Construction of a composite porcine skeletal muscle cDNA microarray

The de novo development of a porcine skeletal muscle cDNA microarray was based on the use of two representative, directionally cloned λZAP-Express cDNA expression libraries; one derived from the back muscle of a 50-day-old foetus and the other from the gastrocnemius of a 3-day-old pig. Genes that are expressed in skeletal muscle are likely to be present in one or both libraries. In our porcine skeletal muscle cDNA libraries, about a third of the clones were found to be from highly expressed genes, based on signal intensity of hybridisation (Fig. 1A). Weakly to moderately expressed genes were selected for microarray assembly; highly expressed clones were avoided (Fig. 1B). On average, between a quarter and a half of the randomly picked colonies from each square agar plate were eventually selected for microarray assembly. A total of 5,500 plasmid clones were selected, of which 3,500 clones were taken from the 3-day-old muscle library and 2,000 clones were taken from the 50-day foetal muscle library. After each production step of miniprep plasmid preparation, insert amplification, and PCR product purification, about a quarter of the selected clones were checked by gel electrophoresis or spectrophotometry. Approximately 10% of all selected clones were found to be without any cDNA insert, a figure expected in a typical cDNA library (data not shown). Each selected clone was printed twice on the microarray (Fig. 2A).

Functional assessment of the porcine cDNA microarray: red-white muscle analysis

To evaluate the performance of our composite porcine skeletal muscle cDNA microarray, a profiling experiment was conducted to determine the differential expression of genes between red (psoas) and white (longissimus dorsi, LD) muscle. Dual-colour hybridisation (Fig. 2A) was performed on seven replicate microarray slides. An intensity-dependent (LOWESS) step was used to normalise data. Fig. 2B plots the median expression of each clone in the psoas against the median expression of the same clone in the LD muscle. Each point is the median of 14 values (2 replicates per slide, 7 slides per muscle). Most points cluster around the middle line indicating similar levels of expression in both muscles. There were, however, a number of clones falling substantially below the line, indicating consistently lower levels of expression in the psoas compared with the LD muscle. Low signal readings for both dyes may indicate the absence of a cDNA insert. About 12% of the printed clones were found to fall below 50 units for both dyes and were considered to be without cDNA inserts. This value was consistent with the earlier estimate of insertless clones in the cDNA libraries.

Quantitative expression of selected genes: validation by TaqMan real time RT-PCR

To assess the validity of the microarray approach to identify differentially expressed genes, quantitative real-time RT-PCR (TaqMan) was performed on four representative clones, GAPDH, MyHC 2b, bin1, and a novel gene (kc2725), each normalised to β-actin. In line with functional expectations, recent quantitative work performed has shown that the mRNA expression of GAPDH and MyHC 2b was higher in the psoas than in the LD within individual animals [20]. In this study, the relative levels of GAPDH and MyHC 2b of pooled total cDNA samples from three 22-week-old Berkshire pigs were measured (Fig. 3). Our results showed that the differential expression of GAPDH and MyHC 2b was sufficiently consistent between the two muscles to be detected in pooled cDNAs. A novel gene kc2725 of unknown function, found by microarray analysis to be more highly expressed in the psoas than in the LD (Table 1), was individually quantified in 4 pigs (Fig. 4A). CK2 phosphorylation sites were also predicted in the deduced kc2725 protein (data not shown). All 4 pigs gave the same pattern of expression as that detected on the microarray. Interestingly, expression of kc2725 was ubiquitous and was much more abundant in other tissues than in skeletal muscle (Fig. 4B).

