For microarray preparation we selected genes coding for known or predicted mitochondrial proteins, genes known to be involved in cell cycle growth and regulation, and genes involved in apoptosis and free radical metabolism.

The final set contained 1632 genes, of which 992 are "mitochondrial" genes, 42 lysosomal genes, 277 genes are associated with apoptosis, and 321 are "oncogenes". For normalization and background correction we included 146 human "housekeeping" genes, 10 Arabidopsis genes and 32 blanks. Full list of selected genes with corresponding symbols, accession and LocusLink codes is provided [see Additional file 1]. Functional annotation of selected genes and comparison of the gene content against whole human genome set is provided [see Additional file 2].

Hybridization properties and performance of designed oligonucleotide probes and control features placed on h-MitoArray were tested by hybridization of fluorescently labeled panomers and fluorescently labeled cDNA prepared from a pool of total RNA isolated from several cell lines (test RNA). Gene expression signal was detected in > 77% of 1820 elements when fluorescently labeled cDNA pool was used.

Following comparison of various labeling strategies and optimization of hybridization conditions a series of self-to-self experiments was performed using test RNA. Data analysis showed acceptable reproducibility with Pearson correlation coefficient ranging 0.987 – 0.991.

Fluorescent cDNA probes labeled with Cy5 were prepared from 13 patient and 9 control cell lines and were hybridized to common reference cDNA probe labeled with Cy3 in two technical replicates for each sample. Following data acquisition, transformation, normalization and replicate averaging, gene expression signals were obtained for 1264 genes. Ratios of Log2 sample gene intensities against Log2 common reference gene intensities (M) were calculated and are provided [see Additional file 3]. Calculated ratios of individual patient Log2 gene intensities against the Log2 of average of controls gene intensities (M) are provided [see Additional file 4].

To reveal biological consequences and to identify pathways potentially involved in the pathogenesis of the studied defects, we extracted from original expression data for each of the three defined groups all genes found to be differentially expressed at unadjusted P < 0.05 and showing expression change |M| > 0.2. Resulting expression datasets were uploaded into the DAVID database 23 and gene enrichment analysis was performed against h-MitoArray gene list. Results are provided in Table 1.

As the enrichment analysis suggested group specific dysregulation of several metabolic and signaling pathways, we further uploaded identical datasets into KEGGArray software (KEGG pathway databases – Kyoto Encyclopedia of Genes and Genomes) and inspected gene expression changes in all the indicated pathways.

In the

M group

In the

N1 group

N2 group

To rule-out platform specific bias, we re-analyzed all RNA samples from the N1 and control groups using the same common reference RNA on Agilent 44 k arrays. We used available annotations and extracted from the Agilent data gene expression values for the genes identified as significantly (P < 0.05), differentially expressed in the N1 group on our platform. Correlation coefficient of expression values of 102 identified genes was 0.925.

ATP synthase deficiency of nuclear genetic origin is characterized at the protein level by pronounced decrease of the individual subunits and the mature ATP synthase protein complex amounts (Table 2). However, our data in patient cell lines, in agreement with previous Q-PCR analyses, did not show pronounced alterations in ATP synthase subunits or of ATP synthase-specific assembly factors mRNA levels that could explain it easily. Only in the M group, the data showed decrease of MTATP6 and MTATP8 mRNA levels which correspond with previously performed Northern blot and Q-PCR analysis showing that this mutation affects processing of ATP8/ATP6/COXIII polycistronic transcript and results in decreased levels and/or stability of mature ATP8/ATP6 mRNA 1924. Many of mitochondrial diseases are associated with compensatory changes in the cellular content of mitochondria and/or the content of one or more OXPHOS complexes. Western blot analysis of fibroblasts with ATP synthase deficiency has previously shown increased mitochondrial content of complex I and complex III 25. In agreement with this observation, our data showed elevated expression of complex I subunit genes in N1 group. Expression of complex III subunit genes was however decreased. Parallel analyses of the fibroblasts with nuclear ATP synthase defects used in this study revealed variable changes in fibroblast COX and/or SDH specific content (Table 2). These changes were not associated with generally elevated expression of COX and SDH subunit genes. Detailed inspection of individual gene expression profiles [see Additional file 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24] however suggested that elevated expression of COX7A2L and SDHA may correlate with this observation (P3, P5, P6, P8, P10).

