Article Text

Download PDFPDF

Original article
A microRNA panel to discriminate carcinomas from high-grade intraepithelial neoplasms in colonoscopy biopsy tissue
  1. Shuyang Wang1,
  2. Lei Wang1,
  3. Nayima Bayaxi1,
  4. Jian Li2,
  5. Wim Verhaegh3,
  6. Angel Janevski4,
  7. Vinay Varadan4,
  8. Yiping Ren2,
  9. Dennis Merkle3,
  10. Xianxin Meng5,
  11. Xue Gao6,
  12. Huijun Wang6,
  13. Jiaqiang Ren7,
  14. Winston Patrick Kuo8,
  15. Nevenka Dimitrova4,
  16. Ying Wu1,2,
  17. Hongguang Zhu1,6,9
  1. 1Department of Pathology, Shanghai Medical College, Fudan University, Shanghai, China
  2. 2Department of Healthcare, Philips Research Asia – Shanghai, Shanghai, China
  3. 3Department of Molecular Diagnostics, Philips Research, Eindhoven, The Netherlands
  4. 4Philips Research North America, New York, USA
  5. 5Shanghai Biochip Company, Shanghai, China
  6. 6Institute of Biomedical Sciences, Fudan University, Shanghai, China
  7. 7Department of Transfusion Medicine, Magnuson Clinical Center, National Institutes of Health, Bethesda, Maryland, USA
  8. 8Developmental Biology, Harvard School of Dental Medicine, Boston, Massachusetts, USA
  9. 9Division of Surgical Pathology, Huashan Hospital, Fudan University, Shanghai, China
  1. Correspondence to Professor Hongguang Zhu, Department of Pathology, Shanghai Medical College, Fudan University, 138 Yi Xue Yuan Road, Shanghai 200032, China; hongguang_702{at}163.com Dr Ying Wu, Philips Research Asia – Shanghai, 888 Tian Lin Road Shanghai 200233, China; yingwuholland{at}yahoo.co.uk Dr Nevenka Dimitrova, Philips Research North America, 345 Scarborough Road Briarcliff Manor, NY10510, USA; nevenka.dimitrova{at}philips.com

Footnotes

  • SW and LW contributed equally to this work.

  • Funding This work was supported by grants from Philips Company Research (2007-062) and from Science and Technology Commission of Shanghai Municipality (06DZ22904).

  • Correction notice This article has been corrected since it was published Online First. The statement ‘SW and LW contributed equally to this work’ has been included.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval The institutional review board of Shanghai Medical College in Fudan University.

  • Provenance and peer review Not commissioned; externally peer reviewed.

View Full Text

Statistics from Altmetric.com

Significance of this study

What is already known about this subject?

  • Colonoscopy biopsy remains the gold standard for the diagnosis of colorectal cancer (CRC). In clinical practice, superficial biopsy samples are often obtained in the intestine due to its thin wall. It is very difficult for a pathologist to distinguish high-grade intraepithelial neoplasms from invasive carcinomas on such biopsy materials without submucosa.

  • The discovery of new molecular markers for early diagnosis and optimal surgical decision-making of CRC would be critical in improving a patient's prognosis.

  • Demonstration of abnormal expression patterns of microRNAs in human disease tissues highlight their potential use as diagnostic and prognostic biomarkers, especially in cancer.

What are the new findings?

  • In the microarray analysis of 133 snap-frozen surgical colorectal specimens, one classifier of 14 microRNAs was identified with a prediction accuracy of 94.1% for discriminating carcinomas from adenomas.

  • A panel of miR-375, miR-424 and miR-92a yielded an accuracy of 94% (AUC=0.968) in discriminating carcinomas from adenomas in formalin-fixed paraffin-embedded surgical tissues.

  • This panel of miR-375, miR-424 and miR-92a has been applied to differentiate carcinomas from high-grade intraepithelial neoplasms in colonoscopy biopsy tissues with an accuracy of 89% (AUC=0.918). The presence of stromal invasion in these biopsy materials was not detectable by microscopy.

Significance of this study

How might it impact on clinical practice in the foreseeable future?

  • We report a microRNA panel that accurately discriminates carcinomas from high-grade intraepithelial neoplasms in colonoscopy biopsy tissues. This microRNA panel has considerable clinical value in the early diagnosis and optimal surgical decision-making of CRC.

