During the drug developmental process, undesired toxicity accounts for about one third of compound failures [40]. However, hepatotoxicity is a common reason for withdrawal of compounds from the market [41], drug-induced toxicity affecting other organs, including kidney, heart and the central nervous system, is a common finding in the preclinical phase of drug development [39,42]. Therefore, it is evident that new technologies are needed as an alternative to classical toxicological tests for prediction of side effects specific to different organs. Toxicogenomics is an emerging technology that uses novel genomic methods to investigate the adverse effects of small molecules at the transcriptome level including DNA microarrays, new generation sequencing and QRT-PCR methods [1–3,5,20]. Among the applied technologies traditional QRT-PCR provides high-throughput and accurate differential expression profiling of usually 10–20 selected genes with high sensitivity, reproducibility, and large dynamic range. However, one of the drawbacks to apply traditional QRT-PCR in toxicogenomics is the relatively low throughput and the small number of genes that can be analyzed on multiple samples simultaneously. Because of the relatively high number of samples that are required to be analyzed and because of the better predictive value of larger gene sets (50–100 genes) for organ-specific toxicity, a high-throughput QRT-PCR approach is needed. Previously we confirmed the analytical performance of a novel, nanocapillary-based QRT-PCR, the OpenArray^™ system (Applied Biosystems, previously Biotrove Inc.) for toxicogenomic screening of 668 compounds for their gene expression profiles in HepG2 cells [2]. This high-throughput QRT-PCR has a capacity of running about 18,000 reactions per day and it is optimal for analyzing numerous samples over 56–112 gene markers.

It is clear that there are a number of limitations using in vitro approaches such as the functional differences observed in primary cells relative to the intact organs, the absence of interactions with biological borders and matrices (i.e., for ADME effects) under in vitro conditions, which are representative of an in vivo situation. Therefore, our objective was to develop an in vivo toxicogenomic screening to analyze multiple organs after systemic administration of the tested compound. Because of the relatively high number of samples needed for our test (multiple organs from numerous biological replicates) the OpenArray^™ platform was adopted to determine relative changes in expression of 56 toxicology-related genes.

Based on our previous, and on literature, data from DNA-microarray experiments, 56 gene markers were selected coding for proteins having roles in acute phase response, inflammation, oxidative stress, metabolic processes, heat-shock response, cell cycle/apoptosis regulation and detoxification. The ideology of gene selection was that transcriptional regulation of the genes should be observed in response to drug treatment that had been suggested as markers for early stages of toxic effects, therefore they could be used as predictive markers. Some of the genes are induced upon xenobiotic or toxic compounds in a specific organ, while others are general indicators of toxicity of multiple organs. Previous studies showed that gene expression profiling of samples having isolated at multiple time-points resulted in very similar alterations to control samples regardless of treatment duration [33]. Based on this observation we applied different tested compounds for 16 h and brains, hearts, kidneys and livers were isolated afterwards. Mouse was used as model organism for our toxicogenomic study because relatively high number of animals is needed. For QRT-PCR we used Taqman chemistry, instead of SybrGreen protocol. Taqman probes were designed for mouse genes; however the same set of genes could be designed for other organisms as well. Because of the open design of the OpenArray^™ plates our discovery gene set can be further optimized having novel genes inserted or replacing genes responding only in very specific cases. The number of selected marker genes can be increased and specific genes can be inserted, especially if the mechanism of action is known for a compound and other pathways are affected.

To test our toxicogenomic screening platform relative gene expression changes after systemic administration of known toxic reference compounds (doxorubicin, sulfasalazin, rotenone, aniline, dihydrocoumarin and 2,4-diaminotoluene) were determined. The number of modulated genes differed between the various treatments. The more genes there were affected by a compound in a certain organ, the more toxic effects could be verified.

