Abstract

Background

   The detection rate of thyroid cancer (TC) has been continuously
   improved due to the development of detection technology.
   Epithelial–mesenchymal transition (EMT) is thought to be closely
   related to the malignant progression of tumors. However, the
   relationship between EMT-related genes (ERGs) characteristics and the
   diagnosis and prognosis of TC patients has not been studied.

Methods

   Four datasets from Gene Expression Omnibus (GEO) were used to perform
   transcriptomic profile analysis. The overlapping differentially
   expressed ERGs (DEERGs) were analyzed using the R package “limma”.
   Then, the hub genes, which had a higher degree, were identified by the
   protein–protein interaction (PPI) network. Gene expression analysis
   between the TC and normal data, the disease-free survival (DFS)
   analysis of TC patients from The Cancer Genome Atlas Thyroid Cancer
   (TCGA-THCA) cohort, function analysis, and immunohistochemistry (IHC)
   were performed to verify the importance of the hub genes. Finally, a
   prognostic risk scoring was constructed to predict DFS in patients with
   the selected genes.

Results

   A total of 43 DEERGs were identified and 10 DEERGs were considered hub
   ERGs, which had a high degree of connectivity in the PPI network. Then,
   the differential expressions of FN1, ITGA2, and KIT between TC and
   normal tissues were verified in the TCGA-THCA cohort and their protein
   expressions were also verified by IHC. DFS analysis indicated
   upregulations of FN1 expression (P<0.01) and ITGA2 expression (P<0.01)
   and downregulation of KIT expression (P=0.01) increased risks of
   decreased DFS for TCGA-THCA patients. Besides, by building a prognostic
   risk scoring model, we found that the DFS of TCGA-THCA patients was
   significantly worse in high-risk groups.

Conclusion

   In summary, these hub ERGs were potential biomarkers for diagnosis and
   prognosis of TC, which can provide a basis for further exploring the
   efficacy of EMT in patients with TC.

   Keywords: thyroid cancer, EMT, signature, bioinformatic analysis,
   immunohistochemistry

Introduction

   Thyroid cancer (TC) is the most common endocrine malignancy. In the
   past 40 years, the incidence of TC has risen steadily in many countries
   around the world.[34]^1 With the improvement of detection technology,
   the incidence of recent TC in China in 2014 was 12.40/100,000, making
   it one of the fastest-growing malignant tumors.[35]^2 Although the
   incidence of TC is increasing year by year, the mortality rate of TC is
   stable or decreased slightly due to early diagnosis.[36]^3^,[37]^4
   Ultrasonography is the main method for stratifying the risk of early
   cancerous thyroid nodules. Patients with suspected TC could take a
   fine-needle aspiration biopsy (FNAB) or surgical resection, and
   followed by a pathological examination to assess the nature of
   cancer.[38]^5 However, in some cases, due to the overlapping of
   cytological features of malignant and benign thyroid nodules,
   approximately 15%-30% of biopsies are uncertain.[39]^6 And the
   uncertain diagnosis may bring inappropriate treatment leading to poor
   quality of life, such as the surgical removal of the thyroid will
   seriously affect the quality of life of patients. Ultrasound and FNAB
   can only be diagnosed and cannot respond to the question of whether to
   perform surgery. Therefore, supplementing and improving traditional
   diagnostic methods of TC patients is still an urgent problem to be
   addressed.

   Epithelial–mesenchymal transition (EMT) is one of the markers of
   carcinogenesis, which includes the re-differentiation of epithelial
   cells into mesenchymal cells, making the cell phenotype
   malignant.[40]^7–9 At present, some studies have found that EMT-related
   molecular markers are significantly correlated with TC occurrence and
   adverse outcomes.[41]^10−13 However, these studies only focused on a
   single genetic biomarker, which has been less effective in predicting
   outcomes. Studies have shown that gene signatures composed of multiple
   genes may be the better predictors of patients.

