CDK8 and CDK19: positive regulators of signal-induced transcription and negative regulators of Mediator complex proteins

Abstract We have conducted a detailed transcriptomic, proteomic and phosphoproteomic analysis of CDK8 and its paralog CDK19, alternative enzymatic components of the kinase module associated with transcriptional Mediator complex and implicated in development and diseases. This analysis was performed using genetic modifications of CDK8 and CDK19, selective CDK8/19 small molecule kinase inhibitors and a potent CDK8/19 PROTAC degrader. CDK8/19 inhibition in cells exposed to serum or to agonists of NFκB or protein kinase C (PKC) reduced the induction of signal-responsive genes, indicating a pleiotropic role of Mediator kinases in signal-induced transcriptional reprogramming. CDK8/19 inhibition under basal conditions initially downregulated a small group of genes, most of which were inducible by serum or PKC stimulation. Prolonged CDK8/19 inhibition or mutagenesis upregulated a larger gene set, along with a post-transcriptional increase in the proteins comprising the core Mediator complex and its kinase module. Regulation of both RNA and protein expression required CDK8/19 kinase activities but both enzymes protected their binding partner cyclin C from proteolytic degradation in a kinase-independent manner. Analysis of isogenic cell populations expressing CDK8, CDK19 or their kinase-inactive mutants revealed that CDK8 and CDK19 have the same qualitative effects on protein phosphorylation and gene expression at the RNA and protein levels, whereas differential effects of CDK8 versus CDK19 knockouts were attributable to quantitative differences in their expression and activity rather than different functions.

In se v eral cases, Mediator kinase acti vity was found to affect (possibl y indirectl y) the phosphorylation of the Cterminal domain (CTD) of RN A pol ymerase II (Pol II). The Pol II CTD phosphorylation-based mechanism has been implicated in downstream potentiation of the serum response network ( 17 ), HIF1 α ( 6 ), ER ( 9 ) and NF B ( 10 ) by CDK8 / 19. Importantly, Mediator kinase inhibition suppresses CTD phosphorylation not globally but only in the conte xt of ne wly acti vated genes, and CDK8 / 19 inhibitors suppress de novo induction of Mediator kinase co-regulated signal-stimulated genes ( 10 ). This pattern suggested that CDK8 / 19 regulate transcriptional reprogramming ( 1 , 10 ). Tr anscriptional reprogr amming is critical for se v eral biological and pathological processes that are suppressed by CDK8 / 19 inhibition, including embryonic de v elopment ( 2 , 18 ), cancer metastasis ( 19 ) and drug resistance ( 20 , 21 ).
Since Mediator kinases have been implicated in many tumor-promoting activities, the development of CDK8 / 19 inhibitors has become a burgeoning area in cancer therapeutics ( 22 ). CDK8 / 19 inhibitors were also found to have therapeutic activities beyond oncology, such as inhibiting viral replication ( 23 ) and ameliorating autoimmune responses ( 24 , 25 ). Almost all the reported Mediator kinase inhibitors have similar potency against CDK8 and CDK19, and it is unknown if selecti v e inhibition of one of the paralogs would be advantageous or detrimental for therapeutic purposes. CDK8 and CDK19 differ in their relati v e e xpr ession in differ ent tissues, CDK19 being tissue-specific and CDK8 relati v ely ubiquitous ( 26 ). Transcriptomic effects of CDK8 and CDK19 knockdown in HeLa cells were reported to be similar although some genes appeared to be pr efer entially affected by either CDK8 or CDK19 ( 26 ). CDK8 and CDK19 were found to cooperate with each other in supporting leukemia cell growth ( 13 ), stimulating NF B-induced transcription ( 10 , 27 ) and Dengue virus replication ( 23 ). In contrast, another study ( 12 ) concluded that CDK8 and CDK19 have mechanistically distinct functions in IFN ␥ -tr eated cells, wher ein CDK8 kinase mediates IFN ␥ -induced transcription and STAT1 phosphorylation at S727 but CDK19 has no effect on STAT1 S727 phosphoryla tion and af fects transcription in a kinase-independent manner. These conclusions were based principally on the findings that CDK8 knockout or inactivation mimicked the effects of a Mediator kinase inhibitor, whereas CDK19 knockdown in mouse embryo fibroblasts (MEF) had no significant effect on STAT1 S727 phosphorylation or on the expression of IFN ␥ -inducible genes ( 12 ).
We have now investigated the transcriptomic, proteomic and phosphoproteomic effects of CDK8 and CDK19 in human cells, using isogenic cell populations expressing either wild type (WT) or kinase-inacti v e v ersions of CDK8 or CDK19, highly selecti v e Mediator kinase inhibitors and a CDK8 / 19-degrading PROteolysis TArgeting Chimera (PROTAC). Our results demonstrate that CDK8 and CDK19 have the same qualitative effects on gene expression and protein phosphorylation, including STAT1 S727 phosphorylation. The differences between the phenotypic effects of the knockout of CDK8 or CDK19 alone could be explained by quantitative differences in the expression and activity of the corresponding proteins. Transcriptomic effects of CDK8 / 19 were kinase-dependent, but CDK8 and CDK19 protected their binding partner CCNC fr om pr oteolytic degradation in a kinase-independent manner. Analysis of the effects of Mediator kinase inhibition on gene expression affected by different signals re v ealed that CDK8 / 19 maximize the expression of the most strongly signal-inducible genes. Mediator kinase inhibition in unstimulated cells initially leads to downregulation of a small number of genes, most of which were signal-inducible. Prolonged Mediator kinase inhibition or genetic inactivation led to upregulation of a larger gene set and a posttranscriptional increase of the protein components of both the core Mediator complex and its kinase module, suggesting a previously unknown mechanism for negative regulation of gene expression by Mediator kinases.

Materials and reagents
All the key r esour ces used in this study (reagents, cell lines, antibodies , vectors , kits , softwar e) ar e listed in Supplementary Table S1.

Cell culture
Cell lines HEK 293 (and its deri vati v es), HAP1, HCT116, HeLa and HT1080 were maintained in DMEM-high glucose media (Thermo-Fisher Scientific) supplemented with 10% fetal bovine serum (FBS) (Atlanta Biologics), 1% penicillin-streptomycin and 2 mM L -glutamine. Cell line MV4-11 was cultured in RPMI-1640 supplemented with 10% FBS, 1% penicillin-streptomycin and 2 mM Lglutamine. Cell line 22Rv1 and its deri vati v es were cultured in RPMI-1640 supplemented with 10% FBS, 1% penicillinstreptomycin, 2 mM L -glutamine, 1 mM sodium pyruvate, 0.15% sodium bicarbonate, 10 mM HEPES and 25 mM D -glucose. All cell lines were confirmed mycoplasma-free (My coAlert PLUS my coplasma detection kit, Lonza). For RN A anal ysis (RN A-Seq or qPCR), cells were seeded in 12-well pla tes a t appropria te numbers (5 × 10 4 -3 × 10 5 cells per well) to allow cells to grow to ∼90% confluence at the endpoint. For inhibitor treatment, cells were seeded 24 h before being treated with vehicle control (0.1% DMSO) or indicated chemicals at the stated concentrations and time periods (up to 3 days). For long-term treatment (15 days), cells were cultured in presence of vehicle or inhibitors and passaged e v ery 3 days with replacement of fresh vehicle or inhibitors. For serum stimulation, cells were serumstarved for 48 h in serum-free media before adding FBS to final serum concentration of 10%. For signal stimulation, cells were seeded in regular culture media for 24 h and then treated with different stimulants (10 ng / ml TNF, 5-20 ng / ml IFN ␥ or 30 nM PMA). CDK8 / 19 inhibitor (Senexin B, 1 M) was added 1 h before signal stimulation and maintained till the end of experiment.

