The Mediator CDK8-Cyclin C complex modulates Dpp signaling in Drosophila by stimulating Mad-dependent transcription

Dysregulation of CDK8 (Cyclin-Dependent Kinase 8) and its regulatory partner CycC (Cyclin C), two subunits of the conserved Mediator (MED) complex, have been linked to diverse human diseases such as cancer. Thus, it is essential to understand the regulatory network modulating 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 developing wing imaginal discs. We identified 26 genomic loci whose haploinsufficiency can modify these CDK8- or CycC-specific phenotypes. Further analysis of two overlapping deficiency lines and mutant alleles led us to identify genetic interactions between the CDK8-CycC pair and the components of the Decapentaplegic (Dpp, the Drosophila homolog of TGFβ, or Transforming Growth Factor-β) 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 (Mad homology 2) transactivation domain. Besides CDK8 and CycC, further analyses of other subunits of the MED complex have revealed six additional subunits that are required for Mad-dependent transcription in the wing discs: Med12, Med13, Med15, Med23, Med24, and Med31. Furthermore, our analyses confirmed the positive roles of CDK9 and Yorkie in regulating Mad-dependent gene expression in vivo. These results suggest that CDK8 and CycC, together with a few other subunits of the MED complex, may coordinate with other transcription cofactors in regulating Mad-dependent transcription during wing development in Drosophila.


Introduction
Composed of up to 30 conserved subunits, the Mediator complex plays critical roles in modulating RNA polymerase II (Pol II)-dependent gene expression by functioning as a molecular bridge linking transcriptional activators and the general transcription machinery in almost all eukaryotes [1][2][3][4][5]. Biochemical purification of the human Mediator complex has revealed the Cyclin-Dependent Kinase 8 (CDK8) module, composed of CDK8 (or its paralogue CDK19, also known as CDK8L), CycC, Med12 (or Med12L), and Med13 (or Med13L), and the small Mediator complex, composed of 26 subunits that are divided into the head, middle, and tail modules [6][7][8][9]. CDK8 is the only Mediator subunit with enzymatic activities. The CDK8 kinase module (CKM) has been proposed to function in two modes. First, it can reversibly bind with the small Mediator complex to form the large Mediator complex, thereby physically blocking the interaction between the small Mediator complex and the general transcription machinery (notably with RNA Pol II itself). Second, CDK8 can function as a kinase that phosphorylates different substrates, particularly transcriptional activators such as E2F1 [10,11], N-ICD (intracellular domain of Notch) [12], p53 [13], Smad proteins [14,15], SREBP (sterol regulatory element-binding protein) [16], and STAT1 (signal transducer and activator of transcription 1) [17]. These characterized functions of CDK8 highlight fundamental roles of the CKM in regulating transcription.
Besides its roles in specific developmental and physiological contexts, the CKM subunits are dysregulated in a variety of human diseases, such as cancers [18][19][20][21][22]. For example, CDK8 has been reported to act as an oncoprotein in melanoma and colorectal cancers [10,23,24]. Moreover, CDK8 and CDK19 are overexpressed in invasive ductal carcinomas, correlating with shorter relapse-free survival in breast cancer [25]. Gain or amplification of CDK8 activity is sufficient in driving tumorigenesis in colorectal and pancreatic cancers in human, as well as in skin cancer in fish [14,23,[26][27][28]. Because of these discoveries, there is a considerable interest in developing drugs targeting the CDK8 kinase for cancer treatment in recent years [29,30]. However, exactly how CDK8 dysregulation contributes to tumorigenesis remains poorly understood. Thus it is essential to reveal the function and regulation of CDK8 activity in different developmental, physiological, and pathological processes.
The major bottleneck for addressing these critical gaps in our knowledge is the lack of in vivo readouts for CDK8-specific activities in metazoans. We overcame this challenge by generating tissue-specific phenotypes caused by varying CDK8 activities in Drosophila. After validating the specificity of these phenotypes using genetic, molecular, and cell biological approaches, we performed a dominant modifier genetic screen to identify factors that interact with CDK8 in vivo based on these unique readouts for CDK8-specific activities. From the screen, we identified Dad (Daughters against dpp), which encodes an inhibitory Smad in the Dpp (Decapentaplegic)/TGFβ (Transforming Growth Factor-β) signaling pathway, as well as additional components of the Dpp signaling pathway including dpp, tkv (thickveins, encoding the type I receptor for Dpp), Mad (Mothers against dpp) and Medea (encoding the Smad1/5 and Smad4 homologs, respectively) in Drosophila. Consistent with the previous biochemical analyses suggesting that CDK8 may phosphorylate Drosophila Mad or human Smad1 [14,15,31], thereby regulating their transcriptional activities [14,15,31], our results have validated and further advanced our understanding of this conserved regulatory mechanism in vivo. Furthermore, our analyses have revealed additional Mediator subunits and protein kinases involved in regulating the Mad/Smad-dependent transcription. These results, together with previous studies, suggest that concerted recruitment of the Mediator complexes and other cofactors play a pivotal role in regulating Mad/Smad-dependent gene expression, a critical process for TGFβ signaling to function in a variety of biological and pathological contexts.

