Molecular Determinants of Thyroid Nodules with Indeterminate Cytology and RAS Mutations

Abstract

Background: RAS gene family mutations are the most prevalent in thyroid nodules with indeterminate cytology and are present in a wide spectrum of histological diagnoses. We evaluated differentially expressed genes and signaling pathways across the histological/clinical spectrum of RAS-mutant nodules to determine key molecular determinants associated with a high risk of malignancy.

Methods: Sixty-one thyroid nodules with RAS mutations were identified. Based on the histological diagnosis and biological behavior, the nodules were grouped into five categories indicating their degree of malignancy: non-neoplastic appearance, benign neoplasm, indeterminate malignant potential, low-risk cancer, or high-risk cancer. Gene expression profiles of these nodules were determined using the NanoString PanCancer Pathways and IO 360 Panels, and Angiopoietin-2 level was determined by immunohistochemical staining.

Results: The analysis of differentially expressed genes using the five categories as supervising parameters unearthed a significant correlation between the degree of malignancy and genes involved in cell cycle and apoptosis (BAX, CCNE2, CDKN2A, CDKN2B, CHEK1, E2F1, GSK3B, NFKB1, and PRKAR2A), PI3K pathway (CCNE2, CSF3, GSKB3, NFKB1, PPP2R2C, and SGK2), and stromal factors (ANGPT2 and DLL4). The expression of Angiopoietin-2 by immunohistochemistry also showed the same trend of increasing expression from non-neoplastic appearance to high-risk cancer (p < 0.0001).

Conclusions: The gene expression analysis of RAS-mutant thyroid nodules suggests increasing upregulation of key oncogenic pathways depending on their degree of malignancy and supports the concept of a stepwise progression. The utility of ANGPT2 expression as a potential diagnostic biomarker warrants further evaluation.

Introduction

Thyroid nodules are palpable in 5–10% of the population, but they can be identified in >50% of women aged 50 years or older with current ultrasound systems. It is estimated that around 5% of all thyroid nodules are malignant (1). When clinically indicated, the most accurate and cost-effective method for evaluating thyroid nodules is fine-needle aspiration (FNA) biopsy (2). Nonetheless, up to 25–30% of the FNAs render an indeterminate cytology (atypia/follicular lesion of undetermined significance [Bethesda III], follicular neoplasm/suspicious for follicular neoplasm [Bethesda IV], and suspicious for malignancy [Bethesda V]) (3). Many of these nodules end up with a diagnostic surgery, which finds 70% of them to be histologically benign. Thus, thousands of patients are “unnecessarily” exposed each year to potential surgical risks and complications and often to life-long need for thyroid hormone replacement therapy (4).

Commercially available oncogene panels used to evaluate cytology specimens have become a common tool to supplement the preoperative risk assessment of malignancy in indeterminate thyroid nodules (2). The most common molecular events identified in thyroid nodules with indeterminate cytology are point mutations in the RAS gene family (H-RAS, K-RAS, and N-RAS), and the prevalence of malignancy of a RAS mutant thyroid nodule has traditionally been estimated at 80–90% (5,6). Unfortunately, the presence of a RAS mutation in a thyroid nodule does not confirm the malignant biology of any given follicular cell-derived thyroid lesion because these alterations can be harbored by a broad spectrum of histopathological diagnoses ranging from adenomatous/hyperplastic nodules (AHN) to high-risk thyroid cancers. Therefore, a RAS mutation in a thyroid nodule could also be regarded as an indeterminate molecular result (7).

In this study, we evaluated differentially expressed genes and signaling pathways among RAS-mutant thyroid nodules with different histological diagnoses and clinical behaviors, which could uncover key biological processes leading to invasion and generate hypotheses to develop biomarkers with diagnostic, prognostic, or therapeutic utility.

