Multiplatform molecular test performance in indeterminate thyroid nodules

Abstract Background Approximately 25% of thyroid nodule fine‐needle aspirates (FNAs) have cytology that is indeterminate for malignant disease. Accurate risk stratification of these FNAs with ancillary testing would reduce unnecessary thyroid surgery. Methods We evaluated the performance of an ancillary multiplatform test (MPTX) that has three diagnostic categories (negative, moderate, and positive). MPTX includes the combination of a mutation panel (ThyGeNEXT®) and a microRNA risk classifier (ThyraMIR®). A blinded, multicenter study was performed using consensus histopathology diagnosis among three pathologists to validate test performance. Results Unanimous consensus diagnosis was reached in 197 subjects with indeterminate thyroid nodules; 36% had disease. MPTX had 95% sensitivity (95% CI,86%‐99%) and 90% specificity (95% CI,84%‐95%) for disease in prevalence adjusted nodules with Bethesda III and IV cytology. Negative MPTX results ruledout disease with 97% negative predictive value (NPV; 95% CI,91%‐99%) at a 30% disease prevalence, while positive MPTX results ruledin high risk disease with 75% positive predictive value (PPV; 95% CI,60%‐86%). Such results are expected in four out of five Bethesda III and IV nodules tested, including RAS positive nodules in which the microRNA classifier was useful in rulingin disease. 90% of mutation panel false positives were due to analytically verified RAS mutations detected in benign adenomas. Moderate MPTX results had a moderate rate of disease (39%, 95% CI,23%‐54%), primarily due to RAS mutations, wherein the possibility of disease could not be excluded. Conclusions Our results emphasize that decisions for surgery should not solely be based on RAS or RAS‐like mutations. MPTX informs management decisions while accounting for these challenges.


| INTRODUCTION
Although thyroid nodules are extremely common, thyroid cancer is relatively infrequent, with only approximately 16 new cases diagnosed per 100 000 adults per year in the United States. 1 Given that most thyroid nodules are benign, it is beneficial to preoperatively distinguish nodules that are likely benign from those that are likely malignant in order to minimize unnecessary surgery for benign nodules and reserve surgery for clinically significant malignancy. This risk assessment is also important for preoperative patient counseling, discussion, and surgical planning. IV, or V, which are considered indeterminate for malignancy. 3 The average rate of malignancy (ROM) associated with the Bethesda III, IV, and V categories is 18% (range 6%-30%), 25% (range 10%-40%), and 60% (range 45%-75%), respectively. 2 As a result, many patients with benign nodules undergo surgery that could potentially be avoided.
Molecular tests have been increasingly used in the clinical setting as adjuncts to further risk stratify nodules with indeterminate cytology. The objective is to distinguish patients who are more likely to benefit from conservative management (ie, continued surveillance) from those who are more likely to benefit from surgical intervention. Some commercially available molecular tests, such as ThyroSeq ® and ThyGeNEXT ® , are oncogenic driver mutation panels. 4,5 Strong driver mutations that are highly predictive of malignancy, such as BRAF V600E mutations, RET fusions, and TERT promoter mutations, have proven useful in surgical decision making. [6][7][8][9][10][11][12][13] However, the most common mutations in indeterminate nodules are RAS, which are weak driver mutations with a lower PPV. In recent studies of RAS mutations, the PPV ranged from only 10% to 37% across multiple institutions. 14,15 Although some have reported a higher PPV for RAS mutations, others have suggested that this higher PPV may reflect RAS performance in distinct benign and malignant histopathologic subtypes included in those studies. 16 In addition, many other mutations and fusions that are included in commercial mutation panels occur at very low frequencies, making their predictive value for malignancy difficult to study and consequently uncertain.
RNA-based risk classifier approaches have also been increasingly used in the clinical setting to risk stratify thyroid nodules with indeterminate cytology. A messenger RNA-based genomic sequencing classifier (GSC, Afirma ® ) can ruleout the need for surgery through its reported high NPV (96%) but cannot effectively rulein the need for surgery due to its suboptimal PPV (47%). 17 In contrast, a multiplatform test (MPT) approach that combines a mutation panel test (ThyGenX ® ) and a microRNA risk classifier test (ThyraMIR ® ) has been shown to provide both high NPV and high PPV for malignancy and have clinical utility. 18,19 Negative MPT test results have been associated with the same low rate (11%) of nodule surgical resection as other tests that effectively ruleout the need for surgery, including benign cytology and GSC. 19,20 Positive MPT results have been associated with high rates (84%) of surgical resection, consistent with a test that effectively rulesin the need for surgery. 19 In the current version of MPT, designated MPTX, an analytically validated expanded, next generation sequencing test (ThyGeNEXT) is used in combination with the microRNA risk classifier test (ThyraMIR). 4,21,22 The expanded mutation panel includes NTRK and ALK fusions that have targeted therapies, as well as TERT and RET protooncogene mutations that are markers of aggressive disease. 6,[23][24][25][26][27][28] In MPTX testing, samples with no detectable mutational change and those that have weak driver mutations are further risk stratified using the microRNA classifier, which incorporates two thresholds for malignancy risk. 29 The first threshold was designed to optimize sensitivity for malignancy while the second threshold was designed to maximize specificity. 18 A corresponding representative follow-up formalin fixed, paraffin embedded, H&E stained tissue section from the surgically resected nodule was also provided. Only slides with adequate thyroidal epithelial cellularity (ie, at least 80 cells per slide) that had been archived for fewer than 10 years and were from patients greater than 18 years of age were accepted. The study was approved by a central independent ethics review board with informed consent waived due to minimal risk (Advarra IRB #33697).

