How are short exonic sequences recognized within the vast intronic oceans in which they reside? Despite decades of research, this remains one of the most fundamental, yet enigmatic, questions in the field of pre‐mRNA splicing research. For many years, studies aiming to shed light on this process were focused at the RNA level, characterizing the manner by which splicing factors and auxiliary proteins interact with splicing signals, thereby enabling, facilitating and regulating splicing. However, we increasingly understand that splicing is not an isolated process; rather it occurs co‐transcriptionally and is presumably also regulated by transcription‐related processes. In fact, studies by our group and others over the past year suggest that DNA structure in terms of nucleosome positioning and specific histone modifications, which have a well established role in transcription, may also have a role in splicing. In this review we discuss evidence for the coupling between transcription and splicing, focusing on recent findings suggesting a link between chromatin structure and splicing, and highlighting challenges this emerging field is facing.
Mammalian RNA is composed of short stretches of exonic sequences, typically around ∼140 nt in length, interrupted by long, non‐coding introns that are typically thousands of nucleotides long. In order for RNA polymerase‐II (RNAPII) transcripts to mature into mRNA molecules, a sophisticated machinery known as the spliceosome excises the introns and ligates exons in a process termed splicing. One of the major and most fundamental challenges facing the splicing machinery is to correctly identify and ligate the short exons and precisely remove the long intronic sequences. This challenge is compounded by the fact that the splicing machinery may recognize a given sequence as an exon in one tissue or disease state, but as an intron in another, in a process termed alternative splicing (Smith and Valcarcel, 2000; Black, 2003; Nilsen and Graveley, 2010). Classically, attempts to understand the mechanism and regulation of splicing have focused on the pre‐mRNA molecules and researchers have sought to characterize the cis and trans factors that facilitate and regulate exon recognition. Four signals residing at the exon–intron boundaries have a well‐characterized role in helping direct the splicing machinery: The 5′ and 3′ splice sites (5′ss and 3′ss), at the respective ends of the introns, the polypyrimidine tract (PPT), which is upstream from the 3′ss, and the branch site (BS), located upstream from the PPT. These signals, along with a vast array of splicing‐regulatory elements (SREs) in the exons and introns, direct the spliceosomal machinery to the exon–intron boundaries and allow precise recognition of exons (Berget, 1995; Smith and Valcarcel, 2000; Graveley, 2001; Black, 2003; Nilsen and Graveley, 2010). However, these signals are far from containing sufficient information to allow precise differentiation between exonic and intronic sequences (Lim and Burge, 2001; Schwartz et al, 2009a), implying that there are additional regulatory layers providing additional information, remaining to be discovered. Another open puzzle is the mechanism underlying tissue‐specific patterns of alternative splicing, despite the identity in terms of the underlying sequence. In some cases this specificity can be explained through tissue‐specific expression of splicing factors. However, in other cases the involved splicing factors are ubiquitously expressed (see references Kornblihtt et al, 2004 and Wang and Cooper, 2007), again indicating that there are additional regulatory layers leading to tissue‐specific regulation of splicing patterns.
Where is this additional layer of information? One attractive possibility is that it lies therein that splicing in vivo does not occur as an isolated process; rather, it occurs co‐transcriptionally. Although transcription‐independent splicing has been documented both in vitro and in vivo (Green et al, 1983; Wetterberg et al, 1996), and it was found that in yeast the majority of splicing occurs post‐transcriptionally (Tardiff et al, 2006), the emerging consensus is that splicing is initiated co‐transcriptionally and that introns are removed while the nascent transcript is still tethered to the DNA by RNAPII (Neugebauer, 2002; Proudfoot et al, 2002; Allemand et al, 2008; Pandya‐Jones and Black, 2009; Singh and Padgett, 2009; Wada et al, 2009). This is based on observations indicative of co‐transcriptional splicing in human (Wuarin and Schibler, 1994; Tennyson et al, 1995; Roberts et al, 1998; Dye et al, 2006; Listerman et al, 2006; Pandya‐Jones and Black, 2009), in Drosophila (Beyer and Osheim, 1988; LeMaire and Thummel, 1990), and in other systems and organisms (for review see references Neugebauer, 2002; Allemand et al, 2008; Perales and Bentley, 2009). In itself, the temporal overlap between transcription and splicing is of no functional consequence in terms of guiding the splicing machinery towards the exon boundaries. However, it sets the foundation for coupling, or crosstalk, between these two processes, thereby providing possibilities for splicing to be controlled and regulated by transcription.
