ATF3 coordinates anti-tumor synergy between epigenetic drugs and protein disulfide isomerase inhibitors

Abstract

Histone deacetylase inhibitors (HDACi) are largely ineffective in the treatment of solid tumors. In this study, we describe a new class of protein disulfide isomerase (PDI) inhibitors that significantly and synergistically enhance the anti-tumor activity of HDACi in glioblastoma and pancreatic cancer preclinical models. RNA-seq screening coupled with gene silencing studies identified ATF3 as the driver of this anti-tumor synergy. ATF3 was highly induced by combined PDI and HDACi treatment as a result of increased acetylation of key histone lysine residues (H3K27-ac, H3K18-ac) flanking the ATF3 promoter region. These chromatin marks were associated with increased RNA Polymerase II recruitment to the ATF3 promoter, a synergistic upregulation of ATF3, and a subsequent apoptotic response in cancer cells. The HSP40/HSP70 family genes DNAJB1 and HSPA6 were found to be critical ATF3-dependent genes that elicited the anti-tumor response after PDI and HDAC inhibition. In summary this study presents a synergistic anti-tumor combination of PDI and HDAC inhibitors and demonstrates a mechanistic and tumor suppressive role of ATF3. Combined treatment with PDI and HDAC inhibitors offers a dual therapeutic strategy in solid tumors and the opportunity to achieve previously unrealized activity of HDACi in oncology.

Introduction

Changes in the epigenetic landscape are underlying hallmarks of cancer. Silencing of tumor suppressor genes via global changes in histone acetylation and DNA methylation are examples of epigenetic alterations that are observed early in tumorigenesis.^1-4 Epigenetic regulation involves a complex interplay between multiple classes of enzymes, DNA binding elements, transcriptional cofactors, and posttranslational modifications that affect chromatin structure. Therapeutic platforms targeting epigenetic regulators have evolved in parallel with our increased understanding of this field. Emerging drug classes include DNA methyltransferase inhibitors, histone deacetylase inhibitors (HDACi), BET/bromodomain inhibitors, lysine specific demethylase 1 inhibitors, and others. Epigenetic targets have proven druggable with potent, specific, and isoform selective small molecules.⁵ Some of these agents are FDA approved while others are in clinical trials for cancer and other diseases. However, despite the emergence of promising epigenetic drug candidates, few have translated into successful clinical programs. HDACi are perhaps the most successful to date with four drugs (panobinostat/FARYDAK®, vorinostat/ZOLINZA®, romidepsin/ISTODAX®, and belinostat/BELEODAQ®) receiving approval for the treatment of cutaneous and peripheral T-cell lymphoma (CTCL and PTCL) and multiple myeloma (MM).^6-8 Aside from their efficacy in CTCL, PTCL, and MM, which account for less than 1% of all cancers,⁹ HDACi remain largely ineffective in other cancer types, particularly solid tumors.

Therapeutic resistance has been implicated in the broad lack of clinical activity that is observed with HDACi monotherapy, and a variety of molecular mechanisms that confer resistance have been reported. These include hyperactivation of the LIFR-JAK1-STAT3 anti-apoptotic cascade,¹⁰ dysregulation of HDAC proteins,¹¹ protective NF-κB signaling,¹² and disruptions in redox homeostasis.¹³ The most effective use of HDACi appears to be in combination therapy.^11,14,15 However, it is unclear which combinations will be most effective nor is there a precise molecular understanding for how to potentiate the effects of HDACi.

Protein disulfide isomerase (PDI) inhibitors are an emerging class of anti-tumor agent. The overall function of PDI family members is to coordinate oxidative protein folding in the endoplasmic reticulum. PDI isoforms share a common thioredoxin-like catalytic redox center consisting of pairs of reactive cysteines that catalyze disulfide bridge formation and isomerization between thiol groups of newly synthesized polypeptides. As PDI utilizes redox chemistry to regulate protein folding, it serves a bifunctional role as mediator of both protein and redox homeostasis. Proteotoxic and oxidative stress are recognized as phenotypic hallmarks of cancer cells and are heightened in a wide range of cancer types.¹⁶ Therefore, the unique dual role PDI plays as a mediator of both proteotoxic and oxidative stress coupled with the fact that PDI isoforms have been linked to poor prognosis and resistance to therapy in several cancer types make this family an appealing oncology drug target.^17,18

In the present study we used combination drug library screening and uncovered a highly synergistic combination between an IND-stage PDI inhibitor candidate (E64FC26)^19,20 and HDACi. Synergy was evident in hematological as well as solid tumor cells and was remarkably high in some cases, potentiating the effects of HDACi by >200-fold. We demonstrate the preclinical potential of this novel combination in multiple tumor types, including glioblastoma and pancreatic cancer, which are two of the deadliest forms of cancer, carrying dismal 5-year survival rates of less than 10%.²¹ We propose this new strategy as a means to maximize the potential of PDI inhibitors in future translational clinical studies and to rescue previously unrealized anti-tumor activity of HDACi. We delineate the molecular mechanism guiding this combination regimen, critical to which is the transcriptional induction of the ATF3 gene. ATF3 upregulation is initiated by PDI inhibitor-induced ER stress and then further potentiated by HDACi-induced chromatin remodeling. Thus, we establish ATF3 and its pro-apoptotic downstream transcriptional program as a mechanistic pathway to enhancing epigenetic therapy.

