Transcription-Associated Cyclin-Dependent Kinases as Targets and Biomarkers for Cancer Therapy

The cyclin-dependent kinases (CDK) are a family of approximately 20 serine/threonine kinases that regulate fundamental cellular processes. They are broadly divided into two major subclasses: (i) cell cycle–associated CDKs (including CDK1, CDK2, CDK4, and CDK6) that directly regulate progression through the phases of the cell cycle and (ii) transcription-associated CDKs (including CDK7, CDK8, CDK9, CDK12, and CDK13). The transcription-associated CDKs regulate gene transcription by phosphorylating the carboxy-terminal domain (CTD) of the DNA-directed RNA polymerase II subunit (RPB1) of RNA polymerase II (RNA Pol II), as well as other targets. However, their precise mechanisms of action related to transcription remain relatively obscure (1). In addition, there remains a class of CDKs for which the underlying functions are largely unknown. Each of the CDKs is bound to a specific cyclin, which directs the activity of the CDK. Given that CDKs control processes critical for cancer cell survival and growth, they have been viewed as promising therapeutic targets. Indeed, multiple CDK inhibitors have been developed and tested in a number of cancer types (reviewed in ref. 1). Recently, inhibitors that target CDK4/CDK6 (e.g., palbociclib, ribociclib, and abemaciclib) have become widely used in hormone receptor–positive [i.e., estrogen receptor (ER) and/or progesterone receptor–expressing] breast cancer, and have shown impressive improvements in progression-free survival (PFS) and overall survival (OS; refs. 2–4). Moreover, inhibitors that target CDK1 and CDK2 are currently in clinical trials for multiple cancer types (5).

In contrast, the transcription-associated CDKs are less developed as therapeutic targets, and small-molecule inhibitors of transcription-associated CDKs have not yet entered routine clinical use. However, several recent studies implicate these CDKs in driving and maintaining cancer cell growth, particularly in cancers primarily driven by dysregulated transcription factors, such as those dependent on MYC (e.g., neuroblastoma) or the EWS–FLI1 fusion oncoprotein (Ewing sarcoma). Mounting evidence suggests that inhibiting this class of CDKs may have important therapeutic relevance. In addition, some transcription-associated CDKs, such as CDK12, are inactivated in ovarian and prostate cancers, supporting a tumor-suppressive role for these transcriptional CDKs. Here, we review our current understanding of the transcription-associated CDKs and highlight the genomic features of tumors carrying transcriptional CDK loss-of-function mutations and potential synthetic lethal approaches. We discuss the utility of using transcription-associated CDKs as biomarkers for targeted therapies, potential combination strategies with checkpoint immunotherapy, and the recent development of small-molecule inhibitors against transcriptional CDKs in various cancer types.

Each of the transcription-associated CDKs binds to an activating cyclin partner to regulate gene transcription (6). CDK7 (also known as CDKN7) is a 346 amino acid protein (Fig. 1A) that binds to cyclin H and the accessory protein MAT1 to function as a CDK-activating kinase (CAK), and has a general role in transcription: by phosphorylating the CTD of RNA Pol II, CDK7 regulates the initiation of transcription and promoter escape (7, 8). The CDK7/CAK complex, which associates with the core TFIIH complex, can activate CDK9 (PITALRE, CDC2L4, CTK1) by phosphorylating the threonine-186 residue within the activating T-loop. This cascade of events controls the switch from transcriptional initiation to elongation of RNA Pol II (9). Several recent reviews on have been published on the functions of CDK7 (10, 11).

CDK9, a 372 amino acid protein, binds to either cyclin T or cyclin K for its kinase activity (12, 13). The CDK9 complex with cyclin T, referred to as the positive transcription elongation factor b (P-TEFb), is a general transcription factor (GTF) that is required for efficient expression of most genes and phosphorylates the CTD of RNA Pol II, as well as the DRB sensitivity–inducing factor (DSIF) and negative elongation factor (NELF), to relieve promoter pausing and promote transcription elongation (14). CDK9 preferentially localizes to the nonnucleolar nucleoplasm with significant enrichment at nuclear speckles and is thought to also function in RNA processing and replication stress responses (15, 16). The structures of CDK7 and CDK9 have a typical kinase fold forming the amino terminal lobe (consisting of a β-sheet and α-helix), in addition to an α-helix–rich carboxy-terminal lobe (17).

CDK8, a 464 amino acid protein, associates with cyclin C, MED12, and MED13 to form a 600-kDa complex known as the CDK8 module. CDK8 can associate with the Mediator complex, which is a multimeric transcriptional coactivator complex that transmits signals from transcription factors to RNA Pol II (18–20). The CDK8/Cyclin C complex also targets CDK7/cyclin H via phosphorylation, repressing the ability of TFIIH to activate transcription and its CTD kinase activity (21). CDK8 has also been described to restrain activation of superenhancers, affecting global gene expression in a cell type–specific fashion (22), and mediates responses to serum stimulation, inflammation, and hypoxia (23–26). CDK19 is a paralog of CDK8 and also binds to cyclin C (27). CDK19 can also associate with the Mediator complex to regulate transcription in a gene-specific manner (25) and has kinase-independent roles in regulating the p53 stress response (28). A large-scale proteomics study recently identified more than 60 proteins phosphorylated by CDK8 and CDK19, many of which are associated with chromatin modification, DNA repair, and transcription (29).

CDK12 was discovered as a CDC2-related kinase with an arginine/serine rich (RS) domain (also known as CRKRS; ref. 30). The CRKRS/CDK12 gene encodes a protein of 1,490 amino acids and is one of the largest CDKs, encompassing a carboxy-terminal kinase domain, two proline-rich motifs (PRM) involved in protein–protein interactions, and an RS domain that is commonly found in splicing factors of the serine/arginine-rich family (ref. 30; Fig. 1A). Within the nucleus, CDK12 localizes in a speckled pattern, overlapping with spliceosome components (30). CDK12 is essential during embryonic development, and knockout of the gene in mouse embryonic stem (ES) cells leads to lethality shortly after implantation; Cdk12^−/− blastocysts fail to undergo outgrowth of the inner cell mass due to apoptosis and exhibit increased spontaneous DNA damage (31). The search for cyclin partners initially identified cyclins L1 and L2 as cognate cyclins for CDK12 (32), but cyclin K was later shown to be the bona fide CDK12-associating cyclin critical for its kinase activity (33–35).

The closest related CDK to CDK12 is CDK13 (also known as CDC2L5, CHED), which is also a large CDK consisting of 1,512 amino acids. CDK12 and CDK13 share 50% amino acid identity overall, having unrelated amino and carboxy-terminal domains. CDK13 contains a carboxy-terminal serine-rich (SR) domain and two alanine-rich (AR) domains, which are not found in CDK12 (ref. 36; Fig. 1A). However, the kinase domains of CDK12 and CDK13 are nearly 92% identical.

