Cyclin-Dependent Kinase-9 Is a Therapeutic Target in MYC-Expressing Diffuse Large B-Cell Lymphoma Free

Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma (NHL) that accounts for >10,000 deaths per year in the United States alone (1). The distinction of DLBCL into cell-of-origin (COO) subsets based on gene expression patterns reminiscent of germinal center B-cell (GCB) and activated B-cell (ABC) has biological implications. Although inhibitors of B-cell receptor (BCR)–associated kinases (e.g., ibrutinib) have transformed the therapeutic paradigm in chronic lymphocytic leukemia (CLL), they show modest activity in other NHL subtypes, including DLBCL. Furthermore, the emergence of ibrutinib resistance as well as the inferior efficacy of novel targeted agents in aggressive lymphomas adds to the unmet clinical need (2). Finally, DLBCLs with rearrangement of MYC, particularly coupled with rearrangement of BCL2/6 (i.e., double-hit lymphomas) are associated with particularly poor outcomes and chemo-refractoriness (3).

Cyclin-dependent kinases (CDK) are serine/threonine kinases first discovered for their role in cell-cycle regulation. Each of the known CDKs possesses a catalytic subunit that associates with specific regulatory cyclin subunits. Such CDK–Cyclin complexes become active during the different phases of the cell cycle and regulate its coordinated progression (4). Yet CDK activity is not restricted to cell-cycle proteins. CDKs fall into a number of subfamilies mainly represented by cell-cycle–related (e.g., CDK1/2/4/6), and transcriptional CDKs (CDK7/8/9/11/20). Among the latter, CDK7/cyclin H, components of the general transcription factor TFIIH, and CDK9/cyclin T, the catalytic core of the positive transcription elongation factor b (pTEFb) facilitate initiation and elongation of RNA transcription, respectively.

Pan-CDK inhibitors (flavopiridol, dinaciclib) demonstrated clinical efficacy in lymphoid malignancies. Downmodulation of Mcl-1, a short-lived pro-survival Bcl-2 family protein has been considered a key event accounting for their pro-apoptotic activity in neoplastic B cells (5–7). Yet we and others have shown that cell-cycle– and pan-CDK inhibition–induced apoptosis is not strictly linked to Mcl-1 in cancer, and that alternate mechanisms such as mitotic catastrophe and ER stress may be responsible (8–10). Here, we investigated the mechanistic underpinnings of pharmacologic inhibition of CDK9 using AZ5576 in DLBCL models. We found that in addition to Mcl-1, selective targeting of CDK9 downmodulated MYC mRNA and protein, and facilitated MYC protein turnover. Susceptibility of MYC-expressing tumors to CDK9 inhibition opens a novel avenue for targeting MYC-driven lymphomagenesis.

The following cell lines were obtained from the ATCC: Raji, Nu-DUL-1, SU-DHL4, SU-DHL6, SU-DHL10, and SU-DHL16. OCI-LY18, U-2932, and VAL cells were obtained from DSMZ (Braunschweig, Germany, in March 2013); OCI-LY3 and OCI-LY19 cells were a kind gift from Dr. Andrew Evens (Tufts University, in July 2011). Raji^T58A mutation was confirmed by Sanger sequencing. Cell were used up to 10 passages and every six months underwent testing with MycoAlert Mycoplasma Detection Kit (Lonza; latest testing in February 2019).

Following approval by the Institutional Review Board, primary DLBCL cells were obtained from lymph node biopsies of patients treated at the Center for Hematologic Malignancies at Oregon Health and Science University. Written informed consent was obtained from patients (OHSU IRB#4422 and #4918). Studies were conducted in accordance with the Declaration of Helsinki. Isolation of peripheral blood mononuclear cells (PBMC) was performed using standard Ficoll-Hypaque technique (Amersham). Primary cells were cultured in RPMI-1640 medium supplemented with 15% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L l-glutamine, 25 mmol/L HEPES, 100 μmol/L non-essential amino acids, and 1 mmol/L sodium pyruvate (Lonza). All experiments were performed with freshly isolated cells.

Mouse fibroblast cell line (L cells) engineered to express CD40L (L4.5) was kindly provided by Dr. Sonia Neron (Hema-Quebec, Quebec, Canada; ref. 11). They were maintained in DMEM-1640 medium with 10% FBS, 100 U/mL penicillin and 100 μg/mL streptomycin. Primary DLBCL cells were cultured under standardized conditions on stroma as previously described (12). Briefly, stromal cells were seeded to achieve 80% to 100% confluence; on the following day, cells were plated at a 50:1 ratio and incubated at 37°C in 5% CO₂. At harvest, lymphoma cells were gently washed off the stromal layer.

