Background

The advent of drug-eluting stent (DES) technology has dramatically altered the care and management of patients with cardiovascular disease [₁,₂,₃]. Percutaneous interventions with DES can delay (and in some cases, prevent) the need for more invasive coronary artery bypass grafting by removing obstruction and reducing the incidence of in-stent restenosis. Nevertheless, questions remain about the optimal anti-proliferative drugs for delivery, the appropriate choice of drug/stent design based on clinical presentation of the patient, and in which settings DES is superior to bypass grafting [₁]. Although drug-eluting vascular stents containing rapamycin (RAPA) or paclitaxel (PTXL) have revolutionized the treatment of coronary artery disease, a complete understanding of the mechanisms of action of these drugs is necessary to fully realize the benefits of this new technology.

Medial smooth muscle proliferation and migration in response to stent-induced injury after DES deployment are generally considered the pathophysiological underpinnings of restenosis [₄]. PTXL is a semisynthetic diterpenoid that binds microtubules and prevents their depolymerization and mitotic spindle formation, blocking the cell cycle at the G1 and G2/M junctions [₅]. RAPA is a macrocyclic lactone that engages FK506-binding protein 12, inhibiting the mTORC1 complex, effectively arresting the cell at the G1 to S phase transition [₆, ₇]. Both are inhibitors of smooth muscle cell proliferation over a wide range of doses, and now have a proven track-record in human patients when delivered via the DES platform [₂]. Despite these advances, diabetic patient populations have proven resistant to percutaneous interventions with DES, with a higher incidence of restenosis and requirement for repeat intervention [₃].

In this report we describe the use of gene transcription profiling to explore the patterns of gene expression within an in vitro model of insulin resistance, the sine qua non of type 2 diabetes mellitus. Using this model system, we have explored the range of transcriptional changes induced by PTXL and RAPA, the two drugs approved for use with DES in the United States. In spite of widely diverging mechanisms of action, both drugs are highly effective at delaying or preventing neointimal development, and loss of stent viability.

Previous work from this laboratory has shown common effects of both drugs in an in vitro model of glycemic insulin hypo-responsiveness [₈]. At higher drug doses, RAPA and PTXL both exert a strong dampening effect on the AKT axis of the insulin stimulatory pathway. Our previous results also demonstrated differential effects of these drugs at very low dose and in the presence of high glucose concentrations that lead to a paradoxical stimulation of the insulin signaling pathway by RAPA. Because subtherapeutic concentrations of RAPA could be present in tissues at the margins of the stent (or at the end of pharmacologic life of the stent) acting to stimulate the proliferation and migration of RASMC in the high glucose, insulin hyporesponsive conditions that mimic the diabetic state, we have expanded our characterizations of the plieotropic effects of these drugs on vascular smooth muscle cells. In this report we explore the global patterns of gene expression related to the varied effects of PTXL and RAPA on RASMC, and identify both common pathways and patterns of gene regulation that are unique signatures of each compound.

Gene transcription profiling of drug effects on cells, tissues, organs or whole organisms has the potential to provide tremendous amounts of data about the molecular changes induced by the pharmacotherapy. One can obtain a gestalt of the response to a drug when tens of thousands of genes are interrogated at once. This allows an unbiased view of the totality of the impact of the agent on all of the systems of a cell. While pharmacogenomics is in its infancy, it is poised to dramatically alter the way in which new therapeutics are developed, studied and approved [₉].

Results

Experimental design The goals of this study were two-fold: 1) to gain an overview of the genome-wide effects of pharmacologic agents used in modern drug-eluting stents, and 2) to develop a large database for the further exploration of the behaviors of PTXL and RAPA in an in vitro model of glucose-induced insulin resistance. To accomplish these goals, we designed the experimental protocol to allow the comparison of a large number of variables (glucose concentration, insulin stimulation, drug type and concentration) at two time points. The fully-crossed experimental factor design is outlined in Figure 1. This design was biologically replicated four times to provide a solid statistical base.

