The goal of the current study was to provide complete coverage of the liver transcriptome with human probes corresponding to every gene expressed in embryonic, adult, and/or cancerous liver. We developed dedicated tools, namely, the Liverpool nylon array of complementary DNA (cDNA) probes for approximately 10,000 nonredundant genes and the LiverTools database. Inflammation-induced transcriptome changes were studied in liver tissue samples from patients with an acute systemic inflammation and from control subjects. One hundred and fifty-four messenger RNAs (mRNA) correlated statistically with the extent of inflammation. Of these, 134 mRNA samples were not associated previously with an acute-phase (AP) response. The hepatocyte origin and proinflammatory cytokine responsiveness of these mRNAs were confirmed by quantitative reverse-transcription polymerase chain reaction (Q-RT-PCR) in cytokine-challenged hepatoma cells. The corresponding gene promoters were enriched in potential binding sites for inflammation-driven transcription factors in the liver. Some of the corresponding proteins may provide novel blood markers of clinical relevance. The mRNAs whose level is most correlated with the AP extent (P < .05) were enriched in intracellular signaling molecules, transcription factors, glycosylation enzymes, and up-regulated plasma proteins. In conclusion, the hepatocyte responded to the AP extent by fine tuning some mRNA levels, controlling most, if not all, intracellular events from early signaling to the final secretion of proteins involved in innate immunity. Supplementary material for this article can be found on the HEPATOLOGY website (http://interscience.wiley.com/jpages/0270-9139/suppmat/index.html). (HEPATOLOGY 2004;39:353–364.)
Three major objectives drive the current interest in human transcriptome analysis. First, messenger RNA (mRNA) profiling in a pathologic context is used as a molecular signature of diagnostic and/or prognostic relevance, regardless of the functions of the cognate proteins. Second, some mRNA changes may point to proteins that are directly involved in a disease process. Third, mRNA changes can be studied in an integrative context in which the so-called “guilt-by-association” approach aims at identification of all the participants in a given signaling cascade or metabolic pathway.1–3 Progress in any of these areas with array technology4 depends on probe diversity. Therefore, the use of a pan-genomic set of probes is the ideal option. However, in practice, such an array is impractical given the current uncertainties about the number and identity of human genes,5, 6 the exponentially growing complexity of data analysis that results from a linearly increasing number of probes,7 and the cost of a pan-genomic probe set. Nevertheless, maintaining a high probe diversity for studies in a given cell type or tissue context is required for an integrative approach and should also help discover many candidate genes whose hallmark appears to be a tissue-restricted expression.8 Overall, a probe selection that results in a virtually complete coverage of the transcriptome in a given cell type or tissue is likely to be the best choice for the time being. Yet, such a tool has seldom been developed.
The liver contains a large number of transcribed genes whose products participate in a vast array of vital and organ-specific functions as well as organ-restricted properties such as a high capacity to regenerate.9 In response to chronic alcohol abuse or hepatitis B or C virus infection, the liver may undergo major tissue modifications that result in cirrhosis and subsequent hepatocellular carcinoma.10 Therefore, the liver ranks high among those whose transcriptome richness may help decipher as yet unknown disease markers, critical gene regulations, and novel protein functions.11, 12 In addition, the acute phase (AP) of a systemic inflammation up-regulates or down-regulates many liver-expressed genes involved in innate immunity and coding, for instance, for the positive or negative plasma acute phase proteins (APP).13–16 However, a global view of the AP-induced changes in the liver transcriptome has not yet been obtained in humans in vivo. We report the development of an array based on selected human complementary DNA (cDNA) probes that correspond to approximately 10,000 nonredundant genes and specifically cover the liver transcriptome. This allowed us to identify the liver mRNAs whose abundance best correlates with the extent of an acute, systemic inflammation in humans.
mRNA, messenger RNA; AP, acute phase; APP, acute phase protein; cDNA, complementary DNA; Q-RT-PCR, quantitative reverse-transcription-polymerase chain reaction; Hs., Homo sapiens; TF, transcription factor; CRP, C-reactive protein; SAA, serum amyloid A; Clqβ, β chain of complement Clq; TSF, transferrin; MIAME, minimum information about a microarray experiment; ALB, albumin; IL, interleukin; C/EBP, CCAAT-enhancer binding protein; NK4, natural killer cell transcript 4; IGFBP2, insulin-like growth factor binding protein 2; MAP4K4, mitogen-activated protein kinase kinase kinase kinase 4; NFκB, nuclear factor κB; STAT, signal transducer and activator of transcription; TRAF5, tumor necrosis factor receptor-associated factor 5.
