CDK12 inhibition reduces abnormalities in cells from patients with myotonic dystrophy and in a mouse model

CDK12 inhibition reduces mutant transcripts and nuclear foci in DM1 cells and produces splicing correction and phenotypic benefit in a mouse model. Muscling in on myotonic dystrophy Myotonic dystrophy type 1 (DM1) is a genetic muscular disorder caused by accumulation of nuclear foci due to CTG repeat expansion in the DMPK gene. Current therapies only target individual symptoms. Now, Ketley et al. set out to identify targets for reducing the accumulation of nuclear foci. The authors showed that cyclin-dependent kinase 12 (CDK12) is increased in muscle biopsies from patients with DM1 and its inhibition with a small molecule reduced the accumulation of nuclear foci and mutant transcripts. In a mouse model of DM1, CDK12 inhibition reduced nuclear foci and improved myotonia. The results suggest that CDK12 might be a good therapeutic target for treating DM1. Myotonic dystrophy type 1 (DM1) is an RNA-based disease with no current treatment. It is caused by a transcribed CTG repeat expansion within the 3′ untranslated region of the dystrophia myotonica protein kinase (DMPK) gene. Mutant repeat expansion transcripts remain in the nuclei of patients’ cells, forming distinct microscopically detectable foci that contribute substantially to the pathophysiology of the condition. Here, we report small-molecule inhibitors that remove nuclear foci and have beneficial effects in the HSALR mouse model, reducing transgene expression, leading to improvements in myotonia, splicing, and centralized nuclei. Using chemoproteomics in combination with cell-based assays, we identify cyclin-dependent kinase 12 (CDK12) as a druggable target for this condition. CDK12 is a protein elevated in DM1 cell lines and patient muscle biopsies, and our results showed that its inhibition led to reduced expression of repeat expansion RNA. Some of the inhibitors identified in this study are currently the subject of clinical trials for other indications and provide valuable starting points for a drug development program in DM1.


INTRODUCTION
Myotonic dystrophy type 1 (DM1) is the most common form of adult muscular dystrophy, which affects 1 in 8000 people (1). It is caused by a CTG repeat sequence in the 3′ untranslated region of the dystrophia myotonica protein kinase (DMPK) gene (2)(3)(4), which is greatly expanded in patients who may have 50 to several thousand repeats on affected chromosomes compared to between 5 and 37 repeats on unaffected chromosomes. The expanded repeat is transcribed, and despite being correctly spliced, the repeat expansion transcripts remain sequestered in the nucleus forming distinct foci (5)(6)(7). These foci interact with cellular proteins, such as muscleblind-like splicing regulator 1 (MBNL1), a key splicing regulator, which, in turn, leads to downstream splicing abnormalities (8,9). In addition to the sequestration of proteins, the mutant RNA causes activation of CUGBP, Elav-like family member 1 (CELF1), which is also implicated in splicing (10). Additional molecular pathways are thought to be affected by the toxic RNA, including repeat-associated non-AUG (RAN) translation and inhibition of translation (11,12).
There is currently no treatment for DM1, and disease management relies on a fragmented approach using specific drugs for the piecemeal treatment of particular symptoms such as mexiletine to treat myotonia and modafinil to address daytime sleepiness (13,14). However, due to the complex and variable nature of DM1, management of individual symptoms is not an efficient way to manage the condition and an effective treatment is required. Drug development for an RNA-based disorder such as DM1 represents a major challenge due to the lack of a suitable protein target (15)(16)(17)(18)(19)(20)(21). Previously, we reported an optimized high-content screening assay to test the effect of small molecules on nuclear foci in DM1 and identified the possible role of a kinase as central to disease pathophysiology (22). Thus, we set out to identify the specific kinase involved in reducing the accumulation of CUG nuclear foci as a cellular target for DM1 therapy.
at least 300 cells were imaged for each compound concentration (22). Six compounds that share a pyrazolo [1,5b]pyridazine core were found to reduce the number of nuclear foci following 24-hour treatment of DM1 cells (Fig. 1, A to F, and fig. S1). This result was consistent across two DM1 fibroblast lines (table S1). We then analyzed the known selectivity profiles of these six compounds to identify the common kinase targets. The log half-maximal inhibitory concentration (pIC 50 ) values generated from the foci assays were compared to the compound inhibition profiles against 224 kinase targets (23). A partial least squares (PLS) model was used to cluster the data, which suggested that the common target was likely to be a member of the cyclin-dependent kinases (CDKs), mitogen-activated protein kinases (MAP kinases), glycogen synthase kinases (GSKs), and CDK-like kinases (CMGC) family (Fig. 1G). The CMGC family members were then further analyzed to identify the most likely target candidates within this group ( fig. S2). We compared the kinase percentage inhibition value for all tested compounds, categorizing the six nuclear foci reducing compounds as active and the remaining compounds that did not result in nuclear foci reduction as inactive ( fig. S2). The resulting scatterplots were then evaluated to identify if any of the kinase targets showed a clear difference in activity between the actives and the inactives. Of the targets covered by the PKIS collection, CDK family members appeared most likely to be involved, specifically CDK4, which showed that the six nuclear foci reducing compounds had higher inhibitory activity on this target than all other compounds tested ( fig. S2D). To evaluate the potential role of CDK family members on nuclear foci modulation, we tested additional small-molecule CDK inhibitors with well-annotated selectivity profiles (figs. S1 and S3 and table S2) in the nuclear foci assay. CDK family inhibitors dinaciclib and SNS-032 reduced nuclear foci (Fig. 1, H and I), further suggesting that CDKs could play a role in the pathophysiology of DM1.