Introduction

Colorectal cancer (CRC) is a major cause of cancer mortality worldwide. The discovery of new molecular markers for the early diagnosis of CRC would be critical in improving a patient's prognosis. Ideally, such molecular markers would also enable the identification of carcinomas in superficial biopsy tissues where the presence of stromal invasion cells is not detectable by microscopic analysis.

Colonoscopy biopsy remains the gold standard for the diagnosis of CRC. In clinical practice, superficial biopsy samples are often obtained in the intestine due to its thin wall. This problem often persists even after repeated biopsies. It is very difficult for a pathologist to determine whether the lesion is invasive on such biopsy material without submucosa. The current WHO classification of tumours1 states that the morphology of adenoma with high-grade intraepithelial neoplasm, including severe degree of dysplasia and carcinoma in situ, is similar to that of invasive carcinoma, with carcinoma differentiated only by the presence of stromal invasion. When superficial biopsy samples without submucosa limit the ability of microscopic analysis to determine stromal invasion, a pathologist cannot accurately distinguish high-grade intraepithelial neoplasms from invasive carcinomas. It has been reported that 95% of high-grade intraepithelial neoplasms diagnosed by colonoscopy biopsy tissues were carcinomas when re-diagnosed after surgical resection whereas only 5% were confirmed as high-grade intraepithelial neoplasms in the surgical tissue.2

Once the tumour has stromal invasion, the carcinoma cells may spread systemically. An adenoma with high-grade intraepithelial neoplasms is therefore a local and curable disease while invasive carcinoma is a systematic disease. Failure to make this distinction on biopsy material presents the surgeon with a dilemma at the time of resection. For example, if a very low rectal tumour is an adenoma with high-grade intraepithelial neoplasm, a local resection should be performed and the anus should be preserved; however, when the diagnosis is an invasive carcinoma, a radical operation should be performed and the anus should be resected.3 Removal of the anus will significantly influence the quality of life. The identification of new molecular markers that can reliably discriminate invasive carcinoma from adenoma with high-grade intraepithelial neoplasm in biopsy tissues would therefore have great clinical utility in early diagnosis and supporting optimal surgical decision-making.

Given the therapeutic and prognostic potential of microRNAs as biomarkers in CRCs,4 ,5 we evaluated microRNA profiles in the transformation of colorectal carcinogenesis to identify new biomarkers for early diagnosis and optimal clinical decision-making of CRC. We applied laser capture microdissection (LCM)6 to 133 snap-frozen surgical specimens for isolating pure epithelial cell populations and performed microRNA microarray analysis to identify candidate biomarkers for discriminating carcinomas from adenomas. We then validated the microRNA candidates on an independent cohort of 91 formalin-fixed paraffin-embedded (FFPE) surgical tissue samples and further applied a microRNA panel with miR-375, miR-424 and miR-92a to differentiate carcinomas from adenomas with high-grade intraepithelial neoplasms on 58 FFPE colonoscopy biopsy tissue samples with stromal invasion cells undetectable by microscopy. Additionally, we determined the expression profiles of 17 putative targeted genes on 62 snap-frozen colorectal tissue samples for miR-375, miR-424 and miR-92a to pave the way for further functional studies.

Methods

Clinical specimens

The study was approved by the institutional review board of Shanghai Medical College in Fudan University and written informed consent was obtained from all the patients. For selection of study samples we used routine histological classification according to the WHO classification of tumours.1 All the lesions were diagnosed by two pathologists and independently reviewed by an expert CRC pathologist. Individuals with known inherited syndromes and patients who had received preoperative radiotherapy or chemotherapy were excluded. Figure 1 depicts three different phases of the study design and table 1 summarises the clinical characteristics of the patients and tumours in the study.

Figure 1

Study design. The microRNA profiles of 282 colorectal tissue samples from 250 patients were used to generate outcomes in three different phases. The candidate microRNAs found on 133 snap-frozen surgical specimens using microarrays were validated in an independent cohort of 91 formalin-fixed paraffin-embedded (FFPE) surgical tissue samples using quantitative RT-PCR. The microRNA panel identified in the FFPE surgical tissue samples was further applied to 58 FFPE biopsy specimens from colonoscopy.