Materials and Methods

Cell Lines, Reagents, and Antibodies

Cell lines were purchased from American Type Culture Collection (ATCC) and were confirmed to be mycoplasma free at the time of experimentation using the LookOut Mycoplasma Detection Kit (Sigma). T-cells were expanded from human peripheral blood monoleukocytes (PBMCs) purchased from AllCells LLC (PB006F). All cells were maintained at 37°C and 5% CO₂. Experiments were performed in a 1:1 mixture of the recommended supplemented medium and RPMI-1640 (HyClone SH30027.01) with cells between passage 2-15. Antibodies used for western blotting, chromatin immunoprecipitation assays, and reverse transcriptase qPCR are found in Table S1. E64FC26 was synthesized as described previously.^19,20 E64FC26 was >95% pure as determined by HPLC, and target product identity was confirmed by NMR and LC-MS. The NCI Approved Oncology Drug Set was provided by the National Cancer Institute Developmental Therapeutics Program. Primary patient glioblastoma neurosphere cultures were maintained as described previously.²² Cell viability was measured in 96-well culture plates using Cell Titer-Glo Luminescent Cell Viability Assay (Promega) according to the manufacturer’s instructions. The EC₅₀ and potentiation were calculated as described before.¹⁹

Confocal Microscopy

Cells were seeded in an 8-well Nunc™ LabTek™ II Chamber Slide™ (ThermoFisher) and treated with the indicated drug for 16 hours. Cells were fixed using BD Cytofix/Cytoperm™ solution as per the manufacturer’s protocol. Slides were stained with the indicated antibody in BD Perm/Wash™ (Table S1). Primary antibodies incubated for 45 minutes followed by secondary antibodies or actin stain for 30 minutes. All samples were stained with Hoechst 33342 nuclear stain for 15 minutes. Slides were mounted with ProLong™ Diamond Antifade Mountant. Confocal microscopy was performed using the Zeiss LSM 880 NLO laser scanning confocal microscope (Zeiss). Each sample was scanned with a 63x oil objective to obtain 3-7 fields for analysis. Image analyses were performed using ZEN-blue and ZEN-black software (Zeiss), and IMARIS Microscopy Image Analysis Software (Bitplane) was used to create 3D-reconstructions of z-stacked images to calculate the number of ubiquitin aggregates per cell and the total area of the ubiquitin aggregates.

Production of shRNA Lentiviral Particles and Infection

HEK-293T cells were infected with 5 μg DNA (Table S2), 2.5 μg pCMV-dR8.91, and 0.5 μg pCMV-VSV-G diluted in Opti-MEM and Lipofectamine 2000. Seventy-two hours post transfection, supernatant containing viral particles was collected and stored at −80°C or used immediately to infect cell lines. Viral particles were titered using the qPCR Lentivirus Titration (Titer) Kit (Applied Biological Materials Inc.) as per the manufacturer’s protocol. Cells were infected with viral particles at a MOI of 10 with 8 μg/mL polybrene. Cells were seeded for experiments after stable knockdown was confirmed. Cells used for transient overexpression experiments were transfected with 2.5 μg DNA (Table S2) diluted in Opti-MEM and Lipofectamine 2000. Cells were seeded for experiments 96 hours post transfection.

Reverse transcriptase qPCR

Total RNA was isolated and purified using the RNeasy Plus Mini Kit (Qiagen). RNA was reversed transcribed and quantified using the Luna Universal One-Step RT-qPCR Kit (New England BioLabs) as per the manufacturer’s protocol. Each reaction contained 56 ng RNA and was amplified with 500 nM forward and reverse primers. Gene expression was normalized to GAPDH, which was constitutively expressed and did not change based on treatment conditions. Fold change was quantified following the Double Delta CT method (2^−ΔΔCT). Primers are listed in Table S3.

RNA-Sequencing

Cells were treated as indicated for 16 hours. RNA was extracted and purified using the RNeasy Plus Mini Kit (Qiagen). Samples were sequenced at the MUSC Hollings Cancer Center Genomics Shared Resource on the Illumina HiSeq2500. A TruSeq mRNA library preparation was used on extracted mRNA, and a rapid mode single-end 1×50 cycle protocol was used for transcript expression quantification. Biological replicates were run in triplicate. Sample data were analyzed at Creighton University using the Tuxedo tools suite.²³ In brief, samples were mapped to the hg19 human reference genome (ftp://igenome:G3nom3s4u@ussd-ftp.illumina.com/Homo_sapiens/UCSC/hg19/Homo_sapiens_UCSC_hg19.tar.gz; downloaded 7 Sept 2018) using TopHat v2.1.0 (Bowtie2 v2.2.6) followed by transcript assembly using cufflinks v2.2.1. Transcript differences were called using cuffdiff on merged transcriptome replicates by condition/treatment. Data visualization and mining was performed using the cummeRbund program for R (v3.5). GEO Accession Number GSE142210.

Chromatin Immunoprecipitation (ChIP)

Cells were treated as indicated then crosslinked with 1% formaldehyde for 10 minutes. Formaldehyde was quenched with the addition of 125 mM glycine for 5 minutes. Cells were lysed in Membrane Lysis Buffer (10 mM TRIS pH 7.4, 3 mM MgCl₂, 10 mM NaCl, 0.1% Igepal) for 20 minutes on ice, and then the lysate was incubated with 40 U/mL micrococcal nuclease for 20 minutes at 37°C to generate DNA fragments 300-500 base pair sizes. Chromatin was extracted using Nuclear Lysis Buffer (50 mM TRIS pH 8.0, 10 mM EDTA, 1% SDS) for 20 minutes on ice. Chromatin was precleared for non-specific binding with Protein A/G agarose beads for 1 hour at 4°C. Beads were removed and 10-25 μg of the chromatin was incubated overnight at 4°C with the indicated antibody (Table S1). The next day, Protein A/G agarose beads were added to each sample, incubating for 1 hour at 4°C. Immunoprecipitation samples were washed with Low Salt Buffer (20 mM TRIS pH 8.0, 2 mM EDTA, 1% Triton X-100, 150 mM NaCl), High Salt Buffer (20 mM TRIS pH 8.0, 2 mM EDTA, 1% Triton X-100, 500 mM NaCl), and Wash Buffer (10 mM TRIS pH 8.0, 2 mM EDTA). Chromatin was eluted from the antibodies with Elution Buffer (100 mM NaHCO₃ and 1% SDS) for 1 hour at 65°C. Crosslink reversal was performed in the presence of 20 mg/mL Proteinase K and 200 mM NaCl for 4 hours at 65°C. DNA was purified with the Qiagen DNA purification kit (#27104). Quantitative PCR was performed using PowerUp SYBR Green Master Mix (ThermoFisher A25741). Data was analyzed using the percent input method. The input samples represented 10% of the chromatin. All experiments were repeated 2-5 times with samples separately crosslinked for reproducibility. Primers were designed to flank gene promoters based on published reports or by mapping the genes using the open-access Ensembl database for vertebrate genomes (Table S3).^24,25 BLAST searches were performed to ensure no off-target binding of the primers would occur.