In case of doxorubicin treatment most of the induced genes could occur in the heart and the kidney which is in good concordance with the known toxicity of this chemotherapeutic agent. The most important cardiotoxic mechanisms proposed for doxorubicin include oxidative stress with its resultant damage to myocardial elements, changes in calcium homeostasis and decreased ability to produce ATP [34]. In our study out of 10 genes that are involved in the oxidative stress response six were induced in the heart and five in the kidney (Figure 2). When doxorubicin was applied at two different concentrations dose-depentend gene expression alteration could be detected.

Sulfasalazine is a drug commonly used in the treatment of inflammatory bowel diseases such as ulcerative colitis and Crohn’s disease and rheumatoid arthritis. Frequent incidence of side effects limits therapy with sulfasalazine, which is due to its effects on oxidative stress [43]. Sulfasalazine induced dramatic changes in the expression of marker genes in the liver, in the kidney and in the heart suggesting severe toxicity when systemically applied at high doses.

Rotenone, a pesticide and mitochondrial complex I inhibitor, triggers general toxicity when administered to animals [35]. Interestingly, brain, liver and kidney toxicity could be registered and no or very slight cardiotoxicity based on the number of genes altered in the study (Figure 2). In case of aniline and dihydrocoumarin we found substantial gene expression modification in the liver and kidney, however they generated completely different gene expression profiles: aniline induced 16 gene markers, while dihydrocoumarin down-regulated 10 genes, out of which 6 were in common with those affected by aniline. Genes specifically altered by treatment of 2,4-diaminotoluene, which induce DNA damage, DNA repair and micronucleus formation in hepatoma cells [36], could be detected in the liver and not in any other organs.

From these results we could conclude that based on early gene expression changes the present genomic approach is able to predict organ-specific transcriptional response.

To demonstrate the utility of the strategy different drugs and drug candidates were profiled. In in-house anticancer drug discovery programs at Avidin we were able to demonstrate the high toxicity of a fatty acid derivative cytotoxic agent, ID9637 and cardio- and hepatotoxicity of a novel lipid-droplet binding thalidomide analog [37].

However, gene expression changes may represent organ adaptation to chemical exposure without acute toxicity. The advantage of toxicogenomic screen over classical methods is to identify genetic elements that could be correlated and even to predict toxic insult when there is still no pathological readout. If a compound induces several genes that are part of the organ adaptation, one could expect organ toxicity or induced activity, which could end in organ failure and organ toxicity upon chronic administration of the drug. Histological analysis was performed from those samples where significant number of altered genes could be registered (in case of ID9637 and sulfasalazin). Although in the sulfasalazin treated organs, a high number of toxicology-related genes were induced, no pathological alteration could be observed in the brain, heart, kidney, spleen, lungs and liver. In the ID9637 treated animals, no signs of toxic effects could be observed in the histological sections of the organs, except the liver, where acute toxic side effects could be registered (medium portal-periportal inflammation and gathered fibrin-pus in the fibrotic liver capsule). From these results we could conclude that early gene expression changes cannot be accurately compared with pathological alterations, mainly because of the different time scale of the methods.

In the present toxicogenomic test Trisequens N was also studied, which is a hormone replacement therapy preparation. It consists of estradiol hemihydrate alone and in combination of norethisterone acetate. These ingredients are forms of the main female sex hormones, estrogen and progesterone [39]. Estradiol hemihydrate is a naturally occurring form of estrogen and norethisterone acetate is a synthetic form of progesterone. By applying these drugs to mice we were interested in whether a single injection results in any changes in the expression of our marker genes. No significant changes between the treated and the control groups were found. Although very small changes in the model of hormone replacement therapy were registered, interestingly, in the kidney down-regulation of both fabp4 and its transcription regulator, pparg were detected. FABP4 is a lipid binding protein playing a role in intracellular lipid transport and metabolism, as well as in signal transduction and its expression is regulated by PPAR-dependent transcriptional mechanism [44]. Both gene products are associated with metabolic syndrome, type 2 diabetes mellitus, cancer and atherosclerosis [44–46]. Although in the acute experiment any significant changes could be found in the expression profiles of the treated animals, further studies on the expression alteration of fabp4, pparg or other gene markers could possibly highlight the effects of chronic hormone replacement therapy applied at different doses and could define the risk population.

