Heatmaps and hierarchical clusters were generated using MultiExperiment Viewer (MeV v. 4.6.0) using the TM4 microarray software suite (Saeed et al., 2003). QRT-PCR statistical analyses were performed with SAS 9.2 (SAS Institute, Cary, NC). Unless stated otherwise, all data were analyzed by analysis of variance (ANOVA) followed by Dunnett’s post hoc test. Differences between treatment groups were considered significant when p < 0.05. Gene expression dose–response changes using 8 and 91 day data were examined using ToxResponse Modeler (Burgoon and Zacharewski, 2008). ToxResponse Modeler identifies the best fit within five different mathematical models (linear, exponential, Gaussian, sigmoidal, and

quadratic). The algorithm then identifies the best-fit from the five best in-class models to calculate half maximal effective concentration (EC50) values. Microarray datasets were first screened to identify genes differentially expressed LDK378 clinical trial (± 2-fold (P1(t) > 0.999))

in the 520 mg/L SDD group. Probes with a best fit sigmoidal model were retained for EC50 calculations. EC50 values were not calculated for probes exhibiting other dose–response profiles. Benchmark dose (BMD) modeling was also performed using BMDExpress v1.4 (Yang et al., 2007). In contrast to ToxResponse Modeler, BMDExpress uses Hill, power, linear and polynomial models to fit differential gene expression responses, and determine a benchmark response (see below). The microarray data were modeled, with modification, using a previously published procedure check details (Thomas et al., 2007). Raw signals from the microarray data were extracted and data were normalized using a semi-parametric approach (Eckel et al., 2005) and log2 transformed. Missing signal values in the array data were imputed as follows. For each probe and treatment group, an average of the signal data was computed if there were at least three values. That average signal for the probe/dose combination was imputed for all missing signals. Probes with a treatment group with two or fewer signal Rho values were not examined. Hill, power, linear and 2° polynomial models were

run assuming constant variance and the benchmark response (BMR) factor was set to 1.349 (Yang et al., 2007). For analysis of the distribution of BMD values, probes with poor model fits (i.e. p < 0.1) and/or BMD values outside the range of exposure (0.3–520 mg/L SDD) were removed. A select number of genes identified as differentially expressed in the microarray analysis, were verified by QRT-PCR. Total RNA (1 μg) was reverse transcribed by SuperScript II (Invitrogen) using an anchored oligo-dT primer as described by the manufacturer. The cDNA (1 μl) was used as a template in a 30 μl PCR reaction containing 0.1 μM of forward and reverse gene-specific primers, 3 mM MgCl2, 1 mM dNTPs, 0.025 IU AmpliTaq Gold, and 1 × SYBR Green PCR buffer (Applied Biosystems, Foster City, CA).