Recent studies have boosted our understanding of long noncoding RNAs (lncRNAs) in numerous biological processes, but few have examined their roles in somatic cell reprogramming. Through expression profiling and functional screening, we have identified that the large intergenic noncoding RNA p21 (lincRNA-p21) impairs reprogramming. Notably, lincRNA-p21 is induced by p53 but does not promote apoptosis or cell senescence in reprogramming. Instead, lincRNA-p21 associates with the H3K9 methyltransferase SETDB1 and the maintenance DNA methyltransferase DNMT1, which is facilitated by the RNA-binding protein HNRNPK. Consequently, lincRNA-p21 prevents reprogramming by sustaining H3K9me3 and/or CpG methylation at pluripotency gene promoters. Our results provide insight into the role of lncRNAs in reprogramming and establish a novel link between p53 and heterochromatin regulation.
Induced pluripotent stem cells (iPSCs) have major implications for regenerative medicine, in vitro disease modeling and toxicology screening1. Yet, to fulfill these expectations, it is necessary to understand how the reprogramming machinery erases and rewrites the somatic epigenetic code. This is important because, among other reasons, reprogramming is prone to errors that could hamper biomedical applications of the derived cells2.
Reprogramming by exogenous factors is nowadays considered as a multistep process in which cells must overcome a series of roadblocks to ultimately acquire pluripotency3,4, though under certain conditions it can also be deterministic5,6,7. Two fundamental reprogramming roadblocks are the activation of a p53-mediated response that promotes apoptosis and cell senescence, and the inefficient removal of preexisting somatic-like epigenetic marks. Accordingly, depletion of p538 or the H3K9 methyltransferases Setdb1, Suv39h1 and Suv39h29,10,11, and inhibition of the maintenance DNA methyltransferase DNMT112 improve reprogramming efficiency.
Noncoding RNAs have important roles in establishing or removing different reprogramming roadblocks. However, despite the extensive characterization of microRNA (miRNA) networks in reprogramming13, only 2 long noncoding RNAs (lncRNAs) are known to modulate this process. LncRNAs are noncoding RNA species longer than 200 nucleotides and act as decoys, guides or scaffolds14. LincRNA-RoR was the first lncRNA implicated in reprogramming15 and partly functions as a 'sponge' that titrates down miRNAs targeting pluripotency regulators (e.g., miR-145)16. Conversely, lincRNA TUNA forms a complex with several RNA-binding proteins at the promoters of Nanog, Sox2 and Fgf4, where it facilitates the deposition of the active histone mark H3K4me317.
In this report, we identify that the p53-induced large intergenic noncoding RNA p21 (lincRNA-p21; also termed Trp53cor1)18,19,20 acts as a barrier for reprogramming by sustaining heterochromatic state at pluripotency gene promoters.
Functional screening reveals roles of lincRNAs in the preiPSC to iPSC conversion
Somatic cells undergoing reprogramming can be trapped in a partially reprogrammed state termed the preiPSC state21. These cells have overcome the senescence barrier and downregulated multiple somatic markers, but fail to fully activate the pluripotency network. Importantly, preiPSCs are stable and can be readily converted to fully reprogrammed iPSCs through a variety of approaches10,21. Because reprogramming cells consist of heterogeneous populations progressing at different speeds3, preiPSCs provide a relatively well-defined model for dissecting the chain of events eventually leading to the full acquisition of pluripotency10,11. Using preiPSCs, we thus sought to identify lncRNAs that control the late stage of mouse somatic cell reprogramming.
