Background
The transcription of genes is highly regulated by epigenetic chromatin modifications, including the acetylation of lysine residues protruding from nucleosomal histones. Thus, histone acetylation status is maintained by the opposing actions of histone acetyl transferase and histone deacetylase (HDAC) enzymes [
1,
2]. HDACs modify gene expression via multiple mechanisms. The deacetylation of histones causes general chromosome condensation, and also plays a role in transcriptional regulation by forming a combinatorial 'histone code' that regulates downstream responses [
2,
3]. Additionally, a variety of non-histone targets such as transcription factors, structural and chaperone proteins are targeted by HDAC enzymes [
4]. The Zn
2+-dependent mammalian HDAC isoenzymes are divided into three classes based on their homology to yeast deacetylase proteins. Class I HDAC isoforms include HDAC1, -2 and -3 that are ubiquitously expressed as well as the low-abundance HDAC8. Class II (HDAC4, 5, 6, 7, 9, 10) and IV (HDAC11) isoforms display a more restricted tissue pattern of expression [
1]. A number of cofactors are required for HDAC activity; indeed, they reside in multi-protein complexes including co-regulators and other chromatin-modifying enzymes [
2].
Recent advances into the biology of HDAC enzymes reveal a substantial division of labor between HDAC subtypes [
2,
5]. Modulating HDAC expression demonstrates that class I HDACs are essential for proliferation and survival. Hence, HDAC1 and HDAC3 are believed to be important for proliferation [
6‐
9], whereas HDAC2 is likely involved in the regulation of apoptosis [
10,
11]. HDAC8 has been implicated in smooth muscle cell contractility [
12], though its knockdown (KD) also affects proliferation in tumor cells [
13]. Class II HDACs are mainly involved in cell differentiation and development [
14], while selective HDAC6 inhibition by tubacin also induced cytotoxicity without accompanying gene-expression changes [
15]. Aberrant expression of HDAC1, 2, 3 and 6 has been observed in various tumor types [
16‐
21], and HDAC2-mutant mice display reduced tumor development [
22]. Further, the transformed epigenome of neoplastic cells includes specific hypo-acetylation of histone H4 [
23]. Together, these findings provide the rationale for the targeted inhibition of HDAC enzymes. HDACi treatment increases global acetylation levels, which ultimately results in cell cycle arrest, apoptosis or terminal differentiation of transformed cells. A considerable variation in the gene-expression response to HDACi depending on cell line and structural class of drug has been demonstrated, and because HDACi treatment potentially affects the entire transcriptome, it is interesting that pan-HDAC inhibition changes the expression of a relatively small percentage of genes [
24,
25]. There are several structurally distinct HDACi currently in clinical trials for the treatment of solid and hematological cancers, of which the hydroxamate Zolinza (vorinostat, SAHA), recently gained approval for the treatment of cutaneous T-cell lymphoma [
26].
Despite several reports on the effects of HDAC KD in human and other species, a direct comparison of global gene-expression changes between individual class I HDAC KD and HDACi treatment has not previously been performed on human cancer cell lines. In this report, we examined viability parameters and transcriptional profiles of human HDAC1, 2 and 3 KD, and directly compared expression profiles with treatment of near-IC50 doses of two structurally distinct HDACi; the pan-inhibitory hydroxamate belinostat (PXD101) and the class I selective short-chain fatty acid valproic acid (VPA) [
26]. Further, we compared HeLa class I HDAC KD microarray data with that obtained in a recent similar study on U2OS cells.
Discussion
Targeting cancer through epigenetic control mechanisms is an area of growing interest. While HDACi show promise in clinical trials, the contribution of each HDAC isoenzyme in the anti-proliferative response of HDACi is unknown. In the present study, we directly compared gene-expression profiles between the two modes of HDAC inhibition; single class I HDAC protein depletion by siRNA and enzymatic HDACi treatment in a human cancer cell line. It is recognized, that HDACs function in multi-protein complexes and their depletion therefore might have a dissimilar outcome to HDACi treatment [
27], however this has not been directly addressed previously.
