Singling out genes that mark the deadly onset of sepsis

The activity of eleven genes may provide a means to make an early diagnosis of sepsis, the leading cause of hospital deaths in the United States, according to the authors of a study in Science Translational Medicine. Stanford researchers used a bioinformatics approach to sort through tens of thousands of genes whose activity fluctuates during the body’s inflammatory response, finding eleven that marked sepsis out from sterile inflammation, a condition whose clinical symptoms match those of sepsis.


Dr. Greg S. Martin, Professor of Medicine, Emory University School of Medicine (webpage):

Expertise: clinical and translational research in critically ill patients; primary interests in sepsis, organ dysfunction syndromes and acute respiratory distress syndrome (ARDS). 

“Sepsis is among the most common and lethal conditions seen in hospitals throughout the world. In the United States sepsis affects more than a million people each year, with the severe forms killing upwards of 30% of those afflicted. Sepsis is the leading cause of death in hospitalized patients and the 3rd leading cause of death overall in the United States, but remains largely outside of the public eye.

“Sepsis doesn’t lend itself to easy recognition. In fact, sepsis was presciently described by Machiavelli in 1513: “…at its inception, is difficult to recognize but easy to treat; left unattended it becomes easy to recognize and difficult to treat.” That’s an unfortunately adept characterization of sepsis: early in its course it is easy to treat but difficult to recognize, and once it has become more recognizable it is incredibly difficult to treat.

“The onset of sepsis is difficult to recognize clinically, particularly in critically ill patients where other aspects of the acute illness may falsely masquerade as sepsis, or mask a real episode of sepsis. Biomarkers have long held promise for discerning real sepsis. However, like sepsis therapeutics, identifying biomarkers has been a “Bermuda triangle” with many losing their way in the vast complexity. Sweeney and colleagues have made a substantial advance by using multiple data sources to identify and validate a pattern of gene expression that represents sepsis.

“The key to this technical advance is including the time of sepsis onset in their analysis of all the data sources; what they call an “integrated time-course-based multicohort.” By calibrating according to sepsis onset, the team was able to dissect the various factors involved in sepsis and identify 11 genes (of ~20,000) that accurately predict sepsis up to 90% of the time. In fact, based on the time-dependent profiles, it appears these markers may “herald” the onset of sepsis hours or days in advance, rather than diagnosing the condition once it has begun. This is incredibly valuable for patients at high risk for contracting sepsis.

“The caveat to sepsis biomarkers is in the heterogeneity of the syndrome. This study advances the field dramatically, but the patient population on which it is derived rests heavily on patients who were known to NOT be infected at the outset and later to develop infection that progressed to sepsis.

“Unfortunately the vast majority of sepsis cases enter the health system with a known infection (e.g. pneumonia, urinary infection, sinusitis) that has been present for an unknown duration. Whether these or different genes will perform in the more common group of patients with infection and concern for deterioration towards sepsis remains to be seen.”


Dr. David Maslove, Assistant Professor, Dept. of Medicine & Critical Care Program, Queen’s University & Kingston General Hospital (webpage):

Expertise: critical care informatics; high-resolution physiologic data capture from bedside monitors and time series gene expression profiling in sepsis and septic shock.

“Sweeny et al. identified an 11-gene signature that was able to distinguish patients with acute inflammation due to infection, from those with sterile inflammation due to other causes. This distinction can be difficult to make at the bedside using traditional clinical methods, highlighting the value of the complex computational approach that was used in this study. Since sterile inflammation (the systemic inflammatory response syndrome, or SIRS) and infectious inflammation (known as sepsis) require different treatment strategies, these findings could eventually form the basis of a change in clinical practice.

“One of the strengths of the methodology used is the extensive validation of the gene signature, which was tested in a number of independent datasets and found to perform well in each of these. What’s more, the authors accounted for the time of blood sampling in relation to the onset of infection, which they show to be a key factor in accurately identifying cases of sepsis. This aspect of the methodology is crucial since sepsis evolves quickly – over the course of hours and days – through various physiologically distinct states. 

