Hepatitis C virus and liver disease: global transcriptional profiling and identification of potential markers

Hepatitis C Virus and Liver Disease: GlobalTranscriptional Profiling and Identification

of Potential MarkersMaria W. Smith,1 Zhaoxia N. Yue,1 Marcus J. Korth,1 Hao A. Do,2 Loreto Boix,3 Nelson Fausto,4 Jordi Bruix,3

Robert L. Carithers, Jr.,2 and Michael G. Katze1

Microarray analysis of RNA from hepatitis C virus (HCV)-infected cirrhotic livers wasperformed to identify a gene expression signature of liver disease. The expression levels ofapproximately 13,600 genes were analyzed using surgical material and core biopsy speci-mens from HCV-infected cirrhotic liver explants in comparison with reference samples ofnormal nondiseased liver. In addition, normal liver samples were compared with each otherto determine normal physiologic variation in gene expression. A set of genes, including someassociated with stress, acute-phase immune response, and hepatic stellate cell activation, hadvariable expression levels in normal livers. These genes were subtracted from the sets of genesdifferentially expressed in cirrhotic livers. To exclude cancer-related genes from our markersets, we subtracted genes that also were expressed differentially in hepatocellular carcinomas.The resultant HCV- and liver disease–associated gene set provided a molecular portrait ofseveral processes occurring in the HCV-infected liver. It included (1) genes expressed inactivated lymphocytes infiltrating the cirrhotic liver, and activated liver macrophages; (2)genes involved in remodeling of extracellular matrix-cell and cell-cell interactions associatedwith cytoskeleton rearrangements; (3) genes related to the anti-apoptotic pathway of Bcl-2signaling; and (4) genes involved with the interferon response and virus-host interactions. Inconclusion, our microarray analysis identified several potential gene markers of HCV-asso-ciated liver disease and contributed to our rapidly expanding database of experiments de-scribing HCV pathogenesis. (HEPATOLOGY 2003;38:1458-1467.)

Chronic viral hepatitis C has a highly variable clin-ical course that results in development of liverfibrosis and cirrhosis in about 20% of patients.1

Early stages of cirrhosis remain compensated or shownonspecific symptoms,2 and there are no accurate nonin-vasive markers of disease progression at this stage.3 Thus,liver biopsy is recommended for patients with chronichepatitis C to establish the histologic stage of liver dis-

ease.3 Analysis of liver biopsy specimens may benefitgreatly from the addition of expression microarray profil-ing to examine messenger RNA (mRNA) populations asmarkers of liver disease progression.4,5 However, to usemicroarray data successfully as a diagnostic or prognosticindicator, it is necessary to define potential marker genesets associated with certain pathologic processes or with aparticular stage of liver disease.

This approach was used in a recent expression microar-ray profiling study that examined liver biopsy specimensfrom hepatitis C virus (HCV)- and hepatitis B virus–infected patients, and that found significant differences ingene expression patterns caused by these 2 chronic infec-tions.6 In another series of studies, liver explants fromliver transplant patients were used to compare 4 differenttypes of cirrhosis, including cirrhosis associated withHCV infection.7,8 However, these studies analyzed theexpression of only about 1,000 genes and used the tech-nique of pooling several (4-8) samples for a single exper-iment. Obviously, a single sample with strongly alteredgene expression patterns could affect the whole pool, par-

Abbreviations: mRNA, messenger RNA; HCV, hepatitis C virus; HSC, hepaticstellate cell; MF, myofibroblast.

From the 1Department of Microbiology, and 2Hepatology Section, Department ofMedicine, and the 4Department of Pathology, School of Medicine, University ofWashington, Seattle, WA; and the 3Barcelona Clinic Liver Cancer Group, LiverUnit, Hospital Clinic IDIBAPS, Barcelona, Spain.

Received May 16, 2003; accepted September 22, 2003.Supported in part by grants 5R01CA074131, 5U19AI48214, and

1P30DA01562501 from the National Institutes of Health.Address reprint requests to: Maria W. Smith, Ph.D., Department of Microbiol-

ogy, University of Washington, Box 358070, Seattle, WA 98195-8070. E-mail:[email protected]; fax: 206-732-6055.

Copyright © 2003 by the American Association for the Study of Liver Diseases.0270-9139/03/3806-0019$30.00/0doi:10.1016/j.hep.2003.09.024

1458

ticularly if the number of pooled samples is small, poten-tially masking some important gene expression changes.

An additional serious caveat of these studies is the lackof data on normal physiologic variation of gene expressionin nondiseased livers.5 The importance of this problem isclear from a study of 6 genetically identical mice, whichshowed that expression of about 0.8%, 1.9%, and 3.3%of all transcripts normally is variable in the liver, testis,and kidney, respectively.9 Apparently, genetically diversehumans are likely to show much higher normal variationin gene expression than may be observed with in-bredanimals.9 Normal variation in human liver gene expres-sion has been estimated only in one study, which analyzed5 normal liver samples.10 However, these data were ob-tained using filter-based complementary DNA microar-rays and radioactive probe labeling, a technique that hassince been replaced with more advanced microarray for-mats. Thus, a more complete database of information onnormal variation in gene expression, to be subtracted insilico from gene sets associated with pathologic processesor experimental manipulations, would be of great valuefor ensuring accurate interpretation of microarray data.

