MicroRNA expression profiling for the prediction of resistance to neoadjuvant radiochemotherapy in squamous cell carcinoma of the esophagus

FFPE tumor samples from 53 patients with locally advanced esophageal squamous cell carcinoma were investigated. The patients were treated between 1996 and 2009 at the Department of Radiation Oncology and Department of Surgery at the Klinikum Rechts der Isar der Technischen Universität München, Germany. The median age of the patients was 57.3 (range: 29–71 years) with 11 female (20.8%) and 42 male patients (79.2%), reflecting the expected gender distribution for this tumor. Tumor differentiation was G2 (moderately differentiated) in 20 cases (37.7%) and G3 (poorly differentiated) in 33 cases (62.3%).

Preoperative RCT consisted of simultaneously applied cisplatin (15 patients) or oxaliplatin (38 patients), and 5-fluorouracil based chemotherapy and external-beam radiotherapy radiation with overall doses of 30–60 Gy, predominantly 45 Gy (45 patients; single doses of 1.8–2 Gy) [8, 19]. Surgery was performed 4–6 weeks thereafter with esophagectomy as described in detail previously [8, 20]. Tumor regression grade (TRG) was histologically assessed in posttreatment resection specimens in a standardized way by microscopic evaluation of the complete previous tumor bed, as described previously [8, 21]. Cases with no residual tumor were classified as “responders” (TRG 1a) and cases with > 50% residual tumors were classified as “non-responders” (TRG 3). Tumors with incomplete or partial regression (1–50% residual tumors) were not included in this study because the data about the prognostic impact of incomplete and partial regression in the literature do not show consistent results [8, 22]. Postoperative histopathological findings after neoadjuvant radiochemotherapy are given in Table 1. Overall survival (OS) was calculated from the day of surgery to death.

Preoperative biopsies from 31 patients containing sufficient tumor material was used for microarray analysis. Confirmation by quantitative RT-PCR (qRT-PCR) was performed on an extended cohort including the cases from the microarray analysis and biopsy tissue from another 22 patients.

The tumor cell content of a minimum of 80% of the biopsies was required, and manual microdissection was performed to enrich the tumor content when necessary. Total RNA, including small RNAs, was extracted using the FFPE miRNeasy Kit (Qiagen, Hilden, Germany). For each sample, between 2 and 10 unstained 10-µm sections were manually microdissected after deparaffinization, and RNA was extracted using 150 µl of proteinase K digestion buffer according to the manufacturer’s instructions. Total RNA was measured using the NanoDrop photospectrometer (NanoDrop, Wilmington, DE, USA) and was further processed if the A260/A280 ratio was ≥ 1.8 and the A260/A230 ratio was ≥ 1.4. All samples were analyzed using RNA 6000 Nano LabChip Kits (Agilent Technologies).

The total RNA samples were spiked with in vitro-synthesized oligonucleotides (MicroRNA Spike-In Kit, Agilent Technologies). The spiked total RNA was treated with alkaline calf intestine phosphatase (CIP). Subsequently, the dephosphorylated RNA was labeled (miRNA Complete Labeling and Hyb Kit, Agilent Technologies) using the T4 RNA ligase, incorporating Cyanine-3-pCp. Next, 100 ng of total RNA per sample were introduced into the labeling reaction. After clean-up, Cyanine-3-labeled miRNA samples were prepared for one-color-based hybridization (Complete miRNA Labeling and Hyb Kit; Agilent Technologies). Each Cyanine-3-labeled miRNA sample was hybridized for 20 h at 55 °C on separate Human miRNA Microarrays, Release 16.0 (Agilent Technologies; AMADID 031181, 8 × 60 K format), containing probes for 1205 human and 144 human viral miRNAs. Thereafter, the microarrays were washed with increasing stringency using Gene Expression Wash Buffers (Agilent Technologies) followed by drying with acetonitrile (SIGMA). Fluorescent signal intensities were detected using Scan Control A.8.4.1 Software (Agilent Technologies) in the Agilent DNA Microarray Scanner.

For validation of the microarray data, differentially expressed miRNAs were investigated by qRT-PCR using the miRCURY LNA™ Universal RT microRNA PCR system (Exiqon A/S, Denmark). Details regarding the commercially available miRNA primer sets used and sequences for individual Custom PCR primer sets are given as Additional file 1. The cDNA samples were prepared using 20 ng of total RNA and the Universal cDNA synthesis kit (Exiqon A/S, Vedbaek, Denmark) according to the manufacturer’s recommendations. A 10-µl volume of a 50× dilution of cDNA was used in each of the real-time PCR reactions with SYBR^® green master mix and miRNA LNA™ PCR primer sets (both from Exiqon A/S) following the manufacturer’s instructions. All samples were run in triplicate. Real-time PCR was carried out on a Light Cycler 480 instrument (Roche Diagnostics), and the arithmetic mean of each triplicate measurement was used for further analysis. The relative quantification of miRNA expression was performed using a non-linear algorithm (second derivative maximum, Roche LightCycler Software V1.5) with a standard curve for each individual assay to calculate and correct for the amplification efficiency and using SNORD 44 as a reference for normalization.

To minimize the data variation in separate runs, a pool of 10 non-tumor samples was examined on the same runs. A no-reverse-transcriptase control (no RT) was included for each run of real-time RT-PCR to ensure that the RNA samples were not contaminated with genomic DNA.