The reduced viability that we observe upon individual HDAC1, -2 and -3 knockdown has been published on class I HDAC KD in cancer cells, especially via proliferation for HDAC1 and -3, and via apoptosis for HDAC2 [
6,
8,
11,
19]. We also detected an increased subdiploid population of HDAC2 and less for -3 KD cells, whereas caspase activity was increased for HDAC1 and -2 KD cells. Thus, mediators of apoptosis following HDAC KD might be dissimilar between the isoforms examined. Caspase 3 as a mediator of apoptosis in HDAC1 KD cells was recently reported [
9], as was an increased subdiploid population for HDAC2 KD [
10,
19] and HDAC3 KD [
6,
19,
28], but not in HDAC1 KD [
19], thus supporting our results. Further, we found no major alterations in cell cycle distribution in response to class I HDAC KD, which is in agreement with other reports [
10,
11,
28]. To conclude, class I HDAC KD causes a reduction in viability and an increase in apoptosis, however at much lower levels than detected for HDACi treatment, as this is not transferred to alterations in cell cycle distributions.
Published data suggest a wide range in the proportion of genes deregulated in response to HDACi treatment; between 1–22%. This depends on factors such as class of compound, dosage, incubation time and choice of cell line [
24,
25,
29‐
31]. Hence, our data on belinostat and VPA in HeLa cells are within this broad range. Between belinostat and VPA, the shared proportion of genes of 30% probably correspond to the overlapping functions as HDAC inhibitors as both drugs affect some typical HDACi-induced genes, whereas differences are attributed to structural dissimilarities, HDAC class specificity, and non-HDACi functions of VPA. Other reports comparing the transcriptional response of different HDACi compounds find approximately 45% similarities between trichostatin A (TSA) and either tributyrate or vorinostat and 77% identical genes between tributyrate and vorinostat treatment, when examining three cancer cell lines [
25], while vorinostat and depsipeptide had very similar responses in one cell line, especially in the first hours of treatment [
31]. Further, of the limited 'core' set of 13 genes universally affected by HDACi treatment [
24], 5 were reproduced by both drugs in this study. In response to single class I HDAC down-regulation, none of these 13 genes were altered, however the expression of a considerable amount of genes were altered that included genes involved in proliferation, apoptosis or adhesion. For HDAC1, this corresponds to data on
C. elegans in which 2.2% were altered by ≥ 1.8-fold [
32], albeit lower than the 7% observed in HDAC1 knockout of untransformed murine embryonic stem cells at 2-fold or more [
33], probably due to the complete abrogation of HDAC1 in this system. HDACi treatment and individual HDAC KD have been shown to cause both up- and down-regulation of multiple gene targets [
5,
32,
33]. The knockdown of class I HDAC enzymes in this report showed that near equal proportions of genes were induced as were repressed by HDAC KD, with a slight overweight of induced genes for HDAC1 and -2 KD and a slight overweight of down-regulated genes for HDAC3, possibly separating this isoform as mainly a transcriptional activator. As HDAC1 and -2 reside in the same co-repressor complexes, the disruption of these might have more similar outcomes. Moreover, we found that HDAC1 KD altered the greatest number of genes, and hence might affect gene transcription to a larger extent than HDAC2 and -3. Between the three KD conditions, we found most genes (73–80%) to be uniquely deregulated upon individual HDAC KD, with HDAC1 having the least degree of overlap. This suggests distinctive transcriptional targets for HDAC enzymes from the same class, and could thus provide the basis for discrete functions between class I HDACs [
11]. In comparison with genes affected by HDAC1, -2 or -3 KD by siRNA in human U2OS cells in a recent study [
9], the majority were not reproduced herein, and generally point to cell-line specific responses to HDAC depletion. This emphasizes the importance of comparing HDAC KD with HDACi treatment in the same cell line.
Finally, we compared individual KD of class I HDAC members with two dissimilar HDACi compounds at near-IC50 doses. At the treatment regimens chosen, three times more genes were deregulated by HDACi treatment than by individual class I HDAC KD. As these drugs target multiple HDACs, this is not unexpected. The overlap of genes between HDACi treatments and between individual HDAC KD was in a similar range; 20–30%. When looking into the genes whose expression overlapped between HDACi treatment and individual KD of the target HDACs of these compounds, a surprisingly low degree of similarity was observed, namely less than 4% of regulated genes. The reason for the low degree of overlap could have several explanations. First, some degree of redundancy might occur after individual HDAC KD. A prior study in Drosophila showed an overlapping proportion of 20% (469 of 2347 genes regulated in total) between DHDAC1 KD and TSA treatment, each for 5 days post-treatment. However, reducing TSA treatment to 6 hours also reduced the overlap to 4.5% (52/1151), thus differences in experimental set-up probably account for a large variation in these numbers. For DHDAC3 KD, the overlap with TSA treatment was 2%, and the authors conclude that especially DHDAC1 affected gene expression in a similar manner to TSA [
34]. The closer resemblance between DHDAC1 and TSA profiles might be because Drosophila has fewer HDAC enzymes and DHDAC1 is orthologous to both human HDAC1 and -2. Second, depleting HDAC levels most likely interferes with the multi-protein complexes in which they reside in a different manner than by enzymatic drug inhibition of HDAC, causing differential cellular responses. It has previously been shown in Drosophila, that DHDAC1 deficiency and point mutations had dissimilar phenotypic outcomes, the latter presumably by altering HDAC complexes rather than disrupting them [
35]. Third, we showed that the transcriptional profile obtained by individual HDAC KD is not simply elaborated by inhibiting multiple HDAC enzymes but altered altogether, and thus other mechanisms might contribute to the HDACi effects other than targeting individual class I HDAC enzymes. These differences might explain why single class I HDAC KD is not as toxic as pan-inhibitory HDACi treatment and fails to produce identical phenotypic effects, despite the probable effects of HDACi mainly via class I HDAC enzymes.
