The RNA exosome nuclease complex regulates human embryonic stem cell differentiation

Embryonic stem cells (ESCs), which derive from the inner cell mass of the blastocyst, are able to self-renew indefinitely while maintaining the ability to differentiate into all three germ layers (Thomson et al., 1998). Many mechanisms that underlie these unique features have been studied extensively. Pluripotency is controlled by a network of transcription factors that includes OCT4, NANOG, and SOX2 (De Los Angeles et al., 2015). This stem cell–specific transcription factor network is associated with a less condensed “open” chromatin state that is thought to allow rapid changes in gene expression upon differentiation (Gaspar-Maia et al., 2011). Open chromatin is associated with elevated levels of chromatin-remodeling factors and increased diversity of nascent RNAs relative to differentiated cells (Efroni et al., 2008; Fort et al., 2014). In addition, transcripts from many repetitive elements, including retrotransposons such as the long interspersed nuclear elements (LINEs) and short interspersed nuclear elements, are present at increased levels (Efroni et al., 2008; Santoni et al., 2012). Thus, ESCs must balance the need to keep many genes transcriptionally competent with the need to protect themselves from deleterious consequences of promiscuous transcription.

Several mechanisms contribute to reducing levels of unwanted and potentially harmful RNAs in ESCs. The chromatin surrounding the transcription start sites (TSSs) of many developmental regulators contains active and repressive histone modifications, a “bivalent” state that may both silence these genes and promote their activation during development (Bernstein et al., 2006). RNA interference contributes to epigenetic silencing of centromeric repeats in mouse ESCs (Kanellopoulou et al., 2005). The proteasome removes preinitiation complexes from tissue-specific promoters (Szutorisz et al., 2006). However, the role of surveillance pathways in degrading differentiation-related and deleterious RNAs is poorly understood. Although in human ESCs (hESCs), down-regulation of the nonsense-mediated decay (NMD) pathway promotes differentiation into endoderm (Lou et al., 2016), the RNA targets responsible were not identified. The roles of other RNA surveillance pathways have not been investigated in hESCs.

The RNA exosome, a multiprotein nuclease complex, is the central effector of a major RNA surveillance pathway in eukaryotes (Zinder and Lima, 2017; Ogami et al., 2018). The core exosome consists of nine subunits that form a hexameric ring topped by three RNA-binding subunits. In human cells, the core exosome lacks catalytic activity, which is conferred by three associated nucleases that differ based on subcellular location (Tomecki et al., 2010). The major catalytic subunit in nucleoli is the 3′ to 5′ exoribonuclease EXOSC10 (also called hRRP6), while the nucleoplasmic exosome also contains DIS3 (hRRP44), which has both 3′ to 5′ exoribonuclease and endonuclease domains. The cytoplasmic exosome contains DIS3 or DIS3L, a related 3′ to 5′ exoribonuclease (Tomecki et al., 2010). The exosome has been best studied in yeast, where it degrades aberrant pre–ribosomal RNAs (rRNAs), pre-tRNAs, small nucleolar RNAs (snoRNAs), small nuclear RNAs (snRNAs), antisense RNAs, and “cryptic unstable transcripts” that arise from bidirectional transcription from RNA polymerase II promoters. The exosome is also required for 3′ maturation of 5.8S rRNA and many snoRNAs and contributes to mRNA decay. Many of the same RNAs are targets in human cells (Morton et al., 2018; Ogami et al., 2018).

Studies of mammalian progenitor cells support a role for the exosome in maintaining these cells in the undifferentiated state. Depletion of the EXOSC9 subunit from human epidermal progenitors results in decreased proliferation, premature differentiation, and loss of epidermal tissue (Mistry et al., 2012). The exosome promotes epidermal progenitor cell self-renewal and prevents differentiation by degrading mRNA encoding the GRHL3 transcription factor. Similarly, depleting EXOSC8 and EXOSC9 subunits from mouse erythroid precursors results in increased mature erythroid cells (McIver et al., 2014). However, progenitor cells differ from ESCs in that they are committed to a specific developmental program and undergo only a few divisions before differentiating. Despite the importance of hESCs to regenerative medicine and the need to understand the mechanisms by which they differentiate into specific lineages, the role of the exosome in regulating hESC pluripotency has not been explored.

Here, we determine the role of the exosome in maintaining the unique properties of hESCs. We show that the exosome restrains differentiation of hESCs into endoderm, mesoderm, and ectoderm. Consistent with a key role in maintaining pluripotency, down-regulation of exosome function occurs early in differentiation. We identify roles for the exosome in decreasing the levels of potentially active LINE-1 RNAs and reducing levels of specific mRNAs and noncoding RNAs (ncRNAs) that either encode key developmental regulators or regulate their expression. The exosome represses induction of mesendoderm (ME), the precursor to mesoderm and endoderm, by degrading pre-mRNA encoding forkhead box protein H1 (FOXH1), a critical transcription factor. Our studies establish the exosome as a novel regulator of hESC differentiation.

We used piggyBac transposons (Ding et al., 2005) to generate hESC lines that stably express doxycycline-inducible shRNAs against mRNA encoding the core EXOSC3 subunit (Fig. 1 A). We targeted EXOSC3 (hRRP40), since its depletion destabilizes other subunits, including catalytic subunits EXOSC10 and DIS3L (Tomecki et al., 2010), allowing efficient reduction of activity. After selecting clonal cell lines, we evaluated depletion by Western blotting. Maximal depletion occurred by 5 d in doxycycline, resulting in 70% reduction in EXOSC3 (Fig. 1 B). As described previously (Tomecki et al., 2010), EXOSC10 also decreased (Fig. 1 B). Depletion reduced exosome function, as promoter upstream transcripts (PROMPTs), short-lived transcripts upstream of RNA polymerase II promoters (Preker et al., 2008), and the 7S precursor to 5.8S rRNA (Mitchell et al., 1997) accumulated. Maximal accumulation of both RNAs occurred by 5 d in doxycycline, and the RNAs remained elevated at 7 d (Fig. 1, C and D).

We analyzed the effects of exosome depletion on hESC proliferation and expression of pluripotency markers. After 5 or 7 d in doxycycline (Figs. 1 E and S1 A), the cell cycle of EXOSC3-depleted cells was similar to that of control cells. No major effects on colony size were observed (Fig. 1 F), although some colonies expressing both EXOSC3 and control nonsilencing shRNAs (shNS) showed minor morphological changes after 7 d (Fig. 1 F, arrows), which may represent a nonspecific doxycycline effect. No differences in viability were detected after 5 or 7 d of EXOSC3 depletion (Fig. S1 B). The lack of strong effects on viability, at least during the time examined, is consistent with a study in which EXOSC3 was depleted from human HeLa cells (Tomecki et al., 2010). Levels of mRNAs encoding the pluripotency markers OCT4, NANOG, and SOX2 were unchanged upon EXOSC3 depletion (Fig. S1 C). Experiments in which we stained hESCs to detect the pluripotency marker alkaline phosphatase also did not reveal differences between control and exosome-depleted cells (Fig. S1 D).

