MicroRNAs (mirnas) are highly conserved small

Transcription

1 MINERVA BIOTEC 2008;20:23-30 Using mirna expression data for the study of human cancer N. MASCELLANI 1, L. TAGLIAVINI 1, G. GAMBERONI 1, S. ROSSI 1, J. MARCHESINI 1, C. TACCIOLI 1, G. DI LEVA 1, M. NEGRINI 2, C. CROCE 2, S. VOLINIA 1 Aim. Mature micrornas (mirnas) belong to a very large group of small non-coding RNAs that regulate gene expression. mirnas are differentially expressed in human cancers, such as colorectal neoplasia, pediatric Burkitt lymphoma, lung cancer, large cell lymphoma, glioblastoma and B-cell lymphomas. We systematically employed a microarray platform containing probes for mature mirna as well as precursor molecules, to investigate the mirna profiles in hematological and solid cancers. Methods. We dealt with mirnas profiling in a large number of human tissues and tumors to identify specific signature of mirnas. The microarrays results were verified with the TaqMan Array Human MicroRNA Panel v1.0 (Early Access) in conjunction with Multiplex RT. To efficiently mine data from thousands of mirna experiments we developed a specialized bioinformatic system. Results. We characterized novel mirnas and determined the changes which underline tumorigenesis from normal precursors to cancer and from primary to metastatic tumors. We designed and built the AdmiR MySQL database and a JAVA interface for the management and analysis of mirnome in human tissues. The more than mirna chips stored and annotated in the system allow us to investigate the association of mirna with prognosis and survival across different types of tumors and patients groups. Conclusion. The expression data in AdmiR indicate that This work was supported by PRIN, PRRIITT Regione Emilia Romagna (GebbaLab) and Italian Association for Cancer Research (AIRC) grants. Received on May 19, Accepted for publication on May 28, Address reprint requests to: S. Volinia, Dipartimento di Morfologia ed Embriologia, Università degli Studi di Ferrara, Via Fossato di Mortara 64/b, Ferrara, Italy. s.volinia@unife.it. 1Department of Morphology and Embryology and DAMA (DAta Mining for Analysis of microarrays) University of Ferrara, Ferrara, Italy 2Department of Experimental and Diagnostic Medicine University of Ferrara, Ferrara, Italy mirnas are involved in cancer pathogenesis and support their function as either dominant or recessive cancer genes. Key words: Neoplasms - Microarray analysis - Database. MicroRNAs (mirnas) are highly conserved small non-coding 18- to 24-nt RNAs that regulate gene expression by targeting messenger RNAs (mrnas) and inducing their translational repression or degradation. 1 They were described for the first time in 1993 in C. elegans, 2 and today 928 mature human mirnas are known (mirna database, sanger.ac.uk). A mature mirna is generated from a long primary transcript (pri-mirna) of several thousand base pairs long. In the nucleus the pri-mirna is capped, polyadenylated and finally cleaved by the endonuclease Drosha in the stem-loop precursor pre-mirna of 60- to 75-nt. The pre-mirna is then actively exported into the cytoplasm by exportin 5 and is further processed by another RNase III, Dicer, to create the mature mirna duplex. One strand of the duplex is incorporated into the RNA-induced silencing complex (RISC) and the other is eliminated by Vol No 1. MINERVA BIOTECNOLOGICA 23

2 MASCELLANI USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER cleavage. Finally, the mirna leads the RISC to the mrna target. The mechanism of mrna regulation depends by the complementarity between its 3 UTR region and the mirna: perfect base matching induces RISC to cleave the mrna, whereas imperfect matching induces mainly translational silencing of the target. mirnas are involved in the regulation of gene expression during development, cell proliferation, apoptosis, glucose metabolism and stress resistance. 3-8 The study of solid tumors and hematological malignancies revealed that mirnas are differentially expressed in human cancers: in colorectal neoplasia, pediatric Burkitt, lymphoma, lung cancer, large cell lymphoma, glioblastoma and B-cell lymphomas. 9, 10 For example, the mir-17/92 cluster cooperates with c- MYC to accelerate tumor development. 11 The development of mirna microarrays allowed to highlight the involvement of mirnas in cancer pathogenesis. 