The Hepatic Transcriptome as a Window on Whole-Body Physiology and Pathophysiology

Abstract

Transcriptomics can be a valuable aid to pathologists. The information derived from microarray studies may soon include the entire transcriptomes of most cell types, tissues and organs for the major species used for toxicology and human disease risk assessment. Gene expression changes observed in such studies relate to every aspect of normal physiology and pathophysiology. When interpreting such data, one is forced to look “far from the lamp post,” and in so doing, face one’s ignorance of many areas of biology. The central role of the liver in toxicology, as well as in many aspects of whole-body physiology, makes the hepatic transcriptome an excellent place to start your studies. This article provides data that reveals the effects of fasting and circadian rhythm on the rat hepatic transcriptome, both of which need to be kept in mind when interpreting large-scale gene expression in the liver. Once you become comfortable with evaluating mRNA expression profiles and learn to correlate these data with your clinical and morphological observations, you may wonder why you did not start your studies of transcriptomics sooner. Additional study data can be viewed at the journal website at <www.toxpath.org>. Two data files are provided in Excel format, which contain the control animal data from each of the studies referred to in the text, including normalized signal intensity data for each animal (n = 5) in the 6-hour, 24-hour, and 5-day time points. These files are briefly described in the associated ‘Readme’ file, and the complete list of GenBank numbers and Affymetrix IDs are provided in a separate txt file. These files are available at http://taylorandfrancis.metapress.com/openurl.asp?genre=journal&issn=0192-6233. Click on the issue link for 33(1), then select this article. A download option appears at the bottom of this abstract. In order to access the full article online, you must either have an individual subscription or a member subscription accessed through <www.toxpath.org>.

Introduction

Liver structure and function in health and disease (Vidal-Vanaclocha, 1997; Arias et al., 2001; Belizikain, 2001), the role of the pathologist in the application of large-scale gene expression technology (microarrays) to toxicology (toxicogenomics) (Boorman et al., 2002), and the underlying skills needed in the latter field (Burczynski et al., 2000; Causton et al., 2003; Hamadeh and Afshari, 2004) have been reviewed in detail elsewhere. The liver plays many roles in whole body function, such as the control and/or synthesis of critical blood constituents including glucose, free-fatty acids, ketone bodies, amino acids, hormones, clotting factors, and inflammatory mediators. The liver is critical in immune function (Parker and Picut, 2004) and a first line of defense against certain infectious organisms and toxins entering from the gastrointestinal tract, to which it is intimately linked functionally, embryologically, and evolutionarily. In this respect, the liver represents but one of many components of an integrated system, the properties of which have been introduced by Bertallanfy (von Bertallanfy, 1968) and relevant dynamics which were previously reviewed in this journal (Morgan et al., 2004).

The current article addresses the influence of feeding, diet, and other extrahepatic activity on the hepatic transcriptome. A data set derived from rats is provided to introduce newcomers to the field of transcriptome interpretation. The data reveal the combined influence of circadian rhythm and fasting on large-scale hepatic gene expression. The interpretation of transcriptional responses to fasting and time of necropsy in these data will be greatly assisted by reference to Harper’s Biochemistry (Murray et al., 2000) and by articles on the effects of circadian rhythm on the rat transcriptome (Kornmann et al., 2001; Stokkan et al., 2001; Akhtar et al., 2002; Kita et al., 2002; Storch et al., 2002; Oishi et al., 2003), respectively. The data from this study can be downloaded at <see bottom of abstract for website>.

Materials and Methods

Animals

Male Harlan Sprague–Dawley rats were received at 5 to 6 weeks of age from Harlan Laboratories (Oregon, Wisconsin), were acclimated for 2 weeks and were 7 to 8 weeks of age at study initiation. The rats, 5 per group, were housed individually in hanging stainless steel wire mesh cages (10″ × 7″ × 7″) on a 12-hour light-dark cycle (6 AM–6 PM), and provided with water and feed (Lab Diet 5002) ad libitum until food was withheld as noted below. The experiment was designed for a single test article exposure at 3 dose levels followed by necropsy after 6 hours and 24 hours or repeat dosing for 5 days with necropsy at 24 hours after the last dose. Control rats that received vehicle only were used as time-matched controls and are the subject of this study. The rats were anesthetized with pentobarbitol, euthanized by exsanguination from the abdominal aorta, and samples collected from the cranial portion of the medial lobe of the liver. Samples were immediately flash-frozen in liquid nitrogen and stored at − 80° C until RNA isolation. The 6-hour rats were not fasted overnight, but food was removed at 8:00 AM on the day of necropsy, which occurred at 2:00 PM ± 30 minutes. The 24-hour and 5-day rats were fasted from 3:00 PM on the day prior to necropsy, and necropsied at 8:00 AM ± 30 minutes.