Although the levels of mRNAs encoding pluripotency markers were unaffected in EXOSC3-depleted hESCs, we detected increased levels of mRNA encoding EOMES, a transcription factor required for endoderm formation (Teo et al., 2011; Fig. S2 A). We also detected slightly increased MIXL1 mRNA, a mesoderm marker, although this did not reach statistical significance (Fig. S2 A). To explore a role for the exosome in curtailing differentiation, we used siRNAs directed against other sites on EXOSC3 mRNA to deplete the exosome. Depletions with either pooled or individual siRNAs were slightly more effective than shRNA depletion, resulting in 75% reduction with pooled siRNAs and one of the two individual siRNAs (siEXOSC3-2; Fig. 2 A). With these siRNAs, PROMPTs increased up to sixfold, compared with the four- to fivefold increases seen with shRNAs (Figs. 2 B and S2 B). Importantly, mRNAs encoding MIXL1, GATA4, and DLX5, markers of mesoderm, endoderm, and ectoderm, respectively, all increased in siEXOSC3-treated cells (Figs. 2 B and S2 B).

To further assess this role of the exosome, we allowed EXOSC3-depleted hESCs to differentiate into embryoid bodies (EBs), 3D structures that contain all three germ layers (Brickman and Serup, 2017). After culturing in doxycycline to deplete EXOSC3, hESCs were induced to differentiate into EBs. We used our stable shRNA-expressing cell lines for EB formation, since by culturing in doxycycline, we could maintain shRNA expression during differentiation. After 5 d of differentiation, EXOSC3-depleted EBs expressed mRNAs encoding early ectoderm (DLX5), mesoderm (MIXL1, EOMES, and TBXT), and endoderm markers (GATA4 and SOX17) at 2.5- to 15-fold higher levels than control cells (Figs. 2 C and S2 C). Examination at earlier times revealed that EOMES mRNA was elevated in EXOSC3-depleted EBs within 2 d of differentiation, with significant changes by day 4 (Fig. S2 D). Similar increases in mRNAs encoding differentiation markers (MIXL1, EOMES, GATA4, SOX17, and PAX6, an early ectoderm marker) were observed in EBs formed from a second hESC line expressing shRNAs against EXOSC3 (Fig. S2 E).

As our results suggested that the exosome inhibits hESC differentiation, we determined if down-regulation of this nuclease complex occurs during differentiation of WT hESCs. We compared the levels of several subunits in H1 hESCs (which are male) and their corresponding EBs by immunoblotting. Within 1 d of differentiation, EXOSC3, EXOSC4, and EXOSC10 proteins decreased (Fig. 2 D). Similar changes were observed when EBs were formed from a second line, female H9 hESCs (Fig. S2 F). Down-regulation occurred after transcription as the mRNAs encoding the nine core subunits and three catalytic subunits did not decrease (Fig. S2 G).

To determine if high levels of exosome subunits correlate with the pluripotent state, we examined reprogramming of differentiated cells into induced pluripotent stem cells (iPSCs). Comparison of differentiated cells derived from H1 hESCs (Park et al., 2008) with the resulting iPSCs revealed that EXOSC3, EXOSC4, and EXOSC10 all increased at least twofold upon reprogramming (Fig. 2 E). All three subunits also increased 2–15-fold when two primary human fibroblast lines (BJ1 and Detroit 551) were reprogrammed (Fig. 2 E), consistent with reports that EXOSC1, EXOSC2, EXOSC7, EXOSC9, and DIS3 mRNAs are up-regulated early in reprogramming of fibroblasts to iPSCs (Polo et al., 2012). Our results reveal that the exosome restrains differentiation of hESCs and support a model in which elevated exosome levels are important for maintaining pluripotency.

To identify RNAs that are direct exosome targets, we performed in vivo cross-linking. We cross-linked proteins to their targets with UV (254 nm) followed by immunoprecipitation and HITS-CLIP (high-throughput sequencing of RNA isolated by cross-linking immunoprecipitation; Licatalosi et al., 2008). We used antibodies against EXOSC3, part of the exosome cap, and EXOSC4, part of the hexameric ring. Nonimmune IgG was a control (Fig. S3 A). More EXOSC4 tags were recovered than EXOSC3 tags (2.3 million vs. 1.5 million) and were derived from a greater variety of RNAs (Fig. 3 A), possibly because RNA substrates make multiple contacts with EXOSC4 (Weick et al., 2018). Both 45S pre-rRNA and 5S rRNA were prominent targets, consistent with a study in yeast (Schneider et al., 2012). Some tags derived from known pre-rRNA targets, such as the 5′ external transcribed spacer (5′ETS) and internal transcribed spacers 1 and 2 (Fig. 3, B–D). In addition to other known targets, such as mRNAs, introns, and ncRNAs, we obtained tags from repeat element transcripts and from mitochondrial rRNAs (mt-rRNAs). Levels of all tags were strongly reduced in the control sample. As expected, EXOSC3 and EXOSC4 tags were enriched for the cross-linking–induced mutation sites (Moore et al., 2014) that occur during reverse transcription across the residual cross-linked oligopeptide (Fig. 3, E and F).

Since many exosome targets are expected to be short lived, we also used 4-thiouridine to label nascent RNA, followed by cross-linking with UV (365 nm) and immunoprecipitation (Hafner et al., 2010). Since HITS-CLIP and PAR-CLIP (photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation) rely on different cross-linking chemistries, use of both protocols recovers a wider range of targets (Castello et al., 2012). To maximize our yield of cross-linked RNAs, we used the iCLIP (individual nucleotide resolution cross-linking and immunoprecipitation) protocol (Huppertz et al., 2014), which recovers cDNAs that truncate prematurely at the site of the cross-link. Using this PAR-iCLIP protocol with antibodies to EXOSC4, the fraction of tags that mapped to the ribosomal DNA and 5S rRNAs declined, while the fraction of repeat transcripts and ncRNAs increased (Fig. 3 A). As expected for a protocol that recovers newly made RNAs, tags derived from pre-45S rRNA spacers were abundant in PAR-iCLIP (Fig. 3, C and D).

Since the exosome functions with oligo(A) polymerases to degrade some RNAs (Zinder and Lima, 2017; Ogami et al., 2018), we examined tags for nontemplated A tails. Although known targets such as the 5′ETS were present in both HITS-CLIP and PAR-iCLIP (Fig. 3, G and H), 81% of the EXOSC3 and 73% of the EXOSC4 tags with A tails in HITS-CLIP mapped to 16S mt-rRNA, with most tags from the 3′ end (Fig. 3, G and I). This RNA is transcribed as a polycistronic precursor, with the 3′ end generated by RNase P cleavage of the downstream tRNA^Leu. Since most tails began after the first G of the tRNA, these RNAs may represent aberrant cleavage products (Fig. 3 I). As expected for bona fide targets, A-tailed 16S mt-rRNAs accumulated when EXOSC3 was depleted with siRNAs (Fig. 3 J).