Table 1

Characteristics of patients and tumours in the study*

In the discovery phase we used 133 independent snap-frozen colorectal tissues from 101 patients who were surgically resected at Shanghai Huashan Hospital between 2007 and 2009. The samples included 28 adenomas (21 low-grade and 7 high-grade) and 105 invasive carcinomas (20 Dukes' A, 21 Dukes' B, 28 Dukes' C, 21 Dukes' D, six lymphatic invasion, nine hepatic metastases). Among the 28 adenomas, seven were from patients who also had CRC. LCM6 was performed on each sample specifically to isolate epithelial cell populations. Examples of histological images and LCM isolated cells from colorectal tissues are shown in supplementary figure 1.

In the validation phase we used an independent cohort of 91 FFPE surgical tissue samples from patients who consecutively underwent surgery in Shanghai Huashan Hospital between 2007 and 2009. The surgical specimens included 41 adenomas and 50 carcinomas. H&E-stained sections on each tissue block were prepared to check the proportion of tumour material. If a tumour had >75% neoplastic cells it was deemed suitable for analysis without further purification of tumour cells. If, however, histology showed the tumour had <75% neoplastic cells it was selected and marked for manual macrodissection.

In the application phase we used an independent cohort of 58 FFPE biopsy tissue samples from patients who were initially diagnosed as having adenomas with high-grade intraepithelial neoplasms. The presence of the stromal invasion in these biopsy materials was not detectable by microscopic analysis. All the adenomas were completely excised during the surgical resections. Of the surgical specimens, 25 were confirmed as adenomas with high-grade intraepithelial neoplasms while 33 were re-diagnosed as carcinomas. These 58 cases were selected from about 1000 patients who had both colonoscopy biopsies and surgical resections in Shanghai Huashan Hospital between 2006 and 2009. Manual macrodissection was performed on each tissue block to ensure >75% neoplastic cells. No tissue sample overlapped all of the phases.

In addition, we used 62 snap-frozen colorectal tissues from the discovery phase to determine the expression profiles of 17 putative targeted genes for miR-375, miR-424 and miR-92a. The surgical specimens included 19 adenomas and 43 carcinomas from 62 patients. Manual macrodissection was performed on each tissue block to ensure >75% neoplastic cells.

Laser capture microdissection

Tissue preparations on snap-frozen surgical specimens for LCM were performed as previously described.6 The target cell populations (∼200 000 cells) were selected and captured using ultraviolet laser cutting following the manufacturer's recommended protocol. Additionally, H&E-stained sections were prepared for each tissue to guide the area of interest for LCM.

RNA isolation

Total RNA of the frozen tissue sections was extracted using mirVana miRNA isolation kit according to the manufacturer's instructions (Applied Biosystems, Foster City, California, USA). The concentration was quantified by NanoDrop 1000 Spectrophotometer (NanoDrop Technologies, Waltham, Massachusetts, USA). The quality control of RNA was performed by a 2100 Bioanalyzer using the RNA 6000 Pico LabChip kit (Agilent Technologies, Santa Clara, California, USA). The quality was measured by using RNA integrity number (RIN). RNA samples were discarded from further analysis if the RIN score was <5.0.

Total RNA of the FFPE tissue sections was isolated using the RecoverAll Total Nucleic Acid Isolation Kit according to the manufacturer's instructions (Ambion, Austin, Texas, USA). The concentration of RNA was quantified by NanoDrop 1000 Spectrophotometer (NanoDrop Technologies). RNA samples were discarded from further analysis if OD 260/280 ratio was <1.8.

Microarray hybridisation

Human microRNA microarrays (Agilent Technologies) were used in the discovery phase. The microarray contains probes for 723 human microRNAs from the Sanger database V.10.1. Total RNA (100 ng) derived from LCM-selected cells was labelled with Cy3. Microarray slides were scanned by XDR Scan (PMT100, PMT5). The labelling and hybridisation were performed according to the protocols in the Agilent microRNA microarray system. The microarray image information was converted into spot intensity values using Feature Extraction Software Rev V.9.5.3 (Agilent Technologies). The signal after background subtraction was exported directly into the GeneSpring GX10 software (Agilent Technologies) for quantile normalisation. Log transformation with base 2 was then performed.

Quantitative RT-PCR

In the validation and application phases, quantitative reverse-transcriptase PCR (qRT-PCR) using Taqman MicroRNA Assays (Applied Biosystems) was performed on FFPE colorectal tissues according to the manufacturer's instructions. The assays were first performed on 91 FFPE colorectal surgical tissue samples for 14 candidate microRNAs (miR-125b, miR-144, miR-17, miR-193a-5p, miR-218, miR-365, miR-375, miR-424, miR-451, miR-494, miR-7, miR-801, miR-92a and miR-99a). The data acquired on the 91 FFPE surgical tissues were randomly separated into a training set (n=46) and a test set (n=45) for data analysis. The assays were further applied to 58 FFPE biopsy tissue samples from colonoscopy for three significant microRNAs (miR-375, miR-424 and miR-92a). The expression level of U47 small nuclear RNA was used as an endogenous control. All assays were carried out in triplicate.

