Short-term arginine deprivation results in large-scale modulation of hepatic gene expression in both normal and tumor cells: microarray bioinformatic analysis

Cell growth is dependent upon availability of essential amino acids for protein synthesis and this relationship makes amino acid-dependent control of gene expression an important area of study [1‐3]. We previously reported that levels of the tumor-associated glycoprotein amino acid transporter, TA1/LAT1/CD98 light chain/SLC7A5, increase in normal hepatic cells under low arginine conditions, while levels are constitutive and high in hepatic tumor cells [4, 5]. Upregulation of this gene is associated with multiple cancer types and we and others have hypothesized that increased expression may provide an adaptive advantage in the tumor microenvironment where nutrients are limiting [6, 7]. Loss of nutrient-sensitive regulation may comprise a subset of the loss of responsiveness to negative growth signals or autonomy from positive growth factors, characteristic of cancer cells in general [8]. We thus manipulated arginine concentration in a culture medium that maintains hepatic differentiation as a means to investigate, using a paired normal and tumorigenic cell type, how many and what types of hepatic genes are responsive to a transient change in amino acid levels. Affymetrix gene chips and bioinformatic analyses were used in a nutrigenomics approach. The goals of the study were to: 1) assess the scope of arginine-responsive hepatic gene expression using a well-defined in vitro rat model of normal and tumorigenic cells, 2) determine to what extent amino acid responsive regulation is retained upon transformation, and 3) provide a microarray dataset predicting novel genes and pathways subject to amino acid (arginine) regulation.

Triplicate RAE 230 arrays containing a total of 15,923 rat genes were hybridized for both normal and tumor cells grown under arginine-sufficient (+) and arginine-deficient (-) conditions for 18 hours in three independent pair-wise comparison experiments. Microarray Suite™ 5.0 software analysis of the resulting datasets revealed a mean of 8611 genes or 54% of the total rat array scored "present" (expressed) for normal cells versus 8355 genes or 53% in tumor cells (Table 1). Of the genes expressed, pair-wise comparisons of Arg + and - conditions revealed no change in expression for 7042 (82%) of genes in normal cells versus 6173 (74%) in tumor cells. Using a p-value of 0.05 or less and pair-wise comparisons for the three experiments in normal cells, expression of 811 genes increased, 608 decreased and 97 showed marginal change relative to arginine-sufficient controls. For tumor cells, transient arginine deficiency resulted in increased expression of 1249 genes, decreased expression of 926 genes, and marginal changes in 109. Thus, expression of as many as 2175 genes was scored as arginine-sensitive when the data were analyzed as independent pairs. This represented 8.9% of genes in normal cells and 13.6% of genes in tumor cells. At the time of the data collection, nearly half of these genes lacked complete functional annotation in the rat genomic database.

Because the number of genes in the preliminary analysis comprised a large portion of the rat genome, additional filters were applied to the dataset using GeneSpring™ 6.1 and 7.0 software analysis in order to identify the most reproducible and significant alterations resulting from transient arginine deficiency. For each cell type, all + samples were compared to all – samples, resulting in 9 comparisons. A concordance threshold of 66% or greater was set and 'arginine-responsive' increases or decreases were defined as a change of 2-fold and higher under arginine-deficient conditions. All other genes expressed were defined as "no change".

Using these more stringent parameters, the total number of arginine-responsive genes in normal cells and tumor cells was significantly reduced, to 56 and 162 respectively. A summary of the data comparing normal and tumor cells by functional gene class and direction of change is shown in Table 2. All 56 and 162 genes in normal and tumor cells respectively listed by accession numbers and corresponding direction of change relative to arginine-sufficient conditions (I = increase; D = decrease) are listed in supplemental Tables 4 and 5 [see additional files 1 and 2]. The entire 12 microarray dataset is available in MIAME format [9] in the NCBI Gene Expression Omnibus (GEO) database, confirmation #GSE2275. Figure 1 presents Venn diagrams of overlap between the arginine-responsive gene subsets of normal and tumor cells. Just over half the arginine-responsive genes in normal cells are also responsive in tumor cells. A surprising finding was the large number (133) of additional genes demonstrating arginine-responsive expression in tumor cells. More than one third of these were listed as transcribed sequences with no listed functional information or homology to known sequences at the time of analysis. Not surprisingly, genes with increased expression upon arginine depletion in both normal and tumor cells include a large number of genes involved in cell cycle and growth regulation, cell death and apoptosis, and stress response. Several of these were known amino acid-responsive genes such as asparagine synthetase [10]. Among other genes scoring as increased in both cell types was growth-arrest DNA damage-inducible 45 alpha (GADD45) and caspase 12. Genes down-regulated in both normal and tumor cells include plasminogen activator urokinase and low density lipoprotein receptor. Genes that exhibited arginine-responsive regulation in normal cells but a loss in tumor cells (or vice versa) may represent 'interesting' subsets in which the potential contribution of amino-acid regulation to the malignant phenotype could be explored. Genes whose expression was altered in tumor cells but not normal cells included tumor suppressor retinoblastoma-like 2, stress-response protein 70 kDa heat shock protein precursor and cell-surface linked signal transducer MAP-kinase phosphatase, cell cycle regulator cyclin dependent kinase inhibitor 2C (p18) and Ras. Genes scoring as changed in normal but "no change" in tumor cells were mainly in cholesterol and steroid biosynthesis pathways. Related to these were farnesyl diphosphate synthase and insulin-induced gene 1/growth response protein CL-6 (Insig-1/CL-6), an ER-residing membrane protein that blocks the proteolytic activation of sterol-regulatory element binding protein (SREBPs) transcription factors that activate the synthesis of fatty acids and cholesterol [11‐13].