Inhibitor treatment as a therapeutic for DM1
Both dinaciclib and SNS-032 have been used in early clinical trials for cancer therapy (24)(25)(26)(27)(28). To investigate the potential use of such compounds as a therapy in DM1, we sought to establish the time course of inhibition on nuclear foci and to determine the minimal exposure time required for beneficial effect. To determine a dosing regimen with these inhibitors, we exposed DM1 fibroblasts to SNS-032 for 2 hours, after which time the cells were washed thoroughly and allowed to recover in complete growth media. Quantification of untreated DM1 cells showed that 68% had five or more foci per cell and only 5% had no detectable foci ( Fig. 2A). When cells were treated with SNS-032 for 2 hours, with no recovery time, this distribution shifted to 28% of cells with five or more foci and 10% cells with no foci (Fig. 2B). With increased recovery times of 48 and 72 hours, the proportion of cells without nuclear foci increased further to 14 and 36%, respectively (Fig. 2, C and D). A short (2-hour) treatment with inhibitor, followed by a prolonged (72-hour) recovery, led to a significant reduction in nuclear foci ( 2 test: P < 0.0001 for all dosing regimen compared to untreated cells), suggesting that pulsatile treatment could be an efficacious approach to DM1 therapy with suitable inhibitors.
Nuclear foci are a key cellular feature in DM1 and act as a useful biomarker for screening. Ultimately, we wanted to understand the effect of compounds on the repeat expansion transcript directly and to assess any downstream effects. Therefore, we next sought to establish the effect of inhibitor treatment on DMPK transcripts.
Following treatment with dinaciclib, we used a previously reported assay based on the presence or absence of a coding Bpm I polymorphism [single-nucleotide polymorphism (SNP) rs527221] that allows us to distinguish between wild-type and mutant DMPK transcripts in informative patient cell lines (7). Analysis of nuclear and cytoplasmic cell extracts showed that the repeat expansion transcripts were still retained within the nuclear fraction (Fig. 2E). However, quantification using GeneScan analysis showed a 59% decrease in the relative proportion of repeat expansion transcripts compared to unexpanded DMPK transcripts in the nucleus following inhibitor treatment [two-way analysis of variance (ANOVA), P < 0.0001; Fig. 2F]. To confirm this result, we used digital droplet polymerase chain reaction (ddPCR) to quantify the wild-type and mutant DMPK transcript numbers in total RNA from two DM1 fibroblast cell lines treated with our most potent compounds, SNS-032 and dinaciclib (Fig. 2, G to J, and fig. S4). Treatment reduced nuclear foci and did not result in cellular toxicity. The data show a reduction between 20 and 60% in mutant repeat-containing transcripts. Only in the HK DM1 fibroblast cell line treated with SNS-032 was a reduction seen in the wild-type transcript (Fig. 2J). Thus, exposure to these CDK inhibitors appears to reduce predominantly the repeat-containing transcript.
To examine if this beneficial effect was translated into an in vivo model, we treated human skeletal actin, long repeat (HSA LR ) transgenic mice, a mouse model of DM1, with dinaciclib (20 mg/kg) by intraperitoneal injection. The dosing regimen of dinaciclib was established based on previously tested doses and tolerability studies in mice and human clinical cancer studies (29). The HSA LR mice express repeat expansion RNA at amounts fivefold to eightfold higher than human DM1 muscle (30). Blinded analysis was performed 1 day after the final injection. Mice were age-and sex-matched to assess the effect on nuclear foci, HSA transgene amounts, and key splice isoform profiles in vehicle-and dinaciclib-treated samples. Following dinaciclib treatment, we observed a reduction in the number of cells containing nuclear foci (Fig. 3, A and B). Likewise, when we quantified the repeat-containing transcript directly by ddPCR, we found a significant reduction in the relative amounts of the HSA transgene in dinaciclib-treated animals ( Fig. 3C; t test, P < 0.0001). This is consistent with the data observed in DM1 fibroblast lines. To assess any beneficial effect on missplicing, we studied nine transcripts, known to be dysregulated in DM1 (31)(32)(33)(34)(35)(36)(37). Across the panel of genes, we consistently observed improvements in dinaciclib-treated mice ( Fig. 3D and fig. S5). In addition, we collected muscle samples from vehicle-and dinaciclib-treated mice. Staining of muscle fibers highlighted a reduction in the presence of centralized nuclei within muscle fibers following inhibitor treatment (fig. S6, A and B; t test, P = 0.0162). To understand if the observed molecular changes resulted in any functional improvement, we conducted electromyography (EMG) and graded the animals with a myotonia grade from 0 to 3, with 3 being the most severe. Dinaciclib treatment improved myotonia in all muscles analyzed except the tibialis anterior ( Fig. 3E; Mann-Whitney, P = 0.0443; additional data: fig. S6, C and D).