Total RNA Extraction

Total RNA was isolated from liver samples using RNeasy maxi kits from QIAGEN (Valencia, CA) following the manufacturer’s animal tissue protocol in the QIAGEN RNeasy Midi/Maxi Handbook. Briefly, frozen liver samples were lysed and homogenized for 45 seconds in the presence of a highly denaturing guanidine isothiocyanate buffer using an Ultra-Turrax T25 homogenizer (IKA, Wilmington, NC). Tissue lysates were centrifuged to remove cell debris and applied onto the RNeasy column. Total RNA was bound to a silica-gel-based membrane. Following several stringent wash steps, total RNA was eluted from the membrane using RNase-free water and concentrated by ethanol precipitation overnight. Quantification of RNA samples was conducted by measuring optical density at 260 nm and 280 nm on a spectrophometer. The RNA integrity and purity were confirmed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA) following the Agilent Technologies Reagent Kit Guide for RNA 6000 Nano Assay.

GeneChip Array Hybridization

According to the suggested protocols in the Affymetrix GeneChip Expression Analysis Manual, first and second strand cDNA were synthesized from total RNA samples using the Superscript Choice System (Gibco BRL Life Technologies, Carlsbad, CA). Following double-stranded cDNA clean-up and quality check, an in vitro transcription reaction was carried out to generate biotin-labeled cRNA. The cRNA was purified, fragmented, and allowed to hybridize onto the chip while rotating in a hybridization oven for 16 hours at 37° C. Following several washing and staining steps, the chips were scanned and each complete probe array image was collected. Scan quality was evaluated using background and noise values, Affymetrix internal controls, and percentage of present genes (Affymetrix GeneChip Expression Analysis Manual). No samples were eliminated from the study analysis as outliers.

Discovery of Fasting and Circadian Rhythm

The analysis of a toxicogenomics study in rats for a drug candidate using a large-scale gene expression platform (Affymetrix RG U34A GeneChip) revealed clear differences in gene expression between the early exposure duration time point (6 hours) and the later 2 time points (24 hours and 5 days) in both the control and treated rats. This difference which was unexpected, was first detected using principal component analysis (Figure 1) and confirmed for several hundred genes using a range of analytical techniques including singular value decomposition (Matlab, The MathWorks, Inc.), relevance networks analysis (PathlinX, Xpogen), Eisen Elser et al. (1998) and Spotfire clustering tools, and manual triage with Excel. Results derived from the Eisen clustering tool are shown in Figure 2 as a heat map for all animals and genes on the left (the clustering trees have been removed). One node highlighted by yellow shows a marked difference between 6 hour and other groups. This is expanded on the right in Figure 2. The expanded node clearly illustrates that both the treated and control 24-hour and 5 day rats are different from the 6-hour rats. The software for clustering is available for download, along with a users manual and certain restrictions, from <http://rana.lbl.gov/index.htm?software/manuals/ClusterTreeView.pdf>. The value of such tools is the grouping or clustering of transcripts exhibiting similar patterns of expression across treatment groups and the generation of lists of genes in such clusters. It provides a clear visualization tool demonstrating the differences between the control groups. The dendogram can be used to identify which genes are responsible for the differences and their extraction from the data is a major function of these bioinformatics tools (Caustont et al., 2003; Irwin et al., 2004; Malarkey et al., 2004).

There were 2 key differences between the 6-hour control rats and those from the 24-hour and 5-day groups. The 6-hour rats were not fasted overnight and they were necropsied at 2:00 PM, while the 24-hour and 5 day groups were fasted for 17 hours prior to necropsy at 8:00 AM. These differences are reflected in the amount of glycogen present in the livers, which is considerable in the 6-hour group and apparently absent in the 24-hour and 5-day animals, as assessed by light microscopy of hematoxylin and eosin (H&E) stained paraffin sections (Figures 3A and 3B). This experiment was repeated approximately 2 years later, with similar results, and data derived from both studies is provided for your investigation (Tables 1 and 2 online, see bottom of abstract for website). A brief interpretation of the gene expression data derived from these studies follows, starting with genes associated with glycogen storage, leading then into general energy metabolism, circadian effects, and finally selected aspects of hepatic function are examined and discussed.

