The Mediator CDK8-Cyclin C complex modulates vein patterning in Drosophila by stimulating Mad-dependent transcription

Dysregulations of CDK8 and its regulatory partner CycC, two subunits of the conserved Mediator complex, have been linked to diverse human diseases such as cancer, thus it is essential to understand the regulatory network mobilizing the CDK8-CycC complex in both normal development and tumorigenesis. To identify upstream regulators or downstream effectors of CDK8, we performed a dominant modifier genetic screen in Drosophila based on the defects in vein patterning caused by specific depletion or overexpression of CDK8 or CycC in wing imaginal discs. We identified 26 genomic loci whose haploinsufficiency can modify these CDK8-specific phenotypes. Further analysis of two deficiency lines and mutant alleles led us to identify interactions between CDK8-CycC and the components of the Decapentaplegic (Dpp, the Drosophila homolog of TGFβ) signaling pathway. We observed that CDK8-CycC positively regulates transcription activated by Mad (Mothers against dpp), the primary transcription factor downstream of the Dpp/TGFβ signaling pathway. CDK8 can directly interact with Mad in vitro through the linker region between the DNA-binding MH1 (Mad homology 1) domain and the carboxy terminal MH2 transactivation domain. Besides CDK8 and CycC, further analyses of other subunits of the Mediator complex have revealed six additional Mediator subunits that are required for Mad-dependent transcription in the wing discs, including Med12, Med13, Med15, Med23, Med24, and Med31. Furthermore, CDK9 and Yorkie also positively regulate Mad-dependent gene expression in vivo. These results suggest that the Mediator complex may coordinate with other transcription cofactors in regulating Mad-dependent transcription during the wing vein patterning in Drosophila. Significance CDK8 is a conserved subunit of the transcription cofactor Mediator complex that bridges transcription factors with RNA Polymerase II in eukaryotes. Here we explore the role of CDK8 in Drosophila by performing a dominant modifier genetic screen based on vein patterning defects caused by alteration of CDK8-specific activities. We show that components of the Dpp/TGFβ signaling pathway genetically interact with CDK8; CDK8 positively regulates gene expression activated by Mad, the key transcription factor downstream of Dpp/TGFβ signaling, by directly interacting with the linker region of Mad protein. Given the fundamental roles of Dpp/TGFβ signaling in regulating development and its misregulation in various diseases, understanding how Mad/Smad interacts the Mediator complex may have broad implications in understanding and treating these diseases.

nubbin-Gal4 line (nub-Gal4), which is specifically expressed in the wing pouch area of the wing imaginal discs (31), results in the formation of ectopic veins in the intervein region, especially around L2 and L5 (Fig. 1B). Similar phenotypes were observed with the depletion of CycC (Fig.   1C), or both CDK8 and CycC (Fig. 1D). In contrast, overexpression of wild-type CDK8 (UAS-Cdk8 + ) disrupts the L3 and L4 veins, as well as the crossveins (Fig. 1E), opposite to the phenotypes caused by depleting CDK8, or CycC, or both. However, overexpression of a kinasedead (KD) CDK8 form (UAS-Cdk8 KD ) using the same approach does not affect the vein patterns ( Fig. 1F), suggesting that the effects of CDK8 on vein phenotypes are dependent on the kinase activity of CDK8. These observations show that CDK8 and CycC are involved in regulating the vein patterning in Drosophila.

Validation of the depletion or overexpression phenotype specificity:
To verify the specificity of these phenotypes, we recombined the nub-Gal4 line with the CDK8-RNAi, CycC-RNAi, or CDK8-overexpression lines, and then tested whether these vein phenotypes could be dominantly modified by cdk8 K185 , a null allele of cdk8 (32). As shown in Fig. S1A, reducing CDK8 by half in 'cdk8 K185 /+' heterozygous background suppresses the vein defects caused by CDK8 overexpression. However, heterozygosity of cdk8 K185 does not further enhance the vein phenotype caused by CDK8-RNAi (Fig. S1B), indicating that RNAi of CDK8 may have depleted most of the CDK8 protein pool.
