Ebola virus causes severe and often lethal hemorrhagic fever in humans and nonhuman primates. Ebola virus (EBOV) is one of two genera that comprise the family Filoviridae. The EBOV genus consists of four distinct species: Ivory Coast Ebola virus, Reston Ebola virus, Sudan Ebola virus, and Zaire Ebola virus (ZEBOV) 1. Sudan Ebola virus and ZEBOV have been associated with human disease outbreaks in Central Africa, with case fatality rates averaging about 50% for Sudan Ebola virus and ranging from 75% to 90% for ZEBOV 2. Although Reston Ebola virus is highly lethal in nonhuman primates 34, the few data available suggest that it is nonpathogenic in humans 5. The pathogenic potential of Ivory Coast Ebola virus is unclear because there has only been a single confirmed nonfatal human case 6 and a second suspected nonfatal case 7. In addition to natural outbreaks, EBOV is an important concern as a potential biologic threat agent of deliberate use because these viruses have low infectious doses and clear potential for dissemination by aerosol route 8. Currently, there are no approved preventive vaccines or postexposure treatments for EBOV hemorrhagic fever, but recent advances have led to the development of several candidate therapeutics and vaccines for EBOV 91011.

The mechanisms of EBOV pathogenesis are only partially understood, but dysregulation of normal host immune responses (including destruction of lymphocytes 2 and increases in levels of circulating proinflammatory cytokines 12) is thought to play a major role. Several animal models of EBOV hemorrhagic fever have been developed, notably a cynomolgus macaque (Macaca fascicularis) model 1314, which closely resembles human infection 215. ZEBOV infection in cynomolgus macaques results in uniform lethality at days 6 to 7 after infection 16171819.

The majority of studies conducted in nonhuman primates have focused on end-point examination when animals are in the final stages of disease, and have restricted their analyses to small numbers of cytokines or mRNA transcripts. cDNA microarrays have been used by our group to study mechanisms of viral pathogenesis in a nonhuman primate model of an agent, albeit unrelated, that also causes overwhelming, systemic infection 2021. In order to understand better the early events in EBOV pathogenesis, we examined global changes in gene transcript abundance, using cDNA microarrays, in sequential blood samples from 21 cynomolgus macaques over the entire time course of ZEBOV infection.

In a series of studies we recently analyzed the pathology of lethal ZEBOV infection in cynomolgus macaques using a sequential sacrifice design 1314. In the present study, we examined the genome-wide transcriptional responses in sequential samples of peripheral blood from 15 of these cynomolgus macaques. Nonhuman primates infected with ZEBOV exhibited a highly homogeneous, time-dependent pattern of gene expression (Figure 1). Given the massive pathologic changes, physiologic instability, and widespread tissue damage, as well as the commonly observed variability in genome-wide transcript abundance patterns among different individual hosts ex vivo, it was surprising that the animals displayed such uniform patterns. Perhaps because of the overwhelming nature of the infection and the relatively short time frame between the first appearance of signs and death, these patterns are highly homogenous due to an effect akin to temporal compression. It is very likely that the observed gene expression patterns reflect many physiologic changes caused by systemic filoviral infection (for example, bystander lymphocyte apoptosis, fibrin deposition, and anti-viral IFN response). With a longer time frame or lower mortality rate, it is possible that individual host responses might show more variation; nonetheless, the homogeneity of this response allowed us to analyze the characteristic gene expression patterns with minimal noise from animal-to-animal variation. The underlying molecular changes echo the uniform lethality of the animal model, and may provide better predictors of morbidity/mortality than a model with high levels of inter-individual variation.

