The Sterolgene v0 cDNA microarray: a systemic approach to studies of cholesterol homeostasis and drug metabolism

Production of the Sterolgene v0 microarray

The Sterolgene array contains 95 mouse genes known to be involved in cholesterol homeostasis and drug metabolism. Among these are members of the cytochrome P450 superfamily, the nuclear receptor superfamily, the ABC transporters and other transporter families, genes for cholesterol biosynthesis and SREBP signaling and some other genes linked to cholesterol homeostasis (Figure 1). The complete gene list with corresponding accession numbers is available in Additional file 1. The positions of probes have been selected according to a priori knowledge of each gene. Probes detect all gene transcripts known at the time of design. In case of major differences between transcripts more probes per gene were selected. Each gene probe is spotted in 6 replicas, three in two blocks. Each block contains 23 types of positive controls (most of them spotted in 6 replicas) and 3 types of negative controls (Figure 2). These serve as control of quality; positive controls are also used for data normalization. Level of unspecific cross-hybridization, background intensity and uniformity were used to optimize hybridization and washing protocols.

Normalization of Sterolgene data

Normalization of Sterolgene data was done using LOWESS fit to normalization spike in RNAs according to their average intensity (A = log₂√(R*G), where R and G represent signal intensities). Figure 3A shows a MA plot of Sterolgene data and a LOWESS curve fitted to the normalization controls. However, data from normalization controls were first adjusted according to concentrations in which normalization RNAs are added to sample RNAs: log₂ ratios of concentrations are subtracted from the log₂ ratios of measured intensities (M = log₂(R/G)). In Figure 3C the same data as in 3A is shown but with normalization controls adjusted according to their RNA concentrations. Notice the difference between the normalization curves at large A values. Such adjustment results in better fit of the normalization curve and consequently in more accurate normalization. Figures 3B and 3D show MA plots of data normalized according to the LOWESS curves shown in Figures 3A and 3C, respectively. Notice the difference in M for genes with large A values. Sterolgene data was normalized in Orange software [13] using a dedicated normalization widget, which was developed specifically for normalization of custom low-density arrays.

Acute TNFα inflammation and fasting affect cholesterol homeostasis and drug metabolism

The Sterolgene microarray was first evaluated in studies of how acute inflammation affects cholesterol homeostasis and drug metabolism. Mouse liver RNAs from the fasting and TNF-α/fasted experiments were hybridized to Agilent 10 K cDNA microarrays (G4104A) and Sterolgene arrays according to the recommended protocols. Differential expression was determined by two tailed Student's t-test. For Sterolgene data, probability of type I error α_S = 0.05 was used. For Agilent data, a complementary probability of type I error was calculated according to Bonferroni correction for multiple testing (α_A = α_S*n_S/n_A = 0.00055 where n_S and n_A represent the number of genes included in Sterolgene and Agilent arrays, respectively), but final analysis was made using a more relaxed criteria α_A = 0.01.

Using the Sterolgene array we show that fasting turns off most of anabolic processes due to energy conserving shifts in the liver and we observe a down-regulation of a majority of genes. These genes code for drug metabolizing enzymes (Cyp3a41, Cyp1a2, and Cyp3a25), different nuclear receptors (Pparα, Rorγ, Rorα, and Lxrα), cholesterol biosynthesis (Acat2, Mvk, Sc4mol, Cyp51a1, Nsdhl, Sc5d,) and transcriptional regulation of cholesterol biosynthesis (Srebp2) (Table 1). We also observed down-regulation of Alas1 (delta-aminolevulinate synthase 1) and Cyp11a1, which has previously been reported [14]. However, bile acid synthesis (Cyp8b1) is up-regulated by fasting. Three down-regulated (Cyp51a1, Sc4mol, Srebp2) and one up-regulated (Cyp8b1) gene were confirmed by RT-PCR [15] (Figure 4A).