Structural studies of CDK12 and CDK13 have demonstrated that the amino-terminal lobe of the CDK12 kinase domain interfaces with the cyclin box of cyclin K, creating CDK12:cyclin K heterodimers, which then further dimerize (35). Similar heterodimers between CDK13 and cyclin K have also been observed (37). However, despite the similarities between CDK12 and CDK13, each seems to regulate the expression of a distinct set of genes (37, 38). These data suggest shared but nonoverlapping functions between CDK12 and CDK13.

In addition to the CDKs that we have described above, CDK10 and CDK11 also have presumed roles in transcription (39). CDK11 associates with splicing machinery as well as proteins involved in transcriptional initiation and elongation (40–42). Cyclin M is the cyclin partner for CDK10 (43), and CDK11 associates with L-type cyclins (40). Whereas CDK10 has been implicated as a tumor-suppressive kinase in estrogen-driven cancers (44), CDK11 is highly expressed in triple-negative breast cancer (TNBC), multiple myeloma, and liposarcoma (45, 46).

The transcription-associated CDKs play major roles in the multistep process of RNA Pol II transcription. Functionally, transcriptional CDKs and their cyclin partners control phosphorylation of the RNA Pol II CTD (Fig. 1B and C). The CTD is composed of multiple heptapeptide repeats (YSPTSPS) which can be phosphorylated (in humans, 52 repeats vs. in yeast, 26 or 27 repeats; ref. 47). The length of the CTD partially determines the efficiency of processing different pre-mRNA substrates (47), whereas the pattern of tyrosine, serine, and threonine phosphorylation dictates the transition between the initiation, elongation, and termination phases of transcription (48).

Preceding transcription is the assembly of the preinitiation complex (PIC), a complex of roughly 100 proteins that docks to the transcription start sites of genes and facilitates DNA entry into the active site of RNA Pol II for transcription to start (49). Formation of the PIC requires the recruitment of several GTFs, which include TFIIA, TFIIB, TFIID, TFIIE, TFIIF, and TFIIH (6, 50). The final GTF to be recruited to the PIC is TFIIH, which contains multiple subunits, including xeroderma pigmentosum type B (XPB) and CDK7 (51). Following recruitment, XPB functions as an ATP-dependent DNA helicase that enables promoter opening for transcription to occur (51, 52), whereas CDK7-mediated phosphorylation of the RNA Pol II CTD at Ser5 may be involved in TFIIH-mediated promoter escape (11, 53). However, the precise role CDK7 plays in transcription remains controversial. For example, it was recently shown that selective inhibition of CDK7 with YKL-5-124 does not significantly downregulate phosphorylation of the CTD or global gene expression, but instead primarily inhibits E2F-driven gene expression and causes G₁–S cell-cycle arrest (54). In addition, CDK7 may also indirectly regulate transcription by phosphorylating and regulating transcription factors (TF), including the ER and androgen receptor, two TFs that play critical roles in hormone-driven cancers such as breast and prostate cancers (49–51).

Following initiation, RNA Pol II enters transcriptional pausing, during which NELF and DSIF are loaded onto RNA Pol II (48). Transcriptional pausing ensures gene-specific regulation, RNA quality control, and 5′ mRNA capping by the human capping enzyme. The P-TEFb/CDK9 complex is then recruited to paused RNA Pol II and cooperates with bromodomain-containing protein 4 (BRD4) and the super elongation complex to release RNA Pol II for active transcription (Fig. 1B). In this process, CDK9 phosphorylates the CTD, as well as DSIF and NELF, to facilitate transcriptional elongation and recruitment of the 3′-end processing and splicing factors necessary for mRNA maturation (55, 56). Several TFs (such as MYC) have been shown to recruit P-TEFb/CDK9 to promote transcription of their targets (57). CDK9 also binds to other transcriptional mediators, such as the recombining binding protein suppressor of hairless (RBPJ) to target gene promoters, which maintains tumor-initiating cells in glioblastoma (58) and phosphorylates other important targets, including the exoribonuclease XRN2, to regulate transcription termination (59).

In addition to CDK7 and CDK9, CDK12 and CDK13 also phosphorylate the CTD at Ser2 and Ser5 in vitro and preferentially phosphorylate the CTD when it is prephosphorylated at Ser7 (refs. 30, 35, 37; Fig. 1B). However, there are conflicting data as to whether knockdown of CDK12 or CDK13 globally decreases Ser2 and Ser5 CTD phosphorylation (38, 60, 61). Specific sets of genes may be more sensitive to genetic CDK12 inhibition, resulting in decreased Ser2 phosphorylation only at some gene loci (62), whereas chemical inhibition of CDK12 and CDK13 appears to more globally dampen Ser2 phosphorylation (60). In addition to the CTD, CDK12 and CDK13 phosphorylate other substrates, which may be critical for their functions in both transcription and other processes; these targets have yet to be fully characterized but phosphoproteomics and affinity-purification mass spectrometry studies have identified multiple pre-mRNA processing and RNA-splicing factors as potential kinase substrates and protein-binding partners, although the exact consequences of phosphorylation remain unknown (38, 63).

Interestingly, CDK12 depletion does not affect transcription globally, but instead alters a select subset of genes involved in the DNA damage response (DDR) and DNA replication (61, 64). Blazek and colleagues showed that loss of CDK12 decreases expression of predominantly long genes (spanning >10 kb) with high numbers of exons, which includes those involved in regulating genomic stability such as BRCA1, ATR, and FANCD2 (64). Loss of CDK12 results in reduced transcription of the BRCA1 gene and increases sensitivity to DNA-damaging agents (64). Moreover, mutations in the CDK12 kinase domain lead to an impaired ability to repair DNA double-strand breaks via homologous recombination (HR; refs. 64, 65). Recently, George and colleagues demonstrated that inhibiting CDK12 and CDK13 using a small-molecule inhibitor, THZ531, results in gene length–dependent elongation defects, and induces premature cleavage and polyadenylation of DDR genes, leading to decreased DDR gene expression (63). CDK12 may also regulate other CDKs, as disruption of Cdk12 in knockout mice and P19 mouse teratocarcinoma cells reduces Cdk5 expression (66). Consistent with the notion that CDK12 selectively regulates a subset of genes, studies in Drosophila cells showed that CDK12 is not required to maintain overall transcription, but is required for the expression of NRF2-dependent stress-response genes (67). Furthermore, gene expression changes after knockdown of CDK13 or CDK12 are markedly different, with enrichment of growth signaling pathways after CDK13 loss (37). In addition, although both CDK12 and CDK13 directly interact with splicing machinery, they affect the processing of different sets of genes and noncoding RNAs (38). The basis for this specificity remains poorly understood.

