Myc and SAGA rewire an alternative splicing network during early somatic cell reprogramming

A functional RNAi screen for epigenetic regulators of reprogramming initiation

We previously reported that cellular reprogramming is accompanied by a phased series of gene expression changes (Samavarchi-Tehrani et al. 2010). However, little is known about how epigenetic regulatory pathways initiate massive reorganization of the chromatin landscape that is required for the broad transcriptional alterations that underlie changes in cellular plasticity associated with reprogramming. To identify epigenetic regulators that function in the earliest stages of reprogramming, we performed a systematic RNAi screen during the initiation phase of reprogramming using a secondary MEF model. Our RNAi library contained all known histone-modifying enzymes, chromatin remodelers, histone chaperones, enzymes associated with DNA methylation, epigenetic readers, and additional components of epigenetic modifying complexes as well as family members closely related to the above (652 siRNAs) (Supplemental Fig. S1A). Control siRNAs _targeting Oct4, Sox2, Klf4, Myc, Nanog, and Smad1 were also utilized. For screening, secondary MEFs were transfected with siRNA 1 d prior to OSKM transgene induction with Dox. After 5 d, the cells were then fixed; stained for alkaline phosphatase (AP) activity, which is an early marker of pluripotency; counterstained with DAPI; and imaged by automated image analysis that quantified reprogramming based on AP and DAPI colony costaining (Fig. 1A). As expected, individual knockdown of each of the OSKM transgenes greatly impaired reprogramming, with the most significant reductions observed upon loss of Oct4, Sox2, or Myc (Fig. 1B). We performed two biological replicate screens of our epigenetic factor library and rank-ordered the average of the median center AP and DAPI colony areas (median rank variability ±5.1%) (Fig. 1C; Supplemental Fig. S1B). Given that the screen was tailored specifically to epigenetic regulators, we found that knockdown of many factors impacted reprogramming to some degree. Therefore, we set a stringent cutoff that classified only the lowest 15% from AP and DAPI colony staining (76 _targets) as key factors required for reprogramming, which is similar to the effect of Oct4 knockdown (Supplemental Fig. S1C; Supplemental Table S1). This stringently defined list concentrates factors that were previously identified as facilitators of reprogramming, such as Ezh2, Suz12, Wdr5, Sirt6, and Prmt5 (Ang et al. 2011; Nagamatsu et al. 2011; Onder et al. 2012; Sharma et al. 2013; Ding et al. 2014), thus validating our screen. However, it also included many factors that have not been previously implicated in reprogramming.

We focused our attention initially on histone-modifying enzymes with previously undefined roles in somatic cell reprogramming. Of particular note, loss of the HAT Gcn5 (encoded by Kat2a) greatly inhibited colony formation, while knockdown of the closely related enzyme Pcaf (encoded by Kat2b) had little effect (Fig. 1C; Supplemental Fig. S1B). Furthermore, loss of Tip60 (encoded by Kat5), which plays a role in ESC maintenance (Fazzio et al. 2008), or other HATs (Kat6a, Kat6b, Kat7, Kat8, or p300) had no appreciable effect on reprogramming (Fig. 1C; Supplemental Fig. S1B, blue), indicating a specific requirement for Gcn5 in this process. Gcn5 was the first transcription-related HAT enzyme identified (Brownell and Allis 1995; Brownell et al. 1996) and functions in the context of multisubunit complexes, including SAGA and ATAC (Grant et al. 1997; Martinez et al. 1998; Guelman et al. 2006, 2009; Koutelou et al. 2010). Importantly, consistent with a role for Gcn5 in reprogramming, our screen identified three other components of the SAGA complex—Trrap, Ccdc101, and Taf12—as regulators of reprogramming (Fig. 1C; Supplemental Fig. S1B, orange).

