Proteins rather than mRNA are the major effectors of cellular and tissue function. The study of mRNA expression assumes that changes in mRNA expression reflect changes in protein expression. There are many examples, such as post-translational modifications, where protein expression or function is not controlled by mRNA expression 1213. In the intact non-diseased liver tissue, approximately 25% of the changes in the mRNA transcript expression are not accompanied by changes in the expression of the corresponding protein. Studies comparing mRNA and protein expression in liver are few. Anderson et. al. showed a poor correlation of the liver tissue abundance of 19 proteins and corresponding mRNA transcripts (correlation coefficient of only 0.48). Further, they isolated 50 abundant mRNA transcripts of which 29 encoded secreted proteins 14. However, of the 50 most abundant proteins they isolated none were secreted 14. There is a bias in mRNA pools, when compared to protein expression, towards an over-representation of both secreted protein transcripts and high abundance mRNA transcripts, such as glyceraldehyde-3-phosphate dehydrogenase, has been repeatedly demonstrated 14151617. Human proteome and transcriptome comparisons are rare and complicated by our heterogenous multi-organ structure. In prototype organisms (yeast Saccharomyces cerevisiae) comparison of the proteome and transcriptome showed a correlation of 0.94 for all proteins and genes but a correlation of only 0.36 for lowly expressed proteins and genes 18. The same study went on to show that mRNA expression varied up-to 20 fold with constant protein expression and that the protein expression could differ by greater than 30 fold when mRNA expression remained constant 18. However, another study in Saccharomyces cerevisiae, showed a good correlation (r = 0.76) between lowly expressed proteins and mRNA expression with an estimated mRNA abundance as low as 10 transcripts per cell 19. These two studies do shed light on some of the pitfalls inherent in human transcriptome analysis. Specifically, the importance of estimating the frequency of mRNA transcript and protein expression within homogenous (i.e. cell lines) compared to heterogenous (i.e. organs) cell populations.

A prototype human cell such as a lymphoblast has a mass of 670 pg, contains 67 pg of protein (a total of 10⁹ polypeptide molecules representing 4,000 different proteins) and 0.024 pg of mRNA (4 × 10⁴ transcripts representing 5,000 different mRNA) 20. Rare mRNA transcripts present at a frequency of one copy per cell may well be missed based on the cell population frequency and size of the mRNA pool being analysed 20. This is an important consideration for hepatologists utilizing biopsy specimens where the non-parenchymal cell subpopulation abundance is low and subject to sampling error 212223. Additionally, it is apparent that if mRNA amplification is utilized in functional genomic experiments then every effort needs to be made to ensure gene, especially low abundance gene, representation is retained. Further, in functional genomic methods that "sample" the transcriptome, such as serial analysis of gene expression (SAGE), differential display and subtractive hybridisation many low abundance transcripts will be missed without sampling of a very large numbers of clones identified. Therefore, most of these "sampling" methods are used as a means of identifying differential expressed mRNA rather than profiling a transcriptome. Detection of protein expression differs as the analytically threshold is the limiting factor. Currently proteomic detection requires picomole amounts of protein (10¹² – 10¹³ peptides) 20. In a homogenous cell population 90% of the cellular protein mass is due to the 100 most abundant proteins and a further 1200 proteins account for another 7% of the protein mass that is detectable on typical proteomic analysis (from a lysate of 10⁶ cells) 20. However, the remaining 3% of the protein mass includes 2800 proteins (over 50% of the different protein species) that fall below the threshold of detection for typical proteomic analysis 20. Therefore, differential protein expression or proteomics can document changes in protein expression but is currently restricted to moderate to high abundance proteins as well as being technically demanding. Ultimately, the protein expression in every cell is controlled by the transcriptome although the relationship between individual gene transcripts and the corresponding protein expression may not at first, be clear.

Array analysis has become a commonly published methodology. Microarray analysis leads the way with more published experiments than other forms of array experiments. Array experiments offer "...a whole new way of looking at cellular connections" 44. Array analysis has targeted identification of "candidate" genes in pathobiology, identification of gene networks, characterisation of gene ontologies and phylogeny as well as the classification of pathobiology 44. The diagnostic potential of array analysis is further suggested by two early pivotal studies conducted in human leukemia and lymphoma patients. Golub and colleagues were able to distinguish acute myeloid leukemia from acute lymphoid leukemia (ALL) based on the gene expression pattern on microarrays 64. Further, they were able to identify important "class predictor" sets of genes and were further able to divide ALL samples into those of B- or T-cell origin 64. The microarray analysis of Alizadeh et. al. showed that diffuse large B-cell lymphoma could be divided into two broad groups based on the pattern of gene expression consistent with either a germinal centre origin or a peripheral B-cell origin 65. Further, patients with the peripheral B-cell origin gene profile had a significantly greater long-term survival 65. Following these initial studies there are now a number of manuscripts that correlate clinical outcome in malignancy with tumor mRNA transcript expression. In a number of studies microarray profiling of transcript expression was the best predictor of clinical outcome. Therefore, gene expression profiles are now moving from the research laboratory into clinical practice as a diagnostic and prognostic tool in malignancy. However there are very few studies correlating clinical outcome with gene expression in non-malignant disease. This is consistent with diseases in which there is a marked difference in gene expression (i.e. in malignancy) being the easiest to discriminate by gene array analysis. This underscores the need to understand normal variability and to address the issue of reproducibility if gene array analysis is to be used for predicting outcome in non-malignant disease.