To further validate the specificity of the CDK8-directed phenotypes at the cellular level, we analyzed the protein levels of CDK8 and CycC in wing discs at the third instar wandering stage by immunostaining with CDK8 or CycC specific antibodies. For this, we used the apterous-Gal4 (ap-Gal4) line, which is specifically expressed within the dorsal compartment of the wing discs ( Fig. 1G) (33), allowing us to use the ventral compartment of the same discs as the internal control. Normally, both the CDK8 (Fig. 1H) and CycC (Fig. 1I) proteins are uniformly distributed in the nuclei of all wing disc cells. Tissue-specific depletion of CDK8 (Fig.   2J), or CycC (Fig. 2K), or both (Figs. S1C, S1D) using the ap-Gal4 line led to significantly reduced protein levels for CDK8 or CycC in the dorsal compartment. In contrast, overexpression of either wild-type (Fig. 1L) or kinase-dead (Fig. S1E) CDK8 driven by ap-Gal4 specifically increased the levels of CDK8 protein in the dorsal compartment. Taken together, these genetic and cell biological analyses validate the specificity of both the antibodies and transgenic lines, demonstrating that these vein phenotypes are caused by specific gain or reduction of CDK8 activity in vivo.
Identification of deficiency lines that can dominantly modify the vein phenotypes caused by varying CDK8: Based on these CDK8-specific vein phenotypes, we performed a dominant modifier genetic screen to identify gene products that can functionally interact with CDK8 in vivo. This approach has been successfully employed to reveal the regulatory networks for proteins of interest in Drosophila (34). The approach posits that if a protein interacts with CDK8-CycC in vivo in defining the vein patterns, then reducing its level by half may either enhance or suppress the sensitized vein phenotypes caused by specific alteration of the CDK8 activities. Accordingly, we can survey through the fly genome to search for factors that interact with CDK8-CycC by single genetic crosses.
To facilitate this screen approach, we generated three stocks with the following genotypes: "w 1118 ; nub-Gal4; UAS-Cdk8-RNAi" (designated as "nub>Cdk8-i" for simplicity), "w 1118 ; nub-Gal4; UAS-CycC-RNAi" ("nub>CycC-i"), and "w 1118 ; nub-Gal4, UAS-Cdk8 + /CyO" ("nub>Cdk8 + "). We then conducted the primary screen by crossing these three lines in parallel with a collection of 490 deficiency (Df) lines (Table S1) From these screens, we identified 57 suppressor and 90 enhancer Df lines for the CDK8-RNAi phenotype, and 62 suppressor and 98 enhancer Df lines for the CycC-RNAi phenotype. In addition, we identified 63 enhancer and 98 suppressor Df lines for the CDK8-overexpression phenotype (Fig. 2G, 2H). The results for all of these Df lines are summarized in Table S1. Of these dominant modifier Df lines, four of them suppressed the CDK8-RNAi and CycC-RNAi phenotypes but enhance the CDK8-overexpression phenotype (Fig. 2G, Table S2), while 22 of them enhance the CDK8-RNAi and CycC-RNAi phenotypes but suppress the CDK8-overexpression phenotype (Fig. 2H, Table S2). To further validate this genetic approach, we generated a transgenic line that allowed us to deplete both CDK8 and CycC simultaneously ("w 1118 ; nub-Gal4; UAS-Cdk8-RNAi, CycC-RNAi", referred to as "nub>Cdk8-i CycC-i") with nub-Gal4, and observed the identical phenotypes to the ones caused by depleting either Cdk8 or CycC alone (Fig. 1D). With the exception of one Df line, the rest of these 25 Df lines have consistently modified the ectopic vein phenotype caused by depletion of both CDK8 and CycC: four of the Df lines behaved as suppressors and 21 of them as enhancers (Table S2). These results show that the CDK8-specific vein phenotypes are modifiable and can be used to identify factors that functionally interact with CDK8 in vivo.
Identification of Dad as an enhancer of the nub>Cdk8i and nub>CycCi phenotypes but a suppressor of the Cdk8-overexpression phenotype: To identify the specific genes uncovered by these dominant modifier Df lines, we analyzed these 25 genome regions with partial overlapping Df lines (Table S2). Interestingly, two partially overlapping Df lines, Df(3R)BSC748 and Df(3R)Exel6176, enhanced the CDK8-RNAi and CycC-RNAi phenotypes, but suppressed the CDK8-overexpression phenotype (Fig. 3A). The overlapping region uncovers one specific gene, Dad (Daughter against Dpp), encoding the Drosophila homolog of Smad6/7 (Fig. 3A).