   Some researchers have studied the relationship between EMT-related
   signatures and the prognosis of patients with gastric
   adenocarcinoma,[42]^14 endometrial cancer,[43]^15 and glioma[44]^16 by
   analyzing public databases. However, there has been no bioinformatics
   research in the field of thyroid cancer. In this study, through the
   analysis of existing genetic data, a transcriptomic profile of thyroid
   cancer of EMT-related genes (ERGs) was constructed, and further
   analysis was performed, as well as exploring the potential signatures
   for diagnosis and prognostic guidance.

Materials and Methods

Data Collection

   Data screening was completed independently by two researchers, and a
   consistent comparison was conducted after completion. We searched in
   Gene Expression Omnibus (GEO) database
   ([45]http://www.ncbi.nlm.nih.gov/geo) for all published data related to
   gene expression in thyroid cancer until August 1, 2020, by using the
   following keywords: (thyroid cancer OR cancerous goiter OR carcinoma of
   thyroid OR thyroid carcinoma). Finally, a total of 4 thyroid cancer
   data sets from the GEO database ([46]GSE29265, [47]GSE33630,
   [48]GSE60542, and [49]GSE65144) were included in this study.[50]^17
   Datasets inclusion criteria: (1) they contained human TC samples and
   normal tissue samples. (2) their sample sizes at least 20. The
   information of four GEO datasets was shown in [51]Supplementary Table
   S1.

Differentially Expressed ERGs Analysis

   A list of ERGs containing 1184 ERGs was downloaded from the EMT gene
   database ([52]Supplementary Table S2).[53]^18 All data were processed
   using the R software (version 3.6.1). The limma package was used to
   identify the differentially expressed ERGs (DEERGs) between the TC
   samples and normal samples.[54]^19 The adjusted P-value < 0.05 and the
   absolute fold change (FC) ≥ 2 were considered statistically
   significant. Overlapping DEERGs between GEO datasets were selected
   using a Venn diagram
   ([55]https://bioinfogp.cnb.csic.es/tools/venny/index.html).

Functional Enrichment Analysis

   Gene Ontology (GO) can annotate genes, gene products and sequences with
   potential biological phenomena in three aspects: biological process
   (BP), cellular component (CC), and molecular function (MF);[56]^20
   Kyoto Encyclopedia of Genes and Genomes (KEGG) is a comprehensive
   database for understanding advanced functions and utilities of
   biological systems.[57]^21 In this study, GO and KEGG analysis of the
   DEERGs were performed using the R package clusterProfiler.[58]^22

Hub ERGs Selection and Analyses

   The online tool Search Tool for the Retrieval of Interacting Genes
   database (STRING) ([59]https://string- db.org/) was used to analyze the
   interactive relationships of the DEERGs.[60]^23 The combined score of
   ≥0.4 was the cut-off value. Cytoscape software (version 3.6.0) was then
   used to make PPI networks.[61]^24 And top 10 DEERGs with higher node
   degree scores were identified hub ERGs using the Cytoscape plug-in
   cytoHubba.

Verification of Differential Expression of Hub ERGs

   Raw read counts for hub ERGs of 497 TC samples and 56 normal ones from
   the TCGA database were extracted using R package TCGAbiolinks.[62]^25
   They were loaded into R package DESeq2 and normalized by
   variance-stabilizing transformation (VST).[63]^26^,[64]^27 VST counts
   were used for Student’s t‐test analysis by R package
   ggstatsplot.[65]^28 The statistical significance cut‑off level was
   P<0.01.