Gene name and function	Gene symbol	Log2 ratio	GeneBank Acc. No.
Cholesterol biosynthesis
Acetyl-CoA acetyltransferase 2	Acat2	-1.13	NM_009338
Mevalonate kinase	Mvk	-0.56	NM_023556
Lanosterol 14α-demethylase	Cyp51a1	-1.32	BC031813
Sterol-C4-methyl oxidase-like	Sc4mol	-2.05	NM_025436
NAD(P) dependent steroid	Nsdhl	-1.47	BC019945
dehydrogenase-like
Sterol C5 desaturase	Sc5d	-1.75	BC024132
Bile acid biosynthesis
Sterol 12α-hydroxylase	Cyp8b1	0.49	NM_010012
Cytochrome P450 superfamily
Cytochrome P450, 1a2	Cyp1a2	-1.35	NM_009993
Cytochrome P450, 3a25	Cyp3a25	-0.78	BC028855
Cytochrome P450, 3a41	Cyp3a41	-1.91	NM_017396
Cholesterol side chain cleavage enzyme	Cyp11a1	0.41	NM_019779
Retionic acid hydrolase	Cyp26a1	-0.83	NM_007811
Nuclear receptor superfamily
Peroxisome proliferator-activated	Nr1c1	-1.03	NM_011144
receptor alpha
Retinoic acid-related orphan	Nr1f3	-0.60	NM_011281
receptor gamma
Retinoic acid-related orphan	Nr1f1	-0.54	NM_013646
receptor alpha
Liver X receptor beta	Nr1h2	-0.46	NM_009473
Constitutive androstane receptor	Nr1i3	0.22	NM_009803
Transcription factors
Sterol regulatory element binding transcription factor 2	Srebp2	-0.60	NM_033218
CCAAT/enhancer binding protein alpha	Cebpa	-0.38	BC028890
Transcription factor 1	Tcf1	0.17	NM_009327
Other
Peptidylprolyl isomerase A	Ppia	-0.65	NM_008907
Organic anion transporter 1	Slco1a1	-0.55	AY195868
Delta-aminolevulinate synthase 1	Alas1	0.98	NM_020559

Genes in bold – RT-PCR analysis is consistent with microarray data. Cyp3a41 probe cross-hybridizes with Cyp3a11.

Note. Three mice were used in this experiment.

With the Agilent array we were not able to detect any change in cholesterol homeostasis and drug metabolism caused by TNF-α and fasting (Additional files 2 and 3). Agreement between both platforms is low with only one gene in common in TNF-α/fasted experiment (Cyp2f2) and with no genes in common in fasting experiment.

Additionally, we have selected only genes that are present in both platforms (we found 54 such genes) and determined their differential expression using the same p-value cut-off as for Sterolgene data (α = 0.05). The rationale behind was not to tighten the analysis conditions for Agilent data just due to the fact that the Agilent array contains more genes than the Sterolgene array. In fasting experiment we found 5 genes differentially expressed (Additional file 4) and of these Cyp1a2 was confirmed by Sterolgene and RT-PCR, Actb by RT-PCR, while Apoa1 was not confirmed by RT-PCR [15] (Figure 4A). In TNF-α/fasted experiment we found 4 genes differentially expressed (Additional file 5) and of these Cyp2f2 was confirmed by the Sterolgene (RT-PCR not performed), and Actb and Apoa1 by RT-PCR [15].

Low correspondence in differential expression may be due to the fact that of 54 genes in common only 20 showed significant signal on the Agilent arrays. For these genes we calculated Pearson's product moment correlation coefficient between log₂ ratios in both experiments. Figure 5A shows a scatterplot of that data. Though the correspondence in differential expression is low, we observed a significant correlation between the platforms (r = 0.508, p = 0.001, N = 40), which is comparable to correlations reported by other cross-platform comparison studies [12, 17–19].

Phenobarbital and cholesterol feeding affect cholesterol homeostasis and drug metabolism

The Sterolgene array was also tested in studies of how phenobarbital and cholesterol feeding affect cholesterol homeostasis and drug metabolism. Mouse liver RNAs were hybridized to Affymetrix MOE430A GeneChips and Sterolgene arrays according to the recommended protocols. Differential expression was determined by two tailed Student's t-test. For Sterolgene data, probability of type I error α_S = 0.1 was used. For Affymetrix data, a complementary probability of type I error was calculated according to Bonferroni correction for multiple testing (α_Af = α_S*n_S/n_Af = 0.00043 where n_S and n_Af represent the number of genes included in Sterolgene and Affymetrix arrays, respectively), but final analysis was made using a more relaxed criteria α_A = 0.001.

Additionally, we have selected only genes that are present in both platforms (we found 92 such genes) and determined their differential expression using the same p-value cut-off as for Sterolgene data (α = 0.1), with the same rationale as for the analysis of Agilent data. In phenobarbital experiment we found 21 genes differentially expressed (Additional file 8) and of these Alas1 and Cyp2b10 were confirmed by Sterolgene and RT-PCR, Cyp2a4 and Nr2c1 by Sterolgene (RT-PCR not performed), Scap by RT-PCR, while for Apoa1 and Scarb1 no confirmation was found (Figure 4C).

