The molecular basis of astrocyte differentiation and maturation is poorly understood. As microRNAs have important roles in cell fate transitions, we set out to study their function during the glial progenitor cell (GPC) to astrocyte transition. Inducible deletion of all canonical microRNAs in GPCs in vitro led to a block in the differentiation to astrocytes. In an unbiased screen, the reintroduction of let‐7 and miR‐125 families of microRNAs rescued differentiation. Let‐7 and miR‐125 shared many targets and functioned in parallel to JAK‐STAT signaling, a known regulator of astrogliogenesis. While individual knockdown of shared targets did not rescue the differentiation phenotype in microRNA‐deficient GPCs, overexpression of these targets in wild‐type GPCs blocked differentiation. This finding supports the idea that microRNAs simultaneously suppress multiple mRNAs that inhibit differentiation. MicroRNA‐regulated transcripts exhibited concordant changes during in vivo differentiation and were enriched for a gene set upregulated in glioblastomas, consistent with validity of using the in vitro model to study in vivo events. These findings provide insight into the microRNAs and the genes they regulate in this important cell fate transition.
The coordinate action of multiple miRNAs is required to shape gene expression profiles in glial progenitor cells and allow differentiation to astrocytes, thereby underscoring the central role for miRNAs in cell fate transitions.
Loss of Dgcr8 and subsequent loss of miRNAs leads to separable defects in astrocyte differentiation and survival.
The differentiation defect is characterized in part by reduced JAK‐STAT signaling, which can be rescued with exogenous LIF.
Let‐7 and miR‐125 add‐back can rescue parts of differentiation, independent of JAK‐STAT signaling.
Let‐7 and miR‐125 have many shared targets in glial progenitor cells, some of which act as barriers to differentiation.
mRNAs regulated by let‐7 and miR‐125 are similarly modulated during astrocyte differentiation in vivo and are misexpressed in glioma stem cells.
Astrocytes are a dominant cell type in the central nervous system (CNS), intimately associated with neuronal synapses and CNS function (Zhang & Barres, 2010). They have come under renewed focus due to discoveries of their roles in driving evolutionary adaptations for central nervous system function and increased cognition in humans (Han et al, 2013; Xie et al, 2013). While the number of studies examining mature astrocyte function is growing, very little is known about the molecular mechanisms underlying their specification and differentiation from glial precursors during development.
During embryogenesis, neuroepithelial stem cells give rise to radial glial cells, which are initially restricted to a neurogenic fate. Near the end of embryogenesis and in early postnatal stages, the progeny of many of these cells become restricted to a glial progenitor fate and give rise to astrocytes and oligodendrocytes (Rowitch & Kriegstein, 2010). The switch from neurogenesis to astrogliogenesis is mediated by a number of mechanisms including activation of JAK‐STAT signaling and silencing of the neurogenic bHLH factors. Recent studies have also identified the transcription factors Sox9 and NF1A/B as being instructive for astrogliogenesis (Molofsky et al, 2012). In vitro, embryonic stem cells (ESCs) can be induced to follow the same sequential steps as seen in vivo providing a powerful tool to dissect the transcriptional and post‐transcriptional events underlying each step in astrocyte differentiation (Okabe et al, 1996; Brüstle & McKay, 1999; Krencik et al, 2011).
MicroRNAs (miRNAs) play a major role in post‐transcriptional regulation. They are small non‐coding RNAs that destabilize and inhibit translation of their target mRNAs (Krol et al, 2010). Once transcribed, canonical miRNAs are processed by a complex composed of the RNA binding protein Dgcr8 and RNaseIII enzyme Drosha, into hairpin structures called pre‐miRNAs. They are then exported to the cytoplasm and further processed by the enzyme Dicer into 18–21 nucleotide duplexes, one strand of which is loaded into a silencing complex consisting of Argonaute proteins. Once loaded into the silencing complex, miRNAs bind targets through partial base pairing with mRNAs through a 6–8 nucleotide seed sequence at the 5′ end of the miRNA. This minimal base‐pairing requirement enables a single miRNA to target hundreds of mRNAs. Additionally, miRNA targets form networks of genes that collaborate to promote or inhibit cell fate transitions (Judson et al, 2013). Therefore, identifying the major miRNAs and their downstream target gene network can provide novel insight into cell fate transitions.
In the neural lineage, detailed information on the function of miRNAs has been mostly limited to neuronal and oligodendrocyte differentiation (Krichevsky et al, 2009; Zhao et al, 2009, 2010b; Dugas et al, 2010). For example, it has been shown that distinct sets of miRNAs, miR‐9/124 in neurons and miR‐219/338 in oligodendrocytes, are highly upregulated during differentiation of progenitors and that these miRNAs play key roles in driving differentiation by inhibiting mRNAs that maintain the progenitor state. Two studies have suggested that miRNAs are required for production of astrocytes by deleting Dicer in spinal cord progenitors and in in vivo derived multipotent neural stem cells (Andersson et al, 2010; Zheng et al, 2010b). However, further characterization in these experimental settings is complicated by the inability to isolate large numbers of astrocyte progenitors. In particular, in vivo astrocyte progenitors are transient, non‐uniform in their spatial and temporal differentiation, and so far lack distinct markers to allow purification (Freeman, 2010).
In this study, we sought to overcome this roadblock by using an established in vitro counterpart to in vivo astrocyte differentiation (Brüstle & McKay, 1999). The in vitro method leads to the production of homogenous populations of glial progenitor cells (GPCs) that can be induced to differentiate in synchrony into astrocytes under defined culture conditions. Here, we characterize a differentiation phenotype associated with the loss of Dgcr8 and identify miRNAs and their target genes that in part underlie this phenotype. Additionally, we show that the miRNA targets are modulated during in vivo astrocyte development and during glioma progression. These studies provide a detailed characterization of how miRNAs promote the GPC to astrocyte transition, a critical aspect of the development of a functional nervous system.