In Vivo Studies

All xenograft studies with SCID Hairless Outbred (SHO) mice were conducted under the approval of the Institutional Animal Care and Use Committee (IACUC) of the Medical University of South Carolina. Two-month-old SHO mice were injected with 1 × 10⁶ human U87 cells (100 μL total volume) in the lower flank of the mouse. Cells were mixed in a 1:1 ratio of RPMI-1640 and Matrigel. Mice were randomly assigned to treatment groups and dosed with drugs as indicated once tumors showed growth for 3 consecutive measurements and reached a tumor volume of 100-200 mm³. Vehicle treatments contained 0.5% (v/v) DMSO and 5% (v/v) Kolliphor EL (200 μL total in DPBS). Tumor measurements and analyses were performed by a blinded investigator, and unblinding occurred after all the experimental data was analyzed.

For pharmacokinetic and tissue distribution studies in CD-1 mice, animals were fasted overnight before dosing, with food returned after the 6-hour blood samples were obtained. Three animals were dosed via gavage needle for oral administration at 5 mg/kg or via tail vein injection for intravenous administration at 2 mg/kg. Blood samples (30-50 μL) were taken via the submandibular vein at 5, 15, and 30 minutes and 1, 2, 4, 6, 8, and 24 hours after dosing. For tissue distribution studies, plasma or indicated organs were harvested at 10 and 30 minutes, and 1, 2, and 4 hours after dosing. Tissues were homogenized in three volumes of PBS buffer (pH 7.4) to obtain each tissue homogenate sample. Subsequently, three volumes of acetonitrile containing internal standard was added to one volume of each tissue homogenate, and the mixture was vortexed, centrifuged (3000 xg for 10 minutes) and supernatant removed for analysis by LC-MS/MS. Calibration standards were made by preparation of a 1 mg/mL stock solution and subsequently a series of working solutions in methanol:water (1:1, v/v) which were spiked into blank tissue homogenate to yield a series of calibration standard samples in the range of 1 ng/mL to 10 μg/mL. LC-MS/MS bioanalysis was performed using a Shimadzu HPLC and AB/MDS Sciex MS/MS system fitted with Phenomenex Kinetex C18 column in MRM negative ion mode.

Results

Indene PDI inhibitors potentiate the anti-tumor effects of HDACi

In this study we screened for synergistic combinations between PDI inhibitor E64FC26 and FDA approved cancer agents by screening the NCI Approved Oncology Drug Set. We conducted combination screening in cell lines from pancreatic cancer (PANC-1), glioblastoma (T98G), ovarian cancer (SKOV3), and AML (MV4-11) using a phenotypic screening approach similar to the one we reported previously where synergy is calculated using the schema illustrated in Figure S1.²⁶ The screening returned 7 positive hits that were synergistic with 1 μM E64FC26 in all cell lines. Notably, these drugs represented only two mechanistic classes - proteasome inhibitors and HDACi. Proteasome inhibitors were expected, given our previous work,^19,20 and all three inhibitors in the set (bortezomib, carfilzomib, and ixazomib) were positive hits (Figure 1A, S2-4). HDACi demonstrated the greatest synergy, and all four HDACi in the set (vorinostat, belinostat, panobinostat, and romidepsin) were positive hits (Figure 1A, S2-3).

Figure 1. E64FC26 potentiates the activity of HDAC inhibitors.

(A) Synergistic hits in PANC-1 (left) and T98G cells (right). Cells were treated for 48 hours with 1 μM E64FC26 and the indicated drug. Combination index scores less than 1.0 are indicative of synergistic potential. (B) Representative cell viability response curves of PANC-1 cells treated with panobinostat and T98G cells treated with vorinostat alone or in combination with 1 μM E64FC26 for 48 hours. Viability was normalized to the RLU values in the absence of the HDACi to account for cell death induced by E64FC26 alone. Thus, separation of the curves indicates synergistic drug interaction. (C) Isobolograms of PANC-1 cells treated with E64FC26 in combination with panobinostat (top) or gemcitabine (bottom) and (D) isobolograms of T98G cells treated with E64FC26 in combination with panobinostat (top) or temozolomide (bottom). Numbers inside the boxes represent cell viability (%). The dotted line represents a combination index score of 1.0. Curves that fall below the dotted line are indicative of synergy. (E) EC₅₀ values of PANC-1 cells and (F) T98G cells treated with various epigenetic protein inhibitors alone or in combination with 1 μM E64FC26 (Pano: panobinostat, Vorino: vorinostat, Romi: romidepsin, Belino: belinostat, Rico: ricolinostat, Tuba: tubastatin, Entino: entinostat, 5-Aza: 5-azacytadine, Taz: tazemetostat). EC₅₀ values were calculated from 3 individual dose curves. (G) Waterfall plot showing the change in pano sensitivity in the presence of 1 μM E64FC26 and (H) 500 nM E64FC26 for the specified cell lines. The “Pano Potentiation” score was calculated by dividing the EC₅₀ for pano in the absence of E64FC26 by the EC₅₀ for pano in the presence of E64FC26. EC₅₀s for each cell line were extrapolated from independent 8-dose cell viability curves. Based on this equation, a score above 1-fold is indicative of synergy.

