|Home | About | Journals | Submit | Contact Us | Français|
MicroRNAs (miRNAs) have emerged as novel cancer genes. In particular, the 17~92 cluster of miRNAs is highly expressed in haematopoietic cancers and promotes lymphomagenesis in vivo1,2. Clinical use of these findings hinges on isolating the oncogenic activity within the 17~92 cluster and defining its relevant target genes. Here we show that miR-19 is sufficient to promote leukaemogenesis in Notch1 induced T-cell lymphoblastic leukaemia (T-ALL) in vivo. Consistent with the pathogenic importance of this interaction, we report a novel translocation targeting the 17~92 miRNA cluster coinciding with a second rearrangement that activates Notch1 in T-ALL. To identify the miR-19 targets responsible for its oncogenic action, we conducted a large-scale short-hairpin RNA (shRNA) screen for genes whose knockdown could phenocopy miR-19. Strikingly, the results of this screen were enriched for miR-19 target genes, and include Bim (Bcl2L11)1,3, AMP-activated kinase (Prkaa1), and the tumour suppressor phosphatases Pten and PP2A (Ppp2r5e). Hence, an unbiased, functional genomics approach reveals a coordinate clamp down on several regulators of PI3K-related survival signals by the leukaemogenic miR-19.
MiRNAs are ubiquitous regulators of biological processes involved in normal development, differentiation, and diseases including cancer. They act by regulating gene expression at transcriptional and translational levels4. Several methods have been devised to identify genes that are regulated by miRNAs, e.g. programs like Targetscan can identify potential miRNA targets based on conserved sequence complementarity5,6,7. Experimental tools can define the actual effects of miRNAs on gene expression, e.g. expression profiles and comparative proteomics (SILAC)8,9, or measurements of incorporation into the miRNA machinery (HITS-CLIP)10. However, most likely only a limited number of miRNA targets is directly responsible for a specific phenotype11. An unbiased way of identifying these functionally relevant miRNA targets is needed to understand the biological basis of miRNA activity.
The oncogenic 17~92 cluster of miRNAs is of eminent importance in human hematopoietic cancers1,2,12. It has two paralogues and together these clusters encode 15 miRNAs with overlapping functions in development3. While miR-19 is important in Burkitt’s lymphoma13,14, which particular miRNA(s) deliver the oncogenic activity of these clusters is not established. Moreover, while comprehensive lists of potential targets of the 17~92 cluster are available, and some individual candidates have been confirmed experimentally3,15,16, the mechanism responsible for its oncogenicity remains unknown.
To identify the oncogenic activity within the 17~92 cluster and its paralogues we used an in vitro assay of hematopoietic transformation17. We tested all miRNAs representing families of seed sequences, specifically miR-17, miR-18a, miR-19b-1 (miR-19), miR-20a, miR-106a, miR-106b and miR-25 (Figure 1a)3. The assay is based on the interleukin-3 (IL3) dependence of FL5-12 lymphocytes, which undergo apoptosis when removed from IL3. We partially transduced FL5-12 cells with the individual miRNAs and GFP or empty vector (MIG), and monitored these mixed populations by FACS for changes in their relative proportions (Figure 1b). FL5-12 cells expressing miR-19 were rapidly enriched over parental cells upon IL3 depletion, and none of the other miRNAs had a similar protective effect (Figure 1c) (miR-19: p < 0.002, all other miRNAs and vector p > 0.05 by t-test). Accordingly, miR-19 showed direct protection from cell death in FL5-12 cells and had little effect on proliferation (Suppl. Figure 1). Hence, within the 17~92 cluster and its paralogues, miR-19 has a distinct ability to enhance lymphocyte survival in vitro.
Next, we assessed the expression of miR-19 in human lymphatic malignancies. Compared to lymphocytes from tonsils, we found a 5-17 fold increase in miR-19 expression in T-ALL, and less for other miRNAs in the 17~92 cluster (Figure 2a, Suppl. Fig. 2 and Suppl. Table 1). MiR-19 was also highly expressed in B-ALL and aggressive lymphomas. Consistent with previous profiling studies18,19, we found lower levels in indolent follicular lymphomas. Retroviral expression of miR-19 in the FL5-12 cells produced levels comparable to some tumour specimens (red bar in Figure 2a).