For 17 predicted target genes of miR-375, miR-424 and miR-92a (CCND1, MAP2K1, MAP2K3, MAPK3, PIK3R1, SMAD3, WNT1, WNT3A, WNT7A, MAP2K4, APPL1, FZD10, JUN, PIK3R3, FZD8, MAP3K5 and WNT5A), qRT-PCR using SYBR Green PCR Master Mix kit (Applied Biosystems) was performed on 62 macrodissected frozen colorectal tissue samples according to the protocol provided by the manufacturer. The expression level of GAPDH was used as an endogenous control. All assays were carried out in triplicate. The primer sequences are available from the authors upon request.

Analysis of microarray data

The data acquired from the microarrays of different tissue types from 133 discovery specimens were randomly separated into a training set (n=72) and a test set (n=61) for further analysis (see supplementary data).

Differential microRNA expression analysis

The mean normalised signal from biological replicates was used for comparative expression analysis. Unpaired unequal variance t test with Benjamini–Hochberg correction7 (p<0.00001) was performed to determine differentially-expressed microRNAs between adenoma and carcinoma tissues.

Prediction analysis

In each of the two datasets on the microarrays, three classification algorithms (prediction analysis of microarray,8 genetic algorithm-SVM9 and one-loop Naïve Bayesian10) were used to select feature subsets and build classification models. Cross-validation methods were all used within these machine learning processes (see supplementary data).

Nearest centroid classification

Nearest centroid (NC) classification was performed on the selected 14 candidates in the discrimination of carcinoma and adenoma using a Euclidian distance metric.11 For the microarray data we first trained an NC classifier on the 72 training samples and then applied this classifier to the 61 test samples. For the results of qRT-PCR on FFPE surgical tissues, we first trained an NC classifier on the 46 training samples and then applied the classifier to the 45 test samples.

Analysis of diagnostic accuracy

For the 14 candidate microRNAs selected for discriminating carcinoma from adenoma, receiver operating characteristic (ROC) curve analysis12 was applied to the data from microarrays and qRT-PCR. We further determined the area under curve (AUC) and the corresponding p values from a Wilcoxon signed rank test. Stepwise logistic regression analysis13 was used to identify the best combination of multiple diagnostic biomarkers. MedCalc software was used to perform ROC curve and regression analysis.

Target prediction and pathway analysis

For miR-375, miR-424 and miR-92a, the targets of the microRNAs were predicted from the gateway miRecords (http://mirecords.biolead.org/). To increase the accuracy of the prediction, the genes that were predicted by at least four of 11 databases (Diana, microinspector, miranda, mirtarget2, mitarget, nbmirtar, pictar, pita, rna22, rnahybrid and targetscan) were selected as targets. The KEGG database (http://www.genome.jp/kegg/tool/search_pathway.html) was used to map the predicted targets of microRNAs to CRC-associated pathways.

Results

MicroRNA expression profile

We measured the expression of 723 human microRNAs in LCM-selected epithelial cells derived from 133 snap-frozen surgical tissues in the transformation of colorectal carcinogenesis. Hierarchical clustering analysis showed that, using 21 differentially expressed microRNAs, 26 of 28 adenomas and 103 of 105 carcinomas were correctly classified (supplementary figure 3 and table 1). Of the 21 significant microRNAs, seven had been reported in the literature in colorectal tumour tissues14 and 14 were newly identified in this study.

MicroRNA classifier for prediction of colorectal tumours

We used three well-known classification algorithms (PAM, GA and Bayesian) to analyse the training and test sets acquired on the microarrays. All of the algorithms exhibited highly balanced predicted accuracies (88–99%) for discriminating adenomas from carcinomas (supplementary tables 2–4). We identified 14 microRNAs that were shared by any two of the classifications used (supplementary table 5). Of the 14 significant microRNAs, six had been reported in the literature in colorectal tumour tissues14 and eight were newly identified in this study.

