Characterization of a dual BET/HDAC inhibitor for treatment of pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is resistant to virtually all chemo‐ and targeted therapeutic approaches. Epigenetic regulators represent a novel class of drug targets. Among them, BET and HDAC proteins are central regulators of chromatin structure and transcription, and preclinical evidence suggests effectiveness of combined BET and HDAC inhibition in PDAC. Here, we describe that TW9, a newly generated adduct of the BET inhibitor (+)‐JQ1 and class I HDAC inhibitor CI994, is a potent dual inhibitor simultaneously targeting BET and HDAC proteins. TW9 has a similar affinity to BRD4 bromodomains as (+)‐JQ1 and shares a conserved binding mode, but is significantly more active in inhibiting HDAC1 compared to the parental HDAC inhibitor CI994. TW9 was more potent in inhibiting tumor cell proliferation compared to (+)‐JQ1, CI994 alone or combined treatment of both inhibitors. Sequential administration of gemcitabine and TW9 showed additional synergistic antitumor effects. Microarray analysis revealed that dysregulation of a FOSL1‐directed transcriptional program contributed to the antitumor effects of TW9. Our results demonstrate the potential of a dual chromatin‐targeting strategy in the treatment of PDAC and provide a rationale for further development of multitarget inhibitors.

Pancreatic ductal adenocarcinoma (PDAC) is the most common and lethal form of pancreatic cancer. It is now the fourth leading cause of cancer death in men and women. Due to late diagnosis and metastasis, most patients with PDAC undergo conventional chemotherapy and obtain improved albeit limited benefits from intensified and toxic chemotherapy regimens, inevitably leading to secondary resistance. 3 In 2015, we reported a promising epigenetic-based therapeutic strategy for PDAC that combined BET and HDAC inhibitors. 4 We have now generated a dual inhibitor simultaneously targeting BET and HDAC proteins. Compared to combination therapies of two or more drugs, multitarget drugs may achieve the same goal but utilize a single compound. The potential advantages of dual targeting are a linked pharmacokinetic profile, reduced risk of drug-drug interactions and simplified dosing scheduling. Here, we present the design and synthesis of dual BET/HDAC inhibitors, resulting in the identification of the potent dual inhibitor TW9. Importantly, we examined its biological effects in PDAC tumor cell lines, showing its potency as a novel chromatin-targeting approach for future clinical development for treatment of PDAC.

| Protein expression, purification and crystallography
The first two bromodomains of BRD4, BRD4(1) and BRD4(2), were each expressed in Escherichia coli as His-tagged proteins and were purified by nickel-affinity and gel-filtration chromatography as described. 5 After the final gel filtration step on a Superdex-75 column in 10 mM HEPES, pH 7.5, 150 mM NaCl, 0.5 mM TCEP and 5% glycerol, the protein was concentrated to 10 mg/mL, flash-frozen in liquid nitrogen and stored at −80 C. For setting up co-crystallization trials, aliquots of the purified BRD4 (1)  and scaled with AIMLESS, 7 which is implemented in the CCP4 package. 8 The structures were solved by difference Fourier analysis in PHENIX 9 using PDB entry 3MXF as a starting model with initial rigidbody refinement. The structural models were then refined using iterative cycles of manual model building with COOT 10 and refinement in PHENIX. Dictionary files for the TW9, TW12 and TW22 ligands were generated using the Grade Web Server (http://grade.globalphasing. org). There was excellent electron density for the (+)-JQ1 moiety in all three structures. For TW9 and TW12, there was however not sufficient electron density to unambiguously model the solvent-exposed HDACi moiety. The disordered parts of the ligands were therefore deleted in the final model. In contrast, the HDACi moiety of TW22 was well resolved due to interaction with a symmetry-related molecule. Data collection and refinement statistics are listed in Table S1.

