Introduction
The clinical management of patients with cytologically indeterminate thyroid nodules represents a challenging issue. Ultrasound (US) examination is an essential instrument for the initial evaluation of nodules and also for patient follow-up. In order to better characterize thyroid nodules, several US-based stratification systems have been, but their overall performance in the setting of indeterminate cytology is poor [
1]. Indeterminate nodules, namely TIR3A and TIR3B according to the Italian System [
2] and classes III and IV according to the Bethesda system [
3], generally prove to be follicular architecture lesions after surgery, most frequently follicular adenoma, noninvasive follicular neoplasm with papillary-like nuclear features (NIFTP), follicular variant papillary thyroid carcinoma (FV-PTC), and rarely follicular thyroid carcinoma (FTC) [
2,
4]. Malignant lesions arising as indeterminate nodules are generally low-risk tumors for which, in most cases, lobectomy represents a definitive treatment [
5‐
7]; patients rarely need re-operation for completion thyroidectomy. However, on histological diagnosis the rate of benign lesions is high, and most patients are subjected to unnecessary surgical procedures. Therefore, indeterminate nodules should be characterized as well as possible preoperatively, so as to be able to select the most appropriate treatment strategy.
The use of ancillary molecular tests including—at least the analysis of the most frequently mutated genes—can improve the diagnostic accuracy of cytology. In detail, the presence of gene mutations in
BRAF and
RAS genes increases the risk of malignancy up to 80% [
8,
9]. Recently, scientific societies have indicated that a molecular testing with the so-called 7-gene panel can effectively integrate the evaluation of indeterminate cytology [
3,
10]. According to the current European Thyroid Association (ETA) and American Thyroid Association (ATA) guidelines, molecular testing performed on Bethesda III and IV nodules can supplement their clinical assessment by providing information useful to refine the risk of nodule malignancy [
11,
12].
If the presence of a genetic mutation is rather specific for malignancy, a negative result must be considered as not informative, owing to the low sensitivity of
BRAF and
RAS molecular testing. MicroRNAs (miRNAs) are emerging as biomarkers whose deregulated expression is associated with pathological conditions. Several miRNAs have been investigated as possible diagnostic markers in indeterminate thyroid cytology, and some PTC-specific markers have been identified so far [
13,
14]. Some of these miRNAs not only show upregulation in PTC compared to normal tissue or benign thyroid lesions, but they have also been correlated to patient prognosis. For instance, overexpression of
miR-146,
miR-221, and
miR-222 have been associated with a higher risk of recurrence in PTC patients [
15]. However, miRNAs specifically deregulated in benign and malignant follicular-patterned lesions are still under investigation.
The aim of this study is to identify miRNAs useful to the differential diagnosis of indeterminate thyroid nodules; specifically, we analyzed nodules with homogeneous cytological diagnosis (TIR3A and TIR3B), absence of point mutations in BRAF and RAS genes and available histological diagnosis of benign and malignant low-risk thyroid lesions, namely follicular adenoma, NIFTP and follicular variant of PTC.
Materials and methods
Study cohort
In the period January 2018 – December 2020 institutional archives were searched for cytology cases matching the following criteria: diagnosis of TIR3A or TIR3B; availability of cytological slides for pathological revision; availability of histological diagnosis of follicular-patterned thyroid lesions, specifically follicular adenoma, follicular variant PTC, or NIFTP. Cytological slides were retrieved and carefully revised by three cyto-pathologists with experience in thyroid pathology. One cytological smear stained with Papanicolaou per case was selected for miRNA isolation; at least ten groups of follicular cells had to be present on a single slide; more than six slides per case had to be available, so that sufficient material could be left for future evaluation. Coverslips were removed after 24 h immersion in xylene, and the slides were then rehydrated through a descending series of ethanol to water. miRNA extraction was performed after manually dissecting areas containing follicular cells, according to the manufacturer’s instruction (miRNeasy FFPE kit, Qiagen, Hilden, Germany). Total RNA quantity and quality were assessed by the spectrophotometer XPose (Trinean, Gentbrugge, Belgium). A standardized protocol for the simultaneous recovery of miRNA and DNA was not available, therefore, to avoid the use of an additional cytological slide, mutation analysis was conducted on DNA purified from the paraffin tissue specimens corresponding to the analyzed nodules. To allow a selective dissection of material, areas containing tumor cells were marked on the hematoxylin-eosin slides; the percentage of tumor cells was also determined. After standard deparaffination of three 10 µm-thick unstained tissue slides per case, DNA was extracted by using the Qiamp DNA Mini Kit (Qiagen).
Ultrasound evaluation
Neck ultrasound was performed in all patients by Real-time instrument (Technos, Esaote Biomedica, Genova, Italy) by using a 7.5 MHz linear transducer [
16] in the context of a dedicated center for thyroid nodule pathology (Center of Endocrinology, University Hospital of Pisa). Nodules were scored with the EU-TIRADS system as recommended by the ETA [
17].
