HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Free Via Open Access Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Cogburn, L. A. Articles by Burnside, J. Search for Related Content PubMed PubMed Citation Articles by Cogburn, L. A. Articles by Burnside, J.

Functional Genomics of the Chicken—A Model Organism

L. A. Cogburn^*,1, T. E. Porter, M. J. Duclos, J. Simon, S. C. Burgess, J. J. Zhu^||, H. H. Cheng^#, J. B. Dodgson^** and J. Burnside^*,

^* Department of Animal and Food Sciences, University of Delaware, Newark 19717; Department of Animal and Avian Sciences, University of Maryland, College Park 20742; Station de Recherches Avicoles, Institut National de la Recherche Agronomique, 37380 Nouzilly, France; College of Veterinary Medicine, Mississippi State University, Mississippi State 39762; ^|| Foreign Animal Disease Research Unit, USDA-Agricultural Research Service, Plum Island Animal Disease Center, Greenport, NY 11944; ^# Avian Disease and Oncology Laboratory, USDA-Agricultural Research Service, East Lansing, MI 48823; ^** Department of Microbiology and Molecular Genetics, Michigan State University, East Lansing 48824; and Delaware Biotechnology Institute, University of Delaware, Newark 19717

¹ Corresponding author: cogburn{at}udel.edu

	ABSTRACT

TOP ABSTRACT INTRODUCTION DEVELOPMENT OF GENOMIC RESOURCES HIGH-THROUGHPUT GENOME-WIDE... BIOINFORMATICS AND COMPUTATIONAL... CONCLUSIONS AND OUTLOOK REFERENCES

 
Since the sequencing of the genome and the development of high-throughput tools for the exploration of functional elements of the genome, the chicken has reached model organism status. Functional genomics focuses on understanding the function and regulation of genes and gene products on a global or genome-wide scale. Systems biology attempts to integrate functional information derived from multiple high-content data sets into a holistic view of all biological processes within a cell or organism. Generation of a large collection (600K) of chicken expressed sequence tags, representing most tissues and developmental stages, has enabled the construction of high-density microarrays for transcriptional profiling. Comprehensive analysis of this large expressed sequence tag collection and a set of 20K full-length cDNA sequences indicate that the transcriptome of the chicken represents approximately 20,000 genes. Furthermore, comparative analyses of these sequences have facilitated functional annotation of the genome and the creation of several bioinformatic resources for the chicken. Recently, about 20 papers have been published on transcriptional profiling with DNA microarrays in chicken tissues under various conditions. Proteomics is another powerful high-throughput tool currently used for examining the dynamics of protein expression in chicken tissues and fluids. Computational analyses of the chicken genome are providing new insight into the evolution of gene families in birds and other organisms. Abundant functional genomic resources now support large-scale analyses in the chicken and will facilitate identification of transcriptional mechanisms, gene networks, and metabolic or regulatory pathways that will ultimately determine the phenotype of the bird. New technologies such as marker-assisted selection, transgenics, and RNA interference offer the opportunity to modify the phenotype of the chicken to fit defined production goals. This review focuses on functional genomics in the chicken and provides a road map for large-scale exploration of the chicken genome.

Key Words: transcriptome • proteome • metabolome • systems biology • gene network

	INTRODUCTION

TOP ABSTRACT INTRODUCTION DEVELOPMENT OF GENOMIC RESOURCES HIGH-THROUGHPUT GENOME-WIDE... BIOINFORMATICS AND COMPUTATIONAL... CONCLUSIONS AND OUTLOOK REFERENCES

 
The completion of sequencing, assembly, and annotation of the chicken genome (International Chicken Genome Sequencing Consortium, 2004) represents a monumental achievement for biologists in basic and applied research (Burt, 2007; Dodgson, 2007). The chicken embryo has been used as a model system for embryology and developmental biology for more than 2 millennia (Stern, 2004, 2005). For more than 8 millennia, the chicken has flourished as a domesticated livestock species deeply integrated into human culture (Dohner, 2001; Price, 2002). Therefore, it is fitting that the chicken would be the first domestic animal chosen for complete genome sequencing. The fact that the chicken genome sequence was completed a century after the birth of classic poultry genetics is another remarkable coincidence (Dodgson, 2003). Functional genomics is a relatively new discipline spawned by the technological revolution (Venter et al., 1996; Rowen et al., 1997), the momentum of inquisitiveness, and the spirited competitiveness (Collins et al., 2003b) that drove early completion of the human genome sequence (International Human Genome Sequencing Consortium, 2001; Venter and et al., 2001). The sequence of the human genome, with its complement of 20,000 to 25,000 protein-encoding genes, was not truly finished until 2004 (Human Genome Sequencing Consortium, 2004).

The chicken genome sequence was completed within a year—a historic accomplishment for avian biologists (International Chicken Genome Sequencing Consortium, 2004; Siegel et al., 2006; Burt, 2007; Dodgson, 2007) empowered by the high-throughput technologies developed in quest of the human genome sequence (Collins et al., 2003a). A critical step toward sequencing of the chicken genome was high-throughput DNA sequencing of expressed sequence tags (EST) from dozens of tissue-specific cDNA libraries generated from several international projects (Abdrakhmanov et al., 2000; Tirunagaru et al., 2000; Boardman et al., 2002; Cogburn et al., 2003c; Carré et al., 2006). This feat has advanced the chicken to 14th place (with 599,330 EST) among all model organisms represented in the dbEST division of GenBank. Furthermore, the large international collection of chicken EST, and the subsequent full-length sequencing of 19,626 cDNA (Hubbard et al., 2005) has enabled functional annotation of the assembled chicken genome sequence. The sequencing of the chicken genome and the development of high-throughput screening platforms (microarrays) and bioinformatic tools clearly advanced the chicken to model organism status (Burt, 2005, 2007). As many have recognized, completion of the genome sequence simply marks the "end of the beginning" (Brenner, 2000; Stein, 2004; Dodgson, 2007) of genome exploration in that species. Functional genomics attempts to bridge the gap between the blueprint (genome sequence or genotype) and the living organism (trait or phenotype; see Figure 1).