Materials and Methods

Thyroid nodules and clinical data collection

After institutional review board approval, the electronic medical records of all patients evaluated with oncogene panels at the Moffitt Cancer Center between 2012 and 2017 were reviewed. A total 473 FNA were analyzed with oncogene panels, and molecular alterations were identified in 151 cases. Among the 151 cases, 76 nodules harbored a RAS mutation, and 64 were resected, but only 61 nodules had available specimens for the study. All hematoxylin and eosin (H&E)-stained slides were reviewed by one board-certified pathologist with subspecialty expertise in thyroid pathology (J.C.H.-P.) who confirmed histological diagnoses following the 2017 World Health Organization (WHO) Classification of Tumours of Endocrine Organs (8). The pathology review was done blinded to the gene expression profile analysis, but not from the RAS mutation status. In addition, the malignant nodules were subsequently classified based on the American Thyroid Association (ATA) postoperative risk stratification schema (2). Based on the histological diagnosis and biological behavior, the nodules were grouped into five categories indicating their degree of malignancy: non-neoplastic appearance (AHN), benign neoplasm (follicular adenomas, FAs), indeterminate malignant potential (noninvasive follicular thyroid neoplasms with papillary-like nuclear features [NIFTPs]), low-risk cancer, or high-risk cancer. Data on participant demographics, thyroid function, ATA sonographic pattern, tumor size, Tumor–Node–Metastasis staging (American Joint Committee Cancer Eighth edition), treatment, and clinical outcome were retrospectively abstracted from electronic medical records. Gene expression and mutation data were also downloaded from The Cancer Genome Atlas (TCGA), and differences in average expression were compared between RAS-, BRAF-, and non-RAS/BRAF-mutant tumors using t-tests.

Gene expression analyses

One formalin-fixed paraffin-embedded (FFPE) tissue block representative of each case was selected for macrodissection to minimize normal thyroid tissue contamination based on the corresponding H&E-stained slide marked by the pathologist. Six 20 μm thick tissue samples from each nodule were subjected to nucleic acid extraction using All Prep DNA/RNA FFPE Kit (Catalog No. 80234; Qiagen, Valencia, CA). The RNA samples were quantified using Nanodrop and Qubit Fluorometer (Thermo Fisher Scientific, Waltham, MA) instruments, and nucleic acid quality control was performed on Tape Station platform (Agilent, Santa Clara, CA). Gene expression from the extracted RNA was determined using NanoString nCounter analysis for the PanCancer Pathways and PanCancer IO 360 Panels (NanoString Technologies, Seattle, WA). The PanCancer Pathways Panel evaluates 770 genes from 13 canonical cancer pathways and selected reference genes, while the PanCancer IO 360 Panel evaluates 770 genes that cover the key pathways at the interface of the tumor, tumor microenvironment, and immune response. The samples were tested according to the manufacturer's protocol. Briefly, 100 ng of RNA was hybridized to the reporter and capture probes in a thermal cycler for 16 hours at 65°C. Washing and cartridge immobilization were performed on the NanoString nCounter Prep Station, and the cartridge was scanned at 550 fields of view on the nCounter Digital Analyzer. The resulting .RCC files containing raw counts were checked for quality in the NanoString nSolver Analysis Software v4.0 and then exported for analysis.

Bioinformatics and statistical analyses

The raw data were normalized across the five categories (non-neoplastic appearance, benign neoplasm, indeterminate malignant potential, low-risk cancer, and high-risk cancer) using the NanoString nSolver Analysis Software v4.0 and log2-transformed. First, all genes and samples were examined using unsupervised hierarchical clustering. To determine differentially expressed genes, each gene expression was correlated with an increasing degree of malignancy across the five categories using Pearson's correlation. The genes that were differentially expressed across the five categories (p < 0.05) and all samples were examined again by hierarchical clustering. These unsupervised analyses were done using “ggdendro” (9). Analyses were performed separately for each NanoString panel. To generate the box plot for close evaluation of the selected genes, adjustment for multiple testing was performed using false discovery rate correction (q < 0.20). All analyses were performed in R version 3.5.0. Data and code for reproducing the analysis can be found at https://github.com/GerkeLab/ThyroidRAS

Immunohistochemical staining and analysis of Angiopoietin-2

Immunohistochemical staining for Angiopoietin-2 was performed on 4 μm thick representative tissue sections. Briefly, slides were subjected to endogenous peroxidase inactivation (0.6% hydrogen peroxidase in methanol for 30 minutes), antigen retrieval (buffer citrate, pH 6.0, for 30 minutes at 100°C), and nonspecific protein interaction block (2.5% whole goat serum for 20 minutes). Sections were incubated with the primary antibody Angiopoietin-2 (Clone JM71-34, rabbit monoclonal, 1:200; Invitrogen, Waltham, MA) for 1 hour, followed by incubation with the secondary anti-rabbit Histofine Simple Stain MAX PO horseradish peroxidase-conjugate polymer (Nichirei Biosciences, Inc., Tokyo, Japan). 3,3′-Diaminobenzidine (DAB; Vector Laboratories, Burlingame, CA) was used as chromogen, and hematoxylin (SH26-500D; ThermoFisher, Waltham, MA) was applied as a counterstain.

