Fine-Needle Aspiration and Molecular Analysis

Molecular Alterations in Thyroid Cancer

Over the past two decades, research into the molecular pathogenesis of thyroid cancer has elucidated many of the genetic drivers of carcinogenesis. Although thyroid cancers have long been known to exist on a spectrum of clinical severity that parallels the degree of cellular differentiation, the molecular alterations behind these phenotypic changes are increasingly being defined (and are more thoroughly reviewed in Chapter 18 , Molecular Pathogenesis of Thyroid Neoplasia). These discoveries, coupled with novel technologies that enable molecular profiling of individual tumors, have led to the burgeoning field of molecular diagnostic pathology. The application of molecular profiling to the evaluation of cytologically indeterminate nodules is a rapidly evolving field that leverages the unique biologic signature of each thyroid cancer subtype compared with normal thyroid tissue and each other.

The most common thyroid cancers are collectively referred to as “differentiated thyroid carcinomas” (DTC) due to their usually indolent nature and close resemblance to the thyroid follicular epithelial cell from which they originate. DTC is subdivided into papillary (PTC) and follicular thyroid carcinomas (FTC), both of which have many subtypes with unique phenotypic features and behaviors. PTC and its variants are by far the most common type of thyroid cancer (80%) and harbor a relatively simple and quiet genome. The majority of mutations are in the mitogen-activated protein kinase (MAPK) pathway, which regulates cell growth and differentiation. BRAF V600E is the most prevalent alteration, followed by alterations in the signaling of tyrosine kinase receptors such as RET and NTRK. FTC, on the other hand, is associated with mutations in RAS, PTEN , and PIK3CA, and the PAX8-PPARγ rearrangement.

Molecular diagnostics have shed light on the disparate clinical behavior exhibited by several DTC subtypes and may enhance traditional pathology and lead to revised classification schema. For instance, RAS mutations, which are mutually exclusive of BRAF V600E, can be found in both FTC and the follicular variant of papillary thyroid cancer (FVPTC, the most common PTC subtype) but almost never in classical PTC. FVPTCs have long been recognized to be heterogeneous tumors; use of integrated genetics to articulate “BRAF-like” and “RAS-like” tumors has shed light on this diversity in clinical behavior. Indeed, BRAF V600E mutant tumors are typically locally invasive and behave like classical PTC, whereas RAS mutant FVPTCs tend to be encapsulated and behave more like follicular adenomas or carcinomas. Because RAS mutant FVPTCs have more in common with follicular tumors than they do classical PTC, they should be managed and perhaps reclassified as such. Another example is Hürthle cell carcinoma (HCC), which has traditionally been classified as a variant of FTC. However, data suggest it has a distinct genomic signature, including chromosomal copy number alterations and novel mutations in mitochondrial DNA.

Poorly differentiated and undifferentiated (anaplastic thyroid carcinoma, ATC) thyroid carcinomas are rarer, have more complex and widespread alterations, and behave more aggressively. Increased MAPK pathway output, combined with mutations in the TERT promoter, EIF1AX, DNA repair genes ( CKEK2, SWI/SNF complex, etc.), and P53 have all been implicated in the process of de-differentiation and genomic instability. Clinical consequences of these changes can include loss of thyroglobulin expression and failure to respond to radioactive iodine, along with increasing aggressiveness; ATC carries a dismal prognosis. Medullary thyroid carcinoma is a biologically unique malignancy that arises from the neuroendocrine parafollicular “C” cells and, compared with follicular cell tumors, has a higher rate of being hereditary. The most common mutations are in the RET gene, although RAS mutations are also common in sporadic tumors.

Molecular Techniques

Molecular profiling relies on several different NGS and gene expression technologies, each of which imparts unique characteristics to testing platforms and affects performance as a “rule in” or “rule out” test. Fundamentally, techniques differ in their testing substrate (DNA versus RNA) and whether they employ gene sequencing and/or expression methodologies. All methods rely on some form of nucleic acid extraction and preservation, often using proprietary preservatives, followed by amplification and detection. Sequencing is typically performed using oncogene or fusion-specific primers by polymerase chain reaction (PCR, for DNA), reverse transcription polymerase chain reaction (RT-PCR, for RNA), or using microarrays.