To assess whether reduced colony formation upon loss of the SAGA components was a consequence of decreased cell proliferation or survival during reprogramming, we measured the area of nonreprogramming DAPI single cells. Knockdown of Gcn5, Trrap, Ccdc101, and Taf12 as well as the reprogramming factors had variable effects on the DAPI single-cell area in reprogramming cells (Supplemental Fig. S1D) but minimal effects in nonreprogramming MEFs (Supplemental Fig. S1E), However, even after normalizing the AP area to the DAPI single-cell area, we observed a dramatic defect in reprogramming upon loss of Gcn5, Trrap, Cdcc101, and Taf12 (Fig. 1D). Furthermore, we manually validated these primary screen hits by knocking down Gcn5, Trrap, Ccdc101, and Taf12. We confirmed that at least two individual siRNAs from each pool (Supplemental Table S1) effectively suppressed expression of the _target gene that correlated with inhibition of reprogramming. We also confirmed that knockdown of Pcaf using either pools or individual siRNAs while efficiently decreasing levels of this factor did not affect reprogramming (Fig. 1D; Supplemental Table S1).

Importantly, knockdown of the SAGA components (Gcn5, Trrap, Ccdc101, and Taf12) 1 d after inducing reprogramming also compromised AP colony formation, while knockdown of these factors later in reprogramming (day 5 and day 9) and secondary iPSCs had a lesser effect (Supplemental Fig. S1F), indicating that these SAGA components have a selective function during early reprogramming. Supporting the generality and specificity of these effects, defective reprogramming caused by loss of Gcn5 was partially restored by transient expression of a siRNA-resistant Gcn5 cDNA construct (Supplemental Fig. S1G), and knockdown of Gcn5 in OSKML mRNA-mediated reprogramming of human BJ fibroblasts also suppressed AP colony formation (Supplemental Fig. S1H; Mandal and Rossi 2013). Together, these results indicate that successful transition through the initiation phase of reprogramming is dependent on Gcn5 and the SAGA complex and is conserved in mouse and human systems.

Gcn5 is a coactivator of the Myc transcriptional network in pluripotent cells

Gcn5 is essential for embryonic survival in mice (Xu et al. 2000; Yamauchi et al. 2000; Bu et al. 2007) and is highly expressed in mESCs compared with differentiating cells (Fig. 2A), but little is known of the gene expression programs regulated by Gcn5 in pluripotent cells and during development. To better address Gcn5 functions in mESCs, we preformed chromatin immunoprecipitations (ChIPs) to identify direct _target genes. Available Gcn5 antibodies proved unsuccessful in ChIP assays (data not shown). Therefore, we used an in vivo biotinylated and Flag-tagged form of Gcn5 expressed at levels similar to that of the endogenous protein (Supplemental Fig. S2A; Kim et al. 2008, 2009, 2010). Coupling biotin:streptavidin-mediated ChIPs with massive parallel sequencing (bioChIP-seq) identified 7499 common Gcn5-bound sites in duplicate experiments that were highly enriched relative to control cells (which expressed an unfused biotin construct) (Fig. 2B; Supplemental Table S2). The majority (53%) of Gcn5-bound sites were located within 1 kb of a transcription start site (TSS), and the average distribution of Gcn5 across the gene bodies correlated strongly with that of RNA polymerase II (Pol II) (Fig. 2C,D). Furthermore, when compared with publicly available mESC histone mark ChIP-seq data sets, unsupervised clustering revealed Gcn5 sites partitioned into five distinct clusters (Fig. 2E; Supplemental Table S2). The majority (59%) were enriched for modifications associated with active promoters, including H3K9/14ac, H3K27ac, H3K4me3, and Pol II, supporting a role for Gcn5 in their activation (Supplemental Fig. S2B,C). We termed these clusters Active-a and Active-b (Fig. 2E), with the latter displaying a broader distribution of active marks. Interestingly, cluster 3 was associated with bivalent marks (Supplemental Fig. S2D), while cluster 4 included transcript elongation, and cluster 5 comprised undefined intergenic sites. Altogether, these results are in accordance with a set of previous studies defining Gcn5 as a gene-specific coactivator (Lee et al. 2000; Krebs et al. 2011) rather than a global regulator of Pol II transcription (Bonnet et al. 2014).

