Introduction

West Nile virus (WNV) is one of the leading causes of arboviral encephalitis in the United States for both horses and humans. The virus can be devastating in its ability to cause long term neurological deficits and even death. Since WNV was introduced into the United States in 1999, 25,791 clinical cases of disease have been confirmed in horse [1]. Similarly, in humans, 30,625 confirmed cases of clinical disease have been reported with 1,202 deaths [2]. The arbovirus season of 2010, ten years after introduction, was notably active with WNV activity detected in all continental states. Human cases were detected in 42 states with 89 infections reported in New York and 105 infections detected in the opposite extreme geographically in Arizona. Most of the cases reported were of the neuroinvasive form and 45 deaths occurred. The US blood supply continues to be at risk with 144 detections of viremic blood obtained from donors in 2010 alone [2].

Characterization of WNV infection provides an opportunity to enhance our understanding of the pathophysiology and immunopathology of viral encephalitis in general. This is particularly important in regards to host-pathogen interactions during infection. During neuroinvasive disease, WNV mainly infects the neurons of the midbrain, hindbrain, and spinal cord, with limited infection of glial cells. Direct viral infection of CNS cells followed by induction of the inflammatory response allows analysis of the virus from systemic infection, to neurological spread, through stages of innate and adaptive immunity for the purposes of biomarker and interventional strategies. Investigation of one of the two most commonly affected natural mammalian hosts allows modeling of this disease under both experimental and field settings.

WNV infection in natural mammalian hosts is characterized clinicopathologically by an increasing number of lesions progressing from the diencephalon through the hindbrain and down through the spinal cord. Congestion of the meninges and hemorrhagic foci may be seen. Comparatively, mice develop disease that is primarily focused in a cerebral and cerebellar localization of virus. Histopathologically, inflammatory lesions characterized by layer of monocellular perivascular cuffing and gliosis are present in horses, humans, and mice, while necrosis is the primary lesion in hamsters [3], [4]. This pathology is likely a result of many complex factors, as there is increasing evidence that the host response to viral infection is one of the main causes of pathology and survival during WNV infection in the CNS [5], [6]. In humans and horses, long term recovery responses are variable, with evidence of moderate to severe congnitive, emotional, and motor deficits in humans, and behavioral and motor deficits in horses [7]. Thus there is a gap in our understanding of the interactions between various components of the natural host response to infection, and how these interactions can lead to pathology, long term deficits, or full recovery. Given the similarities between disease in horses and humans, equine tissue analysis offers an unparalleled opportunity to profile gene expression and pathway interactions between pathogens and the CNS.

Although there are many new methods for profiling gene expression, there is limited development of de novo deep sequencing strategies due to limited financial resources and species specific bioinformatics in veterinary research fields. In terms of adaptability and computational resources, microarrays allow rapid acquisition of tissue specific expression data for many non-model species. Microarrays have facilitated the study of the Flaviviridae in multiple applications including detection of variants of Dengue virus (DV) in human samples [8], differentiation between different flaviviral and other viral infections [9], [10], and mutations in the structural regions of the WNV genome [11]. Microarrays have been used to analyze gene expression at both the cell culture and organism level for DV, JEV, and yellow fever virus (YFV) infection [12], [13], [14], [15], [16], [17]. Although several species specific arrays exist, there is limited development of tissue specific arrays in veterinary medicine. The microarray currently developed in the horse is based on global gene expression in which multiple tissues of the equine transcriptome were sequenced [18]. While a valuable tool, only cerebrum, cerebellum and spinal cord were utilized in this work to generate transcriptome data.

Two studies have used microarrays to examine gene expression changes in response to WNV infection. In one, human glioblastoma cell culture transcriptional responses to WNV were analyzed, and 23 genes involved in neurodegenerative disorders were shown to be changed in expression [19]. In the other, a microarray was used to analyze whole organism gene expression response to WNV strains of different neurovirulence in a mouse model [20]. Genes involved in immunological, neurological, and apoptotic functions were differentially regulated (with forty-seven genes shown to be upregulated in highly neuroinvasive strains).

No studies are available that profile gene expression in the CNS of animals infected with WNV that are considered natural, susceptible hosts. These data could provide detailed information on the host response to infection and on a pathogen's specific manipulation of the host response, and will also allow more efficient analysis of essential pathways in model species such as mouse and hamster. This report provides sequencing data from the equine brain transcriptome and lymphoid system from naïve horses experimentally infected with WNV, vaccinated horses experimentally infected with WNV, and negative controls. Also described is the construction of a custom, validated equine high density microarray, in which pathways of the CNS and immune system were enriched. The microarray was used to profile gene expression changes in the thalamus and cerebrum of naïve and vaccinated horses during experimental WNV infection (and negative controls) and common gene pathways were identified. These data were used to detect differences in gene expression associated with exposure to WNV, survival from WNV infection, and location of WNV infection in the brain.

Materials and Methods

Sample Selection

All tissues used in this study for library creation and array analysis were derived from horses used in an experimental intrathecal challenge model wherein naïve horses developed grave West Nile (WN) encephalitis (100% nonsurvivorship) and all vaccinated horses did not develop clinical disease (100% survivorship). Samples for pyrosequencing and array analysis consisted of 1) naïve horses infected intrathecally with 1×10⁵ WNV/NY99/crow, 2) non naïve horses vaccinated utilizing a modified live attenuated Yellow Fever (YF) chimera vaccine for protection against WNV (Prevenile, Intervet-Schering-Plough) and infected intrathecally with 1×10⁵ WNV/NY99/crow , and 3) horses that were not infected or vaccinated. Experimental infection and vaccination of horses occurred according to previously published data [4], [21], [22]. Horses from groups 1 and 2 were euthanized (University of Florida IACUC protocols #F077, #F093, #D163) if demonstrating clinical signs or at the end of the study (day 21) if not demonstrating clinical signs. Horses from group 3 were normal healthy horses, not infected with WNV and were euthanized due to other causes (4 of these horses were less than 4 years of age, while 2 were over the age of 9). These causes included lameness (angular limb deformities, chronic lameness) and humane reasons as surrender by owner. None of the horses were ill or demonstrating neurological signs at the time of euthanasia (all horses underwent neurological exams). All horses were necropsied immediately upon euthanasia. Tissues were snap frozen in dry ice and ethanol and stored at −80°C until RNA extraction was performed. Eight tissues were collected from each horse and included cerebrum, cerebellum, thalamus, midbrain (rostral and caudal colliculus, tectum, and tegmentum), hindbrain (pons and medulla oblongata), cervical spinal cord, lumbar spinal cord, and spleen.