| XTT cell viability assay
XTT cell viability assays were performed as described, 11 following the XTT kit manufacturer's instructions. Briefly, HEK293T cells were seeded into flat-bottomed 96-well plates 24 hours before the addition of test compound at a density of 2.5 × 10 5 cells/mL (25 000 cells/ well). Then, 100 μL of test compound was added at a concentration range from 10 μM to 170 pM as final concentrations and incubated for 24 hours. For XTT measurement, 50 μL of a solution containing XTT (Sigma) and phenazine monosulfate (Sigma) were added to give a final concentration of 300 μg/mL and 10 μM, respectively, and the reaction was incubated for 30 minutes before measuring absorbance at 475 nm (specific signal) and 660 nm (nonspecific signal). The signal was background-corrected using the following formula: specific absor-

| NanoBRET target engagement intracellular assay
NanoBRET target engagement assays were performed following published protocols. 12,13 Briefly, plasmids for full-length HDAC1 and BRD4 as well as the isolated bromodomains of BRD4 containing either a C-terminal (HDAC1) or an N-terminal (BRD4) placement of NanoLuc were obtained from the manufacturer (Promega, Madison, Wisconsin).
To lower intracellular expression levels of the reporter fusion, the NanoLuc/kinase fusion constructs were diluted into transfection carrier DNA (pGEM3ZF-, Promega) at a mass ratio of 1:10 prior to forming

| CellTiter Glo cell viability assay
The assay was performed according to the manufacturer's instructions

| RNA extraction and quantitative RT-PCR analysis
Total RNA was isolated using Maxwell RSC simplyRNA Cells Kit (Promega) according to the manufacturer's protocol. cDNA was synthesized using PrimeScript Reverse Transcriptase (TaKaRa, Kusatsu, Japan). cDNA was amplified using LightCycler 480 SYBR Green I Master (Roche Diagnostics, Indianapolis, Indiana) and the amplicon was detected by SYBR Green I using LightCycler 480 instrument (Roche).
PCR conditions were 5 minutes at 95 C, followed by 45 cycles of 95 C for 10 seconds, 59 C for 10 seconds and 72 C for 20 seconds.
The relative gene expression levels were normalized to GUSB or GAPDH and calculated using the 2 −ΔΔCt method. The primer sequences are given in Table S2.

| Crystal violet staining for colony formation assay
Attached cells were rinsed once with PBS and fixed with ice-cold methanol for 10 minutes. Fixed cells were then stained by crystal violet solution (containing 0.1% crystal violet and 25% methanol) for 30 minutes. Cells were rinsed twice with H 2 O and dried overnight.

| Cell proliferation assay
MIA PaCa-2 cells (1200 cells/well) were seeded in 12-well plates (Corning). On the days of measurement, whole-well images were captured and cell confluence was analyzed by NYONE Image Cytometer (Synentec, Elmshorn, Germany).

| Kaplan-Meier survival analysis
The analysis was performed using the human protein atlas data set, 14 which contains a pancreatic cancer patient cohort (n = 176). The

FPKMs (number Fragments Per Kilobase of exon per Million reads)
were used for quantification of expression of FOSL1 gene. The cohort was stratified into two expression groups with the FPKM cut-off that yields the lowest P score. The correlation between the expression levels of FOSL1 and patient survival was examined by Kaplan-Meier survival estimator.
F I G U R E 1 Legend on next page.