To identify the effects of exosome depletion on a transcriptome-wide scale, we sequenced rRNA-depleted RNA from shEXOSC3 and shNS-expressing cells after 5 and 7 d in doxycycline, as well as from cells grown without doxycycline. Since RNAs that increase are most likely to be direct targets, we focused on transcripts that increased >1.2-fold upon EXOSC3 depletion, compared with control cells (Table S1). As expected, known targets, such as 3′ extended forms of snRNAs, snoRNAs, and telomerase RNA (Lubas et al., 2015; Macias et al., 2015; Nguyen et al., 2015; Tseng et al., 2015), increased upon EXOSC3 depletion (Fig. S3). Although most mature ncRNAs were not strongly affected, the minor spliceosome U4atac snRNA (Tarn and Steitz, 1997) increased 2.5-fold (Fig. S3), suggesting exosome-mediated degradation of pre-U4atac could affect pre-mRNA splicing in hESCs.

Notably, transcripts derived from L1 LINEs were targets in HITS-CLIP and PAR-iCLIP (Fig. 4 A) and increased upon EXOSC3 depletion. These elements, which comprise ∼17% of the genome, are the only known active autonomous retrotransposons in humans. Although full-length L1 RNA is ∼6 kb, most genomic copies are truncated and/or contain inactivating mutations, and only 80–100 members of the human-specific L1Hs subfamily remain capable of retrotransposition (Beck et al., 2011). L1 RNAs were prominent targets, comprising 3.9% and 5.0% of the EXOSC4 tags in HITS-CLIP and PAR-iCLIP, respectively. Mapping the tags against a retrotransposition-competent L1Hs sequence revealed that the tags were 27- and 30-fold enriched, respectively, compared with the control (Fig. 4 B). The tags mapped throughout the RNA, including the 5′UTR (Fig. 4 C), the region most likely to be truncated (Beck et al., 2011).

In support of the hypothesis that active LINEs are exosome targets, L1Hs RNAs increased upon EXOSC3 depletion (Fig. 4 D). The increase in L1Hs RNAs was accompanied by increased ORF1, the more easily detected of the two L1-encoded proteins (Fig. 4 E, lane 5). Using quantitative RT-PCR (RT-qPCR), we confirmed that L1Hs RNAs increased in EXOSC3-depleted cells (Fig. 4 F) and that levels of other retrotransposons, such as the Alu family of short interspersed nuclear elements, were unchanged (Fig. 4 G). L1Hs RNAs also increased when siRNAs targeting other sites on EXOSC3 or the catalytic subunits were used (Fig. 4, H and I). As depleting EXOSC10 together with DIS3 resulted in accumulation of L1 RNAs to levels seen when EXOSC3 was depleted (Fig. 4 I), degradation may occur within nuclei. Although attempts to determine if exosome depletion increased retrotransposition using an engineered element were inconclusive due to the low efficiency of hESC transfection (Garcia-Perez et al., 2007), our experiments identify L1 RNAs as exosome targets.

Although primary miRNA transcripts (pri-miRNAs) are not known exosome targets, pri-miR-205 was bound by the exosome in HITS-CLIP and PAR-iCLIP and was up-regulated on exosome depletion (Fig. 5, A and B; and Table S1). The encoded miRNA regulates the epithelial-to-mesenchymal transition by down-regulating the ZEB1 and ZEB2 transcriptional repressors (Gregory et al., 2008) and overexpression of miR-205 in mouse ESCs increases differentiation into extraembryonic endoderm (Li et al., 2013). The pre-miRNA spans the intron–exon junction of the MIR205 host gene (MIR205HG) and is cleaved from unspliced MIR205HG RNA (i.e., pri-miRNA), while the spliced form encodes a lncRNA. Both unspliced and spliced RNAs increased in EXOSC3-depleted hESCs (Fig. 5 B). As the mature miRNA also increased (Fig. 5 C), miR-205 levels are regulated by the exosome.

We also identified a way in which the exosome could contribute to pluripotency by modulating lncRNA levels. Several lncRNAs encoded antisense to the TSSs of coding genes were both targets and up-regulated on exosome depletion (Fig. 5 D and Table S1). Up-regulation of lncRNA loc644656 was associated with increased reads mapping to the first intron of the adjacent ZNF143 (Fig. 5 E), which encodes a transcription factor important for mouse and hESC pluripotency (Chen et al., 2008; Chia et al., 2010). In addition to this ESC-specific role, ZNF143 regulates expression of many mRNAs and ncRNAs (Ngondo-Mbongo et al., 2013) and is important for establishing chromatin interactions at promoters (Bailey et al., 2015). ZNF143 levels are partly maintained by an autoregulatory loop involving an alternative TSS. When ZNF143 is high, it binds downstream of the normal TSS (TSS1), resulting in initiation in the first intron (Fig. 5 E). The resulting TSS2 mRNA is poorly translated (Ngondo and Carbon, 2014). Using siRNAs, we confirmed that exosome depletion results in decreased TSS1 mRNA, increased TSS2 mRNA, and reduced ZNF143 (Fig. 5, F and G). Our data support a model in which exosome degradation of the lncRNA regulates ZNF143 levels.

Numerous mRNAs increased upon EXOSC3 depletion (Table S1). Gene Ontology (GO) analyses revealed that many mRNAs involved in regulation of development and tissue formation increased, while genes associated with maintaining pluripotency were largely unaffected (Fig. 6, A and B). Since our CLIP experiments collectively identified >5,500 mRNAs, we focused on mRNAs that were identified by both HITS-CLIP and PAR-iCLIP as high-confidence targets (1,114 mRNAs; Fig. 6 C). Using this stringent definition, some mRNAs associated with regulation of development were both exosome targets and increased upon exosome depletion (Fig. 6 B, names in bold; and Fig. S4). Consistent with findings that the exosome degrades pre-mRNAs and mature mRNAs (Lubas et al., 2011, 2015), introns, 3′ UTRs, and exons were all targets in HITS-CLIP and PAR-iCLIP (Fig. 6 D).