A pooled tissue strategy was used to create a library that was normalized for rare sequences that might not have been annotated in the EqCab2 database, and that allowed for longer reads than that accorded by directly sequencing short reads from each individual horse. A cDNA library was constructed from a pool of tissues mentioned above from six different horses including two gravely ill naïve horses infected with WNV; two horses vaccinated, infected, and recovering from WNV(vaccinated, intrathecally infected with WNV, and survived without showing clinical signs that warranted euthanasia according to IACUC protocols); and two control horses that were not infected or vaccinated. Eight tissues taken from each horse were pooled together for a final total of 5 samples from individual horses, each containing the eight tissues. Brain tissues used to create the cDNA for dye labeling on the array were obtained from the archived tissues of three groups of six horses each (total of 18 individuals) mentioned above. Tissues used in the array included cerebrum and thalamus (one section from each horse for a total of 36 samples).

Three analyses were established to test the hypothesis that there are gene pathways whose expression changes in a significant and consistent manner due to WNV as a result of exposure status, survival/immune status, and CNS location (Table 1). With respect to the experimental analyses, three subhypotheses were generated to analyze if there was a difference in gene expression between the nonvaccinated/exposed and untreated horses (exposure), the nonvaccinated/exposed and vaccinated/exposed horses (survival), and the nonvaccinated cerebrum and nonvaccinated thalamus (location). In particular, the “survivors” represent the gene expression status of those animals that recover (vaccinated, intrathecally infected with WNV, and survived without showing clinical signs that warranted euthanasia according to IACUC protocols) from grave WN encephalitis through induction of vaccine mediated immunity, and the “non-survivors” represent the gene expression status of naive animals undergoing grave encephalitis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Samples and analyses for the array.

https://doi.org/10.1371/journal.pone.0024371.t001

Results

Annotation and Analysis of the Sequenced Equine CNS Transcriptome

Library Sequencing and Annotation.

The quality and purity of the RNA samples (average RIN 7.1–7.6) and cDNA libraries can be seen in Table S2. Successful normalization was confirmed via visualization of the bands on a gel and a titration sequencing run. In total, 308 contigs were identified with an average length of 208.1 base pairs. Only 4 large contigs were identified, confirming that normalization was effective.

After linkers and repetitive sequences were removed, 514,462 sequences were assembled from a total of 49,857,586 bases using Newbler Assembly filter software. Fully assembled reads, partially assembled reads, singletons, repeats, outliers and contigs were identified. A total of 16,895 contigs were identified (1,902 large contigs) (Table S3). The sequences obtained from the three assembly analyses included 16,895 contigs obtained from the Newbler assembly, 443,584 unannotated sequence sets from the 454 sequencing run, and 22,748 sequences from Ensembl. These sequences were databased and assembled using PTA for a total of 188,885 final sequences (Table 2). Clusters were determined (overlap 100 base-pair minimum) and 61,499 sequences matching with ‘seed’ sequences were clustered. Sequences that were not identified as ‘seed’ sequences (127,386) were clustered in three steps via partitioning, pairwise comparison, and clustering.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Data from 454 Sequencing Runs Paracel Transcript Assembler.

https://doi.org/10.1371/journal.pone.0024371.t002

Combining the results from Newbler and PTA analyses and utilizing known sequences from the equine genome databases, 19,987 contigs and 21,053 genes were assembled. Unassembled sequences were not considered for further analysis. In total, 41,040 sequences were used in the BLAST analysis [19,679 PTA contigs ( = 11,621 non-seeded contigs+8,058 seeded contigs), 308 newbler contigs and 21,053 unassembled EquCabv2 genes]. These sequences have been submitted to GenBank (NIH) for public access (study # SRP000619).

Library Analysis.

A BLAST search was run against five separate databases (e-value 10⁻⁴ ) and 31,357 good sequence hits were obtained (Table 3). Approximately 73.7% of the sequences identified in this project were matched in the equine chromosome database and at least one of the equine predicted genes databases, while 23.1% of the sequences recognized in this project were only identified in the equine chromosome database. Completely novel genes yet to be identified by the equine genome project were represented by 3.1% (1,278) of the sequences identified. The average HspScores and bit scores for all equine databases indicated a high degree of alignment of the sequences with those in the equine databases. All sequences demonstrated significant length for matching. In addition, the majority of sequences for all databases demonstrated positive identity greater than 95% (Table S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of BLAST Results for Five Separate Databases.

https://doi.org/10.1371/journal.pone.0024371.t003

Gene ontology analysis was performed on the library. Consistent with the utilization of CNS and immune tissues, a heavy bias towards those genes involved with the immune system, the CNS, and programmed cell death (apoptosis) was identified by GO analysis of the assembled contigs and genes. This methodology also allowed the identification of novel genes involved in the immune response and neurologically specific genes in equine tissues (Figure S1a, S1b).

The sequences were analyzed using the NCBI database and matching to known species in the database made. With an e-value≤10⁻⁴ for all databases, there were 39,257/41,040 accurate hits for both the NR and NT NCBI databases. For the NR database alone, there were 30,011 accurate hits with 26,955 ‘perfect’ (e-value = 0) hits, while for the NT database, there were 39,217 accurate hits with 32,825 ‘perfect’ hits. For both databases, matches to horses comprised the greatest number of hits with the species groups humans, primates, canines, and bovines containing the next greatest number of hits (Figure S2a, S2b).