To investigate which transcription factors might recruit Gcn5 to promoters, we performed de novo motif enrichment analysis. This revealed remarkable concordance with E2f1-binding sites (P = 5.9 × 10⁻⁶⁷) and Myc/Max-binding sites (P = 7.7 × 10⁻⁶²) (Fig. 3A). No such enrichment was observed upon comparing the Gcn5-bound sites with random sequences of the same GC content and length (Supplemental Table S2). Furthermore, Gcn5-bound sites also revealed strong association with sites bound by n-Myc, c-Myc, E2f1, and H3K9/K14ac (Fig. 3B; Supplemental Table S2 for values) but not Oct4, Sox2, or Nanog (Chen et al. 2008; Rosenbloom et al. 2012). In addition, these sites primarily fell into the Active-a and Active-b clusters (Supplemental Fig. S3A,B), suggesting that Gcn5 coactivates Myc and E2f1 transcriptional networks in pluripotent cells. Indeed, when we analyzed altered gene expression by RNA sequencing (RNA-seq) in Gcn5^flox/flox mESCs with and without Gcn5 deletion upon expression of Cre recombinase (Supplemental Fig. S3E,F), we identified 2239 genes that were down-regulated and only 92 genes that were up-regulated upon Gcn5 loss (Fig. 3C; Supplemental Table S3; Loven et al. 2012). Of these, 474 genes were direct _targets of Gcn5, as they were also identified as Gcn5-bound in our ChIP experiments. These genes are mainly involved in cell cycle regulation (Fig. 3D) and are also mostly Myc/E2f1 _targets (Fig. 3E; Supplemental Fig. S3G). Interestingly, Gcn5^−/− mESCs display no obvious abnormalities and are capable of differentiating into three germ layers (Lin et al. 2007). Furthermore, they stably express pluripotency markers and show normal cellular morphology and growth kinetics (Supplemental Fig. S3H,I). Together with our functional screen (Fig. 1), these results indicate that Gcn5 is necessary to establish, but not maintain, pluripotency.

Myc directly activates early expression of SAGA components during reprogramming

Our ChIP-seq analyses indicate that Gcn5 is recruited to Myc _target genes in mESCs, raising the possibility that Gcn5 might also engage with Myc during somatic cell reprogramming. To further address how Gcn5 contributes to reprogramming, we monitored Gcn5 expression during this process and found that Gcn5, but not Pcaf, mRNA levels were rapidly induced upon expression of OSKM (Fig. 4A). Furthermore, siRNA depletion of individual transgenes during the first 2 d of reprogramming revealed that Myc was primarily responsible for the early spike-in Gcn5 expression (Fig. 4B; Supplemental Fig. S4C), and ChIP analysis of Myc binding to the TSS of Gcn5 revealed a strong increase upon induction of reprogramming (Fig. 4C). In addition to Gcn5, Ccdc101 mRNA levels also increased abruptly after just 1 d of reprogramming (Supplemental Fig. S4A), and Myc binding to the TSS of this gene was increased similar to the increase in binding observed at Gcn5 (Supplemental Fig. S4B–D). In contrast, Trrap and Taf12 levels remained constant during reprogramming. To further determine whether OSK also had the potential to activate Gcn5 and Ccdc101 expression in the absence of Myc, we overexpressed the individual reprogramming factors as well as an OSK cocktail in wild-type MEFs by lentiviral infection (Supplemental Fig. S4E). In agreement with the above data, infection with Myc alone, but not OSK, was sufficient to up-regulate Gcn5 and Ccdc101 expression comparable with levels detected during early reprogramming (Fig. 4D; Supplemental Fig. S4F). Conversely, Myc inhibited Pcaf expression. Interestingly, our analysis in mESCs also revealed that Gcn5 as well as other components of the SAGA complex (Ccdc101, Taf12, and Atxn7l3) are direct _targets of Myc (see Supplemental Fig. S3C,D for examples of Gcn5 and Ccdc101). Collectively, these data suggest that Myc drives a Gcn5-SAGA feed-forward loop early in reprogramming by directly regulating expression of SAGA components and recruiting SAGA proteins for activation of Myc _target genes.