Exosome targets that encode proteins with critical roles in gene regulation and/or development included FOXH1, which encodes a transcription factor required for anterior–posterior patterning and endoderm development in mice and ME induction in hESCs (Labbé et al., 1998; Hoodless et al., 2001; Yamamoto et al., 2001; Beyer et al., 2013); FOXD1, which encodes a transcription factor important for kidney and retina development (Quintero-Ronderos and Laissue, 2018); SOX12, which functions redundantly with SOX4 and SOX11 in organogenesis (Bhattaram et al., 2010); and CDKN1A, a known target that encodes the cyclin-dependent kinase inhibitor that regulates the G1/S transition. Using RT-qPCR, we confirmed that the mRNAs and pre-mRNAs increase when EXOSC3 mRNA is depleted (Fig. 6, E and F; and Fig. S4). Other mRNA targets that increased upon EXOSC3 depletion encoded ITGB5, an integrin important for keratinocyte proliferation (Duperret et al., 2015), and ZCCHC8, a subunit of the nuclear exosome targeting complex that recruits the exosome to PROMPTs, pre-mRNAs, and pre-snoRNAs (Lubas et al., 2011, 2015). Both ZCCHC8 mRNA and protein increased upon exosome depletion (Fig. S4), consistent with a possible regulatory loop in which decreased exosome activity is compensated by increases in the levels of its cofactors.

To confirm the association of these RNAs with the exosome, we performed UV cross-linking of hESCs, followed by immunoprecipitation with antibodies against EXOSC4, and analyzed RNAs in the immunoprecipitates. RNAs identified as exosome bound by both HITS-CLIP and PAR-iCLIP, including FOXH1, ITGB5, and CDKN1A mRNAs and L1 RNAs, were all enriched in anti-EXOSC4 immunoprecipitates compared with control immunoprecipitates, while NANOG mRNA, which is not a high-confidence exosome target, was not enriched (Fig. 6 G).

Since the exosome restricts hESCs from differentiating (Fig. 2), we examined which targets could potentially mediate this function. Consistent with reports that CDKN1A mRNA is regulated by miRNAs in ESCs (Wang et al., 2008; Dolezalova et al., 2012), we did not detect CDKN1A in EXOSC3-depleted or control hESCs, and the fraction of cells in G1 was unchanged (Fig. 1 E). Although ZNF143 depletion was reported to reduce NANOG transcription in mouse ESCs and cause cells to differentiate (Chen et al., 2008), NANOG mRNA levels were similar in EXOSC3-depleted hESCs and control cells (Figs. S1 C and 6 B), suggesting the reduction in ZNF143 upon exosome depletion (Fig. 5 G) was insufficient to down-regulate NANOG.

We next determined if exosome-mediated degradation of FOXH1 mRNA was important for restraining differentiation. Although FOXH1 protein is present in hESCs, transcription of its target genes is suppressed by a protein complex that sequesters its SMAD2/3 DNA-binding partners (Beyer et al., 2013). Upon induction of differentiation by Nodal/Activin/TGF-β signaling, SMAD2/3 interacts with FOXH1 to induce transcription of genes involved in forming ME, the precursor to the mesoderm and endoderm. Target genes include NODAL, LEFTY1, CER1, and the ME marker MIXL1 (Kim et al., 2011b; Beyer et al., 2013; Reid et al., 2016). As expected if the exosome modulates ME formation by degrading FOXH1 mRNA, expression of each of these FOXH1 targets increased in exosome-depleted EBs (Figs. 2 C and 7 A). To determine if increased levels of FOXH1 are sufficient to activate transcription of mRNAs encoding ME markers, we established stable hESC lines overexpressing FOXH1 (Fig. 7, B and C). Consistent with the hypothesis that increased FOXH1 is sufficient to drive ME formation, CER1, NODAL, and MIXL1 mRNA levels increased in three independent FOXH1-overexpressing clones (Fig. 7 C).

We also confirmed that the increase in FOXH1 mRNA upon exosome depletion was due to reduced degradation. EXOSC3-depleted hESCs were incubated with the transcription inhibitor 5,6-dichlorobenzimidazone-1-β-D-ribofuranoside (DRB) and the decay rates of the pre-mRNA and mRNA measured. As expected, FOXH1 pre-mRNA was stabilized in EXOSC3-depleted hESCs, as its half-life increased to 30.0 ± 4.3 min, compared with 20.8 ± 0.4 min in hESCs treated with control siRNAs (Fig. 7 D). FOXH1 mRNA half-lives were not significantly different between EXOSC3-depleted and control hESCs during the 30-min DRB incubation, indicating that the exosome may primarily exert its effects by degrading the pre-mRNA (Fig. 7 D). Control RNAs behaved as expected, as exosome depletion increased the half-life of a 3′ extended snoRNA but did not affect the decay rate of NANOG pre-mRNA (Fig. 7 D).

To test if exosome regulation of FOXH1 mRNA enhances ME formation, we depleted EXOSC3 and/or FOXH1 in hESCs using siRNAs and incubated the cells with activin A, which induces formation of ME, followed by formation of definitive endoderm (D’Amour et al., 2005). We chose activin A because differentiation markers are detected within hours (D’Amour et al., 2005), making this system less susceptible to indirect effects than EB formation. We confirmed that EXOSC3 depletion from hESCs was >70% effective (Fig. 8 A, compare lanes 2 and 4 with lanes 1 and 3). Although FOXH1 protein was not significantly increased in EXOSC3-depleted hESCs (Fig. 8 A, lane 2), the protein increased 2.3-fold in EXOSC3-depleted cells after activin A addition (lane 6). Examination of FOXH1 targets revealed that CER1, NODAL, LEFTY1, and MIXL1 mRNAs were detectable within 2 h of adding activin A (Fig. 8 B). Importantly, all these mRNAs were significantly elevated in EXOSC3-depleted cells within 4 h (Fig. 8 B). The increased mRNA levels required FOXH1, as the effects were abolished in cells depleted of EXOSC3 and FOXH1 (Fig. 8 B). These results were specific to FOXH1 targets, as two nontarget mRNAs, encoding AFP, a marker of visceral endoderm, and GAPDH, were unaffected (Fig. 8 B). Thus, by degrading FOXH1 mRNA, the exosome restrains ME formation.

To confirm that the exosome inhibits hESC differentiation, we tested if an shRNA-resistant transgene would rescue the phenotypes detected upon exosome depletion. Using hESC lines that express doxycycline-inducible shRNAs against EXOSC3 mRNA as the parent cells, we generated two clonal cell lines that also stably express shRNA-resistant EXOSC3 mRNA. As EXOSC3 mRNA increased two- to fourfold in the shRNA-resistant cells, but the protein remained at or below WT levels (Fig. S5, A and B), EXOSC3 may be unstable when not assembled with other subunits. Exosomes from cells expressing the shRNA-resistant mRNA were functional, as the elevated PROMPTs detected upon EXOSC3 depletion decreased in the presence of the transgene (Fig. S5 C). L1 RNAs, lncRNAs and 3′ extended snoRNAs were also reduced (Fig. S5, C and D). Since depletions of hESCs with siRNAs were slightly more effective than shRNAs, we depleted two hESC lines expressing the shRNA-resistant EXOSC3 transgene with an siRNA identical in sequence to the mature siRNA produced by the shRNA. Using this approach, levels of PROMPTS, L1 RNAs, lncRNAs, and FOXH1 mRNA were all reduced in cells expressing the transgene (Figs. 9 A and S5 E).