Materials and Methods
Human Subjects and RNA Sources.
Total RNA samples from human fetal liver specimens (15-24 week-old fetuses) or adult brain were obtained from Clontech (Palo Alto, CA). Sections of normal tissue taken from surgically resected livers were obtained from the digestive surgery unit of Charles Nicolle Hospital (Rouen, France) under strict anonymity. The diagnosis was either a primary tumor or a hepatic metastasis that was detected in the follow-up of a nonhepatic carcinoma. A matched blood sample was obtained before surgery. In some instances, a hepatocyte-enriched fraction (<3% contamination by nonparenchymal cells) was immediately prepared as described previously.17 Other tissue samples were obtained from several units in the same hospital. According to French law and ethical guidelines, no informed consent is requested before analysis of RNA samples from resected tissue specimens that would otherwise be discarded. The culture and stimulation of Hep3B hepatoma cells and quantitative reverse- transcription polymerase chain reaction (Q-RT-PCR) of mRNA are fully detailed as a supplementary material on the HEPATOLOGY website (http://interscience.wiley.com/jpages/0270-9139/suppmat/index.html), as well as on our website (www.lille.inserm.fr/u519/coulouarnetal03.html).
Selection of Homo sapiens Clusters and Promoter Sequences.
From the Unigene database (ftp://ftp.ncbi.nih.gov/repository/Unigene), a parsing of Homo sapiens (Hs.) data files (build 129, June 2003) was performed with locally imported flat files. The parsing algorithm implemented in PERL (available upon request) allowed us to select a series of Hs. clusters with the single following criterion: expression must occur at least in the liver. Elsewhere, the Unigene Library Browser along with the whole set of cDNA libraries listed in Unigene allowed us to select every library that obeyed the following single criterion: the library must be constructed from a human liver-related tissue sample. Every Hs. cluster contained in any such library was first retained. A PERL program was used to establish a list of nonredundant Hs. clusters from all the above sources. Finally, a bibliographic search made with standard online tools still identified further Hs. clusters. Our final Hs. cluster selection also encompassed the cDNAs for 38 housekeeping genes.18 It is available upon request.
Promoter sequences from Hs. cluster-defined genes were retrieved from mRNA / gene alignments with the Evidence Viewer tool (http://www.ncbi.nlm.nih.gov/LocusLink/). A set of control promoter sequences were made of Hs. cluster-defined genes that were not AP responsive in the current study and were chosen on the sole basis of promoter sequence availability. This analysis was performed over the first 5 kb of DNA sequence upstream of the transcription start site, provided these 5-kb sequence were available. Potential transcription factor (TF) binding sites were searched in the promoter sequences with the MatInspector and TRANSFAC tools.19
Probe Selection, Array Preparation, and Hybridization.
The selection, amplification, and arraying of cDNA clones, [α33P]dCTP labeling, and hybridization of total RNAs and image analysis are detailed as a supplement on the HEPATOLOGY website and our website.
Data Normalization, Filtering, Statistical Analysis, and Final Data Handling.
To allow for comparison between images, normalization was based on the mean of the signals provided by the complete set of spots per image. All data in the current study were obtained from at least three separate hybridizations per RNA sample and the genes were identified as expressed if at least two hybridizations provided a positive signal. For every probe, the signals obtained under two different conditions (i.e., A vs. B) were expressed as the difference (normalized signal in condition A − normalized signal in condition B) and considered to be significantly induced or repressed (folds) if this difference was outside a CI (P < .05) calculated20 from the entire data set. To select the liver mRNAs that were regulated in patients with an acute inflammation versus controls, the value for any given mRNA from every individual with an inflammation was compared with the mean value obtained from the control set. All statistical analyses were performed with the R software.21 Hierarchical clustering was performed with the Cluster and Tree View software and the uncentered correlation and complete linkage clustering options were used.22
This database utilizes a MySQL relational database server, an Apache web server, and PHP. Accordingly, it can be queried via an internet browser under any operating system. The data are gathered within constraint-linked tables. A module written in the PERL language allows the set of data contained in the “cDNA probes / array design” section of LiverTools to be updated weekly by a connection to NCBI (National Center for Biotechnology Information). LiverTools complies with minimum information about a microarray experiment (MIAME) recommendations23 and is accessible upon request. The functions of proteins encoded by various mRNAs (see the Results section) were retrieved from the LocusLink (www.ncbi.nlm.nih.gov/LocusLink) and OMIM (www.ncbi.nlm.nih.gov/omim/) databases.