View larger version (91K): [in this window] [in a new window]   Figure 1. Road map of functional genomics in the chicken and use of these resources for large-scale exploration of the avian genome. Functional genomics focuses on understanding the function and regulation of genes, proteins, and metabolites on a genome-wide scale. All information derived from the analyses of genome sequence, large-scale analysis of gene and protein expression, and metabolite profiles is integrated into knowledge databases. High-throughput DNA sequencing of cDNA libraries from various tissues has provided a comprehensive catalog of chicken genes and high-density microarrays for transcriptional analyses. High-throughput sequencing of genomic DNA has provided the chicken genomic sequence, with its structure and genetic variation. Bioinformatics and computational analyses allowed assembly and annotation of the genome, whereas various genome browsers (Ensembl, Entrez, UCSC, and ChickVD) enable detailed views of the annotated genome and links to other databases (i.e., Gene Ontology, UniProt, KEGG, and TIGR). Biological samples taken from the embryo or chick facilitate high-throughput analysis of the transcriptome, proteome, metabolome, or their combination. Computational analysis of transcriptional data sets provides information on gene expression patterns and transcriptional control over clusters of coexpressed genes in gene networks. Transcriptional profiling coupled with QTL analysis allows identification of expression QTL (eQTL), which encompasses both positional and functional candidate genes. Analysis of the transcriptome and proteome requires access to well-annotated gene (Gene Ontology) and protein (UniProtKB) databases. Systems biology represents the integration of high-content data sets from the transcriptome, proteome, and metabolome into functional maps of biological pathways that determine the phenotype of the chicken. The premise of functional genomics is that if the genotype is known, then we have the knowledge base and tools to change the phenotype of the chicken. Once functional candidate genes, proteins, or both are identified, MAS or other emerging technologies (transgenics and RNA silencing) can be used to modify the phenotype of the bird to fit defined production goals. Recent improvements in avian transgenics enable expression of human biologicals in eggs and provide a vehicle for targeting genes to improve production traits. For example, green fluorescent protein is highly expressed in the beaks, legs, and feet of second-generation transgenic chicks. [This figure of transgenic chicks expressing green fluorescent protein is reprinted by permission from Macmillan Publishers Ltd. (EMBO Reports; McGrew et al., 2004)]. EST = expressed sequence tags; TIGR = The Institute for Genomic Research; NCBI/EBI = National Center for Biotechnology Information/European Bioinformatics Institute; SAGE = serial analysis of gene expression; MPSS = massively parallel signature sequencing; SOMs = self-organizing maps; MALDI-TOF/MS = matrix-assisted laser desorption/ionization time-of-flight/mass spectrometry; LC/MS = liquid chromatography/mass spectrometry; GC/MS = gas chromatography/mass spectrometry; KEGG = Kyoto Encyclopedia of Genes and Genomes.

	DEVELOPMENT OF GENOMIC RESOURCES

TOP ABSTRACT INTRODUCTION DEVELOPMENT OF GENOMIC RESOURCES HIGH-THROUGHPUT GENOME-WIDE... BIOINFORMATICS AND COMPUTATIONAL... CONCLUSIONS AND OUTLOOK REFERENCES

 
Assembly of a Comprehensive Catalog of Expressed Genes
The most efficient method for gene discovery, before completion of a genome sequence, is high-throughput sequencing of cDNA (or EST) from normalized cDNA libraries (Soares et al., 1994; Bonaldo et al., 1996). Prior to 1998, only a few hundred chicken EST were present in the dbEST division of GenBank. Figure 2 shows the progression of EST discovery in the chicken during the last decade. The most prolific growth of chicken EST sequences in GenBank occurred during a 4-yr period (1999 to 2002) due to completion of several international EST sequencing efforts (for reviews, see Cogburn et al., 2003c, 2004; Carré et al., 2006). The largest number of chicken EST submitted to GenBank in a single year was 359,674 sequences in 2002 (Figure 2A). The first large-scale chicken EST sequences were derived from primary lymphoid tissues, where 7,409 EST from B-lymphocytes (Abdrakhmanov et al., 2000) and 2,770 unique EST (out of 5,251 clones sequenced) from concanavalin A-activated T cells (Tirunagaru et al., 2000) were entered into Gen-Bank. A large number of EST (35,407) sequenced from single and multitissue cDNA libraries (representing the immune, metabolic-somatic, neuroendocrine, and reproductive systems) were entered into GenBank between 2001 and 2004 (Cogburn et al., 2003c, 2004; Carré et al., 2006). The largest collection of 330,096 chicken EST was sequenced from 22 normalized cDNA libraries constructed from major organs and cell types of egg- and meat-type chickens and embryos across a wide range of developmental stages (Boardman et al., 2002). An additional 4,998 chicken EST were derived from unnormalized cDNA libraries constructed from whole embryos (stage 26), somites, and limb buds (Jorge et al., 2004). Another project yielded 21,285 EST from brain and testis cDNA libraries made from Red Junglefowl and White Leghorn chickens (Savolainen et al., 2005). An additional 13,132 EST came from native Korean chicken testis cDNA libraries (Shin et al., 2005). A set of 14,409 EST was obtained from a cDNA library constructed from intestinal tissue of Eimeria-infected chickens (Min et al., 2005). Another group sequenced 8,729 EST from lipopolysaccharide and Escherichia coli-stimulated peripheral blood lymphocytes for construction of an avian macrophage microarray (Bliss et al., 2005). Recently, 10,848 EST were sequenced from a cDNA library constructed from chicken primordial germ cells isolated from White Leghorn embryos at d 6.5 (Han et al., 2006). Collectively, nearly 600,000 chicken EST sequences have been deposited in public databases for computational identification of expressed chicken genes.

View larger version (17K): [in this window] [in a new window]   Figure 2. The progression of gene discovery in the chicken during the last decade. This graph shows the number of chicken expressed sequence tags (EST) entered per year (A) and the total accumulation of chicken EST (B) in the dbEST division of GenBank (National Center for Biotechnology Information).

Several reviews have described the availability of chicken genomics resources, including EST collections and microarrays (Burt, 2004; Cogburn et al., 2004; Antin and Konieczka, 2005; Fadiel et al., 2005). Initial overviews of functional genomics in the chicken have been published (Brown et al., 2003; Cogburn et al., 2003b,c; Burt, 2005; Moore et al., 2005). A list of functional genomics resources for the chicken and links to useful Web sites are provided in Table 1.

Functional Annotation of the Genome Sequence
The next critical step after genome sequencing is the rigorous functional annotation of the genome sequence. Assignment of the major functional units (genes) to the chicken genome sequence was enabled by the comprehensive catalog of chicken EST from chicken tissues (Abdrakhmanov et al., 2000; Tirunagaru et al., 2000; Boardman et al., 2002; Carré et al., 2006) and by the finished or complete cDNA sequencing of most chicken genes (Hubbard et al., 2005). The most common functional annotation of a genome uses the unified Gene Ontology (GO; http://www.geneontology.org/) assignment of gene and protein function to 3 broad categories: cellular component, molecular function, and biological process (Ashburner et al., 2000). Currently, the GO Annotation (GOA) database at the European Bioinformatics Institute (http://www.ebi.ac.uk/GOA/) contains 33,796 distinct human proteins, whereas only 16,146 distinct proteins have been identified in the chicken. Both transcriptomic and proteomic data are excellent sources for structural and functional annotation of chicken genes. At present, the majority of "known" genes in the chicken genome are annotated electronically by sequence homology or by ab initio gene prediction algorithms (Eyras et al., 2005). Experimental evidence of expression from transcriptomic and proteomic studies allows rapid verification of predicted gene transcripts and proteins to support functional annotation of the chicken genome sequence.

	HIGH-THROUGHPUT GENOME-WIDE SCREENING

TOP ABSTRACT INTRODUCTION DEVELOPMENT OF GENOMIC RESOURCES HIGH-THROUGHPUT GENOME-WIDE... BIOINFORMATICS AND COMPUTATIONAL... CONCLUSIONS AND OUTLOOK REFERENCES

 
Gene Expression Profiling with Microarrays
Through the use of cDNA microarrays, investigators can measure mRNA levels for thousands of genes simultaneously, rather than one gene at a time. This high-throughput approach has been widely adopted in biological research. Various chicken tissues have been used for large-scale transcriptional analysis during normal development or after specific perturbations. This section reviews these studies to illustrate the power of DNA microarrays for transcriptional profiling and the discovery of functional genes in the chicken.