Development of in vitro astrocyte differentiation system for the study of miRNA function
To study the role of microRNAs in glia progenitor cells, we took advantage of a conditional knockout model for Dgcr8, which encodes an essential cofactor in the biogenesis of all canonical miRNAs (Wang et al, 2007). An in vitro model was critical for evaluating miRNA function as astrocyte progenitors cannot be isolated in large numbers or as a homogenous population from in vivo tissue. Thus, a ubiquitously expressed tamoxifen‐inducible Cre recombinase was targeted into the Rosa26 locus in mouse embryonic stem cells (ESCs) that were hemizygous for the exon 3 floxed allele of Dgcr8 (R26 CreER; Dgcr8flox/Δ) (Supplementary Fig S1A). The floxed ESCs, which produce Dgcr8 and microRNAs at wild‐type levels, were differentiated into glial progenitor cells (GPCs) using an established neural differentiation protocol (Okabe et al, 1996; Brüstle & McKay, 1999; Wang et al, 2007) (Supplementary Fig S1B). The resulting neural progenitors were initially largely restricted to a neurogenic fate. With continued passaging in the presence of FGF and EGF, they lost neurogenic potential and became restricted to a glial progenitor fate (GPCs). Withdrawal of growth factors along with addition of either serum or BMP‐4 led to an almost exclusive differentiation to GFAP+ astrocyte‐like cells (Supplementary Fig S1C).
Microarray data on the in vitro derived GPCs and differentiated astrocytes were compared to previously published transcriptome data for in vivo purified mature astrocytes, oligodendrocytes and neurons (Lovatt et al, 2007; Cahoy et al, 2008). The in vitro derived cells expressed a majority of genes previously found to be enriched within in vivo purified astrocytes, and to a much lesser degree those expressed in oligodendrocytes and neurons (Supplementary Fig S1D). Many of these genes were expressed in both the GPC and astrocyte populations, although a number of genes further upregulated in the astrocytes, such as Aqp4 (Supplementary Fig S1E), a gene essential for mature in vivo astrocyte function (Xie et al, 2013). The absence of expression of some astrocyte genes in vitro likely reflects incomplete maturation, a common feature of adult cell types derived from ESCs in vitro (Nicholas et al, 2013). Importantly, transplantation of the glial progenitors gave rise to cells that had astrocyte morphology and expressed GFAP when assessed 6 weeks after injection (Supplementary Fig S1F). These results confirm that the in vitro model recapitulates many aspects of in vivo astrocyte differentiation.
Dgcr8 loss in ESC‐derived GPCs leads to separable differentiation and survival defects
To study the effect of conditional loss of miRNAs during the GPC to astrocyte transition, all experiments were done with R26 CreER;Dgcr8flox/Δ GPCs plus/minus tamoxifen after 6–10 passages in bFGF and EGF. Tamoxifen treatment of GPCs for 24–48 h led to 100‐fold reduction of Dgcr8 and approximately 50‐fold reduction of highly expressed miRNAs at 5–7 days post‐treatment consistent with efficient and stable loss of Dgcr8 in the majority of the cell population (Fig 1A and B).
As miRNA knockout GPCs (Dgcr8Δ/Δ) survived and continued to proliferate, we investigated the global requirement for miRNAs in serum‐induced differentiation of GPCs to astrocytes. The Dgcr8Δ/Δ GPCs showed a striking inability to upregulate GFAP (Fig 1C). In contrast, untreated Dgcr8flox/Δ cells and R26 CreER cells treated with tamoxifen (both serving as controls) differentiated normally. The small number of Dgcr8Δ/Δ cells that do differentiate likely represents cells that failed to excise exon 3 of the second allele of Dgcr8. During differentiation of Dgcr8Δ/Δ GPCs, we observed an increase in apoptosis at 8–10 h post‐differentiation, as quantified by flow cytometry using Annexin V and a vital dye (Supplementary Fig S2A). Consistent with the increase in Annexin V‐positive cells, Western blot analysis showed an increase in the apoptotic marker cleaved caspase‐3 (Fig 1D). Together, these results uncover an essential role for Dgcr8 in astrocyte differentiation and a partially penetrant role in cell survival.
The reduced survival raised the possibility that the differentiation defect in Dgcr8Δ/Δ GPCs may be secondary to apoptosis. To address this possibility, we took advantage of an apoptosis‐resistant Dgcr8 conditional model (Wang et al, 2013). R26 CreER; BakΔ/Δ; Baxf/Δ;Dgcr8f/Δ ESC‐derived GPCs were used, in which tamoxifen treatment led to a simultaneous loss of Bax and Dgcr8 (referred to as triple knockout TKO) and a complete absence of apoptosis during differentiation (Supplementary Fig S2B and C). Despite a lack of observable cell death, the knockout cells continued to show a striking defect in differentiation (Fig 1E). In contrast, untreated cells or tamoxifen‐treated cells carrying Dgcr8+/+ in the Baxf/f;BakΔ/Δ background differentiated normally (Fig 1E), consistent with previously published findings (Lindsten et al, 2003). Of note, the removal of Bax and Bak led to increased cell number in wild‐type as well as Dgcr8 knockout cells, showing that appreciable apoptosis occurs in the wild‐type context as well, although to a lesser degree than in the Dgcr8 knockout cells. Together, these data show that Dgcr8 is essential for GPC differentiation independent of its suppression of apoptosis.