Irinotecan, a widely used chemotherapeutic agent is activated by hydrolysis to SN-38, an inhibitor of topoisomerase I. The inhibition of this enzyme by the active metabolite SN-38 leads to inhibition of both DNA replication and transcription and finally apoptosis of cancer cells. Previously Blandizzi and his co-workers described that the antitumor drug irinotecan possesses adverse cardiovascular effects [47], while the same drug was demonstrated to have negative effects on renal functions [48]. The present toxicogenomic platform was applied in order to demonstrate whether the toxic activities of irinotecan and its active metabolite 7-ethyl-10-hydroxycamptothecin (SN-38) could be differentiated. As irinotecan is a prodrug, it was hypothesized that it has less influence on gene expression of toxic markers than SN-38, when applied at the same concentrations.

Gene expression screening results indicate that SN-38 exerts negative effects on both heart and kidney, as determined by altered toxic gene marker expression. Similar effects could be seen in case of irinotecan; however a smaller number of genes were affected. This observation is in good concordance with the different tolerability of the prodrug and the drug. Our results demonstrate that the presented toxicogenomic platform is not only able to detect organ-specific transcriptional response of different harmful chemicals, but also able to distinguish the toxic effects of a prodrug and its active metabolite.

One of the limitations of our gene selection procedure is that although some of the genes might be useful indicators of toxicity in a specific organ, in some cases they cannot be used in other organs. Some genes showed lower, or even undetectable expression by nanocapillary QRT-PCR (e.g., in case of egf, serpinCI, saa3, kap or serpinEI) in some tissues. This organ-specific difference was more pronounced in the brain, where the lesser number of genes showed altered expression, even in those cases where toxicity of the brain could be predicted. This problem could be overcome with precise selection of marker genes in the future and with developing an improved version of nanocapillary QRT-PCR toxicogenomic platform.

Groups of 4 Balb/C female mice, that were kept in a conventional animal house and received conventional food pellets and tap water ad libitum throughout the experiments, were injected intraperitoneally with 400 μL carrier solution (20% DMSO, 25% Solutol (BASF, Germany), 55% saline) for control, or compound dissolved in 400 μL carrier solution in the following doses: doxorubicin (Sigma-Aldrich, Budapest, Hungary): 5 mg/kg and 20 mg/kg; rotenone (Sigma-Aldrich): 30 mg/kg; aniline (Sigma-Aldrich): 150 mg/kg; sulfasalazin: 30 mg/kg; dihydrocoumarin (Sigma-Aldrich): 80 mg/kg; 2,4-diaminotoluene (Sigma-Aldrich): 80 mg/kg; irinotecan (Sigma-Aldrich): 10 mg/kg; SN-38 (Sigma-Aldrich): 10 mg/kg; Ac-915 (Avidin, Szeged, Hungary): 30 mg/kg; ID9637 (Avidin): 20 mg/kg; estradiol: 200 μg/kg and combination of 200 μg/kg estradiol and 100 μg/kg norethisterone (Novo Nordisk, Bagsværd, Denmark). After 16 h brains, hearts, kidneys and livers were isolated and stored in RNA-Later (Ambion, USA) at 4 °C overnight. All animal experiments were performed respecting institutional animal welfare guidelines.

RNA isolation from heart tissue was performed as published [49]. Briefly, our protocol is an improved version of the High Pure miRNA Isolation Kit (Roche, Cat. No. 05080576001) with inserting several additional steps into the standard protocol. Mouse hearts were frozen and homogenized at the temperature of liquid nitrogen. To 50 mg tissue powder 190 μL proteinase K solution was added (prepared as follows: 120 μL Paraffin Tissue Lysis Buffer (Roche, Germany), 20 μL 10% SDS and 50 μL Proteinase K (Roche). Samples were incubated at 55 °C for 30 min. After incubation 325 μL Binding Buffer (Roche) and 320 μL Binding Enhancer (Roche) was added and loaded onto the filter columns (Roche). Next the filters were washed in two steps with 500 and 300 μL of Washing Buffer (Roche) then the RNA was eluted by adding 40 μL Elution Buffer (Roche). The quality and quantity was assessed spectrophotometrically (Nanodrop, USA) and considered acceptable if the absorption ratio of 260/280 was >1.8.

