Identification and validation of tumor microenvironment remodeling markers associated with prognosis in differentiated thyroid cancer

Introduction

Differentiated thyroid cancer (DTC) is a common endocrine malignancy and is the most prevalent cancer of thyroid (1-3). Its etiology involves a multifactorial interplay of genetic mutations (e.g., BRAF, RAS), growth factors, iodine intake, ionizing radiation, sex, and hereditary factors (4). Although DTC usually exhibits indolent behavior and favorable outcomes with timely diagnosis and treatment, pronounced inter- and intra-tumoral heterogeneity continues to limit the precision of current diagnosis, monitoring, and risk-stratification strategies.

Growing evidence highlights the importance of tumor microenvironment (TME) in the initiation and progression of tumors (5). Comprising immune and stromal cells, vasculature, extracellular matrix (ECM), and diverse signaling molecules, the TME undergoes remodeling through interactions between malignant and non-malignant components, shaping tumor biology (6,7). Genetic alterations (8) and TME-related factors can promote lipid uptake and de novo lipid synthesis in tumor cells (9-11), while heterogeneity in immune cell infiltration and metabolic rewiring can modulate anti- and pro-tumor responses (12-15). In our previous work, Mendelian randomization analyses linked DTC to lipid-related traits such as total cholesterol, high-density lipoprotein (HDL) cholesterol, and apolipoprotein B, and identified 19 lipid metabolism-related genes (16), suggesting that the TME of DTC, particularly the lipid metabolic milieu, undergoes reprogramming, potentially resulting in distinct lipid metabolic profiles in affected patients.

Despite these advances, integrative analyses of TME-related genes in DTC remain scarce. Most existing prognostic models focus on a single dimension of the TME, such as immune infiltration (17,18), discrete metabolic pathways (19,20), or specific forms of programmed cell death (PCD) (21,22), and thus overlook the interplay among these processes. To address this gap, we systematically characterized TME remodeling in DTC by analyzing differentially expressed genes (DEGs) across molecular subclusters with distinct profiles of immune infiltration, immune modulator expression, lipid metabolic activity, and PCD regulation. This integrative framework captures the multifaceted nature of the TME and enables more accurate, biologically grounded predictions of patient outcomes. Within this context, we identified key TME-related genes involved in DTC remodeling and elucidated their functional roles in tumor progression. Our findings are expected to provide novel theoretical insights and a foundation for both clinical and mechanistic investigations into DTC pathogenesis. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0208/rc).

Methods

Ethics statement

Seven pairs of tumor tissue samples and their corresponding adjacent non-cancerous tissue samples were obtained from patients diagnosed with DTC at the General Surgery Department of Lanzhou University Second Hospital between October 2024 and December 2024. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Lanzhou University Second Hospital (No. 2025A-217). All participants provided written informed consent prior to inclusion in the study.

Data source

RNA sequencing data and clinical information for thyroid cancer samples were obtained from the UCSC Xena database (http://xena.ucsc.edu/). Samples that were unclassified, poorly differentiated, or lacked survival data were excluded. Ultimately, 511 DTC samples and 59 adjacent non-cancerous tissues were retained as controls. The Cancer Genome Atlas (TCGA)-DTC cohort was randomly divided into training and validation sets at a 7:3 ratio, ensuring strict independence between model construction and performance evaluation.

Nineteen genes of interest (LCAT, CETP, PLTP, LPL, LIPC, NR1H3, GSTM1, GSTM4, GSTM3, HECTD4, TRAFD1, ACAD10, NAA25, TMEM116, ALDH2, PPM1G, EIF2B4, NRBP1, and KRTCAP3) were selected based on our previous work (16). Gene sets corresponding to 13 PCD modalities, including alkaliptosis, apoptosis, autophagy, cuproptosis, disulfidptosis, entotic cell death, ferroptosis, lysosome-dependent cell death, necroptosis, netotic cell death, oxeiptosis, parthanatos, and pyroptosis, were obtained from a prior study (23). Additionally, 14 lipid metabolism-related gene sets were retrieved from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://www.kegg.jp/), encompassing pathways such as alpha-linolenic acid metabolism, arachidonic acid metabolism, biosynthesis of unsaturated fatty acids, ether lipid metabolism, fatty acid biosynthesis, degradation, and elongation, as well as glycerolipid, glycerophospholipid, and linoleic acid metabolism, primary bile acid biosynthesis, sphingolipid metabolism, steroid biosynthesis, and steroid hormone biosynthesis.

Identification of genes related to TME remodeling (TER)

We initially performed differential expression and Kaplan-Meier survival analyses on the 19 selected genes. Genes with significantly altered expression between DTC and control samples, and with distinct survival outcomes between high- and low-expression groups, were subjected to consensus clustering analysis. Clustering was performed using the ConsensusClusterPlus R package (v1.70.0), based on a median-centered expression matrix (24). The algorithm employed partitioning around medoids (PAM), with Euclidean distance as the similarity metric and complete linkage for agglomeration. Clustering was iterated 1,000 times (reps =1,000), each using 80% of the samples (pItem =0.8) and all features (pFeature =1). The optimal number of clusters was determined to be k=3, based on two criteria: (I) the consensus matrix heatmap at k=3 exhibited the most distinct and stable clustering boundaries; and (II) the largest relative increase in the cumulative distribution function (CDF) area was observed at k=3, indicating robust clustering stability.