Stoichiometry determination for CDK8 / CDK19 proteins
Cells were grown to 90% confluence in P150 plates, trypsinized, washed by PBS twice, lysed in RIPA buffer and whole cell extracts were prepared as above. For quantification of CDK8 to CDK19 protein ratio in 293 cells, 293 cell extracts or recombinant GST-CDK8 and GST-CDK19 proteins serially diluted with 293-dKO whole cell extracts (to equalize total protein amounts per lane) were analyzed by immunoblotting with anti-CDK8, anti-CDK19 and anti-GST antibodies. Band signal intensities were acquired using ImageLab software (Bio-Rad) and further analyzed in Microsoft Excel to evaluate the linear relationship between band signals and serial dilutions and quantify relati v e CDK8 and CDK19 protein le v els. Only the standard points that gave a good linear r egr ession wer e used to build calibration curves and only the lanes / dilutions of those cell extracts whose band signals fall in the linear range were picked for quantitation. GST band signals were used to normalize the le v els of the two GST recombinant proteins. This normalization was used to adjust the ratio of CDK8 to CDK19 calculated from the quotient of relati v e CDK8 and CDK19 protein le v els determined by mapping CDK8 / CDK19 band signals to standar d curv es generated with GST recombinant proteins. For quantification of CDK8 / CDK19 ratios in other cell lines, serially diluted cell extracts were run in parallel with 293 whole cell extract as an internal standard for quantifica tion. Rela tive abundance of CDK8 and CDK19 proteins between the tested cell lines and 293 was determined by densitometry and used to calculate CDK8 / CDK19 ratios in each cell line.

RNA-seq analysis
Cells seeded in 12-well plates and treated as described previously and in figure legends were lysed for RNA extraction using RNeasy Mini Kit (Qiagen). RNA-Seq libr ary prepar ation, NGS, post-processing of the r aw da ta and da ta analysis were performed by Functional Genomics Core (FGC) of the CTT. RNA-Seq libraries wer e pr epar ed in conjunction with poly(A)-enrichment using either TruSeq Stranded mRNA prep kit (RS-122-2101 / RS-122-2102) or NEBNext Ultra II Directional RNA Library Prep Kit (#E7760). NGS was performed on Illumina NextSeq 500 (at FGC) or HiSeq 3000 / 4000 (at Genewiz, Inc., South Plainfield, NJ) or Illumina No-vaSeq (at MedGenome, Inc., Foster City, CA) platforms for paired end sequencing. Reads were mapped to the Human GRChg38 r efer ence genome using STAR v2.4 ( 28 ). Samtools (v1.5) was used to convert aligned SAM files to BAM files, and reads were counted using the featureCounts function of the Subreads package ( 29 ) with Gencode.v25.basic.annota tion.gtf annota tion file. Only r eads that wer e ma pped uniquel y to the genome were used for gene expr ession analysis. Differ ential expr ession (DE) analysis was performed in R using the DESeq2 ( 30 ) pipeline, where the normalized counts data were fit to a negati v e binomial distribution model using a generalized linear model (GLM) frame wor k and the Benjamini-Hochberg procedure was used to control the false discovery rate (FDR) for multiple testing. The logFC and FDR values calculated from DESeq2 pipelines were utilized to select differentially expressed genes (DEGs). To select high-confidence DEGs from multiple RNA-Seq experiments, the following criteria were applied: (i) FDR < 0.05 in all experiments; (ii) average log 2 FC from multiple experiments > log 2 (1.5). Normalized RNA expression levels of CDK8 and CDK19 in different cancer cell lines were retrie v ed from Cancer Cell Line Encyclopedia (CCLE) database (CCLE RNAseq genes rpkm 20180929.gct.gz). RNA-Seq raw data of MEF treated with IFN ␥ and CA ( 12 ) were downloaded as SRR files from NCBI-SRA w e bsite and converted into Fastq files using SRA Toolkit. Reads were mapped to the Mouse GRCm38.88 reference genome and processed to gene counts with Gencode.vM15 annotation file for DE analysis. All raw RNA-Seq data have been uploaded to GEO (see data availability section) and detailed information about individual RNA-Seq samples (sample title and description, GEO accession number) is listed in Supplementary Table S3.
Quantitative RT-PCR (qPCR). Cells were seeded in 12well pla tes a t the density r equir ed to approach confluence at the end of experiment and treated as indicated in figur e legends befor e being l ysed for RN A extraction using RNeasy Mini Kit (Qiagen). One microgram of total RNA was used to generate cDNA using iScript cDNA synthesis kit (Bio-Rad). Target gene expression was quantified using iTaq Uni v ersal SYBR green super mix in CFX384 Real time system (Bio-Rad). Primers used for RT-PCR are listed in Supplementary Table S2. RT-PCR data files were processed using Bio-Rad CFX Manager softwar e to r etrie v e Ct numbers of qPCR reactions. Relati v e RNA e xpression of specific genes was calculated by the formula: Relati v e Expression = 2 ∧ (Ct r efer ence -Ct gene ), where RPL13A, HPRT1 or GAPDH were used as reference genes.
Pr oteomics and phosphopr oteomics analysis . Tandem Mass Tag (TMT) based proteomic and phosphoproteomic analysis of a total of 30 samples was performed in three TMT-11plex batches. The first batch comprised dKO-CDK8 and dKO-CDK8M (fiv e biological replicates of each), the second dKO-CDK19 versus dKO-CDK19M (5 + 5 replicates), and the third parental 293 cells treated with DMSO (Ctrl), 1 M Senexin B (3 h) or 1 M Senexin B (72 hrs) (4 + 3 + 4 r eplicates, corr espondingly). For each replicate, cells were grown in a P150 plate to ∼90% confluence before being collected for analysis. At the endpoint, cultur e media wer e r emoved and cells wer e rinsed with ice-cold PBS three times and scraped down in 5 ml PBS with protease / phosphatase inhibitor cocktail. Cells were pelleted by centrifugation at 200 g × 5 min at 4 • C, snap-frozen after removal of supernatant and stored at -80 • C before the pr oteomics / phosphopr oteomics analysis at the Proteomics Core Laboratory of the Uni v ersity of Arkansas for Medical Sciences (UAMS). Cells were lysed in RIPA buffer (Thermo-Fisher Scientific PI89901) and 200 g total protein lysates were reduced, alkylated and purified by chloroform / methanol extraction prior to digestion with sequencing grade trypsin and LysC (Promega). The resulting peptides were labeled using a TMT 11-plex isobaric label reagent set (Thermo-Fisher Scientific) in three multiplex batches with a pooled r efer ence sample in each batch, then enriched using High-Select TiO2 and Fe-NTA phosphopeptide enrichment kits (Thermo-Fisher Scientific) following the manufacturer's instructions. Both enriched and unenriched labeled peptides were separated into 46 fractions on a 100 × 1.0 mm Acquity BEH C18 column (Waters) using an UltiMate 3000 UHPLC system (Thermo-Fisher Scientific) with a 50 min gradient from 99:1 to 60:40 buf fer A:B ra tio under basic pH conditions, then consolidated into 18 super-fractions. Each superfraction was further separated by re v erse phase XSelect CSH C18 2.5 um resin (Waters) on an in-line 150 × 0.075 mm column using an UltiMate 3000 RSLCnano system (Thermo-Fisher Scientific). Peptides were eluted using a 75 min gradient from 98:2 to 60:40 buffer A:B ratio. Eluted peptides were ionized by electrospray (2.2 kV) followed by mass spectrometric analysis on an Orbitrap Eclipse Tribrid mass spectrometer (Thermo-Fisher Scientific) using multinotch MS3 parameters with real-time search enabled. MS data wer e acquir ed using the FTMS analyzer in top-speed profile mode at a resolution of 120 000 over a range of 375-1500 m / z . Following CID activation with normalized collision energy of 31.0, MS / MS data were acquired using the ion trap analyzer in centroid mode and normal mass range. Using synchronous precursor selection, up to 10 MS / MS pr ecursors wer e selected for HCD activation with normalized collision energy of 55.0, followed by acquisition of MS3 reporter ion data using the FTMS analyzer in profile mode at a resolution of 50 000 over a range of 100-500 m / z . Proteins and phosphosites were identified and reporter ions quantified by searching the UniprotKB H. sapiens database (July 2020) using MaxQuant (version 1.6.17.0; Max Planck Institute) with a parent ion tolerance of 3 ppm, a fragment ion tolerance of 0.5 Da, a reporter ion tolerance of 0.001 Da, trypsin enzyme with 2 missed cleav ages, v ariable modifications including oxidation on M, Acetyl on Protein N-term, and phosphorylation on STY, and fixed modification of carbamidomethyl on Cterm. Protein and peptide identifications were accepted if established with < 1.0% false discovery. TMT MS3 reporter ion intensity values were analyzed for changes in total protein using the unenriched lysate sample. Phospho (STY) modifications were identified using the samples enriched for phosphorylated peptides. The enriched and unenriched samples wer e multiplex ed using two TMT11-plex batches, one for the enriched and one for the unenriched samples. Following data acquisition and database search, the results were normalized using cyclic loess normalization for both the protein and the phosphopeptide data sets ( 31 ). The normalized protein and phosphorylated peptide data were analyzed for differential abundance using the limma package by a ppl ying 'lmFit' and 'eBayes' functions. A similar approach was used for differential analysis of the phosphopeptides. The phosphosites were filtered to retain only peptides with a localization probability > 75% and log 2 cyclic loess transformed. Limma is also used for differential analysis of single phosphosite peptides. The P -values were adjusted for multiple test correction using the false discovery rate (FDR). The raw and processed proteomics and phosphoproteomics data have been uploaded to MassIVE database (see data availability section).
Statistical analysis. WB experiments were performed at least in duplicates. Means of densitometry signals from WB duplicate images are presented in the bar diagrams. RNA-Seq experiments were carried out in biological replicates ( n ≥ 3) for each treatment condition. Procedures for RNA-Seq and proteomics data analysis are described in the above sections. The significance of the overlap detected in Venn diagrams was assessed by a hypergeometric test. Slope and Pearson correla tion coef ficients were calcula ted by linear regr ession and corr elation anal ysis using Gra phPad Prism 9 softw are. qPCR analysis w as performed in biological triplicates and data were presented as mean ± standard error of the mean (SEM). Statistical significance was tested using ordinary two-way ANOVA and Tukey's multiple comparisons test with GraphPad Prism 9 software.