In addition to regulating the elongation phase of RNA Pol II, CDK12 also regulates transcriptional termination. CDK12 deficiency reduces Ser2 phosphorylation of the CTD as well as levels of cleavage stimulation factor 64, a regulator of polyadenylation, at the 3′ end of the c-FOS gene; this results in impaired 3′-end processing and expression (68). Taken together, these studies suggest that CDK12 plays a role in transcription termination (Fig. 1B); impairing this function could result in the activation or loss of key oncogenes or tumor suppressor genes, respectively, during tumorigenesis.

Beyond regulating RNA Pol II phosphorylation, the transcription-associated CDKs play other important roles (Fig. 2). Because CDK12 colocalizes with SC35 (also known as SRSF2 or SFRS2; ref. 30), a component of the spliceosome, it is thought to play a role in RNA splicing. CDK13 is also enriched in nuclear speckles, where the PRM domain interacts with RNA-binding proteins within peri-nucleolar structures (69). The smaller transcriptional CDKs (i.e., CDK7, CDK8, and CDK9), which lack these additional domains, likely do not participate in this function. Multiple mass spectrometry studies have shown that splicing machinery proteins are associated with CDK12 and CDK13, including SRSF1 (38, 68–70). Support for a role of CDK12 in splicing was further provided from studies in breast cancer cell lines, where analysis of mRNA transcripts demonstrated that CDK12 regulates alternative last exon (ALE) splicing of a subset of genes, including those involved in the DDR, to generate different mRNA isoforms in their 3′ ends (70). Despite the evidence showing that CDK12 and CDK13 are involved in splicing, the precise mechanisms are not well understood. It remains unclear how gene selectivity is achieved and what other protein partners are involved in this process.

Recent evidence suggests that in addition to a role in splicing, CDK12 may also be critical for regulating intronic polyadenylation (IPA; Fig. 2). Polyadenylation is the process by which a poly(A) tail is added to the 3′ ends of eukaryotic mRNAs. This process involves endonucleolytic cleavage of the transcript followed by sequential addition of adenosine residues, which is critical for nuclear export and stability of the transcript, as well as for efficient translation into protein (71). Interestingly, transcriptome sequencing has revealed that many genes can have more than one poly(A) site; similarly to ALE splicing, differential polyadenylation thus allows cells to generate diverse transcript isoforms with different 3′ ends, a process termed alternative polyadenylation (APA; ref. 71). A recent study in patients with chronic lymphocytic leukemia (CLL) uncovered widespread upregulation of truncated mRNAs and proteins that were generated by increased IPA, suggesting that this is an epigenetic mechanism by which cancer cells can remodel their transcriptomes and proteomes without altering their genomes (72). Truncated mRNAs were found predominantly within tumor-suppressor genes, and the isoforms generated by IPA usage lacked the normal tumor-suppressive functions found in the corresponding full-length isoforms. Strikingly, some of the isoforms even gained de novo oncogenic potential, highlighting that global IPA changes within a cell might be a newly recognized driver of tumorigenesis (72).

Dubbury and colleagues recently showed that CDK12 has an important role in globally suppressing IPA usage to enable production of full-length gene products (62). Using a genetic Cdk12 knockout model in mouse ES cells, the investigators demonstrated that Cdk12 loss results in the loss of HR repair genes and negatively affects transcription elongation dynamics. Cdk12 loss also increased expression of p53 target genes, consistent with the induction of DNA damage. Interestingly, although ALE splicing was not dramatically altered genome-wide upon Cdk12 loss, isoform differences from APA usage were identified. In particular, proximal IPA sites were significantly increased in HR genes (e.g., Brca1), with a concomitant decrease in distal polyadenylation sites. This was, at least in part, due to the significantly more IPA sites in HR genes compared with other expressed genes. Interestingly, tumors harboring CDK12 loss-of-function mutations also had increased IPA sites in HR genes, suggesting that the cumulative effect results in a functional HR-deficient phenotype, which may have therapeutic implications (see below). The increase in global intronic polyadenylation can also be recapitulated using THZ531 in neuroblastoma cells and preferentially affects long genes involved in DDR (63). Therefore, CDK12 has a further mechanism of regulating gene expression, namely by suppressing IPA and promoting distal (full-length) isoform expression, particularly in long HR genes (Fig. 2).

Dynamic regulation of chromatin structure can significantly alter gene expression. Two basic forms of chromatin exist in eukaryotes: Euchromatin regions are generally associated with open chromatin configurations and contain transcriptionally active genes, whereas heterochromatin regions are typically less accessible to the transcriptional machinery due to the highly compacted structure. Loss of Cdk12 in Drosophila results in the ectopic accumulation of HP1 on euchromatin regions, which leads to downregulation of target genes (73). Interestingly, the heterochromatin enrichment mainly occurs within long genes, and results in decreased transcription of neuronal genes involved in courtship learning (73). This suggests that CDK12 regulates the conversion of euchromatin to heterochromatin, highlighting another distinct mechanism that CDK12 regulates in gene expression. Future work investigating whether this also occurs in human cancer cells will be important. Furthermore, CDK7 has also been shown to regulate transcription-associated chromatin modifications. CDK7 stimulates the methyltransferase activity of SETD1A/B to regulate H3K4me3 downstream of the transcriptional start site, and may indirectly regulate H3K36me3 positioning through its effects on CDK9 and SETD2 (74, 75). Through these mechanisms, CDK7 appears to dually regulate transcriptional processing and gene methylation, suggesting that dysregulation of CDK7 might lead to heritable changes in gene expression.

Beyond regulating transcription and epigenetics, some of these transcription-associated CDKs may play additional roles in regulating protein expression at the translational level. Choi and colleagues recently described a role for CDK12 in controlling translation of a subset of mRNAs involved in DNA repair, translation regulation, and mitosis (76). For example, CDK12 phosphorylates the translation repressor 4E-BP1, which releases it from eIF4E at the 5′ cap of mRNAs, enabling expression of mTOR complex 1 (mTORC1) targets. Genome-wide ribosome profiling identified several specific CDK12 “translation-only” target mRNAs (76). These data suggest that in addition to the DDR and stress response genes that require CDK12 for mRNA expression and IPA suppression, another set of targets relies on CDK12-mediated translational control, which may affect the maintenance of genome stability. Whether CDK13 or other transcription-associated CDKs also regulate translation is currently unclear. Nonetheless, taken together, it is becoming increasingly evident that the transcription-associated CDKs regulate gene expression not only by phosphorylating RNA Pol II, but also by modifying chromatin structure and controlling translation, thus expanding their functional repertoires (Fig. 2).

