Identification of Mediator kinase substrates in human cells using cortistatin A and quantitative phosphoproteomics

Mediator kinase substrates are largely transcription-associated proteins

The phosphoproteomics workflow in Fig. 1B identified novel phosphosites whose intensities decreased significantly with CA treatment (Fig. 2A). We identified 78 phosphosites, represented in 64 proteins, that we designated as high confidence based upon 1) their quantification in at least two of three biological replicates, 2) a reproducible mean H/L ratio across replicates, and 3) a significant decrease in H/L ratio with CA treatment as determined by an empirical Bayes analysis (Margolin et al., 2009; Ritchie et al., 2015). These high-confidence phosphosites are summarized in Table 1 and Figure 2A, and all quantified phosphosites are shown in Table S1. To ensure that a reduced H/L ratio did not result simply from a change in protein level, we completed a quantitative proteome analysis in parallel with phosphoproteomics. Importantly, very few high confidence phosphosites exhibited any change at the protein level with one hour of CA treatment (Table S2). Those that did change somewhat were FOXC1, MAML1, KDM3A, and ATF2, although some of these changes were not consistent across deep proteome replicates, and most of the phosphosite changes remained significant even after accounting for small changes in protein level. Although phosphosites not designated as high confidence sites could represent bona-fide Mediator kinase substrates (e.g. those that are quantified in only one biological replicate) we will only discuss targets designated as high confidence based on the criteria above.

Open in a separate window

Figure 2

Identification of Mediator kinase (CDK8/19) substrates

(A) Functional categorization of high-confidence Mediator kinase substrates identified.

(B) Volcano plot of statistically significant phosphosite changes with CA treatment using an empirical Bayes analysis.

To determine whether a H/L ratio for a phosphosite changed significantly with CA treatment across replicates, we employed an empirical Bayes statistical approach using the limma software package (Ritchie et al., 2015). An empirical Bayesian framework allowed for the calculation of adjusted p-values for each phosphosite (Fig. 2B). This approach can account for experiment-specific differences, which is advantageous compared to more arbitrary approaches, such as a universal fold-change cutoff (Margolin et al., 2009). We found that more phosphosite ratios decreased than increased upon CA treatment, as expected with targeted kinase inhibition for a short amount of time. This is shown by a higher number of data points on the left side of the volcano plot compared to the right side, using an adjusted p-value cutoff of 0.1 (Fig. 2B; see also Supplemental Note).

We used iceLOGO (Colaert et al., 2009) to determine statistically enriched motifs within the identified Mediator kinase substrates. We found that the majority of the phosphosites contained an S/T-P motif (Fig. S1A). Additionally, a proline at the −2 and −1 positions relative to the phosphorylation site was overrepresented. These data support the notion that many CDK8 phosphorylation sites occur within PX(S/T)P motifs as previously suggested (Alarcon et al., 2009; Bancerek et al., 2013). Serine was more frequently phosphorylated than threonine (Fig. S1A) and we did not see evidence for overrepresentation of basic residues at positions C-terminal to the phosphosite, as observed with other CDK motifs (Ubersax and Ferrell, 2007).

Because few substrates for human CDK8/19 have been identified, the analysis uncovered many phosphosite targets (Table 1). Many targets are DNA-binding transcription factors (TFs), chromatin regulators, or other known regulators of pol II activity (Fig. 2A, Fig. S1B), consistent with the established role of CDK8 in transcription. Additional substrates, including proteins implicated in DNA replication and repair (BRCA1, MDC1) and ubiquitination (HUWE1, CUL4B) suggest biological roles for Mediator kinases beyond transcription. A known CDK8 substrate, STAT1 S727 (Bancerek et al., 2013), was identified as a high confidence target, and other novel phosphosites reside in proteins that interact with CDK8-Mediator and/or the CDK8 module, including AFF4, MAML1, and Mediator subunits (Fig. 2A, Table 1). AFF4 is a core component of the super-elongation complex (SEC; (Luo et al., 2012)), which co-purifies with CDK8-Mediator (Ebmeier and Taatjes, 2010), and MAML1 is a Notch pathway co-activator that recruits CDK8 to Notch-dependent genes where it phosphorylates the Notch ICD (Fryer et al., 2004).