For the microarray analysis, the software tools Feature Extraction 10.7.3.1 and GeneSpring GX 11.5 were used for quality control, statistical data analysis, miRNA annotation and visualization. A detailed description of the statistical workup of the microarray data is given as Additional file 2.

For statistical analysis of the confirmation experiments, SPSS software (IBM SPSS statistics version 24 was used. The associations between groups of patients were given in cross tabs, and differences were determined using χ²-test. Comparisons between groups were performed using the non-parametric Mann–Whitney U test and results were adjusted for multiple testing using the Bonferroni correction, where appropriate. Categorization into high and low miRNA-expressing tumors was conducted according the results of ROC analysis (non-response as reference). For survival analysis, the log-rank tests and Kaplan–Meier curves were used, as well as Cox proportional hazard tests for multivariate analysis. All tests were two-sided, and the significance level was set to p < 0.05 or lower, in line with the correction for multiple testing.

Thirty-one biopsy samples (16 responders and 15 non-responders) passed the quality check and could be used for microarray analysis. In general, the miRNA expression profiles in the ESCC biopsies examined were quite similar, both within the group of responders and non-responders: Correlation coefficients (r) within the group of responders ranged from 0.93 to 0.98 (average r = 0.96) and within the group of non-responders from 0.85 to 0.98 (average r = 0.94). Moreover, the miRNA profiles were rather similar between responders and non-responders with correlation coefficients ranging from 0.88 to 0.98 (average r = 0.95). The correlation (r) of the samples is visualized in a heat-map (Fig. 1).

Despite the quite homogeneous miRNA expression profile in pretherapeutic biopsies, a small number miRNAs fulfilled the strict predefined criteria for being designated as differentially expressed between responders and non-responders. The following eight miRNAs were upregulated in non-responders: hsa-miR-1323, hsa-miR-3678-3p, hsv2-miR-H7-3p, hsa-miR-194*, hsa-miR-3152, kshv-miR-K12-4-3p, hsa-miR-665 and hsa-miR-3659. By contrast, the four miRNAs hsa-miR-126*, hsa-miR-484, hsa-miR-330-3p and hsa-miR-3653 were downregulated in non-responders compared with that in responders (Table 2).

The complete list of analyzed miRNAs, their expression levels and respective p values for the comparison of the expression levels between responders and non-responders are given in Additional file 3.

Real-time quantitative PCR for the confirmation of microarray analysis findings was performed for 12 differentially expressed miRNAs in the expanded collective of 53 biopsy samples. However, four miRNAs (hsa-miR-1323, hsa-miR-3678-3p, hsa-miR-3152 and kshv-miR-k12-4-3p) could be detected at only very low levels by qRT-PCR, making valid quantification impossible, and the miRNAs hsa-miR-126*, hsa-miR-484, hsa-miR-330-3p, hsa-miR-3653, hsa-miR-194*, hsv2-miR-H7-3p, hsa-miR-665 and hsa-miR-3659 could be constantly detected in all samples and quantified. For miR-194* and miR-665 but not the remaining miRNAs, the microarray results could be confirmed with significantly higher miRNA levels in non-responders than in responders (median quantitative gene expression levels: 0.11 vs. 0.03, respectively, and 0.29 vs. 0.06, respectively; p < 0.001 each with a significance level of p < 0.006 after correction for multiple testing, Fig. 2). ROC analysis showed AUC values of 0.811 (95% CI 0.694–0.927) for miR-194* and 0.817 (95% CI 0.704–0.930) for miR-665 for the non-response to RCTX. The combination of miR-194* and miR-665 (sum of relative values in relation to the median expression level) had only a slightly better AUC value of 0.824 (95% CI 0.713–0.935). Crosstab analysis using the cut-offs defined by ROC analysis confirmed the association between high levels of miR-194* and miR-665 and non-response to neoadjuvant RCTX (p < 0.001 and p = 0.006; Table 3). Interestingly, for the miRNAs that were downregulated in non-responders, microarray results could not be confirmed in single qRT-PCR analysis.

Responders showed significantly better overall survival (OS) than non-responders (p < 0.001), and responders significantly more often had lower ypT categories and the absence of lymph node or distant metastases than non-responders (p < 0.001 each). Higher levels of miR-194* and miR-665 defined by ROC analysis were associated only in trend with a worse outcome of the patients in univariate analysis (p = 0.052 and p = 0.197, respectively; Fig. 3). In multivariate analysis encompassing the relevant prognostic pathological parameters and miRNAs, only the ypN category emerged as constant independent prognostic factors (HR: 7.561; 95% CI 1.6–35.1; p = 0.010; Table 4).

Ex vivo and in vitro studies with expression analyses of single miRNAs and miRNA microarray analyses have already been used by others to identify miRNA profiles in esophageal carcinomas that are potentially predictive for the response or resistance to chemo- or radiochemotherapy [23‐32]. There is considerable variety in the platforms used, tissue analyzed (i.e., FFPE vs. fresh (unfixed) tissue), case collection or neoadjuvant treatment. We screened the results of our microarray analysis to compare the results of these studies with the present investigation. An overview of miRNAs that have been described as differentially expressed between responding and non-responding esophageal carcinomas in relation to the results of the present work is given in Additional file 4.