Methods
Cell culture and drugs
Human cervix cancer cells HeLa, CCL-2 (American Type Culture Collection, Manassas, VA) and mammary cancer cells MCF-7 (HTB-22) were propagated in DMEM + glutamax media supplemented with penicillin and streptomycin and 10% FBS; the colon cancer cell line HCT116 (CCL-247) was maintained in RPMI-1640 media supplemented with glutamine, penicillin, streptomycin and 10% FBS (Invitrogen, Carlsbad, CA). All were grown in a humidified atmosphere of 5% CO2 at 37°C and passaged twice a week. Belinostat was synthesized as described in recent patent applications (International publication number US 6,888,027), and valproic acid was purchased from Sigma-Aldrich (St. Louis, MO). Drugs were dissolved in sterile water, aliquoted and stored at -20°C until use.
Transfection of siRNA
Pre-designed targeting siRNA SmartPOOL was purchased from Dharmacon (Lafayette, CO) (non-targeting siRNA D-001206-13, HDAC1 M-003493-02, HDAC2 M-003495-01, HDAC3 M-003496-00). Cells were plated in 6-well plates, 250,000/well in complete media and incubated overnight prior to aspiration of media and replacement with OPTI-MEM (Invitrogen) with a final concentration of 50 nM siRNA complexed with oligofectamine (Invitrogen). Cells were incubated 4–6 hours before addition of 1 ml growth medium with 20% FCS.
Cells were plated in 6-well plates at 250,000/well and left overnight before the transfection procedure, or replaced with fresh media with drug at 0.5 μM for belinostat or 3.0 mM for VPA. 48 hours post-transfection and 24 hours after drug treatment, total RNA was extracted with Trizol according to the manufacturers protocol (Invitrogen). For microarray samples, 5 of the 6 wells pr condition were pooled to minimize well-to-well variation. The 6th well was lysed directly in SDS sample buffer (Invitrogen), and used for protein analysis.
DNA microarray analysis
RNA integrity was quality checked on the Agilent 2100 Bioanalyser (Agilent Technologies, Santa Clara, CA), then processed and hybridized onto Affymetrix arrays according to the manufacturer's protocol. Briefly, 5 μg RNA pr sample was used to to generate biotin-labeled antisense cRNA. After fragmentation, the labeled cRNA samples were hybridized to Affymetrix HG-U133 Plus 2.0 arrays (Affymetrix, Santa Clara, CA), washed and stained with phycoerytrin conjugated streptavidin, and finally scanned in the Affymetrix GeneArray
® scanner to generate fluorescent images, as described in the Affymetrix GeneChip
® protocol. All statistical analyses, including pre-processing of data was carried out in R, version 2.3 (R development core team. R: A language and Environment for Statistical Computing;
http://www.R-project.org). DNA chips were checked for quality assurance parameters such as visual image inspection, replicate scatter plots and RNA degradation plots, before normalization for mean overall expression using the gcrma package (GC robust multiarray algorithm) [
36]. Agglomerative hierachical clustering showed that the biological replicates clustered together as expected (data not shown). The statistical linear model based method of Limma (linear models for microarray data package) [
37] was found to be most sensitive at identifying genes with differential expression between control and each condition. Raw p-values were adjusted for multiple testing using the Benjamini & Hochberg method [
38] to reduce the number of false positives, and a 5% significance threshold applied. Comparisons of gene lists between conditions was performed using VennMapper [
39].