Array analysis in human liver disease has now profiled the intrahepatic gene expression in non-diseased liver, a number of diseases and animal models of disease. Presently there are in excess of 200 published gene array studies of human liver disease or studies that utilize human liver tissue (Table 2). This represents roughly 40% of the greater than 500 published papers utilizing gene arrays to study liver pathobiology. Most of these studies attempt to understand pathobiology by examining mRNA transcript expression. There are few publications in human liver disease where gene expression is correlated with clinical outcome.

Gene array analysis has provided novel insights into the pathogenesis of hepatitis B (HBV) and hepatitis C (HCV) liver injury. Importantly, both of these diseases have been extensively studied using primate models of infection or in human disease. Further, examination of HCV liver injury is restricted to studies in humans and chimpanzees as there are no in-vivo animal models of infection. The insights into viral hepatitis pathogenesis offered by gene array analysis has shown a number of common aspects such as the interferon (IFN) associated gene response as well as distinct differences that are helping us to understand the unique intrahepatic transcript "signature" of each viral infection. Additionally, studies of temporal changes in transcript expression are central to understanding clearance of virus, carrier states and injury in chronic persistent infection.

Acute HBV and HCV infection has been analysed in the chimpanzee 666768. Acute HBV infection is characterised by an absence of intrahepatic differential gene expression during HBV infection and the initial phase of virus replication. This is followed by the differential gene expression associated with T-cell receptor and antigen presentation. Subsequently gene expression associated with T cell recruitment (chemokines), T cell effector function (granzymes) and monocyte activation was observed. A later phase of clearance was associated with the expression of B cell related genes. This is in direct contrast to HCV infection and suggests that HBV infection fails to induce a significant innate immune response by acting in the initial phase of infection as a "stealth virus" 6669. Presently there are no published studies of acute or chronic HBV infection in humans. Our own studies of cirrhotic HBV explants demonstrated less differential gene expression than studies comparing normal liver and HCV associated cirrhosis. Chronic HBV infection is characterised by intrahepatic upregulation of stress response, cell cycle and immune response associated genes. This is similar to findings reported by others (personal communications).

Chimpanzee models of HCV have helped delineate the development of the intrahepatic immune response in acute and chronic HCV infection. Studies of chimpanzees during acute HCV infection which showed a dramatic intrahepatic response with an increase in IFN response genes as early as two weeks post infection 6768. The chimpanzee studies demonstrated an initial response with elevated IFN-alpha/beta associated with a biphasic intrahepatic immune response to HCV resulting in viral clearance 676870. The IFN induced genes such as ISG15 and ISG16-jun were initially strongly upregulated followed by over 50 fold induction of Th1 associated transcripts such as MIG (CXCL9) and IP10 (CXCL10). Subsequent increased CXCL-10 (IFN-induced protein 10 (IP-10)) and midkine (MK) peaking at 6 weeks is consistent with an adaptive, Th1 associated, immune response clearing infected hepatocytes of the virus 68.

Comparison of chimpanzees that cleared acute HCV infection compared to an animal that had virus persistence has provided further insight into the balance between viral clearance and peristance 71. In these experiments, Su et al. 71 observed upregulation of genes associated with the early response (which correlated with viral load) including many IFN alpha induced genes; STAT 1, 2'-5' oligoadenylate synthetase, Mx1, ISP15 and p27. Interestingly, there was the induction of lipid pathway genes such as fatty acid synthetase, sterol response element binding protein (SREBP), down regulation of PPARα as well as hepatic lipase C and flotillin 2. The lipid pathway genes are associated with viral replication and studies using in vitro replicon experiments has demonstrated altered viral replication 72. Further, the reduction in PPARα would be expected to be associated with insulin resistance, a feature of chronic HCV, but prior to this it was not an expected aspect of acute HCV infection. As noted previously 68 clearance of HCV was associated with the late induction of Th1 transcripts such as CXCL9 and CXCL10, MHC expression and T cell molecules such as CD8 and granzyme A. The induction of IFN alpha induced genes early in infection was also observed by Bigger et al but the timing did not correlate with clearance as high levels of these transcripts continued in the animal with viral persistence. This has also been seen in human studies. Further, functional studies in HCV replicon systems has shown that the NS3/4a was able to inhibit interferon alpha antiviral effector function by blocking the phosphorylation of IRF-3 a key protein in the antiviral response 73. Therefore, chronic HCV infection induces a persistent intrahepatic IFN alpha antiviral response but the virus itself escapes this response via inhibition of the effector arm. However, microarray studies of the intrahepatic IFN alpha induced gene response show that this is variable and observed to be higher in patients not responding to pegylated IFN and ribavirin therapy consistent with resistance of the effector arm of the immune response to amplification by exogenous therapy 74. In contrast, patients who had a sustained viral response (SVR) to pegylated IFN therapy had a lower expression of IFN genes consistent with amplification of the effector arm of the immune response by exogenous therapy resulting in viral clearance.

