Mediator kinase module and human tumorigenesis

Abstract

Mediator is a conserved multi-subunit signal processor through which regulatory informatiosn conveyed by gene-specific transcription factors is transduced to RNA Polymerase II (Pol II). In humans, MED13, MED12, CDK8 and Cyclin C (CycC) comprise a four-subunit “kinase” module that exists in variable association with a 26-subunit Mediator core. Genetic and biochemical studies have established the Mediator kinase module as a major ingress of developmental and oncogenic signaling through Mediator, and much of its function in signal-dependent gene regulation derives from its resident CDK8 kinase activity. For example, CDK8-targeted substrate phosphorylation impacts transcription factor half-life, Pol II activity and chromatin chemistry and functional status. Recent structural and biochemical studies have revealed a precise network of physical and functional subunit interactions required for proper kinase module activity. Accordingly, pathologic change in this activity through altered expression or mutation of constituent kinase module subunits can have profound consequences for altered signaling and tumor formation. Herein, we review the structural organization, biological function and oncogenic potential of the Mediator kinase module. We focus principally on tumor-associated alterations in kinase module subunits for which mechanistic relationships as opposed to strictly correlative associations are established. These considerations point to an emerging picture of the Mediator kinase module as an oncogenic unit, one in which pathogenic activation/deactivation through component change drives tumor formation through perturbation of signal-dependent gene regulation. It follows that therapeutic strategies to combat CDK8-driven tumors will involve targeted modulation of CDK8 activity or pharmacologic manipulation of dysregulated CDK8-dependent signaling pathways.

Introduction

A central outstanding question in metazoan biology concerns the means by which developmental, environmental or homeostatic signals are effectively coupled with precise gene expression output sufficient to specify cell fate and function and, further, how disruption of these processes leads to altered patterns of gene expression and pathologic outcomes, including developmental defects and cancer. While the underlying mechanisms have not yet been completely elucidated, this process nonetheless depends to a large extent on intermediary activities that link signal-regulated and chromatin-bound transcription factors with the core RNA Pol II transcription machinery. Central among these so-called “co-regulators” is the multi-protein Mediator, originally discovered in the budding yeast Saccharomyces cerevisiae, and since found to be broadly conserved across a vast spectrum of eukaryotes, ranging from protists to metazoans (Bourbon, 2008; Conaway & Conaway, 2011b; Myers & Kornberg, 2000). Since the identification of a metazoan counterpart to yeast Mediator over 15 years ago, a convergent body of biochemical and molecular genetic studies have confirmed their structural and functional relationship as an integrative hub through which regulatory information conveyed by signal activated transcription factors is transduced to Pol II (Allen & Taatjes, 2015; Chadick & Asturias, 2005; Kornberg, 2005; Malik & Roeder, 2010). Although metazoan Mediator complexes have been shaped during evolution by substantive diversification and expansion in subunit number and sequence, biochemical and structural analyses of yeast and metazoan Mediator complexes have nonetheless revealed a remarkable degree of conservation with respect to their overall structural organization, suggesting the existence of a universal Mediator interface (Bourbon, 2008; Tsai et al., 2013, 2014).

Functionally, Mediator acts to control and coordinate multiple steps in the transcription process, including preinitiation complex (PIC) formation through chromatin reconfiguration and Pol II recruitment, early initiation events linked to Pol II promoter escape, Pol II pausing and elongation and co-transcriptional RNA processing (Figure 1) (Ansari & Morse, 2013; Carlsten et al., 2013; Conaway & Conaway, 2013; Knuesel & Taatjes, 2011; Lewis & Reinberg, 2003; Plaschka et al., 2015; Struhl, 2005; Yin & Wang, 2014). Furthermore, recent studies have identified Mediator as a key marker and functional constituent of super-enhancers that drive high-level expression of genes encoding key regulators of cell identity (Whyte et al., 2013; Yin & Wang, 2014). In this regard, Mediator is required for super-enhancer activity and also promotes physical and functional communion between regulator-bound enhancers and the basal transcriptional apparatus on core promoters (Hnisz et al., 2013; Kagey et al., 2010; Muto et al., 2014; Phillips-Cremins et al., 2013; Whyte et al., 2013). Finally, activating long non-coding RNAs with enhancer-like activity have recently been shown to interact with Mediator and regulate its chromatin localization and kinase activity (Figure 1) (Lai et al., 2013). At least in part, the pleiotropic functional activity of Mediator derives from its dynamic conformational plasticity. Among the constellation of transcriptionally relevant activities with which Mediator physically interacts, many induce demonstrable conformational change in Mediator sufficient to impart specific and consequential functional interactions related to its diverse roles in the transcription process (Asturias et al., 1999; Cai et al., 2009; Davis et al., 2002; Ebmeier & Taatjes, 2010; Meyer et al., 2010; Taatjes et al., 2002; Tsai et al., 2014; Wang et al., 2014b).

Figure 1.

Mediator is a central integrator and processor of Pol II transcription. Mediator transduces regulatory information conveyed by signal-activated transcription factors to effect changes in gene expression programs that control diverse biological processes, including development, differentiation and homeostasis. To achieve this, Mediator functions at multiple steps in the Pol II transcription process. Structurally, it cooperates with cohesion to promote long-range chromatin interactions (via looping) by bridging enhancer-bound activators with the Pol II machinery. Functionally, it facilitates pre-initiation complex formation through chromatin reconfiguration and Pol II recruitment onto core promoters, regulates early initiation events linked to Pol II initiation and promoter escape and regulates Pol II pausing and/or pause release. Mediator also participates in co-transcriptional RNA processing (data not shown). (see colour version of this figure at www.informahealthcare.com/bmg).

Structurally, Mediator is assembled from multiple constituent subunits (25 and 30 in yeast and mammals, respectively) arranged into four distinct modules, including “head”, “middle”, “tail” and “kinase” modules (Figure 2) (Asturias et al., 1999; Bourbon, 2008; Cai et al., 2009; Elmlund et al., 2006; Naar et al., 2002; Taatjes et al., 2002; Tsai et al., 2014; Wang et al., 2014b). Mediator comprising the head, middle and tail modules (hereafter referred to as “core” Mediator) binds tightly to Pol II in the so-called “holo-enzyme”, while the kinase module exists in variable association with core Mediator. Gene-specific transcription factors generally bind Mediator through its tail and kinase domains, and the regulatory information conveyed by these factors is transduced through Mediator middle and head modules to Pol II, with which Mediator interacts primarily through head module contacts. By virtue of its ability to link signal-activated transcription factors with Pol II, Mediator fulfills a central role as a signal processor and transducer. In this regard, the Mediator kinase module represents a major ingress of signal transduction through Mediator. Individual kinase module subunits have been linked genetically and biochemically to a number of pre-eminent developmental and key oncogenic signaling pathways, including the Wnt, SHH, EGF and Notch pathways, among others (Carrera et al., 2008; Firestein et al., 2008; Fryer et al., 2004; Grants et al., 2015; Kim et al., 2006; Lehner et al., 2006; Li et al., 2014b; Morris et al., 2008; Rocha et al., 2010; Spaeth et al., 2011; Yin & Wang, 2014; Zhou et al., 2006, 2012). In addition to its role as a terminal transducer of oncogenic signals promulgated upstream, emerging studies have identified the Mediator kinase module itself to be a direct target of genetic alteration in human tumors, revealing an additional level at which signal dysregulation through the kinase module may be instigated. For example, overexpression of CDK8 through amplification-dependent and -independent mechanisms has been shown to be oncogenic in colorectal carcinoma and melanoma, respectively, in a manner dependent on CDK8 kinase activity (Firestein et al., 2008; Kapoor et al., 2010). More recently, mutations in MED12 leading to impaired Mediator-associated CDK8 activity have been found at high frequency in uterine leiomyomas (Kampjarvi et al., 2014; Makinen et al., 2011b; Turunen et al., 2014). Herein, we review current concepts concerning the structural and molecular biology of the Mediator kinase module, including the nature of activity-dependent subunit interactions and the functional impact of pathogenic alterations in subunit expression and sequence. These considerations highlight the component nature and oncogenic vulnerability of the Mediator kinase module and suggest new therapeutic strategies with possible applications across a range of human tumors.

Figure 2.

Modular organization of human Mediator. The Mediator core (~1.5 mDa; 26 subunits) is comprised of head (11 subunits), middle (9 subunits) and tail (6 subunits). Also shown is the dissociable kinase module (~0.5 mDa; 4 subunits). The Mediator head module is the primary interface for Pol II, whereas transcription factors bind principally to the tail and kinase modules. (see colour version of this figure at www.informahealthcare.com/bmg).

Kinase module function and regulation

The Mediator kinase module (~430 kDa in yeast; ~560 kDa in humans) is nominally comprised of four subunits: MED13, MED12, CycC and CDK8, with paralogs of all but CycC identified in humans and some vertebrates (discussed below) (Borggrefe et al., 2002; Bourbon, 2008; Hengartner et al., 1995; Sato et al., 2004). Individual kinase module subunits were originally identified in yeast as extragenic suppressors of a conditional growth phenotype arising from truncations in the Pol II large subunit carboxyl terminal domain (CTD), and later as suppressors of a defect in glucose repression (Carlson, 1997; Hengartner et al., 1995; Kuchin et al., 1995; Liao et al., 1995; Nonet & Young, 1989; Song et al., 1996). Subsequent biochemical studies confirmed their reversible association with core Mediator (Kim et al., 1994; Knuesel et al., 2009a; Koleske & Young, 1994; Liao et al., 1995; Myer & Young, 1998; Taatjes et al., 2002; Tsai et al., 2013; Wang et al., 2013).

The initial genetic association between kinase module subunits and transcriptional repression was congruent with early biochemical studies in both yeast and humans that revealed a repressive function for the kinase module. In this regard, two generally dominant and relatively stable forms of core Mediator present in nuclear extracts can be distinguished by the additional presence or absence of the kinase module (and MED26, which is preferentially enriched in mammalian core Mediator complexes) (Borggrefe et al., 2002; Liu et al., 2001; Myers et al., 1998; Spahr et al., 2003; Taatjes et al., 2002; Wang et al., 2001). Presumably, these represent parent species from which a group of highly related, yet nonetheless heterogeneous, multiprotein complexes corresponding to initially reported Mediator preparations were derived through disparate biochemical purification schemes (Akoulitchev et al., 2000; Boyer et al., 1999; Fondell et al., 1996; Gu et al., 1999; Hengartner et al., 1998; Kim et al., 1994; Kretzschmar et al., 1994; Malik et al., 2000; Myers et al., 1998; Naar et al., 1999; Rachez et al., 1999; Ryu et al., 1999; Spahr et al., 2003; Sun et al., 1998; Taatjes et al., 2002). Functional analyses of these purified Mediator complexes were generally consistent with the notion that kinase-bound Mediator inhibited, while core Mediator alone supported, reconstituted activator-dependent transcription in vitro. Consistent with these biochemical findings, molecular genetic studies confirmed an apparent repressive function for the kinase module in vivo. Thus, targeted disruption of CDK8 kinase activity in yeast resulted in genome-wide de-repression of a stress-response gene expression program, while in mammalian cells, signal-dependent gene activation on C/EBPβ- and RAR-responsive promoters was correlated with exchange of kinase-proficient with kinase-deficient Mediator (Holstege et al., 1998; Mo et al., 2004; Pavri et al., 2005). A biochemical basis to explain these observations was suggested by the finding that the kinase module precludes an association between core Mediator and the Pol II CTD, and subsequent EM-based structural analyses provided a mechanistic explanation: the yeast kinase module was observed to physically occlude an overlapping Pol II binding surface on core Mediator, while the human kinase module appeared to lock core Mediator into a conformation incompatible with Pol II interaction (Bernecky et al., 2011; Elmlund et al., 2006; Knuesel et al., 2009a). Nonetheless, these early findings were incongruent with subsequent studies that clearly implicated the kinase module in transcriptional activation (Conaway & Conaway, 2011a; Galbraith et al., 2010; Nemet et al., 2014; Poss et al., 2013).

In fact, an emergent body of biochemical and molecular genetic evidence, including large-scale genomic and proteomic, as well as improved structural, analyses support a considerably more nuanced view of the kinase module, one consistent with a role in context-dependent gene activation in addition to its established role in repression. First, CDK8 kinase activity is demonstrably required for activator-dependent transcription in both yeast and mammalian systems. Mechanistically, this requirement has been linked with biochemically discrete steps in the transcription cycle, including initiation, re-initiation and elongation, as well as phosphorylation-dependent changes in activator fate and/or function (Alarcon et al., 2009; Belakavadi & Fondell, 2010; Donner et al., 2007b, 2010; Firestein et al., 2008; Galbraith et al., 2013; Hirst et al., 1999; Kim et al., 2006; Knuesel et al., 2009b; Liu et al., 2004; Morris et al., 2008). Second, many gene-specific activators directly recruit kinase-containing Mediator in vitro, consistent with the fact that several kinase-containing Mediator complexes were initially isolated based on fractionation schemes that incorporated activation domain affinity matrices (Boyer et al., 1999; Cantin et al., 2003; Ebmeier & Taatjes, 2010; Fondell et al., 1996; Galbraith et al., 2013; Ito et al., 1999; Kim et al., 2006; Naar et al., 1999; Rachez et al., 1999; Xu et al., 2011; Zhou et al., 2006). Third, a growing list of transcriptional activators physically and functionally target Mediator through individual kinase module subunits. For example, β-catenin and the Amyloid Precursor Protein (APP) Intracellular Domain (AICD) bind directly to the MED12 interface in Mediator. In both cases, MED12 depletion significantly impairs activator-dependent transcription and Mediator recruitment onto promoter DNA (Kim et al., 2006; Xu et al., 2011). Fourth, a combination of genome-wide binding, chromosome conformation and gene expression analyses in mouse embryonic stem (ES) cells have identified MED12-containing Mediator to be a key architectural and functional constituent of enhancer-driven transcriptional programs. Thus, MED12/Mediator has been shown to co-localize with Pol II on both enhancers and promoters of active genes, and MED12 depletion leads to enhancer collapse, structural disunion of enhancer/promoter units and diminished enhancer-dependent gene expression (Dowen et al., 2014; Kagey et al., 2010; Whyte et al., 2013). Finally, the prevalent notion that Mediator with and without the kinase module harbors inherently different transcriptional activities, forged by early biochemical and genetic studies, is refutable by more recent findings that support context-dependent, as opposed to intrinsic, differences in the function and composition of Mediator. For example, recent biochemical analyses reveal that the association of core Mediator with MED26 and the kinase module (presumed determinants of active and repressive Mediator, respectively) is not mutually exclusive (Ding et al., 2009; Galbraith et al., 2013; Paoletti et al., 2006; Sato et al., 2004). Indeed, CDK8 has even been shown to be functionally required for context-dependent recruitment of core Mediator, including MED26, onto some transcriptionally active gene promoters (Galbraith et al., 2013). Presently, it is unclear how the purified and biochemically stable Mediator complexes thus far studied in vitro relate to the possible range of dynamic Mediator complexes assembled on target gene promoters in vivo. It is possible, for example, that the structurally and functionally distinct Mediator species isolated biochemically represent static extremes across a continuum of heterogeneous Mediator complexes capable of forming in vivo. Regardless, recent work has rendered it clear that the Mediator kinase module functions in both context-dependent repression and activation of transcription.

How might an activating function for the kinase module be reconciled with prior observations that it precludes Pol II binding to core Mediator, an event that presumably would inhibit an early and essential step in PIC formation? Several mutually compatible scenarios are possible. One model, supported by a growing body of evidence, invokes a temporal distinction between early Pol II promoter recruitment and a subsequent requirement for kinase module activity. For example, in yeast, CDK8 cooperates with CDK7 to promote PIC dissolution and formation of the “scaffold” complex, a functional requirement that follows Pol II recruitment (Liu et al., 2004). In mammalian cells, recent work has uncovered an increasingly prominent role for CDK8-Mediator in control of stimulus-specific gene expression programs through regulation of transcription elongation. For example, on hypoxic- and serum-response gene promoters, stimulus-induced and activator-dependent recruitment of CDK8-Mediator is followed, in turn, by CDK8-dependent delivery of P-TEFb-containing Super Elongation Complexes (SECs) that function to restart pre-loaded, but paused, Pol II (Donner et al., 2010; Galbraith et al., 2013). As CDK8 is required for recruitment of core Mediator, including MED26, onto these gene promoters, SEC recruitment may derive from direct CDK8-P-TEFb and/or MED26-EAF interactions (Donner et al., 2010; Ebmeier & Taatjes, 2010; Galbraith et al., 2013; Takahashi et al., 2011). In either case, activator-dependent recruitment of CDK8-Mediator in response to an inducing stimulus ensures that the transcriptional requirement for CDK8 kinase activity is temporally restricted to a point following Pol II promoter pre-loading.