12 Our research group has been involved in profiling mirnas in human tissues and tumors for the last 4 years We systematically employed a microarray platform to investigate the mirnas profile in hematological and solid cancers. We initially studied lung, colon, stomach, prostate, ovary and pancreatic tumors. The goals of our work were the characterization of novel mirnas, the clarification of the role of mirnas in cancer and the identification of their common signature. To further investigate the role of mirnas in solid cancers we profiled a large number of primary and metastatic tissues and studied mirna expression at different tumor stages and therapy. The changes which underline the tumorigenesis from normal precursors to cancer were determined. Then, we selected the mirnas significantly altered during the progression from primary to metastatic tumors and evaluated the association of mirna with prognosis and survival. We designed and developed a specialized system for the storage and management of the mirna expression data. Whereas profiling has shed light on many biological processes and disease states, a significant bottleneck remains the analysis of expression data in the context of biomaterial and reporter annotations. Full annotation of the analyzed samples enables complex and powerful inquiries. In addition to sample source hierarchies, in vivo or in vitro sample treatments and extraction protocols, many other types of annotations can be useful to record and correlate with hybridization data. For instance, the genotype, mutation profile, patient data or status of particular proteins as indicated by immunohistochemistry may be applied to aid in evaluating analysis results. Therefore, we designed the mirna expression database to allow a full integration of expression data and sample annotation. Materials and methods Mature mirna qrt-pcr: single mirna and multiplex cards The TaqMan Array Human MicroRNA Panel v1.0 (Early Access) contains 365 different human mirna assays in addition to 2 small nucleolar RNAs that function as endogenous controls for data normalization. The TaqMan microrna assays take advantage of a novel reverse transcription step utilizing unique hairpin-loop RT primers that are specific to only the mature mirna species. The stem-loop structure, specific to the 3 end of the mature mirna, extends the very short mature mirna molecule and adds a universal 3 priming site for real-time PCR. The Human MicroRNA panel is used in conjunction with multiplex RT consisting of 8 predefined RT primer pools containing up to 48 RT primers each. All 365 mirna targets were reverse transcribed in 8 separate RT reactions and each RT reaction loaded into one of the 8 filling ports on the TaqMan Array. The data for mature mirna expression were also validated by single mirna stem loop qrt-pcr according to manufacturer protocol (ABI, Foster City, CA, USA). One to 5 ng of total RNA were used in the assay. The U6 non-coding RNA was used as normalizer. In the case of novel mirnas, for which reagents were not commercially available, alternative RT-PCR methods were applied. Computational analysis Microarray images were analyzed by using GenePix Pro and post-processing was performed as follows. Average values of the replicate spots of each mirna were background-subtracted and subject to further analysis. Absent calls were thresholded to 1 prior to normalization and statistical analysis. Four different normalization procedures were assessed: global median, housekeeping, quantiles and cyclic loess. 21, MINERVA BIOTECNOLOGICA March 2008

3 USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER MASCELLANI Figure 1. Unsupervised analysis of microrna expression data in 6 solid cancers and corresponding control tissues. mirna profiling of 540 samples covering breast, colon, lung, pancreas, prostate and stomach normal tissues and tumors. The data were subject to hierarchical clustering on both the samples (horizontally oriented) and the features (vertically oriented), with probe names on the left, with average-linkage and Pearson correlation as a similarity measure. Sample names are indicated on the top and mirna names on the left. Cyclic loess and quantiles exhibited the best improvement in the reduction of variability and yielded the highest number of significant mirnas. Cyclic loess and quantiles normalizations were implemented using the Bioconductor package/function affy/normalization. 