Experimental strategy
As the primary cellular model to analyze the transcriptomic and proteomic effects of CDK8 and CDK19, we have used 293 human embryonic kidney cells, the principal cell line used in our earlier study that elucidated the role of CDK8 / 19 in NF B signaling ( 10 ). We hav e pre viously genera ted 293 deriva tives with the CRISPR-mediated knockout of CDK8 alone (8KO), CDK19 alone (19KO) and both CDK8 and CDK19 (double knockout, dKO) ( 27 ). Howe v er, we avoided drawing conclusions from comparisons between these knockout clones and the parental 293 cells for the following reasons: (i) comparison of parental cells to individual subclones reveals numerous differences, especially at the transcriptomic le v el, due to clonal variability and (ii) sgRN A knockout (or siRN A knockdown) causes transcriptomic changes that are not necessarily mediated by the target. Instead, our principal analysis was based on comparing isogenic mass populations of dKO cells reconstituted with wild-type or kinase-inacti v e v ersions of CDK8 or CDK19, with further validation of the conclusions using highly selecti v e small-molecule inhibitors and a novel PROTAC degrader of CDK8 / 19. Cell line deriva tiza tion stra tegy is diagrammed in Figure 1 A. We have transduced dKO cells with a lentiviral vector (pHIV-dTomato) expressing either wildtype CDK8 or CDK19 or their kinase-inacti v e D173A mutants ( 16 , 32 ), obtaining mass populations named dKO-8, deri vati v es were expected to be higher than in the original 293 cells, and because protein ov ere xpression could cause artificial changes, we have introduced an additional layer of controls to account for the effects of the ov ere xpression, by generating and analyzing deri vati v es of the parental 293 cells that were made to ov ere xpress CDK8, CDK19 or their kinase-inacti v e mutants from pHIV-Luc-BlastR (these cell populations were named WT-V, WT-8, WT-8M, WT-19 and WT-19M) (Figure 1 A). This set of 293 cell deri vati v es was used to deri v e our initial conclusions, which were then tested using selecti v e small-molecule kinase inhibitors and a targeted degrader of CDK8 / 19.

Effects of CDK8 and CDK19 on CCNC degradation
Imm unoblotting anal ysis in Figure 1 B verifies the knockout and reconstitution of CDK8 and CDK19 in 293 cells and asks if CDK8 / 19 expression affects the other components of the Mediator kinase module: CCNC, MED12 and MED13. Remar kab ly, the le v els of CCNC, the necessary binding partner of CDK8 and CDK19, were drastically decreased in dKO, whereas reconstitution of either WT or kinase-inacti v e mutant CDK8 or CDK19 in these cells restored CCNC le v els (Figure 1 B). To determine if CCNC stabilization by CDK8 and CDK19 was due to protection from proteasomal degradation, we have tested the effects of three proteasome inhibitors: MG132, MG115 and bortezomib, on CCNC protein le v els in the parental and dKO cells. All three inhibitors had no significant effect on CCNC in the parental cells but greatly increased its le v els in dKO ( Figure  1 C), indicating the role of proteasomal degradation in the regulation of CCNC le v els by Mediator kinases. Notab ly, expression of kinase-inactive CDK8 or CDK19 mutants in dKO cells not only r estor ed CCNC expr ession but did so to a greater extent than their WT counterparts (Figure 1 B). Furthermor e, over expr ession of kinase-inactive (but not WT) CDK8 or CDK19 further increased CCNC in parental cells (Figure 1 B). These results are in agreement with an earlier study that found CCNC to be stabilized by both kinase-acti v e and inacti v e CDK8 ( 33 ). In contrast to CCNC, MED12 and MED13 proteins were not downregulated but in fact slightly increased in dKO or dKO-V cells, and their le v els were decreased by the expression of the WT but not kinase-inacti v e CDK8 or CDK19 in dKO (Figure  1 B). These results suggested that CDK8 and CDK19 protect CCNC (but not MED12 or MED13) from proteolytic degradation in a kinase-independent manner, whereas their kinase activity may have a negative effect on all three of the other components of the Media tor-associa ted CDK module. As described below, proteomic analysis demonstrated a broad negati v e effect of Mediator kinase activity on the protein le v els of not only the CDK module but also the core Mediator components.

