Post-translational modifications of histone proteins in conjunction with DNA methylation represent important events in the regulation of local and global genome functions. Advances in the study of these chromatin modifications established temporal and spatial co-localization of several distinct ‘marks’ on the same histone and/or the same nucleosome. Such complex modification patterns suggest the possibility of combinatorial effects. This idea was originally proposed to establish a code of histone modifications that regulates the interpretation of the genetic code of DNA. Indeed, interdependency of different modifications is now well documented in the literature. Our current understanding is that the function of a given histone modification is influenced by neighbouring or additional modifications. Such context sensitivity of the readout of a modification provides more flexible translation than would be possible if distinct modifications function as isolated units. The mechanistic principles for modification cross-talk can originate in the modulation of the activity of histone-modifying enzymes or may be due to selective recognition of these marks via modification of specific binding proteins. In the present chapter, we discuss fundamental biochemical principles of modification cross-talk and reflect on the interplay of chromatin marks in cellular signalling, cell-cycle progression and cell-fate determination.
The basic unit of eukaryotic chromatin is the nucleosome. The core nucleosome consists of 146 bp of DNA wrapped around an octameric histone protein complex , which contains a central histone (H3/H4)2 tetramer and two histone H2A/H2B dimers. Eukaryotic histones have two domains, a central globular histone fold domain and less structured N-terminal tail regions. In addition, histones H2A and H2B contain unstructured C-terminal tails. The histone fold domains are the major structural components of the nucleosome core. The tail regions, in contrast, protrude out of the nucleosome. There they are targets of various PTMs (post-translational modifications) that are important for the regulation of chromatin function.
The overall arrangement of chromatin in a cell is not uniform but is ordered into particular structures and superstructures [2,3]. In the simplest classification, chromatin is categorized into heterochromatin and euchromatin. The latter is enriched for active genes and characterized by low compaction as well as the presence of certain PTMs and non-canonical histone variants associated with active transcription (reviewed in ). In contrast, more compacted and transcriptionally less or inactive genomic regions constitute heterochromatin, which is also associated with a different set of PTMs and histone variants. Heterochromatin is further categorized as facultative or constitutive. Transcriptional suppression in facultative heterochromatin is directed by different signals such as from extracellular stimuli during cell-cycle progression or stages of development. In contrast, constitutive heterochromatin (e.g. pericentromeres, telomeres) is found at all times throughout cellular differentiation. It frequently contains highly repetitive elements such as pericentromeric satellite repeats.
Epigenetic markers and combinatorial readout
For this chapter ‘epigenetic’ will be used to describe all processes that regulate functional states of chromatin – although the exact definition of this term comprises only heritable changes in gene expression profiles that do not involve alterations of DNA sequence. Factors that contribute to epigenetic mechanisms in this sense are an incorporation of non-canonical histone variants, chromatin remodelling, non-histone protein components, DNA methylation, DNA-binding proteins, RNAs and histone modifications (Figure 1). In this chapter we will focus our discussion on histone modifications and DNA methylation. We will mainly consider a framework for epigenetic phenomena that comprises cellular signalling, cell-cycle progression and cell-fate determination (Figure 1).
Histone and DNA modifications have been designated ‘marks’ owing to the observation that their presence can correlate with a certain chromatin condition such as transcriptional activation or repression. The co-localization of several distinct modifications on the same histones (cis) and on different histones within a nucleosome or defined stretches of chromatin (trans) was suggested to constitute a ‘histone code’ [5–7]. According to this hypothesis, the different modifications exert their function in a combinatorial manner, implying that the placing and/or the effect of one PTM can be positively or negatively influenced by additional modifications. Diverse and context-specific functions that are observed for one and the same histone modification have been used to argue in favour of such a code. There are many different ways that such modification cross-talk can occur. Combinatorial effects for several histone PTMs are now well documented in the literature, however, evidence for a universal ‘histone code’ is still lacking [5,8–10]. We therefore suggest that modification cross-talk may be considered as a context-dependent readout.