The major goal of our study was to create a databasebringing together different types of gene expression mea-surements related both to normal physiologic variation innoninfected liver and to HCV-induced pathologic pro-cesses, including cirrhosis and hepatocellular carcinoma.These data were obtained by using complementary DNAmicroarrays containing approximately 13,600 genes. Weidentified a set of genes showing normal physiologic vari-ation in expression and subtracted it in silico from a set ofgenes differentially expressed in cirrhotic liver. In addi-tion, a set of tumor-specific genes also was subtracted,resulting in the identification of 132 potential HCV andliver disease–associated markers. This report analyzespathologic changes in human liver in the context of nor-mal variation in gene expression.

Materials and Methods

Human Liver Tissue Samples. Surgical material wascollected with prior informed consent from patients at theUniversity of Barcelona, Spain, who were treated accord-ing to published strategies.11 This material included nor-mal, noninfected liver samples from 4 patients(designated NL1-NL4), samples of HCV-infected cir-rhotic liver (Child-Pugh grades B and C) from 8 patients,and samples of matched hepatocellular carcinoma tumorand adjacent nontumor cirrhotic tissue from 20 patients.Two independent pools (designated CR1 and CR2), eachcontaining RNA from 4 samples of HCV-infected cir-rhotic livers, were prepared as described previously.12 Ad-

ditional HCV-infected cirrhotic liver tissue was obtainedfrom core needle biopsy specimens collected from cir-rhotic liver explants of 4 HCV-infected patients who un-derwent liver transplantation at the University ofWashington (designated bCL1-bCL4). All transplant re-cipients gave informed consent in writing according toprotocols approved by the Human Subjects Review Com-mittee of the University of Washington. For an addednormal liver reference standard (designated NLct), totalRNA from normal, nondiseased liver (27-year-old Asianman, sudden death) was purchased from Clontech (PaloAlto, CA).

Total RNA Isolation and mRNA Amplification.Frozen tissues were disrupted in Trizol Reagent (Invitro-gen, Carlsbad, CA) using a Polytron homogenizer(PowerGene 700; Fisher Scientific, Pittsburgh, PA), andtotal RNA was isolated according to the Trizol protocol.All total RNA samples, including the reference normalliver ones, were amplified using the RiboAmp RNA Am-plification kit (Arcturus, Mountain View, CA). The qual-ity of amplified RNA was evaluated by capillaryelectrophoresis using an Agilent 2100 Bioanalyzer (Agi-lent Technologies, Palo Alto, CA).

Expression Microarray Format and Data Analysis.Microarray format, protocols for probe labeling, hybrid-ization, slide treatment, and scanning are described onour web site (http://expression.microslu.washington.edu). A single experiment comparing 2 samples was per-formed using the dye label reverse technique with mi-croarrays (13,597 unique IMAGE complementary DNAclones), thus providing mean ratios between the expres-sion levels of each gene in the analyzed sample pair, SDs,and P values.12 All data were entered into a custom-de-signed database, Expression Array Manager, and then up-loaded into Rosetta Resolver System 3.0 (RosettaBiosoftware, Kirkland, WA) and Spotfire (Spotfire, Som-erville, MA). Data normalization and the Resolver Systemerror model specifically developed for our slide format aredescribed on our web site. This web site also is used topublish all primary data in accordance with the proposedstandards.13

Results and Discussion

Microarray Analysis of Normal and Cirrhotic Liv-ers. Each microarray experiment compared 2 mRNAsamples and provided a ratio of expression levels for ap-proximately 13,600 genes. Genes were selected based on 2criteria, a greater than 95% probability of being expresseddifferentially (P � .05), and a fold change of 2-fold orgreater. To establish variation parameters of the appliedmicroarray format, we performed an experiment compar-

HEPATOLOGY, Vol. 38, No. 6, 2003 SMITH ET AL. 1459

ing the same normal liver RNA sample labeled with bothCy3 and Cy5 dyes (Fig. 1). Totally, the expression ratioswere measured for 6,172 genes, and only 24 of them gotselected as differentially expressed between the identicalsamples based on the first criterion (P � .05). Addition ofthe second criterion of a greater than 2-fold ratio ofchanges decreased the number of false positives to 7, or0.11% of all genes present on the array. This experimentshowed that our statistical approach resulted in less than1% of false positives.

As a starting point for our analyses, we examined nor-mal physiologic variation in liver gene expression by com-paring different samples of normal liver tissue. For theseexperiments, an individual surgical sample (NL1) wasused as a reference for comparison with 3 other surgicalsamples (NL2, 3, and 4) and with a sample obtainedcommercially (NLct). Depending on the experiment, wefound that from 600 to 4,300 genes were expressed dif-ferentially, indicating a significant degree of normal phys-iologic variation. As described later, these data were usedto differentiate between normal variation in gene expres-sion and gene expression changes specifically associatedwith liver disease.

To analyze gene expression changes in cirrhotic liver,we performed a series of experiments in which we com-pared samples of HCV-infected cirrhotic liver with nor-mal liver samples. In 5 of these experiments, we usedindividual core needle biopsy specimens obtained fromcirrhotic liver explants at the time of liver transplantation

(bCL1-bCL4). All biopsy samples were compared withthe same commercial reference sample, NLct, because itwas the only normal liver sample available in quantitiessufficient for multiple experiments. To obtain mRNA, weperformed the same RNA amplification technique for allsamples, including the normal liver references. To assesswhether any bias was introduced by amplification, wecompared the data obtained from 2 experiments that wereperformed using the same initial liver samples (bCL2 andNLct) that both underwent either 1 or 2 rounds of am-plification. As expected, the data from these experimentswere highly consistent. A total of 3,589 and 3,579 geneswere expressed differentially in the experiments using sin-gle- and double-amplified material, respectively. Amongthese genes, over 92% showed the same pattern of changein both experiments, less than 8% were regulated differ-entially in only one experiment, and only 0.3% (10 genes)were anticorrelated (data not shown). Thus, an extraround of RNA amplification had very little effect on geneexpression measurements.