We further evaluated the 14 shared microRNAs in the differentiation of carcinoma and adenoma by a NC classifier and ROC curve analysis. Training a NC classifier on 72 training samples gave a balanced accuracy of 95.6% (supplementary table 6A). Applying this classifier to 61 test samples showed a balanced accuracy of 94.1% (supplementary table 6B). The classification plots on the training and test sets are shown in figure 2A and B. The diagnostic accuracy of the 14 candidates on the microarrays in the training and test sets are shown in table 2.

Figure 2

Nearest centroid (NC) classification of microRNA expression profiles in discriminating carcinomas from adenomas. (A) NC classification on microarrays using 72 frozen surgical training samples. (B) NC classification on microarrays using 61 frozen surgical testing samples. (C) NC classification on quantitative RT-PCR (qPCR) using 46 formalin-fixed paraffin-embedded (FFPE) surgical training samples. (D) NC classification on quantitative RT-PCR (qPCR) using 45 FFPE surgical test samples.

Table 2

ROC curve analysis of 14 candidate microRNAs on data from microarrays and quantitative RT-PCR

Validation of candidates in FFPE surgical tissues

For the 14 candidate microRNAs for discriminating carcinoma from adenoma, we performed qRT-PCR on 91 FFPE colorectal tissue samples from surgical resections. Figure 3A and B shows histological images of an adenoma with high-grade intraepithelial neoplasm and a carcinoma with obvious stromal invasion from a surgical resection.

Figure 3

Histological images and receiver operating characteristic (ROC) curve analysis of the logit model with miR-375, miR-424 and miR-92a on formalin-fixed paraffin-embedded (FFPE) surgical tissue samples. (A) Histological image of a high-grade adenoma without stromal invasion from surgical resection (H&E-stain, magnification 40×10). (B) Histological image of carcinoma with obvious stromal invasion from surgical resection (H&E-stain, magnification 40×10). (C) ROC plot for the logit(p) value on 46 FFPE surgical training samples for discriminating carcinomas from adenomas. (D) An interactive dot diagram of the logit(p) value on the FFPE surgical training samples. (E) An ROC plot for the logit(p) value on 45 FFPE surgical test samples for discriminating carcinomas from adenomas. (F) An interactive dot diagram of the logit(p) value on the FFPE surgical test samples. Logit(p=CRC) = 5.7428−1.5219*(miR-375) + 0.7039*(miR-424) + 1.0336*(miR-92a).

Testing a NC classifier on 46 training samples gave a balanced accuracy of 74.6% in discriminating carcinoma from adenoma (supplementary table 6C). Applying this classifier to 45 testing samples showed a balanced accuracy of 74.5% (supplementary table 6D). The classification plots on the training and testing sets are shown in figure 2C and D.

The diagnostic accuracy of the 14 candidates on qRT-PCR in the training and testing sets are shown in table 2. Applying stepwise logistic regression to the 46 training samples starting from the 14 candidates gave a logit model logit(p) = 5.7428−1.5219*(miR-375) + 0.7039*(miR-424) + 1.0336*(miR-92a) with a cut-off value of −0.7448 (supplementary table 7A). Applying this model to the training samples gave a balanced accuracy of 93.6% with an AUC of 0.967 (supplementary table 7B). Applying the same model and cut-off value to 45 testing samples gave a balanced accuracy of 93.0% with an AUC of 0.991 (supplementary table 7B). The corresponding ROC curve and scatter plot are shown in figure 3C–F.

Application of a microRNA panel in FFPE biopsy tissues from colonoscopy

We next applied the panel of three significant microRNAs (miR-375, miR-424 and miR-92a) to 58 FFPE biopsy specimens from patients who were initially diagnosed with high-grade intraepithelial neoplasms. Interestingly, after surgical resection, 25 cases were confirmed as high-grade intraepithelial neoplasms (figure 4A) while 33 cases were re-diagnosed as carcinomas (figure 4B). No obvious stromal invasion was observed in these 58 biopsy specimens.

Figure 4

Histological images and receiver operating characteristic (ROC) curve analysis of the logit model with miR-375, miR-424 and miR-92a on formalin-fixed paraffin-embedded (FFPE) biopsy tissue. (A) Histological image of a high-grade adenoma from colonoscopy biopsy (H&E-stain, magnification 40×10). The case was confirmed as high-grade adenoma in the surgical tissue. (B) Histological image of a high-grade adenoma from colonoscopy biopsy. After surgical resection the case was re-diagnosed as carcinoma (H&E stain, magnification 40×10). (C) ROC plot for the logit(p) value on 58 FFPE biopsy samples for discriminating carcinomas from high-grade adenomas. (D) An interactive dot diagram of the logit(p) value on the FFPE biopsy samples. Logit(p) = 5.7428−1.5219*(miR-375) + 0.7039*(miR-424) + 1.0336*(miR-92a).

