microRNAs exhibit high frequency genomic alterations in human cancer

microRNAs exhibit high frequency genomicalterations in human cancerLin Zhang*†‡, Jia Huang‡§, Nuo Yang‡¶, Joel Greshock‡§, Molly S. Megraw�, Antonis Giannakakis*,**, Shun Liang*,Tara L. Naylor§, Andrea Barchetti*, Michelle R. Ward§, George Yao*, Angelica Medina§, Ann O’Brien-Jenkins*,Dionyssios Katsaros††, Artemis Hatzigeorgiou�, Phyllis A. Gimotty‡‡, Barbara L. Weber§, and George Coukos*†§,§§

*Center for Research on Reproduction and Women’s Health, Departments of †Obstetrics and Gynecology and ‡‡Biostatistics and Epidemiology, §AbramsonFamily Cancer Research Institute, ¶Cell and Molecular Biology Graduate Program and Department of Genetics, �Department of Genetics and Penn Centerfor Bioinformatics, University of Pennsylvania, Philadelphia, PA 19104; ††Department of Obstetrics and Gynecology, University of Turin, 10126 Turin, Italy;and **Laboratory of Gene Expression, Modern Diagnostic and Therapeutic Methods, Democritus University of Thrace, 69100 Alexandroupolis, Greece

Edited by Carlo M. Croce, Ohio State University, Columbus, OH, and approved April 17, 2006 (received for review October 11, 2005)

MicroRNAs (miRNAs) are endogenous noncoding RNAs, whichnegatively regulate gene expression. To determine genomewidemiRNA DNA copy number abnormalities in cancer, 283 knownhuman miRNA genes were analyzed by high-resolution array-based comparative genomic hybridization in 227 human ovariancancer, breast cancer, and melanoma specimens. A high proportionof genomic loci containing miRNA genes exhibited DNA copynumber alterations in ovarian cancer (37.1%), breast cancer(72.8%), and melanoma (85.9%), where copy number alterationsobserved in >15% tumors were considered significant for eachmiRNA gene. We identified 41 miRNA genes with gene copynumber changes that were shared among the three cancer types(26 with gains and 15 with losses) as well as miRNA genes withcopy number changes that were unique to each tumor type.Importantly, we show that miRNA copy changes correlate withmiRNA expression. Finally, we identified high frequency copynumber abnormalities of Dicer1, Argonaute2, and other miRNA-associated genes in breast and ovarian cancer as well as melanoma.These findings support the notion that copy number alterations ofmiRNAs and their regulatory genes are highly prevalent in cancerand may account partly for the frequent miRNA gene deregulationreported in several tumor types.

genome � noncoding RNA � comparative genomic hybridization

M icroRNAs (miRNAs) are endogenous �22-nt noncodingsmall RNAs, which regulate gene expression in a sequence-

specific manner (1–8). With �300 already identified, the humangenome may contain up to 1,000 miRNAs (8). Vertebrate miRNAtargets are thought to be plentiful in number (9–13). Up toone-third of human mRNAs are predicted to be miRNA targets(12). Each miRNA can target �200 transcripts directly or indirectly(14, 15), whereas more than one miRNA can converge on a singleprotein-coding gene target (9–13). Therefore, the potential regu-latory circuitry afforded by miRNA is enormous. Increasing evi-dence indicates that miRNAs, in fact, may be key regulators ofvarious fundamental biological processes (1–8).

The expression of miRNAs is highly specific for tissues anddevelopmental stages (5–7) and has allowed recently for molecularclassification of tumors (16, 17). Little is known regarding howmiRNA expression is regulated. Primary miRNA transcripts aregenerated by polymerase II (18). These transcripts are capped,polyadenylated, and usually several thousand bases in length (19).A portion of miRNAs are located within introns of pre-mRNAs andare likely transcribed together with the cognate protein-codinggenes (2, 10). Some miRNAs are clustered and transcribed asmulticistronic primary transcripts, but the majority of humanmiRNAs are not clustered and are transcribed independently (5–7).

The biogenesis and function of miRNAs require a common setof proteins. Drosha, an RNase III endonuclease, is responsible forprocessing primary miRNAs in the nucleus and releasing �70-ntprecursor miRNAs (20). Drosha associates with the dsRNA-

binding protein DGCR8 in human (21) or Pasha in flies (22) toform the microprocessor complex. Precursor miRNAs are trans-ported to the cytoplasm by exportin-5 (23, 24) and cleaved by theRNase III endonuclease Dicer, releasing �22-nt mature dsmiRNA(25). One strand of the miRNA duplex is subsequently incorporatedinto the effector complex RNA-induced silencing complex (RISC)that mediates target gene expression. Argonaute2, a key componentof RISC, may function as an endonuclease that cleaves targetmRNAs (26, 27).

