NEW YORK (GenomeWeb) – While next-generation sequencing (NGS) has enabled the analysis of many new genetic variants associated with thyroid cancer, the increased sensitivity this technology can yield to diagnostic tests appears to come at the cost of specificity, according to a newly released study from molecular cytology firm Veracyte.
As a result, the company, which presented its data last week at the annual meeting of the Endocrine Society (ENDO), said that caution should be exercised in interpreting mutational data when it comes to diagnosing thyroid nodules as either benign or malignant.
Meanwhile, presenting at the same meeting, Interpace Diagnostics said that combining data from its oncogene panel with a soon-to-be-launched miRNA expression classifier could increase specificity.
Thyroid cancer is the ninth most common cancer in the US with an estimated 63,000 people diagnosed each year, according to the National Cancer Institute. Incidence of the disease is also increasing more rapidly than any other cancer, although this may be in part attributable to improved detection methods.
Biopsy of thyroid nodules is the most common method of confirming cancer with samples identified as either benign or malignant by a cytopathologist. For the most part this diagnostic approach is effective, but in about 15 to 30 percent of cases nodules are deemed indeterminate, leading to sometimes unnecessary surgical intervention.
In 2010, Veracyte launched its Afirma Gene Expression Classifier, which measures the expression of 167 genes to preoperatively reclassify ambiguous thyroid FNA as either benign or possibly cancerous. According to 2013 study in the New England Journal of Medicine, the test is able to pick out, with up to a 95 percent negative predictive value, low-risk nodules from those FNA cytology finds inconclusive.
As part of an effort to better understand how the number of gene variants analyzed for nodule diagnosis influences specificity and sensitivity, Veracyte conducted a study of 239 thyroid nodules from fine-needle aspiration (FNA) and surgical tissues. Researchers used NGS to measure increasing numbers of interrogated genomic sites for 14 cancer-linked genes, as well as 43 fusion pairs, and found that while sensitivity to detect malignancy improved as the number of genomic loci increased, specificity fell dramatically.
In FNAs, for example, the smallest nine-site panel used had a sensitivity of 53 percent and a specificity of 93 percent. The largest panel, with 3,670 genomic sites, had a sensitivity of 100 percent but a specificity of 10 percent. Meanwhile, in surgical tissues, the smallest panel had 45 percent sensitivity with 89 percent specificity. The largest panel had a sensitivity of 100 percent and zero percent specificity.
"Overall, the two larger panels wrongly called 87 [to] 90 percent of histology benign FNAs as malignant, while the two smaller panels missed 48-58 percent of known cancers," according to the study.
Broadly speaking, the data showed that "if we keep the curation very strict and only look at amino acid changes that have overwhelming support in the literature … we have very high specificity but we miss half the cancers," Veracyte CSO Giulia Kennedy told GenomeWeb. "When we dial down that stringency and start looking at amino acid changes that are less well-annotated in the literature … we find that most of the cancers will have at least one of these mutations, but so do many of the benign nodules."
She stressed that these findings shouldn't be viewed as a commentary on NGS as a technology, which she described as "very powerful," but rather highlight the importance of curation and "suggest that more study is needed to clarify the clinical utility of commonly used panels in physician decision-making."
"It's not the platform, it's how the data are used," Kennedy said.
The study results, she added, also underscore "the important role of the Afirma Gene Expression Classifier, with its high negative predictive value, in ruling out cancer in indeterminate thyroid nodules."
Also presenting data on thyroid nodule diagnosis at ENDO was Interpace Diagnostics, which aims to address specificity issues through the use of microRNA expression profiling alongside oncogene mutational analysis.
The Veracyte study demonstrates that "as you drive further into the genome on next-gen sequencing … you actually begin to lower specificity," Nancy Lurker, CEO of Interpace partner firm PDI, told GenomeWeb. "We agree, which is why we believe you need a combination approach of the microRNA gene expression coupled with next-gen sequencing."
For Interpace, this means joining its ThyGenX thyroid oncogene panel with its soon-to-be-launched ThyraMir miRNA expression classifier.
ThyGenX is based on Asuragen's miRInform Thyroid, which used Sanger sequencing to look for more than 100 genetic alterations that are associated with papillary and follicular thyroid carcinomas and can be used as a prognosticator of thyroid cancer.
PDI acquired miRInform Thyroid in mid-2014 and renamed it ThyGenX after updating it to use NGS and adding the oncogene PIK3CA.
ThyraMir, meantime, analyzes the expression levels of 10 cancer-linked miRNAs — miR-29b-1-5p, miR-31-5p, miR-138-1-3p, miR-139-5p, miR-146b-5p, miR-155, miR-204-5p, miR-222-3p, miR-375, and miR-551b-3p — using RT-qPCR and custom-designed PCR panels.
To demonstrate the utility of this combination, researchers from Interpace and Asuragen, along with collaborators at the University of Michigan, conducted a study to assess the use of miRNA analysis with nucleic acid sequencing to identify benign and malignant thyroid nodules.
The study, which appeared in a poster at ENDO, analyzed a total of 109 nodules with indeterminate cytology and surgical outcomes of primary benign or malignant thyroid lesions collected from 12 endocrinology centers in the US. Following FNA, samples were analyzed using both ThyGenX and ThyraMir, which were found to have a sensitivity of 89 percent and a specificity of 85 percent, resulting in clinically actionable negative predictive value at 94 percent and positive predictive value at 74 percent, according to Interpace.
These results show that "you get better results … from two different approaches rather than just one," Lurker said. "With both approaches, you optimize both specificity and sensitivity … [yielding] very good negative predictive value and much stronger positive predictive value than what's on the market today."
ThyraMir is set to launch in the second quarter.