Wing vein patterning defects caused by varying the levels of CDK8, CycC, or both
To study the function and regulation of CDK8 and CycC in vivo, we have generated transgenic lines to either deplete them by RNAi (RNA interference) or conditionally overexpress the wild-type CDK8 kinase using the Gal4-UAS system [32,33]. Normal Drosophila wings display stereotypical vein patterns, consisting of six longitudinal veins, dubbed L1 to L6, and two crossveins, the anterior crossvein and the posterior crossvein ( Fig 1A). Knocking down of CDK8 using the nub-Gal4 (nubbin-Gal4) line (see Materials and Methods for details), which is specifically expressed in the wing pouch area of the wing imaginal discs [34], 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 vein, the L4 vein, and the crossveins (Fig 1E), opposite to the phenotypes caused by depleting CDK8, CycC, or both. However, overexpression of a kinase-dead (KD) form of CDK8 (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.
Interestingly, depletion of CDK8 (Fig 1B), CycC (Fig 1C), or both ( Fig 1D) increase the size of the wings, correlating to a significant increase of total cell numbers but a reduction of cell sizes (S1 Fig). In contrast, overexpression of wild-type CDK8 reduces the size of wings and total cell numbers, but no obvious effects on cell size (Fig 1E and S1 Fig). The effects of CDK8 on wing size can also be visualized using ap-Gal4 (apterous-Gal4), which is specifically expressed within the dorsal compartment of the wing discs (Fig 2A) [35]. Ap-Gal4-induced depletion of CDK8 and CycC caused the adult wing to curl downwards (S2C Fig), indicating the overgrowth of the dorsal compartment compared to the ventral compartment; while overexpression of CDK8 led to the adult wing to curl upwards (S2D Fig), suggesting reduced growth of the dorsal compartment. We have previously reported that CDK8 inhibits the transcriptional activity of E2F1, a key transcription factor that controls the expression of factors  Representative confocal images of the wing pouch area of a L3 wandering larval wing disc: (A) ap-Gal4/UAS-2XGFP with DAPI (blue) and GFP (green); (B) ap-Gal4/+ with anti-CDK8 (red) staining; (C) ap-Gal4/+ with anti-CycC (red) staining; (D) ap-Gal4/+; UAS-Cdk8-RNAi/+ with anti-CDK8 (red) staining; (E) ap-Gal4/+; UAS-CycC-RNAi/+ with anti-CycC (red) staining; (F) ap-Gal4/+; UAS-Cdk8-RNAi CycC-RNAi/+ with anti-CDK8 (red) staining; (G) ap-Gal4/ +; UAS-Cdk8-RNAi CycC-RNAi/+ with anti-CycC (red) staining; (H) ap-Gal4/UAS-Cdk8 + with anti-CDK8 (red) staining; and (I) ap-Gal4/UAS-Cdk8 KD with anti-CDK8 (red) staining. Note that the gain for confocal imaging in H and I is lower than the others to avoid over saturation of the signals. At least five wing discs were examined for each genotype. The dorsal/ventral (D/V) boundary is shown in A, D and H. Scale bar in I: 25μm.
https://doi.org/10.1371/journal.pgen.1008832.g002 required for the G1 to S-phase transition of the cell cycle [10,11]. Thus, the effects of CDK8 levels on wing size and cell numbers are likely through E2F1-regulated cell-cycle progression.