Liverpool and LiverTools: a Liver-Oriented cDNA Array and Associated Database.
Our goal was to provide extensive coverage of the human liver transcriptome. We used a single stringent criterion to select as many nonredundant genes as possible, namely, they must be expressed at least in the human liver under normal or pathologic conditions. As shown in Fig. 1, the results of our in silico searches (see Materials and Methods) eventually provided 12,638 nonredundant Unigene Hs. clusters. The corresponding genes evenly cover all human chromosomes as their locations, when known, strongly correlate (r = 0.96, n = 5,765, P < 10−4) with the overall gene frequency per human chromosome (www.ncbi.nlm.nih.gov/genome/guide/human/HsStats.html) and do not indicate any preference for given chromosomes (not detailed). This is also true (data not shown) for a subset of 805 located Hs. clusters with a liver-restricted expression (detailed below), which is in keeping with the lack of tissue-specific gene clustering on chromosomes.24
cDNA clones covering the Hs. clusters were purchased from the IMAGE consortium (RZPD, Berlin, Germany). However, for a limited number of Hs. clusters, no IMAGE clone was available and 15.2% of the 12,493 cDNA clones that were purchased did not provide a satisfactory PCR product. Eventually, our nylon array, dubbed Liverpool, typically harbored 13,824 probes. They included a set of usable cDNA probes covering 9,858 nonredundant Hs. clusters (Unigene build 155) as well as a large set of control spots (Fig. 1). Quality controls were performed. First, because many IMAGE clones are suspected of misidentification, a limited number (n = 686) of human cDNA clones was controlled by end sequencing and an overall misidentification rate was calculated. This rate was 6.9% (out of 131 resequenced clones) for the subset of clones pertaining to a limited IMAGE population previously sequence verified at NCBI (ftp://image.llnl.gov/image/clones_verified) whereas it was 11.9% (out of 555 resequenced clones) for the subset of clones pertaining to the major set of non-NCBI–verified clones. These rates compare quite favorably with earlier estimates made from other IMAGE clones.4 Second, our option of a global background substraction is straightforward but requires a reproducible signal to be obtained with multiple copies of a given cDNA probe spotted over the entire array. This was verified over a large range of hybridization signals with various probes whose coefficient of variation usually was 10% to 12% (see supplementary Table s1A online (http://interscience.wiley.com/jpages/0270-9139/suppmat/index.html), as well as on our web site [www.lille.inserm.fr/u519/coulouarnetal03.html]). In addition, we verified (data not shown) that the genes and gene clusters highlighted in the current study were in no way associated with probes spatially clustered onto the filters or to local variations of the background that may increase the false discovery rate.25 Third, many IMAGE clones are partial cDNAs or expressed sequence tags randomly located within a full-length cDNA, whereas our labeled targets primed with an anchored oligo(dT)VN primer preferentially covered the 3′ end of the cognate mRNA. Therefore, we verified that a difference in a given mRNA level between targets was consistantly detected whatever the location of a partial cDNA probe within the corresponding full-length cDNA (see our supplementary Table s1B online). Fourth, when comparing two complex targets in terms of induction/repression of a given mRNA level, relying on an arbitrary cutoff for a significant variation of the ratio (mRNA level in condition A / mRNA level in condition B) has been highly criticized.26 To identify significantly different mRNA levels between samples, we used a statistically valid CI that varies with the absolute signal level (Fig. 2).
To store all the information and other data, such as clinical data, we developed the LiverTools database (Fig. 3). The data are entered into six sections that cover the MIAME recommendations.23
Tissue or Cell Type-Dependent Expression of Liver-Expressed Genes.