CDK8 and CDK19 have similar, kinase-dependent effects on basal gene expression
RN A-Seq anal ysis was used to char acterize the tr anscriptomic effects of wild-type or kinase-inacti v e CDK8 or CDK19 expression in dKO and WT 293 cells. The strategy for the selection of Differentially Expressed Genes (DEG) is diagrammed in Supplementary Figure S1A; fold-change (FC) > 1.5 and False Discovery Rate (FDR) < 0.05 were used as the cutoff criteria for DEG selection. Figure 2 A sho ws v olcano plots of the effects of the expression of WT CDK8 or CDK19 in dKO cells, with black circles marking the high-confidence DEGs that were shared in two independent sets of deri vati v es. Four hundred and three highconfidence DEGs wer e r egulated by WT CDK8 and 220 DEGs by WT CDK19; these overlapping sets comprise a total of 429 high-confidence DEGs regulated by WT CDK8 or CDK19 (Supplementary Table S4). In contrast to their effects in dKO cells, ov ere xpression of WT CDK8 or CDK19 in parental cells had no significant effects on gene expr ession (Figur e 2 B), validating the use of ectopically expressed Mediator kinases in our experimental strategy. Unlike WT CDK8 and CDK19, very few genes were regulated by kinase-inacti v e Mediator kinase mutants in dKO cells, with only 11 high-confidence genes weakly affected by the CDK19 mutant and 2 genes by the CDK8 mutant (Figure 2 C), indicating that regulation of gene expression by CDK8 and CDK19 is largely kinase-dependent. This conclusion was confirmed by PRO TAC anal ysis (see below). On the other hand, kinase-inacti v e CDK8 and CDK19 mutants showed detectable effects on gene expression in parental cells ( (Figure 1 B). This difference is also apparent from the Venn diagram in Figure 2 H comparing DEGs selected from CDK8 or CDK19 expression in dKO cells. While the overlap was highly significant and almost all the CDK19-regulated DEGs were also CDK8regulated, one half of CDK8-regulated DEGs failed to pass the cutoff criteria for the effect of CDK19, despite qualitati v e cor egulation (Figur e 2 F, G). Furthermor e, the dominant negati v e effects of mutant CDK8 or CDK19 e xpression in parental cells were very similar but in this case the effect of mutant CDK19 was stronger (Figure 2 I). Hence, CDK8 and CDK19 have qualitatively the same but quantitati v ely different transcriptomic effects, both as WT kinases and as dominant negati v e mutants.
The effects of the knockout of both CDK8 and CDK19 (dKO) on CDK8 / 19-regulated DEGs were inverse to the effects of CDK8 or CDK19 expr ession (Figur e 2 E), indi-ca ting tha t both CDK8 and CDK19 re v ersed the transcriptomic effects of dKO. The effects of CDK8 knockout (8KO) on CDK8 / 19-regulated DEGs w ere w eaker than but similar to the effects of dKO (Figure 2 E), but the effect of CDK19 knockout (19KO) was very weak and did not show a similar pattern (Figure 2 E). The reasons for the difference between the CDK8 and CDK19 knockouts will be discussed below.

Downregulation of gene expression is an early response and upregulation is a late response to CDK8 / 19 inhibition
To confirm the transcriptomic data obtained from Mediator kinase expression and mutagenesis and to elucidate the time course of the transcriptomic effects of CDK8 / 19 inhibition, we have used Senexin B, a highly selective CDK8 / 19 inhibitor ( 9 , 34 ) that was the first to reach clinical trials ( 35 ).  Table S5). This DEG set significantly overlaps with the DEGs selected on the basis of CDK8 / 19 expression (Figure 3 B) despite the stringent cutoff criteria for DEG selection.
Changes in gene expression upon CDK8 / 19 inhibition can be seen in Figure Figure S3B).
Supplementary Figure S4A shows the effects of Senexin B treatment, WT CDK8 or CDK19 expression in dKO cells, and the knockout of CDK8, CDK19 or both Mediator kinases, on the combined set of 668 DEGs regulated by either CDK8 / 19 expression or Senexin B. As in the individual sets (Figure 3 D, G), CDK8 or CDK19 expression in dKO cells produces the opposite effects to prolonged Senexin B treatment on the combined DEG set (except for a small number of genes affected by the inhibitor but not by Mediator kinase expression). The effects of dKO closely resemble the effects of Senexin B, whereas single knockout of CDK8 but not CDK19 partially reproduces these effects. In contrast, the frequently used approach of defining the effects of genes through their knockouts does not gi v e such a clear pattern. Using the same cutoff criteria (FC > 1.5 and FDR < 0.05), we identified 1122 high-confidence DEGs af fected in dKO rela ti v e to parental cells (Supplementary Figure S1C), 1040 DEGs affected in 8KO and 158 DEGs affected in 19KO and generated heatmaps for these DEG sets under the same conditions. Most of dKO-based DEGs are affected by Senexin B and CDK8 / 19 expression but some dKO-affected DEGs were neither re v ersed by CDK8 or CDK19 expression nor regulated by Senexin B (Supplementary Figure S4B). The number of such non-responsi v e genes is much greater among the 8KO-based DEGs (Supplementary Figure S4C) and especially 19KO-based DEGs (Supplementary Figure S4D). Such genes may reflect the CDK8 / 19-unrela ted ef fects of sgRNA / CRISPR transduction and clonal selection, illustrating the shortcomings of the analysis based on gene knockout.

Targeted degradation of CDK8 / 19 confirms kinase dependence of the transcriptomic effects and indicates compensatory transcriptomic changes in dKO
We have recently developed a potent PROTAC degrader of CDK8 and CDK19 (manuscript in preparation). This PRO-TAC, SNX7886 (Figure 4 A) is based on the CDK8 / 19 inhibitor BI1347 connected to a Cereblon E3 ligase binder pomalidomide via an alkane linker. Treatment of 293 cells with SNX7886 at concentrations as low as 30 nM degrades CDK8 by up to ∼90% and CDK19 by up to ∼80%, with concurrent CCNC degradation (Figure 4 B) that resembles the effects of dKO (Figure 1 B).
To determine if CDK8 / 19 degradation would produce any transcriptomic effects distinct from those of kinase inhibition (i.e. kinase-independent effects), we have carried out  Figure 4 E sho ws, ho we v er, that the same genes were also affected by the PROTAC in dKO (with the exception of a single gene, which, as shown below, was in fact affected by kinase inhibitors). Hence, the effects of the PROTAC that were not shared by the kinase inhibitor were not CDK8 / 19-mediated. Notably, none of the genes that appeared to be weakly affected by kinaseinacti v e CDK19 or CDK8 mutants (Figure 2 C) were differentially affected by the kinase inhibitors and PROTAC. These results, together with the above-described effects of the kinase-inacti v e CDK8 and CDK19 mutants, confirm that the transcriptomic effects of CDK8 and CDK19 are kinase-dependent.
The heatmap in Figure 4 F compares the effects of SNX7886 PROTAC and four CDK8 / 19 kinase inhibitors (Senexin B, dCA, 15w and BI1347), dKO (5 different studies) and CDK8 or CDK19 reconstitution in dKO cells (two different studies) on the expression of 366 DEGs affected by Senexin B or CDK8 / 19 reconstitution (Supplementary Figure S4A) and the PROTAC. The effects of dKO on these DEGs largely resembled the effects of the inhibitors or PROTAC and re v ersely correlated with the effects of CDK8 or CDK19 expression in dKO (Figure 4 F). Venn diagram comparison of the effects of the PROTAC and dKO ( Figure  4 G) re v eals the expected overlap but also more differences than in other pairwise comparisons. While dKO-specific effects are likely to stem from the clonal nature of dKO cells, ther e ar e also many genes affected by the PROTAC but not by dKO, suggesting that such genes could have undergone compensatory changes during the establishment of dKO cell line. For a closer look at such compensatory changes, we have selected a subset of DEGs that were affected (FC > 1.5, FDR < 0.05) by all 4 kinase inhibitors and the PROTAC but not affected by dKO in the same direction in any of the studies. As shown in Figure 4 H, some of the genes unaffected by dKO were still affected by CDK8 or CDK19 expression in dKO, in the direction opposite to the effect of the inhibitors, whereas a few genes were unaffected in dKO cells by CDK8 or CDK19 expression suggesting that the adaptation of these cells involved a switch from CDK8 / 19dependent to CDK8 / 19-independent regulation.
Supplementary Figure S5 Figure  S5D shows that it was still induced by BI1347 and induced e v en stronger by four other kinase inhibitors. Also CCND1 appears to be upregulated by kinase-inacti v e CDK8 and ETV5 downregulated by kinase-inacti v e CDK19 mutant in dKO cells but these genes were not selecti v ely affected by the PROTAC , indica ting tha t their regula tion was not in fact kinase independent.

