Abstract
Different degrees of a toxic response between and within the various lobes of the liver have been observed in rodents following treatment with acetaminophen. This study was designed to compare 2 sampling methods of the rat liver for gene-expression analysis. Ten male Fischer 344/N rats, 12–14 weeks of age, were treated with vehicle (0.5% aqueous ethyl cellulose) or acetaminophen (APAP, 1500 mg/kg) and sacrificed 24 hours following dose administration. Two representative sections were collected from the left liver lobe, stained with hematoxylin and eosin (H&E), and evaluated independently by 2 pathologists. The central core of the left lobe was cubed and frozen. Five random cubes were conserved, while the remaining left lobe core was pulverized. From each of the 10 animals, 2 random cubes and 2 samples from the homogeneous, pulverized samples were prepared for microarray analysis. Histopathologic evaluation revealed a variable response of centrilobular necrosis within the left lobe. Multiple methods used to analyze the microarray data indicated that sampling technique was not a major contributor to the variability observed in the gene expression data; however, only the powdered samples clustered for all animals, even those with disparate histopathologic results. Additionally, a powdered sample provided the advantages of a homogenous sample pool and the ability to use sample aliquots for other analyses to include proteomics, metabonomics, and other molecular techniques.
Introduction
Microarray technology has opened the door to the field of toxicogenomics, leading to discoveries and insights into mechanisms of toxicity and altered pathologic states following acute and chronic doses of various compounds (Irwin et al., 2004; Waring et al., 2004). While the technology has assisted in the collection of new information in other fields, such as drug design and specific disease states (Boverhof et al., 2005, Ganter et al., 2005), the technology has not developed without skepticism (Petricoin et al., 2002, Luhe et al., 2005) with uncertainty directed toward its reliability. Many have asked, is the expression profile truly representative of the experimental treatment?
Recent studies evaluating the similarity in microarray gene-expression profiles of the liver from untreated mice across multiple microarray laboratories revealed poor reproducibility among research facilities (Members of the Toxicogenomics Research Consortium, 2005). While reasons for the variability have focused primarily on the microarray platforms, hybridization protocols, and the multiple aspects of data normalization and analysis, no studies have addressed the possibility of variability originating from improper sampling. Rarely do studies indicate the methodology for tissue collection. Proper collection of tissue samples for microarray studies is imperative for reliable, reproducible data.
Improper sampling refers to the collection of tissue not representative of the effects of the experimental treatment. In general, for hepatotoxicity studies, RNA is isolated from a small, random piece of the total liver. Since inter and intra treatment variation exists in animals, this random sampling method may not accurately reflect the pathologic and genetic profile of the treated liver. In previous studies conducted by our group, histopathology of rats treated with APAP revealed varying degrees of centrilobular necrosis of the liver within animals and between animals (Heinloth et al., 2004; Irwin et al., 2005). Microarray results from a random liver cube within the same lobe of the liver were variable among animals in the same treatment group. Since the histopathology and microarray results were not in concordance, this study was designed not to identify discriminating genes following treatment with a designated compound, APAP, but to assess the variability between 2 tissue-sampling methods of the liver—a random cube versus a homogenous, pulverized powder using various bioinformatics tools. We describe an optimal method of sectioning the primary lobes of the rat liver for histopathological evaluation to correlate with tissue collected for microarray studies, thus providing optimal representation of the experimental conditions.
Materials and Methods
Chemicals
Acetaminophen (APAP, 99% pure) was purchased from Sigma Chemical Company (St. Louis, MO) and suspension formulations prepared by mixing with 0.5% aqueous ethyl cellulose (USP/FCC grade; Fisher Scientific, St. Louis, MO).
Animals
Fischer 344/N male rats (Taconic Farms, Germantown, NY), 12–14 wks of age, were treated with either vehicle or APAP. Ten rats were assigned to each group. The animals were maintained on a 12-hour light/12-hour dark cycle. NTP-2000 open formula (Rao et al., 2001) pellet diet (Ziegler Brothers, Inc., Gardners, PA) and water were provided ad libitum. Experiments were performed according to the guidelines established in the NIH Guide for the Care and Use of Laboratory Animals (National Research Council, 1996), and an approved Animal Study Protocol was on file prior to initiation of the study.