23 Normalization was performed by using the quantiles method, one of the most widely accepted methods for one channel microarrays. mirna nomenclature accorded to the microrna database at Sanger Center (currently version 10.0). 24 We identified mirna differentially expressed among 2 classes using a random-variance t-test, or for more then 2 classes an F-test. The random variance t-test is an improvement over the standard separate t-test as it permits sharing information among genes about within-class variation without assuming that all genes have the same variance. The nearest neighbors and support vector machine classifier was used on mirna expression profiles to predict the class of future samples. The models incorporated the mirnas that are differentially expressed as assessed by the random variance t-test. The prediction error of each model was estimated by using leave one-out cross-validation. The microarray dataset was deposited in Array-Express public database at EBI.The TaqMan Array Human MicroRNA Panel results were normalized according to the producer, by using RNU44, or by using quantiles. Differentially regulated mirnas were identified by using SAM or random-variance t-test. We finally conducted bioinformatic and data mining procedures by supervised and unsupervised methods on classification and discrimination of solid cancer pathways and networks. Results mirnas are de-regulated in solid cancers We described a large-scale detailed analysis of the mirna profiles in 540 samples from 6 solid tumors. The clustering of mirna expression profiles derived from 228 mirs in 363 solid cancer and 177 normal samples is shown in Figure 1. The tree was constructed by using 137 different mirnas, which are expressed in at least 90% of the samples and shows a very good separation between the different tissues. We initially compared all tumors against all normal tissues. We used significance analysis of microarrays (SAM) for the identification of differentially expressed mirnas. This procedure allows the control of false detection rate (FDR). The delta was chosen as to result in an FDR 0.01 (Table I). Twenty-three mirnas are overexpressed and 14 are down-regulated in breast, colon, lung, pancreas, prostate, and stomach carcinomas. This overall classification might not be tailored to identify tissue-specific mirna alterations that are consistently resulting in transformation, because mirna expression is heavily tissue specific. Thus, to identify the mirnas that are prognostic for cancer status without incurring in the bias due to tissue specificity, we used an alternative approach. We first identified the 6 tissue-specific cancer signatures by performing independent prediction analysis of microarray tests, and then selected the mirnas shared among the different histotypes mirna signatures. To compute the P values for this multi-tissue combined analysis, we performed a resampling test with random per- Vol No 1. MINERVA BIOTECNOLOGICA 25

4 MASCELLANI USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER TABLE I. Differentially overall de-regulated mirnas in 6 solid cancer types vs normal tissues (SAM, significance analysis of microarrays). mirna d value SD P value q value Rfold mir mir mir mir-199a mir-29b mir mir-29a mir mir-10a mir mir-200b mir-10b mir mir-128b mir-199b mir-29c mir-34a mir-125b mir-27a mir mir-125a mir let-7g mir mir mir mir-17-5p mir mir mir mir-128a mir-23b mir mir-30d mir mutations on the mirna identities. A score was defined as the sum of the frequencies of the mirnas reaching the sharing threshold (3 tumors). The P value was defined as the relative frequency of simulation scores exceeding the real score. Table II includes 20 deregulated mirnas common to at least 3 types of solid cancers (P value= ) and shows the frequency of each mirna in the 6 solid cancers signatures. The list includes 20 commonly up-regulated micrornas in 3 or more (N) types of solid cancers (P value= ). To maximize concision, we computed the mean of the mirna absolute expression level for the 6 cancer/normal pairs. These mirnas correctly cluster the TABLE II. The mirnas shared by the signatures of the six solid cancers. mir N Tumor type mir-21 6 Breast, colon, lung, pancreas, prostate, stomach