However, the expression of a gene from the same muscle can vary between similar individuals (Fig. 4A and Fig. 5A). Bin1 was identified by microarray analysis as being more highly expressed in the LD than in the psoas (Table 3). However, this pattern of expression was found in only 3 out of 4 pigs (Fig. 5A). In pig 2, there was no significant difference in bin1 expression between LD and psoas. Bin1 was previously found to be ubiquitously expressed by Northern analysis [18, 21]. Quantitative PCR showed that bin1 expression was between 1 to 2 orders of magnitude greater in the heart and brain than in skeletal muscles (Fig. 5B). Therefore, TaqMan real-time quantitative PCR on selected representative genes had (1) demonstrated the functional integrity of our newly constructed porcine cDNA microarray in the appropriate identification of differentially expressed genes between red and white muscles, and (2) highlighted the existence of variation in muscle gene expression among similar individuals.

Development of porcine skeletal muscle cDNA microarray

We have constructed a composite porcine skeletal muscle cDNA microarray consisting of 5,500 clones from two developmentally distinct libraries, one derived from a 50-day foetal longissimus dorsi muscle, and the other derived from a 3-day-old gastrocnemius muscle. The choice of two developmentally distinct libraries was to increase the range of temporally-regulated genes represented in the microarray, to extend its suitability for use in different microarray-based muscle experiments.

The selection emphasis for inclusion on the microarray was on lowly or moderately expressed clones. This clone selection reduced the representation of highly expressed genes, such as those encoding for sarcomeric proteins, and increased the likelihood of including rare transcripts, including those of regulatory importance. An alternative method of normalising clone selection, which has been used by others, is by reassociation of single-stranded library plasmids at relatively low C_ot to remove highly expressed clones [22, 23]. Another commonly used and commercially available method of selection normalisation is subtractive suppression hybridisation [24]. We chose not to use this method for microarray clone selection because of the possibility of excluding lowly expressed non-muscle specific clones.

A possible disadvantage of our porcine microarray, at least at the beginning, is the lack of knowledge of the identity of each clone. However, there is, at present, insufficient sequence information on farm animals to design a comprehensive oligonucleotide-based microarray. To date, we have sequenced about 10% of our microarray clones (data not shown). A major advantage of our microarray is that we are likely to be in possession of the corresponding full length cDNA clones, whose inserts were unidirectionally cloned into a CMV-promoter driven expression plasmid (pBK-CMV vector). These clones could be readily used for downstream expression studies. As the identity of each clone is unknown, we do not know for certain how representative is the porcine skeletal muscle cDNA microarray. On the one hand, considering that there might be fewer than 30,000 human genes [25], and assuming that 50% of all genes are transcriptionally active at one time in a given tissue [26], it is possible that around 20% of the genes that are expressed in skeletal muscle are found on our microarray [25, 27]. On the other hand, a group of 4080 human skeletal muscle genes, which included both skeletal muscle-specific genes and genes expressed in skeletal muscle as well as in other tissues, was found to correspond to 80% of the total number of genes expressed in skeletal muscle as reported so far in Unigene (including foetal muscle and rhabdomyosarcoma) [26]. Hence our microarray may represent substantially more than 20% of all genes expressed in porcine muscle.

Differential gene expression in red and white muscles

One objective of the red-white muscle experiment was simply to test the function of the newly assembled porcine microarray, from which two gene lists were generated (Tables 1 and 2). Genes that were expected to be differentially expressed and genes that were novel were found on each list. The microarray results validated our prior hypothesis of differential gene expression in red and white muscles, thus demonstrating the functional integrity of our newly constructed microarray. One of the well established distinguishing features of red muscle is its relatively high oxidative phosphorylation capacity, reflected by an abundance of mitochondria in red muscles. It is reassuring that genes from the mitochondrial genome were well represented in the red muscle pool of differentially expressed genes (Table 1). White muscles comprise predominantly more fast-glycolytic fibres than red muscles. Our findings were consistent with expectations, in that most of the 45 clones selected as more highly expressed in white muscle (Table 2) were either fast isoforms of structural genes, or enzymes connected with anaerobic glycolysis. With highly expressed genes, such as mitochondrial genes and structural genes, there were detectable levels of redundancies (repeats) on the microarray. However, in the presence of redundancies, no gene was found to be present on both lists. Differentially expressed clones originating from both the foetal and neonatal libraries were found in comparable proportion (data not shown). A recent report comparing red and white murine skeletal muscles on generic commercial oligonucleotide chips (Affymetrix GeneChips), which comprised the equivalent of 3,000 different genes, yielded a differential list of 49 known genes [28]. Our results of known differentially expressed genes were comparable in both number and in the different types of genes found.