Scoring of the immunohistochemical staining in the nodules and surrounding thyroid parenchyma was performed by the pathologist (J.C.H.-P.) using the modified Histoscore (H-score) (10). First, a visual semiquantitative assessment of both the staining intensity (0 = negative, 1 = weak, 2 = moderate, 3 = strong) and the percentage of positive cells (0–100%) was completed. The H-score was later obtained by multiplying the intensity and the percent tumor cells staining to create a score ranging from 0 to 300. This score was then correlated with the five categories based on the degree of malignancy using Pearson's correlation.

Results

Participant characteristics and FNA analysis

Sixty-one RAS-mutant thyroid nodules corresponding to a total of 59 participants were included in this study (two participants had two nodules). Patient demographics are presented in Table 1. All FNA specimens were interpreted following The Bethesda System, and 58 were designated as indeterminate cytology (Bethesda III, IV, or V). The remaining nodules were classified as Bethesda II (two cases) and Bethesda VI (one case). Different commercially available oncogene panels were used to detect the RAS mutations. Thirty-eight nodules were tested with ThyroSeq v2 (UPMC, Pittsburgh, PA), 5 with ThyroSeq v3, 17 with miRInform (Asuragen, Austin, TX), and 1 with ThyGenX (Interpace Diagnostics, Pittsburgh, PA). These next-generation sequencing panels varied in the total number of genetic alterations that are interrogated, but all of them tested for K-RAS, N-RAS, and H-RAS mutations. The list of genes on each panel is provided in Supplementary Table S1. The variability in the Bethesda categories subjected to molecular testing and the different oncogene panels reflect evolving protocols regarding thyroid FNA molecular analysis at the Moffitt Cancer Center as well as Florida State licensing regulations.

Table 1.

Nodule-Level Characteristics

Characteristic	Malignancy
	Non-neoplastic appearance	Benign neoplasm	Indeterminate (NIFTP)	Low-risk malignancy	High-risk malignancy
N (%)	20 (32.79)	21 (34.43)	8 (13.11)	11 (18.03)	1 (1.64)
Age at surgery (years)
Median (range)	53.5 (25–74)	53 (17–75)	46.5 (29–71)	45 (26–64)	67 (67–67)
Sex, n (%)
Female	17 (85.00)	18 (85.71)	4 (50.00)	11 (100)	0 (0)
Male	3 (15.00)	3 (14.29)	4 (50.00)	0 (0)	1 (100)
Thyroid function, n (%)
Euthyroid	17 (85.00)	20 (95.24)	8 (100)	9 (81.82)	1 (100)
Hypothyroid	3 (15.00)	1 (4.76)	0 (0)	2 (18.18)	0 (0)
Nodule size (cm)
Median (range)	2 (1.10–3.20)	2.4 (1.30–4.80)	2.75 (1.20–4.50)	2.6 (1.00–3.80)	6.2 (6.20–6.20)
Sonographic pattern, n (%)
Very low suspicion	1 (5.00)	0 (0)	0 (0)	0 (0)	0 (0)
Low suspicion	12 (60.00)	8 (38.10)	3 (37.50)	3 (27.27)	0 (0)
Intermediate suspicion	0 (0)	1 (4.76)	0 (0.00)	2 (18.18)	0 (0)
High suspicion	1 (5.00)	1 (4.76)	1 (12.50)	1 (9.09)	0 (0)
Non-ATA suspicion	6 (30.00)	11 (52.38)	4 (50.00)	4 (36.36)	1 (100)
Cytological diagnosis, n (%)
Benign	1 (5.00)	0 (0)	1 (12.50)	0 (0)	0 (0)
AUS/FLUS	11 (55.00)	10 (47.62)	1 (12.50)	4 (36.36)	0 (0)
FN/HCN	8 (40.00)	11 (52.38)	5 (62.50)	6 (54.55)	1 (100)
Suspicious for malignancy	0 (0)	0 (0)	1 (12.50)	0 (0)	0 (0)
Malignant	0 (0)	0 (0)	0 (0)	1 (9.09)	0 (0)
RAS gene mutated, n (%)
HRAS	3 (15.00)	7 (33.33)	2 (25.00)	2 (18.18)	0 (0)
NRAS	12 (60.00)	10 (47.62)	6 (75.00)	6 (54.55)	1 (100)
KRAS	5 (25.00)	4 (19.05)	0 (0)	3 (27.27)	0 (0)
Molecular test used on cytological specimens, n (%)
ThyroSeq v2/v3	15a (75.00)	16b (76.19)	4 (50.00)	7c (63.64)	1 (100)
7-Gene oncogene panel	4 (20.00)	5 (23.81)	4 (50.00)	4 (36.36)	0 (0)
ThyGenX+ThyraMIR	1 (5.00)	0 (0)	0 (0)	0 (0)	0 (0)

^a13 v2 and 2 v3.

