Functional annotation of miRNA targets.
To test the utility of microarray profiling for the identification and functional annotation of miRNA targets, we devised and tested the experimental model described in the legend to Fig. . We collected a compendium of microarray profiles of cells transfected with miRNAs (38
). We selected 24 miRNAs (see Table S1 in the supplemental material) and then systematically examined expression profiles for the regulation of predicted miRNA targets and known biological pathways (Fig. ).
FIG. 1. Microarray profiling identifies a family of miRNAs that regulates cell cycle transcripts. (A) Strategy for identifying pathways regulated by miRNAs. HCT116 or DLD-1 Dicerex5 cells were transfected with miRNAs, and RNA was isolated 24 h posttransfection. (more ...)
We experimented with several cell lines, including Dicer hypomorphs of HCT116 and DLD-1 colon tumor cells (HCT116 Dicerex5
and DLD-1 Dicerex5
, respectively) (15
). These lines have homozygous disruption of the Dicer helicase domain and show reduced levels of many endogenous miRNAs (15
). Perhaps because of the loss of endogenous miRNAs, the hypomorphs show ~2-fold more intense expression changes following transfection with exogenous siRNAs or miRNAs than matched Dicer wild-type cells (see Fig. S1 in the supplemental material). We therefore used these Dicerex5
cells for subsequent experiments.
We transfected HCT116 Dicerex5
and DLD-1 Dicerex5
cells with 24 different miRNAs and determined expression profiles relative to those of mock-transfected cells (Fig. ). We measured expression profiles at 24 h, when mRNA silencing is maximal but secondary transcriptional effects due to protein depletion are minimal (28
). Examples of expression profiles for cells transfected with miRNAs are shown in Fig. S2 in the supplemental material. These miRNAs represent several families of miRNAs sharing identity in the miRNA seed region (see Fig. S3 in the supplemental material). Patterns of down-regulated transcripts were largely specific to each family. The family-specific transcripts were highly enriched with hexamer sequence motifs complementary to the seed region for that family (not shown). Nearly identical expression profiles were obtained with miR-141 and miR-200a; miR-17-5p, miR-20a, and miR-106b; and miR-192 and miR-215. Thus, in most cases, miRNAs sharing seed region identity regulated the same transcripts. miR-15a, miR-16, and miR-195 also gave nearly identical expression profiles, but miR-103 gave an overlapping but distinct profile compared to other members of this family, consistent with the 1-nt offset of the miR-103 seed sequence (see Fig. S3 in the supplemental material). Since expression profiles for each miRNA were largely consistent between the two cell lines tested, we utilized an intersection of profiles from the two cell lines for further studies.
We analyzed down-regulated transcripts for enrichment with hexamer motifs matching the miRNA seed region (seed region hexamers) in their 3′UTRs. Virtually all (23/24) miRNA intersection signatures showed enrichment with seed region hexamers highly unlikely to happen by chance [expectation (E
) < 1E(−20)]. We also measured highly significant enrichments with computationally predicted miRNA targets (34
) in miRNA-down-regulated signatures.
We next asked whether transcripts down-regulated by miRNAs were enriched for association with known biological pathways (Fig. ). Only 3 of the 24 intersection signatures showed significant [E < 1E(−2)] enrichment with transcripts associated with a GO biological process term (Fig. ); the three corresponding miRNAs (miR-15a, miR-16, and miR-103) are members of a single miRNA family (see Fig. S3 in the supplemental material). All three miRNAs triggered expression signatures showing significant enrichment with transcripts annotated with the terms “mitotic cell cycle” or “cell cycle.” Functional enrichment was less significant in miR-103 signatures than in those for miR-15a and miR-16.