Different PTM systems target histone molecules
The multitude of different chemical groups added to histones includes mono-, di- and tri-methylation (me1–3), acetylation (ac) and ubiquitination (ub) of specific lysine residues, monomethylation and symmetric- (symme2) or asymmetric (asymme2) dimethylation of arginine residues and phosphorylation (ph) of serine, threonine and tyrosine residues. These modifications are placed and removed by enzymes that are downstream from signalling cascades. We will not discuss all of the different histone PTMs, but will focus on examples where cross-talk with other modifications is well documented. Reversible acetylation of lysine residues has been mainly linked to transcriptionally active regions of the genome. Lysine acetylation is considered to be highly dynamic and is governed by HATs (histone acetyltransferases; also known as KATs)  and HDACs (histone deacetylases). HATs require acetyl-coenzyme A for their activity and frequently associate with transcriptional co-activators. Nevertheless, they are also involved in cell-cycle progression or dosage compensation and acetylation of various non-histone proteins [12,13]. HDACs remove acetyl groups and are common components of multiprotein complexes that mediate transcriptional repression.
Mono-, di- or tri-methylation (Kme1–3) of specific histone lysine residues is catalysed by HMTs (histone methyltransferases; also known as KMTs) that use the coenzyme SAM (S-adenosyl methionine) as the methyl donor . Different degrees of methylation have different functional implications in a wide range of biological processes including transcriptional activation, elongation or repression, imprinting, DNA replication and DNA-damage repair [15–17]. For example, methylation of histone H3 at lysine 9 (H3K9) or H3K27 has mainly been linked to transcriptionally repressive chromatin. In contrast, H3K4 methylation is found at transcriptionally active sites. Correspondingly, HMTs display much higher substrate specificity than HATs. In mammals the two Suv39h1/h2 isoform enzymes, for example, mediate H3K9me2/3 in pericentromeric heterochromatin, whereas H3K9me1/2 in euchromatic regions is placed by the G9a–GLP complex. EZH2, another HMT, trimethylates H3K27. Histone KDMs (lysine demethylases) comprise two major groups, the flavin-dependent amine oxidases such as LSD1/2 that cannot process trimethylated substrates, and JMJDs (JmjC-domain-containing demethylases) that require Fe(II) and α-oxoglutarate as cofactors .
Phosphorylation of serine and threonine residues was found in different histones and has been associated with a multitude of biological processes such as apoptosis, mitosis/meiosis, DNA-damage repair, transcriptional induction, dosage compensation and heterochromatin formation in post-mitotic cells [19–26]. Histone phosphorylation is highly transient and requires the stable presence of kinases due to the constant action of phosphatases [27,28]. Many distinct kinases have been implicated in phosphorylation of histones . Transcription-related recruitment of kinases appears to be mediated mainly via sequence-specific DNA-binding proteins. In contrast, targeting of phosphatases is not well understood . The genome-wide distribution of histone phosphorylation marks differs considerably between diverse eukaryotic lineages and even during different stages of the cell cycle, indicating that it may be frequently involved in combinatorial modification readout (reviewed in [19,25]).
Several histones were found to be ubiquitinated on lysine residues with diverse biological functions . Mono-ubiquitination of yeast histone H2BK123 (H2BK120 in mammals) is conserved throughout eukaryotic organisms and is pivotal in regulation of gene transcription. In yeast, the modification is placed via sequential activity of E1-activating, E2-conjugating (RAD6) and E3-ligase (BRE1) enzymes. Dynamic regulation of histone H2B ubiquitination is essential for gene transcription and requires the activity of the isopeptidase Ubp8, which removes ubiquitin .
Methylation of genomic DNA has been reported in numerous organisms . In mammals, genomic DNA can be methylated at CpG dinucleotides via DNMTs (DNA methyltransferases). These enzymes are grouped into three separate classes: Dnmt1, Dnmt2 and Dnmt3. Dnmt1 is the major enzyme required for maintenance of DNA methylation profiles. Dnmt2 has been implicated in DNA and RNA methylation. However, the functional significance of this enzyme is not well understood. Dnmt3a and Dnmt3b are required for de novo DNA methylation, and Dnmt3L does not possess enzymatic activity, but is required for proper DNA methylation.
How do chromatin modifications have an impact on chromatin structure and thereby regulate its function?
Histone PTMs might direct chromatin conformation in two not necessarily exclusive ways. First, histone modifications can directly influence the compaction status of chromatin by altering inter- or intra-nucleosomal interactions. This scenario has been most extensively discussed for lysine acetylation and serine/threonine phosphorylation, as these modifications may serve to reduce net positive charge of the histone N-terminal tails and may therefore interrupt interactions with negatively charged DNA . Furthermore, acetylated nucleosomes may be inhibitory to the formation of internucleosomal contacts . One notable modification, H4K16ac, was demonstrated to inhibit 30 nm chromatin fibre folding, demonstrating that a single acetylation signal could have an impact on chromatin higher-order structure .