We selected 51 lincRNAs previously reported to regulate embryonic stem cell (ESC) pluripotency (26) or differentiation (25)22 and 16 p53-regulated lincRNAs18 (Figure 1A). The rationale for choosing the latter group is that p53 is an important regulator of ESC differentiation and a major barrier for reprogramming8. We performed quantitative PCR (qPCR) analysis of these 67 lincRNAs in mouse embryonic fibroblasts (MEFs), 4 independent preiPSC clones (generated from Oct4-GFP transgenic MEFs21 using Oct4, Sox2, Klf4 and c-Myc; OSKM), iPSCs and ESCs. The analysis showed distinct patterns of lincRNA expression in the 4 cell types that were grouped into 7 classes (Figure 1B, 1C and Supplementary information, Table S1A). Classes I and II were the largest and contained 17 and 26 lincRNAs, respectively, while class VI included only 1 lincRNA (p53-regulated) (Supplementary information, Table S1A). In contrast to classes I-VI, class VII lincRNAs did not show a consistent pattern among the different cell types. Of note, pluripotency- and differentiation-related lincRNAs were distributed evenly among classes I-IV and VII, but only 1 differentiation-related lincRNA was included in class V. Conversely, p53-regulated lincRNAs were mostly in classes II (8) and V (comprising 4 out of 5 constituents). We also performed qPCR analysis for all 67 lincRNAs in MEFs reprogrammed with OSKM (Supplementary information, Figure S1 and Table S1B), and observed that 58 out of the 67 lincRNAs increase by at least 2-fold during reprogramming (Supplementary information, Table S1B).
We then selected 15 lincRNAs from classes I-VI for knockdown and functional analysis in a validated preiPSC to iPSC conversion model10. We generated 2 individual short hairpin RNA (shRNA) viral vectors for each of the 15 lincRNAs, and identified at least 1 effective shRNA (> 50% depletion) for 11 out of the 15 lincRNAs (Figure 1D, upper panel). Activation of the Oct4-GFP reporter was employed as readout for detecting the full acquisition of pluripotency. After infection with the shRNAs, the preiPSCs were passaged onto feeders and treated with vitamin C (Vc) for 8 days before GFP+ colony counting10. Notably, knockdown of 3 lincRNAs (-1463, -1526, and -p21) with the 2 independent shRNAs showed enhanced GFP+ colony formation compared to the control, while lincRNA-1307 knockdown had the opposite effects (Figure 1D, lower panel and 1E). In summary, our screening successfully identified 4 lincRNAs that modulate the preiPSC to iPSC conversion.
p53-induced lincRNA-p21 impairs reprogramming independent of apoptosis and cell proliferation
Among the 4 lincRNAs identified above, we considered lincRNA-p21 particularly interesting because of its expression profile (class V: higher in preiPSCs compared to iPSCs/ESCs and MEFs) and p53-dependent regulation. LincRNA-p21 mediates the apoptotic effects of p53 upon DNA damage18, although we did not detect any change in apoptosis or cell proliferation during the conversion of preiPSCs to iPSCs when lincRNA-p21 was knocked down (data not shown). Of note, p53 protein levels were higher in preiPSCs compared with MEFs and iPSCs/ESCs (Figure 2A). This was not due to increased p53 mRNA expression (Supplementary information, Figure S2A), indicating that p53 is likely regulated at the post-transcriptional level in this setting. Intriguingly, p53 knockdown similarly enhanced the conversion of preiPSCs to iPSCs (Figure 2B) without influencing apoptosis or proliferation (Supplementary information, Figure S2B and S2C). These findings suggest that p53 and lincRNA-p21 act in the same pathway, independent of apoptosis or proliferation, to sustain the preiPSC state.
