Abstract
Background Multi-drug resistance is a key factor controlling patient outcomes in pancreatic ductal adenocarcinoma (PDAC). A mechanistic understanding of resistance enables development of novel treatments and precision medicine. Chromatin remodeling, mediated by altered expression and function of regulators including HDAC1, is one mechanism of resistance. We describe a multi-omic analysis of HDAC1, a histone deacetylase involved in several chromatin remodeling complexes, to understand how overexpression contributes to multi-drug resistance and patient survival in PDAC.
Methods We overexpressed HDAC1 in MIA PaCa-2 cells using CRISPRa and characterized gene expression and HDAC1 function using RNA-sequencing and ChIP-sequencing for HDAC1 and H3K27 acetylation. Analysis of HDAC1 targets revealed altered GTPase activity, which was confirmed using a biochemical assay.
Results HDAC1 overexpression in PDAC facilitates multi-drug resistance by promoting a stem-like state through global chromatin remodeling. A downstream result is increased GTPase activity which can be reversed by chemical inhibition or knockdown of HDAC1. HDAC1 target genes identified using in vitro and in vivo analysis led to the development of a clinically relevant nine-transcript prognostic signature that predicts patient prognosis.
Conclusions Integrative genomic analysis demonstrates HDAC1’s role in promoting drug resistance. These findings fuel the advancement of novel strategies for treatment and precision oncology.
Background
Pancreatic ductal adenocarcinoma (PDAC) is the most common form of pancreatic cancer and one of the most lethal cancers with a five-year survival rate of 11.5%. Due to the lack of early-stage symptoms, 52% of patients are diagnosed with unresectable, locally advanced, metastatic cancer1. Chemotherapeutics such as gemcitabine, abraxane, and combination therapies like FOLFIRINOX are standard of care treatments to improve prognosis, but chemoresistance develops in the majority of patients and contributes to the poor outcomes.
PDAC tumor cells achieve drug resistance through many cellular mechanisms. For example, over-activation of DNA damage repair processes limits the effectiveness of nucleoside analogs like gemcitabine. Tumors acquiring a mesenchymal phenotype, activated through genes involved in the epithelial to mesenchymal transition (EMT), become more invasive and have increased migratory potential2. Additionally, the hypovascular nature of PDAC tumors prevents sufficient delivery of oxygen to the tumor cells. This hypoxic environment promotes overexpression of anti-apoptotic genes (i.e. Bcl-XL and FLIP), suppressing apoptosis and contributing to drug resistance2,3,4,. Reduced blood flow also restricts delivery of drugs to tumors, further exacerbating resistance. Understanding mechanisms of drug resistance can facilitate development of therapeutic strategies to prevent or reverse resistance.
Previous work, including our own publications, linked chromatin remodeling genes to chemoresistance and patient survival in PDAC5–7. Chromatin remodeling is a mechanism of gene regulation through rearrangement of chromatin structure to alter DNA accessibility and influence transcription factor binding. This process can alter gene expression patterns and lead to cellular reprogramming that contributes to chemoresistance. Dysregulation of chromatin remodeling genes leads to global changes in gene expression making it difficult to determine whether chromatin remodeling alters known resistance pathways, novel resistance pathways, or a combination8.
We previously demonstrated that overexpression of HDAC1 contributes to multidrug resistance in pancreatic cancer cells9. While HDACs are canonically members of repressive complexes, binding of HDAC1 has also been associated with transcriptional activation. In some cases, this is explained by HDAC1’s ability to recruit RNA Pol II or regulate transcriptional elongation10,11. HDAC1 participates in several remodeling complexes including Sin3A, NuRD, and CoREST12. In cancer, HDAC1 regulates the acetylation of histone and non-histone proteins which modulate the expression of genes that influence cancer progression8. For example, the Sin3A complex, including HDAC1 is recruited to the promoter of CDH1, an epithelial cell marker, where it silences CDH1 expression during metastasis8. HDAC1 regulates expression of several genes involved in resistance pathways including apoptosis, DNA damage repair, metastasis, and EMT8,13.
The ability of HDAC1 to regulate genes is important for drug resistance, cancer progression, and tumor suppression making it a strong candidate as a drug target. Several HDAC inhibitors (HDACi) have been tested (e.g. entinostat, romidepsin) for the treatment of solid tumors and hematological malignancies14. However, clinical trials for HDACis in PDAC have been largely unsuccessful14,15. HDAC inhibition by commercially available HDACis leads to negative side effects in patients16 since HDACs have a broad impact on expression of genes involved in cancer pathways and normal cellular functions including the function of non-histone proteins; thus, inhibition of multiple HDACs using molecules that target a class of proteins causes global effects. Using genomic analyses of an in vitro system perturbing HDAC1 function, we identify genes and pathways which are regulated by HDAC1 and contribute to chemoresistance. These genes represent future targets for drug development and potential markers of treatment response in patients with elevated HDAC1 expression.