| Bioinformatic analysis
Gene ontology (GO) analysis was performed using Gene Ontology Consortium tool (http://geneontology.org/). Gene set enrichment analysis (GSEA) was conducted using default settings on mean expression values from microarray data. 15 Sets included in the analysis ranged from 30 to 2000 genes, and gene set was used as a permutation type. For expression values for the same gene but different isoforms, values for lower expressed isoforms in control samples were disregarded. H3K27ac signal normalized to input was used to identify super-enhancers using the Ranking of Super Enhancer (ROSE) algorithm set at default settings and ignoring regions within 2500 base pairs of transcriptional start sites. 16,17 Genes associated with superenhancers were identified by Genomic Regions Enrichment of Annotations Tool (GREAT) using default settings. 18 Publicly available datasets were used for PANC-1 [E-MTAB-7034] 19 and healthy pan- 20 Mapping to the hg19 genome was performed using BOWTIE/2.2.5. 21 Bam files were generated by SAMTOOLS/ 1.  (Table S1). As expected, the binding mode of the (+)-JQ1 moiety of all three adducts was essentially the same as that of the parental compound (+)-JQ1 ( Figure 1B), and all key interactions with the bromodomain binding pocket were conserved, including the hydrogen bond of the triazole moiety with the highly conserved asparagine (Asn140). All HDAC moieties were solvent-accessible, suggesting that they are free to interact with HDACs even in the BET-bound state. In the case of the TW9 and TW12 complexes, the HDAC inhibitory moieties were disordered. In the structure of the TW22 complex, however, the HDAC inhibitor, which was derived from panobinostat, was visible due to stabilization by crystal contacts (Figures 1B and S1).  69, 52 and 21 nM, respectively, which was in the range of the K d value determined for the parental molecule (+)-JQ1 (51 nM; Table 1).
A representative ITC experiment for TW9 is shown in Figure 1C. K d values for BRD4(1) were all slightly higher, ranging from 230 nM for TW9 to 54 nM for TW22, which again was similar to the K d of (+)-JQ1. In summary, the ITC measurements demonstrated excellent BET binding activity of our dual inhibitors, indicating that introduction of the HDAC-inhibitor moiety did not significantly alter affinities for the two bromodomains present in BRD4.
Next, we tested the cellular BRD4 activity of the synthesized adducts using nanoBRET assays (Table 1 and Figure 1D). All nanoBRET experiments were carried out in HEK293T cells under nontoxic conditions ( Figure S2). TW9 showed the best BRD4-targeting activity, with an IC 50 value of around 700 nM for BRD4(1) ( Figure 1D).
The affinity of TW12 and TW22 for BRD4(1) in cells was notably weaker, with IC 50 values of 5.5 and 12 μM, respectively. A similar trend was observed for the cellular affinity of our dual inhibitors to BRD4(2). Notably, TW9 had a 10 times higher affinity for BRD4 (2) than for BRD4 (1), with an IC 50 value of 74 vs 720 nM, and bound BRD4(2) more than 20 times more strongly than did TW12 and TW22 ( Table 1). Comparison of the ITC and nanoBRET data suggests that the drastically reduced potency of TW12 and TW22 in cells compared to (+)-JQ1 was most likely due to reduced cellular uptake rather than differences in intrinsic binding affinity. We also measured cellular K d s for the full-length BRD4 protein, FL-BRD4. In all cases, the measured apparent binding constants were as expected between the values obtained for the two isolated bromodomains (Table 1).
After confirming the BRD4-targeting activity, we performed further NanoBRET assays to determine the cellular activity of our compounds against HDACs, the second target protein class of our dual inhibitors ( Table 2 and Figure 1E). TW9 was a potent inhibitor of HDAC1 with an IC 50 value in cells of about 300 nM. As such, it was more potent than the parental HDAC inhibitor CI994, which had an IC 50 of around 1 μM under the same conditions. TW12 was less potent than TW9 in targeting HDAC1, and TW22 displayed an even lower potency in cells with an IC 50 > 20 μM, corresponding to a potency more than three orders of magnitude lower compared to its parental compound panobinostat ( Table 2). We also performed a cellfree fluorogenic HDAC assay, which showed that TW9 and CI994 inhibited HDAC2 activity to a similar degree, with IC 50 values of 2.5 and 1.7 μM, respectively ( Figure S3). Based on the above results, we chose TW9 as the most potent BET/HDAC dual inhibitor for further characterization in cancer cells.

