Signal transducer and activator of transcription (STAT)-3 regulates microRNA gene expression in chronic lymphocytic leukemia cells

To find the putative promoter sites of miR genes, we used data on H2K4me3 enrichment regions published by Baer et al. [11]. To find STAT3 binding sites on these promoters, we used chromatin immunoprecipitation followed by high-throughput DNA sequencing (ChIP-seq) data (Additional file 1: Figure S1), generated as part of the ENCODE project [12] and acquired from the University of California Santa Cruz genome browser. For each putative STAT3 binding event a cluster score representing the ChIP-seq signal strength, ranging from 0 to 1000, was assigned.

After obtaining The University of Texas MD Anderson Cancer Center Institutional Review Board-approved informed consent, we obtained peripheral blood (PB) cells from 3 patients with CLL. To isolate low-density cells, we fractionated PB cells using Ficoll Hypaque 1077 (Sigma-Aldrich). More than 90% of the CLL PB cells were CD19+/CD5+ lymphocytes.

293T cells were co-transfected with GFP-lentivirus STAT3 short hairpin RNA (shRNA) or GFP-lentivirus empty vector and the packaging vectors pCMVdeltaR8.2 and pMDG (generously provided by Dr. G. Inghirami, Torino, Italy) using the Superfect transfection reagent (Qiagen, Inc.). 293T cell culture medium was changed after 16 hours and collected after 48 hours. The culture medium was filtered through a 45-μm syringe filter to remove floating cells, the lentivirus was then concentrated by filtration through an Amicon ultracentrifugal filter device (Millipore, Billerica, MA), and the concentrated supernatant was used to infect CLL cells. CLL cells (5 × 10⁶/mL) were incubated in 6-well plates (Becton Dickinson, Franklin Lakes, NJ) in 2 ml DMEM supplemented with 10% fetal calf serum and transfected with 100 μL of viral supernatant. Polybrene (10 ng/mL) was added to the viral supernatant at a ratio of 1:1000 (v/v). Infection efficiency was measured after 48–72 hours and was found to range between 40% and 70% (calculated on the basis of the ratio of propidium iodide (PI)-negative/GFP-positive cells).

Total RNA was extracted using Trizol reagent (Invitrogen) for microRNA analyses according to the manufacturer's protocol. For mature microRNA expression analysis, total RNA was retrotranscribed with microRNA-specific primers using TaqMan microRNA reverse transcription kit (Applied Biosystems, Foster City, CA), and then quantitative RT-PCR was performed using Taqman miR assays according to the manufacturer's protocol. The comparative cycle time (Ct) method was used to calculate the relative abundance of microRNAs compared with snRNA U6 as an internal control for RNA normalization. The profiling was done in duplicate wells for each sample and in two independent experiments (three measurements each), and the results were presented as mean +/− SD of the four measurements.

Western blot analysis was conducted using mouse anti-human STAT3 antibodies (BD Bioscience, Palo Alto, CA) and horseradish peroxidase-conjugated anti-mouse secondary antibodies (GE, healthcare, Amersham, Buckinghamshire, UK), as previously described [4].

Microarray analysis was conducted to identify miRs that were differentially expressed after silencing of STAT3 by transfection with short hairpin (sh) RNA or control treatment by transfection with empty vector. RNA was labeled and hybridized on miR microarray chips as previously described [13]. Signals in the images were quantified with GenePix Pro 6.0 software (Axon Instruments), and data were analyzed using PartekFlow software (v 2.1, 2012). Following quantile normalization, expression differences between the two sets of transfected CLL cells were tested by application of the Welch’s approximate t-test for two groups (with variances not assumed equal), with a P value cutoff of 0.05. Hierarchical clustering was generated for both genes and conditions, by using standard correlation as a measure of similarity.

The RNA22 pattern discovery algorithm, which utilizes miR sequences to predict miR:mRNA heteroduplexes, was used to simulate the sponge-out process [14].

A histone modification in which histone H3 is trimethylated on lysine 4 residues (H2K4me3) typifies transcription start sites in most gene promoters [15]. To identify miR gene promoters, Baer et al. interrogated these regions in CLL cells, normal B cells, and CLL-related cell lines [11]. Among the 939 miRs annotated to the miRBase15 database [16], they identified putative miR promoters in 781 miR genes [11].