Interpretation of the Study Data Sets

Data interpretation begins with the exploration of the gene lists generated by clustering and principal component analysis. The gene lists are important data but without further exploration and interpretation, add little knowledge just as a list of study diagnoses without a pathology narrative is not very helpful. The pathology narrative does not address every single study diagnosis and similarly it is important not to get lost in the myriad of genes but rather address major themes. In this case, many of the differences in gene expression identified by the bioinformatics tools involved energy and circadian pathways. This seemed a good place to start because of necropsy time and fasting differences between the 6-hour and the other groups. The major reference source for this interpretation work was Harper’s Biochemistry (Murray et al., 2000) and several reviews on circadian genes in the liver (Kornmann et al., 2001; Stokkan et al., 2001; Akhtar et al., 2002; Kita et al., 2002; Storch et al., 2002; Oishi et al., 2003; Porsin et al., 2003; Schrem et al., 2004).

Glycogen Metabolism

Of the major genes related to hepatic glycogen synthesis (glycogenesis) and breakdown (glycogenolysis) present on the RG U34A GeneChip, only glycogen phosphorylase (PYGL) exhibited a strong signal difference between the fasted and nonfasted groups (Figure 3C), with a markedly lower expression level in the fasted animals. Such a difference is consistent with the reduced need for glycogen breakdown as these energy reserves in the liver were apparently depleted (Figures 3A and 3B). Milder changes were observed for other genes in these pathways, including a mild up-regulation of glycogen synthase (GYS2, Figure 3D), but such changes were of dubious biological significance. If critical to the study interpretation, this could be a stimulus for further mechanistic studies and for a further look into the gene list.

Carbohydrate Metabolism

The hepatic glycogen-depleted livers (fasted groups) exhibited evidence of decreased glycolysis, including down-regulation of pyruvate kinase (PKLR, Figure 4A), pyruvate dehydrogenase, and glyceraldehydes-3-phosphate dehydrogenase (GAPDH, Figure 4B). There was no consistent evidence of increased gluconeogenesis, which was surprising given the depletion of liver glycogen combined with the critical role of the liver in maintaining blood glucose levels. The alternative pathway for glucose breakdown, the pentose phosphate shunt, also appeared to be down-regulated in the fasted animals, as indicated by distinctly diminished transketolase expression (TKT, Figure 4C). Such a change also indicates the potential for diminished NADPH generation for reductive synthetic activity in the liver of these animals. It would appear, however, that the major energy source for the fasting animals is not glucose, as is well known for the fasting state (Murray et al.), 2000, but fatty acids and possibly ketones (see following paragraph).

Fat Metabolism

There was clear evidence of increased lipid metabolism, which included up-regulation of genes responsible for mitochondrial long-chain fatty acid uptake (carnitine O-palmitoyltransferase 1, CPT1, Figure 5A) and β-oxidation (acetyl-CoA acyltransferase 3-oxo acyl-CoA thiolase A, ACAA1, Figure 5B). In contrast to the “burning of fatty acids,” the synthesis of fatty acids was concurrently down-regulated (fatty acid synthase, FASN, Figure 5C). Increased use of fatty acids for energy provision can be associated with an almost concurrent generation of ketone bodies (Murray et al., 2000), for which fatty acids are potential precursors. In the present study, there was a mild up-regulation of mitochondrial hydroxymethylglutaryl-CoA lyase in the fasted animals. Such a change indicates that the use of ketone bodies as an energy source may have been initiated.

Circadian Genes

Two circadian genes, aryl hydrocarbon receptor nuclear translocator-like (ARNTL, Figure 6A) and period 2 (PER2, Figure 6B), show distinct group differences. Arntl protein heterodimerizes with the clock protein and activates Per gene expression (Roenneberg and Merrrow, 2003). The activation of clockarntl dimers in turn has a negative feedback on Per transcription (Roenneberg and Merrow, 2003). It is becoming clear that the molecular control of the circadian cycle is much more complex involving multiple Period, Cryptochrome, Dec (members of the basic-helix-loop-helix transcription family) and other genes that are beyond the scope of this article. What is important to toxicologists is that there are significant differences in circadian gene expression in rats necropsied at 8 AM (24-hour and 5-day time points) and at 2 PM (the 6-hour time point). These 2 times fall within most necropsy schedules for a study termination. Up to 10% of the hepatic genes have circadian rhythm in their expression (Storch et al., 2002). Importantly, some of the genes involve xenobiotic metabolism (Panda et al., 2002; Oishi et al., 2003). The clock-controlled gene D site albumin promoter (Dbp) expression was much higher at 2 PM than at 8 AM (data not shown). This is consistent with protein levels where Dbp protein is barely detectable in rat hepatocytes in the morning but rises during the day (Schrem et al., 2004). Dbp protein appears important in many hepatic functions including regulation of xenobiotic metabolism (Schrem et al., 2004).