Mediator kinases potentiate the induction of gene expression by different signals
Extending our previous studies on the potentiation of signal-induced transcriptional activation by CDK8 / 19 activity in 293 cells ( 10 ), we have investigated the effects of CDK8 / 19 inhibition on transcriptomic responses to a va-riety of transcription-altering signals in 293 cells, including serum stim ulation (previousl y shown to be potentiated by CDK8 ( 17 )), NF B activation (potentiated by CDK8 / 19 ( 10 )), protein kinase C (PKC) activation (not previously analyzed for Mediator kinase dependence) and IFN ␥ treatment (reported to be affected by Mediator kinase ( 8 , 12 )). Cells wer e tr eated with the corr esponding signal inducers in the presence or absence of Senexin B (1 M), added 1 h before the signals. DEGs affected by each agent or by Senexin B were selected by the criteria FC > 1.5, FDR < 0.05; the corresponding flow charts and DEG numbers are shown in Supplementary Figure S1D.
The effects of Senexin B on DEGs affected by the first three signals are shown in Figure 5   Data are presented as mean ± SEM ( n = 3). Asterisks: P < 0.01 (two-way ANOVA, Tukey's multiple comparisons test) for the differences between Senexin B-treated and untreated conditions. h ( Figure 5 D). Senexin B-affected DEGs (marked with red dots in Figure 5 A-D) comprised 77% of all the genes that were affected by serum, 21% of TNF-regulated genes, 6% of genes affected by 2-h PMA treatment and 3% of genes affected by 24-h PMA treatment. All or almost all of the genes affected both by the signals and by Senexin B were induced by the signals, whereas Senexin B decreased their induction, indica ting tha t Media tor kinase acts primarily as a positi v e regulator of gene expression induced by these signals. The percentage of Senexin B-regulated DEGs gradually increases if only the top 50%, 20%, 10% or 5% most-strongly signal-induced genes are considered ( Figure 5 A-D), reaching 100%, 100%, 38% and 36% among the top genes induced by serum, TNF, 2-h PMA and 24-h PMA, respecti v ely. Hence, the genes that are most strongly induced by different signals are also most likely to be affected by Mediator kinase inhibition, indicating that their induction is augmented by CDK8 / 19. , indica ting tha t the early response to CDK8 / 19 inhibition may primarily reflect the effect on transcription induced by signals in cell culture media. In contrast, many of the late-response genes that were induced by Senexin B after 24 h were no longer induced in the presence of PMA (Figure 5 H), indicating that long-term treatment with the PKC agonist broadly altered the negati v e regulation of gene expression by CDK8 / 19. As discussed below, this effect may reflect a chromatin rearrangement affecting the distribution of Mediator.
To determine the relati v e contributions of CDK8 and CDK19 kinase activities to signal-induced gene expression, we have used qPCR to measure the effects of CDK8 and CDK19 reconstitution on the basal and signal-induced expression of selected Senexin B-affected genes that are stimulated by serum ( Figure 5  In the absence of acti v e Mediator kinases, almost all the tested genes were still inducible by the corresponding signals but Senexin B had no effect on their induction. Reconstitution of the wild-type (but not mutant) CDK8 or CDK19 in dKO deri vati v es increased the induc-tion of these genes by PMA and TNF but not by serum, whereas the induction by all the signals, including serum, became susceptible to inhibition by Senexin B. This result indica tes tha t Media tor kinase e xpression e xerts a partial switch from Mediator-kinase independent to Mediator kinase-dependent transcriptional activation mechanisms.

Effects of CDK8 and CDK19 on IFN ␥-regulated transcription and ST A T1 S727 phosphorylation
W hile Media tor kinases ar e not known to have a dir ect effect on transcription factors that regulate the induction of gene expression by serum, PMA or TNF, the effect of CDK8 / 19 on IFN ␥ -induced transcription has been linked to a direct effect on STAT1, a transcription factor involved in IFN ␥ response. STAT1 is directly phosphorylated by CDK8 at S727 ( 8 ), and this phosphorylation is inducible by IFN ␥ . STAT1 S727 phosphorylation has become a widely used biomarker of Mediator kinase activity, although it also occurs in the absence of CDK8 / 19, indica ting tha t this phosphorylation is also induced by other kinases ( 34 ). Interestingly, the effects of Senexin B on the transcriptomic effects of IFN ␥ showed a more complicated pattern than with the other signals. Only 12 genes were induced by 4-h treatment with IFN ␥ (10 ng / ml) in 293 cells and the induction of only one of them (STAT1) was significantly suppressed by Senexin B (Figure 6 A). We therefore analyzed the effects of IFN ␥ and Senexin B in a known IFN ␥ -responsi v e cell line, HAP1 leukemia ( 39 ). Many more genes (239) were affected Since IFN ␥ -induced STAT1 S727 phosphorylation was suggested to be mediated by CDK8 but not by CDK19 ( 12 ), we have investigated the effects of CDK8 and CDK19 expression on basal and IFN ␥ -induced STAT1 phosphorylation in 293 cell deri vati v es. Upon the addition of IFN ␥ , STAT1 tyrosine (Y701) phosphorylation was induced (from undetectab le le v els), and serine (S727) phosphorylation was incr eased r elati v e to the basal le v el ( tially decreased by the knockout of CDK8 alone and further decreased in dKO cells but not by the knockout of CDK19 alone (Figure 6 G, H). Ne v ertheless, the reconstitution of either CDK8 or CDK19 (but not their kinaseinacti v e v ersions) in dKO cells increased both basal and IFN ␥ -induced S727 phosphorylation and r estor ed the sensitivity of this phosphorylation to Senexin B (Figure 6 H-J), indica ting tha t both CDK8 and CDK19 can phosphorylate STAT1 at S727.