We observed a marked increase in transcript abundance for genes encoding many cytokines, including IL-1β, IL-6, IL-8, MIP-1α, MIP-1β, macrophage colony stimulating factor, and MCP-1 (Figure 2), which is consistent with a systemic proinflammatory response. Reports of cases of human EBOV infection vary considerably with respect to the cytokines that are associated with fatal as opposed to nonfatal outcome 122529. Increases in IL-1β, IL-6, MIP-1α, and MIP-1β have been reported for human survivors of EBOV infection 25. In vitro infection of human monocytes/macrophages with authentic EBOV or virus-like particles that include membrane-associated GP_1,2 leads to increases in protein levels of IL-1β 303132, IL-6 303132, IL-8 3132, MIP-1α 3033, MIP-1β 33, and MCP-1 30. In monkeys infected with ZEBOV or Reston Ebola virus, increases in IL-1β 1434, IL-6 143334, MIP1-α 1433, MIP-1β 143334, and MCP-1 1434 have been reported. Monocytes and macrophages represent a major cellular target for infection and dissemination of EBOV in monkeys 14353637. Infection of monocytes and macrophages leads to increased production and release of proinflammatory cytokines, leading in turn to recruitment of macrophages to areas of inflammation, which may contribute to viral proliferation and eventually an overwhelming sepsis-like syndrome 143839.

Serum levels of TNF-α, in particular, are demonstrably increased in human 1229, primate 1433, and in vitro 30313233 EBOV infection. Wahl-Jensen and coworkers 40 recently showed that the virus-like particle induced decrease in endothelial barrier function was further enhanced by TNF-α, which is known to induce a long-lasting decrease in endothelial cell barrier function and is hypothesized to play a key role in EBOV pathogenesis 40. We detected an increase not only in TNF-α but also in the downstream transcriptional response that is regulated by TNF-α and NF-κB (Figure 3), providing evidence that the circulating cells are responding to the large amounts of TNF-α that are induced during infection. Induction of the NF-κB pathway by TNF-α usually induces an anti-apoptotic response and cell survival 41, possibly reflecting a mechanism by which EBOV counteracts host apoptotic defenses in the infected cell, thereby contributing to viral spread.

Despite the known role of the NF-κB pathway in an anti-apoptotic response, we found that transcripts for many pro-apoptotic genes were induced (Figure 4). Genes for the Bcl antagonists BCL2-antagonist of cell death, BH3 interacting domain death agonist, BCL-X, and BAK appeared to be induced in the later stages of infection; all of these factors promote apoptosis by inhibiting Bcl-2. Expression levels of IL-1α and IL-1β were also increased; these cytokines are proteolytically processed and released in response to cell injury and induce apoptosis. Both forms of IL-1 are proteolytically processed to their active form by caspase 1, which was also expressed. In addition, transcript levels of TRAIL were markedly increased (Figure 4b). TRAIL expression early during infection and induction by IFN-α may contribute to lymphocyte apoptosis 33. In view of the increased transcript levels for a group of pro-apoptotic genes, the decrease in lymphocyte related transcripts, including CD3, CD8, CD19, CD64, major histocompatibility complex class II, T cell receptor β, integrins, and granzymes (Figure 1) in ZEBOV infection may result from 'bystander' lymphocyte apoptosis and subsequent depletion of lymphocytes in circulating peripheral blood 14232425. Thus, it appears that although the infected monocyte/macrophage lineages can survive and carry virus to secondary infection sites in the tissues, cells important for the adaptive immune response are decimated through bystander lymphocyte apoptosis, preventing an effective adaptive immune response, and enabling further virus propagation and spread.

Although the major transcriptional changes appeared on days 4 to 6, corresponding to the initial appearance of clinical signs, a strong IFN response was evident at day 3 after infection (Figure 5a), and transcripts levels for a subset of IFN genes increased as early as 24 hours after infection (Figure 8). In addition, expression of the classical IFN induced protein MxA was detected in circulating cells (Figure 5b). Several studies have reported the detection of IFN-α in serum from EBOV-infected humans 12 and monkeys 1433, and our results provide evidence that cells in circulating peripheral blood can mount a robust transcriptional response to the IFN stimulus, despite the presence of EBOV proteins (VP24 and VP35), which are thought to function as type I IFN antagonists 4243. This might imply that the major role of the ZEBOV type I IFN antagonists is to act locally to influence the microenvironment of the infected cell, rather than to shut down a systemic IFN response. The majority of cells in the peripheral blood sample (PBMCs) are uninfected, because no evidence of EBOV infection of lymphocytes has been observed 1423 and the circulating population of infected monocytes/macrophages constitutes only 1% to 13% of PBMCs in these primates. Both VP35 and VP24 act in a cell autonomous manner; VP35 blocks activation of the IFN regulatory factor 3 and the transcriptional responses of the IFN regulatory factor 3 responsive promoters 44, and VP24 blocks nuclear accumulation of tyrosine phosphorylated STAT through interaction with karyopherin α₁ 43. Because of the cell autonomous nature of the EBOV IFN antagonists, uninfected cells should still be capable of producing a transcriptional response to the large amounts of circulating IFN, as shown in Figure 5a.