To measure cell apoptosis, cells were resuspended in 150 μL of Annexin V binding buffer containing 1 μL of Annexin V-PE, 1 μL of 7-aminoactinomycin D (7-AAD) and 1 μL of CD19-mAbs (Southern Biotech) followed by flow cytometry on FACSCanto or FACSAria (BD Biosciences). Flow-cytometry analysis was performed using FlowJo software (Tree Star). Where applicable, cells were stained with Ki-67 antibodies with aid of the Inside Stain Kit (both from Miltenyi Biotec).

AZ5576 (13) was provided by AstraZeneca; dinaciclib, seliciclib, and flavopiridol were obtained from Selleck Chemicals. To measure cell proliferation, cells were plated in 96-well plates (3,000/well in 100 μL, 6 wells per sample) with drugs and incubated for 48 hours at 37°C in 5% CO₂. After incubation, relative numbers of viable cells were measured using a tetrazolium-based colorimetric assay (CellTiter Aqueous One Solution Cell Proliferation Assay; Promega).

Cells were lysed in radioimmunoprecipitation (RIPA) buffer (20 mmol/L Tris, 150 mmol/L NaCl, 1 % NP-40, 1 mmol/L NaF, 1 mmol/L Sodium phosphate, 1 mmol/L NaVO3, 1 mmol/L EDTA, 1 mmol/L EGTA), supplemented with protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail 2 (Sigma-Aldrich) and 1 mmol/L phenylmethanesulfonyl fluoride (PMSF). Proteins were analyzed by immunoblotting as previously described (8). The following antibodies were used: β-Actin, Bcl-2, Bcl-xL, GAPDH, phosphor-IKKα/β, c-MYC, phospho-MYC^Ser62, MAX, Mcl-1, NIK, phospho-p65, p105/50, p100/52, PARP and cleaved PARP and phospho-Rb^Ser780 (Cell Signaling Technology), phospho-Rb^Thr821 (Life Technologies), Rb (C-15, Santa Cruz Biotechnology), c-Rel, RelB, phospho-RNA Polymerase II Ser2 (H5) and Ser5 (H14), total RNA polymerase (8WG16, all from Covance) and horseradish peroxidase–conjugated anti-mouse and anti-rabbit antibodies (Bio-Rad). Densitometry was performed using ImageJ software. In each case, a representative image of at least three independent immunoblotting experiments is shown.

For immunoprecipitation experiments, cell protein lysates were pre-cleared and incubated at 4°C overnight with 2 μg of the indicated primary antibody or with rabbit IgG as isotype-specific control (Cell Signaling Technology). Lysates were incubated with 20 μL of 50% protein A agarose beads slurry (Cell Signaling Technology) for 1.5 hours at 4°C. After washes, samples were heated to 95°C for up to 5 minutes and analyzed by immunoblotting. 10% of source protein was used as input control.

Total RNA was isolated using the ENZA Total RNA Kit I and Homogenizer Mini Columns (Omega Bio-Tek). cDNA was synthesized from 500 ng of RNA using the qScript cDNA Supermix (QuantaBio). Quantitative real-time PCR (RT-PCR) was performed using a QuantStudio 7 Flex (Applied Biosystems) using PerfeCTa FastMix II according to the manufacturer's instructions (Quantabio) with template cDNA and gene-specific probes. The following probes were used: Mcl-1, Hs01050896_m1; MYC, Hs00153408_m1; CDK9, Hs00977896_g1; CAD, Hs00983188_m1; NPM1, Hs02339479_ m1; NCL, Hs01066668_m1. Amplification of the sequence of interest was compared with a reference probe (GAPDH, Hs02758991_g1, all from Life Technologies). All samples were analyzed in duplicate. We used the comparative C_t method for relative quantitation (2^−ΔΔCt, where ΔΔC_t = ΔC_tP – ΔC_tK; P = probe and K = reference sample).

