miR-221 overexpression contributes to liver tumorigenesis Pascal Pineaua,1, Stefano Voliniab,1, Katherine McJunkinc,d,1, Agn��s Marchioa, Carlo Battistone, Beno��t Terrisf, Vincenzo Mazzaferroe, Scott W. Lowec,d,g, Carlo M. Croceb,2, and Anne Dejeana,2 aNuclear Organization and Oncogenesis Unit, Institut National de la Sant�� et de la Recherche M��dicale U579, Institut Pasteur, 75724 Paris Cedex 15, France bDepartment of Molecular Virology, Immunology, and Medical Genetics, Comprehensive Cancer Center, Ohio State University Medical Center, Columbus, OH 43210 cCold Spring Harbor Laboratory, Cold Spring Harbor, NY 11724 eSurgery and Liver Transplantation Unit, Istituto Nazionale per lo Studio e Cura dei Tumori, Milan, Italy fLaboratoire d���Anatomie Pathologique, H��pital Cochin, Assistance Publique des H��pitaux de Paris, Paris 75679 Paris Cedex 14, France dWatson School of Biological Sciences, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY 11724 and gHoward Hughes Medical Institute, Cold Spring Harbor, NY 11724 Edited by Doug Hanahan, University of California, San Francisco, CA, and approved November 18, 2009 (received for review July 15, 2009) MicroRNA (miRNAs) are negative regulators of gene expression and can function as tumor suppressors or oncogenes. Expression patterns of miRNAs and their role in the pathogenesis of hepato- cellular carcinoma (HCC) are still poorly understood. We profiled miRNA expression in tissue samples (104 HCC, 90 adjacent cirrhotic livers, 21 normal livers) as well as in 35 HCC cell lines. A set of 12 miRNAs (including miR-21, miR-221/222, miR-34a, miR-519a, miR-93, miR-96, and let-7c) was linked to disease progression from normal liver through cirrhosis to full-blown HCC. miR-221/222, the most up- regulated miRNAs in tumor samples, are shown to target the CDK inhibitor p27 and to enhance cell growth in vitro. Conversely, these activities can be efficiently inhibited by an antagomiR specific for miR-221. In addition, we show, using a mouse model of liver cancer, that miR-221 overexpression stimulates growth of tumorigenic murine hepatic progenitor cells. Finally, we identified DNA dam- age-inducible transcript 4 (DDIT4), a modulator of mTOR pathway, as a bona fide target of miR-221. Taken together, these data reveal an important contribution for miR-221 in hepatocarcinogenesis and suggest a role for DDIT4 dysregulation in this process. Thus, the use of synthetic inhibitors of miR-221 may prove to be a promising approach to liver cancer treatment. hepatocarcinogenesis | microRNA | antagomiRs | mouse model | DDIT4 Hmalignant epatocellular carcinoma (HCC) is one of the most common tumors in the world (1). Among the well estab- lished risk factors for HCC, chronic infection with hepatitis B (HBV) or C (HCV) virus is present in 85% of primary liver cancers (2). Early studies indicated that only a few of the protein changes detected so far can be validated at the RNA level (3). miRNAs are thought to control gene expression at the post- transcriptional level by causing mRNA degradation and/or re- pressing mRNA translation (4). Evidence indicating that miRNAs play an important role in various human cancers is now accumu- lating (5), but only a few microtranscriptome HCC profiles, in- volving limited numbers of samples, have been performed. We present the miRNA expression profile of a large number of paired HCC-nontumor samples as well as liver cancer cell lines. We observed unique miRNA expression signatures that could distinguish malignant from adjacent cirrhotic tissues. We provide evidence for a major contribution of miR-221 in liver cancer cell proliferation and HCC development in vivo and identify DNA damage-inducible transcript 4 (DDIT4) as a direct target of miR-221. Results Twelve miRNAs Define a Signature for Liver Cancer. Our aim was to identify miRNAs that are significantly dysregulated during cancer progression from normal liver to full blown HCC, through the precancerous stage of cirrhosis. Thus, we profiled 21 normal livers, 104 HCC, 90 paired cirrhotic tissues (Table S1), and 35 HCC-derived cell lines. According to our strategy, we made three pairwise comparisons: (i) cirrhosis vs. normal liver, (ii) HCC vs. normal liver, and (iii) HCC tissues