Thus to test whether Dad is the specific gene that accounts for the modification of the CDK8specific phenotypes by these two Df lines, we performed similar genetic tests with a mutant allele of Dad, Dad MI04922 . Indeed, Dad MI04922 dominantly enhanced the CDK8-RNAi (Fig. 3B), CycC-RNAi (Fig. 3C), or CDK8-RNAi plus CycC-RNAi ( Fig. 3D) phenotypes, but suppressed the CDK8-overexpression phenotype (Fig. 3E). These effects on the CDK8-specific vein phenotypes are similar to those observed for Df(3R)BSC748 and Df(3R)Exel6176, suggesting that Dad is the specific gene that genetically interacts with CDK8 in vivo.

Mutants of multiple components of the Dpp signaling pathway genetically interact with
CDK8-CycC: The protein Dad functions as an inhibitory Smad in the Dpp/TGFβ signaling pathway, which plays critical roles in regulating cell proliferation and differentiation during the development of metazoans (35-40). During the development of the wing discs, Dpp spreads from the anterior-posterior boundary to the anterior half and posterior halves (35-37). Upon binding of the Dpp ligand to the Tkv-Punt receptor complex on the cell membrane, the TGFβ type II receptor Punt phosphorylates and activates the type I receptor Tkv. This results in the phosphorylation of Mad by Tkv at its C-terminal SSXS motif, known as the phospho-Mad protein or pMad. Medea, the unique co-Smad protein in Drosophila, associates with pMad in the cytoplasm, and then this heteromeric Smad complex translocates into the nucleus and regulates the expression of its target genes (39, 41-43).
The genetic interactions between CDK8-CycC and Dad prompted us to test whether mutant alleles of other components of the Dpp signaling pathway could also genetically interact with CDK8 and CycC. For this, we crossed multiple mutant alleles of these components with the CDK8-CycC depletion or overexpression lines. As summarized in Fig. 3F, mutants of multiple components of the Dpp signaling pathway could either dominantly enhance or suppress the CDK8-specific vein phenotypes. For example, dpp d6 , tkv 7 , Mad k00237 , and Medea 1 all dominantly suppressed the ectopic vein phenotype caused by depletion of CycC, or both CDK8 and CycC; mutant of dpp d6 could also suppress the CDK8-RNAi phenotype; however, tkv 7 , Mad k00237 , and Medea 1 enhance the CDK8-overexpression phenotype (Fig. 3F). Taken together, these genetic interactions suggest that CDK8-CycC may affect vein patterning by modulating Dpp signaling.

CDK8-CycC positively regulates Mad-dependent transcription:
Given that CDK8 and CycC are known subunits of the Mediator complex, which serves as a scaffold complex mediating the interactions between the RNA Pol II basal transcription initiation apparatus and a variety of gene-specific transcription activators (3,7,44), the most parsimonious model to explain the genetic interactions between Dpp signaling and CDK8-CycC is that the CDK8-CycC complex may directly regulate the transcriptional activity of Mad in the nucleus. To test this model, we analyzed the effects of CDK8-CycC depletion on the expression of spalt (sal), a well-characterized direct target gene of Mad involved in vein differentiation (45-47).
Specifically, the expression of sal-lacZ serves as a reporter for the transcriptional activity of Mad, downstream of the Dpp signaling pathway (48).