Validation experiments with expanded dose curves revealed remarkably high synergy between E64FC26 and HDACi (Figure 1B, S5-6). For example, in pancreatic cancer cells, E64FC26 reduced the panobinostat EC50 from 1,200 nM to 5 nM, a 240-fold increase in sensitivity, and potentiated vorinostat by 24.5-fold in T98G glioblastoma cells, reducing the EC50 from 13 μM to 0.53 μM. Synergy was further demonstrated and quantified using dose matrices and isobologram analyses (Figures 1C-D). Gemcitabine and temozolomide, which are standard-of-care agents in pancreatic cancer and glioblastoma, respectively, are included for comparison and show a general lack of anti-tumor activity as single agents and an absence of synergy with E64FC26. Greater than 10-fold synergy was observed with other pan-HDACi (belinostat) and class I HDAC selective inhibitors (romidepsin, entinostat, LP-411)²⁷ (Figure 1E-F). Synergy with HDAC6 selective inhibitors (tubastatin A and ricolinostat) was also observed, albeit to a lesser extent than the other HDACi. We detected significant enhancement of BET/bromodomain inhibitors (i.e., JQ1, mivebresib, birabresib, and IBET151). However, we did not detect any synergy between E64FC26 and the demethylating agent 5-Azacytidine, the EZH2 inhibitor tazemetostat, or the lysine specific demethylase 1 (LSD1) inhibitor GSK-LSD1 (Figures 1E-F), which shows that indene PDI inhibitors specifically enhance HDACi but not necessarily all epigenetic drug classes.

We broadened the screening to include panels of pancreatic cancer and glioblastoma cell lines as well as ovarian, colorectal, and neuroblastoma and panels of cells of hematological origin including MM, mantle cell lymphoma, cutaneous T-cell lymphoma (CTCL), and acute myeloid leukemia (AML). E64FC26 showed single agent anti-tumor activity across many tumor types (Figure S7A) and broad synergy with HDACi in both solid and hematological cancer cell lines (Figures 1G-H, S7B). Minimal sensitization was observed in normal cell types, and in some cases the interaction was antagonistic/protective (Figure 1G). The combination increased pro-apoptotic signaling, as activation of caspase-3, 8, and 9 were detected (Figure S8).

Synergy between PDI and HDAC inhibitors is UPR/ER stress dependent

We next investigated the molecular mechanism responsible for the synergy between PDI inhibitors and HDACi. We first validated the role of PDI by silencing its gene target P4HB (a.k.a. PDIA1). Similar to co-treatment with E64FC26, knockdown of PDI expression increased sensitivity to panobinostat (Figure 2A), and PDI knockdown further enhanced the synergy between E64FC26 and HDACi. In addition, we found that other PDI inhibitors reported in the literature (e.g., PACMA31) enhanced HDACi sensitivity (Figure S9),²⁸ further validating the molecular connection between PDI inhibition and HDACi potentiation.

Figure 2. PDI inhibition results in ER stress caused by the accumulation of misfolded poly-ubiquitinated proteins.

(A) PDIA1 was stably knocked down in PANC-1 cells. Cells were treated with panobinostat (pano) in the presence or absence of 1 μM E64FC26 for 48 hours. A representative cell viability curve (left) and the change in EC₅₀ values are shown (right). Statistical significance was determined by Student’s t-test (n=6). (B) Indicated cell lines were treated with 1 μM E64FC26 alone or in combination with 50 nM pano or 100 nM romidepsin (romi) for 16 hours. Western blots for ubiquitin (Ub), acetylated lysine (Ac-Lysine), and acetylated α-tubulin (Ac-Tubulin) are shown. (C) Confocal microscopy images of PANC-1 cells treated with 1 μM E64FC26 alone or in combination with 50 nM pano or 100 nM romi for 16 hours. Cells were stained with ubiquitin-FITC and Hoechst 33342 (nucleus). The total area and the total number of ubiquitinated (Ub) aggregates per cell were calculated using IMARIS Microscopy Image Analysis Software (Bitplane). Statistical significance was determined by Student’s t-test (n=12). (D) HDAC6 was overexpressed (OE) in PANC-1 cells treated with pano in the presence or absence of 1 μM E64FC26 for 48 hours. Cell viability curves are shown. (E) PANC-1 cells were treated with pano alone or in combination with 1 μM E64FC26, 5 μM Tunicamycin (Tunica), or 5 μM Thapsigargin (Thapsi) for 48 hours. EC₅₀ values were calculated for the combinations (left) with RLU values normalized to account for the individual effects of the ER stress-inducing agents alone. Statistical significance determined by Student’s t-test (n=6). PANC-1 cells were treated for 16 hours with 50 nM pano and 1 μM E64FC26, 5 μM Tunica, or 5 μM Thapsi. The western blot is shown (right; Ub: ubiquitin). (F) T98G cells were treated with 1 μM E64FC26 and/or 50 nM pano in the presence or absence of 0.5 μg/mL cycloheximide (CHX) for 16 hours. The western blot is shown. (G) PANC-1 (left) and T98G cells (right) were treated with 1 μM E64FC26 (26) and/or 37 nM pano in the presence or absence of 0.5 μg/mL CHX. Cell viability is shown here. Significance was determined by Student’s t-test (p<0.001, n=3).

In previous studies, inhibition of PDI by E64FC26 induced the accumulation of large amounts of misfolded ubiquitinated proteins and a robust ER stress and oxidative stress response in MM cells.^19,20 In solid tumor cell lines, single agent treatment with E64FC26 led to increases in protein ubiquitination that appeared as high molecular weight smears by western blotting and as dense perinuclear ubiquitin-positive aggregates by confocal imaging (Figure 2B-C, S10A). We hypothesized that HDACi might be potentiating levels of E64FC26-induced proteotoxic stress by further increasing the accumulation of misfolded poly-ubiquitinated proteins. This, however, was not the case, as the combination with HDACi did not significantly increase global ubiquitination levels or the size or number of ubiquitinated protein aggregates beyond what was induced by E64FC26 alone (Figure 2B-C, S10B). Synergy was also independent of the HDAC6-aggresome pathway, which channels misfolded proteins to the lysosomal compartment for degradation as a parallel pathway to the ubiquitin-proteasome system.²⁹ In support of HDAC6 independence, we noted strong synergy with class I selective HDACi that do not inhibit HDAC6 (Figure 1A, 1E-F), and HDAC6 overexpression failed to rescue cells from the effects of E64FC26 + HDACi combinations (Figure 2D, S11). Synergy was, however, dependent on ER stress signaling originating from oxidative stress and the UPR pathway. Other ER stress-inducing agents, including tunicamycin and thapsigargin, enhanced the activity of panobinostat (Figure 2E, S12A). Treatment with the protein translation inhibitor, cycloheximide, relieved most of the misfolded protein burden in cells and protected against death by the combination of E64FC26 and panobinostat (Figure 2F-G), suggesting that increased protein load was critical for PDI + HDACi synergy. Furthermore, consistent with previous reports,^19,20 E64FC26 induced oxidative stress characterized by Nrf2 stabilization and nuclear translocation downstream of PDI inhibition (Figure S12A-B). The addition of ROS scavengers N-acetylcysteine and β-mercaptoethanol partially rescued cell viability in combination treated T98G and PANC-1 cells (Figure S12C-D). Taken together, these data show that the potentiation of HDACi by PDI inhibition was not due to further disruption of protein folding, but rather a convergence of ER stress and epigenetic signaling events downstream of protein folding errors.