CDK8 / CDK19 ratios account for different effects of CDK8 and CDK19 knockouts on transcription and ST A T1 S727 phosphorylation
While our results demonstrate that CDK8 and CDK19 expression hav e v ery similar qualitati v e effects on gene expression and STAT1 S727 phosphorylation, both of these readouts were affected by single knockout of CDK8 whereas CDK19 knockout had only a weak effect. We have asked if this could be due to a mechanistic difference between the functions of CDK8 and CDK19, as previously suggested ( 12 ), or to a lower expression of CDK19 relati v e to CDK8. We have ther efor e measur ed the r elati v e CDK8 and CDK19 protein le v els in 293 cells. This analysis was carried out by comparing immunoblotting signal intensity of CDK8-and CDK19-specific bands between serial dilutions of 293 whole cell extract and recombinant human CDK8 and CDK19 proteins tagged with GST at their N-termini; immunoblotting for GST was used to normalize the properly sized signals of the recombinant CDK8 and CDK19 proteins (which differ primarily at their C-termini). The results of replicate experiments are shown in Supplementary Figure S6A; the ratio of CDK8 to CDK19 proteins in 293 cell extract was calculated to be 3.0 ± 0.3. The excess of CDK8 over CDK19 can explain why CDK19 knockout has only a minor phenotypic effect in 293 cells.
We then compared the relati v e le v els of CDK8 and CDK19 between 293 and se v eral other human cell lines (HeLa cervical carcinoma, HCT116 colon carcinoma, HT1080 fibrosarcoma, MV4-11 acute myeloid leukemia, HAP1 chronic myeloid leukemia), as well as 22Rv1 prostate cancer cell line, which r epr esents the only type of cancer where CDK19 is known to be systematically upregulated (40)(41)(42)(43). Using serial dilutions of cell extracts (Supplementary Figure S6B), we found that the ratio of CDK8 to CDK19 was e v en higher in most of the cell lines than in 293 (7.2 in HT1080, 6.6 in HAP1, 6.5 in HCT116 and 3.6 in HeLa), whereas MV4-11 expressed similar le v els of CDK8 and CDK19 (0.9) and 22Rv1 cells expressed 4.5 times more CDK19 than CDK8 (Figure 7 A). We have also determined CDK8 / CDK19 RNA ratios in the same cell lines using our RNA-Seq data for 293 cells and RNA-Seq data of Cancer Cell Line Encyclopedia (CCLE) for all the other cell lines (Figure 7 B). The RNA and protein ratios for CDK8 and CDK19 showed an excellent correlation ( r = 0.8652) among different cell lines (Figure 7 C).
To evaluate the phenotypic effects of CDK8 and CDK19 in other cell lines, we have generated HAP1 leukemia derivati v es with the knockout of CDK8, CDK19 or both CDK8 and CDK19 (dKO). In agreement with the predominance of CDK8 in HAP1 cells, only CDK8 but not CDK19 knockout reduced basal and IFN ␥ -induced STAT1 S727 phosphorylation in HAP1 (Figure 6 D). Howe v er, the knockout of CDK8 alone did not reduce STAT1 S727 phosphorylation to the le v el of dKO and did not completely abolish the inhibitory effect of Senexin B. As in the case of 293 cells, CCNC le v els of HAP1 wer e r educed by CDK8 knockout and further decreased by dKO (Figure 7 D). Similarl y, onl y CDK8 knockout significantly reduced basal and IFN ␥induced STAT1 S727 phosphorylation in HeLa (Figure 7 E) and HCT116 (Figure 7 F) cells but the knockout could not fully abolish the inhibitory effect of Senexin B. To confirm that both CDK8 and CDK19 can induce STAT1 S727 phosphorylation in HCT116, we expressed WT or kinaseinacti v e CDK8 or CDK19 in HCT116 cells with CDK8 knockout. Both WT CDK8 and CDK19 enhanced basal and IFN ␥ -induced STAT1 S727 phosphorylation ( Figure  7 G, H). As e xpected, kinase-inacti v e CDK8 or CDK19 showed no effect.
We also generated deri vati v es of 22Rv1 prostate cancer cells (which ov ere xpress CDK19 relati v e to CDK8) with the knockout of CDK8 and CDK19, individually and in combination (dKO), and analyzed STAT1 S727 phosphorylation with and without IFN ␥ treatment (Figure 7 I). Interestingly, this cell line showed only very weak induction of STAT1 Y701 phosphorylation e v en by a high dose of IFN ␥ (20 ng / ml), with no significant increase in STAT1 S727 phosphorylation. Basal STAT1 S727 phosphorylation in 22Rv1 cells, howe v er, was quite prominent and it was strongly inhibited by Senexin B or by dKO. In contrast to the other tested cell lines, the knockout of CDK19 alone in 22Rv1 cells had a stronger effect on STAT1 S727 phosphorylation than CDK8 knockout, in agreement with the high ratio of CDK19 to CDK8. These results demonstrate that differential effects of CDK8 and CDK19 depletion on basal and signal-induced STAT1 S727 phosphorylation are determined by differences in relati v e protein expression rather than qualitati v e differences between the functions of these two paralogs.
We also extended our analysis of the effects of CDK8 and CDK19 on gene expression beyond 293 cells, using qPCR to analyze the effects of CDK8 and CDK19 modifications on the expression of Senexin B-regulated genes in differ ent cell lines. Figur e 8 A shows that the knockout of CDK8 in HCT116 cells (CDK8 / CDK19 ratio 6.5) largely decreases both the expression and the effect of Senexin B on EGR1, KLF2 and CSRNP1. Howe v er, the e xpression of either CDK8 or CDK19 in the CDK8 knockout HCT116 cells r estor es the expr ession and Senexin B r egulation of these genes. Figure 8 B shows that the induction of MVD and ID3 by Senexin B in HAP1 cells (CDK8 / CDK19 ratio 6.6) is greatly (but not completely) diminished by the knockout of CDK8 but not CDK19, whereas the knockout of both CDK8 and CDK19 abolishes the induction.

Proteomic analysis reveals negative post-transcriptional regulation of mediator complex components by CDK8 and CDK19 kinases
To determine how the transcriptomic effects of CDK8 and CDK19 correlate with their proteomic effects, we have carried out Tandem Mass Tag (TMT) based proteomic and phosphoproteomic analysis of the effects of CDK8 and CDK19 in 293 cells (without signal stimulation). A total of 30 samples were multiplexed across three TMT-11plex ba tches. The first ba tch included dK O-8 and dK O-8M (fiv e biological replicates of each), the second included dKO-19 vs dKO-19M (5 + 5 replicates), and the third included parental 293 cells treated with DMSO (Ctrl), Senexin B (3 h) or Senexin B (72 hrs) (4 + 3 + 4 r eplicates, corr espondingly). Data dependent acquisition was used to quantitate peptides from three separate batches of TMT multiplexed samples, which introduces a TMT ba tch ef fect tha t caused a subset of proteins and phosphoepitopes not to be identified in all batches. Only proteins detected in all the samples compar ed wer e used for the analysis shown in Figur e 9 . Proteins affected by Senexin B treatment or Mediator kinase mutations are listed in Supplementary Table S6. A total of 226 proteins, selected by the criteria FC > 1.5, FDR < 0.05, were differentially expressed in the presence of kinase-acti v e or inacti v e forms of CDK8 (dKO-8 v ersus dKO-8M, 151 out of 7460 detected proteins) or CDK19 (dK O-19 versus dK O-19M, 125 out of 7447 proteins). The effects of the kinase domain mutations on the expression of the corresponding genes at the RNA and protein le v els ar e compar ed in Figur e 9 A (for CDK8) and Figure 9 B (for CDK19). This analysis distinguished between two sets of proteins that either were or were not regula ted a t the RNA le v el (based on RNA-Seq analysis, FC < 1.3 was chosen as the cutoff for lack of RNA regulation ). For the genes that ar e upr egulated or downr egula ted a t the RNA le v el (b lue dots in Figure 9 A, B), the effects of the kinase mutations on the RNA and protein le v els show excellent correlations for both CDK8 and CDK19. Most of the proteins that were not regula ted a t the RNA le v el (red dots in Figure 9 A, B) were   TUT1  MED13  SIAE  MED12  CCNC  GSDME  SLC38A10  MED13L  MED7  MED10  S100A10  CRELD2  MED14  KAT6B  DECR2  MED21  MED6  GRAMD4  RAB3A  MED16  ANXA2  PIGQ  ACADS  TMEM159  GRAMD1B  MTMR6  TIMM9  AFG1L  SAPCD2  CNTNAP2 Figure 9 H shows the effects on all the 33 proteins comprising the kinase module or the head, middle and tail modules of the core Mediator complex. Strikingly, Mediator kinase inhibition leads to stabilization of all the Mediator proteins, except for MED26, the Mediator complex subunit that was reported to be excluded from core Mediator when it is bound to Mediator kinase module ( 44 ) and ther efor e does not associate with CDK8 / 19. None of the Mediator subunits show comparable regulation by CDK8 / 19 inhibition at the RNA le v el, although CDK19, MED12 and MED14 RNA were slightly upr egulated (Figur e 9 H).
The results of proteomic analysis agree with immunoblotting results in Figure 1 B that showed CCNC, MED12 and MED13 to be upregulated by CDK8 or CDK19 kinase domain mutations. To determine if the increase in protein le v els of the Mediator kinase module and core Mediator subunits is a general consequence of CDK8 / 19 inhibition, we have treated parental 293 cells for 24 hrs with 1 M concentra tions of dif ferent CDK8 / 19 inhibitors including 15w, Senexin C ( 35 ) and Senexin B. Figure 9 I shows that all three compounds decrease STAT1 S727 phosphorylation while at the same time increasing the protein le v els of CCNC, MED12, MED13 and MED7 but not MED26, in agreement with the proteomic data. MED12, MED13 and MED7 are also upregulated in dKO cells, where CCNC is degraded (Figure 9 I). Hence, CDK8 / 19 kinase activity exerts a negati v e post-transcriptional regulation on the components of both the CDK module and the core Mediator complex es. Upr egulation of this transcriptional complex in-creases over the time of CDK8 / 19 inhibition (Figure 9 H), offering an explanation for the delayed induction of transcription by CDK8 / 19 inhibitor treatment.