Neuroendocrine and Reproductive Systems.
The pituitary gland and hypothalamus of the brain constitute the central components of the neuroendocrine system. This system plays a dominant role in controlling growth, metabolism, and reproduction. Nutrient availability and peripheral neuroendocrine signals from peripheral receptors and glands are integrated in the central nervous system; reciprocally, the hypothalamus communicates signals by regulating the release of hypothalamic-releasing hormones and release-inhibiting hormones, which reach the anterior pituitary gland via hypophyseal portal circulation. These hypothalamic-releasing and release-inhibiting hormones control secretion of trophic hormones from the anterior pituitary gland. In addition, magnocellular neurons originating in the hypothalamus and terminating in the posterior pituitary gland release stored neurohormones directly into systemic circulation. The hormones secreted from the anterior and posterior pituitary glands include those regulating growth and metabolism [growth hormone (GH) and thyroid-stimulating hormone (TSH), reproduction [luteinizing hormone (LH), follicle-stimulating hormone (FSH), and prolactin (PRL)], stress responses (adrenocorticotropic hormone), and renal function (Arg vasotocin). The hypothalamus also plays a critical role in controlling feed intake via the function of several neuropeptides. Stimulation of feed intake involves neuropeptide Y and agouti-related protein, whereas inhibition of intake involves -melanocyte-stimulating hormone, cocaine- and amphetamine-regulated transcript, and corticotrophin-releasing hormone.

Traditionally, regulation of gene expression within the neuroendocrine system has been studied one gene or several select genes at a time. This focused approach has been very effective in defining multiple integrated pathways involved in neuroendocrine regulation of feed intake, metabolism, and somatic growth (Richards and Proszkowiec-Weglarz, 2007). For example, anterior pituitary levels of mRNA for PRL, GH, TSH, and LH in poultry have been analyzed by Northern blotting, ribonuclease protection assays, PCR, and in situ hybridization (Talbot et al., 1991; Kansaku et al., 1994; Tong et al., 1997; Ramesh et al., 1998; Bossis and Porter, 2000; Fu and Porter, 2004; Muchow et al., 2005). Similar studies have evaluated expression of mRNA for hypothalamic neuropeptides, including neuropeptide Y, vasoactive intestinal polypeptide, corticotrophin-releasing hormone, and gonadotropin-releasing hormone in poultry (Talbot et al., 1995; Boswell et al., 1999; Sun et al., 2001; Chaiseha et al., 2004; Saito et al., 2005; Vandenborne et al., 2005). The development of chicken cDNA microarrays has enabled analysis of gene expression profiles for thousands of genes simultaneously in individual samples of the pituitary gland or hypothalamus.

Microarray analysis of RNA from small tissue samples (i.e., individual pituitary glands) is problematic because less than 25 µg of total RNA (needed for a standard microarray analysis) can be recovered from small tissue samples. To overcome this obstacle, RNA amplification protocols have been developed for use with microarrays. Ribonucleic acid amplification procedures, originally published by Eberwine and colleagues (Phillips and Eberwine, 1996), have recently been adapted for use with chicken pituitary samples (Porter and Ellestad, 2005; Ellestad et al., 2006). These procedures transcribe RNA in vitro by using T7 RNA polymerase and cDNA samples produced with an oligo(dT) primer containing the T7 promoter. This procedure typically yields 5 to 10 µg of amplified RNA (equivalent to mRNA) from 500 ng of starting total RNA, an approximately 300-fold amplification. Performing a second round of amplification allows for analysis of RNA from single cells collected by laser capture microscopy. Readers interested in microarray analysis of small samples, (i.e., pituitary glands or cultured cells) are directed to previous reports for detailed descriptions of this procedure (Phillips and Eberwine, 1996; Porter and Ellestad, 2005; Ellestad et al., 2006).

In the first study to examine global gene expression patterns in the chicken neuroendocrine system, Cassone and colleagues (Bailey et al., 2003) developed a cDNA microarray specific to another component of the neuroendocrine system, the pineal gland. This gland secretes melatonin (MT) and functions in synchronizing daily rhythms of activity and reproductive timing in birds. In that study, RNA samples were analyzed from pineal glands of animals exposed to light-dark cycles or to total darkness. A number of genes were identified whose mRNA levels fluctuated in a rhythmic pattern, corresponding to the prevailing light-dark cycle. These included genes involved in MT synthesis and orthologs of mammalian clock genes, as expected. Genes involved in other processes (i.e., transduction of light, immune and endocrine signaling) were also found to fluctuate rhythmically. In a second study, this group extended their transcriptional analyses to the retina of chicks exposed to light-dark cycles or constant darkness (Bailey et al., 2004). Again, expression of chicken orthologs of mammalian clock genes and genes involved in MT synthesis fluctuated with the prevailing photoperiod. More important, application of cDNA microarray technology to this system allowed identification of a number of novel candidate genes with rhythmic expression in both the pineal gland and the retina. Furthermore, several genes involved in intermediary metabolism and protein degradation exhibited rhythmicity, pointing out the extent and complexity of such coordination. None of these genes had previously been implicated in the regulation of daily rhythms.

A second and more extensive cDNA microarray for the chicken neuroendocrine system was developed with clones sequenced from a cDNA library constructed with RNA pooled from the hypothalamus, pituitary gland, and pineal gland. The EST sequencing from this library and development of the Chicken Neuroendocrine System 5K microarray (GEO Accession No. GPL1744 [NCBI GEO] ) was described earlier (Cogburn et al., 2003c, 2004; Porter and Ellestad, 2005; Ellestad et al., 2006). Porter and colleagues used the Chicken Neuroendocrine System 5K microarrays to examine the ontogeny of hypothalamic gene expression during the perihatch period. Ribonucleic acid was isolated from hypothalami before [embryonic day (e)17 and e19] and after hatching [posthatch day (d)1 and d3] and analyzed with the microarrays. Expression levels of 105 genes changed substantially during this period of development. Transcription profiles for myelin basic protein (MBP), stathmin (STMN1), dopamine, and cAMP-regulated neuronal phosphoprotein (DARPP), 2',3'-cyclic-nucleotide 3'-phosphodiesterase (CNP2), receptor-type protein Tyr-protein phosphatase N2 precursor (PTPN2), bone morphogenic protein 7 (BMP7), and glyceraldehyde phospho-dehydrogenase (GAPDH) were confirmed by quantitative real-time PCR (qRT-PCR). The STMN1 gene was overexpressed in undifferentiated neurons, and hypothalamic STMN1 levels decreased from e17 to d3. In contrast, MBP levels increased from e17 to d3, which agrees with the observation that myelination of the central nervous system occurs primarily after hatch. In addition to these predicted changes, the abundance of DARPP and CNP2 increased dramatically between d1 and d3, indicating a substantial increase in neuronal signaling within the hypothalamus after hatching. In addition, levels of PTPN2 and BMP7 transcripts decreased dramatically but transiently on e19 and d1, respectively, indicating changes in signaling events within the hypothalamus specific to the perihatch period.