   Moreover, the protein levels of hub ERGs in 5 PTC tissues and paired
   normal tissues from the Shanghai Tong Ren Hospital were verified using
   immunohistochemistry (IHC) analysis. These tissues were enrolled from
   October 2018 to October 2019 and all diagnoses were based on
   pathological and/or cytological evidence. Ethical approval was obtained
   from the Ethics Committee of the Shanghai Tong Ren Hospital and
   informed consent was given by all patients before sample collection.
   Five human thyroid normal and tumor tissues were fixed in 4%
   formaldehyde and embedded in paraffin. Paraffin-embedded samples were
   cut into 6 μm thick sections, and then they were placed in xylene and
   ethanol solution for deparaffinating and rehydration in PBS. 3%
   hydrogen peroxide solution and 5% donkey serum were used to block
   endogenous peroxidase activity and nonspecific sites after samples were
   placed in the citric acid solution for antigen retrieval. Sections were
   stained with rabbit antibodies against FIBRONECTIN (Abcam, ab268021,
   1:600), ITGA2 (Abcam, ab133557, 1:200), and KIT (Abcam, ab32363, 1:200)
   for incubating at 4°C overnight, and then incubation in the second
   antibody (Abcam, ab97051, 1:1000) was carried out at room temperature
   for 2h. Antigen-antibody complexes were detected using ChemMateTM
   EnVisonTM/HRP complex (DAB) as a peroxidase substrate (GK500705, Gene
   Tech). Results were visualized under an optical microscope (OLYMPUS
   IX73, JAPAN). All pictures were taken under the same microscope camera
   (OLYMPUS DP80, JAPAN) with uniform parameters (OLYMPUS cellSens
   Standard 1.18) after sections were mounted with xylene sealant. Brown
   staining areas were selected with an eyedropper tool after dropping
   colors represented by fewer than 10 pixels, and the mean option density
   was calculated in the manner of integrated option density (IOD)/area
   using Image Pro Plus 6.0 software.

Survival Analysis of Hub ERGs, Construction of the Risk Score Model, and
Correlation Analysis Between Risk Score and Clinical Characteristics

   Clinical characteristics including age, gender, stage, disease-free
   interval (DFI) time, and DFI status for TCGA-THCA patients were
   downloaded from UCSC Xena datasets.[66]^29 After deleting patients with
   incomplete clinical characteristics and DFI time less than 0, a total
   of 350 patients were included in the disease-free survival (DFS)
   analysis. The optimal threshold was determined by the R package
   Survminer[67]^30 to divide patients into high- or low-expression groups
   and DFS was calculated using the R package survival.[68]^31 ERGs
   associated with DFS (P<0.05) were retained for subsequent multivariable
   Cox analysis to construct the ERG-related prognostic signatures. Then,
   a prognostic risk score model was established as follows: risk score =
   Σ (βn × expression of gene n). Patients were divided into low-risk
   group and high-risk group according to the median risk score, and the
   prognostic difference between the two groups was performed by the
   Kaplan–Meier curve. Harrell’s concordance index and calibration curves
   comparing predicted DFS and observed survival were used to evaluate the
   performance of the prognostic nomogram. Time-dependent receiver
   operating characteristic (ROC) curves and multivariable Cox analysis
   including patients’ clinical characteristics were calculated to
   evaluate the effectiveness of the prediction model. The relationship
   between risk score and age was analyzed through Pearson correlation
   coefficient. And the relationships between risk score and other
   clinical characteristics (gender, stage, and race), which are
   classified variables, were analyzed through t-test and ANOVA. P < 0.05
   was considered statistically significant.

Results

Identification of DEERGs in TC and Data Integration

   There were 116 DEERGs in [69]GSE29265, 147 DEERGs in [70]GSE33630, 114
   DEERGs in [71]GSE60542, 298 DEERGs in [72]GSE65144. Among the DEERGs,
   60, 86, 67, and 155 ERGs were upregulated while 56, 61, 47, and 143
   ERGs were downregulated in [73]GSE29265, [74]GSE33630, [75]GSE60542,
   and [76]GSE65144, respectively ([77]Figure 1A-[78]D). After integrated
   analyses of the four GEO datasets, 43 ERGs (21 upregulated and 22
   downregulated) were identified as the DEERGs ([79]Figure 1E and [80]F),
   and the cluster heatmap of these DEERGs is shown in [81]Figure 1G.

Figure 1.