In this study, we overexpressed HDAC1 in a well described PDAC cell line, MIA PaCa-2, and measured the effects on gene expression, HDAC1 binding, and chromatin structure to better understand how HDAC1 activation contributes to resistance. We found that HDAC1 overexpression leads to altered activity of pathways (e.g. EMT, resistance to apoptosis, altered cell cycle checkpoint, and increased hypoxia) that each have the potential to contribute to resistance. We showed that HDAC1 overexpression leads to a more mesenchymal phenotype in vitro and observed that HDAC1 overexpression in patient tissues alters these pathways in a similar way. Using ChIP-seq, we determined that the genes bound and regulated by HDAC1 were enriched for GTPases and expression of these genes in patient tissues were negatively correlated with patient survival. We showed that HDAC1 overexpression results in altered expression of GTPases and thus increased GTPase activity suggesting that altered GTPase activity contributes to chemoresistance and represents possible targets to reverse resistance. Together these data reveal an improved mechanistic understanding of the role of HDAC1 in chemoresistance and nominate GTPase genes as potential therapeutic targets.
Results
HDAC1 overexpression leads to multi-drug resistance
HDAC1 expression has been previously linked to increased cellular resistance to cytotoxic chemotherapeutic drugs9. To further evaluate the effects of HDAC1 overexpression on chemotherapeutic resistance, we used CRISPRa17 to generate a stable MIA PaCa-2 cell line (MP2_HDAC1_OE) expressing HDAC1 at ~3 times the levels of the levels of the control line (MP2_NTC) which expresses a non-targeting control guide (Supplementary Data Fig. 1a). We also treated the MIA PaCa-2 cell line with a HDAC1 siRNA to evaluate the effects of reducing HDAC1 expression on chemoresistance (MP2_HDAC1_KD). MIA PaCa-2 is a well-characterized line with modest to high expression of HDAC118. HDAC1 is the most abundantly expressed HDAC gene in this line (Supplementary Fig. 1b). Under these three conditions, we measured the effect of treatment with irinotecan, gemcitabine, and oxaliplatin on cell viability (Fig. 1a-c). MP2_HDAC1_OE cells were more resistant to drug treatment than control cells and MP2_HDAC1_KD cells. Since HDAC1 overexpression led to increased resistance to these drugs, we evaluated the effect of HDAC1 protein inhibition on drug response. We treated MP2_HDAC1_OE and MP2_NTC lines with romidepsin, a HDAC1/2 inhibitor, in combination with increasing concentrations of irinotecan. We observed a sensitizing effect of romidepsin on the MP2_HDAC1_OE cells treated with irinotecan (Fig. 1d). Together these experiments show a sensitizing effect of decreasing HDAC1 activity through either chemical inhibition or decreased expression.
HDAC1 overexpression induces expression of markers of EMT in vitro and in human PDAC tissues
We performed RNA-sequencing to measure gene expression in MP2_HDAC1_OE and MP2_NTC cells. We found 1,259 genes that are differentially expressed with overexpression of HDAC1 (padj < 0.1). These differentially expressed genes (DEG) were enriched for pathways involved in drug resistance: apoptosis, EMT, G2-M checkpoint, and hypoxia (Fig. 2a, Supplementary Fig. 2). Alteration of EMT-associated DEG promotes invasion and migration associated with a more drug resistant mesenchymal cell state19. The cell surface marker CD44 is characteristic of the mesenchymal phenotype20,21. We detected a 1.8-fold increase in CD44 expression upon overexpression of HDAC1. Consistent with the expression data, immunohistochemistry showed a comparable 2-fold increase in relative density of CD44 protein in PDAC cell lines with HDAC1 overexpression (Fig. 2b,c).
These data provide evidence that HDAC1 overexpression modulates several known resistance pathways including induction of EMT in an in vitro model. Next, we sought to determine whether HDAC1 overexpression in patient tumors leads to similar changes in gene expression. We used RNA-sequencing data collected from pancreatic tumor tissues by The Cancer Genome Atlas (TCGA-PDAC dataset). We grouped tissues based on their expression of HDAC1. We compared the lowest (n=45, HDAC1HIGH) and highest (n=45, HDAC1LOW) quartiles of HDAC1 expressing samples to identify DEG in these tissues. We identified 10,592 DEG between HDAC1HIGH and HDAC1LOW tissues. We intersected this gene list with the 1,259 genes identified in our in vitro experiment and identified 322 genes that are significantly altered (padj < 0.1) in the same direction as we observed in cell lines. Heatmaps of the DEG in both datasets were clustered by sample which separated high and low HDAC1 expression (Fig. 3a,b). This analysis indicates that there is a similar pattern of expression in the PDAC tissues with high HDAC1 expression as in the MP2_HDAC1_OE cells.