Analysis of BRAF and RAS mutations
Samples were first tested by high resolution melt analysis (HRMA) by using the Type It HRM master mix (Qiagen) to detect mutations in BRAF exon 15, NRAS codon 61, and HRAS codon 61. Samples with altered melt profile were sequenced on an ABI 3130 genetic analyzer. A commercial real-time PCR kit was used on wild-type samples to assess the presence of mutations in codons 12 and 13 of NRAS and in codons 12, 13, and 61 of KRAS (EasyPGX Ready THYROID kit, Diatech Pharmacogenetics, Jesi, AN, Italy). TERT promoter mutations, indicative of high-risk histology and thus virtually absent in this series, were not tested.
Analysis of miRNA expression by nCounter
The analysis of miRNA was performed by using the nCounter Human v3 miRNA Expression Assay (nanoString Technologies, Seattle, WA, USA), targeting 828 miRNAs, on the nCounter system (nanoString Technologies). The panel includes probe pairs for 798 endogenous miRNAs and for 30 controls, comprising positive, negative and ligation controls. Samples were prepared for hybridization following the manufacturer’s instructions. After purification and digital counting, raw data were analyzed by the nSolver analysis software v4.0 (nanoString Technologies). Biological normalization was performed by using the mean expression of the top 100 expressed miRNAs. After adding 0.1 to log2 normalized counts to avoid dealing with log of 0, in order to select and analyze only the miRNAs highly expressed in all samples and those expressed with a large variation among samples, miRNAs were filtered according to the following criteria: (a) miRNAs with mean log2 expression above 4 were selected; (b) of the miRNAs with mean log2 expression below 4, only those with variance higher than or equal to 1.5-fold the mean were considered.
Analysis of miR-7-5p and miR-548ar-5p expression by RT-PCR
Reverse transcription PCR (RT-PCR) was performed on a subset of 17 samples already analyzed by the nCounter. These samples were all negative for gene mutations, and were chosen on the basis of RNA availability: the very same RNA sample was used to avoid any source of variability. The miR-7-5p and miR-548ar-5p expression was tested by real-time RT-PCR along with U6 small nuclear RNA (RNU6-1) as reference by using the TaqMan Assay system (Thermofisher Scientific, Waltham, MA, USA). RT was performed by adding target-specific primers with 10 ng of input RNA according to the manufacturer’s protocol. The experiments were conducted in triplicate on a RotorGene Q instrument (Qiagen).
Statistical analysis
Principal component analysis (PCA) was conducted by using the PCAtools Bioconductor package v.2.4.0. Differential miRNA expression analysis was performed following the procedures of the limma Bioconductor package v.3.48.1. In detail, FAs were used as baseline and the Benjamini-Hochberg method was employed to adjust the P-values. The correlation between miRNAs was tested by the Pearson’s method. The miR-score was built by averaging the expression level of miRNAs with a P-value below 0.05 and an absolute log2 fold change (FC) higher than 2. Receiver operating characteristics (ROC) curves analysis was used to assess diagnostic performance, following the procedures of the pROC R package v.1.17.0.1. Confidence intervals were computed by 2000 bootstrap replicates. For the analysis, NIFTPs were considered malignant since they required surgical excision.
RT-PCR data were analyzed by the ddCt Bioconductor package v.1.50.0; amplification data in the form of threshold cycles (Ct) were exported and relative expression was computed using the 2-ΔΔCt method [
18]. The comparison of the expression levels measured by the RT-PCR and the nCounter method was performed by the Passing-Bablok regression so as to avoid any kind of assumption on data distribution, and following the procedures of the mcr R package v.1.2.2. The nCounter was set as reference method, while RT-PCR was used as the test method.
Discussion
The role of molecular testing in improving the presurgical evaluation of indeterminate thyroid nodules has been widely recognized. As a matter of fact, a better characterization of indeterminate cytology can be useful to avoid both diagnostic surgery and two-step surgery.
A recent meta-analysis has investigated the diagnostic performance of the main commercial tests dedicated to thyroid cytology, including ThyroSeq v3 and Afirma Genome Sequencing Classifier (GSC) [
19]. These two tests were confirmed as having a good NPV, higher than 90%. The performance of a molecular test is strictly related to the risk of malignancy observed in each specific institution for indeterminate thyroid nodules. This is in turn related to the cytopathologist’s experience, to US evaluation, and also to the ways in which all this information is integrated and to how it influences the clinical judgment. For instance, a molecular test with a high NPV could be suitable for an institution showing a low prevalence of cancer among indeterminate nodules. In that case, the reduction of unnecessary diagnostic thyroidectomies would be considerable, and the costs related to molecular tests would be largely justifiable. Both ThyroSeq v3 and Afirma GSC tests are offered by private companies, but the costs are not covered by the National Health Systems. On the other hand, in many molecular pathology laboratories, an approach based on a reduced number of markers like the 7-gene panel, represents a practical and cost-effective option.
US scoring systems, such as the EU-TIRADS system, are helpful clinical tools for the selection of nodules deserving FNA and for the risk assessment of thyroid nodules. A higher US score is generally associated with a higher risk of malignancy. However, cytologically indeterminate nodules often do not show worrisome US characteristics, so that the performance of the EU-TIRADS system is overall poor [
1,
20,
21].
