The Prdm family: expanding roles in stem cells and development

The Prdm family is defined on the basis of a conserved N-terminal domain. This domain was originally identified as shared between two proteins: PRDI-BF1 (positive regulatory domain I-binding factor 1), which is now commonly called Prdm1 (PR domain-containing 1); and RIZ1 (retinoblastoma protein-interacting zinc finger gene 1), now called Prdm2 (Buyse et al., 1995) and named the PR (PRDI-BF1-RIZ1 homologous) domain. PR domains are related to the catalytic SET domains (named after the Drosophila factors Suppressor of variegation 3-9, Enhancer of zeste and Trithorax) that define a large group of histone methyltransferases (HMTs, see Glossary, Box 1; see also Fig. 1) (Huang et al., 1998; Schneider et al., 2002; Sun et al., 2008; Wu et al., 2010).

Prmd1 was cloned as a human cDNA encoding a transcriptional repressor that bound the positive regulatory domain of the interferon β promoter (hence the original name PRDI-BF1) (Keller and Maniatis, 1991). Later, Blimp1 (B-lymphocyte-induced maturation protein 1) was independently identified in mouse as a primary regulator of plasma cell differentiation (Turner et al., 1994). PRDI-BF1 and Blimp1 are homologues (Huang, 1994) and to share a PR domain with Riz1, a retinoblastoma-binding protein with tumour suppressor activity in a broad array of tissues (Buyse et al., 1995; Huang et al., 1998; Kim et al., 2003; Steele-Perkins et al., 2001). The family grew further when the leukaemogenic oncogene Evi1 (ecotropic viral integration 1) was revealed to undergo an alternative splicing event, fusing it with what had previously been considered an independent upstream locus (myelodysplasia syndrome-associated protein 1; Mds1) (Fears et al., 1996). This fusion creates a PR domain-containing transcript called Mds1-Evi1, Mecom (Mds1 and Evi1 complex locus) or Prdm3. The same study noted that Caenorhabditis elegans factor EGL-43 (Garriga et al., 1993) also has a PR domain, demonstrating conservation of this family between invertebrates and vertebrates (Fig. 1B) (Fears et al., 1996).

As additional Prdm family members were identified, they were often implicated in neoplasia, including cervical and breast cancer, colorectal and gastric tumours, and leukaemia (see Box 2) (Fog et al., 2012; Morishita, 2007; Schneider et al., 2002). Additionally, a rapidly expanding set of studies demonstrated that Prdms act in stem cells and during cell differentiation and maturation in a wide range of developmental processes. In this review, we discuss recent studies that illustrate common themes and important differences in Prdm action during development, and we examine how these functions relate to the molecular mechanisms of Prdm action.

Prdm family members control gene expression through modification of the chromatin state at target gene promoters. Members are characterized by the presence of an N-terminal PR domain found at various levels of conservation across species (Fig. 1A,B). In all but one family member (Prdm11 being the exception), the N-terminal PR domain is followed by repeated arrays of zinc fingers (Fumasoni et al., 2007; Kinameri et al., 2008; Sun et al., 2008). Included in the family of HMT SET domains, the PR domains are the most diverged from the canonical SET structure (Fig. 1C) (Fears et al., 1996; Sun et al., 2008; Wu et al., 2010). Furthermore, although most SET family members are enzymatically active, HMT activity has been found only in the PR domains of Prdm2, Prdm8 and Prdm9 (also called Meisetz; meiosis-induced factor containing PR/SET domain and zinc-finger motif) (Derunes et al., 2005; Eom et al., 2009; Hayashi et al., 2005). Activity has also been reported for Prdm6 (also called Prism – Prdm in smooth muscle); however, Prdm6 constructs that lack a PR domain still retain histone H4 methyltransferase activity and the nature of this activity requires further clarification (Wu et al., 2008).

Why some Prdm proteins exhibit intrinsic HMT activity and others do not is still unclear. In the larger SET family, mutations in the conserved H/RxxNHxC motif abolish catalytic activity (Rea et al., 2000). However, all PR domains lack this motif, and most significantly the essential histidine in position 2 of NHxC (Fig. 1C, orange box). In Prdm2, C106T mutations reduce methyltransferase activity (Fig. 1C, purple box). The cysteine in this position of Prdm2 corresponds to that in the NHxC motif in other SET proteins (Fig. 1C, orange and purple boxes). Furthermore, this cysteine is required for activity in other SET domains; mutations of this residue in the SET domain of SUV39H1, a human orthologue of suppressor of variegation 3-9, abolish HMT activity of the SUV39H1 protein (Derunes et al., 2005; Rea et al., 2000). Nevertheless, this important cysteine residue is not conserved among the PR domains of Prdm8 and Prdm9, which also exhibit HMT activity, nor in any other Prdm family members.

As most Prdm family members lack intrinsic enzymatic activity, they recruit histone-modifying enzymes to mediate their function. These enzymes include HMTs, which induce repressive chromatin states by methylation of histone lysines, in particular H3K9 (histone H3 lysine 9) by G9a and SuV39H, and H3K27 (histone H3 lysine 27) by the Polycomb repressive complex 2 (PRC2). They also include the enzyme Prmt5 (protein methyltransferase 5), which methylates arginines to create a repressive state, and Lsd1 (lysine-specific demethylase 1, Kdm1a), which primarily promotes repression through erasure of activating H3K4 (histone H3 lysine 4) methylation marks. In addition, Prdm factors can recruit histone deacetylases (HDACs, see Glossary, Box 1), which promote transcriptional repression, and the histone acetyltransferases (HATs; see Glossary, Box 1) p300/CBP (cAMP-responsive element-binding protein-binding protein) and P/CAF (p300/CBP-associated factor), which are associated with transcriptional activation. The activity of these factors is reviewed by Kouzarides (Kouzarides, 2007).