Quantitative reverse transcription polymerase chain reaction
RNA was reverse transcribed by RT-PCR using 1 volume diluted RNA (100 ng/μl) and 1 volume 2× RT-master mix (High Capacity cDNA archive kit, Applied Biosystems, Foster City, CA) exposed to 25°C 10 minutes and 37°C 2 hours. Samples were analyzed for gene-expression levels by qRT-PCR, performed on ABI PRISM™ 7500 Sequence Detection System (Applied Biosystems). Experiments were done in triplicate by mixing 1 μL probe, 10 μL 2× Taqman master mix and 9 μL cDNA diluted 1:50, and subjecting samples to 40 cycles of amplification (15 seconds denaturation at 95°C, 1 minute annealing and elongation at 60°C), using GAPDH (HDAC KD) or β-Actin (drug treatments) as endogenous controls. All pre-validated FAM-labeled probes were purchased from Applied Biosystems. Subsequent data analysis was performed using DART-PCR version 1.0 [
40].
Western Blotting
Cells were lysed directly in SDS sample buffer (Invitrogen) and electrophoretically separated and transferred to nitrocellulose paper, following the manufacturer's instructions (Invitrogen pre-cast Novex gels). Blots were blocked in 10% non-fat skimmed milk, incubated overnight with primary antibody, washed in Tris-buffered saline with Tween-20 and visualized by HRP-conjugated secondary antibodies (1:2,000) and ECL plus reagent (GE Healthcare, Chalfont St. Giles, Buckinghamshire, UK). Antibodies uses were: rabbit anti-HDAC1 and -3 (1:1,000, Cell Signaling Technology, Danvers, MA), mouse anti-HDAC2 (1:10,000, Abcam, Cambridge, UK), rabbit anti-Actin (1:5,000 Sigma-Aldrich).
CellTiter-Glo Assay
Scrambled control and HDAC KD cells were plated in triplicate at 10,000/well in 96-well format 24 hours post-transfection. Cells were incubated 48 hours, without drug for viability measurements, and within expected belinostat toxicity limits for determination of IC50 values. Cells were lysed directly with CellTiter-Glo luminescent viability assay (Promega, Madison, WI), and luminescence proportional to ATP present hence metabolically active cells was measured (HTS 7000 plus Bioassay reader, PerkinElmer Life and Analytical Sciences, Waltham, MA). Data were normalized to the scrambled control, and IC50 values determined in Prism 4 by generation of a sigmoidal dose-response curve with variable slope (GraphPad Software, San Diego, CA). Significant changes in mean viability or IC50 values in the four groups (control, HDAC1, -2 or -3 KD) were calculated by ANOVA one-way analysis of variance repeated measures test and Dunnett's multiple comparisons test in Prism (GraphPad Software).
Caspase-Glo 3/7 assay
Cells were plated at 104/well, in quadroplicates for each HDAC KD condition and control 48 hours post-transfection, or in triplicates for drug treatments. Plates were incubated for 24 hours prior to direct lysis by Caspase-Glo 3/7 reagent (Promega), and luminescence reading according to caspase 3/7 activity.
Cell cycle analysis
Transfected cells were incubated for 48 hours before analysis. For drug treatments, cells were plated in 6-well format 2.5 × 105/well, incubated overnight, treated with drug for 24 hours and processed as follows. Cells were permeabilized by incubation in ice-cold 70% ethanol, rehydrated in PBS supplemented with Tween-20 and FBS, RNAse treated and DNA stained with propidium iodide. Cells were analyzed using a FACSCalibur instrument and the CellQuest software (BD Biosciences, Mountain View, CA).
Declaration of competing interests
Marielle Dejligbjerg was partly sponsored by a grant from the Danish Ministry of Science, Technology and Innovation, and the biotechnology company TopoTarget A/S. Morten Grauslund is a shareholder and full-time employee at TopoTarget A/S. Thomas Litman is an employee of Exiqon A/S. Laura Collins is a previous employee of TopoTarget A/S. Xiaozhong Qian and Michael Jeffers are previous employees of CuraGen Corp. Henri Lichenstein is a shareholder and full-time employee at CuraGen Corp. Peter B Jensen is CEO of and shareholder in TopoTarget A/S. Maxwell Sehested is CSO of and shareholder in TopoTarget A/S.
Authors' contributions
MD designed and performed siRNA transfections, qRT-PCR, western blotting, viability assays and flow cytometry analyses, helped with secondary microarray analysis and drafted the manuscript. MG participated in the design of the study and drafting of the manuscript. TL and LC performed microarray data analysis. XQ assisted in siRNA transfections, western blotting and flow cytometry analyses. MJ, HL, PBJ and MS conceived of the study, participated in its design and coordination, and provided intellectual discussions and ideas regarding the content of manuscript. All participating authors read and approved the final manuscript.