Supplementary Information
Increased expression of miR-126 and miR-10a predict prolonged relapse-free time of primary ER-positive breast cancer following Tamoxifen treatment
Reiner Hoppe, Joanna Achinger-Kawecka, Stefan Winter, Peter Fritz, Wing-Yee Lo, Werner Schroth, Hiltrud Brauch
Index
1. Methods (pages 1-4)
2. Supplementary tables (pages 5-8)
3. References (page 9)
1. Methods
1.1. Tissue processing and total RNA extraction
Formalin-fixed paraffin-embedded (FFPE) tissue blocks stored at the Department of Pathology at RBK were subjected to total RNA extraction from 6 consecutive 10-micron sections of each tumour block flanked by 3-micron sections used for haematoxilin and eosin (H&E)-staining for the evaluation by an experienced pathologist (PF). For the discovery set (12 patients) marked tumour areas of all six serial tissue sections were macrodissected and pooled. For the validation set (81 patients), total RNA was extracted from a single 10-micron section. All specimens had a mean tumour cell content (TCC) greater than 30% as by H&E staining, with 62 samples (76.5%) having a TCC of greater than 50%.
Total RNA was isolated using RecoverAll Total Nucleic Acid Isolation Kit (Life Technologies, Inc., Grand Island, NY, USA) according to the manufacturer’s specifications with minor modifications. Briefly, de-paraffinization was performed using Microclear (Merck, Germany) and the quantity of total RNA was maximized by incubation at 50°C for 3 hours. RNA quality and quantity were assessed by measuring absorbance on a NanoDrop1000 Spectrometer (Thermo Fisher Scientific, Inc., Waltham, MA, USA). Total RNA yield ranged from 1.92 to 26.34 µg for the discovery set (average 8.52 µg), and 0.28 to 8.39 µg for the validation set (average 2.39 µg). Total RNA was stored at -80°C until further use.
1.2. MicroRNA microarray and TaqMan quantitative reverse transcription-PCR analyses
MicroRNA expression profiles were generated utilizing GeneChip miRNA2.0 arrays (Affymetrix Inc., Santa Clara, CA, USA) with 300 ng of each total RNA sample according to the manufacturer’s instructions. The profiling series included the 12 tumour samples of the discovery set and an additional 31 tumour samples from the entire RBK cohort tumours not selected by a specific therapy (data not shown). Affymetrix software miRNA QC Tool (Version 1.1.1.0, Affymetrix, 2010) was used for quality control of microarrays and preprocessing of expression data by robust multichip average (RMA)1–3. Resulting data were on log2-scale and preprocessed data of the discovery set (n=12) were extracted for this study.
For the validation of differentially expressed candidate miRNAs commercially available TaqMan MicroRNA assays and reagents from Applied Biosystems were used as described with slight modification (Applied Biosystems, Foster City, CA, USA). Briefly, 10 ng total RNA have been reverse transcribed in a 15 µl cDNA reaction. Subsequently, 10 µl qRT-PCR reaction mixtures were prepared and ran on a 7900HT Fast Real Time PCR System (ABI) using 1.33 µl of 1:4 diluted cDNA cocktail. Data were collected and analyzed using RQ Manager and thresholds were manually determined. The mean Ct values of triplicate qRT-PCR reactions were calculated and the outliers were identified based on coefficient of variance larger than 5% among triplicates.
For normalization of qRT-PCR results, we identified 7 microRNAs showing sufficient and most stable expression levels on the microarrays across all samples. All possible weighted averages of these microRNAs (with sum of absolute weights=100%) were tested and the combination of miR-16, miR-23a and miR-24, given by was identified to be most adequate for normalization. Accordingly, this equation was applied to the mean Ct values of miR-16, miR-23a and miR-24 to obtain a normalization factor for each tumour sample. Delta Ct values were then calculated as differences between mean Ct values of each microRNA and normalization factors. These relative amounts were further adjusted by subtracting delta Ct values from the highest Ct value (Ct=40) among all samples.
