Gene expression profiling identifies potential molecular markers of papillary thyroid carcinoma

Abstract

BACKGROUND:

Thyroid cancer is the most common endocrine malignancy worldwide, with the predominant form papillary thyroid carcinoma (PTC) representing approximately 80% of cases.

OBJECTIVE:

This study was addressed to identify potential genes and pathways involved in the pathogenesis of PTC and potential novel biomarkers for this disease.

METHODS:

Gene expression profiling was carried out by DNA microarray technology. Validation of microarray data by qRT-PCR, western blot, and enzyme linked immunosorbent assay was also performed in a selected set of genes and gene products, with the potential to be used as diagnostic or prognostic biomarkers, such as those associated with cell adhesion, extracellular matrix (ECM) remodeling and immune/inflammatory response.

RESULTS:

In this study we found that upregulation of extracellular activities, such as proteoglycans, ECM-receptor interaction, and cell adhesion molecules, were the most prominent feature of PTC. Significantly over-expressed genes included SDC1 (syndecan 1), SDC4 (syndecan 4), KLK7 (kallikrein-related peptidase 7), KLK10 (kallikrein-related peptidase 10), SLPI (secretory leukocyte peptidase inhibitor), GDF15 (growth/differentiation factor-15), ALOX5 (arachidonate 5-lipoxygenase), SFRP2 (secreted Frizzled-related protein 2), among others. Further, elevated KLK10 levels were detected in patients with PTC. Many of these genes belong to KEGG pathway “Proteoglycans in cancer”.

CONCLUSIONS:

Using DNA microarray analysis allowed the identification of genes and pathways with known important roles in malignant transformation, and also the discovery of novel genes that may be potential biomarkers for PTC.

1. Introduction

The American Cancer Society estimates that 53,990 new cases of thyroid cancer are expected in the United States in 2018, making this cancer the most common endocrine malignancy, representing 3.1% of all new cancer cases in United States [42]. Histopathological features distinguish four main types of thyroid cancer: papillary, follicular, medullary, and anaplastic, with papillary thyroid carcinoma (PTC) accounting for 80% of cases [8]. Although PTC is characterized by slow growth and good prognosis, about 10–15% of cases exhibit aggressive behavior, including local invasion, distant metastasis and treatment resistance. In the presence of distant metastasis, considered the most important factor for aggressiveness, median survival estimates is 4.1 years, and the 10-year disease-specific survival rate is 26% [41]. Accurate classification of thyroid cancer is the primary step for assessment of prognosis and treatment selection. However, in some cases, histopathological classification is difficult to accomplish. In this regard, the Cancer Genome Atlas Research Network performed a comprehensive analysis of the PTC genomic landscape, and proposed a reclassification of thyroid cancers into molecular subtypes to better improve pathological classification and management of this disease [1]. According to that study, PTC has been classified as a Mitogen-Activated Protein Kinase (MAPK)-driven tumor with the two major signaling drivers being BRAF ${}^{\text{V600E}}$ and mutated RAS, with approximately 50% of PTC carrying the BRAF ${}^{\text{V600E}}$ mutation [44]. Despite all the advances made in recent years at the genomic and transcriptomic level increasing knowledge on the molecular pathogenesis of PTC and development of molecular-based management strategies [51], a complete understanding of its etiology and molecular basis is still lacking [54]. To date, the only FDA-approved protein tumor marker for thyroid cancer currently used in clinical practice is thyroglobulin, which is used to aid in monitoring therapy [24]. More recently, the FDA has approved four different drugs targeting MAPK signaling pathways in the treatment of advanced thyroid cancers [32]. Thus, more research is needed to find and validate markers able to distinguish PTC from benign thyroid lesions or other thyroid cancer types, and/or that can be used for therapeutic intervention, prognosis, staging and monitoring the disease. Currently, fine needle aspiration biopsy (FNAB) is the most appropriate procedure for evaluation of patients with suspected PTC, but sensitivity of this technique is insufficient and in some cases the results are difficult to interpret [36]. More recently, FNAB has been combined with molecular analysis to increase accuracy of cytological evaluation [35]. BRAF ${}^{\text{V600E}}$ has shown to be a highly specific and unique mutation in PTC, and testing for it in FNAB specimens improves diagnostic accuracy of PTC with indeterminate cytology [44]. Molecular biomarkers identified by transcriptomic techniques could also be used to improve FNAB accuracy. Previous studies have used transcriptomic techniques to identify differentially expressed genes in PTC, however, most of them have not been validated by other techniques nor have proven to be sensitive or specific enough to be clinically useful markers [14, 20, 23, 48]. In this study, DNA microarray analysis was used to identify differential expression between PTC and normal thyroid tissue, focusing on genes suitable as diagnostic and prognostic markers, such as those involved in cell adhesion, extracellular matrix (ECM) remodeling, and inflammatory response. A total of 205 genes were found differentially expressed in PTC, including some genes not previously associated with this type of cancer. A set of the overexpressed genes were further validated, including kallikrein-7 and kallikrein-10. The identification and validation of kallikrein-7 and kallikrein-10 in this study is significant because kallikrein-related peptidases have been considered as very promising biomarkers for cancer personalized medicine, especially for prediction and monitoring of patients’ response to treatment [11, 25] making them suitable as clinical biomarkers for PTC. The data discussed in this publication have been deposited in NCBIs Gene Expression Omnibus with GEO Series accession number GSE3678.

