Circulating microRNAs as diagnostic biomarkers for pancreatic cancer

Abstract

There is an urgent need for novel and reliable biomarkers for the diagnosis and prognostication of pancreatic ductal adenocarcinoma (PDAC). Circulating microRNAs (miRNAs) have been extensively profiled in PDAC blood samples, but few studies have performed adequate validation of candidate markers. The evaluated study by Xu et al. investigated pre-operative plasma miRNAs from PDAC patients over three phases and three surgical centers. They revealed miR-486-5p and miR-938 were able to discriminate PDAC patients from healthy controls and those with chronic pancreatitis. The diagnostic ability of miR-486-5p for identifying PDAC from healthy controls was comparable to that of CA 19-9. This study provides further evidence for the use of blood-based miRNAs as diagnostic biomarkers in PDAC. However, as these have not been identified in previous studies these require further validation and methodology needs to be standardized if these are ever to be used in the clinic.

Keywords:

• miRNA
• plasma
• diagnostic
• biomarker
• pancreatic ductal adenocarcinoma
• PDAC
• chronic pancreatitis

Pancreatic ductal adenocarcinoma (PDAC) remains a major unsolved health problem with a 5-year survival of less than 5%.[1] This poor survival rate has not improved in the last decades and, most concerningly, PDAC has been predicted to become the second cause of cancer-related deaths in the USA by 2030.[2] One of the reasons of this dismal prognosis is the absence of biomarkers for early detection. The differential diagnosis remains challenging without invasive procedures, especially to differentiate PDAC from other periampullary tumors, benign lesions and chronic pancreatitis (CP).[3] The only blood-based biomarker, CA 19-9, is lacking sensitivity and specificity, warranting studies on more accurate circulating biomarkers.[4]

MiRNAs are small (19–24 nucleotides) non-coding RNAs playing an important gene-regulatory role in the post-transcriptional phase.[5] Several studies have identified key miRNAs involved in tumor development and progression [6] Importantly, miRNAs are stable in body fluids,[7] either in secreted microvesicles or bound to carrier proteins. This offers a unique opportunity to exploit miRNAs as potential biomarkers for early detection of various cancers, including PDAC.

Whilst the PDAC miRNome has been extensively profiled, therefore supporting circulating miRNAs as potentially useful for diagnosing PDAC, it remains unclear which differentially expressed miRNAs are most discriminative.[8,9] Small sample size, lack of adequate control/comparison groups, methodological and statistical challenges have limited previous studies.

In the evaluated study, Xu et al. [10] demonstrated 13 dysregulated plasma miRNAs in PDAC. MiR-486-5p exhibited diagnostic value in distinguishing patients with PDAC from healthy controls (HC) and patients with CP, whilst miR-938 showed promising results to identify PDAC from CP and other pancreatic tumors (OPT). The diagnostic value of miR-486-5p was comparable with CA 19-9 when discriminating PDAC from HC and CP.

Summary of the methods and results

Methods

Plasma samples were collected from six hospitals to create a multicenter platform to validate differentially expressed miRNAs. For the discovery experiment, three pooled samples from pre-operative patients with PDAC (n = 7), CP (n = 6) and HC (n = 5) were selected. For the validation phase, two separate cohorts (n = 76 and n = 363, respectively) were created including patients with pancreatic neuroendocrine tumors (PNET) and a OPT subgroup (made up of various pancreatic cystic tumors). The diagnosis of PDAC or OPT was confirmed by histology or cytology, but no information is provided on grading and staging, potentially limiting further analyses on miRNAs relevant for early detection and tumor progression. Furthermore, no other subgroups with periampullary tumors (i.e. arising from the bile duct, ampulla or duodenum) were included to further validate the diagnostic value of emerging miRNAs for PDAC and OPT.

In the discovery phase, RNA was extracted from the three pooled plasma samples from patients with PDAC, CP and HC and were analyzed for 671 human miRNAs using TaqMan low-density arrays (TLDA). Selection criteria included fold change >3 or <0.125 and P-values <0.05. Differentially expressed miRNAs were then evaluated in two validation cohorts by qRT-PCR. Currently, there is much debate how to normalize expression levels of circulating miRNAs.[7] Data was normalized using small nuclear (sn) U6 and miR-16. The sensitivity, specificity and area under the curve (AUC) for CA 19-9 levels of 271 subjects were compared with the new validated miRNAs.

Results

Comparing PDAC and HC, 15 miRNAs were significantly dysregulated, using both normalizers (i.e. snU6 and miR-16). Nineteen miRNAs were significantly dysregulated between PDAC and CP. Combining these results, 29 miRNAs emerged for further evaluation in the second cohort. Four additional miRNAs (i.e. miR-126-3p, miR-19b-3p, miR-486-5p and miR-942) were added for their known diagnostic roles in various cancers. However, no explanation was given as to why these four miRNAs were not identified in the initial discovery experiment. This is especially important since three of these miRNAs appeared after further validation as having good diagnostic accuracy for PDAC (Supplementary Table 1).