Karyotype analysis of an adult patient with T-ALL revealed the co-existence of two clonal translocations involving both TCRA/D loci in chromosome band 14q11. The first rearrangement, t(9;14)(q34;q11) affects the NOTCH1 gene locus in chromosome band 9q34, and 5′RACE confirmed the expression of a constitutively active form of NOTCH1 (exons 29–34) (Suppl. Figure 3)20,21. The second rearrangement - t(13;14)(q32;q11) - also involving the TCRA/D locus is novel and pinpoints a new TALL oncogene at 13q32. FISH analysis using genomic clones RP11-980D6 and RP11-97P7 reveals a split signal between chromosome 13 and the derivative chromosome 14 and maps the breakpoint close to the 17~92 cluster t(13;14)(q32;q11) (Figure 2b, c and Suppl. Fig. 4). Thus, the 17~92 cluster is the target of a novel genomic rearrangement in T-ALL, where miR-19 is highly expressed.
Most cases of T-ALL show mutations or translocations that result in increased expression or constitutive activity of a truncated Notch1 gene (Notch-ICN)20–22. To determine whether miR-19 could enhance Notch1-induced T-ALL in vivo, we co-transduced hematopoietic progenitor cells (HPCs) from fetal livers with Notch-ICN and miR-19 or empty vector23. We then transplanted the HPCs into irradiated recipients and monitored them for leukaemia (Figure 2d). Mice receiving HPCs expressing both Notch-ICN and miR-19 succumbed to ALL in ~ 2 months. At the same time, 80% of the mice receiving Notch-ICN remained healthy (p=0.0003) (Figure 2e). Pathology confirmed the diagnosis of T-ALL with abundant lymphoblasts, marrow infiltration, splenomegaly, lymphomas, and expression of T-cell markers (Figures 2f-h, Suppl. Figure 5 and Suppl. Table 2). T-ALLs expressing miR-19 or vector looked identical (not shown). We saw similar activity of miR-19 in the EμMyc model of Burkitt’s lymphoma (Suppl. Figure 6)13,14. Screening a pool of miRNAs containing all members of the 17~92 cluster and its paralogues for tumorigenesis in this model, revealed enrichment of cells transduced with miR-19 (Suppl. Figure 7). Hence, miR-19 behaves as an oncogene in murine models of lymphoma and Notch1-induced in T-cell lymphoblastic leukaemia (T-ALL).
We now wanted to determine which molecular targets were responsible for miR-19’s oncogenicity. Initially we combined computational target identification and gene expression analyses. Targetscan predicts 938 human and 744 murine miR-19 targets (Suppl. Tables 3 and 4). Gene expression profiles in FL5-12 cells transduced with miR-19 or vector showed only modest changes in expression (mean fold change −0.2 +/− SD 0.39) (Figure 3). Overall, expression levels of all predicted miR-19 targets were more reduced than other genes (p < 2e-04, Kolmogorov-Smirnov test). Among these were also potentially interesting candidates like the Pten tumour suppressor. However, there was no significant enrichment of miR-19 targets among genes showing more pronounced (1.5 or 2 SD) reductions in expression (p > 0.46 at 2 SD; p > 0.077 at 1.5 SD, Fisher’s exact test) (Figure 3 and Suppl. Table 5). Thus, expression analysis and target prediction readily confirm global effects of miR-19 and may lead to promising candidates, but they do not provide strong enough filters to identify the most relevant miR-19 targets.