One problem underlying the analysis of microarray data is the large number of comparisons required, which can produce false positive results [29]. We compared the expression of clones in psoas and LD muscle as the median of 14 comparisons (each microarray clone was printed twice on each slide and 7 slides were used), and restricted further analysis to the clones demonstrating the most consistent and extreme differences. Therefore, the majority of our identified clones are likely to represent real differences between muscles. However, any specific gene assignment should be regarded as provisional until it can be confirmed in an independent study, such as another microarray or in another assay such as quantitative PCR.

The interpretation of microarray results can be complicated. Firstly, members of the same gene family could cross-hybridise to the same spots on the microarray. Interpretation of differential expression of individual isoforms should therefore be made with caution. In the case of MyHC genes (Table 2), a plausible interpretation is that fast MyHC mRNA isoforms (2a, 2x and 2b) were more abundant in the LD than in the psoas. However, we have previously found by quantitative real time RT-PCR that in at least 4 out of 6 pigs, of the same sex, age and breed as the one used in the microarray hybridisation, MyHC 2a and 2x were, in fact, more highly expressed in the psoas than in the LD [20]. The 3 fast MyHC isoforms found in Table 2 might have been the result of cross hybridisation by the relatively large amounts of MyHC 2b specific probe generated from the LD muscle. Secondly, in comparing the profiles of two normal physiological states, such as red and white muscles, large variation in normal gene expression between individual pigs could present a major problem [30]. This variation is mainly attributed to genetic differences that exist between individual pigs; even pigs of the same breed are not genetically the same. The use of pooled porcine mRNA samples could inadvertently increase the genetic variation within each experimental group of animals. Therefore, in the context of porcine red-white muscle microarray analysis, there may be no advantage in pooling mRNAs, derived from the same muscle of several pigs. On the other hand, the use of different muscles from the same pig could minimise the effects of environmental variation, which could exist between individuals. The use of inbred lines in laboratory rodents largely eliminates the problem of genetic variation between individuals of the same line. Hence, the use of labelled cDNA from pooled inbred individuals would enhance experimental reliability without increasing genetic variation. However, inbred pig strains are not widely available and limited to a few lines of mini-pigs. Whether the microarray probes were derived from an individual or pooled from a group of individuals, extensive validation, such as by quantitative PCR or Northern analysis, is necessary to demonstrate that the differential expression of a gene identified on a microarray is consistent in a wider context.

Candidate genes for phenotype determination

From the red-white muscle microarray results, a list of novel candidate regulatory genes that could influence muscle phenotype, such as hypertrophy, differentiation and isoform-specific expression, was identified. Most of the genes listed as unknown (Table 1 and 2) were found with major open-reading frames, suggesting that they code for protein products. Candidate regulatory genes with putative identities were based on homology comparison. Even for these genes, with the possible exception of bin1, their functional roles in skeletal muscle are largely unclear. CK2 α1 subunit, smpx, and tyrosine kinase A6-related gene are particularly interesting. They were found to be more highly expressed in red muscle and are connected to the casein kinase 2 signalling pathway. CK2 is a serine/threonine kinase that has been implicated in cell growth and proliferation [31, 32]. CK2-mediated phosphorylation of Myf-5, a member of a family of myogenic transcription factors, was reported to be required for Myf-5 activity [33]. Presently, the contribution of the CK2 signalling pathway to skeletal muscle function is not known. Its role in muscle phenotype determination requires further evaluation. The gene for heat shock 70 kD protein 8 (HSPA8) seemed to be upregulated in white muscle. Heat shock proteins are considered to be molecular chaperones and indicators of cellular stress [34]. The same gene was found to be upregulated in human hypertrophic cardiomyopathy [35]. It is not clear if this finding was related to cellular stress of myopathy or muscle hypertrophy.