Given our previous finding that HDAC1 overexpression can lead to in vitro drug resistance we wondered whether HDAC1 expression might be associated with patient response to treatment. In the TCGA dataset, we used patient prognosis information, specifically overall survival, as a proxy for treatment response. We hypothesized that given HDAC1’s role in cellular resistance, genes associated with HDAC1 overexpression might have prognostic value. We divided the 322 genes associated with increased HDAC1 expression in both patient tissues and in the MP2_HDAC1_OE line into two groups: upregulated genes (n = 216) and downregulated (n = 106) genes. For each gene set, we calculated a mean gene expression value from TCGA PDAC patient tissues and compared the top and bottom quartiles in a survival analysis. We determined that overall survival was shortened for patients with high expression of the upregulated genes and low expression of the downregulated genes (upregulated genes p = 0.00065, downregulated genes p = 0.0051, Fig. 3c,d). Although overall survival depends on multiple factors, including treatment response, this finding is consistent with our observation that HDAC1 overexpression is associated with drug resistance in vitro and supports the hypothesis that these genes might also impact drug response in patients and lead to decreased survival time. Importantly, overexpression of HDAC1 alone is not predictive of patient survival (p=0.44, Supplementary Fig. 3). Using an independent cohort (Kirby et al. 2016), we observed a similar difference in survival based on HDAC1-regulated genes (Supplementary Fig. 4a,b, upregulated genes p = 0.064, downregulated genes p = 0.19).
We wanted to determine whether combining the data we generated from cell lines with patient tumor data improved survival predictions. We compared prognostic predictions from the 322 DEG associated with HDAC1 expression in both TCGA PDAC tumors and HDAC1_OE cell lines in Figure 3 with the top 322 DEG in TCGA PDAC samples with high and low HDAC1 expression as well as the top 322 DEG in our PDAC cell lines with HDAC1 overexpression and controls. We observe a more significant p-value for overall survival when combining our in vitro data with patient data than when using DEG from cell lines or TCGA PDAC tumors individually (Supplementary Fig. 5a-d).
Gene set enrichment analysis of the 322 DEG revealed multiple cancer processes including cadherin binding, cell-cell adhesion, regulation of cell migration, and GTPase activity (Fig. 3e, Supplementary Table 1). Some of these are well described resistance pathways, but GTPase activity has not been thoroughly explored for its role in resistance. Among the enriched GTPase genes is RAP2B, which was upregulated upon HDAC1 overexpression (Supplementary Fig. 6a,b). When activated, this GTP-binding protein promotes cell migration, cell adhesion, proliferation, and metastasis in cancer22–24.
HDAC1 overexpression alters chromatin accessibility in distal enhancer and promoter regions nearby molecular switches
To identify direct and indirect impacts of HDAC1 overexpression that might contribute to resistance, we measured genome-wide DNA binding of the HDAC1 protein and the presence of the activating histone mark, H3K27 acetylation (H3K27ac) using ChIP-sequencing in the MP2_HDAC1_OE and MP2_NTC cell lines. Using the standard ENCODE ChIP-seq protocol for peak calling25, we identified 17,457 binding sites for HDAC1 (10,033 unique to MP2_HDAC1_OE, 3,789 unique to MP2_NTC). We found 30,961 regions of H3K27ac; 8,392 were unique to MP2_HDAC1_OE and 5,916 were unique to MP2_NTC (Fig. 4a). We observed an enrichment of HDAC1 binding and regions of H3K27ac near the transcription start sites (TSS) of DEG when HDAC1 is overexpressed. The H3K27ac peaks specific to HDAC1 overexpressing cells occurred significantly more near the upregulated genes despite HDAC1’s canonical role as a repressor (Supplementary Fig. 7, p<0.1).
To identify regions with altered HDAC1 binding or H3K27 acetylation directly impacting gene expression, we overlapped 1kb regions centered on all HDAC1 and H3K27ac peaks (overlapping peaks were merged, see Methods) with promoter regions of DEG (2kb upstream of annotated TSS). This revealed 1,857 regions of HDAC1 binding or H3K27 acetylation in promoters of 1,040 DEG (one promoter can have more than one overlapping peak). Gene set enrichment analysis of these 1,040 DEG revealed enrichment for GTPase activity, cadherin binding, and DNA binding (FDR < 0.05) (Supplementary Table 2).