Because the expression of sal-lacZ is symmetric along the dorsal-ventral boundary of the wing pouch area of the wing discs (Fig. 4A), we tested whether specific depletion of CDK8 or CycC within the dorsal compartment of the wing discs could affect the transcriptional activity of Mad by detecting the transcription level of sal using an anti-β-galactosidase (anti-β-Gal) antibody. For this, we depleted CDK8, CycC, or both, using the ap-Gal4 driver, and then  (14,15,30,42,49), but whether and how CDK8 interacts with Smads remains unknown. To determine whether CDK8 directly interacts with Mad, we performed a GSTpulldown assay. As shown in Fig. 4E, purified GST-CDK8 can directly bind with His-tagged full length Mad (Mad-FL, AA1-455) expressed in E. coli. We then further mapped the specific domain of Mad that interacts with CDK8 using His-tagged fragments of the Mad protein (see Materials and Methods for details). We have observed that the "Mad-N2" fragment (AA1-230) and the "Mad-C2" fragment (AA151-455), but not the "Mad-N1" fragment (AA1-150) or the "Mad-C1" fragment (AA231-455), can directly interact with CDK8 (Fig. 4E). We validated the interaction between CDK8 and the linker region using the yeast two-hybrid (Y2H) assay: the "Mad-N2" fragment, but not the "Mad-N1" fragment, as the bait can interact with full length CDK8 as the prey (Fig. 4F). It is not feasible to use this two-hybrid approach test Mad-FL or Mad-C1/C2 fragments as baits, since they auto-activate as the baits; while using full-length CDK8 as the bait is also able to auto-activate (Fig. S4). Taken together, these results suggest that CDK8 interacts directly with part of the linker region of Mad protein (AA151-230). Implications of these physical interactions are further discussed below.  MAPKs (mitogen-activated protein kinases) such as ERK and ERK2 (extracellular signalregulated kinases), have been implicated to phosphorylate and regulate the transcriptional activity of Smads (14,15,49,52). The four phosphorylation sites (Ser/Thr residues) within the linker region of Smads appear to be conserved from Drosophila to mammals (Fig. 6A). The phosphorylation of Smads within the linker region may facilitate the subsequent binding with transcription co-factors, such as YAP (Yes-associated protein) (14). However, it is still unclear whether all of these kinases regulate the Smads activity in vivo, and with the exception of YAP (Yorkie or Yki, in Drosophila), it is also unclear whether these regulatory mechanisms are conserved during evolution.
To validate the relevance of these kinases in regulating Mad-dependent gene expression, we depleted the Drosophila orthologs of CDK7, CDK9, Shaggy (Sgg, the GSK3 homolog in

Discussion
To study the function and regulation of CDK8 in vivo, we have developed a genetic system that yields robust readouts for the CDK8-specific activities in developing Drosophila wings. These genetic tools provide a unique opportunity to perform a dominant modifier genetic screen, which allow us to identify multiple components of the Dpp signaling pathway that can genetically interact with CDK8 and CycC in vivo. Our subsequent genetic and cellular analyses reveal that  (Table S3). Med1 and Med25 are loosely associated to the small Mediator complex in human cell lines (5). A caveat for these negative results is that the RNAi lines may not be effective enough to affect sal-lacZ expression, even though the transgenic RNAi lines for majority of these subunits cam generate phenotypes in the eye, wing or both (Table S3).
Therefore, these results indicate that not all Mediator subunits are required for the Maddependent gene expression in the developing wing discs.

Role of Yki/YAP and different kinases in regulating Mad/Smad-dependent transcription:
Interestingly, Yki/YAP, which can function as a transcriptional co-factor for Mad/Smad, was also reported to associate with several subunits of the Mediator complex to drive transcription.
For example, Med12, Med14, Med23, and Med24 were identified from a YAP IP-mass spectrometry sample from HuCCT1 cells (56). Med23 was also reported to regulate Ykidependent transcription of Diap1 in the Drosophila wing discs (57). In this work, we found that Yki, Med12, Med23, and Med24 were also required for the Mad-dependent transcription of sal-lacZ. Although the exact molecular mechanisms of how Yki interacts with certain Mediator subunits remain unclear, it is plausible that Yki may further strengthen the binding between Mad and Med15 through interactions with other subunits such as Med12, Med23, and Med24.
Based on biochemical analyses of the Smad1 phosphomutants and cell biological analyses using cultured human epidermal keratinocytes (HaCaT cells), several kinases including CDK8, CDK9, and ERK2 were shown to phosphorylate serine residues (S) within the linker region of pSmad1 at S186, S195, S206, and S214, or the equivalent sites in pSmad2/3/5. These modifications were proposed to positively regulate the Smad1-dependent transcriptional activity (14). Of these sites, S206 and S214 are both conserved from Drosophila to humans (Fig. S6). In addition, studies using Xenopus embryos and cultured L cells suggest that MAPKs may phosphorylate the linker region of Smad1 (including S214) and lead to its degradation (52).