Validation of the specificity of the vein defects caused by depletion or overexpression of CDK8-CycC
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 [36]. As shown in Fig 1G, reducing CDK8 by half in a 'cdk8 K185 /+' heterozygous background suppresses the vein defects caused by CDK8 overexpression. However, heterozygosity of cdk8 K185 does not obviously enhance the vein phenotype caused by CDK8-RNAi (Fig 1H), indicating that the 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 larval stage by immunostaining with CDK8 or CycC specific antibodies. Normally, both the CDK8 ( Fig 2B) and CycC ( Fig 2C) proteins are uniformly distributed in the nuclei of all wing disc cells. Depletion of CDK8 (Fig 2D), CycC (Fig 2E), or both ( Fig 2F and 2G) using the ap-Gal4 line significantly reduced CDK8 or CycC proteins in the dorsal compartment. The ventral compartment of the same discs serves as the internal control. In contrast, overexpression of either wild-type ( Fig 2H) or kinase-dead ( Fig 2I) CDK8 using ap-Gal4 specifically increased the levels of CDK8 protein in the dorsal compartment. Taken together, these genetic and cell biological analyses have validated the specificity of both the antibodies and transgenic lines, demonstrating that these vein phenotypes are caused by a 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. The idea of using phenotypic modifications to identify multiple genes involved in determining a specific trait or a phenotypic endpoint was initially developed by Calvin B. Bridges, when he analyzed mutant genes that could interact with the eosin mutant in regulating eye color in flies [37]. This genetic modifier approach has been employed to reveal the functional and inter-molecular networks for proteins of interest in Drosophila (for instances, [38][39][40][41][42]), and to provide insights into the phenotypic and genetic variability in mammals [43,44]. The approach posits that if a protein interacts with CDK8-CycC in vivo in defining the wing vein patterns, then reducing its level by half may either enhance or suppress the sensitized wing 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 using single genetic crosses.
We inspected the vein patterns of the F1 females for enhancers and suppressors based on the following criteria: suppressors of the CDK8-or CycC-RNAi phenotypes are expected to display fewer or no ectopic veins (e.g., Fig 3A and 3C), while enhancers of the CDK8-or CycC-RNAi phenotypes show more or longer ectopic veins (e.g., Fig 3B and 3D). To score the strength of the modifications, we define strong suppressors as the Df lines that eliminate all of the ectopic veins, while the Df lines that only shorten the length of the ectopic veins are scored as weak suppressors. Similarly, we define strong enhancers to cause more or longer ectopic veins than CDK8-or CycC-RNAi phenotypes, while the Df lines causing less severe vein defects are designated as the weak enhancers. Conversely, the strong suppressors of the CDK8-overexpression phenotype are expected to have vein patterns similar to those of wildtype wings (particularly the L3/L4; e.g., Fig 3E, compared to the control shown in Fig 1E). If the Df lines only partially restore the missing veins, then they are scored as the weak suppressors. In contrast, the strong enhancers of the CDK8-overexpression phenotype are defined by further disrupting the vein patterns, with the entire L3 or L4 missing, often accompanied with strong disruption on other veins (e.g., Fig 3F); while the weaker enhancers further disrupted the vein defects compared to the CDK8-overexpression phenotype, but less severe than the strong enhancers.
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 3G and 3H). The results for all of these Df lines are summarized in S1 Table. Of these dominant modifier Df lines, four of them suppressed the CDK8-RNAi and CycC-RNAi phenotypes but enhance the CDK8-overexpression phenotype ( Fig 3G, Table 1), while 22 of them enhance the CDK8-RNAi and CycC-RNAi phenotypes but suppress the CDK8-overexpression phenotype ( Fig 3H, Table 1). To further validate this genetic approach, we generated a transgenic line that allowed us to simultaneously deplete CDK8 and CycC ("w 1118 ; nub-Gal4; UAS-Cdk8-RNAi, CycC-RNAi", referred to as "nub>Cdk8-i CycC-i") with nub-Gal4, and observed 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 1). These results show that the CDK8-specific vein phenotypes are modifiable and can be used to identify factors that functionally interact with CDK8-CycC in vivo.

Identification of Dad as an enhancer of the nub>Cdk8-i and nub>CycC-i phenotypes but a suppressor of the Cdk8-overexpression phenotype
To identify the specific genes uncovered by these dominant modifier Df lines, we analyzed these 26 genome regions with partial overlapping Df lines (Table 1). 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 3B, 3D and 3E; Table 1). The overlapping region uncovers one specific gene, Dad (Daughter against Dpp), encoding the Drosophila homolog of Smad6/7 ( Fig 4A). Thus, to test whether Dad is the specific gene that accounts for the modification of the CDK8-specific phenotypes by these two Df lines, we performed similar genetic tests with two mutant alleles of Dad: Dad MI04922 , a MiMIC (Minos Mediated Integration Cassette) insertion in an intron of the Dad gene [47], and Dad j1E4 , an insertion of the P{lacW} element in an intron of the Dad gene [48]. Indeed, both Dad mutant alleles dominantly enhanced the CDK8-RNAi (Fig 4B), CycC-RNAi ( Fig  4C), and CDK8-RNAi plus CycC-RNAi ( Fig 4D) phenotypes, but suppressed the CDK8-overexpression phenotype ( Fig 4E, Table 2). These effects on the CDK8-specific vein phenotypes