Disseminated intravascular coagulation, caused by over-activation of the coagulation system and resulting in microvascular thrombosis 45, may contribute to the lethal multi-system organ failure in EBOV infection. Over-expression of tissue factor in EBOV infected monocytes/macrophages has been shown to produce fibrin deposition in the spleen, liver, and blood vessels of infected macaques 46, and inhibition of the tissue factor/factor VIIa pathway resulted in a decrease of D-dimers (fibrin degradation products) and an increased survival rate in rhesus macaques 47. In this study we found evidence of cellular responses that would be expected to lead to increased fibrin degradation. There was an increase in both uPA and uPA receptor transcripts in PBMCs (Figure 6a,c,d), accompanied by an increase in serum concentrations of uPA protein (Figure 6b). uPA acts to convert plasminogen to plasmin; the uPA receptor mediates the proteolysis independent signal transduction activation effects of uPA, also promoting plasmin formation. However, we also observed an increase in transcripts encoding plasminogen activator inhibitor, perhaps caused by negative feedback regulation. Thus, the overall impact of the observed transcriptional response on the coagulation cascade is not self-evident. Nevertheless, although the majority of the coagulation and fibrinolytic cascade is regulated at the protein level through processing, transcriptional induction of genes that are involved in fibrin degradation may be a factor in the coagulopathy during EBOV infection.

EBOV GP is regulated by complex transcriptional editing and post-translational cleavage processes. The authentic transcript of the GP gene is expressed as a polypeptide, which is cleaved into soluble glycoprotein (sGP) and the secreted delta peptide 4849. Through RNA editing, the transmembrane form of GP is expressed (GP_1,2) and then cleaved into GP₁ and GP₂ disulfide linked fragments, which are present on the surface of virus particles 505152. The role of EBOV GP and its contribution to pathogenesis has been the subject of much investigation. GP can decrease the expression of cell adhesion molecules, interfering with cell attachment and inducing cytotoxicity 53545556, but mutant viruses that fail to produce sGP are more cytotoxic, suggesting a negative regulation by sGP of the GP induced cytotoxicity 57. In vitro studies suggest that GP_1,2 on the surface of virus-like particles, but not sGP, activates target cells 31 and decreases endothelial barrier function 40. However, EBOV replication does not induce direct cytolysis of endothelial cells either in vitro or in animal models of EBOV infection 13, although cytolytic infection of human umbilical cord vein endothelial cells has been demonstrated with Marburg virus 58.

TNF-α converting enzyme/ADAM-17 was recently found to mediate proteolytic processing and shedding of the ectodomain of Ebola GP (GP_1,2Δ) 28. We found that transcript levels for ADAM-17 increased on days 4 to 6 after infection, peaking at day 5 after infection (Figure 7a,b), which is consistent with a role for ADAM-17 in shedding of GP_1,2Δ during in vivo primate infection. In addition, we also detected elevated concentrations of cleaved GP_2Δ in sera from late-stage ZEBOV infected animals compared with uninfected controls (Figure 7c), demonstrating that cleavage of GP also takes place during in vivo infection in a nonhuman primate model of EBOV hemorrhagic fever. The relationship of shed GP_1,2Δ to pathogenesis/disease severity is unclear, and its role during in vivo infection remains to be investigated. It is possible that GP_1,2Δ can act as a decoy and soak up anti-EBOV antibodies, effectively shielding the virus from the immune system 28.

