Identification and validation of a novel lymph node metastasis-related model for papillary thyroid carcinoma to predict the prognosis

Abstract

Background

Papillary thyroid carcinoma (PTC) is a common malignancy with a good prognosis, but lymph node metastasis (LNM) is associated with a poor prognosis for patients. This study aimed to construct an LNM-related risk model and identify hub genes that could predict the prognosis of PTC.

Methods

Gene expression and clinical information were obtained from The Cancer Genome Atlas (TCGA). Cox analysis was used to select hub genes and construct a risk model. The model was validated through receiver operator characteristic (ROC) curve. Nomogram was constructed for clinical application. Survival analysis was performed by the Kaplan-Meier (K-M) curve. Methylation of hub genes and drug sensitivity were calculated in Gene Set Cancer Analysis (GSCA). The expression levels of four hub genes and their effect on the malignant features of PTC were further validated through cell experiments.

Results

A risk model was constructed by four hub genes ATP2C2, CXCL5, IL11, and TREM1. ROC curve showed that the AUC of the risk model for PTC prognosis at 3-, 5-, and 10-year was 0.91, 0.88, and 0.92, respectively. The nomogram indicated that risk score was more important than some clinical characteristics. High-risk group exhibited lower immune infiltration levels. In PTC, four hub genes might function as oncogenes, but ATP2C2 was lowly expressed in tumors and patients with LNM. Additionally, ATP2C2 overexpression promoted PTC cell migration, invasion, and LNM-related protein expression levels, while knockdown of CXCL5, IL11, and TREM1 inhibited PTC cells’ malignant features.

Conclusions

We constructed an LNM-related risk model based on four hub genes, and targeting these key genes can benefit patients from immunotherapy or chemotherapy.

Highlight box.

Key findings

• We constructed a risk model based on four hub genes related to lymph node metastasis (LNM) of papillary thyroid cancer (PTC) with high accuracy and stability, and targeting these key genes can benefit patients from immunotherapy or chemotherapy.

What is known, and what is new?

• PTC patients with LNM require surgical intervention or radioactive iodine therapy, which may increase the probability of recurrence and diminish quality of life.

• It is vital to explore signatures related to PTC progression and metastasis for improving PTC survival and guiding treatment.

What is the implication, and what should change now?

• The identified hub genes were associated with the tumor microenvironment of PTC. However, the specific association of these hub genes with immune cells in PTC also needs more exploration.

Introduction

Thyroid cancer (THCA) is a common malignant originating from follicular epithelial cells or parafollicular C cells (1). It encompasses four subtypes, namely papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), poorly differentiated thyroid carcinoma (PDTC), and anaplastic thyroid carcinoma (ATC) (2). PTC is a predominant subtype accounting for over 90% of THCA (3). Although the majority of PTC has a favorable prognosis, the incidence of lymph node metastasis (LNM) usually leads to a poor prognosis (4-6). At present, the LNM of PTC is usually detected by ultrasonography, but its sensitivity is not favorable (7). Moreover, PTC patients with LNM necessitate surgical intervention or radioactive iodine therapy, which may increase the probability of recurrence and diminish quality of life (8).

Previous studies claimed that genetic alterations played a pivotal role in the pathogenesis of PTC by mediating some signal pathways and immune response for tumor cell proliferation and metastasis (9,10). Recently, several messenger RNAs (mRNAs) have been identified as being implicated in the metastasis of PTC, such as CAPN8, HN1, and FN1 (11-13). Besides, the targeted treatment was widely applied in the treatment of cancers with high efficiency and low side effects according to the molecular signatures, including THCA (14-16). With the growing availability of molecular biomarkers as diagnostic and therapeutic tools in cancer, it is promising to improve early detection and reduce mortality rates in cancers (17). Hence, it is vital to explore signatures related to PTC progression and metastasis to improve PTC survival and guide treatment.

In this study, we tried to construct a risk model for the clinical diagnosis and identify some hub genes that could independently predict the prognosis of PTC and related to tumor growth and LNM based on The Cancer Genome Atlas (TCGA) database. Besides, we primarily explored the mechanism and functions of signatures in tumor development, LNM, and drug sensitivity. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-102/rc).