We also detected that lincRNA-p21 expression increases significantly (from ∼20 to ∼300 copies per cell) during the reprogramming of MEFs with OSKM (Figure 2C and Supplementary information, Figure S2D and Table S1B). This induction was p53-dependent, as assessed by cotransduction of p53 shRNAs and OSKM into wild-type MEFs and the reprogramming of p53−/− tail tip fibroblasts with OSKM (Figure 2C and Supplementary information, Figure S2E). We then tested the effect of knocking down lincRNA-p21 whilst simultaneously reprogramming Oct4-GFP MEFs with OSKM. This led to a higher number of GFP+ colonies (∼4-6-fold) compared to the control without noticeably influencing apoptosis or cell proliferation, as opposed to the effect of p53 knockdown (Figure 2D-2F). Besides, we observed that lincRNA-p21 depletion does not exhibit synergy with p53 depletion in either cell proliferation or reprogramming efficiency (Supplementary information, Figure S3A and S3B). Of note, the resulting iPSC colonies produced with lincRNA-p21 knockdown were fully reprogrammed, as demonstrated by standard characterization procedures including the generation of chimeric mice (Figure 2G and Supplementary information, Figure S3C and S3D). Likewise, they had integrated the shRNA vectors into their genome (Supplementary information, Figure S3E) and displayed a normal karyotype (Figure 2H). Moreover, we observed that overexpressing lincRNA-p21 impairs OSKM reprogramming efficiency (Figure 2I) and diminishes the enhancing effect of p53 knockdown in a similar manner (Figure 2J). In addition, overexpressing a shRNA-resistant lincRNA-p21 with OSKM reduced the enhanced reprogramming efficiency achieved by lincRNA-p21 knockdown (Figure 2K), ruling out an off-target effect of the shRNA. Therefore, p53-induced lincRNA-p21 acts as a barrier for reprogramming independent of apoptosis and cell proliferation.
LincRNA-p21 associates with SETDB1 and DNMT1 in separate repressor complexes facilitated by HNRNPK
During the DNA damage response, lincRNA-p21 interacts with and directs the RNA-binding protein HNRNPK to specific genomic loci that are subsequently silenced by an unknown mechanism18. We thus speculated that lincRNA-p21 might interact with alternative partners in reprogramming cells to regulate a distinctive set of genes. Specifically, we envisaged that these partners would be chromatin modifiers with a repressive role in reprogramming. To explore this idea, we first performed cellular fractionation to analyze the subcellular localization of lincRNA-p21 during the reprogramming of MEFs with OSKM. LincRNA-p21 was predominantly nuclear (Figure 3A), consistent with a potential role in epigenetic regulation. Then, we employed RNA immunoprecipitation (RIP) with antibodies against a panel of chromatin modifiers detrimental for reprogramming9,10,11,12,23 using lysates from MEFs reprogrammed with OSKM on day 12. Notably, there was a substantial enrichment for lincRNA-p21 in DNMT1 RIP and a less substantial enrichment in SETDB1 RIP, but no enrichment in DNMT3A/B and SUV39H1 RIPs, relative to IgG control (Figure 3B). The enrichment for lincRNA-p21 in DNMT1 and SETDB1 RIPs was also present at an earlier phase of OSKM reprogramming (Supplementary information, Figure S4A). We also observed that both SETDB1 and DNMT1 increase during OSKM reprogramming (Supplementary information, Figure S4B).
To confirm the above interactions, we performed RNA pull down with in vitro transcribed biotinylated lincRNA-p21 and lacZ control using nuclear extracts of MEFs reprogrammed with OSKM. Yet, despite a robust interaction with HNRNPK, we failed to detect SETDB1 or DNMT1 by western blotting (Supplementary information, Figure S4C). This led us to speculate that the interaction between lincRNA-p21, SETDB1 and DNMT1 is facilitated by another protein, potentially HNRNPK. To demonstrate this, we conducted immunoprecipitation of endogenous SETDB1 and DNMT1 in MEFs reprogrammed with OSKM on day 12, and observed that endogenous HNRNPK could be pulled down in both cases (Figure 3C). This interaction was confirmed by reverse immunoprecipitation using antibodies against HNRNPK (Figure 3D). However, we did not detect interaction between SETDB1 and DNMT1, indicating that each of them likely associates separately with HNRNPK (data not shown). We also performed an in vitro interaction assay with purified HNRNPK, SETDB1 and lincRNA-p21. The result showed that SETDB1 interacts with lincRNA-p21 in an HNRNPK-dependent manner (Figure 3E), though we cannot exclude the possibility that additional proteins present in the lysates facilitate the interaction. Besides, we observed decreased enrichment for lincRNA-p21 in SETDB1 or DNMT1 RIPs upon Hnrnpk knockdown in MEFs reprogrammed with OSKM on day 12 (Figure 4A and 4B). Yet, this was not associated with reduced SETDB1 or DNMT1 protein or mRNA expression (Supplementary information, Figure S4D and S4E), which reinforces the notion that HNRNPK facilitates the interaction between SETDB1/DNMT1 and lincRNA-p21 to impair reprogramming. Supporting this, we observed that Hnrnpk expression increases in reprogramming (Supplementary information, Figure S4F and S4G). Moreover, Hnrnpk knockdown using shRNAs enhanced reprogramming efficiency (∼5-6-fold) (Figure 4C) and its overexpression blocked reprogramming (Figure 4D). Likewise, the effect of Hnrnpk knockdown was not synergistic with the effect of lincRNA-p21 knockdown (Figure 4E), further arguing that Hnrnpk and lincRNA-p21 act in the same pathway. In addition, we noticed enhanced conversion of preiPSCs to iPSCs with Hnrnpk knockdown upon treatment with Vc (Figure 4F). Altogether, these experiments indicate that lincRNA-p21 and SETDB1 or DNMT1 associate in separate complexes in reprogramming cells, which are facilitated by HNRNPK.