In our work, we attempted to set up an experimental platform which will allow in a cost-effective way prospective gene expression analysis of cell lines and tissues from patients with various genetically determined OXPHOS defects. We considered availability of biological materials for the analysis, estimated the number of informative genes, evaluated gene content of commercially available microarrays and took into account instrumentation availability and platform related running costs. Since cultured skin fibroblasts are the most accessible, relatively well standardized, and multiple analysis amendable source for gene expression analysis especially in nuclear encoded OXPHOS defects, we estimated (based on Gene Expression Omnibus database data), that in fibroblasts, reliable expression signal may be obtained for approximately 6000 (HG-U95 array) to 10000 (HG-U133 Plus 2.0 Array) genes, of which only part may be meaningful to detect and understand anticipated changes in mitochondrial biology and related basic cell responses. In addition, we also evaluated representation of mitochondria encoded genes on available whole genome arrays. We found that (at the time of project planning) no complete coverage of mitochondrial tRNA, rRNA and OXPHOS structural subunits have been available on() Affymetrix HG_U95Av2 Array (contained just TRNC, TRNY and TRNS1), HG-U133 Plus 2.0 Array (no tRNAs, rRNAs and ND1, ND4L, CYTB) and Agilent 44 k Array (no tRNAs, rRNAs and ND4L). Considering this data and also available instrumentation, we then decided to construct focused an oligonucleotide microarray and employ competitive two-color hybridization approach with common reference experimental design.

Selected gene content allows gene expression analysis of the entire mitochondrial genome and almost all of "mitochondria" related genes in context of key DNA synthesis, growth response, regulatory and apoptotic genes. Hybridization signal was obtained from 78% of the designed oligonucleotides. Vast majority of the oligonucleotides giving no hybridization signal were designed to detect regulatory genes and transcription factor transcripts probably not transcribed in the analyzed materials. Interestingly, we detected hybridization signals for almost all mitochondrial tRNA and rRNA probes which is, in respect to oligo-dT labeling strategy, suggestive that all those transcript are also at least partially polyadenylated 26.

In the work presented herein, we analyzed and compared gene expression profiles in fibroblast cell lines from 9 control individuals and 13 patients with biochemically proven but genetically heterogeneous F₁F_o ATP synthase deficiency. We aimed to identify gene expression changes indicating how affected cells react to and compensate for the common biochemical defect, use gene expression data to assign patients into already defined and/or putative disease subgroups, identify candidate disease causing genes, and define potential pathogenetic mechanisms associated with the disease.

The magnitude of observed expression changes was moderate with only several dozens of genes exceeding 2-fold changes. Comparing all the patient cell lines with all control cell lines, we have not identified any common and meaningful gene expression changes attributable to ATP synthase deficiency per se. It has been suggested recently that the degree and compartmentalization of ATP depletion may be defect specific and may thus have also specific biological consequences 27. Our data support this view.

Cell lines with mtDNA mutation (

M group

The

N1 group

Resulting transcriptional silencing and other ATP depletion mediated disturbances of intracellular signal transduction cascades lead thus to premature senescence phenotype, reduced proteasome activity and accumulation of oxidized proteins, which may explain observed discrepancies between gene expression and Western blot data. Patients forming this group are of common ethnic origin, and this is suggestive that common genetic defect may underlie this specific gene expression profile.