Methods

Data collection and processing

The gene expression profile and clinical information were obtained from TCGA database (https://portal.gdc.cancer.gov/). There were 572 samples, including 59 normal samples and 513 tumor samples. The GSE33630 dataset, including 49 PTC samples and 45 normal samples, was downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). The gene expression difference between the N0 and N1, tumor and normal samples was calculated using the limma R package. The P<0.05 and the fold change ≥1.5 were taken as the threshold of significant difference. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis

The ClusterProfile R package was used for the KEGG enrichment analysis. After the gene list was imported into R software, the enrichKEGG function was performed. Besides, the results were visualized by the ggplot2 R package.

Construction and validation of the risk model

After the gene list was determined based on the Venn plot, the univariate Cox analysis was performed to select genes that were closely related to the prognosis of PTC. Then, the least absolute shrinkage and selection operator (Lasso) Cox analysis was executed to shrink and select variables using the glmmet R package. Next, the hub genes with independent prediction values and corresponding coefficients were determined using multivariate Cox regression. Furthermore, we constructed a risk model based on the expressions and coefficients of hub genes. The receiver operator characteristic (ROC) curve was used to validate the risk model through the timeROC package in R software. Finally, some clinical characteristics and risk scores were integrated to construct a nomogram for clinical application using rms and survival packages.

Immune infiltration analysis

The proportions of 22 immune cell subtypes in the TCGA data were evaluated by the CIBERSORT R package which contained the marker genes of 22 immune cells as same as the data file from CIBERSORTX (https://cibersortx.stanford.edu/). Besides, the results were visualized using the ggplot2 package.

Methylation and survival analysis

The University of Alabama at Birmingham Cancer Data Analysis Portal (UALCAN; https://ualcan.path.uab.edu/) is a comprehensive, user-friendly, and interactive web resource for analyzing cancers based on TCGA data. In this study, UALCAN was used for the methylation analysis of hub genes. The survival analysis was performed using surv and survfit functions in the survival package, and the Kaplan-Meier (K-M) curves were generated using the ggsurvplot function.

Gene set enrichment analysis (GSEA)

To study the role of hub genes in the development of PTC, the GSEA was performed to analyze their function. The expression profile of PTC from TCGA was imported, and c2.cp.kegg.v2023.1.Hs.symbols.gmt was downloaded as a reference. The results with P<0.05 were included.

Drug sensitivity

Gene Set Cancer Analysis (GSCA; http://bioinfo.life.hust.edu.cn/GSCA/#/) is an integrated platform. It integrates over 10,000 multi-dimensional genomic data across 33 cancer types from TCGA and over 750 small molecule drugs from Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Therapeutics Response Portal (CTRP). In this study, we analyzed the drug sensitivity of hub genes in GSCA.

Molecular docking

The 3D structures of protein and ingredients were obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/) and UniProt (https://www.uniprot.org/). Then, these structures were imported into AutodockTools to remove water molecules and add hydrogen, followed by molecular docking. The results were saved in dlg formats and converted into pdb format in Open Bable. Next, the PyMOL was used for the visualization of results after the results were imported in pdb format.

Cell culture and transfection

The PTC cell lines (TPC-1 and IHH4) and normal thyroid cell lines (Htori-3 and Nthy-ori-3-1) were purchased from Procell Life Science & Technology Co., Ltd. (Wuhan, China). TPC-1 and Htori-3 cells were incubated in Dulbecco’s modified Eagle medium (Hyclone, Logan, UT, USA) supplemented with 10% fetal bovine serum (FBS; Solarbio, Beijing, China) and 1% penicillin-streptomycin (Hyclone). IHH4 and Nthy-ori-3-1 cells were cultured in RPMI-1640 medium. All cells were incubated at 37 ℃ in 5% CO₂. According to the cell growth conditions, some media were changed every 2 days.

For cell transfection, IHH4 cells were transfected with small interfering (si)-negative control (NC), si-TREM1, si-IL11, or si-CXCL5 using Lipofectamine 3000 (Ribo-Bio, Guangzhou, China) under the instruction of the manufacturer. Plasmids of overexpression (oe)-ATP2C2 were transfected into IHH4 cells to upregulate the ATP2C2 expression.