The 5K Chicken Neuroendocrine System microarray was recently used to profile developmental changes in gene expression in the pituitary gland from e10 to e17 (Ellestad et al., 2006). This period of embryonic development is characterized by the differentiation of 3 distinct anterior pituitary cell types that produce specific trophic hormones (TSH, GH, and PRL). In that study, 393 genes were differentially expressed during this period of embryonic development (e10 to e17). Self-organizing map (SOM) analysis (Tamayo et al., 1999) enabled clustering of these differentially expressed genes based on their transcriptional profiles during development. The TSHß mRNA levels increased steadily during embryonic development, whereas PRL expression was nearly absent through e14, and then increased dramatically on e17. In contrast, expression of ß-actin decreased in the pituitary during embryonic development, whereas the abundance of GH was low on e10 and e12, but increased dramatically by e17. A previous report indicated that glucocorticoid treatment increased GH mRNA indirectly through induction of another unidentified gene in the chicken (Bossis and Porter, 2003). Interestingly, 2 genes that respond to glucocorticoid treatment in mammals, the glucocorticoid-induced Leu zipper (GILZ) and dexamethasone-induced Ras 1 (DEXRAS1), exhibit expression profiles that are similar to GH. The expression profiles of 33 differentially expressed genes identified by microarray analysis were confirmed by using an independent method, qRT-PCR. These findings demonstrate that microarray analysis can be performed on amplified RNA from individual pituitary glands from chicken embryos as early as e10. Moreover, a number of unique genes were identified that could play a role in regulating the differentiation of anterior pituitary cells.

Microarray technology has also been used to study global gene responses in cultured pituitary cells. In the first study of this type, Porter and colleagues aimed to identify genes directly and rapidly regulated by the adrenal glucocorticoid corticosterone (CS) within the embryonic pituitary gland. Pituitary cells from e11 embryos were treated with CS (for 1.5, 3, 6, 12, or 24 h) in the absence or presence of cycloheximide, a protein synthesis inhibitor. Amplified pituitary RNA was then analyzed with the Del-Mar 14K Integrated Systems microarrays. Expression of 27 genes was affected by CS at 1.5 or 3 h both in the absence and in the presence of cycloheximide; 13 of these genes were induced at least 2-fold by CS within 3 h. None of these direct targets of CS had previously been demonstrated in the anterior pituitary gland.

Transcriptional profiling with cDNA microarrays has also been used to identify differentially expressed genes in the neuroendocrine system in a set of divergently selected chickens. Hallböök and colleagues (Ka et al., 2005) compared hypothalamic gene expression at hatching between 2 chicken lines genetically selected for high or low BW at 8 wk of age (Dunnington and Siegel, 1996). This microarray analysis indicated that 41 genes, including endogenous avian leukosis virus (ALV), were differentially expressed between the 2 lines, although no details were provided.

Another functional genomics project (Cogburn et al., 2003c, 2005) has focused on a different population of broiler chickens genetically selected for either high-growth (HG) or low-growth (LG) BW (Ricard, 1975). The idea was that global transcript profiling in multiple tissues of divergent lines could identify genes that contribute to such large differences in production traits. Body weight in these experimental broiler lines diverges after 3 wk of age, with a greater than 2-fold difference at 11 wk. Gene expression profiles in the anterior pituitary gland were recently compared at 1, 3, 5, and 7 wk of age by using the Del-Mar 14K Integrated Systems microarrays (Porter et al., 2007). The microarray analysis identified 263 genes that were differentially expressed in the pituitary gland between the HG and LG lines in at least one age. These included 4 of the 6 trophic hormones produced by the anterior pituitary gland. Three pituitary hormones (TSHß, LHß, and FSHß) were more abundant in the HG line, whereas the fourth gene (GH) was expressed at higher levels in the LG line. The expression patterns of TSHß, LHß, GH, and 6 other genes were confirmed by qRT-PCR analysis.

An additional study (Porter et al., 2007) examined gene expression profiles in the anterior pituitary gland and hypothalamus of genetically selected fat (FL) and lean (LL) lines of chickens (Leclercq, 1988) during juvenile development. Anterior pituitary glands and hypothalami were collected before (1 and 3 wk) and after (5 to 11 wk) the divergence in weight of the abdominal fat pad. The Del-Mar 14K Chicken Integrated Systems microarrays were used for transcriptional profiling of these samples. Interestingly, differences in gene expression profiles were found in both the anterior pituitary and the hypothalamus between FL and LL chickens at 1 and 3 wk. This indicates early divergence in the expression of genes in the neuroendocrine system between the FL and LL chickens. This system, which regulates feed intake, growth, and metabolism, could be programmed differently between these genetic lines, resulting in large (2- to 3-fold) differences in accumulation of body fat. For the pituitary gland, microarray analysis identified 386 differentially expressed genes between the FL and LL birds by using stringent criteria. For the hypothalamus, 206 genes were differentially expressed between the FL and LL. Several of the differentially expressed genes identified by microarray analysis in the anterior pituitary gland were confirmed by qRT-PCR; one of these genes was a member of the aldo-keto reductase family (AKR), which catalyzes the reduction of the aldehydes to ketones. Expression of AKR was greater in the pituitaries of LL birds compared with FL birds.

Thus, microarray analysis has enabled examination of the ontogeny of gene expression in the chicken neuroendocrine system during late embryonic and early post-hatch development and the effects of drug or hormonal treatments on global gene expression in cultured pituitary cells. Rhythmic patterns of gene expression have been described in the pineal gland and retina of birds exposed to dark-light cycles. Finally, numerous differentially expressed genes have been discovered in the neuroendocrine system of chickens divergently selected for body composition (high vs. low body fat) or growth rate (high vs. low BW). Gene expression profiling with cDNA microarrays has already identified numerous candidate genes in the neuroendocrine system that could control the growth and metabolism of the chicken, and most likely other vertebrates.

Metabolic and Somatic Systems.
To date, there have been only a few reports of gene expression profiling in metabolic (liver, fat, muscle) and somatic (skeletal muscle and bone growth plate) tissue of the chicken. Most studies have focused on the liver because this organ regulates whole-body metabolism of major nutrients (i.e., glucose, amino acids, and lipids). This is particularly evident in avian species, in which the liver is the main site of de novo lipogenesis (Goodridge and Ball, 1967). Several different models have been used for initial transcriptional profiling in metabolic and somatic tissues of the chicken and analysis of gene networks (Cogburn et al., 2003b,c, 2004; Wang et al., 2007). These experimental models include divergent selection (FL vs. LL and HG vs. LG; Cogburn et al., 2003c), metabolic perturbation [the fasting and refeeding response (Duclos et al., 2004), the abrupt embryo-to-hatchling transition (Glass et al., 2002)], and hormonal perturbation (Wang et al., 2007). Some observations from these original gene expression studies are presented below.

The first microarray analysis of chicken liver used a nylon membrane-based array of 1,200 (1.2K) cDNA derived from activated T cells (Morgan et al., 2001) to examine developmental differences (3 to 9 wk) between broiler lines divergently selected for either HG or LG (Cogburn et al., 2003b). Hierarchical clustering with SOM analysis (Tamayo et al., 1999) identified 59 differentially expressed genes in the liver of HG birds that belonged to 4 distinct clusters, and 6 distinct clusters containing 76 genes in the LG. Thyroid hormone-responsive Spot 14 (THRSP) and superoxide dismutase 3 (SOD3) were among the first differentially expressed genes discovered in the liver of HG and LG chickens (Cogburn et al., 2003b) with this early chicken cDNA microarray.