Study Design
To increase the amount of test article absorbed, APAP was administered at a dose of 1500 mg/kg/day to each animal via gavage in 2 sessions, 1 hour apart at a dosing volume of 15 ml/kg body weight. The vehicle control was administered in the same manner. Rationale for dose selection was based on the degree of toxicity observed at this dose in previous studies (Heinloth et al., 2004; Irwin et al., 2005).
Clinical Chemistry
Clinical-chemistry analyses were performed on serum samples using the Roche Cobas Fara chemistry analyzer (Roche Diagnostic Systems, Inc., Montclair, NJ). Specifically, alanine aminotransferase (ALT) analysis was employed to assess hepatocellular damage.
Tissue Collection
Rats were euthanized under CO2 gas and tissue collected 24 hours following the initial APAP dose. Blood samples were collected from the posterior vena cava of all rats for clinical chemistry parameters. The liver was excised and processed within 2 minutes of removal from the body. Two representative sections (3 mm) from the left lobe were collected and fixed in 10% neutral-buffered formalin (NBF) for histopathology. The central core was cubed (2 mm × 2 mm) from each of these respective lobes, flash-frozen in liquid nitrogen (LN2), and stored at −80°C. Five random cubes from the left lobe core were conserved, while the remaining left-lobe core cubes were pulverized (Figure 1).
Histopathology
Representative liver sections were fixed overnight (18–24 hours) in NBF, routinely processed, and paraffin-embedded. Hematoxylin and eosin-stained sections (6 um) were examined in a blinded fashion. The degree of necrosis was scored by assigning a value from 0–4: 0 = the absence of necrosis, 1+ = minimal necrosis, 2+ = mild necrosis, 3+ = moderate necrosis and 4+ = marked necrosis.
RNA Isolation
Total RNA was isolated from 2 of the 5 random cubes collected from the left lobe. The remaining cubes of the left lobe core were pulverized and 2 aliquots (100–150 mg) transferred for RNA isolation. Detailed protocols for tissue pulverization and RNA isolation are located at the NIEHS Microarray Group web site 〈http://dir.niehs.nih.gov/microarray/methods.htm〉. RNA isolation was performed at Integrated Laboratory Systems (Research Triangle Park, NC).
Microarray Hybridization
Equal amounts of RNA (1 ug) from each of the 10 vehicle control rats were pooled and compared with 1 ug of RNA from each cube and each pulverized powder aliquot from individual animals. Following RNA amplification and labeling by a fluorescent dye (either Cy3 or Cy5) using the Low RNA Input Linear Amplification Labeling Kit™(Agilent Technologies, Palo Alto, CA), the amount and integrity of RNA were verified using a Nanodrop ND-1000 spectrophotometer and the Agilent Bio-analyzer (Agilent Technologies). Equal amounts of Cy3- or Cy5-labeled cRNA were hybridized to the Agilent Rat In-Situ 60mer Oligonucleotide microarrays (#F4130A containing 22,575 oligonucleotide probes representing over 20,000 well-characterized rat genes, ESTs, and EST clusters). Fluorescent intensities were measured with an Agilent DNA Microarry Scanner (Model G2565AA). Data were extracted from the resulting images using Agilent’s Feature Extraction Software.
Microarray Data Analyses
To quantify and normalize the signal intensities from the oligo spots on the image files, data from the biological replicates were loaded and combined in the Rosetta Resolver Gene Expression Data Analysis System V5.0 (Seattle, WA) using an error-weighted average. ANOVA analyses and hierarchical clustering of the resulting microarray data were completed using Rosetta Resolver. Principal component analysis (PCA) calculations and plots were performed using the statistical software package R 〈http://www.r-project.org/〉. In addition to these analytical methods, a unique statistical approach was devised to examine the gene expression variability by determining the absolute deviation (A) in the gene expression (g) for each rat (r) between the cube (C) and powder (P) sampling methods using the equation A gr = |C gr 1 – C gr 2| – |P gr 1 – P gr 2|. Permutations of the resulting gene-specific averages (Ag ) were then run using a sign statistic (Hollander and Wolfe, 1973).
Results
Two representative liver sections (A, B) adjacent to the central lobe core of tissue collected for RNA analysis were obtained for histopathology from each of the primary lobes. The collection of tissue from 2 areas of an individual lobe allowed for improved histopathologic evaluation of the experimental conditions. Considerable variability in the degree of centrilobular necrosis in the left lobe existed within an animal and among animals 24 hours following a single toxicdose exposure to APAP (Figure 2). Few animals had the same extent of necrosis present in section A and B of the left lobe. Even within a representative section of the left lobe, the distribution of the centrilobular necrosis was not uniform. Clinical chemistry values confirmed hepatocellular damage 24 hours after administration of 1500 mg/kg of APAP. Hepatocellular damage, indicative of APAP treatment, was evidenced by increased ALT levels, which ranged from 460 to greater than 15,000 U/L, compared to the mean control group value of 60 U/L (data not shown) (Table 1).