In the second effector-mediated readout mode, PTMs are specifically recognized by proteins that either directly alter the function of chromatin or recruit additional protein complexes that initiate further modification steps . Multiple protein-interaction domains that recognize particular modifications have now been extensively characterized, providing a comprehensive understanding of the molecular basis for selective PTM binding (Table 1) [17,34].
Modification cross-talk can occur in different modes
There is expanding experimental evidence that histone PTMs are interdependent, and a particular modification may provide the context for additional modification events. Modification cross-talk can be best understood in the effector-mediated readout mode, but interdependency may also be relevant for modifications that directly alter chromatin structure.
Different histone modifications can influence each other either in a positive or negative manner (Figure 2A). If the interacting modifications are located on the same histone molecule we refer to this as cross-talk in cis, whereas trans cross-talk occurs if the modifications are on different histones or nucleosomes (Figure 2A). The simplest form of cross-talk is the exclusion of modifications that target the same residues, such as histone H3K9 methylation and acetylation. Different correlations between distinct modifications (Figure 2A) are the observable result of cross-talk on two levels: the enzymatic systems that place or remove the modifications (Figure 2B) and the readout of PTMs by modification-dependent binding proteins (Figure 2C).
Pre-placed modifications can directly stimulate or block the enzymatic activity of secondary chromatin modifiers (Figure 2Bi and 2Bii). Cross-talk on this level can occur in cis and in trans. For example, H3K9me demethylases LSD1 and JMJD2C are inhibited by H3S10ph (cis) and histone H2BK120ub stimulates the Dot1 HMT enzyme to methylate histone H3K79 (trans) [35–37]. Correspondingly, removal of a blocking modification by one factor may relieve repulsion of another factor that can now bind or modify a particular site (Figure 2Bi). Also, complex modification patterns can be placed by the sequential activity of enzymes, such as the initial phosphorylation of histone H3S10 that stimulates the HAT Gcn5 to acetylate H3K9/14  (Figure 2Bi). Histone PTMs can also indirectly mediate cross-talk by PTM-dependent enzyme recruitment (Figure 2Biii). This form of cross-talk can involve a single factor that recognizes a modification directly via a PTM-dependent interaction domain (Figure 2Biii). For example, the HMT G9a can bind histone H3K9me2 via ankyrin repeats and might consecutively methylate nearby nucleosomes, thereby propagating the H3K9me2 on the chromatin fibre . In addition, indirect enzyme recruitment can involve the assembly of multiprotein complexes. In this case the modification is recognized by a subcomponent of the complex and not by the enzyme directly (Figure 2Biii). For example the PHD-finger containing ING (inhibitor of growth) proteins can bind histone H3K4me3 and associate with HAT or HDAC enzymes to mediate acetylation or deacetylation of nearby histones . Some enzymes may have to place or remove modifications in a consecutive order as has been suggested for HATs and HDACs (Figure 2Biv) .
More indirect cross-talk is established on the level of modification readout via effector protein binding. Modification patterns can form co-operative or repulsive binding sites for PTM-dependent factors. For example, HP1 (heterochromatin protein 1) proteins are removed from H3K9me2/3 binding by adjacent H3S10ph. Synergistic binding has been observed for 14-3-3 proteins to H3S10phK9/14ac [42,43] (Figure 2Ci and 2Cii). Binding of multiple modifications by one factor or a multiprotein complex can also occur in trans and mediate interaction between different histone tails or even nucleosomes (Figure 2Ciii and 2Civ). For example, Brd4 can simultaneously bind acetylated lysine residues on histones H3 and H4 via its two bromodomains (Figure 2Ciii) . The interaction between different nucleosomes is particularly interesting, as this would enable trans-fibre interactions that may be crucial for chromatin structural organization. Some PTM-dependent binding proteins such as HP1 can form dimers or multimers and therefore have the potential to mediate such trans-fibre interactions (Figure 2Civ).