Kinase inhibitor target deconvolution
To identify the specific CDK target responsible for these beneficial effects, we used the kinobeads methodology, which is based on sepharose beads derivatized with a combination of promiscuous kinase inhibitors (38)(39)(40), to profile 11 compounds, which represent a range of activities in the nuclear foci assay. Target profiles were generated by adding each compound to K562 erythroleukemia cell extract at a concentration of 2 M, followed by incubation with two variations of kinobeads and quantification of beadbound proteins (38,41). The profiling results suggested that members of the CDK family are common targets of the most active compounds (table S3), consistent with the PKIS data.
We sought to expand the target coverage within the CDK family by the immobilization on beads of two of the active compounds containing a suitable secondary amine: SNS-032 and AT7519 ( fig. S7). Beads derivatized with either compound showed good coverage of the CDK family, including family members against which the PKIS collection was not profiled. For an in-depth chemoproteomics study, profiling inhibitor selectivity across the CDK family, we generated dose-response competition-binding profiles for all active and one of the inactive compounds (PD0332991) in K562 cells or in A204 rhabdomyosarcoma cells (data file S1). We also conducted a comparative whole proteome analysis to confirm that the K562, A204, and DM1 cell lines express the same kinases (data files S1 and S2). All CDK/PCTK proteins identified by profiling were also identified in the cells used for the nuclear foci assay ( fig. S8 and data files S1 and S2). The resulting dataset comprises IC 50 curves for 12 CDK family kinases ( fig. S9 and table S4). To identify the most likely kinase target, we plotted pIC 50 values for kinobead binding against the pIC 50 in the foci inhibition assay for each compound. A good correlation of kinase binding affinity with the inhibitory activity on foci was observed for CDK7, CDK9, and CDK12. The best correlation of kinase pIC 50 values with inhibitory activity on foci was observed for CDK12. Dinaciclib was the most potent inhibitor against the CDK family, in particular against CDK12 (Fig. 4).