In some cases, histone-modifying enzyme recruitment occurs through the zinc fingers. In others, it is through additional domains, such as the proline-rich domains reported in Prdm1 (Huang et al., 1998) and Prdm3 (Bartholomew et al., 1997). Similarly, proline-rich sequence stretches can be found in Prdm6, in Prdm16 (also known as Mel1 – Mds1-like gene 1) and in the solitary invertebrate homologues of Prdm3 and Prdm16 – Hamlet in Drosophila and EGL-43 in C. elegans (Fig. 1). Prdm1 recruits the repressive chromatin-modifying factors Hdac2 (Yu et al., 2000) and Lsd1 (Su et al., 2009) through its central proline-rich domains. It also associates with G9a via both its zinc fingers (Gyory et al., 2004) and these proline-rich domains (Su et al., 2009). The PR domain of Prdm1 does not exert HMT activity, but it is also required for maximal repressive activity, suggesting that it could mediate as yet undefined protein-protein interactions (Gyory et al., 2003).

Prdm factors also recruit co-repressors that, in turn, associate with histone-modifying enzymes. Prdm1 recruits Groucho family members through its proline-rich domain (Ren et al., 1999), and Prdm2, Prdm3, Prdm16 and Hamlet all bind CtBP (C-terminal binding protein) through canonical PLDLS CtBP-binding sites (Fig. 1A,B) (Endo et al., 2012; Izutsu et al., 2001; Kajimura et al., 2008; Nishikata et al., 2011; Palmer et al., 2001; Quinlan et al., 2006; Van Campenhout et al., 2006). CtBP is used by Prdm16 to repress white fat cell differentiation (Kajimura et al., 2008), and by Hamlet to mediate repressive histone methylation events (H3K27) at target promoters during neuron diversification (Endo et al., 2012).

There are further motifs found in Prdm family members (Fig. 1A), the functions of which are not well characterized. These include an AWS (associated with SET) domain in Prdm6 that is necessary for PR-mediated repression (Davis et al., 2006), and SSX (synovial sarcoma X) and KRAB (Krüppel-associated box) repressor domains in Prdm7 and Prdm9 (Hochwagen and Marais, 2010). Notably, KRAB domain activity is involved in HAT and HMT recruitment in other proteins (Margolin et al., 1994; Pengue et al., 1994).

Prdm proteins can also bind DNA directly. Sequence-specific DNA binding via the zinc-finger domains has been demonstrated for Prdm1, Prdm3, Prdm5, Prdm9, Prdm14 and Prdm16 (Bard-Chapeau et al., 2012; Baudat et al., 2010; Chia et al., 2010; Delwel et al., 1993; Duan et al., 2007; Funabiki et al., 1994; Kuo and Calame, 2004; Ma et al., 2011; Seale et al., 2007). Zinc fingers one and two of the five zinc fingers of Prdm1 preferentially bind DNA sequences containing a GAAAG motif. This sequence can also be recognized by interferon regulatory factors (IFRs, see Glossary, Box 1), and Prdm1 and IRFs can antagonistically regulate promoter activity (Kuo and Calame, 2004). More recently, DNA binding of Prdm1 was shown to be abrogated by CpG methylation, adding complexity to transcriptional regulation by this protein, and suggesting a model in which Prdm1 could be actively excluded from a subset of its target sites, once silencing at the level of DNA methylation has been established (Doody et al., 2011).

Chromatin immunoprecipitation (ChIP) experiments followed by deep sequencing (ChIP-seq) data have been generated for Prdm1, and also for Prdm3 and Prdm14. These studies reveal binding at many sites across the genome, and a handful of targets has been validated (Bard-Chapeau et al., 2012; Chia et al., 2010; Doody et al., 2011; Ma et al., 2011; von Hofsten et al., 2008). It has also become clear that the same Prdm factor regulates divergent sets of targets in different cell types. Therefore, Prdm target selection is under context-dependent control. Recent ChIP-seq studies of Prdm3 found it to bind at many sites across the genome, and gave more information about how such Prdm context dependency might occur. Prdm3 has two separate domains of multiple zinc fingers with different DNA-binding specificities: the N-terminal zinc fingers predominantly bind a GATA-like motif (Delwel et al., 1993); and the C-terminal zinc fingers bind to an ETS-like sequence (Funabiki et al., 1994). Prdm3-bound ETS-like sites lie close to transcriptional start sites (TSS), whereas the Prdm3-bound GATA-like motifs tend to be localized further from TSS. Interestingly, Prdm3 binds to either of these motifs, but not to both motifs at the same promoter at the same time (Bard-Chapeau et al., 2012). Hence, binding site selection (and at least in the case of Prdm3 competition between different binding sites) represents one mechanism for context-dependent control of Prdm factor activity.