We also tested whether the shRNA-resistant EXOSC3 mRNA would restore the role of the exosome in preventing mesendoderm differentiation. To this end, we allowed shEXOSC3-depleted hESCs and two cell lines expressing the shEXOSC3-resistant transgene to differentiate to EBs. As expected, after 5 d of differentiation, EXOSC3-depleted EBs had increased levels of mRNAs encoding FOXH1 targets (MIXL1, NODAL, LEFTY1, and CER1) and other mRNAs that are markers of endoderm (GATA4 and SOX17) and mesoderm (EOMES; Fig. 9 B). As these mRNAs were present at reduced levels in EBs expressing the shRNA-resistant transgene, we conclude that the exosome curtails hESC differentiation.

Although the roles of transcription factors in controlling hESC identity are well known, the ways in which posttranscriptional processes regulate hESC fate are far less understood. We demonstrated that the RNA exosome restrains the differentiation of hESCs into endoderm, mesoderm, and ectoderm. A crucial target is the mRNA encoding FOXH1, a transcription factor required for ME formation.

Our data support a model in which the exosome contributes to hESC pluripotency by degrading RNAs that encode key developmental regulators. ESCs are unique in that they must both self-renew indefinitely and remain poised to differentiate into all three germ layers. Although multiple mechanisms exist to repress transcription of lineage-specific regulators (Young, 2011), the exosome may function to degrade RNAs that escape other silencing pathways. Consistent with a role in maintaining pluripotency, the exosome is rapidly down-regulated when hESCs differentiate to EBs and up-regulated when fibroblasts are reprogrammed to iPSCs.

Although exosome-mediated degradation of FOXH1 mRNA prevents ME induction, the full role of the exosome in restricting hESC differentiation undoubtedly involves other targets. Our finding that miR-205 increases in EXOSC3-depleted cells, coupled with a report that overexpressing this miRNA in mouse ESCs results in increased endoderm markers (Li et al., 2013), suggests that increased levels of this miRNA could contribute to endoderm formation. Moreover, since ectoderm markers also increase in EXOSC3-depleted hESCs, other targets of the exosome must influence formation of this germ layer. Our description of the RNAs affected by exosome depletion may allow identification of these targets.

Our finding that multiple exosome subunits are rapidly down-regulated during differentiation, while mRNAs that encode the subunits are unchanged, implies that down-regulation occurs at the translational and/or posttranslational level. Notably, siRNA-mediated depletion of any of three tested core subunits (EXOSC3, EXOSC4, and EXOSC5) reduces levels of the other two subunits as well as two of the three catalytic subunits (Kammler et al., 2008; Tomecki et al., 2010). Thus, inactivation of a single core subunit (through reduced translation, enhanced decay, or a posttranslational modification that impairs complex assembly) should be sufficient to rapidly down-regulate exosome activity. The ability of the exosome to undergo rapid inactivation may allow it to function as a developmental switch, allowing hESCs to quickly increase levels of mRNAs encoding key transcription factors such as FOXH1. At later times, transcriptional regulation may contribute to reducing exosome activity, since EXOSC3 and EXOSC5 mRNAs were reported to decrease in EBs after 12 d (Fathi et al., 2009).

Our study, together with the finding that the NMD pathway is important for hESC differentiation (Lou et al., 2016), reveals that RNA decay pathways make key contributions to regulating hESC fate. In hESCs, NMD enhances mesoderm formation and limits definitive endoderm, although the mRNA targets responsible were not identified (Lou et al., 2016). Notably, human cells contain many pathways by which defective and potentially harmful mRNAs and ncRNAs can be targeted for decay (Schoenberg and Maquat, 2012; Belair et al., 2018). These pathways are often functionally overlapping, as it is often necessary to deplete multiple pathways to see significant accumulation of a specific target (Belair et al., 2018). Given that hESCs must both maintain the pluripotent state and retain the ability to rapidly differentiate in response to environmental signals, RNA surveillance pathways may function widely to degrade RNAs whose presence compromises hESC function. Although knowledge of how mammalian RNA decay pathways are regulated in response to environmental stimuli is limited, the ability of surveillance pathways to rapidly alter RNA levels may allow them to function as switches in a wide variety of developmental processes.

Our data implicate the exosome in the defense against retrotransposition and revealed new ways in which the exosome modulates gene expression. Although the exosome had been implicated in pre-miRNA surveillance (Liu et al., 2014), we found that the EXOSC10-containing exosome affects miR-205 levels by reducing the pri-miRNA. The identification of 16S mt-rRNA as an exosome substrate opens the door to studying the poly(A) polymerases involved and raises questions as to the subcellular location where decay occurs. Although lncRNAs were known substrates (Flynn et al., 2011; Sigova et al., 2013), EXOSC3 depletion resulted in increased loc644656 lncRNA levels and resulted in a change in TSS usage for the adjacent ZNF143. Given the many antisense RNAs that are exosome targets, this mechanism of regulation could be widespread.

H1 and H9 hESCs were obtained from the Yale Stem Cell Core Facility. Cells were cultured on Matrigel (BD Biosciences)-coated plates under feeder-free conditions at 37°C in 5% CO₂, 5% O₂, and 90% humidity. Media consisted of DMEM/F12 (Invitrogen) supplemented with 1% MEM-nonessential amino acids (Invitrogen), 2 mM L-glutamine, 0.12 mM monothioglycerol, 0.5 mg/ml bovine serum albumin, 50 ng/ml basic fibroblast growth factor (Millipore), N2 supplements (1×), and B27 supplements (1×; Invitrogen). Media were changed every other day. Cells were passaged weekly by dissociating into small clumps with 1 mg/ml dispase (StemCell Technologies). Cells were used between passages 40 and 60 and tested regularly for mycoplasma. Alkaline phosphatase was detected using the Stemgent staining kit (Reprocell).

To establish hESCs stably expressing shRNAs, we used a modified piggyBac DNA transposon (Ding et al., 2005) in which the XbaI/PmeI fragment from pTRIPz (Dharmacon) containing the tetracycline-inducible promoter, the turboRFP reporter, the shRNA cassette, the human ubiquitin C promoter, the reverse tetracycline-transactivator 3, an internal ribosome entry site, a puromycin selection marker, and the woodchuck hepatitis posttranscriptional regulatory element was cloned between the left and right piggyBac repeat (PBL and PBR, respectively) transposon termini in pBlueScript II SK+ (Stratagene). A rabbit β-globin poly(A) signal was added to ensure transcription termination (Ding et al., 2005). To create the EXOSC3 shRNA-expressing cassette, pTRIPz containing this shRNA (V3THS_306405; Dharmacon) was digested with NheI and XhoI and the shRNA-containing fragment was cloned into the same sites of the piggyBac construct. H1 ESCs were cotransfected with the shEXOSC3 piggyBac construct or a similar one expressing nonsilencing shRNA (gift of T. Xu, Yale University, New Haven, CT) together with the plasmid Act-PBase (Ding et al., 2005), which contains piggyBac transposase under control of the ACTB promoter. After 72 h, stable lines were selected in media containing 0.5 µg/ml puromycin. After 7 d selection, clonal lines were cultured in 0.8 µg/ml puromycin. For each cell line, shRNA transcription was induced with 2 µg/ml doxycycline and TurboRFP expression monitored by epifluorescence.