The transcript annotation results suggest that targets regulated by miR-15a and miR-16, and to a lesser extent miR-103, are involved in the regulation of the cell cycle and cell growth. Both mitotic cell cycle transcripts and putative targets of miR-15a and miR-16 were enriched in 24-h down-regulated signatures. It was unclear, however, whether mitotic cell cycle transcripts were direct or indirect targets of miR-15a and miR-16. To help distinguish between these possibilities, we performed a kinetic analysis of transcript regulation. We hypothesized that if mitotic cell cycle transcripts were direct targets, they would be regulated sooner and would be more enriched with miR-16 seed region hexamers than if they were indirect targets. As shown in Fig. , transcripts enriched with miR-15a-miR-16 seed region hexamers in their 3′UTRs were regulated as early as 6 h following transfection. Mitotic cell cycle transcripts were not enriched with seed region hexamer matches in their 3′UTRs and were regulated later. Thus, it is likely that most mitotic cell cycle genes in miR-15a and miR-16 signatures are not direct targets of miR-15 and miR-16.
Cellular phenotypes triggered by miR-15a and miR-16.
We next investigated whether the gene expression phenotypes described above were manifested in changes in cell cycle distribution. Mock and luciferase siRNA-treated (not shown) or miR-106b-transfected HCT116 Dicerex5 cells showed a normal cell cycle distribution (Fig. ). In contrast, cell cultures transfected with miR-16 had increased numbers of cells in G0/G1 (diploid DNA content) and corresponding decreases in numbers of cells in S and G2/M. This finding suggests that miR-16 induced the accumulation of cells in G0/G1 in this cell line. Similar effects were seen with DLD-1 Dicerex5 cells (not shown). The G0/G1 accumulation phenotype became clearer when the microtubule-depolymerizing drug nocodazole was added 24 to 28 h posttransfection to block cells from reentering the cell cycle after mitosis. This treatment caused nearly all miR-106b-transfected cells to accumulate in G2/M (tetraploid DNA content), whereas a large fraction (~40%) of miR-16-transfected cells remained in G0/G1 (Fig. ).
FIG. 2. miR-15 and miR-16 cause the accumulation of cells at the Golgi stage of the cell cycle. (A) miR-16 triggers the accumulation of cells at a stage of the cell cycle. HCT116 Dicerex5 cells were transfected with miR-106b or miR-16. −Nocodazole, cells (more ...)
The fraction of miR-106b-transfected cells in G0/G1 decreased to a minimum and the fraction in G2/M reached a maximum ~12 h after nocodazole addition (Fig. ). The fraction of miR-16-transfected cells accumulating in G1 decreased and the fraction in G2/M increased more slowly with time in nocodazole (Fig. ). This result suggests that the G0/G1 accumulation phenotype is reversible or that cells are not actually blocked but rather progress more slowly through G0/G1. It is also possible that the effects of miR-16 transfection are transient.
We next asked whether miR-16 caused the G0/G1 accumulation phenotype in cells having wild-type Dicer function. We found that miR-16 induced a similar cell cycle phenotype in wild-type HCT116 and DLD-1 cells (see Fig. S4 in the supplemental material). In other experiments, we found that miR-16 induced a similar cell cycle phenotype in A549 (lung cancer), MCF7 (breast cancer), and TOV21G (ovarian cancer) cells (not shown). In contrast, miR-16 did not induce a measurable cell cycle phenotype in HeLa cells (not shown). Thus, miR-16 triggered the G0/G1 accumulation phenotype in most but not all cells with wild-type Dicer function.
Cellular phenotypes and gene expression profiles triggered by miR-16 that result from miRNA target regulation should be dependent upon the seed region of the mature strand (38
). To test this possibility, we tested the effects of miR-16 base pair mismatches on the cell cycle phenotype (Fig. ). The G0
accumulation phenotype was reversed by paired seed region mismatches at positions 2 and 3 (and 4 and 5; not shown) but not by mismatches outside the seed region at positions 18 and 19 (and 19 and 20; not shown). We also tested the effects of seed region mismatches on the gene expression profiles triggered by miR-16 and other family members (Fig. S5 in the supplemental material). We identified a set (n
= 116) of consensus transcripts down-regulated by miR-16 (see Table S4 in the supplemental material). Consensus miR-16-down-regulated transcripts were also down-regulated by miR-15a, miR-15b, and miR-195 but not miR-106b (see Fig. S5 in the supplemental material). Only a subset of these transcripts was regulated by miR-103 and miR-107. These transcripts were not regulated in cells transfected with miR-16 duplexes having mismatches at positions 2 and 3 or 4 and 5, but their regulation was unaltered in cells transfected with duplexes with mismatches at positions 18 and 19 and 19 and 20. Therefore, cell cycle and gene expression phenotypes induced by miR-16 family miRNAs are seed region dependent.