We submitted the 64 CDK8/19 substrate proteins to the STRING protein-protein interaction database (Szklarczyk et al., 2015) and found that six Mediator complex subunits, three subunits of the TIP60/NuA4 complex (EPC2, DMAP1, MRGBP), and two subunits of the NuRD complex (CHD3 and CHD4) were represented (high confidence score, ≥ 0.7; Fig. S1C). The TIP60/NuA4 and NuRD complexes are multi-subunit assemblies that possess multiple enzymatic activities, including nucleosome remodeling, acetyltransferase, and deacetylase activities. Additionally, this analysis identified a network of interacting proteins involved in DNA damage repair (Fig. S1C), as well as an interaction between XRN2 and SETX. Taken together, these data suggest that Mediator kinases regulate multiple and diverse cellular processes, potentially via several distinct multi-subunit assemblies.

Validation of selected Mediator kinase substrates

To further validate the CDK8/19 substrates identified with SILAC-based phosphoproteomics, we performedin vitro kinase assays, western blots, and MS experiments. We selected proteins representing different classes of substrates (Fig. 2A) for further evaluation. The DNA-binding TF STAT1, a previously identified CDK8 kinase substrate (Bancerek et al., 2013), was probed by western blot in IFN-γ induced HCT116 cells (Fig. 3A). This experiment confirmed STAT1 S727 as a Mediator kinase substrate in HCT116 cells, and also showed CA-dependent inhibition at low nM concentrations, as reported previously (Pelish et al., 2015).

Open in a separate window

Figure 3

In vitro validation of select CDK8/19 substrates

(A) Validation of STAT1 S727 as a Mediator kinase target in HCT116 cells.

(B) Western blot validation of SIRT1 T530 as a Mediator kinase target. Levels of total SIRT1 and other proteins known to regulate CDK8 activity (MED12 or CCNC), remained unchanged. TBP is a loading control.

(C) Quantitation of data in (B). Error bars are SEM; n=2 for technical replicates.

(D) In vitro kinase assay with recombinant CDK8 module and SIRT1. With increasing time, SIRT1 pT530 detection increases, indicating CDK8 is phosphorylating this site. Increase is not seen in no kinase or no substrate (ns) controls.

(E) In vitro kinase assay with GST-tagged TP53BP1 or RIF1 fragments. Alanine mutations at identified phosphorylation sites show reduced phosphorylation by CDK8.

(F) Overview of method for identifying MED12 and MED13 phosphorylation sites using recombinant CDK8 modules.

(G) Verification of MED12 S688 and MED13 S749 phosphorylation sites.

(H) In vitro kinase assay using CA and GST-pol II CTD as a substrate. Whereas each kinase tested phosphorylates this substrate, CA only inhibits the CDK8 module.

Among the chromatin modification and regulation substrates, we examined SIRT1, in part because an antibody against the phosphorylated SIRT1 T530 site was commercially available. When HCT116 cells were treated with CA, we noted a decrease in SIRT1 T530 phosphorylation (Fig. 3B and 3C). Total SIRT1 levels were unaffected by CA treatment, and levels of the CDK8 module subunits CDK8, CCNC, and MED12 were not changed by treatment with the compound (Fig. 3B). The approximately 50% reduction in phospho-SIRT1 did not change with increasing CA concentration, indicative of CA selectivity (Pelish et al., 2015). Although other kinases, such as CDK1 and JNK, are known to phosphorylate this site (Sasaki et al., 2008), treatment with inhibitors of CDK1 (RO-3306) and JNK family kinases (SP600125) did not seem to impact SIRT1 T530 phosphorylation; in fact, we were unable to completely reduce SIRT1 phospho-T530 detection, even when treating with all three inhibitors (Fig. S2). In vitro kinase assays with purified CDK8 module and SIRT1 confirmed CDK8-dependent SIRT1 T530 phosphorylation by western blot (Fig. 3D).

We next tested two different substrates, RIF1 and TP53BP1, linked to DNA replication and repair. Because these proteins are very large (each over 200 kDa), we expressed GST-tagged fragments (ca. 100 residues) surrounding the phosphosite. As shown in Figure 3E, the CDK8 module phosphorylated these substrates, whereas point mutations (S to A) at the identified phosphorylation site(s) greatly reduced substrate phosphorylation, supporting these sites as CDK8 module targets in vitro (Fig. 3E).