Cluster-based comparison of the TME and clinical traits

To evaluate the TME across different clusters, we performed the following analyses: (I) calculated the scores of 13 PCD pathways and 14 lipid metabolism pathways per cluster; (II) assessed the infiltration levels of 28 immune cell types—including activated B cells, activated CD8⁺ and CD4⁺ T cells, activated dendritic cells, CD56bright and CD56dim natural killer (NK) cells, memory B cells, central and effector memory CD4⁺ and CD8⁺ T cells, γδ T cells, immature B and dendritic cells, macrophages, mast cells, myeloid-derived suppressor cells (MDSCs), monocytes, NK T cells, neutrophils, regulatory T cells (Tregs), plasmacytoid dendritic cells, T follicular helper cells, eosinophils, and type 1/2/17 T helper cells—using the single-sample gene set enrichment analysis (ssGSEA) algorithm (25); and (III) analyzed differential expression of immune activators and suppressors (26). Clinical traits were also compared among clusters.

Construction and validation of the TER-related risk model

DEGs between DTC subclusters were identified using adjusted P value <0.05 and |log₂fold change| >1, and defined as TER-related genes (TERRGs). Functional enrichment of TERRGs was conducted via ClueGO (27). Subsequently, TERRGs were subjected to univariate Cox regression to identify those associated with DTC patient survival. Based on the median expression level of each prognostic TERRG, DTC samples were stratified into high- and low-expression groups, and Kaplan-Meier survival analyses were performed to compare overall survival. TERRGs with statistically significant survival differences were retained as candidate prognostic genes for subsequent machine learning (ML) analyses. In this study, 10 ML algorithms, including least absolute shrinkage and selection operator (Lasso), Ridge, elastic net (Enet), stepwise Cox regression (StepCox), survival support vector machine (survivalSVM), CoxBoost, supervised principal components for survival analysis (SuperPC), partial least squares regression for Cox model (plsRcox), random survival forest (RSF), and gradient boosting machine (GBM), were integrated, yielding a total of 101 algorithm combinations for predictive model construction. To maximize data utilization and enhance model stability, 10-fold cross-validation was employed during the training process. The concordance index (C-index) of each model combination was evaluated on training, validation, and entire TCGA-DTC cohorts, and the results were visualized using a heatmap. Gene features from the top two models were intersected to identify key TERRGs associated with the TME of DTC. Correlations among these prognostic TERRGs were evaluated using Spearman’s correlation coefficient. Based on the expression profiles of key TERRGs, a TER score was calculated for each sample. Patients were then categorized into high- and low-TER score groups, and survival differences were visualized using Kaplan-Meier curves. To explore functional pathways associated with TER, gene set variation analysis (GSVA) was conducted using the c2.cp.kegg.v7.4.symbols.gmt reference gene set (28). Clinical characteristics (including age, sex, overall stage, T stage, N stage, and M stage) were compared between the two TER score groups.

Construction of the prognostic predictive model for DTC

Univariate and multivariate Cox regression analyses were conducted to identify independent prognostic factors among the TER score and clinical variables. A nomogram incorporating these independent predictors was developed to estimate DTC prognosis. The accuracy and clinical utility of the nomogram were evaluated using calibration plots, receiver operating characteristic (ROC) curves, and decision curve analysis (DCA).

Detection and validation of key prognostic TERRG expression

Key TERRGs were queried in the Human Protein Atlas (HPA) database (http://www.proteinatlas.org) to assess protein expression differences between DTC and control tissues. Additionally, total RNA was extracted using the TRIeasy™ Total RNA Extraction Reagent and reverse transcribed with HyperScript™ RT SuperMix with gDNA Remover. Quantitative polymerase chain reaction (qPCR) was performed using SYBR Green and gene-specific primers (Table 1), with glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as the internal control. Relative expression levels of key TERRGs were calculated accordingly.

Statistical analysis

All statistical analyses were conducted in R (v4.3.1). Survival differences were assessed using the log-rank test. Comparisons between two groups were conducted using the Wilcoxon rank-sum test or chi-square test, with P<0.05 considered statistically significant.

Results

DTC patients were divided into three clusters

In our previous study, we identified 19 genes involved in DTC development (16). To investigate their functional relevance, we first examined their expression levels in DTC versus control tissues. We found that EIF2B4, KRTCAP3, LPL, NRBP1, and PPM1G were significantly upregulated (Figure 1A). Kaplan-Meier survival analysis showed that DTC patients with high expression of TRAFD1 and KRTCAP3, and low expression of HECTD4, TMEM116, and ALDH2, had significantly better overall survival (Figure 1B). These five genes were thus selected for downstream analyses. Using consensus clustering based on the expression of these five prognostically relevant genes, we stratified DTC patients into three clusters (clusters 1, 2, and 3; Figure 1C-1E). Survival analysis revealed that cluster 1 had the poorest prognosis, while cluster 3 exhibited the most favorable outcome (Figure 1F).