Our current understanding suggests that certain (if not all) histone PTMs are interconnected and the impact of particular modifications on chromatin structure is only comprehensible on the basis of the overall PTM content of a chromatin region. So far we have discussed general modes of modification cross-talk. In the following sections we will have a more detailed look at examples of cross-talk in concrete cellular situations (see Figure 1).
Modification cross-talk during transcription
Transcription constitutes one of the most complex cellular processes and encompasses initiation, reinitiation, elongation and termination. Modification of chromatin is a potent platform for the regulation of transcription.
Trans-modification cross-talk during transcription initiation
Analysis of yeast strains displaying defects in histone H2BK123 ubiquitination revealed a strong reduction of histone H3K4 and H3K79 di- and tri-methylation, which corresponds to a positive trans cross-talk system (Figure 2A). This modification cross-talk is quite complex and involves the HMT enzymes Dot1 and Set1 that methylate histone H3K79 and H3K4 respectively. Histone H2B ubiquitination is not required for the recruitment of Set1 and Dot1, but directly stimulates their processivity in trans (Figure 2Bii) [36,45]. Besides regulating H3K4/K79 methylation, histone H2BK123ub also recruits two additional proteins Rpt4 and Rpt6, which may support the localization of the SAGA–HAT protein complex, corresponding to an indirect recruitment of enzymatic activity (Figure 2Biii). The SAGA complex not only has HAT activity, but also contains the ubiquitin protease Ubp8 that removes ubiquitin from H2BK123, a step that is essential for productive elongation. In addition, the CHD1 subunit of SAGA contains a double chromodomain that has been suggested to bind H3K4me2/3 and thereby might support recruitment of the complex (Table 1). This complicated reciprocal cross-talk system also demonstrates that sequentially placed histone modifications regulate transcription in a dynamic manner.
Cis-modification cross-talk during transcription
Serine/threonine phosphorylation of histone H3 displays opposite patterns in mitosis and interphase. It correlates with chromosome condensation on one side and with gene transcription on the other side . Histone H3S10ph has been implicated in gene activation in response to MAPK (mitogen-activated protein kinase) or nuclear hormone receptor signalling  and appears to positively cross-talk with H3K9/K14ac in cis (Figure 2A) . Numerous H3S10 kinases were described, including RPS6K (ribosomal S6 kinase), MSK1/2 (mitogen- and stress-activated kinase 1/2), JIL1 and PIM1. These enzymes appear to be recruited to their target regions via sequence-specific DNA-binding proteins. So far it is not known whether H3S10ph has a major impact on chromatin higher-order structure .
Phosphorylation of H3S10 relieves HP1 from its H3K9me2/3-binding site and generates a low-affinity interaction site for 14-3-3 proteins [43,48,49]. These factors are then more stably bound upon additional H3K9ac or H3K14ac . Therefore this modification can participate in both repulsion and enhancement of protein binding depending on the context of interacting proteins and additional modifications (Figure 2Ci and 2Cii). 14-3-3 proteins were demonstrated to recruit chromatin remodelling complexes to promoter regions . Another function of 14-3-3 binding to histone H3 appears to be cross-talk with histone H4K16ac via recruitment of the MOF acetyltransferase (Figure 2Biii) . The resulting acetylations of H3K9 and H4K16 constitute a binding platform for the double bromodomains of Brd4 (Figure 2Ciii). Brd4 in turn recruits p-TEFb that is required for RNA polymerase II elongation . Interestingly, co-operative binding of two acetyl groups (H4K5ac/K8ac) can also be mediated by a single bromodomain, as has been demonstrated for the mouse TAF1 homologue Brdt . This observation suggests that acetylation-dependent protein recruitment can occur in a threshold-dependent manner. Another example of phosphorylation-dependent cross-talk includes reduced target binding of the HAT Gcn5 initiated by loss of H3T11ph at DNA-damaged chromatin regions . The regulatory examples demonstrate that gene transcription is strongly effected by modification cross-talk, thereby setting up a (sometimes gene-specific) cause-and-effect chain that serves to fine-tune the complex process of transcription.