An alternative explanation for the discordant observations that the kinase module can activate transcription on the one hand while blocking Mediator–Pol II interactions on the other may relate to the structural plasticity of Mediator itself. In this regard, early structural studies in yeast indicated that the kinase module physically occludes an overlapping Pol II binding surface on core Mediator presumed to encompass the Mediator middle module (Bernecky et al., 2011; Elmlund et al., 2006; Knuesel et al., 2009a). Nonetheless, this finding was incongruent with several empirical observations, including structural and biochemical studies, that clearly established the Mediator head, not middle, module as the principal Pol II binding interface (Lariviere et al., 2012; Robinson et al., 2012b). Inconsistencies emerging from biochemical, functional and early EM structural analyses likely derived from limitations on the quality and interpretation of EM structures as a result of conformational and compositional heterogeneity in Mediator preparations. However, recent improvements in sample preparation and image analysis have permitted EM maps of yeast Mediator to be rendered at improved accuracy, leading to a complete redefinition of its modular organization, one more congruent with prior biochemical observations (Tsai et al., 2014; Wang et al., 2014b). Thus, these redefined structures now clearly establish the head module in core Mediator as the primary interface for Pol II. Furthermore, recent studies have documented concurrent association of Pol II and the kinase module with core Mediator in vivo, and recent structural analyses revealed that Pol II and the kinase module, upon binding independently to core Mediator in vitro, both induce Mediator into an extended open conformation, one presumed to be the active functional state (Turunen et al., 2014; Wang et al., 2014b). Together, these new findings provide a biochemical and structural basis consistent with a role for the Mediator kinase module in transcriptional activation. Nonetheless, details concerning whether and how the kinase module might physically and functionally coordinate with core Mediator and Pol II during early events of the transcription process remain to be fully elucidated.

The signals and mechanism(s) that regulate reversible kinase module association with core Mediator remain to be clarified, but several studies have nonetheless begun to shed light on this issue. Early studies in yeast provided initial clues that regulable association of the kinase module with core Mediator might offer a means to control signal-dependent transcriptional output. For example, CDK8 was found to be physically depleted in yeast cells undergoing diauxic shift in response to nutrient starvation, leading to up-regulation of survival genes linked to dimorphic and stress response pathways (Holstege et al., 1998). Similarly, activation of an oxidative stress-responsive gene expression program in yeast was shown to involve ubiquitin-mediated proteolysis of CycC in a manner requiring MAPK signaling (Cooper et al., 1999; Jin et al., 2014). In mammals, Ras/MAPK and retinoic acid (RA) signals were shown to induce a switch between kinase-proficient and kinase-deficient Mediator complexes on signal-dependent target genes (Mo et al., 2004; Pavri et al., 2005). In the case of RA signaling, PARP-1 was identified as an apparent specificity factor in regulating this switch (Pavri et al., 2005). Presently, however, it remains unclear whether such Mediator switches on promoter DNA reflect reversible kinase module dissociation or, alternatively, exchange of stable kinase-proficient and kinase-deficient Mediator complexes. More recently, a mechanistic basis for dynamic kinase module dissociation has been described that involves phosphodegron-dependent and Fbw7 ubiquitin ligase-mediated proteasomal turnover of MED13/13L, which anchors the kinase module to core Mediator (Davis et al., 2013; Knuesel et al., 2009a; Tsai et al., 2013; Wang et al., 2013). Although the cellular signals that trigger phosphorylation-dependent MED13/13L turnover are currently unknown, the observation that Fbw7 nonetheless regulates kinase module association in unsynchronized cells under normal growth conditions suggests that this pathway is a general regulator of Mediator-kinase module association (Davis et al., 2013).

Kinase module structure and activity

Within the kinase module, an orchestrated network of physical and functional subunit interactions dictates its structural integrity and enzymatic activity. In this regard, recent structural and biochemical analyses have uncovered the hierarchical subunit organization within the kinase module, revealing an evolutionarily conserved molecular architecture and new insight concerning the mechanism of CDK activation. First, EM analysis of the yeast kinase module revealed a structure resembling two bent ends protruding from a central globular density; subsequent use of subunit deletion mutants and subunit labeling techniques unambiguously identified CycC–CDK8 and MED13 as the two ends surrounding a centrally organized MED12 (Figure 3A) (Tsai et al., 2013; Wang et al., 2013). Second, comparative EM analyses of human core Mediator without and with bound kinase module revealed an apparent overall structural similarity between the human and yeast kinase modules, both of which apparently connect to core Mediator principally through an interaction involving MED13 and the tip of the middle module (Tsai et al., 2013). Finally, independent binding studies using recombinant human kinase module subunits confirmed the order of their respective interactions inferred from structural analyses: MED12 occupies a central architectural role as a link between MED13 and CycC–CDK8, and between the latter, MED12, binds directly to CycC (Figure 3A) (Turunen et al., 2014). CycC thus bridges MED12 and CDK8.

Figure 3.

Mediator kinase module subunit organization. (A) Hierarchical subunit organization within the kinase module. (B) Structure of H. sapiens CycC–CDK8 (Schneider et al., 2011) (Protein Data Bank accession number 3RGF). Cyclin C, blue; CDK8, gray. Targeted residues that lie within (W177, N181, D182 and Y238) and outside (W6 and E98) the groove are rendered yellow (Turunen et al., 2014). The MED12-binding site on CycC corresponds to the CycC surface groove as indicated. CDK8 surface residues potentially available for interaction with MED12 in the ternary MED12–CycC–CDK8 complex are rendered in red. (C) Representation of kinase module paralog interactions. Black lines represent permissive interactions and red lines represent mutual exclusivity. (see color version of this figure at www.informahealthcare.com/bmg).

The network of physical interactions deduced from structural and biochemical studies have important implications for CDK8 kinase activity. Canonical CDKs are activated in a series of sequential steps, the first of which involves the binding of its partner cyclin. This binding induces a structural reorientation within the CDK of a αC helix toward its activation segment, resulting in partial CDK activation (Endicott et al., 2012; Huse & Kuriyan, 2002; Nolen et al., 2004). Complete CDK activation typically requires phosphorylation of a conserved Threonine residue within the CDK activation loop (T-loop). The T-loop is located between three conserved amino acids (DFG) on one side and three less well-conserved amino acids (APE) on the other (Huse & Kuriyan, 2002; Nolen et al., 2004). In the inactive conformation, the phenylalanine of the DFG motif is positioned toward the ATP site (DFG-out) thereby preventing proper binding of the ATP molecule. In the active conformation, the CDK undergoes a conformational change (DFG-in) thereby allowing the aspartate of the DFG motif to chelate an MG²⁺ ion in order to facilitate binding of an ATP substrate (Endicott et al., 2012). CDK8, however, is an atypical CDK in several respects, and these unique features underlie its apparent distinct mode of activation. First, CDK8 does not contain a typical DFG motif, but instead contains a methionine in place of the phenylalanine (DMG). The DMG motif in CDK8 does, however, undergo a similar conformational change of DMG-out to DMG-in upon activation (Endicott et al., 2012). Second, CDK8 is unique among CDKs in its binding interaction with CycC. For example, the cyclin-associated element, present in all CDKs (PFTAIRE or PCTAIRE), is significantly divergent in CDK8 (SMSACRE) (Xu & Ji, 2011). Notably, the CDK8 paralog CDK19 (see below) also harbors this unique cyclin-associated element, indicating that CDK19 may interact similarly with CycC (Xu & Ji, 2011). Third, CDK8 contains a unique N-terminal helix (αB), missing in other CDKs, that makes extensive interactions with CycC (Xu et al., 2014). Finally, and most notably, CDK8 activation occurs in the apparent absence of T-loop phosphorylation, invoking an alternative phosphothreonine-independent mechanism by which to achieve requisite conformational change in the ATP binding site (Knuesel et al., 2009a; Schneider et al., 2011; Turunen et al., 2014; Xu et al., 2014). In fact, this possibility is supported by the observation that MED12 is an obligate activator of CycC-dependent CDK8 (Knuesel et al., 2009b; Turunen et al., 2014).

The molecular basis by which MED12 activates CDK8 has been informed by both comparative phylogenetic analyses and experimental biochemistry. First, comparative analyses of yeast and human primary sequence and tertiary structures revealed that CycC orthologs are among the most highly conserved within the cyclin family, consisting of two typical 5-helical cyclin repeats, as well as CycC-specific features, including a negatively charged surface groove removed from its CDK8-binding surface (Hoeppner et al., 2005; Schneider et al., 2011). Recent biochemical studies identified this CycC surface groove as a direct docking site for MED12 (Figure 3B) (Turunen et al., 2014). Reciprocally, and significantly, the CycC-binding interface on MED12 was mapped to within its N-terminal 100 amino acids, encoded largely by exons 1 and 2 that carry frequent hotspot mutations linked to multiple human tumor types (Kampjarvi et al., 2012, 2014, 2015; Lim et al., 2014; Makinen et al., 2011b; Turunen et al., 2014; Yoshida et al., 2015). Notably, mutagenic disruption of the MED12–CycC interface, including reciprocal mutations on either face, uncouples CycC–CDK8 from core Mediator and severely impairs CycC-dependent CDK8 kinase activity (Turunen et al., 2014). These findings thus identify the CycC surface groove as a principal binding interface through which MED12 both anchors and activates CycC–CDK8 in Mediator, and further impute the MED12–CycC interface as a novel therapeutic target in CDK8-driven cancers.

Kinase module paralogs

Considerations concerning kinase module subunit interaction dynamics, both physical and functional, need account for the fact that vertebrate paralogs of CDK8, MED12 and MED13 have been identified, designated CDK19, MED12L and MED13L, respectively. CDK19, (aka CDK11, CDK8L and CDC2L6) (Malumbres et al., 2009; Manning et al., 2002; Sato et al., 2004; Tsutsui et al., 2008) was discovered through the use of a Markov model profile of eukaryotic protein kinase domains and was found to bear 91% amino acid similarity to CDK8; the two proteins share a particularly high degree of sequence conservation in their respective kinase and cyclin-binding domains, but exhibit considerably more divergence in their corresponding C-termini (Manning et al., 2002; Tsutsui et al., 2008), The gene encoding MED12L, originally designated NOPAR, was discovered as a novel gene by positional cloning during a search for genes implicated in the developmental disorder Usher syndrome type 3 (USHS) (Joensuu et al., 2001). Although it was excluded as USH3 on the basis of mutational analysis, NOPAR was nonetheless found to encode a protein with 61% amino acid sequence homology to the amino terminus of MED12, but otherwise lacking a MED12-like OPA domain (Joensuu et al., 2001). Subsequently, NOPAR was re-designated as MED12L, as it was found to share 67% overall amino acid sequence similarity to MED12 and include MED12-like PQL and OPA domains (Vogl et al., 2013). The gene encoding MED13L (originally called PROSIT240, aka THRAP2) was identified by positional cloning as a candidate gene involved in the pathogenesis of the cyanotic heart defect transposition of the great arteries (TGA) (Muncke et al., 2003). Upon further characterization, it was discovered that this novel gene shared 51% amino acid sequence similarity to MED13 and has since been designated as MED13L (Muncke et al., 2003).

Recent mass spectrometric-based proteomic analyses indicate that paralogous kinase module subunits assemble into Mediator in a mutually exclusive manner, raising the possibility that multiple, compositionally distinct and functionally divergent Mediator-kinase complexes may exist in mammalian cells (Figure 3C) (Daniels et al., 2013). Notably, independent genetic disruption of paralogous subunits leads to early embryonic lethality in mice, while their genetic mutation or variation in model metazoans and humans engenders clear developmental phenotypes (Li et al., 2014b; Rocha et al., 2010; Spaeth et al., 2011; Westerling et al., 2007). These considerations exclude the possibility of complete functional redundancy among kinase module paralogs during development.

While the relative functional contribution of individual kinase module paralogs remains to be fully elucidated, emerging studies suggest that paralogous subunits likely confer distinct transcriptional regulatory functions on Mediator. For example, biochemical studies have revealed that CDK8-Mediator supports, while CDK19-Mediator suppresses, herpes simplex virus VP16 transactivation domain function (Tsutsui et al., 2008). Furthermore, comparative genome-wide transcriptome profiling revealed gene sets uniquely regulated by CDK8 or CDK19, further highlighting their independent roles in transcription (Galbraith et al., 2013; Tsutsui et al., 2011). Notably, CDK8 and CDK19 exhibit differential tissue-specific expression; CDK8 is ubiquitously expressed, whereas CDK19 expression is more narrowly restricted to prostate, salivary gland, testis and thymus (Tsutsui et al., 2011). Nonetheless, their high degree of sequence homology also suggests possible overlapping functional roles for CDK8 and CDK19. In this regard, both kinases have been shown to bind to PRMT5, an arginine methyltransferase, and its co-factor WDR77 (Tsutsui et al., 2013). This interaction facilitates recruitment of PRMT5 to C/EBPβ target gene promoters, resulting in histone H4 arginine 3 dimethylation (H4R3me2) and consequent repression of C/EBPβ target gene expression (Tsutsui et al., 2013). In addition, both kinases can phosphorylate the Pol II CTD, adding further support for possible overlapping roles in transcriptional regulation (Tsutsui et al., 2011). Although considerably less is known regarding the relative functional contributions of MED12/MED12L, at least one study has suggested possible functional overlap. In this regard, both proteins were shown to functionally interact with Sox10 to promote differentiation of myelinating glia (Vogl et al., 2013). Clearly, additional studies will be required to clarify the functional relationships among the various kinase module paralogs.

Biological roles of the kinase module

The Mediator kinase module has been functionally implicated and mechanistically defined in a host of physiological processes spanning development and differentiation to the maintenance of cell fate and function (Figure 4). Attempts to separate these natural systems, many of which are highly integrated in organismal space and time, necessarily entail arbitrary distinctions. Nonetheless, for purely descriptive purposes, we review the role of the Mediator kinase module in the following biological processes, cognizant of their overlapping and often highly interrelated natures.

Figure 4.

Biological functions of the Mediator kinase module. Wheel diagram depicting biological functions in which the Mediator kinase module subunits are implicated. (see colour version of this figure at www.informahealthcare.com/bmg).

Developmental signaling

Genetic studies in model metazoans have revealed critical roles for kinase module subunits in signal-dependent developmental gene regulation. In Caenorhabditis elegans (C. elegans), MED12/MED13 suppress Wnt- and Ras-target genes involved in vulval cell fate specification and asymmetric cell division, while in Drosophila melanogaster (Drosophila), Med12/Med13 suppress Wnt- and Hedgehog-target genes critical for differentiation in the eye and antennal discs (Janody et al., 2003; Moghal & Sternberg, 2003; Treisman, 2001; Yoda et al., 2005). Furthermore, in the developing Drosophila wing disc, all four kinase module subunits are required to activate one subclass of Notch-target genes, while only Med12/Med13 are required to regulate, both positively and negatively, a distinct subclass (Janody & Treisman, 2011). This lineage-specific segregation of function appears to also be conserved in the developing Drosophila hematopoietic system, wherein all four kinase module subunits control the emergence and proliferation of crystal cell progenitors, but CDK8/CycC are uniquely dispensable for crystal cell differentiation (Gobert et al., 2010). Finally, in C. elegans, a genetic interaction screen identified med12 as a highly connected “hub” gene linked to multiple developmental signaling pathways, including the EGF/Ras, Notch and Wnt pathways (Lehner et al., 2006).

In zebrafish, Med12/Med13 appear to repress Wnt-target genes during embryogenesis (Lin et al., 2007), and Med12 has been shown to be broadly required for proper development of multiple tissues, organ systems and structures, including the brain and neural crest through co-activation of Sox9, the epithalamus through maintenance of proper FGF signaling, the endoderm (at least in part) through regulation of proper nodal signaling and the hindbrain and hematopoietic system (myeloid lineage) (Hong & Dawid, 2011; Keightley et al., 2011; Rau et al., 2006; Shin et al., 2008; Wang et al., 2006; Wu et al., 2014).

In mice, kinase module subunits are critical for early development, as genetic disruption of CDK8, CycC or MED12 leads to early embryonic lethality. Inactivation of CDK8 through gene trap insertion leads to failure of embryo implantation, indicating that CDK8 is required for preimplantation at E2.5–3 (Westerling et al., 2007). CycC knockout embryos die in utero at E10.5 and exhibit severe developmental retardation and an underdeveloped placental labyrinth layer (Li et al., 2014b). Hypomorphic MED12 mutant embryos (>90% reduction in MED12 protein levels) die by E10, with acute defects in neural tube closure, axis elongation, somitogenesis and heart formation (Rocha et al., 2010). Complete ablation of MED12 through disruption of the first 7 exons leads to full developmental arrest at E7.5 (Rocha et al., 2010).

The underlying basis for these severe developmental phenotypes almost certainly derives from defects in key developmental signaling pathways. For example, hypomorphic MED12 mutant mouse embryos exhibited impaired expression of Wnt/β-catenin target genes and MED12 null embryos recapitulated phenotypes observed in the absence of β-catenin, including failures to establish anterior visceral endoderm, activate Brachyury expression and complete gastrulation, with a severe defect in mesoderm formation (Rocha et al., 2010). Together, these observations indicate that MED12 is essential for canonical Wnt/β-catenin and Wnt/planar cell polarity signaling during embryogenesis. These genetic findings are fully congruent with earlier biochemical and cell biological studies that identified human MED12/Mediator as a physical target and functional transducer of Wnt/β-catenin signaling (Kim et al., 2006). In addition to MED12, CDK8 has also been implicated in Wnt/β-catenin signaling as loss of CDK8 disrupts expression of Wnt target genes (Firestein et al., 2008).