mir-17-5p 5 Breast, colon, lung, pancreas, prostate mir Colon, lung, pancreas, prostate, stomach mir-29b 4 Breast, colon, pancreas, prostate mir Colon, pancreas, prostate, stomach mir-128b 3 Colon, lung, pancreas mir-199a 3 Lung, pancreas, prostate mir-24 3 Colon, pancreas, stomach mir Breast, pancreas, prostate mir Breast, colon, lung mir-181b 3 Breast, pancreas, prostate mir-20a 3 Colon, pancreas, prostate mir Colon, pancreas, stomach mir-32 3 Colon, pancreas, prostate mir-92 3 Pancreas, prostate, stomach mir Pancreas, prostate, stomach mir-30c 3 Colon, pancreas, prostate mir-25 3 Pancreas, prostate, stomach mir Colon, pancreas, stomach mir-106a 3 Colon, pancreas, prostate different tissues, irrespective of the disease status (Figure 2A). Figure 2B shows the differential expression of the common mirnas across the different tumors in relation to the normal tissues. The tree displays the different cancer types according to the fold changes in the mirna subset. These data on 6 solid cancers were obtained with a mirna chip containing less than 300 mature mirna, identified in an early mirbase version. We used microarrays with probes for almost twice as many mature mirnas. Melanoma We recently investigated mirnas associated to melanoma tumorigenesis. We found that mirnas are correctly linked to the melanoma phenotype. For example, mir-204 and mir-211 mirnas were consistently and highly expressed in melanocytes but not in melanomas. mir-346 was down-regulated exclusively in the normal melanocytes. mir-203 and mir-205 had very high expression levels in epidermis and keratinocytes, in agreement with previous report on mouse skin mirna cloning. 25 mir-146a, mir-181a/c, mir-130a/b and mir-21 were highly expressed in melanoma tissues, but not in the cell lines. On the other hand, a group of mirnas, including mir-93, mir-95 and mir-30 family members, were characteristic both of melanoma cell lines and melanoma tissues. mir-218, mir-373 and mir-145 appeared as 26 MINERVA BIOTECNOLOGICA March 2008

5 USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER MASCELLANI Primary cutaneous melanoma The unsupervised analysis and the study of melanoma cell lines revealed a meaningful and specific mirna profile (Figure 3); therefore, we moved forward to identify the mirnas specifically regulated during melanoma establishment and progression. We thus followed with the study of primary tumors. The significant mirnas were validated by the classifiers after leave-one-out cross validation and are displayed in Figure 4. The primary melanoma mirna signature obtained here was highly informative, with a prediction rate of 95%. Five mirnas including mir- 21, mir-346, mir-130a, mir-107 and mir-181b were strongly up-regulated during primary tumorigenesis. An additional over-expressed microrna was mir- 106a. The 2 most down-regulated mirnas in melanoma tissues were mir-34b and mir-373. Other downregulated mirnas during primary transformation included mir-34b, let-7c, mir-145. Figure 2. mirna expression signature in six solid cancers. A) Expression of the differentially regulated mirnas across solid cancers. Sixty-one mirnas are present in at least 90% of the tissues. The tree displays their average absolute expression values after log2 transformation. The mean was computed over all samples from the same tissue or tumor histotype. Genes and arrays were mean-centered and normalized. Average linkage clustering was performed by using Euclidean distance. B) Fold changes (cancer vs normal) of the mirnas present in at least 75% of the solid tumors with at least one tumor absolute value higher than 2. The tree displays the log2 transformation of the average fold changes (cancer over normal). The mean was computed over all samples from the same tissue or tumor histotype. switched off in most melanomas (tissues and cell lines), but on in normal cells (melanocytes and keratinocytes) and tissues (epidermis). Metastatic melanoma After having identified the key mirnas deregulated in tumorigenesis, we moved onto the identification of those related to metastatic colonization. At first, we studied the expression of mirnas in metastatic and primary melanomas, by using unsupervised clustering. Sixty-four mirnas passed the filter threshold of 1.5 SD and resulted in a clear separation of primary and metastatic melanomas (Figure 5). A large portion of the primary melanomas were clustered on the right hand side of the tree and displayed high level of mir-203 and mir-205, 2 mirnas present in keratinocytes. These 2 mirnas were not detectable in the metastatic melanomas. The metasta- Figure 3. Clustering of human melanoma and control cell by mirna expression. Unsupervised clustering analysis of 71 samples represents melanocytes and melanoma samples and cell lines. 