Other Observations

There was evidence of down-regulation of a number of hepatic synthetic functions, including cholesterol biosynthesis (HMGCS, Figure 7A, HMGCR, Figure 7B FDPS, Figure 7C). The transcript in Figure 7B, 3-hydroxy-3-glutaryl CoA reductase, is an important target for certain lipid lowering drugs, the study of which should therefore take into account the effects of fasting. Links between hepatic and thyroid function are revealed in altered gene expression (data not shown) for type I iodothyronine deiodinase (DIO1) and thyroid hormone responsive protein (THRSP). The DIO1 gene product provides the major portion of the circulating T3 (Maia et al., 1995) while THRSP is postulated to play a role in lipogenesis (Wang et al., 2004). Finding these 2 genes might prompt one to read further about thyroid effect on the liver (Feng et al., 2000).

The regulatory subunit of glutamate-cysteine ligase (GCLM) that catalyzes the first step of glutathione synthesis was down-regulated with fasting. This indicates a diminished synthesis of glutathione (Yang et al., 2001), an essential regulator of cellular redox state, and might lead to the potential for redox imbalance (oxidative stress). Examination of key genes that are known to respond to such redox imbalance, such as catalase, superoxide dismutase (1 and II), and thioredoxin reductase, indicates no evidence of such imbalance, suggesting that oxidative stress has probably not been induced. The majority of the changes described above were seen in a second study conducted about 2 years later. The data from both studies are included for your review (see bottom of abstract for website), however no systematic comparison is reported here. It was interesting to note that the expression patterns of the circadian genes, Arntl and Per2 in the Sprague–Dawley rats in this study mirror the results from a National Toxicology Program study in F344/N rats (Boorman personal communication) even though it involved a different strain of rat and a different microarray platform (NTP used the Agilent Rat Oligo Chip). It is encouraging that the gene expression data appears consistent across studies and between laboratories.

Application of PathlinX and Ingenuity Pathways Analysis

There are many other areas of biology for you to explore in microarray data sets, and many new tools are becoming available to carry out such exploration. We will give a brief example of the use of 2 such tools, PathlinX relevance networks (Xpogen, Inc.) and Ingenuity. The first product, which uses a sophisticated statistical algorithm (Butte et al., 2000), was used to generate a list of genes exhibiting similar expression behavior, and this list was then imported (as a text file) into another commercial package, Ingenuity Pathways Analysis Tool (Mountain View, CA). The latter tool uses a “knowledge database” to link the list of “similarly behaving” genes to an expression network. The generated networks are comprised of genes that have regulatory circuitry in common. The principal network from this procedure is shown in Figure 8, and it is apparent that it comprises two key components: (a) a small set of genes associated with circadian rhythm (top of figure), of which Per2 and Arntl are examples, and (b) a group of genes associated with lipid metabolism and other aspects of energetics (bottom of figure), the central controller of which is the nuclear transcription factor, peroxisome proliferator activated receptor gamma (PPAR-γ).

Such findings can be used to lead your search into the underlying mechanisms of fasting-induced alterations of gene expression and the potential impact of circadian rhythm on your study designs. It is apparent that such tools provide a structured approach to the difficult task of large-scale gene expression interpretation and can often point where you may want to begin your investigation. However, during this process there is no substitute for knowledge, which should not be confused with information (Morgan et al., 2003). Your knowledge of biology of the animal and pharmacology of the compound, coupled with the gene lists, can provide clues as to meaningful areas for further reading. Finally, the altered expression of a gene may not always demonstrate altered function, the confirmation of which is a critical aspect of transcriptomics (Merrill et al., 2002).

Brief Review of Hepatic Transcriptomics

Gene expression array technology is increasingly being used, and a number of selected studies are addressed briefly here, to introduce you to the technology and its applications and in some cases, to provide you with further data sets for exploration.