Independent verification of expression was sought for a subset of genes via Northern blot analysis (Figure 2) and RT-PCR (Figure 3). Genes were selected in which cDNA probes were available with appropriate known controls to validate selected cases in which arginine-deprivation predicted an increase, decrease or no significant change in expression. Though the level of expression for some genes was found to differ substantially between normal and tumor samples, between arginine + and - conditions respectively, or between independent experimental replicates, trends predicted in microarray data analysis were generally validated by the confirmatory RNA expression analysis and densitometry. That is, LAT1, GADD45 and 4F2 increased in both cell types under conditions of arginine deprivation; LDLr decreased in both cell types upon arginine deprivation; LAT2 and GAPDH showed no significant change in either cell type, though LAT2 expression was very low in tumor cells; p21 decreased in normal cells but was below detection limits in tumor cells; farnesyl diphosphate synthase (FDPS) decreased in normal cells but not in tumor cells upon arginine deprivation; Insig 1 showed large variability with decreases in some replicates but not in others. The use of different probe sets between the microarray and Northern or PCR analyses, differences in quantitation and normalization methodology or undetermined other variables may account for differences in fold change values observed between these techniques.

Given its enormous potential for insight into regulatory physiology, nutrigenomics, the use of genomic tools to study nutrient-gene interactions, is an area of expanding interest in health and disease [14‐16]. We found global changes in gene expression when hepatic cells were subjected to short term deficiency of arginine. Arginine, a precursor of proteins and other compounds critical to normal metabolism, is considered essential for growth and development of many mammals [17, 18]. Our experimental system was one of amino acid imbalance, but not of total arginine starvation, since the medium was supplemented with 5% non-dialyzed fetal bovine serum and the exposure to deficient conditions was less than 24 hr. In fact, arginine-free medium has been widely used as a means to deplete primary hepatocyte cultures of small numbers of contaminating nonparenchymal cells because all the enzymes necessary for arginine synthesis are expressed by the normal hepatocyte [17]. Importantly, viability was well preserved over the experimental timeframe (data not shown). Thus, the observed changes in gene expression were not the result of impending cell death but are more likely a response to nutrient stress.

Stress caused by changes in the cellular environment including nutrients is recognized as an important physiological parameter to which eukaryotic cells have evolved complex cellular responses [19‐22]. Although it is not possible to differentiate primary effects from secondary during this timeframe, overall, cell growth and maintenance, metabolism, stress-related and apoptotic gene classes appear most predominant. Genes associated with metabolism and stress response that were present in the rat chip and showed a significant p-value for altered expression upon arginine deprivation are presented in Table 3. Because many genes associated with stress response in human cells were not present or identifiable on the RAE 230 chip however, it is not possible to fully assess arginine response of these gene sets in this model. With the exception of a subset of genes involved in cholesterol and steroid biosynthesis, there was no overall loss of amino-acid responsive regulation in the tumor cells compared to normal cells. In fact, to our surprise, in assessing the entire rat transcriptome a larger proportion of genes was found to be arginine-responsive in tumor cells compared to the normal cell from which they were derived. The use of this paired cell set in the experimental design is an important consideration as it minimizes variation due to other factors such as genetic strain or viral exposure. Our data not only add to the nutrigenomics literature and database in general, but also extend the finding of others in which changes in amino acid composition or starvation have been associated with specific patterns of hepatic gene response both in vivo and in vitro [23‐25]. The data suggest that arginine deprivation of hepatic normal and tumor cells may induce an initial stress response very similar to the well known endoplasmic reticulum (ER) stress-response described in eukaryotic cell types [19]. The location of products of several arginine-responsive genes including caspase 12, LDLr, Insig-1 and cholesterol/steroid biosynthesis all in the ER are consistent with the induction of a potent luminal ER stress response in both normal and tumor cells. Interestingly, although both cell types appear able to modulate LDLr, a regulator of cholesterol and steroid biosynthesis, in response to arginine deprivation, at some point downstream of the stress pathway, tumor cells appear to deviate by losing regulation of Insig-1 and metabolic genes directly involved in cholesterol and steroid biosynthesis. While it is tempting to speculate that loss of regulation within this gene cluster may be adaptive in a tumor microenvironment of limited nutrient availability, permitting malignant cells to persist in membrane synthesis required for repeated cell divisions, endocytosis, fusion or other processes, the relevance of these observed changes in gene expression in vitro to potential alterations in hepatic function and tumor growth in vivo are currently unknown. This hypothesis is testable but beyond the limits of the current project. The generation and reporting of nutrigenomic data is undeniably descriptive. However, datasets such as this one are valuable in that they facilitate hypothesis generation and further research. Much more study is needed to define specific genes and mechanistic pathways linking amino acids and other dietary nutrients to cancer risk.