PDI and HDAC inhibitor combinations synergistically induce ATF3, HERPUD1, and CHOP

HDACi alter chromatin structure and gene expression by increasing histone acetylation.^30,31 We hypothesized that HDACi-induced changes in chromatin topography were enhancing pro-apoptotic transcriptional programs set in motion by PDI inhibition. To test this theory, we conducted comparative RNA-Seq profiling to identify transcripts and pathways that were induced specifically by the combination. We used multiple cancer cell lines (PANC-1 and T98G) to identify critical pathways across cells from different genetic backgrounds and tissues of origin. Cells were treated as indicated in biological triplicates and sequenced. Sample quality overall was high as assessed by TIN score (RSeQC v3.0.1; median TIN among samples = 54), and biological replicates clustered tightly by transcriptional changes through principle component analysis (PCA; R v3.5; Figure S13). Gene sets were screened using the decision tree outlined in Figure S14A. Relative to DMSO treated controls, the individual single agents and the combination of these two drugs induced an RNA-Seq profile with varying degrees of overlap (Figure 3A). PDI inhibition alone induced transcripts principally related to ER stress and oxidative stress pathways, including GADD45A, ATF3, DDIT3, and HMOX1 (Figure 3B). Major repressed transcripts included ATF5 and TNS3. The gene sets affected by single agent panobinostat varied widely between the two cell lines and were not associated with any one pathway; however, a common subset of transcripts (i.e., DHRS2, PHOSPHO1, TUBB4A, SLC17A7, and others) were significantly upregulated in both cell lines, and ATF5, JADE2, and TNS3 were commonly repressed. Most importantly, in the combination we observed a gene set that was common to both cell lines and discovered that three of those gene transcripts, ATF3, HERPUD1, and DDIT3 (gene encoding CHOP), were upregulated synergistically (Figure 3B-C; Table S4-5). Results were confirmed by qRT-PCR (Figure 3D, S14B-C), western blotting (Figure 3E), and fluorescence confocal microscopy (Figure S15). ATF3, as an example, was induced 16-fold by single agent E64FC26 and 3-fold by single agent panobinostat. A purely additive interaction would produce an expected fold change of 19-fold; however, we observed a highly synergistic 113-fold induction in the combination. Synergy at the mRNA level translated into even more dramatic differences at the protein level in response to PDI inhibition combined with both panobinostat and romidepsin (Figure 3E). These data implicate ATF3, HERPUD1, and CHOP as potential mediators of the pro-apoptotic response to PDI and HDAC inhibition.

Figure 3. RNA-Seq reveals ATF3, CHOP, and HERPUD1 as downstream regulators of PDI and HDAC inhibition.

(A) T98G (top) and PANC-1 cells (bottom) were treated with 1 μM E64FC65, 50 nM panobinostat (pano), the combination (combo), or DMSO vehicle control for 16 hours. RNA-Seq was performed and the number of transcripts that significantly changed in each treatment condition compared to DMSO alone are shown in the Venn Diagram (adjusted p-value<0.05). (B) Transcripts were filtered for Log2(fold change)>∣3.5∣ and FPKM>∣10∣ compared to DMSO alone. The number of genes that fit these criteria after the indicated treatment is shown in the Venn Diagrams with the top 5 significantly changed transcripts listed for each treatment. (C) The most significant transcripts affected in both PANC-1 and T98G cell lines as determined by the criteria in 3B are shown in the heatmap. (D) PANC-1 cells were treated with 1 μM E64FC26, 50 nM pano, the combination, or DMSO vehicle control for 16 hours. RNA was extracted and RT-qPCR performed. Graphs show the fold change in mRNA expression. Significance determined by Student’s t-test (n=9). (E) Indicated cell lines were treated with 1 μM E64FC26 in the presence or absence of 50 nM pano or 100 nM romidepsin (romi) for 16 hours. Western blots are shown.

ATF3 is the key driver of PDI and HDAC inhibitor synergy

We next set out to determine whether ATF3, HERPUD1, and CHOP were drivers or passengers of the anti-tumor synergy between PDI and HDAC inhibitors. To accomplish this, we used RNAi to inhibit their expression and evaluated the impact on the synergistic phenotype. Knockdown of ATF3 nearly completely reversed E64FC26 potentiation of panobinostat in PANC-1 cells and partially in T98G cells (Figure 4A-4C, S16A). Likewise, ATF3 was critical for the apoptotic response, as knockdown reduced activation of caspase-3 and −8 and PARP cleavage (Figure 4D). Conversely, knockdown of HERPUD1 and CHOP had no effect on HDACi sensitization by E64FC26 in either cell line (Figure 4E-4H, S16B-C). We also observed robust upregulation of ATF4 upon E64FC26 and panobinostat treatment, although this occurred predominantly at the protein level and did not meet our RNA-Seq hit criteria of Log2(fold change)>∣3.5∣. However, similar to what we observed with CHOP and HERPUD1, ATF4 knockdown showed no effect on PDI and HDAC inhibitor synergy (Figure S17). ATF4 induction was independent of ATF3 (Figure S18), suggesting that ATF4 and ATF3 function in parallel but independent pathways in the context of PDI and HDAC inhibition. Altogether, these findings demonstrate an essential role of ATF3 in driving the anti-tumor synergy between PDI and HDAC inhibitors and suggests a dispensable or passenger function for HERPUD1, ATF4, and CHOP.