| Molecular testing
Prior to the present study, Interpace Diagnostics had performed molecular testing on a cohort of archived cytology slides from thyroid nodule FNAs to assess DNA and RNA degradation over time after cytology fixation. Based on these results, it was expected that approximately 23% of cytology slides that had been archived for a median of 5 years would fail to provide molecular results due to DNA and RNA degradation(SD Finkelstein, unpublished observation).
In the current study, all molecular testing was performed by Interpace Diagnostics using archived thyroid cytology slides that had been collected by FNAs performed and processed in accordance with usual clinical and laboratory practice. Interpace was blind to the histopathology outcomes of subjects when molecular testing was performed. All molecular testing was performed using standard clinical procedures for the ThyGeNEXT mutation panel and ThyraMIR microRNA risk classifier commercial tests (Interpace Diagnostics). All molecular test results were stored and finalized in a secure laboratory information system (LIMS) database that was password protected and separated from that which harbored de-identified baseline characteristics and follow-up histopathology diagnoses of subjects.
The expanded mutation panel test (ThyGeNEXT) utilizes targeted next-generation sequencing (NGS) (MiSeq, Illumina) to detect messenger RNA fusion transcripts and DNA mutation variants listed in

| MPTX test results
MPTX test results were recorded as negative, moderate, or positive in accordance with Interpace's standard clinical procedures, blind to histopathology outcomes. All MPTX test results were stored in the LIMS database that was password protected and separated from that which harbored de-identified histopathology outcomes. In MPTX, samples with strong driver mutations do not undergo microRNA risk classification, while those with weak drivers or no detectable mutations are further risk stratified by microRNA levels. MPTX is resulted as negative when no mutations are detected and the microRNA test is negative (Level-1); as positive when a strong driver mutation is detected or when the microRNA test is positive (Level-3); and as moderate when a weak driver mutation is detected and the microRNA test is negative or moderate (Levels 1-2), or when no mutations are detected and the microRNA test is moderate (Level-2). The development of negative, moderate, and positive microRNA levels (Levels 1-3) have been previously described, with the current study serving as a validation of this approach. 18,29 The threshold for negative MPTX results was designed to optimize sensitivity, while the threshold for positive MPTX results was designed to maximize specificity. In MPTX, BRAFV600E, TERT,   Table S1.

| Observed expanded mutation panel test performance
We first evaluated performance of the expanded mutation panel test by itself, without the addition of the microRNA risk classifier test (  (Table S1).
Given the high false positive rate of individual RAS mutations and resulting low PPV in our cohort, we verified the analytical accuracy of the next NGS based platform used for detecting RAS mutations.
Thirty-five RAS positive subjects, for which sufficient residual nucleic acid remained, were randomly selected and tested using a different analytical platform (ie, competitive allele-specific PCR). There was 100% (95-100%, 95%CI) qualitative agreement between the two platforms confirming the high analytical specificity of RAS testing using the NGS based platform.