Finally, we asked whether the defect in differentiation was associated with defective upregulation of astrocyte genes. Therefore, we performed RT‐qPCR on a panel of markers enriched in astrocytes during differentiation of TKO and control cells. The astrocyte markers Aqp4, ApoE, Clu, Atp1a2 and Cyp4f14 failed to be upregulated in the TKO in stark contrast to control cells (Fig 1F). However, AldoC and Pla2g7 were upregulated similar to control cells. Therefore, Dgcr8 is required for the upregulation of many, but not all, genes associated with astrocyte differentiation.
Induction of JAK‐STAT pathway rescues Dgcr8Δ/Δ differentiation defect
Normal astrogliogenesis requires activation of BMP signaling and JAK‐STAT signaling (Bonni et al, 1997; Fan et al, 2005; He et al, 2005). For a number of astrocyte genes such as GFAP, phosphorylated STAT3 forms a complex with SMAD and p300 and binds to their promoter to activate transcription (Fukuda et al, 2007). In cultures of in vivo derived cortical neural stem cells, phosphorylation of STAT3 at tyrosine 705 increases coincident with astrocyte differentiation of cortical neural stem cells and inhibition of STAT3 abrogates astrocyte differentiation.
To determine whether BMP or JAK‐STAT signaling is disrupted upon miRNA loss, we examined the downstream mediators of the two pathways, phosphorylated SMAD and STAT3 (pSMAD and pSTAT3), in control and Dgcr8Δ/Δ GPCs during differentiation. The addition of BMP‐4 robustly activated pSMAD in control Dgcr8f/Δ GPCs. The activation of pSMAD was unaffected in Dgcr8Δ/Δ GPCs under the same conditions (Fig 2A). In contrast to pSMAD, pSTAT3 was significantly reduced in Dgcr8Δ/Δ cells during both basal growth and differentiation conditions (Fig 2B and Supplementary Fig S3A). Approximately 6–12 h following the shift to differentiation conditions, pSTAT3 in Dgcr8Δ/Δ cells showed an increase in levels but still only reaching basal levels of Dgcr8f/Δ GPCs. Dgcr8f/Δ showed a three‐ to fourfold increase under the same conditions. Total STAT3 levels were unchanged. Similar results were seen in the Bak/Bax/Dgcr8 TKO background showing that this defect in STAT3 phosphorylation occurs independently of apoptosis (Fig 2B).
To further test the role of STAT3 signaling in the differentiation defect, we introduced activators and inhibitors of the pathway. First, to confirm that JAK‐STAT signaling was necessary for astrocyte differentiation in ESC‐derived GPCs, a small molecule inhibitor targeting JAK (JAKi) was added during differentiation of wild‐type cells. JAK inhibition abrogated the expression of GFAP during differentiation (Supplementary Fig S3B). Next, to activate the JAK‐STAT pathway in Dgcr8Δ/Δ cells, leukemia inhibitory factor (LIF), a canonical activator, was introduced along with FBS to increase STAT3 phosphorylation (Supplementary Fig S3C). Addition of LIF led to a consistent increase in the number of GFAP‐positive cells in both Dgcr8Δ/Δ and TKO backgrounds (Fig 2C). Together, these data suggest that a decrease in JAK‐STAT signaling is at least partially responsible for the defect in astrocyte differentiation seen upon Dgcr8 loss.
Let‐7 and miR‐125 families of miRNAs are highly expressed in GPCs and rescue the Dgcr8Δ/Δ differentiation defect
As the predominant role of Dgcr8 is in miRNA biogenesis (Preiss, 2009; Barad et al, 2012; Seong et al, 2014), we next sought to identify specific miRNAs that may underlie the differentiation defect of Dgcr8Δ/Δ cells by simultaneously profiling and functionally screening miRNAs. In a number of lineages, miRNAs that promote differentiation increase upon induction of differentiation and promote downregulation of the progenitor transcriptional program (Yoo et al, 2009; Dugas et al, 2010; Goljanek‐Whysall et al, 2011). Therefore, we performed small RNA sequencing of GPCs in proliferation conditions and following 48 h of serum differentiation. Surprisingly, levels of the most abundant miRNAs were largely unchanged following 48 h of differentiation, even though the cells express GFAP and other markers of differentiated astrocytes as well as develop characteristic astrocyte morphology by this time point (Fig 3A, Supplementary Figs S1E and S4). This finding suggested that miRNAs already expressed in the proliferating progenitors are required for the production of astrocytes.
To uncover the impact of miRNA loss on the transcriptome of GPCs, we profiled mRNAs in Dgcr8Δ/Δ versus control Dgcr8f/Δ GPCs in proliferation conditions. As has been previously shown in other cell types, the global loss of miRNAs led to hundreds of misregulated mRNAs (Fig 3B) (Sinkkonen et al, 2008; Yi et al, 2009; Gurtan et al, 2013). To determine whether these mRNAs represent targets of expressed miRNAs, the downregulated and upregulated gene sets were evaluated for enriched 7‐mer motifs in an unbiased fashion. Among upregulated genes, 104 enriched motifs were uncovered (P < 0.01). Many of the most highly enriched motifs perfectly matched the seed region of four of the most highly expressed miRNA families including let‐7, miR‐9, miR‐181 and miR‐30, consistent with these miRNAs having a major effect on the GPC transcriptome (P < 2.2e−16, Fig 3C, Supplementary Table S1). These data show that Dgcr8 loss leads to misregulation of many genes, a large fraction of which can be directly linked to expressed miRNAs.