Prdm factors can complex with a wide variety of partners (summarized in Table 1), and partner selection might contribute to the context dependency of Prdm action. For example, the above-mentioned Prdm3 ChIP-seq studies, moreover, have revealed a strong enrichment for Ap1- (activator protein 1), Pax2- (paired box gene 2) or Yy1- (yin-yang 1) binding sites in close proximity to Prdm3 binding. As each of these factors is able to bind Prdm3 directly, binding partner selection may also control context-dependent activity. Notably, the Ap1 transcription factor commonly partners the proto-oncogene Fos, and knockdown of Prdm3 weakens Fos occupancy at Ap1 sites found in close proximity to the Prdm3-binding site, suggesting a regulatory interaction between these two factors (Bard-Chapeau et al., 2012).

In summary, Prdm factors act as direct HMTs or recruit a suite of additional histone-modifying enzymes to specific target promoters. They act simultaneously at multiple sites across the genome, and they exhibit context-dependent activity. These features create the capacity to drive and maintain cell state transitions, and to modify the transcriptional output of pleiotropic developmental signalling pathways.

A rapidly expanding body of work has now established Prdm factors as central players in development (summarized in Tables 2 and 3). Here, we discuss some of these recent studies, which illustrate that Prdms often function to promote or repress developmental transitions in cell state by directing a series of chromatin modifications at target loci. We also discuss accumulating evidence that suggests Prdm factors drive cell fate by both promoting a cell type-specific program and by repressing alternative states that the cell can acquire.

The Prdm family first appeared in metazoans (Fumasoni et al., 2007). Three family members have been described in Drosophila, four in C. elegans and seven in sea urchin (Strongylocentrotus purpuratus) (Table 2). There is a base group of 15 Prdm family members in vertebrates (Fig. 1, Table 2). However, gene duplication and loss events have changed this number in different species. Prdm9 is lost from Xenopus and birds, but is duplicated in primates to create Prdm7 (Fumasoni et al., 2007; Oliver et al., 2009). Further paralogue groups (see Glossary, Box 1) of Prdm1a, Prdm1b and Prdm1c, and also Prdm8a and Prdm8b are found in zebrafish (Sun et al., 2008).

Prdm1 is the family member examined most extensively across species (Bikoff et al., 2009). It controls endomesoderm (see Glossary, Box 1) formation in sea urchin and starfish (Livi and Davidson, 2006), may perform a similar role in lamprey (Nikitina et al., 2011), and controls the formation of anterior endomesoderm in Xenopus and zebrafish (Baxendale et al., 2004; de Souza et al., 1999; Wilm and Solnica-Krezel, 2005). Further examples of putative evolutionarily shared function can be found in Tables 2 and 3, although the extent to which Prdm1-dependent developmental mechanisms are conserved between analogous developmental events remains to be determined (Bikoff et al., 2009).

Evolutionary conservation of Prdm1 action is evident in the retina: disruption of Prdm1 in both zebrafish and mouse leads to loss of photoreceptor cells (see Glossary, Box 1) (Brzezinski et al., 2010; Katoh et al., 2010; Wilm and Solnica-Krezel, 2005), and studies in the mouse have addressed the mechanism of Prdm1 action. Retinal bipolar cells (see Glossary, Box 1) and photoreceptor cells (see Glossary, Box 1) derive from a shared Chx10 (visual system homeobox 2, Vsx2)-positive proliferating precursor cell (Fig. 2). Prdm1 is transiently expressed in nascent photoreceptors. Ectopic expression of Prdm1 reduces the number of bipolar cells and increases the number of photoreceptors (Brzezinski et al., 2010; Katoh et al., 2010). Conversely, ectopic expression of Chx10 converts photoreceptors into bipolar cells (Livne-Bar et al., 2006). Prdm1 binds multiple canonical Prdm1-binding sites in the Chx10 promoter in vivo, and represses expression of a Chx10-luciferase reporter construct in vitro (Brzezinski et al., 2010; Katoh et al., 2010). In Prdm1 mutants, a nascent Chx10-negative photoreceptor forms, but as differentiation proceeds, Chx10 is re-expressed and the cell becomes re-specified to a bipolar or proliferating precursor state. These data suggest that the normal role of Prdm1 is not to induce photoreceptor formation, but rather to stabilize or maintain photoreceptor identity by repressing alternative differentiation programs (Fig. 2).

Further recent reports emphasize the role of Prdm1 in maintaining one developmental state by repressing the transition to another. Neonatal mice feed solely on milk, and this requires enterocytes (see Glossary, Box 1) that are metabolically tuned for processing lactose-rich fluid. During weaning onto solid food, this requirement is changed. Thus, adult enterocytes exhibit reduced expression of factors required for milk processing and enhanced expression of factors required for breakdown of complex carbohydrates. Interestingly, Prdm1 is found only in neonatal and not in adult enterocytes, and, during weaning, neonatal Prdm1-positive enterocytes are systematically replaced by Prdm1-negative adult ones (Harper et al., 2011; Muncan et al., 2011). Lineage tracing shows that the adult Prdm1-negative enterocytes derive from a Prdm1-positive population, presumably a subset of neonatal enterocytes that have downregulated Prdm1 and undergone expansion (Harper et al., 2011). Importantly, Prdm1-deficient neonatal enterocytes show precocious activation of the adult gene expression program and simultaneous loss of the neonatal one (Harper et al., 2011; Muncan et al., 2011). Thus, Prdm1 represses adult and promotes neonatal enterocyte metabolism, and its loss is a trigger for the programmed developmental transition in enterocytes required for weaning.