Figure 1 Effect of the genes on prognosis of DTC patients and the consensus clustering analysis of DTC patients. (A) The expression levels of five genes that are significantly highly expressed in DTC patients. (B) Kaplan-Meier survival analysis of five genes significantly associated with survival. (C) Consensus clustering matrix for k=3 in DTC patients. (D) Consensus clustering CDF for k=2 to k=5. (E) Relative change in area under the CDF curve according to various k values. (F) Kaplan-Meier curves of survival for the DTC patients in different subtypes. ***, P<0.001; ****, P<0.0001; ns, not significant (P>0.05). CDF, cumulative distribution function; DTC, differentiated thyroid cancer.

TME varied across different clusters

Understanding the heterogeneity of the TME within DTC is critical to elucidating survival differences among patient clusters. We first identified 14 lipid metabolism pathways that significantly differed between clusters. Kaplan-Meier analysis further revealed that seven of these pathways (fatty acid degradation, ether lipid metabolism, arachidonic acid metabolism, α-linolenic acid metabolism, steroid biosynthesis, primary bile acid biosynthesis, and glycerolipid metabolism) were significantly associated with patient prognosis (Figure 2A). Random forest analysis ranked fatty acid degradation, ether lipid metabolism, and arachidonic acid metabolism as the top three pathways most strongly linked to clinical outcomes (Figure 2B). Second, significant differences were observed in the abundance of 25 immune cell types across clusters. Notably, cluster 1 exhibited the lowest proportions of activated dendritic cells, macrophages, MDSCs, NK cells, and type 17/2 T helper cells (Figure S1). Consistently, the expression of immune activators such as CD70, MICB, RAET1E, and TMIGD2 was also lowest in cluster 1 (Figure S2). These findings underscore the distinct immunological landscapes of DTC clusters. Given the influence of the TME on cellular behavior, we next investigated PCDs across clusters. Thirteen PCDs exhibited differential enrichment, with cuproptosis most prominent in cluster 1 and autophagy in cluster 3 (Figure S3). These results suggest a close association between the TME and PCDs, with potentially divergent regulatory effects on distinct PCD modalities.

Figure 2 The importance of differentially enriched lipid metabolism pathways between DTC subclusters in the survival. (A) The Kaplan-Meier curves showed that seven pathways were associated with DTC’s survival. (B) The importance ranking of seven pathways by random forest. DTC, differentiated thyroid cancer.

Twelve survival-related TERRGs were identified in DTC

The distinct TMEs among clusters also provided a basis to explore TERRGs in DTC. We identified 336 differentially expressed TERRGs between clusters 1 and 3 (Figure 3A). Functional enrichment analysis indicated these genes were involved in key biological processes, including cellular responses to zinc ions, triglyceride catabolism, cell junction assembly, angiogenic endothelial proliferation, cell fate commitment, and organic hydroxy compound transport (Figure 3B). KEGG pathway analysis further showed enrichment in pathways such as metallothionein-metal binding, thyroxine biosynthesis, neuronal signaling, neuroactive ligand-receptor interactions, NODAL signaling regulation, and lipase complex assembly (Figure 3C). Univariate Cox regression identified 45 TERRGs significantly associated with DTC survival [P<0.05, hazard ratio (HR) ≠1; Figure 3D]. Kaplan-Meier analysis refined this list to 12 survival-related TERRGs: PLCD4, LMX1B, DDIT4L, CALN1, CAMP, NAT16, TENM2, LEFTY2, ISL1, GATA5, WFDC6, and VGLL2 (Figure 3E).

Figure 3 Identification of 12 survival-related TERRGs in DTC. (A) Volcano plot of the DEGs in cluster 1 and cluster 3. (B) Biological processes enrichment analyses of DEGs. (C) KEGG enrichment analyses of DEGs. (D) Univariate Cox screening for differential genes associated with prognosis. (E) Kaplan-Meier curve analysis of 12 genes independently associated with survival. CI, confidence interval; DEG, differentially expressed gene; DTC, differentiated thyroid cancer; KEGG, Kyoto Encyclopedia of Genes and Genomes; TERRG, tumor microenvironment remodeling-related gene.