Figure 1

Figure 1: Fifty-six 100 mm dishes of RASMC were seeded and treated as shown. Each experiment was replicated four times, for a total of 224 microarrays.

Effects of PTXL on the transcriptional profiles of RASMC— The EDGE [₁₀] technique was used to identify dose-dependent patterns of gene expression over time. Loess normalized expression data for 3 doses of PTXL and the cognate vehicle-only control were analyzed in EDGE using time as the covariate (32 microarray slides). At a significance threshold of p<0.01, 129 probes were identified that PTXL had in common between the two time points, and 191 probes were identified which discriminated the two time points. These gene lists were further restricted by filtering for genes displaying a greater than 2-fold change in at least 4 samples, resulting in 103 probes characterizing genes in common (Table 1), and 162 probes characterizing the genes that were different (Table2). Hierarchical clustering of these gene lists was performed, and visualization of the heat maps is presented in Figure 2 (common) and Figure 3 (different).

Figure 2

Figure 2: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 6 and 24 hours (N=4 for each PTXL concentration at each time point). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Figure 3

Figure 3: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 6 and 24 hours (N=4 for each PTXL concentration at each time point). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Effects of RAPA on the transcriptional profiles of RASMC— The EDGE technique was next used to identify dose-specific patterns of gene expression in RASMC in HG conditions following insulin stimulation and treated with RAPA. Loess normalized expression data for 3 doses of RAPA and the cognate vehicle-only control were analyzed in EDGE using time as the covariate (32 slides). At a significance threshold of p<0.01, 466 probes were identified that RAPA had in common between the two time points, and 120 probes were identified which discriminated the two time points. These gene lists were subsequently restricted by filtering for genes displaying a greater than 2-fold change in at least 4 samples, resulting in 134 probes characterizing the genes in common between the two time points (Table 3), and 122 probes characterizing the genes differentially regulated at the two time points (Table 4). Hierarchical clustering of these gene lists was performed, and visualization of the heat maps is presented in Figure 4 (common) and Figure 5 (difference). Notable among the genes up regulated in common between the two time points were the members of the Kruppel-like family of transcription factors, with Klf2, Klf4, Klf6 (Copeb) and Klf10 (Tieg), all strongly up-regulated.

Figure 4

Figure 4: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 6 and 24 hours (N=4 for each PTXL concentration at each time point). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Figure 5

Figure 5: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 100 ng/ml). Four dishes of each treatment were studied at 6 and 24 hours (N=4 for each PTXL concentration at each time point). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Common Effects of PTXL and RAPA on the transcriptional profiles of RASMC— Analysis of the transcriptional profiles of both RAPA and PTXL as a function of the dose of drug in RASMC in HG conditions following insulin stimulation revealed patterns of genes that were modulated similarly by each drug. The EDGE technique was used to identify dose-dependent patterns of gene expression that were common to both drugs. Loess normalized expression data for all doses of PTXL, RAPA and the cognate controls were co-analyzed in EDGE using time as the covariate (28 slides). At a significance threshold of p<0.01, 471 probes were identified for 6 hours, and 451 probes were identified for 24 hours. These gene lists were further restricted by filtering for genes displaying a greater than 2-fold change in at least 4 samples, resulting in 300 probes characterizing the 6 hour time point (Table 5), and 300 probes characterizing the 24 hour time point (Table 6). Hierarchical clustering of these gene lists was performed, and visualization of the heatmaps is presented in Figure 6 (6 hrs) and Figure 7 (24 hours). A strong transcriptional dependence on time is indicated by the limited overlap between these genelists, only 3 genes are present on both lists: Gkap1, Fkbp1a, and Tmem24.