Different lines of evidence suggest that such cross‐talk exists. First, there is evidence for coordinated regulation of transcription and splicing (Auboeuf et al, 2002). Furthermore, there are a large number of transcriptional regulators and elongation factors with documented roles in splicing (Allemand et al, 2008), and similarly there is an increasing number of splicing factors with documented roles in transcription (e.g., Bres et al, 2005; Lin et al, 2008). However, the most dominant factor implicated in this cross‐talk between transcription and splicing is RNAPII as indicated by the finding that that in the absence of intact RNAPII, synthetic pre‐mRNAs is spliced less efficiently (Bird et al, 2004; Das et al, 2006; Hicks et al, 2006).
How does RNAPII facilitate splicing? Evidence exists for at least two mechanisms. The first mechanism by which RNAPII affects splicing is by serving as a ‘piggyback’ or ‘landing pad’ for various splicing factors. In particular, the C‐terminal domain (CTD) of RNAPII recruits a wide range of proteins, including splicing factors, to the nascent transcript (McCracken et al, 1997; Goldstrohm et al, 2001; Howe, 2002; Fong et al, 2003; de Almeida and Carmo‐Fonseca, 2008). For example, human U1 snRNP components co‐immunoprecipitate with RNAPII (Das et al, 2007) and presence of U1 snRNP at a 5′ss can promote recruitment of RNAPII and general transcription factors to the promoter (Damgaard et al, 2008). Also SR proteins, which help mediate spliceosomal assembly by binding to the U1 and U2 components of the spliceosome (Ram and Ast, 2007; Long and Caceres, 2009), co‐immunoprecipitate with RNAPII (Yuryev et al, 1996; Das et al, 2007) and SRp20 regulation is dependent on the presence of the CTD in transcribing RNAPII (de la Mata and Kornblihtt, 2006). The second mechanism by which RNAPII is thought to regulate splicing is through kinetic coupling. According to this model, decreased RNAPII elongation rates facilitate the recognition of weakly defined exons that would otherwise not be recognized by the splicing machinery. There are different sources of support for this model: First, a mutant RNAPII with lower transcription rates than the wild‐type polymerase increases the inclusion levels of certain alternatively spliced exons (de la Mata et al, 2003; Kornblihtt, 2006; Schor et al, 2009). Second, the topoisomerase‐I inhibitor camptothecin, which stalls elongating RNAPII, increases co‐transcriptional splicing factor accumulation and splicing (Listerman et al, 2006). Third, factors that increase the rate of RNAPII elongation, such as specific splicing enhancers or factors acting in trans, lead to increased exon skipping, whereas the elongation inhibitor dichlororibofuranosylbenzimidazole (DRB) enhances exon inclusion (Nogues et al, 2002; Kornblihtt et al, 2004). Finally, upon UV‐induced damage, the CTD becomes hyper‐phosphorylated leading to slowing down of transcription and subsequent alterations in splicing patterns (Munoz et al, 2009).
The coupling of splicing with transcription raised the possibility that factors classically known to regulate transcription could also have a role in the regulation of splicing. One such factor is chromatin structure. By chromatin structure we refer to two main elements: nucleosome positioning and histone modifications. Approximately 147 base pairs of DNA are wrapped around protein octamers, termed nucleosomes, that are separated by linker regions of variable length (Kouzarides, 2007). Positioning of nucleosomes along the genome is determined, to a large extent, by the genome sequence (Ioshikhes et al, 2006; Segal et al, 2006; Kaplan et al, 2009), but can also be subjected to modulation by chromatin remodellers (Vignali et al, 2000). In turn, nucleosome positioning modulates the properties of the sequence coiled around the nucleosome. For example, promoter DNA sequences covered by nucleosomes are less accessible to transcription factors (Kornberg and Lorch, 1999; Cairns, 2009). The chemical properties of nucleosomes can be subjected to fine‐tuning, through post‐translational histone modifications. The N‐terminal tails of histones are subjected to at least eight types of modifications, including acetylation, methylation, phosphorylation, and ubiquitination; these modifications alter the chromatin structure and accessibility, and subsequently gene expression levels (Bernstein et al, 2007; Jones and Baylin, 2007; Kouzarides, 2007).