All sequences that were not previously identified (novel) were analyzed. Of the 1,280 (3.1%) sequences not recognized by any of the equine databases, 709 (55.4%) were classified into gene ontology databases. With overlap, 579 could be classified into biological processes, 592 into cellular component, and 619 into molecular function. The average length of all these sequences was 595 base pairs with a range of 50–8,802 base pairs (Figure S3).

Finally, the sequences were compared to the human expressed sequence tag database (e-value≤10⁻⁴ ) (Table 4). A high degree of sequence homology between the human and equine transcriptome was found, with percentage matches of 69.27% (contigs), 78.13% (seed contigs), and 80.17% (singlets). Equine sequences demonstrated good match to the human EST database with an average percent identity of 90.17%, an average bit score of 512.91, and an average alignment length of 424.85.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Summary of BLAST Analysis of Sequenced Equine Transcriptome to the Human Expressed Sequence Tag Database.

https://doi.org/10.1371/journal.pone.0024371.t004

Gene Expression Analysis

With respect to the experimental groups, three analyses were performed to determine if there was a difference in gene expression based on exposure status (the thalamic nuclei of the infected vs noninfected, nonvaccinated horses), immune status/survivorship (the thalamic nuclei of the infected vs infected, vaccinated horses), and location (the thalamic nuclei of the infected nonvaccinated cerebrum and infected nonvaccinated thalamus). These experimental groups and the tissue types tested are outlined in Table 1.

Loess normalization was performed on all arrays and confirmed by distribution analysis (Figure S4). For all arrays, the majority of variance was accounted for in the first three principal components (x, y, and z) with Eigenvalues (percents of variability) of the each component, 11.09 (30.81%), 4.94 (13.71%, and 3.21 (8.94%), respectively. In addition, the mean of the R² was 0.939392 (range 0.8781–0.9871) for all arrays. A heat map and dendrogram was generated between all arrays (Figure S5).

Statistical analysis was performed using analysis of mean relative difference in gene expression. An ANOVA with interactions between treatment and location revealed significant differences in gene expression (p<0.05) for all comparisons, including those between exposure status, survival status, and CNS location. For the normal, non-challenged horses, there were 6,911 genes that showed showed significantly different levels of expression between cerebrum and thalamus after duplicate removal. Therefore, for the exposure and nonsurvival groups, only thalamus was compared. The three analyses were compiled and 3,421 of the same genes were significantly altered in all comparisons.

For exposure status, significant differences in gene expression in the thalamus were seen between nonvaccinated/exposed and normal, nonexposed horses for 9,020 genes (Table S5). When analyzed solely by fold change, 2,936 genes decreased by <−1.0 (395<−2.0) and 2,084 increased by >+1.0 (749>+2.0). For immune/survival status, significant differences in gene expression in the thalamus were seen between nonvaccinated/exposed horses (nonsurvivors) and vaccinated/exposed horses (survivors) in 7,395 genes (Table S5). In the nonvaccinated, nonsurvivors, 2,123 genes were decreased by <−1.0 (225<−2.0) while 1,800 were increased by >+1.0 (666>+2.0) compared to the vaccinated, survivors. Finally, where analyzed by location in the brain, significant differences in gene expression were seen between the cerebrum and thalamus of nonvaccinated horses exposed to WNV (location) for 7,649 individual genes (Table S5). For the location analysis, 2,053 genes were decreased by <−1.0 (609<−2.0) while 1,827 were increased by >+1.0 (406>+2.0).

Exposure Status.

Gene ontology was explored for the exposure status. The first subhypothesis asked whether there was a difference in gene expression due to exposure to WNV between nonvaccinated horses exposed to WNV and normal horses not exposed to WNV. Genes that were found to have significant differences in expression were classified into the gene ontology (GO) categories of biological process, cellular component, and molecular function (Figure S6a). Within these categories, 2,022 genes mapped to components of transcription and RNA processing, 1,081 mapped to neurological categories, 983 genes mapped to immunological categories, and 420 mapped to cell death or apoptosis (430) (Figure S6b).

Canonical pathways were first examined for interactions between multiple significant transcripts in the exposure status. The majority of pathways were involved with some aspect of cell signaling for all groups (Table S6). Seventeen neurological pathways were identified with significantly changed transcriptional levels when nonvaccinated horses exposed to WNV were compared with normal horses not exposed to WNV (Figure 1). The canonical pathways involved consisted mainly of neurotransmitter and signaling pathways with extensive involvement of glutamate receptor signaling (Figure 2a, Table 5) and dopamine receptor signaling (Figure 2b, Table 5). Neurological pathways containing the most molecules when the WNV infected horses were compared to the noninfected horses (15 or more) included axonal guidance signaling, CREB signaling in neurons, synaptic long term depression, and Huntington's disease signaling (Figure 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Neurological canonical pathways for all analyses.

Canonical pathways identified as significant for each analysis were selected. There was a high degree of overlap between all three analyses in neurological pathways. The location analysis contained transcripts that mapped to the most neurological pathways. The green line represents significance. For the purposes of this study, ‘exposure’ represents the difference in gene expression between the nonvaccinated/exposed-normal, ‘survival’ represents the difference in gene expression between the nonvaccinated/exposed-vaccinated/exposed, and ‘location’ represents the difference in gene expression between the thalamus and cerebrum of the nonvaccinated/exposed.

https://doi.org/10.1371/journal.pone.0024371.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Glutamate and Dopamine Receptor Signaling Pathways.