Figure 6

Figure 6: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 100 ng/ml) and PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 6 hours (N=4 for each RAPA or PTXL concentration). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Figure 7

Figure 7: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 100 ng/ml) and PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 24 hours (N=4 for each RAPA or PTXL concentration). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Differential Effects of PTXL and RAPA on the transcriptional profiles of RASMC— Similarly, an analysis of the transcriptional profiles of RAPA and PTXL as a function of the dose of drug in RASMC in HG conditions following insulin stimulation uncovered patterns of gene expression that were distinct for each drug. Again, the EDGE technique was used, this time to identify dose-specific patterns of gene expression that were opposing in PTXL and RAPA treatment. Raw data for all doses of PTXL and RAPA were co-analyzed in EDGE using time and type of drug as covariates. This analysis revealed genes with transcriptional behavior that inversely correlated between the two drugs. With a significance threshold of p<0.01, 203 probes were identified for 6 hours, and 62 probes were identified for 24 hours. As before, these lists were further restricted by filtering for genes displaying a greater than 1.2-fold change in at least 4 samples, resulting in 146 probes characterizing the 6 hour time point (Table 7), and 62 probes characterizing the 24 hour time point (Table 8). Hierarchical clustering of these gene lists was performed, and visualization of the heatmaps are presented in Figure 8 (6 hrs) and Figure 9 (24 hours). Only three genes were found on both lists: Klf2, Tuba4, and Cyr61.

Figure 8

Figure 8: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 100 ng/ml) and PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 6 hours (N=4 for each RAPA or PTXL concentration). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Figure 9

Figure 9: EDGE was performed on RNA extracted from dishes of cells grown in various concentrations of RAPA (0, 0.01, 1 and 100 ng/ml) and PTXL (0, 0.01, 1 and 1000 ng/ml). Four dishes of each treatment were studied at 24 hours (N=4 for each RAPA or PTXL concentration). Significance threshold of p < .01. Genelist was filtered for genes displaying a fold change of > 2 in at least 4 samples.

Discussion

Gene expression profiling has allowed us to define patterns of expression that may have important implications for the use of cytostatic drugs delivered via stent-based platforms. Analysis of the spectra of genes that are differentially expressed following treatment with PTXL and RAPA revealed common themes of regulation that are shared by both drugs, consistent with their phenotypic response in RASMC. This analysis also allowed us to identify patterns of transcriptional regulation that were idiosyncratic for each drug.

The gene lists identified by EDGE for each comparison group were subsequently subjected to analysis by GATHER [₁₁] and Chilibot [₁₂]. Gather provides a statistical functional analysis of genes, giving a variety of outputs. These analyses include Gene Ontology, MEDLINE Words, MeSH, KEGG Pathway, Protein Binding, Literature Net, miRNA, TRANSFAC, and ChromosomeGenes. We focused on the output from the Gene Ontology (GO) assignments, looking for functional relationships among the genes modulated by drug treatment.

PTXL modulates genes in RASMC that are, for the most part, involved in the production and maintenance of the cytoskeleton, consistent with its primary mode of action. A list of genes that were differentially expressed in response to PTXL treatment was identified by EDGE. Notably, the vast majority of genes were down-regulated. As expected, various isoforms of tubulin were the most significantly up-regulated genes, and displayed a dose-dependent expression profile (Figures 2, 3, 6, 7, 8, and 9). The increased expression of tubulin genes is likely to be a compensatory mechanism, representing the cell's attempt to replenish supplies of tubulin monomer depleted through the primary mechanism of PTXL action- the prevention of tubulin depolymerization. Previous studies have shown that drugs known to increase the pool of tubulin monomers also decrease transcription of tubulin genes, whereas drugs that deplete tubulin monomer levels increase transcription [₁₃], consistent with our results. MIZ1 (Msx-2 interacting protein 1) was one of the few genes upregulated by PTXL treatment that was not a member of the tubulin family. Miz1 is a member of the PIAS [protein inhibitor of activated STAT (signal transducer and activator of transcription)] protein family [₁₄,₁₅]. PIAS proteins are transcription factor cofactors that interfere with the JAK/STAT pathway, can act as E3 ligases for sumoylation mediated by a ring-finger domain in their C-terminal region, and have DNA binding properties of their own [₁₆,₁₇]. Analysis of the GO categories by GATHER revealed a variety of categories that were dominated by the effects of PTXL on tubulin.