Cell cycle dependent cross-talk
Mitosis constitutes a special section of the cell cycle, with major alterations in chromatin conformation. Several histone PTMs are subjected to cyclical patterns of appearance and removal during mitotic progression, including methylation, acetylation and phosphorylation . Indeed, massive phosphorylation of several serine residues in histone H3 constitutes a hallmark of M-phase chromatin. Aurora B and NIMA kinases phosphorylate histone H3S10 and H3S28 whereas Haspin modifies H3T3 . Furthermore, mitotic phosphorylation of H3T11 has been described. Although the exact function of these simultaneous phosphorylation events is still not fully clear, it is tempting to speculate that these are causal for the displacement of a large number of factors involved in chromatin functional regulation from mitotic chromosomes including HATs, HDACs and HP1. In addition, transcription factors such as Sp1/3, E2F1, TFIIB/D and RNA polymerase II are released at the onset of M-phase.
What is the purpose of these massive chromatin alterations? The major function of mitotic chromatin modifications is most likely to generate a chromatin conformation that is compatible with chromosome segregation. However, at the same time it is necessary to propagate epigenetic markers throughout cell division to re-establish parental chromatin states in daughter cells. Accordingly, not all genes are completely silenced during mitosis and some chromatin-associated proteins such as CTCF (CCCTC-binding factor), TBP (TATA-binding protein) or Brd4 remain bound to mitotic chromosomes, but in a locally restricted manner. These factors are thought to mark genes for re-establishing gene expression after mitosis . The mitotic marking function of Brd4 is particularly interesting, as binding of this factor may require simultaneous histone H3 and H4 acetylation. Mitotic chromatin is globally hypoacetylated, nevertheless Brd4 is selectively bound to genes that are immediately expressed after mitosis . This observation indicates that during mitosis a particular pathway locally maintains histone H3/H4 acetylation to regulate marking via Brd4.
Conversely, negative cross-talk of modifications has been described at the onset of mitosis for H3S10ph at silenced pericentromeric heterochromatin that directly interrupts the association of HP1 proteins with H3K9me2/3 . It was shown in fission yeast that displacement of the HP1 homologue Swi6 results in transcription of centromeric repeats. These transcripts are further processed into siRNAs (small interfering RNAs) that redirect H3K9me2/3 and restoration of heterochromatin . Paradoxically the propagation of constitutive heterochromatin is therefore dependent on its transcription . Consequently, H3S10ph may function as a signal to direct either gene transcription in interphase (see above) or heterochromatin transcription in mitosis to mediate epigenetic heterochromatin preservation. Together, the addition or removal of histone PTMs and chromatin-associated proteins may support the propagation of gene expression while at the same time these mechanisms allow chromosome compaction and chromatid separation.
Cross-talk in cell-fate decisions
The switch from cellular pluripotency to lineage-committed or terminally differentiated cells involves major alterations in chromatin structure, including DNA and histone methylation. These epigenetic adaptations are critical for the maintenance of the identity of a cell. Conversely, cellular transformation entails faulty epigenetic changes that complement genetic alterations in the progression of cancer [58,59] (Figure 3).
DNA methylation and histone methylation cross-talk
DNA methylation constitutes a central parameter in the establishment of cell-type-specific chromatin states. Cross-talk of DNA methylation with histone modifications, especially histone H3K9me2/3, has been observed in different model systems . The mammalian Suv39h1 enzyme methylates histone H3K9 and this modification was shown to mediate localization of the DNMT Dnmt3b and accordingly DNA methylation to pericentromeric repeats. In contrast, binding of Dnmt3L to the histone H3 N-terminal tail is inhibited by H3K4 methylation in vitro . In this way de novo methylation via Dnmt3a/b may be positively and negatively regulated by pre-placed histone modifications. Another histone H3K9 methyltransferase, G9a/GLP, is required to direct DNA methylation at euchromatic regions (reviewed in ). Both systems appear to cross-talk in a reciprocal manner, as Dnmt1 and Dnmt3b mutant cells also show changes in the H3K9 methylation profile .
In addition, histone H3S10 phosphorylation appears to be involved in this system. Loss of the phosphatase PP1 results in increased histone H3S10ph in Neurospora crassa concomitant with loss of histone H3K9me3 and DNA methylation . In Drosophila, histone H3S10ph counteracts heterochromatin formation and has an impact on chromatin structural organization . These findings place H3S10ph upstream from histone and DNA methylation, thereby preventing transcriptionally repressive heterochromatin formation. Nevertheless, histone H3S10 phosphorylation by Aurora B together with H3K9me3 was reported to mark heterochromatin in the course of terminal differentiation . Furthermore, DNA methylation has been implicated in Aurora-B-mediated histone H3S10ph at pericentromeric heterochromatin . These contradictory observations indicate that this cross-talk system could be used in different manners in distinct organisms. Alternatively, additional modifications and factors that have not yet been described may participate in the cross-talk between DNA and histone methylation and phosphorylation.
