MicroRNA expression values depend on a selected internal control
Pancreatic ductal adenocarcinoma is one of the most frequently occurring solid cancers and it carries an extremely poor prognosis. As such, an extensive search for biomarkers of early disease is undergoing, miRNAs may have the ideal characteristics to fulfil this role. Due to their stability and resistance against RNase degradation, they are viable in a wide range of samples. Viable miRNAs for PDAC diagnosis may be isolated from frozen and paraffin embedded tissue samples, stool, blood plasma, or pancreatic juice, [24, 28, 30, 31].
For our analysis we have selected miRNAs, frequently described to be dysregulated in various types of PDACs samples. Studies mapping microRNA expression using microarrays have proven considerable heterogeneity in pancreatic carcinomas. Zhang et al. have demonstrated relative expression values miRNAs spanning 6-logs (from 0.01–10,000) among individual cases . Among tumor samples we determined up to 45-fold variability in both miR-21 and miR-155 levels. During our brief review of literature we have noticed that the mean values of miRNA-levels in tumors varied among studies. There are many factors including differences in reagents/materials for miRNAs quantification protocols and data-processing algorithms, which can contribute to the variation. One of these factors is a variety of controls, which were used for normalization. Thus, the differences in the mean expression of miRNAs may be at least partially explained by the choice of controls for normalization.
For example, when normalizing with snoRNA U6, Bloomston et al. measured a median 2.2-fold increase in tumor miR-21 levels . Zhang et al., using the same internal control, found that expression of miR-21 was up-regulated up to 6888-fold in several tumors . Hong et al. reported about up to 550-fold increase in PDACs, normalizing relative to U6 . When using both U6 and 5S as endogenous controls, du Rieu et al. detected a 20.1-fold tumor miR-21 up-regulation . In our study, when normalizing with U6, a mean 5.5-fold increase in miR-21 in tumors was present. However, when normalizing to miR-16 a 7.03-fold increase was present (p < 0.01, Table 4). Wang et al. detected in plasma samples with miR-16 only a mean 2.42-fold up-regulation . On the other hand, when normalizing to the artificial spike or with U91 we detected a mean 14.56-fold and 17.71-fold increase (p < 0.01, Table 4).
The data about miR-96 expression in PDAC are controversial. Several groups of authors reported about miR-96 expression fold increase in experiments with microarray [24, 32]. For example, Bloomston et al. measured an average 1.77-fold increase, when determining miR-96 levels in PDACs . Kent et al. also demonstrated 2.7-fold upregulation of miR-96 in pancreatic cancer cell lines . On the other hand, miR-96 has been shown to be frequently down-regulated in experiments, utilizing Northern blot or RT-qPCR [13, 15, 25, 31, 33]. Szafranszka et al. determined in their study a −4.35-fold decrease in tumor miR-96 expression, when normalizing to miR-24 . Hong et al. as well as Feng et al. showed that miR-96, normalized to U6, was downregulated in PDAC samples up to −8-fold [31, 33]. With U6, miR-16 and combination of U6 + U91 respectively, we demonstrated a statistically significant mean −3.22-fold, −2.32-fold and −1.85-fold decrease in tumor tissue (p < 0.01, Table 4). However, expression analysis with the artificial spike and U91 alone yielded a statistically insignificant alteration in miR-96 expression in tumors in comparison with normal tissues (p > 0.05, Table 4).
miRNA miR-148a expression is described to be down-regulated in PDAC due to promoter hypermethylation, which represents an early event in pancreatic carcinogenesis . Bloomston et al. as well as Jamieson et al. measured an average −5.5-fold and −7.14-fold decrease respectively, when determining miR-148 expression with a microarray [24, 29]. In experiments, utilizing RT-qPCR, Szafranszka et al. demonstrated −6.15-fold decrease of miR-148a levels, with miR-24 as normalizer . However, Ma et al. and Zhang et al., normalizing to U6, determined a respective −2.86-fold and −2.5-fold downregulation in PDAC samples [34, 35]. Hanoun et al. also reported about miR-148a down-regulation, using U6 like the endogenous control . In our study, tumor miR-148a levels were −2.04-fold and −1.33-fold decreased with U6 and miR-16 as a normalizers, respectively (p < 0.01 and p < 0.05, Table 4). On the other hand, analysis of miR-148a expression, normalized to the alien spike, U91 and a combination of U6 and U91 did not determine statically significant differences in expression between cancerous and non-cancerous tissues (p > 0.05, Table 4).
The miR-155 is an onco-miR, overexpressed in early pancreatic adenocarcinoma precursors and invasive PDAC . The miR-155 expression in PDACs and pancreatic cancer cell lines, measured by microarray, ranged from 1.8 to 2.9-fold in different studies [24, 28, 29, 36, 37]. On the other hand, Habbe et al. found a mean 11.6-fold increase in intraductal papillary mucinous neoplasms, which was measured by RT-qPCR relative to U6 . Zhang et al. also used U6 like a normalizer in their study. They reported about up to 52-fold increase in individual cases . In our pancreatic carcinomas a mean 5.05-fold increase was present, when normalizing to U6 (p < 0.01, Table 4). However, Ma et al. measured only a 2.11-fold increase with the same endogenous control . Wang et al. determined a 3.74-fold increase in serum miR-155 levels in cancer, when normalizing with miR-16 . Our PDAC samples showed, on the other hand, a mean 6.39-fold increase with miR-16 as internal control (p < 0.01, Table 4). However, the expression levels were several times higher, measured relative to the alien spike or U91 - 15.1 and 13.36-fold respectively (p < 0.01, Table 4).