A prototype 3.2K liver-specific microarray (GEO Accession No. GPL1742 [NCBI GEO] ) was developed and first used to examine hepatic gene expression during the abrupt embryo-to-hatchling transition period (Cogburn et al., 2004). Gene cluster analysis, using a spanning tree clustering method (Rejto and Tusnady, 2006), revealed 756 differentially expressed genes that formed 32 distinct expression patterns. These clusters of coexpressed genes are involved in the metabolic switch from embryonic to terrestrial life in the perihatch chick. For example, one group of 49 genes [Cluster (C) 6, C25, and C32] showed higher levels of expression in embryos, whereas 3 other clusters (C15, C22, and C29) showed higher expression after hatching (Figure 3). Expression of genes in other clusters increased sharply just before (C28) or just after (C14) hatching, whereas the genes in cluster C31 progressively declined in the late embryo (e16 to e20) and newly hatched (d1) chick and then sharply increased thereafter. Several genes, expressed at higher levels in embryos, are directly involved in the fat catabolism, transcription, or signal transduction pathways. In contrast, the transcriptional pattern of several other clusters of hepatic genes increases sharply after hatching. These gene clusters encode numerous metabolic enzymes, transcription factors, inflammatory factors, transporters, and signaling proteins. Many of the up-regulated genes in the newly hatched chick reflect enhanced lipogenesis [THRSP, fatty acid synthase (FASN), stearoyl-coenzyme A (CoA) desaturase 1 or ⁹-desaturase (SCD1), cytosolic malic enzyme 1 (ME1), adipose differentiation-related protein (ADFP), and fatty acid-binding protein 1 (FABP1)] after the initial ingestion of carbohydrate- and protein-enriched feed. Furthermore, qRT-PCR analysis has confirmed similar patterns of gene expression first revealed by microarray analysis. Some of these functional genes have been integrated into a working model of transcriptional control of the citric acid and fat biosynthesis pathways in the liver of the chicken (see Figure 4 in Cogburn et al., 2004). Thus, microarray analysis and clustering of coexpressed genes have provided the first global view of the transcriptional control over metabolism during the abrupt embryo-to-hatchling transition.

View larger version (33K): [in this window] [in a new window]   Figure 3. Clusters of coregulated hepatic genes during the embryo-to-hatchling transition with distinct expression patterns. A spanning tree clustering method (Rejto and Tusnady, 2006) revealed 32 distinct expression patterns from 756 differentially expressed genes. Transcriptional profiles were determined with individual liver-specific 3.2K microarrays (GEO Platform No. GPL1742) for 4 embryos at embryonic day (e)16, e18, and e20 and 4 chicks at posthatch day (d)1, d3, and d9. The dotted vertical line indicates hatching, and the horizontal line in each graph denotes the mean expression level (centroid) of that cluster. The number of genes within the cluster is shown, and some examples of cluster members are given. For inclusion in a cluster, a gene must be differentially expressed at one or more ages.

View larger version (19K): [in this window] [in a new window]   Figure 4. Expression of peroxisome proliferator-activated receptor (PPAR) genes in liver during the embryo-to-hatchling transition (A) and the fasting-refeeding response (B). The abundance of PPAR, PPARß, and PPAR transcripts were determined by quantitative reverse transcription-PCR (qRT-PCR) analysis by using gene-specific primers and SYBR Green chemistry. Gene expression levels, calculated by the delta-delta threshold crossing point method, are presented in arbitrary units (AU). Each bar represents the average of 4 birds. For the embryo-to-hatchling transition study, liver samples were collected from embryos at embryonic day (e)16, e18, and e20 and hatchlings at posthatch day (d)1, d3, and d9. For the fasting and refeeding experiment (B), 6-wk-old high-growth (HG) or low-growth (LG) cockerels were assigned to 5 different nutritional states: fully fed (fed), fasted for 16 h (F16), fasted for 48 h (F48), refed for 4 h (R4), or refed for 24 h (R24). Four birds from each genotype were killed under each nutritional state for collection of liver samples. Total RNA was isolated using Qiagen Midi-RNeasy kits for microarray and qRT-PCR analyses.

The influence of nutritional state (fasting vs. refeeding) on the liver transcriptome was examined in the divergently selected HG and LG chickens (Ricard, 1975) by using the 3.2K liver-specific arrays (Duclos et al., 2004). In total, 429 differentially expressed genes that form 21 unique gene clusters were identified in the HG genotype, compared with 346 differentially expressed genes that form 16 clusters in the LG genotype. A number of functional gene clusters were down-regulated with fasting, whereas after refeeding, the expression of these hepatic genes was increased. One cluster contained FASN, which was depressed by prolonged fasting and sharply up-regulated during refeeding. The transcriptional response of FASN to abrupt changes in feed intake was confirmed by independent qRT-PCR analysis and is consistent with an earlier report on transcriptional control of FASN (Back et al., 1986). In contrast, some genes followed an opposite pattern, with an increase during starvation and a decrease during refeeding; L-lactate dehydrogenase-ß (LDHß) illustrates this pattern of feeding-induced repression. Adipophilin or ADRP is a marker of fat accumulation that belongs to a gene cluster induced by refeeding. The peroxisome proliferator-activated receptors (PPAR) belong to a family of ligand-activated transcription factors that control key metabolic pathways (i.e., adipogenesis, fat metabolism, and insulin signaling; Lee et al., 2003a). Transcriptional profiling in the liver of chickens clearly shows that the PPAR respond to abrupt changes in metabolism during the embryo-hatchling transition and the fasting-re-feeding response (Figure 4). In the perihatch chick, expression of PPARß and PPAR transcripts was highest in the liver after hatching, whereas PPAR levels were lowest in day-old hatchlings (Figure 4A). Hepatic PPAR mRNA levels increased sharply in fasting chickens, which reflects an increase in catabolism of stored fat (Figure 4B). In contrast, hepatic expression of PPAR increases after hatching and refeeding, which suggests that this transcription factor supports lipogenesis in the chicken. When compared with the HG chickens, hepatic expression of PPAR was lower in the LG line, except at 24 h after refeeding. The abundance of PPARß declined in 6-wk-old HG and LG chickens during prolonged fasting but returned to normal following refeeding. The PPAR play a central role in transcriptional control of energy balance via their activation by lipid ligands, subsequent interaction with coactivators or corepressors, and binding of these heterodimers to PPAR response elements in the promoter of numerous metabolic enzymes (Feige and Auwerx, 2007).

A large number of hepatic genes showed differences in mRNA levels between the 2 genotypes (HG vs. LG) in at least one of the metabolic states. Most of the differences between genotypes were apparent in the fed (44 genes) or refed state at 24 h (308 genes). However, the genes that showed a consistent difference between the 2 genotypes were less numerous than those that responded to nutritional state. In the HG genotype, only 3 genes [trans-ketolase (TKT), methionyl-tRNA formyltranferase (MTFMT), and aminopeptidase (ANPEP)] were consistently expressed at higher levels, whereas 2 genes [XAP-5 (XAP5) and ribosomal protein S27 (RPS27)] were consistently expressed at lower levels. Hepatic insulin-like growth factor-1 (IGF-I) mRNA levels were higher in HG than in LG, as previously reported (Beccavin et al., 2001). Three genes involved in the control of lipid metabolism [ADRP, glutathione S-transferase A1 (GSTA1), and PPAR] were higher in the HG than in the LG genotype in at least one nutritional condition. This is consistent with the higher percentage of abdominal fat observed in the HG compared with the LG chickens. In particular, the relative abdominal fat weight (percentage of BW) of HG birds was about 6-fold greater than that in the LG at 6 wk.