Figure 4. ATF3 drives the synergy between E64FC26 and panobinostat.

(A) PANC-1 cells were stably transduced with control shRNA (shCTL) or ATF3 shRNA. Cells were treated with 1 μM E64FC26 and 50 nM panobinostat (pano) or DMSO for 16 hours. The western blot comparing ATF3 expression is shown here. (B) Stable PANC-1 ATF3 knockdown and shCTL cells were treated with pano in the presence or absence of 1 μM E64FC26 for 48 hours. Cell viability curves are shown (n=3). (C) Fold change in pano potentiation was calculated from PANC-1 (left) and T98G (right) cell viability curves. Significance was determined by Student’s t-test (n=9). (D) PANC-1 control and ATF3 knockdown cells were treated with 1 μM E64FC26 and 50 nM pano or DMSO for 16 hours. The western blot is shown (Casp3: caspase-3; Casp8: caspase-8). Cleaved bands are indicated by the arrow. (E) CHOP (a.k.a. DDIT3) was stably knocked down in PANC-1 cells. Cells were treated for western blot (left) and cell viability (right; n=3). (F) Change in pano potentiation in PANC-1 (left) and T98G (right). Significance was determined by Student’s t-test (n=6). (G) PANC-1 cells were stably transduced with HERPUD1 shRNA or shCTL and treated for western blot (left) or cell viability (right; n=3). (H) Change in pano potentiation in PANC-1 (left) and T98G (right). Significance was determined by Student’s t-test (n=6).

Reports of ATF3 in the literature are sparse with some studies demonstrating tumor suppressor function and others suggesting pro-tumorigenic roles.^32,33 The function of ATF3 in the context of cancer therapy, though also limited, points to a pro-apoptotic function.^34-36 Our data show a clear anti-tumor function of ATF3 within the specific therapeutic framework of PDI + HDACi therapy. To more completely understand this role, we set out to characterize the downstream effects of ATF3. We conducted RNA-Seq in parallel with ATF3 knockdown to identify ATF3-dependent target genes following PDI and HDACi treatment. The combination induced a RNA-Seq profile with several transcripts uniquely affected in ATF3 knockdown cells (Figure 5A). Many of these transcripts are implicated in ER stress, the unfolded protein response, and ER-associated degradation pathways (Figure S19A). These include multiple members of the heat shock protein 40 (HSP40) family (DNAJB1 and DNAJA4) and the HSP70 family members HSPA6, HSPA1A, and HSPA1B (Figure 5B, Table S6). RNA-Seq results for HSPA6 and DNAJB1, the two most significant ATF3-dependent genes, were confirmed by qRT-PCR and western blotting (Figure 5C-D). We then confirmed the functional role of HSPA6 and DNAJB1 in mediating cell death following PDI and HDAC inhibition using RNAi (Figure 5E-F). These data highlight a pro-apoptotic role for HSP40 and HSP70 downstream of ATF3 in the context of PDI and HDAC inhibition.

Figure 5. ATF3 induces the transcription of heat shock protein family members.

(A) PANC-1 cells were stably transduced with control shRNA (shCTL) or ATF3 shRNA (shATF3). Cells were treated with 1 μM E64FC26 alone or in combination with 50 nM panobinostat (pano) for 16 hours. RNA was extracted and RNA-Seq performed. The Venn Diagram shows the number of transcripts that changed significantly when compared to DMSO treated cells (adjusted p-value<0.05). (B) Transcripts were filtered for Log2(fold change)>∣2.5∣ and FPKM>∣10∣. The heatmap shows expression (FPKM) of the indicated transcripts. (C) Stably transduced PANC-1 cells were treated with 1 μM E64FC26 and 50 nM pano for 16 hours. The western blot is shown. (D) Stable PANC-1 ATF3 knockdown and control cells were treated with 1 μM E64FC26 alone or in combination with 50 nM pano for 16 hours. Graphs show the fold change in mRNA expression. Significance was determined by Student’s t-test (n=3). (E) PANC-1 cells were stably transduced with shCTL or DNAJB1 shRNA. Cells were treated with 1 μM E64FC26 and 50 nM pano for 16 hours and collected for western blot (left; arrow indicates cleaved band). Fold change in pano potentiation was calculated from cell viability curves after treatment for 48 hours (right). Significance was determined by Student’s t-test (n=6). (F) PANC-1 cells were stably transduced with shCTL or HSPA6 shRNA. Cells were treated with 1 μM E64FC26, 50 nM pano, the combination (26+pano), or DMSO control for 16 hours. RNA was extracted and RT-qPCR performed. Fold change in gene expression is shown (left; n=3). Fold change in pano potentiation was calculated from cell viability curves after treatment for 48 hours (right). Significance was determined by Student’s t-test (n=6).

HDAC inhibitors increase acetylation of histones flanking the ATF3 promoter and enhance RNA polymerase II recruitment to the ATF3 transcriptional start site.