Brain tissue was homogenized in Trizol reagent (Sigma), liver and kidney was homogenized in RA1 buffer (Machery-Nagel, USA). Total RNA was purified from drug treated and control organs with AccuPrep^™ RNA purification kit (Bioneer, Daeleon, Korea) according to the manufacturers’ protocol, except that DNase I treatment was incorporated. Homogenized tissues were centrifuged through NucleoSpin Filters (Machery-Nagel 740606, 13,000 rpm, 3 min in Eppendorf centrifuge). Pellet was suspended in RA-1 lysis buffer (Machery-Nagel, 740961.500) supplemented with β-mercaptoethanol. Equal volume of 70% ethanol was also added, samples were vortexed, and were loaded onto extraction columns (Bioneer Viral RNA Extraction Kit, KA-1111). Columns were centrifuged with 13,000 rpm, 1 min in Eppendorf centrifuge, washed with 80% ethanol, than treated with DNase for 15 min at RT. Reaction was stopped with RA1:EtOH (1:1), centrifuged, then washed twice with Wash Buffer 2 (Bioneer, KB1052). RNA was eluted with 50 μL RNase free water at 55 °C, and the concentration was determined by Nanodrop. After addition of RNase inhibitor, samples were stored at −80 °C.

For QRT-PCR total RNA (750 ng) was converted into cDNA with the High-Capacity cDNA RT Kit (Applied Biosystems, Foster City, CA, USA) and without purification the mixture was diluted with RNase-free water and applied to QRT-PCR analysis.

Amplification of the samples was followed in real time with an OpenArray NT Cycler (BioTrove Inc., Woburn, MA; Applied Biosystems, Foster City, CA, USA). For our discovery gene set individual Taqman assays were specified (Table 1). An aliquot of each Taqman assay was sent to BioTrove (Woburn, MA, USA) for loading in their OpenArray plates. Taqman assays are purchased individually and loaded by BioTrove (now at Applied Biosystems, Life Technologies) in a customer-specified layout. Recently, the list of the genes is available to prepare the custom-designed plates by the company. A third fluorescent dye (ROX), present in the Taqman assay mixture, was imaged to provide quality assessment of manufacturing and loading of the arrays.

The reverse transcribed samples (or water for no template controls) were added to a 384-well plate containing GenAmp Fast PCR Master Mix (Applied BioSystems, Foster City, CA, USA) and OpenArray DLP 5× Remix Solution (BioTrove Inc., Woburn, MA, USA) for OpenArray amplification. The OpenArray autoloader transfers the cDNA/master mix from the plate to the array through-holes by capillary action. Each subarray was loaded with 5.0 μL of master mix containing 1.2 μL of reverse transcribed cDNA. The array is manually transferred to the OpenArray slide case and sealed. The plates were cycled in the OpenArray NT cycler (up to three arrays simultaneously) under the following conditions: 50 °C for 15 s, 91 °C for 10 min, followed by 50 cycles of 54 °C for 170 s and 92 °C for 45 s.

The Biotrove OpenArray NT Cycler System software (version 1.0.2) uses a proprietary calling algorithm that estimates the quality of each individual threshold cycle (C_T) value by calculating a C_T confidence value for the amplification reaction. In our assay, C_T values with C_T confidence values below 300 (average C_T confidence of the non-target amplification reactions plus 3 standard deviations) were considered background signals. Higher C_T confidence levels were considered positive and were analyzed further. Normalization was done by using the average C_T values of three house-keeping genes (ppia, pgk1 and rplp0) and gene expression changes were calculated from the average of four replica experiments. Average values were accepted when the STD was below 0.5-fold of the average.