We performed two independent human viral challenge studies (using influenza H1N1 and H3N2) to define the host-based peripheral blood gene expression patterns characteristic of the response to influenza infection. The results provide clear evidence that a biologically relevant peripheral blood gene expression signature can distinguish influenza infection with a remarkable degree of accuracy across the two strains. We have also defined the performance of the blood gene expression signature over time throughout the complete course of human influenza infection. Furthermore, despite arising from a controlled experimental challenge setting, we demonstrate that an influenza signature is able to accurately identify individuals presenting with naturally-occurring, RT-PCR confirmed H1N1 infection during the 2009 pandemic.
Defining the etiology of clinical syndromes in which infection is suspected remains challenging. Currently available influenza diagnostic tests exhibit highly variable sensitivity, ranging from 53 to 100% in various studies 
. Importantly, even those with powerful test characteristics such as RT-PCR are dependent upon sampling technique and inclusion of virus-specific components leading to reduced effectiveness with emerging viral strains 
. In addition to being less susceptible to sampling error, genomic signatures are not viral antigen or nucleic acid-dependent, and unlikely to be as strain-specific as pathogen-based platforms. Therefore, in addition to high sensitivity in the cohorts studied [92% (95% CI 79–99% for 2009 H1N1)], influenza gene signatures have the added potential of being able to identify, in the acute phase of illness, likely cases of infection with emerging influenza strains for which a specific diagnostic platform has yet to be developed and distributed. The nature of challenge studies limits our ability to make direct comparisons to other infected states – however, our previous work has demonstrated that genomic signatures similarly derived from viral challenges are capable of distinguishing upper respiratory viral infection from pneumonia due to Streptococcus pneumoniae .
These findings are promising but additional testing of these signatures in other models, including acute human cases of bacterial infection, will need to be performed to better delineate their specificity.
The unique design and frequent sampling involved in two experimental challenge studies has also given us the singular ability to examine the dynamics of temporal development of the genomic responses following exposure to infectious virus. We have shown that when viewed through the lens of the genomic response, it is possible to correctly distinguish individuals as infected or uninfected with influenza well before they have clinically relevant symptoms or would be ill enough to present for clinical evaluation. The potential power of this approach is manifested by full discriminative ability of the genomic signature as early as 53 hours post-viral exposure, at a time when the average clinical score of symptomatic individuals is only 2.4. Symptoms of this nature and severity are clinically vague and would be typical of very mild allergies 
or even symptoms due to sequelae of chronic smoking 
. Therefore, genomic analyses demonstrate the potential to identify viral infection either before symptoms emerge or among what otherwise are common, nonspecific upper respiratory symptoms, when early intervention with antiviral medications could have profound impact on both individual symptoms and disease transmission 
. Furthermore, we show that the overall trajectory of the Influenza Factor tracks closely with symptom scores over time, but also that the observed genomic response tends to significantly precede changes in clinical scores in symptomatic individuals. None of our affected individuals developed severe infection, but the characteristics of the timing and development of these signatures suggest that, similar to recent work with Dengue infections 
, genomic signatures may potentially prove invaluable for predicting clinical outcomes. However, further longitudinal studies with patients who eventually exhibit more severe disease will be required to fully assess this potential.
The nature of the individual components of the genomic response to influenza infection and the biological pathways they represent lend plausibility to this discovery. In particular, interferon stimulated pathways such as those including RSAD2, IRF7, MX1, OAS3, MDA-5, RIG-I and others are incorporated and thought to drive both innate and, to a lesser degree, adaptive immune responses to viral infection 
. Many of these pathways are consistent with those identified in acutely ill pediatric influenza subjects 
and recent studies of the genomic response following vaccination with live, attenuated influenza vaccine reported a profile of ‘immune activation’ which shares a number of genes with the Influenza Factor described here 
. Interestingly, a few genes which consistently feature prominently in the Influenza Factor are not clearly tied to inflammatory or immunologic pathways, and their significance remains unclear. Previously published work with bacterial respiratory infections has yielded quite different genomic results 
suggesting that some aspects of the host response are specific at least for major classes of pathogens (e.g., viral vs. bacterial). The genomic pathways identified suggest we are largely measuring indicators of the development/amplification of the immune response to the virus similar to previous work 
, and that these indicators parallel (and usually precede) clinical symptom development in time. The immunologic pathways observed in these studies that are known to be commonly activated early on at the primary site of infection (i.e., respiratory epithelium) 
, exhibit relatively delayed appearance in the periphery. This delay seems logical, as early innate responses at the site of infection would be expected to have an initially minor impact on global peripheral gene expression. At very early time-points (<53 hrs following exposure) insufficient numbers of peripheral cells are undergoing the conserved stimulation required to produce a significant change in global gene expression, at least as detected by microarray analysis. This raises the possibility that more sensitive methods of detecting genomic changes, such as individual cell-type sampling or RT-PCR of select genes, will prove to be even more precise at early time points in the evolution of viral infection. Additional work will be essential (and is underway) to further define the nature and biological implications of these data, as well as to work towards development of a more practical means of assaying these changes in the clinical setting, such as RT-PCR of select ‘core’ genes from signatures like the one described herein.
Clearly, great care must be taken when analyzing and applying host genomic data from human challenge studies where the means of transmission of the virus is experimentally designed rather than ‘natural’, and the degree of illness which follows is not always typical of the severity seen in naturally acquired infection in subjects who present for clinical care, even though it does tend to mimic the overall character of natural clinical disease 
. Hosts in these studies are universally young, healthy individuals at minimal risk for developing severe complications, which may limit the broad applicability of such findings, although this is somewhat mitigated by the strong performance of the gene signatures despite significant clinical variability in infected subjects. It is also important to note that while this type of factor analysis allows for description of conserved biological pathways indicative of influenza infection, a given factor only represents a limited interrelated subset of all genes that are globally up- or down-regulated in response to a given condition, and thus does not describe the entirety of the genomic response.
Despite these limitations, we have for the first time defined the temporal dynamics of a genomic signature driving the host response to influenza infection in humans. These molecular and statistical techniques combined with the ability to longitudinally study exposed human hosts have given us the opportunity to examine periods of human disease which have previously been largely unexplored. Moreover, despite being developed in an experimental challenge model, this host genomic signature performs at a high level of accuracy in the setting of naturally acquired pandemic 2009 H1N1 infection. This work demonstrates that analyses of the temporal development of gene expression signatures shows promise both for creating diagnostics for early detection, as well as providing insight into the biology of the host response to influenza and other pathogens.