Previous studies have shown that pancreatic carcinogenesis is a multistep process involving well-defined precursor lesions (1
). Recent evidence suggests that the time frame from the onset of first genetic hit within pancreatic ducts until the development of metastatic disease spans about 2 decades, suggesting that clinicians have a reasonable diagnostic window for early detection strategies to help ameliorate the outcome of this deadly disease (33
). Modern imaging is a central component in diagnosing PDAC. However, the microscopic precursor lesions that a majority of PDACs are believed to arise from are usually undetectable on EUS, MR, or CT, hampering early detection. On the contrary, IPMNs are cystic precursor lesions of PDAC and can be readily identified by abdominal imaging (34
). IPMNs account for the most frequently diagnosed detectable cystic precursor lesion of PDAC and are currently the second most common indication for primary pancreatic surgery at The Johns Hopkins Hospital after PDAC. In fact, cystic lesions of the pancreas, in general, are being increasingly diagnosed. Recent imaging studies report prevalence rates of pancreatic cysts in asymptomatic individuals from about 2.6% on CT scans (35
) up to 13.5% on MRI (36
). Autopsy studies even suggest that up to one-fourth of the general population might harbor undiagnosed cystic lesions in their pancreata (37
Despite the advances in modern imaging techniques, an incidentally diagnosed pancreatic cyst is often a clinical challenge for both the patient and the treating physicians. The prevailing diagnostic uncertainty has resulted in accurate preoperative diagnoses in only up to two-thirds of asymptomatic pancreatic cysts that have been subsequently resected (26
), with the resulting potential for both over- and undertreatment in any given individual.
The primary objective of this study was to identify miRNA biomarker candidates that allow a prediction of dysplasia (high-grade vs. low-grade) in IPMNs. IPMNs with underlying high-grade dysplasia require surgical resection in light of the associated risk for synchronous or metachronous invasive neoplasia, whereas those harboring low-grade dysplasia can be followed conservatively. A secondary objective of this study was to determine whether additional cystic entities of the pancreas that can be followed with watchful waiting (e.g., SCAs) and can be differentiated from entities such as cystic PanNETs or SPNs that mandate surgery, using miRNA biomarkers.
Aberrant miRNA expression data established with well annotated, microdissected FFPE tissue specimens, which underwent a thorough histopathologic review represented “proof-of-concept” for use of miRNAs as biomarkers for pancreatic cystic lesions. With a high-throughput expression profiling approach, we were able to interrogate expression levels of 750 miRNAs, even in specimens with limiting RNA yield and preserve enough material in majority of the samples to carry out a verification step. Although target preamplification bias is a concern when using high-throughput expression platforms, data generated at Asuragen (unpublished) and from other research groups (38
) show that preamplification of miRNA-containing cDNA improves sensitivity of miRNA detection, while maintaining the relative expression levels. To our knowledge, this study is the first comprehensive demonstration of miRNAs aberrantly expressed in tissue specimens collected from patients with cystic lesions who have high-risk versus low-risk of malignancy.
Although we initially hypothesized that malignant epithelial cells shed from the cyst wall to the internal fluid will be detected in the cystic fluid, a majority of miRNA candidates identified through the FTS and CFS were nonoverlapping. This result may appear as a surprise, considering that millions of cells reported to be shed daily from a tumor into circulation (40
). One of the reasons for this lack of correlation may be that in contrast to strong aberrant expression signal observed in carefully microdissected neoplastic epithelium, the signal from malignant cells in the aspirated cyst fluid is likely to be weakened by contaminating nonneoplastic epithelium, endothelial cells, and others and therefore more difficult to detect. Our results also underscore the importance of directly profiling the most proximate biologic sample to be used for biomarker analyses (in this case, aspirated cyst fluid), rather than extrapolating candidates from tissue profiling experiments alone, as prior studies have done.