In cholesterol feeding experiment we found 41 genes differentially expressed (Additional file 9) and of these Fdft1 was confirmed by Sterolgene and RT-PCR, Acss2, Cyp51a1, Nsdhl, Pmvk and Sqle by Sterolgene (RT-PCR not performed), Cyp26a1 by RT-PCR, while for Ppara and Scap no confirmation was found (Figure 4D).

From the 92 genes that are present in both platforms we selected those that showed signal on both platforms and calculated a Pearson's product moment correlation coefficient between log₂ ratios for data from both experiments. Figure 5B shows scatterplot of that data. We observed a similar correlation as for the Agilent-Steroltalk comparison (r = 0.499, p = 0.000, N = 319) and comparable to correlations reported by other cross-platform comparison studies [12, 17–19]. Since more probes per gene are available on Affymetrix array we included all in the correlation analysis.

Chemicals

All chemicals were purchased from Sigma (St Louis, MI, USA) if not stated otherwise.

Mouse experiments and sample preparation

In the TNF-α (tumor necrosis factor alpha) and fasting experiments male mice C57BL/6 (Harlan, Italy) age between 10 and 12 weeks were housed in normal light-cycle room, maintained on standard rodent chow, and were allowed water and food ad libitum [15]. Three groups of mice, 5–8 animals per group, were injected i.v at the same time point either with vehicle (saline) or with human recombinant TNF-α 30 μg/animal in saline (a gift from V. Menart, Lek Pharmaceuticals d.d.) and food was withdrawn, except for the control group. After 20 h mice were sacrificed using CO₂ suffocation. Total RNA was isolated from liver using TRI-reagent (Sigma, St Louis, MI, USA) and purified by RNeasy Mini Kit (Qiagen GmBH, Hilden, Germany) according to the manufacturer's protocols.

In phenobarbital treatment and cholesterol feeding experiments male mice C57BL/6 (RCC, Ittingen, Switzerland) age between 9 and 11 weeks were maintained in a 12-hour light/12-hour dark cycle and fed ad libitum standard rodent chow diet or the same diet supplemented with 1% (w/w) cholesterol (Provimi Kliba SA, Kaiseraugst, CH). Mice were divided in three groups, 8 animals per group: the control group which was injected i.p. vehicle (5% DMSO-corn oil) and maintained on standard diet; the cholesterol group which was maintained on cholesterol diet for week prior injection and injected i.p. vehicle; and the phenobarbital group which was maintained on standard diet and injected i.p. with 50 mg/kg phenobarbital sodium in a 5% DMSO-corn oil. After 10 hours, animals were sacrificed using CO₂ suffocation. The livers were snap-frozen in liquid nitrogen, total RNA was isolated using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer protocol and purified by RNeasy Mini Kit (Qiagen GmBH, Hilden, Germany). The total RNAs of two animals were pooled in equal amounts resulting in four pools per group. Quality and purity of total RNAs from both studies were checked using the RNA 6000 Nano LabChip Kit on Agilent 2100 Bioanalyzator (Agilent Technologies, Pato Alto, CA, USA). All animal experiments followed the Amsterdam Protocol on Animal Protection and Welfare and were approved by the local ethics committees.

cDNA microarray production

For 95 selected genes cDNA microarray probes of a similar length (400 – 500 bp), melting temperature and located less than 1500 bp from the polyA tail were designed using ProbeWIZ software [35]. Three probes of firefly luciferase in sequential order from the 5'-end to the polyA tail have been designed as controls of the reverse transcriptase reaction and as normalization controls. Chloramphenicol acetyltransferase (CAT) and Renilla luciferase gene probes were designed as negative controls. For housekeeping genes two to three probes were designed enabling control of RNA degradation rate. PCR products were prepared and cloned using the Qiagen PCR Cloning kit (Qiagen GmBH, Hilden, Germany). Plasmids were sequenced to verify the correct insert sequence. From these plasmids PCR products were amplified (Qbiogene, Morgan Irvine, CA, USA) and purified using the QIAquick 96 PCR BioRobot purification kit (Qiagen GmBH, Hilden, Germany). PCR products were dried and dissolved in 50% formamide and 1% CHAPS (3-[(3-cholamidopropyl)-dimethylammonio]-1-propane sulfonate) spotting buffer in final 200 ng/μl concentration. DNAs were spotted on GAPS II-coated slides (Corning Inc. Corning, NY, USA) using a GeneTAC G3 robot (Genomic Solutions, Huntington, UK) in a controlled environment (25°C to 26°C and 25–29% humidity). Each PCR product was spotted in triplicates in two blocks yielding 6 replicas. Firefly luciferase probes were spotted in each block. Commercial "spike in controls" ArrayControl Spikes (Ambion, Austin, TX, USA) were also spotted, two in each block. After spotting slides were dried for 8 hours at room temperature, protected from light. The DNA was fixed by UV cross-linking at 250 mJ/cm² (Hoefer UVC500, Amersham Pharmacia Biotech, San Francisco, CA, USA). Slides were stored in a dry and dark place.