Previous studies have shown that the addition of a single miRNA is sufficient to rescue individual phenotypes seen with global miRNA loss (Giraldez et al, 2005; Wang et al, 2008; Dugas et al, 2010; Melton et al, 2010; Zhao et al, 2010b; Steiner et al, 2011). We therefore performed a genome‐wide screen for individual miRNAs that could rescue the capacity of the cells to differentiate to GFAP+ cells. Specifically, 570 chemically synthesized miRNA mimics were added to individual wells of proliferating Dgcr8Δ/Δ GPCs, which were then transferred to differentiating conditions with serum (Fig 4A). Using the criteria outlined in Materials and Methods, we identified let‐7 and miR‐125 families of miRNAs as positive hits (Fig 4B). Validation experiments with independently synthesized mimics confirmed let‐7 and miR‐125 as positive hits, showing a striking rescue of GFAP expression in Dgcr8Δ/Δ GPCs in differentiation media (Fig 4C, Supplementary Fig S5E). The let‐7 and miR‐125 miRNA families have highly distinct seed sequences suggesting potential interdependent effects. Indeed, the simultaneous introduction of let‐7 and miR‐125 mimics led to a synergistic increase in the number of GFAP‐positive Dgcr8Δ/Δ cells, suggesting that they cooperate to promote astrocyte differentiation (Fig 4D). Rescue of GFAP was also seen with introduction of let‐7 and miR‐125 into Bax/Bak/Dgcr8 TKO GPCs (Supplementary Fig S5A). Additional markers confirmed rescue of astrocyte differentiation following let‐7 or miR‐125 transfection including quantitative RT‐qPCR for GFAP, Aqp4, Clu and ApoE (Fig 4E). However, there was no detectable increase in Atp1a2 and Cyp4f14. The lack of complete rescue of levels may reflect lower than complete transfection efficiency or partial rescue by the microRNAs. Levels of AldoC and Pla2g7 were up as seen even in the absence of miRNAs in Fig 1E. Importantly, cells transfected with let‐7 and miR‐125 did not upregulate markers of the oligodendrocyte (MBP) or neuronal (Tuj1) fate (Supplementary Fig S5B). Rescued cells also expressed the astrocyte marker vimentin (Supplementary Fig S5C). Together, these data show that let‐7 and miR‐125 rescue many aspects of astrocyte differentiation in otherwise miRNA‐deficient cells.
Let‐7 and miR‐125 are highly expressed in both GPCs and astrocytes suggesting that they enable rather than induce astrocyte differentiation (Fig 3A). That is, they suppress mRNAs that encode proteins that when overexpressed in GPCs block differentiation. If true, these miRNAs should not be sufficient to induce differentiation. To test this directly, let‐7 and miR‐125 mimics were introduced into Dgcr8Δ/Δ GPCs that were maintained in proliferation conditions. Under these conditions, Dgcr8Δ/Δ GPCs remained undifferentiated (Supplementary Fig S5D), indicating that these miRNAs are not sufficient to induce differentiation. From these data, we conclude that let‐7 and miR‐125 repress targets in Dgcr8Δ/Δ GPCs that otherwise form a barrier to astrogliogenesis.
Identification and functional testing of mRNA targets of let‐7 and miR‐125
Next, we aimed to uncover the pathways and mRNAs regulated by these two miRNAs. Initially, we took a candidate approach and evaluated the JAK‐STAT pathway based on our finding that the STAT3 signaling was reduced in Dgcr8Δ/Δ under both proliferation and differentiation culture conditions and that introduction of exogenous LIF could at least partially rescue differentiation (Fig 2). First, we asked whether the miRNAs could increase the levels of pSTAT3 in differentiation conditions, placing the miRNAs upstream of STAT3 phosphorylation (Fig 4F). However, the introduction of let‐7 and miR‐125 failed to show a significant increase in pSTAT3. Next, to confirm that let‐7 and miR‐125 did not bypass JAK‐STAT signaling when added to Dgcr8Δ/Δ GPCs, we asked whether let‐7 and miR‐125 could rescue differentiation even in the absence of JAK‐STAT signaling (Supplementary Fig S6). The introduction of JAKi to the differentiation media along with transfection of let‐7 and miR‐125 completely blocked the ability of the miRNAs to rescue GFAP expression in Dgcr8Δ/Δ cells. These data suggest that there are two alternative means of rescuing Dgcr8Δ/Δ GPC differentiation: let‐7/miR‐125 introduction or activation of JAK‐STAT by exogenous LIF. Therefore, the let‐7/miR‐125 miRNAs and JAK‐STAT appear to act in parallel but interdependent pathways.
Next, we took the direct approach of identifying targets of let‐7 and miR‐125 in GPCs. While these two miRNAs have distinct seeds, they have been predicted to have a statistically significant overlap in targets, suggesting the ability of each of these miRNAs to rescue differentiation could be through the downregulation of a common set of mRNAs (Tsang et al, 2010). To identify direct and indirect targets of let‐7 and miR‐125 in GPCs, microarray analysis was performed on RNA isolated 24 h following transfection of Dgcr8Δ/Δ GPCs with let‐7 and miR‐125. Passage‐matched mock‐transfected Dgcr8Δ/Δ GPCs were used as control. As expected, hundreds of genes were up‐ and downregulated with an adjusted P‐value cutoff of 0.01 (Fig 5A). To distinguish direct and indirect targets, the mRNAs were evaluated for sequences complementary to the seed sequence of the two miRNAs (Supplementary Fig S7A). There was a strong enrichment for the seed matches for both miRNAs within the open reading frame (ORF) and 3′ untranslated region (3′UTR) of downregulated genes (Fig 5B). In contrast, there was a depletion of seed matches in the upregulated genes. To further validate the target identification, we examined the levels of a number of known let‐7 targets (Supplementary Fig S7B and C). A number of these genes that have previously been shown to be important in other contexts including pluripotency and earlier neural development (N‐Myc, Hmga2, LIN28, Igf2bp1) were undetected in Dgcr8Δ/Δ GPCs (detection P‐value < 0.5) (Supplementary Fig S7B) (Melton et al, 2010; Nishino et al, 2013; Patterson et al, 2014). However, other known targets identified as being part of a mid‐gestation network of genes silenced by let‐7 were expressed in GPCs, upregulated in Dgcr8Δ/Δ cells and downregulated upon the addition of let‐7 (Supplementary Fig S7C) (Gurtan et al, 2013). Therefore, the combination of downregulation upon miRNA introduction and seed matches in the 3′UTR/ORF uncovers a set of targets within the GPC context. Furthermore, many of the gene changes in the Dgcr8Δ/Δ GPCs could be reversed by simple addition of either let‐7 or miR‐125, consistent with these two miRNAs representing a large fraction of the miRNA pool in these cells (Figs 5C and 3A).