^b15 v2 and 1 v3.

^c5 v2 and 2 v3.

ATA, American Thyroid Association; AUS/FLUS, atypia of undetermined significance/follicular lesion of undetermined significance; FN/HCN, follicular neoplasm/Hürthle cell neoplasm; NIFTPs, noninvasive follicular thyroid neoplasms with papillary-like nuclear features.

RAS mutation distribution

The majority of nodules harbored an N-RAS mutation (35/61), followed by H-RAS (14/61) and K-RAS (12/61). The most common alteration (27/35) for N-RAS mutation was an amino acid substitution at codon 61 from a glutamine (Q) to an arginine (R). In addition, three nodules analyzed by ThyroSeq v2 had concomitant mutations in EIF1AX and one nodule had EIF1AX and TERT promoter mutations. One case tested by miRInform exhibited a simultaneous RET-PTC1 gene fusion. The distribution of RAS mutations is summarized in Figure 1. The information regarding the allelic frequency of these mutations is presented in Supplementary Table S2.

FIG. 1.

Heat map of gene mutations for the 61 RAS-mutant nodules. Some nodules have multiple mutations present. Samples in orange were tested using the ThyroSeq panels, blue using the miRInform, and green using the ThyGenX panel. Not all cases were tested for TERT, EIF1AX, and RET mutations. Supplementary Table S1 lists which genes were evaluated in each panel.

Histopathological diagnoses and characterization

The RAS-mutant nodules showed a broad spectrum of histopathological diagnoses, all characterized by exclusive or predominant follicular architecture and are summarized in Table 1. Morphologically, 20 nodules designated as AHNs were characterized by poor encapsulation, variable micro- and macrofollicular architecture, and occurred in the background of a goitrous thyroid gland. The 21 cases diagnosed as FAs were noninvasive solitary lesions that showed a well formed fibrous capsule or were well circumscribed from the surrounding parenchyma, uniform microfollicular architecture, and lack of nuclear atypia diagnostic for papillary thyroid carcinoma (PTC). Four of those nodules were variably composed of oncocytic cells and subcategorized as oncocytic (Hürthle cell) FA.

The diagnosis of NIFTP was rendered in eight cases that showed follicular pattern growth, absence of papillary architecture, absence of psammoma bodies, and absence of invasion, but included the presence of the tumor cells exhibiting variable degrees of nuclear atypia diagnostic for PTC. Three cases with nuclear atypia that also displayed well-formed papillae were excluded from the NIFTP category and were designated as classical variant of PTC while the diagnostic term follicular variant of PTC (FVPTC) was utilized for two cases showing exclusive follicular architecture, nuclear features of PTC, and invasion. The diagnosis of minimally invasive follicular thyroid carcinoma (FTC) was established in five nodules by the identification of capsular invasion at the microscopic level, and one case with the oncocytic appearance of the tumor cell and capsular and vascular invasion (<4 foci) was classified as oncocytic (Hürthle cell) FTC, angioinvasive. One case displayed extensive invasive growth, grossly and microscopically, that included invasion of thyroid parenchyma as well as more than four foci of vascular invasion. Therefore, it was designated as widely invasive FTC. This case also showed high-grade histological features including 2 mitoses per 10 high-power fields and focal solid growth but lacked necrosis or other diagnostic features of poorly differentiated thyroid carcinoma. Of note, this nodule had co-existing N-RAS, EIF1AX, and TERT promoter mutations.

Categorization of the thyroid nodules based on the malignant potential and biological behavior

The histopathological diagnoses of the thyroid nodules were based on the morphological appearance. For further analyses, we grouped the histopathological diagnoses into five categories based on their malignant potential and biological behaviors to reflect the biology rather than the morphology (Table 1). Twenty AHNs were categorized as non-neoplastic appearance. Twenty-one FA nodules were categorized as benign neoplasm. Eight nodules that fulfilled the diagnostic criteria of NIFTP were categorized as tumors of indeterminate malignant potential. The 12 nodules that were diagnosed as carcinoma following the WHO Classification of Tumours of Endocrine Organs were further classified as low-risk (n = 11) and high-risk (n = 1) cancers based on the ATA postoperative risk stratification schema. No intermediate ATA risk tumors were identified. None of the low-risk cancers required further radioactive iodine therapy and have not presented biochemical and/or structural locoregional recurrence after 31 months of mean follow-up. The high-risk malignant RAS-mutant nodule was the widely invasive FTC with the TERT promoter co-mutation. This patient had distant metastases to the lungs at diagnosis, was treated with radioactive iodine, and remains alive with stable disease after 27 months of follow-up.