The

N2 group

As gene expression changes may be used for selection of candidate disease causing genes 961, we evaluated group specific and individual gene expression profiles. This approach was successful in both patients from the M group, in whom detected alterations clearly indicated the involvement of ATP6/ATP8/COXIII transcript. In other patients we first focused on expression of ATP synthase subunits. Inspection of this expression profile (Figure 5B) suggested involvement of ATP synthase assembly factor ATPAF2 in several patients from N1 group. Mutation of ATPAF2 has been found in the case with ATP synthase deficiency 62 and this warrant sequence analysis of this gene in this group of patients. Other candidate genes may be ATP5G2, the expression of which is decreased in P5 and possibly also ATP5C1 found lowered in P3. From other genes, no clear candidates for immediate sequence analysis may be defined yet. However, more focused interpretation will be possible once candidate disease genomic intervals are defined by ongoing linkage studies.

Lists of "mitochondrial" and "mitochondria related" genes were extracted and merged from various public databases such as Mitomap 63, Mitop 64, Migenes 65, Mitoproteom 66, Molecular Signature Database 67, OMIM, RefSeq and Unigene sections at NCBI 68, Gene Ontology database 69 and UniProt resource 70. Full annotation of selected genes has been obtained and deposited in a locally installed database BASE 71.

For each of the selected 1632 genes, a single 5'-aminomodified 40-mer oligonucleotide was designed using Oligopicker software 72. Blast searches were performed with each candidate probe to exclude possibility of cross hybridization with homologous genes prior to the synthesis of oligonucleotide probes. Synthesized oligonucleotides, Generi Biotech (Czech Republic) and Illumina (San Diego, CA), were resuspended at 20 μM concentration in 3 × SSC, printed in triplicates on aminosilane modified slides, and immobilised by standard technique using combination of baking and UV cross-link as previously described 61. Qualities of arrays from individual printing series were assessed using fluorescently labelled panomers (Invitrogen, Carlsbad, CA).

As a standard for microarray optimisation, standardization and validation total RNA was isolated from HeLa G, ECV 304, 293, U 937, JURKAT and A 301 cell lines using the TRIZOL solution (Invitrogen, Carlsbad, CA). Isolated RNA samples were pooled, and aliquots were stored at -80°C until the analysis.

As a common reference RNA for gene expression studies, total RNA from cultured HeLa cells was chosen. Total RNA was extracted as above. Concentration was determined spectrophotometrically at A 260 by NanoDrop (NanoDrop Technologies, Wilmington, DE) and quality was checked on Agilent 2100 bioanalyser – RNA Lab-On-a-Chip (Agilent Technologies, Santa Clara, CA). Aliquots of isolated RNA were stored at -80°C until the analysis.

Selected control fibroblasts cell lines were used repeatedly in previous diagnostic biochemical tests and showed no signs of any mitochondrial or other metabolic defect.

Fibroblast cell lines from 13 patients were used in this study. All the patients showed major clinical symptoms associated with OXPHOS defect. Biochemical diagnosis of ATP synthase deficiency was based on absence or significant decrease of mature ATP synthase complex and of its subunits in electrophoretic analysis of OXPHOS complexes in cultured fibroblasts and other available tissues 73. Mitochondrial genome sequencing performed in all patients revealed disease causing mitochondrial DNA mutations in two patients (P1, P2, M group) 19. Molecular basis of defect in the other patients has not yet been defined. Relevant clinical, biochemical and molecular data and references on individual patients included in this study are provided in Table 2.

Growth characteristics of the cell lines used in this study are provided in Table 3. Skin fibroblasts were cultured in the Dulbecco's modified Eagle's medium supplemented by 10% fetal calf serum, 20 mM HEPES pH 7.5, 0.2% NaHCO₃ and gentamycin 0.02 mg/ml at 37°C in a 5% CO₂ humidified atmosphere. For experiments, confluent cell were harvested using 0.05% trypsine and 0.02% EDTA. Detached cells were diluted in ice-cold culture medium, sedimented by centrifugation (600 g) and washed twice in phosphate buffered saline (140 mM NaCl, 5.4 mM KCl, 8 mM Na₂HPO₄, 1.4 mM KH₂PO₄, pH 7.2).