Recently, conditional ablation of CycC in living mice has uncovered a role for CycC-dependent CDK8/19 kinase activity in the negative regulation of Notch1 signaling, a pre-eminent developmental pathway that controls self-renewal and differentiation in multiple cell and tissue types (Li et al., 2014b). Notch1 is a binary cell-fate determinant, one whose regulatory control is unleashed by regulated intramembrane proteolysis to produce a transcriptionally active Notch intracellular domain (ICN1) that traffics to the nucleus and regulates gene expression programs involved in proliferation, differentiation and apoptosis (Yamamoto et al., 2014). Unexpectedly, targeted ablation of CycC in the hematopoietic lineage of mice was accompanied by a dramatic increase in the level of the ICN1 in both thymocytes and bone marrow, the latter of which exhibited precocious differentiation along the T cell lineage (Li et al., 2014b). The marked elevation in ICN1 levels accompanying CycC disruption was attributed mechanistically to increased ICN1 protein stability, and subsequent analyses confirmed CycC, in association with CDK8/19, to be a major determinant of ICN1 levels through its ability to phosphorylate and promote ICN1 degradation via the ubiquitin-proteasome pathway (Li et al., 2014b). Thus, Cyc-CDK8/19 is a physiological suppressor of ICN1-dependent gene regulation during development. This finding is concordant with prior biochemical and cell biological studies showing that CycC–CDK8 phosphorylates and thereby primes the ICN1 for proteolytic turnover on active promoters (Fryer et al., 2004). However, whether CycC–CDK8-instigated ICN1 degradation in this context is truly repressive for ICN-dependent gene regulation or, alternatively, a means to ensure a sustained oscillatory response through clearance of promoter-bound enhancer complexes at each round of transcription remains to be established. Nonetheless, the notion that CycC–CDK8/19 primes the ICN1 for proteolytic destruction is consistent with, and further supports, a more general role for CycC–CDK8/19 in the control of inducible activators through signal-dependent phosphorylation (Chi et al., 2001; Hirst et al., 1999; Nelson et al., 2003).

CycC–CDK8 have also been linked to TGFβ signaling, another prominent developmental pathway that controls a plethora of biological processes, including cell fate specification, differentiation, proliferation and apoptosis (Shi & Massague, 2003). The TGFβ family of growth factors, including TGFβ and BMP among others, initiate signaling by binding to cell surface receptor kinases that phosphorylate and activate receptor regulated SMADS (R-SMADS) that function as downstream transcriptional effectors in the nucleus (Shi & Massague, 2003). Recent work has uncovered a critical role for CycC–CDK8 in the instigation of a phosphorylation-dependent SMAD action turnover switch that regulates specific developmental responses to defined TGFβ signals (Alarcon et al., 2009; Aragon et al., 2011; Gao et al., 2009). In this regard, BMP-activated SMAD1 and TGFβ-activated Smad2/3, when delivered to the nucleus, were found to undergo targeted serine-specific phosphorylation by CDK8 (and CDK9). These phosphoserines serve as ligands for SMAD-binding partners whose temporally ordered associations serve to couple transcriptional activation by SMADS with their regulated destruction. Specifically, CDK8/9-specified SMAD phosphoserines bind Smad-specific co-activators (YAP and PIN1 for SMAD1 and SMAD2/3, respectively) as well as GSK3 kinase. Once bound, GSK3-dependent SMAD phosphorylation displaces co-activators and promotes binding of SMAD-specific ubiquitin ligases (SMURF1 and NEDD4L for SMAD1 and SMAD2/3, respectively) that target SMADS for proteolytic destruction. In this way, CDK8 fulfills a critical role in the initiation of a phosphoserine code on SMADS whose recognition by biologically active effectors provides an effective means to modulate the amplitude and duration of TGFβ-driven and SMAD-dependent transcriptional responses. Biologically, this SMAD action turnover switch effectively limits SMAD2/3-dpendent induction of mesodermal fates in response to TGFβ signaling (Gao et al., 2009).

The biological significance of the Mediator kinase module in human development, suggested by genetic studies in lower metazoans, is illustrated by a growing number of developmental disorders)/multiple congenital anomaly disorders, including FG, Lujan and Ohdo (MKB type) syndromes (Risheg et al., 2007; Schwartz et al., 2007; Spaeth et al., 2011; Vulto-van Silfhout et al., 2013). Originally considered distinct entities on the basis of clinical criteria alone, these three disorders are instead allelic, being linked by mutations in MED12. Accordingly, these disorders share broad overlapping clinical manifestations, including intellectual disability and some craniofacial dysmorphisms (Fryns & Buttiens, 1987; Lujan et al., 1984; Opitz et al., 2008; Vulto-van Silfhout et al., 2013). FG and Lujan syndromes further share a more restricted set of clinical phenotypes, including macrocephaly, corpus callosal defects, hypertolerism and syndactyly (Fryns & Buttiens, 1987; Lujan et al., 1984; Opitz et al., 2008). Beyond this relatively narrow set of overlapping features, however, each disorder is also characterized by its own unique and heterogeneous spectrum of clinical phenotypes that variously include musculoskeletal, behavioral, gastrointestinal and nervous system defects. The etiological basis for such broad phenotypic variation arising from mutations in a common gene is likely explained by the unique character of MED12, clearly elaborated in model metazoans, as a regulatory hub that functions to integrate and transduce signals from diverse developmental pathways to specify unique programs of Poll-dependent gene transcription. In this regard, several recent studies have begun to identify key developmental pathways dysregulated as a consequence of these pathogenic mutations in MED12, suggesting possible bases to explain cognitive and perhaps other phenotypes associated with these syndromal disorders.

Mutations linked with all three MED12-related disorders have been shown to disrupt epigenetic repression of neuronal gene expression imposed by the RE1 silencing transcription factor/neural restrictive silencer factor (REST/NRSF), a master regulator of neuronal fate (Ding et al., 2008, 2009; Vulto-van Silfhout et al., 2013). In this regard, the MED12 interface in Mediator was found to link chromatin-bound REST with G9a histone methyltransferase to silence REST-targeted neuronal genes through the imposition of transcriptionally repressive histone H3K9 di-methylation. Notably, pathogenic mutations in MED12 linked to FG (R961W), Lujan (N1007S) and Ohdo (R1148H; S1165P) syndromes were shown to disrupt its REST-specific co-repressor function, leading to unscheduled de-repression of REST target genes. As REST and MED12 are both implicated in neuronal development, misregulation of REST target genes arising as a consequence of pathogenic mutations in MED12 could affect neuronal differentiation and possibly contribute to XLID.

More recent work has revealed that the FG/R961W and Lujan/N1007S mutations in MED12 disrupt a Mediator imposed constraint on GLI3-dependent SHH signaling (Zhou et al., 2006, 2012). Mechanistically, the FG and Lujan mutations in MED12 were shown to impair its gene-specific recruitment into Mediator of CDK8, an enzymatically active suppressor of GLI3 transactivation activity. Furthermore, in FG and Lujan patient-derived cells, enhanced SHH pathway activation and GLI3-target gene induction was found to be coincident with impaired recruitment of CDK8 onto promoters of GLI3-target genes, but not non-GLI3-target genes. Notably, many of the digit, craniofacial, corpus callosal and anorectal malformations that typify FG and/or Lujan syndromes are similarly observed, to varying extents, in congenital anomaly syndromes arising from mutations in GLI3 (Johnston et al., 2010). Together, these findings suggest that dysregulated GLI3-dependent SHH signaling contributes to phenotypes of individuals with FG and Lujan syndromes and further reveal a basis for the gene-specific manifestation of pathogenic mutations in a global transcriptional co-regulator (Berk, 2012).

In addition to MED12, several other kinase module subunits have been linked to developmental disorders in humans. Haploinsufficiency of CDK19 has been implicated in a clinical syndrome characterized by congenital retinal folds, microcephaly and mental retardation (Mukhopadhyay et al., 2010). Both allelic disruption and missense mutations in MED13L have been attributed to transposition of the great arteries (TGA), a condition characterized by ventriculoarterial discordance, and more recently, to MED13L syndrome, characterized by intellectual disability and developmental delay as well as multiple congenital anomalies, including craniofacial, musculoskeletal and behavioral defects (Adegbola et al., 2015; Asadollahi et al., 2013; Muncke et al., 2003). In addition, syndromic MED13L haploinsufficiency has been linked to intellectual disability and congenital anomalies ranging from musculoskeletal and heart defects to craniofacial dysmorphisms (Asadollahi et al., 2013; Cafiero et al., 2015; Utami et al., 2014; van Haelst et al., 2015). Finally, MED13L was recently reported to be a candidate causative gene for idiopathic autism spectrum disorder (Codina-Sola et al., 2015). The molecular bases by which these various genetic alterations lead to specific pathological defects are presently unclear. Nonetheless, the established links between subunit dysfunction and human pathology further support a critical role for the Mediator kinase module in developmental signaling.

Cell cycle regulation

The eukaryotic cell division cycle is precisely orchestrated by the oscillatory activity of CDKs in conjunction with their partner cyclins. Cyclin-CDKs drive major cell cycle transitions (G1/S and G2/M) through phosphorylation-mediated changes in the stability or activity of key substrates and thus promote cell cycle progression (Hunt et al., 2011; Nurse et al., 1998). Notably, periodic transcription is closely linked with cell cycle progression, and there is now significant evidence to implicate CDK8, a transcriptional CDK, in control of both G1/S and G2/M cell cycle transitions.

In budding yeast, a wave of periodic transcription in early G1 produces cell cycle regulators, including G1 cyclins, contribute to “START”, a point of irreversible commitment to cell cycle entry (Hartwell, 2002; Johnson & Skotheim, 2013; Spellman et al., 1998). Among the established transcriptional activators of these early cell cycle regulators, Swi5p recruits Mediator for high-level gene activation (Bhoite et al., 2001; Cosma et al., 1999, 2001). Somewhat paradoxically, temporally coordinated proteolytic turnover of promoter-bound Swi5p is necessary for continued high-level gene induction and S phase entry (Tebb et al., 1993). Notably, Mediator-bound CDK8p has been shown to phosphorylate and thereby mark Swi5p for subsequent SCF^Cdc4 ubiquitin ligase-mediated proteasomal degradation (Kishi et al., 2008). In this manner, CDK8p thus fulfills an important function in coordinating periodic transcription control with orderly transit through G1/S.

In mammalian cells, the restriction point, much like “START” in yeast, is a G1 phase cell cycle checkpoint beyond which the cell is irreversibly committed to cell cycle progression absent extracellular growth signals (Johnson & Skotheim, 2013). When engaged by these stimuli, an elaborate network of signaling pathways converges at this checkpoint to control the formation of active Cyclin D-CDK (4 and 6) heterodimers that promote an E2F-mediated transcriptional program dedicated to DNA replication initiation and cell cycle commitment (Hunt et al., 2011; Nurse et al., 1998). As a major conduit of cellular signaling through Mediator, the kinase module is uniquely positioned to impact these pathways, and substantive evidence now suggest that CDK8 indeed contributes to regulated transit through G1/S.

A key mediator of p53-dependent cell cycle arrest in response to cellular stress is p21, which inhibits CDK1/2, and thus, regulates G1 progression through the cell cycle (Reinhardt & Schumacher, 2012). Notably, p21 is transcriptionally regulated by p53 in a stimulus-specific manner. For example, p53 only weakly activates p21 in response to UV-induced DNA damage, while ionizing radiation and certain chemotherapeutic agents [including 5-fluorouricil (5-FU) and doxorubicin] induce strong p53-dependent p21 activation (Donner et al., 2007b). Recently, the mechanistic basis for this differential p53-dependent regulation has been clarified (Donner et al., 2007a,b). A series of biochemical and molecular cell biological studies revealed that the differential transcriptional response of p21- to p53-induced induced by diverse stimuli was strongly correlated with the assembly of stimulus-specific transcriptional complexes, characterized by the presence of CDK8-Mediator, on the p21 promoter (Donner et al., 2007a,b). Furthermore, functional studies revealed that p21 transcription is stimulated and inhibited, respectively, by over- and under-expression of CDK8. Together, these findings identify CDK8 as a stimulus-specific co-activator of p53-dependent target genes, including p21 (Donner et al., 2007a,b; Galbraith et al., 2010). Accordingly, CDK8 contributes significantly to a p53-dependent stress-induced checkpoint that can override the restriction point and control G1/S transit. Reciprocally, recent studies have revealed that p21 stimulates CDK8 kinase activity through a direct interaction, suggesting that a positive feedback loop can be formed between p21 and CDK8 in cell cycle regulation (Porter et al., 2012).

CDK8, as well as CycC and MED12, also indirectly control mammalian G1/S through regulation of canonical Wnt/β-catenin signaling, which promotes G1/S phase progression through direct induction of key mitogenic genes including MYC and CCND1 (Niehrs & Acebron, 2012; Talluri & Dick, 2012). Canonical Wnt/β-catenin signaling conforms to a classic two-state model for signal activation. In the absence of secreted Wnt signals, cytoplasmic β-catenin is primed for proteasomal degradation through phosphorylation by GSK3-β within a destruction complex that also includes the adenomatous polyposis coli (APC) tumor suppressor and the scaffold protein AXIN (MacDonald et al., 2009). Binding of Wnt ligand to Frizzled-LRP5/6 co-receptor complexes leads to inactivation of the destruction complex and stabilization of β-catenin. Stabilized β-catenin then translocates into the nucleus, whereupon its interaction with members of the TCF/LEF family of DNA-binding proteins positions it to activate transcription from the promoters of Wnt target genes that function in cell proliferation and fate, as well as cell–cell and cell–matrix interactions (MacDonald et al., 2009). The Mediator kinase module was originally linked to Wnt/β-catenin signaling based on the observation that the C-terminal transactivation domain in β-catenin physically and functionally targets the MED12 interface in Mediator to activate transcription (Kim et al., 2006). This observation has important mechanistic implications for β-catenin-dependent gene activation since MED12 activates CycC-dependent CDK8 kinase. Taken together, these findings suggested the involvement of enzymatically active CDK8 in β-catenin-driven gene regulation, a possibility confirmed by subsequent work notable for its revelation of a direct link between the Mediator kinase module and colorectal cancer (Firestein et al., 2008). Taken together, these findings reveal a network of physical and functional interactions within the Mediator kinase module critical for mitogenic Wnt signal transduction.

In addition to its role as a direct transducer of Wnt/β-catenin signaling, CDK8 also stimulates β-catenin activity through suppression of E2F1, a negative regulator of β-catenin signaling. In this regard, an antagonistic relationship between β-catenin and E2F1 was initially uncovered during the course of a genetic screen in Drosophila designed to identify regulators of E2F1-induced apoptosis (Morris et al., 2008). Importantly, E2F1-induced apoptosis was found to be dependent upon its ability to inhibit β-catenin-dependent gene activation. In a consequent genetic screen for upstream regulators of E2F1, CDK8 was identified as a potent E2F1 suppressor, and biochemical analyses revealed that CDK8 kinase activity was required to suppress the inhibitory effect of E2F1 on β-catenin-dependent transcription. Together, these studies identify two independent regulatory roles for CDK8 in the promotion of Wnt/β-catenin signaling – one as a co-activator of β-catenin-dependent gene activation and the other as a direct suppressor of E2F1 (Figure 5). Importantly, and as discussed subsequently, both of these activities have important implications for CDK8 in human tumorigenesis.

Figure 5.

CDK8 regulates oncogenic Wnt/β-catenin signaling through two different mechanisms. 13q12-encoded CDK8, overexpressed through amplification-dependent means in a colorectal cancer subset, controls Wnt/β-catenin signaling directly as a β-catenin co-activator and indirectly as a suppressor of the β-catenin inhibitor E2F1. Both of these CDK8 regulatory functions require its kinase activity and both promote oncogenic Wnt/β-catenin signaling (Firestein et al., 2008; Morris et al., 2008). (see colour version of this figure at www.informahealthcare.com/bmg).

In addition to its role in regulating cell cycle transit through G1/S, recent studies support an additional role for CDK8 in control of the G2/M transition. In fission yeast, CDK8 has been shown to regulate mitotic commitment through periodic transcriptional control of key mitotic genes. Among ~500 genes in fission yeast identified to be transcribed in a cell-cycle dependent manner, a fraction are specifically activated during mitosis and encode activities required for cell separation, subsequent G1/S transition and eventual replication (Marguerat et al., 2006; Oliva et al., 2005; Peng et al., 2005; Rustici et al., 2004). Active repression of mitotic genes during other cell cycle phases in enforced by the forkhead protein Fkh2, whose activity is regulated by protein turnover (Buck et al., 2004; Rustici et al., 2004; Whitfield et al., 2002). In this regard, precisely timed destruction of Fkh2 at the onset of mitosis is essential to ensure the coordinated and timely activation of mitotic genes in concert with mitotic-specific transactivators. Notably, CDK8 was recently shown to function as a key determinant of Fkh2 stability; CDK8 phosphorylates Fkh2, thereby preventing its degradation prior to the onset of mitosis (Szilagyi et al., 2012). Accordingly, forced expression of a dominant-negative (kinase-dead) CDK8 derivative delayed mitotic entry in fission yeast, while a phosphomimetic Fkh2 mutant conversely accelerated this process. Together, these findings establish CDK8 as a key regulator of mitotic entry through regulation of Fkh2 cell cycle oscillatory behavior. More recent studies implicate Med12/Med13 as negative regulators of CDK8-dependent Fkh2 phosphorylation and the timing of mitotic commitment. While the precise details of this regulation remain to be clarified, physical sequestration by, and temporally regulated release from, Med12/Med13 in Mediator has been imputed as a basis for control of CDK8 in this setting (Banyai et al., 2014).

