Transcriptional gene silencing in humans

It has been over a decade since the first observation that small non-coding RNAs can functionally modulate epigenetic states in human cells to achieve functional transcriptional gene silencing (TGS). TGS is mechanistically distinct from the RNA interference (RNAi) gene-silencing pathway. TGS can result in long-term stable epigenetic modifications to gene expression that can be passed on to daughter cells during cell division, whereas RNAi does not. Early studies of TGS have been largely overlooked, overshadowed by subsequent discoveries of small RNA-directed post-TGS and RNAi. A reappraisal of early work has been brought about by recent findings in human cells where endogenous long non-coding RNAs function to regulate the epigenome. There are distinct and common overlaps between the proteins involved in small and long non-coding RNA transcriptional regulatory mechanisms, suggesting that the early studies using small non-coding RNAs to modulate transcription were making use of a previously unrecognized endogenous mechanism of RNA-directed gene regulation. Here we review how non-coding RNA plays a role in regulation of transcription and epigenetic gene silencing in human cells by revisiting these earlier studies and the mechanistic insights gained to date. We also provide a list of mammalian genes that have been shown to be transcriptionally regulated by non-coding RNAs. Lastly, we explore how TGS may serve as the basis for development of future therapeutic agents.


The history of RNA-directed transcriptional gene silencing (TGS)
Almost three decades ago, Marjorie Matzke et al. observed that over-expression of a transgene led to DNA hypermethylation and transcriptional silencing in doubly transformed tobacco plants (1) (Figure 1). Mechanistically, this type of silencing in plants was found to be the result of small non-coding RNAs directing epigenetic changes, specifically DNA methylation, to those loci containing sequences homologous to the small RNA. The phenomenon was termed small RNA-directed transcriptional gene silencing (TGS). TGS was later shown in Arabidopsis to require the action of RNA-dependent DNA methylation (2,3) and members of the Argonaute protein family (4). A few years later RNA interference (RNAi), mediated by double-stranded RNAs, was discovered as a powerful post-TGS (PTGS) system against messenger RNAs (mRNAs) in plants (5), and a few months later in Caenorhabditis elegans (6).

Transcriptional gene silencing in humans
The study of small non-coding RNA-directed TGS has been carried out in various model organisms such as plants (Arabidopsis thaliana), yeast (Saccharomyces pombe), flies (Drosophila melanogaster) and worms (C. elegans) (reviewed extensively in (7,8)). A decade ago, the first report of RNA-directed TGS in human cells was observed when exogenous siRNAs were used to silence a transgenic elongation factor 1 ␣ promoter driving a Green Fluorescent Protein (GFP) reporter gene (9) (Figure 1). Importantly, the observed silencing was clearly at the transcriptional level, as indicated by nuclear run-on analysis. Moreover, silencing was also epigenetic: inhibition was abrogated by 5 Aza-cytadine (5 AzaC) and Trichostatin A (TSA), com- pounds involved in inhibiting DNA methylation and histone deacetylalion, respectively (9). This early observation was soon followed by other studies (10,11), all of which confirmed that small non-coding RNAs could functionally control gene transcription and epigenetic states in human cells. But the underlying mechanism of action remained unknown.

