The derivation, differentiation, viral infection, and immunostaining of primary mouse NSCs were detailed in our previous study (Gao et al., 2011). Briefly, mouse NSCs were dissociated from dissected hippocampi and lateral ventricles from 3-week old REST loxP/loxP or WT littermate mice. Pooled hippocampal and SVZ NSCs (“neurospheres”) were cultured in DMEM:F12 supplemented with N2 and B27 (Invitrogen) in the presence of FGF2 (20 ng/ml), EGF (20 ng/ml), and heparin (5 μg/ml; proliferation conditions) on uncoated tissue culture plates. To delete REST, neurospheres were trypsinized and infected in suspension with GFP (control) or Cre-GFP adenovirus (1/10,000 of 1 × 10¹⁰ pfu/ml stock; University of Iowa) in N2/B27 medium containing FGF2/EGF/heparin. Three days after Ad-Cre infection, where over 95% of the NSCs were GFP+, the recombination rate was determined to be greater than 80–90% (Gao et al., 2011). NSCs were induced to differentiate with retinoic acid (RA; 1 nM) and forskolin (5 nM) for indicated time points. Semi-quantitative RT-PCR was carried out to confirm induction of select neuronal and/or REST target genes (Gao et al., 2011).

For miRNA array profiling, qRT-PCR and validation, total RNA samples were prepared using the miRNeasy kit (QIAGEN). Five micrograms of total RNA (from control and cKO cells) were sent to LC Sciences (Houston, TX, USA) for miRNA array analysis using Sanger miRBase Release 16.0. Multiple redundant regions were included on the chip and each region further comprised a miRNA probe region plus multiple control probes. Statistics were done with ANOVA analysis. Hierarchical clustering analyses of miRNA array profiling data were done using the software CLUSTER (Eisen et al., 1998). Bioinformatics prediction of RE1 sites was accomplished using TFbind (Tsunoda and Takagi, 1999).

To confirm the miRNA array profiling results, one microgram of total RNA was used for reverse transcription using NCode miRNA first-strand synthesis and QPCR kits (Invitrogen). QPCRs were performed using an ABI 7900HT instrument. The PCR program consisted of 2 min at 50°C and 2 min at 95°C, followed by 40 cycles of 15 s at 95°C and 60 s at 58°C. Primer quality was analyzed by dissociation curves. MiRNA expression was normalized to the RNA input (Lossos et al., 2003). Statistics were performed using unpaired Student’s t-test. Values of p ≤ 0.05 were considered significant.

For the array profiling experiment, we conducted the following quality control steps before performing in-depth analyses. As expected, we did not detect any significant global differences in the total expression of miRNAs between the control and cKO cells (Figure A1 in Appendix). For both groups, out of the total 1040 miRNAs examined, only a quarter have significant expression (defined as having an intensity above 32 as the service provider recommends) in each experimental condition (Figure A2 in Appendix). About 8% have an expression value above 500 but account for more than 90% of the total miRNA expression (Figure A3 in Appendix). About 2% had an intensity score above 5000 but account for more than 60% of the total expression. The top expressed miRNAs are either ubiquitous (e.g., let-7 family) across lineages or neural enriched (e.g., miR-9, -21, -132, -16, -26a, -99b, and -125; Cao et al., 2006; Table A1 in Appendix). When we examined the expression of miRNA gene clusters including miR-17-92, -106b-25, -106a-363, and -297-669 loci, we observed a strong correlation among miRNAs from the same loci (Figure 2; Figure A4 and Table A2 in Appendix). In particular, from the miR-297-669 gene locus that comprised more than 70 miRNAs, all miRNAs with an intensity of 32 or above displayed similar expression dynamics throughout the experimental series (Figure A4 in Appendix).

As previously detailed (Gao et al., 2011), control and cKO cells were cultured with the desired treatments at indicated time points, crosslinked, and harvested. The chromatin immunoprecipitation (ChIP) assay was then performed following manufacturer’s instructions (Invitrogen). Antibodies used for ChIP were as follows: REST/NRSF (Upstate), mSin3A (Santa Cruz), and CoREST (Upstate). ChIP experiments were from independent chromatin preparations. QPCR and quantification of the results were performed as previously described (Gao et al., 2011). All primer sets were subjected to a dissociation curve analysis. The relative enrichment was determined by taking the absolute quantity ratios of specific IPs normalized to the input groups.