Myc and Gcn5 cooperate to activate a network of RNA processing genes during initiation of reprogramming

The requirement for Gcn5 and, more specifically, the SAGA complex during the initiation phase suggests that a Myc–SAGA-mediated regulatory module is one of the earliest epigenetic mechanisms activated during reprogramming. To directly evaluate the Myc–Gcn5 axis, we next performed Myc ChIP-seq on secondary MEFs grown for 2 d in the absence (MEFs) or presence (day 2 [D2]) of Dox. As expected, Myc bound to more genes in D2 reprogramming cells (∼3.2-fold more) than in MEFs and bound a number of genes comparable with that observed previously in early human reprogramming cells (Supplemental Fig. S5A; Soufi et al. 2012). Strikingly, the Myc-bound genes in D2 reprogramming cells (96%) incorporated virtually all genes bound by Gcn5 in mESCs (Fig. 5A). In stark contrast, the overlap was much less in nonreprogramming MEFs (43%). Furthermore, c-Myc coimmunoprecipitation experiments demonstrated a physical interaction between Gcn5 and c-Myc in both reprogramming D2 cells and mESCs (Supplemental Fig. S5B). These data suggest that Myc and Gcn5 cooperate to regulate gene expression profiles early in the reprogramming process through physical interaction. To determine how this interaction contributes to establishment of pluripotency, we depleted either Gcn5 or Myc in secondary MEFs and D2 reprogramming cells, performed duplicate RNA-seq analyses (Supplemental Fig. S5C,D; Supplemental Table S4), and globally compared gene expression profiles by principal component analysis (PCA). These comparisons revealed that the gene expression profiles in Myc versus Gcn5 knockdown MEFs were distinct (Fig. 5B, blue dots). In contrast, in D2 reprogramming MEFs, Myc or Gcn5 knockdown profiles were strikingly similar to each other but were distinct from D2 siControl cells (Fig. 5B, pink dots). Indeed, of 2246 genes that were dependent on Gcn5 during reprogramming, 78% (1760) were also Myc-responsive (Supplemental Fig. S5E). Collectively, these results indicate that in MEFs, Myc and Gcn5 have distinctive roles in maintaining the global gene expression program but that during reprogramming, they form a tightly coupled functional module that launches a transcriptional program required for acquisition of pluripotency.

We next identified reprogramming-specific transcriptional programs governed by the Myc–Gcn5 module. For this, we identified genes specifically ≥1.4-fold up-regulated by Myc and Gcn5 during reprogramming but not in MEFs. This revealed 2262 and 657 genes up-regulated by Myc and Gcn5, respectively. Of these Myc reprogramming-responsive genes, 94% were also directly bound by Myc, indicating that almost the entire program is a direct _target (Fig. 5C, top). Interestingly, although technical challenges of performing bioChIP in reprogramming cells prevented an analysis of Gcn5 genome occupancy, analysis of our mESC results revealed that 32% of Gcn5-dependent genes are occupied by Gcn5 in mESCs (Fig. 5C, bottom). Since almost all Gcn5-bound genes in mESCs are occupied by Myc during reprogramming (Fig. 5A), these data suggest that Myc and Gcn5 are tightly coupled early during reprogramming.

We next analyzed Myc–Gcn5 _targets in reprogramming cells by gene ontology (GO) analysis, which revealed a strong enrichment for RNA processing and RNA splicing among the top functional categories (Fig. 5D; Supplemental Table S4). Among these genes are several general splicing factors, including Snrpd1, U2af1, Isy1, Skiv2l2, Prpf4, Pnn, and Snrpg. In accordance, genes cobound by Myc and Gcn5 were also enriched for these categories (Supplemental Fig. S5F). Importantly, this group of genes was distinct from non-Myc/Gcn5-induced genes, which were enriched for MET genes (functional categories were epidermis development, ectoderm development, epidermal cell differentiation, and keratinization) that are regulated by Klf4 and BMP–Smad signaling (Fig. 5D; Supplemental Table S4; Mikkelsen et al. 2008; Li et al. 2010; Samavarchi-Tehrani et al. 2010; O'Malley et al. 2013). Furthermore, genes bound by Myc and up-regulated by Myc at D2 of reprogramming (Fig. 5C, top) were also incredibly enriched for RNA processing functions (P value = 1.62 × 10⁻⁸⁹), while Myc-suppressed genes were enriched for skeletal system development and cell adhesion-related processes (Supplemental Table S4). These findings imply that Myc may play a more direct role in stimulating RNA processing events than cellular proliferation during early reprogramming. We also assessed whether disruption of the other SAGA components identified in our screen interfered with expression of a panel of splicing and/or RNA processing-associated genes affected by Myc/Gcn5 induction. Indeed, loss of either Trrap, Ccdc101, or Taf12 disrupted induction of the majority of genes tested (Supplemental Fig. S5G). These results demonstrate that Myc and Gcn5 cooperate to control the expression of key splicing and RNA processing genes during reprogramming.