Applying the model derived from the 46 FFPE surgical training samples and the corresponding cut-off value to the FFPE biopsy tissues gave a balanced accuracy of 89.0% (AUC=0.918) in discriminating high-grade intraepithelial neoplasms from carcinomas (figure 4C,D).

Association of a microRNA panel with colorectal carcinogenesis

It is of interest to evaluate the putative targets of miR-375, miR-424 and miR-92a further as a preliminary to functional studies. We predicted the targets for each of the three microRNAs through gateway miRecords and examined their target genes in the KEGG database. We identified 17 putative target genes that were key members of established signalling pathways in CRC, such as Wnt, MAPK, p53 and TGF-beta (supplement table 8). There were 10, 5 and 3 predicted targets for miR-424, miR-92a and miR-375, respectively. One target gene (MAP2K4) was duplicated between miR-424 and miR-92a (supplement table 9).

We further determined the expression levels of 17 putative targeted genes in 62 macrodissected frozen colorectal tissues. Of the 17 targeted genes, 11 had significantly differential expression in the comparison of adenomas with carcinomas while the other six genes did not show differential expression in this comparison (supplementary table 9). All of the 11 significant genes were downregulated in CRCs. For microRNA-424 we observed a strong negative correlation with its six putative target mRNAs (MAP2K1, MAPK3, SMAD3, WNT1, WNT7A and MAP2K4; figure 5A). Similarly, there was a significant negative correlation between the expression of miR-92a and its three putative target mRNAs (MAP2K4, JUN and PIK3R3; figure 5B). On the other hand, a strong positive correlation was detected between the expression of miR-375 and its two putative target mRNAs (MAP3K5 and WNT5A; figure 5C).

Figure 5

Expression relation of microRNA/target mRNA. (A) microRNA-424/target mRNA expression relation in colorectal carcinoma. (B) microRNA-92a/target mRNA expression relation in colorectal carcinoma. (C) microRNA-375/target mRNA expression relation in colorectal carcinoma. Green circle, downregulation; red circle, upregulation; white circle, no significant differential expression between carcinoma and adenoma.

Discussion

Both colorectal adenomas and carcinomas are epithelial neoplasms. Since the amount of non-cancer tissue in colorectal tumours is highly variable (10–80%), analysis of such complex tissues may conceal the specific signature of the epithelial cells. In contrast to 20 reported studies on microRNAs in CRCs using resected tumour tissues,14 our study is unique for the following reasons. First, we have defined the utility of microRNA biomarkers via microarrays using LCM-selected epithelial cells derived from a large number of specimens which enabled us to have a better chance of identifying potential diagnostic microRNA markers. Furthermore, we used three classification algorithms and cross-validation methods8–10 for analysis of the microarray data to derive reliable candidates. In addition, the identified microRNA panel from our study was validated in an independent cohort of macrodissected FFPE surgical tissue samples and further applied to colonoscopy biopsy material where the presence of stromal invasion cells was not detectable by microscopic analysis. Finally, we determined the expression profiles of the putative targeted genes for the identified microRNA panel to pave the way for further functional studies.

As LCM is not a widely available method for clinical diagnosis, we have shown, using macrodissection on FFPE colorectal tissues, that 8 of 14 microRNA candidates had significantly differential expression in carcinomas compared with adenomas. Our results demonstrate that the microRNA biomarkers detected by qRT-PCR using tumour tissue with >75% neoplastic cells are highly reliable. In situ hybridisation allows specific nucleic acid sequences to be detected in morphologically preserved tissue sections.15–17 Such techniques may further speed up the transition from microRNA validation to clinical implementation.

A major goal of our study was to identify a panel of microRNAs which could be used to discriminate carcinomas from high-grade intraepithelial neoplasms in colonoscopy biopsy tissues. The identified microRNA classifiers co-identified by any two of the three classification algorithms could have significant potential as diagnostic biomarkers. We therefore integrated these analysed results and identified 14 shared microRNAs for the differentiation of carcinomas and adenomas. We believe that this small panel of microRNAs could be applied clinically as molecular biomarkers in the early detection of CRC. Moreover, the clinical follow-up of patients from whom discovery samples were obtained is only 3–4 years, limiting our current ability to explore the prognostic value of these associations.