Gene expression profiles of the RAS-mutant thyroid tumors

Of the 61 RNA samples, 57 passed quality control metrics for the gene expression analyses. To evaluate the potential value of the gene expression in differentiating histopathological diagnoses among RAS-mutant thyroid nodules, unsupervised cluster analyses of the genes included in the PanCancer Pathways and IO 360 Panels were performed (Fig. 2A, C). The expression was relatively homogeneous in majority of the nodules, and no discrete group with a significant sample size could be separated based on the expression patterns for further analyses. This overall homogeneous gene expression pattern in majority of the nodules contrasts with the heterogeneous profile of histopathological diagnoses.

FIG. 2.

Heat maps of gene expression from the NanoString Pathways Panel and IO 360 Panel with additional annotations for RAS mutation, Bethesda category, histology, nuclear atypia, and tumor size. (A) Unsupervised hierarchical clustering of all genes in the PanCancer Pathways Panel. (B) Unsupervised hierarchical clustering of only differentially expressed genes within the PanCancer Pathways Panel. These genes were selected using Pearson's correlation for differential expression (p < 0.05) across the five categories based on the histological diagnosis and biological behavior. (C) Unsupervised hierarchical clustering of all genes in the PanCancer IO 360 Panel. (D) Unsupervised hierarchical clustering of only differentially expressed genes within the PanCancer IO 360 Panel. These genes were selected using Pearson's correlation for differential expression (p < 0.05) across the five categories based on the histological diagnosis and biological behavior. For ease of reading, the figure can be viewed online.

To delineate differentially expressed genes, we analyzed the gene expression using the five categories based on their morphology and biological behaviors as the supervising parameters (Pearson's correlation, p < 0.05). Again, we performed hierarchical clustering analyses using these genes, and no discrete group with a significant sample size was observed (Fig. 2B, D). After adjusting for multiple testing (q < 0.20) of these differentially expressed genes, we found a significant correlation between the degree of malignancy and several genes involved in cell cycle and apoptosis (BAX, CCNE2, CDKN2A, CDKN2B, CHEK1, E2F1, GSK3B, NFKB1, and PRKAR2A) and the PI3K pathway (CCNE2, CSF3, GSKB3, NFKB1, PPP2R2C, and SGK2) in the PanCancer Pathway Panel (Fig. 3A and Supplementary Table S3). Moreover, the individual gene analysis also exhibited a correlation between the five categories and the expression of genes associated with stromal factors (ANGPT2 and DLL4) in the PanCancer IO 360 Panel (Fig. 3B and Supplementary Table S4).

FIG. 3.

Box plots of differentially expressed genes with statistical significance after adjusting for multiple testing (q < 0.20) that correlated with the five categories based on the histological diagnosis and biological behavior. (A) Differentially expressed genes in the PanCancer Pathways Panel. (B) Differentially expressed genes in the PanCancer IO 360 Panel.

We further evaluated an angiogenic factor, ANGPT2. In TCGA analysis of 496 differentiated thyroid cancers, the mean ANGPT2 expression was much higher in RAS-mutant tumors than in BRAF-mutant tumors (p = 0.0005) but was not significantly different from non-RAS/BRAF-mutant tumors (p = 0.12) (Fig. 4).

FIG. 4.

Box plot of ANGPT2 expression by mutation types. ANGPT2 expression was significantly higher on average in RAS-mutant tumors than in BRAF-mutant tumors (p = 0.0005) but not significantly different from non-BRAF/RAS-mutant tumors in TCGA data set (p = 0.12) using t-tests. ANGPT2, Angiopoietin-2; TCGA, The Cancer Genome Atlas. ***p ≤ 0.001.