| Mutation panel test performance after histopathologic subtype prevalence adjustment
Compared to a recent prospective study of indeterminate thyroid nodules, our study cohort had a disproportionally large number of benign adenomas relative to hyperplastic nodules. 5 Adenomas accounted for 81% of all benign nodules tested in our study (Table S1), while they accounted for less than half (45%) of benign nodules tested in the prospective study. 5 Additional differences were also noted in the proportions of malignant or NIFTP subtypes in our study compared to that reported by Steward et al 2019. Due to these differences, we examined performance of the expanded mutation panel after proportions of these distinct histopathologic subtypes observed in our study were prevalence adjusted to match those reported by Steward el al 2019. The prevalence adjustment improved specificity and PPV of the mutation panel from 69% to 77% and from 56% to 64%, respectively ( Table 2). It also marginally improved sensitivity and NPV. The PPV of individual RAS mutations increased from 32% to 46%, with NRAS having the highest PPV of all RAS mutation subtypes, which improved from 37% to 51% after the prevalence adjustment (Table S1 vs Table S2).   (Table S3D).
The expected NPV and PPV of MPTX and the expected ROD in moderate MPTX results based on prevalence adjusted performance in Bethesda III and IV nodules are shown over variable disease prevalence in Figure 2. At the highest disease prevalence expected in nodules with Bethesda IV cytology (40%), the NPV and PPV of MPTX was 96% and 82%, respectively (Figure 2A). 2 Moderate MPTX results had a 49% ROD ( Figure 2B). At the highest disease prevalence expected in nodules with Bethesda III cytology (30%), the NPV and PPV of MPTX was 97% and 75%, respectively. 2 Moderate MPTX results had a 39% ROD.
A prior observational study has shown that the rate of malignancy in Bethesda III and IV nodules that have undergone clinically prescribed molecular testing is only 14%. 19 Since the ROD observed in our study was much higher for such nodules, we examined the expected rate at which each MPTX categorical result would occur at this lower disease prevalence (Figure 3). At a 14% rate of malignancy  In addition to validating the performance of the MPTX multiplatform approach, our study supports the utility of mutation panels that include key therapeutic and prognostic markers. An NTRK3 fusion, which is a known therapeutic target, was detected in one nodule with malignancy (Table S1). Multiple coexisting mutations were also detected. All coexisting mutations occurred in malignant nodules and were paired with TERT promoter mutations. Such coexisting mutations have the potential to promote aggressive tumor behavior and have been associated with poor patient survival. [9][10][11]43 Consistently, coexisting mutations were found in aggressive cancer types, including Hürthle cell carcinoma, poorly differentiated thyroid carcinoma, and widely invasive follicular thyroid carcinoma (Table S1).
Remarkably, two subjects had benign adenomas with individual TERT promoter mutations that did not coexist with other mutations.
Although infrequent, others have detected TERT mutations in follicular adenomas. [44][45][46] Some have suggested that these mutations may be an early genetic event in follicular tumors that have yet to show morphological signs of malignancy. 46 21 In addition, our study had a large number of benign adenomas. To address this, MPTX test performance observed in our study was prevalence adjusted to match the proportions of distinct histopathologic subtypes observed in a recent prospective study in which adenomas were less frequent. 5 After the prevalence adjustment, MPTX test performance was similar or improved. We also did not encounter any cases of medullary thyroid carcinoma (MTC) in our study, and consequently we were unable to further validate the reported utility of using the mutation panel in combination with microRNA analysis to identify MTC. 47,48 Our results emphasize that decisions for surgery should not solely high risk when RAS is detected, it is less effective at rulingout disease in these cases, and as a result a small portion of nodules that have moderate risk of malignancy will be reported as such in clinical practice.

ACKNOWLEDGMENT
The study was sponsored by Interpace Diagnostics Corporation. Gyanendra Kumar, Alidad Mireskandari, Sydney D. Finkelstein, and Shikha Bose each participated in study design, generating data for the study, review of data analysis, and preparation of the manuscript into its final form. Thomas J. Giordano and Peter M. Sadow provided histopathology diagnoses, reviewed data analysis, and participated in preparation of the manuscript into its final form. All authors approved the manuscript in its final form prior to submission.