(A) The diagrams represents signaling in the synaptic cleft (square = pre-synaptic neuron, oval = post-synaptic neuron, circle = glial cell). Green represents downregulation of pertinent receptors, red represents upregulation of pertinent receptors. WNV induces downregulation of glutamate receptors on the post-synaptic cleft as well as glutamate uptake receptors on glial cells. (B) Dopamine receptor signaling pathway. The graphic shown is for the exposure analysis, but was similar in all analyses. Green represents down-regulation of transcripts, red represents upregulation. Dopamine receptors (DRD5) and downstream signaling pathways as well as enzymes that create dopamine (TH) were downregulated, while MAO (enzyme that degrades dopamine) was upregulated. This network was generated through the use of IPA (Ingenuity Systems, www.ingenuity.com).

https://doi.org/10.1371/journal.pone.0024371.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Levels of Expression of Transcripts in Neurological Pathways of Note.

https://doi.org/10.1371/journal.pone.0024371.t005

Regarding immunological canonical pathways and WNV exposure status, 47 pathways were identified (Figure 3) and involved in both the innate and adaptive response. Pathways in which there was upregulation in horses exposed to WNV included the IL-15 signaling pathway, the IL-22 signaling pathway, the IL-9 signaling pathway, and the Interferon Signaling Pathway (Figure 4a,b, Table 6). Multiple pathways involved in apoptosis were dysregulated (both up and down) in horses with exposure to WNV and these included the retinoic acid mediated apoptosis signaling, calcium-induced T lymphocyte induced apoptosis, cytotoxic T lymphocyte mediated apoptosis of target cells, induction of apoptosis by HIV1, and April mediated signaling.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Immunological Canonical Pathways for All Analyses.

The location analysis contained transcripts that mapped to the most immunological pathways. The green line indicates significance. For the purposes of this study, ‘exposure’ represents the difference in gene expression between the nonvaccinated/exposed-normal, ‘survival’ represents the difference in gene expression between the nonvaccinated/exposed-vaccinated/exposed, and ‘location’ represents the difference in gene expression between the thalamus and cerebrum of the nonvaccinated/exposed.

https://doi.org/10.1371/journal.pone.0024371.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. IL-15 Production Pathway and IL-22 Signaling Pathway.

(A) The diagrams represent the different methods of IL-15 production. Green represents downregulation of pertinent molecules, red represents upregulation of pertinent molecules. WNV induces upregulation of the production of IL-15. (B) Red represents upregulation of gene expression, green represents down-regulation of gene expression. The JAK/STAT portion of the pathway is upregulated during viral infection, demonstrating an innate immune response. However, the SOCS3 molecule is also upregulated, indicating that infection with the virus may lead to subsequent suppression of the JAK/STAT pathway and evasion of the innate immune response. This network was generated through the use of IPA (Ingenuity Systems, www.ingenuity.com).

https://doi.org/10.1371/journal.pone.0024371.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Levels of Expression of Transcripts in Immunological Pathways of Note.

https://doi.org/10.1371/journal.pone.0024371.t006

Genomic functions analysis links the top transcripts in each pathway to reported disease states and normal function. The functions were distributed amongst many analyses and, in particular, neurological, immunological, and cell death pathways (Table S7) were represented. In horses exposed to WNV compared to normal horses, four categories were identified involving neurological functions (2,326 transcripts), 10 categories were identified involving immunological functions (1,830 transcripts for exposure), and 1 category was identified as involving cell death (1,153 transcripts exposure).

The genes in the functions from neurological categories were grouped mainly under neurological disease when compared to nervous system development and function, behavior, and psychological disease. When further analyzed by specific disease, genes mapped to mental disorders (including bipolar affective disorder, Alzheimer's, and schizophrenia), as well as degenerative neuropathies (including progressive motor neuropathy, Huntington's disease, Parkinson's disease, amytrophic lateral sclerosis, and multiple sclerosis) (Figure 5 a,b).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Neurological Functions for All Analyses.

(A) Neurological disease and psychological disorders categories, and (B) Nervous system development and function and behavior categories. The majority of transcripts mapped to neurological disease. The red bars represent the exposure analysis, the purple bars the survival analysis, and the orange bars the location analysis. The diamonds represent the p-value for exposure, the triangles the p-value for nonsurvival, and the squares the p-value for location. For the purposes of this study, ‘exposure’ represents the difference in gene expression between the nonvaccinated/exposed-normal, ‘survival’ represents the difference in gene expression between the nonvaccinated/exposed-vaccinated/exposed, and ‘location’ represents the difference in gene expression between the thalamus and cerebrum of the nonvaccinated/exposed.

https://doi.org/10.1371/journal.pone.0024371.g005

The functions involving immunological pathways were grouped with cell death/apoptosis for analysis. The most genes were categorized under inflammation (992 transcripts). Both innate (inflammatory response, antigen presentation, immune cell trafficking) and adaptive (humoral immune response, cell-mediated immune response, cytotoxicity, immune cell trafficking) aspects of immunity were identified. Cell death and apoptosis categories were also seen for the exposure analysis, with 1,299 total genes involved with cell death, and 1,006 total genes involved with apoptosis (Figure S7).

Transcripts that were increased in expression by 1-fold or more or decreased in expression by -1-fold or less and mapped to the IPA database were included. There were 176 neurological transcripts downregulated and 43 neurological transcripts upregulated (Table 7) for the exposure analysis. Transcripts involved with neurotransmitter pathways including glutamate receptor signaling (Figure 2a, Table 5) and dopamine receptor signaling (Figure 2b, Table 5) were of particular note. Catenin (cadherin-associated protein), delta 2 (neural plakophilin-related arm-repeat protein) (CTNND2) was also highly upregulated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Transcripts for all analyses mapped to neurological CPs.

https://doi.org/10.1371/journal.pone.0024371.t007

There were 176 immune transcripts downregulated and 130 immune transcripts upregulated (Table S8). The most notable was pentraxin 3 (PTX3), upregulated over 9-fold. Transcripts that mapped to specific immunological pathways of interest that were significantly changed in expression included those that mapped to the IL-15 pathway and those that mapped to the IL-9, IL-22, and JAK/STAT pathways (Figure 4a,b, Table 6). Transcripts involved in apoptosis were also upregulated in WNV infected horses compared to normal horses, including poly (ADP-ribose) polymerase family member 14 (PARP), caspase 4 (CASP4) retinoid receptor (RXR), and retinoic acid nuclear receptor (RAR).

