Identification of the long non-coding RNA POU3F3 in plasma as a novel biomarker for diagnosis of esophageal squamous cell carcinoma

Esophageal squamous cell carcinoma (ESCC) is one of the most fatal malignancies in humans, causing more than 400000 deaths per year [1]. Patients trend to present with dysphagia at an advanced stage, and the 5-year survival rate is less than 15% [2]. Current biomarkers such as serum squamous cell carcinoma antigen (SCCA), carbohydrate antigen (CA) 19–9, and carcinoembryonic antigen (CEA) are the classic tumor markers commonly used in the management of patients with ESCC. However, these tumor markers, have limited utility in the early detection of ESCC due to lack of sufficiently high diagnostic sensitivity and specificity [3]. Although CYFRA 21–1 has been reported to have the higher sensitivity for diagnosing ESCC than the traditional tumor markers, the sensitivity is less than 10% for early detection of ESCC [4]. Therefore, the significance of exploration of new biomarkers with high sensitivity and specificity in early detection of ESCC should be emphasized.

Long non-coding RNAs (also known as lncRNAs), which are longer than 200 bases with lack of protein-coding capability, play critical roles in tumor initiation, progression and metastasis by modulating oncogenic and tumor-suppressing pathways [5, 6]. Previous studies have proved that lncRNAs are frequently dysregulated expression in different kinds of tumors, including ESCC [7]. The aberrant expression of lncRNAs have been reported to serve as potential diagnostic or prognostic biomarkers for many human malignancies such as breast, lung, liver, and colon cancers [8‐11]. Although these lncRNAs have shown great promise as a new kind of tumor markers, they cannot be used for clinical screening purposes because of difficulty in getting biopsies of tissue from patients suspected to be ESCC.

Circulating RNA in plasma or serum has been an emerging field for noninvasive diagnostic applications [12]. More recently, MicroRNAs (miRNAs) have been detected in human peripheral blood, being remarkably stable in spite of the high amounts of endogenous ribonuclease in the blood of cancer patients [13]. Moreover, several clinical trials have been approved by the FDA to assess the value of serum miRNAs in cancer diagnosis (http://​clinicaltrials.​gov) [14]. At present, several lncRNAs have been characterized as potential tumor markers in human fluids. For example, MALAT1 was found to be significantly up-regulated in plasma of prostate cancer patients, and could be used to discriminate cancer patients from healthy controls [6, 15]. In gastric cancer patients, plasma AA174084 levels dropped markedly on day 15 after surgery compared with preoperative levels and were associated with invasion and lymphatic metastasis [6, 15]. However, to our knowledge to date, no study has been performed regarding the circulating lncRNAs for early detection of ESCC patients.

In the present study, 10 lncRNAs (HOTAIR, AFAP1-AS1, POU3F3, HNF1A-AS1, 91H, PlncRNA1, SPRY4-IT1, ENST00000435885.1, XLOC_013104 and ENST00000547963.1) that were previously reported with deregulated expression in esophageal cancer were selected as candidate diagnostic makers [16‐23]. They were examined in tissues and plasma, and their potential use as tumor markers for ESCC detection were evaluated. We hypothesized that these ESCC-related lncRNAs might be released into the circulation during ESCC initiation and could be utilized to detect and monitor ESCC. To test the hypothesis, the following crucial questions need to be addressed: 1) the stability of circulating lncRNAs in plasma and serum, 2) the relationship between tumor tissue lncRNAs and circulating lncRNAs, 3) and the source of circulating lncRNAs (from cancer cells or from normal blood cells).

Recently, it has been demonstrated that the cell-free nucleic acids are detectable in plasma and serum of cancer patients and therefore may be utilized as a tool for cancer diagnosis [26]. Although numerous studies have focused on miRNAs as potential tumor markers for cancer diagnosis and prognosis prediction, the diagnostic utility of plasma lncRNAs in ESCC has never been studied.