173 mirnas were variable in expression (Filter SD gene = 1) and differentiated normal melanocytes, melanoma tissues, and short term (MT) and long term (CL) melanoma cell lines. Average linkage clustering was performed by using uncentered correlation metric. Vol No 1. MINERVA BIOTECNOLOGICA 27

6 MASCELLANI USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER tic melanomas themselves were separated in 2 branches. The metastatic tumors in the right branch harbor high level of expression of mir-10a and mir-10b. In the same melanoma group other mirnas were also prominent, like mir-93, mir-21, mir-100, mir-106, mir-146a, mir-181a/b/c, mir- 30a/b/c/d and mir- 125a/b. In the left branch of the metastatic melanomas, though, these mirnas were not over-expressed and an alternative group of mirnas was activated. This second group, besides mir-22, unexpectedly included only minor starred mirna forms, such as mir- 125b*, mir-25*, mir-29*, and mir-130b*. Figure 4. Plot of differentially expressed in mirna primary melanomas. Size is proportional to the log inv of P value and color to the log inv of the expression ratio. The mean intensity values of mirna expression in the primary melanoma tissues are displayed on the abcissa and on the ordinate those of the control (melanocyte cell cultures). The font size of the labelled mirnas is proportional to the log inv of P value. RT-PCR validation in independent melanoma cell lines We verified the melanoma tissue signature by using the TaqMan mature mirna assay. Most importantly, we validated the metastatic signature in an independent set of melanoma cell lines. The stem loop qrt-pcr assays confirmed the microarray measures (Figure 6). Figure 5. Unsupervised clustering analysis of primary melanoma and metastatic melanomas. Sixty-four mirnas passed the filter threshold of 1.5 SD and resulted in a consistent separation of primary and metastatic melanomas. The metastatic melanomas are subdivided into 2 groups. The first branch, on the right, displays high levels of mir-93, mir-222, mir-10a/b, mir-125b and other related micrornas. The second, somewhat smaller group has extensive down-regulation of mirnas, and over-expression of a small group, spanning mir-22 and other starred minor forms of mature mirnas. 28 MINERVA BIOTECNOLOGICA March 2008

7 USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER MASCELLANI Relative expression P1 M1 P2 M2 P3 M3 P4 M4 P5 M5 P6 M6 P7 M7 Melanoma cell lines Figure 6. Metastatic melanoma mirna validation by stem loop qrt-pcr in an independent set of melanoma cell lines. Paired cell lines derived from the same patients were used in the analysis: primitive (P1-P7) and metastatic (M1-M7) melanomas were unrelated to those used in the microarray experiment. The panel shows the relative expression levels (2-DeltaCt 10) for the mature form of mir-93. U6 was used as a normalizer. The experiment was replicated 3 times (two-tailed paired t-test, P value =0.047). Paired cell lines derived from the same patients were used for qrt-pcr: primitive (P1-P7) and metastatic (M1-M7) melanomas were unrelated to those used in the microarray experiment. The panel shows the relative expression levels (2-DeltaCt 10) for the mature form of mir-93. U6 was used as a normalizer (twotailed paired t-test, P=0.047). Finally, we used the multiplex card system to run genome wide experiments by RT-PCR system. Again, we used primary and metastatic cell lines derived from the same patients, and unrelated to those used for the microarrays. Three of the mirna significant in the microarray paired t-test and activated in one of the two metastatic signatures, mir-93, mir-30a and mir- 222, were also positive in this assay (data not shown). The mirna database Our bioinformatics system performs a range of different elaborations on a list of experiments, like a list of.gpr files and produces an output file, we called the Dataset. To