To better understand how HDAC1 overexpression impacts H3K27ac and influences gene expression, we divided regions of HDAC1 binding and H3K27 acetylation based on whether they were increasing or decreasing across the regions described above (500bp up and downstream centered on the peak). Using the sequencing reads collected in a region called as a peak in any of our ChIP-seq experiments, we calculated a fold-change to determine whether there was evidence of increased or decreased binding. Given HDAC1’s canonical role as a repressor, we expected that increased HDAC1 binding would be associated with decreased H3K27 acetylation, however, we only identified 235 DEG with increased HDAC1 binding and reduced H3K27 acetylation in the promoter regions (+/− 2kb from TSS). In contrast, the promoters of 597 DEG had increased HDAC1 binding and increased H3K27 acetylation (fold-change > 1) upon HDAC1 overexpression (Supplementary Table 3). Since previous studies have shown that HDAC1 binding can be found near active genes10, we tested whether expression of DEG where we also measured changes in altered HDAC1 binding and H3K27 acetylation were associated with overall patient survival. We performed survival analysis comparing outcomes of patients with the top 25% and bottom 25% mean tumor gene expression of these 597 genes. Patients with the highest mean expression of upregulated genes and lowest mean expression of downregulated genes have worse overall survival (upregulated: p = 0.0021, Fig. 4b, downregulated: p = 0.023, Supplementary Fig. 8a). We also showed that HDAC1 expression is significantly higher in patients with the top 25% of mean tumor gene expression of the 597 genes (Supplementary Fig. 8b).
A nine-gene signature of HDAC1 regulated genes predicts PDAC patient survival
Identification of prognostic signatures in PDAC could be of clinical utility. The analyses above identified 597 genes regulated by HDAC1 that predict patient outcomes using genes identified from in vitro and in vivo signatures of HDAC1 overexpression, although expression of HDAC1 alone is not prognostic. We calculated a simplified signature of patient prognosis using a multivariate logistic regression with L1 penalized log partial likelihood (LASSO) for feature selection26. From the 597 genes, LASSO identified a 9-transcript model sufficient to differentiate TCGA PDAC tumors with high and low HDAC1 expression (Supplementary Fig. 8c,d). To determine the clinical relevance of the genes selected using the LASSO model, survival analysis was performed comparing the patients in the top and bottom quartile of predictor values from the regression and the group with the highest predictor values had worse overall survival (Fig. 4c, p = 0.037).
HDAC1 overexpression leads to increased GTPase activity and chemoresistance
The identification of many DEG upon HDAC1 overexpression led us to explore pathways that have not been previously linked to chemoresistance. Pathway enrichment analysis of DEG with increased HDAC1 binding and H3K27 acetylation in promoters (n = 597) identified an enrichment for many known cancer pathways (Supplementary Table 4). Included on this list was Ras signaling, regulation of apoptotic signaling pathway, chromatin binding, and GTPase activity. GTPase activity was also significant in enrichment analyses described in Figure 3 driven by overexpression of the GTPases and associated proteins (e.g. RALB, RAB27B, and RAC1) which have increased expression with HDAC1 overexpression and are associated with worse overall patient survival (Supplementary Fig. 9a). ARHGAP5, a Rho family-GTPase activating protein (Fig. 5a), was also increased with HDAC1 overexpression and we observed increased HDAC1 binding and H3K27 acetylation near its TSS (Fig. 5b). Since activation of GTPase activity is not a well described mechanism of drug resistance, we used RNA-sequencing data from 14 unmodified PDAC cells lines with varying response to gemcitabine (Data from Kirby 2016) to further support the hypothesis that GTPase activity alters cellular response to chemotherapy. We found that cell lines with increased expression of genes influencing GTPase activity had higher levels of resistance to gemcitabine (Fig. 5c). We identified a total of 11 genes that modulate GTPase activity and they all show a similar pattern of increased expression in resistant cell lines (Supplementary Fig. 9b). Increased GTPase activity activates the MAPK and PI3K pathways which promote proliferation and survival in cancer27. KEGG28 pathway mapping of these 597 DEG confirmed the increased expression of genes in the MAPK and PI3K pathways upon HDAC1 overexpression (Supplementary Fig. 10).
Given the increased transcript levels of several GTPases (Fig 5d), we tested whether there was a measurable difference in GTPase activity upon HDAC1 overexpression. GTPase activity was measured through the detection of GTP remaining after a GTP hydrolysis reaction catalyzed by cell lysates from the MP2_HDAC1_OE line compared to the MP2_NTC line. MP2_HDAC1_OE cell lysates have significantly increased GTPase activity (Fig. 5e) compared to the MP2_NTC control line. Conversely, treatment of MP2_HDAC1_OE and MP2_NTC cells with romidepsin, a HDAC1/2 inhibitor, decreased GTPase activity (p < 0.05) (Supplementary Table 5). We also observed decreased GTPase activity in MP2_NTC cells with a siRNA targeting HDAC1 (MP2_NTC_siRNA_HDAC1) compared to MP2_NTC cells with a non-targeting siRNA (MP2_NTC_siRNA_Control). These data demonstrate that HDAC1 overexpression increases GTPase activity and that inhibition of HDAC1 reverses the effect.