2. Methods

2.1 Patients and ethics statement

Patients diagnosed with stage II/III PTC during thyroidectomy at the New York Eye and Ear Infirmary (New York, NY, USA) and Westchester Medical Center (Valhalla, NY, USA), were considered eligible for the study. Exclusion criteria included previous chemotherapy or radiotherapy for any carcinoma. The study protocol was approved by the Institutional Review Board committee at each institution and written informed consent was obtained from each patient.

2.2 Tissue samples and total RNA isolation

Tumor/normal paired thyroid tissue samples were obtained from PTC patients. The samples were harvested and frozen in dry ice/ethanol bath immediately after surgical removal and stored at $-$ 80 ${}^{\circ }$ C until use. PTC was diagnosed according to WHO criteria by a certified pathologist, with each tumor sample containing more than 90% tumor tissue and each normal sample containing more than 90% normal tissue. Total RNA was prepared from frozen tissue specimens by homogenization with a Brinkmann Polytron at top speed in Trizol solution (Invitrogen, Carlsbad, CA, USA) and stored as an ethanol precipitate. RNA concentration and quality were assessed spectrophotometrically by the 260/280 ratio.

2.3 Microarray procedures

RNA quality from each sample was first assessed by electrophoresis using the Agilent Bioanalyzer 2100 and spectrophotometric analysis. Then, RNA amplification, and cDNA synthesis and labelling were performed at Functional Genomics Center, University of Rochester, according to Affymetrix techniques (Santa Clara, CA, USA). One $\mu$ g of cDNA was converted to biotinylated cRNA and hybridized to Affymetrix high-density oligonucleotide GeneChip Human Genome U133 plus 2.0 arrays. Arrays were automatically washed after hybridization; stained with streptavidin-phycoerythrin conjugate in the GeneChip Fluidics Station and scanned using Hewlett-Packard GeneArray Scanner equipped with an argon ion laser.

2.4 Microarray data preprocessing and analysis

Arrays were classified into Tumor group (from PTC tumor tissue samples) and Normal group (from normal paired thyroid tissue samples). Affymetrix CEL files were subjected to quality control, pre-processing and GC-RMA (GeneChip robust multi-array average) normalization, using the automatic R pipeline AffymetrixQC (http://www.arrayanalysis.org) [10]. Statistical analysis of expression data was also performed in www.arrayanalysis.org using the statistical analysis module [2]. Probeset IDs were converted to HUGO gene symbols using BioMart data mining tool. Gene lists were generated for genes that were differentially expressed between the two groups, either over-expressed or under-expressed, with a criterion of fold change of 2.0 ( $⩽$ or $⩾$ 2.0), and FDR-corrected $p$ value $<$ 0.05 (Benjamini and Hochberg method).

