response of two risk groups. (A-L)Showed the immunotherapy sensitive drugs between high- and low-risk groups

Discussion

Thyroid cancer (THCA) is a common malignancy typically associated with a favorable prognosis; however, a subset of patients experience aggressive disease with poor clinical outcomes [25]. Recent advancements in single-cell sequencing have revolutionized our understanding of tumor heterogeneity and the dynamics of the tumor microenvironment (TME) in various cancers, including THCA [26, 27]. In this study, we employed comprehensive single-cell sequencing to elucidate the cell composition of THCA, with a particular focus on macrophage and cuproptosis-associated long non-coding RNAs (lncRNAs), and developed a novel prognostic model to predict patient outcomes.

Single-cell sequencing analysis identified eight major cell types within THCA samples, including thyroid follicular cells, T cells, epithelial cells, fibroblasts, endothelial cells, plasma cells, macrophages, and B cells. Macrophages, in particular, have been widely implicated in cancer biology due to their dual roles in tumor progression and inhibition, depending on their phenotypic polarization (M1 vs. M2 macrophages) [28, 29]. The interactions between macrophages and tumor cells within the TME are crucial for tumor growth, metastasis, and response to therapy [23, 30]. Our analysis not only confirmed the presence of macrophages but also highlighted a significant correlation between macrophage gene expression profiles and patient prognosis, identifying 4,789 macrophage-related genes.

Cuproptosis is a recently identified form of cell death linked to copper-induced cytotoxicity and is rapidly becoming a focal point in cancer biology [31]. In this study we calculated cuproptosis scores for previously characterized cell types and found them to be tightly associated with specific long non-coding RNAs (lncRNAs). Through rigorous statistical analyses—Wilcoxon tests and LASSO regression—we selected 11 lncRNAs that are co-expressed with macrophage- and cuproptosis-related genes and used them to build a robust prognostic risk model. As shown in the risk equation, three lncRNAs carried negative coefficients and were associated with favorable prognosis, whereas the remaining eight, with positive coefficients, predicted poor outcome. Kaplan–Meier survival and ROC curve analyses demonstrated that the model accurately stratified patients into high- and low-risk groups: the AUC approached 1.0 in the training set and exceeded 0.75 in all validation cohorts, indicating excellent discriminatory power [32, 33]. Subgroup Kaplan–Meier analyses (age, sex, overall stage, T, N, M) revealed that high-risk patients consistently exhibited significantly shorter overall survival (OS) than low-risk patients in every stratum. Nomograms translate complex statistical models into intuitive visual tools that allow clinicians to rapidly estimate patient risk and guide clinical decisions; they have previously been reported for thyroid cancer and lymph-node metastasis [34, 35]. To facilitate clinical translation of our risk signature, we integrated the 11-lncRNA risk score with readily available clinical variables (age, sex, anatomic TNM stage) to construct a pragmatic prognostic nomogram. The resulting chart is simple, user-friendly, and requires no specialized bioinformatics platforms; clinicians can promptly and accurately estimate 1-, 3-, and 5-year overall survival probabilities. In routine practice the workflow is straightforward: the clinician records the patient’s clinical parameters (e.g., age ≥ 65 or < 65, TNM stage, sex) and calculates the 11-lncRNA risk score; locates the corresponding points for each variable on the nomogram; sums these points to obtain a total score; and finally reads the predicted survival probability directly from the survival axis. Univariate and multivariate analyses confirmed that the risk score, age, and M stage are independent prognostic factors, underscoring their value for individualized patient management. Calibration curves showed close agreement between nomogram-predicted and observed survival, further validating the model’s accuracy [36, 37].