We also confirmed phosphorylation sites in MED12 and MED13 using in vitro kinase assays using the recombinant CDK8 module (containing CDK8, CCNC MED12, MED13) purified from insect cells. Incubation of the CDK8 module with ATP and subsequent TiO₂ enrichment and MS analysis confirmed both S688 on MED12 and S749 on MED13 as substrates (Fig. 3F and 3G). We did not identify the CCNC site from these experiments because the site identified from CA-treated HCT116 cells is not present in the canonical CCNC isoform used for recombinant CDK8 module expression and purification.

The data summarized in Figure 3 verified each of seven high-confidence Mediator kinase sites, representing about 10% of all high-confidence sites listed in Table 1. These results, combined with previous data demonstrating CA potency and specificity (Pelish et al., 2015), support the substrates listed in Table 1 as Mediator kinase targets. Although extensive kinome profiling has demonstrated CA specificity, we conducted in vitro kinase assays using a shared substrate, the pol II CTD, and found that CDK1, ERK2, and GSK3β activity was unaffected by CA treatment, even at concentrations ten-fold above those used for proteomic and gene expression analyses (Fig. 3H).

Mediator kinase inhibition has limited and specialized effects on transcription

As Mediator-associated kinases, it was plausible that inhibition of CDK8 and CDK19 activity could affect expression of large numbers of genes. We analyzed gene expression (RNA-Seq) data from CA treated HCT116 cells (Table S3). To minimize secondary or indirect effects resulting from long-term Mediator kinase inhibition, we completed RNA-Seq after a three-hour CA treatment (100 nM); this also helped match mRNA changes with measured phosphorylation changes that were determined after one-hour CA treatment. RNA-Seq analysis identified 150 genes whose expression changed significantly with CA treatment (Fig. 4A). Among these genes, the magnitude of change in expression was modest (largely 1.2 – 2 fold), indicating that CDK8/19 activity per se is not a major driver of their transcription, at least in the context of this analysis (HCT116 cells under normal growth conditions). Such modest gene expression changes were also observed in CA-sensitive cell lines (e.g. MOLM-14), although the genes affected were distinct (Pelish et al., 2015).

Open in a separate window

Figure 4

Mediator kinase inhibition is functionally distinct from CDK8 or CDK19 knockdown

(A) Heat map of differentially expressed genes (RNA-Seq) after 3 hr CA treatment. Green font represents transcription or chromatin regulator.

(B and C) Comparison with microarray data (Galbraith et al., 2013) using stable CDK8/19 knockdown (shRNA) vs. 3h CA treatment (B) or 24h treatment (C) in HCT116 cells under normal growth conditions. A 1.5-fold cutoff was used for microarray data and Cufflinks was used for CA-treated cells (no specific fold-change cutoff).

(D) TFBS analysis of promoters for genes whose expression changed with 3h CA treatment (listed in A). Promoters (±2kb from the TSS of the canonical isoform) were analyzed using F-Match, part of the Transfac database. Overrepresented sites with at least 1.5 fold increase vs. control promoters are shown for Transfac vertebrate matrices. Matrix name at left.

Gene expression changes in CA-treated cells compared to CDK8 or CDK19 knockdown

Because CA inhibits both CDK8 and CDK19 kinases, we used previously published HCT116 microarray datasets (normal growth conditions), in which either CDK8 or CDK19 had been stably knocked down (Donner et al., 2010; Galbraith et al., 2013), as a comparison to CA-treated HCT116 cells. Only genes exhibiting 1.5-fold change in expression or greater, with p-values <0.05, were used from the microarray data; these genes were compared to our RNA-seq analysis in which cells were treated with 100 nM CA for three hours. We observed only a modest overlap among genes differentially expressed (Fig. 4B; note that because CA inhibits both CDK8 and CDK19, gene sets for CDK8 or CDK19 knockdown were combined). Because cellular knockdown experiments take over 24 hours to manifest, the modest correlation in gene expression changes could reflect the short time of CA treatment. However, RNA-seq analyses after 24 hour CA treatment revealed similarly low numbers of shared gene expression changes (Fig. 4C). These results, further summarized in Figure S3, suggest that the physical presence of the CDK8 or CDK19 protein has distinct effects on transcription compared to targeted kinase inhibition.