Chronic HCV infection has been studied in a number of ways using gene array analysis. The study of Honda et al. profiled gene expression in liver biopsy material in individuals with chronic hepatitis B (HBV) and chronic HCV comparing them to a single non-diseased control 75. The authors concluded that chronic HCV infection was associated with a predominant anti-inflammatory, pro-proliferative, anti-apoptotic intrahepatic gene profile 75. However, the presented data demonstrated widespread upregulation of pro-inflammatory genes such as IL-2 Receptor, CD69, CD44, IFN gamma inducible protein, MHC Class 1 genes and monokine induced by gamma IFN. These findings were similar to another study of HCV cirrhosis in which a pro-inflammatory Th1 associated transcript expression predominated 76. The Th1 immune response is thought to be responsible for the accelerated fibrogenesis of HCV liver injury 77. Fibrosis associated gene expression in HCV associated fibrosis has included upregulation of a wide variety of genes including PDGF, TGF-beta 3 78. Additionally, gene array analysis has identified potential novel mediators of HCV associated injury such as Frizzled related proteins, discoid domain related protein1 (DDR1), EMMPRIN and SARP-3 76.

The premaligant potential of intrahepatic HCV infection has been studied by our group by comparing HCV cirrhosis with and without HCC by gene array analysis 79. The upregulation of many oncogenes (i.e. TEL oncogene), immune genes (IFN gamma associated), fibrosis genes (integrins) as well as cell signalling (G coupled receptor kinase) and proliferation associated genes (cyclin K) was demonstrated in cirrhosis complicated by HCC. This is consistent with a premalignant cirrhotic response in HCV infection. Further, the data suggest that there is more cellular proliferation, immune activation and fibrosis in the liver of patients with HCC than those with cirrhosis alone. A key area of future research will be to ascertain wether such a profile can be recognised before HCC develops. This approach has a direct clinical application in identifying and screening high-risk patients.

Gene array studies of HCV infection has revealed new insights into the development of HCC in HCV, structural analysis of the HCV RNA genome and identified novel markers of HCV intrahepatic injury and HCV associated HCC. The study of Smith et al. utilized 13 600 gene microarrays to profile patients with HCV cirrhosis, HCV and HCC and normal liver 80. The results identified 87 upregulated and 45 down-regulated genes that appear to be markers of HCV liver injury 80. Importantly, the analysis aimed to exclude genes expressed in normal liver, other forms of cirrhosis or HCC. Genes such as ILxR (IL-13 receptor a2), CCR4 and cartilage glycoprotein 39 (GP-39) were identified 80. However, the study of Smith et al. highlights the problems with the interpretation of these large data sets using small numbers of patient samples; does the identified gene expression represent unique disease or phenotype associated gene expression or the stochastic probability of identifying a small cohort of genes from the many thousands being analysed? Cleary studies such as these, as powerful as they are, need to be validated by alternative methodologies in large patient groups. Our own approach to validation has been to confirm important gene expression identified in these studies by real-time RT-PCR in a larger cohort of patients 7881. Indeed our own results have confirmed the increased expression of GP-39 in HCV associated cirrhosis identified following subtractive hybridization and confirmed this finding by real-time RT-PCR 82.