Due to the ability of let‐7 and miR‐125 to cooperate in rescuing differentiation in Dgcr8Δ/Δ cells, we sought to determine whether the mRNAs downregulated by the two miRNAs overlap greater than expected at random for expressed genes (Supplementary Fig S7D). Indeed, the overlap was highly significant suggesting that these miRNAs affect many common downstream genes and pathways.
To identify targets of let‐7 and miR‐125 whose upregulation might block GPC differentiation, we knocked down predicted targets in the Dgcr8Δ/Δ background and transferred cells to differentiation conditions. We selected candidate target genes by the following criteria: (i) downregulated with addition of let‐7 and miR‐125, (ii) contain a 3′UTR seed match for let‐7 or 125, (iii) upregulated in Dgcr8Δ/Δ cells relative to control and (iv) unchanged or decreased during differentiation. This resulted in a list of 62 mRNAs (Supplementary Table S2). Cells were plated in 96‐well format, and each well was transfected with an siRNA targeting each of the 62 mRNAs and placed in differentiation conditions the following day (Fig 6A). Surprisingly, in stark contrast to let‐7 and miR‐125, not a single siRNA pool to either the target set or a control set increased GFAP staining in Dgcr8Δ/Δ cells (Fig 6B). This finding suggests that either no single target can recapitulate the effects of the miRNAs or that the critical target did not fulfill the criteria for selection described above.
While knockdown of any one of these targets failed to rescue GFAP expression in Dgcr8Δ/Δ cells, it remained possible that the increased expression of any of these targets in knockout cells prevents normal differentiation of these cells. To test this hypothesis, we overexpressed candidate targets in wild‐type GPCs and assessed differentiation capacity. From twenty‐seven mRNAs co‐targeted by both miRNAs (i.e., requiring predicted target sites for both miRNAs) (Supplementary Table S3), four were selected for overexpression. cDNAs encoding these proteins were cloned into expression constructs co‐expressing mCherry. Luciferase was introduced as a negative control. Overexpression of two of the four targets, Plagl2 and Igf2bp2, robustly inhibited the number of GFAP‐positive cells, while control, Ddx19b and Cbx2 overexpression did not affect differentiation (Fig 6C and Supplementary Fig S8A). These results combined with the results of individual target siRNA knockdown are consistent with a model where miRNAs target many mRNAs to block differentiation. However, we cannot exclude the possibility that a single untested target or combination of a small number of targets underlies the miRNA requirement.
mRNAs regulated by let‐7 and miR‐125 show consistent changes during in vivo astrocyte maturation
To assess the relevance of data from ESC‐derived GPCs to in vivo astrocyte development, we overlapped let‐7 and miR‐125‐regulated genes with profiling data of maturing astrocytes isolated from mouse spinal cords (Molofsky et al, 2013). Globally, mRNAs upregulated by addition of let‐7 and miR‐125 to Dgcr8Δ/Δ GPCs were upregulated during normal astrocyte maturation (Fig 6D and Supplementary Fig S8B). Conversely, mRNAs downregulated by let‐7 and miR‐125 were downregulated by addition of let‐7 and miR‐125 during astrocyte maturation. Together, these data support a role for let‐7 and miR‐125 modulating a molecular program consistent with astrocyte development in vivo.
It has been proposed that dedifferentiation of astrocytes via oncogene overexpression can lead to the formation of malignant gliomas (Friedmann‐Morvinski et al, 2012). Since let‐7 and miR‐125 promote astrogliogenesis, we predicted that their targets may be misregulated in glioma, a tumor type associated with astrocyte dysfunction. Indeed, Plagl2 was previously identified as a negative regulator of astrocyte differentiation and is commonly amplified in glioma (Zheng et al, 2010a). To determine whether other mRNAs downregulated by let‐7 and miR‐125 in Dgcr8Δ/Δ cells are upregulated in gliomas, we overlapped let‐7/miR‐125‐regulated genes with a previously identified 30‐gene molecular signature of glioma stem cells (Sandberg et al, 2013) that is positively correlated with tumor severity. mRNAs downregulated by both miRNAs were strikingly enriched for the glioma stem cell molecular signature (Fig 6E). The highest enrichment was for mRNAs downregulated by both miRNAs, further providing support to the notion that let‐7 and miR‐125 co‐target mRNAs inhibitory to astrocyte differentiation.