“While the extensive validation used by the authors considerably increases the strength of their conclusions, they note that a dedicated prospective study is still required. Further work is also needed to translate this gene signature from the time consuming gene expression technologies used in this study, to a more rapid test that can be deployed at the point of care. Such a test might make use of molecular barcoding technology, or methods to measure specific protein levels in the blood of patients with suspected sepsis. Timely diagnosis and treatment are crucial, as each hour of delay confers a worse prognosis.”


Dr. David Flannery, Medical Director, American College of Medical Genetics & Genomics (webpage):

Expertise: Board Certified MD Clinical Geneticist and Board Certified Pediatrician.

“Inflammation is a general phenomenon in the human body. For example if you get a cut in your skin the process of healing involved inflammation. All kinds of injury to cells can cause an inflammatory reaction, so what the study authors were trying to do was figure out if there are specific patterns of genes that are expressed when you have inflammation that’s related to sepsis as opposed to sterile inflammation. The standard way to diagnose sepsis is to take blood cultures and wait for the bacteria to grow, which takes a day or more.

“The study authors took large datasets of gene expressions and started “mining” data and looking for correlations. Once they settled on the 11 genes identified, then they applied them to other datasets to see if they held up. And they held up pretty well.

“The big issue is that they haven’t yet got valuable information on what the predictive value of the test is – how often if a person has sepsis is the test correct, compared to how often false positives occur? They have analyzed a set of data in which they know the outcome of whether people have sepsis or not. But if you take people who present with symptoms of sepsis and run the test based on the expression of these 11 genes, you need to know what the likelihood of a patient having sepsis is when the test is positive. The authors mention that they want to do some prospective studies on this key issue.

“The other question is how long it takes to do the gene expression analysis. That becomes important in discriminating how good this test is compared to others. There are laboratory companies using proteomics to look at different proteins expressed, and there are others looking at PCR and molecular techniques for identifying the bacteria involved in sepsis rather than signs that the body is sliding into infection. A PCR test that is looking for a panel of commonly known bacteria that cause sepsis can be done in a matter of hours. If someone could get the time for a gene expression test down to an afternoon then it could potentially have a huge impact”


Dr. Daniel Henning, Acting Instructor, Division of Emergency Medicine, University of Washington (webpage):

Expertise: Sepsis/Shock, including the discrimination of infection and non-infection using clinical data and biomarkers, and monitoring of resuscitation.

“Early identification of infection is critical to the administration of time-dependent interventions, like antibiotics. For all of the classic signs and symptoms used to differentiate infection from non-infection, clinicians often find difficulty in establishing an infectious diagnosis, especially early in clinical care.  Improving the diagnostic process is an on-going effort, and the inclusion of novel biomarkers and gene expression assays provide promise for improving our clinical accuracy.

“This study employed publically available datasets exploring gene expression in patients with and without infection to determine if gene assays could improve the diagnostic process.  Using derivation and validation datasets, they constructed an 11-gene group that performed well discriminating the two groups in regression models.

“This is a hypothesis generating study that sets the stage for future investigation.  The future clinical use of gene assays to discriminate infection from non-infection faces several hurdles:  

  1. Assay results must be rapidly available.  
  2. These genes must be tested prospectively in a clinical context.  
  3. The test must be assessed for potential redundancy between expressed genes and clinical symptoms.   

“Overall, the study shows promise for new diagnostic modalities for a common and costly disease.  The actually use of gene expression to determine the presence of infection in a clinical setting still has a way to go, however.”


‘A comprehensive time-course–based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set’ by Sweeney et al, published in Science Translational Medicine on Wednesday 13 May, 2015.


Declared interests (see GENeS register of interests policy):

Dr. Greg S. Martin: Dr. Martin led a sepsis biomarker trial for which funds were paid from Abbott Laboratories to his institution, Emory University

No further interests declared

Please feel free to leave your comments below, but be aware that by doing so you agree to our Terms & Conditions.