Myc–Gcn5-regulated splicing factors are necessary for reprogramming

To determine whether any of the RNA splicing factors directly regulated by Myc and Gcn5 are functionally required for reprogramming, we performed a small-scale RNAi screen _targeting these factors in early reprogramming cells and performed a parallel screen in mESCs to distinguish factors required for establishment versus maintenance of pluripotency. Our library included siRNAs _targeting 20 RNA processing genes expressed in MEFs and mESCs and directly coregulated by Myc and Gcn5 as well as three additional RNA processing genes that were not identified as _targets of Myc and Gcn5 (Larp4, Celf1, and Hnrnpa1). Also, siRNAs _targeting Gcn5, Myc, and Oct4 were used as controls, along with a non_targeting siControl. The establishment screen was performed in reprogramming cells similar to the epigenetic regulator screen (Fig. 1A), while, in the maintenance screen, mESCs were transfected with siRNAs and fixed after 2 d in culture (Fig. 6A). We performed three biological replicates and analyzed the average relative AP and DAPI colony area as above. As expected, knockdown of Gcn5 or Myc suppressed colony formation selectively in reprogramming cells compared with mESCs, whereas Oct4 knockdown prevented colony formation in both reprogramming cells and mESCs (Fig. 6B; Supplemental Fig. S6A–C). To define RNA processing genes that were required for the initiation phase of reprogramming, similar to Gcn5, a cutoff for the lowest 15% based on AP area was set (Fig. 6B, bottom dotted black line). This analysis revealed a total of 13 splicing/RNA processing genes coregulated by Myc and Gcn5 as mediators of reprogramming and hence associated with the establishment of pluripotency. The requirement for each of these genes was further validated using the four individual siRNAs that made up the pools, and all except Naa38 were confirmed as hits based on at least two individual siRNAs inhibiting expression of the _target gene and abrogating reprogramming (Supplemental Table S5). Conversely, the cutoff for the mESC screen was set at 25% AP area reduction (similar to siOct4) compared with siControl transfected cells (Fig. 6B, top dotted black line). This revealed that six of the RNA processing factors implicated in early reprogramming were also necessary to maintain pluripotent cell growth (Utp6, U2af1, Isy1, Snrpd1, Phf5a, and Snrpg). These proteins thus play multifaceted roles in pluripotency, whereas the six remaining RNA processing factors identified as Myc–Gcn5 _targets are primarily associated with inducing pluripotency (Tra2b, Prpf4, Snrnp70, Hnrnpc, Skiv2l2, and Pnn). Together, these data reveal that the Myc–SAGA module regulates a network of RNA splicing and processing factors that is required for reprogramming during early initiation.

The Myc-SAGA regulated splicing program _targets cell migration

Regulated AS has recently emerged as an important event in the control of stem cell pluripotency and somatic cell reprogramming (Gabut et al. 2011; Han et al. 2013; Ohta et al. 2013; Venables et al. 2013; Lu et al. 2014). Our data suggest that Myc–SAGA-driven expression of a splicing-associated network is critical early in reprogramming to modulate AS events. To explore this possibility, we first assessed whether knockdown of Gcn5 or Myc impacted alternative exon splicing within the first 2 d of reprogramming. For this, we quantified AS events using our RNA-seq data from secondary MEFs and D2 reprogramming cells in which Myc or Gcn5 was knocked down by siRNAs (Supplemental Fig. S5C). Splicing levels were quantified by measuring the percentage of transcripts with the alternative exon spliced in (percent spliced in [PSI]). In addition, AS patterns were monitored in mESCs and matched secondary iPSCs. We identified 59 differential AS events that changed by a PSI value ≥15% by D2 of reprogramming when compared with MEFs, with the majority of events (48) associated with exon exclusion during reprogramming (Fig. 7A; Supplemental Fig. S7A; Supplemental Table S6). In particular, genes that mediate cell migration—including such processes as cytoskeletal organization, cellular polarization, and cell adhesion (Fig. 7A, blue)—were identified among these early reprogramming AS events, while signaling factors (Fig. 7A, red), transcriptional regulators (Fig. 7A, purple), and RNA processing factors (Fig. 7A, orange) were less represented. Myc and/or Gcn5 modulated 27 out of the 59 AS events within 2 d of reprogramming (Supplemental Fig. S7B), and, by unsupervised clustering, we noted that Myc and Gcn5 knockdown reprogramming cells corresponded better with nonreprogrammed secondary MEFs than D2 reprogramming cells (Fig. 7A). Interestingly, the latter clustered with mESCs and secondary iPSCs, indicating that Myc–Gcn5 induces a pluripotency-associated AS program at an early stage during reprogramming. Furthermore, we observed that Myc and Gcn5 coregulated a number of reprogramming-specific AS events associated with exon exclusion (Fig. 7A; Supplemental Fig. 7A; Supplemental Table S6), six of which are associated with genes involved in cell migration and are necessary for reprogramming (Slain2, Plod2, Fat1, Pxdn, Myo5a, and Pcm1) (Supplemental Fig. S7C).