Therefore, we wondered if an unbiased genetic screen might be an alternative means to functionally identify miR-19 targets involved in lymphocyte survival. Specifically, a short-hairpin RNA (shRNA) screen24,25 for genes whose knockdown phenocopies miR-19 in lymphocytes should also identify the genes responsible for miR-19 action. We conducted extensive control experiments using an shRNA against Bim (Bcl2L11), a known target of the 17~92 cluster1,2, and also increasing dilutions of miR-19 to establish the exact behaviour of miR-19 and a known target in our screening assay (Suppl. Figures 8a-d). We then devised a screening protocol involving transduction of FL5-12 cells with pools of ~1,000 shRNAs, and two cycles of IL3 depletion or continued passage in complete media (Figure 4a). We used custom half-hairpin arrays to measure changes in the abundance of shRNAs in treated (−IL3) versus untreated (+IL3) samples at two time points (T1 and T2, equal one or two cycles of IL-3 depletion, respectively)26. Unsupervised clustering showed good reproducibility between biological replicates (A-C; mean correlation r=0.60+/−0.17), and revealed a progressive shift in shRNAs across subsequent cycles of IL3 depletion (Figure 4b). Statistical tools helped us identify biologically significant signals. A Significance Analysis of Microarrays (SAM) identified changes in individual shRNAs (Figure 4c, see Suppl. Table 6 for the full analysis). A Gene Set Analysis (GSA) defined groups of shRNAs targeting the same genes. The GSA identified 14 genes, each targeted by at least 2 and up to 5 different shRNAs, and the top ‘hit’ was the alpha subunit of AMP-activated kinase (Prkaa1) targeted by 5 independent shRNAs (Figure 4d and Suppl. Table 7 for the full analysis).
Next, we individually validated candidate genes in the same experimental system. We included nearly all genes identified from the GSA analysis, and also tested protein-coding genes from the SAM analysis above an arbitrary threshold (≥ 1.65 fold increase, p < 0.05). In total, we re-tested > 70 genes and typically 3 shRNAs against each (all genes included in the validation are highlighted in Suppl. Tables 6 and 7). Ultimately, shRNAs against eight genes positively validated and produced a survival benefit in FL5-12 cells depleted from IL3 (in triplicate experiments p(vector/not shown) > 0.2 and p(for all shRNAs) < 0.05; t-test) implying a positive validation rate of ~ 10% (Figure 4e). Strikingly, in the human genome five of these genes harboured the miR-19 seed sequences. Besides the pro-apoptotic Bim1, these were the tumour suppressor gene Pten1, the alpha subunit of AMP-activated kinase (Prkaa1), the epsilon isoform of PP2A (Ppp2r5e), and the dedicator of cytokinesis-5 gene (Dock5). In addition, we validated three other genes that did not contain miR-19 binding sites in their 3′UTRs, the FoxO transcription factors FoxO1 and FoxO3, and Bnip3, a regulator of Rheb/mTOR and Bcl2 binding protein. Taken together, the results of this unbiased genetic screen for shRNAs that behave like miR-19 showed a highly significant enrichment for genes that contain miR-19 binding sites (p < 7.2 exp-07 in Fisher’s exact test; Suppl. Figure 9a). Unlike the human DOCK5 gene, the mouse gene is not a miR-19 target, and calculation of the enrichment statistics for the murine genome confirmed similarly high significance levels (p < 3.2 exp-05 by Fisher’s exact test; Suppl. Figure 9b). The prediction of miR-19 sites is based on Targetscan and did not take into account potential non-classical miRNA binding sites, e.g. those in the coding sequence (CDS). However, some of the eight genes also harbour potential CDS sites and these are shown in Suppl. Table 13. Thus, our unbiased, large-scale shRNA screen identified five human and four murine genes whose knockdown phenocopies miR-19 action in lymphocyte survival and that harbour miR-19 binding sites in their 3′UTRs.
Next, we wanted to test whether miR-19 actually regulates the expression of these genes. Reporter assays confirmed direct 3′UTR inhibition mediated by the predicted binding sites for these genes (Suppl. Figure 10 and Suppl. Table 9 and 12). Using qRT-PCR we measured the effect of miR-19 on mRNA levels of Pten, Ppp2r5e, Prkaa1 and Bim in miR-19 expressing murine T-ALL cells (Suppl. Figure 11) and in FL5-12 cells (Figure 5a), where we consistently observed up to 2 fold reductions (p (vector vs. miR-19) < 0.05 for all four genes, t-test) (Suppl. Table 8). Conversely, an antagomir against miR-19 caused an increase in mRNA levels in vitro (p(vector vs. anti-19) < 0.05 for all four genes) (Figure 5b, Suppl. Table 8), and both miR-19 and the antagomir had measurable effects on protein levels. For example, miR-19 produced a clear reduction in PPP2R5E, PRKAA1, BIM, and resulted in overall activation of PI3K signalling as indicated by increased phosporylation of AKT and the ribosomal S6 protein (Figure 5c, Suppl. Figure 12). While we did not detect an effect of miR-19 on PTEN protein, its mRNA was clearly decreased, and the antagomir produced increases in all miR-19 target proteins including PTEN (Figure 5b and c). As expected neither miR-19, nor the antagomir had an effect on Bnip3, which does not contain a 3′UTR site for miR-19 (Figure 5c). The larger effects of the antagomir compared to miR-19 expression may indicate that these genes are tonically suppressed by the endogenous miR-19 in proliferating cells. Hence, we confirm that the expression of Bim, Pten1, Prkaa1, and Ppp2r5e is regulated by miR -19 in lymphocytes.