LincRNA-p21 sustains H3K9me3 and CpG methylation at pluripotency genes
To identify the genes targeted by lincRNA-p21 during reprogramming and better understand the global interaction with HNRNPK, we performed RNA-sequencing (RNA-seq) of MEFs reprogrammed with OSKM with or without lincRNA-p21 or Hnrnpk knockdown (2 independent shRNAs for each target and 2 shRNA control) at 3 different time points. Notably, most of the genes differentially regulated by lincRNA-p21 knockdown at each time point (compared with the control) were among those genes differentially regulated by Hnrnpk knockdown (Figure 5A, 5B and Supplementary information, Figure S5A and Table S2). This result indicated that lincRNA-p21 mainly exerts its role in reprogramming through HNRNPK. KEGG pathway analysis of genes upregulated by lincRNA-p21 or Hnrnpk knockdown included the p53 signaling pathway (Supplementary information, Figure S5B). In addition, principal component analysis of the RNA-seq time course similarly confirmed the direction of global gene expression, with a dramatic divergence between the control and the lincRNA-p21, as well as Hnrnpk knockdown (Figure 5C).
Importantly, mesenchymal and epithelial genes showed no difference upon lincRNA-p21 or Hnrnpk knockdown in the RNA-seq time course compared with the control (Supplementary information, Table S2), which was validated by qPCR (Supplementary information, Figure S6A). We did not observe any change in the cell cycle repressor p21 in the RNA-seq time course either (Supplementary information, Table S2), which supports our data that lincRNA-p21 knockdown does not increase cell proliferation in the preiPSC to iPSC conversion or in reprogramming. However, many pluripotency regulators such as Esrrb, Lin28a, Utf1, Nr5a2, Sall4, Lefty2 and Nanog were significantly upregulated by lincRNA-p21 or Hnrnpk knockdown (Figure 5B and Supplementary information, Table S2). We performed qPCR analysis to confirm that the pluripotency network is severely affected by lincRNA-p21 knockdown (Figure 5D). Interestingly, Lin28 had been previously described as a target for lincRNA-p21 in MEFs subjected to DNA damage18. As a major function of LIN28 is to block the maturation of let-7 miRNAs, we measured mature let-7 miRNAs by qPCR and observed that they were downregulated (Supplementary information, Figure S6B).