Increasing evidence shows that expression of miRNA genes isderegulated in human cancer (28–31). Specific over- or underex-pression has been shown to correlate with particular tumor types(16, 17, 32–34). miRNA overexpression could result in down-regulation of tumor suppressor genes, whereas their underexpres-sion could lead to oncogene up-regulation (28–31). For example,let-7, down-regulated in lung cancer (35–37), suppresses Ras (36);mir-15 and mir-16, deleted or down-regulated in leukemia (38),suppress BCL2 (39); mir-17-5p and mir-20a control the balance ofcell death and proliferation driven by the proto-oncogene c-Myc(40). Clear evidence indicates that miRNA polycistron mir-17-92serves as an oncogene in lymphoma (33) and lung cancer (41);mir-372 and mir-373 are novel oncogenes in testicular germ celltumors by numbing p53 pathway (42). Most importantly, miRNAexpression signatures can predict outcome (35, 37, 43). These datastrongly suggest that miRNAs play an important role in humancancer.

The mechanisms underlying miRNA gene deregulation in cancerare not well understood. Although genomic alterations are criticalin oncogenesis (44, 45), studies so far have focused mostly onprotein-coding genes. Because more than one-half of the miRNAshave been aligned to genomic fragile sites or regions associated withcancers (30), genome copy abnormalities could involve miRNAgenes. Over the past several years, array comparative genomichybridization (aCGH) has proven its value for analyzing DNA copynumber variations (46). In the present study, we investigatedgenomewide DNA copy number abnormalities of genomic regionscontaining 283 known human miRNA genes in 227 human cancersamples by using a recently described high-resolution aCGH (47).This approach was generated by using BAC clones with: (i) unam-biguous mapping data across genome builds based primarily onrange-defining sequence anchors; (ii) the capacity for reproducible,sensitive copy number determination; and (iii) even spacing acrossthe genome, so that the collection has no gap �2 Mb and a mean

Conflict of interest statement: No conflicts declared.

This paper was submitted directly (Track II) to the PNAS office.

Freely available online through the PNAS open access option.

Abbreviations: aCGH, array comparative genomic hybridization; miRNA, microRNA.

‡L.Z., J.H., N.Y., and J.G. contributed equally to this work.

§§To whom correspondence should be addressed. E-mail: [email protected].

© 2006 by The National Academy of Sciences of the USA

9136–9141 � PNAS � June 13, 2006 � vol. 103 � no. 24 www.pnas.org�cgi�doi�10.1073�pnas.0508889103

spacing of �1 Mb (47). An aCGH platform with this resolution isa powerful tool to screen cancers for genetic changes and providesa direct link to the human genome sequence for immediateidentification of genes in regions exhibiting copy number changes.

Here we provide experimental genomewide documentation ofDNA copy alterations involving miRNA genes in epithelial cancers.We demonstrate a high frequency of copy number abnormalities inregions containing miRNA and their associated genes in breastcancer, ovarian cancer, and melanoma. Importantly, we show thatmiRNA gene copy changes are concordant with miRNA genetranscriptional expression. Given that miRNA genes may targetgenes critical for oncogenesis, the present data support the notionthat genomic alterations of miRNA and associated genes are partlyresponsible for miRNA deregulation in cancer and may constitutea critical step in cancer development.

ResultsHigh-resolution aCGH (47) was used to determine the DNA copynumber abnormalities of genomic regions containing knownmiRNA genes in 227 cancer specimens, including 109 ovariancancer specimens (93 primary tumors and 16 cell lines), 73 breastcancer specimens (55 primary tumors and 18 cell lines), and 45primary cultured melanoma cell lines. Tumor DNA and referenceDNA were labeled with Cy3 and Cy5, respectively (Fig. 5A, whichis published as supporting information on the PNAS web site).Reverse dye experiments were done to validate results for allsamples. The fluorescence intensity ratio of tumor to reference