Cellular phenotypes are triggered by natural forms of miR-16 and reversed in loss-of-function experiments.
It is conceivable that the cellular phenotype induced by miR-16 is an artifact of the overexpression of the miR-16 mature form. To increase our confidence that miR-16 phenotypes were not artifactual, we determined whether the G0
cell cycle accumulation phenotype could be induced by miR-16 expressed under more natural conditions (i.e., requiring processing from a precursor). We approached this problem in two ways. First, we expressed an miR-16-encoding short hairpin RNA (shRNA) (46
). We also expressed miR-16 from a genomic fragment encoding the endogenous miR-15a-miR-16-1 locus. In both cases, HCT116 Dicerex5
cells were transfected with plasmids expressing precursor forms of miR-16 under the control of an H1 promoter, and cell cycle distributions were measured as described in the legend to Fig. . In these and other experiments, we found large variations in the absolute numbers of G0
cells in identically treated samples in different experiments. When it was necessary to compare results obtained from different experiments, we normalized percentages of cells in G0
so that the calculations for background cells and cells transfected with miR-16 (100 nM) gave 0% and 100% of cells in G0
, respectively (see the legend to Fig. for details). These normalized G0
values varied less than the raw values for identically treated samples between experiments. Both raw and normalized G0
values are given in Table S3 in the supplemental material. Levels of miR-16 were measured in the same samples by using a quantitative primer extension PCR assay (43
). This assay preferentially detects mature miR-16 over duplex and hairpin forms. We also determined cell cycle phenotypes and miR-16 levels induced by increasing concentrations of the miR-16 duplex.
FIG. 3. Natural forms and levels of miR-16 regulate cell cycle progression. (A) miR-16 expressed from hairpin precursors can trigger the phenotype of accumulation of cells at a stage of the cell cycle. HCT116 Dicerex5 cells were transfected with increasing concentrations (more ...)
As shown in Fig. , the percentage of cells in G0/G1 increased with rising concentrations of miR-16. The distinct accumulation of cells in G0/G1 was achieved with miR-16 duplex concentrations of 0.5 to 1 nM, which resulted in miR-16 levels of ~3,500 to 5,600 copies/20 pg, respectively. This compared with ~300 copies/20 pg measured with mock- or luciferase-transfected cells. Both the miR-16 shRNA and miR-16 locus consistently triggered a ~2-fold increase in the number of cells in G0/G1 and an increase in miR-16 levels to ~3,000 copies/20 pg. The miR-16 shRNA triggered identical results in HCT116 wild-type cells (not shown), suggesting that the processing of the hairpin does not require full Dicer activity. Transfection with empty vector and an shRNA for miR-106b did not result in the accumulation of cells in G0/G1 or miR-16 levels greater than those in mock-transfected cells (not shown). At equivalent miR-16 levels, the duplex was slightly less efficient at inducing the accumulation of cells in G0/G1, consistent with the idea that not all the duplex with which cells were transfected was accessible for incorporation into the RNA-induced silencing complex. The overexpression of miR-16 from hairpin precursors therefore triggered the accumulation of cells in G0/G1 with a level of efficiency similar to that of the miR-16 duplex.
It was also important to demonstrate that the cellular phenotype induced by miR-16 in gain-of-function experiments was reversed in loss-of-function experiments. We therefore performed loss-of-function analysis using LNA- and 2′-O
-methyl-modified oligonucleotides (anti-miRs) to mediate the specific inhibition of miRNA function (26
). To demonstrate the effectiveness of transfection with anti-miRs, we first sought to demonstrate that endogenous miR-16 and miR-106b levels were reduced by this treatment. We found, however, that anti-miRs interfered with miRNA quantitation by a PCR assay (43
). We therefore tested the ability of anti-miR-16 to block the accumulation of cells in G0
induced by transfection with miR-16. Wild-type HCT116 cells were first transfected with miR-16 and then transfected with anti-miR-16 or anti-miR-106b 24 h later. We found that anti-miR-16, but not anti-miR-106b, abolished the accumulation of cells in G0
induced by miR-16, indicating the effectiveness of our anti-miR-16 transfection protocols (not shown).