Two additional experiments were performed usingpooled samples of RNA. Aliquots of RNA from surgicalsamples of HCV-infected cirrhotic liver (obtained fromthe University of Barcelona) were mixed to generate 2independent pools of 4 samples each (CR1 and CR2).12 Apool of RNA prepared from normal livers (pNL) was usedas a reference sample for these experiments.12 Comparedwith our analyses using individual biopsy specimens, the

Fig. 1. Estimation of the false-positive rate of the applied microarray format using an experiment that compares the same normal liver referencesample pNL labeled in 2 dyes, Cy3 and Cy5. The Resolver System (Rosetta Biosoftware) plots show the log intensity (average of 4 normalizedmeasurements for 2 dyes) on the x-axis and the log ratio on the y-axis. Each gene/data point is shown with a plus symbol. The left panel plots all6,172 genes analyzed in this experiment. The right panel plots only false positives: genes selected as differentially expressed between the identicalsamples based on a P value of .05 or less and a fold change of 2 or greater. Thus, the number of false positives for this experiment is 7, or 0.11%of all genes present on the microarray.

1460 SMITH ET AL. HEPATOLOGY, December 2003

number of differentially expressed genes was reduced by4- to 5-fold in experiments using pooled samples.

The experiments with pooled samples were used fordata validation by reverse-transcription polymerase chainreaction described in our previous work.12 Briefly, for 50microarray measurements that fit our criteria of a P valueof .05 or less and a fold change of 2 or greater, the false-positive number was 0.

Construction of Gene Sets Specific for HCV-Asso-ciated Liver Disease. Our experiments involved diversecirrhotic liver samples from different patient cohorts,however, our goal was to detect common gene expressionchanges associated with HCV-induced liver disease. Webegan by using hierarchic 2-dimensional clustering,which identified a set of 2,589 genes that were expresseddifferentially in at least 3 of 7 experiments comparingcirrhotic with normal livers (Fig. 2). We then performed aseries of analytical steps to narrow this set and to identifygenes that might serve as potential microarray markers ofHCV-associated liver disease. The first step was to selectgenes that showed common patterns of expression (shownin red in Fig. 2), resulting in the identification of 1,226up-regulated and 1,147 down-regulated genes (Fig. 3,step 1).

In our experiments, about 60% of all genes showingvariation in gene expression among normal livers alsowere expressed differentially in cirrhotic livers; thus, thenext step was to subtract these genes (Fig. 3, step 2). Weeliminated genes that were expressed differentially in atleast 2 of 4 experiments comparing normal liver samples(NL1-NL4). In addition, because all cirrhotic liver biopsysamples (bCL1-bCL4) were compared with the NLct ref-erence sample, genes that were expressed differentially in acomparison of NLct with NL1 also were subtracted. Al-though genes with normally variable expression levelstruly might be associated with a pathologic process, wetook this conservative approach to prevent misinterpreta-tion of the microarray data. For example, mouse calmod-ulin 3 previously was reported to be expresseddifferentially with age14; however, its expression subse-quently was observed to vary normally.9 In our experi-ments, the expression of its human homolog, CALML3,also was highly variable in normal liver (Table 1).

In the next steps of our analysis, we subtracted genesthat also were expressed differentially in hepatocellularcarcinoma (Fig. 3, steps 3 and 4). These genes were iden-tified in previous experiments that compared tumor withnormal tissue and tumor with adjacent cirrhotic tissue.12

Again, although some genes that are expressed differen-tially in tumors also may play a role in cirrhosis, they wereremoved from further analysis to make our final set ofgenes as specific for noncancerous liver cells as possible. As

a result, 501 up-regulated and 206 down-regulated genesremained. The expression of these genes did not vary innormal liver, and these genes were not expressed differen-tially in hepatocellular carcinoma tumors.

As a final step to identify potential markers of HCV-associated liver disease, we required that all genes selectedhave at least some annotation that could be used to classifythem into functional categories. We also required thatthey be expressed differentially in at least 5 of 7 experi-ments that compared cirrhotic with normal liver. Thisresulted in a set of 132 differentially expressed genes (Fig.3, step 5). Among these genes, 44 were up-regulated re-gardless of whether experiments used individual biopsyspecimens or pooled samples (Fig. 4), 43 were up-regu-

Fig. 2. Expression profiles of 2,589 genes differentially expressed inat least 3 of 7 experiments comparing cirrhotic with normal liver.Two-dimensional hierarchic clustering was performed by using ResolverSystem software (Rosetta Biosoftware) with an agglomerative algorithm,complete link heuristic criteria, and correlation without mean subtractionmetric. A full-matrix view is shown and each vertical column representsan independent experiment. CR-v-pNL indicates a pooled cirrhotic liversample versus a pooled normal liver sample. bCL-v-NLct indicates anindividual needle biopsy specimen from a cirrhotic liver versus thecommercial normal liver sample. The fold changes in mRNA levels in thecirrhotic liver relative to normal liver are represented by green bars andred bars, showing decreased and increased levels, respectively. Thecolor scale indicates the magnitude of change. Black bars indicate nochange in gene expression level. Gene clusters showing common expres-sion patterns in most experiments are shown in red. Up-regulated anddown-regulated gene sets are divided by the dotted white line.

HEPATOLOGY, Vol. 38, No. 6, 2003 SMITH ET AL. 1461

lated primarily in the experiments using individual biopsyspecimens (Fig. 5), and 45 were down-regulated (Fig. 6).In general, our analytical strategy might have seriouslylimited the size of the resultant gene set, however, weconsidered it to be essential for accurate data interpreta-tion.