DNA �0.8 was considered copy number loss, whereas �1.2 was again. A circular binary segmentation algorithm (48) was applied toraw log2 ratio data. This algorithm recursively identifies breakpointsand splits chromosomes into subsegments based on a maximum tstatistic. A permutation-generated reference distribution is used todefine the statistical significance of the estimated splits and decidewhether to split at each stage (48). This process removed clone-based dependence and allowed accurate estimations of copy num-ber for the entire genome. Fig. 5B depicts the bioinformaticapproach, with transitions between diploid DNA content andcircular binary segmentation calls of gain or loss regions. Thegenomic loci of 283 known human miRNA genes located inautosomes were identified in the miRNA registry (49) (Fig. 5 C andD). Copy number alterations observed in �15% tumors wereconsidered significant for each miRNA gene. In summary, 37.1%(105 of 283) of miRNA genes were located in regions that exhibitedDNA copy number abnormalities in ovarian cancer (Fig. 1; see alsoTables 1–3, which are published as supporting information on thePNAS web site), 72.8% (206 of 283) in breast cancer (Fig. 6 andTables 4 and 5, which are published as supporting information onthe PNAS web site; see also Table 1), and 85.9% (243 of 283) inmelanomas (Fig. 7 and Tables 6 and 7, which are published assupporting information on the PNAS web site; see also Table 1).

Overall, the genomic alterations involving miRNA genes weredistinct among tumor types as evidenced by a heat map conditiontree developed by using cluster analysis (Fig. 2). Despite differentprofiles, some novel miRNA genomic alterations were shared by

Fig. 1. High frequency miRNA gene copy number alterations in ovarian cancer. aCGH frequency plots of ovarian cancer specimens are shown. Green representsgain, and red represents loss. Stars indicate miRNA genes.

Zhang et al. PNAS � June 13, 2006 � vol. 103 � no. 24 � 9137

GEN

ETIC

S

the three cancer types (Fig. 3). Ovarian and breast cancer shared 41miRNA genes located in regions with copy number gains and 19with losses. Furthermore, 26 miRNA genes located in regions withcopy number gains and 15 with losses were shared by all three typesof cancer. Bioinformatic target prediction for the 41 miRNAs withgenomic alterations shared by the three cancer types suggests animportant role in oncogenesis (Tables 8 and 9, which are publishedas supporting information on the PNAS web site). For example,mir-9-1, whose mature miRNA was reported recently to be over-expressed in breast cancer (50), is located in loci amplified in allthree tumor types. Its target, hairy and enhancer of split 1 (HES1),predicted by MIRANDA, TARGETSCANS, and PICTAR, is a potentialtumor suppressor gene that inhibits breast cancer cell proliferation(51). Furthermore, mir-320 is located in regions with DNA copynumber loss in all of the three cancer types. A notable mir-320 targetpredicted by two independent programs is methyl CpG bindingprotein 2 (MECP2), which is overexpressed in breast cancer andserves as an oncogene promoting cell proliferation (52).

This study revealed novel genomic alterations involving miRNAgenes in epithelial cancers, which are distinct from previous studiesin hematologic malignancies. For example, the mir-17-92 polycis-tron is frequently amplified in B cell lymphoma (33). In contrast, wefound that the region containing this polycistron rather was deletedin 16.5% of ovarian cancers, 21.9% of breast cancers, and 20.0% ofmelanomas. This result suggests that specific miRNA gene alter-ations underlie specific malignancies. On the other hand, selectmiRNA alterations can be shared among epithelial and hemato-logic malignancies. For example, mir-15a and mir-16-1, locatedwithin one cluster at 13q14, are deleted or down-regulated in �50%chronic B cell lymphocytic leukemias (38). Our data showed a copynumber loss of the regions containing mir-15a and mir-16-1 in23.9% of ovarian and 24.7% of breast cancers. These miRNAsrecently have been shown to negatively regulate BCL2 protein at aposttranscriptional level, and their overexpression induces apopto-sis in leukemic cells (39).

Among 283 miRNA genes analyzed, at least 47 miRNAs arelocated within introns of protein-coding genes, which are likelytranscribed together with the cognate protein-coding genes (2, 10).Of these genes, 21 were in regions that exhibited significantlyaltered copy number in ovarian cancers, 23 in breast cancers, and38 in melanomas (Table 10, which is published as supportinginformation on the PNAS web site). This result suggests that copynumber changes may occur simultaneously in miRNA and thehosting protein-coding genes in these cancers (5–7, 10). For exam-ple, mir-218-1 is located within the tumor suppressor gene SLIT2

(human homologue of Drosophila Slit2), which is frequently inac-tivated in breast, lung, and colorectal cancer because of allelic loss(53). We found copy number losses of the region containingmir-218-1 and SLIT2 in 15.5% of ovarian cancers, 35.6% of breastcancers, and 33.3% of melanoma lines. Thus, miRNA genes maygain or lose copy numbers simultaneously with respective onco-genes or tumor suppressor genes in cancer.