Hepatocellular carcinoma proliferation in HCV associated liver injury has been studied by array analysis. This has resulted in a plethora of potentially novel tumour markers being identified. These include the serine/threonine kinase 15 (STK15) and phospholipase A2 (PLA2G13 and PLA2G7) that were shown to be increased in over half of the tumours identified 80. However, a different study implicated different gene groups in HCV associated HCC; cytoplasmic dynein light chain, hepatoma derived growth factor, ribosomal protein L6, TR3 orphan receptor and c-myc 83. The clustering analysis in this study showed that the expression of 22 genes in HCC related to differentiation of the malignancy with over half of these genes being transcription factors or related to cell development or differentiation 83. Although many of these genes can be implicated in HCC development they often identified in large gene sets in end stage disease. Therefore, whether these genes represent cause or effect is unknown. Additionally, the number of differing gene sets being examined by the gene arrays being utilized is almost as great as the number of studies using them. Further as these gene sets still only represent a fraction of the transcriptome being examined they selectively identify differentially expressed genes. The same group went on to examine the expression of genes in hepatocyte cell lines expressing α-fetoprotein (α-FP) 84. Comparison with non α-FP producing cell lines showed that the hepatocyte lines had a similar pattern of gene expression 84. Further, the cell lines Huh-7, Hep3B and HepG2 clustered together suggesting that these cells lines are closely related 84. Indeed these cell lines shared 254 genes (out of 930) that were commonly up-regulated 84. Genes increased in the α-FP producing cell lines included α-FP, ephrin-A1, TGF-α2, MMP-2 and IGF-II 84. This is entirely consistent with the known role of these genes in carcinogenesis and more specifically with transcriptional regulation of MMP-2 by IGF-II 85. The oncogenic potential of viral components, especially core protein, has been implicated by transcriptome profiling in the pathogenesis of HCV associated HCC.

Gene array analysis of HCV recurrence in transplant allografts has provided novel insights into the molecular mechanisms of viral recurrence 8687. HCV recurrence in the liver graft is associated with expression of IFN-γ associated genes such as CXCL10 (IP-10), CXCL9 (HuMIG) and RANTES 87. Further, antiviral IFN-α associated gene expression is seen in chronic HCV recurrence and during acute rejection associated with HCV recurrence 87. Additionally, upregulation of NF-kappa β pathway during acute rejection in association with HCV recurrence appears to alter cellular apoptosis via changes in the expression of TRIAL associated genes 87. Importantly chronic HCV recurrence in grafts is associated with Th1 associated gene expression similar to that seen in chronically HCV infected individuals that have not been transplanted 87. In contrast, cholestatic HCV recurrence, which follows an aggressive course, is associated with a Th2 cytokine profile 87. This suggests that the Th1 immune response suppresses viral replication whilst being profibrogenic 788788. In cholestatic HCV recurrence the unchecked viral replication is directly fibrogenic 8788.

Alcoholic liver injury is an example of both a classical and atypical hepatotoxin. Microarray analysis has been applied to studies of neural tissue in an attempt to understand ethanol addiction. Intrahepatic gene profiling using microarrays in ethanol feed baboons has identified increased expression of 14 different annexin genes (including A1 and A2) that were not previously implicated in the progression of fibrosis in alcoholic liver disease 89. Further, the intrahepatic transcriptome profile in alcoholism shares some similarity with LPS administration but in general is significantly different from other forms of liver disease 8990. Additionally, the hepatocyte transcript response to ethanol is significantly different compared to other hepatotoxins such as anticancer drugs 91.

Cluster analysis has allowed differentiation of alcoholic hepatitis from alcoholic steatosis. Genes known to be involved in alcohol injury such as alcohol dehydrogenases, acetaldehyde dehydrogenases, interleukin-8, S-adenosyl methionine synthetase, phosphatidylethanolamine N-transferase and several solute carriers have been shown to be differentially expressed in alcoholic hepatitis versus alcoholic steatosis. Many novel differentially expressed genes were identified, including claudins, osteopontin, CD209, selenoprotein and genes related to bile duct proliferation 89. The most prominent categories of differentially expressed genes involved cell adhesion/extracellular matrix proteins, oxidative stress and coagulation that were also common to end-stage alcoholic liver disease. Genes associated with fibrosis/cell adhesion/ECM were the most prominent category in human advanced ALD, consistent with the fibrotic nature of ALD. However, these were not specific to alcohol, and have been reported in primary biliary cirrhosis and other forms of liver cirrhosis 7681.

Non-alcoholic steatohepatitis (NASH) is the clinicopathological syndrome in non-alcoholic fatty liver disease (NAFLD) that has been most widely studied using gene array analysis of transcript expression. Studies have identified differentially expressed genes in end stage NASH cirrhosis compared to other disease states 929394. Decreased expression of genes associated with mitochondrial function and increased expression of genes associated with the acute phase response were observed 92. The latter increases were speculated to be associated with insulin resistance, a feature of NAFLD 92. Further differential expression of genes involved in lipid metabolism, extracellular matriz (ECM) remodelling, regeneration, apoptosis and detoxicification have all been observed in NASH following microarray analysis 93.

Biliary liver injury has only been examined in a limited number of studies utilizing array analysis. The induction of Wnt pathway genes including Wnt13, Wnt5A, and Wnt12 was a striking and confirmed finding of gene array studies of primary biliary cirrhosis (PBC) 81. Further novel PBC associated transcript expression included upregulation of Transcription initiation factor 250 kDa subunit (TAFII 250), PAX3/forkhead transcription factor and patched homolog (PTC). An unexpected but consistent feature of the gene array analysis of PBC was the repeated identification of differentially expressed Drosophila genes homologues (Wnt genes, hedgehog pathway, notch pathway) 81.

