Abstract
High throughput molecular analysis of veterinary tissue samples is being applied to a wide range of research questions aimed at improving survival, development of diagnostic assays, and improving the economics of commercial production of animal products. Many of these efforts also, implicitly or explicitly, have ramifications for the clinical care of humans and, potentially, animals. Here we provide an overview of applications of gene expression profiling in veterinary research and practice. We then focus on the current state of quality control and quality assurance efforts in gene expression profiling studies, underscoring lessons learned from such analysis of human samples. Finally, we propose practices aimed at ensuring the reliability and reproducibility of such assays. The implementation of quality assurance practices by a trained pathologist is an essential link in the chain of events leading ultimately to reliable and reproducible research findings and appropriate clinical care.
High throughput molecular analysis of clinical tissue samples is now being performed in a variety of contexts including research into the pathogenesis of veterinary diseases and animal modeling of human diseases. In this review we describe the various ways in which gene expression profiling is being applied to veterinary research, as well as research related to human disease using animal models. Despite the broad application of gene expression profiling, quality control measures are not consistently described in the literature. We describe here the current state of quality control and quality assurance efforts in gene expression profiling studies, drawing on our own experience with gene expression analysis of human samples to identify practices that will maximize the validity of such studies whether applied to veterinary or human disease processes. We believe that the quality assurance and quality control practices we describe are of great importance to both veterinary and human pathology in order to ensure the highest quality of research and, eventually, clinical care in both settings.
Gene Expression Profiling and Veterinary Disease
Examples of veterinary diseases studied by means of gene expression analysis are diverse. Many of these studies focus on conditions affecting dogs, including cardiomyopathy, 27,28,30 degenerative mitral valve disease, 29 atopic dermatitis, 22 pancreatic acinar atrophy, 5 and malignancies of the breast 34 and central nervous system. 43 Disease processes in animals other than dogs studied by microarray analysis include bovine mastitis 42 and osteoarthritis in horses. 41 Among the more esoteric examples is the microarray analysis of a cell line derived from a breast tumor arising in a female Tammar Wallaby. 36
While the preceding studies have sought to identify transcriptional changes unique to a disease process in a particular tissue type, a large body of literature exists focusing on the interaction between host and infectious organisms. The goal of these studies is identification of diagnostic markers, therapeutic targets, or factors leading to resistance. Such examples include mycobacterial infection 6,39 and trypanosomiasis in cows, 13 Salmonella infection 45 and avian coccidiosis 17,23 in chickens, amoebic gill disease 25,48 and infectious anemia in Atlantic salmon, 15 gastrointestinal nematodes in sheep, 16 and vaccination response in Japanese flounder, 3,20,49 rainbow trout, 31 and cows. 26 Finally, there are numerous studies examining the expression of genes encoded by the infectious organisms themselves (see, for example, Jack et al. 14 ).
The results of these studies improve our understanding of the biology of these diseases and point to novel diagnostic, therapeutic, and breeding strategies that could prolong the life of pets and/or improve the efficiency of commercial production of animal products.
Gene Expression in Animal Models of Human Disease
In addition to providing a window into the pathobiology of veterinary disorders, gene expression analysis is frequently employed to gain insight into human disease processes using animal models as a surrogate. Examples include atrial fibrillation, 4 heart failure, 9 basilar artery vasospasm, 35 chronic kidney disease, 11 and myositis 37 in dogs. And countless studies have made use of both transgenic and xenograft mouse models combined with gene expression to gain insight into the pathobiology of human tumors (see, for example, reviews by Shoushtari et al., 38 Ahmad et al., 1 and Knostman et al. 18 ).
Pitfalls in gene expression analysis of clinical specimens
There have been proposals for incorporating gene expression profiling into certain aspects of the clinical care of animals, 8 and this is already taking place to a limited extent in the clinical care of humans—most notably with respect to breast cancer prognosis (Bao, 2008). Molecular medicine holds the promise to augment anatomic pathology, by increasing its objectivity—an important goal in light of documented inter-site variation in subjective interpretation of pathology samples and inherent discrepancy in semiquantitative methods. 2,7,10,40 The combination of histology and gene expression analysis has the power to improve the accuracy of clinical diagnostics and prognostics by precise segregation of individuals into meaningful groups, allowing for better informed therapeutic decision making. 32,33 However, molecular techniques themselves have been plagued with reproducibility challenges, 44 including the fact that parallel studies frequently identify different gene sets as a result of variations in sample processing, array platform, endpoints of interest, or computational analysis method. 19,46,47 These documented sources of variability highlight the need for standardized procedures to ensure that appropriate samples are being analyzed when molecular techniques are employed. The need for standardization of reporting and analysis has been recognized in related contexts including clinical trials and tumor marker reporting. 21,24 Details regarding tissue identity, viability, and homogeneity must be verified when molecular profiling data are reported to ensure valid interpretation of results.
Histologic quality control in gene expression studies
Pathologists are well aware of the need for rigorous application of standards and quality control to ensure that accurate diagnostic information is obtained from clinical samples and communicated to caregivers and investigators. While this seems intuitive, we have found that systematic quality controls are the exception, rather than the rule, when it comes to clinical research involving high throughput analysis of clinical samples. For example, our analysis of the literature revealed that among 100 of the most recent articles reporting results from microarray analysis on human malignancies (a context in which one would arguably expect the highest level of quality assurance by a trained pathologist), only 13% described serial sectioning of samples destined for molecular profiling to verify the histology of those samples, and 40% documented no quality control measures whatsoever to ensure accurate pathologic designation of the actual sample destined for molecular analysis. Simply put, if one does not look, there is no assurance that what was on the label (e.g., cancer, normal, infected, uninfected, etc.) is in fact what was selected.