A total of 2 × 10⁵ cells were fixed in ice-cold 70% ethanol while being vortexed, incubated on ice for 15 minutes, washed in PBS and resuspended in 250 μL of staining solution containing 20 ng/mL propidium iodide, 200 ng/mL RNAse A (Sigma-Aldrich), 0.1% Triton-X 100 in PBS. Cells were incubated for 15 min and submitted to flow cytometry. Cell-cycle analysis was performed using FlowJo software (Tree Star).

To measure EdU incorporation into newly synthesized DNA, 2 × 10⁵ cells were treated with drug for 4 hours followed by 2 hours of incubation with 1 μmol/L EdU. Cells were then collected, washed with PBS, fixed in 2% paraformaldehyde and stored at 4°C until analysis. Subsequently, cells were processed using Click-iT EdU AlexaFluor 647 Assay Kit (Life Technologies) according to the manufacturer's instruction, and subjected to flow cytometry. Analysis was performed using the FlowJo software.

Please see Supplementary Materials and Methods. Full results are available at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE113035.

HEK293 cells were co-transfected with experimental plasmid (pcDNA3-V5-MYC, a kind gift from Dr. Sun, Oregon Health and Science University; ref. 14) or empty vector control and pGL2M4-luc reporter plasmid (ref. 15; containing four CACGTG binding sites, a canonical E-box) or control (pBV-luc, Promega) along with pCMVβ vector (Clontech Laboratories). The amount of DNA per transfection was kept constant by using empty pcDNA3.1 vector. Cells were harvested 48 hours post-transfection and luciferase assays were performed using the Dual-Luciferase Reporter Assay System (Promega). Relative light units were determined using a luminometer (Synergy II, BioTek) for firefly luciferase. β-galactosidase activity was determined using a colorimetric method to normalize transfection efficiency.

All animal studies were carried out in accordance with the institutional guidelines and approved by OHSU IACUC. A total 5 × 10⁶ OCI-LY3 or VAL cells were mixed with 50% Matrigel (BD Biosciences) and injected subcutaneously into the right flank of 5–8-week-old NOD/SCID/γC_null (NSG) mice. When tumors reached 10 mm, approximately 10 to 14 days post-implantation, mice were treated with 60 mg/kg AZ5576 (in 20% cyclodextrine) by oral gavage twice weekly or vehicle control. Tumor volume was assessed every other day and animals were sacrificed when tumor diameter exceeded 20 mm or after loss of greater than 10% body weight. The data are expressed as average tumor diameter (mm) per mouse as a function of time, normalized to pre-treatment value. Each group had 8 to 10 mice.

Two hours after the last injection, mice were sacrificed. Tumors were passed through a cell strainer. Cell suspension was then assayed for apoptosis (Annexin V) and proliferation (Ki-67) by flow cytometry. Whole-cell protein lysates and RNA was isolated as described elsewhere.

293T17 cells (ATCC) were transiently transfected with MYC-expressing vector (pCDH-puro-cMyc, Addgene #46970), lenti-sh1368 knockdown c-myc (Addgene #29435), or vector control (16, 17). Lentiviral particles were produced by three-plasmid transfection system using Δ8.9 and VSVG plasmids (both from Addgene) and JetPrime transfection reagent according to the manufacturer's protocol (Polyplus-transfection), as previously described (18). OCI-LY3 and U-2932 cells were infected with viral particles by spinoculation with 1 μg polybrene and 1 M HEPES for 90 minutes at 2500 RPM and 37°C. 48 hours thereafter, MYC-overexpressing and control OCI-LY3 cell pools were selected in 2 μg/mL puromycin, whereas U-2932 shMYC and control pools were selected by flow sorting for GFP (FACSAria). MYC expression was tested by RT-PCR and immunoblotting.

Statistical analysis was performed with Student t test in GraphPad Prism software. P < 0.05 was considered to be statistically significant. *, P < 0.05 and **, P < 0.01 throughout the article.

We and others have previously demonstrated the pre-clinical activity of pan-CDK inhibitors in B-cell neoplasia, which has been in part explained by global transcriptional shutdown via inhibition of CDK9 (5, 6). Hence, we evaluated the in vitro activity of the selective CDK9 inhibitor AZ5576 (Supplementary Fig. S1), in DLBCL. Upon 24-hour treatment, CDK9 inhibition induced apoptosis of the DLBCL cells in a dose-dependent manner, independent of COO (Fig. 1A). By contrast, AZ5576 exerted minimal cytotoxicity toward healthy donor lymphocytes (Supplementary Fig. S2A). Drug washouts confirmed cytotoxicity of CDK9 inhibition in DLBCL cells, indicating that short-term drug exposure induced lasting effects (Supplementary Fig. S2B). Meanwhile, long-term exposure to AZ5576 in vitro decreased proliferation of the DLBCL cell lines, including OCI-LY3, likely indicating non-specific effects (Supplementary Fig. S2C).

