© 2007 American Thoracic Society doi: 10.1164/rccm.200707-972ED
Opening the Window on Genome-wide Expression Analyses in SepsisUniversity of Florida College of Medicine, Gainesville, Florida With the completion of the first phase of the Human Genome Project in 2001, there was a general expectation that our understanding of complex diseases would rapidly expand. Francis Collins wrote, "the true payoff from the Human Genome Project will be the ability to better diagnose, treat, and prevent disease, and ... with these immense data sets ... in hand, we are now empowered to pursue those goals in ways undreamed of a few years ago" (1). For polygenic diseases, like sepsis and critical illness, it was assumed that "discovery science" using genomewide genetic analyses and expression profiling (transcriptome) would unlock the riddles that had evaded earlier reductionist approaches focused on the contribution of individual genes and proteins (2, 3). The general gestalt was that complex diseases like sepsis would require a complex understanding, and that genomic analyses would provide unique insights into pathogenesis, and identify new diagnostics and potential therapeutic targets. However, those benefits and results have been generally slow in coming, and that has been due to a number of both theoretical and technical hurdles. In the current issue of the Journal (pp. 676–684), Tang and colleagues have made a potentially significant contribution to the diagnosis of sepsis using genome-wide expression analysis on neutrophils obtained from 94 critically ill patients, 71 of whom had the reference diagnosis of sepsis (4). The hurdles that these investigators had to overcome were considerable. The cost of performing these widescreen genetic analyses or microarrays has declined precipitously in recent years, but still remains beyond the reach of most individual investigators, and hinders the widespread use of these technologies. In recent years, however, most major academic institutions have committed to the purchase of analytical stations that are now being used as a shared-institutional resource. Early on, institutional review boards took a very proactive stance about the collection of personalized data and their need for protection, and have moved forward cautiously in granting approval. However, in recent years, institutional review boards seem to be more comfortable with the potential benefits and risks associated with the collection of genetic and genomic data. Obtaining high-quality genomic data from critically ill patients has been more problematic than originally anticipated. Preliminary studies demonstrated that the process of collecting the target cell population and isolating nucleic acids significantly influenced the pattern of gene expression (5, 6), as did methods of amplification and labeling of RNA products (7, 8). It became evident that incorrect sample collection generated uninterpretable data, but when performed properly by skilled personnel, analytical variance could be readily controlled, and disease-related signatures could be identified (6). With the vast amount of data being generated, there was also a requirement for new statistical approaches to avoid type I and type II errors (9). In their study, Tang and colleagues used training and validation sets to confirm their selection of informative genes, limiting the false-positive rate to less than 1 in a 1,000. Once gene lists are identified, extracting biological information has proven to be one of the most perplexing challenges. In human subjects administered endotoxin, the number of genes whose expression changed in blood leukocytes was greater than 4,000 (10), and in severely traumatized patients, the expression of over 6,000 different genes changed in total leukocytes, lymphocytes, and monocytes (11). It became evident very quickly that tools had to be developed that could categorize these genes and responses into "functional modules," "interactome maps," and signaling pathways.
Tang and colleagues identified a set of 50 signature genes from blood neutrophils that could accurately predict sepsis in 91% and 88% of their training and validation sets of critically ill patients, respectively. Importantly, the expression of these 50 signature genes was a stronger predictor of sepsis than physiologic indices and cytokines, such as procalcitonin. Interestingly, many of these signature genes were contained in the nuclear factor- There have been a few earlier publications that attempted to use genomewide expression analysis to identify sets of genes that were predictive in trauma, sepsis, and surgical injury (12–14). Most of these studies were grossly underpowered in their predictive ability and were more anecdotal than assurative. The present study, too, is at the lower limits for their own predictive sample size measurements. Nevertheless, Tang and colleagues' study shows what potential benefits are offered by genomewide expression analyses. Identifying a set of genes that is diagnostic of sepsis is an important first step, but a number of questions come to mind. Validation is, of course, required by other investigators, but one can already begin to speculate on subsequent questions. Is there a subset of these 50 signature genes that contains the same amount of diagnostic information in sepsis? Could an equally predictive model be generated from 1, 5, 15, or 25 of these genes, and could a more analytical multiplex reverse transcriptase–polymerase chain reaction approach be used that is more rapid and less expensive? The authors demonstrate that the genomic signatures were more class-predictive than physiologic and cytokine measurements for the diagnosis of sepsis, but could a combined model of genomic and physiologic parameters be created that has increased specificity and sensitivity? Such a combined genomic and physiologic model has been recently advanced as an improved diagnostic for outcomes from breast cancer (15). Diagnosis and prognosis can be considered the low-hanging fruit, and an additional question is what can the genomic signatures tell us about the response by the blood neutrophils to sepsis in critically ill patients? Here, caution must be exercised because we are looking only at the transcriptome and not the proteome or physiome. But, the preliminary data are evocative in suggesting that multiple genes involved in the basic signaling of inflammation are underexpressed in critically ill patients with the diagnosis of sepsis. It has now been almost 7 years since the first phase of the Human Genome Project was completed. The promises and expectations are still unfulfilled. But Tang and colleagues' report in this issue of the Journal shows us a glimpse of what the future may look like when genome-wide expression analyses are fully integrated into the diagnostic and research armamentariums. The window is opening and "discovery science" is beginning to demonstrate its potential utility in the clinic. FOOTNOTES Conflict of Interest Statement: The author has no financial relationship with a commercial entity that has an interest in the subject of this manuscript. REFERENCES
Related articles in AJRCCM:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
B, MyD88, and MAPKKK families, and were relatively underexpressed in the neutrophils from critically ill patients identified with sepsis.