Considerable evidence has accumulated over the past years in support of a link between nucleosome organization and splicing. One of the first studies suggesting such a link was based on changes in splicing patterns after perturbation with drugs that inhibit histone acetylases (Nogues et al, 2002). Additional support for such a link came from the fact that different chromatin remodellers have a role in splicing. For example, the Brahma subunit of the chromatin‐remodelling complex SWI/SNF interacts with splicing factors and regulates alternative splicing, presumably by pausing RNAPII elongation and thereby leading to inclusion of otherwise skipped exons (Batsche et al, 2006). Also, one isoform of the coactivator‐associated arginine methyltransferase‐1 associates with the U1 component U1C and affects 5′ ss selection (Ohkura et al, 2005). In addition, the CHD1 chromatin‐remodelling ATPase has a role in splicing (Sims et al, 2007) and conversely the splicing factors SRp20 and ASF/SF2 are associated with chromatin (Loomis et al, 2009). Moreover, it was recently found that Gcn5, which encodes the histone acetyltransferase (HAT) activity of the SAGA complex, affects splicing (Gunderson and Johnson, 2009). Is there, then, a direct cross‐talk between chromatin structure, exon–intron architecture, and splicing?
Link between nucleosome positioning and splicing
Studies of several groups, including ours, published over the last year have found large‐scale evidence for a link between nucleosome positioning and exon–intron architecture (Andersson et al, 2009; Hon et al, 2009; Nahkuri et al, 2009; Spies et al, 2009; Tilgner et al, 2009; Schwartz et al, 2009b; Chen et al, 2010). These studies relied on analysis of MNase‐digested chromatin followed by next‐generation sequencing in cells from human, Drosophila melanogaster, Caenorhabditis elegans, and Japanese killifish, and consistently found an approximately 1.5‐fold higher level of nucleosome occupancy in exons than in introns. By nucleosome occupancy we refer to the average levels of nucleosome positioning at a given genomic position. The factor underlying this differential nucleosome occupancy appears to be differences in sequence composition and, in particular, the higher GC content in exons than in introns. Two main lines of evidence support the hypothesis that differences in nucleosome occupancy are sequence‐based: First, higher nucleosome occupancy in exons versus introns can be precisely reproduced by computational models predicting nucleosome positioning based solely on DNA sequence (Tilgner et al, 2009; Schwartz et al, 2009b). Second, this pattern is also reproduced when measuring experimentally derived nucleosome occupancy levels in intergenic regions with exonic sequence composition flanked by sequences with intronic sequence composition (Spies et al, 2009). As nucleosomes are drawn by GC‐rich sequences and repelled by GC‐poor ones (Tillo and Hughes, 2009), and exons are characterized by higher GC content than introns (Tilgner et al, 2009; Schwartz et al, 2009b), it appears that the juxtaposition of intronic sequences to exonic ones serves as a force that drives nucleosomes to bind exons preferentially.
While positive correlation between exons and nucleosome positioning is indicative of a statistical association between the two, no experimental studies to date have established a causal link between nucleosome occupancy and splicing. Notably, some bioinformatics analyses do link nucleosome occupancy with splicing‐related characteristics of the exons: More prominently defined nucleosomes were found in exons with weak splicing signals (Spies et al, 2009; Tilgner et al, 2009) and in isolated exons (Spies et al, 2009), suggesting that nucleosomes ‘mark’ these exons to insure their identification by the splicing machinery. In addition, the length of DNA wrapped around a mononucleosome, 147 nt, elegantly correlates with the evolutionarily conserved average length of metazoan exons (Tilgner et al, 2009; Schwartz et al, 2009b). However, somewhat contradicting these results, nucleosome occupancy levels are higher in longer exons than in shorter ones (Spies et al, 2009; Tilgner et al, 2009; Schwartz et al, 2009b), are higher in exons with stronger PPTs than in weaker ones, and were found to be lower in alternatively spliced exons as compared with constitutively spliced ones (Schwartz et al, 2009b). Thus, experimental studies that simultaneously manipulate nucleosome positioning while evaluating splicing patterns are required to fully evaluate the presence of a cross‐talk between the nucleosome positioning and splicing. This is particularly important, because various alternative scenarios come to mind, which may underlie the association between nucleosome positioning and exon–intron architecture. For example, this association may merely reflect the gradient of GC content between exons and introns without having any regulatory role whatsoever. Alternatively, nucleosomes may confer protection to the exonic sequences coiled around them (Kogan and Trifonov, 2005), as sequences coiled around nucleosome are less sensitive to UV irradiation and chemical attack (Gale et al, 1987; Thrall et al, 1994) and less prone to indels longer than 1bp base pair (Sasaki et al, 2009), although nucleosome‐bound sequences are more prone to 1‐nt indels than are unbound sequences (Washietl et al, 2008; Sasaki et al, 2009).