The cell culture and quantification of gene expression by quantitative real-time polymerase chain reaction (qRT-PCR)

Total RNA extractor (Sangon, Shanghai, China) was applied to extract total RNA from cells. The Cary 3500 UV-Vis Spectrophotometer (Agilent Technologies, Inc., Santa Clara, CA, USA) was employed to measure the ratio of A260/A230 and A260/A280. After detection, reverse transcription of RNA samples and qRT-PCR were performed with One Step RT-qPCR Kit (Sangon) in the Mx3000P qPCR System (Agilent Technologies, Inc.). MicroRNA (miRNA) was detected with miRNAs qPCR kit (Sangon). The mRNA expression was calculated by the 2^−ΔΔCt method. The internal reference was glyceraldehyde-3-phosphate dehydrogenase (GAPDH).

Polymerase chain reaction (PCR) conditions were: 1 min at 95 ℃, then 20 s at 95 ℃, and 45 s at 58 ℃ for 40 cycles. The primer sequences are shown in Table 1.

Table 1. All primers in qPCR experiments in this study.

ID	Forward sequence (5'-3')	Reverse sequence (5'-3')
ATP2C2	TTCCTCTACTCCGTCCTGGG	CTCTTGGGGCTGCAACAGTA
CXCL5	GTGCAATTAACAAAGCTACTGCAAG	GGCATCTAAAAAGCTCAGCAATG
IL11	AGGTGGCTCTTCCCTGAA	GGGTCACAGCCGAGTCTT
TREM1	TCCGAATGGTCAACCTTCAAGTGG	GAACAGCATGTGAGGCTCCTTGG
GAPDH	GAACGTCGAAAAGAAAAGTCTCG	CCTTATCAAGATGCGAACTCACA

GAPDH, glyceraldehyde-3-phosphate dehydrogenase; qPCR, quantitative polymerase chain reaction.

Western blotting

Cells were lysed using radioimmunoprecipitation assay (RIPA) buffer (Beyotime, Shanghai, China) and centrifuged at 12,000 rpm for 5 min at 4 ℃. Proteins were resolved through sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene fluoride (PVDF) membranes. After blocking with 5% nonfat milk for 2 h, the membranes were incubated overnight at 4 ℃ with primary antibodies specific for ATP2C2 (1 µg/mL; #ab176279; Abcam, Cambridge, UK), TREM1 (1:1,000; #ER1918-03; Huabio, Hangzhou, China), IL11 (1:1,000; #55169-1-AP; Proteintech, Rosemont, IL, USA), CXCL5 (1:1,000; #HA722561; Huabio), E-cadherin (1:5,000; #ET1607-75; Huabio), N-cadherin (1:5,000; #ET1607-37; Huabio), vimentin (1:20,000; #ET1610-39; Huabio), VEGF-C (1:1,000; #22601-1-AP; Proteintech), VEGF-D (1:1,000; #ER65298; Huabio), VEGFR3 (1:1,000; #ER65750; Huabio), LYVE-1 (1:2,000; #ET1702-29; Huabio), PROX1 (1:1,000; #HA722318; Huabio), and GAPDH (1:10,000; #10494-1-AP; Proteintech). This was followed by a 2 h incubation with secondary antibodies (1:20,000; #SA00012-1; Proteintech). Protein detection was performed using an enhanced chemiluminescence (ECL) reagent (Millipore, Burlington, MA, USA).

Wound healing

Cells (1×10⁵) were seeded into 6-well plates and incubated for 16 h. Once the cells reached 90% confluence, the complete medium was replaced with fresh medium. A uniform scratch was then made across each well using a 10 µL pipette tip. To eliminate detached cells, the wells were gently rinsed twice with PBS, followed by the addition of serum-free medium. Images of the scratch areas were captured using an inverted microscope under bright-field conditions at 0 and 24 h.

Transwell invasion analysis

Transwell invasion analysis was assessed using a Boyden chamber assay with an 8 µm pore size. IHH4 cells (1×10⁵) were suspended in 200 µL of serum-free medium and seeded into the top chamber pre-coated with Matrigel (BD Biosciences, San Jose, CA, USA). The lower chamber contained a medium supplemented with 10% FBS to act as a chemoattractant. After 24 h of incubation, the cells were fixed with 4% paraformaldehyde for 15 min and stained with 0.1% crystal violet for 20 min. The number of invading cells in six randomly chosen fields was quantified using a microscope.

Statistics analysis

In this study, all statistics analyses were performed in SPSS 25, R software (version 4.3.0), and GraphPad (version 10.1.2). Student’s t-test was used to compare the differences between the two groups, and ANOVA was used for comparison among three or over groups. Log-rank test was used to compare the survival probability in K-M curves. Pearson correlation was used to analyze the linear relationship between two variables. The P<0.05 was considered statistically significant.