The only available data on primary sclerosing cholangitis (PSC) cirrhosis comes from a comparison to PBC cirrhosis 81. Compared with PBC there were a far greater number of genes showing differential expression in PSC versus non diseased liver. These include genes associated with epithelial biology (Amphiregulin, Bullous pemphigoid antigen), inflammation (T-cell Secreted Protein P I-309, CTLA4), apoptosis related genes (Bcl-2 interacting killer, Bcl-x, Death associated protein 3) and intracellular kinases such as CDK7 and JAK1.

Biliary atresia (BA) has been studied by microarray analysis by comparing embryonic and perinatal forms of the disease 95. Gene profiling clearly separated these two conditions. The most remarkable difference was in the expression of so-called regulatory genes. In Embryonic BA 45% of differentially expressed genes were in this category versus 15% in the perinatal form. Included in these genes were imprinting genes, genes associated RNA processing and cell cycle regulation that were not present in the perinatal form of BA.

Experiments with cells in culture offer the advantage of a controlled environment in which to test specific hypotheses without influences at the level of the organ, organism or environment. Gene array technology has been used to characterise cells in culture in greater detail than previously possible. The techniques have been applied to study cellular differentiation and behaviour in response to toxins and various disease states. Despite the advances derived from this novel technology, limitations of the cellular models apply to these studies as well. For example, cellular function is altered by the culture conditions and may not represent those found in vivo.

Gene expression array analysis has been applied to identify the multiple signals involved in cellular differentiation. Corticosteroids, hepatocyte growth factor (HGF), and epidermal growth factor are associated with mature histology in the organoid culture model 96. Array analysis identified that these factors stimulate hepatocyte nuclear factor 4alpha (HNF-4alpha) expression in hepatocytes 96. HNF-4alpha is a recognized liver-specific transcription factor and its effects on hepatic gene expression have also been studied. Almost half of the induced genes were metabolism genes many related to lipid metabolism which is frequently altered in liver disease 97. Yamashita and colleagues determined target genes for hepatocyte differentiation and found that the Oct-3/4 transcription factor was upregulated while the early growth response-1 (EGR-1) transactivator was down-regulated 98. Gap junctions are considered to play a central role in differentiation of hepatocytes. Connexin 32 (Cx32) is closely related to tight junctional proteins and can induce expression and function of tight junctions. To investigate the mechanisms of induction of tight junctions, cells transfected with Cx32 were analysed by cDNA microarray 99. Expression of membrane-associated guanylate kinase with inverted orientation-1 (MAGI-1) was increased. MAGI-1 is known to be localized at adherens and tight junction regions. MAGI-1 was expressed in the apical-most regions at cell borders of Cx32 transfectants and co-localized with occludin, claudin-2, ZO-1, and F-actin 99.

Cultured primary hepatocytes have proven to be a valuable resource and extensively used research tool but questions remain regarding functional differences observed in these hepatocytes relative to the intact liver 9798100101102103. One study characterized cultured hepatocyte cell lines, primary hepatocytes in conventional monolayer or in sandwich culture, and liver slices based on mRNA expression profiles in comparison to gene expression in liver tissue 103. Liver slices exhibited the strongest similarity to liver tissue regarding mRNA expression, whereas the two cell lines clustered together and were quite different from the whole liver. For selected cytochrome P450s the differences observed on the mRNA expression level there was a marked effect with the duration of culture. Expression patterns changed most rapidly soon after cell isolation and culture initiation and stabilized with time in culture 103. A second study of cultured hepatocytes over time revealed time-dependent regulation of phase I and phase II metabolizing enzymes 100. In general, cytochrome P450 mRNA expression was repressed, but expression of phase II metabolizing enzymes varied by class (upregulation of glucuronidation, down-regulation of sulfation). Progressive induction of several genes associated with the cellular cytoskeleton and extracellular matrix was observed in accord with physical changes in cell shape and connectivity associated with cellular adhesion 100.

Gene expression profiling has been utilised to define the molecular mechanism underlying epithelial non-parenchymal interactions in hepatocyte cocultures. Primary rat hepatocytes were cocultivated with closely related murine fibroblast cell types and revealed functional responses that correlated with fibroblast gene expression profiles. Two candidates playing an important role in functional differentiation were the cell surface protein neural cadherin (N-cadherin) and decorin 102.