   [82]Figure 1
   [83]Open in a new tab

   Identification of the DEERGs in TC and data integration. (A) Volcano
   plot of [84]GSE29265, (B) volcano plot of [85]GSE33630, (C) volcano
   plot of [86]GSE60542, (D) volcano plot of [87]GSE65144, (E) Venn plots
   of overlapping upregulated DEERGs, (F) Venn plots of overlapping
   downregulated DEERGs, and (G) the cluster heatmap of the overlapping
   DEERGs. Red, white, and blue represent higher expression levels, no
   expression differences, and lower expression levels, respectively. The
   rows and columns of the heatmap represent DEERGs and datasets,
   respectively, and the numbers in it represent the logFC between TC
   tissue and normal ones from each GEO dataset.

Significant Functions and Pathway Enrichment Analysis

   For 43 DEERGs, a pathway and process enrichment analysis were carried
   out using R package clusterProfiler. Terms with an adjusted P-value
   <0.05 were collected. GO analysis demonstrated that DEERGs were mainly
   related to the process of ossification, extracellular matrix (ECM), and
   cell growth ([88]Figure 2A). KEGG pathway analysis showed that DEERGs
   were mainly related to the PI3K−Akt signaling pathway, ECM−receptor
   interaction, and focal adhesion ([89]Figure 2B).

Figure 2.

   [90]Figure 2
   [91]Open in a new tab

   GO and KEGG pathway enrichment analysis. (A) GO circle plot of
   overlapping DEERGs, (B) KEGG circle plot of overlapping DEERGs.

Hub DEERGs Identification in the PPI Network

   We used the STRING database and Cytoscape to construct a PPI network to
   further analyze the interaction between the DEERGs ([92]Figure 3A).
   There were 43 nodes and 78 edges in the network. As a result,
   fibronectin 1 (FN1), the tissue inhibitor of metalloproteinase 1
   (TIMP1), integrin alpha 2 (ITGA2), receptor tyrosine kinase (KIT),
   secreted phosphoprotein 1 (SPP1), cadherin 2 (CDH2), runt-related
   transcription factor 2 (RUNX2), galectin 3 (LGALS3), bone morphogenetic
   protein 2 (BMP2), and versican (VCAN), which have a higher degree of
   connectivity according to the “Degree” algorithm, were identified as
   hub DEERGs ([93]Figure 3B).

Figure 3.

   [94]Figure 3
   [95]Open in a new tab

   The PPI network of the DEERGs. The PPI network of (A) the overlapping
   DEERGs, (B) top 10 hub DEERGs. Red nodes and blue nodes represent
   upregulated DEERGs and downregulated DEERGs, respectively in. (A). The
   size of the interaction score between genes was positively correlated
   with the depth of nodes color in (B).

Verification of Differential Expression of Hub DEERGs

   Using mRNA expression data of 497 TC samples and 56 normal ones from
   the TCGA-THCA dataset, upregulations of FN1 expression (P<0.01) and
   ITGA2 expression (P<0.01) and downregulation of KIT expression (P<0.01)
   was identified in TC samples compared with normal samples ([96]Figure
   4A-[97]C). To confirm the reliability of the hub DEERGs, the IHC was
   used to verify the protein expression levels of the 3 hub DEERGs.
   Consistent with mRNA expression, the IHC results showed that FN1 and
   ITGA2 proteins were upregulated, while KIT protein was downregulated in
   the TC tissues ([98]Figure 5A-[99]D)

Figure 4.

   [100]Figure 4
   [101]Open in a new tab

   Validation of hub DEERG expressions in the TCGA-THCA dataset. (A) FN1,
   (B)ITGA2, and (C) KIT gene expression differences between TC tissues
   and normal ones.

Figure 5.

   [102]Figure 5
   [103]Open in a new tab

   Immunohistochemical staining results of the protein expressions of hub
   DEERGs from the clinical specimens. Brown staining represents the
   expression of 3 proteins and blue staining represents nucleus. Bar:
   50um. A zoomed-in 40 magnification of each IHC picture is shown in
   better resolution at the lower left corner. (A) IHC of FN1, (B) IHC of
   ITGA2, (C) IHC of KIT, and (D) IHC of Blank.