2.5 Functional group and pathway enrichment analysis

To identify functional groups and biological pathways implicated in PTC, the gene lists of differentially expressed genes were subjected to Gene Ontology (GO) functional and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses using the online Database for Annotation, Visualization and Integrated Discovery (DAVID) (https://david. ncifcrf.gov) [9], with the criterion of $p$ -value $<$ 0.05 and enrichment gene count $>$ 5. The online DAVID platform has been used previously for systematic and integrative analysis of large gene lists [17, 18].

2.6 Quantitative real time RT-PCR (qRT-PCR)

qRT-PCR was performed for a group of selected genes associated with ECM remodeling and the inflammatory response: kallikrein-related peptidase-7 (KLK7), kallikrein-related peptidase-10 (KLK10), arachidonate 5-lipoxygenase (ALOX5), growth differentiation factor-15 (GDF-15), and secreted frizzled-related protein-2 (SFRP2). Gene-specific primers were designed with Primer3 software (http://frodo.wi.mit.edu/ primer3/) and their sequences are shown in Table S1 (Supplementary material). One microgram of total RNA was used as template for cDNA synthesis using the Reverse Transcription System kit (Promega, CA, USA). qRT-PCR reactions and measurements were performed with LightCycler DNA Master SYBR Green system (Roche Molecular Biochemicals, Mannheim, Germany) in a LightCycler 2.0 instrument (Roche, Mannheim, Germany) with a 2 min pre-incubation step at 95 ${}^{\circ }$ C followed by 35 cycles of amplification steps. PCR products were subjected to melting curve analysis to verify absence of amplification of non-specific products. In each qRT-PCR run, a standard curve was generated using a serial dilution of a cDNA standard identical in size and sequence to the target gene fragment to be amplified. Serial dilution ranged from 0.01 to 100 pg of the standard cDNA fragment. The standard curve was calculated with LightCycler 5.32 software (LC-Run Version 5.32, Roche) by regression of the crossing points of the PCR curves from the dilution series of the standard cDNA fragment. Changes in gene expression were quantified by calculating the average value of the reactions from the PTC tumor samples group and comparing that to the average derived from the reactions from the normal thyroid tissue group. GAPDH was used as reference gene for normalization, since no change in GAPDH mRNA levels was observed across samples (tumor and normal samples). Statistical significance of changes was determined with $t$ -test.

2.7 Western blot (WB)

To correlate mRNA expression levels with protein levels, WB analysis was performed using standard protocols with antibodies against protein markers kallikrein-7, kallikrein-10, growth/differentiation factor-15 (GDF-15), and secretory leukocyte peptidase inhibitor (SLPI protein). Tumor/normal paired thyroid tissue samples from the PTC patients were coarsely grounded in a mortar under liquid nitrogen and washed twice in PBS containing 1 $\mu$ M complete protease inhibitor (Sigma Chemical Company, St. Louis, MO, USA) to remove excess hemoglobin. Total protein extraction was performed with RIPA buffer. Protein concentration was determined by Bio-Rad Protein Assay (Bio-Rad, Hercules, CA, USA). Twenty five micrograms of total protein were loaded and resolved on 12% SDS-PAGE gels. Gels were transferred at 200 mA for 2 h onto PVDF membranes (Millipore, Bedford, MA, USA). Membranes for kallikrein-10, growth/differentiation factor 15, and secretory leukocyte peptidase inhibitor were blocked with 4% milk, and membranes for kallikrein-7 with 4% BSA. Blocked membranes were subsequently incubated overnight with respective antibodies (R&D Systems, Minneapolis, MN, USA): anti-kallikrein-7 rabbit polyclonal (1:500), anti-kallikrein-10 goat polyclonal (1 $\mu$ g/ml), anti-GDF-15 goat polyclonal (1 $\mu$ g/ml), and anti-SLPI mouse monoclonal (2 $\mu$ g/ml). Membranes were incubated at room temperature with either secondary donkey anti-goat HRP-conjugated antibody (R&D Systems, Minneapolis, MN, USA) (1:5,000) or anti-mouse HRP-conjugated antibody (Pierce, Rockford, IL, USA) (1:10,000) for kallikrein-10, GDF-15, and SLPI; or anti-rabbit alkaline phosphatase conjugated secondary antibody (KLP, Gaithersburg, MD, USA) for kallikrein-7. To visualize bands for kallikrein-10, GDF-15 and SLPI, membranes were developed with Super Signal ${}^{\mathrm{\circledR }}$ West Pico Chemiluminescent Substrate (Pierce, Rockford, IL, USA) and bands for kallikrein-7 were developed with Western Blue ${}^{\mathrm{\circledR }}$ Stabilized Substrate for Alkaline Phosphatase (Promega, Madison, WI, USA). $\beta$ -actin protein was used as loading control for normalization. Band intensity was quantified by densitometry with ImageJ software (https://imagej.nih.gov/ij/index.html) and signals were normalized to $\beta$ -actin values.