To further highlight the importance of the role of the pathologist in molecular diagnostics, we randomly selected 18 paraffin blocks of tumor and adjacent normal samples—as designated by the surgeon or pathology assistant—employed in a recent cancer study. Upon microscopic examination, several of the “tumor” samples proved in fact to be admixtures of normal and tumor tissue, and one “normal” tissue was clearly mislabeled and was in fact tumor. Analysis based on the original classifications would have been biased in the direction of type II errors (i.e., concluding there were no differences in expression of a given gene when in fact there were). Histologic verification of the sample designations will promote accurate and reproducible results of such analysis.
For microarray analysis to be a reliable tool in either research or clinical decision making, it is imperative that the classification of the source sample be confirmed by histologic analysis. As tissues are subdivided for various analytic processes (immunohistochemistry, gene expression profiling, etc.), the portions of tissue obtained may or may not reflect the characteristics of the whole specimen. And yet, procedures for verifying the histologic characteristics of samples utilized in gene expression profiling studies are explicitly described in only a minority of these studies.
This situation parallels the instructive historic example of breast tumor cell surface marker testing. Initially, the determination of estrogen and progesterone receptor status in breast cancer specimens was made by radioligand assay in one aliquot of tumor, while a second aliquot was used for histologic diagnosis. While this methodology yielded helpful information with respect to neo-adjuvant therapy planning, it was known to be prone to errors resulting from 1) significant differences between tumor and normal expression of receptors and 2) nonviable or inappropriate tissue in sample for quantification (e.g., adipose or muscle tissue) as well as 3) contaminating benign cells—leading eventually to the adoption of immunohistochemical analysis that enable simultaneous detection of receptor status and visualization of the histology. 12 Without the appropriate involvement of pathologists in assuring the suitability of tissue samples used for high throughput molecular assays, this historic example is destined to repeat itself.
Not only does anatomic pathology have an important role to play in ensuring accurate results from molecular assays, but the results of high-throughput molecular profiling can also provide a check on sample handling and classification. For example, we have found the excessively high correlation of gene expression data between purportedly “normal” and “tumor” samples is indicative of mislabeling or misclassification (i.e., the indication is that both samples are either normal or tumor). For any given high-throughput clinical assay, the laboratory performing the analysis would do well to define a standard (specific to tissue of origin and assay platform) using appropriate control specimens such that thresholds for excessively high or low correlation can be defined and applied as a quality control measure.
Quality Assurance Practices: A Proposal
We propose that the anatomic pathology community adopt a standard for quality control and quality assurance of specimen aliquots destined for high-throughput molecular analysis. To this end, we here outline an initial draft of standard practices for pathologic quality assurance in the course of high-throughput assays of tissue samples. These practices are particularly important where “diseased” and “normal” tissues are being compared as well as in cases where the disease tissue has a high probability of being heterogeneous. These proposed procedures will promote assay validity and will assist in the planning of genomic analysis by enabling clinicians and researchers to accurately determine the number of replicates that can reliably be provided by the sample and the appropriateness of the array analysis as it relates to comparing the paired samples.
Draft standard operating procedure for multiplex molecular testing of clinical samples utilized in clinical trials or clinical practice:
The histologic characteristics of the sample submitted for molecular assay must be verified by a trained pathologist by either of the 2 protocols described below. In either case, the protocol for the molecular assay must include parameters describing to what extent the sample analyzed must be histologically homogeneous and what is to be done if those parameters are exceeded. These parameters must be experimentally defined—the duty of the pathologist is to determine whether the parameters are met, not to define them. The 2 acceptable methods for verifying the histologic characteristics of material sent for molecular assay are as follows:
Serial cryostat sectioning: A section from the frozen specimen is taken for histologic examination, followed by 5 to 10 sections of tissue (10 µm each) to be submitted for molecular assay. This process (alternating sections for histologic examination and molecular assay) can be repeated as necessary to obtain the required mass of tissue for assay. After the last set of tissue sections are obtained for molecular assay, a final section will be obtained for histologic examination to ensure consistency throughout the samples submitted for assay. Gross sectioning: At the time of acquisition, a thin gross specimen can be obtained and placed, freshly cut surface up, in a histology cassette and submitted for fixation and embedding. A second thin gross specimen is then taken from the same site as the previous specimen and submitted for molecular assay. A third thin gross specimen is then obtained from the same site and placed, freshly cut surface up, in a histology cassette and submitted for fixation and embedding. After processing, the 2 fixed specimen blocks are faced, and a section is taken for hematoxylin and eosin (HE) staining and histologic examination to ensure that the tissue immediately adjacent to the material sent for molecular assay was histologically equivalent. The molecular assay protocol must identify suitable positive and negative control samples. The laboratory performing the assay will process these samples accordingly and ensure that the analytic results obtained from subsequent test samples fall within the defined ranges identified by the control samples (e.g., that results from tumor specimens do not correlate too highly with normal or other reference tissue).
Conclusions
High-throughput molecular assays are playing an ever-growing role in research into the pathobiology of human and animal diseases and, to a growing extent, clinical practice. The implementation of quality assurance practices by a trained pathologist is an essential link in the chain of events leading ultimately to reliable and reproducible research findings and appropriate clinical care.