Prognostic Value of Hub DEERGs, Construction of the Risk Score Model, and
Correlation between Risk Score and Clinical Characteristics

   A total of 350 TCGA-THCA patients were enrolled in our survival
   analysis. The univariate Cox regression results showed that the high
   expression of FN1 (P<0.01) and ITGA2 (P<0.01) and low expression of KIT
   (P=0.01) were significantly associated with the worst DFS in TC
   patients ([104]Figure 6A-[105]C). Then, the prognostic signature
   including the above-mentioned DEERGs was developed using the
   multivariate Cox regression model and risk scores for every patient
   were determined using the coefficients (−1.32 for FN1, −0.32 for ITGA2,
   and 0.69 for KIT). Then, according to the median risk score, all
   patients were divided into high-risk (n=175) or low-risk groups (n=175)
   ([106]Figure 7A). And high-risk group had worse DFS compared with
   low-risk group using the Kaplan–Meier curve (P<0.01) ([107]Figure 7B).
   The C-index of the 3-ERG signature was 0.67, and the calibration curve
   further revealed that the signature had a good performance in
   predicting the DFS of TC patients ([108]Figure 7C). ROC curves show
   that the risk score accurately predicts TC patient’s DFS ([109]Figure
   7D). In addition, the risk score calculated from the 3-ERG signature
   was an independent prognostic factor using multivariate Cox regression
   analysis ([110]Figure 7E). The correlation analysis showed that the
   risk score was negatively correlated with age (P<0.01, r=−0.17) and not
   with other clinical characteristics (gender, stage, and race)
   ([111]Supplementary Figure 1S).

Figure 6.

   [112]Figure 6
   [113]Open in a new tab

   Survival analysis of hub DEERGs using TCGA-THCA dataset. Differences in
   DFS in patients with high and expression of (A) FN1, (B) ITGA2, and (C)
   KIT.

Figure 7.

   [114]Figure 7
   [115]Open in a new tab

   Survival analysis of the 3 ERGs signature using TCGA-THCA dataset. (A)
   Risk score based on 3 ERGs signature, survival status, and heatmap for
   the expressions of 3 ERGs for each patient, (B) DFS difference between
   high-risk and low-risk groups using the Kaplan–Meier curve, (C)
   calibration curves for 1-, 3-, and 5-year from the 3 ERGs signature,
   (D) ROC curves for 1-, 3-, and 5-year DFS predictions for the 3 ERGs
   signature using the ROC curves, and (E) forest plot of the risk score
   using multivariate Cox regression analysis.

Discussion

   Thyroid cancer (TC) is the most common endocrine malignant tumor, and
   its incidence increases year by year.[116]^32 Despite good prognosis,
   patients with recurrence or metastasis still have a high risk of
   death.[117]^33^,[118]^34 Some studies have demonstrated that the
   aggressiveness of thyroid cancer is closely related to processes such
   as epithelial–mesenchymal transformation (EMT).[119]^34–37 At present,
   many studies have constructed ERG signatures to predict the survival of
   cancer patients. Zhong et al developed a reliable EMT-related lncRNA
   risk signature, which can independently predict the OS and DFS of
   ccRCC.[120]^14 In hepatocellular carcinoma, a total of 5 prognostic
   correlated ERGs were included to develop a novel prognostic risk
   model.[121]^38 Cao et al developed a novel EMT-related gene signature
   as an independent prognostic factor for bladder cancer.[122]^39
   However, researches on the relationship between ERGs and the diagnosis
   and prognosis of TC patients are still very limited. In this study, for
   the first time, we used the ERG signature with higher prediction
   accuracy to predict the survival of TC patients and achieved good
   prediction results.