cDNA microarray was used to identify genes upregulated in activated hepatic stellate cells (HSCs) in culture 104105. In one study, a number of novel and previously recognised genes were identified including osteopontin (OPN) 104. In another, a total of 835 differentially expressed genes were identified in an array comparison of activated and quiescent HSC. The differentially expressed genes included those involved in protein synthesis, cell-cycle regulation, apoptosis, and DNA damage response 105. Functional expression of the telomerase catalytic subunit (human telomerase reverse transcriptase; hTERT) in human activated hepatic stellate cells (HSCs) rescues them from death with immortalization and maintains an activated HSC phenotype 106. Senescent HSCs expressed reduced levels of extracellular matrix proteins, including collagens, tenascin, and fibronectin. Maintenance of telomere length represents an important survival factor for activated human HSCs 107. Using this information, Schnabl and colleagues have created an immortalized human HSC line by infecting primary human HSCs with a retrovirus expressing hTERT 106. Telomerase-positive HSCs did not undergo oncogenic transformation and exhibit morphologic and functional characteristics of activated HSCs. Microarray and RT-PCR analysis showed that mRNA expression patterns in telomerase-positive HSCs are very similar to those in activated human HSCs 106. The immortalized HSC lines LX-1 and LX-2 were characterized by microarray analysis and determined to have a gene expression profile very similar to that of activated primary HSC 108. These newly developed cell lines are proving to be valuable tools to study the biology of human HSCs.

A particular challenge in the study of the effect of viruses on liver cells is the difficulty in infecting liver cells with virus. The studies described below have involved models in which cultured cells are infected with viral proteins or viral genome. Progress in this field has been rapid and most recently, a cellular model of HCV infection has been reported that is likely to stimulate further study 109110.

To determine the oncogenic role(s) of HBx protein Hepatitis B virus (HBV) in the development of HCC, gene expression profiles in primary adult human hepatocytes and an HCC cell line (SK-Hep-1) ecotopically expressing HBx via an adenoviral system. Many genes including a subset of oncogenes (such as c-myc and c-myb) and tumour suppressor genes (such as APC, p53, WAF1 and WT1) were differentially expressed and cluster analysis showed distinctive gene expression profiles in the two cell types. HBx protein altered gene expression as an early event that favours hepatocyte proliferation that may contribute to liver carcinogenesis 111.

Interferon-alpha is currently the leading treatment for viral hepatitis. Several studies have used microarray analysis to identify the mechanisms by which interferon-alpha (IFN-alpha) acts on hepatocytes and the hepatitis C virus. IFN-alpha activated the multiple signal transducer and activator of transcription factors (STAT) 1, 2, 3, 5 in cultured hepatocytes 112. Other up-regulated genes include a variety of antiviral and tumour suppressors/proapoptotic genes. Down-regulated genes include c-myc and c-Met and the HGF receptor 112. In a second and comparable study, IFN-alpha antiviral efficacy was associated with 6–16 (G1P3) expression. Involvement of STAT3 in IFN-alpha signalling was confirmed 113. Resistance to IFN-alpha antiviral activity may be mediated the hepatitis C viral protein, NS5A. To identify the mechanisms through which NS5A blocks interferon activity, gene expression profile was studied in IFN-treated Huh7 cells expressing NS5A. The strongest effect of NS5A on interferon response was observed for the OAS-p69 gene 114. Another key response of hepatocytes to the HCV virus is cellular proliferation. Gene array studies identified upregulation of growth-related genes, in particular wnt-1 and its downstream target gene WISP 115. In another study, CDK activity, hyperphosphorylation of Rb, and E2F activation was shown to be associated with hepatocyte proliferation induced by a full-length HCV clone 116.

A number of studies have investigated the effects of toxins and therapeutic drugs in cultured liver cells using array techniques. Perhaps surprisingly, studies of the effects of alcohol on liver cell gene expression in vitro have not been reported. It has been proposed that transcription profiling can generate the information needed to assign a compound to a mode-of-toxicity class. Primary rat hepatocytes have been exposed to a variety of different hepatotoxins on the basis of their variety of hepatocellular effects 117118. A low-density DNA microarray containing 59 key genes was selected. All tested drugs generated a specific gene expression profile. Even with a relatively limited gene set, gene expression profiling allowed a certain degree of classification of compounds with similar hepatocellular toxicities such as cholestasis or necrosis. Clustering analysis linked the compounds known to cause hepatic steatosis. Drugs inducing necrosis and cholestasis clustered together and drugs classified as the CYP450 inducers formed individual clusters 119.

The pathways underlying ursodeoxycholic acid (UDCA) action were investigated by array analysis in primary rat hepatocytes 120. In cells exposed to UDCA, >440 genes were modulated by >1.5-fold. Genes affected by UDCA included new regulatory molecules, such as Apaf-1. Other altered genes were directly involved in cell cycle (cyclin D1, cadherin 1, HMG-box containing protein 1) and apoptosis (prothymosin-alpha) events. The E2F-1/p53/Apaf-1 pathway appears to be targeted by UDCA 120.