The aim of this study was to find miRNAs able to differentiate benign from malignant follicular-architecture lesions in BRAF- and RAS-negative nodules. Analysis of miRNA expression analysis was conducted directly on the cytological slides, which represent the only diagnostic material available in the presurgical setting.
By comparing wild-type FAs versus wild-type FV-PTCs, none of the 99 expressed miRNAs showed significant differences after statistical correction for multiple comparisons caused by low statistical power. However,
miR-7-5p and
miR-548ar-5p were downregulated in FV-PTC with
P-value <0.05 and fold change >2. ROC analysis for these two miRNAs demonstrated that their combination (miR-score) best performed in terms of AUC (0.79), sensitivity (93%) and NPV (92%) in wild-type cases. To confirm the potential value of these markers, the miR-score was also tested in all the samples analyzed with the nCounter (including mutated cases) and was then compared with the performance of gene mutations. As shown in Fig.
3 and Table
5, the miR-score displayed a better performance in terms of AUC, reaching a sensitivity of 94% and a NPV of 85%. It should also be noted that, despite a considerably high sensitivity, NPV was affected by the high prevalence of malignancy in our series (48%), as a result of the selection criteria adopted.
miR-7-5p has been previously described as suppressor miRNA across multiple cancer types [
22], including lung, colon, and breast cancer. In particular,
miR-7-5p acts as tumor suppressor in signaling pathways with a crucial role in cancer (EGFR/MAPK, PI3K/Akt, Wnt, and others), inhibiting cell survival, proliferation, and migration.
In thyroid cancer,
miR-7-5p has been described as downregulated in PTC compared to benign thyroid tissue [
23,
24]. In particular, an intriguing result obtained by Jahanbani et al. raised the possibility that
miR-7-5p could serve as a NIFTP-specific marker [
25]. The authors found that NIFTPs showed significant downregulation of this miRNA compared to thyroid hyperplasia; however, downregulation of miR-7-5p was also observed also in PTCs. In our series the
miR-7-5p and
miR-548ar-5p (miR-score) expression in NIFTPs was variable, likely depending on their mutational status: the miR-score was higher in three wild-type NIFTPs, while it was below the cutoff in three
RAS-mutated NIFTPs. It can be hypothesized that wild-type NIFTPs maintain
miR-7-5p and
miR-548ar-5p expression similarly to FAs, while mutated NIFTPs tend to lose these miRNAs similarly to FV-PTCs. However, the number of NIFTP cases in this series makes it difficult to draw definitive conclusions in this regard.
It is important to note that the results herein provided represent the first evidence—obtained directly from fine-needle aspiration cytology specimens – that miR-7-5p is downregulated in follicular-patterned thyroid malignancies.
Only a few reports have investigated the role of
miR-548ar-5p in cancer. This miRNA has hundreds of predicted target genes. It belongs to the
miR-548 family, which is involved in several biological processes, including MAPK and Wnt signaling pathways [
26]. For instance,
miR-548-3p has been described as significantly downregulated in breast cancer; its tumors suppressor role has also been confirmed in vitro [
27]. Further studies are needed to investigate the role of
miR-548ar-5p in thyroid tumors and to experimentally validate its target genes.
Nodules with high cyto-histological, US, and molecular similarity were selected for this cohort study. Indeed, all nodules presented with a cytologically indeterminate diagnosis and with benign or low-risk US features; histologically, all nodules were low-risk follicular-architecture thyroid neoplasms. In this context, the presurgical characterization of nodules should be as accurate as possible, in order to select the optimal clinical and surgical management of patients. This work demonstrates that the analysis of two miRNAs is highly capable of ruling-out malignancy both in BRAF/RAS-mutated and in wild-type nodules.
This study has some limitations. Molecular testing was limited to BRAF and RAS genes. Screening for additional RAS-like alterations, for instance, EIF1AX mutations and PPARG fusions, would have added important pieces of information and strengthened our findings. Moreover, miRNA expression analysis and mutational screening were conducted on nucleic acids purified from different types of material, i.e., cytological smear and paraffin tissue, respectively. This depended on the unavailability of an optimized protocol for the simultaneous recovery of miRNA and DNA, and could represent a potential bias. Finally, this retrospective cohort study allowed us to perform a careful selection of samples according to specific histological types. Although the samples were morphologically and molecularly similar, this selection constituted a bias, and may have influenced the performance of miRNA expression analysis. A confirmation on a prospective series of indeterminate thyroid nodules is therefore warranted. In this context, an RT-PCR assay for testing few target miRNAs would represent a practical and cost-effective strategy to be applied for validation purposes. Nonetheless, as shown by method comparison regression analysis, RT-PCR and nCounter data present an imperfect correlation, characterized by proportional differences, possibly due to technical biases related to reverse transcription and amplification processes. This aspect should be taken into account for the future selection of an appropriate validation methodology, along with a careful cost analysis.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.