For comparative purposes, we also identified a set of128 genes that were expressed differentially in at least 3 of4 experiments that compared samples of normal liver (apartial listing is shown in Table 1). Included in this nor-mal physiologic variation set are genes that encode com-

ponents of signaling pathways, enzymes of livermetabolism, transcription factors, various plasma compo-nents, and stress-response proteins (Fig. 7). These datareflect the metabolic plasticity and functional heterogene-ity of the normal liver, which is produced by hepatocytesunder the influence of a variety of circulating signals.15 Inthis regard, we observed normal physiologic variation inthe expression of many genes associated with signalingpathways, including several encoding hormones, chemo-kines, cytokines, and growth factors (Table 1).

Compared with the normal physiologic variation set,the HCV- and liver disease–associated gene set showed alarger proportion of differentially expressed genes encod-ing DNA- and RNA-binding proteins (Fig. 7), possiblysuggesting the induction of certain new pathways in thecirrhotic liver. In addition, genes associated with apopto-sis, the cytoskeleton, and lymphocyte activation were rep-resented much more abundantly in the liver disease setthan in the normal variation set. To better define poten-tial markers of HCV-induced liver disease, we furtheranalyzed the different functional categories, as well as spe-cific genes associated with particular cell types, in bothnormal physiologic variation and disease-associated sets.

Gene Expression Patterns Associated With theStress Response. Oxidative stress induced by HCV in-fection is thought to play a key role in the pathogenesis ofliver disease.16 However, we observed that many stressresponse genes were expressed differentially in compari-sons of normal liver samples (Table 1). This observation isconsistent with previous results obtained using either in-bred mice9 or normal human liver tissue10 and supportsthe idea that stress response genes can be a significantsource of individual variation in gene expression. Indeed,it is likely that normal liver is subjected to some type ofstress regardless of the sampling process. Therefore, wesuggest that in the search for potential markers of liverdisease, stress response genes as a group should be avoideddue to the inability to determine a baseline for their ex-pression in normal liver.

Gene Expression Patterns Associated With HepaticStellate Cell Activation and Myofibroblast-Like Cells.The liver contains diverse populations of mesenchymalcells, which include hepatic stellate cells (HSCs), myofi-broblasts (MF), and MF-like cells.17,18 Stress and injuryactivate the conversion of HSCs to MF-like cells, a pro-cess that involves proliferation, acquisition of fibrogenicand contractile capacity, and cytokine release.19 In ourexperiments, a group of genes associated with HSC acti-vation showed normal physiologic variation in expression(Table 1), including 2 genes, CRYAB and CSRP2, shownto be expressed exclusively during early steps of the acti-vation process.20-22 Such normal physiologic variation in-

Fig. 3. Flowchart outlining the analysis strategy used to define poten-tial HCV- and liver disease–associated marker genes. Each of the first 4steps involved 2 procedures, both performed by using Resolver Systemsoftware (Rosetta Biosoftware). The first procedure used 2-dimensionalhierarchic clustering to construct a set of genes that were expresseddifferentially in certain experiments (e.g., comparisons of normal liversamples). The second procedure used the bioset compare feature withthe minus set operator to subtract the constructed gene set from thegene sets up- and down-regulated in cirrhosis. Step 5 was performed asselection of genes having at least some functional annotation anddifferentially expressed in at least 5 of 7 experiments comparing cirrhoticwith normal liver.

1462 SMITH ET AL. HEPATOLOGY, December 2003

dicative of HSC activation may be correlated with thestress response observed in normal livers and describedearlier. On the other hand, the genes associated with HSCactivation were not up-regulated in our experiments withcirrhotic liver, probably because the samples were ob-tained from late stages of liver disease. Thus, to use thegenes related to HSC activation as potential microarray

markers, it would be necessary to analyze their expressionin early stages of cirrhosis.

Differential gene expression in samples of normal liveralso may be associated with variation in the number ofMF-like cells present. Liver MF-like cells express proteinscharacteristic of myogenic and/or neural crest cells,17,18

and the normal physiologic variation set included several

Table 1. Normal Physiologic Variation in Gene Expression Among Normal, Nondiseased Human Livers: A Selected Setof Genes That Are Annotated Sufficiently for Attributing Them to Particular Functional Categories