CDK12 in DM1 pathogenesis
To examine the potential involvement of CDK12 in DM1 pathogenesis, we assessed the endogenous amounts of this protein in vastus lateralis muscle biopsy samples from four patients with DM1 and four healthy volunteers using Western blot (Fig. 5A). There was an increase in CDK12 in DM1 biopsies versus those from healthy volunteers, with 48% more CDK12 protein detected in DM1 samples (t test, P = 0.02; Fig. 5B). Next, we used immunohistochemistry to examine the location of CDK12 in DM1 fibroblasts. Consistent with previously published data from non-DM1 cells that show CDK12 colocalization with SC35 speckles, we observed nuclear staining in granular structures in both DM1 and non-DM cells (42). Previous work has also identified a localization of CUG repeat expansion foci with nuclear speckles (43). Quantification of the CDK12 granular structures in 400 cells showed that the number of granules was significantly elevated in DM1 cells, with an average number of 18.74 (±3.88), compared to 12.07 (±2.27) in non-DM1 cells (P < 0.0001; Fig. 5C). To understand further the relationship between CDK12 protein and repeat expansion foci and to confirm that specific inhibition of this protein was responsible for nuclear foci reduction, we used short hairpin RNA (shRNA) and small interfering RNA (siRNA) to reduce CDK12 in DM1 cells. shRNA (m) Lentiviral Particles, sc-44531-v] expressing three to five shRNAs against CDK12, we observed a 56% reduction in the number of CDK12 granules and a 69% reduction in repeat expansion foci compared to scrambled shRNA-treated cells (Fig. 5, D to F, and fig. S10). This was verified by siRNA knockdown of CDK12 and quantification by Western blot analysis ( fig. S10). Conversely, CDK12 overexpression with a full-length open reading frame (ORF) clone resulted in increased numbers of nuclear foci in a dosedependent manner (one-way ANOVA, P = 0.0017; Fig. 5G). CDK12 is a known regulator of transcription, specifically involved in transcriptional elongation, rather than at the initiation of transcription (44). To understand the mechanism by which CDK12 inhibition leads to nuclear foci reduction, we tested the effect of CDK12 inhibition on repeat expansion transcription. For this, we used a bidirectional inducible plasmid expressing enhanced green fluorescent protein (eGFP) in one direction and DMPK exons 10 to 15, containing 960 interrupted CTG repeats, in the other (45). This construct was transfected into mouse P19 cells in the presence or absence of the highly selective and potent CDK12 inhibitor, THZ531 ( Fig. 5H and fig. S11) (46). This compound was chosen for this analysis as, in contrast to dinaciclib, it is selective for CDK12 (46). Following doxycycline induction of the transgene, cells were harvested, and reverse transcription PCR (RT-PCR) and ddPCR analyses were conducted to establish the amounts of CUG 960 RNA and eGFP transcription. Transcript amounts were normalized to endogenous glyceraldehyde-3-phosphate dehydrogenase (GAPDH). In untreated cells, both transcripts accumulated at the same rate (fig. S12), whereas in THZ531-treated cells, there was a clear reduction in the amount of CUG repeat expansion RNA following CDK12 inhibition compared to eGFP RNA (Fig. 5, I and J). The amount of eGFP transcript did not show any differences in the presence or absence of CDK12 activity, suggesting that the inhibition of CDK12 leads to a preferential reduction in transcription of the expanded repeat (Fig. 5, I and J). To assess if CDK12 inhibition leads to increased rates of transcript decline, we halted transcript production from the plasmid by removal of doxycycline and monitored the transcripts over 24 hours. This showed a natural reduction of both the CUG repeat expansion and eGFP transcripts at a similar rate (Fig. 5, K and L, and fig. S12). The differential profiles for transcript accumulation suggest that the effect of CDK12 inhibition on the repeat expansion transcript is specific at the level of transcript production, rather than a generalized effect on global transcription. To confirm this, we conducted a nuclear run-on experiment to examine the transcription of wild-type and mutant transcript in untreated and THZ531-treated DM1 cells. These data support the notion that, in DM1 cells, THZ531 inhibition affects transcription of the mutant transcript more than it affects the wild-type transcript (t test, P = 0.0017; Fig. 5M).