Using those reports’ and the ChIP-seq ENCODE’s data [12], we searched for STAT3-binding sites in miR promoters. In our search, we identified 52,348 plausible STAT3-binding sites; STAT3 binding sites were found in 160 putative promoters of nearly 25% of the miR genes examined (N=200) with binding scores ranging from 100 (the lowest score) to 1000 (the highest score). Binding sites were found in 132 promoters of single pre-miR genes and in 28 promoters of pre-miR clusters of 2 to 6 genes (Table 1 and Additional file 2: Table S1). High STAT3-binding scores of 1000 and 710 were identified in miR-21 and miR-181b, consistent with the data of Iliopoulos et al. [9]. Remarkably, putative STAT3-binding sites were also detected in the promoters of miR-15-1 and miR-16a, which are deleted in most CLL patients with del(13q) [17].

Because putative STAT3 binding sites were identified in a significant number of miR promoters and STAT3 is constitutively activated in CLL, we sought to determine whether STAT3 would affect miR levels in unstimulated PB CLL cells. Transfection of CLL with STAT3-shRNA downregulated STAT3-mRNA and protein levels (Figure 1A). Microarray analysis showed that RNA levels of STAT3-shRNA-transfected cells were significantly different (P ≤ 0.05) from RNA levels of empty vector-transfected cells in 152 probes from 78 non-coding RNA genes, including 72 miR genes; the levels of 63 of the 72 miR genes were significantly downregulated by STAT3-shRNA, suggesting that STAT3 upregulates the levels of those miRs (Additional file 3: Table S2). For 60% of the 63 downregulated miR genes (n = 38) ChIP-seq data confirmed STAT3 binding in a putative promoter upstream of the gene location, significantly more than expected by chance (p<0.0001) (Additional file 2: Table S1).

To confirm these data, we studied miR levels of CLL cells transfected with STAT3-shRNA or empty vector using RT-PCR. Analyzed were miRs with either high levels of STAT3-binding sites, as reported in the ChIP-seq database (i.e. miR-21), reduced STAT3 levels following transfection with STAT3-shRNA in our studies (miR-181a), high STAT3-binding scores (213 for miR15-1 and miR-16a and 1,000 for miR-21), or a known pathogenetic role in CLL (miR15-1, miR-16a, miR-155, and miR-21) [18‐20]. As shown in Figure 1B, RT-PCR confirmed the array data in 5 of the 6 studied miRs.

Phosphorylated STAT3 binds to and activates the STAT3 gene, inducing the production of STAT3 protein [4]. Our current analysis revealed that STAT3 binds several miRs whose levels are overexpressed in CLL such as miR-21, an oncomiR of various neoplasms [20], and miR-29 and -181a, which affect CLL cell survival [21]. The levels of most miRs we studied were downregulated by transfection with STAT3-shRNA; however, the levels of 9 miRs were upregulated (Figure 1; Additional file 3: Table S2), suggesting that STAT3 downregulates those miRs’ levels. Several pathways can lead to STAT3 mediated suppression of transcription. Direct binding of unphosphorylated STAT3 suppresses transcription [22]. For example, miR-451, whose levels are downregulated by STAT3, carries a very high STAT3 binding score in its promoter. In addition, STAT3 mediates epigenetic silencing of a variety of genes [23], in particular genes that affect histone deacetylases [24, 25]. Therefore, we theorized that STAT3 induces epigenetic silencing of various miRs in CLL cells. Finally, STAT3 may also operate in a transcription-independent fashion. Single-stranded RNAs were shown to sequester (“sponge”) miRs in a sequence-specific manner [26]. Because constitutively phosphorylated STAT3 activates the STAT3 gene, single-stranded 4.3-kb STAT3-mRNA is generated at high levels in CLL cells. Those 4.3-kb RNA strands may bind complementary miRs, thereby interrupting their interaction with their corresponding target genes. For example, miR-18-a antisense (AS), upregulated by STAT3-shRNA has a complementary sequence in STAT3 mRNA, consistent with the “sponge regulation” hypothesis. Conversely, the miR-18-a sense form, downregulated by STAT3-shRNA, does not have a complementary sequence in STAT3 or the 3’UTR-STAT3.

To test whether this mechanism may be operative in CLL cells, we performed a simulation analysis using the 9 miRs whose levels were upregulated by STAT3-shRNA. We compared the theoretical energy released by single-stranded STAT3 mRNA sequences to that of random 4.3-kb mRNA sequences with an identical base content. Because miRs preferentially bind 3’-untranslated region sequences, we also calculated the energy release of shorter mRNA sequences. Using the RNA22 algorithm, we found higher energy release by full-length and short 3’-untranslated region STAT3 mRNA than by random single-stranded RNA sequences, suggesting that complementary binding of single-stranded STAT3 mRNA to miRs sequestered the circulating miRs, generating a stable, energetically superior state (Additional file 3: Table S2).

Taken together, our data suggest that STAT3 directly and indirectly modulates miR expression in CLL cells. Further studies to explore the mechanisms that affect miR expression in CLL are warranted.