AS of the Slain2 and Plod2 genes was recently described in somatic cell reprogramming and has been implicated in cytoplasmic microtubule elongation and collagen cross-link stabilization, respectively (van der Slot et al. 2003; van der Vaart et al. 2011; Han et al. 2013; Ohta et al. 2013; Venables et al. 2013). However, the mechanism that contributes to AS of these genes has not been described. Interestingly, we identified Slain2 and Plod2 as the two most differentially spliced genes coregulated by Myc and Gcn5 that functioned in reprogramming. To validate the involvement of Gcn5 and Myc in these AS events, RT–PCR was performed using primers bracketing the AS exons in Slain2 and Plod2 transcripts, and PSI values were quantified. Furthermore, we monitored splicing of two additional genes linked to cell migration with previously undescribed roles in reprogramming: Fat1 and Pcm1 (Moeller et al. 2004; Ge et al. 2010). Comparison of PSI values for these four genes confirmed that exon skipping was more prevalent in D2 reprogramming cells compared with MEFs and was altered to variable degrees upon loss of Myc or Gcn5 (Fig. 7B). To determine whether exon exclusion at these genes was also affected by loss of other components of the SAGA complex identified in our screen as regulators of reprogramming (Fig. 1C,D), we knocked down Trrap, Ccdc101, and Taf12. Similar to Gcn5, these SAGA components also influenced exon exclusion of the Slain2, Plod2, Fat1, and Pcm1 genes during early reprogramming, although their ability to modulate these splicing events was quite variable, with knockdown of Pcaf having little effect (Fig. 7C). Knockdown of Trrap generally yielded the most striking changes, while Ccdc101 and Taf12 knockdowns were less severe, implying that Trrap-mediated recruitment of the SAGA complex is vital for initiating the AS network, although SAGA may retain some residual activity in the absence of the Ccdc101 and Taf12 subunits. Additionally, we confirmed that different combinations of the RNA processing factors directly regulated by Myc and SAGA also had the capacity to control these AS events (Supplemental Fig. S7D). However, RNA processing factors that control AS in reprogramming cells generally did not regulate the same events in mESCs (Supplemental Fig. S7E), suggesting that the Myc/Gcn5 axis may not be maintained in pluripotent cells but rather may serve as an immediate means to activate RNA processing in early reprogramming. Together, these results thus demonstrate that Myc cooperates with Gcn5 in the context of the SAGA complex to control expression of key genes required for regulated splicing events that _target cell migration during the first days of reprogramming.