Saturated fat plays a role in common debilitating diseases yet the molecular mechanisms of saturated fatty acid signalling in the liver are poorly defined. Hepatic gene expression analysis was undertaken in a human hepatocyte cell line after incubation with palmitate 121. Altered transcription profiles were observed in a wide variety of genes, including genes involved in lipid and cholesterol transport, cholesterol catabolism, cell growth and proliferation, cell signalling, beta-oxidation, and oxidative stress response 121. Streptozotocin (SZ) is known to exert toxic effects not only on pancreatic islet beta cells but also on other organs including the liver. Primary cultured hepatocytes were treated with SZ 122. Gene expression analysis revealed up-regulation in cell proliferation/apoptosis related genes, and down-regulation of lipid metabolism related genes. These results indicate that many of hepatic alterations are direct effects of SZ rather than by the secondary effect of the hyperglycaemia or hypoinsulinemia 122.

Microarray analysis was conducted on isolated human hepatocytes to understand the mechanisms underlying the idiosyncratic toxicity induced by trovafloxacin 123. The results clearly distinguished trovafloxacin from other marketed quinoline agents and identify unique gene changes induced by trovafloxacin that are involved in mitochondrial damage, RNA processing, transcription, and inflammation 123. The mechanism(s) by which LPS stimulates cultured hepatocytes was studied by DNA microarray analysis. LPS modulates the selective expression of more than 80 genes and expressed sequence tags including members of relevant signalling pathways. NF-kappa B activation was reduced in TLR4-mutant or -null hepatocytes compared to control hepatocytes, and this defect was partially restored by adenoviral transduction of mouse TLR4. This study provided additional evidence that hepatocytes respond to LPS through a TLR4 response pathway 124.

Animal model studied include acute liver regeneration, drug toxicity, liver fibrosis, fatty liver, biliary obstruction, liver transplantation and carcinogenesis. Drug toxicity studies are numerous and include effects induced by clofibrate, PPAR alpha agonists, carbon tetrachloride, amiodarone, arsenic and methotrexate 117125126127128129130131132133134. In one study, a novel cDNA library highly enriched for genes expressed under a variety of hepatotoxic conditions was created and used to develop a custom oligonucleotide library 135.

An expression signature for rat liver fibrosis was identified using a cDNA 14,814 gene microarray 136. The "genetic fibrosis index" identified consisted of 95 genes (87 upregulated, 8 down-regulated). These included genes associated with cytoskeletal proteins, cell proliferation and protein synthesis. Bile obstruction in the mouse identified 3 sequential main biological processes. At day 1, enzymes involved in steroid metabolism were over expressed. This was followed by an increase in cell cycle/proliferation associated genes at day 7, occurring at a time of maximum cholangiocyte proliferation. From days 14–21 genes associated with the inflammatory response and matrix remodelling were identified. Similar temporal gene expression was identified in the model of acute liver regeneration. Steroid and lipid metabolism genes were down-regulated as early as 2 hours post hepatectomy whilst genes associated with cytoskeletal assembly and DNA synthesis became upregulated by 16 hours and remained elevated at the 40 hour time point at the peak of S phase.

Carcinogenic foci in experimental animals have been isolated and studied using gene array technology 137. Approximately 8% of 2000 transcripts were differentially expressed in one study. These included genes with roles in signal transduction, detoxification and cytoskeletal assembly. Over 30 genes were identified as being dysregulated in these foci as well as in neonatal liver. Small for size liver allografts in rats showed upregulation of adhesion molecule, inflammatory mediators and apoptosis-associated genes together with a down regulation of energy metabolising genes.

Alcoholic liver disease has been studied in the chronic enterogastric ethanol infusion model in a mouse, a total of 12,422 genes were analysed 138. Several cytochrome P450 genes were shown to be upregulated, whilst several genes involved in fatty acid metabolism (stear-a oil co-aid saturase 3-hydroxy-assile co-aid dehydrogenase) and fatty acid synthesase were down regulated. In contrast, genes associated with glutathione-s-transferase were markedly upregulated. Interestingly, a novel molecule intestinal factor was 50-fold down-regulated. It was postulated that alcohol may be affecting the healthy intestinal epithelium and down-regulation of this gene may be associated with permeability changes in the intestine associated with chronic alcohol ingestion.

Although array analysis is the predominant technique for examining differential gene expression other techniques such as differential display, SAGE 139 and subtractive hybridisation 34 are useful complementary methods of examining transcriptomes.