Gene Description Functional Category Range of Fold Change

ILxR: interleukin 13 receptor, �2 Signaling 6-43.7CCR4: chemokine (C-C motif) receptor 4 Signaling �2.6-4.5CALML3: calmodulin-like 3 Signaling �3-3.5PIASX-BETA: Protein inhibitor of activated STAT X Signaling �2.8-3.5IK: IK cytokine, down-regulator of HLA II Signaling �2-3.5DEPP: decidual protein induced by progesterone Signaling 2.1-15.9EBAG9: estrogen receptor binding site associated, 9 Signaling �2.5-3.1CCR3: chemokine (C-C motif) receptor 3 Signaling �2.5-2.9MST1R: macrophage stimulating 1 receptor Signaling �1.9-2.4RGS7: regulator of G-protein signalling 7 Signaling �5-1.35PDGFRB: platelet-derived growth factor receptor, � Signaling �3.6-�1.8HSF4: heat shock transcription factor 4 Stress response �3.1-3.6MAP3K1: MAP kinase kinase kinase 1 Stress response �3.4-3MAP4K1: MAP kinase kinase kinase kinase 1 Stress response �1.9-3.1CRH: corticotropin-releasing hormone Stress response �2.2-2.9CANPX: calpain-like protease Stress response �2.1-2.8AGPS: alkylglycerone phosphate synthase Stress response �2.1-2.5STIP1: stress-induced phosphoprotein 1 Stress response 2-2.5PLA2G2A/phospholipase A2, group IIA Stress response �5-5.1APP: amyloid � (A4) precursor protein Acute phase response �3.1-�2.1APLP2: amyloid � (A4) precursor-like protein 2 Acute phase response �3.9-�1.7CHIT1: chitinase 1 (chitotriosidase) Macrophages �2.2-3MRC1: mannose receptor, C type 1 Macrophages �4.5-�2.3CD163: CD163 antigen, haemoglobin scavenger Macrophages �4.2-�2.6CHI3L1: chitinase 3-like 1 (cartilage glycoprotein-39) Macrophages �2.6-2.4CAPG: capping protein (actin filament), gelsolin-like Macrophages 2.1-3ENG: endoglin, part of TGF � receptor Endothelial cells 8.3-31.7CRYBA1/crystallin, � A1 HSC activation 2-2.8CRYAB/crystallin, � B HSC activation 2-2.4CSRP2/cysteine and glycine-rich protein 2 HSC activation 2-2.5NOGOR: nogo receptor MF-like cells �2.5-3.4PCDH9: protocadherin 9 MF-like cells �3.7-4.1DSCAM: Down syndrome cell adhesion molecule MF-like cells �1.8-3.1ACTN3: actinin, � 3 MF-like cells �2.1-3.2SEMAY: semaphorin Y MF-like cells �2.3-3.2VLDLR: very low density lipoprotein receptor MF-like cells 2.3-2.5EPHB4: Eph-related receptor tyrosine kinase B4 MF-like cells 1.7-3CRI1: CREBBP/EP300 inhibitory protein 1 MF-like cells 2-8CRB1: crumbs (Drosophila) homolog 1 MF-like cells 2-2.9SYN2: synapsin II MF-like cells 3.6-20MYOZ2: muscle-specific protein MF-like cells �20.9-�2.9MEF-2: myelin gene expression factor 2 MF-like cells �5.3-2.8TNFSF13 or APRIL: tumor necrosis factor superfamily, 13 Apoptosis �2.7-4AVEN: cell death regulator aven Apoptosis �2-3SFRP4: secreted frizzled-related protein 4 Apoptosis �3.8-3FZD8: frizzled (Drosophila) homolog 8 Apoptosis �1.8-2.9SH3GLB2: SH3-containing protein SH3GLB2 Apoptosis 2.3-2.8TNFSF10 or TRAIL: tumor necrosis factor superfamily, 10 Apoptosis �8.5-�2.2COL2A1: collagen, type II, � 1 Extracellular matrix �2.7-3.9KRT7: keratin 7 Extracellular matrix 2.3-3.2

HEPATOLOGY, Vol. 38, No. 6, 2003 SMITH ET AL. 1463

of such genes (Table 1). Our HCV- and liver disease–associated set also contained genes associated with MF-like cells (shown in gray in Figs. 4 and 5), the up-regulation of which might indicate an increased numberof MF-like cells in the cirrhotic liver. Thus, our data sug-gest that differences in the number of MF-like cells may

contribute significantly to variations in gene expression inboth normal and cirrhotic livers.

Gene Expression Patterns Associated With Extra-cellular Matrix Components. Activated HSC and MF-like cells play a major role in extracellular matrixdeposition and remodeling during liver fibrosis and cir-rhosis.17 However, our experiments with cirrhotic liverswere performed using samples obtained at end-stage liverdisease, when collagen synthesis may be normal or evenabsent.23 In our work, we did not observe up-regulation ofcollagen genes. The major changes in extracellular ma-trix–associated gene expression observed at end-stage cir-rhosis (shown in brown in Figs. 4-6) were related to cell-cell and cell-matrix interactions rather than to collagenfibril deposition and extracellular matrix remodeling.

Gene Expression Patterns Associated With ImmuneSystem Cells. The liver contains a large resident and mi-gratory population of immune system cells, includinglymphocytes, macrophages, and Kupffer cells.24 Differ-ences in the number of resident immune system cells alsomight be a significant source of normal physiologic vari-

Fig. 4. Potential gene markers specifically up-regulated in cirrhosisboth in experiments with individual biopsy specimens and pooled sam-ples. The 2-dimensional hierarchic clustering of genes and experimentswas performed using Spotfire software with an agglomerative algorithm,complete link heuristic criteria, and Euclidean metric. Each verticalcolumn represents an independent experiment. CR(1 and 2)-v-pNLindicates a pooled cirrhotic liver sample versus the pooled normal liverreference. bCL(1, 2, 3, and 4)-v-NLct indicates an individual needlebiopsy specimen from cirrhotic liver versus the individual commercialnormal liver reference. NL (2, 3, 4, or ct)-v-NL1 indicates an individualnormal liver sample versus the normal liver reference NL1. EaT-v-pNL andAdT-v-pNL are comparisons of pooled samples of early and advancedtumors versus pooled normal liver, respectively. The clustering of exper-iments with HCV-infected cirrhotic liver is shown with blue lines. The foldchanges in mRNA levels in the experimental sample versus the referenceare represented by green squares and red squares, showing decreasedand increased levels, respectively. The color scale indicates the magni-tude of fold changes. Black squares indicate no change in geneexpression level. Gray squares indicate no data. Genes are indicated bytheir Human Genome Organization names and brief gene descriptions.MF-like cell–associated genes are shown in gray; macrophage-associatedgenes are shown in red; activated lymphocyte-associated genes areshown in purple; extracellular matrix–associated genes are shown inbrown; Cdc42 (cytoskeleton)-associated genes are shown in green;apoptosis-associated genes are shown in blue.