Prdm1 action has been extensively examined in the B-cell lineage during the differentiation of antibody-secreting plasma cells (Turner et al., 1994). Prdm1 is not only needed to acquire plasmocytic properties, it is continually required for the maintenance of long-lived plasma cells, suggesting that, at some loci, it maintains a repressive state (Shapiro-Shelef et al., 2005). During plasma cell differentiation, Prdm1 directly represses the proliferation factor Myc through recruitment of HDACs (Yu et al., 2000); this leads to a strong reduction of histone 3 acetylation (H3Ac) at the Myc promoter. Prdm1 also represses Ciita (MHC class II trans-activator) through Lsd1 (Su et al., 2009). RNAi knockdown experiments showed that both Prdm1 and Lsd1 repress H3Ac and histone 3 lysine 4 dimethylation (H3K4me2) at this locus; Lsd1 also suppressed H3K4me3 (Su et al., 2009). Other targets of Prdm1 in plasma cell differentiation are Pax5 (paired box 5) (which is required for B cell fate specification, must be downregulated during plasma cell differentiation and is also active in the germinal centre) and Bcl6 (B-cell lymphoma 6), which promotes germinal centre B cell identity (reviewed by Belz and Kallies, 2010; Crotty et al., 2010; Martins and Calame, 2008). Notably, Bcl6 also acts as a transcriptional repressor and forms a negative-feedback loop with Prdm1 (reviewed by Crotty et al., 2010).

Prdm1 also acts as a repressor in subsets of effector CD4⁺ and CD8⁺ T cells (Cretney et al., 2011; Kallies et al., 2006; Kallies et al., 2009; Martins et al., 2006; Nutt et al., 2008; Rutishauser et al., 2009) and Bcl6 is once more a target (Crotty et al., 2010). Nevertheless, Prdm1 appears to modulate predominantly different sets of targets between B- and T-cell lineages (Martins and Calame, 2008). One repressed target in T cells is the cytokine Il2 (interleukin 2). Il2 regulates the initial expansion of naïve T cells then, acting through Stat5 (signal transducer and activator of transcription 5), it induces Prdm1 expression creating a feedback loop that downregulates its own expression during the later stages of T-cell differentiation (Martins et al., 2008). Prdm1 also functions in natural killer (NK) cells (Kallies et al., 2011; Smith et al., 2010), dendritic cells (Chan et al., 2009) and macrophages (Chang et al., 2000). (The extensive literature covering Prdm1 function in immunology is beyond the scope of this review; thus, for comprehensive recent reviews, see Belz and Kallies, 2010; Crotty et al., 2010; Martins and Calame, 2008; Nutt et al., 2007.)

Interestingly, although Prdm1 is characterized as a transcriptional repressor, it can also promote transcriptional activation in combination with Irf4 (interferon regulatory factor 4) in regulatory T cells. At the Il10 (interleukin 10) locus, Irf4 removes repressive H3K27 and Prdm1 promotes upregulation of activating H3K4 histone methylation (Cretney et al., 2011). Such bi-directional transcriptional control by Prdm1 may occur in other contexts. In particular, examination of putative Prdm1 targets (defined by the presence of canonical Prdm1 binding sites) in Prdm1 mutant embryonic enterocytes noted a significant number of genes that showed downregulation, suggesting they may be positively regulated by this factor (Harper et al., 2011).

How can a single Prdm act both as a repressor and activator in the same cell during a single developmental event? Recent studies of Prdm16 activity illustrate a mechanism through which this can occur. Prdm16 promotes brown adipose tissue (BAT) identity and simultaneously represses alternative fates (Kajimura et al., 2009; Kajimura et al., 2008; Seale et al., 2008; Seale et al., 2011; Seale et al., 2007). BAT is specialized to dissipate chemical energy by creating heat in defence against cold. It is common in human babies and small mammals. By contrast, the common form of human adult fat is white adipose tissue (WAT), which stores chemical energy. Weight gain is primarily caused by an energy imbalance; in other words, the excess of energy intake over energy consumption. Therefore, understanding the molecular principles underlying brown adipogenesis is a step towards BAT gene program manipulation, which could be used to combat human obesity (Fruhbeck et al., 2009). Lineage tracing shows that BAT derives from Myf5 (myogenic factor 5)-positive precursor cells that also give rise to skeletal muscle (Seale et al., 2008) (Fig. 3A). Removal of Prdm16 from cultured brown adipocyte (see Glossary, Box 1) precursor cells by RNAi leads to loss of BAT-specific genes and upregulation of the myocyte (see Glossary, Box 1) program, leading to their re-specification as skeletal myocytes (see Glossary, Box 1). A similar, albeit less robust, phenotype is observed in Prdm16 mutant mice (Seale et al., 2007; Seale et al., 2008) (Fig. 3A).