1.3. Statistical analyses
Fisher’s exact tests, Wilcoxon-Mann-Whitney tests, and Cochran-Armitage trend tests were applied as appropriate to check for differences in clinicopathological parameters between relapse and non-relapse in the validation set (Table 1). In addition, associations of clinicopathological parameters and relapse-free time (RFT) were assessed by univariate Cox regression in the validation set. Here, RFT was defined as the time from date of diagnosis to date of relapse (censored event) or last date of follow-up (uncensored event). Wilcoxon-Mann-Whitney tests, Kruskal-Wallis tests, and Spearman correlation tests were used as appropriate to study univariate correlations between clinicopathological parameters and candidate microRNAs in the validation set. Moreover, for each candidate microRNA, a linear model with 9 clinicopathological parameters (age at diagnosis, tumour size, nodal status, grade, radiotherapy, PgR-status, ER-status (IRS), HER2/neu-status, and mean TCC) was calculated to study multivariate interrelationships.
Pearson correlations were computed to compare i) microRNA expression levels generated by microarray profiling and TaqMan (discovery set) and ii) expressions of different microRNA candidates in the validation set. Differences in microRNA expression between cases with and without relapse were investigated by i) paired t-tests for the 6 matched pairs in the discovery set or ii) Welch’s t-tests for the patients in the validation set (34 relapse vs. 47 non-relapse).
Associations between microRNAs and RFT in the discovery and validation set were analyzed by multivariate Cox regression with correction for age at diagnosis, tumour size, nodal status, radiation treatment, and mean TCC. For the validation set, the obtained results were checked and confirmed by multivariate Cox regression with correction for all nine clinicopathological parameters.
For each of the candidate microRNAs measured in the validation set (n=81), receiver operating characteristics (ROC) curve analysis was applied to study their performance in the classification of relapses and non-relapses. Corresponding area under curves (AUC) along with 95% confidence intervals were calculated: Quoted sensitivities and specificities refer to cut-offs determined by maximizing Youden’s index (=sensitivity+specitivity-100%) of ROC curves4. These cut-offs were used to perform Kaplan-Meier analyses of RFT for the candidate microRNAs, where groups of patients with high and low microRNA expression were compared by log-rank tests.
All analyses were conducted by R-2.15.05 and additional packages coin 1.0-21 and survival-2.36-126, GraphPad Prism version 5.00 for Windows (San Diego, California, USA) or IBM SPSS Statistics for Windows, Version 20.0 (Armonk, NY). Where indicated P-values were adjusted for multiple testing by Holm’s procedure7. Statistical significance was defined as P< 5%.
2. Supplementary tables
Supplementary Table 1 Pearson correlation coefficients among microRNA candidates in the validation set (n=81).
miR-10a / miR-126 / miR-375 / miR-210 / miR-31miR-10a / 1 / 0.28 / 0.06 / 0.31 / 0.42
miR-126 / 0.28 / 1 / 0.16 / 0.16 / 0.34
miR-375 / 0.06 / 0.16 / 1 / 0.25 / -0.01
miR-210 / 0.31 / 0.16 / 0.25 / 1 / 0.27
miR-31 / 0.42 / 0.34 / -0.01 / 0.27 / 1
Supplementary Table 2 Univariate Cox regression of clinical and histopathological characteristics for relapse-free time (RFT) in the validation set (n=81).
Age at diagnosis
Tumour size a
Nodal status
Grade b
Radiotherapy
PgR status
ER score (IRS)
Her2/neu status
Mean TCC / 1.03 (0.99 – 1.08)
1.17 (0.66 – 2.06)
1.40 (0.68 – 2.87)
1.90 (1.10 – 3.25)
1.05 (0.51 – 2.13)
2.56 (0.60 – 10.96)
0.90 (0.78 – 1.04)
1.88 (0.65 – 5.46)
1.03 (1.00 – 1.05) / 0.160
0.597
0.361
0.020
0.897
0.205
0.158
0.109
0.042
a T1&T2 vs T3&T4
b Grade 1&2 vs Grade 3
Abbreviations: CI, confidence interval; ER, estrogen receptor; Her2/neu, human epidermal growth factor receptor 2; IRS, Immunoreactive Remmele Score; PgR, progesterone receptor; TCC, tumour cell content.