Opposing histone modifications mark genes in ESCs (embryonic stem cells)
Another interesting example of chromatin cross-talk in cell-fate decision involves histone H3K4 and H3K27 methylation in ESCs. Histone H3K27 is methylated by the EZH2 methyltransferase subunit of the PRC2 (Polycomb repressive complex 2). It recruits the PRC1 complex for repression of major developmental regulator genes (reviewed in ). In ESCs, H3K27me3 covers a large portion of the genome, but surprisingly the majority of these regions are also marked with histone H3K4me3 that is normally found at sites of active transcription . The sites of co-existence of these two opposing histone modifications have been termed ‘bivalent domains’  as they may poise a gene for either transcriptional activation or repression, depending on the particular fate of the ESC (Figure 3). This means that upon differentiation into a particular cell lineage, a gene whose regulatory region is marked in a bivalent manner will either become transcriptionally active and will lose the repressive H3K27me3 mark, or will maintain this modification, but will lose H3K4me3 (reviewed in ). Two additional factors demonstrate cross-talk with this system. Recruitment of the PRC1 complex mediates histone H2AK119 ubiquitination via its RING1 component, which is crucial to keep RNA polymerase II located at the bivalently marked genes in a non-elongating form . In addition, a specialized histone variant, H2A.Z that was shown to negatively correlate with DNA methylation is located at these regions. Together these mechanisms appear to ensure that differentiation-specific genes are not stably silenced in ESCs, but at the same time they inhibit erroneous expression, which would interfere with the pluripotent state. It is not yet clear if or to what extent histone H3K4 and K27 methylation exert cross-talk, or even if they occur on the same histone. However, the resolution of these domains into transcriptional active sites may require the activity of H3K27-specific demethylases along with H3K4 methyltransferases. Accordingly, the stable silencing of these domains may require H3K4-specific demethylase together with H3K27me3-mediated polycomb silencing and DNA methylation. Perhaps both systems exert cross-talk to exclude the opposing state. Indeed, recruitment of H3K27 demethylases via H3K4 methylation in the course of cell differentiation might block reconversion into a bivalent state .
Histone PTMs are important regulators during all processes that involve alteration of chromatin behaviour, such as transcription, DNA-damage repair, cell-cycle progression and apoptosis. Experimental evidence that chromatin modifications are interdependent is continuously increasing. In this chapter we have discussed examples of cross-talk between histone modifications and between histone PTMs and DNA methylation in different cellular processes. Cross-talking systems offer the advantage of increased flexibility and functional control due to combinatorial modification usage. Despite clear examples for an interconnection between different histone modifications, our understanding of this issue is far from complete. Clearly, there is much more to be discovered in the near future. We would like to encourage the interested reader to take the cited literature as a starting point for further and continuous reading.
• Histone PTMs serve as marks for particular epigenetic states.
• Histone modifications have an impact on chromatin structure either directly, by altering internucleosomal or nucleosome–DNA interactions, or by recruiting modification-dependent binding factors (effector proteins).
• Multiple histone modifications do not function as isolated signals but cross-talk with additional marks to mediate a particular biological readout.
• Modification cross-talk occurs on different levels. First, PTMs have a direct impact on the activity of enzymes that place additional modifications and second, they modulate effector protein binding (Figure 2).
• Cross-talk can occur either in cis, if modifications on the same histone tail are involved or in trans, if cross-talk involves modifications on different histone tails, different nucleosomes or DNA methylation. The readout of combinatorial modification patterns can be either positive or negative.
• The combinatorial readout of histone modifications is essential for multiple cellular processes such as transcription, mitosis or cellular differentiation.
We apologize to many colleagues whose work could not be cited owing to space limitations. We are grateful to Kathy Gelato for comments on the manuscript. Work in the Fischle laboratory is funded by the Max Planck Society, Deutsche Forschungsgemeinschaft (DFG) and the European Union (EU, FP6). Stefan Winter is supported by the EMBO long-term fellowship programme.
- © The Authors Journal compilation © 2010 Biochemical Society