2.8 Enzyme-linked immunosorbent assay (ELISA)

ELISA assay was performed to validate selected candidate serum protein biomarkers in the sera of PTC patients and healthy controls using commercial kits for kallikrein-10 (IBEX Pharmaceuticals, Montreal, Quebec, Canada) and SLPI (R&D Systems, Minneapolis, MN, USA). ELISA assay kits were run following manufacturer’s instructions. Individual samples were assayed in triplicates, and absorbance of each well was measured at 450 nm with a ELISA plate reader. Concentrations were calculated using a four parameter logistic (4-PL) curve fit of the standard curves using GraphPad Prism 5.0 (GraphPad Software, Inc., San Diego, CA, USA). Kallikrein-10 and SLPI serum levels were analyzed in 13 PTC patients and 8 age-matched healthy donors. Serum samples from PTC patients were collected pre-surgically and stored at $-$ 80 ${}^{\circ }$ C until use. Serum samples from 5 PTC patients were additionally collected one month after surgery. Patients used for ELISA assays were different from the patients used in the DNA microarray study.

2.9 Statistics

Statistics were performed using GraphPad Prism ${}^{\mathrm{\circledR }}$ v5.00 (GraphPad Software Inc, San Diego, CA, USA). Data normality was assessed using D’Agostino/Pearson Omnibus normality tests. For qRT-PCR, statistical significance of differences in gene expression for each target gene between PTC tumor and paired normal samples was analyzed by Student’s $t$ -test; all observations were derived from at least 3 independent experiments. For WB results, differences between groups (PTC and Normal Tissue) were assessed using Mann-Whitney test (non-normally distributed data). For ELISA results, differences between groups (Healthy Controls, PTC, and PTC-PS) were assessed using a Student’s $t$ -test (normally distributed data). Statistical significance was defined as $p<$ 0.05.

3. Results

3.1 Patients and pathological findings

Age range was 23–74 years (average: 43.6); 6 females and 1 male; four were diagnosed with stage III, two with stage II, and one with stage I; four patients tested positive for BRAF ${}^{\text{V600E}}$ mutation.

Table 1

Genes validated in PTC

Gene name	Protein name	Fold change
		Microarray	qPCR	* $p$ -value	Protein level (WB)	${}^{+}p$ -value
KLK7	Kallikrein-7	78,4	19.0	0.03990	3.7	0.02390
KLK10	Kallikrein-10	17,1	31.6	0.03721	3.7	0.04926
ALOX5	Arachidonate 5-lipoxygenase	15,1	6.1	0.04905	N.P	–
GDF15	Growth/differentiation factor 15	4,9	2.5	0.03069	1.8	0.03800
SFRP2	Secreted frizzled-related protein 2	3,4	5.5	0.03124	N.P	–
SLPI	Secretory leukocyte protease inhibitor	8,9	N.P	–	2.4	0.03970

Statistical significance was determined using: * $t$ test and ${}^{+}$ Mann-Whitney test. N.P: assay not performed.

Figure 1.