4
Supplementary Table 3 Associations between candidate microRNAs and clinical and histopathological characteristics in the validation set (n=81).
Unadjusted P (univariate tests) / Unadjusted P (multivariate linear models) aCharacteristics / miR-126 / miR-375 / miR-210 / miR-10a / miR-31 / miR-126 / miR-375 / miR-210 / miR-10a / miR-31
Age at diagnosis / 0.609 b / 0.390 b / 0.341 b / 0.891 b / 0.498 b / 0.785 / 0.114 / 0.367 / 0.910 / 0.936
Tumour size / 0.502 c / 0.670 c / 0.133 c / 0.207 c / 0.684 c / 0.832 / 0.739 / 0.186 / 0.258 / 0.863
Nodal status / 0.375 d / 0.387 d / 0.045 d / 0.988 d / 0.635 d / 0.105 / 0.835 / 0.031 / 0.742 / 0.657
Grade / 0.038 d / 0.283 d / 0.306 d / 0.335 d / 0.336 d / 0.008 / 0.913 / 0.285 / 0.847 / 0.925
Radiotherapy / 0.836 d / 0.387 d / 0.366 d / 0.961 d / 0.607 d / 0.502 / 0.221 / 0.469 / 0.79 / 0.910
PgR status / 0.104 d / 0.224 d / 0.767 d / 0.862 d / 0.719 d / 0.101 / 0.398 / 0.216 / 0.394 / 0.699
ER score (IRS) / 0.089 b / 0.688 b / 0.514 b / 0.004 b / 0.914 b / 0.048 / 0.986 / 0.836 / 0.013 / 0.890
Her2/neu status / 0.985 c / 0.370 c / 0.290 c / 0.844 c / 0.694 c / 0.449 / 0.817 / 0.637 / 0.992 / 0.920
Mean TCC / 0.864 b / 0.242 b / < 0.0001b / 0.769 b / 0.924 b / 0.461 / 0.573 / 0.01 / 0.569 / 0.596
a For each microRNA, a multivariate linear model with the 9 listed characteristics was computed
b Spearman’s Correlation test
c Kruskal-Wallis test
d Wilcoxon-Mann-Whitney test
Abbreviations: PgR, progesterone receptor; ER, estrogen receptor; Her2/neu, human epidermal growth factor receptor 2; IRS, Immunoreactive Remmele Score; TCC, tumour cell content.
4
Supplementary Table 4 Cox regression of candidate microRNAs in the validation set (n=81) with correction for the five clinicopathological parameters used in the analysis of the discovery set.
MicroRNA probe set / Multivariate Cox regression aHazard ratio (95% CI) / P (Wald test) / Holm-adj. P
miR-210 / 0.63 (0.43-0.94) / 0.025 / 0.075
miR-31 / 0.80 (0.64-0.98) / 0.035 / 0.075
miR-10a / 0.55 (0.37-0.80) / 0.002 / 0.008
miR-375 / 0.95 (0.81-1.13) / 0.579 / 0.579
miR-126 / 0.63 (0.45-0.89) / 0.009 / 0.036
a Endpoint RFT; correction for: Age at diagnosis, tumour size, nodal status, radiation treatment and mean tumour cell content
Abbreviations: CI, confidence interval.
3. References
1. Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. 2003;31(4):e15.
2. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003;4(2):249–64.
3. Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics. 2003;19(2):185–93.
4. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32–5.
5. Alao JP. The regulation of cyclin D1 degradation: roles in cancer development and the potential for therapeutic invention. Molecular Cancer. 2007;6(1):24.
6. Hothorn T, Hornik K, van de Wiel MA, Zeileis A. A Lego System for Conditional Inference. The American Statistician. 2006;60(3):257–63.
7. Holm Sture. A simple sequentially rejective multiple test procedure. Scand J Statist. 1979;6:65–70.
9