Probe labeling and hybridizations

In TNF-α/fasted and fasting experiments, individual samples from fasted and TNF-α-treated groups were hybridized versus a reference pool of samples from normal-fed control group using Sterolgene (3 biological replicas) and Agilent arrays (2 biological replicas, identical to those used in Sterolgene hybridizations). The hybridizations were performed in a single laboratory according to the recommended protocols.

In phenobarbital and high cholesterol feeding experiments, samples from 8 mice were pooled by two into 4 pools for each control, cholesterol fed and phenobarbital treated group. All 4 pools from each group were hybridized to Affymetrix GeneChips. 2–3 pools, identical to those used in Affymetrix hybridizations, were hybridized together with a reference pool (pool of all samples) to Sterolgene arrays. Hybridizations of each of the platforms were performed in two different laboratories according to the recommended protocols.

In Sterolgene experiments, 10 μg of total RNA was used for labeling using Agilent Fluorescent Direct Label Kit (Agilent Technologies, Pato Alto, CA, USA) according to the manufacturer's protocol. Before labeling spike RNAs were added to the total RNA in following quantities: 500 pg of Firefly luciferase mRNA (Promega, Madison, WI, USA); and one of the two mixes (reference mix and sample mix) of 8 RNAs from ArrayControl Spikes (Ambion, Austin, TX, USA). The reference spike mix was a mixture of 500 pg of each RNA. The sample spike mix was a mixture of RNAs ranging from 100 pg to 1500 pg. In the TNF-α and fasting experiment, the control group was labeled with Cy3 (PerkinElmer Life Sciences, Boston, MA, USA), the fasted and the TNF-α treated groups were labeled with Cy5 (PerkinElmer Life Sciences, Boston, MA, USA). In the phenobarbital and cholesterol experiment the control, cholesterol and phenobarbital groups were labeled with Cy5 and the reference was labeled with Cy3. Sterolgene slides were washed before hybridization at room temperature for 5 min in 2 × SSC and 0.1%SDS, 3 min in 0.2 × SSC and 2 min in MilliQ water and dried by centrifugation. Samples were denatured and hybridized in 3 × SSC and 0.2% SDS hybridization buffer at 65°C for 16 hours using humidified hybridization chambers (HybChambers, GeneMachines, San Carlos, CA, USA). After hybridization slides were washed at room temperature for 5 min in 2 × SSC and 0.5% SDS, followed by 5 min in 0.5 × SSC, then 5 min in 0.05 × SSC and dried by centrifugation. Slides were scanned using a Tecan LS200 scanner (Tecan Group Ltd., Maennedorf, Switzerland).

In Agilent experiments, 10 μg of total RNA was used for labeling by Agilent Fluorescent Direct Label Kit (Agilent Technologies, Pato Alto, CA, USA). Preparation of samples, hybridization and washing protocols were done according to Mouse cDNA Microarray Kit (Agilent Technologies, Santa Clara, CA, USA) using the mouse Agilent cDNA microarrays (G4104A). The control group was labeled with Cy3, the fasted and the TNF-α treated group were labeled with Cy5. Hybridizations were performed using hybridization chambers (HybChambers, GeneMachines, San Carlos, CA, USA) and slides scanned by Tecan LS200 scanner (Tecan Group Ltd., Maennedorf, Switzerland).

In Affymetrix experiments, total RNA was analyzed by the MOE430A GeneChip (Affymetrix, Santa Clara, CA, USA) according to manufacturer's recommendations and slides scanned by a laser scanner (Affymetrix, Santa Clara, CA, USA) as described in [36].