Next, we wanted to assess the contribution of these target genes to miR-19 action. The pro-apoptotic BIM protein opposes BCL2 and is an important regulator of lymphocyte survival27. To measure if miR-19 had Bim-independent effects, we conducted complementation studies with FL5-12 cells engineered to stably express Bcl2 (FL5-12/Bcl2 cells). As expected the Bim shRNA conferred no additional benefit, whereas miR-19, and the shRNAs against Pten and Pp2r5e showed continued enrichment in the presence of Bcl2 (Figure 5d). Conversely, an antagomir against miR-19 produced anti-proliferative effects and FL5-12/Bcl2 cells expressing the antagomir were lost upon continued passage (Figure 5e). Similarly, miR-19 maintained its protective effect in the presence of Metformin-activated AMP kinase or enforced expression of Pten or Pp2r5e (Suppl. Figure 13). This indicates that multiple target genes contribute to miR-19 action in vitro. In vivo we find that individual knockdown of some targets is sufficient to accelerate leukaemogenesis, this is most striking for Pten (n=4, vector n=10, p < 0.001) and Bim (n=6; p < 0.001), an shRNA against Prkaa1 produces a modest effect (n = 5; p=0.3), while knockdown of Pp2r5e alone is unable to drive leukaemogenesis (n=4) (Figure 5f). Together, our data identify multiple miR-19 targets, which together act on PI3K-related signals and affect lymphocyte survival in vitro and leukaemogenesis in vivo (Figure 5g).
Our results provide new insights into the oncogenic activity of the miR-17~92 cluster in T-cell leukaemia. MiR-19, expressed at levels seen in human tumours, enhances lymphocyte survival and is sufficient to cooperate with Notch1 in T-cell lymphoblastic leukaemia (T-ALL) in vivo. Similarly, miR-19 can enhance c-Myc induced B-cell lymphomagenesis (Suppl. Figure 6)13,14. Given the conservation of the entire 17~92 cluster it is likely that these miRNAs carry out important functions in other tissues or biological processes12. The interaction of Notch1 and miR-19 is a relevant pathogenic mechanism in T-ALL. This is most strikingly demonstrated by the co-existence in the same leukaemic clone of two translocations activating the NOTCH1 gene - t(9;14)(q34;q11) - and targeting the 17~92 cluster - t(13;14)(q32;q11), respectively. It is likely that other mechanisms, for example transcriptional activation by c-Myc and Notch1 are more common causes of the increased miR-19 expression we observed in T-ALL specimens. Together, miR-19 is an oncogene that contributes to the molecular pathogenesis of T-ALL.
We identify a pattern of gene regulation by miR-19. Using an unbiased shRNA screening approach we find eight genes whose knockdown recapitulates miR-19 effects in vitro. Intriguingly, in the human genome five of these ‘hits’ harbour miR-19 binding sites and are regulated by miR-19. It is likely that these genes are among the key effectors of miR-19 action in lymphocyte survival, and that the regulation of multiple genes contributes to the overall activity of miR-19. These include Bim, a pro-apoptotic Bcl2 protein and target of the 17~92 cluster3, as well as AMP-activated kinase (Prkaa1), the tumour suppressor phosphatases Pp2a (subunit Ppp2r5e) and Pten11 (Figure 5g). Other miRNAs encoded in the 17~92 cluster are predicted to target some of the same genes, suggesting a potential mechanism for the functional cooperation between these clustered miRNAs11. These results suggest that miR-19 coordinates a PI3K pathway related program of cell survival in lymphocytes that contributes to leukaemogenesis.