A TER risk model for predicting the prognosis of DTC was constructed and validated

To construct a prognostic model, we applied multiple ML algorithms and excluded models containing fewer than two variables, among which RSF performed best, followed by Ridge regression (Figure 4A). Both models were selected for validation. In both, patients with low TERRG-based risk scores exhibited significantly better survival (Figure 4B,4C), and each model demonstrated strong prognostic accuracy (Figure 4D,4E). For instance, the 3-year area under the curve (AUC) for the RSF model was 0.96 in the training set, 0.89 in the validation set, and 0.93 across the entire DTC cohort. We identified five key TERRGs (CAMP, DDIT4L, LMX1B, NAT16, and CALN1) by intersecting the RSF and Ridge model outputs (Figure 5A). Based on these genes, a TER index was calculated to stratify patients into low- and high-index groups. GSVA revealed that tryptophan metabolism, primary bile acid biosynthesis, and arginine and proline metabolism were enriched in the high TER index group, whereas tight junction, adherens junction, and mammalian circadian rhythm pathways were enriched in the low-index group (Figure 5B). Univariate and multivariate Cox analyses confirmed that TER index, age, and tumor stage were independent prognostic factors (Figure 5C,5D). A nomogram incorporating these variables was developed to predict patient survival (Figure 5E). Calibration and ROC curves demonstrated excellent predictive performance, with AUCs of 0.96, 0.99, and 1.00 for 1-, 2-, and 3-year survival, respectively (Figure 5F-5G), noting that the AUC of 1.00 at 3 years may be partially attributable to the limited sample size and number of events. However, DCA indicated that the nomogram provided higher net benefit across a wide range of threshold probabilities compared to individual clinical variables (TER index, age, or tumor stage alone), supporting its potential clinical utility (Figure 5H).

Figure 4 Construction of TER risk model for predicting the prognosis of DTC. (A) The optimal model was fitted using the LOOCV. (B,C) Kaplan-Meier survival curve analysis of RSF and Ridge models. (D,E) ROC curve analysis of RSF and Ridge models. AUC, area under the curve; C-index, concordance index; DTC, differentiated thyroid cancer; Enet, elastic net; GBM, gradient boosting machine; Lasso, least absolute shrinkage and selection operator; LOOCV, leave-one-out cross-validation; plsRcox, partial least squares regression for Cox model; ROC, receiver operating characteristic; RSF, random survival forest; StepCox, stepwise Cox regression; SuperPC, supervised principal components for survival analysis; survivalSVM, survival support vector machine; TER, tumor microenvironment remodeling.

Figure 5 Evaluation and validation of TER risk model. (A) Venn diagram of prognostic genes. (B) GSVA enrichment analysis in low- and high-index groups. (C,D) The univariate and multivariate Cox analyses. (E) A nomogram construction. (F) Calibration curves of the nomogram. (G) ROC curve of the 1-, 2-, and 3-year survival. (H) The DCA curves of the nomogram. AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; GSVA, gene set variation analysis; M, metastasis; N, node; ROC, receiver operating characteristic; RSF, random survival forest; T, tumor; TER, tumor microenvironment remodeling.

Expression of key TERRGs was detected and validated in DTC

We first examined the expression patterns of key TERRGs and found that CALN1 was significantly upregulated in tumor tissues compared to controls, whereas the other four TERRGs were downregulated (Figure 6A). Moreover, positive correlations were observed among these TERRGs; for example, DDIT4L showed a moderate correlation with LMX1B (correlation coefficient =0.58; Figure 6B). We further assessed their protein expression levels and found that CALN1 protein was elevated, while NAT16 protein levels were reduced in tumor tissues (Figure 6C), consistent with their transcriptional trends. Finally, we validated these findings using clinical samples from our own cohort. The expression levels of CAMP, DDIT4L, LMX1B, and NAT16 were markedly decreased, while CALN1 was significantly increased in DTC samples (Figure 6D), in alignment with observations from the TCGA-DTC cohort.

Figure 6 The expression level of key TERRGs in DTC. (A) The expression level of prognostic genes. (B) Spearman correlation analysis of prognostic genes. (C) The protein expression levels analysis of prognostic genes by the HPA database. From left to right, the HPA images depict the expression of CALN1 in thyroid cancer, CALN1 in normal thyroid tissue, NAT16 in thyroid cancer, and NAT16 in normal thyroid tissue, respectively, sourced from the following websites: (https://www.proteinatlas.org/ENSG00000183166-CALN1/tissue/thyroid+gland#img), (https://www.proteinatlas.org/ENSG00000183166-CALN1/cancer/thyroid+cancer#img), (https://www.proteinatlas.org/ENSG00000167011-NAT16/tissue/thyroid+gland#img), and (https://www.proteinatlas.org/ENSG00000167011-NAT16/cancer/thyroid+cancer#img). (D) The verification of prognostic genes by qRT-PCR. *, P<0.05; **, P<0.01; ***, P<0.001; ***, P<0.0001. DTC, differentiated thyroid cancer; HPA, Human Protein Atlas; qRT-PCR, quantitative real-time polymerase chain reaction; TERRG, tumor microenvironment remodeling-related gene.

Discussion

Current prognostic assessment in DTC still relies heavily on histopathological and imaging-based approaches, which often lack the precision needed to capture the underlying biological heterogeneity. Recognizing that the development of DTC may be determined by the dynamic TME (integrating lipid metabolic rewriting, immune modulation, and PCD), we systematically dissected these dimensions and developed an integrated prognostic model with high predictive accuracy.