Since great differences were observed in the histopathology of the left liver lobe, microarray studies were conducted on biological replicates to observe whether differences in gene expression also existed in tissue from the central core. To assess genomic variability in the left lobe, the central core was cubed and frozen. Five random cubes were conserved, while the remaining left-lobe core was pulverized. From the 5 cubes, 2 cubes were selected and compared to 2 powdered aliquots from the remaining pulverized left lobe. Multiple data-grouping methods were used to discriminate group separation by assessing sequence variability within the liver following APAP treatment. Principal component analysis (PCA) of all genes was used to allow a global view of the variability across all samples (Figure 3). Seventy-three percent of the variation fell in the X direction of principle component (PC) 1. The absence of a bimodal distribution in the PC 1 separation plot indicated no distinction between the 2 groups. Further breakdown of the principle components at PC 10 and PC 11 began to show a possible segregation of the groups; however, the percentage of variation comprising these components was less than 1%.
Since group separation by a global gene perspective was absent, data analysis focused on sequences showing a significant difference in their gene expression. Using Rosetta Resolver, signature sequences were identified from an entire chip using an ANOVA analysis to compare the variability of the arrays within the powdered and cubed groups to the variability between these groups for each animal at a significance value of p < 0.01. The number of signature sequences for each animal ranged from 0–414 with the mean equal to 68 (Table 2), and accounted for ≤2% of the total number of sequences present on the array. Considering the variability comparison for animals #5 and #50, we observed that the percentage (≤2%) of differing genes on the array between the cubed and powdered samples was small. Animal #5 did exhibit a greater number of signature sequences compared to animal #50—414 versus 38, respectively, which was consistent with the histopathologic observation of increased centrilobular necrosis. The analysis was then expanded to include all the arrays, a total of 40. When all cubed and powdered samples were analyzed together, the number of signature genes found at p < 0.01 was 23 or 0.11% of the total number of sequences. The same analysis was then performed with a randomized assignment of arrays to the powdered and cubed groups. The randomly assigned cubed and powdered samples yielded a similar result, with 16 signature sequences observed or 0.08% of the total number of sequences. Since a low number of signature sequences was detected by ANOVA and array randomization resulted in a flat curve similar to the nonrandomized arrays (data not shown), a difference between the powdered and cubed sample could not be established.
Another analytical method used to help differentiate the powdered from the cubed sampling groups comprised hierarchial clustering (Figure 4). Using the following criteria of ratio values greater than 1.5-fold and p < 0.01 in 25% of all arrays, 2,823 signature sequences were identified. When clustered across all 40 arrays, an obvious separation of powdered and cubed arrays was not observed. When clustering with only the powdered arrays, both samples from a given animal clustered in all cases. When clustering with only the cubed arrays, the samples from 2 animals (#5 and #18) did not cluster. These animals also exhibited the maximum and minimum number of signature sequences in the ANOVA analysis (Table 2), in addition to disparate degrees of centrilobular necrosis between the 2 representative sections.
One final statistical method was devised to evaluate the absolute deviation of the gene-expression variability between the 2 sampling methods. Using Rosetta Resolver, we identified 6,761 signature sequences utilizing an ANOVA analysis to compare the variability of the arrays within the powdered and cubed groups for each animal at a significance value of p < 0.001 in at least one array. For this analysis, 4 intensity values must have been recorded for each animal. If an intensity value was absent for a signature sequence, the respective gene was not included in the analysis; therefore, a total of 6,228 signature sequences were utilized for analyzing the absolute deviation of the gene-intensity values. Again, the cubed sampling method failed to show greater variability in the differential gene expression when compared to the powdered samples ( p = 0.176).
Discussion
Proper sampling of tissues is important in many ways for microarray studies. In conducting a hepatotoxicity study, capturing details of the histopathologic observations and correlating them with signature sequences identified in microarray studies allow an investigator to determine whether these parameters indeed represent the experimental conditions used. Standardization of all aspects of a microarray study, from tissue collection to data preparation for statistical analysis, reduces the variability among studies and increases an investigator’s confidence in study results. The purpose of this study was to determine whether gene-expression differences exist between a random tissue cube used in a microarray study and a more homogenous pulverized, powdered sample.