compared to their respective cirrhosis tissues. We hypothesized that a miRNA must be in- volved in liver tumorigenesis if it is consistently dysregulated through each step going from normal liver to cirrhosis and cir- rhosis to HCC. miRNAs satisfying these requirements were then examined in HCC cell lines to ensure that the miRNA signature was truly a characteristic of expression in transformed liver cells per se. The heat map of the miRNAs dysregulated in the last step of progression (cirrhosis to HCC) is shown in Fig. 1. This com- binatorial strategy identified the miRNAs significantly dysregu- lated in liver tumorigenesis: miR-106b, miR-21, miR-210, miR- 221, miR-222, miR-224, miR-34a, miR-425, miR-519a, miR-93, and miR-96 were increased, and let-7c was lost during liver tumor progression (Fig. 2A and Table S2). We then determined which of the 12 microRNAs (miRNAs) had the highest diagnostic value. Our purpose was to distinguish between the tumors (139 tumors and cell lines) and the non- tumor tissues (111 cirrhotic and normal livers). We used the prediction test in BRB-ArrayTools to identify the classifier sig- nature with the lowest misclassification error. As few as seven miRNAs could classify (15% misclassification error). The cross- validation ROC curve for the seven miRNA classifier is shown in Fig. S1A. The miRNAs used in the ROC classifier are under- lined in Fig. 2A. There were no significant differences in miRNA expression associated with p53 mutations, but we detected a signature asso- ciated with ��-catenin mutation (Table S2). Among the five miR- NAs up-regulated 2- to 3-fold in cases of ��-catenin gene mutation was miR-34a, known previously as a p53 target, and miR-122, the miRNA most highly expressed in hepatocytes (6, 7). miR-221/222 Target p27. To validate the miRNA microarray data, we carried out qPCR on RNA from 12 pairs of HCC and corre- sponding nontumorous livers and tested four different members of the progression signature (Fig. 2B and Fig. S1B). Statistically significant differences of expression levels were observed for all Author contributions: C.M.C. and A.D. designed research P.P., S.V., K.M., A.M., C.B., B.T., V.M., and S.W.L. performed research P.P., S.V., S.W.L., C.M.C., and A.D. analyzed data and P.P., C.M.C., and A.D. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. Freely available online through the PNAS open access option. Data deposition: Data have been deposited in the Array Express database, www.ebi.ac. uk/arrayexpress (accession no. E-TABM-866). 1 P.P., S.V., and K.M. contributed equally to this work. 2 To whom correspondence may be addressed. E-mail: anne.dejean@pasteur.fr or carlo. croce@osumc.edu. This article contains supporting information online at www.pnas.org/cgi/content/full/ 0907904107/DCSupplemental. 264���269 | PNAS | January 5, 2010 | vol. 107 | no. 1 www.pnas.org/cgi/doi/10.1073/pnas.0907904107

miRNAs investigated. In addition, we confirmed trends observed on microarray by qPCR for four different putative ��-catenin targets (Fig. 2B and Fig. S1C) (8). WethendecidedtoinvestigatetherolesofmiR-221andmiR-222 in detail. These miRNAs were chosen because of their consistently high levels of expression in tumors, suggesting that they act as Fig. 1. Heat map of miRNA expression in HCC: Tree showing the miRNAs deregulated in HCC vs. cirrhotic tissues (paired t test, P 0.01). Each column in- dicates the normalized log 2 ratio of the HCC/cirrhosis pairs. Clustering was performed with average linkage and uncentered correlation. Fig. 2. miRNA signatures in HCC. (A) Intersection of SAM analysis for three profiles. The intersection is shown among the deregulated miRNAs in the three classes of tissues: HCC, cirrhosis, and normal liver. Only the miRNA deregulated also in the HCC cell lines were considered. Red, up-regulated miRNAs green, down-regulated miRNAs.(B)ValidationofmicroarraybyqPCRofdifferentiallyexpressedmiRNAsinnontumoral(NT)vs.tumoral(T)livertissuesand ��-cateninmutated(��-catm)vs.wild- type (��-catwt) specimens. The relative expression fold of each miRNA is normalized to RNU6B and miR-140. P values were obtained by using a two-sided Student t test. Pineau et al. PNAS | January 5, 2010 | vol. 107 | no. 1 | 265 MEDICAL SCIENCES