From 19 candidate genes identified in our previous work, we derived five survival-associated markers that stratified DTC into three molecular subtypes with distinct prognosis. These subtypes displayed marked differences in lipid metabolism, immune infiltration, and PCD activity. Among metabolic pathways, alterations in arachidonic acid metabolism, fatty acid biosynthesis and degradation, glycolipid metabolism, primary bile acid biosynthesis, and steroid hormone biosynthesis were prominent. Metabolic reprogramming is now regarded as possibly being a characteristic manifestation of cancer (29), enabling tumor cells to adapt to nutrient-deprived and hypoxic conditions through enhanced glycolysis and lipid remodeling (30,31). Such lipid metabolic shifts can actively reshape the microenvironment by modulating immune and stromal cell recruitment (32,33). Our findings support the concept we speculate that lipid metabolism not only fuels tumor growth but also influences immune contexture, ultimately affecting patient outcome.

The immune landscape differed sharply between subtypes. Cluster 1, with the poorest prognosis, exhibited depletion of activated dendritic cells, macrophages, and NK cells, which are key mediators of antigen presentation and cytotoxicity (34-36). We speculate that this profile is consistent with an immunosuppressive, ‘cold’ TME, potentially amendable to immune-enhancing interventions such as dendritic cell vaccines, NK cell-based therapies, or macrophage reprogramming. Prior studies have shown that thyroid tumors can form localized immune niches (37,38), with immune composition closely linked to aggressive and differentiation status (39-43). Mechanisms of immune evasion, such as elevated immunosuppressive cytokines (e.g., IL-10, TGF-β), reduced antigen-presenting cell infiltration, and tumor-intrinsic metabolic adaptations, may further limit therapeutic response (44-46). Integrative TME profiling, as demonstrated here, could refine patient selection for immunotherapy in DTC.

PCD profiling revealed subtype-specific enrichment patterns across 13 types, linking cell death regulation to TME states. PCD, including apoptosis, necroptosis, pyroptosis, ferroptosis, anoikis, and autophagy, modulated immune effector recruitment and function (47-49). The interplay between death pathways and immune contexture may partly explain subtype-specific prognosis and offers opportunities for therapeutic targeting.

By intersecting outputs from 101 ML algorithms, we identified five key genes, including CAMP, DDIT4L, LMX1B, NAT16, and CALN1, with strong prognostic value. Chen et al. reported that low expression of CAMP was associated with poor histological differentiation and lymph node metastasis in gastric cancer, suggesting a role in tumor progression (50). In oral squamous cell carcinoma, human CAMP/LL-37 may exert anticancer effects, with DNA methylation potentially contributing to tumorigenesis by downregulating CAMP promoter activity. A previous study reported that knockdown of CAMP could promote the proliferation and inhibit the apoptosis of DTC cells (51), indicating its important role in DTC. In addition, the expression of CAMP was regulated during macrophage M1/M2 polarization, and was mainly expressed in M2 phenotype in breast cancer (52). Lee et al. found that CAMP could promote type 17 T helper cells to express CD73, thus leading to an immune-suppressing environment for tumors (53). In prostate cancer, CAMP was found to promote the differentiation and polarization of immature myeloid progenitors to pro-tumorigenic macrophages (54). These results indicate the interaction between CAMP and TME. Michalski et al. demonstrated that DDIT4L regulates mitochondrial function and may suppress innate immune activity in bone marrow cells during development (55). Koga et al. further confirmed that DDIT4L promoter methylation increases moderately in early-stage and significantly in advanced melanoma, as determined through bisulfite sequencing and real-time fluorescence qPCR (56,57). Ozdemir Kutbay et al. found a marked reduction in DDIT4L expression in undifferentiated thyroid carcinoma, supporting its role as a tumor suppressor (58). During development, Michalski et al. reported that DDIT4L may inhibit the innate immune activity of myeloid cells (55), suggesting a possible immunomodulatory role for DDIT4L in myeloid-lineage cells. Meng et al. identified that LMX1B binds to the GFRA1 promoter and regulates circGFRA1 expression in prostate cancer cells. In turn, circGFRA1 enhances HECT1 expression by sponging miR-3064-5p (59). Lu et al. found that miR-206b may promote the progression of PTC by targeting LMX1B by bioinformatic analyses (60). Bae et al. developed a predictive model for gastric cancer based on ¹⁸F-fluorodeoxyglucose (FDG) uptake by NAT16, and found significant correlations between glucose uptake, tumor mutational burden, and genomic alterations using positron emission tomography (PET)-derived scores in the TCGA dataset (61). Takagi et al. identified CALN1 as a prognostic biomarker for bladder cancer, with CALN1 hypomethylation significantly associated with higher tumor stage, poorer histological grade, and increased risk of recurrence (62). However, the role of DDIT4L, LMX1B, NAT16, and CALN1 in thyroid cancer and TME remains poorly characterized. Our enrichment analysis suggests that these genes were involved in the pathways including metallothionein-metal binding, thyroxine biosynthesis, neuroactive ligand-receptor interactions, NODAL signaling, and the assembly of LPL and LIPC lipase complexes, offering mechanistic hypothesis for future validation.