Analysis of the patterns of gene expression following RAPA treatment in the in vitro model of insulin resistance suggest that the full mechanism of action of this drug is still not completely understood. Significantly, the most striking effects involved many members of the Kruppel family of transcription factors. These transcription factors play a role in a wide variety of cellular processes, not least of which are vasculogenesis, smooth muscle cell activation and inflammation [₁₈]. The fact that Klf2, 4, 6 and 10 are all strongly induced by RAPA treatment of smooth muscle cells suggests that RAPA has a potent effect on the pathways regulated by these factors. As RAPA is demonstrably an inhibitor of the proliferation and migration of vascular smooth muscle cells, the activation of members of the Kruppel-like family of transcription factors implies that a complex interaction of these factors and other cellular stimuli are required for smooth muscle cell mitogenesis.

Conclusions

Our results show striking similarities in the patterns of transcriptional activation induced by both RAPA and PTXL. This observation, while not entirely unexpected in light of the utility of these drugs in preventing restenosis, points to the inter-relatedness of a variety of cellular pathways that are key to the proliferation and migration of vascular smooth muscle cells. Of particular interest are the patterns of transcriptional regulation that are common to both drugs, and those that are different. The implications of this data are that the therapeutic value of each drug may vary based upon the biological status of the patient. Further analysis of the contrasting mechanisms of action of these two drugs, guided by our initial genomic screen, may point out specific indications for each drug, improving the therapeutic outcome.

Methods

Reagents: PTXL, RAPA, insulin, dimethyl sulfoxide and glucose were obtained from Sigma-Aldrich (St. Louis, MO, USA). Dulbecco's Minimum Essential Medium (DMEM), Trypsin/ EDTA, antibiotics/antimycotics, tissue culture grade phosphate-buffered saline (PBS) and trypan blue were obtained from GibCO (Grand Island, NY, USA). Filtered, mycoplasma-/endotoxin-free fetal bovine serum (FBS) was purchased from Gemini Bioproducts (Woodland, CA, USA). RNeasy Mini Kits and Rnase-free DNase I were obtained from Qiagen Sciences (Maryland, USA). Microarray slides, hybridization chambers, hybridization buffer, Low Input Linear RNA Amplification/Labeling kits and free-radical scavenging wash buffer were obtained from Agilent Technologies (Palo Alto, CA, USA). Cyanine-5 and cyanine-3 CTP was purchased from Perkin Elmer/NEN (Wellesley, MA, USA). 20X standard saline citrate (SSC) buffer was obtained from the Promega Corporation (Madison, WI, USA). All other salts, buffers and detergents were obtained from Sigma-Aldrich (St. Louis, MO, USA).

Cell culture: RASMC were explanted from aortas of 1 day old rat pups, purified and cultured essentially as previously described [₈,₁₉]. The cells were screened by imunoflourescence assay, and stained positive for smooth muscle α-actin and negative for von Willebrand (Factor VIII) antigen. RASMC were maintained in DMEM containing 5 mM glucose, 10% FBS and an antibiotic/antimycotic cocktail (penicillin 100 IU/ml; streptomycin, 100 µg/ml; and amphotericin-B, 2 µg/ml) at 37°C in a 5% CO₂ /100% humidity atmosphere. RASMC were used between passage level 8 and 11. Cells were plated into 100 mm polystyrene dishes (Falcon cat num 35100) and grown to 85% confluence. Dishes of RASMC were quiessed for 72 hrs prior to stimulation or drug treatment by reduction of the FBS concentration to 0.1%. Normo- or hyper-glycemic conditions (5 or 25 mM glucose, respectively) were established in the final 24 hrs of quiessesence by re-feeding with medium containing 5mM or 25mM glucose supplemented with 0.1% FBS and antibiotic/antimycotic cocktail. RASMC were stimulated with 100nM insulin (or PBS vehicle) for 30 min at 37°C prior to drug treatment. Un-stimulated and insulin stimulated cells grown under normal and high glucose conditions were treated with PTXL (0.01ng/mL, 1ng/mL, or 1000ng/mL final concentration), RAPA (0.01ng/mL, 1ng/mL or 100ng/mL final concentration), or vehicle alone (dimethylsulfoxide, DMSO) for either 6 or 24 hours at 37°C prior to harvesting RNA.