Here, we have shown that Dgcr8 is essential for astrocyte differentiation in vitro, consistent with in vivo Dicer knockout‐derived models (Andersson et al, 2010). Unlike past work, our in vitro system enabled discovery of specific miRNAs, pathways and targets that underlie major components of the differentiation defect. Let‐7 and miR‐125 rescued differentiation of Dgcr8Δ/Δ GPCs into cells with astrocyte‐like morphology and expression of multiple mature astrocyte markers. Similarly, activation of the otherwise repressed JAK‐STAT pathway was sufficient to rescue differentiation of the knockout GPCs to GFAP‐positive cells. However, let‐7 and miR‐125 did not result in increased STAT3 phosphorylation placing these miRNAs in a parallel pathway to JAK‐STAT signaling. The two miRNAs suppressed hundreds of targets in the GPCs, presumably priming them to respond to differentiation cues. Overexpression of a subset of individual targets in wild‐type cells blocked astrocyte differentiation, but knockdown of any single candidate target in Dgcr8Δ/Δ cells was insufficient to recapitulate the effect of let‐7 and miR‐125. Therefore, let‐7 and miR‐125 appear to be functioning through numerous targets that otherwise cooperatively block differentiation. Interestingly, the genes directly and indirectly modulated by these miRNAs are similarly regulated during in vivo astrocyte maturation. Furthermore, mRNAs downregulated by let‐7 and miR‐125 are enriched for a gene signature correlated with glioma progression, a cancer associated with defects in astrocyte differentiation. Therefore, the in vitro findings presented here are clearly relevant to in vivo astrocyte differentiation, an area of research that has been relatively intractable due to the difficulty of isolating and manipulating GPCs within the organism.
This work provides insight into the role of specific miRNAs and their target genes in astrogliogenesis. The difficulty in isolating astrocyte progenitors in vivo has left a large gap in knowledge about astrocyte development, despite the dramatic rise in studies showing novel roles for astrocytes in CNS function and evolution (Han et al, 2013; Xie et al, 2013; Molofsky et al, 2014). Mechanisms underlying early astrocyte commitment of a gliogenic progenitor are largely unknown (Freeman, 2010). As miRNAs have evolved to target cooperative gene networks, it is possible to identify novel components of such networks by analyzing the targets of a given miRNA. In this study, we identify two miRNA families whose shared targets provide a rich candidate set of regulators of astrogliogenesis. Indeed, two out of four genes (Igf2bp2 and Plagl2) tested that are co‐targeted by let‐7 and miR‐125 inhibit astrogliogenesis.
Previous studies in neural development have suggested diverse and age‐dependent roles for let‐7. Let‐7 plays a role in promoting the switch between neurogenic and gliogenic potential of neural stem cells during fetal development by repressing Igf2bp1, Hmga2 and LIN28 (Nishino et al, 2013; Patterson et al, 2014). These mRNAs are lowly expressed in GPCs and not derepressed upon loss of Dgcr8. This suggests that during the neurogenic to gliogenic transition, increasing levels of let‐7 may provide robustness to the regulatory mechanisms that lead to post‐transcriptional silencing of these genes. As glial fate is specified and these targets are epigenetically silenced, let‐7 targets gene networks inhibitory to terminal astrocyte differentiation. Consistent with let‐7 targeting of the Igf2bp family members, two other family members of the Igf2bp family, Igf2bp2 and Igf2bp3, are actively repressed by let‐7 and miR‐125 in GPCs. In adult neural stem cells, one study has suggested that let‐7 promotes both increased neurogenesis and gliogenesis via suppression of the stemness factor Tlx in neural stem cells (Zhao et al, 2010a). The fact that Tlx is lowly expressed and is unchanged with addition of let‐7 in the context of Dgcr8 KO GPCs suggests context dependence of let‐7 targets. Taken together, these studies suggest evolving roles for let‐7 during gliogenesis.
Our study highlights how stem/progenitor cells utilize multiple, parallel and redundant pathways to ensure robust differentiation. First, characterizing shared targets of functionally overlapping miRNAs increases the likelihood of identifying a network of genes involved in a critical regulatory pathway. Let‐7 and miR‐125 are orthologs of let‐7 and lin‐4, heterochronic miRNAs in Caenorhabditis elegans. In C. elegans, these are among the small number of miRNAs that lead to a phenotype of delayed differentiation when deleted (Miska et al, 2007). While the two miRNAs are a bioinformatically predicted co‐targeting pair, to date only a small number of common targets have been confirmed experimentally (Maller Schulman et al, 2008; La Torre et al, 2013). Two members of the families, let‐7i and miR‐125b, are co‐transcribed, further suggesting a functional link. Here, we identify the largest set of mRNAs that are targeted by both miRNAs. As let‐7 and miR‐125 function through distinct seed sequences, this high degree of target overlap indicates that some evolutionary redundancy exists not only within miRNA families but also between miRNA families. Why such a large degree of redundancy exists remains to be determined. Further, as let‐7 and miR‐125 are broadly expressed, it remains to be determined whether this redundancy as well as their target gene networks is conserved across cell types.
We demonstrate that let‐7 and miR‐125 additionally share overlapping function with the JAK‐STAT pathway canonically associated with astrogliogenesis. The activation of JAK‐STAT or introduction of let‐7/miR‐125 could induce GFAP expression during differentiation of Dgcr8Δ/Δ GPCs. Multiple factors affect the transcriptional activity at the GFAP locus during differentiation, providing a means for multiple regulatory mechanisms to converge to ensure robust GFAP activation (Cheng et al, 2011). It will be interesting to determine in future studies how the JAK‐STAT pathway and miRNAs converge to promote astrogliogenesis.