Methods
Cell Culture
MiaPaCa-2 cells (ATCC #CRM-CRL-1420) were cultured in D10 media: DMEM (Lonza #12-614Q) supplemented with 10% FBS (GELifeSciences #SH30071.03), and 0.5% penicillin-streptomycin (ThermoFisher #15140122). All cell lines were maintained at 37 °C and 5% CO2. Cells were cryopreserved with the addition of 10% DMSO (EMD #MX1458-6).
Plasmids
LentiCRISPRv2 (Addgene #52961) or lentiSAMv2 (Addgene #92062) and lenti-MS2-p65-HSF1-Hygro (Addgene #89308) were used to generate stable cell lines for gene knockout and activation, respectively. pMD2.G (Addgene #12259) and psPAX2 (Addgene #12260) were used to facilitate viral packaging of sgRNAs and single vector plasmids.
sgRNA Cloning
gRNA oligos were designed and cloned into their respective plasmids as described previously17.
DsiRNA
IDT TriFECTa RNAi kit was used per manufacturer’s protocol. 100,000 cells were seeded in 1 well of a 12 well tissue culture treated plate 24 hours prior to transfection. Cells were transfected using RNAiMax (ThermoFisher #13778-030) following manufacturer’s recommended protocol. As indicated in the TriFecta kit (IDT #hs.Ri.HDAC1.13), TYE 563 transfection efficiency control, positive HPRT-S1 control, and negative (DS NC1) scrambled sequence control were utilized. Further assays were performed 48 hours after transfection. Expression was validated with each transfection with the IDT PrimeTime qPCR Assay system on an Agilent QuantStudio 6 Flex Real-Time PCR system.
DsiRNA Sequences
qPCR Sequence IDs
GTPase-Glo Assay Using Cell Lysates
In vitro GTPase activity was measured using the GTPase-Glo assay (Promega #V7681). We followed the protocol as described29 with modifications for use of cell lysates. Cell lysates were made from the following cell lines: MP2_HDAC1_OE, MP2_NTC, MP2_NTC with siRNA targeting HDAC1 (MP2_NTC_siRNA_HDAC1), MP2_NTC with a non-targeting siRNA (MP2_NTC_siRNA_Control), MP2_HDAC1_OE treated with 0.01 μM romidepsin, an HDAC1 inhibitor, and MP2_NTC treated with 0.01 μM romidepsin. Cells (2 × 106 per tube) were lysed in a lysis buffer containing 50 mM HEPES at pH 7.6, 150 mM NaCl, 10% Glycerol, 0.1% NP-40, and 2 mM MgCl2. To generate the lysate, 10μL of lysis buffer per 100,000 cells was added to each cell pellet and resuspended. Lysates were mixed for 30 minutes at 4°C, vortexed in three 10 second intervals, then centrifuged at 4°C for 30 minutes at 16.1x RCF. A 2X GTP solution was prepared and the reaction was initiated following the manufacturer’s protocol.
Modifications for cell lysates required background wells for each cell line. GTPase-Glo Buffer was added to cell lysates at a final concentration of 1 μL per 10,000 cells. After the GTPase reaction, 20μL was added to each respective background well. Luminescence was measured using a BioTek Synergy H5 plate reader. To calculate GTPase activity for each cell type, we calculated the difference between the luminescence of the experimental wells and background wells. GraphPad Prism 9 (version 9.3.1) was used for plotting bar charts and t-tests performed in GraphPad were unpaired, parametric, two-tailed with 95% confidence interval.
ChIP-sequencing
MP2_HDAC1_OE and MP2_NTC cells (2 × 107) were cross-linked, harvested, and DNA was precipitated using a commercial H3K27ac antibody (Abcam, ab4729). Libraries were constructed, pooled, and sequenced using an Illumina NovaSeq instrument with 75bp single-end reads. These data were generated and analyzed using published ENCODE protocols (https://www.encodeproject.org/documents/).