Transcriptionally active gene loci are associated with histone lysine residues with specific acetylation marks, and HDACi are known to alter these acetylation patterns.^10,37,38 We hypothesized that ATF3 gene induction following PDI inhibition was being potentiated by increased histone acetylation and transcriptional activation in the presence of HDACi. To evaluate the acetylation status of histones surrounding the ATF3 promoter, we performed chromatin immunoprecipitation (ChIP) assays with antibodies targeted to acetylated histone 3 lysine 27 (H3K27-ac), histone 3 lysine 18 (H3K18-ac), and histone 3 lysine 9 (H3K9-ac), which are histone marks associated with active chromatin.^37,38 We found significant enrichment of H3K27-ac and H3K18-ac marks flanking the regulatory regions upstream of the ATF3 transcription start site (Figure 6A), which contains two well-defined promoter sites - “A” and “A1”.²⁴ Interestingly, we observed striking enrichment of H3K27-ac at Promoter “A”, particularly with the combination (p=0.001); whereas H3K18-ac surrounded both promoter regions (p=0.001). By comparison, none of the treatments, single agents or the combination, altered the H3K9-ac status (Figure S20). To understand the functional consequence of histone acetylation flanking the ATF3 promoter regions, we conducted additional ChIP assays to evaluate the recruitment of RNA polymerase II (RNA Pol II) to the ATF3 promoters. We found that single agent E64FC26 led to RNA Pol II occupancy on both ATF3 promoters (Figure 6A), which is consistent with previous reports demonstrating usage of both promoters following induction of ER stress.^24,39 Consistent with gene and protein expression analyses, panobinostat alone failed to recruit RNA Pol II to the ATF3 promoter. Most notably, the combination of E64FC26 with panobinostat led to a selective and significant increase in RNA Pol II occupancy at Promoter “A”, which closely matches the pattern of increased histone acetylation (Figure 6B). These data suggest that HDACi alter the acetylation status of key histone residues surrounding the ATF3 promoter region leading to enhanced RNA Pol II recruitment, ATF3 transcription, and the anti-tumor synergy of the combination. Taken together, our mechanistic studies led us to develop the molecular model presented in Figure 6C. PDI inhibition promotes protein folding errors, the accumulation of misfolded poly ubiquitinated proteins, and an ER stress response. ATF3 is a critical ER stress responsive transcription factor that is induced downstream of PDI inhibition and increased proteotoxic stress. In the presence of HDACi, the histones surrounding ATF3 promoters are acetylated promoting enhanced RNA Pol II recruitment and increased ATF3 transcription. ATF3 then induces a select set of gene targets, including members of the HSP70 and HSP40 family, which drive the synergistic anti-tumor effects of PDI and HDAC inhibitor combinations.

Figure 6. Hyperacetylation surrounding the ATF3 promoter is associated with greater RNA polymerase II occupancy.

(A) Schematic showing the ATF3 gene. Black boxes represent exons and white boxes represent promoters. Primers are indicated as P1, P2, and P3 (top). PANC-1 cells were treated with 1 μM E64FC26, 50 nM panobinostat (pano), the combination (26 + pano), or DMSO control for 16 hours. Chromatin was fragmented and incubated with an IgG control antibody or acetylated histone 3 lysine 27 (H3K27-ac; n=12), H3K18-ac (n=6), or RNA polymerase II (RNA Pol II; n=9) antibody overnight. Co-immunoprecipitation (ChIP) assays were performed. The percent input was calculated for each primer region by comparing the change in expression as determined by the Double Delta CT method (2^−ΔΔCT) between samples that incubated with primary antibody overnight and fragmented chromatin input samples. Thus, 10% input indicates that 10% of the total amount of fragmented chromatin showed acetylation or RNA Pol II occupancy at the primer site. Statistical significance was determined using Student’s t-test. (B) Schematic representing the histone acetylation sites and RNA Pol II occupancy along the ATF3 gene. (C) Representative schematic explaining the mechanism of synergy between PDI and HDAC inhibitors.

In vivo efficacy of PDI and HDAC inhibitor combinations in a mouse model of glioblastoma

Given the unmet need for new therapies in glioblastoma and the promising activity of PDI and HDAC inhibitor combinations in panels of established human cell lines, we decided to screen glioblastoma patient-derived neurospheres for similar activity. We observed significant synergy between panobinostat and two indene PDI inhibitors, E64FC26 and E64FC65 (Figure 7A-B). We next used the U87 glioblastoma subcutaneous xenograft model, given its proven utility in published reports,^40,41 to evaluate PDI + HDACi anti-tumor activity. We conducted a combination study with panobinostat and E64FC65 (intravenous, 10 mg/kg, 2x/week), which, like E64FC26, was highly synergistic with panobinostat in cellular assays (Figure 7A-C, S14B) and demonstrated a favorable pharmacokinetic profile in mice, including high systemic exposure and wide tissue distribution to major organ systems (Figure 7D). We observed CNS accumulation of E64FC65 suggesting blood brain barrier penetration, however the maximum drug concentration (C_max) and duration of CNS exposure were on the low end of the spectrum. Panobinostat was given 3x/week at 5 mg/kg (intraperitoneally), which is considered a low dose based on reports in the literature using the U87 xenograft model.⁴¹ Low dose E64FC65 and panobinostat monotherapies reduced tumor growth, although the effect did not reach statistical significance (vehicle vs. E64FC65 p=0.0549 and vehicle vs. panobinostat p=0.0685). The combination, on the other hand, dramatically reduced tumor growth compared to either single agent alone, reaching the minimum level of statistical significance as early as day 16 (p<0.05; Figure 7E). The anti-tumor effect of the combination became increasingly evident throughout the course of treatment. For example, on day 35, these tumors were considerably smaller by statistical measures compared to the vehicle (p=0.0002), E64FC65 alone (p=0.0069), and panobinostat alone (p=0.0003). Notably, tumors in combination treated mice showed minimal growth during the course of treatment, and tumors only began to progress after treatment was withdrawn on day 42. The anti-tumor response we observe with the combination of E64FC65 + panobinostat is comparable to the effects of standard of care agents such as temozolomide and radiation therapy in the U87 model based on recent published data.^42-44 Notably, we maintained mice on the combination treatment regimen for 42 days without signs of toxicity, demonstrating the tolerability and therapeutic index of the combination regimen.

Figure 7. E64FC26 and panobinostat reduce glioblastoma tumor growth.