Collectively, our study delineates three biologically distinct DTC subtypes with distinct profiles of immune infiltration, lipid metabolism, and PCD, and provides a ML-driven prognostic model with superior predictive power. These insights not only improve risk stratification but also suggest therapeutic avenues, such as metabolic inhibitors for lipid-rewired tumors, immune activation for cold TME, and pathway-specific interventions for cell death modulation. While our findings provide a promising framework, further validation is required. The limited sample size of the TCGA dataset introduces potential statistical limitations. Although we confirmed gene expression patterns, we further analyzed the association between key gene expression and patient prognosis using Kaplan-Meier curves and log-rank tests, and evaluated the predictive performance of the model in an independent cohort. Moreover, the biological functions of DDIT4L, LMX1B, NAT16, and CALN1 in thyroid cancer and the TME remain unexplored, and our current analysis does not establish causal mechanisms. Although CAMP has been reported in immune regulation, its precise role in DTC remains to be elucidated. Thus, in vivo and in vitro studies are planned to investigate how these genes functionally contribute to TME reprogramming and DTC progression, which may help uncover novel regulatory circuits and therapeutic targets.

Conclusions

This study systematically characterized the molecular landscape and biological heterogeneity of DTC, identified five key prognostic markers, and developed a predictive model with potential clinical utility. These findings lay the foundation for further mechanistic studies and precision oncology approaches in DTC.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0208/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0208/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0208/prf