Differential binding analysis was conducted using the “multiBigwigSummary’’ tool from the “deepTools’’ package30. Using this tool, a ChIP-seq score was generated for each sample and region using genomic coordinates defined as +/− 500 bp from the center of peaks defined using the published ENCODE protocol25 and 1 kb upstream of all annotated genes. Regions were merged together if they overlapped. We omitted any regions with a ChIP-seq score less than 1 for both MP2_HDAC1_OE and MP2_HDAC1_NTC. Using the ChIP-seq score, we calculated a fold-change between MP2_HDAC1_OE and MP2_NTC for HDAC1 and H3K27ac in each defined region. The criteria for differentially bound regions was a fold-change greater than or less than one for HDAC1 binding and H3K27 acetylation.
3’ RNA-sequencing
Cell pellets were frozen at −80°C until RNA extraction. For RNA extraction 350 μl of RL Buffer plus 1% β-ME from the Norgen Total RNA extraction kit was added to each cell pellet and extraction proceeded per manufacturer’s instructions including use of the DNase kit (Norgen # 37500, 25720). RNA quality was verified with the Agilent BioAnalyzer RNA Nano 600 kit (cat# 5067-1512) with the RIN range between 9.2-10. RNA-sequencing libraries were made using Lexogen QuantSeq 3’ mRNA-Seq Library Prep Kit FWD for Illumina kit (cat# 015.24) with 250 ng of RNA input. They were pooled and sequenced on an Illumina NextSeq 500 instrument with 75 bp single-end reads. Read counts averaged 4 million reads and an average Q30 of 91.28%. Lexogen’s BlueBee integrated QuantSeq data analysis pipeline was used for trimming, mapping, and alignment and the R package “DESeq2” was used for differential expression analysis.
Drug resistance screening
Cells were seeded in 96-well plates at 2000 cells/well. Seeded cells were dosed with a range of concentrations of each drug: gemcitabine (0-12.5nM), oxaliplatin (0-3.5uM), or irinotecan (0-5uM). Cells were given a second dose of drug at the same concentration as the first 48 hours later. The number of viable cells surviving drug treatment were assayed with CellTiter-Glo (Promega #G7571) 24 hours after the last drug treatment per manufacturer’s protocol using a BioTek Synergy H5 plate reader.
HDAC1 inhibition with romidepsin (Sigma #SML1175-1MG) was performed similarly to above except that cells were dosed every 24 hours with either 0.01 μM romidepsin, a range of gemcitabine, or both. Equal volume DMSO was used as a control in place of romidepsin. The number of viable cells surviving drug treatments were assayed with CellTiter-Glo (Promega #G7571) 24 hours after the last drug treatment per manufacturer’s protocol using a BioTek Synergy H5 plate reader. In both cases, data were plotted using GraphPad Prism 9, version 9.3.1. T-tests performed were unpaired, parametric, two-tailed with 95% confidence interval.
Cell staining
75,000 MIA PaCa-2 cells with non-targeting, HDAC1 OE sgRNAs, or HDAC1 KD with DsiRNA were seeded in 12 well plates. Cells were stained using Alexa Fluor 488 Conjugate kit for live cell imaging (LifeTechnologies #A25618) for CD44 via the manufacturer’s protocol. DAPI (Invitrogen #D21490) was counterstained per manufacturer’s protocols for adherent cells. Presence of CD44 in the cells was quantified using ImageJ 1.53K with measurements (area, mean, and integrated density) for stain and background taken with the freehand selection tool. Relative CD44 intensity or bound CD44 per area was calculated for each cell by: integrated density of cell-integrated density of background for that cell/area of that cell. GraphPad Prism 9 (version 9.3.1) was used for plotting violin plots and t-tests performed in GraphPad were unpaired, parametric, two-tailed with 95% confidence interval.
Enrichment Analysis
Enrichr, a comprehensive gene set analysis web server, and the R package “ClusterProfiler” (version 3.12.0)31 were used for enrichment analysis of the differentially expressed genes32. We focused on the pathways (MSigDB) and gene ontology terms (GO MF, GO BP) reaching the significance threshold of FDR < 0.05.
Survival Analysis
To conduct survival analysis, clinical and RNAseq expression data was retrieved from The Cancer Genome Atlas (TCGA) for 178 PDAC (TCGA-PAAD) patients (https://portal.gdc.cancer.gov/). Data was normalized using the R package “DESeq2”33 and differentially expressed genes with an FDR < 0.1 were used to generate Kaplan-Meier survival curves. We classified tissues based on their mean expression of a given gene set (bottom, middle, and top quartiles of gene expression). We compared the patients with the lowest and highest quartile of mean gene expression and performed survival analysis. Survival curves and analyses were generated using the “ggplot2”, “survminer”, and “survival” R packages34–36.
Clinical Data and Samples
Clinical data and RNA-sequencing data for TCGA PDAC samples were retrieved on 04/01/2020 using the GDC Data Portal. Our analyses included 178 samples in this cohort that had matched clinical and RNA-sequencing data.