Our results demonstrate the power of microarray analysis in identifying candidate genes that influence muscle phenotype. The challenge now is to confirm these associations and demonstrate how these genes are involved in the relevant muscle phenotypes.

Construction of porcine skeletal muscle cDNA microarray

Mass phagemid excision was performed on two porcine skeletal muscle λZAP-Express cDNA libraries (50-day foetal longissimus dorsi muscle and 3-day-old gastrocnemius muscle), developed in-house, according to manufacturer's protocol (Stratagene). Both libraries, each with a million primary plaques, had been extensively characterised by sequencing and screening, and were used to isolate several full-length cDNA clones, including MyHCs (6.0 kb in size) and transcription factors, such as NFAT2 and GATA2. One study, performed on adult human skeletal muscle, found that highly expressed genes represented only 9.1% of the transcript variation, whereas moderately and weakly expressed genes made up 27.5% and 63.5% of the variation, respectively [26]. In order to ensure that the clones chosen for the porcine microarray assembly represented as many different genes as possible, clones were screened by colony hybridisation to assess their relative abundance of expression. About 400 bacterial clones were picked onto each square agar plate (Bio-Assay dish, Nunc) for colony hybridisation with ³²P-labelled cDNA probe, derived from mRNA extracted by oligo(dT)₂₅ Dynabeads (Dynal) from the same muscle type and stage of development as that used for the library construction. Based on signal intensity on the autoradiograms, each clone was classified as a weakly, moderately or highly expressed gene [26]. Clones picked for microarray construction were judged to be lowly or moderately expressed. Miniprep plasmid DNA was prepared from selected clones using the QIAprep 96 Turbo kit (Qiagen).

Insert amplification of each pBK-CMV-based plasmid clone (Stratagene) was performed with T7 primer (5'-GTA ATA CGA CTC ACT ATA GGG C-3') and T3 primer (5'-CGA AAT TAA CCC TCA CTA AAG GG-3'), using a HotStarTaq Master Mix kit (Qiagen). PCR was conducted in a 100 μl volume, using the equivalent of 0.1 μl of miniprep plasmid (about 15 ng), for 35 cycles at 55°C for 45 s, followed by 72°C for 3.5 min, and 94°C for 1 min, with an initial activation step of 95°C for 15 min. PCR products were purified with QIAquick 96 PCR Purification kit (Qiagen) and printed onto CMT-GAPS coated slides (Corning) using a Microgrid II arraying robot (BioRobotics, Cambridge, UK). Spotted DNA was immobilised by baking at 80°C for two hours. Each clone was printed in duplicate on each slide. Appropriate positive controls to help with orientation and vector controls were incorporated on the microrarray.

Red-white muscle microarray hybridisation

Messenger RNA was extracted by oligo(dT)₂₅ Dynabeads (Dynal) from the longissimus dorsi (LD), a white muscle, and the psoas, a red muscle, of a 22-week-old pig (Berkshire breed). Dual Cy-dye labellings were performed with a CyScribe First-Strand cDNA labelling kit (Amersham), which incorporates the use of oligo(dT) primers and random nanomers. Labelled cDNAs were purified in AutoSeq G-50 columns (Amersham). In a typical Cy3 or Cy5 labelling reaction, using 0.5 to 1.0 μg mRNA, a final volume of 32 μl was obtained, which was used on two microarray slides. Prior to pre-hybridisation, microarray slides were denatured in distilled water by heating to 95°C for 2 min. Pre-hybridisation was then performed in a plastic 2-slide holder containing 15 ml of 3 × SSC (sodium chloride and sodium citrate), 2% bovine serum albumin (B4287, Sigma) and 0.1% SDS at 65°C for 20 min. After a brief rinse in distilled water at room temperature, the slides were dehydrated in absolute ethanol and centrifuged at 100 g for 2 min.