Western blots of a set of proteins over-expressed in PTC. Western blot assays were carried for protein markers kallikrein-7, kallikrein-10, growth/differentiation factor 15, and antileukoproteinase/secretory leukocyte protease inhibitor in paired tumor/normal thyroid tissue samples from PTC patients. $\beta$ -actin was used as loading control. Densitometry analysis of bands was carried out to evaluate the difference in expression of protein markers between PTC and Normal samples, and showed that overall signal intensities for all the evaluated proteins were significantly higher in the PTC tumor samples compared to the normal samples, with $p<$ 0.05.

3.2 Identification of differentially expressed genes

Preprocessing and statistical analysis of DNA microarray data generated from paired tumor and normal tissue from seven papillary thyroid carcinoma patients identified 20,111 genes: 205 were differentially expressed between tumor and normal tissue, 104 over-expressed and 101 under-expressed, with $|\text{FC}|$ (absolute value of fold change) $>$ 2 and $p$ value $<$ 0.05. The highest over-expressed and under-expressed genes are shown in Tables S2 and S3 respectively (supplementary material).

3.3 Functional and pathway enrichment analysis

According to DAVID online tool, a total of 26 GO functions belonging to the Biological Process (BP) category were enriched with a count of at least 5 genes in each GO term and $p$ value $<$ 0.05. Over-expressed genes were mainly enriched in 14 GO functions and under-expressed genes were mainly enriched in 12 GO functions. The top GO terms in each category are shown in Table S4 (supplementary material). The most significant GO function in the over-expressed genes was signal transduction, and the most significant one in the under-expressed was oxidation-reduction process. The top KEGG pathways in each category, over-expressed and under-expressed, are shown in supplementary material (Tables S5 and S6, respectively). KEGG pathways of over-expressed genes were significantly enriched in Proteoglycans in cancer ( $p=$ 0.013118572), ECM-receptor interaction ( $p=$ 0.003226258), Cell adhesion molecules (CAMs) ( $p=$ 0.0177183), and Complement and coagulation cascades ( $p=$ 0.012650834). KEGG pathways of under-expressed genes were significantly enriched in Thyroid hormone synthesis ( $p=$ 0.000124594), Insulin resistance ( $p=$ 0.006968337), Oxytocin signaling pathway ( $p=$ 0.025126328), Adipocytokine signaling pathway ( $p=$ 0.013152301), and Thyroid hormone signaling pathway ( $p=$ 0.046572382). Genes overexpressed in PTC belonging to the Proteoglycans in cancer pathway included the proteoglycan encoding genes SDC1 (syndecan-1) and SDC4 (syndecan-4); FN1 (encoding the extracellular matrix glycoprotein fibronectin-1), MET (encoding the proto-oncogene receptor tyrosine kinase MET), ERBB3 (encoding the member of the epidermal growth factor receptor family of receptor tyrosine kinases, erb-b2 receptor tyrosine kinase 3), Tiam1 (encoding the guanine nucleotide exchange factor (GEF) T cell lymphoma invasion and metastasis-1). Some of the over-expressed genes were common to several related pathways, such as SDC1, SDC4, FN1, that also belong to ECM-receptor interaction and Cell adhesion molecules pathways.

Figure 2.

Serum hK10 levels in PTC patients. Serum hK10 levels were measured by ELISA in three groups of individuals: Healthy Controls ( $n=$ 8), PTC patients ( $n=$ 13) and PTC-PS patients (PTC patients post-surgery; $n=$ 5). Box plots for each group show the minimum and maximum values. The lines in the box plots indicate the median serum hK10 levels of each group. The difference in hK10 levels between Healthy Controls and PTC or PTC and PTC-PS groups were analyzed using a Student $t$ -test. A $p$ value $<$ 0.05 was considered statistically significant.