Fig. 5. Potential gene markers specifically up-regulated in cirrhosis inexperiments with individual biopsy specimens. See legend of Fig. 4 fordetails on clustering, the applied color scheme, experiment representa-tion, and gene names. MF-like cell–associated genes are shown in gray;macrophage-associated genes are shown in red; activated lymphocyte-associated genes are shown in purple; extracellular matrix–associatedgenes are shown in brown; apoptosis-associated genes are shown inblue.

1464 SMITH ET AL. HEPATOLOGY, December 2003

ation,9 and, indeed, we observed that a number of genesassociated with macrophage-like cells and Kupffer cellswere expressed differentially in comparisons of normalliver samples (Table 1).

In our analysis of cirrhotic livers, we also observed theup-regulation of several macrophage-associated genes(shown in red in Figs. 4 and 5). Two of these genes areshown to be associated with Kupffer cells, includingLGALS3 (shown in red in Fig. 4), which regulates cellgrowth and apoptosis,25 and SGK (shown in red in Fig.5), which is strongly induced in patients with chronicviral hepatitis.26 The up-regulation of LGALS3 and SGKgene expression might reflect Kupffer cell activation in theHCV-infected cirrhotic liver, providing potential mark-ers for microarray diagnostics.

During inflammatory liver disease, including cirrhosis,the number of lymphocytes infiltrating the liver increasesdramatically.24 As expected, our analysis of cirrhotic liversshowed up-regulation of a gene set expressed in activatedlymphocytes (shown in purple in Figs. 4 and 5). Genesfrom this set encoded components of the lymphocyte-mediated immune response, including proteins from the

class II major histocompatibility complex and the immu-noglobulin family; and the signaling proteins SCYA19and MIG. The observed up-regulation of these genesmost likely is associated with HCV-associated chronicinflammation and liver infiltration with activated lym-phocytes, and may be a useful marker for monitoringthese processes.

Gene Expression Patterns Associated With the Cy-toskeleton and Cdc42 Signaling Components. Theproportion of differentially expressed genes encodingcomponents of the cytoskeleton was almost twice as highin the HCV- and liver disease–associated set than in thenormal physiologic variation set (Fig. 7). Two cytoskele-ton-related genes from the Cdc42 signaling pathway, IQ-GAP1 and SPEC2, were specifically up-regulated in allcirrhotic liver samples (shown in green in Fig. 4). It is wellestablished that Rho guanosine triphosphatase Cdc42plays a central role in polarizing and organizing the cy-toskeleton.27 Both IQGAP1 and SPEC2 bind to Cdc42and work in a complex that reorganizes the cytoskeleton,leading to cell polarization.28,29 The observed up-regula-tion of these 2 Cdc42 effectors might be correlated withthe expression of cell adhesion–related genes describedearlier. Both groups of genes may participate in liver tissueremodeling during cirrhosis.30 We suggest that the genesfrom both groups might be used as potential markers ofHCV-associated liver disease.

Gene Expression Patterns Associated With Apopto-sis and Bcl-2–Related Signaling. Similar to previousstudies,10 our data also showed normal physiologic varia-tion in the expression of several apoptotic genes (Table 1),including TNFSF10 (or TRAIL), which is involved in theapoptosis of activated HSC.31 Two other apoptotic genes

Fig. 6. Potential gene markers specifically down-regulated in cirrhosis.See legend of Fig. 4 for details on clustering, the applied color scheme,experiment representation, and gene names. Extracellular matrix–asso-ciated genes are shown in brown; apoptosis-associated genes are shownin blue.

Fig. 7. Comparison of the gene set showing normal physiologicvariation in expression (hatched bars) with the HCV- and liver disease–associated gene set (black bars). Bars indicate proportions of genesbelonging to particular functional categories.

HEPATOLOGY, Vol. 38, No. 6, 2003 SMITH ET AL. 1465

that showed normal physiologic variation in expressionencoded frizzled-related proteins (Table 1). AlthoughTNFSF10 and some frizzled-related proteins of the Wntpathway previously were found to be up-regulated inHCV-induced cirrhosis,8 our results argue for a more so-phisticated approach toward interpreting apoptotic path-ways in cirrhotic liver by incorporating data on normalvariations in gene expression.

Our HCV- and liver disease–associated set contained 3genes related to the Bcl-2 family (shown in blue in Figs.4-6). Anti-apoptotic functions of Bcl-2 depend on theheterodimerization of this protein with other familymembers.32 In our experiments, Bcl-2 and a related gene,BNIP3L, were up-regulated (shown in blue in Figs. 4 and5). The putative Bcl-2 interactor, pro-apoptotic proteinBid, specifically was down-regulated in the diseased liver(shown in blue in Fig. 6). Bid is thought to counter theprotective effects of Bcl-2 and to induce apoptosis oninteraction with specific viral and cellular proteins.33

Thus, Bcl-2 up-regulation and Bid down-regulationmight indicate a protective anti-apoptotic mechanism in-duced in the HCV-associated disease.

Our data on the coordinated expression of Bcl-2–re-lated apoptotic genes differs from a previous report thatthe Bcl-2 gene is up-regulated only in autoimmune hepa-titis–induced cirrhosis, and not in HCV-induced liverdisease.8 Another discrepancy between our data and theprior report is the down-regulation of the pro-apoptoticgene, ARP3 (shown in blue in Fig. 6), which previouslywas found to be up-regulated in HCV-induced cirrhosis.8

We believe that the sample pooling approach used in theprevious study8 might be responsible for the observeddiscrepancies. We suggest that a preferable approach toimproving the accuracy and reproducibility of microarrayanalyses is to increase the number of individual samplesanalyzed and to bring together experiments of differenttypes. Analysis of additional human liver samples there-fore may help to reveal the role of Bcl-2–mediated anti-apoptotic signaling in liver disease.