The first genotyping assays were designed to detect point mutations and gene fusions but over time have evolved to also identify insertions/deletions, along with copy number alterations. Notably, whereas BRAF V600E, TERT promoter, and P53 mutations have proven highly specific for malignancy, alterations in RAS, PTEN, and others may be found in follicular adenomas and other benign lesions, thereby creating challenges in using these signatures to differentiate benign from malignant specimens. Expression analyses using messenger RNA (mRNA) rely on RNA sequencing (RNA-Seq) and microarray technologies to determine differentially expressed genes between benign and malignant tissues. More recently, microRNA, which are small noncoding RNAs that indirectly regulate gene expression through modulation of mRNA, have also been used. Regardless of the RNA substrate, machine learning algorithms are typically used to perform pattern recognition to distinguish a benign from malignant expression signature.

At the time of writing this chapter, there are three commercially available tools, all of which use some combination of the aforementioned methods: ThyroSeq, Afirma, and ThyGeNEXT/ThyraMIR ( Table 12.1 ).

Clinical Utility

To understand the clinical utility of molecular testing, it is worthwhile to briefly review the statistics associated with test performance. In molecular testing for cytologically indeterminate thyroid nodules, the histopathology of the surgically excised nodule serves as the standard for whether the nodule is benign or malignant. Sensitivity and specificity, as determined by validation studies, are immutable characteristics of the test under consideration. Sensitivity represents the proportion of malignant nodules that will test positive (true positives); specificity represents the proportion of benign nodules that will test negative (true negatives). A highly sensitive test is useful as a “rule out” test because a negative result is unlikely to be a false negative (a test with 100% sensitivity will detect all individuals with the condition). A highly specific test is useful as a “rule in” test because a positive result is unlikely to be a false positive (a test with 100% specificity will be negative for all those without the condition).

When applying a test clinically, the true diagnosis is unknown; what clinicians and patients must know for clinical decision making is the positive predictive value (PPV) and negative predictive value (NPV) of the test. For molecular testing of cytologically indeterminate nodules, the PPV indicates the percentage of patients with a positive result that will actually have a malignancy (true positives); the NPV indicates the percentage of patients with a negative result that will truly not have a malignancy (true negatives). It is important to recognize that PPV and NPV are influenced by the sensitivity and specificity of a test, as well as the prevalence of the disease state in question. For a given sensitivity and specificity, as disease prevalence rises, PPV increases and NPV decreases, affecting the clinical utility of the test. Conversely, decreased disease prevalence decreases the PPV and increases the NPV. Given the wide variability in reported rates of malignancy for cytologically indeterminate specimens, it is critical to know local rates of malignancy to understand test performance in a specific population. Clinicians are urged to recognize how the local prevalence of malignancy in each Bethesda category affects a test’s reported PPV and NPV when it is applied to their individual clinical setting.

It should be noted that the recent reclassification of noninvasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTP) from malignancy to neoplasm has bearing on the interpretation of molecular testing. The term “NIFTP” was introduced in 2015 to describe encapsulated follicular variant papillary thyroid cancers that harbor no capsular or vascular invasion and exhibit indolent behavior with an excellent prognosis. Cytologic specimens from NIFTP lesions are frequently indeterminate (Bethesda III to V); its reclassification as a neoplasm lowers the risk of malignancy in each group. During the development and validation of first- and second-generation molecular tests, nodules that would now be diagnosed as NIFTP would have been considered malignant. It follows that NIFTP renders a “suspicious” or positive result on molecular tests, thereby lowering the PPV. However, although NIFTP is clinically regarded as a low-risk neoplasm, it is a histologic diagnosis and thus requires a surgical excision to be made, and its malignant potential, if left untreated, is unclear. As such, lobectomy directed by a “suspicious” or positive result is in line with current diagnostic and therapeutic recommendations. For this reason, grouping a NIFTP with malignant nodules optimizes test performance and interpretation when results are reported in a binary fashion.