We also examined whether miRNA genes involved by genomicalterations are, in fact, expressed in cancer and whether theirtranscript expression correlates with DNA copy number. First, weexamined expression of mature miRNA transcripts in ovariancancer samples for which aCGH data were available. Because hostcells infiltrating ovarian cancer could contribute to miRNA geneexpression changes without contributing to genomic alterations, wefirst chose to limit our analysis to 16 cell lines. We used TaqManmiRNA arrays, which comprised 140 of the miRNA genes previ-ously analyzed by aCGH. The majority (n � 121) of these genescould be detected at the miRNA level, and only 19 miRNA geneswere undetectable in all ovarian cancer cell lines. Similar data wereobtained with five ovarian cancer cell lines analyzed with miRNAmicroarrays, comprising 167 of the miRNA genes analyzed byaCGH.

To investigate whether abnormalities in DNA copy numberresult in aberrant miRNA expression, we examined the level ofconcordance between miRNA transcript levels and miRNA genecopy number. From the 140 genes analyzed simultaneously byTaqMan miRNA arrays and aCGH, we selected 78 genes for thisconcordance analysis, eliminating genes that were either: (i) neverfound to be involved in genomic alterations (thus their transcrip-tional expression could not be attributed to genomic changes; n �21); (ii) never detected as mature miRNA in any ovarian cancer line(thus genomic alterations could not be interpreted at a transcrip-tional level; n � 19); or (iii) multiple copies existed in differentgenomic loci (thus transcriptional changes could not be linked togenomic alterations of one specific locus; n � 22). For each miRNA,we compared the mature miRNA cell expression level in lines withnormal or abnormal DNA copy number. We found that 73.1% (57of 78) miRNA genes showed concordance between mature miRNAlevels and DNA copy number; i.e., mean mature miRNA levelswere higher in lines with DNA copy number gain relative to lineswithout copy number gain for those miRNAs. Similarly, for miRNAthat were mostly deleted, mean mature miRNA levels were lowerin lines with copy number loss relative to lines without copy numberloss (Fig. 4B).

Next, we examined whether miRNA genomic alterations wereassociated with transcriptional expression in primary tumors. Asproof of principle, we used real-time quantitative PCR to analyzelet-7a3, let-7f2, mir-9-1 and mir-213 precursor miRNA expression in57 ovarian cancers for which aCGH data were available. We foundthat expression of three of four precursor miRNAs (let-7a3, mir-9-1and mir-213) were concordant to corresponding genomic alter-ations (Fig. 8, which is published as supporting information on thePNAS web site). Finally, we compared our breast cancer aCGHdata with an unrelated tumor set that has been analyzed previouslyfor miRNA expression by miRNA microarrays (50). As recentlyreported by Iorio et al. (50), 28 miRNAs were differentiallyexpressed between breast cancer and normal breast tissue (Fig. 4D).Among these 28 miRNA genes, 11 exhibited copy number gains inour study; 9 of these 11 (81.8%) miRNAs had been demonstratedby Iorio et al. (50) to be significantly up-regulated for transcriptabundance. Similarly, five of their significant miRNAs exhibitedcopy number losses in our study; three of these five (60.0%)miRNAs had been demonstrated to be significantly down-regulatedfor transcript abundance by Iorio et al. (50). Collectively, these datasuggest that DNA copy number alterations may be a critical factoraffecting expression of miRNAs in cancer.

For some miRNAs, additional factors may affect miRNA ex-pression in the absence of genomic alterations. To test how fre-

Fig. 2. Genetic aberrations of miRNAs in human cancer. Heat map conditiontree developed by using GENE CLUSTER 2.0 shows aCGH data of all genomic locicontaining miRNAs in ovarian cancer, breast cancer, and melanoma speci-mens. Green and red indicate gain and loss in DNA copy number, respectively.

9138 � www.pnas.org�cgi�doi�10.1073�pnas.0508889103 Zhang et al.

quently protein-coding genes involved in miRNA biogenesis andfunction are involved by genomic alterations in cancer, we analyzedDNA copy number alterations in loci harboring miRNA-associatedgenes. We identified significantly frequent genomic alterationsinvolving the regions containing these genes (Fig. 9, which ispublished as supporting information on the PNAS web site). Forexample, Dicer1 and Argonaute 2 exhibited gains in DNA copynumber by 24.8% and 51.5%, respectively, in ovarian tumors. Thesedata suggest that DNA copy number alterations of miRNA-associated genes also might serve as an alternative mechanism toaffect miRNA expression in human cancers.