We then transfected wild-type HCT116 cells with LNA-modified anti-miR-16 and anti-miR-106b and measured levels of several miR-16 target transcripts using real-time PCR. These preliminary experiments showed increased expression of several miR-16 targets in anti-miR-16-transfected cells compared to anti-miR-106b-transfected cells, but the effects were small (<30% increase for each target).
To better quantify the effects of loss-of-function experiments, we used microarray profiling to measure the effects of anti-miRs on consensus miR-16 and miR-106b transcripts (see Table S4 in the supplemental material). We reasoned that small gene expression changes would assume greater statistical significance if measurements of many transcripts were combined. We transfected HeLa, TOV21G, HCT116 wild-type, and Dicerex5 cells with 2′-O-methyl oligonucleotide inhibitors of miR-16 and miR-106b (anti-miR-16 and anti-miR-106b, respectively) and examined changes in gene expression by using microarrays.
Consistent with the real-time PCR results, consensus miR-16-regulated transcripts were weakly up-regulated by anti-miR-16 but not anti-miR-106b. A heat map of results obtained with HeLa cells is shown in Fig. . miR-16 and miR-106b consensus transcripts were weakly but detectably up-regulated by the specific anti-miR but not the nonspecific anti-miR. For other cells with wild-type Dicer function, the median regulation of consensus miR-16 and miR-106b targets was also slightly increased in cells transfected with the specific anti-miR (Fig. ).
It was of interest to determine the overlap of genes up-regulated in loss-of-function experiments and those down-regulated in gain-of-function experiments. The numbers of up-regulated miR-16 or miR-106b targets were higher in Dicer wild-type cells treated with the specific anti-miR (Fig. ). However, it was difficult to quantify the overlap in gene sets precisely because the magnitude of the regulations induced by anti-miRs was small. In HeLa cells, it appeared that nearly 100% of down-regulated targets were up-regulated by specific anti-miRs (Fig. ).
When considered as a group, miR-16 targets were more significantly up-regulated than miR-106b targets in cells treated with anti-miR-16 (Fig. ). Target regulation was greatly reduced in HCT116 Dicerex5 cells, as expected given the reduced miRNA levels in these cells. Likewise, anti-miR-106b-treated cells showed significant regulation of miR-106b, but not miR-16, consensus targets. Taken together, these results indicate that in cells with wild-type Dicer function, many, if not most, miR-16 and miR-106b targets down-regulated by miRNA duplexes in gain-of-function experiments were up-regulated by specific anti-miRs in loss-of-function experiments.
It was also important to determine the effect of anti-miRNAs on cell cycle progression. miR-16-regulated transcripts annotated with the GO biological process term mitotic cell cycle (Fig. ) were significantly up-regulated by anti-miR-16 in TOV21G and HCT116 wild-type cells (Fig. ). These transcripts were less regulated in HeLa cells, which do not show an miR-16 cell cycle phenotype, and HCT116 Dicerex5 cells, which have reduced endogenous levels of miR-16. Mitotic cell cycle transcripts were not significantly regulated in any cell line by anti-miR-106b (Fig. ). These results support the regulation of cell cycle progression by physiological levels of miR-16. However, analysis of the cell cycles of anti-miR-16-treated TOV21G and HCT116 wild-type cells did not reveal any obvious differences from that of control-treated cells (not shown). Thus, gene expression changes measured by microarray were not sufficient to drive a detectable cell cycle phenotype in these cells.
We hypothesized that cells having higher endogenous levels of miR-16 would be more susceptible to anti-miR-16-induced phenotypic changes detectable by flow cytometry. We therefore screened a number of transfectable cell lines for endogenous miR-16 levels by using a quantitative primer extension PCR assay (43
). These experiments showed that SW1417 cells had elevated levels of endogenous miR-16 (~1,500 copies/cell; not shown). The transfection of SW1417 cells with anti-miR-16 resulted in a significant decrease in numbers of cells in G0
compared to mock-transfected cells [10% ± 1% decrease (three independent experiments); P
< 1E(−3)]. In contrast, anti-miR-106b-transfected cells did not show significant differences from mock-transfected cells [5%+
3% decrease (three independent experiments); P
> 5E(−2)]. Therefore, disruption of physiological miR-16 levels in certain cell types can alter cell cycle distribution.