Data Access
ChIP-sequencing data is available using the GEO accessions GSE209895 (H3K27ac) and GSE158541 (HDAC1). RNA-sequencing is available using the GEO accessions GSE79668 and GSE79669 (gemcitabine resistant and sensitive cell lines)37.
Statistical Testing
Statistical analysis was conducted in R (version 1.2.1335). The following R packages and software were used for analysis:
survival (version 1.2.1335)35
survminer (version 0.4.9)36
ggplot2 (version 3.3.6)34
DESeq2 (version 1.24.0)33
pheatmap (version 1.0.12)38
clusterProfiler (version 3.12.0)31
glmnet (version 4.1-3)39
ROCR (version 1.0-11)40
deepTools (version 3.5.0)30
IGV (version 2.7.2)41
LASSO model selection
A predictive gene signature from transcripts that are differentially expressed (DESeq2 FDR < 0.1) and have increased HDAC1 binding and H3K27 acetylation near their TSS (+/− 2000 bp) was developed using the LASSO regression model. LASSO was performed using the R package “glmnet” (version 4.1-3)39. The TCGA PDAC cohort was split into three groups by HDAC1 expression (top 25%, middle 50%, and bottom 25%). The cohort was further subset by randomly distributing an equal number of samples from the top 25% and bottom 25% of HDAC1 expression into two groups (n = 45). The training cohort and the validation cohort used the same dichotomization threshold (top 25% and bottom 25% of HDAC1 expression). Model performance was evaluated based on the model’s ability to classify patients into the high or low HDAC1 expression group. We generated an area under the curve (AUC) value using the R package “ROCR” (version 1.0-11)40. Kaplan-Meier curves were generated using the R package “survival” (version 3.2-13)35.
Data Analysis
The adjusted p-value was used to determine the top 322 differentially expressed genes in TCGA PDAC samples for comparison of patients in the top and bottom quartile of HDAC1 expression. Using the R package “DESeq2’’ (version 1.24.0)33, differentially expressed genes were excluded from the analysis if baseMean < 10.
Discussion
Pancreatic cancer ranks among the deadliest cancers due to its chemoresistant nature and insufficient treatment options. Understanding what drives chemoresistance is essential to identifying new therapeutic targets and improving patient outcomes. Chromatin remodeling has been established as a critical feature of tumorigenesis and cancer progression, making the pathway an attractive drug target. Our work and others showing the importance of key genes like HDAC1 in chemoresistance have nominated chromatin remodeling genes as possible drug targets. Using genomic and biochemical approaches we revealed potential mechanisms by which HDAC1 OE contributes to chemoresistance and showed that HDAC1 inhibition sensitizes PDAC cells to chemotherapeutic treatment further strengthening the argument that this pathway is a good candidate for treatment, however HDAC inhibitors have faced challenges in clinical trials. Commercial HDAC inhibitors target a class of HDACs and not specific proteins. This cross-reactivity leads to genome-wide off target effects and patient toxicity. That has motivated the current study which aims to better understand how HDAC1 activation contributes to resistance and reveal novel downstream targets that may lead to alternative treatment strategies.
In contrast to other tumor types, multiple large scale drug trials that used targeted therapy were not as successful in pancreatic cancer42,43, thus using a targeted gene panel that can be used to better define potential treatment options for PDAC patients could lead to improved survival and quality of life44,45. In this study, we collected data from an in vitro system testing the impact of HDAC1 overexpression on PDAC cells and combined these results with information from publicly available gene expression data gathered from both tissues of PDAC patients and PDAC cell lines to show that HDAC1 overexpression regulates a set of transcriptomic responses that contribute to chemoresistance and the genes regulated by HDAC1 can also be shown to predict patient outcome. HDAC1 overexpression alone is not significantly prognostic of worse overall survival in PDAC patients. However, the genes altered by HDAC1 overexpression are prognostic. Since HDAC1 is a genome-wide regulator of gene expression, in PDAC, we hypothesize that the genes contributing to a worse overall survival prognosis, resulting from tumorigenesis and drug resistance, are the subset of genes altered by HDAC1 overexpression. These results show that pathways under the control of HDAC1 contribute to patient outcomes and could be used to predict outcomes that may be linked to treatment response. Of the 597 genes altered by HDAC1 expression in vitro, we identified a 9-transcript novel signature that successfully predicted patients with high and low HDAC1 expression and was correlated with patient survival. These analyses provided insight into a subset of genes that are regulated by HDAC1 and are predictive of worse patient survival, suggesting a potential clinical utility. Our panel of biomarkers represents a potential step forward in the development of an assay that is predictive of patient survival which could influence treatment decisions.