DiscussionIn summary, our data suggest that high-frequency copy numberabnormalities occur in miRNA-containing regions throughoutthe genome in a range of human epithelial cancers. We identifiedshared abnormalities in miRNA-containing genomic loci among

ovarian cancer, breast cancer, and melanoma. We also identifieddistinct differences in genomic alterations in miRNA-containinggenomic loci between epithelial cancers and previously describedhematologic malignancies. Furthermore, we found DNA copyalterations of loci containing miRNA genes that were unique tospecific epithelial cancers. Importantly, for many miRNAs, DNAcopy changes correlated with miRNA transcript expression.These findings support the notion that copy number alterationsof miRNAs may account partly for the frequent miRNA genederegulation reported in several tumor types. Clearly, additionalfactors are involved in miRNA deregulation in cancer, as wecould not identify a concordance between copy number andtranscript levels for many miRNAs. Our data show that protein-coding genes involved in miRNA biogenesis, including Dicer1and Argonaute2, also may participate in complex interactionsthat regulate miRNA expression in tumors, together with addi-tional mechanisms that may regulate miRNA at the epigenetic,

Fig. 3. Venn diagrams of miRNA genes with copy number gain and loss shared by two or three types of epithelial cancer.

Zhang et al. PNAS � June 13, 2006 � vol. 103 � no. 24 � 9139

GEN

ETIC

S

transcriptional, posttranscriptional, and translational levels. Weperformed bioinformatic analyses by using available tools topredict targets for miRNAs that emerged as significant in thisstudy, such as the miRNAs that were shared among all tumortypes. We found predicted targets that are known protein-codinggenes implicated in oncogenesis. This notion is in agreementwith previous reports investigating miRNAs in cancer, wheresignificant miRNAs were predicted to target protein-codinggenes involved in malignant transformation. At this point, themechanisms underlying the high frequency alteration of miRNAgenes observed in cancer genome remain unclear. In light of aprevious report that miRNAs are frequently located in fragilesites and genomic regions involved in cancers (30), one potentialexplanation is that genomic aberrations preferentially involveregions containing miRNA genes at a high density. Alternatively,clones with miRNA amplifications or deletions are selectedbecause of the biological advantage that is afforded by thesemiRNA expression changes.

In summary, the present work shows that (i) genomic alterationsinvolving miRNA are highly frequent in epithelial cancers; (ii) theyappear to be shared, in part, among epithelial cancers and, in part,

among tumor-specific cancers; and (iii) they result in changes inmature miRNA expression. Because some of our predicted targetsfor miRNA genes involved in DNA copy number changes areprotein-coding genes with known involvement in oncogenesis, ourwork suggests that genomic alteration of miRNA genes may con-stitute a critical step in cancer development. Genetic alterations ofmiRNAs thus may promote and�or enhance alteration of protein-encoding gene expression in cancer, accelerating malignant trans-formation and�or tumor growth (28–31). Rescued expression ofdown-regulated or functionally deficient miRNAs and�or inhibitionof overexpressed miRNAs may contribute to rebalanced expressionof large gene clusters implicated in oncogenesis and tumor pro-gression. Therefore, our results may contribute to a better under-standing of the pathogenesis and the identification of biomarkersand targets for human cancer. Targeting of miRNAs may providean important therapeutic stratagem for human cancer.

Materials and MethodsPatients and Specimens. We evaluated 93 stage-III and -IV epithe-lial ovarian cancer specimens from untreated patients undergoingdebulking surgery. Sixty-two specimens were provided by D. Kat-

Fig. 4. Correlation analysis between DNA copy number alteration and miRNA expression. (A–C) Expression of mature miRNA transcripts (by TaqMan miRNAassay) and DNA copy number of loci containing the specific miRNA (by aCGH) in 16 ovarian cancer cell lines. (A) Higher magnification of mir-15a and mir-15bmaps. Each spot represents a cell line. Expression levels of mature miRNA transcripts are presented in the upper lane as a heat map in 16 cell lines. DNA statusis presented in the lower lane (red, DNA copy number loss; green, DNA copy number gain). Cell lines are ranked based on DNA status (leftmost, amplified; middle,normal; rightmost, deleted gene copy). (B) miRNA genes (n � 57) showing concordance between DNA copy number alterations and miRNA transcript expression.(C) miRNA genes (n � 21) showing no concordance. (D) Comparison of miRNA aCGH data from the present tumor set with expression data of 28 miRNAsoverexpressed in a different breast cancer set, as reported by Iorio et al. (50). miRNA genes with normal DNA copy number are marked in gray; gains are markedin green, and losses in red. miRNA transcript overexpression (relative to normal breast) is marked in pink; underexpression (relative to normal breast) is in blue (50).