   In this study, we comprehensively analyzed four microarray datasets
   from GEO, which were all published data related to gene expression in
   thyroid cancer and normal ones in the GEO database until August 1,
   2020. A total of 43 DEERGs (21 upregulated and 22 downregulated) were
   identified between TC tissues and normal ones. Functional enrichment
   analyses demonstrated that DEERGs were mainly related to the process of
   ossification, extracellular matrix (ECM), and PI3K−Akt signaling
   pathway. In patients with thyroid cancer, there are changes in
   thyroid-stimulating hormone, which is closely associated with
   osteogenesis.[123]^40–43 The ECM plays an important role in the process
   of tumor shedding, adhesion, movement, degradation, and hyperplasias,
   such as prostate cancer,[124]^44 gastric cancer,[125]^45 breast
   cancer,[126]^46 and anaplastic thyroid cancer.[127]^47 The
   phosphoinositide 3-kinase–protein kinase B/AKT (PI3KPKB/AKT) pathway is
   one of the most prominent molecular signaling pathways implicated in
   cellular growth, proliferation, apoptosis, and metabolism.[128]^48
   Activation of the PI3K-AKT signaling pathway is critical to the
   occurrence and development of thyroid cancer.[129]^49–51

   We finally screened out that FN1, ITGA2, and KIT were hub DEERGs from
   the PPI network, TCGA differential expression verification, and IHC
   verification. Otherwise, the DFS analyses on the TCGA-THCA dataset
   showed that upregulations of FN1 expression (P<0.01) and ITGA2
   expression (P<0.01) and downregulation of KIT expression (P=0.01)
   increased risks of decreased DFS for patients. Fibronectin 1 (FN1)
   participates in cell adhesion and migration processes and is expressed
   in multiple cell types.[130]^52 FN1 has also turned out to be
   associated with the ECM changes.[131]^53 In thyroid cancer, Sponziello
   M et al demonstrated that the overexpression of FN1 mediated thyroid
   tumor cell migration and invasion by 86 patients.[132]^54 Shaohua Zhan
   et al identified FN1 as a novel prognostic biomarker associated with
   sporadic medullary thyroid cancer pathophysiological changes.[133]^55
   Furthermore, some studies found FN1 was highly expressed in TC from a
   series of data sets, but they did not offer a survival analysis of
   FN1.[134]^56–59 Integrin alpha 2 (ITGA2) is an alpha subunit, which
   often combines with the beta subunit to form a heterodimer α2β1, and
   then participates in the adhesion of platelets and other cells to the
   extracellular matrix.[135]^60–62 More studies suggested that ITGA2
   might be closely related to tumor cell migration, invasion, and
   metastasis.[136]^63^,[137]^64 In thyroid cancer, one study identified
   that ITGA2 expression was upregulated in PTC tissues but not in
   BRAF-positive samples,[138]^65 and another study showed that PTC
   patients with high ITGA2 expression had poorer relapse-free survival
   than PTC patients with low ITGA2 expression.[139]^66 Receptor tyrosine
   kinase (KIT), a form of myeloid receptor that binds the stem cell
   factor, plays a major role in cancer occurrence.[140]^67 Papers are
   showing that KIT is highly expressed in small cell lung cancer,[141]^68
   leukemia cells,[142]^69 colon cancer,[143]^70 and
   neuroblastoma.[144]^71 But KIT expression is lower in breast
   cancer[145]^72 and melanoma.[146]^73 Numerous studies have investigated
   the expression of KIT in TC,[147]^74–76 suggesting its role in thyroid
   epithelial cell differentiation and growth control. These findings
   indicate that different ERGs jointly promote the occurrence of EMT, and
   these ERGs may have different functions in different tumors.