We integrated several datasets to better understand how HDAC1 overexpression impacts PDAC cells. These studies revealed several processes known for their role in tumorigenesis, progression and drug resistance. One of these was EMT. In addition to its role in progression, EMT is associated with suppression of proteins involved in drug transport, such as CNT3, allowing the cells to evade the anti-proliferative effects of chemotherapeutics (gemcitabine)46. Cells that undergo EMT also have a more stem cell-like phenotype making them more resistant to chemotherapeutics2. Increased expression of CD44, a cell surface protein important for cell adhesion and migration, is associated with a more mesenchymal-like phenotype which is characteristic of EMT20. Here we have shown that the mesenchymal marker, CD44 transcript and protein are more abundant in cells with HDAC1 overexpression, agreeing with our past work showing that HDAC1 overexpression leads to increased migration9.
Induction of EMT is also associated with drug resistance and we showed that MP2_HDAC1_OE cells are resistant to multiple drugs. Understanding the direct regulatory impacts of HDAC1 binding and H3K27ac occupancy is necessary to determine the impacts of HDAC1 overexpression in pancreatic cancer. Our ChIP-seq experiments revealed altered H3K27 acetylation and HDAC1 binding near HDAC1 regulated genes. Interestingly, we found that the majority of the DEG with HDAC1 binding had an increase in HDAC1 and H3K27ac signals near their promoter. Our findings are in agreement with a previously published study concluding that HDAC1 binding is enriched at actively transcribed genes10. It is still not well understood how HDAC1 binding directly affects the activation of gene expression but Greer et al. did show that HDAC1 binding can regulate RNA Pol II recruitment and that HDAC1 may regulate transcription elongation through interaction with BRD4. Additionally, HDAC1 is known to deacetylate proteins other than histones which might facilitate activation of nearby transcription factors (e.g. MYC)11.
Throughout this study, our gene set enrichment analyses of genes associated with HDAC1 overexpression consistently revealed GTPase activity as an enriched process. A variety of proteins contribute to GTPase activity and many are druggable47, which makes them of potential clinical interest. We have shown that expression of GTPases and GTPase activating proteins are associated with worse overall survival. We used biochemical assays to confirm that cells overexpressing HDAC1 have higher GTPase activity than control cells. This effect was reversible in cells treated with a HDACi, which reduced GTPase activity. GTPase signaling is important for pancreatic cancer initiation, metastasis, and invasion48. Increased GTPase signaling leads to the activation of key signaling cascades, such as MAPK and PI3K, that regulate cell proliferation, migration, and survival in cancer27. We have shown that the expression of genes in the MAPK and PI3K pathways are increased upon HDAC1 overexpression highlighting one known pathway altered by HDAC1 that contributes to PDAC progression49. GTPases, such as RAC1, have been shown to activate EMT pathway in multiple cancers, thus leading to a more invasive and drug resistant phenotype50–52 and we show that RAC1 expression is increased upon HDAC1 overexpression in PDAC cells. Additionally, ARHGAP5 is a GTPase activating protein associated with promoting EMT, a known mechanism of drug resistance in colorectal cancer, but its role in PDAC is not well understood53. We showed that expression of ARHGAP5 and of several other proteins promoting GTPase activity are increased upon HDAC1 overexpression and their transcripts are more abundant in gemcitabine resistant PDAC cell lines.
Despite decades of research, PDAC patients continue to have limited treatment options and suffer poor outcomes. Identifying patients who will benefit from existing treatments or those who need an alternative treatment is a key clinical need. Our work identified sets of genes that may help identify patients with poor prognosis due at least in part to treatment resistance and reveals novel therapeutic targets that may benefit patients experiencing resistance to treatment.
Additional Information
Authors’ contributions
CAW, ERG, and SJC designed the experiments. CAW and ERG collected data. CAW and ERG analyzed the data. CAW and SJC wrote the first draft. All authors contributed to the writing of the paper and read and approved the final manuscript.
Ethics approval and consent to participate
Not applicable.
Data availability
ChIP-sequencing data is available using the GEO accessions GSE209895 (H3K27ac) and GSE158541 (HDAC1). RNA-sequencing is available using the GEO accessions GSE79668 and GSE79669.
Competing interests
None reported.
Funding information
This work was supported by Alabama’s State Cancer Fund (to CAW, ERG, SJC). SJC received support from the UAB CCTS grant (UL1TR003096).
Acknowledgements
We thank the Myers and Cooper labs for helpful feedback. We also acknowledge the data produced by The Cancer Genome Atlas which were extremely valuable and without which this study would not be possible. SJC is supported by UL1TR003096. SJC, ERG, and CAW are supported by Alabama’s State Cancer Fund.
Footnotes
Text updated; Figures updated.