Functional links between gene expression changes and Mediator kinase substrates

Because many Mediator kinase substrates are transcription factors (Fig. 2A), we hypothesized that some of the observed differences in gene expression due to CDK8/19 inhibition might be caused by changes in TF function. To begin to address this hypothesis, we extracted promoter sequences (± 2kb from TSS) for genes that were differentially expressed (increased or decreased expression) 3h CA treatment. F-Match was then used to compare promoter sequences to controls to determine if any TF binding sites, reported as Transfac matrices, were over-represented. The ratio of this increase (CA-treated cells vs. DMSO controls) is displayed for over-represented sites in Figure 4D. We found that many of the identified Transfac matrices for genes whose expression increased or decreased with CA treatment were mutually exclusive (Fig. 4D). That is, an enriched TF binding site in CA increased genes was generally not present in CA decreased genes, and vice versa. A hypergeometric test confirmed a significant overlap (p-value = 1.15E⁻⁵) between Mediator kinase targets identified in Table 1 and Transfac matrices identified in our gene expression promoter analysis (Fig. S3F).

Many of the Transfac matrices identified by F-match (Fig. 4D) can be traced back to Mediator kinase activity, as summarized in Figure S4. For example, the RREB1 TF was identified in both the F-match analysis (RREB1_01) and the SILAC phosphoproteomics (Table 1). Enriched TF binding sites were observed for genes with altered expression in CA-treated HCT116 cells (Fig. 4A), including MYC, a β-catenin target gene, EGR1 (i.e. KROX_Q6), and HES1, a Notch pathway target gene. Moreover, the MGA and NAB2 proteins, each high-confidence Mediator kinase substrates, are known regulators of MYC and EGR1 activity, respectively (Hurlin et al., 1999; Svaren et al., 1996). Transfac matrices representing the AP2 and ATF family of TFs (e.g. AP2alpha_01, ATF1_Q6) were also uncovered in the F-match analysis. The Mediator kinase target KLF12 is a well-established repressor of AP2α activity (gene name: TFAP2A) (Imhof et al., 1999), whereas the ATF2 and ATF7 proteins were each identified as Mediator kinase targets. Finally, enriched binding sites for E2F1 and SREBP, previously identified CDK8 kinase substrates (Morris et al., 2008; Zhao et al., 2012), were found by the F-match analysis shown in Figure 4D. Although these TFs were not identified in our HCT116 phosphoproteomics experiments, several co-regulators of E2F1 or SREBP activity (e.g. MGA, SIRT1; see Discussion) were among the high-confidence substrates listed in Table 1. Thus, there are many functional links between CA-dependent changes in gene expression (Fig. 4A) and the Mediator kinase targets shown in Table 1.

CDK8/19 phosphorylate Mediator subunits and post-initiation transcription regulators

CDK8 can reversibly associate with Mediator to form a CDK8-Mediator complex (Taatjes et al., 2002), and immunoprecipitation-mass spectrometry experiments in HeLa or HEK293T cells suggest CDK19 interacts similarly with Mediator (Daniels et al., 2013; Ebmeier and Taatjes, 2010; Sato et al., 2004). We identified eight high-confidence CDK8/19 phosphorylation sites in six different Mediator subunits: CCNC, MED12, MED13, MED13L, MED14, and MED26. CCNC, MED12, MED13, and MED13L each associate with CDK8 or CDK19 as part of the kinase module of Mediator. MED13 appears to be important for physical interaction between the kinase module and Mediator (Knuesel et al., 2009), and previous studies have shown that increased MED13 or MED13L abundance can increase the proportion of CDK8-Mediator vs. core Mediator in cells (Davis et al., 2013). These previous results were shown in the context of inhibition of the E3 ubiquitin ligase FBW7, which ubiquitylates MED13 and MED13L to promote their degradation. Our quantitative whole proteome data showed that the abundance of MED13 and MED13L were each increased in CA-treated HCT116 cells. FBW7-dependent ubiquitylation of MED13 or MED13L required prior modification at residue T326, a phospho-degron site in MED13 and MED13L (Davis et al., 2013). The CDK8/19 sites identified in MED13 and MED13L are distinct (residue S749 and residue S878, respectively) and do not overlap with known or predicted phospho-degron motifs; thus, it remains unclear how Mediator kinase activity may affect MED13 or MED13L protein levels.