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 [49][50][51][52][53][54]. During the development of the wing discs, Dpp spreads from the anterior-posterior boundary to the anterior and posterior halves [49][50][51]55]. Upon the 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 [53,[55][56][57].
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 Table 2, mutants of multiple components of the Dpp signaling pathway could either dominantly enhance or suppress the CDK8-specific vein phenotypes. For instance, dpp d6 , dpp hr92 , dpp S11 , tkv 7 , tkv k16713 (Fig 4F), Mad 1-2 , Mad 12 (Fig 4H), Mad 8-2 , Mad k00237 , Mad KG00581 , Medea 1 , and Medea 13 ( Fig  4J) all dominantly suppress the ectopic vein phenotype caused by depletion of CDK8, CycC, or both CDK8 and CycC (Table 2). However, tkv 7 , tkv k16713 (Fig 4G), Mad k00237 , Mad 12 (Fig 4I), Mad KG00581 , Medea 1 , and Medea 13 (Fig 4K) enhance the CDK8-overexpression phenotype ( Table 2). Testing additional mutant alleles of these genes have revealed that most of them can also dominantly modify the CDK8-specific phenotypes ( Table 2). Dpp is activated in a specific pattern in the middle part of the wing pouch area, while the nub-Gal4 display a well-characterized pattern in the entire wing pouch area. These two patterns differ, arguing against the possibility that Dpp signaling may affect nub-Gal4 expression pattern. In addition, reducing Mad or Dad by half has little effects on the expression of a UAS-RFP reporter driven by nub-Gal4 (S3 Fig), suggesting that the expression and activity of nub-Gal4 are not affected by Dpp signaling. 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 machinery and a number of gene-specific transcription activators [3,7,58]. Thus, the simplest 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 salm (spalt major), a well-characterized direct target gene of Mad involved in vein differentiation [59][60][61][62]. The sal-lacZ (P{PZ}salm 03602 ) is a enhancer trap line derived from an insertion of a P {PZ} element in the promoter region of the salm gene [63,64], and the expression of sal-lacZ can serve as a reporter for the transcriptional activity of Mad [65].
Because the expression of sal-lacZ is symmetric along the dorsal-ventral boundary of the wing pouch area of the wing discs (Fig 5A), 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 genes of interest using the ap-Gal4 driver, and then compared the β-Gal expression between the dorsal and ventral compartments. As expected, depleting Mad with two transgenic RNAi lines (BL-43183 ( Fig 5B)   To further validate the effects of CDK8-CycC depletion on Mad-activated gene expression, we analyzed the expression of the quadrant enhancer (QE) of the selector gene vestigial (vgQE-lacZ) in wing discs. Similar to Sal-lacZ reporter, vgQE-lacZ also displays a symmetric expression pattern along the D-V boundary in the wing pouch ( Fig 6A) [66,67]. As expected, depleting Mad (BL-31315) using ap-Gal4 driver significantly reduced the expression of vgQE-lacZ in the dorsal compartment ( Fig 6B). Although depleting CDK8 alone only marginally reduced the vgQE-lacZ expression in the dorsal compartment (Fig 6C), a more obvious effect was observed with the depletion of CycC (Fig 6D), and a stronger reduction of the reporter expression was detected with the depletion of both CDK8 and CycC (Fig 6E) using the same approach. We note that the interpretation of the data presented in Fig 6 is compounded by the fact that the transcription of the vg in different compartments of wing discs is controlled by Wingless (Wg) and Dpp signaling, as well as a feed-forward regulation by Vg itself [66,67]. Nevertheless, the most parsimonious model to explain the observations based on Sal-lacZ and vgQE-lacZ reporters is that CDK8-CycC positively regulates Mad-dependent transcription.
One caveat of these analyses is that the CKM could affect ap-Gal4 activities. As shown in S7B Fig, we observed that depleting CDK8 and CycC reduces the ap-Gal4-dependent

Direct interactions between CDK8 and Mad
Since Mad phosphorylation at its C-terminus (pMad) by the Tkv-Punt receptor complex marks the activation of Mad, we tested whether CDK8 affects the pMad level. For this, we depleted CDK8, CycC, or both, with the ap-Gal4 line, and then detected the levels of the activated Mad with an anti-pMad antibody. In the wing pouch area of the control discs, the pMad protein is symmetrically distributed along the dorsal-ventral boundary (S8A Fig). However, depletion of CDK8-CycC did not affect pMad levels when comparing the dorsal compartment with the ventral compartment (S8B- S8D Fig), suggesting that CDK8-CycC does not affect the phosphorylation of Mad at its carboxy terminus in the cytoplasm. These results support the idea that the CDK8-CycC complex directly regulates the transcriptional activity of Mad in the nucleus.
R-Smads are characterized by a highly conserved amino-terminal MH1 (Mad homology 1) domain that binds to DNA and a C-terminal MH2 (Mad homology 2) domain that harbors the transactivation activity, separated by a serine-and proline-rich linker region (Fig 5H) [68]. It was previously reported that CDK8 and a few other kinases (see below) may phosphorylate Smad proteins in both Drosophila and mammalian cells [14,15,31,55,68], but whether and how CDK8 interacts with Smads remain unknown. To determine whether CDK8 directly interacts with Mad, we performed a GST-pulldown assay. As shown in Fig 5H, purified GST-CDK8 can directly bind with His-tagged full length Mad (Mad-FL, AA1-455) expressed in E. coli. We then mapped the specific domain of Mad that interacts with CDK8 using His-tagged fragments of the Mad protein. We 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 interact directly with CDK8 ( Fig 5H). 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  5I). It is not feasible to use this Y2H approach test with Mad-FL or Mad-C1/C2 fragments as bait, since they auto-activate as the baits; while the full-length CDK8 can also auto-activate as the bait (S9 Fig). Taken together, these results suggest that CDK8 directly interacts with part of the linker region of Mad protein (AA151-230). Implications of these physical interactions are discussed below.