Liverpool was probed with total RNA samples from various sources and the presence versus absence of a positive signal was recorded for every gene tested. Overall, the probes for 8,921 Hs. clusters (90.5% of all tested clusters) provided a positive signal with at least one human tissue sample, whether liver or other organ related. As illustrated in Fig. 4, few probes (subset A, 1,359 Hs. clusters) did not provide a positive signal with at least one liver-related sample, a result that supports our liver-oriented probe selection. Most likely, the gene in subset A are expressed in the liver sample to an extent that is below our detection threshold. Another subset (subset C, 880 Hs. clusters) with a relatively limited size comprises genes that exhibit a liver-restricted expression and are mostly involved in amino acid and lipid metabolism, innate immunity, energy transformation, and detoxication. Finally, a major gene subset whose expression is found in every tissue sample tested (subset B) can be defined as being mostly composed of housekeeping genes. In agreement with this, subset B contains 38 known housekeeping genes.18
We also investigated whether the liver transcriptome exhibits any major difference when compared with the transcriptome of a single cell type, namely, the hepatocyte. When matching a crude human liver sample versus a hepatocyte-enriched fraction (<3% nonparenchymatous cells) purified from the same sample, we found that the homologous mRNAs from both sources were expressed to the same relative extent (r = 0.92, n = 12,493, P < 10−4). This is consistent with the finding that the hepatocytes are mRNA-rich cells and account for 80% to 90% of the liver cell mass.27
AP-Driven Genes in Human Liver In Vivo.
We used Liverpool to gain a genomewide insight of the AP-associated events in the liver tissue samples obtained from patients experiencing AP. Based on the number of abnormalities within a set of eight biologic data measured at the time of surgery, the patients listed in Table 1 were given an AP score. They were divided into two subsets with a strong (patients 1–3) or moderate AP score (patients 4–6) and one subset of AP-free, control individuals (patients 7–10). Selecting genes whose mRNA level was abnormally high or low in at least two AP patients compared with its mean level in all four control patients, we identified 772 nonredundant genes. The overall data are presented as a bidimensional hierarchical clustering (Fig. 5A ; the complete data set is available as Table s5 online). These data are reliable based on (1) the presence of numerous Hs. clusters corresponding to hepatic mRNAs whose levels are known to be either up-regulated by the AP such as C-reactive protein (CRP), orosomucoid, serum amyloid A (SAA) 1, fibrinogen α, β, and γ chains, phospholipase A2, cystathionase, annexin A1, the β chain of complement C1q (Clqβ), and the α chain of complement C4-binding protein, or down-regulated such as albumin (ALB), transferrin (TSF), transthyretin, alcohol dehydrogenase, and selenoprotein P14 (also visit our web site for AP genes at www.lille.inserm.fr/u519/thematiques/equipe1/souryetal/index.html); (2) the unequal numbers of up-regulated and down-regulated genes (59% vs. 41%), in excellent agreement with earlier studies in humans and rats28 (also visit our web site for AP genes); and (3) when applicable, the tight clustering of several cDNA probes corresponding to the same Hs. cluster as illustrated with complement C1qβ, SAA 1, TSF, or orosomucoid (Fig. 5A,B). As a quality control, Q-RT-PCR of several mRNAs whose levels measured by arrays exhibited extensive changes between our 10 liver samples was performed. In all instances, an excellent correlation (P ≤ .01) of the values obtained by arrays versus Q-RT-PCR was found: CRP, r = 0.90; haptoglobin, r = 0.76; ALB, r = 0.82.
Table 1. Biologic Data in Patients With an Acute, Systemic Inflammation and Control Patients
The unsupervised clustering based on the complete set of 772 genes identified three major patient groups (patient 1 vs. patients 2, 3, 6 vs. patients 4, 5) that did not match the moderate versus strong AP score (Fig. 5A). We then determined whether one or more functionally defined gene subsets would better cluster patients 1 to 3. Generally known protein functions allow the division of the Liverpool gene list into 42 main subsets29 (see details in our supplementary Figure s1 online). Excluding, from the 772 genes identified above, those coding for orphan or putative proteins and next dividing the remaining 314 genes into functionally defined subsets resulted in a clear-cut clustering of patients 1 to 3 in one instance only, that is, when considering the so-called “anti-pathogen response”-associated genes (47 genes in our study; Fig. 5B). The anti-pathogen function immediately argues against a selection of this subset by chance. Some of the genes in this subset (italicized names in Fig. 5B) code for proinflammatory cytokines (e.g., interleukin [IL]-8), TFs (e.g., CCAAT-enhancer binding protein [C/EBP]-β), and several complement components and APPs (e.g., CRP, SAA 1, orosomucoid) that are AP regulated.14, 30 Some other gene products have not been previously associated with AP (names written in blue block letters in Fig. 5B) and extend the number of AP-regulated anti-pathogen response proteins. These observations underscore that, in the human liver, the so-called anti-pathogen response proteins largely overlap with the AP-regulated proteins.