The composite gene expression pattern assayed in a mixed cell population, such as PBMCs, gives a rich and multidimensional picture of the systemic host responses to infection, reflecting many interconnected responses of a complex system. However, because the observed gene expression pattern represents a composite of diverse influences, data from mixed cell populations can be more difficult to interpret. In samples of whole blood or PBMCs, large variations in the cellular composition are often the largest source of overall variation in the observed gene expression patterns 59. We correlated the gene expression vector for each individual gene in the dataset with complete blood count and differential data on relative levels of individual cell populations (Additional data file 3). Correlation scores were highest between cell populations that increased during the course of infection and the cluster of genes whose transcripts levels were increasing, and also cell populations that decreased and the cluster of genes whose transcripts levels were decreasing. As an example, the depletion of circulating lymphocytes during EBOV infection in vivo 24 correlates with the decrease in lymphocyte-related transcripts in our microarray dataset (Figure 1). However, given the mathematical simplicity of the gene expression and differential parameters (elements either increase or decrease uniformly and uniderectionally with respect to time), the correlation scores could be due to temporal coincidence of processes with similar directionality.

These correlations indicate that there is a certain proportion of the variation in gene expression that might be explained by changes in cell population. However, it is difficult to determine whether the variation in gene expression is directly caused by changes in the composition of the mixed cell population, activation of cells, or coincidence of temporal processes. Although correlative analysis based on numbers of PBMC types may provide the basis for attributing variation in expression to specific cell subsets, future experiments examining filovirus infection in more homogenous cell subpopulations will be essential for reliable identification of cell type specific responses. Likewise, examination of different tissues during the course of infection could provide a more comprehensive picture of the molecular anatomy of host responses to EBOV on an organism-wide basis.

Early detection and diagnosis of EBOV infection would be invaluable, as many of the symptoms and signs are nonspecific at presentation 2, and some recently described interventions have benefit when given early after filovirus infection 4760. To determine whether we could detect any changes in gene expression before appearance of symptoms, we examined the gene expression profile of peripheral blood in the early preclinical stages of infection (days 1 and 2). We observed changes in expression of over 200 genes before any clinical signs were evident and before plasma viremia was detectable in the ZEBOV infected nonhuman primate model (Figure 8a,b). This gene set is a possible starting point for the identification of early diagnostic markers. Any of these genes alone may provide little specificity, but combinations of these and/or others may allow differentiation among different etiologic agents, all of which may otherwise demonstrate similar clinical pictures. Although these early response genes are all responsive to diverse infections and inflammatory conditions and thus are of limited specificity for detection and diagnosis, the pattern of response to EBOV infection, in a setting with a high index of suspicion, may provide useful early warning for triage, aggressive treatment, and/or quarantine.

Analysis of global gene expression as disease pathogenesis unfolds provides a multifaceted picture of the complex interplay between host and pathogen. Examination of early events during infection may help to identify and provide insight into the specific molecular processes that initiate the cascade of host damage during EBOV infection. The ability to detect gene expression patterns before clinical symptoms may provide an opportunity for early diagnosis.

ADAM, α-disintegrin and metalloproteinase; BCL-X, BCL2-like 1; EBOV, Ebola virus; ELISA, enzyme-linked immunosorbent assay; GP, glycoprotein; IFN, interferon; IL, interleukin; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; MX, myxovirus resistance protein; NF-κB, nuclear factor-κB; PBMC, peripheral blood mononuclear cell; sGP, soluble glycoprotein; STAT, signal transducer and activators of transcription; TBS-T, Tris-buffered saline with 0.05% Tween-20; TNF, tumor necrosis factor; TRAIL, TNF related apoptosis inducing ligand; uPA, urokinase plasminogen activator; ZEBOV, Zaire Ebola virus.

KHR conceived, designed, and executed the experiments described in this report and wrote the manuscript. LEH and TWG designed and performed the animal studies. VJ and KMD performed the Western blot experiment. HAY performed the RNase protection assays. POB, DAR, LEH and TWG oversaw completion of the studies as well as the final manuscript. All authors read and approved the final version of the manuscript.

The following additional data are available with the online version of this paper. Additional data file 1 provides animal numbers for blood samples. Additional data file 2 shows the bleed schedule. Additional data file 3 shows the gene expression correlation with immune cell types.