Our results validate a new functional genomics approach to identify miRNA target genes. A genetic screen provides an unbiased tool to pinpoint miRNA targets genes that may be individually responsible for a specific phenotype, such as lymphocyte survival in this study. Some limitations should be noted, for example the screen is based on shRNAs targeting individual genes, as a consequence we may miss weak miRNA targets whose activity will only be apparent in the context of others. Also shRNAs and miRNAs act through somewhat distinct mechanisms6,28, and current technology cannot guarantee efficient protein knockdown by all shRNAs in a large library and this may. Short-hairpin RNA screens are an evolving technology, and limitations of array detection and shRNA validation likely account for the overall low (10%) validation rate. Despite the apparent complexity of gene regulation by miRNAs, our data indicate that phenotype-based genetic screens can identify key targets of miRNA action and thus complement other computational7,29 and genomic/proteomic8,10 methods of target identification. Notably, the cross-comparison of these methods will enable new insights into the biological activity of important miRNAs.
FL5-12 murine lymphocytes, cell cycle, apoptosis studies, viral transductions were as described30,17. All vectors are based on MSCV, and include the miRNA expression vector encoding miRs17, 18a, 19b-1 (miR-19), 106a, 106b, 252, Notch-ICN (a gift from W. Pear)23, Bcl231, the individual shRNA vectors (protein knockdown by shRNAs Suppl. Figure 14, vector constructs Suppl. Figure 15, Suppl. Table 10). The miRNA pool contained 48 miRNAs (all 17~92 and additional control miRNAs; the complete list is available upon request). The antagomirs MZIP19a-PA-1, MZIP19b-PA-1, and scrambled control (MZIP000-PA-1) were from System Biosciences(Suppl. Figure 16).
The mouse T-ALL and EμMyc lymphoma models have been reported23,31. Data were analyzed in Kaplan-Meier format using the log-rank (Mantel-Cox) test for statistical significance. The p53 LOH PCR and surface marker analysis were as described31.
Immunoblots were performed from whole cell lysates as described31. Antibodies were against Prkaa1 (2532, 1:1000, Cell Signaling), Bim/Bcl2L11 (AAP-330, 1:1000, Assay Designs), FoxO1 (94545, 1:1000, Cell Signaling), FoxO3a (9467, 1:1000, Cell Signaling) Phospho-FoxO3a (94665, 1:1000, Cell Signaling), Ppp2r5e (NB 100-845, NOVUS), Pten (9559, 1:1000, Cell Signaling), Tubulin (1:5000; Sigma, B-5-1-2) and Actin (1:5000; Sigma, AC-15), Bnip3 (3769, 1:1000, Cell Signaling), phosphorylated S6 (2215, 1:1000, Cell Signaling), phosphorylated Akt (4058, 1:1000, Cell Signaling), the Dock5 antibody was a gift from Alan Hall (MSKCC).
Total RNA and miRNA enriched RNA was extracted using the Allprep DNA/RNA/Protein and miRNeasy Mini. Pathological diagnosis was by expert haematopathologists at Weill Cornell University. cDNA synthesis and qRT-PCR and analysis by the ΔΔ Ct method as described 30. Taqman Gene Expression Assays: Bcl2l11 (Mm00437796_m1), Dock5 (Mm00555757_m1), FoxO1 (Mm00490672_m1), FoxO3 (Mm01185722_m1), Ppp2r5e (Mm00803759_m1), Prkaa1 (Mm01296695_m1), Pten (Mm01212532_m1), and Mouse GAPD (GAPDH) (4352932, Applied Biosystems), miR-19b (000396, Applied Biosystems) expression was normalized to RNU6B (001093, Applied Biosystems); primer sequences for miRNA detection are available upon request.
The Dock5, Ppp2r5e, Prkaa1, and Bim 3′-UTR fragments (Suppl. Table 11) were generated by PCR and cloned into the psi-CHECK-2 vector (Promega). The assays were performed as described1. The binding site mutants were generated by site-directed mutagenesis (Suppl. Table 12).