Novel Markers of AP Extent.
Cumulated bibliographic data indicate that fewer than 200 AP-regulated genes have been identified in the liver (see our web site for AP genes). Therefore, many of the 772 genes shown in Fig. 5A have not been previously associated with the liver response to AP and provide novel potential markers for this condition. However, it may be argued that some of the mRNA regulations observed in our AP patients may not result from the inflammatory syndrome. Therefore, a positive or negative correlation between the extent of change in a given mRNA level and the extent of inflammation based on the AP score were used to rank the 772 genes. The resulting list of 154 genes for which such a correlation is statistically significant (r > 0.63 or r < − 0.63, P < .05) is illustrated in Table 2 and is accessible as supplementary Table s2 online. It is noteworthy that the identity of all genes identified at this stage was further checked by resequencing the corresponding IMAGE clones used as probes.
Table 2. Correlation Between the Extent of AP and Various Hepatic mRNA Levels
These 154 genes contain a subset of 20 genes that have long been known to be AP regulated. Of these 20 mRNAs, 12 code for positive APPs (60%) and display an excellent correlation with the AP score (e.g., C1qβ, SAA 4, CRP; r ≥ 0.70). In contrast, only 4 of these 20 mRNAs are down-regulated and only one codes for a plasma protein (IL-4). Strikingly, none of the mRNAs for well-known negative APPs (e.g., ALB, TSF, transthyretin) was found.
These 154 genes further contain a set of 134 novel genes. As illustrated in Table 2, this set includes 70 known genes whose expression had not yet been shown to be AP modulated, as well as 64 genes for putative proteins. Many of these 134 novel mRNAs, most of them down regulated, have a correlation value greater than ±0.8. Therefore, they are excellent candidates as novel markers of clinical interest. We then determined whether some of the corresponding proteins are secreted as they would lend themselves to quantitation in body fluids for diagnosis and prognosis purposes. To address this, we took advantage of another study in which protein secretion was suggested from a combination of ontology and sequence-based analyses.31 Matching these data with our 134 AP-regulated mRNAs identified five mRNAs coding for secreted proteins (e.g., natural killer cell transcript 4 [NK4], insulin-like growth factor binding protein 2 [IGFBP2], transforming growth factor-β3, lamininα5, and collagenIVα1).
Hepatocyte Origin and Cytokine-Regulated Abundance of the Novel mRNAs.
We verified that the 134 newly identified mRNAs were synthesized in hepatocytes in a cytokine-dependent manner. This was proven by measuring some of these mRNAs in human Hep3B hepatoma cells stimulated in vitro with a proinflammatory, cytokine-enriched conditioned medium versus a negative control medium. This approach was proven to be reliable by the expected up-regulation of the CRP mRNA level and the down-regulation of the ALB mRNA level after 6 or 16 hours of stimulation (Fig. 6, first 2 panels). Moreover, the levels of the six other mRNAs of interest (NK4, IGFBP2, nuclear receptor subfamily 1, group 2, member 2 [NR1D2], protection of telomeres 1, [POT1], MAP4K4, CGI-41) were up-regulated or down-regulated in the conditioned medium-challenged Hep3B cells (Fig. 6), in agreement with the data obtained for the patients in the current study. Only one mRNA (tumor necrosis factor-associated factor 5 [TRAF5]) did not exhibit a significant modulation in Hep3B cells, which may result from the limited time frame of the stimulation used.
Inflammation-Regulated Sites in the Promoters of the Novel AP-Regulated Genes.
We investigated whether the promoters of the 154 genes whose mRNA levels strongly correlate with the extent of AP exhibit a high frequency of binding sites for AP-regulated TFs. In the liver, these TFs mostly comprise nuclear factor κB (NFκB), activating protein-1, signal transducer and activator of transcription-3 (STAT-3), and some members of the C/EBP family.14 The occurrence of potential binding sites for one or more of these factors in the promoters of some of the 154 genes was compared with their occurence in a random set of control genes transcribed at least in the liver. The results detailed in our supplementary Table s3 point to a significantly higher number of potential sites for NFκB and activating protein-1 in the AP-regulated genes compared with the control gene set. We conclude that the majority of the 154 known or orphan genes (Tables 2 and s2) are likely to be controlled by one or more AP-regulated TFs.