Involvement of additional Mediator complex subunits in regulating the Mad/Smad-dependent transcription
The Med15/ARC105 subunit of the Mediator complex has been previously shown to directly interact with the transactivation MH2 domain of Smad2/3, thereby mediating the Smad2/ 3-Smad4-dependent transcription in Xenopus [69], and Med15 is required for the transcription of Dpp target genes in Drosophila [70]. However, whether other Mediator subunits are involved in regulating the Mad/Smad-dependent transcription remains unknown. To address this question, we depleted individual subunits of the Mediator complex upon conditional expression of interfering RNAs with ap-Gal4, and then analyzed the expression of the sal-lacZ reporter. Of the 30 Mediator subunits tested (Table 3), we have observed that depletion of six additional Mediator subunits, Med12 (Fig 7B), Med13 (Fig 7C), Med15 (Fig 7D), Med23 ( Fig  7E), Med24 (Fig 7F), and Med31 (Fig 7G), by ap-Gal4 significantly reduced the expression of sal-lacZ in cells of the dorsal compartment compared with the cells in the ventral compartment of the same wing discs (Fig 7J); similar to depletion of CDK8 or CycC (Fig 5). The effects of these six Mediator subunits were further validated using the vgQE-lacZ reporter: their depletion using ap-Gal4 also reduces the vgQE-lacZ expression in the dorsal compartment (Fig 6F-6K). These results suggest that these Mediator subunits are required for the Mad-activated gene expression. However, RNAi depletion of the remaining 15 Mediator subunits using ap-Gal4 driver did not significantly affect sal-lacZ expression (Table 3), as β-Gal expression remained symmetric along the dorsal-ventral boundary as exemplified for depletion of Med1 ( Fig 7A and 7J) and Med25 (Fig 7H) on sal-lacZ expression. Similarly, depletion of Med30 using ap-Gal4 does not obviously affect the expression of sal-lacZ and vgQE-lacZ reporters, which remains symmetric along the dorsal-ventral boundary ( Fig 6L, Table 3). Furthermore, depleting the remaining Mediator subunits, including Med7 (Fig 7I), Med8 (S10A

CDK9 and Yorkie also positively regulate the Mad/Smad-dependent transcription
Besides CDK8, several other kinases, such as CDK7, CDK9, GSK3 (Glycogen synthase kinase 3), and MAPKs (mitogen-activated protein kinases) such as ERK (extracellular signal-regulated kinase) and ERK2, have been implicated to phosphorylate and regulate the transcriptional activity of Smads [14,15,68,71] (Fig 8A, see below). The four phosphorylation sites (Ser or Thr residues) within the linker region of Smads appear to be conserved from Drosophila to mammals (Fig 8B; see Discussion). 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 Smads activity in vivo. 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 Drosophila), Rolled, and dERK2 (MAPK/ERK homologs in Drosophila), in the dorsal compartment of wing discs (using ap-Gal4 as above), and then analyzed sal-lacZ expression in the wing pouch. As expected for a positive role of Yki in regulated Mad-dependent transcription [14], depletion of Yki in the dorsal cells significantly reduced the expression of sal-lacZ compared to the cells in the ventral compartment of the same discs (Fig 8C and 8J). Using the same approach, we have observed that depleting CDK9 (Fig 8D and 8J) and its partner CycT (Cyclin T, Fig 8E and 8J; [72]) also reduced sal-lacZ expression. These observations suggest that both Yki and CDK9-CycT are required for Mad/Smad-dependent transcription in Drosophila, which is consistent to the previous reports [14,31]. However, depletion of CDK7 (Fig 8F and Fig 8J) or Drosophila MAPK homologs, either Rolled (Fig 8G) or dERK2 (Fig 8H), did not affect the expression of sal-lacZ. Although depletion of Sgg increased the size of the dorsal compartment, the intensity of anti-β-Gal staining remained similar to the ventral compartment (Fig 8I and 8J). We note that depleting CDK9 (S7D Fig), Med12 (S7E Fig), or Med13 (S7F Fig) have no obvious effects on the expression of UAS-GFP reporter, suggesting that their effects on sal-lacZ expression are independent of the Gal4 activity per se. Together with the previous reports [14,68], our in vivo analyses have validated the conserved roles of CDK8-CycC, CDK9-CycT, and Yki/YAP on Mad/Smad-dependent transcription.

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, allowing us to identify multiple components of the Dpp/TGFβ signaling pathway that can genetically interact with the CDK8-CycC complex in vivo. Our subsequent genetic and cellular analyses reveal that CDK8, CycC, and six additional subunits of the Mediator complex, as well as CDK9 and Yki are required for the Mad-dependent transcription in the wing discs. In addition, CDK8 can directly interact with the linker region of Mad. These results have extended the previous biochemical and molecular analyses on how different kinases and transcription cofactors modulate the Mad/Smad-activated gene expression in the nucleus. Further mapping of specific genes uncovered by other deficiency lines may also open up the new directions to advance our understanding of the conserved function and regulation of CDK8 during development.