Metaphase chromosome preparations made from primary patient lymphoblasts were subjected to karyotype analysis following standard procedures. Genomic clones RP11-97P7 and RP11-980D6 located at 13q32 were from Invitrogen. DNA was labelled by nick-translation using spectrum orange dUTP fluorochrome (Vysis, Downers Grove, IL). A spectrum green-labelled RB1 probe was used as control for chromosome 13q hybridization. FISH was performed by standard methods on cells used for cytogenetic analysis. Hybridization signals were scored on at least 20 metaphase spreads on DAPI-stained slides using the Cytovision Imaging system attached to a Nikon Eclipse 600 microscope (Applied Imaging, Santa Clara, CA).
mRNA was extracted from primary patient lymphoblasts using the Nucleotrap mRNA extraction kit from Clontech, and 5′RACE was performed with the SMART RACE kit (Clontech) using a oligonucleotide primer complementary to the sequence of exon 29 of the NOTCH1 gene (5′-TCGTCCATGAGGGCACCGTCTGAAG -3′).
The pooled shRNA screen and half-hairpin array detection has been described, the library has been cloned from the original PSM2 constructs into the MRP vector and it contains ~ 14,000 distinct shRNAs, pooled into 1,000 shRNAs per pool25,26,30. FL5-12 cells were transduced at low multiplicity of infection in triplicates and each subjected to 2 cycles of IL3 depletion and rescue (T1 and T2). Samples were collected after viability had recovered for DNA isolation, PCR amplification of integrated shRNAs or the library as a reference, labelled and hybridized. Data generated (Nimblescan software) from image scans (Axon 4000BScanner) were imported into R version 2.4 for processing and analysis. Each Nimblegen 12-plex custom array consists of 12,033 half-hairpin probes (sequences are available on request). Most were represented by shRNAs in the library, and the remainder used for background estimation. Negative control spots showed lower intensities than the experimental probes on each array (Suppl. Figure 15c). Signal values less than the background were replaced with the background estimate to dampen large ratios due to low signal. The two channels for each array were then normalized using loess normalization and log ratios of the normalized intensity were utilized in further analyses.
The analyses were performed using the Bioconductor linear models for microarray (limma)32, Significance Analysis of Microarrays (SAM)33 and Gene Set Analysis (GSA)34 libraries in R. Correlations between biological replicates were calculated for each time point, principal component analyses (PCA) to determine the magnitude of experimental effects and biological replicates. Biological replicates did not constitute a major source of variation in the experiment. The mean correlation for biological replicates was 0.60 (+/−0.17) and likely reflected random variations as populations drifted in culture. Biologically significant signals were identified using the SAM software to select probes reflecting differences in relative abundance of shRNAs. We used the unpaired, two class algorithm, which assumes independent samples of each feature across the conditions. Q values are calculated empirically and used for identifying significant features. Overall, a gene was scored as being a potential candidate if its shRNA probe was found to fulfil these criteria: ≥ 1.65 fold change (FC), p < 0.05, and current gene bank annotation with a corresponding protein. Further testing was performed using GSA to identify candidates based on data from multiple hairpins targeting the same gene, as each gene is targeted by 2–4 hairpins their fold changes are summed. This statistic was compared to the same statistic from exhaustive permutation to arrive at empirical p-values for each gene. All gene sets showing positive fold-change in the GSA analysis were included in validations (except olfactory receptor481). These cut-offs are relatively arbitrary and were chosen as a conservative criterion to minimize false positives.