The Tightly Controlled mRNA Levels Cover From Signaling to Positive APPs.
The 154 mRNAs listed in Tables 2 and s2 code for proteins covering prominent functions that are detailed in Table s2. This includes membrane/cytoskeleton organization, protein sorting and secretion, general metabolism, and detoxication. Two down-regulated genes participate in proteolysis and two up-regulated, early response genes protect against proteasome-mediated proteolysis. Other mRNAs code for secreted proteins (e.g., extracellular matrix components; cytokines). Most importantly, four other mRNA subsets are of immediate relevance in an AP context.
One subset codes for TRAF5, calpain 6, MAP4K4, and chemokine-like factor superfamily 6 that all play critical functions at early stages of AP-driven signaling. Specifically, TRAF5 (up-regulated) is an accessory molecule for the tumor necrosis factor-α receptor and is involved in activation of the NFκB pathway. In addition, both calpain 6 and MAP4K4 control the tumor necrosis factor-α/IL-1–activated MAP kinase pathway. Calpains control a protein tyrosine kinase 2-mediated cascade that targets the MAP2K1 protein. MAP4K4 is a serine/threonine kinase that specifically activates c-Jun kinase 1. Its down-regulation, as observed in the current study, limits c-Jun activity and, therefore, is in keeping with the lowered c-Jun mRNA abundance that we observed.
Another subset is involved in transcription and indicates a trend towards transcriptional limitation. Down-regulated mRNAs for activators include c-Jun, c-Myc, and the structure-specific recognition protein 1 that is a p63 coactivator. Other mRNAs correspond to factors involved in transcriptional repression (prospero-related homeobox 1, RPB5-mediating protein, rev-Erb-related NR1D2, Ets variant Etv3) or chromatin rearrangement (MSL3-like1, DNA helicase type 2, structure-specific recognition protein 1). It is noteworthy that none of the mRNAs for transcriptional activators that are known to be up-regulated by the AP in liver, namely, NFκB, C/EBP-β and C/EBP-δ, and STAT-3,14 was found in this list. These mRNAs are up-regulated only after latent NFκB, C/EBP-β, and STAT-3 molecules that preexist in the cytosol of quiescent hepatocytes have been rapidly imported to the nucleus at the onset of AP. Remarkably, our list of up-regulated mRNAs included RAN-binding protein 16, which participates in such a nuclear protein import.
A third subset participates in N-glycan maturation and processing, including the controls of sialic acid addition by sialic acid synthase and ganglioside synthesis by sialyltransferase 9. Sialylation/desialylation of glycoproteins is critical in self-recognition and intercellular communication in innate immunity32 and the sialic acid residues present on gangliosides participate in activation of the AP-triggered JAK-STAT pathway.33
Finally, the mRNAs that are most tightly correlated with the AP score preferentially code for up-regulated plasma APPs. The latter amount to 15 of 30 (50%) of the functionally identified mRNAs whose correlation is greater than or equal to +0.70.
The liver transcriptome in primates or rodents has been increasingly studied using array technology.11, 12 A limited probe diversity (<2,000 genes) prevented complete transcriptome coverage in approximately 50% of the studies. Current estimates of the number of genes expressed in any given vertebrate tissue sample, including the liver, range from 10,000 to 15,000.12, 34 Other studies of the liver transcriptome were conducted using probes for 4,000 to 23,000 genes that were not selected with respect to this organ. This resulted in only 2,500 to 4,500 informative genes35–37 or in a figure that is not publicly available.28, 38–45 As yet, only one study has used 11,000 probes that were specifically tailored to liver transcriptome analysis. However, this analysis resulted in only 30% to 67% informative probes, depending on the target mRNA mixture.46 Therefore, the current study is the first to cover most of the liver transcriptome in humans as judged from the high rate of detected mRNAs (86% of 9,858 Hs. clusters) as well as the relatively high number of mRNAs transcribed by subset C of liver-specific genes. Subset C comprises a limited size compared with the total number of genes transcribed in the liver, which supports earlier observations.34, 41, 47, 48 However, subset C contains 880 Hs. clusters, which is far more than the number of liver-specific genes identified in a study performed with only 10,000 unselected genes.34 Along with comprehensive coverage of the liver transcriptome owing to Liverpool, we have conjointly developed LiverTools. We believe that such a combination dedicated to liver transcriptome analysis has no counterpart.