Data normalization and analysis

Images of Sterolgene arrays were analyzed by Array-Pro Analyzer 4.5 (Media Cybernetics, Bethesda, MD, USA). The median feature and local background intensities were extracted together with the estimates of their standard deviation. Only features with foreground to background ratio higher than 1.5 and coefficient of variation (CV, ratio between standard deviation of the background and the median feature intensity) lower than 0.5 in both channels were used for further analysis. Log₂ ratios were normalized using LOWESS fit [37] to spike in control RNAs according to their average intensity. For normalization, two types of spike in controls were used: custom-made Firefly luciferase (348 such probes per array) and commercial ArrayControl Spikes (Ambion, Austin, TX, USA) (96 such probes per array). Ambion spike in RNAs were added to each sample RNAs in the following concentrations: 1:5, 1:2, 1:1, 2:1 and 3:1. Data from Ambion spike in controls were adjusted prior to fit of normalization curve by subtracting log₂(1/5), log₂(1/2), log₂(2) and log₂(3) from the log₂ ratios of measured intensities according to the concentrations (Figure 3C). In phenobarbital and cholesterol-feeding experiments data were standardized (median-centered and scaled by median absolute deviation) in order to reduce inter-array variability. Mean was used to average replicated microarray measurements. Data normalization and standardization was done in Orange software [13].

Images of Agilent microarrays were analyzed using Array-Pro Analyzer 4.5 (Media Cybernetics, Bethesda, MD, USA). Features, with CV > 0.39, were filtered out and data were normalized using LOWESS fit to all genes according to their average intensity [38]. Filtration and normalization was done in BASE software [39].

Affymetrix data were normalized by the Robust Multichip Average (RMA) algorithm as described [40]. After transformation to non-logarithmic data, the expression estimates were scaled to the average expression levels in the control group analyzed using GeneSpring software (Silicon Genetics, Redwood, USA).

Classification of differentially expressed genes was done in Orange using two tailed Student's t-test for independent samples. Between the platforms genes were matched by unigene or refseq codes, or by gene symbols. Pearson's product moment correlation coefficients were calculated and scatterplots produced in SPSS 14.0. All data have been submitted to GEO (Gene expression omnibus) under accession codes: GSE6271 (Affymetrix data), GSE6317 (Agilent data), GSE6447 (Sterolgene phenobarbital and high-cholesterol diet data), and GSE6423 (Sterolgene fasting and inflammation data).

Real-time polymerase chain reaction (RT-PCR)

RT-PCR analyses were performed in triplicates on pools of 3 mice from each group in TNF-α and fasting experiments, and on 4 biological replicas in phenobarbital and high-cholesterol diet experiments. In TNF-α and fasting experiments 1 μg of total RNA was reverse-transcribed using Superscript™ II or III reverse transcriptase (Invitrogen, Carlsbad, CA, USA). Analyses were done using Platinum^® SYBR^® Green qPCR SuperMix-UDG (Invitrogen, Carlsbad, CA, USA), except in case of Sc4mol where Assay-on-demand from Applied Biosystems was used, and were performed on Applied Biosystems ABI PRISM 7900 HT or 7500 (PE Applied Biosystems, Foster City, CA, USA) according to the manufacturer's protocol. In phenobarbital and high-cholesterol diet experiments 1 μg of total RNA was reverse-transcribed using MMLV reverse transcriptase (Roche Molecular Biochemicals, Rotkreuz, Switzerland). Analyses (for transcripts Cyp3a11, Cyp2b10, Alas1) were done using the TaqMan PCR Core Reagent Kit (PE Applied Biosystems, Rotkreuz, Switzerland), and performed on ABI PRISM 7700 Sequence Detection System (PE Applied Biosystems, Rotkreuz, Switzerland) according to the manufacturer's protocol. Additional analyses (for transcript Scap, Apoa1, Car, Cyp7a1, Cyp1a2, Scarb1, Ppara, Cyp26a1) were done using Platinum^® SYBR^® Green qPCR SuperMix-UDG (Invitrogen, Carlsbad, CA, USA) and 7500 (PE Applied Biosystems, Foster City, CA, USA) according to the manufacturer's protocol. Relative transcript levels were determined using the comparative Ct (cycle threshold) method or using 2^-ddCt values [41]. Internal control used in the first set of experiments is 18s and in the second set Gapdh. Statistical significance was determined by two tailed Student's t-test (α = 0.05). Primer sequences are provided in additional material (Additional file 10).