In a more extensive study, hepatic gene expression profiles were examined in the HG and LG chickens during juvenile development (1 to 11 wk) with the Del-Mar 14K Integrated Systems cDNA microarray (Cogburn et al., 2004). Surprisingly, the most highly expressed gene in the liver of the LG line, relative to the HG line, across all ages was an endogenous retrovirus related to the ALV envelope protein (PR57). This time-course study revealed 532 hepatic genes that showed a significant interaction between genotype and age (Table 2). The largest number of differentially expressed genes was found at 7 wk of age, when 199 hepatic genes were up-regulated in the HG line and 176 genes were up-regulated in the LG line. A large number of up-regulated genes found in the liver of HG birds at 7 wk are involved in the synthesis, transport, and metabolism of lipids, which supports their phenotype of higher abdominal fat content (Table 2). The differential expression (higher in HG birds) of an original candidate gene, THRSP, was confirmed by qRT-PCR analysis. Thyroid hormone-responsive Spot 14- is an important transcription factor that controls expression of several lipogenic genes (Towle et al., 1997). Furthermore, Cogburn and colleagues have identified insertion-deletion polymorphisms in chicken THRSP that are associated with QTL for abdominal fatness on chicken chromosome 1 (GGA1; Wang et al., 2004). The "leaner" LG line shows higher expression of several genes involved in energy metabolism, signal transduction, and hematopoiesis.

Some differentially expressed genes in breast muscle are involved in muscle hypertrophy [integrin-ß1 (ITGB1) and glycogen synthase kinase-3ß (GSK3ß)]. Other genes associated with metabolic pathways involved in muscle growth include protein phosphatase 2A- (PPP2R1A), TRAF4-associated factor 2 (TRAF4AF2), and prohibitin (PHB). Additional differentially expressed genes include regulators of protein catabolism such as cullin 2 (CUL2) and ubiquitin-conjugating enzyme E2D 3 isoform 1 (UBE2D3). Another differentially expressed gene, annexin-V (ANXA5), is a marker of muscle growth and is regulated by a selection for growth potential in cattle (Sudre et al., 2003). Additionally, several unknown genes are differentially expressed in breast muscle between the HG and LG lines.

The FL and LL chickens, introduced earlier, represent unique models available to identify genes in metabolic pathways that contribute to excessive fatness or leanness. In agreement with the fat phenotype, genes coding for several lipogenic enzymes [adenosine triphosphate citrate lyase (ACLY), acetyl-CoA carboxylase (ACC), FASN, ME1, and SCD1] were also found to be more abundant in the liver of FL chickens (Assaf et al., 2004). Surprisingly, sterol response element-binding protein 1 (SREBP1), a transcription factor that governs the expression of these lipogenic enzymes, was expressed at similar levels in the liver of FL and LL (Assaf et al., 2003).

Gene expression in the liver of FL and LL chickens was recently examined with a low-density "focused" microarray of 323 cDNA (Bourneuf et al., 2006). The spotted chicken cDNA represent genes involved in or related to carbohydrate and lipid metabolism, including some signaling and transcription factors and 195 cDNA previously identified by differential mRNA display analysis in the liver of FL and LL birds (Carré et al., 2001, 2002). Hepatic expression of several enzymes involved in lipogenesis [ACC, FASN, SCD1, apolipoprotein A1 (APOA1), SREBP1, and mitochondrial malate dehydrogenase 2 (MDH2)] were overexpressed in the genetically fat chickens. In contrast, 10 genes were down-regulated in the liver of FL chickens [peroxisomal 2,4-dienoyl CoA reductase (DCER2), activating transcription factor 4 (ATF4), cyclic adenosine monophosphate-response element-binding protein 2 (C/EBP2), pyruvate carboxylase (PC), -amylase (AMY1A), cytochrome B (CYTB), ras homolog gene family member F (RHOF), Ran GTPase-activating protein 1 (RANGAP1)], as were other transcripts that belong to the cytochrome P450 family. The differential expression between the FL and LL lines of most genes identified by the low-density arrays was confirmed by qRT-PCR analysis.

Recently, the developmental profiles (1 to 11 wk) of hepatic gene expression in the FL and LL chickens were examined with the high-density Del-Mar 14K Integrated Systems microarray. Across 6 ages, 1,805 hepatic genes were differentially expressed in the FL and LL chickens (Table 2). Similar to the HG and LG lines, most of the differentially expressed hepatic genes were found at 7 wk of age, well after the phenotypic differences are established. Sixteen genes involved in mitogen-activated protein kinase signaling were differentially expressed between the FL and LL at 7 wk, when 12 genes were up-regulated in the liver of FL birds and only 4 genes were expressed at higher levels in the LL. Gene network analysis with Pathway Miner software (Pandey et al., 2004) showed that 9 genes in the wingless signaling pathway and 7 genes in the phosphatidyl inositol signaling pathway were differentially expressed in the liver of FL and LL birds. This gene association network of cellular and regulatory pathways showed a high representation of genes in several important signaling systems [mitogen-activated protein kinase (16 members), inositol trisphosphate (7 members), wingless (9 members), transforming growth factor-ß (9 members), and Toll-like receptor (TLR; 6 members)], in addition to apoptosis and integrin-mediated cell adhesion pathways (see the gene network in Figure 1).

An analysis of adipose tissue from an egg-laying breed and a "fat" grandsire broiler breed at 10 wk of age with a 9K chicken cDNA microarray has identified 67 differentially expressed transcripts, although only 42 EST correspond to known genes (Wang et al., 2006). This undefined chicken 9K cDNA microarray was obtained from the Beijing Genomics Institute. Surprisingly, only 3 of the 42 differentially expressed genes are directly related to lipid metabolism [APOA1, lipoprotein lipase (LPL), and leptin receptor gene-related protein (LEPR-GRP)]. Nonetheless, this is the first paper published on a microarray analysis of abdominal fat tissue in the chicken. Perhaps transcriptional profiling of adipose tissue in the FL and LL chickens during juvenile development would afford a higher resolution of the genes and metabolic pathways controlling excessive fattening in the broiler chicken.

Thus, divergently selected chickens represent a unique model for identification of genes that control important production traits. As a whole, transcriptome studies should provide a better understanding of the genes and their regulatory networks that control growth, tissue development, and ultimately body composition. The intensive genetic selection (almost exclusively for growth) applied to broiler chickens for many years could have isolated a relatively limited number of general mechanisms or pathways. In contrast, 176 cases of obesity in humans result from single mutations in only 11 genes. In genome-wide scans, the number of QTL for obesity-related phenotypes in humans continues to increase. In the 2006 Obesity Map update, 253 QTL involved in the development of obesity in humans were identified from 61 scans (Rankinen et al., 2006). It is of particular interest that these putative obesity loci are found on all human chromosomes except chromosome Y (HSAY).