DISCUSSION
The highly annotated PKIS collection provides an excellent resource for kinase inhibitor target deconvolution. We have used this cheminformatic collection in conjunction with kinobead mass spectrometry to identify CDK12 as a cellular target for DM1 therapy.
We used a phenotypic screen based on nuclear foci, a key molecular feature of this disease, as no clear molecular target was known and foci provide a visual biomarker of the condition. Our work has established a clear biological link between CDK12 and the molecular signature of DM1. CDK12 is a transcription elongation-associated C-terminal repeat domain (CTD) kinase, which has been shown to regulate phosphorylation of serine-2 in the C-terminal domain of RNA Pol II, a step required for productive elongation of transcripts, rather than being involved in the initiation of transcription (44,47). The Saccharomyces cerevisiae gene CTK1, an ortholog of metazoan CDK12, is not an essential gene, and reports suggest that CDK12 is not part of the core transcriptional apparatus (42,48). Assessment of  cell essentiality in human cell lines showed that CDK12 is essential in only one (KBM7 haploid cells) of the four cell lines tested (KBM7, K562, Rao, and Jiyoye) (49). Expression analysis showed that only 2.67% of genes, producing primarily long and complex transcripts, had altered amounts following CDK12 depletion, thereby providing a potential link between CDK12 and transcription of the DM1 expanded repeats (50). As CDK12 is not required at the start of transcription and its inhibition does not result in global transcriptional arrest, alongside our data that show that continuous treatment is not necessary for beneficial effects, it may be suitable as a target for long-term DM1 treatment.
Our results show that genetic knockdown of CDK12, or treatment with inhibitor, leads to reduced numbers of nuclear foci, which could be due to two possible mechanisms: reduced production of the mutant transcripts or their increased degradation. Using ddPCR, we show that selective inhibition of CDK12 results in a reduction in mutant RNA transcription, which points to a requirement for CDK12 to produce repeat expansion transcripts. Other than RNA Pol II, the targets of CDK12 phosphorylation are largely unknown, particularly in muscle, although it has been shown to suppress genes involved in supporting metabolic functions during stress and reactive oxygen species (ROS)-induced gene activation (51).
Here, we identify CDK12 as a druggable target for DM1. The development of specific CDK12 inhibitor molecules has been achieved, and some of the molecules presented here have been the subject of clinical trials in other indications (24)(25)(26)(27)(28). Our data suggest that a pulsatile treatment resulting in a temporary transcriptional block in both cell lines and in vivo leads to beneficial downstream effects reducing nuclear foci, improving key splicing profiles and functional improvements in myotonia. The functional benefit observed in the HSA LR mouse model demonstrates that the effects of CDK12 inhibition translate from cell-based assays to an in vivo model and provide evidence for the application of CDK12 inhibitors in DM1 therapy.
Further development of more selective CDK12 inhibitors will be required to fully understand the functional benefit of inhibiting this target for DM1 therapy. Furthermore, the HSA LR mouse does have limitations, despite being the best studied DM1 model, in that expression of the repeat expansion transcript is driven from the skeletal actin promoter, so these mice do not recapitulate all features of the disorder. The model has a repeat number of 250 CTG repeats, which is limited in size compared to what is seen in DM1 patients. Testing the effect of small molecules on other models should help understand the efficacy of CDK12 inhibition in the brain, for example. Future work will be required to refine the targeting of this protein in DM1, and the compounds presented here will act as valuable starting points and tool compounds for future drug development opportunities.