The dysfunction of astrocytes has been linked to the development of gliomas, including the malignant glioblastoma multiforme (GBM) (Friedmann‐Morvinski et al, 2012). Indeed, the miRNAs and a number of targets that we identified as promoting astrogliogenesis are misregulated in gliomas. For example, Plagl2 is amplified in gliomas and inhibits differentiation of tumor‐initiated cell lines (Zheng et al, 2010a). Additionally, in a number of cancers including gliomas, LIN28 has been identified as an oncogene, suggesting a misregulation of let‐7 (Mao et al, 2013; Zhou et al, 2013; Qin et al, 2014). In cell lines derived from GBMs, overexpression of let‐7 and miR‐125 inhibits proliferation (Lee et al, 2010; Wu et al, 2012). The ability of these miRNAs to suppress proliferation in glioma cells may be in part associated with promoting differentiation.
In summary, the findings presented here demonstrate how two families of miRNAs can downregulate a common network of targets and cooperate with signaling pathways to prime a progenitor cell for differentiation. Importantly, these results provide a window into the molecular mechanisms that underlie the elusive GPC to astrocyte cell fate transition. Future work will need to bring together the multiple regulatory mechanisms driving this cell fate transition including signaling, transcriptional/epigenetic regulation, post‐transcriptional and post‐translation regulation. The in vitro GPC model provides an important entryway into these questions, which can then be further evaluated using in vivo models.
Materials and Methods
Cell culture and GPC derivation
ES cell culture has been previously described (Wang et al, 2007). Briefly, ES cells were maintained in media supplemented daily with 1,000 units/ml LIF. For neural differentiation, the protocol was adapted from a previously described method (Okabe et al, 1996; Brüstle & McKay, 1999). Briefly, ES cells were dissociated and grown as embryoid bodies (EBs) culture in media without LIF. EBs were plated in ITSF media as described by Okabe et al for 5–8 days and then dissociated into N2 media supplemented with 10 ng/ml FGF on to laminin‐coated plates. Following 1–2 passages, 10 ng/ml EGF was added to the media along with FGF to stimulate glial specification. For differentiation, cells were grown to 70–80% confluency unless described otherwise. Growth factors were withdrawn, and N2 media supplemented with 1% FBS and B27 was added for 24–48 h.
To induce Dgcr8 loss, GPCs were treated with 1 μM tamoxifen for 16–24 h and, 2 days later, treated with 200 nM tamoxifen for 16–24 h to stimulate efficient Cre‐lox‐mediated recombination. To minimize cell death following loss of Dgcr8, cells were supplemented daily with growth factors until inducing differentiation.
For experiments with inhibitors, JAKi (AG490; Tocris Biosciences) was resuspended in DMSO and used at a final concentration of 1 μM. LIF was used at a concentration of 2,500–5,000 units/ml of media.
A library of 570 miRNA mimics described previously was transfected into Dgcr8 KO GPCs plated in 96‐well format in proliferation conditions (+EGF/FGF) (Wang et al, 2008). All mimics were transfected at a final concentration of 50 nM using DharmaFECT3. Differentiation was induced 3 days post‐transfection with low serum addition and growth factor withdrawal. Following 48 h in differentiation conditions, plates were fixed and processed for GFAP immunofluorescence (anti‐GFAP; Dako Cytomation Z0334). A DAPI counterstain was performed. All plates were imaged at 10× magnification using a high‐throughput high‐content fluorescent microscope system (InCell Analyzer, GE). The InCell Workstation software was used for analysis using the multitarget analysis module to count number of total cells (DAPI) and percent of GFAP+ cells. Hits identified by this system were then validated by eye. Significance of change in GFAP staining was calculated using strictly standardized mean difference (SSMD) score. Criteria for positive hits chosen for follow‐up studies were (i) SSMD score > 2.5, (ii) rescue by multiple members of a miRNA family and (iii) expressed in GPCs or astrocytes.
Cells were passaged for flow cytometry and resuspended at a concentration of 150,000 cells/500 μl in Annexin‐binding buffer. Annexin‐APC (Invitrogen A35110, 1:500) and Sytox Blue (1:1,000) were added. Cells were analyzed on a BD LSRII instrument. Cells double‐positive for Annexin and Sytox Blue were scored as apoptotic.
miRNA mimics were ordered from Dharmacon, resuspended in sterile H2O and transfected using the DharmaFECT3 reagent at a final concentration of 50–100 nM. Media was changed 24 h later and differentiation was induced 72 h following transfection. For the miRNA target knockdown screen, Dharmacon siGenome smartpool siRNAs were used at a final concentration of 50 nM.
All animal experiments described here were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Francisco.
For overexpression of miRNA target genes, cDNAs were cloned into a vector driven by EF1alpha and contained a T2A‐mCherry element downstream of the cloned insert. Nucleofection was performed using the Glial cell kit with the ‘A033’ program to express the plasmid in GPCs. Differentiation was induced 24 h following nucleofection and cells were fixed 30–40 h later and processed for immunofluorescence.
For immunofluorescence, cells were fixed in 4% PFA for 10–15 min, permeabilized with PBS containing 0.2% Triton‐X100 and blocked in PBS containing 2% BSA and 1% goat serum. Primary and secondary antibody incubation was done in blocking solution, and cells were counterstained with DAPI. Secondary antibodies (Alexa Fluor) were used at a dilution of 1:500. Rat mCherry antibody was used at a dilution of 1:500 (Clontech).
RNA for all microarray samples was extracted from cells following lysis with Trizol (Invitrogen) and processed for Illumina bead chip arrays at the UCLA genome core facility. Microarrays of control and Dgcr8 knockout GPCs (mock transfected) were performed 7 days following tamoxifen treatment. Quantile normalization was performed using the R/Bioconductor package beadarray, and differential expression analysis was performed using the R package Limma (Dunning et al, 2007). For discovery of let‐7 and miR‐125 targets, microarrays were performed 24 h post‐transfection with let‐7 and miR‐125. Arrays were quantile normalized using the beadarray package. As one pair of mock and let‐7 arrays was performed at a separate time, differential expression analysis was performed using the RankProd R package. Seed enrichment analysis was performed as previously described using custom python scripts and R code (Melton et al, 2010).