To stably overexpress FOXH1 in hESCs, FOXH1 cDNA was amplified by RT-PCR from H1 ESC RNA using the Superscript III First Strand Synthesis System (Thermo Fisher) and the oligonucleotides listed in Table S2, digested with MluI, and cloned into a piggyBac transposon-based vector containing an ACTB promoter and a blasticidin S resistance cassette (Gayle et al., 2015; gift of T. Xu). H1 ESCs were cotransfected with either the FOXH1 or empty piggyBac together with the plasmid Act-PBase using the P4 Primary Cell 4D-Nucleofector X Kit (Program CA137; Lonza). After 72 h, stable cell lines were selected in media containing 2 µg/ml Blasticidin. After 7 d, single colonies were isolated and analyzed.

To express the shEXOSC3 RNA-resistant transgene, EXOSC3 cDNA was amplified from the plasmid pCMV6-EXOSC3 (RC200035; ORIGENE) with oligonucleotides EXOSC3-MluIF and EXOSC3-MluIR (Table S2), digested with MluI, and cloned into the piggyBac vector used to overexpress FOXH1. Silent mutations were introduced using the Q5 site-directed mutagenesis kit (New England Biolabs) using oligonucleotides EXOSC3-mut-F and EXOSC3-mut-R (Table S2). H1 ESCs were cotransfected with the shRNA-resistant EXOSC3 in the piggyBac vector or the empty vector together with the plasmid Act-PBase using the P4 Primary Cell 4D-Nucleofector X Kit (Program CA137). After 72 h, stable cell lines were selected in media containing 2 µg/ml Blasticidin. After 7 d, single colonies were isolated and analyzed.

To produce EBs, hESCs were dissociated into clumps with 1 mg/ml dispase (StemCell Technologies) and cultured in DMEM/F12 media supplemented with 20% KnockOut Serum Replacement (Life Technologies), 2 mM L-glutamine, 0.1 mM MEM-nonessential amino acids, and 0.1 mM β-mercaptoethanol at 37°C in 5% CO₂, 21% O₂, and 90% humidity. Media were changed every other day. To form endoderm, cells were incubated with 100 ng/ml Activin A (Peprotech) in RPMI-1640 medium containing 0.2% FBS, 2 mM L-glutamine, and 0.1 mM MEM-nonessential amino acids at 37°C in 5% CO₂, 21% O₂, and 90% humidity.

To derive iPSCs, primary fibroblast cell lines were purchased from the Coriell Cell Repository and cultured in DMEM supplemented with 15% fetal bovine serum, 1 mM L-glutamine, 1% nonessential amino acids, and penicillin/streptomycin. To drive reprogramming to iPSCs, the cells were transduced with four retroviral reprogramming factors (OCT4, SOX2, KLF4, and MYC; Park et al., 2008; Kim et al., 2011a). The iPSCs were shown to have all features of pluripotent hESCs, including the ability to differentiate into all three germ layers. The iPSCs were maintained on Matrigel in mTeSR1 medium (StemCell Technologies). To monitor decay, hESCs were incubated with 50 µM DRB (Sigma) dissolved in DMSO. At the indicated times, TRIzol reagent (Thermo Fisher) was added to the culture dish and the RNA was extracted.

Cells were transfected with siRNAs and plasmids as described previously (Ma et al., 2010). Briefly, after incubating for 1 h in media supplemented with 10 µM ROCK inhibitor Y-27632 (Selleckchem), hESC colonies were dissociated into single cells with Accutase (StemCell Technologies) and cells were harvested by centrifugation. After resuspending 2 × 10⁶ cells in 200 µl siRNA (10 nM)/lipid mix and incubating 10–15 min at room temperature, 3 ml of prewarmed media containing 10 µM ROCK inhibitor Y-27632 was added, and the cells were transferred into a 60-mm Matrigel-coated dish and incubated overnight at 37°C in 5% CO₂, 5% O₂, and 90% humidity. Media were replaced the next day. For siRNAs, the first transfection was performed in duplicate, and the two sets of transfected cells were pooled for a second transfection after 48 h. Cells were harvested 72 h after the second transfection. A list of siRNAs is in Table S2. For plasmids, a single round of transfection was performed using 1.5 µg DNA.

After incubating hESCs in fresh media containing 10 µM BrdU for 25 min, hESC colonies were dissociated into single cells with Accutase and fixed in 70% ethanol at 4°C overnight. Fixed cells were pelleted and incubated in 1 ml of 2 N HCl and 0.5% Triton X-100 for 30 min. After pelleting, cells were resuspended in 0.1 M Na₂B₄O₇, pH 8.5, pelleted, and resuspended in 75 µl of a mixture of 65 µl PBS, 0.5% Tween 20, 1% BSA, and 10 µl FITC-conjugated mouse anti-BrdU antibodies (clone 3D4; BD Biosciences) and incubated overnight at 4°C. Cell pellets were collected and resuspended in 500 µl PBS supplemented with 1 µg/ml DAPI for ≥30 min. Afterward, fluorescence was analyzed on a LSRII flow cytometer at the Yale FACS facility. To examine viability, hESC colonies were dissociated into single cells with Accutase, mixed with 1 vol Trypan Blue 0.4% Solution (Gibco), and counted with a hematocytometer.

RNA was extracted with TRIzol reagent. For Northern blots, RNA was fractionated in 7% polyacrylamide-7 M urea gels and transferred to Hybond-N (GE Healthcare) in 0.5× TBE at 150 mA for 16 h. Membranes were hybridized at 42°C in 180 mM Na₂HPO₄, 80 mM NaH₂PO₄, 7% SDS, and 1 mM EDTA with 5′-³²P-labeled oligonucleotides. Blots were analyzed using a phosphorimager (Typhoon FLA 7000; GE Healthcare). For RT-qPCR, cDNA was synthesized with the Superscript III First-strand cDNA synthesis system (Thermo Fisher) and subjected to quantitative PCR with iTaq Universal SYBR Green Supermix (BioRad). To measure poly(A)-tailed RNAs, cDNA was made using Oligo(dT)_12-18 and Superscript III Reverse transcription (Invitrogen). Probes and primers are listed in Table S2.