Immunohistochemical expression of Angiopoietin-2 in RAS-mutant thyroid tumors

Due to positive correlation between the 5 biological categories and the expression of ANGPT2, we further analyzed the expression of Angiopoietin-2 protein by immunohistochemistry in 60 RAS-mutant thyroid tumors (1 case classified as “non-neoplastic appearance” did not have available material for testing). The immunohistochemical reactivity for Angiopoietin-2 was observed in 91.7% (55/60) cases and varied in intensity and percentage of positive cells. The distribution of H-score across the five categories (Fig. 5) demonstrated increased expression of Angiopoietin-2 compared with the surrounding thyroid parenchyma in high-risk cancer with a p-value of <0.0001 (Pearson's correlation). This pattern of expression mirrored the trend observed at the RNA transcript level. The immunohistochemical expression of Angiopoietin-2 was significantly higher in all RAS-mutant nodules in comparison to the surrounding thyroid parenchyma (Fig. 5). Interestingly, in eight cases, there was histological evidence of chronic lymphocytic thyroiditis, and those cases had increased expression of Angiopoietin-2 in the non-neoplastic follicular cells (median H-score 60, 20–150 range).

FIG. 5.

Box plot of Angiopeitin-2 immunohistochemical staining in RAS-mutant tumors and surrounding normal thyroid parenchyma. H-score is compared across the five categories based on the histological diagnosis and biological behavior using Pearson's correlation (p < 0.0001).

Discussion

This study analyzed the gene expression profile of 57 RAS-mutant thyroid nodules looking for differences among tumors with different histological diagnoses and clinical behaviors. Our unsupervised cluster analysis shows that RAS-mutant nodules have a relatively homogenous gene expression. This is consistent with the findings reported in TCGA of PTC where the RAS-mutant tumors are distinct compared with the non-RAS-mutant tumors (11). However, after adjusting for multiple testing, we were able to identify genes involved in cell cycle, apoptosis, and angiogenesis that are differentially expressed within the five categories based on morphology and biologic behavior. This pattern of gene expression supports the proposed concept of an adenoma–carcinoma sequence in thyroid oncogenesis (12). Our study further shows that this sequence starts early in thyroid lesions that morphologically have a non-neoplastic appearance such as AHN (Fig. 6). Despite the known RAS mutations, it is notable that one-third of our resected nodules were classified as AHN on histological evaluation. This finding highlights the limitations of light microscopy in differentiating benign clonal (FA) from hyperplastic (AHN) follicular pattern lesions, and this distinction often is subtle and may be arbitrary. However, if the mutation status of a nodule is unknown, morphology remains the gold standard for the diagnosis of thyroid nodules, and more recently, additional mutations have been identified in nodules reported as AHNs (8,13). In any event, as in other proposed stepwise tumorigenesis models, the driving mutations precede the phenotypic alterations (14,15). If this paradigm is applied to RAS-mutant thyroid nodules, one should expect to encounter AHNs harboring oncogenic clonal mutations.

FIG. 6.

Morphological spectrum of RAS-mutant nodules. (First row: H&E-stained slide at 4 × magnification; second row: H&E-stained slide at 40 × magnification; third row: whole-slide image of Angiopeitin-2 immunohistochemical staining.) H&E, hematoxylin and eosin.

Consistent with prior reports, our study shows that the majority of RAS-mutant thyroid carcinomas with indeterminate cytology are low-grade malignancies (6). Consequently, the incorporation of gene expression analysis focusing on cell cycle, apoptosis, and angiogenesis may improve the risk stratification of oncogene panels. The potential clinical benefit of this approach cannot be underestimated as RAS-mutant nodules are commonly encountered in clinical practice, and current management guidelines of thyroid nodules advocate for a more conservative approach. Current versions of commercially available molecular tests do not risk stratify RAS-mutant tumors. Therefore, gene expression signatures could be developed to that end, resulting in improved diagnostic performance and fewer diagnostic surgeries that could be avoided by more robust molecular risk stratification. However, it remains uncertain if this stepwise thyroid oncogenesis model represents a true biological continuum in which increasing upregulation of key oncogenic pathways leads to progression of a benign RAS-mutated nodule to high-risk malignancy. If this concept proves to be true, serial biopsies to confirm molecular stability of the nodules for follow-up might be required unless other reliable markers of progression are identified.

It has been shown that ultrasound can risk stratify thyroid nodules with indeterminate cytology (16,17). However, as seen in our cohort, there is an overlap in sonographic patterns of RAS-mutant nodules regardless of histological diagnosis, and a significant proportion of them are not adequately classified into any of the sonographic patterns proposed by the ATA. The non-ATA suspicious sonographic pattern includes nodules with heterogeneous echogenicity with or without other suspicious features; and isoechoic or hyperechoic nodules with at least one suspicious feature (irregular-microlobulated/infiltrative-margins, taller-than-wide shape, microcalcifications or interrupted rim calcifications, or extrathyroidal extension) (16,17). In our cohort, heterogeneous echogenicity is the most common reason to regard a nodule as non-ATA suspicious sonographic pattern.