Mechanisms of small non-coding RNA-directed TGS
TGS is mechanistically distinct from the abundantly studied PTGS pathway of RNAi. One notable difference is that TGS results in long-term stable epigenetic modifications to gene expression that can be passed on to daughter cells during cellular division (reviewed in (12)). Early observations postulated that siRNA-directed TGS functioned through an epigenetic nuclear mechanisms distinct from RNAi-mediated PTGS in the cytoplasm (13). For instance 5 AzaC and TSA were functional in reverting the siRNA targeted TGS, indicating epigenetic modes of gene regulation were at play in siRNA-directed TGS, and not via a PTGS-based mechanism (9). Indeed, recent studies have observed that two different siRNAs, one targeted to the promoter and one targeted to exon 1 of the coding transcript, can functionally repress the targeted gene in a TGS or PTGS based manner (14). A lot has been gleaned over the last decade regarding the mechanism of action for RNAdirected TGS in human cells. Studies carried out to determine the underlying mechanism of siRNA-directed TGS revealed that RNA-mediated TGS is operative through RNAdirected methylation of histone 3 lysine's 9 and 27 (H3K9 and H3K27, respectively) and DNA methylation at the targeted promoter (9,11,(15)(16)(17)(18)(19)(20)(21) (Figure 2). These promoterdirected siRNAs interact with a low level expressed (∼1-2%) promoter associated RNA, which is essentially the 5 UTR of the protein coding gene (16,22) (Figure 2). It is worth noting that most genes and gene promoters appear to be transcribed to some extent (23,24) and experimental observations suggest that non-coding RNAs interact with target loci via Watson-Crick-based RNA:RNA hybridization (16,22) and not by double-stranded DNA invasion. Temporal studies have determined that exogenously introduced siRNAs targeted to a promoter region interact first with Argonautes 1 and 2 (AGO1 and AGO2) (17,25,26). SiRNA and AGO interactions is found within the first 24 h, at the siRNA targeted promoter and is followed shortly thereafter with the recruitment of the H3K9me2 and H3K27me3 silent state epigenetic marks (17), and later by the recruitment of DNA methyltransferase and DNA methylation at 72-96 h for some genes (14). It should be noted, however, that the role of DNA methylation in TGS in human cells is not as clearly understood as in plants; DNA methylation at the targeted promoter is not always observed in human TGS applications (Table 1). These effects may be explained by the duration of RNA targeting to the promoter, the presence of robust siRNA targeting (e.g. delivery to the nucleus), the presence and abundance of promoter-occupied RNAs and/or the dynamic interplay of proteins interacting with the promoter. Despite differences in the various experimental observations, a key consistent feature has been the observations that promoter-directed small RNAs can modulate gene transcription and that some level of epigenetic based silencing is ongoing in the observed silenced genes.

The endogenous pathway of TGS in human cells; rise of long non-coding RNAs
While small RNAs were observed early on to regulate gene transcription in human cells by the targeting of epigenetic silencing complexes to those loci containing complementarity to the small RNAs (Figure 1), the endogenous mechanism(s) driving this form of gene regulation in the context of human cells remained largely unknown. MicroRNAs (miRNAs) have been shown to be endogenous drivers of TGS with some genes in human cells (27)(28)(29)(30) (Table 1). In 2005, through the efforts of the FANTOM and ENCODE consortia, it started to become apparent that a large fraction of the human genome was generating long non-coding RNAs (lncRNAs) and that many of these transcripts were antisense to protein-coding counterparts (31,32). Several of these sense/antisense or bidirectionally-transcribed genes are evolutionarily conserved, suggesting some functional cues for retention of these elements (33,34). Indeed, studies with imprinted genes and X-inactivation found that cis acting long non-coding RNAs (lncRNAs) were actively in-  volved in epigenetic regulation of these, dosage-dependent, regulated loci (35).