Labelled probe was not quantified. Equal volumes of Cy3 and Cy5 labelled probes (16 μl each) were mixed with 1.0 μl (8.0 μg/μl) of poly(dA) oligonucleotide (27–7836, Amersham), heated to 95°C for 2 min, and mixed with 33 μl of 2 × hybridisation buffer (GHB-200, Genpak). A microarray slide was layered with the hybridisation mixture, then covered with a 22 mm × 64 mm cover glass (BDH), to the exclusion of air bubbles, and placed in a hybridisation chamber (ArrayIt hybridisation cassette AHC-1) in a 45°C dry incubator for 24 h. Slides were washed once in 300 ml of 1 × SSC and 0.2% SDS for 10 min at 45°C, twice in 300 ml of 0.2 × SSC and 0.2% SDS for 10 min at 45°C, and twice in 300 ml of 0.1 × SSC for 10 min at 37°C. After drying in a centrifuge at 100 g for 2 min, scanning was performed with an Affymetrix 428 scanner. A total of seven microarray slides (replicates) were subjected to two-colour Cy-dye hybridisation and scanned for the red-white muscle experiment.

Microarray expression analysis and clone identification

Image analysis (grid generation and dye quantification) of scanned slides was performed with ImaGene v4.2 (BioDiscovery). Data mining was conducted with GeneSpring v4.2 (Silicon Genetics), in which 3 normalisation steps were performed: per spot, intensity-dependent, and per chip (slide). In per spot normalisation, after background subtraction, the fluorescent intensity (e.g. Cy3) of each clone was divided by its control or reference channel intensity (e.g. Cy5). Values of the control channel that fell below 10 were adjusted to 10 prior to taking the ratio between signal and control value. Intensity-dependent (non-linear or LOWESS) normalisation was applied to correct for artefacts caused by differential Cy3 and Cy5 dye incorporation and non-linear rates of fluorescence between the two dyes. Per chip normalisation was used to take into account intensity variation across the entire slide, by dividing the signal strength of a clone on a slide by the 50^th percentile signal of all of the measurements taken from the same slide. Additionally, the GeneSpring Global Error Model, based on the replicate measurement samples of all genes on the microarray was used to estimate differences between medians, the standard errors among the different clones, and the extreme tails of the distribution of differences. Differences among muscles in expression were standardised for different clones by dividing the difference between muscle medians by the square root of the common variance. Sequencing reactions with T7 and T3 primers were conducted with an ABI BigDye Terminator v3.0 PCR-based sequencing kit (Applied Biosystems). Performa DTR gel filtration cartridges (EDGE Biosystems) were used for purification of PCR products prior to loading onto the ABI PRISM 3100 Genetic Analyzer. All sequence data were subjected to BLAST (Basic Local Alignment Search Tool) searches for gene identification by sequence similarity.

Quantitative real-time RT-PCR

Quantitative real-time RT-PCR was performed using TaqMan (Applied Biosystems) on five selected porcine genes: β-actin, GAPDH, MyHC 2b, bin1, and a novel gene (kc2725) (Table 3). The protocol, based on the use of the relative standard curve, was as previously described [20]. In the relative standard curve method of quantification, comparisons of relative expression should only be made between samples of the same gene. It is not appropriate to compare expression levels between two different genes. A reference cDNA panel, comprising a number of different tissue templates (LD muscle, psoas, heart, uterus, brain, liver and spleen) of a 7-week-old pig, along with LD and psoas cDNAs, from three additional 22-week-old Berkshire pigs, were used to evaluate the quantitative distribution of selected genes.