9140 � www.pnas.org�cgi�doi�10.1073�pnas.0508889103 Zhang et al.

saros (University of Turin). Additional ovarian cancer (n � 31) andbreast ductal carcinoma (n � 55) specimens were provided byB.L.W. All ovarian and breast tumors were from primary sites.Specimens were immediately snap-frozen and stored at �80°C.Primary melanoma cell lines (n � 45) were obtained from M.Herlyn (Wistar Institute, Philadelphia). Ovarian (n � 16) andbreast (n � 18) cancer cell lines were provided by G.C. and B.L.W.

BAC Array Platforms. BAC clones included in the ‘‘1 Mb–array’’platform were described in ref. 47. Detailed information is providedin Supporting Methods, which is published as supporting informationon the PNAS web site.

aCGH and Circular Binary Segmentation. Genomic DNA was isolatedfrom frozen tumors or cultured cells by overnight digestion, phenol-chloroform extraction, and ethanol precipitation. One microgramof tumor and reference DNA were labeled with Cy3 or Cy5,respectively (Amersham Pharmacia, Piscataway, NJ), by using theBioPrime random-primed labeling kit (Invitrogen). In parallelexperiments, tumor DNA and reference DNA were labeled withthe opposite dye to account for the difference in dye incorporationand provide additional data for analysis. Labeled tumor andreference DNA were combined and precipitated with human Cot-1DNA to reduce nonspecific binding. DNA was resuspended andhybridized to the array for 72 h at 37°C on a rotating platform.Images were scanned with an Axon 4500 microarray scanner (AxonInstruments, Union City, CA) and analyzed with GENEPIX (AxonInstruments). Tumor�reference DNA fluorescent intensity ratios�0.8 or �1.2 were considered as alteration. For each sample, copynumber estimates were made by the circular binary segmentationmethod with the DNAcopy package in R programming language(48). Additional analyses and visualization of aCGH data weredone by using the CGHANALYZER suite described in ref. 54.

miRNA Database. The genomic loci of human miRNA genes wereidentified in the miRNA registry (The microRNA Registry, release7.1, October 2005) (49).

Low Molecular Weight RNA Isolation and miRNA Microarray. Detailedinformation is provided in Supporting Methods.

Total RNA Isolation and Quantitative Real-Time RT-PCR. Detailedinformation is provided in Supporting Methods.

TaqMan miRNA Assay. Total RNA was isolated from 1 � 106

cultured cells with TRIzol reagent. Expression of 155 maturemiRNAs in 16 human ovarian cancer cell lines were analyzed byTaqMan miRNA Assay (Applied Biosystems) under conditionsdefined by the supplier. Detailed information is provided in Sup-porting Methods.

Bioinformatic Analysis. mRNA targets were predicted for 41miRNAs of interest by using four well known miRNA targetprediction programs: DIANA-MICROT, TARGETSCANS, MIRANDA,and PICTAR. Detailed information is provided in SupportingMethods.

Protein Isolation and Western Blot. Detailed information is pro-vided in Supporting Methods.

We thank Dr. M. Herlyn for melanoma cell lines. This work wassupported by the Ovarian Cancer Research Fund (OCRF), the Abram-son Family Cancer Research Institute, the Pennsylvania Department ofHealth, the Breast Cancer Research Foundation, and Specialized Pro-gram of Research Excellence (SPORE) on Skin Cancer Grant P50-CA093372. L.Z. was supported by SPORE P50-CA083638 and theOCRF. D.K. was supported by Associazione Italiana per la Ricerca sulCancro. A.H. and M.S.M. were supported by National Science Foun-dation Grant DBI-0238295.

1. Lee, R. C., Feinbaum, R. L. & Ambros, V. (1993) Cell 75, 843–854.2. Lagos-Quintana, M., Rauhut, R., Lendeckel, W. & Tuschl, T. (2001) Science 294, 853–858.3. Lau, N. C., Lim, L. P., Weinstein, E. G. & Bartel, D. P. (2001) Science 294, 858–862.4. Lee, R. C. & Ambros, V. (2001) Science 294, 862–864.5. Bartel, D. P. (2004) Cell 116, 281–297.6. Ambros, V. (2004) Nature 431, 350–355.7. He, L. & Hannon, G. J. (2004) Nat. Rev. Genet. 5, 522–531.8. Zamore, P. D. & Haley, B. (2005) Science 309, 1519–1524.9. Lewis, B. P., Shih, I. H., Jones-Rhoades, M. W., Bartel, D. P. & Burge, C. B. (2003) Cell 115,

787–798.10. John, B., Enright, A. J., Aravin, A., Tuschl, T., Sander, C. & Marks, D. S. (2004) PLoS Biol.