3.4 Identification of genes with potential use as biomarkers for PTC

To identify novel genes not previously mentioned in PTC, a literature search was performed for genes with the highest over-expression in PTC compared to normal thyroid tissue (Table S2, supplementary material). Many of the genes listed have been previously identified in other published microarray expression studies [23, 43], validating our results. However, most of these genes have not been validated by other methods. Thus, we selected six of these genes and corresponding encoded proteins for validation by qPCR and/or WB: KLK7, KLK10, ALOX5, SLPI, GDF-15, and SFRP2. qRT-PCR and WB analysis confirmed the increased levels of these genes in PTC compared to normal paired thyroid tissues (Table 1; Fig. 1). Consistent with microarray results, qPCR experiments showed that expression of KLK7, KLK10, ALOX5, GDF-15, and SFRP2, were significantly increased in PTC compared to normal tissue. Densitometry analysis of bands from WB analysis showed that signal intensities for all the evaluated proteins were significantly higher in the tumor samples compared to the corresponding paired normal samples, with overall fold changes of 3.7 for kallikrein-7 ( $p=$ 0.0239); 3.7 for kallikrein-10 ( $p=$ 0.04926); 1.8 for GDF-15 ( $p=$ 0.0380); and 2.4 for SLPI ( $p=$ 0.0397). KLK-10 over-expression was found in PTC tissue samples compared to normal paired tissue, both at transcript and protein levels. Levels of two secreted protein biomarkers, kallikrein-10 and SLPI protein, were further measured in serum of PTC patients and healthy controls.

3.5 Serum levels of selected protein markers

ELISA assays were performed twice and similar results were obtained. Serum kallikrein-10 levels in healthy controls were consistent with previous reports [28]. Kallikrein-10 serum level was significantly higher in PTC patients relative to healthy controls (Fig. 2). The mean and median values of serum kallikrein-10 in healthy controls were 230 and 142 ng/lt, respectively. Pre-surgery serum kallikrein-10 levels in the PTC group was significantly higher than levels from the Healthy Control group, with mean and median of 457 and 449 ng/liter, respectively ( $p=$ 0.0465; Student $t$ -test). Post-surgery serum kallikrein-10 levels in PTC patients were lower than pre-surgical levels, although the difference was not statistically significant ( $p>$ 0.05). Higher serum kallikrein-10 levels in PTC patients (pre-surgery and post-surgery) compared to healthy controls suggests the potential of kallikrein-10 as a diagnostic or prognosis indicator. The sample size used in this study was rather small, limiting this study. Larger sample size is required to determine the real value of kallikrein-10 as biomarker for PTC. Evaluation of serum SLPI levels in PTC patients (mean $=$ 64 ng/ml; median $=$ 66 ng/ml) and normal controls (mean $=$ 70 ng/ml; median $=$ 75 ng/ml) did not find significant differences ( $p>$ 0.05) making this protein not suitable for PTC screening in serum samples.

4. Discussion

Whole transcriptome analysis has potential to identify novel biomarkers for diseases like cancer, and is beginning to impact real-time clinical decision-making [38]. In the present study, DNA microarray analysis identified genes and biological pathways differentially expressed between PTC and normal thyroid tissue. According to our results, the most apparent feature of PTC was upregulation of extracellular activities, such as proteoglycans, ECM-receptor interaction, and cell adhesion molecules, while downregulated pathways were involved with thyroid hormone biosynthesis and signaling pathways. Among over-expressed genes encoding components of the ECM were SDC1, SDC4, FN1, c-MET, erbB3, and Tiam1; upregulation of these genes has been previously reported in PTC [5, 15, 19, 34, 47], supporting the reliability of our microarray data. Syndecan-1 is mainly found on the basolateral surface of epithelial cells in adult tissues and syndecan-4 is ubiquitously expressed on most tissues [4]. Syndecan-4 binds to fibronectin leading to the activation of focal adhesion kinase (FAK), a tyrosine kinase that regulates focal adhesions [49, 50]. Binding of syndecan-1 and syndecan-4 to fibronectin has been shown to regulate tumor cell ECM attachment and growth [46]. c-MET is a receptor tyrosine kinase that regulates different cellular signaling pathways, including those involved in proliferation, motility, migration and invasion, which has been found to be aberrantly activated in many human cancers, including thyroid cancer [33, 43]. ERBB3 has been implicated in feedback regulation of MEK/ERK signaling in PTC driven by BRAF ${}^{\text{V600E}}$ mutation [13]. Tiam1 is a guanine nucleotide exchange factor (GNEF) that specifically activates Rac1, member of Rho family of GTPases [30].