Histopathologic evaluation revealed a variable hepatic response of centrilobular necrosis within the left lobe of some animals and among animals 24 hours following treatment with a toxic dose (1500 mg/kg) of APAP. That food was not removed from the animals prior to APAP dosing may in part be responsible for the varied degree of centrilobular necrosis; however, because not all animals respond in an analogous manner to toxicity testing, differences among tissue sampling methods for microarray studies must be identified (McLean et al., 1975; Richardson et al., 1986; Irwin et al., 2005). Collecting liver sections adjacent to either side of the central lobe core used for genomic studies complements correlation of the experimental conditions between the histopathologic observations and microarray analysis.
Multiple statistical approaches failed to distinguish differences between the powdered and cubed sampling methods that increased variability would occur in the gene-intensity values from the random cubes compared to the homogenous powdered aliquots was speculated; however, this was not the case. Traditional bioinformatics tools used for categorizing groups, such as PCA and unsupervised hierarchial clustering proved unable to separate the 2 sampling methods. While major components of the variability did not differentiate the powdered from the cubed samples, minor components of the gene-expression variability may be attributed to sampling determined by PCA. Cluster analysis applied individually to the powdered and cubed samples also failed to categorize the sampling groups. While the more homogenous powdered group clustered for each animal, this occurrence was not observed in the cubed group. On the contrary, cubed samples from 2 animals (#5 and #18) did not cluster. In addition, these animals displayed moderate variability in their clinical-chemistry ALT values and extreme differences in their degree of centrilobular necrosis within the left lobe. Clearly, the cohesiveness of the cluster was disturbed by the random sampling method.
ANOVA analysis yielded a small number of signature sequences that were different between the powdered and cubed samples; however, when the same analysis was performed with a random assignment of arrays to the 2 groups, a similar result was achieved again, implying that no detectable difference could be discerned. This lack of difference was also the case for our statistical method, which evaluated the absolute deviation of each gene. Our screening method was designed to eliminate genes with high levels of variability, regardless of the sampling method. Removal of these genes with levels of increased variability should have made the differences in the variation between the sampling methods easier to detect if it was present but should not have created false differences between sampling methods when none existed.
A definitive reason could not be found to explain why there was no difference in gene expression, even though the degree of necrosis ranged from absent to marked with-in and between treated animals. Although necrotic hepatocytes were present in the sample, sufficient amounts of intact, non-degraded RNA were present to perform microarray. If the powdered sample had been from the remaining left lobe instead of the middle lobe core, separation between the two sample populations might have been more distinct.
Clear advantages and disadvantages exist between the powdered and cubed sampling methods. For all studies involving microarray, obtaining non-degraded RNA of the highest integrity constitutes the primary goal. A pulverized powder sample allows for multiple molecular procedures to be run from a more uniform sample, therefore minimizing study variability. Preparation of the pulverized powder sample is laborious, and great care must be taken to maintain a non-degraded RNA sample. Evaporation of LN2 in the powdered preparation must not occur and provisions ensuring sample integrity during the measurement and distribution of the powdered aliquot must be maintained; otherwise, the risk of degraded RNA becomes greatly increased. While isolating RNA from a cubed sample is simple and rapid if using a column-based assay from a commercial kit, two disadvantages are clear. First, a small, single representation of the target tissue is assayed to determine whether the total experiment is represented. If a commercial product is used for RNA preservation, the sample can be used solely for the molecular analysis of RNA. While, therefore, more labor is required and great care must be exercised in the preparation of a pulverized powder sample, the advantages of greater representation and the ability to investigate other downstream genomic applications (i.e., proteomics and metabonomics) outweigh the benefits of a random cubed sample.
Although this study did not detect conspicuous differences and establish a need for the utilization of one sampling method over another, we do advise our tissue collection method for histopathologic and microarray assessment of hepatocellular toxicity following treatment with a chemical compound. Collecting liver sections adjacent to either side of the central lobe core used for genomic studies allows improved correlation of the experimental conditions by histopathology and microarray. While the effect of the powdered samples in reducing variability is minimal, any improvements in the precision of assessing gene expression levels contribute to the potential identification of key toxicity markers following exposure to specific chemical compounds.