Multiple subunits of the Mediator complex are required for Mad/Smaddependent transcription
The Mediator complex functions as a molecular bridge between gene-specific transcription factors and the RNA Pol II general transcription apparatus, and diverse transactivators have been shown to interact directly with distinct Mediator subunits [4,[6][7][8][9]73]. However, it is unclear whether all Mediator subunits are required by different transactivators to regulate gene expression, or whether Mediator complexes composed of fewer and different combinations of Mediator subunits exist in differentiated tissues or developmental stages. Gene-specific combinations of the Mediator subunits may be required in different transcription processes, as not all Mediator subunits are simultaneously required for all transactivation process [74]. For instance, ELK1 target gene transcription requires Med23, but lacking Med23 does not functionally affect some other ETS transcription factors, such as Ets1 and Ets2 [75]. Similarly, Med15 is required for the expression of Dpp target genes, but does not appear to affect the expression of EGFR (epidermal growth factor receptor) and Wg targets in Drosophila [70].
It has been previously reported that the Med15 subunit is required for the Smad2/3-Smad4 dependent transcription, as its removal from the Mediator complex abolishes the expression of Smad-target genes and disrupts Smad2/3-regulated dorsal-ventral axis formation in Xenopus embryos [69]. Further biochemical analyses showed that increased Med15 enhances, while its depletion decreases, the transcription of Smad2/3 target genes, and that the Med15 subunit can directly bind to the MH2 domain of Smad2 or Smad3 [69]. In Drosophila, loss or reduction of Med15 reduced the expression of Dpp targets, resulting in smaller wings and disrupted vein patterning (mainly L2) [70]. We also observed that depletion of Med15 or CDK8 reduces the expression of a Mad-target gene. These observations support the idea that CDK8 and Med15 play a conserved and positive role in regulating Mad/Smad-activated gene expression.
Aside from Med15 and CDK8, it remains unclear whether other Mediator subunits are also involved in Mad/Smad-dependent transcription. We identified six additional Mediator subunits that are required for the Mad-dependent transcription, including CycC, Med12, Med13, Med23, Med24, and Med31 (Fig 5, Fig 6, Fig 7 and Table 3). Interestingly, aside from Med23 and Med24 being specific to metazoans, counterparts of the other six subunits are not essential for cell viability in the budding yeast [5]. The similar effects of the four CKM subunits on Mad-activity suggest that they may function together to stimulate Mad-dependent transcription. We note that depletion of seven Mediator subunits, Med7, Med8, Med14, Med16, Med17, Med21, and Med22, severely disrupts the morphology of the wing discs (Fig 7I and  S10 Fig), making it difficult to assay their effects on the transcriptional activity of Mad in vivo. Consistently, all corresponding subunits, except Med16, are critical for cell viability in the budding yeast [5]. In contrast, reducing expression of the 15 remaining subunits of the Drosophila Mediator complex did not significantly alter the expression of a Mad-dependent reporter ( Table 3). Med1 and Med25 are loosely associated to the small Mediator complex in human cell lines [5]. A caveat for these negative results is that depleting these subunits using the existing RNAi lines may not be sufficient to affect sal-lacZ expression, even though the majority of these transgenic RNAi lines can generate severe phenotypes in the eye, wing, or both (Table 3). Further analyses are necessary to validate these negative data in the future. Taken together, our results indicate that not all Mediator subunits are required for the expression of the Mad-target genes that we tested 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. Specifically, Med12, Med14, Med23, and Med24 were identified from a YAP IP-mass spectrometry sample in HuCCT1 cells [76]. Med23 was also reported to regulate Yki-dependent transcription of Diap1 in wing discs [77]. In this work, we found that Yki, Med12, Med23, and Med24 were also required for 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 (Ser, or 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 regulate positively Smad1-dependent transcriptional activity [14]. Of these sites, S206 and S214 are both conserved from Drosophila to humans ( Fig  8B). 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 [71]. Nevertheless, analyses with Drosophila embryos and wing discs indicate that S212 (equivalent to human pSmad1 S214) is phosphorylated by CDK8, while S204 (unique in Drosophila) and S208 (equivalent to human pSmad1 S210) are phosphorylated by Sgg/GSK3 [15]. These studies suggest the following model in explaining how Smads activate the expression of their target genes and how this process is turned off (Fig 8A, Fig 9): after Smads are phosphorylated at their C-termini and translocated into the nucleus, CDK8 and CDK9 (potentially also MAPKs) act as the priming kinases to further phosphorylate pSmads in the linker region at S206 and S214. This may facilitate the interaction between pSmads and transcriptional cofactors such as YAP, stimulating the expression of Smads target genes. Overexpression of Yki in Drosophila wing disc increases the expression of the vgQE-lacZ reporter [14], which validates the role of Yki/YAP in activating Mad/Smad1-dependent gene expression in vivo. Subsequently, pSmads are further phosphorylated by GSK3 within the linker region at T202 and S210, which may facilitate Smad1/5 binding to E3 ubiquitin ligases such as Smurf1 and Nedd4L, causing the degradation of Smads through the ubiquitin-proteasome pathway [14,15,31,55,68].
Although this model (Fig 9) is still rather speculative, it serves as a conceptual framework to explain how transactivation of Smads is coupled to its degradation, similar to other transcriptional activators [78]. It is challenging to determine whether these kinases act redundantly or sequentially 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 role of TGFβ signaling in regulating the normal development of metazoans and the dysregulation of this pathway in a variety of human diseases such as cancers [54,[79][80][81].
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 [59,62]. 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 expression of the Mad-dependent sal-lacZ reporter, suggesting that CDK8 positively regulates Maddependent transcription. This is consistent with the effects of CDK8 on Smad1/5-dependent transcription in mammals [14,82]. Depleting CDK8 does not affect the phosphorylation of Mad at its C-terminus as revealed by pMad immunostaining (S8 Fig), nor does it affect the physical interaction between CDK8 and the linker region of Mad, supporting the idea that CDK8 may only affect subsequent phosphorylation of Mad, presumably within the linker region.
Besides CDK8-CycC, depleting CDK9-CycT also decreases the expression of the sal-lacZ reporter, supporting the notion that CDK8-CycC and CDK9-CycT may play non-redundant roles in further phosphorylating pMad in the nucleus. However, we did not observe any effects of depletion of CDK7 or MAPKs on sal-lacZ expression, suggesting that their role in regulating the transcriptional activity of Smads may not be conserved in Drosophila. Alternatively, the two MAPK/ERK homologs, Rolled and ERK2, may act redundantly in regulating Mad-dependent transcription. Lastly, depleting Sgg/GSK3 in the dorsal compartment of the wing disc increases the size of this compartment, yet the expression level of the sal-lacZ reporter is similar to the ventral compartment. These observations are consistent with previous reports that phosphorylations of Mad/Smad in the linker regions by CDK8-CycC and Sgg/GSK3 regulate the level and range of Mad-dependent gene expression [14,15,31,55,68].
Together with the previous reports [14,15,31,55,68,83], our data support that CDK8-CycC and CDK9-CycT 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 at least through the interaction between its Med15 subunit and the MH2 domain of Mad (Fig 9). 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 9).
One of the challenges is to illustrate 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 [84]. Thus, it will be important to test whether additional Mediator subunits that we identified in Drosophila play similar roles in mammalian cells. It will also be interesting to determine whether a partial Mediator complex, composed of a subset of the Mediator subunits, exists and regulates Mad/Smad-dependent gene expression. Furthermore, detailed biochemical analyses may yield mechanistic insights into how CDK8 and Med15 act in concert in stimulating the Mad/Smad-dependent gene expression.