Next, we employed chromatin immunoprecipitation (ChIP) to assess the binding of HNRNPK to the promoters of a panel of pluripotency genes in MEFs reprogrammed with OSKM on day 12. There was a significant enrichment for HNRNPK compared to the IgG control in most of the selected targets (Supplementary information, Figure S7A), and this was reduced in MEFs cotransduced with shRNA for lincRNA-p21 compared to control shRNA (Figure 6A). A similar pattern was observed when we analyzed the binding of SETDB1 and DNMT1 to the promoters of the same panel of genes (Figure 6B and 6C), though DNMT1 binding to Nanog was more enriched compared to SETDB1. Likewise, lincRNA-p21 knockdown reduced H3K9me3 enrichment in the same genes bound by HNRNPK (except for Nanog) compared to the shRNA control (Figure 6D). In addition, Setdb1 and lincRNA-p21 knockdown produced a similar reduction in H3K9me3 at target promoters and their combination was not synergistic (Figure 6D). On the other hand, analysis of CpG methylation by bisulfite sequencing demonstrated that lincRNA-p21 and Dnmt1 knockdown accelerate the demethylation of the Nanog promoter compared to the control and their combination is not synergistic (Figure 6E). However, lincRNA-p21 knockdown had little effect on CpG methylation of the Oct4 promoter (Supplementary information, Figure S7B), in agreement with the observation that DNMT1 binding to the Oct4 promoter was not affected by lincRNA-p21 knockdown (Figure 6C). We could not detect significant methylation in the Lin28a promoter region that we amplified (data not shown).
Collectively, our data imply that lincRNA-p21 preferentially employs a repressive mechanism through H3K9me3 maintenance at the promoters of a subset of pluripotency genes (e.g., Lin28a) and CpG methylation at the promoter of other set of genes (e.g., Nanog), which are both mediated by HNRNPK (Figure 6F).
The mammalian genome encodes thousands of lincRNAs, many of which participate in a plethora of biological processes ranging from normal embryonic development to tumorigenesis14. However, only 2 lincRNAs have been reported to promote reprogramming efficiency, lincRNA-RoR and TUNA15,17, whilst lincRNAs that act as a barrier to reprogramming have not yet been identified. Understanding the repertoire of lincRNAs that regulate reprogramming and their roles is relevant because it could help to improve the reprogramming and also shed light on the role of the same lincRNAs in unrelated contexts, particularly in development. Notably, our functional screening has shown that lincRNA-1463 and lincRNA-1526 prevent the preiPSC to iPSC conversion, while lincRNA-1307 facilitates it. These 3 lincRNAs are known for their roles in preventing ESC differentiation along particular lineages22. One possibility is that they funtion by changing the balance of lineage specifiers24,25 that are highly expressed in incompletely reprogrammed colonies12.
We have also shown that a p53-induced lincRNA, lincRNA-p2118, not only sustains the preiPSC state but also poses a barrier for reprogramming. This finding reinforces the notion that p53 impairs reprogramming not exclusively through apoptosis or cell senescence26. Mechanistically, lincRNA-p21 blocks reprogramming by preserving high levels of H3K9me3 and CpG methylation at pluripotency genes. This finding is important, as it may help to understand how lincRNA-p21 represses gene transcription in the DNA damage response18 and potentially other contexts. Moreover, given the role of DNA methylation in sustaining the original cell's epigenetic memory2,27, manipulating lincRNA-p21 expression may help to produce iPSCs of higher quality.
Our results indicate that lincRNA-p21's effects on reprogramming are almost entirely mediated through HNRNPK. In combination, the 2 factors form repressive complexes containing either SETDB1 or DNMT1 at pluripotency gene promoters. However, HNRNPK undoubtedly has additional functions during reprogramming, particularly as it regulates multiple cellular processes including RNA splicing and translational control28. In agreement with this, Hnrnpk knockdown has a more potent enhancing effect on reprogramming compared to lincRNA-p21 knockdown. Likewise, the number of genes regulated by Hnrnpk knockdown in reprogramming exceeds those regulated by lincRNA-p21 knockdown. In this regard, HNRNPK and other members of the hnRNP family interact with multiple lincRNAs29,30. This makes it tempting to speculate that hnRNP proteins are widespread mediators of lincRNA function in reprogramming, which act by facilitating the assembly of various chromatin complexes (activating or repressive) at specific loci. In this scenario, the particular combination of hnRNP proteins and partner lincRNAs at different stages of reprogramming would determine the ultimate effects on this process.
As for the 2 other proteins interacting with lincRNA-p21 identified in our study, it is interesting to note that DNMT1 had been previously shown to interact with non-poly(A) RNAs (lincRNAs are frequently polyadenylated). This mechanism has been proposed to stimulate gene transcription by preventing access of DNMT1 to loci genome-widely31. Our study, in line with another report on Kcnq1ot132, thus expands the concept of DNMT1-interacting RNAs, and suggests that, as opposed to non-poly(A) RNAs, lncRNAs (potentially not just lincRNA-p21 or Kcnq1ot1) promote DNMT1 activity at particular loci.