Funding: This study was supported by the Natural Science Fund Project of Gansu Province under Grants (No. 21JR1RA133) and the Cuiying Scientific and Technological Innovation Program of Lanzhou University Second Hospital under Grants (No. CY2023-BJ-16).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0208/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Lanzhou University Second Hospital (No. 2025A-217). All participants provided written informed consent prior to inclusion in the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Haugen BR, Alexander EK, Bible KC, et al. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid 2016;26:1-133. [Crossref] [PubMed]
2. Baloch ZW, Asa SL, Barletta JA, et al. Overview of the 2022 WHO Classification of Thyroid Neoplasms. Endocr Pathol 2022;33:27-63. [Crossref] [PubMed]
3. Ringel MD, Sosa JA, Baloch Z, et al. 2025 American Thyroid Association Management Guidelines for Adult Patients with Differentiated Thyroid Cancer. Thyroid 2025;35:841-985. [Crossref] [PubMed]
4. Houten PV, Netea-Maier RT, Smit JW. Differentiated thyroid carcinoma: An update. Best Pract Res Clin Endocrinol Metab 2023;37:101687. [Crossref] [PubMed]
5. Arneth B. Tumor Microenvironment. Medicina (Kaunas) 2019;56:15. [Crossref] [PubMed]
6. Spill F, Reynolds DS, Kamm RD, et al. Impact of the physical microenvironment on tumor progression and metastasis. Curr Opin Biotechnol 2016;40:41-8. [Crossref] [PubMed]
7. Del Prete A, Schioppa T, Tiberio L, et al. Leukocyte trafficking in tumor microenvironment. Curr Opin Pharmacol 2017;35:40-7. [Crossref] [PubMed]
8. Chen Y, Huang W, Ouyang J, et al. Identification of Anoikis-Related Subgroups and Prognosis Model in Liver Hepatocellular Carcinoma. Int J Mol Sci 2023;24:2862. [Crossref] [PubMed]
9. Zhao J, Zhi Z, Wang C, et al. Exogenous lipids promote the growth of breast cancer cells via CD36. Oncol Rep 2017;38:2105-15. [Crossref] [PubMed]
10. Röhrig F, Schulze A. The multifaceted roles of fatty acid synthesis in cancer. Nat Rev Cancer 2016;16:732-49. [Crossref] [PubMed]
11. Li S, Han H, Yang K, et al. Exosome-mediated metabolic reprogramming: effects on thyroid cancer progression and tumor microenvironment remodeling. Mol Cancer 2025;24:247. [Crossref] [PubMed]
12. Nagarsheth N, Wicha MS, Zou W. Chemokines in the cancer microenvironment and their relevance in cancer immunotherapy. Nat Rev Immunol 2017;17:559-72. [Crossref] [PubMed]
13. Sadik CD, Luster AD. Lipid-cytokine-chemokine cascades orchestrate leukocyte recruitment in inflammation. J Leukoc Biol 2012;91:207-15. [Crossref] [PubMed]
14. Yang K, Wang X, Song C, et al. The role of lipid metabolic reprogramming in tumor microenvironment. Theranostics 2023;13:1774-808. [Crossref] [PubMed]
15. Seok J, Choi H, Lee EK, et al. Tumor microenvironment remodeling across thyroid cancer differentiation states revealed by spatial transcriptomics. Cancer Immunol Immunother 2025;74:357. [Crossref] [PubMed]
16. Ma Q, Li Y, An L, et al. Assessment of causal association between differentiated thyroid cancer and disordered serum lipid profile: a Mendelian randomization study. Front Endocrinol (Lausanne) 2023;14:1291445. [Crossref] [PubMed]
17. Qin R, Li C, Wang X, et al. Identification and validation of an immune-related prognostic signature and key gene in papillary thyroid carcinoma. Cancer Cell Int 2021;21:378. [Crossref] [PubMed]
18. Li X, Jian J, Zhang A, et al. The role of immune cells and immune related genes in the tumor microenvironment of papillary thyroid cancer and their significance for immunotherapy. Sci Rep 2024;14:18125. [Crossref] [PubMed]
19. Jin T, Ge L, Chen J, et al. Identification of iron metabolism-related genes as prognostic indicators for papillary thyroid carcinoma: a retrospective study. PeerJ 2023;11:e15592. [Crossref] [PubMed]
20. Wu M, Ou-Yang DJ, Wei B, et al. A Prognostic Model of Differentiated Thyroid Cancer Based on Up-Regulated Glycolysis-Related Genes. Front Endocrinol (Lausanne) 2022;13:775278. [Crossref] [PubMed]
21. Xie D, Huang L, Li C, et al. Identification of PANoptosis-related genes as prognostic indicators of thyroid cancer. Heliyon 2024;10:e31707. [Crossref] [PubMed]
22. Lin R, Fogarty CE, Ma B, et al. Identification of ferroptosis genes in immune infiltration and prognosis in thyroid papillary carcinoma using network analysis. BMC Genomics 2021;22:576. [Crossref] [PubMed]
23. Wang J, Wang G, Hu T, et al. Identification of an ADME-related gene for forecasting the prognosis and responding to immunotherapy in sarcomas. Eur J Med Res 2024;29:45. [Crossref] [PubMed]
24. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics 2010;26:1572-3. [Crossref] [PubMed]
25. Bindea G, Mlecnik B, Tosolini M, et al. Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer. Immunity 2013;39:782-95. [Crossref] [PubMed]
26. Liu X, Xi X, Xu S, et al. Targeting T cell exhaustion: emerging strategies in non-small cell lung cancer. Front Immunol 2024;15:1507501. [Crossref] [PubMed]
27. Bindea G, Mlecnik B, Hackl H, et al. ClueGO: a Cytoscape plug-in to decipher functionally grouped gene ontology and pathway annotation networks. Bioinformatics 2009;25:1091-3. [Crossref] [PubMed]
28. Guo Q, Dong D, Qiao X, et al. Hub genes, diagnostic model, and predicted drugs related to ferroptosis in chronic rhinosinusitis with nasal polyps. Medicine (Baltimore) 2024;103:e40624. [Crossref] [PubMed]
29. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011;144:646-74. [Crossref] [PubMed]
30. Koundouros N, Poulogiannis G. Reprogramming of fatty acid metabolism in cancer. Br J Cancer 2020;122:4-22. [Crossref] [PubMed]