These 33 miRNAs were validated by qRT-PCR in a second cohort of patients (29 PDAC, 16 CP and 33 HC). A total of 13 were significantly dysregulated using both normalization methods. However, the authors reported that dysregulation of miRNAs could be contradictive between the two normalization methods, sometimes being up-regulated or down-regulated. Further validation of these emerging 13 miRNAs in a third, cohort (156 PDAC, 57 CP, 65 HC, 27 PNET, and 58 OPT) identified the potentially diagnostic miRNAs (Supplementary Table 1). Comparing PDAC to HC, miR-486-5p, miR-126-3p and miR-938 exhibited diagnostic value with AUC of 0.861, 0.618 and 0.693, respectively. Three other miRNAs, namely miR-663b, miR-19b-3b and miR26b-3p, showed promising results toward distinguishing CP and PNET from PDAC (Supplementary Table 1). However, miR-938 was the only marker able to identify PDAC from OPT with an AUC of 0.618. Probably, the different diseases included in the OPT group compromised the ability to find valuable miRNAs. Further sub-grouping is needed to identify more discriminative miRNAs useful for following premalignant lesions (e.g. intraductal papillary mucinous neoplasms). Furthermore, given the large sample size of the third cohort, it seems unfortunate that the authors did not correlate any of the candidate miRNAs with clinico-pathologic factors (e.g. lymph-node status, resection margin status, stage, etc.) or survival outcomes. Indeed, some PDAC-specific miRNAs are not only diagnostic, but also predictive/prognostic.[11–13] In this third cohort, 72.5% of the patients were resectable (i.e. stage I/II), so it would have been interesting to see if miRNA levels could have identified these patients pre-operatively.

The AUC values of some individual miRNAs were comparable to those for CA 19-9. MiR-486-5p was able to diagnose PDAC from HC and CP, and miR-938 levels could distinguish between PDAC and CP with similar accuracies to CA 19-9. Some studies have also combined circulating miRNA expression levels with CA 19-9 to improve diagnostic accuracy.[9,14,15] However, the diagnostic combination of miR-486-5p and miR-938 with CA 19-9 was not performed in this study.

Expert commentary

Lack of sensitive and specific biomarkers for PDAC adds to the existing problems of increasing incidence and poor prognosis of this lethal disease. Early detection of high-risk premalignant lesions such as pancreatic intraepithelial neoplasia or mucinous neoplasms, or early stage cancers would be the best option to improve patient survival.

The clinical use of serum CA19-9 as a biomarker is limited by its lack of sensitivity and specificity, whilst circulating miRNAs are gaining momentum as novel blood-based diagnostic tools. Nine previous studies have performed microarrays to identify differential circulating miRNAs in PDAC and found a total of 44 significantly differentially expressed miRNAs in blood from PDAC patients. Notably, only 2 miRNAs of the 6 most promising miRNAs in the present study (i.e. miR-126-3p, miR-26b-3p) emerged in previous circulating miRNA profiles of PDAC (Supplementary Table 1), stressing the importance of extensive validation studies.[16] An explanation for this may be the use of plasma instead of whole blood or serum.[9]

The methodology to evaluate miRNA profiles is another critical point. Whilst pooling of samples may be required due to the limited RNA availability or to help reduce the cost of microarray experiments, there are several implications.[17,18] If well designed, pooling can provide equivalent power and minimize the biological variance, but at a cost of losing all independent replication.[17,18] Furthermore, some feel that pooling should not be performed when the goal is to identify profiles that help to classify individuals and predict their membership in a group (e.g. cancer vs. healthy/benign patients).[18] There are limited details about the RNA pooling in the evaluated study and so it cannot be completely assessed. Due to missing basic information of the stage/grade of PDAC, disease duration, severity of CP, etc., the “type” of patients in the initial profiling cannot be gauged.

Many different platforms are available to evaluate differentially expressed miRNAs; for instance, the TLDA identifies only 671 of the 1,881 known human miRNAs (miRBase 21). With the increasing possibilities of Next-Generation Sequencing (NGS), we believe that miRNA-Seq should be the next step for blood-based biomarker discovery. Moreover, there is no consensus on optimal normalization for circulating miRNAs. Small nuclear U6 is less stable in fluids than in tissues,[7] whilst miR-16 is a red blood cell expressed miRNA, and variations in blood cell count and/or any hemolysis could have important implications.[19] Interestingly, highly expressed circulating miR-16 [14] and peripheral blood mononuclear cell miR-16 [15] levels have themselves been shown to discriminate PDAC from HC, and those with benign disease. Therefore, the combined use of snU6 and miR-16 might not be appropriate as endogenous control, whereas a miRNA with the least change in expression across the various samples in the array could have been chosen as a normalizer. Other studies have also used cel-miR-39 spike-in and then quantification in order to normalize the differences in RNA extraction efficiency and reverse transcription efficiency amongst samples.[20–22]

Notably, this is the first study presenting a multicenter, multistep validation of circulating miRNAs, and candidates were tested in a large cohort including CP, PNET and OPT. These approaches should increase the value and specificity of the dysregulated miRNAs. However, next to the absent overlap with previous studies, another limitation to be conquered is the specificity of the expression of miRNAs in tumors. Chen et al. [7] reported the overlapping expression of serum miRNAs in colorectal and lung cancer patients, suggesting some common serum tumor-related miRNAs. Importantly, the majority of circulating miRNAs might originate from blood rather than cancer cells.[19] Furthermore, co-expression of miRNAs in different diseases might reflect a general inflammatory response. Hence, blood-based miRNA expression could be a non-specific indicator of a tumor, rather than tissue-origin-specific.