Following treatment of DLBCL cells with AZ5576, we observed a rapid decrease in phosphorylation of RNA polymerase II (RNAPII) at Ser2 position, consistent with inhibition of CDK9 (Fig. 1B; Supplementary Fig. S3A). A 4-hour incubation of DLBCL cells with 0.033–0.1 μmol/L AZ5576 led to a pronounced decrease, whereas 8-hour exposure to 0.3 μmol/L AZ5576 led to loss of RNAPII phosphorylation at that site. Meanwhile, phosphorylation of RNAPII at Ser5, which is targeted by CDK7, was not affected. In contrast with non-selective CDK inhibitors, Rb phosphorylation at residues T821 and S780 remained unchanged, indicating that interphase CDKs (2/4/6) were not inhibited (8, 10). AZ5576 did not abrogate NFκB signaling (Supplementary Fig. S3B), further confirming its selectivity. Measurement of EdU incorporation as well as cell cycle analysis demonstrated that, like dinaciclib, short-term treatment with AZ5576 suppressed DNA synthesis in DLBCL cells and diminished S phase entry (Fig. 1C; Supplementary Fig. S4).

We confirmed sensitivity of primary neoplastic B cells to CDK9 inhibition. B cells derived from the lymph nodes of patients with DLBCL underwent apoptosis upon exposure to AZ5576 (Fig. 1D; white bars). Given that the tumor microenvironment is known to play a critical role in DLBCL (19), we have conducted a partial reconstitution of the lymph node microenvironment by using the CD40L-expressing stroma, as previously reported by us (12). Under those conditions, primary neoplastic B cells are partially rescued from apoptosis, in part via upregulation of the anti-apoptotic proteins Mcl-1 and Bcl-xL. Primary DLBCL cells co-cultured with the CD40L-expressing stroma acquired partial resistance to AZ5576 (Fig. 1D, black bars). Of note, 5/10 samples harbored aberrant MYC: three patients had double-expressor lymphoma (high MYC protein expression by immunohistochemistry) and two patients had double-hit lymphoma (MYC and BCL2 rearranged).

We then conducted a study of the pre-clinical efficacy of AZ5576 in vivo. Mouse xenograft models using GCB-DLBCL (double-hit) cell line VAL, which readily underwent apoptosis with AZ5576, and ABC-DLBCL cell line OCI-LY3, a relatively resistant cell line (<15% apoptosis with 1 μmol/L AZ5576 at 24 hours), were established as described in the Methods. Twice-weekly treatment with AZ5576 60 mg/kg delayed growth of DLBCL tumors (Fig. 2). Xenografted cells from mice treated with AZ5576 demonstrated increased apoptosis and decreased proliferation (Fig. 2). Although proliferation of both VAL and OCI-LY3 xenografts was significantly reduced, apoptosis was more pronounced in VAL xenografts, reminiscent of the in vitro experiments (Fig. 1A). Treatment with AZ5576 did not result in weight loss or any apparent toxic effects in mice (Supplementary Fig. S5; ref. 13).

Thus, selective targeting of CDK9 induces apoptosis of DLBCL cells independent of COO in vitro and in vivo.

We next employed gene expression profiling by RNA-Seq to determine which pathways were deregulated by CDK9 inhibition in DLBCL cells. VAL and OCI-LY3 cells were treated with 0.3 μmol/L AZ-5576 for 3 or 6 hours and comparisons in gene expression were made to vehicle-treated cells. Expression of 16,451 protein coding genes was detected in both OCI-LY3 and VAL cells. Using a cutoff of at least 1.5-fold change we identified between 3,000 and 6,000 genes whose expression was significantly downregulated by AZ5576 (padj<0.05; Fig. 3A, Supplementary Table S1) depending on the cell line and duration of treatment. We analyzed the downregulated genes for functional significance. After adjustment for multiple comparisons using Benjamini and Yekuteli method it was revealed that the “Generic transcription pathway” was most significantly associated with the downregulated genes in VAL and OCI-LY3 cell lines at both timepoints (P < 10⁻¹⁶). Additional pathways commonly downregulated in both cell lines included MAPK signaling, PI3K signaling, NFκB and tumor necrosis factor receptor signaling amongst others (P < 0.05; Fig. 3B). Interestingly, in the face of inhibited global transcription, we found a significant number of genes (>1,000) and pathways (>50) upregulated following treatment with AZ5576 (Supplementary Table S1). Among those, post-translational protein phosphorylation pathway was the most significantly upregulated (P < 0.001). The integrin-mediated pathway was upregulated in OCI-LY3 cells following 6 hours of drug treatment.