| TW9 preserves both BETi and HDACi activities in cancer cells
Both MYC and HEXIM1 are often used as readout biomarkers of BET inhibitor activity in cancer cells, while acetylated histone H3 is used as a direct readout for HDAC inhibitor activity. Expression analysis by RT-PCR showed prompt (6 hours) down-regulation of MYC and upregulation of HEXIM1 by TW9 to similar levels as (+)-JQ1 ( Figure S4A). Immunoblot analysis showed that MYC and histone H3 acetylated at lysines 9 and 14 were regulated in a dose-and timedependent manner by TW9 (Figure 2A,B), comparable to (+)-JQ1 and CI994, respectively. Consistent with the nanoBRET assays for HDACtargeting activity of TW9 (Table 2 and Figure 1E), histone H3 acetylation at lysine 9 and 14 was enhanced by TW9 at a much higher level, compared to the parental HDAC inhibitor CI994 (Figures 2B and   S4B). Intriguingly, when we treated cells with a higher dose (4 μM) of  Figure S4F). However, TW12 and TW22 failed to accumulate acetylated histone H3, indicating decreased activity of the HDAC-inhibitor moiety ( Figure S4F).  Figure 2F). Furthermore, TW9 induced gene expression of Involucrin, a differentiation marker, to a similar level as (+)-JQ1
3.4 | TW9 is more potent in suppressing growth of pancreatic cancer cells than its parental molecules We next performed cell viability assays in a panel of PDAC cell lines (MIA PaCa-2, DAN-G and HPAC) and observed that TW9 decreased cell survival in a dose-dependent manner ( Figure 3A). Furthermore, TW9 showed a more potent effect on cell survival when compared to (+)-JQ1, CI994 alone or the combination of (+)-JQ1 and CI994. Colony formation assay also showed that TW9 was more potent in suppressing cell growth ( Figures 3B and S5A,B). Immunoblot analysis clearly showed that TW9 treatment led to increased expression of cleaved caspase 3, consistent with apoptotic cell death ( Figure 3C).
In current clinical trials, single-agent BET inhibition is challenged by limited effectiveness, toxicity issues and still largely unknown optimal scheduling strategies. 27 To evaluate alternative scheduling strategies early on, we evaluated the long-term effects on cell survival of TW9 when administered only once for a time period of 24 hours.
Compared to single-agent or combination treatment with (+)-JQ1 and CI994, TW9 had more sustained effects on suppressing cell growth, and cells started to grow much later (Day 10; Figure 3D, left panel).
Colony formation assay at the end time point also showed much fewer colonies formed by TW9-treated cells ( Figure 3D, right panel).
Next, we applied a schedule for TW9 (1 day on and 6 days off for 3 cycles), which further prolonged the suppressive effects ( Figure S5C). Thus, dosing of TW9 may be reduced with ongoing tumor-suppressive activity.

| Sequential administration of gemcitabine and TW9 shows synergistic effects
Gemcitabine is a standard-of-care chemotherapeutic agent in the treatment of PDAC. Previous combination of gemcitabine with the HDAC inhibitor CI994 resulted in increased toxicity and no beneficial effect of this combination in a phase II clinical trial. 28 Thus, we evaluated the efficacy of combining TW9 and gemcitabine using different scheduling routes. Gemcitabine and TW9 were administered either at the same time or sequentially. Serial dilutions of each compound were combined, and the effect on the viability of HPAC cells was measured by cell viability assay. A synergistic effect between gemcitabine and TW9 was observed and most prominent when gemcitabine was given 24 hours before TW9 (G1T2 schedule, Figures 4A,B and S6A). Cleaved caspase 3 staining induced by G1T2 schedule supports highly increased apoptosis ( Figure 4C). After the same schedule, we also performed synergistic test for JQ1/CI994 when combined with gemcitabine ( Figure S6B). The ZIP score (8.96) of JQ1/CI994 is lower than that of TW9 (10.20), proving benefits of TW9 as a dual inhibitor. Strikingly, simultaneous or sequential administration of TW9 first followed by gemcitabine showed no combinatorial benefits or even detrimental effects in colony formation ( Figures 4B and S7A), indicating that TW9 may reduce sensitivity to gemcitabine in these schedules. Consistent with previous reports, we found that (+)-JQ1 as well as TW9 induced accumulation of the cellcycle regulator p21 ( Figure S7B), which led to G1 arrest ( Figure S7C).
We assumed that the attenuated S-phase entry by (+)-JQ1 or TW9 inhibited gemcitabine incorporation into replicating DNA. To support this hypothesis, we found that simultaneous administration of gemcitabine and (+)-JQ1 or TW9 retained more cells in G1 phase compared to gemcitabine treatment alone, which induced early Sphase arrest ( Figure S7D).
As known, gemcitabine induces acute replication stress as indicated by the induction of phospho-CHK1, and withdrawal of gemcitabine releases cells from such stress ( Figure S7E). Simultaneous treatment of TW9 followed by gemcitabine largely prevented cells from replication stress induced by gemcitabine ( Figure S7F). However, G1T2 scheduling enhanced the gemcitabine-induced replication stress ( Figure 4D). Withdrawal of gemcitabine relieved replication stress and cells returned to normal cell-cycle distribution ( Figure 4E). However, after the G1T2 schedule, cells maintained S-phase arrest after drug withdrawal ( Figure 4E). In summary, G1T2 scheduling showed the most prominent synergistic effects by augmenting gemcitabineinduced replication stress and apoptosis.