After resuspending cells in lysis buffer (60 mM Tris-HCl, pH 7.5, 10% glycerol, and 2% SDS) with 1 mM PMSF and 1× protease inhibitor cocktail (Roche Diagnostics), incubating for 10 min at 4°C, and passing 10–15 times through a 27-gauge needle, lysates were cleared by sedimenting at 20,000 ×g for 15 min at 4°C. After SDS-PAGE, proteins were transferred to Immobilon-P (Millipore). Antibodies are listed in Table S2.

HITS-CLIP was performed largely as described previously (Darnell, 2012). Briefly, after irradiating H1 cells with 600 mJ/cm², cells were harvested in cold 1× Dulbecco’s PBS (Gibco) and sedimented at 300 ×g. After resuspending pellets in 400 µl lysis buffer (1× PBS, 0.1% SDS, 0.5% deoxycholate, and 0.5% NP-40) with EDTA-free protease inhibitors (Roche) and 1 mM PMSF, incubating 10 min on ice, adding 50 U DNase I (Roche), and incubating at 37°C for 10 min, EDTA was added to 1 mM final concentration and lysates were cleared by sedimenting twice at 20,000 ×g for 10 min at 4°C. Antibodies (10 µg) were conjugated with 50 µl Dynabeads Protein G (Life Technologies) for 1 h at 4°C in 400 µl lysis buffer. After preclearing for 10 min at 4°C with 50 µl Dynabeads Protein G, lysates were incubated with antibody-conjugated beads for 1 h at 4°C. Beads were washed three times with PXL + E (5× PBS, 0.1% SDS, 0.5% deoxycholate, 0.5% NP-40, and 1 mM EDTA), three times with H-RIPA (50 mM Tris-HCl pH 7.5, 400 mM NaCl, 1% NP-40, 1% deoxycholate, 0.1% SDS, 1 mM EDTA, and 0.2 M urea), and three times with PNK buffer (50 mM Tris-HCl, pH 7.5, 10 mM MgCl₂, and 0.5% NP-40). Beads were resuspended in 500 µl MNase buffer (50 mM Tris-HCl, pH 8, and 5 mM CaCl₂) with 20 U or 0.2 U micrococcal nuclease and incubated 1 min at 37°C, followed by 10 min on ice. After washing twice in PNK + EGTA (50 mM Tris-HCl, pH 7.5, 20 mM EGTA, and 0.5% NP-40) and twice in PNK, beads were treated with 2 U rAPid alkaline phosphatase (Roche) for 10 min at 37°C and washed twice each with PNK + EGTA and PNK buffers. Next, 3′ linkers were ligated to the RNA overnight at 4°C in 40 µl containing 80 pmol RL3 RNA linker, 4 µg BSA, 1 mM ATP, 40 U RNaseOUT (Life Technologies), and 10 U T4 RNA ligase. After two washes in PXL and PNK buffers, RNAs on beads were 5′-labeled with [γ-³²P]ATP and 20 U T4 polynucleotide kinase in PNK buffer for 30 min at 37°C, 5 µl ATP (1 mM) was added, and the reaction was incubated an additional 10 min at 37°C. After two washes in H-RIPA and PNK buffers, RNPs were eluted by heating beads at 95°C twice for 5 min each in elution buffer (2× NuPage LDS sample buffer [Life Technologies] in PNK + EGTA). Complexes were resolved in a NuPage Novex 10% Bis-Tris gel (Life Technologies) and transferred to BA-85 nitrocellulose (Sigma-Aldrich). Bands corresponding to proteins cross-linked to ∼50 nt RNAs were excised and digested with 4 mg/ml Proteinase K in 200 µl PK buffer (100 mM Tris-HCl, pH 7.5, 50 mM NaCl, and 10 mM EDTA) for 20 min at 37°C. After adding 200 µl of Proteinase K/7M urea in PK buffer, incubating for another 20 min at 37°C, an equal volume of phenol/chloroform/isoamyl alcohol (25:24:1) was added and incubated for 20 min at 37°C. After removing the aqueous phase, RNA was precipitated with 1 ml of 1:1 ethanol/isopropanol. Afterward, 5′ linkers were ligated for 1 h at 16°C in 10 µl containing 20 pmol RL5 RNA linker, 1 µg BSA, 1 mM ATP, and 1 U T4 RNA ligase. After adding 50 U RQ1 DNase I (Roche) and incubating for 20 min, RNA was isolated by extracting with phenol/chloroform/isoamyl alcohol (25:24:1) and precipitated with 2.5 vol of 1:1 ethanol/isopropanol. Next, RNA was resuspended in 10 µl H₂O containing 10 pmol primer DP3 and 10 μmol dNTPs, heated at 65°C for 5 min, and reverse transcribed with 200 U Superscript III (Life Technologies) for 45 min at 50°C and 55°C for 15 min. After heating at 90°C for 5 min, PCR was performed using AccuPrime Pfx SuperMix (Life Technologies), 15 pmol each of primers DP3 and DP5, and 2 µl cDNA. After gel-purifying ∼110–150 bp DNAs and multiplexing by performing PCR with AccuPrime Pfx SuperMix, 10 pmol DSPF5 primer, 10 pmol 3′ index primer, and 1 µl DNAs, DNAs were gel purified and subjected to 75-bp single-end Illumina HiSeq 2000 sequencing at the Yale Center for Genome analysis. Antibodies, RNA linkers, and PCR primers are listed in Table S2.

PAR-iCLIP was performed as described previously (Huppertz et al., 2014). Briefly, after incubating hESCs with 100 µM 4-thiouridine for 7 h, cells were irradiated with 800 mJ/cm² UV light (365 nm). After sonication in 50 mM Tris-HCl, pH 7.4, 100 mM NaCl, 0.1% SDS, 0.5% deoxycholate, 1% NP-40, and 1 mM EDTA, lysates were incubated with 20 U Turbo DNase (Invitrogen) and 100 U RNase I (Invitrogen) at 37°C for 10 min, followed by immunoprecipitation with anti-EXOSC4 antibody (A303-775A; Bethyl) conjugated to Dynabeads Protein G (Invitrogen). Beads were washed with high-salt wash buffer (50 mM Tris-HCl, pH 7.4, 1 M NaCl, 0.1% SDS, 0.5% deoxycholate, 1% NP-40, and 1 mM EDTA) and incubated with T4 PNK (NEB), followed by 3′ linker ligation. After labeling RNA with ³²P, the protein–RNA complex was fractionated in a 10% NuPAGE Bis-Tris gel (Invitrogen) and transferred to nitrocellulose. After autoradiography, protein–RNA complexes were excised and RNAs were eluted from the membrane. After reverse transcription using Superscript III (Invitrogen) and primers containing 9 nt barcodes (4 nt for sample discrimination and 5 nt to identify PCR duplicates), cDNAs were gel purified, circularized with Circligase II (Epicentre), and cut using BamHI and a BamHI site–containing complementary oligonucleotide. Linearized cDNA was amplified using Phusion High-Fidelity PCR Master Mix (NEB) and sequenced on Illumina NEXTSeq.