Immune/Survivor Status.

Gene ontology was investigated for the immune/survivor status. Horses that survived infection (all vaccinated before experimental infection) were compared to horses that succumbed to grave neurological diseases (no vaccine before experimental infection). Because of experimental design to allow for assessment of survivorship, nonsurviving horses were analyzed at an average of 7–9 days post-infection compared to survival horses at 21 days post-infection. Similar to previous analysis performed between exposed and nonexposed horses, genes that were found to be significantly different in expression belonged to three major gene ontology (GO) categories (Figure S6a). The most genes mapped to GO processes of transcription/RNA processing (1,864) with the second most genes mapping to immunological categories (850). Genes also mapped to neurological categories (840) and cell death/apoptosis (338) (Figure S6b).

For this analysis, the majority of canonical pathways engaged cell signaling for a variety of cell types, functions and transcripts (Table S6). Ten of the top 25 pathways (based on the p-value) were classified as neurological pathways composed of 156 transcripts. None of the top 25 pathways were identified as immunological pathways.

The neurological canonical pathways were analyzed for nonsurvivorship and 19 pathways were identified (Figure 1). Like the analysis for exposure status, specific neurological pathways that demonstrated dysregulation included neurotransmitter pathways and signaling pathways with glutamate receptor signaling (Figure 2a, Table 5) and dopamine receptor signaling (Figure 2b, Table 5) highly represented once again. The pathways with the most molecules ( greater than 15) involved included axonal guidance signaling, CREB signaling in neurons, synaptic long term depression, neuropathic pain signaling in dorsal horn neurons, synaptic long term potentiation, and glutamate receptor signaling.

Forty-nine pathways involved in both the innate and adaptive immunity were identified as associated with immune/survivorship status (Figure 3). Some of the major immune pathways that were upregulated included the IL-15, IL-22, IL-9, and IFN signaling pathways (Figure 4a,b, Table 6). Multiple pathways involved in apoptosis were also dysregulated and these included the retinoic acid mediated apoptosis signaling, calcium-induced T lymphocyte induced apoptosis, and death receptor signaling.

Five neurological function categories involving 2,246 transcripts were identified as significantly altered when surviving horses were compared with nonsurviving horses, while nine categories were identified involving immunological functions (1,542 transcripts), and one category was identified involving cell death (1,082 transcripts exposure) (Table S7).

In the analysis of specific neurological categories, more genes grouped under neurological disease when compared to nervous system development and function, behavior, and psychological disease. When further analyzed by specific disease, genes mapped to the similar mental and degenerative disorders identified by the previous analysis (Figure 5 a,b).

Functions similar to those identified in the previous exposure analysis were noted when deeper analyses of immunological and apoptosis functions was performed. Genes were mostly commonly categorized under inflammation (832 transcripts), with both innate and adaptive immune functions identified. In total, 476 total genes were involved with cell death, and 340 genes were involved with apoptosis (Figure S7).

Individual significantly upregulated and downregulated transcripts were analyzed for their association with nonsurvival (naïve horses infected with WNV). A total of 209 neurological transcripts were decreased in expression and 42 neurological transcripts were increased in expression (Table 7). Similar to previous analyses, these transcripts were primarily involved with neurotransmitter pathways including glutamate and dopamine receptor signaling (Figure 2ab, Table 5).

Further analysis of the transcripts involved in the immune response in the nonsurvivors identified 215 downregulated and 116 upregulated immune transcripts (Table S8). The transcript increased most in expression was PTX3 exhibiting a 7.7-fold increase in nonsurviving horses compared to surviving horses. Transcripts that mapped to separate immunological pathways that were significantly changed in expression included those that mapped to the IL-15, IL-9, IL-22, and JAK/STAT pathways (Figure 4a,b, Table 6). Transcripts predominantly involved in apoptosis were also upregulated for the nonsurvivors and examples included (PARP) and caspase 4.

CNS Location.

Gene ontology was analyzed for CNS location. The third subhypothesis asked whether there was a difference in gene expression due to location in the brain during WNV infection between thalamus and cerebrum, and for this only infected, nonvaccinated horses were analyzed. The transcripts were mapped to the three major GO categories (Figure S6a). Most genes mapped to GO processes of transcription/RNA processing (1,664) with the second most genes mapping to immunological categories (798). Genes also mapped to neurological categories (447) and cell death/apoptosis (349) (Figure S6b).

Canonical pathways analysis was performed on the data in the location status. Similar to previous analyses, the majority of pathways affected and differentially expressed were involved with cell signaling (Table S6). Seven of the top 25 pathways were identified as important in neurological processes (125 transcripts) while two of the top 25 pathways were identified as immunological processes (40 transcripts).

Sixteen canonical pathways were significantly affected based on differences in CNS location (Figure 1). Like the exposure and survivor analysis, neurotransmitter pathways and signaling pathways were heavily involved, including glutamate receptor signaling (Figure 2a, Table 5) and dopamine receptor signaling (Figure 2b, Table 5). Pathways with greater than 15 molecules involved included axonal guidance signaling, CREB signaling in neurons, synaptic long term depression, neuropathic pain signaling in dorsal horn neurons, Huntington's disease signaling, synaptic long term potentiation, and neureglin signaling (Figure 1).