The MED26 subunit is generally absent from CDK8-Mediator purifications (Ebmeier and Taatjes, 2010; Sato et al., 2004; Taatjes et al., 2002) and hence, its phosphorylation by CDK8 or CDK19 may promote MED26 dissociation from Mediator. The CDK8/19 modification site on MED26 (S314), however, does not reside in regions required for Mediator association (Takahashi et al., 2011). The MED14 subunit is an important architectural factor within Mediator, and structural studies with reconstituted partial complexes and crosslinking-mass spectrometry (CXMS) revealed MED14 crosslinks with several Mediator subunits, including MED8 and MED7, involving MED14 residues 1256 and 1295, respectively (Cevher et al., 2014). These reside some distance (in sequence space) from the Mediator kinase phosphorylation sites (S1112, S1128, S1136). Furthermore, CXMS and cryo-EM data with reconstituted yeast Mediator and yeast pol II revealed MED14 interactions with pol II and TFIIF (Plaschka et al., 2015). However, the S. cerevisiae Med14 subunit from this study consisted of residues 1-755 (of 1082 residues in yeast Med14) and the human MED14 S1112, S1128, and S1136 residues do not appear to be conserved.

Knockdown experiments have implicated the CDK8 protein in the regulation of transcription elongation and/or pol II pausing or pause release (Donner et al., 2010; Galbraith et al., 2013). Furthermore, ChIP-Seq data from CA-treated MOLM-14 cells indicated a reduced pol II travel ratio (TR; ratio of promoter-bound pol II vs. pol II in gene body) at genes whose expression was up-regulated by CA (Pelish et al., 2015), implicating Mediator kinase activity in pol II pausing or pause release. The reduced TR in CA-treated cells could also reflect inhibition of premature pol II termination. Here, we identified AFF4, NELFA, MED26, POLR2M, SETX, and XRN2 as high-confidence Mediator kinase targets, and each of these factors has been implicated in regulation of pol II pausing, premature termination, or elongation (Brannan et al., 2012; Cheng et al., 2012; Jishage et al., 2012; Kwak and Lis, 2013; Lin et al., 2010; Takahashi et al., 2011; Wagschal et al., 2012).

Mediator kinases as potential metabolic regulators

CDK8 orthologs in Drosophila and yeast have been linked to lipid and glucose metabolism and regulation of cellular responses to nutrient stress (Kuchin et al., 1995; Lindsay et al., 2014; Mousley et al., 2012; Zhao et al., 2012). Upon Mediator kinase inhibition by CA, we observed changes in the abundance of about 200 proteins (Fig. 5), including many involved in basic metabolic pathways such as oxidative phosphorylation, fatty acid metabolism, and cholesterol homeostasis. MED13 and CCNC appear to regulate mitochondrial function in yeast (Cooper et al., 2014; Khakhina et al., 2014), and over-expression of MED13 in mouse cardiac tissue alters fatty acid metabolism, β-oxidation, and mitochondrial content (Baskin et al., 2014). We identified MED13 and CCNC as Mediator kinase substrates and observed an increase in MED13 protein levels upon CA treatment, which could contribute to altered fatty acid metabolism or oxidative phosphorylation observed in CA-treated cells (Fig. 5D).