Gene Expression Patterns Associated With HCVInduction of the Antiviral Response. By using previ-ously published gene expression studies from our labora-tory,34 we constructed a set of 241 genes related to virus-host interactions. These genes were identified asdifferentially expressed in at least one type of in vitroexperiment with cultured human cells infected with var-ious viruses, or treated with double-stranded RNA or in-terferon.35,36 A subset of 24 genes from this set wasexpressed differentially in at least 4 of 7 experiments com-paring cirrhotic with normal liver (Fig. 8). The expressionof many of the genes was up-regulated by 5- to 10-fold inindividual cirrhotic biopsy specimens. Several of these

genes also showed normal physiologic variation in expres-sion, but the ratio of change was always less than 3-fold.

Some expression changes observed for this subset ofgenes might be attributed to the infiltration of cirrhoticliver with activated lymphocytes and natural killer cells(e.g., the human leukocyte antigen genes in Fig. 8). How-ever, up-regulation of other genes might reflect the anti-viral response of liver hepatocytes induced by chronicHCV infection. Previous studies analyzing liver fromHCV-infected chimpanzees37 or humans38 showed up-regulation of several genes from this subset, includingSTAT1, IFI27, MX1, and IFIT1 (shown in red in Fig. 8).Our gene set associated with the anti-HCV responsemight have clinical relevance. However, to define this setaccurately as HCV specific, it is necessary to analyze theexpression of these genes in patients with cirrhosis unre-lated to HCV.

ConclusionsWe have identified a set of differentially expressed

genes associated with HCV and liver disease and haveanalyzed it in the context of normal physiologic variationin liver gene expression. Our analyses provide a molecularportrait of gene expression changes associated with vari-ous cell types and pathologic processes occurring in the

Fig. 8. Differential expression of a subset of genes related to virus-host interactions in HCV-induced cirrhosis. See legend of Fig. 4 for detailson 2-dimensional clustering, the applied color scheme, experimentrepresentation, and gene names. Genes that have been suggestedpreviously to be up-regulated in HCV-induced liver disease are shown inred.

1466 SMITH ET AL. HEPATOLOGY, December 2003

HCV-infected cirrhotic liver. The present work has beenperformed with a relatively small group of cirrhotic liversamples, thus the resultant picture is certainly incomplete,and a different analytical approach applied to the samedataset might produce a somewhat different gene list. Aswe continue to add related experiments to our gene ex-pression database, we should be able to continually rean-alyze our data in relation to new contexts and to developincreasingly sophisticated gene expression indicators ofspecific types or stages of liver disease.

References1. Alter HJ, Seeff LB. Recovery, persistence and sequelae in hepatitis C virus

infection: a perspective on long-term outcome. Semin Liver Dis 2000;20:17-35.

2. Fattovich G, Giustina G, Degos F, Tremolada F, Diodati G, Almasio P,Nevens F, et al. Morbidity and mortality in compensated cirrhosis type C:a retrospective follow-up study of 384 patients. Gastroenterology 1997;112:463-472.

3. Saadeh S, Cammell G, Carey WD, Younossi Z, Barnes D, Easley K. Therole of liver biopsy in chronic hepatitis C. HEPATOLOGY 2001;33:196-200.

4. Branch AD, Walewski JL. The coming impact of gene expression profilingon the diagnosis and treatment of HCV-associated liver disease. AntiviralRes 2001;52:173-179.

5. Shackel NA, Gorrell MD, McCaughan GW. Gene array analysis and theliver. HEPATOLOGY 2002;36:1313-1325.

6. Honda M, Kaneko S, Kawai H, Shirota Y, Kobayashi K. Differential geneexpression between chronic hepatitis B and C hepatic lesion. Gastroenter-ology 2001;120:955-966.

7. Shackel NA, McGuinness PH, Abbott CA, Gorrell MD, McCaughanGW. Identification of novel molecules and pathogenic pathways in pri-mary biliary cirrhosis: cDNA array analysis of intrahepatic differential geneexpression. Gut 2001;49:565-576.

8. Shackel NA, McGuinness PH, Abbott CA, Gorrell MD, McCaughanGW. Insights into the pathobiology of hepatitis C virus-associated cirrho-sis: analysis of intrahepatic differential gene expression. Am J Pathol 2002;160:641-654.

9. Pritchard CC, Hsu L, Delrow J, Nelson PS. Project normal: definingnormal variation in mouse gene expression. Proc Natl Acad Sci U S A2001;98:13266-13271.

10. Yano N, Habib NA, Fadden KJ, Yamashita H, Mitry R, Jauregui H, KaneA, et al. Profiling the adult human liver transcriptome: analysis by cDNAarray hybridization. J Hepatol 2001;35:178-186.

11. Bruix J, Llovet JM. Prognostic prediction and treatment strategy in hepa-tocellular carcinoma. HEPATOLOGY 2002;35:519-524.

12. Smith MW, Yue ZN, Geiss GK, Sadovnikova NY, Carter VS, Boix L,Lazaro CA, et al. Identification of novel tumor markers in hepatitis Cvirus-associated hepatocellular carcinoma. Cancer Res 2003;63:859-864.