Study design
The objective of this study was to identify the specific protein kinase responsible for previously observed beneficial effects in DM1 cell lines treated with kinase inhibitors. The design of the study was based on comparative analysis of compound activities in a cell-based screening assay with biochemical activity in in vitro assays. The PKIS collection was screened and the target was refined using ki-nobead mass spectrometry. Following target identification, we tested key compounds in a DM1 in vivo model to establish the translational benefit of this kinase on key molecular events, transgene amounts and misspliced transcripts, and a key physiological output, myotonia.
All cell-based screening assays were unblinded but performed in triplicate with appropriate controls and analyzed with an automated process. For in vivo experiments, mice were randomized to treatment groups, and all tests were carried out and analyzed in a blinded manner. Power calculations established treatment groups for EMG of nine mice per group [Gpower v3.1, for a Wilcoxon-Mann-Whitney test for a 50% reduction in scores, based on variance from (52) Alpha = 0.05; power = 0.90; d = 1.5]. All experiments were carried out in three biological replicates, unless stated otherwise in the figure legend. Experimental details and the statistical tests used are listed in each figure legend. All data are presented and includes all outliers. Raw data and the corresponding statistical test details for the figures are listed in data file S3.

Sample use
The human biological samples were sourced ethically, and their research use was in accord with the terms of the informed consent. All animal studies were ethically reviewed and carried out in accordance with Animals (Scientific Procedures) Act 1986 and the GSK Policy on the Care, Welfare and Treatment of Animals. All experiments were performed in accordance with relevant guidelines and regulations of the Animal Procedures Act under license number PPL3003449 (Nottingham), issued 5 December 2016, and granted by The Home Office. All studies were reviewed and conducted in accordance with the Institutional Animal Care and Use Committee by the ethical review process at the institution where the work was performed.

Cell culture
Fibroblast cells were grown in Dulbecco's modified Eagle's medium (DMEM) with penicillin and streptomycin and 10% fetal calf serum (FCS) (Sigma-Aldrich). KB Telo MyoD cell line contains 400 CTG repeats, LR-Telo MyoD cell line contains 1200 CTG repeats, and HK Telo MyoD cell line contains 1600 CTG repeats.
In situ hybridization protocol Cells were exposed to compounds for 24 hours, after which in situ hybridization was performed to identify foci using a Cy3-labeled (CAG) 10 probe. Plates were analyzed on a Molecular Devices Micro High Content Imaging system, with nine fields imaged per well to give about 100 cells per well, per compound treatment. The nuclear area was identified by Hoechst stain, and the number, size, and intensity of foci were determined by scoring adjacent pixels that were 80 grayscales or more above background.

Preparation of cell extracts
K562 and A204 cells were obtained from the American Type Culture Collection (ATCC) and cultured in RPMI 1640 containing 10% FCS. Cells were expanded to 1.5 × 10 6 cells/ml. A204 cells were cultured in McCoy's 5A medium containing 15% FCS. Cells were expanded to 100% confluency. Cells were harvested and subjected to three washes with ice-cold phosphate-buffered saline (PBS). Aliquots were snap-frozen in liquid nitrogen and stored at −80°C. Cell extracts were prepared as described (53).

Chemoproteomics
Affinity profiling was performed as described previously (38,53). Sepharose beads were derivatized with SNS-032 or AT7519 at a concentration of 1 mM to generate a bead matrix, or kinobeads were used as a matrix for profiling. Beads (35 l in case of kinobeads or 5 l in case of SNS-032) were washed and equilibrated in lysis buffer at 4°C for 1 hour with 1 ml (5 mg) K562 cell extract, which was preincubated with compound or buffer. Beads were transferred to disposable columns (MoBiTec), washed extensively with lysis buffer, and eluted with SDS sample buffer. Proteins were alkylated, separated on 4 to 12% NuPAGE (Invitrogen), stained with colloidal Coomassie, and quantified by isobaric mass tagging and liquid chromatography tandem mass spectrometry (LC-MS/MS).