Subtractive hybridisation is a method of enriching for differentially expressed genes in one gene pool compared to another 34. The essential feature of this method is that one gene pool is labelled or tagged enabling the separation of unique transcripts from the tester (also called tracer) gene pool following hybridisation to an excess of driver cDNA 34. The starting material is often limited and amplification is performed prior to subtraction using a number of techniques such as poly(A) RT-PCR, Eberwine amplification and SMART cDNA synthesis (Clontech, CA, USA) 34. The technique classically uses a biotinylated driver enabling common sequence in the driver and tester gene pools to be removed using streptavidin precipitation 34. Multiple rounds of subtraction can be performed giving a tester gene pool enriched for differentially expressed genes 34. A number of techniques related to subtractive hybridisation include representational differences analysis (RDA) and suppression subtractive hybridisation (SSH). Representational differences analysis uses PCR amplification of the unique transcripts in the cDNA pool 153154. This can give a greater than 10⁶ fold enrichment of differentially expressed cDNA in three rounds of amplification 153154. RDA has been utilised to find unique genomic DNA mutations as well comparing mRNA transcript pools 153154. SSH utilises PCR amplification (like RDA) combined with the suppression effect of PCR 155156. The suppression effect is mediated by the incorporation of long inverted terminal repeats which, when attached to the ends of DNA fragments, form stable panhandle-like loop structures that are favoured over the shorter PCR primers 155156. Therefore, undesirable sequences are not amplified in the PCR.

All subtractive hybridisation approaches simply enrich for differentially expressed genes. Individual gene differential expression needs to be confirmed by supplemental methodology. The supplemental methodology used varies but includes northern blot analysis and PCR based methods. Northern blot analysis, although an accurate method of quantification, that additionally gives information about mRNA transcript size, is limited by being labour intensive and lacks the throughput necessary to match the subtractive hybridisation approach. Additionally, Northern blot analysis is restricted to moderate to high abundance transcripts. PCR based methods are now the preferred means of confirming differential gene expression. Real-time quantitative RT-PCR is a rapid and effective means of confirming gene differential expression. The multiple methods of performing real-time RT-PCR include the use of FRET probes, molecular beacons and the use of the double strand intercalating flurophore Sybr Green 1 157158159. Real-time RT-PCR has the additional benefit of being able to confirm differential expression of lowly expressed mRNA transcripts 160. Indeed the use of real-time RT-PCR is now used not just to confirm differential gene expression following subtractive hybridisation but also to confirm differential gene expression following cDNA array analysis, differential display and SAGE.

Subtractive hybridisation has been used to investigate the pathobiology of human liver disease. There are now in excess of twenty papers that utilise subtractive hybridisation to investigate liver injury. The SSH variation, described above, is now the most commonly utilised subtractive hybridisation method 156161. Further, SSH unlike mRNA differential display appears to have a greater yield in terms of the number of genes identified that are both known and novel in an individual experiment. The identification of gankyrin in HCC using SSH is a significant finding as this oncoprotein, with ankyrin repeats, has been demonstrated to increase anchorage-independent growth and tumorigenicity in NIH/3T3 cells 162. Gankyrin appears to increase the phosphorylation of the retinoblastoma gene with activation of the E2F-1 transcription factor 162. Studies of iron overload have previously demonstrated iron overload in knockout mice lacking the Usf-2 transcription factor in a pattern similar to the HFE -/- mouse, the murine model of genetic haemachromatosis 163. Comparison of intrahepatic gene expression in Usf-2-/- and Usf-2+/+ mice by SSH showed a marked down-regulation of the mRNA encoding hepcidin (also known as liver expressed antimicrobial peptide) 163. Concurrently, Pigeon et. al. also identified increased hepcidin expression using SSH in carbonyl treated mice that have iron overload 164. Therefore, hepcidin appears to act in conjunction with HFE to regulate intestinal iron absorption and iron storage in macrophages although the exact mechanism is still unknown 163164. The study of hepatitis C infected livers using SSH confirms a Th1 associated immune profile with CXCL-10, IFN regulated MxA, IFN induced p44, and IFN induced p56 (IFI-56 K) all increased greater than 4 fold in chronic HCV infection 165. The identification of increased CXCL-10 in HCV infection has now been demonstrated using both array analysis and SSH 165166. Further, a study of PBC using SSH identified a total of 71 differentially expressed sequences with 62 being known genes and the remaining 9 clones homologous to EST sequences 167. Interestingly, two of the sequences enriched for in PBC included Wnt13 and notch2 suggesting an involvement of these highly conserved Drosophila pathways in PBC pathobiology 167. Unfortunately, many of the studies utilising SSH do not utilise supplemental methodology to confirm differential gene expression.

Our own work with SSH utilized this technique as a method of bio-discovery rather than a means of profiling transcriptomes. By comparing human HCV, AIH, PBC and non-diseased liver in multiple paired comparisons we were able to uncover completely novel sequences involved in the pathogenesis of these human liver diseases 82. The reliance on sequencing clones helped us identify previously unrecognised spliced variants of the gene RERE in HCV cirrhosis 82. Further our results identified genes such GP-39 and 2,5-oligoadenylate synthetase that had been identified previously by other techniques 82. Additionally, we extensively used real-time RT-PCR to validate the observed gene expression 82.