Characteristics of miR-16-down-regulated transcripts.
Consensus miR-16-down-regulated transcripts (see Table S4 in the supplemental material) overlapped with but were not identical to computationally predicted miR-16 targets (see Fig. S6 in the supplemental material). Nearly 60% (65/110) of miR-16-down-regulated transcripts were not predicted by either of two different computational methods (34
). Conversely, >90% of the computational targets were not significantly down-regulated on microarrays. Moreover, most computational targets were unique to the particular method used (55% and 68% unique for TargetScan and PicTar, respectively). Similar results were found (not shown) with miR-16 targets predicted with other computational methods (31
). The poor overlap was unlikely to be due to a lack of target expression in HCT116 Dicerex5
cells, since the calculations were restricted to transcripts expressed at ~≥1 copy per cell.
We next compared the properties of miR-16-down-regulated transcripts with those of a set of unregulated (background) transcripts selected to have similar distributions of expression levels. Hexamer motifs matching the miR-16 seed region (target sites) were found in both CDS and 3′UTRs. The numbers of transcripts with CDS target sites did not differ significantly between the miR-16-down-regulated transcripts and the background set (not shown). In contrast, nearly all miR-16-down-regulated transcripts had hexamer matches in their 3′UTRs, compared to less than half of the background set [P < 5E(−13)] (Fig. ).
FIG. 4. miR-16-down-regulated transcripts contain multiple miR-16 target sites. Properties of consensus miR-16-down-regulated transcripts (see Table S4 in the supplemental material) were compared with those of an expression level-matched background set. The significance (more ...)
The median number of target sites per transcript (Fig. ) was also higher for miR-16 transcriptional targets than for the background set [P < 1E(−10)]. The increased number of target sites per transcript for miR-16-down-regulated transcripts was partially attributable to longer 3′UTRs in these transcripts [median of ~1,150 nt for miR-16-down-regulated transcripts versus ~660 nt for the background set; P < 1E(−2)] but more significantly to a greater number of target sites per kilobase of the 3′UTRs [P < 4E(−10)] (Fig. ).
Furthermore, the longest target site per transcript was significantly longer for miR-16-down-regulated transcripts than for background transcripts [P < 5E(−14)] (Fig. ). The median length of the longest sites in individual miR-16 targets was 8 bases, versus 0 bases for background transcripts. Even when transcripts without hexamer matches were excluded from the background, the median length of the longest sites in individual background transcripts with hexamer matches was only 6 bases [not shown; P < 3E(−10)]. Thus, many miR-16-down-regulated transcripts have multiple target sites matching the miR-16 seed region, generally with at least one site showing extended complementarity to the miR-16 seed.
Many miR-16-down-regulated transcripts regulate G0/G1-to-S cell cycle progression.
We wished to determine which miR-16-down-regulated transcript(s) was essential for the G0/G1 cell cycle accumulation phenotype. We hypothesized that siRNA-mediated silencing of such transcripts would yield a phenocopy of the miR-16-induced phenotype (i.e., trigger the accumulation of cells in G0/G1). To test this idea, we transfected HCT116 Dicerex5 cells with siRNAs (pools of three siRNAs/target) targeting each miR-16-down-regulated transcript (n, 102 individual pools targeting the well-characterized genes from Table S4 in the supplemental material). We then analyzed transfected cells for their cell cycle distributions. Sequences of siRNAs triggering the accumulation of cells at a stage of the cell cycle are shown in Table S1 in the supplemental material. Raw and normalized G0/G1-cell accumulation values and target silencing data are presented in Table S3 in the supplemental material.