31. Strickaert A, Saiselet M, Dom G, et al. Cancer heterogeneity is not compatible with one unique cancer cell metabolic map. Oncogene 2017;36:2637-42. [Crossref] [PubMed]
32. Liu Y, Cao X. Characteristics and Significance of the Pre-metastatic Niche. Cancer Cell 2016;30:668-81. [Crossref] [PubMed]
33. Jin HR, Wang J, Wang ZJ, et al. Lipid metabolic reprogramming in tumor microenvironment: from mechanisms to therapeutics. J Hematol Oncol 2023;16:103. [Crossref] [PubMed]
34. Hilligan KL, Ronchese F. Antigen presentation by dendritic cells and their instruction of CD4+ T helper cell responses. Cell Mol Immunol 2020;17:587-99. [Crossref] [PubMed]
35. Pan R, Ryan J, Pan D, et al. Augmenting NK cell-based immunotherapy by targeting mitochondrial apoptosis. Cell 2022;185:1521-1538.e18. [Crossref] [PubMed]
36. Mantovani A, Allavena P, Marchesi F, et al. Macrophages as tools and targets in cancer therapy. Nat Rev Drug Discov 2022;21:799-820. [Crossref] [PubMed]
37. Agarwal S, Bychkov A, Jung CK. Emerging Biomarkers in Thyroid Practice and Research. Cancers (Basel) 2021;14:204. [Crossref] [PubMed]
38. Cunha LL, Ward LS. Translating the immune microenvironment of thyroid cancer into clinical practice. Endocr Relat Cancer 2022;29:R67-83. [Crossref] [PubMed]
39. Thorsson V, Gibbs DL, Brown SD, et al. The immune landscape of cancer. Immunity 2018;48:812-830.e14. [Crossref] [PubMed]
40. Angell TE, Lechner MG, Smith AM, et al. Circulating Myeloid-Derived Suppressor Cells Predict Differentiated Thyroid Cancer Diagnosis and Extent. Thyroid 2016;26:381-9. [Crossref] [PubMed]
41. Najjar YG, Finke JH. Clinical perspectives on targeting of myeloid derived suppressor cells in the treatment of cancer. Front Oncol 2013;3:49. [Crossref] [PubMed]
42. Febrero B, Ruiz-Manzanera JJ, Ros-Madrid I, et al. Tumor microenvironment in thyroid cancer: Immune cells, patterns, and novel treatments. Head Neck 2024;46:1486-99. [Crossref] [PubMed]
43. Guo J, Si G, Song X, et al. Mediating role of circulating inflammatory proteins in the effect of immune cells on esophageal cancer risk: A Mendelian randomization study. Medicine (Baltimore) 2024;103:e40374. [Crossref] [PubMed]
44. Yang S, Liu H, Zheng Y, et al. The Role of PLIN3 in Prognosis and Tumor-Associated Macrophage Infiltration: A Pan-Cancer Analysis. J Inflamm Res 2025;18:3757-77. [Crossref] [PubMed]
45. Qiu C, Wang W, Xu S, et al. Construction and validation of a hypoxia-related gene signature to predict the prognosis of breast cancer. BMC Cancer 2024;24:402. [Crossref] [PubMed]
46. Guo Q, Zhong X, Dang Z, et al. Identification of GBN5 as a molecular biomarker of pan-cancer species by integrated multi-omics analysis. Discov Oncol 2025;16:85. [Crossref] [PubMed]
47. Leone RD, Powell JD. Metabolism of immune cells in cancer. Nat Rev Cancer 2020;20:516-31. [Crossref] [PubMed]
48. Xiao Z, Dai Z, Locasale JW. Metabolic landscape of the tumor microenvironment at single cell resolution. Nat Commun 2019;10:3763. [Crossref] [PubMed]
49. Liu J, Hong M, Li Y, et al. Programmed Cell Death Tunes Tumor Immunity. Front Immunol 2022;13:847345. [Crossref] [PubMed]
50. Chen X, Qi G, Qin M, et al. DNA methylation directly downregulates human cathelicidin antimicrobial peptide gene (CAMP) promoter activity. Oncotarget 2017;8:27943-52. [Crossref] [PubMed]
51. Zhang T, Zhang H, He L, et al. Potential Use of 1-25-dihydroxyvitamin D in the Diagnosis and Treatment of Papillary Thyroid Cancer. Med Sci Monit 2018;24:1614-23. [Crossref] [PubMed]
52. Chen J, Shin VY, Ho JC, et al. Functional Implications of Cathelicidin Antimicrobial Protein in Breast Cancer and Tumor-Associated Macrophage Microenvironment. Biomolecules 2020;10:688. [Crossref] [PubMed]
53. Lee J, Shin KO, Kim Y, et al. Cathelicidin-Related Antimicrobial Peptide Regulates CD73 Expression in Mouse Th17 Cells via p38. Cells 2020;9:1561. [Crossref] [PubMed]
54. Cha HR, Lee JH, Hensel JA, et al. Prostate cancer-derived cathelicidin-related antimicrobial peptide facilitates macrophage differentiation and polarization of immature myeloid progenitors to protumorigenic macrophages. Prostate 2016;76:624-36. [Crossref] [PubMed]
55. Michalski C, Cheung C, Oh JH, et al. DDIT4L regulates mitochondrial and innate immune activities in early life. JCI Insight 2024;9:e172312. [Crossref] [PubMed]
56. Koga Y, Pelizzola M, Cheng E, et al. Genome-wide screen of promoter methylation identifies novel markers in melanoma. Genome Res 2009;19:1462-70. [Crossref] [PubMed]
57. Ness C, Katta K, Garred Ø, et al. Integrated differential DNA methylation and gene expression of formalin-fixed paraffin-embedded uveal melanoma specimens identifies genes associated with early metastasis and poor prognosis. Exp Eye Res 2021;203:108426. [Crossref] [PubMed]
58. Ozdemir Kutbay N, Biray Avci C, Sarer Yurekli B, et al. Effects of metformin and pioglitazone combination on apoptosis and AMPK/mTOR signaling pathway in human anaplastic thyroid cancer cells. J Biochem Mol Toxicol 2020;34:e22547. [Crossref] [PubMed]
59. Meng M, Wu YC. LMX1B Activated Circular RNA GFRA1 Modulates the Tumorigenic Properties and Immune Escape of Prostate Cancer. J Immunol Res 2022;2022:7375879. [Crossref] [PubMed]
60. Lu H, Zhu C, Ruan Y, et al. hsa-miR-206b Involves in the Development of Papillary Thyroid Carcinoma via Targeting LMX1B. Biomed Res Int 2022;2022:7488708. [Crossref] [PubMed]
61. Bae SW, Berlth F, Jeong KY, et al. Glucose metabolic profiles evaluated by PET associated with molecular characteristic landscape of gastric cancer. Gastric Cancer 2022;25:149-60. [Crossref] [PubMed]
62. Takagi K, Naruse A, Akita K, et al. CALN1 hypomethylation as a biomarker for high-risk bladder cancer. BMC Urol 2022;22:176. [Crossref] [PubMed]