Phosphoproteomic analysis of the effects of CDK8 and CDK19 kinase inhibition
We have carried out phosphoproteomic analysis of TMT data to compare dKO cells expressing WT or kinaseinacti v e forms of CDK8 or CDK19. Phosphoepitopes affected by Senexin B or Mediator kinase mutations (FC > 1.5, FDR < 0.05) are listed in Supplementary Table S7. Figure 10 A-D sho ws v olcano plots and enriched motif analysis using iceLogo ( 45 ) for phosphoepitopes regulated by CDK8 kinase activity (Figure 10 A), by CDK19 kinase activity (Figure 10  Together with the results of STAT1 S727 phosphorylation analysis, the phosphoproteomic data re v eal that CDK8 and CDK19 hav e qualitati v ely the same effect on protein phosphorylation. We have asked if phosphoepitope changes could be due to changes in the total protein le v els or, conv ersely, if proteomic changes could reflect changes in protein stability consequential to CDK8 / 19-mediated phosphorylation. Figur e 10 H-K compar e fold changes in the protein levels to changes in the expression of the most strongly affected phosphoepitopes for the same protein. In these plots, red dots mark phosphoproteins that were affected with FC < 1.3 at the protein le v el, b lue dots mar k phosphoproteins affected with FC > 1.3 at the protein le v el in the same direction as phosphoprotein changes, and green dots mark phosphoproteins affected with FC > 1.3 at the protein le v el in the opposite direction to phosphoprotein changes. Most of the affected phosphoproteins were not altered > 1.3-fold at the protein le v el (red dots), suggesting that the observed effects on such proteins were at the le v el of phosphorylation. Among proteins affected at both proteomic and phosphoproteomic le v els, the majority showed changes in the same direction by both parameters (blue dots). A few proteins showed opposite directions of proteomic and phosphoproteomic changes, including MED14 (phosphorylated at S1112) and TP53BP1 (phosphorylated at S265), phosphorylation of which was strongly decreased both after 3 or 72 h of Senexin B treatment, while the le v els of these proteins became increased only after 72 h treatment. This pattern suggests that such proteins could be destabilized by CDK8 / 19-media ted phosphoryla tion.
We have also compared the results of our phosphoproteomic analysis of CDK8 / 19 kinase domain mutations or Senexin B treatment of 293 cells with the data of Poss et al. ( 46 ) based on 1-h CA treatment of HCT116 cells (Supplementary Table S8). Among 75 proteins whose phospho-   Figure 10. Phosphoproteomic analysis of the effects of CDK8 and CDK19 kinase inhibition. Tandem Mass Tag (TMT) based phosphoproteomic analysis was carried out for the effects of CDK8 and CDK19 kinase inhibition across the same three TMT-11plex batches as in Figure 9 .   (Figure 11 A), suggesting that such phosphoepitopes could potentially pro vide biomark ers of Mediator kinase activity in different cell types. The commonly affected phosphoepitopes are found in nuclear phosphoproteins OGFR-S349, MED14-S1112, RREB1-S1653, TP53BP1-S265, STAT1-S727, NELFA-S363, AFF4-S814, BRD9-S588, TAF10-S44 and CHD3-S1601, phosphorylation of which is reduced by Mediator kinase inhibition, as well as MED26-T323, ZNF768-S97 and GATAD2A-S100, phosphorylation of which is increased by Mediator kinase inhibition. Figure  11 B shows a heat map of 24 r epr esentati v e phosphoepitopes (discovered in at least three comparisons) that were not identified in the study on HCT116. Enriched motif analysis of the downregulated phosphoepitopes in Figure 10 A and B showed very similar (P / A)PSP moti v es, suggesting that the newly identified downregulated phosphoepitopes are likely to be Mediator kinase phosphorylation substrates.  ( 48 ). Enriched motif analysis of the downregulated phosphoepitopes is presented below the heatmaps.

DISCUSSION
rylation expr essed gr eater le v els of CDK8 than CDK19. In contrast, CDK19 knockout had a stronger effect than CDK8 knockout on STAT1 S727 phosphorylation and a similar effect on gene expression in 22Rv1 prostate carcinoma cells that ov ere xpress CDK19 relati v e to CDK8. In addition, STAT1 S727 phosphorylation and CDK8 / 19regulated gene expression in HCT116 cells with CDK8 knockout were restored by the expression of either CDK8 or CDK19. Based on these results, along with proteomic and phosphoproteomic studies, we conclude that the differences in the effects of CDK8 and CDK19 on basal gene expression and protein phosphorylation are dictated primarily by differences in their expression and secondarily by quantitati v e differences in their activity, but not by a qualitati v e difference in their functions, at least in the cellular systems that we have analyzed.