Forty-eight immunological canonical pathways (Figure 3) involving both the innate and adaptive response were detected as significantly involved. The immune pathways that were upregulated (Figure 4a,b, Table 6) included the same previously identified signaling pathways (IL-15, IL-22, IL-9 and IFN). Multiple pathways involved in apoptosis were also dysregulated in the location analysis. These included the retinoic acid mediated apoptosis signaling, calcium-induced T lymphocyte induced apoptosis, cytotoxic t lymphocyte mediated apoptosis of target cells, induction of apoptosis by HIV1, and apoptosis signaling, and myc mediated apoptosis signaling.

Five categories were identified involving neurological functions (3,242 transcripts), ten categories were identified involving immunological functions (1,558 transcripts), and one category was identified as involving cell death (719 transcripts) (Table S7). The further analyses of specific neurological function were similar to that mapped for both the exposure and immune status analyses (Figure 5 a,b).

The functions involving immunological pathways were grouped with cell death/apoptosis for analysis. Like the previous analyses, most genes were categorized under inflammation (834 transcripts), with involvement of both innate and adaptive immunity. Cell death and apoptosis categories were also seen for the location analysis, with 210 total genes involved with cell death, and 184 total genes involved with apoptosis (Figure S7).

Neurological transcripts were significantly changed in expression (Table 7). A total of 176 transcripts were downregulated and 43 transcripts were upregulated. Transcripts of note were similar to the previous analyses. This included a decrease in transcripts involved with glutamate signaling and dopamine signaling pathways (Figure 2a,b, Table 5). The protein CTNND2 was also highly upregulated.

In total, 266 immune transcripts were downregulated, while 210 transcripts were upregulated (Table S8). PTX3 was upregulated over 4-fold. Other immunological transcripts significantly changed in expression were similar to the analyses involving exposure and immune status. These included those that mapped to the IL-15 pathway and those that mapped to the IL9, IL22, and JAK/STAT pathways (Figure 4a,b, Table 6). Apoptotic transcripts were also upregulated in the location analysis and were similar to that of immune/survivorship status, including poly (ADP-ribose) polymerase family member 14 (PARP), and caspase 4 (CASP4).

Comparison of Histopathological Data and Clinical Signs with Gene Expression Data.

All horses were evaluated for clinical signs and neurological exams were performed. Infected, non-vaccinated horses (exposure group) all demonstrated ataxia and muscle fasciculations. The majority of these horses also demonstrated a change in mentation (4/6) and/or paresis (5/6) (Table S9). All of the exposed naïve horses developed clinical signs, half developed a fever, and all were humanely euthanized (Table 8). Infected vaccinated horses and normal control horses did not develop clinical signs, signs of neurological disease, or fever (Table 8). Viremia was detected in non-vaccinated exposed horses from days 1–5 post infection, but not in the vaccinated exposed horses [22].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Summary of outcomes of each treatment group analyzed by CNS microarray.

https://doi.org/10.1371/journal.pone.0024371.t008

For histopathological analysis, all horses in the non-vaccinated exposed group demonstrated moderate or severe encephalitis on histopathology (Table 8). This pathology was graded according to the degree of gliosis and/or perivascular cuffing and was more severe in the thalamus of these horses compared to the cerebrum (Table 9). Mild signs of encephalitis were seen in 1 out of the 6 infected, vaccinated horses (Table 8). These inflammatory changes were more severe in the thalamus than in the cerebrum (Table 9). No abnormal histopathology was seen in the cerebrum or thalamus of the normal control horses. The severity of the clinical signs and histopathology in the non-vaccinated horses exposed to WNV, compared to the exposed vaccinates and normal controls, is consistent with the changes in gene expression seen between the analyses of exposure and survival. This is likewise true for the location analysis when examining the histopathological differences between the thalamus and cerebrum of the exposed, non-vaccinated horses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. Histopathology on individual animals.

https://doi.org/10.1371/journal.pone.0024371.t009

Array Validation

The correct sequences (checked against the sequences from the transcriptome) were identified for the primer pairs β-actin, 2′5′OAS, CC1R, IL-6, DEADBox60, Defensin β 4, and TNF-α. β -actin was included as the ‘house-keeping gene’, and the other genes were significantly up-regulated. Primer efficiencies were established for all primer pairs using standard curves analysis and efficiency calculation, with efficiencies ranging between 85 and 97%. Real time relative quantitative PCR was then performed on the thalamus from 6 of the vaccinated and exposed horses and 6 of the non-vaccinated and nonexposed horses, with β actin as the endogenous control (Table 10). There was a relative increase in expression for all primer pairs which correlated with the microarray data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 10. Validation of the array.

https://doi.org/10.1371/journal.pone.0024371.t010

Microarray probe sequences were analyzed by comparison to the most recent version of the EqCab2 genome using the basic local alignment search tool (BLAST, Fisher Cluster, University of Florida, Gainesville, FL) to determine the accuracy of each sequence to detect single genes as opposed to gene families. In total, 42,843 oligonucleotide probes were analyzed and 40,113 probes matched to one sequence with 100% identity (93.6%). Of these, 3,700 (9.2%) matched more than once to a genomic sequence implying possible binding to a gene family. The majority of these which matched to multiple sequences were identified as belonging to one chromosome. In addition, 2,687 probes matched at <100% identity (average 97.5% identity). Forty-three probes did not match, and were most likely present as controls since the sequences could not be detected in the sequenced library.

Discussion

This project sequenced all levels of equine brain and spinal cord transcriptome to analyze global gene expression in the CNS of natural, outbred equine hosts during grave WNV encephalitis. The data generated from this project provides invaluable insight into WN encephalitis in horses and possibly humans, and is a useful platform for future studies for both pathological and non-pathological applications. In the sequencing and analysis of the transcriptome, 41,040 sequences were identified by BLAST analysis in 5 sequence databases. There was overall consensus amongst the NCBI databases as to the hits on species, with the vast majority of sequences matching to the horse. Further analysis of the sequenced transcriptome revealed that 9,504 of the identified sequences were missed by equine predicted databases, and 1,280 of the identified sequences have yet to be discovered in the equine genome project. This is most likely due to the incomplete annotation of the equine genome and differential expression of the equine brain in disease allowing for detection of rare transcripts. Since other species' genomes have undergone more comprehensive annotation and analysis, transcriptomic sequences that should be recognized by the equine databases may be recognized in other organisms. Another possibility for the discrepancy could be differences between the equine genome and transcriptome (i.e. splice variants). These issues are likely to improve with time and more tissue specific transcriptome analyses. This portion of the project also demonstrated high sequence homology between the equine and human EST database, showing that the horse may be useful in the study of the human organism.