SAGA is a gene-specific coactivator complex for Myc

Our data further demonstrate that Gcn5, but not Pcaf, is critical for somatic cell reprograming. These two highly related HATs clearly have both shared (Jin et al. 2014a,b) and unique functions (Xu et al. 2000; Yamauchi et al. 2000). Why particular SAGA components, including Gcn5, Trrap, Ccdc101, and Taf12, are necessary for reprogramming but others are not is intriguing. The SAGA complex comprises distinct multisubunit structural modules that mediate histone acetylation (Gcn5, Ada2b, Ada3, and Ccdc101), transcription activation (Trrap, Spt3, Spt7l, and Spt20 as well as the general transcription factors Taf5l, Taf6l, Taf9, Taf10, and Taf12), and histone deubiquitination (Usp22, Atxn7, Atxn7l3, and Eny2) (Koutelou et al. 2010; Samara and Wolberger 2011; Spedale et al. 2012). Based on our observations, it seems unlikely that the SAGA deubiquitination module has an impact on reprogramming. However, recruitment of the SAGA complex with an intact HAT module appears to be key. Trrap and Gcn5 have both been reported to interact directly with Myc (McMahon et al. 1998, 2000; Zhang et al. 2014), consistent with our findings (Supplemental Fig. S5B), so these factors likely negotiate recruitment of the SAGA complex to Myc _target genes. In addition, the bromodomain of Gcn5 and/or the tudor domain of Ccdc101 may aid in docking the complex to acetylated lysine residues and H3K4me2/3 modifications, respectively (Li and Shogren-Knaak 2009; Bian et al. 2011). Previous studies in yeast and mammalian cells indicated that SAGA is recruited to promoter regions by sequence-specific binding proteins, where it increases acetylation of H3 and other factors to drive expression of select genes (Wang et al. 1997; Nagy et al. 2010). More recent work has suggested a general role for the SAGA complex in transcription, as loss of ADA3 or ATXN7L3 leads to alterations in H3K9ac or H2Bub, respectively, at all active Pol II-driven genes in HeLa cells (Bonnet et al. 2014). However, Bonnet et al. (2014) did not directly examine GCN5 or PCAF recruitment. Our ChIP-seq and RNA-seq data in mESCs clearly indicate that Gcn5 is recruited to and activates a subset of active genes. Rather than acting as a general transcription factor, most Gcn5-binding sites were located near the TSS of select genes and were colocalized with active histone modifications specifically at those sites.

The role of Myc in gene activation has also been controversial. Although Myc and its partners clearly bind to E-box sequences, global gene expression studies suggested that Myc serves as an amplifier of all active genes (Lin et al. 2012; Nie et al. 2012). A more recent study concluded that global amplification of gene expression might be secondary to the effects of Myc in regulating specific gene _targets, especially those related to cell growth and division (Sabo et al. 2014). Our data are consistent with previous studies that indicated a role for Myc and E2f1 in SAGA complex recruitment to specific _target genes in transformed cells (McMahon et al. 1998, 2000; Zhang et al. 2014). Furthermore, Myc facilitates global maintenance of active acetylated chromatin and recruits Gcn5 to cell cycle-related genes in neural stem cells (Knoepfler et al. 2006; Martinez-Cerdeno et al. 2012). Altogether, these studies demonstrate the close connections between Gcn5 and Myc in both normal and transformed cells.

Gcn5 is required for establishing pluripotency

Our studies also uncovered a specific requirement for Gcn5 in the early initiation of a pluripotency-associated AS program but little if any role in the maintenance of pluripotency, as Gcn5^−/− mESCs remain pluripotent with no obvious morphological or growth defects (Lin et al. 2007). This apparent contradiction is quite interesting. There is precedence for such effects, as knockout of Utx or the Tet dioxygenases has no effect on pluripotency or self-renewal of mESCs, but depletion of these factors in MEFs compromises reprogramming (Mansour et al. 2012; Welstead et al. 2012; Hu et al. 2014). Pcaf might at least partially compensate for Gcn5 loss in mESCs, as deletion of Pcaf in Gcn5-null mice leads to more severe developmental defects than deletion of Gcn5 alone (Xu et al. 2000; Yamauchi et al. 2000). Alternatively, Gcn5 may simply play a more pivotal role in mediating cellular plasticity and hence pluripotent induction rather than the pluripotent state per se.