complete this task we designed and implemented a MySQL database and a JAVA interface. To date more than mirna chips, corresponding to published or unpublished experiments, are stored in the system. The mirna database allows seamless statistical comparison of expression data obtained from different platforms and techniques, not only limited to microarrays. We implemented a normalization step by using the quantiles normalization in Bioconductor. The system is populated with biological and clinical annotations and it is currently used to mine the key mirna associated to cancer progression and metastasis. References 1. Bartel DP. MicroRNAs: genomics, biogenesis, mechanism, and function. Cell 2004;116: Lee RC, Feinbaum RL, Ambros V. The C. elegans heterochronic gene lin-4 encodes small RNAs with antisense complementarity to lin-14. Cell 1993;75: Chen CZ, Li L, Lodish HF, Bartel DP. MicroRNAs modulate hematopoietic lineage differentiation. Science 2004;303: Karp X, AmbrosV. Developmental biology, encountering micrornas in cell fate signaling. Science 2005;310: Xu P, Guo M, Hay BA. MicroRNAs and the regulation of cell death. Trends Genet 2004;20: Cheng AM, Byrom MW, Shelton J, Ford LP. Antisense inhibition of human mirnas and indications for an involvement of mirna in cell growth and apoptosis. Nucleic Acid Res 2005;33: Poy MN, Eliasson L, Krutzfeldt J, Kuwajima S, Ma X, MacDonald PE et al. A pancreatic-islet specific mirna regulates insulin secretion. Nature 2004;432: Dresios J, Aschrafi A, Owens GC, Vanderkilsh PW, Edelman GM, Mauro VP. Cold stress-induced protein Rbm3 binds 60S ribosomal subunits, alters microrna levels, and enhances global protein synthesis. Proc Natl Acad Sci USA 2005;102: McManus MT. MicroRNAs and cancer. Semin Cancer Biol 2003;13: Gregory RI, Shiekhattar R. MicroRNA biogenesis and cancer. Cancer Res 2005;65: Tam W, Hughes SH, Hayward WS, Besmer P. Avian bic, a gene isolated from a common retroviral site in avian leukosis virus-induced lymphomas that encodes a noncoding RNA, cooperates with c-myc in lymphomagenesis and erythroleukemogenesis. J Virol 2002;76: Liu CG, Calin GA, Meloon B, Gamliel N, Sevignani C, Ferracin M et al. An oligonucleotide microchip for genome-wide mirna profiling in human and mouse tissues. Proc Natl Acad Sci USA 2004;101: Calin GA, Liu CG, Sevignani C, Felli N, Dumitru CD, Shimizu M et Vol No 1. MINERVA BIOTECNOLOGICA 29

8 MASCELLANI USING mirna EXPRESSION DATA FOR THE STUDY OF HUMAN CANCER al. MicroRNA profiling reveals distinct signatures in B cell chronic lymphocytic leukemias. Proc Natl Acad Sci USA 2004;101: Iorio MV, Ferracin M, Liu CG, Veronese A, Spizzo R, Sabbioni S et al. microrna gene expression deregulation in human breast cancer. Cancer Res 2005;65: Ciafre SA, Galardi S, Mangiola A, Ferracin M, Liu CG, Sabatino G et al. Extensive modulation of a set of micrornas in primary glioblastoma. Biochem Biophys Res Commun 2005;334: He H, Jazdzewski K, Li W, Liyanarachchi S, Nagy R, Volinia S et al. The role of microrna genes in papillary thyroid carcinoma. Proc Natl Acad Sci USA 2005;102: Calin GA, Ferracin M, Cimmino A, Di Leva G, Shimizu M, Wojcik SE et al. A microrna signature associated with prognosis and progression in chronic lymphocytic leukemia. N Engl J Med 2005;353: Volinia S, Calin GA, Liu C-G, Ambs S, Cimmino A, Petrocca F et al. A microrna expression signature of human solid tumors define cancer gene targets. Proc Natl Acad Sci USA 2006;103: Yanaihara N, Caplen N, Bowman E, Seike M, Kumamoto K, Yi M et al. microrna signature in lung cancer diagnosis and prognosis. Cancer Cell 2006;9: Calin GA, Croce CM. MicroRNA-cancer connection: the beginning of a new tale. Cancer Res 2006;66: Wu W, Dave N, Tseng GC, Richards T, Xing EP, Kaminski N. Comparison of normalization methods for CodeLink Bioarray data. BMC Bioinformatics 2005;6: 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: Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S et al. Bioconductor open software development for computational biology and bioinformatics. Genome Biol 2004;5:R Griffiths-Jones S. The microrna Registry. Nucleic Acids Res 2004;32:D Yi R, O Carroll D, Pasolli HA, Zhang Z, Dietrich FS, Tarakhovsky A et al. Morphogenesis in skin is governed by discrete sets of differentially expressed micrornas. Nature Genetics 2006;38: MINERVA BIOTECNOLOGICA March 2008