RAS protein cycles between inactive (GDP-bound) and active (GTP-bound) forms (18). Once an activating mutation occurs, the RAS gene encodes a protein resistant to the hydrolysis of GTPase, which converts GTP into GDP (19). It has been proven that the RAS mutation is a driving molecular event in both benign and malignant thyroid nodules (20–23). Thus, it is plausible that there is a series of genes involved in cell cycle and apoptosis that are deregulated after a thyroid follicular cell acquires an activating RAS mutation and consequently contributes to the progression of the lesion by providing a permissive environment in tumorigenesis. RAS protein triggers several downstream effectors, including the MAPK and the PI3K-AKT-mTOR pathways (24,25). It has been reported that the progression from a benign thyroid nodule to a carcinoma depends upon simultaneous signaling activation and genetic alteration of the PI3K pathway (23,26,27). Our findings also show that expression of genes associated with the PI3K pathway (CCNE2, CSF3, GSKB3, PPP2R2C, and SGK2) is gradually increased across the five categories as tumors progress to more malignant phenotypes, as shown in Figure 3A.

During the last decade, major attention has been given to the biological mechanism and clinical significance of TERT promoter mutations in follicular cell-derived thyroid cancer (28). This alteration affecting the catalytic protein subunit of telomerase has been associated with aggressive phenotypes of thyroid cancers and has been better described in BRAF^V600E-mutant thyroid cancers (29). In contrast, thyroid nodules with coexisting RAS and TERT promoter mutations are rare and therefore poorly characterized (29–31). The co-occurrence of TERT promoter mutations with RAS mutations has been shown to be enriched in aggressive and metastatic thyroid cancers (32). Similarly, enrichment on EIF1AX mutations co-occurring with RAS was observed in poorly differentiated and anaplastic thyroid cancers, whereas mutations in EIF1AX were a rare finding in PTC and usually mutually exclusive with either BRAF or RAS mutations (11,32). In poorly differentiated thyroid cancers, EIF1AX mutations were associated with larger tumors and predicted shorter survival (32). In our series, however, two benign tumors with non-neoplastic appearance and measuring close to 2 cm in size harbored a combination of RAS and EIF1AX mutation. Interestingly, the C-terminal p.A113 splice mutation was detected in both tumors, a genetic alteration that seems unique to thyroid malignancies (32). The only high-grade thyroid cancer identified in our study was a large tumor (>6 cm), had EIF1AX (338TT) and TERT (C250T) promoter co-mutations, and had distant metastatic disease at presentation. Although no strong conclusion can be derived from a single case, this observation could suggest that a TERT promoter co-mutation might be a late genetic alteration for the transformation to a higher grade carcinoma. The role of TERT promoter co-mutation in RAS-mutant tumors needs to be elucidated by further research.

The extent of vascular invasion on the histopathological evaluation has been traditionally recognized by pathologists as one of the most important prognostic factors in differentiated thyroid cancer (33). Our study further supports this observation, as in our cohort, the degree of malignancy correlates with the expression of ANGPT2 (Fig. 6). The encoded protein of the former gene is Angiopoietin-2 (ANGPT2), and our study also shows that the immunohistochemical reactivity of this protein follows the same trend of increasing expression from benign to malignant nodules. ANGPT2 mainly functions as an antagonist of the receptor tyrosine kinase, TIE-2, on endothelial cells. The expression of ANGPT2 has been identified as a key angiogenic factor in different tumor types that also correlates with poor prognosis (34). ANGPT2 increases vascular permeability, destabilizes vasculature, and consequently facilitates angiogenesis (35,36). The role of ANGPT2 in vascular remodeling also has implications in the development of metastasis (37,38). Moreover, the expression of ANGPT2 has been proposed as an independent diagnostic marker of thyroid cancer in FNA biopsies (39,40). The gradual overexpression trend of ANGPT2 and its encoding protein in our RAS-mutant thyroid nodules may also have a role in the development of resistance to anti-angiogenic therapies. Lenvatinib and sorafenib are Food and Drug Administration-approved treatment options for patients with progressive, locally advanced, or metastatic differentiated thyroid cancer refractory to radioactive iodine therapy (41,42). These two multikinase inhibitors target the VEGF receptor family, and it has been reported that ANGPT2 expression confers adaptive resistance to anti-angiogenic therapy targeting VEGF signaling (43,44). Therefore, the ANGPT2 gene and protein expression could potentially emerge as a robust biomarker that helps refine the risk of malignancy in thyroid FNA molecular testing, establish prognosis by indicating metastatic potential, and predict response to approved targeted therapies in differentiated thyroid cancer. Further studies are needed to confirm these observations.