In the present study, the initial ESCC-related lncRNAs screening was performed based on different expression profiling between ESCC tumor samples and matched normal samples that have been demonstrated in previous studies. All lncRNAs of interest were then subjected to qPCR validation. Eight lncRNAs were identified and then further measured their expression levels in plasma from ESCC patients and healthy subjects. The results demonstrated that the levels of POU3F3, HNF1A-AS1 and SPRY4-IT1 were significantly higher in plasma from ESCC patients compared with normal controls, providing strong evidence that ESCC-related lncRNAs could be released into the circulation and that their different expression profiles in plasma could be used as diagnostic markers for ESCC. Among the three lncRNAs, POU3F3 provided the highest diagnostic power for detection of ESCC (AUC = 0.842; sensitivity, 72.8%; and specificity, 89.4%), suggesting that plasma POU3F3 could serve as a promising tumor marker for ESCC detection. Furthermore, use of POU3F3 and SCCA in combination could provide a more powerful differential diagnosis between ESCC patients and healthy controls than use of POU3F3 or SCCA alone. Therefore, both in combination could be used as a diagnostic tool for screening apparently healthy individuals. Most importantly, the results indicated that plasma POU3F3 was more effective than SCCA for early detection of ESCC (69.2% vs 26%), and that the positivity rate of both in combination was significantly improved compared with SCCA or POU3F3 alone. In addition, we also proved the mirVana PARIS Kit (Ambion 1556, USA) approach as the most effective RNA extraction method, and both plasma and serum would be acceptable for evaluation of lncRNAs as blood-based tumor markers. This is the first time to systematically characterize circulating lncRNAs in plasma as diagnostic markers for ESCC.

Recently, many tumor markers such as SCCA, CEA, CA19-9, MMP-9, IL-6, CYFRA 21–1, DKK-1, M-CSF, MiR-18a, MiR-1246 and many other genes were evaluated for ESCC diagnosis [3, 27‐30]. However, there was no sufficient sensitivity and specificity for these biomarkers, even for the most commonly used biomarkers such as serum CEA and SCCA. Mealy K reported that the individual sensitivities of CEA and SCCA for the diagnosis of ESCC were about 28% and 32%, respectively [28]. Yamamoto K study demonstrated that the sensitivity of CYFRA 21–1 was only 47.9%, although the specificity was 100% [31]. MMP-9 has been shown to have higher diagnostic sensitivity of ESCC compared to SCCA, but its diagnostic performance was poorer [31]. Serum MiR-1246 improved the discriminatory power, but sensitivity and specificity were lower [30, 32]. This study was sought to find novel markers that could improve early detection of ESCC. Our results demonstrated that plasma lncRNA POU3F3 was a promising tumor marker, which effectively supplemented the serum SCCA for ESCC detection.

In the past several years, qPCR has been considered to be a reliable method for quantitative gene expression due to its accuracy, sensitivity, specificity, reproducibility and robustness [33]. However, to produce accurate results in qPCR assays the use of robust normalization strategy is important [34]. Generally, the use of reference gene as endogenous controls is the most commonly method for normalizing qPCR gene expression data. Currently, several mathematical algorithms which are specifically developed for reference gene evaluation and selection deliver suitable reference genes with the lowest variation and with high stability across the biological samples. The four commonly used approaches are: (1) NormFinder algorithm which is a statistical model that estimates the overall variation of gene expression for each candidate reference gene and delivers a stability value. (2) GeNorm. (3) BestKeeper, a Microsof Excel-based tool, uses pair-wise correlation and (4) the comparative delta Ct [35]. Therefore, in this study we paid special attention on reference genes selection. All candidate reference genes for ESCC tissue lncRNAs normalization were chosen based on their stably expressed in different human tissues or were once used in ESCC qPCR study. Finally through mathematical algorithms, GAPDH was selected for normalization of tissue lncRNAs measurement because of its best reference performance (Table 1). This was in agreement with Li W [18], using GAPDH normalization lncRNA POU3F3 expression in tissue samples of ESCC patients. As there was no established endogenous control for detection of plasma or serum lncRNAs, GAPDH was selected as a potential reference gene. And we observed that plasma GAPDH expression level was stable between the ESCC group and normal control group and was not affected by age, sex, and pathology. However, in the previous study by Ng EK et al. [36]. GAPDH seemed to be no feasible reference gene because of its expression level in HCC patients was significantly higher than those in health individuals. One possible explanation was that different diseases and selected primer pair was not the best choice for measuring gene expression could influence the expression of GAPDH.