Different studies have shown that high expression of Tiam1 protein is associated with progression of several cancer types [31], and it has been suggested as prognostic factor and potential therapeutic target for thyroid cancers [16, 27]. Tiam1 has also been shown to bind syndecans promoting cell-matrix adhesion and cell migration, establishing a novel link between Tiam1 and syndecans, two previously unrelated signal transduction pathways implicated in cancer [22, 39]. We found over-expression in PTC of all these genes belonging to KEGG pathway “Proteoglycans in cancer”. Our microarray data is strongly supported by the scientific literature on the involvement of these genes in thyroid carcinoma.

Functional group enrichment analysis in the context of GO_BP terms found that many of the significant genes were grouped into categories associated with cell adhesion, extracellular matrix organization, inflammatory response, activation of MAPK activity, among others. Two of the most upregulated genes in PTC were KLK7 and KLK10, two members of the tissue kallikrein (KLK) gene family of conserved serine proteases located in human chromosomal region 19q13.3–q13.4 [7], involved in cell adhesion and ECM remodeling. The encoded serine proteases have diverse expression patterns and physiological roles including activation of other proteases, degradation of extracellular matrix (ECM) components, and activation of growth and angiogenic factors [7, 53].

KLKs have been explored as promising biomarkers for different cancer types, especially for prediction and monitoring of patients’ response to chemotherapy [25]. Human kallikrein-3 (prostate-specific-antigen/PSA) and kallikrein-2 have clinical application as biomarkers for prostatic diseases [29]. Further evaluation of KLK10, both in tissue and serum, in larger cohorts of PTC patients and healthy controls are required to determine the true value of this protein in management of thyroid cancer. Another gene upregulated in PTC was SLPI (secretory leukocyte peptidase inhibitor), which encodes a protein that is a potent inhibitor of leukocyte elastase, cathepsin G, chymotrypsin, and trypsin; SLPI also has been shown to be a specific inhibitor of kallikrein-7 [12]. SLPI is produced and released into mucus by secretory cells in the parotid, bronchus, cervix, and testicular gland, playing physiological roles against proteolytic degradation of these tissues. In cancer tissues, increased expression of kallikrein-7 and decreased expression of SLPI have been found in cervical adenocarcinoma, but in ovarian tumors both proteins have been found overexpressed [40, 45].

This study identified additional genes which may represent potential markers for PTC, as they are associated with cell-extracellular matrix interaction (ECM), which is important in cell growth promotion and survival; these genes included secreted Frizzled-related protein 2 (SFRP2), proprotein convertase subtilisin/kexin type 2 (PCSK2), cathepsin-C (CTSC), and cathepsin-H (CTSH). SFRP2 gene encodes a secreted glycoprotein member of the SFRP family that function as Wnt receptors. Wnt signaling is implicated in cell proliferation, apoptosis, and onset of neoplasia. Methylation of SFRP2 gene has been associated with esophageal adenocarcinoma and colorectal cancer [52, 55].

We also found overexpression of some important genes associated with mechanisms other than epithelial-ECM interaction, such as immune/inflammatory reaction, antiapoptosis and angiogenesis, including ALOX5 and GDF-15. ALOX5 is a lipoxygenase implicated in several metabolic pathways and it has been recently associated with cell proliferation and neo-angiogenesis [37]. We previously reported ALOX5 overexpression in PTC and documented its correlation with invasive tumor histopathology [26]. GDF-15, which has been linked to prostate cancer [21], was first isolated from a differential screen for genes that were induced in macrophage activation [6]; it is member of TGF- $\beta$ superfamily associated with immune surveillance suppression, facilitation of tumor invasion, and promotion of metastasis development in tumors of epithelial origin [3].

Genes identified in this study have been shown to play important roles in malignant transformation, tumor progression and phenotype of PTC. Additional studies with larger sample size and detailed clinical characterization of patients are required to evaluate the biological role and prognostic value in PTC of some of these genes and the encoded proteins.