More recent work supports a functionally conserved, if mechanistically distinct, role for CDK8 in control of the mammalian G2/M phase transition through its participation in a regulatory Skp2–MacroH2A1–CDK8 axis (Figure 6) (Xu et al., 2015). Skp2, the substrate-binding component of a SCF ubiquitin ligase complex, regulates G2/M progression, at least in part, by mediating degradation of the Cdc2 inhibitor p27 (Nakayama et al., 2004; Xu et al., 2015). A biochemical screen for novel Skp2 substrates that might also contribute to G2/M control identified MacroH2A1 (mH2A1), a histone variant typically associated with heterochromatin and transcriptional repression (Xu et al., 2015). Subsequent biochemical and functional studies confirmed mH2A1 as a critical substrate of Skp2 in G2/M control (Xu et al., 2015). Thus, Skp2 was shown to directly target mH2A1 for ubiquitin-mediated degradation, and mH2A1 overexpression in Skp2-proficient (WT) MEFs could recapitulate the G2/M arrest phenotype observed in their Skp2-deficient counterparts. Furthermore, depletion of mH2A1 rescued the G2/M arrest characteristic of Skp2-deficient MEFs. Together, these functional relationships establish Skp2-mediated modulation of mH2A1 protein levels as a mechanism to control G2/M phase progression. Notably, CDK8 was subsequently identified as a target of mH2A1 repression, and forced expression of CDK8 could partially rescue the G2/M arrest phenotype observed in Skp2-deficient MEFs (Xu et al., 2015). These findings indicate that Skp2, through suppression of mH2A1, promotes CDK8 expression to control G2/M transit. Mechanistically, CDK8 was shown to phosphorylate p27, which primes the latter for Skp2-dependent degradation. Thus, Skp2 and CDK8, both components of a common signaling axis, cooperate in the turnover of p27, yet another component of the same axis, thereby providing a positive feedback loop to amplify their regulatory influence over p27 and G2/M phase progression.

Figure 6.

A conserved MacroH2A-CDK8 core regulatory axis controls melanoma and breast cancer progression. Left: During melanoma progression, mH2A expression levels decrease, leading to chromatin decondensation and upregulation of CDK8, which drives tumorigenesis as part of the Mediator kinase module (Kapoor et al., 2010). Right: In breast cancer, Skp2, through suppression of mH2A1, promotes CDK8 expression to control G2/M phase progression, ploidy and tumorigenesis through modulation of p27 levels. Mechanistically, CDK8 phosphorylates p27, which primes the latter for Skp2-dependent degradation. Thus, Skp2 and CDK8, both components of a common signaling axis, cooperate in the turnover of p27, yet another component of the same axis, thereby providing a positive feedback loop to amplify their regulatory influence over p27 and breast cancer progression (Xu et al., 2015). (see colour version of this figure at www.informahealthcare.com/bmg).

Innate immunity

The interferon (IFN) mediated innate immune response represents an ancient and robust first-line defense system against invading pathogens. IFN molecules produced in response to pathogenic stimuli bind to cell surface receptors that, in turn, signal to the nucleus via the Janus kinase signal transducer and activator of transcription (JAK-STAT) pathway (Levy & Darnell, 2002). IFN-stimulated JAK-STAT signaling drives programmed changes in the expression of hundreds of interferon-stimulated genes (ISGs) that shape the anti-viral response, effective against viruses, bacteria and parasites, as well as the adaptive immune response (Stark & Darnell, 2012; Schoenborn & Wilson, 2007). IFN–receptor interactions at the cell membrane activate JAKs bound to cytoplasmic receptor chains, resulting in receptor phosphorylation and recruitment of STATs, transcriptional effectors that dimerize with one another in response to JAK-mediated phosphorylation on conserved tyrosines. STAT dimerization promotes their nuclear translocation, whereupon STATs bind to gamma activated sequence (GAS) elements ISGs. In addition to JAK-dependent tyrosine phosphorylation in the cytosol, STAT family members also undergo nuclear phosphorylation at a conserved serine residue (S727 in STATs 1 and 3) within their C-terminal transactivation domain (Shah et al., 2012), a modification termed “canonical” phosphorylation that is required for high level STAT transactivation activity (Staab et al., 2013; Wen et al., 1995). While the requirement for canonical STAT phosphorylation has been established for decades, the identity of the responsible nuclear kinase has remained elusive. In this regard, recent work revealed that canonical phosphorylation occurs solely on promoter-bound STATs, imputing a component of the STAT-assembled transcriptional complex as the relevant kinase (Sadzak et al., 2008; Yang et al., 2010). Notably, Mediator-associated CDK8 was most recently identified as the likely STAT canonical kinase (Bancerek et al., 2013). Mechanistic studies revealed that CDK8 phosphorylates STAT1 S727 in response to IFNγ. Furthermore, genome-wide transcriptome profiling revealed that nearly half of all IFNγ-responsive genes functionally require CDK8-dependent STAT1 phosphorylation. Mechanistically, chromatin immunoprecipitation analysis revealed that CDK8-mediated STAT1 phosphorylation correlated well with Pol II promoter occupancy, suggesting a role for STAT1 S727 phosphorylation, and thus CDK8, in transcription PIC assembly on IFN-regulated genes. Finally, disruption of CDK8 kinase activity through depletion of its partner CycC increased cell sensitivity to vesicular stomatitis virus (VSV). Collectively, these studies identify CycC–CDK8 as a key regulator of STAT1 in the IFNγ-induced antiviral response, and further invoke CDK8 as a possible therapeutic target in the modulation of cytokine responses (Bancerek et al., 2013).

Oxygen homeostasis

Respiratory oxygen (O₂) is essential for multicellular life, and metazoans have therefore evolved elaborate regulatory systems capable of functioning on both cellular and physiological systems levels to ensure proper O₂ homeostasis (Semenza, 2014). Central to integrity of these systems is hypoxia inducible factor (HIF), a master regulator of O₂ homeostasis. HIFs correspond to a family of heterodimeric transcription factors composed of a single constitutively expressed HIF1β subunit and one of three O₂-regulated HIFα subunits (Jiang et al., 1996; Loenarz et al., 2011). Under normoxic conditions, HIFα is hydroxlated by O₂-dependent prolyl hydroxylases (PHDs), marking it for ubiquitin-mediated proteasomal degradation (Kaelin & Ratcliffe, 2008). Upon hypoxia, PHDs are inactivated, and stabilized HIF1α heterodimerizes with HIF1β in the nucleus, where HIF binds to genomic hypoxia response elements (HRE) and directs a gene expression program that functions to reduce cellular O₂ demand and increase O₂ delivery and tissue re-oxygenation (Loenarz et al., 2011; Semenza, 2014). Due to its unique and central role in the integration and execution of the hypoxic response, precise regulation of HIF-driven transcription is required for maintenance of cellular metabolic and physiologic systems function. In this regard, recent work has shown that Mediator-associated CDK8, but not CDK19, is a critical co-activator of HIF1α-driven gene transcription in response to hypoxia (Galbraith et al., 2013). Comparative genome-wide transcriptome profiling revealed that CDK8 depletion significantly blunts the hypoxic response, thus revealing a critical role for CDK8 in this process. Furthermore, ChIP-seq showed enrichment of CDK8 on ~65% of genes induced by hypoxia, confirming its broad role in the hypoxic response. Mechanistic studies based on ChIP from isogenic HIF1α-proficient and HIF1α-deficient cells revealed that on hypoxic response genes, HIF1α recruits both the histone acetyltransferase p300/CBP and CDK8-Mediator, but not Pol II, which was observed instead by GRO-seq to be transcriptionally engaged, but paused, under normoxic conditions. Notably, HIF1α-mediated recruitment of p300/CBP in response to hypoxia was insufficient to activate paused Pol II in the absence of CDK8-Mediator, implying a co-regulatory function for CDK8-Mediator after chromatin reconfiguration, PIC complex formation and Pol II initiation and promoter escape. This was subsequently confirmed by the finding that CDK8-Mediator recruits the super elongation complex (SEC) to release paused Pol II and thus promote hypoxic-responsive transcription. Collectively, these findings reveal a novel mechanistic basis to explain the physical and functional relationship between CDK8 and HIF1α in the transcriptional response to hypoxia, a critical step in the maintenance of O₂ homeostasis (Galbraith et al., 2013).

Growth factor signaling

The mechanistic link between CDK8 and transcriptional elongation, apparent in the hypoxic response, may represent a more commonly conserved theme among inducible transcription systems. Indeed, such a link has also been observed in the cellular response to serum growth factors. Within minutes of growth factor stimulation, quiescent mammalian cells initiate a precisely coordinated response that involves a cascade of protein phosphorylation events along the Ras-MAPK and other axes that converges in the nucleus on a network of DNA-binding and physically interacting transcription factors, collectively termed the “serum response network” (Galbraith & Espinosa, 2011). This network includes the eponymous serum response factor (SRF), a member of the MADS box superfamily, as well as ternary complex factors (TCFs) including Elk1, Ets1, Runx and others (Galbraith & Espinosa, 2011). Stimulus-activated serum responders within this network bind to serum response elements (SREs) within immediate early genes (IEGs), including FOS, that function as further effectors of the serum response, generating a second wave of transcription that culminates in the expression of genes that program cell cycle progression, cell growth, apoptosis and differentiation (Donner et al., 2010; Galbraith & Espinosa, 2011). The IEGs have become a paradigm for regulation of Pol II at post-recruitment steps since a distinguishing feature of these genes is the present of paused Pol II. Like other genes subject to rapid and robust induction, regulated Pol II pause release provides a highly efficient means to achieve this end (Core & Lis, 2008; Core et al., 2008; Guenther et al., 2007; Rahl et al., 2010). Recent work has uncovered a critical role for CDK8 as a positive regulator of Pol II pause release in the serum response network (Donner et al., 2010; Galbraith & Espinosa, 2011). In this regard, comparative microarray analyses of WT and CDK8-depleted human cells identified IEGs encoding AP-1 (JUN, FOS and JUNB) and EGR (EGR1, EGR2 and EGR3) to be significantly reduced in CDK8-depleted versus WT cells upon serum stimulation, revealing that CDK8 functions as a co-activator of these genes (Donner et al., 2010). ChIP analyses revealed that CDK8 is not required for Pol II recruitment, but rather for transcriptional elongation. Increased Pol II occupancy and phosphorylation of the Pol II CTD within the intragenic region of FOS and EGR1, along with a concordant increase in FOS and EGR1 pre-mRNA transcripts, were observed upon serum stimulation. Importantly, this effect on Pol II and FOS/EGR1 transcription was diminished upon CDK8 depletion (Donner et al., 2010). Moreover, recruitment of CDK7, CDK9 and BRD4 to FOS and EGR1 loci was impaired upon CDK8 knockdown. This finding is significant because CDK7, CDK9 and BRD4 also play a positive role in transcriptional elongation (Donner et al., 2010). These studies thus establish a positive role for CDK8 in transcriptional elongation on serum response genes, which is critical to ensure an appropriate and timely response to growth factor signals.

Oncogenic alterations in kinase module subunits

Structural and functional studies, such as those described above, have identified the kinase module as major conduit through which key instructive, inductive and homeostatic cell signals flow through Mediator. Most often, these pathway signals are routed through CDK8/19, which effect chemical change and functional properties in key transcriptional targets that contribute to programmed gene expression changes. Notably, many of these pathways and the cellular processes to which they contribute are associated with human tumorigenesis. Indeed, most have been considered “hallmarks” of cancer, including cell cycle control, immunity, oxygen and energy homeostasis and growth factor signaling (Figure 4) (Hanahan & Weinberg, 2011). In addition, the kinase module is linked with nearly every major developmental signaling pathway, many of which are also dysregulated as a course of oncogenesis. Accordingly, it is perhaps not surprising that that the kinase module has been increasingly implicated in human tumorigenesis. Table 1 provides a catalog of currently reported tumor-associated alterations in kinase module subunits. Note that we include only studies with reported alteration frequencies in excess of 1.5%. Below we highlight specific examples with a focus on those for which mechanistic relationships as opposed to strictly correlative associations are established.

Table 1.

Oncogenic alterations in kinase module subunits.