Prdm16 was originally identified as a leukaemogenic oncogene (Mochizuki et al., 2000; Morishita, 2007; Nishikata et al., 2003). Sequence-specific DNA binding through its two zinc-finger domains is crucial for its transcriptional activity in this pathogenic context (Nishikata et al., 2003). When Prdm16 promotes the BAT-specific genetic program, however, it does not target DNA itself. Rather Prdm16 binds to and is targeted by the sequence-specific DNA-binding protein C/EBPβ (CCAAT/enhancer-binding protein β) in order to initiate the BAT program (Fig. 3B) (Kajimura et al., 2009; Seale et al., 2009; Seale et al., 2007). Forced expression of Prdm16 and C/EBPβ can convert naïve fibroblasts into BAT, an exciting potential therapeutic tool (Kajimura et al., 2009; Seale et al., 2008). Among the factors induced by the Prdm16-C/EBPβ complex are Pparγ (peroxisome-proliferator-activated receptor γ) and Ppargc1α (Pparγ co-activator 1α). After induction of these factors, they now form new complexes with Prdm16 to again target its activity, this time to drive terminal BAT maturation (Fig. 3B).

Prdm16 also induces the formation of beige cells, BAT-like adipocytes that emerge in subcutaneous WAT as adaptive responses to cold exposure and to overfeeding (Seale et al., 2011). Beige cells may form via direct trans-differentiation of WAT; however, their definitive origin remains unclear (Barbatelli et al., 2011; Seale et al., 2011). In cultured adipocytes, Prdm16 is recruited to promoters of BAT-specific genes, activating their expression, and is simultaneously recruited to promoters of WAT-specific genes and represses their expression through recruitment of the co-repressor CtBP (Fig. 3B) (Kajimura et al., 2008). These studies show how Prdm16 acts both as a transcriptional activator and repressor within the same cell by forming different molecular complexes in a promoter-context-dependent manner (Fig. 3B).

An outstanding feature of Prdm16 action in BAT formation is that it is targeted to specific loci via other DNA-binding partners. Prdm3 may have a similar mode of action at the Pbx1 (pre-B-cell leukaemia homeobox 1) locus (Shimabe et al., 2009), and Prdm8 at the Cdh11 (cadherin 11) locus (Ross et al., 2012). Prdm8 is expressed in several regions of the developing nervous system (Table 2) (Kinameri et al., 2008; Komai et al., 2009) and is co-expressed in several places with the DNA sequence-specific bHLH (basic helix-loop-helix) transcription factor Bhlhb5 (Bhlhe22 – Mouse Genome Informatics) (Ross et al., 2012). Mice lacking Prdm8 or Bhlhb5 have very similar behavioural phenotypes and defects in axon outgrowth. Bhlhb5 targets several loci involved in neuronal development, including Cdh11, where it recruits Prdm8 that in turn represses transcription (Ross et al., 2012); whether this is through the intrinsic H3K9 methyltransferase activity of Prdm8 is unclear (Eom et al., 2009).

The above examples demonstrate how Prdms can promote cell fate choice and cell differentiation. Prdm factors can act by repressing the ability of a cell to transition to an alternative fate or to an alternative differentiation state. They can act in a cell context-dependent manner. For example, the targets of Prdm1 are different in B- and T-cell lineages. They can also act both as activators and repressors within the same cell in a promoter context-dependent manner. For Prdm16, this occurs because different binding partners control activator versus repressor activity. Furthermore, the DNA targeting activity of Prdm zinc fingers is well established; nevertheless, as demonstrated for Prdm16 and Prdm8, targeting can also occur through complex formation with sequence-specific DNA-binding partners. During BAT maturation, Prdm16 switches partners, thus enhancing context-dependent transcriptional flexibility.

Pluripotent germ cells (see Glossary, Box 1), as well as pluripotent stem cells in culture- and tissue-specific stem cells, must all generate and maintain highly specialized cell states. Recent evidence demonstrates that several members of the Prdm family play important roles in creating these states.

In the haematopoietic system, defects in stem cell function can result in a wide range of malignant phenotypes, including leukaemia; both Prdm3 and Prdm16 were originally identified because their dysregulation causes myeloid malignancies (see Box 2) (reviewed by Morishita, 2007). The normal role of Prdm3 and Prdm16, however, is to maintain quiescent long-term and proliferating short-term haematopoietic stems cells (HSCs) (Aguilo et al., 2011; Chuikov et al., 2010; Goyama et al., 2008; Zhang et al., 2011). Leukaemic stem cells may share self-renewal machinery with normal HSCs, and elucidating of how Prdm3 and Prdm16 govern HSC self-renewal might provide biological insight into pathogenesis of these leukaemias (Goyama et al., 2008). Care is required in such interpretations; nevertheless, comparison of gene expression in wild-type and Prdm3 mutant HSCs and further assessment against putative Prdm3 targets in acute myeloid leukaemia samples suggest some common pathways (Goyama et al., 2008). Prdm16 and Prdm3 have highly related molecular structures. They also both cause acute myeloid leukaemia and loss of either leads to similar phenotypes in HSCs. However, transcriptome examination of Prdm3 mutant and Prdm16 mutant HSCs suggests they are regulating somewhat different programs (Aguilo et al., 2011). For example, Prdm3 promotes Pbx1 expression, whereas Prdm16 may repress it (Aguilo et al., 2011; Goyama et al., 2008; Shimabe et al., 2009). Interestingly, Prdm16, but not Prdm3, also maintains embryonic cortical stem cells, and this occurs in part through regulating oxidative stress (Chuikov et al., 2010; Kinameri et al., 2008).