We performed several computational analyses to investigate whether the results of the screen might be attributable to off-target effects. We considered two hypotheses: (1) that shRNAs overrepresented in the screen have sequence similarity to miR-19 itself and hence mimic miR-19 activity via sites in 3′ UTRs; (2) that results of the screen are explainable by microRNA-like off-target effects on a few key genes. We ranked all shRNAs by ungapped sequence similarity of (a) the shRNA “seed” (positions 2–8) to the miR-19 seed region and (b) the shRNA to the full-length miR-19a and miR-19b sequences. No shRNAs in the library had the same 7-mer seed region as miR-19; among the shRNAs whose seed region matched the miR-19 7-mer seed at 6 positions (12 shRNAs with 6-mer seeds matching positions 3–8, 2 shRNAs with 6 -mer seeds matching positions 2–7), none were in the top 100 shRNAs overrepresented in the screen. The maximal sequence similarity between shRNAs in the library and the full-length miR-19a/b sequences was a 13 base ungapped match; 23 shRNAs had this degree of similarity to miR-19a and/or miR-19b, but none of them occurred in the top 100 overrepresented shRNAs from the screen. Moreover, we found poor Spearman rank correlation (| rho | < .01) between the ranking of shRNAs by overrepresentation in the screen and these similarity comparisons to miR-19. Next, we considered whether shRNAs that were highly ranked in the screen were more likely to have predicted off-target effects on PTEN, PRKAA1, BCL2L11, PPP2R5E, DOCK5, FOXO1, FOXO3 and BNIP3, compared to the full library. If this were true, overrepresentation of an shRNA in the screen might be explainable through off-target silencing of these key genes. To predict potential microRNA-like off-target effects of shRNAs, we scanned 3′ UTRs in the entire mouse genome for matches to 7-mer seed sequences (positions 2–8, no conservation filter) of shRNAs in the library. A gene with a 7-mer seed match for a given shRNA was considered to be a predicted off-target for this shRNA. We then tested if, for any of the 8 genes, the set of shRNAs that were predicted to “off-target” the gene were enriched in the top K overrepresented shRNAs from the screen, as compared to the full shRNA library. We used values of K ranging from 2 to 250, and we did not observe statistical significance by a Fisher’s exact test for any value of K (p > 0.05 in all cases for each gene). These controls indicate that the shRNAs are not similar to miR-19 itself, and that the screen result could not be explained by off-target effects of top ranked shRNAs on key genes. We computed the enrichment statistic with respect to the number of unique targets of the ~12000 shRNAs in the library and the subset of these targets that were also predicted miR-19 targets using Fisher’s exact test to account for the small number of ‘hits’. This analysis was done for both the human and the murine genome. Target predictions were as per Targetscan 5.1.
RNA isolation, quality control, cRNA synthesis, labelling hybridization, scanning according to standard protocols and manufacturer (Affymetrix) and analyzed using Partek® Genomic Suite™ 6.4. For statistical analysis, we centred the data using the mean log2 (expression change) of genes and normalized to have unit variance in log2 (expression change) across all genes. This normalization results in a modified Z-transformation of the data (Z score).
To compare the expression changes for miR-19 targets versus all genes, we compared their distributions of log(expression change) values using a one-sided Kolmogorov-Smirnov (KS) statistic, which assesses whether the distribution of expression changes for one set is significantly shifted downwards (down-regulated) compared to the distribution for the other set. The KS statistic computes the maximum difference in value of the empirical cumulative distribution functions (cdfs):
where is the empirical cdf for gene set j = 1, 2, based on nj (Z-transformed) log(expression change) values. We used the Matlab function kstest2 to calculate the KS test statistic and asymptotic p-value.
We thank Lin He, S.W. Lowe, and G. Hannon for access to shRNA and miRNA screening technologies. W. Pear, J. Huse, E.C. Holland for plasmids. H. Zhu for technical assistance, the MSK animal facility and RARC, A. Viale of the MSK Genomics Core, H. Zhao of cBIO program, J. Schatz and J. Massague for editorial advice, V. Murty for cytogenetic and FISH analyses. This work is supported by grants to HGW from the American Cancer Society, the Geoffrey Beene Cancer Center, Leukemia Research Foundation, The Louis V. Gerstner Foundation, May and Samuel Rudin Foundation, the NY Community Trust, the William and Blanche Foundation, Andrew Seligson Memorial Clinical Fellowship (JZ), NYStar (PJP), AF is supported by R01CA120196, the WOLF Foundation, the Rally across America Foundation, the Leukemia and Lymphoma Society (grants 1287-08, 6237-08), and AF is a Leukemia & Lymphoma Society Scholar.
Author contributions K.J.M, A.L.W. contributed equally; K.J.M, A.L.W., E.O. experimental design and analysis; K.d.K., T.P., A.F. T-ALL translocation analysis; K.M., J.Z., T.J., K.C., screen and analysis; A.A.K., C.S.L., J.S.P data analysis; P.J.P generated shRNA library; W.T. clinical specimens; H.G.W. designed the study and wrote the paper.
Author information The authors declare no competing financial interests.