For RNA-seq, RNA integrity was verified on a 2100 Bioanalyzer (Agilent Technologies), ERCC ExFold RNA Spike-In mix (Life Technologies) was added, and rRNAs were removed using Ribo-Zero (Epicentre). Libraries were prepared using ScriptSeq v2 RNA-seq Library Preparation kits (Epicentre) and multiplexed with ScriptSeq Index PCR primers (Epicentre). Libraries were submitted for 50- or 75-bp single-end Illumina HiSeq 2000 sequencing at the Yale Center for Genome Analysis and the Yale Stem Cell Genomics Core Facility. 25–50 million reads were obtained per library.

Cells were irradiated and immunoprecipitated as described for HITS-CLIP, except that 1 U/µl RNaseOUT recombinant ribonuclease inhibitor (Thermo Fisher) was added to the lysis buffer, and RNase was omitted. After incubating 1 h at 4°C with anti-EXOSC4 antibodies or control IgG, beads were washed five times with 5× PXL containing 1 mM EDTA, H-RIPA, and PNK+ buffers. RNAs were eluted by incubating the beads for 20 min at 37°C in PK buffer containing 4 mg/ml proteinase K and adding 200 µl 7 M urea PK buffer, followed by a second incubation for 20 min at 37°C, phenol/chloroform extraction, and ethanol precipitation. 5% of the lysate was subjected directly to phenol/chloroform extraction and ethanol precipitation. Following DNase I treatment, cDNA was prepared using SuperScript III First-Strand Synthesis System (Thermo Fisher) and subjected to quantitative PCR with iTaq Universal SYBR Green Supermix (BioRad). Cq (quantitation cycle) values were adjusted to the total quantity of RNAs obtained for each sample (input, IgG, and EXOSC4), and the fold enrichment over IgG control was calculated using the ΔCq method. A Cq value of 40 was used when no signal was detected during quantitative PCR.

After RNA-seq, reads were processed using the FASTX-Toolkit to remove low-quality, short (<16 nt) reads and adapters. After removing reads mapping to rRNAs, >98% of reads aligned to the human genome (hg19/GRCh37) using TopHat (Kim et al., 2013). Differential expression analyses were performed using Cufflinks (Trapnell et al., 2010). Genome coverage was done using Bedtools (genomeCoverageBed; Quinlan and Hall, 2010). As the number of reads that mapped to the genome differed between samples, a scale was applied for each library. This scale corresponds to the ratio between the total number of reads mapped to the genome for each library and the lowest number obtained from each set of libraries (5 or 7 d). RNA-seq reads for repeat elements were quantified using Salmon (Patro et al., 2017). Consensus sequences of repeat elements in the human genome were downloaded from the Repbase database (Bao et al., 2015) and used as reference transcripts for counting.

Gene expression levels were estimated as reads per kilo million by Cufflinks (v1.2.1) using RefSeq genes as the reference (Trapnell et al., 2010). GO analysis was then performed on genes whose difference in expression between doxycycline- treated and untreated cells was >1.5 fold, with GO stats (v2.24.0) library in Bioconductor (Falcon and Gentleman, 2007). The P value for each GO term was adjusted by the Benjamini–Hochberg method with the p.adjust function in R to obtain the false discovery rate. For the heat map, genes were selected from the list of differentially expressed genes which had GO terms related to cell development (e.g., muscle organ development).

HITS-CLIP and PAR-iCLIP reads were preprocessed using Cutadapt (Martin, 2011) and FASTX-Toolkit to remove adaptors and low-quality reads, collapsed to remove duplicates, and split by barcodes. Reads were then aligned to the hg19 genome using Novoalign. Since CLIP experiments can generate multiple mismatches and deletions within reads, the mapping stringency was lowered (-t 15,3 -l 20 -x 4 -g 20). Reads that could be mapped to multiple loci in the genome were allowed to map once to a randomly chosen locus with the best mapping score. For the coverage pie charts, reads were counted for each category using Bedtools (Quinlan and Hall, 2010) based on annotations from the University of California Santa Cruz genome browser. For L1Hs, libraries were mapped against a consensus sequence from Repbase (Bao et al., 2015). Read clusters for PAR-iCLIP were analyzed using Paralyzer v1.5 (Corcoran et al., 2011).

Data are shown as means ± SEMs, with the number of replicates in parentheses. Significances of differences between samples were evaluated using one-way ANOVA or two-tailed unpaired t test. P values <0.05 were considered to represent significant differences. P values are represented as *, **, and ***, corresponding to P ≤ 0.05, P ≤ 0.01, and P ≤ 0.001, respectively.

The data associated with this manuscript have been deposited to the Gene Expression Omnibus and are available under accession no. GSE132469.

Fig. S1 shows the effects of EXOSC3 depletion on proliferation and pluripotency markers. Fig. S2 shows additional evidence that mRNAs encoding differentiation markers increase in EXOSC3-depleted hESCs and EBs, and that exosome subunits are down-regulated when hESCs are differentiated to form EBs. Fig. S3 shows the autoradiograms from the HITS-CLIP and PAR-iCLIP experiments and the effects of exosome depletion on snoRNA, snRNA, and telomerase RNA targets. Fig. S4 provides additional data that FOXH1, CDKN1A, and ZCCHC8 mRNAs are exosome targets. Fig. S5 shows the characterization of hESCs expressing an shRNA-resistant EXOSC3 transgene. Table S1 reports the RNA-seq data for mRNAs and ncRNAs, HITS-CLIP and PAR-iCLIP tag counts, and PAR-iCLIP clusters. Table S2 lists the primers, oligonucleotide probes, antibodies, and siRNAs used in this study.

We thank C. Qiu and the Yale Stem Cell Center hESC Facility for advice; J. Goodier (Johns Hopkins University School of Medicine, Baltimore, MD) and T. Xu for reagents; and A. Alexandrov, X. Chen, M. Boccitto, Y. Leng, and H. Park for comments on the manuscript.

This work was supported by the Connecticut Stem Cell Research grants program (grant 13-SCB-YALE-06 to I.-H. Park and grants 09-SCA-YALE-12 and 11-SCB-YALE-19 to S.L. Wolin) and by National Institutes of Health grant R01 GM111667 to I.-H. Park and grant R01 GM073863 to S.L. Wolin. C. Belair was a Berbecker Scholar in Molecular Medicine. This work was also funded by the National Institutes of Health, National Cancer Institute, Center for Cancer Research Intramural Research Program (C. Belair, S. Sim, and S.L. Wolin).

The authors declare no competing financial interests.

Author contributions: C. Belair and S.L. Wolin conceived the study. K.-Y. Kim, Y. Tanaka, and I.-H. Park provided iPSC pellets, performed GO, and generated the heat map. C. Belair and S. Sim performed all other experiments. C. Belair, S. Sim, and S.L. Wolin analyzed data and wrote the manuscript. All authors provided feedback on the manuscript.