2, e363.11. Kiriakidou, M., Nelson, P. T., Kouranov, A., Fitziev, P., Bouyioukos, C., Mourelatos, Z. &

Hatzigeorgiou, A. (2004) Genes Dev. 18, 1165–1178.12. Lewis, B. P., Burge, C. B. & Bartel, D. P. (2005) Cell 120, 15–20.13. Krek, A., Grun, D., Poy, M. N., Wolf, R., Rosenberg, L., Epstein, E. J., MacMenamin, P.,

da Piedade, I., Gunsalus, K. C., Stoffel, M. & Rajewsky, N. (2005) Nat. Genet. 37, 495–500.14. Bartel, D. P. & Chen, C. Z. (2004) Nat. Rev. Genet. 5, 396–400.15. Lim, L. P., Lau, N. C., Garrett-Engele, P., Grimson, A., Schelter, J. M., Castle, J., Bartel,

D. P., Linsley, P. S. & Johnson, J. M. (2005) Nature 433, 769–773.16. Lu, J., Getz, G., Miska, E. A., Alvarez-Saavedra, E., Lamb, J., Peck, D., Sweet-Cordero, A.,

Ebert, B. L., Mak, R. H., Ferrando, A. A., et al. (2005) Nature 435, 834–838.17. Volinia, S., Calin, G. A., Liu, C.-G., Ambs, S., Cimmino, A., Petrocca, F., Visone, R., Iorio,

M., Roldo, C., Ferracin, M., et al. (2006) Proc. Natl. Acad. Sci. USA 103, 2257–2261.18. Lee, Y., Kim, M., Han, J., Yeom, K. H., Lee, S., Baek, S. H. & Kim, V. N. (2004) EMBO

J. 23, 4051–4060.19. Kim, V. N. & Nam, J. W. (2006) Trends Genet. 22, 165–173.20. Lee, Y., Ahn, C., Han, J., Choi, H., Kim, J., Yim, J., Lee, J., Provost, P., Radmark, O., Kim,

S. & Kim, V. N. (2003) Nature 425, 415–419.21. Gregory, R. I., Yan, K. P., Amuthan, G., Chendrimada, T., Doratotaj, B., Cooch, N. &

Shiekhattar, R. (2004) Nature 432, 235–240.22. Denli, A. M., Tops, B. B., Plasterk, R. H., Ketting, R. F. & Hannon, G. J. (2004) Nature 432,

231–235.23. Yi, R., Qin, Y., Macara, I. G. & Cullen, B. R. (2003) Genes Dev. 17, 3011–3016.24. Lund, E., Guttinger, S., Calado, A., Dahlberg, J. E. & Kutay, U. (2004) Science 303, 95–98.25. Bernstein, E., Caudy, A. A., Hammond, S. M. & Hannon, G. J. (2001) Nature 409, 363–366.26. Tabara, H., Sarkissian, M., Kelly, W. G., Fleenor, J., Grishok, A., Timmons, L., Fire, A. &

Mello, C. C. (1999) Cell 99, 123–132.27. Hammond, S. M., Boettcher, S., Caudy, A. A., Kobayashi, R. & Hannon, G. J. (2001) Science

293, 1146–1150.28. Croce, C. M. & Calin, G. A. (2005) Cell 122, 6–7.29. Gregory, R. I. & Shiekhattar, R. (2005) Cancer Res. 65, 3509–3512.30. Calin, G. A., Sevignani, C., Dumitru, C. D., Hyslop, T., Noch, E., Yendamuri, S., Shimizu,

M., Rattan, S., Bullrich, F., Negrini, M. & Croce, C. M. (2004) Proc. Natl. Acad. Sci. USA101, 2999–3004.

31. McManus, M. T. (2003) Semin. Cancer Biol. 13, 253–258.32. Calin, G. A., Liu, C.-G., Sevignani, C., Ferracin, M., Felli, N., Dumitru, C. D., Shimizu, M.,

Cimmino, A., Zupo, S., Dono, M., et al. (2004) Proc. Natl. Acad. Sci. USA 101, 11755–11760.33. He, L., Thomson, J. M., Hemann, M. T., Hernando-Monge, E., Mu, D., Goodson, S., Powers,

S., Cordon-Cardo, C., Lowe, S. W., Hannon, G. J. & Hammond, S. M. (2005) Nature 435,828–833.