The main hypothesis investigated was that there are families of genes that are changed in a consistent manner in horses undergoing WN encephalitis. In the analysis of the microarray data, three subhypotheses were investigated to explore whether there was a difference in gene expression based on the state of exposure, immunity/survival, and location in the CNS. Because there was high amount of overlap in our findings from these analysis, either these findings support a generalized model of WNV encephalitis based on exposure status, recovery, and CNS pathology, or the state of WN infection without regard to immunity and recovery has been primarily modeled. Alternatively, it is possible that immunity from WN encephalitis after exposure is similar to a completely naïve, nonexposed state. Additionally, it is likely that the time of sample collection (21 days for vaccinated/exposed horses, 7–9 days for nonvaccinated/exposed horses) influenced gene expression in immune horses. However, overall, it appears that horses that are exposed to WNV demonstrate similar changes in gene expression, which are highlighted by the changes in the thalamus. Finally since the results appear to be location dependent with the majority of differentially expressed genes in the gray matter coinciding with disease/survivorship state, study of WNV in the brain is topical and likely best elucidated in a tissue specific manner.

A total of 17 neurological canonical pathways were identified across the three analyses, the majority of which involved cell signaling within the nervous system. Neurotransmitter pathways were one of the top dysregulated pathways for all groups, including glutamate and dopamine pathways. Glutamate is the primary excitatory neurotransmitter in the neurological system. Previous work has demonstrated that an excess of glutamate at the synaptic cleft can lead to apoptosis of neurons through glutamate excitotoxicity as a cause of pathology in many neurological conditions [23], [24], [25], [26], [27]. In this study, the nonvaccinated group of horses exposed to WNV demonstrated gene expression changes consistent with glutamate excitotoxicity when compared to both the vaccinated and normal (non-exposed) control horses. This was also true when comparing the thalamus of the non-vaccinated horses exposed to WNV to the cerebrum of these same horses. Changes consistent with glutamate toxicity included a decrease in the expression levels of NMDA glutamate receptors, metabotropic glutamate receptors, kainate glutamate receptors, ionotropic glutamate receptors, and glutamate clearance receptors. Infection with WNV may lead to a downregulation of glutamate receptors on the post-synaptic neuron as well as glutamate uptake receptors on glial cells, leading to an increase in glutamate levels in the synaptic cleft and pathology associated with glutamate excitotoxicity.

Dopamine was another neurotransmitter pathway that was significantly changed in all three groups. Dopamine is a stimulatory neurotransmitter that functions, among other things, in the control of voluntary movement [28]. In the nonvaccinated group of horses exposed to WNV, a decrease was seen in the expression levels of dopamine receptor D5 as well as the downstream affector transcripts AC, PC, and PP. In addition, tyrosine hydroxylase, which catalyzes the conversion of tyrosine to dopamine, was downregulated. The expression of monoamine oxidase (MAO), which functions to breakdown dopamine, was increased in the nonvaccinated exposed group. Thus exposure to WNV may lead to a decrease in dopaminergic receptors and subsequent downstream signaling, a decrease in enzymes to create dopamine, as well as an increase in MAO. This results in a total decrease in available dopamine, which may explain many of the clinical signs seen in WNV infection that seem to mimic human disorders such as Parkinson's disease. Further study involving the detection and quantification of the transcripts from neuronal cells infected with WNV associated with the glutamic and dopaminergic pathways is necessary before any firm conclusions can be drawn.

Clinical neurological disease in horses caused by WNV is characterized by a combination of spinal cord, midbrain/hindbrain, and mentation abnormalities, with long-term residual neurological deficits [29]. The clinical signs seen in horses during WN infection mimic many of the clinical signs seen in some human neurological disorders, such as Parkinson's disease, progressive motor neuropathy, Huntington's disease, neurodegeneration, amyotrophic lateral sclerosis, and multiple sclerosis. For this study, it was found that many of the pathways and transcripts previously shown to be dysregulated during these diseases are also abnormally expressed during WNV infection in horses. Thus neurological infection with WNV in horses appears to mimic many of the seemingly non-infectious neurological disorders seen in man on both the clinical disease scale and the transcriptomic level. It is possible that seemingly non-infectious neurological disease may have an infectious origin, or that the brain can only behave and react in a certain manner no matter the stimulus or insult. This study demonstrated that infection with WNV leads to dysregulation in known neurological disease gene pathways, including those involved with neurotransmission and downstream signaling. This corresponds with clinical signs of disease in affected hosts, and also suggests a correlate between the neuropathology induced by viral infection of the CNS and the neuropathology seen in non-infectious neurological disease.

The similarities between the three analyses can also be seen when examining the immunological pathways and functions. This study demonstrated that changes in both the innate (inflammatory response, antigen presentation, immune cell trafficking) and adaptive (humoral immune response, cell-mediated immune response, cytotoxicity, immune cell trafficking) immune pathways are present during WNV infection. In general, the majority of immune transcripts and pathways were decreased in expression in the nonvaccinated horses exposed to WNV, providing evidence of downregulation of a balanced immune response during WNV infection at the peak of clinical disease.