Prdm1 and Prdm14 play essential roles in the generation of pluripotent primordial germ cells (PGCs; see Glossary, Box 1), the precursors of oocytes in females and spermatozoa in males. In addition, after germ cell differentiation, Prdm9 positions recombination hotspots during meiosis (reviewed by Hochwagen and Marais, 2010) (see Box 3). Mouse PGCs are induced from proximal epiblast cells at the embryonic/extra-embryonic boundary in response to a bone morphogenic protein 4 (Bmp4) signal from the extra-embryonic ectoderm. Bmp4 induces both Prdm1 and Prdm14, and both Prdm1 and Prdm14 mutants cannot form PGCs (Ohinata et al., 2009; Ohinata et al., 2005; Vincent et al., 2005; Yamaji et al., 2008). However, in both mutants, a limited number of transient PGC-like cells do form and this has allowed the use of transcriptome profiling to determine the genetic programs controlled by these factors during PGC specification (Kurimoto et al., 2008; Yamaji et al., 2008). During PGC specification, the somatic epiblast gene expression program must be repressed, PGC-specific factors upregulated and a pluripotency program acquired. Prdm1 and Prdm14 coordinate these processes. In Prdm1 mutant PGC-like cells, almost all somatic genes are not downregulated; instead, they remain expressed at the same level as in adjacent somatic cells (Kurimoto et al., 2008). By contrast, Prdm14 mutant PGC-like cells repress the somatic program normally (Yamaji et al., 2008). However, both Prdm1 and Prdm14 are required for the proper initiation and coordination of the PGC-specific gene expression program, and initial evidence suggests they may control divergent sets of targets that remain to be fully characterized (Kurimoto et al., 2008; Yamaji et al., 2008). Finally, both Prdm1 and Prdm14 promote PGC pluripotency. The pluripotency regulator Sox2 (SRY box-containing 2) is expressed in the epiblast before PGC induction; however, nascent PGCs and surrounding somatic epiblast cells both lose Sox2 expression. Sox2 then becomes re-expressed in the PGCs, and this requires both Prdm1 and Prdm14 activity (Kurimoto et al., 2008; Yamaji et al., 2008). Prdm1, but not Prdm14, is also required for the PGC expression of Nanog, a second pluripotency factor (Kurimoto et al., 2008). (For further reviews of this subject, see Bikoff and Robertson, 2008; Saitou, 2009.)

PGC specification also involves genome-wide epigenetic reprogramming (Saitou, 2009; Saitou et al., 2012). Early in PGC formation, Prdm1 associates with the arginine-specific HMT Prmt5, and this association results in high levels of dimethylated histone 2A and histone 4 arginine 3 (H2A/H4 R3me2) (Ancelin et al., 2006). As PGC differentiation continues, translocation of the Prdm1-Prmt5 complex from the nucleus into the cytoplasm correlates with loss of H2A/H4R3me2 levels (Ancelin et al., 2006). Prdm14 is required for the reduction of PGC H3K9me2 levels during the early steps of epigenetic reprogramming. This activity presumably involves Prdm14-mediated repression of Ehmt1 (euchromatic histone methyltransferase 1), a HMT that promotes H3K9me2 in early embryonic tissues. Prdm14 also acts via an undetermined mechanism to upregulate H3K27me3 levels 1 day later (Yamaji et al, 2008).

In culture, stem cells must actively maintain a pluripotent state, and they must simultaneously repress programs that will promote commitment and differentiation into specific cell types. These are roles carried out by Prdm1 and Prdm14 in the forming PGCs, so do these factors also function in cultured stem cells? PGCs can be changed by culturing into pluripotent stem cells known as embryonic germ cells (EGCs; see Glossary, Box 1). The formation of these EGCs requires Prdm14 activity and the loss of Prdm1 expression (Ancelin et al., 2006; Durcova-Hills et al., 2008; Yamaji et al., 2008). Somatic cells, such as adult fibroblasts, can also be artificially converted into a stem cell-like state in culture. These induced pluripotent stem cells (iPSCs; see Glossary, Box 1) cells are produced using a cocktail of Sox2, Oct4 (Pou5f1; POU domain, class 5, transcription factor 1), Myc and Klf4 (Kruppel-like factor 4). Notably, Prdm14 also has iPSC induction activity, and can enhance the activity of this cocktail, even if c-Myc or Klf4 are removed (Chia et al., 2010).

Conversely, neither Prdm1 nor Prdm14 are required for the derivation of mouse embryonic stem cells (ESCs; see Glossary, Box 1), although lineage tracing has shown that ESCs commonly arise from Prdm1-positive precursors (Chu et al., 2011). Nonetheless, once formed, both mouse and human ESCs continuously require Prdm14 to maintain the ESC state. Prdm14 RNAi knockdown leads to downregulation of ESC genes and upregulation of the extra-embryonic endoderm program along with morphological signs of differentiation. The phenotype is slightly different between mouse and human ESCs, possibly reflecting the differing tissues from which these cells are derived (Chia et al., 2010; Ma et al., 2011; Tsuneyoshi et al., 2008). Genome-wide ChIP-seq studies followed by target validation found that Prdm14 binds and activates the Oct4 promoter, and co-binds with Sox2, Nanog or Oct4 at many other sites (Chia et al., 2010; Ma et al., 2011).

In summary, Prdm factors promote and maintain germ and stem cell fate. The extent to which the same programs are controlled between different factors remains to be determined. Although very similar in structure, Prdm3 and Prdm16 control differing gene expression programs in HSCs. In PGCs Prdm1 and Prdm14 also control complementary programs rather than a shared one. To what degree Prdm14 function in PGCs relates to its activity in cultured stem cell types is also unclear.