Genomic alterations affecting CDK7, CDK8, CDK9, or CDK13 are relatively uncommon in human cancers. The association of these kinases to tumor development has generally been through large-scale genetic screening approaches or hypothesis-driven investigation, rather than direct observation of recurrent somatic events in human cancers (77–79). However, copy-number changes as manifested by gene amplifications as well as deletions in these transcription-associated CDKs can be found, albeit rarely in human cancers (Fig. 3A). For example, the genomic locus harboring CDK8 is gained or amplified in roughly 20% of colorectal adenocarcinomas, where CDK8 expression has been linked to β-catenin signaling (78). CDK12 has also been found to be coamplified with the ERBB2/HER2 oncogene in subsets of breast cancer (80). Interestingly, deletions and mutations in the cyclin-binding partners for these CDKs have also been identified in human cancers. For example, mutations in cyclin C (encoded by the CCNC gene) are found in a subset of acute lymphoblastic leukemias (ALL; refs. 81, 82). Cyclin C, which binds to CDK8 as part of the Mediator complex, functions as a haploinsufficient tumor suppressor by controlling Notch1 levels, and loss of CCNC accelerates tumor development in a mouse model of T-cell ALL (T-ALL; ref. 82). The relative scarcity of recurrent somatic alterations, either activating or inactivating, observed in these kinases may be related to their dramatic influence on transcriptional activity of large numbers of downstream targets, which may both positively and negatively regulate cell growth, depending on the cellular context. However, CDK12 is exceptional in this regard: Of the transcription-associated CDKs, it is the only one where inactivating somatic alterations (i.e., frameshift, nonsense, and missense mutations) have been recurrently observed in prostate and high-grade serous ovarian carcinomas (Fig. 3B and C) and where gene inactivation is linked to unique genomic alterations.

Numerous studies have identified mutations in the CDK12 gene in human cancers. For example, in The Cancer Genome Atlas (TCGA) analysis of 489 high-grade ovarian adenocarcinomas, CDK12 was found to be one of 9 recurrently mutated genes, occurring in 3% of primary ovarian tumors (83). BRCA1, BRCA2, and CDK12 mutations are mutually exclusive (84–86), but despite the mounting evidence that CDK12 regulates the expression of HR genes, the genomic signature of CDK12-mutated tumors is unique from those with BRCA1 or BRCA2 mutations (87, 88). This suggests that although CDK12 loss may share overlapping features with BRCA1 or BRCA2 loss, our understanding is likely incomplete. Elucidating the roles of CDK12 in DNA replication may be important in understanding the genomic signatures seen in CDK12-mutated tumors.

Genomic studies have uncovered tumors with a distinctive tandem duplicator phenotype (TDP), whereby tumors harbor hundreds of small copy-number gains, without an obvious selection for a single coding or noncoding location (89–92). Tandem duplications (TD) are a structural rearrangement that produces physically contiguous, head-to-tail duplications of a segment of DNA. In contrast to tumors that develop focal DNA copy-number gains targeting driver genes such as ERBB2/HER2, the apparent lack of selective pressure targeting a single locus in tumors with the TDP suggests a systemic etiology rooted in a defect in DNA repair or replication. Interestingly, tumors with the TDP lack germline or somatic inactivation of BRCA2, indicating another unknown defect was responsible.

Using retrospective analysis of copy-number data from high-grade serous ovarian carcinomas (HGSOC), Ng and colleagues showed that approximately 12% of HGSOC was characterized by the TDP phenotype (90). CDK12 inactivation was associated with 200 to 800 DNA copy-number gains per tumor, and the distribution of the TD sizes was bimodal, with modes of 300 kb and 3 Mb (93). Some cases showed an acquired somatic mutation in CDK12 and loss of heterozygosity of the remaining functional allele (93). Subsequently, Menghi and colleagues reanalyzed 992 TCGA genomes and formally reidentified the TDP by statistical analysis, reporting TDs in three recurrent, narrowly defined duplication spans present in 12% of TCGA tumors (94). The TDP occurs in up to 50% of triple-negative breast adenocarcinoma, ovarian carcinoma, and endometrial carcinoma; 10% to 30% of adrenocortical, esophageal, stomach, and lung squamous cell carcinoma, and less frequently in other tumor types (94).

Interestingly, in localized prostate cancer, both alleles of CDK12 are inactivated in approximately 1% to 2% of cases (89, 95). However, more than 5% of advanced prostate adenocarcinomas harbor biallelic CDK12-inactivating mutations, emphasizing that advanced prostate cancer is a genomically distinct entity from primary disease (96). Loss of CDK12 in prostate cancer is also associated with features of more aggressive disease, including higher Gleason score and shorter time to developing castration resistance and metastasis (97). Focusing on CDK12 in metastatic castration-resistant prostate cancer (mCRPC), Wu and colleagues reported biallelic inactivation of CDK12 in 7% of mCRPC cases (98). CDK12 inactivation in this setting was also associated with a TDP, with a bimodal size distribution similar to that reported in ovarian carcinoma (modes of 400 Kb and 2.4 Mb). In contrast to HGSOC, where a majority of CDK12-inactivated tumors were tetraploid, and single mutations with loss of heterozygosity was observed, a typical CDK12-inactivated mCRPC genome was diploid and frequently harbored two distinct inactivating CDK12 somatic alterations. CDK12 inactivation was mutually exclusive with other established mCRPC genomic subclasses such as BRCA2 mutation, activating ETS-family gene fusions, or mutations in SPOP (98). This study was complemented by whole-genome studies of mCRPC that noted that whereas BRCA2 inactivation produced both a characteristic structural signature of frequent deletions and a nucleotide mutation signature, biallelic CDK12 inactivation was associated only with a structural TDP signature (87, 88).

The mechanism by which these CDK12-associated TDs are generated remains largely unknown. The cell-cycle CDKs in general prevent reinitiation of replication origins through multiple overlapping mechanisms, so that the genome is replicated only once in each cell cycle (99). Deregulation of the replication initiation proteins MCM2-7 and CDC6 and induction of rereplication aberrantly in yeast are sufficient to start the initial steps of TD formation (100). Rereplication-induced gene amplifications appear to be mediated by nonallelic homologous recombination between repetitive elements. Interestingly, cyclin K and CDK12 promote prereplication complex (pre-RC) assembly in G₁ by phosphorylating and restricting cyclin E1 activity, and knockdown of cyclin K or CDK12 prevents assembly of the pre-RC (101). One possibility is that loss of CDK12 causes dysregulation of replication origins, leading to TDs that accumulate over cell divisions. Interestingly, CDK12 was recently shown to be critical for proper chromosome alignment and progression through mitosis by regulating mitotic regulators such as the structural maintenance of chromosome complexes, centromere proteins (CENP), and the kinetochore protein NDC80 (76). Furthermore, BRCA1 (but not BRCA2) was recently shown to suppress TDs, and microhomology-mediated TDs of approximately 10 kb are commonly found in BRCA1-mutated cancers (102). Willis and colleagues showed that in the setting of BRCA1 loss, TDs might arise by a replication restart–bypass mechanism terminated by end joining or by microhomology-mediated template switching, the latter leading to the formation of complex TD breakpoints (102).