Because the mechanism of action of pan-CDK inhibitors has previously been linked to downmodulation of the anti-apoptotic protein Mcl-1 (5, 6, 8), we investigated the effect of AZ5576 on Mcl-1 and MYC, short-lived proteins with T_1/2 of approximately 30 minutes (5, 20). CDK9 inhibitor quickly downregulated Mcl-1 and MYC protein abundance across multiple tested DLBCL cell lines and in primary DLBCL cells (Fig. 4A). A decrease in Mcl-1 and MYC mRNA and protein in DLBCL cells was evident upon short-term drug exposure (Fig. 4B and C; Supplementary Fig. S6), including in the above RNASeq experiment. Interestingly, there was poor correlation between Mcl-1 expression and DLBCL COO in our panel of cell lines; in addition, DLBCL cells abundantly expressed other anti-apoptotic Bcl-2 family members, including Bcl-xL and Bcl-2 (Fig. 4D).

Primary DLBCL cells co-cultured with CD40L-expressing stroma demonstrated loss of Mcl-1 upon treatment with AZ5576 (Fig. 4E). MYC protein decrease was less pronounced in CD40L-stimulated cells upon treatment with AZ5576, compared with off stroma control (Fig. 4E).

Finally, in vivo treatment with AZ5576 resulted in decreased phosphorylation of RNAPII at Ser2, downmodulation of Mcl-1, and robust reduction of MYC levels in mouse xenografted tumors, whereas Rb phosphorylation was similar between experimental and control groups (Fig. 4F).

Thus, CDK9 inhibition induces loss of Mcl-1 and MYC in DLBCL cells.

MYC is expressed in almost all cases of DLBCL and its high expression predicts therapeutic resistance, particularly when accompanied by overexpression of Bcl-2 (double-hit or double-expression lymphomas; ref. 3). Robust pharmacologic disruption of MYC-driven oncogenesis has thus far been elusive. Because we observed rapid loss of MYC mRNA transcript and protein following CDK9 inhibition, we determined the effect of MYC protein levels on AZ5576-mediated apoptosis. We noted that MYC protein expression levels of unsynchronized DLBCL cells correlated with sensitivity to AZ5576-induced apoptosis, but not proliferation (Fig. 5A; Supplementary Fig. S7). Furthermore, engineered overexpression of MYC in OCI-LY3 cells, which exhibit relatively low MYC protein levels, sensitized them to CDK9 inhibition as indicated by a statistically significant increase in apoptosis (Fig. 5B; Supplementary Fig. S8A). To address non-specific effects, we treated DLBCL cells with doxorubicin and with etoposide (chemotherapeutic agents commonly used in treatment of DLBCL), and found that MYC overexpression had no effect on apoptosis induced by these agents (Supplementary Fig. S8B). In addition, genetic downregulation of MYC in U-2932 and OCI-LY19 cells resulted in diminished susceptibility to AZ5576 (Fig. 5C; Supplementary Fig. S9).

We next examined whether targeting CDK9 downregulated MYC-driven transcription. AZ5576 resulted in dose-dependent downregulation of MYC transcription of a luciferase reporter containing multiple canonical MYC-MAX–binding sites, an effect more pronounced compared with multi-CDK inhibitor dinaciclib (Fig. 5D). We mined our RNA-Seq experiment to evaluate the effect of AZ5576 on MYC transcriptional targets. We analyzed >1,100 known MYC transcriptional targets (21). The expressed MYC target genes were predominantly downregulated following treatment with AZ5576. 155 and 189 genes were downregulated in OCI-LY3 cells and 225 and 254 genes were downregulated in VAL cells ≥1.5-fold at 3 and 6 hours, respectively (P < 0.05; Supplementary Fig. S10). By contrast, only a minority MYC target genes were found upregulated upon short-term exposure to AZ5576. We then employed GSEA to assess the distribution of a list of 58 canonical MYC targets within our sequencing dataset. Similar to the gene ontology analysis, we observed the majority of the genes to be strongly underrepresented with a normalized enrichment score less than −1 (P< 0.001; Fig. 5E; Supplementary Fig. S11A). The rapid downmodulation of several MYC transcriptional targets (CAD, NPM1, and NCL) was confirmed in DLBCL cells (Supplementary Fig. S11B).