CDK8 kinase activity has previously been linked to cholesterol metabolism and fatty acid synthesis via regulation of SREBP. In particular, CDK8-dependent phosphorylation of SREBP residue T402 correlated with SREBP degradation in Drosophila and mouse cells (Zhao et al., 2012). GSEA analysis of whole proteome data identified changes in the cholesterol homeostasis, adipogenesis, and fatty acid metabolism hallmark signatures in CA-treated cells (Fig. 5D). Moreover, F-match identified SREBP binding motifs as over-represented among genes whose expression changed upon CA treatment (Fig. 4D). Whereas phosphorylation of SREBP T402 was detected in our phosphoproteomics experiments, its level was not altered in CA-treated cells, suggesting alternate means of Mediator kinase-dependent SREBP regulation in HCT116 cells. Other kinases, including GSK3 (Sundqvist et al., 2005), are known to target SREBP T402 and we have confirmed that CA does not inhibit GSK3β in cell lysates (Pelish et al., 2015) or in in vitro kinase assays with the purified protein (Fig. 3H). Therefore, the SREBP T402 phosphorylation level may remain constant in CA-treated cells due to other kinases targeting this site. Alternately, SREBP may not be a substrate for CDK8 in HCT116 cells. SIRT1, a validated Mediator kinase target, can negatively regulate SREBP activity through deacetylation (Walker et al., 2010). The Mediator kinases phosphorylate SIRT1 at residue T530, and phosphorylation at T530 has been shown to activate the SIRT1 deacetylase (Sasaki et al., 2008). Thus, via SIRT1 and potentially other substrates, Mediator kinases may regulate cholesterol or fatty acid metabolism independent of direct SREBP phosphorylation in HCT116 cells.

Human Mediator kinases and TF turnover

Previous studies revealed that phosphorylation of nutrient-responsive TFs Gcn4, Ste12, or Phd1 by yeast Cdk8 promoted their degradation (Chi et al., 2001; Nelson et al., 2003; Raithatha et al., 2012). Studies in metazoans have shown evidence for CDK8-dependent phosphorylation of the TFs SMAD1, SMAD3, Notch ICD, SREBP, E2F1, and STAT1 (Alarcon et al., 2009; Bancerek et al., 2013; Fryer et al., 2004; Morris et al., 2008; Zhao et al., 2012). Among these, increased degradation of the Notch ICD, SMAD1, SMAD3, and SREBP correlated with phosphorylation. For these reasons, we anticipated that inhibition of CDK8 and CDK19 kinase activity would affect the protein levels of a subset of their targets. Whole proteome data revealed no evidence that TF phosphorylation by Mediator kinases affected their stability, even with analyses at 1, 3, 6, 18, or 24 hours of CA treatment. In fact, we found little evidence for altered stability of any high-confidence Mediator kinase targets, with the notable exception of MED13 and MED13L. Despite this result, cell type or context may be key factors that dictate the effect of Mediator kinase phosphorylation on protein turnover. Here, we evaluated HCT116 cells in normal growth conditions whereas Mediator kinases may in fact more generally regulate substrate protein turnover during stress responses or at different developmental stages.

Whereas no changes in TF turnover were evident from the whole proteome data in CA-treated vs. control cells, we identified many links between the gene expression changes and the phosphoproteomics data (Fig. S4). These results are consistent with Mediator kinases affecting TF activity in HCT116 cells under normal growth conditions, rather than TF turnover.

CDK8 as a colon cancer oncogene: Mediator kinase inhibition vs. subunit knockdown

CDK8 was identified as a colon cancer oncogene in part through an shRNA screen for genes required for HCT116 cell proliferation (Firestein et al., 2008). CDK8 was one of 166 candidates in this screen; CDK8 was also identified in a screen for factors required for activation of a β-catenin-driven reporter in a different colon cancer line, DLD-1 (Firestein et al., 2008). Our analyses with CA indicate that, in contrast to CDK8 knockdown, Mediator kinase inhibition does not affect HCT116 cell growth (Pelish et al., 2015). These findings highlight the distinction between physical loss of a protein vs. targeted inhibition of its enzymatic activity.

As a TF, β-catenin assembles with the DNA-binding proteins TCF and LEF-1 to activate genes that drive cell proliferation. HCT116 cells are heterozygous for a mutant β-catenin protein that is resistant to degradation (Morin et al., 1997). Consequently, HCT116 cells have increased β-catenin levels and are considered “β-catenin-dependent”. Consistent with an oncogenic function for CDK8, CDK8 knockdown prevented activation of β-catenin target genes in colon cancer cell lines (Firestein et al., 2008). In addition, the E2F1 TF has been shown to be an important negative regulator of β-catenin stability (through unknown mechanisms), and elevated levels of the CDK8 protein, as observed in HCT116 cells (Firestein et al., 2008), can block E2F1-dependent inhibition of β-catenin target gene expression (Morris et al., 2008). Thus, in colon cancer cells, the CDK8 protein appears to up-regulate β-catenin target gene expression in two ways: as a β-catenin co-activator and as an inhibitor of E2F1 activity.