There are conflicting reports on whether the TDP preferentially targets specific loci for gene amplification or consistently disrupts tumor suppressors. The balance of evidence suggests that although the TDP affects the entire genome at random, variants in cells that inactivate tumor suppressors or amplify relevant drivers will be selected and expanded. Menghi and colleagues demonstrated that the smaller TDs associated with functional inactivation of BRCA1 and TP53 were more likely to disrupt genes by double transection within the coding region, whereas longer duplications such as those associated with CDK12 were more likely to amplify coding and noncoding gene-regulatory elements (103). Wu and colleagues observed TD-mediated amplification of the driver genes MCM7, RAD8A, CDK18, and CCND1 more frequently than expected by chance, with a dose-dependent increase in CCND1 expression (98). Gene rearrangement may also be an important consequence of the TDP, as the TDP in mCRPC was associated with an increase in TD-associated fusion events, whereas in CDK12-intact tumors, fusions typically derive from translocations or large intragenic deletions (98).

Synthetic lethality has been utilized as a strategy to target cancers with germline or somatic loss-of-function mutations in BRCA1 or BRCA2. Indeed, PARP inhibitors (PARPi) are FDA-approved drugs that target tumors with defects in HR, including those with BRCA1 or BRCA2 mutations (104, 105). Given the success of this approach to target molecular subsets of cancers, this therapeutic strategy has been extended beyond BRCA and PARP, with the hope of targeting additional tumors characterized by the loss of tumor-suppressor genes. However, with the exception of CDK12, deleterious loss-of-function mutations in the other transcription-associated CDKs have not been described in cancer.

A number of studies have now suggested that CDK12 mutation or deficiency may lead to sensitivity to PARPi and platinum chemotherapy as well as agents that target cell-cycle checkpoints such as CHK1 (refs. 84, 106, 107; Fig. 4). For example, a high-throughput genome-wide short-hairpin RNA (shRNA) screen previously identified CDK12 as one of the most significant genes enhancing sensitivity to PARPi when depleted (84). Given the role of CDK12 in the maintenance of genomic stability, it seems likely that PARPi target CDK12-mutant tumors by further interfering with DNA repair. However, the possibility that PARPi block the transcriptional function of PARP1, which may be critical in the absence of CDK12, may also be important. Indeed, PARP1 was previously shown to regulate androgen receptor transcriptional activity and function (108). In addition, data from TCGA showed that patients with CDK12 mutations have better responses to platinum-based chemotherapy (84). Joshi and colleagues also demonstrated increased sensitivity to platinum and the PARPi veliparib in ovarian cancer cell lines in which CDK12 was silenced. Several CDK12 point mutations within the kinase domain disable the kinase activity of CDK12, suggesting that this domain is important for its tumor-suppressive function (106).

CDK12-deficient cells also appear to depend on the S-phase checkpoint kinase CHK1, and CHK1 inhibitors can selectively kill CDK12-deficient cells regardless of TP53 status (107). The mechanism may involve CDK12 regulation of CHK1 itself; CDK12-mutant ovarian cancers appear to have reduced CHK1 expression, which may underlie their sensitivity to CHK1 inhibition.

In breast cancer, ERRB2/HER2 is an oncogene that is frequently amplified and overexpressed, and can be therapeutically targeted with the antibody trastuzumab. Because the CDK12 gene is located close to the HER2 gene at chromosome 17q12-q21, it is often coamplified with ERRB2/HER2 and therefore may also be overexpressed. Naidoo and colleagues assessed the frequency of CDK12 overexpression by IHC in breast cancer. Twenty-one percent of primary unselected breast cancers were CDK12-high, and the protein was absent in approximately 10% of cases. As might be expected, CDK12 expression correlated with HER2 expression and the absence of CDK12 expression correlated with the TNBC phenotype and reduced expression of a number of DNA-repair proteins (80). Proteomic analysis of breast cancers demonstrated that amplification of CDK12 is associated with enhanced CDK12 phosphorylation, suggesting that CDK12 may be an important therapeutic target in HER2-amplified breast cancers (109, 110). However, functional validation in tumor cell lines overexpressing CDK12 is required to demonstrate dependency on CDK12. Complicating matters, in some cases, the CDK12 gene may become rearranged within the HER2 amplicon, potentially leading to the loss of CDK12 function (111). In addition, a CDK12 gene fusion has been described in a rare micropapillary form of breast cancer (111). Therefore, whether CDK12 gene amplification is an oncogenic driver (or simply a passenger) and whether this genetic alteration has any therapeutic significance remains unclear at the moment.

There has been recent interest in using CDK12 mutational status as a biomarker for PARPi response (Table 1). TRITON2 (NCT02952534) is a phase II study of rucaparib in patients with mCRPC with HR gene alterations. The preliminary results were recently reported, which included 13 patients with CDK12 alterations, in addition to 45 patients with BRCA1/BRCA2 alterations and 18 patients with ATM alterations (112). In the patients with CDK12 mutations, only 1 of 13 patients had a confirmed prostate-specific antigen (PSA) response (defined as a sustained 50% decrease in baseline PSA), and no patients had an objective response as defined by RECIST. The exact CDK12 mutations were not reported, and will be critical in determining how mutations in different domains of CDK12 may affect sensitivity to PARPi.

In addition to TRITON2, another trial, BRCAAway, is testing abiraterone (an inhibitor of androgen synthesis) versus olaparib (a PARPi) versus the combination of abiraterone plus olaparib in patients with mCRPC. This trial will include a cohort of patients with CDK12 mutations (NCT03012321). A trial testing the combination of olaparib and copanlisib (a PI3K inhibitor) plus an anti–PD-1/PD-L1 checkpoint inhibitor in molecularly selected patients will soon open (NCT03842228). One cohort will also include patients harboring CDK12 mutations. Finally, a phase Ib trial is testing the combination of rucaparib plus irinotecan in patients with DNA-repair mutations, including CDK12 (NCT03318445; Table 1). It remains to be determined whether the laboratory data in cell lines showing PARPi sensitivity in the setting of CDK12 loss will yield meaningful clinical advances in CDK12-mutated cancers.

Given the focal TDs and genomic instability seen within CDK12-mutated prostate cancer genomes, it was hypothesized that these TDs might generate neofusion peptides that could stimulate and be recognized by the immune system. Wu and colleagues demonstrated that the genomic signature of CDK12 deficiency in advanced prostate cancers was different from that generated by either BRCA2 loss or mismatch repair deficiency (MMRD; ref. 98). They showed that TDs in coding regions generated a large number of gene fusions that could potentially function as neoantigens. Although prostate cancer is generally thought of as a immunologically “cold” tumor type (with the exception of prostate cancers harboring MMRD and microsatellite instability), prostate cancers harboring CDK12 mutations had a significantly increased number of T-cell infiltrates and expanded T-cell clonotypes, similar to MMRD prostate cancers (98). Consistent with the hypothesis that CDK12-deficient prostate cancer might be more immunogenic, in a preliminary cohort of 4 patients with mCRPC, 50% of patients with CDK12 mutations responded to an anti–PD-1 checkpoint inhibitor. These data suggest that the genomic instability resulting from CDK12 deficiency leads to an increase in neoantigens, and that this subtype of cancer may benefit from immune-checkpoint inhibition (ref. 113; Fig. 4). This hypothesis is being tested in multiple clinical trials, including a phase II study of ipilimumab (targeting the CTLA4 checkpoint) plus nivolumab (targeting the PD-1 checkpoint) in patients with CDK12-mutated metastatic cancers (NCT03570619; Table 1). Moreover, whether the combination of immunotherapy with a PARPi might be efficacious in CDK12-deficient cancers awaits further investigation.