MYC protein stability is tightly regulated: although phosphorylation at Ser62 enhances MYC stability, pMYC^Thr58 is primed for degradation (22). It has been previously suggested that CDKs may be responsible for MYC phosphorylation at Ser62 (23). Such an interaction is expected to result in lengthening of MYC half-life. By contrast, CDK inhibition would be expected to enhance MYC turnover. To investigate the effect of AZ5576 on MYC half-life, we conducted cycloheximide chase experiments. We found that MYC degradation indeed was accelerated in VAL cells treated with AZ5576 (Fig. 6A). By contrast, targeting CDK9 did not increase Mcl-1 protein turnover (Supplementary Fig. S12A). We then used AZ5576 as well as dinaciclib and seliciclib (non-specific CDK inhibitors which also inhibit CDK9 with IC₅₀ values of 4 nmol/L and 0.79 μmol/L, respectively) and evaluated their effect on MYC phosphorylation. All drugs led to loss of Ser62 phosphorylation, accompanied by downregulation of MYC (Fig. 6B). Loss of RNAPII phosphorylation preceded those events. To ascertain that MYC phosphorylation was not lost simply due to protein degradation, we treated DLBCL cells with a proteasome inhibitor prior to adding AZ5576. In this context, CDK9 inhibition rapidly decreased pMYC^Ser62 in DLBCL cells, despite total MYC levels being stable (Fig. 6C, left). A similar pattern of dose-dependent reduction in MYC^Ser62 and MYC was found in OCI-LY3 cells engineered to overexpress MYC, without an effect on MYC partner MAX (Supplementary Fig. S12B). By contrast, CDK9i failed to similarly dephosphorylate MYC^Ser62 in Raji cells, which carry a stabilizing MYC^T58A mutation (Fig. 6C). Correspondingly, Raji cells demonstrated delayed MYC degradation and intermediate sensitivity to AZ5576 (Fig. 6C; Supplementary Fig. S12C), suggesting that MYC turnover may play a role in sensitivity to apoptosis triggered by CDK9 inhibition in DLBCL.

Finally, CDK9 was previously shown to form a complex with MYC (24). We therefore examined whether enhanced MYC degradation by AZ5576 might correspond to disruption of the CDK9-MYC complex. Immunoprecipitation with MYC monoclonal antibodies showed MYC in a complex with CDK9 in DLBCL cells, but AZ5576 had no effect on association between MYC and CDK9 (Fig. 6D).

Together, these results suggest that inhibition of CDK9-dependent MYC phosphorylation at Ser62 by AZ5576 may contribute to accelerated MYC turnover and, together with RNA PolII inhibition, cause selective DLBCL cytotoxicity and inhibited tumor growth.

We and others have shown that pan-CDK inhibitors (flavopiridol, dinaciclib and others) exert antitumor effects via a number of mechanisms, including induction of ER stress and the unfolded protein response, inhibition of autophagy, NFκB pathway, and the induction of mitotic catastrophe (8–10, 25, 26). Thus, they lack a refined mechanism of action and therefore have a narrow therapeutic window. The emergence of a new generation of selective inhibitors has enabled targeting tumors with improved effectiveness and fewer adverse effects. For example, CDK4/6 inhibitors (e.g., palbociclib), disrupt G₁-to-S-phase cell-cycle transition with surgical precision, allowing for decreased toxicity in comparison with their pan-CDK counterparts (27). We postulated that selective targeting of transcriptional CDKs will be associated with a similarly enhanced therapeutic profile and offer an alternative approach in DLBCL.