We recently demonstrated conditional deletion of REST led to reduced proliferation and precocious neuronal differentiation in hippocampal and SVZ mixed NSC cultures (Gao et al., 2011). Consistent with these findings, we observed dramatic upregulation of neuronal and/or REST target genes (Gao et al., 2011). To begin evaluating REST’s possible function in miRNA regulation, we performed miRNA microarray profiling using the same culture system (Gao et al., 2011). Floxed REST NSCs under proliferation conditions expressed the progenitor markers Nestin and Sox2 and generated Tuj1-positive neuronal cells and GFAP-positive astroglial cells under differentiation conditions consistent with their multipotentiality (Figure 1A). As indicated in Section “Materials and Methods,” we deleted REST by Ad-Cre-GFP infection of floxed REST NSCs to generate REST conditional knockout (cKO) NSCs. Consistent with REST deletion at a recombination efficiency of 80–90%, we confirmed robust upregulation of REST target neuronal genes NeuroD1 and Tuj1 in cKO NSCs, as well as precocious neuronal differentiation under proliferation conditions (Figure 1B; Gao et al., 2011). Upregulation of the immature neuroblast marker doublecortin (DCX) also confirmed that REST deletion precociously induced neuronal differentiation (Figure 1B; Gao et al., 2011).

First, we validated the microarray profiling analysis by performing quality control steps (see Materials and Methods; Figures A1–A4 in Appendix and Tables A1– A3 in Appendix). We then performed qRT-PCR analysis to confirm a subset of miRNAs (either with the highest expression and/or largest fold change) from the array data and confirmed ∼70% of the array results (Table A3 in Appendix). We identified a subset of miRNAs that are significantly changed (p < 0.01) in cKO NSCs compared to control cells (Table A4 in Appendix). Among them, there were 29 miRNAs with over twofold expression increase and 35 miRNAs with over twofold expression decrease in cKO NSCs (Table A4 in Appendix). The only two known REST targets that were significantly changed in cKO NSCs compared to control cells were miR-30a (Cre-GFP/GFP = 0.43; p-value 2.95E−03) and miR-9* (Cre-GFP/GFP = 0.6; p-value 1.67E−03). However, instead of being upregulated, they were downregulated by 2.3 and 1.6 fold, respectively, in cKO NSCs compared to control cells. Since there are currently over 40 known REST miRNA targets (Wu and Xie, 2006; Otto et al., 2007), it is surprising that we did not observe expression changes of additional REST miRNA targets, at least under proliferation conditions.

We hypothesized that the failure to observe significant upregulation of a large number of REST target miRNAs may be due to the lack of differentiation stimuli, which would be consistent with our previous findings that REST deletion results in enhanced neuronal differentiation when primed with RA and forskolin (FSK; Gao et al., 2011). Thus, we treated cKO and control NSCs with RA and FSK and performed miRNA profiling. We performed hierarchical cluster analysis and identified expression changes of known REST miRNA target genes, which were grouped into 6 clusters (Figures 1C–I). Cluster 1 contained miRNAs that were upregulated during differentiation and further enhanced upon REST deletion (Figure 1D). Clusters 2 and 3 contained miRNAs whose expression was initially downregulated and eventually restored during differentiation (Figures 1E–F). Loss of REST blocked this transient downregulation by augmenting their expression throughout the differentiation process (Figures 1E–F). The difference between cluster 2 and 3 was that cluster 2 contained highly expressed miRNAs. Cluster 4 contained miRNAs that were upregulated upon differentiation induction, which was blocked by REST ablation (Figure 1G). Cluster 5 contained miRNAs whose expression either persisted or slightly decreased upon RA and FSK treatment, which was further reduced with REST deletion (Figure 1H). Finally, cluster 6 contained miRNAs which did not display altered expression, regardless of REST ablation, either in proliferation or differentiation conditions, suggesting that REST may not be required for their expression in this context (Figure 1I). These results confirm that a subset of known REST target miRNAs were indeed subjected to REST regulation in mouse NSCs under differentiation, but not proliferation conditions. These results are also consistent with previous findings where downregulation of REST failed to globally activate or de-repress REST target gene expression (Chen et al., 1998; Ballas et al., 2005; Conaco et al., 2006), highlighting the context-dependent and dynamic nature of REST-dependent miRNA changes.