While this manuscript was in preparation, Dimitrova et al.33 used a genetic deletion approach to demonstrate that lincRNA-p21 also acts in cis to induce p21 expression and diminish MEF proliferation. This contrasts with our work presented here and the work by Huarte et al.18, who could not detect any reduction in p21 expression upon lincRNA-p21 knockdown in MEFs. Dimitrova et al.33 also postulated that lincRNA-p21 blocks OSKM reprogramming of MEFs (measured through alkaline phosphatase staining and qPCR on day 6) by increasing p21, though they did not quantify cell proliferation in this setting. A potential explanation for the discrepancy is that complete suppression (as achieved by genetic deletion) of lincRNA-p21 is necessary to deregulate p21. Additional experimentation will be needed to clarify this issue and it is possible that lincRNA-p21 regulates reprogramming through multiple mechanisms.
Altogether, our study constitutes a resource that may guide others to investigate the functions of lincRNAs in reprogramming, and also provides a new mechanism by which p53, HNRNPK and lincRNA-p21 regulate gene expression. Future work will further delineate the networks controlled by lincRNAs during reprogramming, which will form new connections with the roadblocks identified up to this point.
Materials and Methods
RIP and RNA pull-down
RIP was performed as described34. Briefly, cells were crosslinked with 1% (w/v) formaldehyde (Sigma) and suspended in lysis buffer. Lysates were incubated overnight with the indicated antibodies at 4 °C and RNA-protein complexes were recovered with protein A/G Dynabeads (Life Technologies). RNA pull-down was performed as described34. Briefly, biotinylated lincRNA-p21 and lacZ were synthesized with the mMESSAGE mMACHINE kit (Ambion). Biotinylated RNA was folded and incubated with nuclear extracts. RNA-protein complexes were recovered with Streptavidin agarose beads (Invitrogen). Further details for both procedures are provided in Supplementary information, Data S1.
RNA-seq, bioinformatics and accession number
Sequencing was performed at Guangzhou RiboBio Co., Ltd. with the Illumina HiSeq 2500. RNA-seq data was aligned to the Ensembl v73 transcript annotations using bowtie and RSEM as previously described35. All other bioinformatic analysis was performed using glbase36. These data were deposited in GEO with accession number GSE57967. Further details are provided in Supplementary information, Data S1.
Gene Expression Omnibus
We thank all members of the Esteban Lab for their support. This work was supported by the National Basic Research Program of China (973 Program; 2011CB965201 and 2011CBA01106), the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA01020106), the National Natural Science Foundation of China (31371513 and 81202010), the Cooperation Program of the Research Grants Council (RGC) of the Hong Kong Special Administrative Region and the National Natural Science Foundation of China (81261160506), the RGC of the Hong Kong Special Administrative Region, China (476310), the Natural Science Foundation of Guangdong Province (S2013010015487), the Bureau of Science, Technology and Information of Guangzhou Municipality (2012J5100040) and the International Science and Technology Cooperation Program of China (2013DFE33080).
Expression of lincRNAs in reprogramming (related to Figure 1)
LincRNA-p21 is induced by p53 (related to Figure 2)
Comparison of cell proliferation and reprogramming efficiency in reprogrammed MEFs with lincRNA-p21 or/and p53 knockdown and iPSC clone characterization (related to Figure 2)
LincRNA-p21 is a subset of HNRNPK function in reprogramming (related to Figure 5)
LincRNA-p21 influences the maturation of let-7 miRNAs but not the expression of mesenchymal or epithelial genes (related to Figure 5)
HNRNPK binds to pluripotency genes in reprogramming and lincRNA-p21 regulates CpG methylation at the Nanog promoter (related to Figure 6)
Extended Materials and Methods
About this article
(Supplementary information is linked to the online version of the paper on the Cell Research website.)