Subunit	Tumor Type	Alteration	%	Mechanism	Study
CDK8	Bladder cancer	Mutation in 4/127	3.1	–	Cancer Genome Atlas Research (2014b)*
	Breast cancer	Amplification-independent overexpression	6.9	–	Eirew et al. (2015)*
	Colorectal cancer	Amplification-dependent overexpression in 13/50	26.0	Activation of CDK8 kinase activity: promotes Wnt/β-catenin signaling	Firestein et al. (2008) Schneider et al. (2011) Firestein et al. (2010) Seo et al. (2010) Morris et al. (2008)
		Amplification-dependent overexpression in 9/220	4.1		Cancer Genome Atlas Network (2012a)*
		Mutation in 4/72	5.6	–	Seshagiri et al. (2012)*
	Lung adenocarcinoma	Mutation in 4/230	1.7	–	Cancer Genome Atlas Research Network (2014c)*
		Underexpression through deletion in 4/230	1.7	–	Cancer Genome Atlas Research Network (2014c)*
	Melanoma	Amplification independent overexpression in 29/38	76.0	Activation of CDK8 kinase activity: promotes epigenetic control through macroH2A	Kapoor et al. (2010)
	Prostate cancer	Underexpression through deletion in 3/61	4.9	–	Grasso et al. (2012)*
	Small cell lung cancer	Mutation in 1/29	3.4	–	Peifer et al. (2012)*
CycC	Adenoid cystic carcinoma	Underexpression through deletion in 2/60	3.3	–	Ho et al. (2013)*
	Bladder cancer	Underexpression in through deletion in 2/127	2.4	–	Cancer Genome Atlas Research Network (2014b)*
	Breast cancer	Amplification-dependent overexpression in 11/29	37.9	–	Eirew et al. (2015)*
	Colon cancer	Amplification-dependent overexpression in 56/208	26.9	–	Bondi et al. (2005)
	Gastric cancer	Amplification-independent overexpression	80.0	–	Galamb et al. (2007)
	Leukemia	Underexpression through deletion in 12/13	92.0	Disruption of CDK8/19 kinase activity: promotes upregulated Notch	Li et al. (1996)
		Underexpression through deletion in 13/150	8.7		Li et al. (2014b)
	Lung adenocarcinoma	Underexpression through deletion in 3/182	1.6	–	Imielinski et al. (2012)*
	Peripheral nerve sheath tumor	Underexpression through deletion in 1/15	6.7	–	Lee et al. (2014)*
	Osteosarcoma	Underexpression through deletion	62.0	–	Ohata et al. (2006)
	Ovarian cancer	Amplification-dependent overexpression in 7/316	2.2	–	Cancer Genome Atlas Research Network (2011)*
	Prostate cancer	Underexpression through deletion in 3/56	5.4	–	Baca et al. (2013)*
		Amplification-dependent overexpression in 2/56	3.6	–	Baca et al. (2013)*
		Underexpression through deletion in 2/103	1.9	–	Taylor et al. (2010)*
		Underexpression through deletion in 2/61	3.3	–	Taylor et al. (2010)*
	Stomach adenocarcinoma	Underexpression through deletion in 6/287	2.1	–	Cancer Genome Atlas Research Network (2014a)*
MED12	Adrenocortical carcinoma	Mutation in 6/45 (dispersed)	13.0	–	Assie et al. (2014)
	Adenoid cystic carcinoma	Mutation in 1/60 (dispersed)	1.7	–	Ho et al. (2013)*
	Bladder cancer	Mutation in 10/127 (dispersed)	7.9	–	Cancer Genome Atlas Research Network (2014b)*
		Mutation in 4/109 (dispersed)	3.7	–	Kim et al. (2015)*
	Breast cancer	Mutation in 2/65 (dispersed)	3.1	–	Shah et al. (2012)*
		Mutation in 3/103 (dispersed)	2.9	–	Banerji et al. (2012)*
	Chronic lympocytic leukemia	Mutation in 37/709 (exon 1/2)	5.0		Kampjarvi et al. (2014)
				Disruption of CDK8 kinase activity
	Colorectal cancer	Mutation in 3/72 (dispersed)	4.2	–	Seshagiri et al. (2012)*
		Mutation in 10/212 (dispersed)	4.7	–	Cancer Genome Atlas Network (2012a)*
	Diffuse gastric cancer	Mutation (dispersed)	48.0	–	Majewski et al. (2013)
	Fibroepithelial tumors			Disruption of CDK8 kinase activity
	Fibroadenomas	Mutation in 58/98 (exon 2)	59.0		Lim et al. (2014)
		Mutation in 17/26 (exon 2)	65.0		Piscuoglio et al. (2015)
		Mutation in 6/9 (exon 2)	67.0		Nagasawa et al. (2015)
		Mutation in 36/58 (exon 2)	62.0		Yoshida et al. (2015)
	Malignant phyllodes tumors	Mutation in 2/5 (exon 2)	40.0		Cani et al. (2015)
		Mutation in 5/11 (exon 2)	45.0		Nagasawa et al. (2015)
		Mutation in 1/13 (exon 2)	8.0		Piscuoglio et al. (2015)
		Mutation in 10/13 (exon 2)	77.0		Yoshida et al. (2015)
	Benign phyllodes tumors	Mutation in 22/25 (exon 2)	88.0		Piscuoglio et al. (2015)
		Mutation in 15/18 (exon 2)	83.0		Yoshida et al. (2015)
		Mutation in 4/5 (exon 2)	80.0		Cani et al. (2015)
	Borderline phyllodes tumors	Mutation in 7/9 (exon 2)	78.0		Piscuoglio et al. (2015)
		Mutation in 12/15 (exon 2)	80.0		Yoshida et al. (2015)
		Mutation in 4/5 (exon 2)	80.0		Cani et al. (2015)
	Lung adenocarcinoma	Mutation in 8/182 (dispersed)	4.4	–	Imielinski et al. (2012)*
		Mutation in 13/230 (dispersed)	5.7	–	Cancer Genome Atlas Research Network (2014c)*
	Lung squamous cell carcinoma	Mutation in 5/178 (dispersed)	2.8	–	Cancer Genome Atlas Research Network (2012)*
	Medulloblastoma	Mutation in 1/37 (dispersed)	2.7	–	Robinson et al. (2012a)*
	Melanoma	Mutation in 1/25 (dispersed)	4.0	–	Berger et al. (2012)*
		Mutation in 7/121 (dispersed)	5.8	–	Hodis et al. (2012)*
	Prostate cancer	Mutation in 6/111 (dispersed)	5.0	Predicted through disruption of CDK8 kinase activity	Barbieri et al. (2012)
		Mutation in 1/56 (dispersed)	1.8		Baca et al. (2013)*
		Mutation in 5/109 (dispersed)	4.6		Barbieri et al. (2012)*
		Mutation in 2/66 (dispersed)	3.3		Grasso et al. (2012)*
		Overexpression in 5/66: amplification-independent	8.2	Through TGFβ signaling	Grasso et al. (2012)*
		Overexpression in 47/160: amplification independent	29.4		Shaikhibrahim et al. (2014)
	Small cell lung cancer	Mutation in 2/42 (dispersed)	4.8	–	Rudin et al. (2012)*
		Mutation in 2/29 (dispersed)	6.9	–	Peifer et al. (2012)*
	Stomach adenocarcinoma	Mutation in 4/100 (dispersed)	4.0	–	Wang et al. (2014a)*
		Mutation in 13/287 (dispersed)	4.5	–	Cancer Genome Atlas Research Network (2014a)*
		Mutation in 1/30 (dispersed)	3.3	–	Kakiuchi et al. (2014)*
		Mutation in 1/22 (dispersed)	4.5	–	Wang et al. (2011)*
	Uterine tumors
	Abdominal leiomyoma	Mutation in 1/6 (exon 2)	17.0	Disruption of CDK8 kinase activity	de Graaff et al. (2013)
	Atypical leiomyoma	Mutation in 4/42 (exon 2)	10.0		Zhang et al. (2014a)
	Bizarre leiomyoma	Mutation in 1/4 (exon 2)	25.0		Matsubara et al. (2013)
	Cellular leiomyoma	Mutation in 3/22 (exon 2)	14.0		Zhang et al. (2014a)
		Mutation in 2/6 (exon 2)	33.0		Matsubara et al. (2013)
	Typical uterine leiomyoma	Mutation in 73/88 (exon 1/2)	83.0		Heinonen et al. (2014)
		Mutation in 65/76 (exon 1/2)	86.0		Heinonen et al. (2014)
		Mutation in 11/19 (exon 2)	58.0		de Graaff et al. (2013)
		Mutation in 15/28 (exon 2)	54.0		Schwetye et al. (2014)
		Mutation in 159/225 (exon 2)	70.0		Makinen et al. (2011b)
		Mutation in 100/148 (exon 2)	67.0		McGuire et al. (2012)
		Mutation in 35/45 (exon 2)	80.0		Matsubara et al. (2013)
		Mutation in 30/40 (exon 2)	75.0		Li et al. (2014a)
		Mutation in 35/67 (exon 2)	52.0		Je et al. (2012)
		Mutation in 41/69 (exon 2)	59.0		Makinen et al. (2013)
		Mutation in 6/9 (exon 2)	67.0		Perot et al. (2012)
		Mutation in 9/12 (exon 2)	75.0		Ravegnini et al. (2013)
		Mutation in 14/28 (exon 2)	50.0		Makinen et al. (2011a)
		Mutation in 92/143 (exon 2)	63.0		Halder et al. (2015)
		Mutation in 12/36 (exon 2)	33.0		Di Tommaso et al. (2014)
		Mutation in 133/178 (exon 2)	75.0		Bertsch et al. (2014)
	Leiomyosarcoma	Mutation in 2/10 (exon 2)	20.0		Perot et al. (2012)
		Mutation in 1/7 (exon 2)	14.0		de Graaff et al. (2013)
		Mutation in 6/20 (exon 2)	30.0		Schwetye et al. (2014)
		Mutation in 3/27 (exon 2)	11.0		Ravegnini et al. (2013)
		Mutation in 3/41 (exon 2)	7.0		Kampjarvi et al. (2012)
		Mutation in 2/12 (exon 2)	17.0		Matsubara et al. (2013)
		Mutation in 4/38 (exon 2)	10.0		Zhang et al. (2014a)
		Mutation in 1/25 (exon 2)	4.0		Schwetye et al. (2014)
		Mutation in 3/32 (exon 2)	9.7		Bertsch et al. (2014)
	Mitotically active leiomyoma	Mutation in 6/7 (exon 2)	86.0		Zhang et al. (2014a)
	Pelvic/retroperitoneal leiomyoma	Mutation in 10/29 (exon 2)	34.0		Schwetye et al. (2014)
	STUMP	Mutation in 1/9 (exon 2)	11.0		Perot et al. (2012)
		Mutation in 1/12 (exon 2)	8.0		Schwetye et al. (2014)
		Mutation in 2/18 (exon 2)	11.0		Zhang et al. (2014a)
	Uterine corpus endometrioid carcinoma	Mutation in 24/240 (dispersed)	10.0	–	Cancer Genome Atlas Research Network et al. (2013)*
MED13	Adenoid cystic carcinoma	Amplification-dependent overexpression in 1/60	1.7	–	Ho et al. (2013)*
	Bladder cancer	Amplification-dependent overexpression in 2/97	2.1	–	Iyer et al. (2013)*
		Amplification-dependent overexpression in 3/127	2.4	–	Cancer Genome Atlas Research Network (2014b)*
		Mutation in 7/127	5.5	–	Cancer Genome Atlas Research Network (2014b)*
	Breast cancer	Mutation in 2/100	2.0	–	Stephens et al. (2012)*
		Amplification-dependent overexpression in 38/482	7.9	–	Cancer Genome Atlas Network (2012b)*
		Amplification-dependent overexpression in 7/29	24.1	–	Eirew et al. (2015)*
		Amplification-dependent overexpression	–	–	Monni et al. (2001)
	Colorectal cancer	Mutation in 4/72	5.6	–	Rudin et al. (2012)*
		Mutation in 7/212	3.3	–	Cancer Genome Atlas Network (2012a)*
	Head and neck squamous cell carcinoma	Mutation in 3/74	4.1	–	Stransky et al. (2011)*
	Intrahepatic cholangiocarcinoma	Mutation in 2/40	5.0	–	Jiao et al. (2013)*
	Kidney renal clear cell carcinoma	Mutation in 10/418	2.4	–	Cancer Genome Atlas Research Network (2013)*
	Liver cancer	Mutation in 7/231	3.0	–	Ahn et al. (2014)*
	Lung adenocarcinoma	Mutation in 6/182	3.3	–	Imielinski et al. (2012)*
		Mutation in 7/230	3.0	–	Cancer Genome Atlas Research Network (2014c)*
		Amplification-dependent overexpression in 7/230	3.0	–	Cancer Genome Atlas Research Network (2014c)*
	Lung squamous cell carcinoma	Mutation in 5/178	2.8	–	Cancer Genome Atlas Research Network (2012)*
	Peripheral nerve sheath tumor	Amplification-dependent overexpression in 1/15	6.7	–	Lee et al. (2014)*
	Medulloblastoma	Mutation in 1/37	2.7	–	Robinson et al. (2012a)*
	Melanoma	Mutation in 2/25	8.0	–	Berger et al. (2012)*
		Mutation in 6/121	5.0	–	Hodis et al. (2012)*
	Ovarian cancer	Amplification-dependent overexpression in 3/316	1.9	–	Cancer Genome Atlas Research Network (2011)*
	Pharyngeal cancer	Mutation in 1/56	1.8	–	Lin et al. (2014)*
	Prostate cancer	Amplification-dependent overexpression in 2/103	1.9	–	Taylor et al. (2010)*
		Amplification-dependent overexpression in 3/61	4.9	–	Grasso et al. (2012)*
	Sarcoma	Amplification-dependent overexpression in 5/207	2.4	–	Barretina et al. (2010)*
	Small cell lung cancer	Mutation in 2/42	4.8	–	Rudin et al. (2012)*
	Stomach adenocarcinoma	Mutation in 2/100	2.0	–	Wang et al. (2014a)*
		Mutation in 14/287	4.9	–	Cancer Genome Atlas Research Network (2014a)*
		Amplification-dependent overexpression in 7/287	2.4	–	Cancer Genome Atlas Research Network (2014a)*
		Mutation in 1/22	4.5	–	Wang et al. (2011)*
	Uterine corpus endometrioid carcinoma	Mutation in 16/240	6.7	–	Cancer Genome Atlas Research Network et al. (2013)*

^*Studies collected from TCGA Bioportal.

A dominant theme to emerge from these reported studies is the loss or gain of CDK8 kinase activity, which appears to function in a context-dependent manner to either promote or suppress tumorigenesis. Mechanistically, this can derive from oncogenic alterations to CDK8 itself. For example, CDK8 gain of function (kinase activity) has been implicated in colorectal cancer, where CDK8 is often overexpressed or amplified and is also a potent oncogene. Furthermore, CDK8 overexpression has been documented, and CDK8 inferred an oncogene, in melanoma, as well as pancreatic, breast and gastric cancers. In contrast, CDK8 loss of function has been noted and CDK8 thus linked with tumor suppression, in uterine leiomyomas, T-cell acute lymphoblastic leukemias and endometrial cancers.

As CDK8 activation is triggered by a precise network of subunit interactions within the kinase module, CDK8 activity change could also derive from oncogenic alteration in any of its companion subunits, particularly MED12 and CycC, both of which contribute directly to kinase activity. Indeed, mutations and expression changes in these two subunits continue to be identified in human tumors at a growing pace. For some of these pathogenic alterations, experimental data supports altered kinase activity as a likely mechanistic basis for their tumorigenic potential. Specific examples of such alterations are discussed below. For others, functional impact on kinase activity may be inferred, but experimental data will be required to validate these predictions. Nonetheless, these studies point to an emerging picture of the Mediator kinase module as an oncogenic unit, one in which pathogenic activation/deactivation through component change drives tumor formation through perturbation of signal-dependent gene regulation (Figure 10).

Figure 10.

Mediator kinase module as an oncogenic unit. A dominant theme to emerge from studies of the kinase module in human cancers is loss or gain of kinase activity, which appears to function in a context-dependent manner to either promote or suppress tumorigenesis. Mechanistically, kinase activity change could derive from oncogenic alteration in any one of the kinase module subunits, particularly MED12, CycC and CDK8, all of which contribute directly to kinase activity. In this respect, the kinase module operates much like an oncogenic unit, one in which pathogenic activation/deactivation through component change drives tumor formation through perturbation of signal-dependent gene regulation. The large arrow represents CDK8 activity increase (green) or decrease (red). The specific cancer settings in which these changes occur are listed in the green or red boxes above or below the arrow. Some established CDK8 substrates are listed to the right of the arrow, and those identified to be effectors of oncogenic signaling are rendered in color (green: substrate phosphorylation suppresses tumorigenesis; red: substrate phosphorylation promotes tumorigenesis; black: impact of phosphorylation on tumorigenesis not yet established). (see color version of this figure at www.informahealthcare.com/bmg).

CDK8

Colorectal cancer

The first reported oncogenic alterations in CDK8 involved overexpression through amplification-dependent or -independent mechanisms, leading to the prediction that CDK8 is an oncogene (Firestein et al., 2008; Kapoor et al., 2010). However, loss of CDK8 and/or its associated kinase activity has since been linked with endometrial cancer, uterine leiomyomas (through MED12 mutations) and T-ALL (through CycC deletion), implying a tumor suppressive role for CDK8 kinase activity in these settings (Gu et al., 2013; Li et al., 2014b; Makinen et al., 2011b; Turunen et al., 2014). It therefore appears that CDK8 can function to promote or suppress tumorigenesis in a context-dependent manner. Nonetheless, one setting in which CDK8 clearly drives oncogenesis is colorectal cancer.

Worldwide, colorectal cancer is the fourth most common and the fifth leading cause of cancer-associated deaths (Ferlay et al., 2015). While early surgical excision of non-invasive tumors is essentially curative, few effective treatment options are available for advanced stage disease, which carries a grave prognosis. In 2008, CDK8 was reported to be a colorectal cancer oncogene (Firestein et al., 2008). In this regard, CDK8 suppression was shown to inhibit proliferation in colon cancer cells characterized by high levels of CDK8 and β-catenin hyperactivity. Furthermore, CDK8 kinase activity was found to be required for cell transformation, as well as β-catenin-regulated target gene transcription and colon cancer cell proliferation. Notably, CDK8 resides on chromosome 13q12.13, a region of recurrent copy number gain in a substantial fraction of colon cancers, and subsequent work confirmed that CDK8 is amplified or overexpressed in a large fraction of tumors (Firestein et al., 2008, 2010; Martin et al., 2007; Morris et al., 2008; Schneider et al., 2011; Seo et al., 2010; Tsafrir et al., 2006). CDK8 overexpression is associated with high colon cancer-specific morbidity. Five-year survival rates for patients with CDK8 overexpression is significantly lower than those who do not overexpress CDK8, and this association is independent of disease stage and other predictors of disease outcome (Firestein et al., 2010). Thus, CDK8-overexpressing tumors identify a subset of colorectal cancers that are predictive of poor disease-free survival.

Colon tumors arise from intestinal crypts, from whence progenitor-derived epithelial cells commence differentiation as they initiate their ascent up intestinal villi (de Lau et al., 2007). Maintenance of the crypt progenitor phenotype is dependent upon the expression of genes programmed by the canonical Wnt/β-catenin pathway, and constitutive activation of this pathway is a driving force in the immortalization of intestinal epithelia and the initiation of colorectal cancer (Behrens & Lustig, 2004; de Lau et al., 2007; Kinzler & Vogelstein, 1998). Indeed, aberrant activation of the canonical Wnt/β-catenin pathway occurs in >90% of colorectal cancers. As discussed previously, CDK8 is functionally implicated in the Wnt/β-catenin pathway. CDK8 represses E2F1, a negative regulator of β-catenin-dependent gene activation (Morris et al., 2008). In addition, CDK8 is a β-catenin co-activator (Firestein et al., 2008). Both of these functions require CDK8 kinase activity, and their net effect is to promote oncogenic Wnt/β-catenin signaling (Figure 5). These twin functions are both likely to contribute to the role of CDK8 as a colorectal cancer oncogene.

Recent work also reveals an oncogenic role for CDK8 through maintenance of tumor dedifferentiation (Adler et al., 2012). Mechanistically, this function appears to derive, at least in part, from CDK8-mediated regulation of MYC-dependent downstream target genes (Adler et al., 2012). In this regard, targeted depletion of CDK8 was shown to inhibit growth and promote differentiation of xenograft tumors derived from human colon cancer cells. Furthermore, genome-wide transcriptome profiling in human colon cancer cells revealed that CDK8 activates an ES cell-like gene network enriched in MYC target genes, providing a molecular basis to explain its proclivity to restrict tumor cell differentiation. Notably, enhanced expression of a CDK8-regulated, ES cell MYC-target gene signature was correlated with loss of differentiation and poor outcome in primary human breast cancers. Collectively, these findings suggest that CDK8 functions, at least in part, through MYC to maintain colorectal tumors in an undifferentiated state (Adler et al., 2012).

Melanoma

Melanoma is the least common but the most deadly form of skin cancer, accounting for only 4% of skin cancer cases but >75% of skin cancer deaths (Jerant et al., 2000). Over the past 40 years, the incidence of melanoma has increased more rapidly than any other cancer (Weinstock, 2001). While highly curable in its early stages, there is no effective treatment for metastatic melanoma, and a greater appreciation of the genetic and epigenetic changes involved in melanoma progression is therefore essential to advance these limited options. Work in this regard has uncovered an oncogenic function for CDK8 in melanoma involving epigenetic control and the histone variant macroH2A (mH2A) (Kapoor et al., 2010). To determine if histone variant exchange is a possible epigenetic determinant of melanoma progression, the H2A variant profile, including macroH2A (mH2A) and H2A.Z, isoforms typically associated with transcriptionally silent and active chromatin states, respectively, was probed during the course of melanoma progression. Strikingly, this analysis revealed a global decrease in mH2A expression, along with chromatin decondensation, coincident with a reciprocal increase in H2A.Z expression, suggesting possible H2A variant exchange during the course of melanoma progression. Notably, forced reduction of mH2A in minimally malignant melanoma cells enhanced their proliferative and migration capacities in vitro as well as their growth and metastatic potential in vivo, consistent with a role for mH2A as a suppressor of malignant melanoma. Subsequent efforts to assess the impact of chromatin decondensation elicited by mH2A loss involved gene expression profiling and led to the identification of CDK8 as a mH2A-repressed gene and, thus, a potential mediator of malignant melanoma progression. Accordingly, CDK8 knockdown in mH2A-depleted melanoma cells suppressed the proliferative advantage provoked by mH2A loss, and an inverse correlation was observed between mH2A and CDK8 expression in melanoma patient samples. Together, these findings implicate CDK8 as a major effector of mH2A-mediated melanoma progression (Figure 6) (Kapoor et al., 2010).