With regard to the candidate miRNAs identified in the current study, miR-486-5p has been reported to be also dysregulated in gastric,[23] cervical [24] and non-small-cell lung cancers.[25] Except for lung cancer, all the results show an up-regulation of miR-486-5p in plasma/serum in cancer compared to HC. Both oncogenic and tumor suppressive functions have been described in different tumors.[26] Recently, it was shown that miR-486-5p can modulate the NF-kB pathway by targeting negative NF-kb regulators in gliomas.[27] The inhibition of this pathway by the histone deacetylase inhibitor, belinostat, can suppress PDAC growth in vitro and in vivo, indicating a possible role of miR-486-5p as an oncomiR. Another functional target is CD40, which is carried by antigen-presenting cells, which activated macrophages that rapidly infiltrated tumors, killed them, and depleted tumor stroma.[28] Thus, miR-486-5p could be important to enable immune evasion of tumor cells.

Over-expression of miR-938 has been associated with chemoresistant glioblastomas.[29] So far there are no functionally validated targets for miR-938. However, Butz et al. [30] predicted SMAD3 as possible target of miR-938 in non-functioning pituitary adenomas. This can decrease TGF-β pathway activation, skewing towards a more activated alternative pathway like Ras-MAPK, p38, c-Jun and PI3K-Akt, which are all known core pathways in PDAC. Another predicted target is TCF7L2, whose polymorphisms have been correlated to type 2 diabetes mellitus (T2DM).[31] Functional down-regulation of this protein via miR-938 could induce dysregulation of glucose homeostasis. Similarly, dysregulation of circulating miR-126-3p has been reported as a marker for impaired glucose intolerance and new-onset T2DM.[32] The role of these miRNAs in T2DM should be further investigated, also as a confounder in miRNA studies concerning PDAC, considering the correlation between T2DM and PDAC. The other miRNAs described by Xu et al. [10] have been detected in blood-based microarrays in different cancer types, questioning their role as a specific biomarker for PDAC.

Five-year view

Blood-based biomarkers are attractive for diagnostic purposes due to their non-invasive character. So far, the main problem is the lack of high specificity in diagnosing patients with PDAC and the need for validation in large prospective multicenter randomized trials and case–control studies.

Currently, there are tissue-based miRNA signatures with well-described functions in various tumor types. However, the functions of many dysregulated miRNAs have not been characterized. Thus, circulating miRNAs may be influenced by other processes like immune responses and might also not be cancer-specific. Therefore, another challenge lies in elucidating the biological processes in which emerging miRNAs play a role and how they might be used as diagnostic biomarkers, prognostic factors and/or innovative therapeutic targets.

Most likely, a panel of miRNAs will help to increase the AUC in the future instead of a single discriminating miRNA. Moreover, a pan-cancer miRNA panel [33] combined with other genomic or protein validated biomarkers should be investigated, considering also the trend towards certain miRNAs which are dysregulated in multiple cancers. For this purpose, NGS can be used to simultaneously discover different genomic alterations that may allow a better molecular understanding of a patient’s disease, as well as the identification of new biomarkers. Another extension to the usefulness of circulating microRNAs can be the isolation of actively secreted microRNAs in exosomes. Cancer-related microRNAs in vesicles might indeed be more specific.[34]

So far, the step towards the clinic has not yet been made for blood-based miRNAs in PDAC due to contradictive results and extensive validation that is needed. With increasing large-scale studies and optimization/standardization of technical procedures in this field, miRNAs as biomarkers could open up a new opportunity for diagnostic tools.

Key issues

• The study evaluated is the first multiphase, multicenter study to investigate plasma miRNAs in PDAC.

• MiR-486-5p and miR-938 were able to differentiate PDAC from HC and CP with good areas under the curve (0.861 and 0.706; and 0.693 and 0.754 respectively).

• They were unable to distinguish PDAC from PNET or other pancreatic tumors.

• Correlating pre-operative circulating miRNA levels in PDAC with clinico-pathologic factors may have revealed these miRNAs to also be useful predictive/prognostic biomarkers.

• MiR-486-5p and miR-938 had diagnostic accuracy comparable to CA 19-9, although their combination with CA 19-9 was not investigated.

• MiR-486-5p has been reported to be dysregulated in several cancers, suggesting a possible role as oncomiR, but further functional studies on targeted signaling pathways are warranted

Financial & competing interests disclosure

This work was kindly supported by The Bennink Foundation (Laren, the Netherlands). The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Supplementary material available online

Supplementary Table 1.

ORCID

Adam E Frampton http://orcid.org/0000-0002-1392-2755