Among the miRNAs whose expression were dynamically altered during differentiation induction, several groups are worth mentioning. The first class represented “high-expressers” (Figure 2A and Table A5 in Appendix). MiR-709, previously implicated in tumorigenesis and the DNA damage response (Li et al., 2011; Tamminga et al., 2008), was the overall highest expressed single miRNA accounting for 5–14% of the total miRNAs expressed. The let-7 family of miRNAs was the highest expressed class and together comprised 9–32% of the total miRNA expression. Other high-expressers included miR-9, -9*, and -21 which are previously described REST targets that function in neural differentiation (Gao, 2010). We also observed robust upregulation of miR-1937a and -1937b, small miRNA-like RNAs of tRNA origin, immediately after RA and FSK treatment. They were the two highest expressed miRNA species, together accounting for 21% of the total miRNA expression in control cells 3 h after RA and FSK treatment, a time point when NSCs transiently increased their proliferation as they differentiate. Interestingly, this dramatic upregulation was abolished by REST ablation, which simultaneously prevented the transient increase in proliferation (Gao et al., 2011). This observation is in line with the finding that increased tRNA synthesis promotes proliferation and oncogenesis (Berns, 2008; Marshall et al., 2008), perhaps through an antagonistic mechanism to miRNA and siRNA silencing (Haussecker et al., 2010). Besides high-expressing miRNAs, another noteworthy class of miRNA changes is the miR-17-92 class and its two paralogous clusters, highly conserved between species and recently identified as oncomiRs (He et al., 2005; Mendell, 2008; Figure 2B and Table A6 in Appendix). The expression of the miR-17-92 family was downregulated immediately after RA and FSK treatment but subsequently restored with extended time points (Figure 2B). Interestingly, REST ablation prevented their initial downregulation, but instead led to their upregulation in the presence of differentiation stimuli.

Since concordantly regulated genes will cluster together with genome-wide profiling (Eisen et al., 1998), we reasoned there may be potential novel REST target miRNAs residing in the same cluster with known REST miRNA targets. To examine this, we used the CLUSTER software (Eisen et al., 1998) and performed unsupervised clustering analysis with all the miRNAs that have an expression value above 32 in the majority of the ten samples. We first validated the clustering strategy using the miR-297-669 cluster that contained dozens of miRNA genes under the same transcriptional regulation (Calabrese et al., 2007). As expected, regardless of the clustering algorithm used, miRNAs from this cluster readily grouped together (Figure A5 in Appendix), confirming the validity of our approach.

Next, we focused our analysis on clusters containing multiple known REST miRNA target genes. As shown in Figures 3A,B and Figure A6 in Appendix, two neighboring clusters contained at least nine known REST miRNA target genes together with dozens of other miRNAs. We used the TFBind software (Tsunoda and Takagi, 1999) to search for putative consensus RE1 site(s) flanking these miRNA genes. Together with a REST ChIP-sequencing (seq) study in our laboratory (unpublished results), we identified many potential REST target miRNAs within these two clusters (Figure 3C). In total, we have identified over 80 potential novel REST miRNA target genes through bioinformatics analysis and genome-wide profiling. The expanded list is shown in Table A7 in Appendix.

Finally, we randomly selected nine putative novel REST miRNA targets and performed ChIP-QPCR of cKO and control NSCs under proliferation conditions. Notably, for all of them except two, we were able to detect 2 fold or more enrichment in control cells compared to cKO NSCs (Figure 3C). For comparison, miR-129 and -149 were included as known miRNA target controls. In addition to REST, its corepressors CoREST and Sin3a are recruited to target promoter regions in a REST-dependent manner (Gao et al., 2011). We performed CoREST and Sin3a ChIP-QPCR analysis and found decreased recruitment of both CoREST and Sin3a (data not shown) in cKO NSCs (Figure 3D). Taken together, miRNA profiling expression data and ChIP-QPCR demonstrated we can identify known as well as novel REST-dependent miRNA targets in NSCs.