Mechanisms of small and long non-coding RNA-directed TGS
To date there are ∼55 reports of small RNA-directed TGS and ∼10 of antisense lncRNA-directed TGS (Table 1). Mechanistically, much of what we know about how small non-coding RNAs, such as siRNAs, miRNAs and small antisense RNAs (sasRNAs)-directed TGS, have been determined from cell culture studies. The promotertargeted small RNAs interact with various proteins to guide TGS, beginning in the first 24 h, with direct interactions with AGO1 and AGO2 (17,18,25) followed shortly thereafter by interactions at the targeted promoter with DNMT3a (14,18,77,78), HDAC1 (14,20) and resulting ultimately in histone 3 lysine 9 di-methylation and histone 3 lysine 27 tri-methylation (H3K9me2 and H3K27me3, respectively)(14,16-18,20,79,80) (Figure 3). SiRNA-directed TGS has also been observed to occur in the absence of DNA methylation, suggesting that alternative routes may be present for RNA-mediated transcriptional and epigenetic silencing (10). Small RNA-directed TGS appears to require a template or target transcript at the corresponding targeted promoter (16,22), similar to the method by which plants utilize RNA Polymerase V transcribed and processed siRNAs to regulate DNA methylation and TGS (reviewed in (81)). Notably, in plants there is a requirement for RNAdependent RNA polymerase (RdRP) activity to amplify RNA polymerase V transcript-directed TGS (81), whereas humans lack such a polymerase, which opens up a methodology for specific RNA-directed epigenetic modes of regu-lation. Curiously, this is exactly what lncRNAs appear to be doing in human cells via cis and trans-specific targeting of epigenetic complexes to particular loci ( Figure 3 and Table  1), similar to what is also observed in Saccharomyces cerevisiae, which also lacks RdRP activity (82,83).
Early studies carried out with S. cerevisiae indicated that antisense non-coding RNAs function endogenously to direct epigenetic gene silencing in place of RdRP-mediated mechanisms (82,83). The parallels between S. cerevisiae and previous observations of small antisense RNA-directed TGS in human cells (18) have emerged, suggesting that antisense transcripts also function to direct TGS (Figure 3). Most notable are the observations that particular antisense lncRNAs, first observed in tumor suppressor genes, p15 (38) and p21 (39), function to epigenetically modulate their protein-coding counterparts ( Figure 1, Table 1). One interesting, and surprisingly overlooked, early study found antisense transcription was involved in DNA methylation in Thalassemia (40), and even early work linked antisense transcripts and DNA methylation in regulating HIV (51) and MYC (84,85).
Mechanistically, far less is known about how antisense lncRNAs direct epigenetic silencing in human cells. Studies carried out with the lncRNA, HOTAIR, indicate that bimodal chromatin modifying complexes can be localized to the HOX locus via the action of this lncRNA (86). A common theme is also evident with Kcnq1ot1 (45) and the p53 regulatory lincRNA p21, which indicates that the entire p53 expressed pathway is controlled by the action of this lncRNA at the p53 locus (57). Indeed, many lncR-NAs have been observed to be associated with chromatin (87), but mechanistic insights into the process of lncRNAdirected gene regulation remain less clear. Interesting insights into the mechanism of action of lncRNA-directed TGS came from a recent study looking at the PTEN pseudogene. It had been reported previously that the PTEN pseudogene functions as a miRNA 'sponge' (64), similar to the CEBPA lncRNA that acts to sponge DNMT1 away from the CEBPA promoter (88). Studies to interrogate the PTEN pseudogene in greater detailed determined that this pseudogene also expressed an antisense lncRNA in trans which functions to direct TGS to the PTEN promoter and control PTEN expression epigenetically (53). Mechanistically, the PTEN pseudogene expressed antisense lncRNA modulated PTEN transcription by recruiting DNMT3a and EZH2 to the PTEN promoter. The parallels between the functions of the PTEN pseudogene and previous observations with small antisense ncRNA-directed TGS are notable, as both involved the action of DNMT3a (Figure 3). It is noteworthy that DNMT3a is the only known de novo DNA methyltransferase in human cells (89) and has been observed previously to be the only DNA methylatransferase to bind non-coding RNAs including small ncRNAs, both antisense and double stranded RNAs (18,77,78,90), and lncR-NAs (53,91). There is an interesting connection between DNMT3a and epigenetic silencing, which including studies indicating DNTM3a co-immunoprecipitates with HDAC1 (92,93) and EZH2 (94), as well as early predictions that DNA methylation is an active participant in X-inactivation (95), one of the first bona fide lncRNA regulatory pathways described. Collectively, a paradigm is emerging in human cells, which proposes that non-coding RNAs, both small and long forms (Figure 3), function through the action of DNMT3a to modulate chromatin and epigenetic states of gene expression. While there are several other mechanisms of action described for lncRNAs in human cells, the interactions with DNMT3a and targeting of transcriptional and epigenetic states is of particular interest, as this mode of gene regulation has the potential to be long-lasting, heritable and may be of significant relevance to the development of targeted therapeutics (reviewed in (96)).