Link between histone modifications and splicing
Considerable advance has been made over the past year in understanding the role of specific chromatin modifications in the context of splicing. An initial study by Kolasinska‐Zwierz et al (2009) found the H3K36me3 modification, known to mark regions undergoing transcription (Talasz et al, 2005; Vakoc et al, 2006; Bell et al, 2007; Du et al, 2008), to be globally enriched in exons. Furthermore, this enrichment was greater in constitutive exons than in alternative ones, although this result was questioned by a subsequent study (Spies et al, 2009). As Kolasinska‐Zwierz et al did not examine the underlying nucleosome positioning patterns, the question remained open whether the H3K36me3 exonic enrichment simply mirrors the underlying nucleosome enrichment, or whether it reflects an independent enrichment. Addressing this issue proved technically challenging, as nucleosome occupancy levels and histone modifications levels are typically derived on the basis of two completely different sets of experiments (MNase‐digested chromatin followed by deep sequencing versus chromatin immunoprecipitation followed by deep sequencing or ChIP‐seq, respectively) and comparison of data from these two types of experiments is not straightforward. Using different normalization schemes, some studies concluded that H3K36me3 appears to mirror nucleosome occupancy (Tilgner et al, 2009; Schwartz et al, 2009a), whereas others concluded that H3K36me3 is enriched even after taking into account nucleosome occupancy (Andersson et al, 2009; Spies et al, 2009). All studies are in agreement that the enrichment is dramatically reduced, once nucleosome occupancy is accounted for.
In parallel, these studies identified additional histone modifications enriched in exons. Surprisingly, although these studies all analysed the same data set of histone modifications in human T cells (Barski et al, 2007; Schones et al, 2008), they did not consistently identify the same set of histone modifications. In addition to H3K36me3, one study found H3K79me1, H4K20me1, and H2BK5me1 modifications to be enriched in exonic sequences (Schwartz et al, 2009b), another found H3K79me1, H2BK5me1, H3K27me1, H3K27me2, and H3K27me3 (Andersson et al, 2009), whereas the main candidates of another study were H3K27me2 and H3K4me1(Spies et al, 2009). Yet another analysis showed an anti‐correlation between H4K20me1 and nucleosome levels (Tilgner et al, 2009). The differences among the studies appear to originate from different approaches to analysis of deep‐sequencing data and in the different normalization schemes used.
As was the case for nucleosome occupancy, also for histone modifications it remained to establish a causal relationship between histone modifications and splicing; and until recently, only limited experimental data were available. The CHD1 chromatin‐remodelling ATPase, which binds to H3K4me3, is known to bind to spliceosomal components, and knockdown of CHD1 and reduction of H3K4me3 levels were shown to alter the efficiency of splicing (Sims et al, 2007). In addition, two histone modifications, H3K9 acetylation and H3K36me3, were experimentally shown to be associated with exon skipping (Schor et al, 2009). An additional study found an association between splicing patterns and changes in H3K9me2 and H3K27me3 (Allo et al, 2009). However, as none of these studies monitored the underlying nucleosome occupancy patterns, it remained possible that changes in splicing attributed to specific histone modifications could stem from nucleosome positioning.