Results

Identification of differentially expressed genes (DEGs)

Firstly, the volcano plot showed that there were 2,169 upregulated genes and 3,253 downregulated genes between normal and tumor samples (Figure 1A), while 917 gene expressions were upregulated and 545 gene expressions were downregulated in N1 compared with N0 (Figure 1B). The KEGG enrichment analysis showed that the upregulated genes in the tumor group were enriched in p53, PI3K-Akt, calcium, phospholipase D, and MAPK signal pathways (Figure 1C). The genes with upregulated expression in the N1 group were mainly involved in the IL-17, chemokine, PI3K-Akt, tumor necrosis factor (TNF), and nuclear factor (NF)-κB signal pathways (Figure 1D). Based on the Venn diagram, 1,152 genes were screened out which were both differentially expressed in the two datasets (Figure 1E).

Figure 1.

The identification of DEGs in the TCGA database. (A) The volcano of DEGs plots showed the DEGs between normal and PTC samples. (B) The volcano of DEGs plots showed the DEGs between N0 and N1 samples. (C) The KEGG enrichment analysis based on the upregulated genes between normal and PTC samples. (D) The KEGG enrichment analysis based on the upregulated genes between LNM and non-LNM of lymph node samples. (E) The Venn plot showed a common list of 1,152 DEGs. DEGs, differentially expressed genes; KEGG, Kyoto Encyclopedia of Genes and Genomes; LNM, lymph node metastasis; PTC, papillary thyroid carcinoma; TCGA, The Cancer Genome Atlas.

Construction and validation of the risk model

After the gene list was obtained through the Venn diagram, 197 genes closely related to the overall survival (OS) time were selected using univariate Cox analysis. Then, Lasso Cox analysis was performed on these genes, and the lambda was determined as the optimal lambda value by 10-fold cross-validation (Figure 2A). Five prognostic genes with optimal lambda were successfully identified (Figure 2B). Next, the multivariate Cox analysis was further performed based on these five genes to select the hub genes with independent prediction value. As a result, four genes, namely ATP2C2 (also called SPCA2), CXCL5, TREM1, and IL11, were determined as risk-related genes (Table 2). The risk model was determined according to the hub gene expression and β coefficient as follows: risk score = 0.534 × CXCL5 + 0.102 × TREM1 + 1.099 × ATP2C2 + 0.766 × IL11. Furthermore, according to this formula, the risk score of every patient was calculated, and these patients were divided into low- and high-risk score groups based on the optimal truncation value. The survival analysis exhibited that the high-risk score group was closely related to the poor prognosis of PTC (Figure 2C). Besides, the ROC result showed that the AUC of the risk model for 3-, 5-, and 10-year prognosis was 0.91, 0.88, and 0.92, respectively (Figure 2D). Based on the risk score and clinical characteristics in the TCGA database, we built a prognostic nomogram for clinicians to predict 3-, 5-, and 10-year OS probabilities of PTC patients (Figure 2E). It was obvious that risk score played most important role in the prediction of prognosis compared with other clinical characteristics.

Figure 2.

The risk model was constructed using Cox regression analysis and validated using the ROC curve. (A) The confidence interval under each lambda. (B) The trajectory of each independent variable. (C) The K-M survival curve showed the prognosis of the high- and low-risk groups. (D) The ROC curve of the risk score. (E) The nomogram to predict the 3-, 5-, and 10-year OS time of PTC patients. AUC, area under the curve; CI, confidence interval; K-M, Kaplan-Meier; M, metastasis; N, node; OS, overall survival; P, pathological; PTC, papillary thyroid carcinoma; ROC, receiver operator characteristic; T, tumor.

Table 2. Univariate and multivariate Cox regression analysis.

Genes	Univariate			Multivariate
	HR (95% CI)	P value		HR (95% CI)	P value
ATP2C2	2.408 (1.815–3.195)	<0.001		2.815 (1.892–4.188)	<0.001
CXCL5	1.819 (1.452–2.278)	<0.001		1.919 (1.405–2.622)	<0.001
HTR1E	1.841 (1.399–2.421)	<0.001		1.614 (0.996–2.615)	0.052
TREM1	1.118 (1.015–1.232)	0.02		1.919 (1.405–2.622)	<0.001
IL11	2.302 (1.565–3.388)	<0.001		2.266 (1.259–4.079)	0.006

CI, confidence interval; HR, hazard ratio.