Tumor mutational burden (TMB) analysis did not reveal significant mutations correlated with risk stratification, although genes such as BRAF and NRAS, well-documented in thyroid cancer pathogenesis, were identified [38, 39]. These results suggest that while TMB may not directly impact risk classification per se, it drives tumor aggressiveness, as evidenced by poorer survival in patients with high TMB.Differential expression analyses revealed upregulation of 237 genes in high-risk groups, implicating pathways related to extracellular matrix organization and cytokine-cytokine receptor interactions, involved in tumor invasiveness and aggression. These findings resonate with existing literature underscoring the roles of these pathways in numerous cancers [40–42]. Gene Set Enrichment Analysis (GSEA) validated these observations, pinpointing significant pathways—such as chemokine signaling and ECM-receptor interaction—that reflect more pronounced oncogenic signaling in high-risk groups [43, 44].The TME landscape further emphasized that high-risk patients had elevated stromal and immune scores, suggesting a more complex and potentially treatment-resistant tumor milieu. Increased expression of immune checkpoint-related genes and enhanced macrophage infiltration in these patients support hypotheses of immune evasion and necessitate novel therapeutic strategies, such as macrophage-targeted therapies or immune checkpoint inhibitors, as potential interventions [9, 45, 46].The landscape of the tumor microenvironment in THCA, as revealed by GSEA and ssGSEA, highlights the enrichment of immune-related pathways and the infiltration level of immune cells in high-risk groups. The association between high-risk scores and increased immune cell infiltration, particularly macrophages, suggests a role for these cells in THCA progression and immune evasion. Macrophages in the TME were not only abundant in high-risk patients but also displayed enhanced immunological functions, as shown by our single-sample gene set enrichment analysis (ssGSEA). This is in line with existing literature suggesting that tumor-associated macrophages (TAMs) are critical for tumor immune evasion and progression. Recent studies have shown that targeting TAMs or modulating their phenotype could significantly impact tumor progression and patient outcomes [47, 48].

The differential sensitivity of THCA to various therapeutics, as revealed by IC50 analyses, offers valuable insights for personalized medicine. High-risk groups showed sensitivity to drugs like Afuresertib and Nilotinib, aligning with markers of therapeutic susceptibility observed in other cancers [49, 50]. Conversely, low-risk groups exhibited responsiveness to agents such as Osimertinib and Olaparib, known to have efficacy against genomic-driven cancers [51, 52]. The delineation of chemotherapy sensitivity profiles could greatly enhance treatment strategies, tailoring regimens based on individual genetic and molecular landscapes.

Despite these promising findings, our study has limitations. The sample size is limited, necessitating further validation in larger, more diverse cohorts. Additionally, functional studies are required to elucidate the precise roles of identified lncRNAs in THCA pathogenesis. Experimental validation of therapeutic targets suggested by our analyses is crucial for transitioning from theoretical models to clinical applications.

Conclusion

In conclusion, this study conducted a comprehensive analysis of THCA at the single-cell level, providing novel insights into uncovering the characteristics of THCA cellular heterogeneity, screening prognostic biomarkers, and identifying therapeutic targets. The integration of these research findings with clinical data lays a solid foundation for developing personalized treatment strategies for THCA and improving patient outcomes. Although our model demonstrates favorable predictive performance in both the training set and validation set, we also acknowledge the limitations of this study: the prognostic model was constructed and validated only based on the single TCGA-THCA dataset, which may pose a risk of overfitting and thereby limit the model’s generalizability. Future research will focus on further validating this model through multi-center external cohorts.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CAF

Cancer-Associated Fibroblast

C-index

Concordance index

DCA

Decision Curve Analysis

DEGs

Differential Expressed Genes

DEGs

Differentially Expressed Genes

GEO

Gene Expression Omnibus

GSEA

Gene set enrichment analysis

IC50

Half Maximal Inhibitory Concentration

IFNG

Interferon-Gamma

KEGG

Kyoto Encyclopedia of Genes and Genomes

KM

Kaplan Meier

LASSO

Least Absolute Shrinkage and Selection Operator

LncRNA

Long Non-coding RNA

MDSC

Myeloid-Derived Suppressor Cell

MSI

Microsatellite Instability

OS

Overall survival

ROC

Receiver Operating Characteristic

scRNA

Single cell sequencing

ssGSEA

Single-sample Gene set enrichment analysis

TAM

Tumor-Associated Macrophage

TCGA

The Cancer Genome Atlas Program

THCA

Thyroid Carcinoma

TIDE

Tumor Immune Dysfunction and Exclusion

TME

Tumor Microenvironment

Author contributions

JW and FY designed and conducted this study, performed the data processing and wrote the initial manuscript. YX downloaded the raw data online and preprocessed it. GH performed data supervision and writing-review. All of the authors have read and approved the final manuscript.

Funding

No funding for this study.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not available.

Consent for publication

All authors agree to publish.

Competing interests

The authors declare no competing interests.