Recently, Luco et al (2010) made a considerable advance in establishing the causal role of histone modifications in splicing. The authors found that a range of histone modifications were differentially associated with the splicing pattern of a set of mutually exclusive exons regulated by the PPT‐binding protein (PTB). These modifications included H3K36me3 and H3K4me1, acting in one direction, and H3K27me3, H3K4me3, and H3K9me1, acting in the opposite direction. Through modulations of the levels of SET2 and ASH2, the methyltransferases leading to H3K36me3 and H3K4me3 modifications, respectively (Du et al, 2008; Edmunds et al, 2008), Luco et al showed a causal link between the histone modifications and splicing. Furthermore, they showed that the effect was mediated through recruitment of PTB to H3K36me3‐modified chromatin through the histone‐tail‐binding protein MRG15. Thus, this study not only established the causal link between histone modifications and splicing, but also provided experimental evidence of the mechanism by which these processes are linked. It is important to point out that this study did not attempt to examine the effect of nucleosome occupancy thus leaving open the question of to what extent chromatin modifications mirror nucleosome occupancy.
Another major question is whether there is a combinatorial cross‐talk between different histone modifications in the context of splicing or whether the effect of each histone modification is independently exerted. Notably, this question is not specific to splicing, but also remains an open question in the field of histone modifications around promoters and within genes (Kouzarides, 2007). Studies using sequential ChIP‐seq experiments may begin to shed light on these questions and refine our understanding of the precise chromatin structure around exons.
In addition to the questions discussed above, the link between chromatin structure and splicing opens up a plethora of questions pertaining to the characteristics of chromatin around exons, just as the nucleosome‐free region upstream from the transcription start sites opened up questions pertaining to the chromatin organization at these sites. These questions include the following:
Assuming that nucleosome positioning and histone modifications can be unequivocally linked with splicing, the biological mechanism linking these two processes must be uncovered. RNAPII may mediate the link by either of the two mechanisms described above. Nucleosomes could serve as speed‐bumps for RNAPII, slowing down the transcription rates in the vicinity of exons (Schwartz et al, 2009b), thereby improving their recognition by the splicing machinery through kinetic coupling (de la Mata et al, 2003; Kornblihtt, 2006; Schor et al, 2009) (Figure 1A). The potential of nucleosomes to slow down RNAPII was recently shown by Hodges et al (2009a). However, it is also possible that it is not, or not only, the nucleosomes, but also specific histone modifications, that slow down RNAPII. Conversely, they may also speed up RNAPII, as is thought to be the case with histone acetylations (Nogues et al, 2002; Figure 1B). Alternatively, by means of its ‘landing pad’ characteristics, RNAPII may recruit splicing factors to the pre‐mRNA; these splicing factors may act in conjunction with splicing factors bound by chromatin (Figure 1C). Specific chromatin landscapes may recruit or interact directly with splicing factors, just as chromatin directs the interaction of transcription regulators with promoters (Cairns, 2009). SR proteins, for instance, may directly interact with chromatin as suggested by at least two studies (Kress et al, 2008; Loomis et al, 2009) and thereby contribute to splicing. Another possibility is that RNAPII may recruit chromatin modifiers altering the chromatin organization along exons, to allow recruitment of splicing factors to the exon either directly or indirectly, as was elegantly shown recently by Luco et al (2010) (Figure 1D). Yet another important question to be considered is the directionality of the association between chromatin structure and splicing. Although the currently favoured model is that nucleosome occupancy precedes splicing and thus helps guide the splicing machinery, the opposite model, in which chromatin structure is altered by the splicing machinery, cannot be ruled out.
Role of histone variants
Although one means of altering the chemical properties of nucleosomes are through histone modifications, a similar effect can be achieved by using different histone variants. Most core histone proteins have several variants (e.g., H2B.1, H2A.Z, H2A.Bbd, H3.3, CENP‐A) differing in their amino‐acid composition and manifesting different biological characteristics (Segal and Widom, 2009; Talbert and Henikoff, 2010). It will be interesting to find out whether there are differences in the binding profiles of histone variants in exons and introns, and to understand the biology underlying these differences.