Logistic regression analysis led to the final identification of a 9-miRNA model, which allowed separation of all but one high-grade IPMNs from low-grade IPMNs/SCAs with 89% Se and 100% Sp. We re-reviewed the histology of the misclassified IPMN case and did not find any histologic features to the contrary. As we have recently shown (3
), independent locules within the same IPMN might harbor distinct genetic alterations, and it is possible that the cyst fluid was aspirated from a locule lined by lower grade dysplastic epithelium. As for the intermediate-grade specimens, they were separated by the model into those exhibiting high- and low-risk of malignant transformation (). Future prospective studies should clarify whether this separation of intermediate-grade IPMNs according to our miRNA panel reflects true differences in natural history within this IPMN subcategory or results from the inherent imprecision associated with grading epithelial dysplasia. Our miRNA signature also distinguished low-grade IPMNs/SCAs from uncommon cysts such as SPNs and cystic PanNETs. PanNETs are a highly heterogeneous group of tumors with diverse biologic behavior and occasional cystic appearance, for which a clear staging system has been historically challenging. Although they generally have a better prognosis than PDAC, even localized PanNETs will, over time, metastasize and thus, their surgical resection at diagnosis is warranted (41
). Solid pseudopapillary neoplasms (SPN) are lesions that present with an admixture of solid and cystic features and predominantly arise in younger women (43
). Although not an overt malignancy, the diagnosis of an SPN is generally regarded as an indication for resection as well. As both PanNETs and SPNs imply resection, it is beneficial that the miRNA candidates selected via logistic regression herein allow clustering of those lesions with high-grade IPMNs.
The most important miRNAs in the regression model included miR-24, miR-30a-3p, miR-18a, miR-92a, and miR-342-3p. Of these, miR-92a and miR-18a were previously reported to be involved in pancreatic cancer (18
). In agreement with previous reports (20
), in our study miR-21 was upregulated in low-grade/intermediate-grade/high-grade IPMNs, SPNs, and PanNETs as compared with SCAs. However, it did not seem to predict the risk level of malignant transformation and therefore, is likely to be less useful for this purpose.
With respect to therapeutic stratification, logistic regression correctly predicted indication for surgery in so-called higher risk lesions, as defined by current recommendations for surgery (high-grade IPMNs, PanNETs, and SPNs) versus those in which conservative management is considered appropriate (low-grade IPMNs, SCAs). Despite our promising findings, we recognize that our study has limitations. First, the model parameters, including the malignancy threshold, needs to be further evaluated in a large prospective study. In addition, here we examined differential miRNA expression using resected specimens only, which included lesions that would not have been surgically removed if diagnosed accurately (BD-IPMNs and SCAs). The resection of those lesions was likely mandated by either cyst size (>3 cm) or worrisome features on imaging. Furthermore, there are additional rare cystic lesions (e.g., MCNs, lymphangiomas) in which the expression of our miRNA candidates needs to be interrogated to determine their diagnostic/predictive potential. And finally, although SPNs and NETs are important diagnostic entities to differentiate from SCAs and low-grade/high-grade IPMNs, the numbers of specimens used in this study were rather low.
In conclusion, we identified a 9-miRNA model that may aid in diagnosis and surgical treatment decisions for patients with pancreatic cystic lesions, such as high-grade IPMNs, cystic PanNETs, and SPNs. We recommend that this signature is prospectively validated in larger series of consecutively collected cysts that have not been resected, with clinical follow-up or subsequent surgical findings serving as the gold standard. The validation study must follow CLIA and CAP regulations to ensure that this signature can be introduced into clinical use as an LDT. We also recommend that reference cystic fluid analyte, such as carcinoembryonic antigen, and the emerging genetic markers, such as GNAS
mutation status (3
) are incorporated into the study design. In addition, aspiration of multiple IPMN locules in patients with multifocal IPMNs should be considered, to reduce the probability of specimen bias and false negative results. A study that follows the abovementioned recommendations should allow establishing whether this 9-miRNA signature can be applied clinically toward stratification of patients with pancreatic cysts for surgical intervention or watchful waiting.