Background

The first version of ThyroSeq was developed for clinical use at the University of Pittsburgh Medical Center in 2007. This panel (ThyroSeq v0) initially evaluated seven genes for mutations ( BRAF , NRAS61 , HRAS61 , KRAS12 / 13 ) or rearrangements ( RET / PTC1 , RET / PTC3 , PAX8 / PPARγ ). The 7-gene panel had high PPV but low NPV for the detection of malignancy in indeterminate nodules. Since then, there have been multiple iterations of the ThyroSeq test, which have employed the use of NGS of DNA and RNA. Each successive iteration has evaluated a larger number of genes; ThyroSeq v1 was a 15-gene panel and v2 was a 56-gene panel. Although initially developed as a “rule in” test, the sensitivity and NPV have increased with each new version of the test.

The most recent version, ThyroSeq v3, was developed in 2017 and analyzes 112 genes for four classes of genetic abnormalities, including mutations, gene fusions, gene expression alterations, and copy number variations. A proprietary algorithmic analysis of all detected genetic abnormalities is used to generate a genomic classifier (GC) score; each abnormality detected is assigned a value from 0 to 2 commensurate with its association with malignancy, and these individual values are summed. The test is reported as positive if the GC score is ≥ 1.5 or negative if the GC score is < 1.5. Additionally, the test report provides information about the specific genetic abnormalities detected.

Clinical Validation

ThyroSeq v3 has been validated in a prospective, blinded, multicenter study. In this study, 286 FNA samples from nodules with indeterminate cytology (Bethesda III to V) were evaluated with ThyroSeq v3 and compared with histologic findings after thyroidectomy. Ten percent of samples were inadequate for evaluation, due to either low total nucleic acids or insufficient thyroid cells present. Of the remaining 257 samples, 152 (59%) tested negative and 105 (41%) tested positive. One hundred and forty-seven of the 152 nodules that tested negative were benign on final pathology. Of the five nodules with false negative results, four were PTC and one was a minimally invasive FTC; all fell into the ATA low risk category. For the subset of nodules with Bethesda III or IV cytopathology, the disease prevalence (cancer/NIFTP) as determined by histologic evaluation of the thyroidectomy specimen was 28%. In these 247 samples, the sensitivity, specificity, NPV, and PPV were 94%, 82%, 97%, and 66%, respectively.

As previously discussed, test utility is dependent on the prevalence of the disease in a studied population, with NPV decreasing as disease prevalence increases. For ThyroSeq v3, the NPV remains 95% or greater for a disease prevalence of up to 40% for Bethesda III and 60% for Bethesda IV nodules. It has limited applicability as a “rule out” test for Bethesda V lesions (SFM) due to the higher cancer prevalence in patients with this FNA finding. In these patients, a negative test result carries a risk of malignancy of 20%. For nodules with Bethesda III/IV cytopathology, the 3% to 5% cancer risk (at typical rates of disease prevalence) with a negative test is comparable with the cancer risk in nodules diagnosed as Bethesda II by FNA cytology.

The benign call rate (BCR), defined as the percentage of cytologically indeterminate specimens that are subsequently found to be benign on molecular analysis, is an important measure of test utility. In applying a molecular test with a high NPV, the BCR determines the number of patients that can be managed nonoperatively. One study found the BCR of ThyroSeq v3 to be 74.1% for patients with Bethesda III/IV cytopathology.

Studies have demonstrated the validity of ThyroSeq v3 in the evaluation of Hürthle cell thyroid lesions. Notably, all 10 HCCs were correctly identified with positive test results in the initial validation study. In another analysis of the performance of ThyroSeq v3, 39 of 42 (93%) Hürthle cell malignancies were correctly identified; among all oncocytic nodules, sensitivity and specificity were 93% and 69%, respectively. It has been suggested that ThyroSeq v3 can be used to help determine the extent of surgery for Hürthle cell neoplasms.

Regarding NIFTP, the prospective multicenter study for ThyroSeq v3 grouped NIFTP with cancerous nodules. Previous validation studies occurred before the introduction of NIFTP and therefore overestimated cancer prevalence within the Bethesda III/IV groups. A study specifically evaluating the effect of the classification of NIFTP as a benign lesion showed a decrease in the PPV of ThyroSeq v2 when NIFTP was classified as benign.