Developmental processes produce a huge diversity of cell types. Surprisingly, in order to generate this diversity, nature uses only a handful of signalling systems, which function over and over again in different developmental contexts to create different outcomes (Pires-daSilva and Sommer, 2003). The output of signalling systems such as the transforming growth factor β (TGFβ) and Notch pathways is to regulate gene transcription through the action of dedicated sequence-specific transcription factors. Therefore, in order to ensure that these pleiotropic factors do not repeatedly activate the same targets in different developmental or cellular contexts, molecular mechanisms must exist to modulate the transcriptional activity in a ‘context-dependent’ manner. Recent studies, summarized below, suggest that Prdms can modify the transcriptional output of such developmental signalling pathways.

TGFβ signalling controls transcriptional regulation via sequence-specific Smad (MAD homologue) transcription factor activity (Moustakas and Heldin, 2009). A number of studies have shown that Prdm3 and Prdm16 bind Smads in order to repress Smad-mediated transcriptional activation (Alliston et al., 2005; Bjork et al., 2010; Sato et al., 2008; Takahata et al., 2009). This Prdm-mediated TGFβ repression requires recruitment of CtBP, which in turn recruits HDACs to deacetylate histones at Smad target promoters (Alliston et al., 2005; Izutsu et al., 2001). Smads usually recruit p300 to acetylate histones and thereby activate transcription. Surprisingly, however, binding of Prdm3 to Smads is also facilitated by p300. This means that the Prdm3-Smad complex, although primarily acting as a histone-deacetylating repressor complex, also contains putative histone-acetylating activity, providing a possible strategy for tuning Prdm3-Smad complex activity in different developmental contexts (Alliston et al., 2005).

In cells where Notch signalling is activated, the Notch intracellular domain (NICD) translocates to the nuclei where it binds to the DNA sequence-specific transcription factor CSL [C-promoter binding factor 1 (also called recombination signal binding protein J), Suppressor of Hairless and Lag-1)]. The resultant NICD/CSL complex activates Notch-target genes (Bray and Bernard, 2010). Recent studies in invertebrates have demonstrated a close relationship between Prdm function and NICD/CSL-mediated transcriptional regulation.

In C. elegans, EGL-43 (Fig. 1B,C) and Notch cooperate in two sequential signalling events during vulval development (Hwang et al., 2007; Rimann and Hajnal, 2007). Four precursor cells give rise to the vulva: one pair of cells directly adopts a ventral uterine precursor cell (VU) fate and the second pair undergoes a Notch-mediated binary cell fate choice. The cell with high Notch becomes another VU, whereas the cell with low Notch becomes the anchor cell (AC). EGL-43 is required to implement this Notch-mediated decision, as in egl-43 mutants both cells adopt an AC fate. In the nascent VU cell, Notch is subsequently required to properly maintain EGL-43 expression. Finally, as the VU cells differentiate into π cells, Egl43 and Notch co-activate the lin-11 homeobox gene (Hwang et al., 2007; Rimann and Hajnal, 2007). EGL-43 is additionally required in a Notch-independent manner for AC invasion of the vulval epithelium.

In Drosophila, the four cells of the notum mechanosensory organs (see Glossary, Box 1) derive from a single precursor via a stereotyped lineage (Lai and Orgogozo, 2004). The fates of all four cells are determined by the Notch signalling pathway. Thus, Notch-mediated binary cell fate decisions guide the formation of an external sensory (ES) neuron (low Notch) versus a glial cell (high Notch), and also the formation of a hair cell (low Notch) versus a socket cell (high Notch) (Fig. 4). Hamlet (Fig. 1B,C) is expressed in the IIIB intermediate precursor cell that gives rise to ES neurons and glia, and also in the nascent ES neurons and glia (Fig. 4) (Moore et al., 2002; Moore et al., 2004). When hamlet function is lost, the ES neuron and glia form, and they initially express neuronal and glial markers. However, as differentiation progresses, they re-specify into a second hair and socket cell (Moore et al., 2004). One peculiar aspect of this phenotype is that in hamlet mutants, the high-Notch glial cell transforms into a hair cell, but the hair cell is usually defined by a low-Notch state. Similarly, the low-Notch ES neuron converts into a socket, and this is a cell fate that is usually defined by high Notch activity (Fig. 4) (Moore et al., 2004).

Studies of Drosophila olfactory receptor neuron (ORN; see Glossary, Box 1) diversification have better illuminated the interaction between Hamlet and Notch (Endo et al., 2012). Hamlet acts as an epigenetic ‘reset switch’ that enables the Notch pathway to induce multiple distinct developmental outcomes for olfactory neurons. In Drosophila, there are three primary ORN identities, all of which are defined by Notch state (Endo et al., 2012). Two, Naa (high Notch) and Nab (low Notch), are sibling cells; the third, Nba (high Notch), is their cousin. Importantly, the Naa and Nab ORNs derive from a precursor cell, which was itself formed as the result of high Notch activity. Hamlet acts in the nascent Naa and Nab ORNs to erase the high-Notch transcriptional state of the parental cell. This then enables a successful new Notch signalling event to be used in the nascent ORNs. In hamlet mutants, the parental cell Notch state is not properly reduced, and the usually low Notch Nab ORN undergoes a fate switch to a high-Notch Nba identity (Endo et al., 2012). At the molecular level, Hamlet forms a complex with CtBP, and alters histone methylation status and organization at the promoter of the Notch target Enhancer of split m3 [E(spl)m3]. At this locus, Hamlet upregulates H3K27me3, downregulates H3K4me3 and increases histone density at the CSL-binding site; correspondingly, the NICD/CSL activator complex is unable to bind properly to the E(spl)m3 promoter and the locus becomes refractory to Notch-mediated induction (Fig. 5A) (Endo et al., 2012).