We focused on the liver response to acute, systemic inflammation because this condition strongly regulates numerous genes in the liver. Therefore, deciphering the underlying mechanisms is of interest for our understanding of liver biology as well as for clinical purposes. This is the first study of the AP-dependent modulations of the liver transcriptome in humans in vivo that is amenable to statistical analysis. It could be argued that in some patients the altered abundance of some mRNAs resulted from the underlying cancerous disease and was inappropriately ascribed to the AP. However, this is unlikely given the heterogeneity of the cancers involved, the presence of liver samples from noncancerous patients, the correlation of mRNA abundance with the inflammation index, and our controls with cytokine-challenged cell cultures. The data obtained for our patients and in cytokine-challenged Hep3B cells and our search for enrichment in binding sites for AP-regulated TFs in the liver indicate that the current study deals mostly, if not exclusively, with the hepatocyte transcriptome. This is in keeping with our observation that the mRNA levels in whole liver and isolated hepatocytes are strongly correlated. However, our probe selection was made, at least partly, from whole liver cDNA libraries and, therefore, lends itself to transcriptome analysis in nonparenchymatous liver cells as well.
A major finding of the current study is the selection of 20 known and 134 novel human mRNAs whose hepatic level significantly correlates with the extent of AP. This correlation value allowed us to rank the resulting series of 134 newly identified mRNAs as novel AP markers and, in this respect, some of them outperformed some well-known markers. Remarkably, haptoglobin, ALB, and TSF are classically used as AP markers14, 15 and they participated in the initial assignment of an AP score to our patients. Haptoglobin, ALB, and TSF mRNAs were also listed in our preliminary selection of 772 genes but they did not pass our final selection for genes whose mRNA abundance correlates with the AP score. A high level of albuminemia or transferrinemia observed in patients 3, 4, and 6 in Table 1 supports this finding. We conclude that markers such as haptoglobin, ALB, and TSF are of interest in the detection of an AP but they poorly perform beyond their use in an all-or-none fashion (as in Table 1). On the contrary, some novel genes also code for secreted proteins (collagenIVα1, IGFBP2, lamininα5, NK4, transforming growth factor-β3) but with the added bonus of an up-regulated or down-regulated mRNA level that strongly correlates with the AP score. The levels of the corresponding proteins deserve extensive studies in body fluid samples. They will help solve the current need for novel APPs as sensitive diagnosis and prognosis tools.15 It will also be worth investigating to which extent the mRNA versus protein levels correlate. In this respect, the plasma protein versus hepatic mRNA correlation was searched for all APPs listed in Table 1 and was found to be highly variable between APPs (0.2 < r < 0.8), in agreement with other studies.49
One could have expected that the abundance of the mRNAs that are controlled most directly by cytokine-triggered receptors and associated cascades (e.g., some mRNAs for TFs) would best correlate with the AP extent. However, the functions of the proteins coded by the 154 mRNA samples in Table 2 indicate that the liver is able to adapt its response to the AP extent by virtue of a fine-tuned change in abundance of functionally diverse mRNAs. This is well illustrated by four prominent mRNA sets that code for early signaling molecules TFs, N-glycosylation enzymes, or positive APPs. These four sets summarize the exquisite precision of the AP-driven hepatocyte response. The liver is able to follow the extent of AP owing to a fine tuning of some mRNA levels controlling most, if not all, intracellular events from early signaling to the final secretion of proteins involved in innate immunity.
Dr. A. Vente (Deutsches Ressourcenzentrum für Genomforschung [RZPD], Berlin) is acknowledged for his assistance in the early stage of IMAGE clone selection. The authors are indebted to Dr. G.M. Hampton for providing a list of secreted proteins, to Dr. S. Claeyssens for plasma protein determinations, and to M. Hiron for excellent technical assistance. The assistance of C. Bansard, F. Caillot, and G. Saint-Auret in clone resequencing is appreciated. The authors also thank Prof. R.P. Erickson for a critical reading of the manuscript.