34. Cummins, J. M., He, Y., Leary, R. J., Pagliarini, R., Diaz, L. A., Jr., Sjoblom, T., Barad, O.,Bentwich, Z., Szafranska, A. E., Labourier, E., et al. (2006) Proc. Natl. Acad. Sci. USA 103,3687–3692.

35. Takamizawa, J., Konishi, H., Yanagisawa, K., Tomida, S., Osada, H., Endoh, H., Harano,T., Yatabe, Y., Nagino, M., Nimura, Y., et al. (2004) Cancer Res. 64, 3753–3756.

36. Johnson, S. M., Grosshans, H., Shingara, J., Byrom, M., Jarvis, R., Cheng, A., Labourier,E., Reinert, K. L., Brown, D. & Slack, F. J. (2005) Cell 120, 635–647.

37. Yanaihara, N., Caplen, N., Bowman, E., Seike, M., Kumamoto, K., Yi, M., Stephens, R. M.,Okamoto, A., Yokota, J., Tanaka, T., et al. (2006) Cancer Cell 9, 189–198.

38. Calin, G. A., Dumitru, C. D., Shimizu, M., Bichi, R., Zupo, S., Noch, E., Aldler, H., Rattan,S., Keating, M., Rai, K., et al. (2002) Proc. Natl. Acad. Sci. USA 99, 15524–15529.

39. Cimmino, A., Calin, G. A., Fabbri, M., Iorio, M. V., Ferracin, M., Shimizu, M., Wojcik, S. E.,Aqeilan, R. I., Zupo, S., Dono, M., et al. (2005) Proc. Natl. Acad. Sci. USA 102, 13944–13949.

40. O’Donnell, K. A., Wentzel, E. A., Zeller, K. I., Dang, C. V. & Mendell, J. T. (2005) Nature435, 839–843.

41. Hayashita, Y., Osada, H., Tatematsu, Y., Yamada, H., Yanagisawa, K., Tomida, S., Yatabe,Y., Kawahara, K., Sekido, Y. & Takahashi, T. (2005) Cancer Res. 65, 9628–9632.

42. Voorhoeve, P. M., le Sage, C., Schrier, M., Gillis, A. J., Stoop, H., Nagel, R., Liu, Y. P., vanDuijse, J., Drost, J., Griekspoor, A., et al. (2006) Cell 124, 1169–1181.

43. Calin, G. A., Ferracin, M., Cimmino, A., Di Leva, G., Shimizu, M., Wojcik, S. E., Iorio, M. V.,Visone, R., Sever, N. I., Fabbri, M., et al. (2005) N. Engl. J. Med. 353, 1793–1801.

44. Albertson, D. G., Collins, C., McCormick, F. & Gray, J. W. (2003) Nat. Genet. 34, 369–376.45. Bentley, D. R. (2004) Nature 429, 440–445.46. Pinkel, D. & Albertson, D. G. (2005) Nat. Genet. 37, Suppl. 1, S11–S17.47. Greshock, J., Naylor, T. L., Margolin, A., Diskin, S., Cleaver, S. H., Futreal, P. A., deJong,

P. J., Zhao, S., Liebman, M. & Weber, B. L. (2004) Genome Res. 14, 179–187.48. Olshen, A. B., Venkatraman, E. S., Lucito, R. & Wigler, M. (2004) Biostatistics 5, 557–572.49. Griffiths-Jones, S. (2004) Nucleic Acids Res. 32, Database issue, D109–D111.50. Iorio, M. V., Ferracin, M., Liu, C. G., Veronese, A., Spizzo, R., Sabbioni, S., Magri, E.,

Pedriali, M., Fabbri, M., Campiglio, M., et al. (2005) Cancer Res. 65, 7065–7070.51. Hartman, J., Muller, P., Foster, J. S., Wimalasena, J., Gustafsson, J. A. & Strom, A. (2004)

Oncogene 23, 8826–8833.52. Muller, H. M., Fiegl, H., Goebel, G., Hubalek, M. M., Widschwendter, A., Muller-Holzner,

E., Marth, C. & Widschwendter, M. (2003) Br. J. Cancer 89, 1934–1939.53. Dallol, A., Da Silva, N. F., Viacava, P., Minna, J. D., Bieche, I., Maher, E. R. & Latif, F.

(2002) Cancer Res. 62, 5874–5880.54. Margolin, A. A., Greshock, J., Naylor, T. L., Mosse, Y., Maris, J. M., Bignell, G., Saeed, A. I.,

Quackenbush, J. & Weber, B. L. (2005) Bioinformatics 21, 3308–3311.

Zhang et al. PNAS � June 13, 2006 � vol. 103 � no. 24 � 9141

GEN

ETIC

S