Approximately 95% of high-grade intraepithelial neoplasms diagnosed in colonoscopy biopsy tissue would be re-diagnosed as invasive carcinoma after the surgical resection.2 In Shanghai Huashan Hospital between 2006 and 2009 approximately 1000 patients were initially diagnosed with high-grade intraepithelial neoplasms in the colonoscopy biopsy tissue. After surgical resection about 3% were confirmed as high-grade intraepithelial neoplasms while approximately 97% were re-diagnosed as carcinomas. In contrast, the combination of miR-375, miR-424 and miR-92a differentiated carcinomas from high-grade intraepithelial neoplasms with an accuracy of 89% on colonoscopy biopsy samples where the presence of stromal invasion was not detectable by microscopic analysis. This study is the first step towards the clinical application of early diagnosis and guiding treatment decisions in CRC. Further large-scale studies are needed to ensure the clinical applicability of this biomarker panel.

The differentiation of submucosal invasive T1 CRCs from high-grade intraepithelial neoplasms is very important in clinical practice. T1 CRCs account for about 6% of all colon cancers.18 Unfortunately, in our study there were only three cases of T1 CRCs, which limited our ability to evaluate the performance of the microRNA panel in discriminating between these two pathological entities.

Studies in zebrafish, mouse and human samples have shown that miR-375 is important in insulin secretion and is critical to the development of the pancreas.19–22 In this study we found significant downregulation of miR-375 in human CRC tissue. Recently, Tsukamoto et al 23 also found that miR-375 is downregulated in gastric carcinoma and regulates cell survival by targeting PDK1 and 14-3-3ζ. We propose that miR-375 is a candidate tumour suppressor in gastrointestinal carcinomas.

MicroRNA-92a controls the growth of new blood vessels (angiogenesis).24 Huang et al 25 recently found that plasma miR-92a has diagnostic value for advanced neoplasia. We found significant upregulation of miR-92a in CRC tissue, suggesting that the microRNA in plasma may be released from CRC tissue.

MicroRNA-424 regulates cell autonomous angiogenic functions in endothelial cells by targeting VEGFR-2 and FGFR -1.26 Downregulation of miR-424 contributes to abnormal angiogenesis via MEK1 and cyclin E1 in senile haemangioma.27 However, our study is the first to report an association of the miR-424 expression profile with CRC.

We further predicted the target genes of miR-375, miR-424 and miR-92a and then filtered the putative targeted genes through KEGG pathways. We observed that the 17 putative targeted genes are key members of established signalling pathways in CRC including the Wnt, MAPK, p53 and TGFβ pathways. We therefore determined the expression levels of the 17 putative targeted genes in the colorectal adenomas and carcinomas. We identified 11 genes (MAP2K1, MAPK3, SMAD3, WNT1, WNT7A, MAP2K4, FZD10, JUN, PIK3R3, MAP3K5 and WNT5A) that showed significant differential expression between carcinomas and adenomas. Further functional studies in vitro and in vivo will enrich our knowledge of this interesting field of targeting treatment of CRC.

It is well accepted that the expression levels of microRNAs and their direct mRNA targets should be negatively correlated in their regulation.28 ,29 Intriguingly, we observed a significant positive correlation in the expression levels of miR-375 and its two putative targets. It has been shown that microRNAs can switch from repression to activation.30 ,31 Further functional studies are needed to confirm the role of the putative targets exhibiting positive correlation with miR-375.

In conclusion, we have identified a microRNA panel in a large number of colorectal tissue specimens that accurately discriminates between carcinomas and high-grade intraepithelial neoplasms in colonoscopy biopsy tissues. Our study shows that this microRNA panel has considerable clinical value in the early diagnosis and surgical decision-making of CRC.

Acknowledgments

We thank Tengfang Zhu, Qiong Li, Jing Gu and Fengyun Zheng for technical assistance.

References

View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

Footnotes

  • SW and LW contributed equally to this work.

  • Funding This work was supported by grants from Philips Company Research (2007-062) and from Science and Technology Commission of Shanghai Municipality (06DZ22904).

  • Correction notice This article has been corrected since it was published Online First. The statement ‘SW and LW contributed equally to this work’ has been included.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval The institutional review board of Shanghai Medical College in Fudan University.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.