Relationship between risk score and clinicopathological characteristics in PTC patients

Clinical characteristics may affect the prognosis of patients. Thus, we compared risk scores in different clinicopathological characteristics. As shown in Figure 3, it was obvious that the difference in risk scores between different clinical groups was not significant. It meant our risk model was steady because it was not influenced by clinical characteristics.

Figure 3.

Relationship between risk score and clinical characteristics. (A) Age, the old indicated the age of patients >50 years, while the young indicated the age of patients ≤50 years; (B) gender; (C) M stage; (D) N stage; (E) T stage; (F) P stage. ns, not significant. M, metastasis; N, node; P, pathological; T, tumor.

Immune infiltration and the expression of LNM-related genes in different risk groups

As we all know, the abnormality of some immune cell infiltration levels leads to tumor deterioration and metastasis. Besides, the change of some metastasis-related gene expression also promotes the progress of LNM. Our results exhibited that the CD8⁺ T cell, natural killer (NK) cell, monocytes, and M1 macrophages have a higher infiltration level in the low-risk group compared with the high-risk group, while the infiltration level of the dendritic cell (DC) was lower in the low-risk group (Figure 4).

Figure 4.

Immune infiltration levels and four LNM-related gene expressions were obviously different in the high- and low-risk groups. *, P<0.05; ***, P<0.001. LNM, lymph node metastasis; NK, natural killer.

Expression and survival analysis of hub gene

Then, we explore the role of hub genes in the development and LNM of PTC (Figure 5A-5H). The ATP2C2 was downregulated in tumor and N1 samples (Figure 5A,5E), while CXCL5, IL11, and TREM1 were upregulated in tumor samples and N1 samples (Figure 5B-5D,5F-5H). In addition, the survival analysis revealed that the high expressions of ATP2C2, CXCL5, IL11, and TREM1 were all closely related to poor prognosis in PTC (Figure 5I-5L). In the validation GEO dataset, expression levels of these four hub genes between PTC and normal samples showed consistent trends with those in the TCGA cohort (Figure S1).

Figure 5.

Expression and prognosis of four hub genes. (A-D) The expressions of (A) ATP2C2, (B) CXCL5, (C) IL11, and (D) TREM1 in normal and tumor samples. (E-H) The expression of (E) ATP2C2, (F) CXCL5, (G) IL11, and (H) TREM1 in N0 and N1 samples. (I-L) The K-M curve of (I) ATP2C2, (J) CXCL5, (K) IL11, and (L) TREM1 for the OS of PTC patients based on the optimal truncation value. *, P<0.05; **, P<0.01; ***, P<0.001. K-M, Kaplan-Meier; N, node; OS, overall survival; PTC, papillary thyroid carcinoma.

Methylation level and role of hub genes

The abnormal mRNA expression may be caused by methylation. Thus, we analyze the methylation level of hub genes in normal and tumor samples. From Figure S2A-S2D, it could be seen that there was no significant difference in the methylation of ATP2C2 between normal and tumor samples. Additionally, the methylation levels of CXCL5 and IL11 were significantly upregulated, while the methylation level TREM1 was significantly downregulated in tumor samples. Furthermore, the correlation analysis revealed that the mRNA expression of TREM1 was negatively and closely related to their methylation levels, while the Spearman correlation of ATP2C2, CXCL5, and IL11 were respectively −0.23, −0.13, and 0.09, which were regarded as irrelevant (Figure S2E). However, we found the methylation of four hub genes was not closely related to the OS in THCA (Figure S2F).

GSEA function enrichment analysis of hub genes

Subsequently, the GSEA was performed for the function enrichment analysis of four hub genes. The results showed that APT2C2 was enriched in the p53 signal pathway, NK cell-mediated cytotoxicity, and cell adhesion molecule (CAM) (Figure S3A-S3C). Besides, CXCL5 was enriched in the chemokine signal pathway, cytokine-cytokine receptor interaction, and JAK-STAT signal pathway (Figure S3D-S3F). Additionally, the main function of IL11 contained transforming growth factor (TGF)-β, Notch, and MAPK signal pathways (Figure S3G-S3I). Furthermore, TREM1 was mainly involved in the JAK-STAT, MAPK, and VEGF signal pathways (Figure S3J-S3L).