In contrast, some immune pathways, such as the interleukin-15 signaling pathway, were upregulated during WNV infection in nonvaccinated horses exposed to WNV. IL-15 has been shown to be particularly important in providing a protective immune response to viral infection [30], [31], [32], [33]. For all three analyses, IL-15 was upregulated over 2-fold, as well as the transcription factor STAT1, which was upregulated over 2–3 fold. Interestingly, the downstream elements of IL-15 were downregulated in the unvaccinated horses exposed to WNV. The virus, either directly or indirectly, may be blocking the downstream effector elements of the IL-15 pathway to prevent the host immune response to the virus. There could also be other elements in the IL-15 pathway that are not yet elucidated. It is also possible that this finding is only a reflection of the timing when the naïve horses exposed to WNV were euthanized (at the onset of clinical signs) and a beneficial response from IL-15 to viral infection could not be realized in these horses. Thus it appears that IL-15 is upregulated in response to WNV infection, and while it may play a key role in recovery from viral infection, its dysregulation may be a key component of the immunopathology of this disease. Continued work targeting the quantification of IL-15 levels during viral infection at different time points is necessary for further clarification of this data.

Other pathways that were upregulated in non-vaccinated horses exposed to WNV were the IL-22, the IL-9, and the interferon signaling pathways with IL-22 and IL-9 activating similar transcripts. Both of these pathways activate JAK and TYR transcripts, which in turn phosphorylate and activate STAT (Signal Transducers and Activators of Transcription)- specifically STAT1, STAT3, and STAT5. These STAT transcripts induce the expression of ISGs (interferon stimulated genes) through a variety of mechanisms, and lead to the induction of an innate antiviral response [34]. As expected, expression of these JAK/STAT transcripts is upregulated during WNV infection in the unvaccinated horses exposed to WNV. Of interest as well is the finding that the SOCS3 (suppressor of cytokine signaling 3) is also upregulated in the exposure and survival analyses. SOCS3 functions as a negative feedback inhibitor on the JAK/STAT pathway, thereby inhibiting the innate immune response [35]. Transcripts of SOCS1 and SOCS3 have been shown to increase during WNV infection of the murine brain [36]. Upregulation of SOCS3 allows the virus to escape the innate immune response and has also been shown to lead to chronic infection and inflammation. This is further supported by decreased expression of the transcriptional regulators ASB1 and ASB5, which function to suppress SOCS expression. Thus it is possible that while the JAK/STAT pathway is upregulated in response to WNV infection for the activation of innate immunity, WNV may induce the expression of the SOCS3 molecule to suppress this pathway and evade the innate immune response.

The transcript increased the most in expression for all analyses was pentraxin 3 (PTX3). This was the case for all analyses when comparing the unvaccinated exposed horses to the vaccinated exposed and normal horses, as well as the thalamus to the cerebrum (thus the highest levels of expression were in the unvaccinated controls). This molecule has many functions, including an integral role in the pathway of pattern recognition receptors in recognition of viruses and bacteria [37], [38]. This gene is induced by IL-1b, and functions in the phagocytosis and opsonization of antigens, as well as in the inflammatory response. Thus infection with WNV and recovery from disease may be associated with an increase in this molecule that plays an integral role in innate immunity. Another transcript that was highly increased in expression for all analyses (greatest level of expression in the non-vaccinated controls compared to other groups and in the thalamus compared to the cerebrum) was the brain specific molecule CTNND2, which functions to connect cell junctions and cytoskeletal architecture with signaling pathways [39]. This may provide evidence that dysregulation of neurological tissue, such as that induced during WNV infection, leads to re-arrangement of neuronal architecture and the induction of signaling. This may also be important in viral entry into the cell. Apoptotic transcripts that were upregulated in all analyses included PARP and CASP4. Some apoptotic transcripts were upregulated in only the exposure analysis, including RXR and its receptor RAR. Understanding which transcripts are upregulated or downregulated during viral infection provides a glimpse into the affect of the virus on individual transcripts and, with further studies, could lead to the elucidation of many unanswered questions. These could include an understanding of how the virus invades the cell, as well as which cell molecules the viruses uses for replication, transcription, and translation.

The changes in gene expression data were also compared to clinical signs and histopathological data. Horses that were infected and not vaccinated demonstrated abnormal neurological and clinical signs as well as viremia, warranting humane euthanasia by day 9 of the study. On histopathological analysis, these horses also demonstrated moderate to severe signs of encephalitis, worse in the thalamus than in the cerebrum. This was in comparison to the vaccinated, exposed horses and normal horses that did not display any abnormal clinical or neurological signs and did not have a detectible viremia. Only a mild encephalitis was seen in 1 of the 6 vaccinated and exposed horses, with no abnormal histology in the normal controls. This correlates well to the gene expression data, as the non-vaccinated exposed horses demonstrated the greatest changes in gene expression when compared to the other groups of horses. The severity of the clinical signs and histopathology in the non-vaccinated horses exposed to WNV, compared to the exposed vaccinates and normal controls, is consistent with the changes in gene expression seen between the analyses of exposure and survival. This is likewise true for the location analysis when examining the histopathological differences between the thalamus and cerebrum of the exposed, non-vaccinated horses.

In summary, the microarray proved to be a useful tool to understand changes in gene expression patterns during WNV infection utilizing a custom fabricated microarray enriched for neurological and immunological sequences for study in the clinically affected host. Significant changes were identified in neurological, immunological, and apoptotic pathways with associations made between viral encephalitis and non-infectious neurological disease based on a systems biology approach. This information will eventually be integrated with other components of a systems biology approach, combining interdisciplinary scientific fields to validate these findings. This information could eventually be used to combat not only outbreaks of WNV, but also as a model to understand and reduce the impact of viral encephalitis in general.

Supporting Information

Acknowledgments

We would like to thank the University of Florida Interdisciplinary Centers for Biotechnology Research (UF-ICBR) for their invaluable sequencing and bioinformatics expertise. We would like to specifically thank Li Liu, Gigi Ostrow, and Regina Shaw at the UF-ICBR.