B-cell lymphomagenesis depends heavily on the activities of the MYC oncoprotein, a transcription factor that regulates cell growth and proliferation (28, 29). In mouse models, deregulated expression of MYC is both sufficient for tumor initiation and indispensable for tumor maintenance, and thus inhibiting MYC function is an attractive therapeutic strategy (30–35). Importantly, Eμ-driven MYC induces lymphoid malignancies in transgenic mice (30), and second hits affecting regulators of apoptosis (BCL2, TP53) enhance MYC transformation (36, 37). Commensurate with this, high expression of MYC is associated with chemoresistance and poor outcomes in multiple NHL subtypes, whereas “double-hit” lymphomas co-expressing MYC and Bcl-2 or Bcl-6 are most aggressive (3). Use of bromodomain protein inhibitors that act in part by suppressing MYC transcriptional activity demonstrate the potential of MYC suppression in pre-clinical models (38). BRD4 protein levels are asymmetrically loaded across super-enhancers, and their targeting with bromodomain inhibitors such as JQ1 results in preferential downmodulation of oncogenic transcriptional programs, particularly MYC and E2F1 (39). Although BRD4 inhibitors show strong antitumor properties pre-clinically, the development of specific BRD4 inhibitors has remained a challenge. Pan-BRD inhibitors such as JQ1 are toxic, and even newer compounds such as PLX511107 show inhibition of multiple BRDs due to the conserved nature of the BD1 binding region across the family. This in turns leads to modulation of multiple biologic processes such as insulin production, T cell differentiation and adipogenesis (40, 41). The rapid downregulation of MYC by a specific CDK9 inhibitor to target transcription, as shown here, suggests an alternative therapeutic approach. In this context, it is important to note that sustained CDK9 inhibition may in fact lead to compensatory induction of MYC via BRD4-dependent mechanism, and thus simultaneous inhibition of CDK9 and BRD4 can be an efficient approach (42); however, clinical applicability of this approach is unclear.

We verified that AZ5576 acted as a selective CDK9 inhibitor in DLBCL. AZ5576 demonstrated a selective effect on RNAPII phosphorylation at Ser2, a CDK9-specific site, yet did not inhibit CDK7, a CDK involved in initiation of transcription. CDK9 inhibition restricted the viability of multiple DLBCL cell lines independent of COO, both in vitro and in vivo. We and others have already reported on the pro-apoptotic effects of voruciclib, a pan-CDK inhibitor, in lymphoid malignancies, where it led to downregulation of Mcl-1 when used in μmol/L concentrations (7, 8). By contrast, AZ5576 was active at much lower concentrations, with 33–100 nmol/L leading to loss of pRNAPII^Ser2 in DLBCL cells at 4 to 8 hours. As previously reported, targeting CDK9 led to downmodulation of Mcl-1 mRNA transcript and protein. Importantly, we found that CDK9 inhibition concurrently decreased MYC. In fact, MYC protein levels were strongly decreased by AZ5576 in primary DLBCL cells in vitro, including under the conditions which partially mimic tumor microenvironment and in xenografted mouse tumors. Although DLBCL cells with high MYC levels were more susceptible to AZ5576-induced apoptosis, and engineered overexpression of MYC sensitized DLBCL cells to apoptosis, the drug also had a pronounced anti-proliferative effect in cells with relatively low expression of MYC, likely indicating dependence of DLBCL on both MYC transcriptional program, as well as due to MYC-independent effects. Huang and colleagues (43) have shown that dual CDK7/9 inhibition leads to robust antitumor effects in an MYC-driven model of hepatocellular carcinoma, accompanied by the disruption of MYC-dependent transcription. Meanwhile, a selective CDK9 inhibitor (LDC67) was shown to downmodulate MYC in carcinoma models (44). In those models, high MYC expression levels were required to establish sensitivity to CDK inhibition. Here for the first time we show that endogenous MYC levels play an important role in determining sensitivity to a CDK9-specific inhibitor.

The effect of enhanced MYC turnover following CDK9 inhibition is unexpected and appears to be dependent on MYC phosphorylation. Loss of MYC phosphorylation at Ser62, a residue associated with enhanced MYC protein stability, preceded degradation of MYC. MYC phosphorylation is tightly regulated, where signaling via RAF/MEK/ERK and PI3K/AKT pathways leads to ERK-mediated phosphorylation of MYC at Ser62 and inhibition of GSK3β with concomitant loss of phosphorylation at Thr58 (which labels MYC for degradation). Together, those events enhance MYC stability. The MYC^T58A mutation is oncogenic as it renders MYC resistant to degradation (22), and is found in Burkitt lymphoma and, rarely, in DLBCL (45). It remains unclear how exactly CDK9 might regulate MYC phosphorylation. We found that CDK9 inhibition does not disrupt the MYC–CDK9 complex. Likewise, direct phosphorylation of Ser62 by CDK9 is unlikely given that treatment with AZ5576 does not lead to loss of pMYC^Ser62 in Raji cells that carry a T58A mutation. Although phosphorylation of MYC at Thr58 requires prior phosphorylation of Ser62 (20), the reverse is not known to be true, that is, loss of CDK9 activity should lead to loss of pMYC^Ser62, if CDK9 phosphorylated MYC directly. It is likely that CDK9 regulates an unknown intermediary involved in MYC phosphorylation. Nevertheless, CDK9 inhibition-mediated loss of Ser62 appears an important contributor to subsequent MYC degradation and cell apoptosis. Given the pre-clinical efficacy of AZ5576 in MYC-expressing DLBCL, CDK9 inhibition may hold promise as a therapeutic approach in DLBCL with MYC rearrangement and/or overexpression, an area of unmet clinical need.