Drug sensitivity of hub genes based on CTRP and GDSC database

To determine the clinical value of hub genes in THCA treatment, we analyzed the drug sensitivity of four hub genes based on the CTRP and GDSC databases. The results showed that IL11 was positively related to most drugs in the CTRP database (Figure S4A), while it was negatively related to most drugs in the GDSC database (Figure S4B). Besides, CXCL5 and ATP2C2 were positively related to most drugs (Figure S4). In particular, ATP2C2 was negatively associated with austocystin D, with the highest correlation coefficient. TREM1 was only negatively related to teniposide and canertinib in the CTRP database (Figure S4A), while it was only positively related to elesclomol in the GDSC database (Figure S4B).

Molecular docking

To identify the feasibility of the targeted therapy, ATP2C2 and austocystin D which had the highest negative correlation were selected for the molecular docking. As shown in Figure S5, the austocystin D could bind to ATP2C2 by forming a hydrophobic interaction with GLU-701, and the binding energy is −5.10, which indicates a good interaction. Therefore, ATP2C2 may be targeted by austocystin D in treating PTC.

Validation of hub genes expression using qRT-PCR

Subsequently, qRT-PCR was used to verify the expression of hub genes in PTC cells. The results exhibited that ATP2C2, TREM1, and IL11 were upregulated in PTC cells compared with the normal cells (Figure 6A). Besides, there was not a significant difference in CXCL5 expression between PTC and thyroid cells (Figure 6A). Due to the inconsistent expression of ATP2C2 and CXCL5 with the previous bioinformatics analysis results, we further verified the expression of these two genes in different cell lines. As shown in Figure 6B, compared with normal thyroid cells, the expression level of ATP2C2 was significantly downregulated in IHH4 cells, while the expression of CXCL5 was significantly upregulated. We preliminarily speculate that the expression difference of these two genes in different cell lines is due to cell specificity, but this requires further exploration.

Figure 6.

Expression of hub genes in PTC and thyroid cells. (A) Expression levels of ATP2C2, TREM1, IL11, and CXCL5 in PTC (TPC-1) and thyroid (Htori-3) cells; ns means no significance. (B) Expression levels of ATP2C2 and CXCL5 in PTC (Nthy-ori-3-1) and thyroid (IHH4) cells. ns, not significant. mRNA, messenger RNA; PTC, papillary thyroid carcinoma.

Effect of hub genes on malignant characteristics of PTC cells

To further explore the role of hub genes in PTC cells, the expression levels of four hub genes were reversed in the PTC cell line IHH4 cells. As shown in Figure 7A,7B, transfection of oe-ATP2C2 significantly elevated the mRNA and protein expression of ATP2C2 compared to the control group. ATP2C2 overexpression promoted cell migration and invasion of IHH4 cells (Figure 7C,7D). To further investigate the effects of the hub genes on tumor LNM, we examined the expression of epithelial-mesenchymal transition (EMT)-related and LNM-related proteins. As shown in Figure 7E,7F, overexpression of ATP2C2 significantly decreased E-cadherin expression while increasing the expression of N-cadherin and vimentin (Figure 7E). In addition, the expression levels of LNM-related proteins, including VEGF-C, VEGF-D, VEGFR3, LYVE-1, and PROX1, were also significantly elevated following ATP2C2 overexpression (Figure 7F).

Figure 7.

Effect of ATP2C2 overexpression on malignant features of IHH4 cells. (A) mRNA expression of ATP2C2. (B) Protein expression of ATP2C2. (C) Wound healing detected the migration of IHH4 cells. The images were captured under bright-field conditions using an inverted microscope at 0 and 24 h after scratch creation; scale bar =500 μm. (D) Transwell analysis detected the invasion of IHH4 cells; cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet before imaging; scale bar =200 μm. (E) EMT-related protein expression. (F) LNM-related protein expression. #1, #2, and #3 represent three repeated experiments. IHH4 cells were transfected with control vectors or oe-ATP2C2. *, P<0.05. EMT, epithelial-mesenchymal transition; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; LNM, lymph node metastasis; mRNA, messenger RNA; oe, overexpression.