The miR-196a is an onco-miR reported to be frequently dysregulated in PDAC [27, 30]. Zhang et al. measured, when normalizing to U6, 0.35-1557-fold variations in tumor miR-196 expression . In our tumors we determined a mean −2.2-fold decrease, when normalizing to U6 (p < 0.01, Table 7). Wang et al. demonstrated 16.05-fold increase in plasma samples with miR-16 as the endogenous control . On the other hand, for miR-16, as well as for the alien spike or U91 we did not find significant differences in miR196a expression between cancer and normal tissues (p > 0.05, Table 4).
The miR-217 inhibits in vitro tumor cell growth and it functions as a potential tumor-suppressor by influencing the Akt/KRAS signaling pathway, therefore, miR-217 is frequently down-regulated in PDAC. MicroRNA miR-217 was down-regulated 10-fold in the study by Szafranszka et al., normalized relative to miR-24 . However, Greither et al. determined only a mean −2-fold decrease with 18S as internal control . Ma et al. demonstrated −3.91-fold decrease, using U6 for normalization . On the other hand, Hong et al. found, that expression of miR-217 was down-regulated up to −62.5-fold in PDACs. They also used U6 like internal control . In our samples, miR-217 expression was significantly down-regulated across all internal controls, with a maximum −24.39-fold decrease with U6 and a minimum −7.19-fold decrease with U6 + U91 combination (p < 0.01, Table 4).
Thus, for miRNAs with high positive or negative expression levels, such as miR-21, mir-155 or miR-217, fold change values may differ up to several times, depending on selected internal controls. Moreover, different internal controls can produce controversial results for miRNAs quantification, as it was demonstrated for miR-96, miR-148a or miR-196a.
Comparing internal controls: U91 is a new endogenous control for microRNAs quantification in pancreatic cancer
RT-qPCR quantification of tumor miRNA expression may play an essential role in PDAC diagnostics, chemotherapy resistance and survival prediction. RT-qPCR-based experiments require endogenous controls for the result normalization, reliability and reproducibility. U6 small nuclear RNA [8–11, 14, 15, 27, 30, 37, 38], 18S  and 5S ribosomal RNAs [8, 15, 39, 40], small nucleolar RNAs RNU48, RNU43, RNU44 – commercial available Applied Biosystems assays , or miR-16 [30, 42, 43] were often used as the endogenous controls for miRNAs expression evaluation. However, there are data indicating, that expression levels of these reference genes may differ significantly in neoplastic and normal tissues [17–19]. This expression instability may introduce bias, when determining miRNA dysregulation in tumors. For example, U6 small nuclear RNA was the most common internal control [8–11, 14, 15, 27, 30, 38] for the quantification of miRNAs expression in PDAC. However, there are data, implying that U6 expression may be unstable in breast and cervical cancers [17, 19, 42]. Also, the amount of U6 may vary significantly in serum samples from patients with breast and colorectal cancers [18, 42]. According to our findings, U6 expression may show as high as an 8-fold difference in PDAC and normal pancreatic tissue (Table 8). On the other hand, U6 was determined as the second most stable gene by the NormFinder algorithm (Table 5). U6 also demonstrated greater expression stability in breast carcinoma tissue samples when compared with the snoRNAs RNU44, RNU48 and RNU43. Furthermore, changes in levels of these snoRNAs correlated with tumor morphology and patient prognosis . However, U6 alongside 5S and miR-16 showed remarkable expression variability in tissue samples from patients with breast carcinoma .
The data about miR-16 expression in serum samples from the breast cancer patients are controversial. On the first look, this miRNA demonstrated significant expression variations [18, 42]. On the other hand, analysis with the geNorm algorithm has identified miR-16 as well as miR-425 as the most stable normalizers . According to our measurements, expression of miR-16 varied significantly in pancreatic carcinomas (p < 0.01, Table 7). In addition, miR-16 was marked by the NormFinder algorithm as the least stable of the analyzed endogenous controls (Table 5).
Another possibility for RT-qPCR result normalization is the use of alien spike RNAs, such as miR-39 from C. elegans [18, 44], as internal controls. Also, these spike RNAs should be selected while taking into consideration that the same RNA sequences may already exist in the human genome. Surprisingly, according to the NormFinder analysis, the artificial spike was the least stable control (stability 0.085; Table 5). It must be taken into consideration, that the NormFinder algorithm can calculate variations between two groups including all normal and cancer samples, but it is unable to evaluate the differences between normal and cancer tissues from the individual patients. Accordingly, this may be the reason for the “instability” of the alien spike.
In this study, we compared the expression of 4 internal controls to determine the most stable of them. On the first look, the best internal control is the artificial spike, due to its amplification curves and threshold cycles, which have demonstrated to be very close for cancers and normal tissues (Table 7). On the other hand, according to the results, yielded by the NormFinder analysis, the best normalizer is the combination of U6 and U91. This combination has the best stability value, but as normalization results show, it differs significantly from the artificial spike (p < 0.01, Table 6). The most stable gene, determined by NormFinder, was U91 (Table 5). Each miRNA normalized relatively to the spike or U91, demonstrated similar expression values. Thus, statistically significant and insignificant differences between tumors and normal tissues for miRNAs were equal for the spike and the U91 (Table 4). Also, the differences between the spike and U91 were statistically insignificant for all of miRs except of miR-217 (Table 6). Among three endogenous controls, the U91 had the lowest average expression values and standard deviation in cancer tissues (Table 8).
Thus, we recommend U91 as a new normalizer of miRNA expression in pancreatic adenocarcinoma.