Potential role of CDK8-CycC in regulating cross-talks among different signaling pathways
Wing pouch-specific alteration of CDK8 activity results in two major phenotypes: disrupted vein patterns and altered size of wing blades. While the effects on wing size and cell numbers can be explained by the role of CDK8 in regulating cell proliferation through E2F1 [10,11], the effects of CDK8 on vein patterning are more complex. The stereotypical wing vein patterns in adult flies are gradually defined by elaborated spatiotemporal interplays among different signaling pathways, including Dpp, EGFR, Hedgehog (Hh), Notch (N), and Wingless (Wg), in the developing wing discs [55,59,60,62]. During the larval and pupal stages, these signaling pathways and their downstream transcriptional targets coordinately control the cell proliferation and differentiation of cell in different parts of the wing disc to form individual veins.
It is noteworthy that varying CDK8 activities has different effects on different veins: gain of CDK8 causes the loss of the L3 and L4 veins, but the vein patterns of L2 and L5 appear thicker and more diffusive; while the ectopic veins caused by reduction of CDK8 are mainly intertwined with the L2 and L5 veins (Fig 1). Our analyses on the genetic interactions between CDK8 and the components of the Dpp signaling pathway led us to discover the role of the Mediator complex in Mad-stimulated transcription of sal. However, there is a gap in our understanding of how reduced expression of sal in wing discs is linked to the vein defects in adult wings. It is known that salm and salr (spalt-related), two members of the spalt gene family that encode zinc-finger transcriptional repressors, function downstream of the Dpp signaling pathway during development of the central part of the wing [85]. Depletion of either salm or salr alone resulted in ectopic vein formation around L2 in adult wings, yet depletion or loss of both salm or salr caused loss of vein phenotype [61,86]. In addition, elimination of L2 in ventral-anterior and ectopic L5 in dorsal-posterior were observed in salm/salr clones at different region of the wing [61]. These observations suggest that the dosage of salm and salr in wing discs does not have a linear relationship with the wing vein patterning at the adult stage.
Interestingly, it is known that the CKM complex regulates the transcriptional activities of the key transcription factors of these pathways, including N-ICD downstream of N signaling [12], Mad/Smad proteins ( [14,15] and this work). In addition, Med12 (Kohtalo, or Kto in Drosophila) and Med13 (Skuld, or Skd in Drosophila) subunits of the CKM interact with Pangolin (the lymphoid-enhancing factor (LEF)/T cell factor (TCF) homolog in Drosophila), the key transcription factor downstream of Wg signaling, through the transcriptional cofactors such as Pygopus, Legless, and Armadillo [87]. In mammalian cells, Med12 is also known to regulate the activities of Gli proteins, the key transcription factors downstream of Hh signaling [88,89]. Furthermore, the Mediator subunit Med23 interacts with ETS (E-twenty six transcription factor) proteins, a family of key transcription factors downstream of the EGFR signaling pathway [75]. However, whether CDK8-CycC also regulates TCF-, ETS-or Gli-dependent transcription is still not understood. Nevertheless, these studies in other biological contexts suggest that the effects of CDK8 on wing vein patterning are not likely solely through the Dpp signaling pathway. Therefore, we speculate that the potential interactions between CDK8 and the aforementioned signaling pathways may contribute to these differential effects on distinct veins. Further analyses of these cross-talks, as well as further mapping of other Df lines that modify the CDK8-specific vein phenotypes, may yield the insights into the molecular and dynamic mechanisms underlying these vein phenotypes.