Immune System.
Neiman and collaborators (2001) were the first to develop a chicken immune system array. This immune array, containing 2,200 elements, was used to analyze myc-oncogene-induced lymphomagenesis in the chicken bursa of Fabricius. Genes whose expression levels correlated with myc expression in transformed follicles and metastatic tumors were identified, including genes involved in nucleolar function, ribosome biogenesis, and protein synthesis. Subsequently, this immune array was expanded to 3,451 cDNA and used to compare the transcriptional signature of chick bursal lymphomas resulting from ALV insertional mutation of c-myb vs. transformation by v-Rel (Neiman et al., 2003). These arrays were also used to identify genes regulated by the v-jun oncogene in chick embryo fibroblasts (Black et al., 2004), and a pattern of expression was observed that is strikingly similar to the one produced by the Marek’s disease virus (MDV) meq oncogene (Levy et al., 2005). Another immune system array was used to study the host response to infection with MDV (Morgan et al., 2001) and herpes virus of turkeys infection (Karaca et al., 2004), and to catalog gene expression in the developing chick thymus (Cui et al., 2004). As expected, many of the genes identified in the viral studies responded to interferon. A number of differences worthy of additional study were also detected; these could contribute to the pathology of MDV or the vaccination response to herpes virus of turkeys. Collectively, these studies have established common mechanisms for the transformation of chicken cells and point to differences that are characteristic of individual pathogens.

In a follow-up study on c-myc-induced tumorigenesis, Neiman and colleagues used the 13K chicken cDNA microarray (Burnside et al., 2005), which is enriched for chicken immune system EST, for comparative genomic hybridization (Neiman et al., 2006). Gene amplification and chromosomal instability were detected in myc-transformed bursal follicles and lymphomas and were mapped by using the arrays. These data established the relationship between a copy number change and RNA expression patterns. The study showed that cDNA microarrays are useful for determining both gene expression and gene copy number.

An avian macrophage-specific array containing nearly 5,000 genes expressed in peripheral blood lymphocytes has been used to examine the transcriptional response of macrophages to gram-negative bacteria in comparison with the response to lipopolysaccharide (Bliss et al., 2005). Bacteria elicit a more complex response, and there is common signaling through TLR 4, although additional pathways are activated by whole bacteria. This avian innate immune microarray has been used in experiments with several avian cell or tissue types as well as a variety of pathogens (Bliss et al., 2005; Dalloul et al., 2007). Changes in expression of innate immune genes have been observed in vivo (with intestinal epithelial and splenic tissues) and in vitro [with peripheral blood monocytes, heterophils, nonadherent blood lymphocytes, and avian macrophage cell lines (HD11, HTC)]. In addition, innate immune responses have been elucidated after stimulation with bacteria (Salmonella, E. coli, and Mycoplasma), viruses (avian influenza), intestinal parasites (Eimeria), bacterial components (lipopolysaccharide), and immune modulators (interferon-). Another avian immune system array has been used to evaluate the host response to different respiratory pathogens (Munir and Kapur, 2003; Dar et al., 2005). As would be expected, infection of chickens with respiratory viruses leads to a marked increase in expression of genes related to interferon activation and inflammatory and protein trafficking. A microarray analysis of chicken intestinal lymphocyte genes induced or repressed in response to infection with Eimeria has also been reported (Min et al., 2003). Infection of chickens with Eimeria parasites stimulates transcription of interferon-, interleukin-15, and several cytokines in intestinal intraepithelial lymphocytes. The transcriptional responses of chickens to challenge with 2 important enteric pathogens (Eimeria and Salmonella) have been reviewed in detail (Lillehoj et al., 2007).

Microarrays have a potential use in identifying candidate genes for desired traits. Liu et al. (2001a) used microarrays to identify differentially expressed genes in MDV-resistant lines of birds, and they successfully integrated gene expression with genetic mapping data to identify functional candidate genes for disease resistance (see the expression QTL or "genetical genomics" section below). In another application of microarray technology, Degen et al. (2003) used microarrays to identify host-derived natural adjuvants. Enhancement of the immune response is a major issue in vaccine development, and the use of natural adjuvants is more desirable than commonly used chemical adjuvants. Global gene expression profiling can be used not only to identify candidate genes, but also to evaluate their effectiveness.

An immune system array has been used to study the response to infection with infectious bursal disease virus and also to identify differences in gene expression between resistant and susceptible lines (Ruby et al., 2006). Genes involved in the inflammatory response were induced in both lines; however, differences between the 2 lines were observed and a model for resistance was established in which a more rapid and robust inflammatory response serves to limit infection and pathology. Using microarrays, van Hemert et al. (2006) found differences in gene expression in Salmonella-resistant and Salmonella-susceptible lines as well. In response to Salmonella infection, the resistant, fast-growing chicken broiler line induced genes that affected T-cell activation, whereas in the more susceptible, slow-growing broiler line, genes involved in macrophage activation seemed to be more affected at d 1 postinfection. These studies point to the value of microarrays in identifying genes associated with disease-resistance phenotypes. Once verified, the candidate gene(s) could be the goal of marker-assisted selection.

A 5K chicken immuno-microarray developed by ARK Genomics (Roslin Institute) has been used to study the immune response to vaccination with avian influenza and provides insight into virus-host interaction (Degen et al., 2006). As expected, genes associated with a strong immune response were highly elevated in infected, naive birds compared with immunized birds. This study also identified genes affected by an immune adjuvant, demonstrating another utility of microarrays in evaluating adjuvant influence.

Cellular and Gene Networks
The advent of cDNA microarrays for global gene expression analysis (Schena et al., 1995, 1996) created the need to interpret vast data sets and to organize genes by their temporal expression patterns. Hierarchical clustering (Eisen et al., 1998) and SOM (Tamayo et al., 1999; Toronen et al., 1999) were developed to visualize and understand gene expression patterns. The assumption is that genes with a similar function cluster together, presumably due to common transcriptional regulation, and they usually belong to similar metabolic or regulatory pathways. The features and limitations of these early, unsupervised clustering methods have been reevaluated to provide more informed choices for potential users (Yin et al., 2006).

Perhaps the greatest challenge of functional genomics has been to extract useful information on genetic interactions from large data sets (van Someren et al., 2002). Detection of the genetic interactions that determine phenotype, which themselves are related to protein and metabolite interactions, requires a mixture of computational and experimental approaches (Carter, 2005). Several early papers (Wagner, 2001; Brazhnik et al., 2002; de la Fuente et al., 2002) introduced the concepts and mathematical methods for reconstructing gene networks from gene expression profiles. Some biologists believe that reconstructing gene networks from genetic perturbation experiments represents the "holy grail of functional genomics" (Wagner, 2001). The major requirement for gene network analysis is the systematic perturbation of each gene in a network or pathway to determine the interaction of each gene with other members (Brazhnik et al., 2002; de la Fuente et al., 2002). The strengths and direction (positive or negative) of gene interactions are determined by perturbing the rate of transcription, one gene at a time. Although gene-by-gene perturbations are easily accomplished in simple organisms (e.g., yeast), genetic interactions in higher organisms (e.g., birds and mammals) are more difficult to demonstrate. The perturbation approach for identifying gene networks involves the integration of computational models with experimental perturbations and global (or genome-wide) measurements of the responses of the biological system (Tegner and Bjorkegren, 2007). The architecture and dynamics of gene networks are best revealed when prior knowledge of the biological system is incorporated into the inference algorithm. Similarly, pathway analysis focuses on identifying a defined set of biochemical reactions (i.e., metabolism, apoptosis, or growth factor signaling; Klamt and Stelling, 2003; Papin et al., 2003). Metabolic pathway analyses reveal complex biochemical-reaction networks, which ultimately define biological systems. A recent review has described several popular bioinformatic methods and Web-based resources used for incorporating genome-wide transcriptional data into metabolic, cellular, and regulatory networks or pathways (Cavalieri and De Filippo, 2005).