Role of DNA methylation
Several studies point to a link between nucleosome positioning and DNA CpG methylation. The emerging consensus is that CpG methylation decreases DNA bending flexibility and hence influences nucleosome positioning (Tippin and Sundaralingam, 1997; Nathan and Crothers, 2002), although it cannot be excluded that the link between the two is mediated indirectly through a range of processes triggered by CpG methylation (Segal and Widom, 2009). In parallel, recent studies indicate that DNA methylation and histone modification pathways are also linked through a wide range of mechanisms (Cedar and Bergman, 2009; Choy et al, 2010). For instance, in early embryo development, DNA methylation patterns are established in regions lacking H3K4 methylation, as its presence inhibits DNA methyltransferases (Ooi et al, 2007; Cedar and Bergman, 2009). At pericentromeric satellite repeats, histone methyltransferases are responsible for heterochromatization through H3K9me3 histone modification and for recruitment of DNA methyltransferases (Lehnertz et al, 2003). Inactivation of Polycomb target genes and the X chromosomes are additional examples of mechanisms in which DNA methylation is believed to be induced by histone modifications (for review see references Cedar and Bergman, 2009). The opposite mechanism is also believed to exist: In this case, DNA methylation induces specific repressive histone modifications (Eden et al, 1998; Hashimshony et al, 2003). Thus, if exons differ from introns in terms of nucleosome positioning and in terms of histone modifications, they may also differ from each other in terms of DNA methylation status. In support of this conjunction, higher methylation levels were indeed found in exons as compared with introns by two recent studies (Lister et al, 2009; Hodges et al, 2009b). Are these differential levels of functional consequence or do they only reflect indirect links between the processes? Do they impose an epigenetic layer of regulation on splicing? It will be fascinating to decipher the cross‐talk between chromatin structure and DNA methylation in the context of exon–intron architecture and splicing.
Role of RNAPII kinetics
Recent studies indicate that the rate of transcription by RNAPII is 3100–3800bp per minute, whereas exon splicing is thought to be a considerably slower process, possibly in the timeframe of minutes (Singh and Padgett, 2009; Wada et al, 2009). Thus, by the time an exon is spliced, RNAPII may be several thousands of nucleotides downstream. In light of the different time scales between transcription and splicing, it will be important to assess to what extent the slowing down of RNAPII by nucleosomes (Hodges et al, 2009a) can be ‘meaningful’ in terms of splicing.
Prokaryotes lack both nucleosomes and splicing, whereas in eukaryotes both exons and nucleosomes, each of which involves sequences in the range of ∼150 nt, have come to exist. This raises speculations pertaining to an evolutionary link between the two. Did nucleosomes and splicing arise concurrently? Did one set the ground for the other? Furthermore, among mammals, and in particular primates, intron length has gradually expanded, whereas the length of exons has remained relatively constant. This and other findings has led to speculations that in mammals the spliceosome primarily recognizes exons in a process termed exon definition, as opposed to that in fungi in which introns are kept short and are thought to be the recognized unit in a process termed intron definition (Berget, 1995; Ast, 2004; Ram and Ast, 2007). Can this shift from intron definition to exon definition be linked with improved positioning of nucleosomes along mammalian exons?
We opened this paper by stating that it remains an enigma how short exons are recognized within their overwhelming intronic environment. We then summarized recent advances pertaining to our understanding of the potential role of chromatin organization in this process. Does, then, nucleosome positioning contain a substantial amount of information, which, in conjunction with previously found features, can account for the ability of the splicing machinery to recognize exons? Apparently not. First, the relatively low fold change (∼1.5) in terms of nucleosome occupancy between exons and introns suggests that the information borne by nucleosome occupancy is limited. Second, in a study assessing this using a computational model, nucleosome occupancy was found to improve recognition of exons, but to a very limited degree (Spies et al, 2009). Thus, nucleosome occupancy brings us one small step forward in understanding the workings of the splicing machinery, but we remain far from ‘explaining away’ this remarkable process. While this is suggestive of additional features remaining to be discovered or quantified (including histone modification levels), we close with the speculation that perhaps it is precisely this inherent stochasticity in the genome sequence, this intrinsic absence of determinative information (see also Fox‐Walsh and Hertel, 2009), that sets the ground for alternative splicing and for enriching our transcriptome with previously non‐existing transcripts.
Conflict of Interest
The authors declare that they have no conflict of interest.
We are grateful to Dror Hollander for help with preparing the figure. SS is a fellow of the Edmond J Safra bioinformatics program at Tel Aviv University. Funding for open access charge: Israel Science Foundation (ISF 61/09); Joint Germany‐Israeli Research Program (ca‐139); Deutsche‐Israel Project (DIP MI‐1317); and European Alternative Splicing Network (EURASNET). This work was performed in partial fulfilment of the requirements for a PhD degree of SS (Sackler Faculty of Medicine, Tel Aviv University, Israel).
- Copyright © 2010 European Molecular Biology Organization