Given the critical roles that transcription-associated CDKs play in regulating gene expression, they have emerged as putative targets in cancer and other diseases (reviewed in ref. 57). The recognition that certain tumors are dependent on persistent transcription for oncogenesis (i.e., “transcriptionally addicted” cancers) has fueled interest in directly targeting these CDKs. Moreover, increased global transcription is a putative mechanism for oncogene-induced DNA damage, suggesting that high levels of transcription may be linked to genomic stability (114).

Gray and colleagues initially discovered and characterized a covalent CDK7 inhibitor called THZ1, which targeted a remote cysteine residue located outside the canonical kinase domain, providing some selectivity for CDK7, although THZ1 also has weak activity against CDK9, CDK12, and CDK13 (115). Recently, Gray and colleagues described a more potent and selective covalent CDK7 inhibitor, YKL-5-124. Interestingly, although treatment with YKL-5-124 led to a G₁–S cell-cycle arrest, there was little effect on RNA Pol II phosphorylation status, suggesting some functional redundancy among the transcriptional CDKs (54). A subset of cancer cell lines, including human T-ALL, demonstrated remarkable sensitivity to CDK7 inhibition, in part by affecting RUNX1 transcription (115). The potential to disrupt transcriptionally addicted cancers via CDK7 inhibition was later supported by studies showing that MYCN-amplified neuroblastoma cells and MYC/MYCL–amplified small-cell lung cancer (SCLC) cells were also sensitive to CDK7 inhibitors, leading to tumor regression in mouse models (77, 116). This was mediated, at least in part, by downregulating MYC-driven transcription and correlated with preferential downregulation of superenhancer-associated genes, indicating that CDK7 inhibitors can target the mechanisms that promote global transcription amplification and might be a useful strategy for targeting transcription factor oncoproteins (116). THZ1 also inhibits androgen receptor signaling in CRPC cells, reverses hyperphosphorylation of MED1 (which is associated with a drug-resistant phenotype), and induces tumor regression (117). Moreover, THZ1 and THZ2, an analogue with improved pharmacokinetic properties and longer plasma half-life, were also efficacious in targeting transcriptional addiction in TNBC cells (79). THZ2 inhibited the growth of TNBC patient-derived xenograft (PDX) models by targeting a cluster of transcription factor networks required for TNBC survival, including FOXC1, MYC, and SOX9. shRNA and CRISPR/Cas9–mediated silencing in TNBC cells confirmed a high dependency on CDK7 as well as CDK9 but not CDK12 or CDK13 (79). In addition, chordoma cell lines, which are a primary bone cancer with a dependency on transcription factor T (brachyury; TBXT) were sensitive in vitro and in vivo to inhibitors targeting CDK7, as well as CDK9, CDK12, and CDK13 (118). Taken together, these studies demonstrate that cancers dependent on oncogenic transcription factors and their downstream networks can be therapeutically targeted by CDK7 inhibitors. Several trials are currently testing CDK7 inhibitors in advanced solid malignancies (Table 2).

The mediator-associated kinases CDK8 and CDK19 may also be important drug targets, although the oncogenic and tumor-suppressive functions of CDK8 appear to be context-dependent. In acute myeloid leukemia (AML), CDK8 restrains key superenhancer-associated genes involved in growth inhibition. Cortistatin A selectively inhibits CDK8 (IC₅₀ = 12 nM) and CDK19 (which are 94% identical in the catalytic domain), but not the other transcription-associated CDKs (22). Inhibition of CDK8 and CDK19 upregulates transcription of genes that inhibit AML cell proliferation. Importantly, in leukemia mouse models, cortistatin A has antitumor effects (22). CDK8 also plays a role in the hypoxic response and is a critical downstream mediator of hypoxia-inducible factor 1 (HIF1A) target gene expression (23). In addition, CDK8 promotes transcription of genes involved in glycolysis, and inhibition of CDK8 reduces glucose uptake and sensitizes cancer cells to glycolysis inhibitors (119).

In colon cancer, CDK8 is located in a region of the genome that shows recurrent copy-number gains, and is an oncogene that acts in part by regulating β-catenin activity and WNT signaling (78). CDK8 expression is associated with poor clinical outcomes in colon cancer (120). In addition, CDK8 protects β-catenin–dependent transcription from inhibition by the E2F1 transcription factor, which antagonizes the WNT pathway (121). However, conflicting data exist on whether CDK8 is essential for survival, as genetic ablation has little effect on proliferation in cell-line models of colon cancer, calling into question the observations from the initial shRNA screen (122). Regardless, inhibitors of CDK8 and, to a lesser extent, CDK19 have been developed. Senexin A inhibits CDK8 with an IC₅₀ = 0.28 μM, although activity against other CDKs was not reported (123). This agent inhibits β-catenin transcriptional activity, and a derivative, Senexin B, suppresses estrogen-dependent transcription in ER-positive breast cancer (124). Other compounds targeting CDK8 and CDK19 have also been generated (122, 125), although as a class, dual CDK8/19-targeting compounds may be too toxic at therapeutic dose levels (126). Combinations of CDK8 inhibitors with drugs targeting orthogonal pathways, for example, glycolysis, may generate synergy and allow for more tolerable levels of CDK8 inhibitors (119). Taken together, although the evidence shows that CDK8 and CDK19 may be important targets, efforts to further develop these molecules clinically will likely require improved targeting approaches and a better understanding of whether pharmacologic inhibition will lead to cell death. SEL 120, a CDK8/19 inhibitor, is currently in early-phase clinical trials for hematologic malignancies (Table 2).

One of the first-generation CDK inhibitors, flavopiridol (alvocidib), was a nonselective CDK inhibitor that most potently inhibited CDK9, with a reported K_i of 3 nM against human cyclin T1-CDK9 in vitro (14, 127). In addition, flavopiridol also inhibited other transcription-associated CDKs including CDK7, as well as CDK1, CDK2, CDK4, and CDK6 with varying potencies (1). Despite promising in vitro activity, there was limited in vivo activity except in hematologic malignancies such as mantle cell lymphoma and CLL (1, 128, 129). In addition, nonselective CDK inhibitors suffer from narrow therapeutic windows causing adverse effects including bone marrow suppression, nausea, and gastrointestinal effects. Subsequently, development of flavopiridol was halted.