Circulating lncRNAs were thought to be unstable because of the high level of RNase activity in plasma, and in cancer patients, increased plasma RNase has been detected [12]. In the present study, we confirmed that circulating lncRNAs were remarkably stable even when treated directly with RNase A digestion. These findings were consistent with those in patients with prostate cancer [6]. However, the precise mechanism used to explain why circulating lncRNAs are resistant to endogenous RNase digestion remains largely unknown. One explanation is that they are packaged in some kinds of microparticles, such as exosomes, microvesicles, apoptotic bodies, and apoptotic microparticles [14, 37]. Our results of the blood-processing assay support this explanation. Recently, it has been hypothesized that circulating RNAs could be modified in certain ways, including methylation, adenylation, and uridylation which make them resistant to RNase digestion [38]. Another possible explanation is that lncRNAs often forms secondary structures, and relatively more stable, which could facilitate their detection as free nucleic acids in body fluids such as blood and urine [39]. The mechanism of resistance of circulating lncRNAs to RNase deserves further study. In the present study, we also demonstrated that plasma lncRNAs were resistant to multiple freeze-thaw cycles, strong acid and base treatment. However, when plasma or unprocessed blood was subjected to extended room temperature incubation, the concentrations of lncRNAs were even more variable. Such artifactual fluctuations in lncRNAs concentrations may be attributable to two factors: the secreted lncRNA from necrotic and/or apoptotic blood cells; the stability of original and the newly secreted lncRNA [12]. Taking the unprocessed EDTA-blood for instance (Figure 5F); an initial increase in lncRNAs concentrations at 6 h may be the result of the newly released lncRNAs from blood cells. After that, the lncRNA concentrations started to decline which may be attributable to the degradation of the newly secreted lncRNAs. Based on the above results, we recommend that the unprocessed EDTA-blood should be stored at 4°C.

The release of nucleic acids into the blood is thought to be associated with apoptosis and necrosis tumor cells from the tumor microenvironments and is also the results of secretion [40, 41]. In the previous study of circulating lncRNA LIPCAR in heart failure patients from Kumarswamy and his colleges, who suggested that a good proportion of mitochondrial lncRNAs detected in circulation might come from the heart [42]. On the contrary, Prichard et al. provided the pioneering evidence that blood cells were the major contributor to the circulating miRNAs and that perturbations in blood cell counts and hemolysis could alter plasma miRNAs levels by up to 50-fold [25]. Another possible origin may be that the circulating miRNAs of cancer patients were not just a result of cancer but may actively contribute to cancer defense, because the population of circulating miRNAs correlated tightly with the immune response [40]. Currently, there were three major hypotheses for circulating miRNAs to enter into the circulation: energy-free passive leakage of cellular miRNAs into circulation; active and selective secretion of miRNAs in response to various stimuli as microvesicle-free miRNAs; and active secretion via cell-derived microvesicles [38]. These theories could also be utilized to explain the origin of circulating lncRNAs. In this study, we found that ESCC-related lncRNAs could enter into the cell culture medium at a measurable level. Furthermore, ESCC-related lncRNAs were derived from tumor cells as evidenced with the xenograft assay, the relationship of ESCC-related lncRNAs levels between the tumor samples and plasma matched for the same individuals, and the changes of plasma ESCC-related lncRNAs concentrations between pre-Op and post-Op. However, the delayed blood processing assay suggested that the circulating lncRNAs could also be released from blood cells although their contribution was minor. Therefore, to avoid the origin of lncRNAs from blood cells, the unprocessed blood should be stored at 4°C and further processed as soon as possible. In addition, the stable levels of the three lncRNAs in the filtered plasma could be used to explain why increased plasma lncRNAs detected in unfiltered blood samples were largely from microparticles. Some patients, however, exhibited a different pattern of lncRNAs expression, low expression in plasma with high levels in ESCC tissues. At present, it was difficult to explain such phenomena; one possible hypothesis for that phenomenon could be the heterogeneity of primary tumor. At present, the exact biological roles of circulating lncRNAs in cancer patients remain unclear. Are they merely molecular remnants of necrotic tumor cells or do they play important roles in cell-to-cell communication? Obviously, more investigations will be needed to solve the exciting questions.

In summary, our results suggested that increased plasma lncRNA POU3F3 can be used as ideal tumor biomarker for ESCC detection, and that combination of POU3F3 and SCCA has a higher positive diagnostic rate of ESCC than POU3F3 or SCCA alone, in particular at its early stage.

Yu-suo Tong, Xiao-Wei Wang and Xi-lei Zhou considered as joint first authors.