Discovery of New Target Genes

The wealth of information provided by large-scale gene expression array technology is leading to the discovery of new physiological pathways and gene functions. For instance, a recent study of the murine hepatic transcriptome led to the identification of 6 novel genes that are regulated by dietary cholesterol (Maxwell et al., 2003). In another study, 45 previously unknown thyroid hormone-responsive genes were detected in the livers of mice using array technology (Feng et al., 2000), further emphasizing links between liver function and the thyroid gland as indicated by data derived from fasted rats in the present study.

Microarrays have proven to be excellent tools for discovery of novel circadian genes (Akhtar et al., 2002; Duffield, 2003). Instead of 1 gene at a time, thousands of genes can be evaluated at multiple time points during the circadian cycle. Often genes that follow a similar pattern to known genes may have similar or related function. Genes that appear to have an interaction with known circadian genes can then be selected for further study. Several recent studies of the mouse liver show that a significant portion of the hepatic genome has a circadian cycle in gene expression (Stokkan et al., 2001; Panda et al., 2002; Storch et al., 2002). A complexity of the liver is that the circadian rhythm in gene expression is affected both by light through the suprachiasmatic nucleus and by feeding (Stokkan et al., 2001). Thus, in evaluating the hepatic transcriptome both time and feeding patterns need to be kept in mind.

Inflammation

The liver is an important part of the immune system having the largest population of fixed macrophages, producing acute phase proteins as part of nonspecific immunity, deleting activated T lymphocytes and playing a role in extrathymic lymphocyte development (Parker and Picut, 2004). Ongoing studies in our laboratory (Aventis, Inc.) demonstrated dysregulation of multiple pro-inflammatory mediators in the livers of rats treated with a proprietary compound, including tumor necrosis factor (TNF), TNF receptor super-family member 1a, intercellular adhesion molecule-1, and CD14. These changes were in addition to dysregulation of the complement and anti-complement systems, and in the absence of any evidence of liver pathology in H&E-stained sections. Such changes were present in the hepatic transcriptomes of animals treated with the test compound by subcutaneous injection, and were absent in animals treated orally. The former animals exhibited moderate to severe subcutaneous inflammation and microabscessation, which was considered responsible for the hepatic transcriptional changes. These observations are consistent with responses by the hepatic transcriptome to cutaneous burn-injury in rats (Vemula et al., 2004). Recent studies have demonstrated distinct changes in the hepatic transcriptome of a rat model of sepsis (Chinnaiyan et al., 2001), and similar responses have been reported for humans during acute systemic inflammation (Coulouarn et al., 2004). The hepatic transcriptome may thus provide an indication of injury at remote extrahepatic sites, which could play an important role in the understanding of mechanisms of extrahepatic inflammatory disease, and the role played by the liver in response to such conditions.

Conclusions

We have only just started to explore responses by the hepatic transcriptome to exogenous and endogenous factors. There are reports of the influence of diet, including protein quality (Endo et al., 2002), soy protein (Iqbal et al., 2002), atherogenic diets (Vergnes et al., 2003), fish oil (Takahashi et al., 2002), and caloric restriction (Cao et al., 2001) on the hepatic transcriptome. Many such articles may be found in the literature, and many of them provide complete data sets for study, or they can be requested from the authors. Complete interpretation of a single such study could take several yrs to complete if appropriate confirmatory work is carried out. We may need to rethink our approach to this technology if we are to avoid becoming completely overwhelmed by data. As in this study, one may need to focus on one or several major themes, knowing that other important data, as yet unexplored, resides in the data. Making the data available to the scientific community, as was done for the yeast cell cycle (Spellman et al., 1998), allows other investigators with different interests and expertise to mine the data.

Whatever approach is developed, it is clear that pathologists and other morphologists have an important role to play. For instance, if you prefer to have animals fasted prior to necropsy as the liver sections are easier to read, you might want to consider relaxing this restriction for studies of compounds that have profound effects on energy metabolism. Timing of necropsies also becomes an important consideration. Studies should be designed to generate the most valuable data, not just to satisfy a protocol for optimal morphological assessment. These new technologies are challenging and they may alter the conduct of biological studies. Do not be put off by such mysteries as heat maps and principal component analysis. Work with a bioinformatician, and ask them to transform the data into a form with which you are familiar, such as mean and standard deviation (see bottom of abstract for website). With a little work (and a lot of reading) you could contribute in a major way to the evolution of the field of transcriptomics, and many other “omics,” for that matter.

The liver is a window through which one may view much of the functions of the body (Boyer, 1994). Scientists with an understanding of the whole animal in health and disease are crucial to placing gene expression in context if we are going to take advantage of the rich, but complex view of pathophysiology provided by the hepatic transcriptome.