Mapping of gene and regulatory networks requires high-throughput analysis of transcriptional scans, clustering of coregulated genes, and computational searches for functional motifs [i.e., cis-regulatory elements and transcription factor (TF) binding sites (TFBS); Banerjee and Zhang, 2002]. Multiple transcriptional snapshots taken in a time series provide a dynamic dimension of gene expression (Hoffman et al., 2003). Bayesian modeling seems best suited for reconstruction of gene networks from global expression data (van Someren et al., 2002). More recent gene network models integrate meta-analyses of gene expression profiles from different studies and multiple microarray platforms in humans and mice (Kyoon Choi et al., 2004; Hackl et al., 2005; Stahlberg et al., 2005; Stathopoulos and Levine, 2005; Estrada et al., 2006; Mulligan et al., 2006). Rigorous interrogation of the cis-regulatory regions and GO annotation of genes provides a more comprehensive view of genetic control over metabolic and developmental processes. Large-scale transcriptional profiling and refined bioinformatics analyses have revealed exquisite detail of the biological processes and molecular networks involved in transcriptional regulation of fat cell development (Hackl et al., 2005). A systems biology approach that integrates gene expression, QTL analysis, and modular gene network modeling in a segregating population affords the greatest power in detecting major genes controlling complex traits (Ghazalpour et al., 2006). This novel approach, called modular QTL analysis, combines expression QTL (eQTL) analysis and gene coexpression networks to identify key regulatory loci that control expression of phenotypic traits.

A powerful new computational approach for detecting network motifs in coexpressed gene networks uses graph theory to extract groups of highly interconnected transcripts (cliques) from genetic correlation matrices derived from high-throughput transcriptional scans (Baldwin et al., 2005; Chesler et al., 2005). Genes with similar expression patterns are clustered together, presumably under a common transcriptional control mechanism. Correlations are the most prevalent measure of coexpression and allow construction of a "graph" with genes forming vertexes connected with strength equal to the correlation coefficient. Cliques are completely intercorrelated groups of genes, an ideal definition for a cluster of potentially functionally related and commonly regulated genes (Voy et al., 2006). Tightly connected regions of the correlation graph represent subsets of genes with strong correlations among members, and thus are likely to represent biologically significant interactions. Another component of graph structure exploits the likelihood that coregulated genes share some common TFBS (Allocco et al., 2004). Genes linked by physical interactions in a network, such as TF-gene interactions, have strongly correlated expression levels (Ideker et al., 2001).

Only a few attempts have been made to apply gene network modeling in the chicken, although the importance of gene-gene interactions in determining phenotype has been clearly demonstrated (Carlborg et al., 2003, 2006). Two strong metabolic perturbations—the embryo-to-hatching transition (Glass et al., 2002) and the fasting and refeeding response (Duclos et al., 2004)—have been used to take time-series transcriptional snapshots of the chicken liver. A dynamic Bayesian model for analysis of microarray data and a spanning tree clustering method (Rejto and Tusnady, 2006) have been developed for mapping "functional" clusters of genes that respond to these metabolic perturbations. Some of the metabolic enzymes and transcription factors identified by gene cluster analysis in the liver of the perihatch chick (Cogburn et al., 2003c, 2004) or fasting and refed chickens (Duclos et al., 2004) have been integrated into a working model of transcriptional control of the tricarboxylic acid cycle and fat biosynthesis pathway (Cogburn et al., 2004). Thus, time-series perturbation studies and gene cluster analysis provide powerful methods for revealing the major topography of gene networks that control metabolic pathways in the chicken. Identification of conserved motifs (i.e., TFBS) in the promoter region of coexpressed genes should enhance the identification of genetic regulatory loci that control important production traits in poultry.

Expression QTL or "Genetical Genomics"
Several techniques have been developed to help assess gene function on a genome-wide scale. The most common method is to monitor gene expression levels with cDNA microarrays, whereby genes found to be differentially expressed may contribute to the trait being examined. Proteomics with mass spectrometry provides essentially the same type of information except at the protein level. With a growing number of genomic techniques, it is not surprising to find that 2 or more high-throughput methods have been integrated to harness more power and information. This section reviews attempts to merge transcriptional profiling with traditional QTL analysis to reveal functional genes controlling expression of important phenotypic traits.

An early example was the identification of positional candidate genes for Marek’s disease (MD) resistance QTL (Liu et al., 2001a). Fourteen QTL for MD resistance were identified by using an F₂ cross between inbred experimental lines that were relatively resistant or susceptible to MD, a virus-induced lymphoma of chickens (Vallejo et al., 1998; Yonash et al., 1999). Because it is extremely difficult to find the causative gene(s) for each QTL, it was hypothesized that resistance to MD could be accounted for by differences in gene expression. In other words, a positional candidate gene for MD resistance QTL would be one that is within a QTL and is differentially expressed between the resistant and susceptible parental lines following challenge with the MDV. Because this work was done prior to release of the chicken genome sequence, each differentially expressed gene had to be mapped to determine its genomic location. Despite monitoring only 1,200 cDNA spotted on nylon membrane arrays and mapping 15 candidates on the unassembled genome sequence, Liu et al. (2001b) identified a single MD resistance gene, GH, and several other promising candidate genes with this approach. With the current availability of an assembled genome sequence and genome-wide DNA microarrays, this approach could be more powerful, simpler to implement, and used for any quantitative trait (Wayne and McIntyre, 2002).

Thus, genetical genomics or eQTL is simply the marriage of traditional linkage analysis and global transcript profiling with cDNA microarrays (de Koning et al., 2005, 2007; Haley and de Koning, 2006). Initially conceived by Jansen and Nap (2001), gene expression, as measured by transcript abundance, is considered as another quantitative trait or phenotype and, in combination with genetic markers spaced throughout the genome, QTL are revealed that account for variation in gene expression. The result is that QTL can be in either cis or trans with respect to the gene of interest. The simplest interpretation for cis-acting QTL is that sequences flanking the gene (e.g., the promoter region) regulate gene expression or transcript stability. On the other hand, trans-acting QTL are thought to involve transcription factors or other modulators. Genomic regions with a high proportion of eQTL could represent areas with genes that have common transcriptional regulators, which control important biological pathways. The ability to identify trans-acting loci is particularly attractive because it is difficult to identify expression regulators even with a complete genome sequence.

The first study to implement this approach was performed by using yeast (Brem et al., 2002), with additional reports with more information appearing soon afterward (Yvert et al., 2003;