Myc–SAGA regulate a distinct RNA processing network during reprogramming

Temporal proteomic profiling of reprograming cells indicates that RNA processing factors are robustly up-regulated early in reprogramming, but the mechanism underlying this increase is not clear (Hansson et al. 2012). Subsequently, the RNA-binding proteins MBNL1, MBNL2, U2af1, and Srsf3 have emerged as important regulators of AS during reprogramming (Han et al. 2013; Ohta et al. 2013). Our studies now reveal that the Myc–SAGA module directly activates expression of U2af1. Moreover, we identified six additional RNA processing/splicing factor genes (Tra2b, Prpf4, Snrnp70, Hnrnpc, Skiv2l2, and Pnn) similarly coregulated by Myc and SAGA that are necessary for inducing pluripotency, further indicating that Myc and SAGA nucleate regulation of a RNA splicing pathway that is distinct from that found in MEFs. Moreover, we found that genes induced by Myc and Gcn5-SAGA regulate AS of a number of genes associated with cell migration, including Slain2, Plod2, Fat1, Pxdn, Myo5a, and Pcm1. These results are consistent with previous reports that Plod2 and Slain2 are alternatively spliced during reprogramming (Han et al. 2013; Ohta et al. 2013). Interestingly, the longer isoform of Plod2 decreased during reprogramming increases collagen cross-links (Mercer et al. 2003; Walker et al. 2005). Therefore, it is tempting to speculate that Myc and SAGA regulate an AS gene network that ultimately mediates reorganization of cellular structures and migration to facilitate pluripotency. Future studies will be aimed at dissecting the functionality of these AS events and their mechanistic roles in establishing pluripotency.

The ability of Myc to trigger a positive feed-forward loop and couple with SAGA to modulate a novel set of AS events defines a second key pathway within the initiation phase that is entirely distinct from the BMP-driven MET mediated by Klf4 and Smad1 (Li et al. 2010; Samavarchi-Tehrani et al. 2010). These studies not only reveal a new role for Myc in boosting reprogramming potential but also convey that reprogramming factors carry out distinct functions that are collectively required for successful reprogramming to a pluripotent state.

In summary, our studies highlight a previously undefined role for a Myc–Gcn5 feed-forward loop in driving Myc _target genes important for self-renewal in stem cells and for RNA processing and splicing during the earliest stage of somatic cell reprogramming. This pathway is distinct from but parallel to the induction of MET genes during the initiation phase of reprogramming. Moreover, our findings imply that the Myc–Gcn5 RNA processing module identified here may also play a fundamental role in Myc-driven oncogenesis. Forthcoming studies will address this important question as well as the role of other epigenetic regulators at different stages of reprogramming.

Generation of stable mESCs and bioChIP assays

AB1 ESCs stably expressing in vivo biotinylated Gcn5 (BirAV5-FLBioGcn5) were generated, and bioChIPs were performed as previously described with minor modifications (Kim et al. 2009).

siRNA screening

RNAi-mediated knockdown was performed with 40 nM siRNA pools using Lipofectamine RNAiMAX (Invitrogen) in secondary 1B MEF cells or mESCs upon seeding as previously described (Samavarchi-Tehrani et al. 2010; Golipour et al. 2012). After 24 h, Dox was added and replaced daily. Five days later, cells were fixed and stained for AP activity (Vector red; AP substrate kit, Vector Laboratories) and counterstained for DAPI. All siRNAs are listed in Supplemental Table S7.

Quantitative RT–PCR (qRT–PCR)

Total RNA was isolated using the RNeasy minikit (Qiagen). RNA was converted to cDNA prior to qPCR using SYBR green PCR master mix (Roche). Primer sequences are detailed in Supplemental Table S7.

ChIP-seq

Libraries were prepared using a modified version of the Illumina TruSeq ChIP sample preparation protocol. See the the Supplemental Material for complete details. Each library (10 pM) was sequenced on an Illumina HiSeq 2000. The raw reads in ChIP-seq data sets were mapped to NCBI build 37 (University of California at Santa Cruz [UCSC] mm9) using Bowtie (version 0.12.8) (Langmead et al. 2009). Peaks were called using MACS (version 2.0.10) (Zhang et al. 2008) at a Q-value threshold of 0.05 using input as a control. The enriched regions from each biological replicate of samples were intersected with BEDTools (version 2.13.3) (Quinlan and Hall 2010) to form the final peak set for Gcn5 bioChIPs.

RNA-seq

RNA-seq libraries were prepared according to the Illumina TruSeq standard total RNA sample preparation kit (with Ribo-Zero GoldRS-122-2301). Each library (10 pM) was sequenced on an Illumina HiSeq 2000. RNA-seq reads were aligned to NCBI build 37 (UCSC mm9) using TopHat 2.0.9 (Trapnell et al. 2009). Complete details regarding spike-in normalization for the mESC RNA-seq are described in the Supplemental Material. EdgeR was applied to determine differentially expressed genes (Robinson et al. 2010).

Data for ChIP-seq and RNA-seq have been submitted to Gene Expression Omnibus; accession numbers are pending and will be available on request.