The clinical limitation of RAS mutations, when interpreted alone, is mainly attributable to the fact that thyroid lesions harboring this molecular abnormality predominately (if not exclusively) exhibit follicular architecture (6,45). It is well established that follicular-patterned lesions of the thyroid are a main source of diagnostic challenges and tumor classification controversy (46). For instance, the diagnosis of encapsulated follicular-patterned lesions showing nuclear atypia, diagnostic of PTC, FVPTC, has been particularly a subject of a major debate, and different studies have highlighted the subjectivity and lack of consensus in the diagnosis of this histological entity (47,48). Additionally, it has been shown that the clinical behavior of tumors regarded as FVPTC exhibiting an invasive growth pattern is different from their noninvasive counterparts since the metastatic potential of noninvasive tumors has been reported as low or nonexistent in the majority of publications (49,50). These differences are further supported by the high rate of RAS mutations in noninvasive FVPTC in contrast to infiltrative tumors (51). All these clinical, pathological, and molecular evidence lead to the introduction of the diagnostic term “NIFTP” to replace the terminology “noninvasive FVPTC” with the purpose of reducing the psychological and clinical consequences associated with the term “cancer” (52). The concept of no longer considering noninvasive FVPTC as cancer and the true biological existence of NIFTP have been challenged (53–55). In fact, the WHO classifies NIFTP as a borderline tumor with very low malignant potential (8). To this end, our study shows that the RAS-mutant nodules classified as NIFTP show a different pattern of expression that falls between benign and low-grade malignant tumors. To the best of our knowledge, this gene expression analysis provides evidence of NIFTP as a separate molecular entity. Pathological classification of thyroid tumors traditionally follows the dichotomous paradigm of benign and malignant; however, nature is a continuum and our gene expression analysis reflects this reality. Thyroid tumor classification based on molecular characteristics has already been proposed, and our study supports that this concept is feasible (11).

Our study is limited by its retrospective nature. However, we included >95% of all known RAS-mutant tumors resected during the study period mainly identified by oncogene panel testing of thyroid nodules with indeterminate cytology. Therefore, we might have skewed our cohort toward less aggressive nodules. In fact, there was only one high-risk cancer in our cohort, which might have limited our ability to find significant differences in the unsupervised gene expression analysis. However, we believe that the cohort is representative of the population in which oncogene panels are currently being used and thus of the population of tumors with known RAS mutation before surgery. We included nodules tested with different oncogene panels; thus, the mutational status of several genes, such as EIF1AX and TERT, remains unknown in some of the nodules in this cohort. Nonetheless, the co-occurrence of these, or other, mutations seem to be enriched in high- risk cancers and low in benign nodules and low-risk cancers (28,32). The supervised analysis of our study relied on categories derived from the histological interpretation of the nodules blinded from gene expression analyses. RAS-mutant tumors are mostly, or exclusively, follicular-pattern thyroid lesions, for which interobserver agreement is known to be poor (47,48). Thus, the histological diagnoses could have been different if another pathologist interpreted the same cases. However, despite this intrinsic limitation, the gene expression correlated with the five categories.

In summary, our study has identified differentially expressed genes involved in cell cycle, apoptosis, and the PI3K pathway in RAS-mutant thyroid nodules as they progress from AHN to widely invasive follicular carcinoma. The expression of these genes could potentially be used as biomarkers to accurately establish the risk of malignancy of lesions with indeterminate cytological results and validate the concept of a stepwise progression in follicular cell-derived thyroid tumorigenesis. In particular, we demonstrated the important role of angiogenic factors, such as ANGPT2, as a potential diagnostic biomarker and therapeutic target. Moreover, the gene expression pattern identified in our study confirms the true biological nature of NIFTP, as an entity that lies between benign and low-risk malignant thyroid neoplasms. Clinical utility of our findings will require further prospective validation in independent patient cohorts.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This study was funded by the Team Science Award at the Moffitt Cancer Center. This work has been supported in part by the Molecular Genomics Core Facility at the H. Lee Moffitt Cancer Center and Research Institute, a National Cancer Institute designated Comprehensive Cancer Center (P30-CA076292).

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3

Supplementary Table S4