Identification of novel genomic loci that genetically interact with CDK8 in vivo
To understand how dysregulated CDK8-CycC contributes to a variety of human cancers, it is essential to elucidate the function and regulation of CDK8 in vivo. Given that the CDK8-CycC pair and other subunits of the Mediator complex are conserved in almost all eukaryotes [5], Drosophila serves as an ideal model system to identify both the upstream regulators and the downstream effectors of CDK8 activity in vivo. Our dominant modifier genetic screen is based on the wing vein phenotypes caused by specific alteration of CDK8 activity in the developing wing disc, which serves as a unique in vivo readout for the CDK8-specific activities in metazoans. This screen led us to identify 26 genomic regions that include loci whose haplo-insufficiency could consistently modify CDK8-CycC depletion or CDK8-overexpression phenotypes. Identification of Dad and genes encoding additional components of the Dpp signaling pathway provides a proof of principle for this approach. Since each of the chromosomal deficiencies uncovers multiple genes, further mapping of the relevant genome regions is expected to identify the specific genetic loci encoding factors that may function either upstream or downstream of CDK8 in vivo. It is hoped that further analyses of the underlying molecular mechanisms in both Drosophila and mammalian systems will advance our understanding of how dysregulation of CDK8 contributes to human diseases, thereby aiding the development of therapeutic approaches.

Fly strains
Flies were raised on a standard cornmeal, molasses and yeast medium, and all genetic crosses were maintained at 25˚C. The UAS-Cdk8 + and UAS-Cdk8 KD lines were generated using the pUASt vector [36]. The construct allowing conditional expression of a kinase-dead CDK8 form (D173A; [90]) was generated through site-specific mutagenesis by double PCR, using the overlap extension method. The UAS-Cdk8-RNAi and UAS-CycC-RNAi lines were generated using the pVALIUM20 vector [91], and the UAS-Cdk8-RNAi CycC-RNAi line was generated using the pNP vector [92]. The vgQE-lacZ line was received from Gary Struhl [66,67].

Adult Drosophila wing imaging
The wings from adult females were dissected onto slides, briefly washed using isopropanol, and then mounted in 50% Canada balsam diluted in isopropanol. Images were taken under 5X objective of a microscope (Leica DM2500) and then processed by Adobe Photoshop CS6 software.
Quantification of anti-β-galactosidase was performed with Nikon NIS software and Microsoft Excel: a single section of the wing discs was selected for the following quantification based on the DAPI channel, which indicates the cell nucleus are on the same focal plat. Three lines around 50μm long, 10-15μm apart, were drawn along the dorsal-ventral boundary. The linescan profile of intensity for each line was calculated along each line (S5A and S5B Fig; genotype: ap-Gal4, sal-lacZ/+; UAS-Cdk8-i/+). The area below the intensity index profile represents the Sal-lacZ expression levels along the line (S5B Fig). To obtain the average intensity of dorsal or ventral compartment, the dorsal or ventral compartment index area was divided by the dorsal or ventral length of the line (S5C Fig). The intensity for three lines was normalized and averaged in dorsal and ventral compartments (S5C Fig, S5D Fig, S2 Table). Following this approach, five wing discs for each genotype were analyzed to quantify the expression of Sal-lacZ in dorsal and ventral compartments, and statistical significance was calculated using Student's one-tailed t-test (S5E Fig, S2 Table).
To validate the afore described quantification method, we also measured the signaling intensity by selecting 20x20μm squares in the dorsal and the ventral compartments of the same wing disc using the Nikon NIS software (S6A and S6B Fig; genotype: ap-Gal4, sal-lacZ/+; UAS-Med15-i/+). We then calculated the dorsal to ventral ratios of the signal intensities of three different discs (S6B Fig), followed by statistical analyses using the Student's one-tailed t-test (S6C and S6D Fig). We obtained similar results to the quantification based on the line profiles as described above.

Statistical analysis
Standard deviation and Student's one-tailed t-tests were performed using Microsoft Excel. Statistical significance ( � p<0.05; �� p<0.01; ��� p<0.001) was shown in figures and all error bars indicate standard deviation.