Therapeutic applications of RNA-directed epigenetic regulation of gene expression
The utility of small RNA-induced TGS as a therapeutic has been largely ignored, mainly due to the pervasiveness of using RNAi targeted approaches to degrade mRNAs. The main concern with RNAi and post-transcriptional mechanisms of gene silencing (Figure 2) is the duration of their therapeutic effect. The effector siRNAs required to drive RNAi must be administered continuously to repress a therapeutic target gene. This is not the case with RNA-induced TGS, where stable, long-term, silencing can be achieved following a relatively short duration of promoter targeting with the siRNAs (19,20,(97)(98)(99)(100) or small antisense RNA (14,101). This is because the mode of action for the observed gene silencing is transcriptional and driven ultimately by epigenetic silencing (79,102) and not 'slicing' of the genes messenger RNA as is the case with RNAi. One universal hurdle that both RNAi and RNA-induced TGS face with is the targeted delivery of the effector RNAs to those cells requiring treatment. One approach is to utilize synthetic antisense oligonucleotides targeted to promoters of interest. This approach has worked with regards to blocking transcription (103) but was not found to induce robust epigenetic silencing, unless the particular oligonucleotides were RNA based (104). However, it may be that better interrogation of each non-coding RNA targeted promoter is required to delineate the best promoter-associated transcripts to target and that many of the earlier studies may have neglected this notion. Indeed, establishing TGS in the absence of a target promoter RNA has not been reported and attempts by some groups, including ours, have proven fruitless. Another approach might be to deliver the effector RNAs using receptor targeted aptamers, which has shown promise for targeting HIV infected cells (105,106). While delivery remains an important concern, the notion that one needs to only target a particular gene for 2-4 days to instill stable epigenetic silencing is promising with regards to minimizing the need for sustained delivery. Recent studies suggest that small RNA-directed TGS is feasible and that stable epigenetic marks can be imposed at small RNA target loci in vivo (99,107).
Another area of therapeutic utility can be found in the plethora of lncRNAs that are appearing to be involved in various diseases. Emerging evidence suggest that noncoding RNAs play a wide role (108) in various disease states in humans. Genome-wide observations of diseased states, such as heart failure (109), indicate significant differential and discordant expression between protein-coding and non-coding antisense and pseudogenes is prevalent (110). To date the list of those lncRNAs involved in human diseases is expanding at an unprecedented rate. LncR-NAs have been observed in disease ranging from Cancer (57,86,105,106), to HIV (111,112), to autism (113), to pluripotency and differentiation (114)(115)(116). It is worth underscoring that many of the disease relevant lncRNAs have been observed to be antisense to particular protein coding genes. A significant obstacle to using RNAi and other post-transcriptional effectors for targeting antisense lncR-NAs is the fact that double stranded siRNAs have an ability to target both sense and antisense transcripts (117). The use of RNA-directed TGS avoids this issue by targeting the lncRNA promoter with single stranded antisense transcripts (52). The targeting of endogenous effector antisense lncRNAs can result in the de-repression and subsequent transcriptional activation of the lncRNA targeted locus (Figure 4). Using this mode of action, it becomes feasible to activate gene expression to affect those proteincoding genes under sustained lncRNA-directed TGS (Figure 4). This has proven an effective approach to inducing genes both in vitro (39,52,(118)(119)(120)(121) and in vivo (42,107,122), but presupposes that there are known antisense lncRNAs regulating the therapeutic target gene. Collectively, the advantages to using RNA-directed TGS as a therapeutic are many and include: (i) strand specific targeting of a gene, (ii) stable long-term epigenetic based silencing can be established to particular genes of therapeutic interest and (iii) antisense RNA-based approaches work as well, if not better than double stranded RNAs, as the endogenous pathway of RNA-directed TGS appears to contain significant overlap with small antisense RNAs and antisense lncRNAs ( Figure  3).

CONCLUSION
It has been roughly 10 years since the first observation that promoter-directed RNAs can affect gene transcription (Figure 1 and Table 1). This seminal observation in 2004 (9) was indicative of a role for RNA in regulating gene expression, a notion proposed ∼5 decades ago but largely overlooked (123,124). Possible reasons for the poor early adoption of RNA-directed TGS (Table 1) are varied but may include (i) the unfortunate retraction of a similar paper published in Nature (125), and/or (ii) the overwhelmingly positive response to PTGS and the rejection of any RNAi-related phenomena occurring in the nucleus, despite the fact that RNAi was shown to be functional in the human nucleus in 2005 (126) and confirmed in many subsequent studies (25)(26)(27)(127)(128)(129)(130)(131).
The notion that RNA may function as the master gene regulator in the cell was something proposed by Britten and Davidson in 1969 (123), which at the time was largely neglected by the broad scientific community. With the advent of high-throughput technologies and the findings from EN-CODE, that most of the human genome is transcribed and likely plays a functional role (132)(133)(134)(135)(136)(137)(138)(139), it is becoming apparent that Britten and Davidson's theory should be reappraised. Certainly, lncRNAs are abundantly active in the nucleus, and many of them are active modulators of transcriptional and epigenetic modes of gene expression (reviewed in (37,76) and appear to share many of the mechanistic characteristics observed in small RNA-directed TGS ( Figure 3). Collectively, the mounting observations that an- tisense non-coding RNAs, both small and long RNAs, directed to gene promoters can affect transcription by the recruitment of silent state epigenetic complexes suggests that a pervasive and underappreciated role for non-coding RNAs is part of the basic fabric of life. Knowledge of this molecular pathway may prove incredibly insightful with regards to the development of disease, including epigenetic silencing of gene expression and the development of newtargeted therapeutics aimed at specifically affecting gene ex-pression. The next decade could prove an exciting time for our understanding of non-coding RNAs in the transcriptional gene expression and their application as novel therapeutics.