As shown in Fig. , 25/102 siRNA pools (~25%) targeting miR-16-down-regulated transcripts triggered the accumulation of ≥20% of the cells in G0/G1. In contrast, only 4/51 (~8%) of siRNA pools targeting transcripts that did not match the miR-16 seed triggered G0/G1-cell accumulation (P < 0.01; Fisher's exact test). Examples of cell cycle phenotypes triggered by siRNA pools are depicted in Fig. . These experiments provided further evidence that many miR-16 family targets regulate G0/G1 transition.
FIG. 5. miR-16-down-regulated transcripts cooperatively regulate cell cycle progression. (A) miR-16-down regulated transcripts are enriched with targets whose disruption causes the accumulation of cells in G0/G1. HCT116 Dicerex5 cells were individually transfected (more ...)
The siRNA pool targeting CARD10 was the only one that induced the accumulation of cells in G0/G1 to the same extent as miR-16 (see Table S1 in the supplemental material; Fig. ). However, siRNA titration experiments showed that the phenotype induced by an individual CARD10 siRNA required greater CARD10 gene silencing than was achieved with miR-16 (not shown). Individual siRNAs targeting CDK6, CDC27, and C10orf46 genes also triggered less G0/G1-cell accumulation than miR-16, despite silencing their target mRNA more strongly than miR-16 (not shown and Table S1 in the supplemental material). HCT116 Dicerex5 cells transfected with miR-16 (100 nM) had CDK6, CARD10, CDC27, and C10orf46 gene transcript levels ranging from ~45% to 75% of the maximum (i.e., levels in control-transfected cells; not shown). In contrast, cells transfected with the best siRNAs corresponding to CDK6, CARD10, CDC27, or C10orf46 genes (33 nM) had transcript levels of <40% of the maximum (see Table S1 in the supplemental material). Therefore, the silencing of individual miR-16-down-regulated targets can produce phenocopies of the miR-16-induced phenotype, but the effects are weaker.
One explanation for these findings is that the stronger cell cycle phenotype elicited by miR-16 resulted from the coordinate silencing of multiple targets. To test this possibility, we devised a strategy for comparing phenotypes caused by siRNAs corresponding to different targets alone and in combination (i.e., pools of siRNAs corresponding to different targets). If the miR-16 phenotype results from the coordinate silencing of different targets, then siRNAs corresponding to miR-16 targets should be more effective when added together than when tested individually. A potential limitation of this approach is that pooled siRNAs can compete for RNA-induced silencing complex binding, which may reduce silencing by the individual siRNAs (23
We first determined whether the siRNA pools we tested were representative of the individual siRNAs comprising them. Individual siRNAs from the 25 siRNA pools that gave an accumulation of ≥20% of cells in G0/G1 were tested for their abilities to produce a phenocopy of the miR-16 phenotype. In some cases, additional siRNAs corresponding to the same targets were tested. For 24/25 pools, at least one member of the siRNA pool gave a phenotype as strong as that given by the pool itself (see Table S3 in the supplemental material).
We then performed experiments to determine whether pooled siRNAs corresponding to different miR-16 targets were more effective than the individual siRNAs at triggering arrest in G0/G1. For these experiments, we transfected cells with individual siRNAs at a concentration at which most gave phenotypes only slightly above the background (0.25 nM) (Fig. ). We tested whether a pool of siRNAs representing each target whose disruption caused ≥20% of cells to accumulate in G0/G1 (pool 1) (see Table S3 in the supplemental material) triggered a stronger phenotype than the individual siRNAs. To quantify these comparisons, we normalized percentages as described in the legend to Fig. . None of the individual siRNAs triggered a normalized value of ≥20% of cells to accumulate in G0/G1 (at 0.25 nM), whereas a pool of all 24 siRNAs (pool 1; 0.25 nM [each]; total concentration, 6 nM) triggered a normalized value of 73% of cells to accumulate.
In another experiment, we compared individual siRNAs (0.25 nM) targeting the three targets that gave the strongest phenotypes as indicated in Fig. (CARD10, KIAA0317, and C9orf91 genes) (Table S3 in the supplemental material) to a pool of all three siRNAs (pool 2; 0.25 nM [each]; total concentration, 0.75 nM). Each individual siRNA triggered a normalized value of <30% of cells to accumulate in G0/G1, but the pool of three siRNAs triggered a normalized value of ~50% of cells to accumulate (see Table S3 in the supplemental material). These experiments supported our hypothesis that the robust cell cycle phenotype elicited by miR-16 results from coordinate silencing of multiple targets.