RNA extraction and microarray analysis: Total RNA was extracted from RASMC using the Qiagen RNeasy Minikit in accordance with the manufacturer's instructions including the optional DNase step. RNA integrity was verified by assay on an Agilent BioAnalyzer 2100. Five hundred nanograms of RASMC total RNA was labeled with Cyanine-5 CTP in a T-7 transcription reaction using the Agilent Low Input Linear RNA Amplification/Labeling System. Labeled cRNA from test samples was hybridized to Agilent G4130A Rat 22K microarray slides in the presence of equimolar concentrations of Cyanine-3 CTP labeled rat reference RNA prepared from pools of 1 day old rat pups [₂₀].

Statistical Methods: Microarray data (N=224 arrays) were loess normalized [₂₁,₂₂] and probes were filtered for features having a normalized intensity of < 30 aFU in either channel, then a probe was removed if <70% of the data were present across all samples. Missing data points were imputed using the k nearest-neighbors algorithm (k = 3). 11,575 probes passed these filters, and were subsequently used for analysis. Principal component analysis (PCA) was used to identify eigenvectors in the data matrices that contributed to differences between the four replicate runs that were unrelated to the biological conditions under study [₂₁,₂₂,₂₃], and were removed from the dataset using scripts written in the R Statistical Language and Environment (“R”; Version 2.2.1, build r36812, release date 2005-12-20.). The samples were standardized (µ = 0, σ =1) with a custom Perl script (ActiveState Perl 5.8.1, build 807, release date 2003-11-6).

Lists of differentially expressed genes were identified using the statistical analysis of microarray algorithm [₂₂,₂₃,₂₄,₂₅,₂₆] (“SAM”, Version 2.21, release date 2005-8-24. Typical false discovery rate of < 5%), EDGE [₁₀] (significance threshold of p<0.01), and custom R scripts written in our laboratory. Unsupervised, semi-supervised and supervised clustering analysis was performed on genelists essentially as described [₂₇] using Cluster (Version 2.11, http://rana.lbl.gov/EisenSoftware.htm). Heatmaps of cluster analyses were visualized with JavaTreeView (Version 1.0.12, release date 2005-3-14; http://sourceforge.net/projects/jtreeview/).

High-level pathway analysis and mapping to gene ontology (“GO”, http://www.geneontology.org/) categories were performed on genelists using the Expression Analysis Systematic Explorer (“EASE” Version 1.21, released date 2003-6-9; http://david.niaid.nih.gov/david/ease.htm) and GATHER [₁₁] (http://gather.genome.duke.edu) analysis environments. The Chilibot [₁₂] (http://www.chilibot.net/) contextual datamining algorithm was used for text mining of the PubMED database with selected genes and keywords.

Micorarray experiments were performed in compliance with the minimum information about a microarray experiment (MIAME) guidelines [₂₈]. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) and are accessible through GEO Series accession number GSE 5337.

Author's Contributions

Acknowledgements

Correspondence to

Peter C. Charles; Carolina Cardiovascular Biology Center, CB# 7126, School Of Medicine, University of North Carolina, Chapel Hill, North Carolina, 27599-7126. 011.919.843.1610 pcharles@med.unc.edu