Additionally, TREM1, IL11, and CXCL5 mRNA expression levels decreased after transfecting with si-TREM1, si-IL11, and si-CXCL5, respectively (Figure 8A). These genes’ protein expression levels were also significantly reduced when knockdown TREM1, IL11, and CXCL5 (Figure 8B). As exhibited in Figure 8C,8D, downregulation of TREM1, IL11, and CXCL5 significantly inhibited migration and invasion of IHH4 cells. Conversely, knockdown of TREM1, IL-11, or CXCL5 markedly increased E-cadherin expression while reducing the levels of N-cadherin, vimentin, VEGF-C, VEGF-D, VEGFR3, LYVE-1, and PROX1 (Figure 9). These results suggest that ATP2C2, TREM1, IL11, and CXCL5 may be closely associated with tumor LNM by regulating the expression of EMT- and LNM-related proteins.

Figure 8.

Effect of TREM1, IL11, and CXCL5 on malignant features of IHH4 cells. (A) mRNA expression of TREM1, IL11, and CXCL5. (B) Protein expression of TREM1, IL11, and CXCL5. (C) Wound healing detected the migration of IHH4 cells. The images were captured under bright-field conditions using an inverted microscope at 0 and 24 h after scratch creation; scale bar =500 μm. (D) Transwell analysis detected the invasion of IHH4 cells; cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet before imaging; scale bar =200 μm. IHH4 cells were transfected with si-NC, si-TREM1, si-IL11, or si-CXCL5. *, P<0.05; ***, P<0.001; ****, P<0.0001. GAPDH, glyceraldehyde-3-phosphate dehydrogenase; mRNA, messenger RNA; NC, negative control; si, small interfering.

Figure 9.

Effect of TREM1, IL11, and CXCL5 on EMT- and LNM-related protein expression. (A) EMT-related protein expression. (B) LNM-related protein expression. #1, #2, and #3 represent three repeated experiments. IHH4 cells were transfected with si-NC, si-TREM1, si-IL11, or si-CXCL5. *, P<0.05. EMT, epithelial-mesenchymal transition; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; LNM, lymph node metastasis; NC, negative control; si, small interfering.

Discussion

It is all known that the prognosis of PTC is affected by various clinical factors, including age, gender, and TNM stage (18). The presence of metastasis hinders the effectiveness of treatments for PTC patients (19). PTC metastasis was affected by some genetic alterations, including changes in mRNA expression, methylation of genes, and other special modifications (20). In this study, a total of 1,152 genes with differential expression were identified, potentially contributing to the growth and LNM of PTC. Subsequently, a risk model was developed using Cox regression analysis, incorporating four hub genes (ATP2C2, TREM1, IL11, and CXCL5) that are associated with the prognosis of PTC. In addition, there was no significant difference in risk scores between different clinical characteristic groups, and our model had a high AUC value according to the ROC results. It indicated that our risk model has high stability and accuracy.

ATP2C2 is a protein-coding gene involved in the transport of Ca²⁺ in secretory granules (21). ATP2C2 has been shown to inhibit cancer cell migration and metastasis by reversing EMT (22). Zhao et al. identified ATP2C2 as an independent prognostic factor for THCA (23), revealing that patients with high ATP2C2 expression had worse prognoses. Similarly, our results indicate that high ATP2C2 expression predicts poor prognosis in PTC patients. Further cellular experiments demonstrated that overexpression of ATP2C2 enhanced PTC cell migratory and invasive capabilities. TREM1 is a cell surface receptor expressed on neutrophils, monocytes, and some tissue macrophages, with its overactivation linked to cancer progression (24). Xie et al. discovered that TREM1 is highly expressed in PTC and may promote THCA progression through immune-related pathways (25). IL11 is an inflammatory factor that initiates mesenchymal programs in stromal cells, epithelial cells, and cancer cells, with notable tumor-promoting and metastasis-promoting effects (26,27). CXCL5 is a member of the angiogenic CXC chemokine family and plays a pro-tumor role in various types of malignancies. CXCL5 is upregulated in THCA tissues and serves as a marker of poor prognosis in THCA (28). In our study, high expression of TREM1, IL11, and CXCL5 was associated with poor prognosis in PTC. Furthermore, we found that these three genes were significantly upregulated in PTC cells. Knockdown of these genes resulted in a marked suppression of migration and invasion in IHH4 cells.