Total RNA was extracted from cultured cells and QC controlled as described above.

Five μg of total RNA was reverse transcribed and labeled by Array 900 Expression Detection Kit (Genisphere, Hatfield, PA) according to the manufacturer protocol. The slides were pretreated by baking at 80°C, UV cross-linked and washed twice in 0.1% SDS for 2 minutes, twice in 0.2 × SSC for 2 min, four times in MilliQ water, followed by denaturation in boiling water for 2 minutes. Prehybridization was performed using hybridization buffer (Genisphere, Hatfield, PA) according to the manufacturer protocol. All hybridizations were performed in humid hybridization chamber, ArrayIt Hybridization Cassette chamber (TeleChem International, Sunnyvale, CA).

The hybridized slides were scanned with GenePix 4200A scanner (Axon Instruments, Union City, CA) with PMT gains adjusted to obtain highest intensity unsaturated images. GenePix Pro software (Axon Instruments, Union City, CA) was used for image analysis of the TIFF files, as generated by the scanner.

All 13 patient samples and 9 controls were hybridized to common reference (HeLa cell lines) in two replicates of each sample. All arrays were hybridized with a Cy5-labeled sample cDNA and a Cy3-labeled reference cDNA.

Expression data were obtained using GenPix Pro software. Comparative microarray analysis was performed according to MIAME guidelines 74. Normalization was performed in R statistical environment 75 using Limma package 76 which is part of the Bioconductor project 77. Raw data from individual arrays were processed using Loess normalization and normexp background correction. Gquantile function was used for normalization between arrays. The correlation between 3 replicate spots per gene on each array was used to increase the robustness. Linear model was fitted for each gene given a series of arrays using lmFit function. The empirical Bayes method was used to rank differential expression of genes using eBayes function. Multiple testing correction was performed using Benjamini & Hochberg method 78.

Variation among feature replicates on the array was calculated by conversion of raw data to log-ratios. Data were further normalized using Loess function. Features with less than double background intensity (A < 8.5) were removed. For each feature on the array the deviation from the mean computed as the difference between the ratio of the feature and the mean of the set of feature replicates was calculated. Standard deviation of the error distribution using all of the replicates was calculated and converted to coefficient of variability using equation.

CV = exp ⁡ [ ( ln ⁡ 2 ∗ SD 2 ) 2 ] − 1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGacaGaaiaabeqaaeqabiWaaaGcbaGaee4qamKaeeOvayLaeyypa0ZaaOaaaeaacyGGLbqzcqGG4baEcqGGWbaCcqGGBbWwcqGGOaakcyGGSbaBcqGGUbGBcqaIYaGmcqGHxiIkcqqGtbWucqqGebarcqaIYaGmcqGGPaqkdaahaaWcbeqaaiabikdaYaaakiabc2faDjabgkHiTiabigdaXaWcbeaaaaa@42C3@

The variability between the duplicate spots ranged from 8.1% to 27.5%. Arrays with variability higher then 18% were removed from the analysis.

Principal component analysis, hierarchical clustering, ANOVA and SAM analyses were performed in TIGR Multiexperiment Viewer (MeV), version 4.0 79, available 80. Significant gene expression changes between defined subgroups were identified using t-test in R statistical environment 75. Applied parameters are provided in corresponding result sections.

Functional annotation and pathway enrichment analysis was performed in DAVID (The Database for Annotation, Visualization and Integrated Discovery 23). Visualization of gene expression changes along affected pathways was performed in KEGGArray software (KEGG pathway databases – Kyoto Encyclopedia of Genes and Genomes) 81.

Description of h-MitoArray platform and gene expression data reported in this study are stored and available in Gene Expression Omnibus repository under accessions GPL5150 and GSE8648.

The project was approved by the Scientific Ethics Committee of the 1st Faculty of Medicine of Charles University of Prague under reference NR/8069-3. Patient participation in the project was made on a voluntary basis after oral and written information and consent according to the Helsinki V Declaration.