To address coordinate silencing and the G0
-cell accumulation phenotype more quantitatively, we focused on a more limited subset of targets (Fig. ). We first needed to demonstrate that the cell cycle phenotypes triggered by siRNAs corresponding to these targets resulted from the silencing of the intended targets rather than unintended targets (28
). For CDK6, CARD10, CDC27, and C10orf46
genes, we identified at least two siRNAs that triggered the accumulation of cells in G0
. We then tested these siRNAs for their abilities to silence their target genes (see Table S3 in the supplemental material). Generally, siRNAs triggering phenotypes also silenced their targets well.
We next explored whether representative individual siRNAs corresponding to each target were more effective when pooled together than they were alone. Individual siRNAs targeting CDK6, CDC27, CARD10, and C10orf46 genes triggered minimal levels (normalized value of <12%) of cells to accumulate in G0/G1 when tested at 0.25 nM (Fig. ). However, when these siRNAs were pooled (pool 3; total concentration, 1 nM), a much stronger phenotype was observed (a normalized value of ~61% of cells in G0/G1) (Fig. ). (Equivalent results were obtained when luciferase siRNA was added to individual siRNAs to maintain a total concentration of 1 nM.) Measurements of target silencing by quantitative PCR or immunoblotting showed that the silencing of target transcripts was maintained or slightly reduced when siRNAs were pooled (see Fig. S7 in the supplemental material).
The effects of the four siRNAs tested as described in the legend to Fig. were greater than additive. If the effects were additive, we would have expected to find ~34% more pool-transfected cells (normalized value) in G0/G1 than control-transfected cells (19% + 11% + 8.3% − 2.7% = ~36%) (see Table S3 in the supplemental material). This compares with the normalized value of ~61% of cells in G0/G1 that we actually observed (Fig. ; see Table S1 in the supplemental material). In four independent experiments, 67% ± 15% of pool-transfected cells were in G0/G1 versus the expected value of 37% ± 13% [P < 8.8E(−7); chi-square distribution]. Taken together, these findings demonstrate that miR-16 coordinately regulates targets that collaborate to regulate cell cycle progression from G0/G1 to S.
Levels of miR-16 family-down-regulated transcripts negatively correlate with miR-195 levels in human tumors.
We wished to determine whether the phenotypes we showed for miR-16 family members in cell culture are relevant to an in vivo setting. We reasoned that steady-state levels of transcripts down-regulated by miR-16 family members would inversely correlate with miRNA levels. To examine this possibility, we compared levels of an miR-16 family member with levels of miR-16-down-regulated transcripts in a panel of human tumors and matched adjacent normal tissue samples (see Table S6 in the supplemental material). We chose miR-195 for comparison because it is the only member of the family for which we have reliable tumor atlas expression data. mRNA and miRNA expression levels in tumors and adjacent normal tissues were expressed as ratios of these levels to expression levels in a pool of normal samples of each tissue type. As shown in Fig. , we observed significant negative correlations between miR-195 levels and the levels of transcripts down-regulated by miR-16 at 24 h posttransfection. miR-16-down-regulated transcripts were significantly more likely to be negatively correlated with miR-195 levels than would be expected by chance [P < 1.5E(−12); Wilcoxon rank-sum P value for a difference in median correlation coefficient versus random permutations of expression ratios]. Thus, tumors with high levels of miR-195 tended to have low levels of transcripts that were down-regulated by transfection with miR-16, and vice versa. These results, therefore, show that gene expression changes triggered by miRNA transfection in our in vitro model reflect the relationship between levels of the transcripts and the miRNA in human tumors.
FIG. 6. Levels of miR-16-down-regulated transcripts negatively correlate with miR-195 levels in human tumors. RNA was isolated from a series of 29 tumors and 28 adjacent uninvolved normal tissues. mRNA expression was measured using microarrays, and miR-195 levels (more ...)