Dinaciclib (SCH727965) was initially developed as a more potent follow-up molecule to flavopiridol (130). It is a single-digit nanomolar inhibitor of CDK1, CDK2, CDK5, and CDK9 but also shows activity against CDK12 and CDK13. Dinaciclib inhibits cell-cycle progression in >100 cell lines tested (130) and clinical trials are ongoing, although preliminary results are less encouraging than initially anticipated (131). Interestingly, dinaciclib also promotes immunogenic cell death and enhances anti–PD-1 checkpoint blockade by activating dendritic cells and increasing T-cell infiltrates, suggesting a potential rationale for combining dinaciclib with anti–PD-1 checkpoint inhibitors (132). Additional preclinical and clinical studies will be required to determine whether this combination is tolerable and leads to improved tumor control.

The search for more specific CDK9 inhibitors continues, because it remains a promising target given its central role in transcription elongation (133). For example, CDK9 was identified in a screen of more than 1,500 kinase inhibitors that could specifically kill BRD4–NUT-rearranged NUT midline carcinoma (NMC) cells (134). NMC cells depend on CDK9 and cyclin T1 expression, and their inhibition induces apoptosis. A highly selective and potent CDK9 inhibitor (i-CDK9) was recently identified that exhibits more than 600-fold selectivity toward CDK9 (135). In addition, CDK9 inhibition with the kinase inhibitor PIK-75 in AML cells represses MCL1, a key AML survival factor. PIK-75, which also inhibits the p110α isoform of PI3K, reduces leukemia burden in vivo (136). Pharmacologic strategies to degrade CDK9 have also been explored. These compounds consist of a CDK-binding ligand (SNS-032) linked to a thalidomide derivative that binds to the E3 ubiquitin ligase Cereblon to promote degradation (137). Finally, a highly selective CDK9 inhibitor, MC180295 (IC₅₀ of 5 nM), was recently shown to have broad anticancer effects in vitro and in vivo. Interestingly, the mechanism of tumor suppression was not attributed to inhibition of transcription elongation, but instead to reactivating epigenetically silenced genes, thereby restoring tumor suppressor gene expression and cell differentiation (138). CDK9 inhibition decreased phosphorylation of the SWI/SNF protein BRG1, and led to genome-wide epigenetic derepression, suggesting that CDK9 is a novel epigenetic target (138). Moreover, CDK9 inhibition also activated an IFN response, as well as endogenous retroviruses and increased sensitivity to immune checkpoint blockade (138). Several clinical trials testing CDK9 inhibitors are currently being conducted in advanced solid and hematologic malignancies (Table 2). As CDK9 inhibition may reactivate tumor-suppressor genes and induce cellular immune responses that increase sensitivity to checkpoint inhibition, the development of additional, more specific CDK9 inhibitors and combination therapies will be of significant interest.

In addition, CDK12 has also attracted considerable attention as a drug target. CDK12 is critical in Ewing sarcoma, which is sensitive to CDK12 inhibitors and appears to synergize with PARPi (139). The tumor-specific expression of the EWS–FLI oncofusion mediates the sensitivity to the combination. Although the mechanism for this sensitivity remains to be fully elucidated, it may relate to the preexisting sensitivity of Ewing sarcoma to DNA-damaging agents such as doxorubicin and etoposide, which are used to treat Ewing sarcoma. It remains to be seen whether simultaneous, systemic suppression of CDK12 and PARP is a tolerable combination in patients; bone marrow toxicity has been a significant issue when combining many PARPi with cytotoxic chemotherapy regimens (104). In addition, CDK12 inhibitors might also have activity against another type of sarcoma: osteosarcoma—a disease with limited therapeutic options (140).

A synthetic lethal interaction between CDK12 and MYC was previously demonstrated by Grandori and colleagues, who performed a synthetic lethal siRNA screen of approximately 3,300 druggable genes using fibroblasts overexpressing MYC, and identified CDK12 as a synthetic lethal target (141). In addition, ovarian cancer cells dependent on MYC are sensitive to CDK12 inhibition. Treatment of 11 PDX models derived from patients with heavily pretreated ovarian cancer with THZ1 suppressed tumor growth and abrogated MYC expression. Interestingly, MYC downregulation requires the combined inhibition of CDK7, CDK12, and CDK13 in these ovarian cancer models (142). In osteosarcoma, the sensitivity to CDK12 inhibitors may also be explained by MYC levels (140). Targeting CDK12 with compound 919278 also inhibits osteosarcoma cells by transcriptionally regulating components of the noncanonical NFκB signaling pathway (143). Taken together, these data support the notion that in cancers dependent on certain transcription factors (i.e., MYC, EWS–FLI1, NFkB), targeting CDK12 and other transcription-associated CDKs may be a viable strategy. However, specifically targeting CDK12 has been challenging due to the similarities with other CDKs, especially CDK13; the kinase domains of CDK12 and CDK13 are 92% identical, making it difficult to generate specific CDK12 or CDK13 kinase inhibitors through conventional routes.

THZ531 is a THZ1 derivative with 50-fold greater potency against CDK12/CDK13 than CDK7 or CDK9 (e.g., CDK12 IC₅₀ = 158 nM vs. CDK7 IC₅₀ = 8,500 nM; ref. 60). However, THZ531 is subject to drug efflux through upregulation of the ABCB1 and ABCG2 transporters, a potential resistance mechanism (144). THZ531 covalently binds the cysteine residue at position 1039 as well as CDK13 at cysteine 1017 and CDK7 at cysteine 312. Compound E9 was developed to overcome resistance to the THZ series of transcriptional CDK inhibitors by protecting it from drug efflux (144). Compound E9 also targets cysteine residues in CDK7/12/13, and so cells harboring a C1039S mutation in CDK12 are resistant to compound E9 (Supplementary Table S1).

Ito and colleagues (Takeda) recently developed Compound 2, in an effort to derive a CDK12 inhibitor with higher specificity and inhibitory properties as well as physiochemical properties (145). This compound inhibits Ser2 phosphorylation in the CTD of RNA Pol II and growth of SK-BR-3 breast cancer cells, which overexpress HER2. In addition, inhibition of CDK12 enhances HER2-targeting therapies, and HER2-amplified tumors may be sensitive to CDK12 inhibitors (146). Johannes and colleagues (AstraZeneca) used structure-based design to develop noncovalent inhibitors of CDK12 with moderate potency (147). Their data suggest that the window of inhibition with CDK12 may be too small to achieve a therapeutic index in the tumor over normal tissue. Surprisingly, they did not observe any synergy with PARPi; the explanation for this is currently unclear. In contrast, SR-4835, an ATP-competitive kinase inhibitor against CDK12 and CDK13, has been shown to suppress the expression of multiple DNA-repair genes and regulate intronic polyadenylation, which increases sensitization to platinum chemotherapy and PARPi in a model of TNBC (148). Although several research-grade tool compounds are currently available (Supplementary Table S1), no clinical-grade CDK12 or CDK13 inhibitor is yet available.