   Several previous studies have constructed prognostic signatures of TC
   patients. Most research has focused on immune-related genes, like Gan
   et al[148]^77 Li et al[149]^78 Peng et al,[150]^79 and so on. They
   established different immune-related signatures to predict the
   prognosis of patients with TC. The remaining thyroid prediction models
   mainly focused on the study of characteristic gene sets are mainly
   m6A-related signature[151]^80^,[152]^81 and autophagy-related
   signature.[153]^82 These results indicate that the characteristic
   signature not only has a good prediction effect on the prognosis of
   thyroid cancer patients but is more conducive to the comprehensive
   interpretation of the occurrence and development of TC and provides a
   theoretical basis for the use of drugs that act on this characteristic.
   However, no studies have been conducted to analyze the EMT
   characteristics of TC, which is closely related to the occurrence and
   development of TC, and establish relevant prediction signatures. In
   this study, a 3 EMT-related genes were used to construct a prognostic
   risk signature for TC patients. The univariate and multivariate Cox
   regression results showed that high-risk group had worse DFS in TC
   patients.

   The study has several advantages. First, to our knowledge, this is the
   first study to analyze the EMT gene expression profile of thyroid
   cancer and its function. In addition, our research has constructed an
   excellent ERG signature as a potential predictor of TC patients, which
   helps to develop new strategies for diagnosing and predicting the
   prognosis of TC patients. Finally, calibration curves were used for
   internal validation of the signature, and the results showed that the
   signature’s prediction effect was good. However, our observations still
   have some limitations: 1) as the cumulation of TC samples, subtype
   analysis was required to perform; 2) because patients with TC have a
   longer survival period and fewer patients have observed death outcomes,
   this study only performed DFS analysis; 3) and the constructed
   predictive model has not been validated in a prospective cohort.

Conclusions

   In conclusion, this study used existing data to construct a more
   comprehensive transcriptomic profile of TC and found potential
   biomarkers for TC diagnosis and prognosis. However, it is necessary to
   further study the specific role of these genes in TC.

Funding Statement

   This work was supported by Startup Fund for Youngman Research at SJTU
   (to TF, 17×100040015); Shanghai Jiao Tong University Medical and
   Industrial Cross Project (to TF, YG2017QN70); Young Talent Program of
   China National Nuclear Corportion (to SJ, CNNC201948); and The
   Scientific Project of Shanghai Municipal Health Commission (to BY,
   Grant No.2018ZHYL0202).

Abbreviations

   TC, thyroid cancer; EMT, epithelial–mesenchymal transition; ERGs,
   EMT-related genes; GEO, Gene Expression Omnibus; DEERGs, differentially
   expressed ERGs; PPI, protein–protein interaction; DFS, disease-free
   survival; TCGA-THCA, The Cancer Genome Atlas Thyroid Cancer; IHC,
   immunohistochemistry; FNAB, fine-needle aspiration biopsy; PTC,
   papillary thyroid cancer; ATC, anaplastic thyroid carcinoma; FC, fold
   change; GO, Gene Ontology; BP, biological process; CC, cellular
   component; MF, molecular function; KEGG, Kyoto Encyclopedia of Genes
   and Genomes; STRING, Search Tool for the Retrieval of Interacting
   Genes; VST, variance-stabilizing transformation; DFI, disease free
   interval; ROC, receiver operating characteristic; ECM, extracellular
   matrix; FN1, fibronectin 1; TIMP1, the tissue inhibitor of
   metalloproteinase 1; ITGA2, integrin alpha 2; KIT, receptor tyrosine
   kinase; SPP1, secreted phosphoprotein 1; CDH2, cadherin 2; RUNX2,
   runt-related transcription factor 2; LGALS3, galectin 3; BMP2, bone
   morphogenetic protein 2; VCAN, versican; PI3KPKB/AKT, phosphoinositide
   3-kinase–protein kinase B/AKT.

Author Contributions

   All authors made a significant contribution to the work reported,
   whether that is in the conception, study design, execution, acquisition
   of data, analysis and interpretation, or in all these areas; took part
   in drafting, revising or critically reviewing the article; gave final
   approval of the version to be published; have agreed on the journal to
   which the article has been submitted; and agree to be accountable for
   all aspects of the work.

Disclosure

   The authors report no conflicts of interest in this work.

References