Whereas E2F1 and β-catenin activity or stability is known to be regulated by phosphorylation, we did not observe significant changes in E2F1 or β-catenin protein or phosphopeptide levels in CA-treated HCT116 cells. An F-match analysis based upon gene expression changes in CA-treated cells, however, identified E2F binding motifs as over-represented (Fig. 4D), and Ingenuity Pathway Analysis (IPA) of upstream regulators identified β-catenin target genes as over-represented among those whose expression increased or decreased upon CA treatment (Table S7). GSEA of our proteomics data (CA-treated vs. untreated, 0 – 24h) revealed up-regulation of both the E2F1 and β-catenin pathways (Fig. 5D). Furthermore, numerous high-confidence Mediator kinase substrates are known to directly regulate β-catenin or E2F activity, and these are summarized in Table S8.

Although these results implicate Mediator kinase activity in the regulation of E2F1 and β-catenin transcription networks in HCT116 cells, the effects of Mediator kinase inhibition are clearly distinct from CDK8 or CDK19 knockdown (Donner et al., 2010; Firestein et al., 2008; Galbraith et al., 2013). This was not unexpected, as the physical presence of an enzyme typically serves structural roles, such as maintaining the integrity of a multi-protein complex. For example, ablation of the CDK7 ortholog in yeast (Kin28) abolishes essentially all pol II transcription (Holstege et al., 1998), in contrast to targeted inhibition of Kin28 activity (Kanin et al., 2007). Direct comparison of the transcriptional changes resulting from physical loss of the CDK8 or CDK19 protein versus targeted inhibition of kinase activity (i.e. with protein levels remaining intact) revealed stark differences in both the genes affected and in the magnitude of gene expression changes. These differences highlight the importance of a structural or “scaffolding” role of the CDK8 or CDK19 proteins. Indeed, CDK8 knockdown decreases MED12 levels and increases CDK19 protein levels in HCT116 cells (Donner et al., 2010; Galbraith et al., 2013), which likely contributes to the distinct gene expression and anti-proliferative effects of CDK8 knockdown (Firestein et al., 2008) compared with kinase inhibition by CA. Because CA inhibits CDK19 as well as CDK8 (Pelish et al., 2015), this may also result in compensatory effects that distinguish the CDK8 knockdown phenotype from CDK8/CDK19 inhibition. Future studies are needed to more precisely establish the roles of CDK8 vs. CDK19 in regulating the elaborate E2F1, β-catenin, and other inter-related signaling networks that contribute to HCT116 survival and proliferation.

SILAC labeling

HCT116 cells were cultured in DMEM lacking arginine and lysine (Pierce-88420) supplemented with either Arg10 (33.6 μg/ml) and Lys8 (73 μg/ml) or Arg0 and Lys0 for heavy and light treatment, respectively. After six passages at 1:3 ratio, cells were tested for Arg/Lys incorporation and were subsequently supplemented with 200mg/L of proline (Sigma-P5607) as a small amount of Arg→Pro conversion was detected. Cells were maintained in 10% dialyzed FBS and penicillin/streptomycin.

TiO₂ phosphopeptide enrichment, ERLIC chromatography, and LC/MS/MS

Protocols were carried out as described (Stuart et al., 2015). An Orbitrap LTQ (Thermo Fisher) was used for phosphoproteomics, and an Orbitrap Velos (Thermo Fisher) was used for quantitative proteome analysis.

In vitro kinase assays

Assays were done essentially as described (Bancerek et al., 2013). Additional details provided in Supplemental Experimental Procedures.

Z.C.P. would like to thank P. Kovarik for STAT1 antibodies, J. Liddle and S. Stuart for assistance and phospho-ERK, and C. Poss and R. Burgundy for useful discussions. We thank J. Balsbaugh at the Mass Spectrometry and Proteomics core facility at UC-Boulder. Expression of CDK8 module was completed at the Tissue Culture Shared Resource at the UC Cancer Center, supported by P30-CA046934. RNA-seq was carried out at the Genomics Shared Resource at the UC Cancer Center, supported by P30-CA046934. This work was supported by R01 CA170741 (DJT), R21 CA175448 (DJT and WMO), F31 CA180419 (ZCP) and T32 GM08759 (ZCP).