Additionally, we identified that the high-risk score was related to lower immune infiltration of CD8⁺ T cells, NK cells, monocytes, and M1 macrophages. These results indicated that the PTC development may be related to the immune response. Monocyte cells play a dual role in tumors (29). They could promote tumor metastasis and development through various mechanisms, such as promoting angiogenesis and inhibiting T cell function (30,31). Additionally, monocyte cells possessed cell-mediated cytotoxicity and phagocytic activity to directly kill malignant tumor cells through cytokine-mediated mechanisms (32). On the other hand, monocyte cells could differentiate into either anti-tumor M1 or pro-tumor M2 macrophages, as well as DC, depending on the influence of different chemotactic factors, thereby participating in tumor development (33,34). DCs consist of two subgroups, known as type one (cDC1) and type two (cDC2) conventional DCs (35). Among them, cDC1 enhances antigen presentation and activates CD8 T cells for anti-tumor function, while cDC2 is mainly responsible for antigen presentation to CD4 T cells (36). It was noteworthy that Kalinski et al. found that DCs could also serve as mediators for the interaction between NK cells and CD8 T cells, and the activation of NK cells and CD8 T cells induced polarization of DCs towards cDC1 (37). In adoptive T cell therapy, cDC1 cells could promote T cell infiltration by producing CXCL9/CXCL10 (38). These findings suggested the presence of some positive feedback regulatory mechanisms among immune cells. Our results on immune infiltration indicated a significant decrease in the infiltration levels of monocyte cells and M1 macrophages in the high-risk group, while M2 macrophages showed no significant difference. This implies that monocytes may exert an inhibitory effect on tumor development in THCA. This may be achieved through their differentiation into cell types directly targeting tumor cells, and the differentiation process towards M1 macrophages may be suppressed. This regulation of monocytes with other immune cells in PTC is worth more experimental validation.

Furthermore, the highly expressed IL11 in tumors was enriched in the TGF-β signaling pathway. Previous studies have shown that TGF-β could influence the phenotypes of macrophages derived from monocyte cells and inhibit NK cell-mediated cancer surveillance through the secretion of SMAD4 (39). Mouse experiments have demonstrated that TGF-β inhibitors can increase the infiltration levels of CD8 T cells in tumors, suggesting an inhibitory role of TGF-β on CD8 T cell infiltration (40). In PTC, high expression of IL11 may inhibit CD8 T cell infiltration by activating the TGF-β pathway. Moreover, ATP2C2 was negatively correlated with the cytotoxicity of NK cells and the p53 signaling pathway. p53 pathway was reported to target the activation of gene expression, such as CEACAM1, thus inhibiting the immune cytotoxicity of NK cells (41). Our immune infiltration results indicated that NK cells’ infiltration levels were inhibited. Therefore, the low expression of ATP2C2 may play a significant role in the activation of the p53 signaling pathway. We speculated that ATP2C2 inhibited the function of NK cells by promoting the activation of the p53 signaling pathway in PTC. On the other hand, the highly expressed CXCL5 and TREM1 are both enriched in the JAK-STAT signaling pathway. STAT proteins, including STAT1, STAT2, STAT3, STAT4, STAT5A, STAT5B, and STAT6 (42), can be downstream signaling molecules of JAK. STAT3 can indirectly impair the function of NK cells by regulating the expression of NK cell activation receptors and immune checkpoint proteins (43,44). It can also inhibit the proliferation of effector T cells by inducing the production of IL-23 (45). Similarly, STAT1 could inhibit T cell-mediated immune responses and induce T cell apoptosis (46). Combining these studies, our results suggested that CXCL5 and TREM1 may inhibit the proliferation and cytotoxicity of immune cells by activating the JAK-STAT pathway in PTC.

However, our results have some limitations. Although we identified some hub genes related to the development and LNM of THCA, the potential mechanism should be identified through experiments in vitro. The specific association of these hub genes with immune cells in PTC also needs more exploration.

Conclusions

In conclusion, we identified four key genes closely related to the metastasis and prognosis of PTC. These genes showed differential expression levels between normal and tumor cells. Regulation of their expression can significantly alter the malignant characteristics of PTC cells. These genes were also associated with the tumor microenvironment of PTC. These results suggest that these four hub genes are expected to serve as potential therapeutic targets for PTC.

Supplementary

The article’s supplementary files as

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.

Data Sharing Statement

Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-102/dss

gs-14-08-1456-dss.pdf ^{(26.3KB, pdf)}

DOI: 10.21037/gs-2025-102