The ability to suppress adaptive responses to targeted cancer therapies by repressing transcription has been proposed, as therapy resistance remains a major hurdle. Resistance appears to be at least partially due to the acquisition of adaptive cellular programs. Inhibition of CDK7/CDK12/CDK13 with THZ1 blocks transcriptional programs that facilitate resistance. Although the exact mechanism remains to be fully elucidated, it may involve remodeling of enhancers and other signaling outputs required for tumor cell survival in the setting of targeted therapy. Accordingly, genetic ablation of either CDK7 or CDK12 reduces the outgrowth of resistant clones (149). THZ531 also impairs the emergence of resistance. Furthermore, acquired resistance to BET inhibitors in CRPC results in reactivation of androgen receptor signaling (150). This is mediated by CDK9-mediated androgen receptor phosphorylation, as inhibition of CDK9 with dinaciclib or LDC000067 suppresses androgen receptor and its transcriptional activity in BET inhibitor–resistant cells (150). The addition of transcriptional CDK inhibitors to targeted cancer therapies may therefore delay resistance and lead to more durable responses, although more robust preclinical work needs to be done prior to testing these combinations in clinical trials.

Resistance to PARPi also remains an important limitation, even in patients with BRCA1/BRCA2-mutated cancer (104, 151, 152). CDK12 inhibitors might reverse de novo and acquired PARPi resistance in BRCA1-mutant breast cancer cells (153). Dinaciclib (which inhibits CDK9, CDK12, and CDK13 as well as CDK1, CDK2, and CDK5) reduces levels of BRCA1 and RAD51 mRNAs and levels of HR DNA repair. This sensitizes BRCA1-mutant and wild-type TNBC cells to PARPi, which can be phenocopied by genetic CDK12 depletion. In models of PARPi resistance, CDK12 inhibition restores PARPi sensitivity (153). Whether more specific CDK12 inhibitors will be efficacious in treating or delaying PARPi resistance, and whether the combination of CDK12 inhibitors and PARPi is tolerable in patients remains to be determined.

Beyond oncology, transcription-associated kinase inhibitors may also find indications in other disease areas, including infectious diseases and cardiology (14). Indeed, a recent study showed that a novel compound active against visceral leishmaniasis acts principally by inhibiting CDC2-related kinase 12 (CRK12), the Leishmania homolog of CDK12 (154). In addition, cardiac hypertrophy is characterized by global increases in mRNA and protein synthesis, and inactivation of Cdk9 in mice dampens hypertrophic signals (155). These data suggest that inhibitors against the transcription-associated CDKs may have broad indications in human disease.

The transcription-associated CDKs are emerging as important targets and biomarkers in oncology. In contrast to the cell cycle–associated CDKs, this family of CDKs plays critical roles in regulating gene expression at multiple levels, including transcription, splicing, intronic polyadenylation, and epigenetics. Recent evidence also points to a role in regulating protein expression at the level of translation, although additional work will be required to sort out whether this is an indirect effect, and whether as a family the transcription-associated CDKs affect these processes. How these CDKs interact with each other to regulate complex molecular processes also remains to be more fully elucidated. What intrinsic redundancies are in place, and how do these CDKs work together to affect cancer cell properties such as growth, invasion, and apoptosis? These answers will likely be cell context–dependent, and so understanding which cells rely on which particular transcription-associated CDKs will be critical.

Loss of particular transcription-associated CDKs or their cyclin partners may also uncover potential therapeutic vulnerabilities and synthetic lethal interactions. To date, most transcriptional CDKs are not frequently mutated or lost, with the exception of CDK12. Preclinical evidence demonstrates that CDK12 deficiency increases sensitivity to PARP and CHK1 inhibitors, and, because of the genomic instability that results from CDK12 loss, may also increase susceptibility to checkpoint immunotherapy. However, our mechanistic understanding of this phenomenon and the precise role CDK12 plays in DNA replication and TDs remains poor. Nonetheless, several of these targeted strategies are currently being tested in clinical trials, particularly in advanced prostate cancer. Yet our understanding of how specific CDK12 mutations affect their function remains limited, and how other molecular alterations occur in context (whether in cis or in trans) will be important to work out, which will likely affect the efficacy of these targeted approaches.

Additional work on developing more specific chemical inhibitors that target the transcription-associated CDKs will also be important, as the side-effect profiles of these inhibitors may limit patient tolerability. Several studies point to the use of CDK7, CDK8, CDK9, CDK12, and CDK13 inhibitors in various solid and hematologic cancer types, including TNBC, MYCN-driven neuroblastoma, colon cancer, Ewing sarcoma, AML, and SCLC. These cancers are primarily driven by dysregulated transcription, and are highly dependent on high levels of basal transcription mediated through transcription factors such as MYC, β-catenin, or EWS–FLI1. Several clinical trials with CDK7 and CDK9 inhibitors are currently under way to determine the MTD and dose-limiting toxicities, to determine whether these can be successfully administered to patients. In addition, for WNT-driven tumors and those harboring CDK8 amplification, CDK8 inhibition may hold therapeutic promise, but this awaits to be tested clinically. Moreover, these transcriptional CDK inhibitors may delay adaptive resistance to targeted therapies, including to PARPi. Therefore, combination strategies will likely be important, including with immunotherapy, although the therapeutic index and cumulative toxicities will need to be carefully monitored. Finally, inhibitors of other lesser-known transcription-associated CDKs, such as CDK11, may also be important targets in cancer (156), and underscore the need for additional research beyond the canonical transcription-associated CDKs.

In summary, understanding the transcription-associated CDKs is an exciting area of great promise in oncology, one that will likely continue to receive considerable attention in the coming years. We eagerly await the entry of transcription-associated CDK inhibitors into the clinical armamentarium against cancer, and anticipate the identification of new synthetic lethal strategies to target tumors characterized by the loss of these important regulators.

F.Y. Feng has paid consultant/advisory board relationships with Janssen, Sanofi, Astellas, Bayer, Genentech, EMD Serono, Clovis, and Celgene, and an unpaid consultant/advisory board relationship with PFS Genomics. A. Ashworth is a cofounder of Tango Therapeutics, Azkarra Therapeutics, and Ovibio, is an advisor for Gladiator, Prolynx, Earli, and Genentech, reports receiving commercial research grants from AstraZeneca and SPARC, and has ownership interest in patents on the use of PARP inhibitors, held jointly with AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

J. Chou is supported by the A.P. Giannini Foundation, the Quinlan and Rosenberg Fellowships in Genitourinary Oncology, and a training grant from the NCI (T32 CA108462). D.A. Quigley is supported by Young Investigator awards from the Prostate Cancer Foundation and the BRCA Foundation. F.Y. Feng and A. Ashworth are supported by Challenge Awards from the Prostate Cancer Foundation. This work was supported by the NCI (1R01CA230516; to F.Y. Feng and A. Ashworth).