Digital droplet polymerase chain reaction
Primers and probes used in ddPCR assays were manually designed and synthesized by Integrated DNA Technologies Inc. Sequences of primers and probes are listed in tables S5 and S6. All reactions were prepared using Bio-Rad reagents, and assays were performed with Bio-Rad equipment. After reverse transcription, ddPCR solution was prepared to a final volume of 25 l containing 1× ddPCR supermix for probes, 250 nM gene-specific primers, 125 nM probes (for ex-ON and ex-OFF), and complementary DNA (cDNA) (diluted from 20× to 40×). No template control and no reverse transcriptase control (RT−) were included in each ddPCR run to detect possible contaminations. The ddPCRs were loaded to a DG8 cartridge along with 70 l of droplet generation oil to form droplets in a QX100 droplet generator. Forty microliters of partitioned emulsion containing droplets was then slowly transferred to 96-Well twin Semi-Skirted PCR Plate (Eppendorf). After heat-sealing with foil, the plate containing the droplets was PCR-cycled to the final point under conditions at 95°C for 10 min, 95°C for 30 s, and 60°C for 60 s for 40 cycles, 98°C for 10 min, then held at 4°C (for details about annealing temperature for each gene, see table S6). Following PCR, samples were read on a droplet reader, which automatically reads the droplets from each well of the plate. Last, data were analyzed using QuantaSoft software to determine the number of positive droplets. A manual selection of "+/−," "−/+," "+/+," and "−/−" counts was done using the Lasso function in the two-dimensional plots. The counts were then used by the software to calculate the copy numbers of FAM-positive droplets, HEX-positive droplets, and FAM and HEX double-positive and double-negative droplets in the four quadrants. For each ddPCR assay, serial dilutions of cDNA were used to obtain the lowest number of double-positive droplets; annealing temperature gradients were used to optimize PCR conditions and to determine the best separation between negative and positive reactions.

Assay for repeat expansion transcripts
Reverse transcription was performed using 1 g of total RNA from compound-treated and untreated cells. PCR was carried out using 1 / 20 of the synthesized cDNA with primers N11 (5′-CACTGTCGG-ACATTCGGGAAGGTGC) and 133 (5′-GCTTGCACGTGTG-GCTCAAGCAGCTG). For GeneScan analysis, primer N11 was labeled with FAM. Amplification was performed with a T m of 58°C. The PCR product was subsequently heated to 95°C for 2 min followed by cooling to 4°C. For Bpm I restriction digestion analysis of DMPK PCR products, 8 l of PCR mixture was digested overnight with restriction enzyme Bpm I (NEB) in a total reaction volume of 20 l at 37°C. The final products were analyzed by electrophoresis at 90 V with 3% agarose gels, and the density of bands was quantified using ImageJ software or by fragment analysis on an ABI377 sequencer followed by GeneScan quantification.

Western blots and detection
Western blotting was performed using a commercial NuPAGE system (Invitrogen) according to the manufacturer's instructions. The primary antibodies used in this study were human CDK12 (Abcam; 1:400 dilution) and human -tubulin (obtained from Santa Cruz Biotechnology and used at a dilution of 1:500). Anti-mouse IgG-horseradish peroxidase (HRP) was used as the secondary antibody. ImageJ software was used for the quantification of bands on Western blots.

CDK12 shRNA knockdown and overexpression
Cells were plated at 40% confluency the day before infection in 96-well format. Lentiviral titer (Santa Cruz Biotechnology, sc-44343-V) was added at a multiplicity of infection (MOI) of 10 in polybrene (5 g/ml) diluted in DMEM. Cells were spin-inoculated by centrifugation at 2500 rpm for 30 min. Following 24-hour incubation, the virus was removed and replaced with fresh DMEM. The infection was repeated on day 4, and cells were collected on day 7 for immunohistochemistry and in situ hybridization analysis. For overexpression analysis, cells were electroporated on day 1 with CDK12 full-length ORF clone (BC150265; GeneCopoeia) and in situ hybridization was conducted 48 hours later.