| TW9 blocks cell-cycle progression through super enhancer-associated transcription factor FOSL1
To explore the molecular mechanisms elicited by TW9, transcriptomic profiling was performed in the established PDAC cell line PANC-1.
F I G U R E 4 Administration of TW9 with chemotherapy. A, Combination response to TW9 and gemcitabine for HPAC cells treated with three administration schedules: (1) simultaneous administration of gemcitabine and TW9 for 24 hours and incubation for another 4 days; (2) Sequential administration of TW9 and gemcitabine (each for 24 hours) and incubation for another 3 days; (3) Sequential administration of gemcitabine and TW9 (each for 24 hours) and incubation for another 3 days. CellTiter Glo cell viability assay was performed to measure cell viabilities for all the indicated dose combinations. Synergy effects were evaluated using SynergyFinder (synergyfinder.fimm.fi). 39 The ZIP synergy score is averaged over all the dose combination cells. B, Colony formation assay for HPAC cells treated with different administration schedules indicated above.
Here Further, we aimed to identify TW9-regulated master transcriptional regulators (TFs) that control cell-cycle progression. Previous studies suggested that master TFs could be driven by super-enhancers (SEs). 29 We thus performed ChIP-seq using H3K27ac to identify the genomic enhancer landscape. Bioinformatic analysis of the genomewide occupancy of H3K27ac resulted in the identification of 453 SEs in PANC-1 cells ( Figure 5F). We next analyzed SE-associated gene expression of TW9-or (+)-JQ1-treated cells compared to controls.
Venn diagram analysis shows that 41 TW9-downregulated genes were associated with SEs ( Figure 5G). Compared to (+)-JQ1 treatment, 12 SE-associated genes were identified to be selectively targeted by TW9, including the transcription factor FOSL1. FOSL1 has previously been linked to KRAS-associated mitotic progression. 30 To confirm the downregulation of FOSL1 by TW9, we performed RT-PCR and found that FOSL1 was indeed downregulated by TW9 to a larger extent, compared to (+)-JQ1, CI994 and combined treatment ( Figure 5H).
Moreover, GSEA analysis revealed that FOSL1 targets 30 were significantly downregulated by TW9 ( Figure 5I), implying that TW9 dysregulated the transcriptional program of FOSL1. Consistent with the elevated levels of FOSL1 in PDAC, 30 we found SEs around the FOSL1 gene locus in PANC-1 cells but not in healthy pancreas ( Figure 5J), supporting the notion that cancer cells establish super-enhancers at oncogenes during tumor pathogenesis. 29 Statistical analysis of a pancreatic cancer cohort from the Human Protein Atlas data set 14 revealed that FOSL1 is a prognostic marker and that high FOSL1 expression levels are unfavorable for patient survival ( Figure 5K).
These results suggest that the antiproliferative phenotype observed in Intriguingly, we observed that TW9 had enhanced and more sustained HDAC-inhibiting effects when assessing histone H3 acetylation at lysines 9 and 14 by western blot analysis and in the drug washout experiment, suggesting that the CI994 moiety on TW9 has a longer residence time on its targets. While this needs further characterization, it may explain why TW9 has more prolonged growth-inhibitory effects in the long-term proliferation assay.
Chemotherapy remains the standard-of-care treatment for PDAC patients with limited long-term effectiveness. In our study, we addressed if TW9 administration could improve the efficacy of gemcitabine, a well-tolerated chemotherapeutic agent with limited single-drug activity in PDAC. Strikingly, different schedules for administration yielded highly different results. We found that the cell-cycle arrest by cotreatment with TW9 interferes with the incorporation of gemcitabine into newly synthesized DNA and suppresses gemcitabine-induced replication stress. In contrast, TW9 treatment after gemcitabine administration sustains gemcitabine-induced Sphase arrest and replication stress. This result may explain why we did not previously observe synergism between (+)-JQ1 and gemcitabine in a PDAC mouse model, in which twice daily treatment of (+)-JQ1 plus a Q3Dx4 schedule (every third day for four cycles) for gemcitabine administration was applied. 4 Previous studies implied that super-enhancers play key roles in the control of cell identity and disease. 29 In our study, we discovered As a proof of concept, our data support future efforts at developing chromatin-acting multitarget drugs harboring potential benefits for clinical application: a linked pharmacokinetic profile, reduced risk of drug-drug interactions and simplified dosing schedules. We are now designing a second generation of BET/HDAC dual inhibitors that integrate the two pharmacophores to achieve a lower molecular weight and more drug-like physicochemical properties.