T r anscriptional and post-transcriptional regulation by CDK8 / 19 is kinase-dependent
Se v eral studies concluded the existence of kinaseindependent phenotypic activities for both CDK8 ( 14 , 15 ) and CDK19 ( 12 , 16

Mediator kinases potentiate transcription induced by most signals
The primary effect of CDK8 / 19 inhibition on gene expression in cells exposed to serum, NF B inducer TNF, and PKC agonist PMA was the reduction of signal-induced gene expression. CDK8 / 19 inhibition had the grea test ef fect on the most strongly signal-induced genes, indica ting tha t Mediator kinases act as positi v e co-factors amplifying the effects of signal-activated transcription factors. Among the genes that were downregulated by CDK8 / 19 inhibitor under basal or signal-stimulated conditions, the majority were inhibited to a gr eater degr ee in the presence than in the absence of the signal, including many signal-inducible genes that were 'silent' (expressed at very low levels) in unstimulated cells. This analysis confirms and extends our previous conclusion, based on the analysis of a small number of signal-stimulated genes ( 10 ), that Mediator kinase is a pleiotropic regulator of signal-stimulated transcriptional reprogramming. This function of Mediator kinase is not limited to multicellular organisms, as a recent study in yeast has also concluded that CDK8 kinase activity is r equir ed for gene activation under stress but not under stead y-sta te growth conditions ( 50 ). Similar positi v e co-regulation of thr ee differ ent signals by CDK8 / 19 is likely consequential to the regulation of Pol II CTD phosphorylation in the selecti v e conte xt of signal-acti vated genes, as pre viously demonstrated for the serum response network ( 17 ), NF B ( 10 ), HIF1 ␣ ( 6 ) and ER ( 9 ). The effects of CDK8 / 19 on IFN ␥ signaling were more complica ted. Media tor kinase inhibition did not show prefer ential suppr ession of IFN ␥ induced gene expression in human 293 or HAP1 cells. Howe v er, analysis of the RNA-Seq data of Steinparzer et al. ( 12 ) showed that the CDK8 / 19 inhibitor pr efer entially downr egulated IFN ␥inducible genes in MEF cells, resembling our results with the other signals in 293 cells. IFN ␥ signaling is regulated to a large extent by STAT1, which is directly (but not e xclusi v ely ( 34 )) phosphorylated by Mediator kinase at S727. STAT1 S727 phosphorylation modulates rather than merely activates STAT1 activity ( 12 , 51 ), which may explain the complica ted ef fects of CDK8 / 19 inhibition on IFN ␥regulated transcription and the differences between its effects in different cell types. Furthermore, we found that IFN ␥ induced (but not basal) STAT1 RNA expression was reduced to some extent by Mediator kinase inhibition in all three tested cell lines, suggesting that some of the effects of Mediator kinase on IFN ␥ regulated genes could be due to the reduced expression of this transcription factor.

Early and late responses to CDK8 / 19 inhibition: relation to signal stimulation and to post-transcriptional upregulation of the mediator complex
Under basal cell culture conditions, CDK8 / 19 inhibitor trea tment af fected only a small number of genes at an early (3 hrs) time point (46 genes by our cutoff criteria), and all these genes were downregulated. Remar kab ly, most of these early-r esponse genes wer e induced upon short-term serum stimulation or 2-h treatment with a PKC agonist PMA (Figure 5 I). Serum stimulation and PMA addition mimic signals that ar e pr esent and have fluctuating activity in conventional cell culture media. In particular, PKC signaling (which is activ ated b y PMA) is controlled by fluctuation in diacylglycerol and Ca levels and interactions with proteins tha t regula te its activity and sta bility, through ela borate feedback mechanisms ( 52 ). This suggests that most if not all the early responses to Mediator kinase inhibition may be mediated by transcription-stimulating signals present in cell culture, which are positi v ely regulated by CDK8 / 19.
In contrast to the early inhibition of gene expression, the primary effect of prolonged CDK8 / 19 inhibitor treatment was the upregulation of a larger number of genes ( ∼400 by our cutoff criteria), and this upregulation was also observed upon long-term genetic inactivation of the Mediator kinase. Surprisingly, we found that prolonged CDK8 / 19 inhibition or mutagenesis of the kinase domain not only upregulated a set of genes at the RNA le v el but also induced posttranscriptional upregulation of a group of proteins, most of w hich directl y or indirectl y interact with CDK8 / 19. Interestingly, one of these proteins is TUT1 implicated in nucleolar integrity ( 53 ), a process that we found to be regulated by CDK8 via its interaction with p21 (CDKN1A) ( 54 ). The largest group of post-transcriptionally upregulated proteins comprises almost all the components of Mediator complex, with a notable exception of MED26, which is displaced from the Mediator by the Mediator kinase module ( 44 ). The increased le v els of Mediator, a coacti vator of transcription ( 55 ), can explain why the late response to CDK8 / 19 inhibition comprises upregulation of a relati v ely large set of genes. Furthermor e, this r esult can explain why CDK8 / 19 inhibition in leukemia cells increased the expression of genes associated with super-enhancers (which are characterized by increased Mediator binding), leading to leukemia suppression ( 13 ).
Remar kab ly, long-term (24 h) treatment of cells with a PKC agonist had a drastic effect on late-response genes that wer e upr egulated by CDK8 / 19 inhibition under basal conditions, as most of such genes were no longer induced by Senexin B in the presence of PMA (Figure 5 H). We hypothesize that this drastic change reflects PMA-induced chromatin rearrangement that includes redistribution of Mediator complexes, which regulate the genes that are upregulated by CDK8 / 19 inhibition. The association of Mediator with the late-response genes remains to be tested in future studies.

Dir ect and indir ect effects of CDK8 / 19 on protein phosphorylation and stability
Inhibition of Mediator kinase activity affected hundreds of phosphoepitopes. Remar kab ly, similar numbers of proteins showed either decreased or increased phosphorylation upon CDK8 / 19 inhibition, e v en after the shortest period of inhibitor treatment (3 h), indicating that many and probably most of the phosphoproteomic effects were indirect. Our phosphoproteomic analysis was not aimed at Nucleic Acids Research, 2023, Vol. 51, No. 14 7311 identifying direct Mediator kinase phosphorylation targets, but enriched motif analysis of the phosphoepitopes that were found here to be downregulated by CDK8 / 19 inhibition matches the previously identified CDK8 / 19 phosphorylation motifs ( 46 ), suggesting that many of these could be direct Mediator kinase substrates. Furthermore, a number of phosphoproteins affected by CDK8 / 19 inhibition in 293 cells (such as OGFR, MED14, RREB1, TP53BP1, NELFA, AFF4, BRD9, TAF10, CHD3 and STAT1) were previously identified as likely targets of Mediator kinase in HCT116 colon carcinoma ( 46 ). These phosphoepitopes could potentially be used as general markers of CDK8 / 19 activity in different cell types, but unfortunately antibodies specific to Media tor kinase-regula ted phosphoepitopes identified here are not currently available.
Post-transcriptional negati v e regula tion of Media tor kinase-interacti v e proteins by Mediator kinase activity seems unlikely to be exerted thr ough pr otein synthesis, gi v en the nuclear localization of CDK8 / 19. Alternati v ely, this effect could be mediated by the enhancement of protein degradation (which has not been directly tested in this study). We have asked if protein phosphorylation by CDK8 / 19 could be responsible for changes in the protein le v els, via stabilization or destabilization of the phosphorylated proteins. Consistently with this hypothesis, se v eral proteins (such as TP53BP1 and MED14) showed a strong decrease in phosphorylation after 3-hr treatment with a Mediator kinase inhibitor, followed by an increase in protein le v els at the 72-hr treatment point, whereas se v eral proteins (including MED13) showed the same direction of changes in their phosphorylation and expression upon CDK8 / 19 inhibition. On the other hand, most of the proteins that showed post-transcriptional regulation by Mediator kinase were not identified as CDK8 / 19-affected phosphoproteins, suggesting that their regulation is more likely to be mediated by differential pr otein-pr otein interactions of kinase-acti v e v ersus inacti v e CDK8 / 19.

DA T A A V AILABILITY
The RNA-Seq data are available in GEO database (accession numbers GSE154357, GSE149432, GSE101629, GSE199331, GSE200523, GSE200732, GSE200735, GSE200747, GSE221160, GSE200753 and GSE231891). The proteomics and phosphoproteomics data are available in MassIVE (accession number MSV000089494). All the source codes (R scripts) and raw data tables (both processed RNASeq and proteomics / phosphoproteomics data) used for data analysis and plotting in this manuscript are included in the supplemental file named 'Source Codes and RawData Files.zip'.