(A) Primary patient glioblastoma neurospheres (GBM12 and GBM28) were treated with pano in the presence or absence of 1 μM E64FC26, 1 μM E64FC65 for 48 hours. A representative cell viability curve (left) and the change in EC₅₀ values are shown (right). Statistical significance was determined by Student’s t-test (n=6). (B) Representative images of GBM12 neurospheres treated with 1 μM E64FC65, 50 nM pano, the combination (65 + pano), or DMSO for 48 hours. (C) PANC-1 and T98G cells were treated with a dose range of panobinostat (pano) in the presence or absence of 1 μM E64FC65 for 48 hours (n=6). (D) Mice were treated with E64FC26 intravenously (IV) or orally (PO) for the indicated time. Plasma concentration (Plasma Conc), half-life (t_1/2), steady state volume of distribution (V_ss), plasma clearance (CL_p), and bioavailability (%F) were calculated (left). E64FC65 concentration at the time of IV injection was calculated for the indicated tissues (right). (E) U87 subcutaneous xenograft mice were treated with 10 mg/kg E64FC65 (65) intravenously 2 times per week (n=10), 5 mg/kg panobinostat (Pano) intraperitoneally 3 times per week (n=9), the combination (65 + Pano; n=11), or DMSO vehicle control (n=10). The days of treatment (days 0-42) are highlighted. Statistical significance was determined by student’s t-test comparing tumor growth in the combination group to vehicle, E64FC65 monotherapy, and panobinostat monotherapy (*P<0.01). Data points are shown as mean ± S.E.M.

Discussion

HDACi have been largely ineffective in solid tumors despite the widely recognized role that epigenetics play in tumor formation and progression. Pairing epigenetic drugs with select agents in combination regimens has generated anti-tumor efficacy in hematological cancers, although effective combinations in solid tumors have remained elusive. Here we present previously uncharacterized synergy between a new class of indene PDI inhibitors and HDACi. This combination was effective across a broad range of hematological and solid tumors, including some of the deadliest cancer types like glioblastoma and pancreatic cancer.

A potential challenge facing the clinical translation of this combination is that HDACi have known dose-limiting toxicities.^14,45 Pharmacodynamic (PD) studies suggest that the current dosing level of HDACi at or near the maximum tolerated dose may not be necessary, though. These studies have shown effective induction of HDACi PD biomarkers at doses far below current prescribing practices.^46,47 In representative glioblastoma and pancreatic cancer cell models we show levels of synergy between PDI and HDAC inhibitors ranging from 25- to 250-fold. Because of the potentiation observed, we used a low dose and intermittent dosing schedule of panobinostat in our preclinical glioblastoma mouse model and were able to demonstrate efficacy and tolerability of the combination with our lead PDI inhibitor. Our results reinforce the rationale for targeting epigenetics in cancer but suggest that optimal combination regimens are required. If successful, these combinations have the potential to bring forth previously unrealized anti-tumor efficacy of HDACi and may allow for reduced dosing to alleviate toxicity concerns.

Mechanistically, we demonstrate that PDI and HDAC inhibitor synergy is driven by ATF3, which is a point of convergence downstream of PDI and HDAC inhibition. ATF3 is a member of the CREB/ATF family of transcription factors that acts as an adaptive response protein, regulating genes involved in inflammation, cell cycle progression, cellular stress, and apoptosis. The outcome of ATF3 transcriptional regulation appears to be both cell-type and stress-type dependent. In oncogenesis, reports in the literature are conflicting with studies suggesting a pro-tumorigenic role while others demonstrate a tumor suppressor function.^32,33 There is more agreement on the role of ATF3 within the scope of response to therapy, as studies overwhelmingly support an anti-tumor function,^48-53 findings that are consistent with our observations. We propose the molecular model in Figure 6C whereby ATF3 is transcriptionally upregulated in response to protein folding errors and ER stress that are induced by PDI inhibition. In the presence of HDACi, we detect increased acetylation of histone lysine residues surrounding the ATF3 promoter region. These acetylation events, which mark transcriptionally active chromatin, are associated with increased RNA Pol II occupancy at the ATF3 promoters, heightened ATF3 transcription, and a pro-apoptotic signaling cascade that also involves upregulation of HSP40 and HSP70 gene family members DNAJB1 and HSPA6. To our knowledge, this is the first characterization of an ATF3-HSP40/70 apoptotic signaling axis. Additional studies are warranted to determine if this is a generalized apoptotic mechanism or one that is specific to the unique combination of PDI and HDAC inhibitors.

We identified a relatively small and highly specific set of ATF3-dependent genes belonging to the HSP40 and HSP70 families. This was somewhat unexpected given that heat shock proteins (HSPs) are thought to play a pro-survival function and maintain cellular proteostasis by stabilizing protein folding. HSP40 family members recognize and escort misfolded polypeptides to HSP70 proteins. This association stimulates the ATPase function of HSP70 to refold proteins into their native conformations.⁵⁴ It is therefore not clear how HSPs drive an anti-tumor response. However, a major distinction is that our study is focused specifically within the context of PDI and HDAC inhibition. It is possible that HSPs function differently in the face of different forms of cellular stress. In support of this hypothesis, we observe different regulation of HSPs following heat shock stress compared to combined treatment with PDI and HDACi. We found that HSP70 and DNAJB1 were induced independently of ATF3 under heat stress compared to their ATF3-dependence following PDI and HDAC inhibition (Figure S19B). A pro-apoptotic role for HSP70 where NF-kB signaling is inhibited following TNF-α stimulation has been reported^55-58 as well as DNAJB1 and HSP70 stabilization of caspase-activated DNase (CAD) downstream of T-cell activation-dependent apoptosis.⁵⁹ Future studies are required to determine if these pro-apoptotic mechanisms are recruited by HSP70 and DNAJB1 and contribute to the anti-tumor effects we report here.

Altogether, our study provides a novel combination strategy that enhances the anti-tumor efficacy of both HDAC and PDI inhibitors, classes of drugs that have either performed poorly in the clinic or are emerging agents in oncology, respectively. Pancreatic cancer and glioblastoma have some of the highest mortality rates among all cancers, and the current standards of care lack efficacy. Even with the number of new agents in development, pancreatic cancer remains in the top five most deadly and prominent cancers and glioblastoma survival rate is dismally low. Despite clinical trial failures, HDAC inhibition may be an active mechanism in solid tumors, but one that requires amplification through combinations with synergistic targeted therapies. Our findings support the use of PDI inhibitors to enhance the anti-tumor activity of HDACi in glioblastoma, pancreatic cancer, and other tumor types.

Significance.

This study uses a first-in-class PDI inhibitor entering clinical development to enhance the effects of epigenetic drugs in some of the deadliest forms of cancer.