13. Brazma A, Hingamp P, Quackenbush J, Sherlock G, Spellman P, Stoeck-ert C, Aach J, et al. Minimum information about a microarray experiment(MIAME) toward standards for microarray data. Nat Genet 2001;29:365-371.

14. Lee CK, Klopp RG, Weindruch R, Prolla TA. Gene expression profile ofaging and its retardation by caloric restriction. Science 1999;285:1390-1393.

15. Christoffels VM, Sassi H, Ruijter JM, Moorman AF, Grange T, LamersWH. A mechanistic model for the development and maintenance of por-tocentral gradients in gene expression in the liver. HEPATOLOGY 1999;29:1180-1192.

16. Erhardt A, Hassan M, Heintges T, Haussinger D. Hepatitis C virus coreprotein induces cell proliferation and activates ERK, JNK, and p38 MAPkinases together with the MAP kinase phosphatase MKP-1 in a HepG2Tet-Off cell line. Virology 2002;292:272-284.

17. Eng FJ, Friedman SL. Fibrogenesis I. New insights into hepatic stellate cellactivation: the simple becomes complex. Am J Physiol 2000;279:G7-G11.

18. Cassiman D, Libbrecht L, Desmet V, Denef C, Roskams T. Hepatic stel-late cell/myofibroblast subpopulations in fibrotic human and rat livers.J Hepatol 2002;36:200-209.

19. Friedman SL. Molecular regulation of hepatic fibrosis, an integrated cel-lular response to tissue injury. J Biol Chem 2000;275:2247-2250.

20. Lang A, Schrum LW, Schoonhoven R, Tuvia S, Solis-Herruzo JA, Tsuka-moto H, Brenner DA, et al. Expression of small heat shock protein alphaB-crystallin is induced after hepatic stellate cell activation. Am J Physiol2000;279:G1333-G1342.

21. Cassiman D, Roskams T, van Pelt J, Libbrecht L, Aertsen P, Crabbe T,Vankelecom H, et al. Alpha B-crystallin expression in human and rathepatic stellate cells. J Hepatol 2001;35:200-207.

22. Weiskirchen R, Moser M, Weiskirchen S, Erdel M, Dahmen S, BuettnerR, Gressner AM. LIM-domain protein cyst. Biochem J 2001;359:485-496.

23. Rojkind M. Fibrogenesis in cirrhosis. Potential for therapeutic interven-tion. Pharmacol Ther 1992;53:81-104.

24. Lalor PF, Shields P, Grant A, Adams DH. Recruitment of lymphocytes tothe human liver. Immunol Cell Biol 2002;80:52-64.

25. Liu FT, Patterson RJ, Wang JL. Intracellular functions of galectins. Bio-chim Biophys Acta 2002;1572:263-273.

26. Fillon S, Klingel K, Warntges S, Sauter M, Gabrysch S, Pestel S, TanneurV, et al. Expression of the serine/threonine kinase hSGK1 in chronic viralhepatitis. Cell Physiol Biochem 1903;12:47-54.

27. Brunner D. How to grab a microtubule on the move. Dev Cell 2002;3:2-4.28. Fukata M, Kaibuchi K. Rho-family GTPases in cadherin-mediated cell-

cell adhesion. Nat Rev Mol Cell Biol 2001;2:887-897.29. Pirone DM, Fukuhara S, Gutkind JS, Burbelo PD. SPECs, small binding

proteins for Cdc42. J Biol Chem 2000;275:22650-22656.30. Welsh CF, Assoian RK. A growing role for Rho family GTPases as inter-

mediaries in growth factor- and adhesion-dependent cell cycle progression.Biochim Biophys Acta 2000;1471:M21-M29.

31. Taimr P, Higuchi H, Kocova E, Rippe RA, Friedman S, Gores GJ. Acti-vated stellate cells express the TRAIL receptor-2/death receptor-5 andundergo TRAIL-mediated apoptosis. HEPATOLOGY 2003;37:87-95.

32. Gross A, McDonnell JM, Korsmeyer SJ. BCL-2 family members and themitochondria in apoptosis. Genes Dev 1999;13:1899-1911.

33. Belov GA, Romanova LI, Tolskaya EA, Kolesnikova MS, Lazebnik YA,Agol VI. The major apoptotic pathway activated and suppressed by polio-virus. J Virol 2003;77:45-56.

34. Geiss GK, An MC, Bumgarner RE, Hammersmark E, Cunningham D,Katze MG. Global impact of influenza virus on cellular pathways is medi-ated by both replication-dependent and -independent events. J Virol 2001;75:4321-4331.

35. Geiss G, Ge J, Guo J, Bumgarner R, Katze MG, Sen GC. A comprehensiveview of gene expression by double-stranded RNA-mediated cell signaling.J Biol Chem 2001;276:30178-30182.

36. Geiss GK, Salvatore M, Tumpey TM, Carter VS, Wang X, Basler CF,Taubenberger JK, et al. Cellular transcriptional profiling in influenza Avirus-infected lung epithelial cells: the role of the nonstructural NS1 pro-tein in the evasion of the host innate defense and its potential contributionto pandemic influenza. Proc Natl Acad Sci U S A 2002;99:10736-10741.

37. Lanford RE, Bigger C. Advances in model systems for hepatitis C virusresearch. Virology 2002;293:1-9.

38. Patzwahl R, Meier V, Ramadori G, Mihm S. Enhanced expression ofinterferon-regulated genes in the liver of patients with chronic hepatitis Cvirus infection: detection by suppression-subtractive hybridization. J Virol2001;75:1332-1338.

HEPATOLOGY, Vol. 38, No. 6, 2003 SMITH ET AL. 1467