Nevertheless, analyses with Drosophila embryos or wing discs indicate that S212 (equivalent to human pSmad1 S214) is phosphorylated by CDK8, and S204 (unique in Drosophila) or S208 (equivalent to human pSmad1 S210) are phosphorylated by Sgg/GSK3 (15). These studies suggest the following model to explain how Smads activate the expression of their target genes and how this process is turned off (Figs. 6A, 6H): after Smads are phosphorylated at their Ctermini and translocate into the nucleus, CDK8 and CDK9 (potentially alsoMAPKs) act as the priming kinases to further phosphorylate pSmads in the linker region at S206 and S214, which may facilitate the interaction between pSmads and transcription cofactors such as YAP, thereby stimulating the expression of Smads target genes. Subsequently, pSmads are further phosphorylated by GSK3 within the linker region at T202 and S210, which may facilitate Smad1/5 binding to E3 ligases such as Smurf1 and Nedd4L, thereby causing the degradation of Smads through the ubiquitin-proteasome pathway (14,15,30,42,49).
This model explains how the transactivation of Smads is coupled to its degradation, similar to other transcriptional activators (58). However, it is rather challenging to determine whether these kinases act redundantly or specifically for different phosphorylation sites, the exact orders of these phosphorylation events, as well as their biological consequences in vivo.
Moreover, it remains unexplored whether these regulatory mechanisms are conserved during evolution. The importance of these issues is highlighted by the critical roles of TGFβ signaling in regulating the normal development of metazoans and the dysregulation of this pathway in a wide variety of human diseases such as cancers (40, 59, 60).
The precise spatiotemporal activation of the Dpp signaling pathway in the wings discs is critical for proper formation of the stereotypical vein patterns in Drosophila (45). This model system provides an ideal opportunity to dissect the dynamic regulation of the Mad-activated gene expression in the nucleus. Indeed, depleting CDK8 in wing discs reduces the expression of the Mad-dependent sal-lacZ reporter, suggesting that CDK8 positively regulates Mad-dependent transcription, which is consistent to the effects of CDK8 on Smad1/5-dependent transcription in mammals (14,61). Depleting CDK8 does not affect the phosphorylation of Mad at its Cterminus as revealed by pMad immunostaining (Fig. S3), as well as the physical interaction between CDK8 and the linker region of Mad, supporting the idea that CDK8 may only affect the subsequent phosphorylation of Mad.
Besides CDK8, depleting CDK9 also decreased the expression of the sal-lacZ reporter, supporting the notion that CDK8 and CDK9 may play non-redundant roles in further phosphorylating pMad in the nucleus. However, we did not observe any effects of depletion of Together with the previous reports (14,15,30,42,49,62), our data support that CDK8 or CDK9 may phosphorylate pMad at the linker region, which may facilitate the binding between Yki and Mad. We speculate that this interaction may synergize the recruitment of the Mediator complex, presumably through at least the interaction between its Med15 subunit and the MH2 domain of Mad. Alternatively, Yki may also facilitate the recruitment of the whole Mediator complex through its interactions with Med12, Med23, and Med24. The synergistic interactions among Mad, Yki, the Mediator complex, and RNA Pol II may be required for the optimal transcriptional activation of the Mad-target genes (Fig. 6H).
One of the important future challenges is to illuminate the dynamic interactions between these factors and diverse protein complexes that couple the transactivation effects of Mad/Smads on gene transcription with their subsequent degradation at the molecular level. Smad3 phosphorylation strongly correlates with Med15 levels in breast and lung cancer tissues, together they potentiate metastasis of breast cancer cells (63). Thus it will be important to test whether similar effects of the additional Mediator subunits that we identified here can be observed in mammalian cells. It will also be interesting to determine whether a partial Mediator complex, To facilitate the dominant modifier genetic screen and the subsequent analyses, we generated the following strains using the standard Drosophila genetics: "w 1118 ; nub-Gal4>UAS-Cdk8 + /CyO" (i.e., "nub>Cdk8 + /CyO" line), "w 1118 ; nub-Gal4; UAS-Cdk8-RNAi" (i.e., "nub>Cdk8-i" line), "w 1118 ; nub-Gal4; UAS-CycC-RNAi" (i.e., "nub>CycC-i" line), "w 1118 ; nub-Gal4; UAS-Cdk8-RNAi CycC-RNAi" (i.e., "nub>Cdk8-i CycC-i" line), and "w 1118 ; ap-Gal4, sal-lacZ/T(2:3)". All deficiency (Df) lines, listed in Table S1 and Table S2, were obtained from the Bloomington Drosophila stock center.
Adult Drosophila wing imaging. The wings from adult females were dissected onto slides, briefly washed using isopropanol, and then mounted in 50% Canada balsam in isopropanol.
Images were taken under 5X objective of a microscope (Leica DM2500) and then processed by Adobe Photoshop CS6 software.