Transition from a precursor to an immature neuronal state also occurs during mouse cortical neurogenesis and is controlled by Notch signalling. During this transition, cells switch from Prdm16 to Prdm8 expression (Kinameri et al., 2008). However, whether these factors also influence Notch pathway activity is yet to be addressed.

A further study links the bi-directional transcriptional activity that can be associated with Prdm factor activity to the modification of signalling pathway activity. This study shows that Prdm2 acts at an oestrogen-activated locus, first as a repressor in advance of signalling and then, acting at the same locus, it switches to an activator during signalling (Fig. 5B). In this way, it accentuates the transcriptional response to oestrogen action. In line with this, Prdm2 mutant mice have reduced reproductive capabilities and impaired responses to oestrogen signalling, such as markedly reduced mammopoiesis during pregnancy and in response to experimental hormone administration (Carling et al., 2004). At the molecular level, Prdm2 is associated with the oestrogen target pS2 promoter in advance of oestrogen receipt. In this state, it mediates promoter repression via H3K9 methylation. Notably, Prdm2 is in a complex with p300, but p300 does not exert its activity when in this state. The activity of the Prdm2-p300 complex changes upon oestrogen activation. The estrogen receptor (ER) enters the nucleus, Prdm2 dissociates from the DNA, it binds directly to the ER, and together they create a Prdm2-p300-ER activation complex that drives histone acetylation, resulting in a transcriptionally active chromatin state (Carling et al., 2004) (Fig. 5B).

Taken together, these studies show that the molecular activity of Prdm factors makes them good candidates to modify the transcriptional output of signalling system activation. EGL-43 in C. elegans and Hamlet in Drosophila are linked to the Notch pathway (Endo et al., 2012; Hwang et al., 2007; Moore et al., 2004; Rimann and Hajnal, 2007), and Prdm3 and Prdm16 to TGFβ (Alliston et al., 2005; Bjork et al., 2010; Sato et al., 2008; Takahata et al., 2009). Bi-directional transcription regulation activity of Prdm2 at the pS2 promoter provides one solution of how to amplify the outcome of signalling at a specific target promoter (Carling et al., 2004).

In summary, Prdm factors can act as direct HMTs or they can recruit a suite of histone-modifying enzymes to target promoters. Besides being associated with tumour formation, they function in many developmental contexts. They specify cell fate choice, and drive cell differentiation and maturation events. They can also maintain specialized cell fates, including that of HSCs and pluripotent ESCs (Aguilo et al., 2011; Chia et al., 2010; Chuikov et al., 2010; Goyama et al., 2008; Ma et al., 2011; Tsuneyoshi et al., 2008). Prdm9 has a divergent role in controlling meiotic crossing over (Hayashi et al., 2005; Hochwagen and Marais, 2010).

Importantly, Prdm factors show strong context dependency. Prdm-controlled genetic programs vary significantly for the same factor from cell type to cell type, e.g. Prdm1 in B and T cells (Martins and Calame, 2008). They also vary for closely related factors in the same cell type e.g. Prdm3 and Prdm16 in HSCs (Aguilo et al., 2011; Goyama et al., 2008). Even so, contained within these divergent programs may be key shared factors. For example, Prdm1 repression of Bcl6 occurs in both B and T cell lineages (Crotty et al., 2010; Martins and Calame, 2008), and also in osteoclasts (see Glossary, Box 1) (Miyauchi et al., 2010; Nishikawa et al., 2010). Prdms can also have bi-directional transcriptional activity in the same process. For example, Prdm16 acts simultaneously at different loci as an activator or repressor during BAT specification (Kajimura et al., 2008) and Prdm2 transitions from repressor to activator at a single locus in response to oestrogen (Carling et al., 2004). What cellular contexts and promoter-specific mechanisms determine whether a Prdm factor will control gene transcription (i.e. how is cell type context dependency set) and whether it will act either as activator or repressor at a particular locus? Elucidating this will be central to our future understanding of how Prdm factors coordinate developmental processes.

Finally, recent evidence shows that Prdm factors can regulate the outcome of developmental signalling pathway activity. At the transcriptional level, this is best illustrated by EGL-43- and Hamlet-mediated regulation of Notch in invertebrates (Endo et al., 2012; Hwang et al., 2007; Rimann and Hajnal, 2007), and by Prdm3- and Prdm16-mediated regulation of Smad activity in vertebrates (Alliston et al., 2005; Bjork et al., 2010; Izutsu et al., 2001; Sato et al., 2008; Takahata et al., 2009). Because Prdm factors can create global changes in chromatin state, they have interesting potential to construct cell-type-specific chromatin contexts upon which developmental signalling pathways can operate.

We thank members of the Moore laboratory for many discussions, and we apologize to authors whose work was omitted from this article because of space limitations.