Breast cancer

CDK8 also exerts an oncogenic function in breast cancer through its participation in a regulatory axis that serves to coordinate cell cycle progression and tumorigenesis. Strikingly, this signaling axis shares conserved regulatory nodes, including mH2A and CDK8 itself, with the pathway described above that operates to control melanoma progression (Figure 6). As discussed earlier, CDK8 acts as a downstream effector in a Skp2–mH2A1–CDK8 axis to control p27 levels and cell cycle progression through G2M (Xu et al., 2015). As it turns out, this axis also controls ploidy and breast tumorigenesis, providing a built-in checkpoint to couple control of the G2/M transition with tumorigenesis. As described previously, a combination of molecular, genetic and biochemical approaches were used to establish the hierarchy of functional interactions within the Skp2–mH2A1–CDK8 axis. These studies revealed that Skp2, through suppression of mH2A1, promotes CDK8 expression to control G2/M phase progression through modulation of p27 levels (Figure 6) (Xu et al., 2015). Notably, studies have shown that Skp2 deficiency suppresses, while Skp2 overexpression promotes, tumorigenesis (Chan et al., 2012; Inuzuka et al., 2012; Lin et al., 2010). To determine if the Skp2–mH2A1–CDK8 axis that controls G2/M progression also regulates ploidy and tumorigenesis, efforts were undertaken to examine the impact of modulating mH2A1 and CDK8 levels on the extent of polyploidy and tumor formation provoked by Skp2 deficiency using both cell and xenograft tumor models (Xu et al., 2015). Strikingly, low levels of polyploidy and tumor formation observed in Skp2-deficient cells could be reversed by either mH2A1 knockdown or CDK8 overexpression, thus confirming the same epistatic relationships among individual axis components that operate in control of G2/M transit (Figure 6). Furthermore individual axis components were significantly correlated (in directions predicted) with one another and also with human breast cancer progression and patient survival outcomes. Taken together, these findings identify the Skp2–mH2A1–CDK8 axis as a key pathway to orchestrate the G2/M cell cycle transition with breast tumorigenesis, and also suggest its utility as a predictive marker of patient survival (Xu et al., 2015).

Endometrial cancer

Whereas CDK8 exerts an oncogenic activity in melanoma, as well as breast and colorectal cancers, it is nonetheless linked with tumor suppressive potential in endometrial cancer (Gu et al., 2013). Endometrial cancer is the fifth most common cancer in women worldwide and one of the most frequent gynecological malignancies (Ferlay et al., 2015). While localized disease is responsive to treatment and most patients face good prognosis, systemic relapse is associated with poor survival despite treatment and patient survival averages only 7–12 months (Werner & Salvesen, 2014). Like other gynecological malignancies, endometrial cancer is responsive to ovarian steroid hormones, including estrogen and progesterone. Accordingly, major risk factors for endometrial cancer include unopposed post-menopausal estrogen therapy, obesity and nulliparity (Yang et al., 2013). Recent work suggests that CDK8 may play a tumor suppressive role in endometrial cancers, although the mechanistic details for this activity presently remain obscure (Gu et al., 2013). In this regard, prior work had established an apparent evolutionarily conserved role for CDK8 in control of lipogenesis (Zhao et al., 2012). Based on epidemiological data linking obesity with endometrial cancer risk, the role of CDK8 in endometrial cancer cell growth and tumorigenicity was examined (Gu et al., 2013). Forced expression or targeted depletion of CDK8 in endometrial cancer cells harboring low or high endogenous CDK8 expression levels, respectively, were observed to impact tumorigenic properties in opposite ways. CDK8 overexpression reduced, while CDK8 depletion enhanced, cell proliferation, cell migration and invasion, cell cycle arrest and tumor formation in a murine xenograft model (Gu et al., 2013).

Other cancers and considerations

CDK8 has been linked with oncogenic potential in additional cancer types, but most of these reports are largely correlative with few mechanistic details presently elucidated. For example, in gastric adenocarcinoma CDK8 has been correlated with β-catenin, thought to be an early driving force in the development of these tumors (Kim et al., 2011). Analysis of primary human gastric adenocarcinoma samples (early and late stage) revealed that CDK8-expressing tumors exhibited distinct clinical, prognostic and molecular features, particularly when correlated with β-catenin intracellular localization. In this regard, although CDK8 and β-catenin were both expressed in early and late stage disease, CDK8 expression was significantly positively correlated with β-catenin delocalization (including intranuclear localization) and activation during disease progression, particularly in lymph node metastases. Furthermore, CDK8 depletion in human gastric cancer cells reduced β-catenin levels. These findings are consistent with the established role of CDK8 and the kinase module as a functional transducer of Wnt/β-catenin signaling, and suggest that combination screening for CDK8 expression and β-catenin localization could offer prognostic value in pathological evaluation of gastric adenocarcinomas (Kim et al., 2011).

The oncogenic activity of CDK8 is further supported by recent work showing that it is involved in chemotherapy-induced paracrine signaling in colon, breast and ovarian cancers (Porter et al., 2012). Exposure of human colon cancer cells to doxorubicin, a DNA-damage inducing chemotherapeutic drug, results in increased cell survival and increased expression of secreted tumor promoting factors; however, this effect can be reversed upon inhibition of CDK8 by Senexin A, a newly identified inhibitor of CDK8 (Porter et al., 2012). These findings were recapitulated in vivo, in which CDK8 inhibition by Senexin A was found to reduce xenograft tumor growth in mice after pre-treatment with doxorubicin. Notably, meta-analyses of microarray data acquired from independent breast and ovarian cancer studies revealed that high CDK8 expression was strongly correlated with poor relapse free survival in both breast and ovarian cancer patients, particularly ovarian cancer patients treated with DNA-damaging chemotherapeutic drugs. Collectively, these findings suggest that CDK8 inhibition might increase the efficacy of cancer chemotherapy (Porter et al., 2012).

CYCLIN C

T-ALL

T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive hematologic malignancy arising from the neoplastic transformation of T-cell progenitors. T-ALL represents a high-risk subtype of ALL and accounts for 10–15% of pediatric and 25% of adult ALL cases (Roti & Stegmaier, 2014). Among the established oncogenic alterations linked to the development of T-ALL, constitutive activation of NOTCH1 is the most prominent. Indeed, nearly 50% of T-ALLs carry activating mutations in NOTCH1 (Weng et al., 2004). Many of these mutations disrupt amino acids in the NOTCH1 intracellular domain (ICN1) that code for its regulated destruction when ICN1-dependent gene activation it is no longer developmentally required. These mutations thus permit ICN1 to escape programmed proteolytic destruction, resulting in its pathological stabilization and enhanced proliferation and survival of leukemic cells (Roti & Stegmaier, 2014; Weng et al., 2004). Notable recent work identified CycC as a haploinsufficient tumor suppressor in T-ALL, and further linked this function to its role as an activator of CDK8/19-mediated phosphorylation and consequent proteolytic turnover of oncogenic ICN1 (Li et al., 2014b).

CycC was originally linked to T-ALL almost 20 years ago based on the observation that CycC deletions occurred at high frequency in T-ALL patient samples (Li et al., 1996). More recent findings confirmed that CycC is a tumor suppressor in the setting of T-ALL (Li et al., 2014b). The use of gene-targeted mice carrying deletions in either or both CycC alleles revealed that CycC ablation or heterozygosity can collaborate, to the same extent, with other oncogenic alterations to accelerate development of T-ALL, thus establishing CycC as a haploinsufficient tumor suppressor. Consistent with this conclusion, a significant fraction of human T-ALLs carry heterozygous deletions in the gene encoding CycC (CCNC), and CCNC expression is reduced by 2-fold in these tumors (Li et al., 2014b). Notably, CDK19 resides in relative close genomic proximity (<10 Mb) to CycC, and chromosomal deletions that encompass CycC consequently often also include CDK19 (Li et al., 2014b).

The molecular basis by which CycC suppresses tumorigenesis in T-ALL is directly linked to its role as an obligate activator of CDK8/19-dependent regulation of ICN1 levels. As discussed previously, CycC–CDK8/19 function as physiological suppressors of ICN1-dependent gene expression during development. Mechanistically, this occurs because CycC–CDK8/19 phosphorylates, and thereby primes, ICN1 for degradation via the ubiquitin-proteasome pathway (Li et al., 2014b). This biological activity provides a mechanistic basis to explain the function of CycC as a tumor suppressor in the lymphoid system; CycC–CDK8/19 control ICN1 levels, and therefore, its activity as an oncogenic transcription factor (Figure 7). This was confirmed by both biochemical and molecular cell biological analyses in CycC-deficient cells and animals (Li et al., 2014b). Thus, conditional ablation of CycC in vivo triggered a delay in ICN1 degradation in thymocytes, bone marrow cells, hematopoietic cells and embryonic stem cells. Furthermore, CDK8 activity was lost in CycC-deficient MEFs and, as expected, no interaction of CycC with MED12/MED13 could be observed. Finally, acute ablation of CycC in thymocytes and knockdown of CycC in human T-ALL cells were both found to diminish phosphorylation of endogenous ICN1. Taken together, these findings reveal that CycC suppresses tumorigenesis through CDK8/19-mediated modulation of oncogenic ICN1 levels, and further, that oncogenic onset through CycC deletion occurs as a consequence of diminished CDK8/19 activity (Figure 7) (Li et al., 2014b).

Figure 7.

CycC is a haploinsufficient tumor suppressor that antagonizes oncogenic Notch1. Notch1 is a binary cell-fate determinant. Signal-regulated intramembrane proteolysis of Notch1 produces a transcriptionally active Notch intracellular domain (ICN1) that traffics to the nucleus and regulates gene expression programs involved in proliferation, differentiation and apoptosis. Left: In CycC-proficient (CycC^+/+) cells, CycC–CDK8/19 function as physiological suppressors of ICN1-dependent gene expression by phosphorylating and thereby priming ICN1 for degradation via the ubiquitin-proteosome pathway (Li et al., 2014b). Right: In CycC-deficient (CycC^+/− or CycC^−/−) cells, oncogenic ICN1 escapes proteolytic destruction resulting hyperactivated signaling that drives tumorigenesis (Li et al., 2014b). Note that proteolytic turnover of promoter-bound ICN1 shown here is based on prior findings (Fryer et al., 2004). The location of ICN1 turnover as described by Li et al. (2014b) has not been described. (see colour version of this figure at www.informahealthcare.com/bmg).

Osteosarcoma

Osteosarcoma is the most common pediatric bone cancer in children and young adults. It arises from mesenchymal cells, is genetically complex, and characterized pathologically by spindle cells and aberrant osteoid formation (Gill et al., 2013). CycC has been reported to undergo allelic loss in osteosarcomas and exert growth suppression in osteosarcoma cell lines consistent with possible tumor suppressive activity in this disease setting (Ohata et al., 2006). In this regard, comparative genomic hybridization studies identified 33% (4/12) of osteosarcomas that displayed partial loss of chromosome 6q (where CCNC resides). Furthermore, loss of heterozygozity analyses using 21 microsatellite markers spanning chromosome 6q15–24 identified 74% (23/31) that displayed allelic loss of at least one of the 21 markers. Notably, CCNC lies within a region that is deleted in up to 62% of these cases. Congruent with these findings, CCNC mRNA was significantly reduced in both osteosarcoma cell lines and tissue samples. Apparently, reduced CCNC expression was due to allelic loss as no mutations were detected in 11 sequenced osteosarcoma tissues. Furthermore, forced overexpression of CycC in CycC-underexpressing osteosarcoma cell lines significantly inhibited cell growth (Ohata et al., 2006). These studies provide evidence that CycC may harbor tumor suppressive potential in osteosarcomas. Nonetheless, osteosarcomas are notoriously genetically complex and prone to chromosomal derangement (Gill et al., 2013). Therefore, whether CycC loss is cause or consequence of tumorigenesis remains to be definitively established through additional studies.

MED12

Alterations in both the sequence and expression of MED12 have been observed in human tumors. In particular, MED12 mutations have been reported with increasing frequency in a variety of tumors, consistent with its recent identification through large scale genomic analyses as one of only several hundred so-called “cancer driver genes” (Lawrence et al., 2014; Vogelstein et al., 2013). Interestingly, this function may extend to MED12L, which was recently implicated as a potential cancer driver gene through a cross-species genomic comparison of murine and human oral squamous cell carcinomas (Onken et al., 2014). The type and distribution of mutations within MED12 is notable (Figure 8). The vast majority of reported mutations are missense mutations that cluster in a hotspot region within exon 2. These mutations alter highly conserved amino acids residues (L36, Q43 and G44) within the encoded protein, implying an important biological function for this region (Turunen et al., 2014). Notably, MED12 exon 2 mutations are largely tumor-restricted and occur at very high frequency (59–80%) in estrogen-dependent benign stromal tumors of the uterus (uterine leiomyomas) and breast (fibroadenomas and phyllodes tumors). In addition, exon 2 mutations occur recurrently, albeit less frequently, in malignant uterine leiomyosarcomas (7–30%), chronic lymphocytic leukemias (5%) and colorectal cancers (0.5%). Outside of exon 2, additional MED12 missense mutations have been reported at intermediate frequency (5%) in prostate cancer. Recent genetic and biochemical studies have provided initial clues regarding the impact of MED12 exon 2 mutations on MED12 function and the molecular basis for their tumorigenic potential (see below). Nonetheless, additional studies will be required for a deeper understanding of how these and other mutations in MED12 drive tumorigenesis in different disease settings.

Figure 8.

Tumor-associated exon 1/2 mutations in MED12 disrupt Mediator-associated CDK activity. Schematic diagram of CycC, CDK8, MED13 and MED12, with chromosomal location of each indicated. Pathogenic mutations and their approximate locations in MED12 are colored and annotated in the legend. The CycC/CDK8 binding interface on MED12, corresponding to its N-terminal 100 amino acids, is disrupted by exon 1/2 mutations, leading to loss of Mediator-associated CDK activity. Shown here is the predicted disposition of Mediator in myometrium and MED12 WT UL compared to MED12 exon 1/2 mutant UL (Turunen et al., 2014; Kampjarvi et al., 2014). FG, Lujan, Ohdo and prostate cancer associated mutations generally cluster in the middle region of MED12 and are therefore positionally distinct from exon 1/2 mutations. Nonetheless they may confer a similar defect (indicated by the dashed line), as FG and Lujan mutant MED12/Mediator complexes are known to suffer a gene-specific defect in the recruitment of Cyc-CDK8 (Zhou et al., 2012). (see colour version of this figure at www.informahealthcare.com/bmg).

Uterine leiomyomas

Uterine leiomyomas (UL; also called fibroids) are benign monoclonal neoplasms of the myometrium and represent the most common solid tumor in reproductive age women (Bulun, 2013). Tumors occur in ~77% of women overall and are clinically manifest in ~25% by age 45 (Walker & Stewart, 2005). Although benign, these tumors are nonetheless associated with significant morbidity. They are the primary indicator of hysterectomy, and represent a major source of gynecological and reproduction dysfunction, ranging from profuse menstrual bleeding and pelvic discomfort to infertility, recurrent miscarriage and pre-term labor (Bulun, 2013; Stewart, 2001; Walker & Stewart, 2005).

Despite their high prevalence and clinical significance, the etiological underpinnings of UL have heretofore remained largely obscure. In this regard, recurrent genetic aberrations have been observed in UL, including chromosome 7q deletions, rearrangements affecting the HMGA2 gene on chromosome 12q14, and biallelic loss of the fumarate hydratase (FH) gene (Bulun, 2013; Stewart, 2001; Walker & Stewart, 2005). Nonetheless, these alterations occur at low combined frequency and cannot explain the vast majority of tumors. In a landmark 2011 study, however, recurrent somatic mutations in MED12 exon 2 were found to occur at very high frequency (~77%) in UL (Makinen et al., 2011b). Most of these (~60%) are missense mutations that precipitate substitutions at three highly conserved amino acids within MED12: L36R (5% of tumors), Q43P (4%) and G44X (49%; where X=R, S, C, A, D, V corresponding to all six possible underlying base substitutions). The remaining ~40% of exon 2 mutations correspond to small in-frame insertions/deletions that maintain protein reading frame. These initial findings have since been validated by multiple independent studies worldwide, encompassing patients of diverse racial and ethnic origins. To date, ~70% of (>1500) UL have been found to harbor MED12 exon 2 mutations (Je et al., 2012; Kampjarvi et al., 2012; Makinen et al., 2011b, 2013; Markowski et al., 2012, 2013; McGuire et al., 2012; Mehine et al., 2013; Ono et al., 2012; Perot et al., 2012). More recently, additional UL-linked mutations occurring in MED12 exon 1 were identified (Kampjarvi et al., 2014). These mutations (all in-frame insertions/deletions/) occur at low frequency (~0.8%), and thus only slightly increase the overall proportion of MED12-mutant UL. Notably, comparative transcriptome profiling of MED12 WT and mutant UL revealed that MED12 exon 1 and 2 mutant tumors cluster together but separately from MED12 WT tumors, revealing: (i) MED12 exon 1 and 2 mutations are functionally equivalent; (ii) MED12 WT and mutant fibroids are unlikely related by driver mutations occurring in a common MED12-dependent pathway; and (iii) uterine fibroids may be stratified into at least four molecular subtypes based on their distinct recurrent genetic alterations (i.e. driver mutations) and their corresponding unique gene expression profiles, indicative of separate pathways to tumorigenesis (Kampjarvi et al., 2014). Among these, MED12 mutant tumors are the quantitatively dominant subtype, occurring in ~70 of tumors. These findings confirm and extend prior observations that MED12 mutations and HMGA2 overexpression, another well-established alteration in fibroids, occur mutually exclusive of one another (Bertsch et al., 2014; Mehine et al., 2013). Regarding gene expression alterations, prominent pathways linked to MED12 mutant tumors were found to include estrogen receptor α (ERα), TGF-β and Wnt/β-catenin signaling, possibly suggesting dysregulation of these pathways as a course of tumorigenesis. Finally, MED12 mutation status was independently correlated in a recent study with IGF-2 overexpression (Di Tommaso et al., 2014). Taken together, these findings identify mutant MED12 dysregulated pathways that might be therapeutically targeted for non-surgical intervention in UL.