Transgenic mice
Homozygous HSA LR transgenic mice were previously described (30). Mice are routinely genotyped to confirm the presence of the CTG repeat expansion. HSA LR mice, aged 8 to 12 weeks, housed in standard conditions, in groups of four in individual ventilated cages with standard laboratory food were used for EMG assessments (n = 12 males per group). They were randomly assigned to receive dinaciclib or vehicle alone [2-hydroxypropyl--cyclodextrin (200 mg/ml); Sigma-Aldrich] by intraperitoneal injection at a dose of 20 mg/kg. Mice were dosed every other day for four injections, and EMG was performed as previously described (17) by blinded examiner. Two mice were excluded from the treatment group because of adverse effects, and three EMG datasets were excluded due to electrical interference. One day after the final injection, mice were euthanized and muscle tissue was snap-frozen and used for analysis of HSA transgene amounts and alternative splicing (eight samples were chosen at random from each group).
In situ hybridization of HSA LR mouse muscle samples Eight female HSA LR mice (12 to 17 weeks of age) were dosed three times with dinaciclib (20 mg/kg) or vehicle alone [2-hydroxypropyl--cyclodextrin (200 mg/ml); Sigma-Aldrich] by intraperitoneal injection over 24 hours. Mice were sacrificed 2 hours after the final injection to allow assessment of dinaciclib on foci dissipation in snap-frozen gastrocnemius muscles samples. Twelve-micrometer cryostat sections were thawed onto a Superfrost Plus slide. Slides were fixed in 2% paraformaldehyde (PFA) in PBS for 30 min at 4°C. After two brief washes in PBS at room temperature, slides were permeabilized in 2% acetone in PBS for 5 min at 4°C, followed by two brief washes in PBS. Prehybridization (2× SSC/30% formamide) was conducted at room temperature before hybridization at 42°C [2× SSC, 30% formamide, BSA, vanadate, yeast transfer RNA (tRNA) (1 mg/ml), Cy3-labeled (CAG) 10 probe (500 ng/ml)] for 2 hours. Posthybridization washes were in 2× SSC/30% formamide at 45°C for 30 min followed by 2× SSC at room temperatures for 5 × 5 min. Slides were incubated in PBS with 5 mM MgCl 2 for 15 min at room temperature. After two brief washes in PBS/5 mM MgCl 2 , the stained sections were embedded in Vectashield and stored at 4°C.

Image analysis
Microscopy was performed on a Zeiss 200M widefield fluorescence microscope using the DAPI filter for Hoechst-stained nuclei and the tetramethyl rhodamine isothiocyanate (TRITC) filters for the CUG repeats detected with the Cy3 probe. Images were captured from eight evenly spread locations with 20× objective as 16-bit Z stacks with 0.7-m step size over the full thickness of the sections for both the DAPI (nuclei) and the TRITC (CUG repeats) channel. Maximal projection images were generated for both channels, and nuclei were selected using an IJ-Isodata threshold. Individual nuclei were segmented using rolling circle background subtraction and generated a binary mask to define the DAPI files. Background signal on the TRITC channel was subtracted from the signal generated using the triangle method. The two channels were merged to allow quantification of the relative amount of CAG probe signal per nuclei.

Assay to quantify transcription rates
Mouse P19 cells were treated with THZ531 at 300 nM concentration for 24 hours before transfection (THZ531 was a gift from N. Gray). Treated and untreated cells were transfected with pBItetDT960GFP (1.5 g) and pTet-One inducer plasmid (1 g) using Polyfect transfection reagent, and transcription was induced with the addition of doxycycline to the media at 1 g/ml [pBItetDT960GFP was a gift from T. Cooper (Addgene plasmid # 80419)]. Twenty-four hours after induction, cells were harvested and total RNA was extracted. cDNA was synthesized using SuperScript III and processed by RT-PCR and ddPCR.

Nuclear run-on assay
Nuclear run-on analysis was conducted following the protocol described by Gardini (54). KB Telo MyoD (DM1 fibroblasts) were used, and transcript analysis was carried out using ddPCR, as described above.

Statistical analysis
All raw data presented in this article were assessed for normality using Shapiro-Wilk tests (or residuals in the case of ANOVA tests). Normally distributed data were tested for significance using twotailed t tests and one-or two-way ANOVAs followed by Sidak's, Dunnett's, or Tukey's multiple comparison tests. Non-normal data were tested using a Mann-Whitney test or chi-square analysis. P values were assumed to be significant if less than 0.05. Bonferroni correction was applied to the data in Fig. 3D due to multiple testing, which reduced the significant P value threshold to <0.005 for this dataset. Statistical tests were carried out using GraphPad Prism version 8.