Previous reports suggested that ABC-DLBCL may be a prime target for Mcl-1 inhibitors, as Mcl-1 expression was more frequently expressed in ABC-DLBCL tumors (48%), compared with GCB-DLBCL tumors (25%; ref. 46). The anti-apoptotic proteins Bcl-2 and Bcl-xL are known to contribute to MYC-driven lymphomagenesis and thus are expected to compensate for the loss of Mcl-1 (47, 48). Thus, combination strategies employing BH3-mimetics and CDK9 inhibitors may lead to enhanced tumor killing. In this context, we have previously demonstrated a cooperative effect of voruciclib (a pan-CDK inhibitor) and navitoclax, a Bcl-2/X inhibitor, in primary neoplastic B-cells (8).

No single subset of MYC targets accounts for its oncogenic activity, suggesting a role distinct from that of conventional transcription factors. Indeed, MYC is known to bind to and increase transcription of thousands of genes. Inducing MYC expression in cells leads to global increases in gene expression. This is accomplished in part by MYC functioning as a general transcriptional amplifier, in which its binding to the core promoter of actively transcribed genes stimulates RNAPII activity, thereby enhancing the existing gene expression program in healthy and neoplastic cells (49, 50). MYC's interaction with CDK9/Cyclin T1 within the pTEFb elongation complex is the key mechanism underlying its ability to stimulate RNAPII. Thus, the role of MYC as a transcriptional amplifier and oncogenic transcription factor is anticipated to be highly susceptible to disruption within the pTEFb complex. Here, we have not investigated this aspect of CDK9 inhibition, but the studies to elucidate MYC–chromatin interactions are ongoing in our laboratory.

In sum, we provide pre-clinical evidence that MYC-expressing aggressive lymphoma is susceptible to selective targeting of CDK9. In this context, it is hoped that CDK9 inhibitors will have an improved therapeutic window and enhanced clinical activity compared with pan-CDK inhibition, and are thus worthy of clinical investigation in NHL.

L. Drew has ownership interest (including stock, patents, etc.) at AstraZeneca. A.V. Danilov reports receiving a commercial research grant from Astra Zeneca, Gilead Sciences, Takeda Oncology, Bristol Myers Squibb, Verastem, and Bayer Oncology; and is a consultant/advisory board member for Astra Zeneca, Verastem Oncology, Gilead Sciences, Curis, Seattle Genetics, and Genentech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Hashiguchi, N. Bruss, C.J. Paiva, P. Hurlin, A.V. Danilov

Development of methodology: T. Hashiguchi, N. Bruss, A.V. Danilov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Hashiguchi, N. Bruss, S. Best, V. Lam, O. Danilova, C.J. Paiva, J. Wolf, E.W. Gilbert, A.V. Danilov

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Hashiguchi, N. Bruss, O. Danilova, P. Hurlin, A.V. Danilov

Writing, review, and/or revision of the manuscript: T. Hashiguchi, N. Bruss, C.J. Paiva, P. Kaur, L. Drew, J. Cidado, A.V. Danilov

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Hashiguchi, C.Y. Okada, L. Drew, J. Cidado, A.V. Danilov

Study supervision: T. Hashiguchi, A.V. Danilov

This study was supported by the Leukemia and Lymphoma Society Translational Research Program Award #6542-18, by the Knight Cancer Institute Albert and Elaine Borchard Foundation Pilot Project Award, and by Astra Zeneca (all to A.V. Danilov). A.V. Danilov is a Leukemia and Lymphoma Society Scholar in Clinical Research (#2319-19).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.