Along with their combined high frequency occurrence, several additional lines of genetic evidence suggest that MED12 exon 1 and 2 mutations contribute to the genesis of UL. First, all identified mutations affect evolutionarily conserved and functionally important residues in the encoded MED12 protein (Makinen et al., 2011b; Turunen et al., 2014). Second, monoallelic expression of the mutant MED12 allele has been confirmed in all tumors duly examined (Makinen et al., 2011b). Finally, stem/progenitor cells isolated from patient-derived MED12-mutant UL are tumorigenic in murine transplantation xenograft models (Ono et al., 2012). These latter findings point to MED12-mutant UL stem cells as a likely cell of origin for most UL (Moravek et al., 2015; Ono et al., 2013; Yin et al., 2015). Thus, genetic evidence strongly supports the notion that MED12 exon 1 and 2 mutations are pathogenic. Recent biochemical studies, on the other hand, have revealed the functional impact of these mutations on kinase module activity (Turunen et al., 2014).

As described previously, MED12 activates CycC–CDK8 through a direct interaction involving MED12 amino acids 1–100 (encoded by exons 1 and 2) and a conserved surface groove on CycC. This indicates that all UL-linked mutations in MED12 affect its CycC-binding interface, and subsequent biochemical studies confirmed that exon 1 and 2 mutations in MED12 disrupt its ability to bind to CycC and activate CDK8 in Mediator (Figure 8) (Kampjarvi et al., 2014; Turunen et al., 2014). Collectively, these findings identify a common molecular defect associated with UL-linked mutations in MED12, and implicate aberrant CDK8 activity in the pathogenesis of UL. It is important to note that, in addition to CDK8, CDK19 is also expressed in myometrium and leiomyomas (Turunen et al., 2014). As these paralogous subunits assemble into the kinase module in a mutually exclusive manner, the extent to which the oncogenic potential of MED12 mutations derives from loss of Mediator-associated CDK8 versus CDK19 activity is presently unclear.

Breast fibroepithelial tumors

Fibroepithelial tumors of the breast comprise a heterogeneous group histologically characterized by a mixture of stromal and epithelial cells (Tan & Ellis, 2013; Yang et al., 2014). Within this group, fibroadenomas and phyllodes tumors represent the dominant tumor types. Fibroadenomas represent the most common breast tumors in women under 30 (Foster et al., 1988). Although benign, they are nonetheless associated with a 2-fold increase in the relative risk of developing future breast adenocarcinoma (Dupont et al., 1994). Phyllodes tumors, in contrast, are rare and classified as benign, borderline or malignant based on specific histological criteria (Reinfuss et al., 1996; Yang et al., 2014). Recent studies have identified MED12 exon 2 to be recurrently mutated at high frequencies in both fibradenomas (59–65%) and phyllodes tumors (63–80%) (Cani et al., 2015; Lim et al., 2014; Nagasawa et al., 2015; Piscuoglio et al., 2015; Yoshida et al., 2015). Notably, in both tumor types, the mutation spectrum is nearly identical to that of UL in both exon location and variant codon preferences. Thus, codon 44 is a major mutational hotspot, accounting for the majority of lesions, while codons 43 and 36 are also significantly affected, albeit at lower relative frequencies. These observations imply a common underlying genetic basis among UL, fibroadenomas and phyllodes tumors. In this regard, comparative expression profiling and gene set enrichment analyses from MED12 WT and mutant fibroadenomas revealed estrogen receptor and TGF-β signaling pathways to be particularly prominent, consistent with findings in UL. As observed for UL, exon 2 mutations in both fibroadenomas and phyllodes tumors disrupt the MED12/CycC interface and Mediator-associated CDK8/19 activity.

Prostate cancer

Prostate cancer is the second most frequently diagnosed cancer and the fifth leading cause of cancer deaths in men worldwide (Ferlay et al., 2015). Prostate cancer onset is driven by circulating androgens through their interactions with the androgen receptor (AR), a ligand-activated transcription factor that promotes high levels of transcription from androgen-responsive genes that control prostate cancer cell growth and survival (Lupien & Brown, 2009). Accordingly, most prostate cancer patients respond to initial androgen deprivation therapy, but invariably succumb to lethal castration resistant prostate cancer (CRPC), a disease state characterized by AR-dependent tumor growth despite castrate levels of circulating androgens (Chen et al., 2008; Heidegger et al., 2013). Mechanisms purported to underlie CRPC include tumor-specific expression of ligand-independent AR isotypes, intratumoral androgen production and stimulation of AR activity via crosstalk with alternative signaling pathways (Karantanos et al., 2013; Yuan et al., 2014). In this regard, recent work has revealed an intriguing link between MED12 overexpression and TGF-β signaling in CRPC (Shaikhibrahim et al., 2014). MED12 expression was observed to be significantly elevated in 70 distant metastatic CRPC samples and 90 local recurrent CRPC samples compared to androgen-sensitive prostate cancer and benign prostatic tissues. Immunohistochemical analyses of these samples revealed that MED12 expression correlated significantly with cell proliferation, and overexpression of nuclear MED12 correlated with activated TGF-β signaling as assayed by phosphorylated SMAD. In vitro, MED12 depletion significantly impaired androgen-independent prostate cancer cell proliferation, partially blocked cell cycle progression and reduced expression of the TGF-β-target gene VIMENTIN. Thus, enhanced accumulation of nuclear MED12 occurs frequently in CRPC where MED12 is also implicated in TGF-β signaling (Shaikhibrahim et al., 2014). These findings are possibly consistent with an established biological role for CycC–CDK8 in the instigation of a phosphorylation-dependent SMAD action turnover switch, since CycC–CDK8 is activated by MED12 in Mediator. Furthermore, previous reports have shown that SMAD3, a downstream effector in TGF-β signaling, is required for growth of prostate cancer tumors in nude mice, possibly highlighting the role of TGF-β signaling in prostate cancer progression (Lu et al., 2007).

In addition to overexpression, MED12 is also recurrently mutated in prostate cancers. The first such study identified 7/152 primary tumor samples (4.6%) harboring MED12 mutations, all of which were missense mutations (Barbieri et al., 2012). Among these, five (~70%) represented recurrent alterations in exon 26 that led to the same amino acid substitution (L1224F), while the remaining two corresponded to D727E and P1310Q substitutions. These findings have since been substantiated by independent study projects in The Cancer Genome Atlas (TCGA), wherein two additional L1224F mutations and four new mutations (V1223L, E1144Q and N1845T) have been reported. Notably, all prostate cancer-associated mutations in MED12 occur in the central region of the gene, and are thus positionally distinct from exon 2 mutations linked to UL. Presently, it is unclear whether the two classes of mutations differentially impact MED12 function or if they reflect distinct pathways to tumorigenesis. Interestingly, most prostate cancer mutations in MED12 lie close to MED12 missense mutations responsible for the XLID disorders FG, Lujan and Ohdo syndromes (Figure 8) (Risheg et al., 2007; Schwartz et al., 2007; Vulto-van Silfhout et al., 2013). As discussed previously, both FG/R961W and Lujan/N1007S mutations in MED12 impair gene-specific recruitment into Mediator of CycC–CDK8, leading to hyperactivated SHH signaling, which is notably linked with prostate cancer (Chen et al., 2011; Zhou et al., 2012). Therefore, should prostate cancer and XLID mutations in MED12 alter its function in equivalent ways, the former, like the latter, would be predicted to disrupt Mediator-associated CDK activity, with potential implications for oncogenic signaling. Further study will be required to test this prediction.

Other cancers and considerations

It is now well established that MED12/Mediator and cohesion cooperate to enforce long-range chromosomal interactions (via looping) that are essential for enhancer-driven Pol II transcription (Figure 1) (Newman & Young, 2010). As a major architectural activity in chromatin, cohesion directly promotes loop formation, while Mediator facilitates this process by bridging enhancer and core promoter binding factors. Recent work suggests that this functional collaboration, initially established for the ES cell pluripotency gene expression program (Kagey et al., 2010), also extends to regulation of the ERα transcriptome in human breast cancer cells (Prenzel et al., 2012). In this regard, comparative global expression profiling following knockdown of the SMC3 subunit in cohesion or MED12 in Mediator revealed that both were similarly required for proper ERα-dependent transcription control genome-wide, including auto-regulation of ESR1 gene itself. Furthermore, ChIP analyses following knockdown of SMC or MED12 in breast cancer cells established a reciprocal requirement for their own occupancy on ESR1 as well as that of Pol II across the ESR1 gene. Together these findings establish a critical role for MED12 in global ERα-dependent gene regulation through probable control of large-scale chromatin organization in human breast cancer cells (Prenzel et al., 2012).

MED13

Little is currently known regarding the involvement of MED13 in human tumorigenesis. Nonetheless, MED13 appears to be a target of multiple alterations in a variety of tumors. Further work will be required to elucidate the impact and contribution of these lesions, if any, to MED13 function and tumor formation. Amplification-dependent overexpression of MED13 has been reported in breast cancer (Monni et al., 2001). In this regard, efforts to identify target genes on chromosome 17q23, a region of frequent amplification with poor prognosis, led to the identification within this locus of two regions of common amplification in breast cancer cell lines, and these results were confirmed by FISH in 184 primary breast tumors. A total of 99 transcripts were localized to these regions, and the expression levels of their corresponding genes in a panel of six breast cancer cell lines were subsequently determined. Notably, MED13 was one of these genes. Although the functional significance of this finding remains unknown, it was speculated that MED13 might contribute to the more aggressive clinical course observed in breast cancer patients with 17q23-amplified tumors (Monni et al., 2001). MED13 has also been reported to be a target of altered adenosine-to-inosine (A-to-I) RNA editing in human brain cancer. This editing process, with established links to human disease, is catalyzed by the ADARs (adenosine deaminases acting on RNA) and involves the conversion of a genetically encoded adenosine (A) to an inosine (I) in double-stranded RNA, including both coding and non-coding RNAs (Dominissini et al., 2011; Paz et al., 2007; Slotkin & Nishikura, 2013). It was reported that editing within MED13 was significantly reduced in brain cancer compared to normal tissue. However, the biological relevance of this finding and its implications for MED13 and brain tumorigenesis are presently unclear.

Clinical implications

Tumor biomarkers

The recent explosion in molecular genetics and cancer genomics has contributed to the evolving definition of a biomarker, which now includes any quantifiable disease-associated molecular changes (Biomarkers Definitions Working Group, 2001; Poste, 2011). Validated biomarkers provide clinicians with important information with which to determine risk, monitor disease progression and predict patient survival and response to therapy (Ludwig & Weinstein, 2005; Mayeux, 2004). Presently, no kinase module subunit has qualified for FDA approval as a clinically validated tumor biomarker. However, this is not surprising, given their relatively recent discoveries, their even later (and ever expanding) links with human cancers, and the rigorous and lengthy process associated with biomarker validation, which entails replication and qualification, the latter focused on factors (sensitivity and specificity) critical to assess the diagnostic utility of a candidate. Nonetheless, emerging studies highlight the potential clinical use of kinase module subunits as biomarkers. For example, CDK8 expression has been shown to predict patient disease-free survival in colorectal and breast cancers (Firestein et al., 2008, 2010; Xu et al., 2015). Furthermore, MED12 exon 2-mutation status has been shown to correlate with markers of poor prognosis in CLL (Kampjarvi et al., 2015). Thus, in the near term, we expect additional studies in a range of cancers to be focused on the positive and negative predictive power of kinase module subunits, either alone or as multivariate classifiers along with established tumor biomarkers, for their discriminative and prognostic potentials. Beyond these measures, improved molecular insight concerning the functional impact of oncogenic alterations in kinase module subunits might help inform therapeutic options. For example, CDK8 amplification in colorectal cancer and CycC deletion in T-ALL are linked with activated Wnt/β-catenin and NOTCH1 signaling, respectively, implying possible therapeutic benefit from the application of WNT and NOTCH1 inhibitors (Firestein et al., 2008; Li et al., 2014b). In the longer term, it thus seems likely that kinase module subunits will enter the clinical realm, where they will offer predictive, prognostic and/or possible discriminative value in human cancers.

CDK8 inhibitors

The discovery of CDK8 as an oncogene with inherent kinase activity has spurred obvious interest in the development CDK8 inhibitors as possible therapeutic agents. With the exception of the FDA-approved Sorafenib, which was initially developed to target tyrosine kinases and only later found to inhibit CDK8, no established CDK8 inhibitor has thus far progressed beyond the pre-clinical phase. Nonetheless, as discussed below, several compounds show potential for further development (Figure 9) (Guerzoni et al., 2014; He et al., 2013; Kaur et al., 2010; Porter et al., 2012; Verlinden et al., 2000). Notably, with exception of Senexin A, all of these compounds have been either developed or tested against CDK8, but not CDK19. Nonetheless, the high degree of sequence conservation between these two paralogs implies that CDK8 inhibitors might be similarly effective against CDK19. Further studies are clearly warranted to confirm prediction.

Figure 9.

Inhibitors and microRNAs that target the kinase module. Compounds and microRNAs (miR) that have been established to inhibit (blunted line) or activate (arrow) CDK8 are shown. Notably, miR-107 has been shown to both inhibit and promote CDK8 expression in a cancer cell-type specific manner. The question marks denote CDK8 activity-dependent subunit interfaces (CycC–CDK8 and MED12–CycC interfaces) that can putatively be targeted for CDK8 kinase inactivation. (see colour version of this figure at www.informahealthcare.com/bmg).

Senexin A

Senexin A is a selective ATP binding site inhibitor with nanomolar affinity for CDK8 and CDK19 (Porter et al., 2012). It was identified initially through a cell-based compound library screen for small molecules capable of inhibiting p21-induced tumor-promoting factors (Porter et al., 2012). Among hits from this screen, five carried a 4-aminoquinozoline center, and subsequent structural optimization led to the lead compound Senexin A. In biological assays, Senexin A inhibited β-catenin-driven transcription and induction of transcription factor EGR1, consistent with the established role of CDK8 in activating Wnt/β-catenin signaling and promoting expression of EGR1 via the serum response network (Donner et al., 2010; Firestein et al., 2008; Porter et al., 2012). Notably, CDK8 inhibition by Senexin A also blocked DNA damage-induced tumor-promoting paracrine activities both in vitro and in vivo (Porter et al., 2012). In cell-based assays, Senexin A decreased the expression of secreted tumor-promoting factors and DNA damage-induced anti-apoptotic activity (Porter et al., 2012). In a mouse xenograft model of lung cancer, administration of both Senexin A and doxorubicin, compared to doxorubicin alone, reduced tumor growth and increased the number of tumor-free animals (Porter et al., 2012). Thus, Senexin A is a novel inhibitor of CDK8/19, and this activity could improve the efficacy of chemotherapeutic agents by minimizing their collateral tumor-promoting effects.

Sorafenib and deep pocket binders

Sorafenib (BAY-43-9006; Nexavar^®) is an oral multi-tyrosine kinase inhibitor that received fast-track FDA approval in 2005 for the treatment of renal cell carcinoma, and is also effective against hepatocellular carcinoma and thyroid cancer (Schneider et al., 2011; Smalley & Flaherty, 2009; Wilhelm et al., 2008). Originally developed to target tyrosine kinase receptors, sorafenib is an ATP-competitive type II inhibitor whose binding mode extends from the kinase ATP binding site to an adjacent “deep pocket” (Schneider et al., 2011; Smalley & Flaherty, 2009; Wilhelm et al., 2008). This deep pocket binding mode is believed responsible for the clinical success of sorafenib as well as another small molecule tyrosine kinase inhibitor imatinib. Although Cyclin–CDK complexes were thought incapable of accommodating sufficiently dynamic conformational change within the ATP site to render deep pocket access, sorafenib was nonetheless found to bind uniquely to CycC–CDK8 by inducing a deep pocket binding mode with a “DMG-out” conformation characteristic of inactive kinases (Schneider et al., 2011). Sorafenib binds CycC–CDK8 with high affinity (nanomolar K_d) and, importantly, an elongated residence time, which is considered a key success factor for compound optimization in vitro and enhanced drug potency in vivo (Schneider et al., 2013). Sorafenib is therefore a promising CDK8 inhibitor, and a model for the rationale design and optimization of new inhibitors based on structure–activity and structure–kinetic relationships.