For analysis of Affymetrix data from Molofsky et al, microarrays were first normalized by RMA method using the ‘Affy’ R package. The fold change between e13 and e18 embryonic day astrocyte samples was used for further analysis. Triplicate array experiments were averaged. To overlap in vitro (Illumina) and the in vivo (Affymetrix) data, all probes that crossed a detection P‐value of 0.05 in the Illumina array were overlapped by gene name in the Affymetrix array. For multiple probes with the same gene name, probes exhibiting the highest change were used.
Motif analysis was performed by matching and extracting 3p‐UTR sequences from mm9 RefSeq mRNA transcripts for each Illumina probe ID. Redundant probes were removed so that no two probes mapped to a shared RefSeq transcript. For probes matching to multiple transcripts, the matched UTR with the largest length was retained. Transcripts were divided into experimental and control groups, and within each group, all UTRs were scanned and counts of every observed 7‐mer motif were generated. For each unique 7‐mer motif, the probability of observing k or more counts in the experiment was determined using Fisher's exact test.
RNA for all qPCR experiments was extracted using Trizol and quantified using a nanodrop spectrophotometer. One microgram of RNA was DNase‐treated and reverse‐transcribed using random hexamer primers and the Superscript III first strand synthesis kit (Invitrogen). qPCRs were performed on an ABI 7900 system. RT–qPCR primer sequences are listed below.
For protein extraction, cells were lysed in first lysed in weak lysis buffer (low NP40 lysis buffer—12.5 mM Tris pH 7.5, or 7.9/150 mM NaCl, 100 μM EDTA, 10% glycerol, 0.1% NP‐40) containing protease inhibitors (Roche) and phosphatase inhibitors (PhosphSTOP; Roche), followed by 3 freeze–thaw cycles in liquid nitrogen. The supernatant was collected following a high speed spin at 12,000 g and used for Western blots. Proteins were quantified using a Bradford assay with BSA standards. Approximately 20–25 μg of protein was used per lane. Antibodies were used at the following concentrations: pSTAT3 (Cell Signaling # 9145)—1:1,000, total STAT3 (Cell Signaling #9139), pSMAD1/5/8 (Cell Signaling #9511) and actin (Sigma, A4700). Primary antibody incubation was performed at room temperature for 2 h or overnight at 4°C. Secondary antibodies (Thermofisher) were used at a concentration of 1:10,000. Imaging was performed using an Odyssey LICOR scanner. Quantification was performed using ImageJ software.
An initial Western blot for pSTAT3 in Dgcr8f/Δ cells showed levels increase significantly with a latency of > 1 h and are apparent at 5 h following the addition of BMP‐4 (data not shown). Thus, a time point of 5 h or later was used for all subsequent experiments to ensure observable activation.
Small RNA sequencing and analysis
Total RNA was extracted using Trizol Reagent (Invitrogen). Small RNAs were cloned and multiplexed using the Illumina Tru‐Seq kit protocol and in house reagents. 3′ adapter was adenylated prior to small RNA cloning as described previously (Babiarz et al, 2008). All adapters and PCR primers were purified on a 22% urea‐acrylamide gel prior to use. Libraries were diluted to 10 nM and sequenced on a HiSeq 2000 at UCSF.
Small RNA sequencing data were processed using custom perl scripts and mapped to miRNAs using Bowtie (‐p4 ‐f ‐v0 ‐k5 ‐m5 options) (Langmead et al, 2009). Sequences were trimmed of the 3′ adapter by requiring 8 nt of perfect match to the adapter sequence. Sequences were mapped to the mouse pre‐miRNA hairpins in miRBase v20.0. All reads perfectly mapping no more than to five hairpins were included in counts. Reads mapping to multiple miRNAs were divided by the number of mappings and then assigned to individual miRNAs. Read counts for each miRNA seed family were summarized and reported as a percentage of total miRNA reads. Targetscan miRNA families were used to group miRNAs into seed families.
All microarray data and small RNA sequencing data have been submitted to the GEO repository. Microarray data are available under Accession Number GSE64637 and small RNA sequencing data under GSE64661.
RHB and AS conceived the project, designed experiments and wrote the manuscript. AS contributed to experiments in all figures. MD contributed to experiments in Figs 1 and 2 and Supplementary Figs S2 and S3.
Conflict of interest
The authors declare that they have no conflict of interest.
Source Data for Supplementary Figure S2
Source Data for Supplementary Figure S3
We would like to thank members of the Blelloch laboratory, especially Amy Chen, Jake Freimer, Raga Krishnakumar, Ronald Parchem, Mayya Shveygert and John Vincent for critical reading of the manuscript. We would like to thank Collin Melton for assistance with statistics and bioinformatic analysis, and Nicole Moore, Jason Liu and Glencijoy David for technical assistance. We would like to thank Luis C Fuentealba (Alvarez‐Buylla laboratory) for training and advice on transplantation experiments. We would like to thank Justin Ellis (Rowitch laboratory) for brain sections and antibodies. We would like to thank Fong Ming Koh (Ramalho‐Santos lab) for overexpression constructs. This work was funded by a CIRM training grant TG2‐01153 (A.S.) and grants to R.H.B. from the US National Institutes of Health (NINDS R01 NS057221 and NIGMS R01 GM101180) and the California Institute of Regenerative Medicine (New Faculty Award, RN2‐00906‐1).
FundingCIRM training grant TG2‐01153
- © 2015 The Authors