To identify potential CDK8 inhibitory compounds with elongated residence times similar to sorafenib in complex with CycC–CDK8, Schneider et al. (2013) recently employed a reporter displacement assay to screen a kinase-focused fragment compound library. The screen led to the discovery of seven lead-like compounds that were subsequently evaluated according to their unique structure–kinetic relationships with CycC–CDK8. Notably, this analysis revealed that compound residence time was not significantly influenced by the flip to the DMG-out conformation or by the displacement of the CDK8 T loop, but rather by large hydrophobic complementarities and hydrogen bonding (Schneider et al., 2013).

Kinase module interface disruption

CDK8 activity-dependent subunit interactions within the kinase module represent alternative targets to inhibit CDK8. Compounds that specifically disrupt CDK8 activity-dependent CycC–CDK8 and MED12–CycC interactions could potentially offer greater specificity than notoriously promiscuous ATP binding site inhibitors. In this regard, a virtual ligand screen of (~14 000) ChemBank anti-cancer compounds was recently employed to identify molecules predicted capable of docking onto the CDK8-binding surface of CycC (Rajender et al., 2011). This screen yielded six aryloxy compounds predicted to bind to CycC and competitively inhibit its association with CDK8 (Rajender et al., 2011). The biological relevance of these compounds is presently unclear, however, as no biochemical characterization or experimental validation was described.

Regarding the MED12–CycC interface, recent structural and biochemical studies revealed that MED12 binds to a conserved surface groove on CycC, one that is far removed from the CycC–CDK8 binding interface (Hoeppner et al., 2005; Schneider et al., 2011). Notably, mutagenic disruption of the MED12/CycC interface severely impairs CycC-dependent CDK8 kinase function, providing proof-of-principle for disruption of CDK8 (and possibly CDK19) activity-dependent subunit interactions as a basis for kinase inhibition (Turunen et al., 2014).

Indirect inhibitors

An increasing number of natural compounds have been reported to harbor CDK8 inhibitory activity, although their mechanism(s) of action are complex, indirect and often pleiotropic in nature. Silibinin, an active ingredient of milk thistle, has been reported to reduce expression of β-catenin, CDK8 and CycC, in APC-mutant colorectal cancer (SW480) cells (Kaur et al., 2010). Silibinin-mediated suppression of Wnt/β-catenin pathway components was accompanied by reduced β-catenin-dependent gene activation, reduced proliferation and increased apoptosis in cultured SW480 cells (Kaur et al., 2010). In vivo, silibinin treatment suppressed the growth of SW480 tumor xenografts through a mechanism involving reduced β-catenin-driven gene expression, reduced proliferation and an increased apoptosis. These studies identify silibinin as a multi-target inhibitor of oncogenic Wnt/β-catenin signaling.

Wogonin, a bioactive plant flavonoid, maltonis, a soluble maltol-derived hydroxypyrone and TX522 and TX527, 14-epi-analogues of 1,25-dihydroxyvitamin D3, have also been reported to inhibit CDK8 or CycC (Guerzoni et al., 2014; He et al., 2013; Verlinden et al., 2000). Wogonin was reported to induce G1/S phase cell cycle arrest in HCT116 colon cancer cells. Mechanistically, this was attributed to down-regulation of Wnt/β-catenin signaling, characterized by reduced expression of Wnts, enhanced proteolysis of β-catenin and inhibition of CDK8 kinase activity leading to impaired β-catenin-dependent gene activation (He et al., 2013). Maltonis was reported to inhibit proliferation and increase apoptosis of osteosarcoma cells through apparent modulation of multiple downstream effectors, including suppression of CDK8 mRNA levels (Guerzoni et al., 2014). TX522 and TX527 were reported to downregulate CycC protein expression in MCF7 breast cancer cells, which may contribute to the ability of these compounds to inhibit MCF7 cell growth both in vitro and tumor xenografts (Verlinden et al., 2000). Further studies will be required to clarify the mechanisms and establish the specificities of these compounds as inhibitors of CDK8 (and possibly CDK19) function.

MicroRNAs as potential therapeutic agents

Beyond the realm of traditional pharmacological drugs, RNA interference (RNAi) offers the potential for novel therapeutic approaches in cancer (Burnett & Rossi, 2012). Initially discovered as a cellular defense mechanism, this unique machinery has been exploited to great effect for the targeted manipulation of cancer-associated genes and pathways (Davidson & McCray, 2011; Kim & Rossi, 2007). In this regard, the development of RNAi-based therapeutic agents to target aberrant CDK8 activity could be of clinical value as CDK8 kinase activity is implicated in a variety of cancers. Recent reports have identified several endogenous microRNAs with specificity for CDK8 (Figure 9). MicroRNA-107 and miR-26b (miR-107 and miR-26b, respectively) bind to the CDK8 3′ UTR thereby targeting it for degradation. Notably, both miRNAs have been implicated in CDK8-dependent tumorigenesis (Li et al., 2014a,c). For example, in human breast and non-small cell lung cancer (NSCLC), miR-107 and miR-26b appear to oppose the oncogenic activity of CDK8, suggesting that these miRNAs are tumor suppressive in these settings. In the breast, basal expression of miR-107 and miR-26b is significantly decreased in breast cancer tissues compared to normal adjacent breast tissue. Delivery of miR-107 or miR-26b mimics into MDA-MB-231 human breast cancer cells reduced CDK8 protein expression and led to cell cycle arrest as well as reduced proliferation, colony formation and migration (Li et al., 2014a,c). In the lung, basal expression of miR-107 expression was found to be significantly lower in A549 human NSCLC cells compared to normal bronchial epithelial cells. Notably, CDK8 mRNA and protein expression levels were downregulated in response to miR-107 mimics (Zhang et al., 2014b). Although NSCLC is commonly treated with cisplatin, its clinical efficacy is limited, as some tumors show resistance or become resistant after treatment (Eaton & Martins, 2010). Significantly, ectopically expressed miR-107 rendered NSCLC cells significantly more sensitive to cisplatin therapy. Mechanistically, this effect was attributed to miR-107-mediated suppression of CDK8, since RNAi-mediated CDK8 knockdown conferred similar cisplatin sensitivity (Zhang et al., 2014b). Together, these studies indicate that reduced levels of miR-107 and miR-26b in breast and NSLC cells, respectively, lead to overexpression of oncogenic CDK8, possibly contributing to tumorigenesis and drug resistance.

In contrast to breast and lung cancers, wherein miR-107 is tumor-suppressive and functionally opposed to oncogenic CDK8, and in gastric cancer, miR-107 appears to be oncogenic and antagonistic to tumor suppressive CDK8. Thus, miR-107 was significantly increased in SGC7901 human gastric cancer cells compared to normal gastric epithelial cells (Song et al., 2014). Inhibition of miR-107 in SGC7901 cells led to reduced CDK8 expression, which was accompanied by reduced proliferation and colony formation, as well as increased apoptosis (Song et al., 2014). Taken together, these studies indicate that miR-107 harbors context-dependent oncogenic or tumor suppressive properties determined by the tumor-specific disposition of its direct target CDK8.

MED12 in drug resistance

The development and use of targeted pharmacological agents has gained traction in the fight against cancer; however, long-term use often results in drug resistance that renders tumors unresponsive to once-effective targeted agents. Therefore, a critical unmet need exists to elucidate the molecular mechanisms underlying drug resistance. As described previously, oncogenic MED12 mutations leading to MED12 dysfunction have been identified in several different tumor types, including uterine leiomyomas, fibroadenomas, phyllodes tumors and prostate cancer (Barbieri et al., 2012; Cani et al., 2015; Je et al., 2012; Kampjarvi et al., 2012, 2014; Lim et al., 2014; Makinen et al., 2011b, 2013; Markowski et al., 2012, 2013; McGuire et al., 2012; Mehine et al., 2013; Ono et al., 2012; Perot et al., 2012; Piscuoglio et al., 2015; Yoshida et al., 2015). It has recently been proposed that loss of MED12 function might represent a general mechanism by which cancer cells escape treatment with multiple therapeutic agents, a finding with significant clinical implications for the development of strategies to circumvent drug resistance (Huang et al., 2012).

In this regard, it was reported that MED12 knockdown in a variety of cancer cell types promotes MEK/ERK activation and elicits resistance to several therapeutic agents, including crizotinib, cisplatin, 5-Fluorouracil (5-FU) and ALK and EGFR inhibitors (Huang et al., 2012). Drug resistance in MED12 knockdown (MED12^KD) cells was found to be dependent on TGF-β signaling, as suppression of TGF-βR2 in combination with crizotinib led to significant inhibition of cell proliferation in MED12^KD cells post-resistance (Huang et al., 2012). Congruent with these findings, TGF-β signaling was found to promote MEK/ERK activation and confer resistance to ALK and EGFR inhibitors (Huang et al., 2012). To delineate the mechanistic link between MED12 and TGF-β signaling, biochemical and immunohistochemical analyses revealed that MED12 appears to play a direct role in inhibiting TGF-β signaling by interfering with proper TGF-βR2 glycosylation thereby preventing cell-surface expression of the receptor (Huang et al., 2012; Kim et al., 2012). It is well established that TGF-β signaling is critical for EMT as evident by the induction of mesenchymal markers such as Vimentin (VIM) and N-cadherin (CDH2) upon TGF-β stimulation (Thiery, 2009). Interestingly, VIM and CDH2 were also induced in MED12^KD cells, indicating that suppression of MED12 likely promotes an EMT-like process (Huang et al., 2012). Consistent with this idea, 31 MED12-regulated genes (of 237) were found to overlap with those that characterize a general EMT signature (229 total) (Huang et al., 2012). Interestingly, the MED12^KD signature was correlated poor survival in colorectal cancer patients (Huang et al., 2012). Furthermore, colorectal cancer patients with a MED12^KD-like signature showed a poor response to 5-FU-based chemotherapy and, in addition, drug resistance to MEK inhibitors was correlated with a MED12^KD-like signature in a heterogeneous panel of cancer cell lines (Huang et al., 2012). Significantly, the MED12^KD signature appears to be linked directly to drug resistance since a related gene expression program was shown to be activated in NSCLC tumors after acquisition of drug resistance to gefitinib (Huang et al., 2012). The observation that TGF-β signaling is activated upon MED12^KD led to the prediction that TGF-β inhibitors were likely to synergize with commonly used anti-cancer agents in MED12^KD tumors. In fact, this prediction was confirmed by the observation that the TGF-βR2 inhibitor LY2157299 showed strong synergy with crizotinib and gefitinib in MED12^KD cells. Mechanistically, this was shown to involve suppression of ERK, leading to inhibition of cell proliferation (Huang et al., 2012). Therefore, the combination of TGF-βR2 inhibitors with other commonly used anti-cancer agents could prove to be an effective strategy for treating cancers with low MED12 expression and/or elevated TGF-β signaling. Together, these findings suggest that MED12 expression level or functional status might be useful as an indicator of drug resistance, and further identify effective clinical approaches to circumvent this problem.

Conclusion and future perspectives

As the only component in Mediator with established enzymatic activity, the kinase module is replete with regulatory potential. Indeed, a confluence of genetic and biochemical evidence points to this module as both an integrative hub and central conduit in Mediator for developmental, facultative and homeostatic signals that control diverse aspects of cell fate and function. Accordingly, deregulated kinase activity has been increasingly identified as a key driver of tumorigenic events, consistent with the idea that these Mediator-dependent control nodes are vulnerable to oncogenic subversion. Notably, oncogenic change in kinase activity has been linked with alterations in MED12 and CycC, as well as CDK8, consistent with the fact that all three subunits contribute directly to kinase activity. From a functional perspective, the kinase module may thus be viewed as a single oncogenic unit, one in which the level of catalytic activity is determined by pathologic changes in subunit expression or function (Figure 10). This notion conforms to the established structural and functional character of the kinase module, as well as a substantive and growing body of functional and mechanistic data. It does not, however, exclude the possibility that kinase module subunits, particularly MED12/MED13, might contribute to tumorigenesis through kinase-independent mechanisms. This is particularly salient point in light of prior findings that MED12 can execute important cellular functions in a manner independent of kinase module activity (Ding et al., 2008).

Our steady advances thus far in unraveling the complexity of Mediator have precipitated new questions whose answers will no doubt offer deeper insight into fundamental properties of eukaryotic gene control and how these processes are subverted as a course of oncogenesis. We therefore conclude with what we consider to be emerging questions of academic interest and clinical significance.

First, what is the basis for the apparent context-dependent oncogenic versus tumor suppressive function of the kinase module? CDK8 kinase activity is oncogenic in melanoma, as well as breast and colorectal cancers. In contrast, loss of CDK8 kinase activity elicits tumorigenesis in T-ALL (via loss of CycC) and UL (via MED12 mutation), suggesting a tumor suppressive role in these tissues. The basis by which kinase module activity functions in a context-dependent manner to promote or suppress tumorigenesis is presently unclear, but likely derives in part from the tissue-specific expression and/or biological activity of its target substrates. In other words, CDK8 kinase activity, while an important means to an end, is less a determinant of oncogenic potential in a specific tissue-type than are its key substrates and the relevant cell fate-determination pathways in which they operate. Hence, in colorectal cancers, where Wnt/β-catenin signaling is a driving force, CDK8 represses E2F1, a negative regulator of β-catenin. In addition, CDK8 is a β-catenin co-activator. Both of these functions require CDK8 kinase activity, and their net effect is to promote oncogenic Wnt/β-catenin signaling (Figure 5). On the other hand, in T-ALL, the NOTCH1 intracellular domain (ICN1) escapes degradation primed by CDK8-mediated phosphorylation (due to deletion of CycC) and drives oncogenesis (Figure 7). Thus, the perceived functional dichotomy ascribed to kinase module activity as an apparently oncogenic in colorectal cancer and tumor suppressive in T-ALL is, at root, determined by the distinct biological activities of its respective substrates, E2F1 and ICN1.

Second, can our growing knowledge of kinase module structure and function be translated to effective therapies for CDK8-driven tumors? As described previously, CycC and MED12 both contribute to CDK8 activation. This distribution of labor within the kinase module obviously increases its susceptibility to dysregulation through oncogenic alteration in multiple subunits. On the other hand, it also increases the number of potentially druggable targets. Thus, advances in high-throughput cell-based and biochemical screening approaches have already yielded CDK8 active site inhibitors with good potential for further development. However, the CycC–CDK8 and MED12–CycC interfaces remain untapped in terms of therapeutic potential. Future studies should reveal whether these interaction surfaces represent actionable drug targets. Furthermore, the use of CDK8 activators to restore kinase function disrupted UL and T-ALL should merit future consideration. Concerning the clinical utility of CDK8 modulators, considerations relating to the tissue-specific and context-dependent function of CDK8 in tumorigenesis suggest caution. Thus, while CDK8 inhibition could well inhibit colorectal and breast tumor growth, it nonetheless might promote tumor formation in lymphoid and uterine compartments. The fact that CDK8 inhibitors (or activators) could have unintended oncogenic consequences in non-target tissues could therefore limit their clinical utility, and this issue will require serious consideration and further study.

Third, to what extent are the oncogenic properties of individual subunits restricted to their function within the kinase module or intact Mediator? In other words, do kinase module subunits function in isolation, and does the kinase module itself function independently of Mediator? Unfortunately, relevant data in this regard are currently limited, and additional investigation will be required to clarify this issue.

Fourth, concerning MED12, what is the basis for the striking observation that high-frequency exon 2 mutations are restricted primarily to benign estrogen-dependent tumors with mesenchymal stem cell origins? Presumably, this implies an important role for MED12/Mediator in the development and/or differentiation of this progenitor lineage, but data in this regard is presently lacking. Furthermore, what is the nature of the relationship between somatic MED12 exon 2 mutations and those in the middle region of MED12 linked to prostate cancer? Does their positional distinction in MED12 predict differences in functional impact and mechanisms of tumorigenesis? Notably, both classes of mutations are linked with steroid hormone-dependent tumors, and future studies will be required to clarify whether and how MED12 might interface with receptor-dependent transcriptional programs.

Fifth, what are the structural and functional relationships among paralogous subunits within the Mediator kinase module? Other than CDK8 and CDK19, for which initial investigation has commenced, very little is known regarding the extent of functional convergence and divergence as they relate to MED12/12L and MED13/13L. Genetic ablation of MED12 or MED13 in mice similarly confers early embryonic lethality, excluding the possibility of functional redundancy among their corresponding paralogs during development. However, beyond these narrow distinctions, much additional work is clearly required to delineate their cell- and tissue-type specific functions.

Finally, concerning the biological processes in which Mediator has been functionally implicated, it will be important to establish whether and how the kinase module contributes to their integration in space and time on both the cellular as well as physiological systems levels. For example growth factor signaling is closely linked with cell cycle progression and differentiation, all of which in turn are tightly controlled by the kinase module. How might the kinase module orchestrate and facilitate communion between these individual programs to enforce a coordinated functional network? As an established signal processor and transducer, the kinase module would seem to be evolutionarily internally hard-wired and structurally configured to fulfill this function, yet we currently know little about how this occurs. Here the emerging discipline of genomic systems biology and its intent to integrate high-throughput multi-layer genomics datasets with systems biology approaches will likely find a challenging problem for its application. In conjunction with advanced bioinformatics-based capabilities, these complimentary and wholly integrated methods should provide a unique and powerful tool-kit with which to deconstruct the complexity of Mediator-dependent gene regulation and dysregulation in development and disease.