Evaluation of the host transcriptional response to human cytomegalovirus infection

Jean F. Challacombe1, Andreas Rechtsteiner2, Raphael Gottardo1, Luis M. Rocha2, Edward P. Browne3, Thomas Shenk3, Michael R. Altherr1 and Thomas S. Brettin1

1 Bioscience Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545
2 Computing and Computational Sciences Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545
3 Department of Molecular Biology, Princeton University, Princeton, New Jersey 08544-1014


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
Gene expression data from human cytomegalovirus (HCMV)-infected cells were analyzed using DNA-Chip Analyzer (dChip) followed by singular value decomposition (SVD) and compared with a previous analysis of the same data that employed GeneChip software and a fold change filtering approach. dChip and SVD analysis revealed two clusters of coexpressed human genes responding differently to HCMV infection: one containing some genes identified previously, and another that was largely unique to this analysis. Annotating these genes, we identified several functional categories important to host cell responses to HCMV infection. These categories included genes involved in transcriptional regulation, oncogenesis, and cell cycle regulation, which were more prevalent in cluster 1, and genes involved in immune system regulation, signal transduction, and cell adhesion, which were more prevalent in cluster 2. Within these categories, we found genes involved in the host response to HCMV infection (mainly in cluster 1), as well as genes targeted by HCMV’s immune evasion strategies (mainly in cluster 2). As the second group of genes identified by the dChip and SVD approach was statistically and biologically significant, our results point out the advantages of using different methods to analyze gene expression data.

gene expression analysis; herpesvirus; pathogenesis


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
GLOBAL GENE EXPRESSION ANALYSIS using DNA microarray technology makes it possible to simultaneously monitor the expression levels of a very large number of mRNAs in cells (16, 26). Microarray analysis is particularly useful in studying host-pathogen interactions, because the levels of mRNAs from infected and uninfected cells can be compared. One pathogen that has been studied by this approach is human cytomegalovirus (HCMV), a member of the herpesvirus subfamily betaherpesvirinae. HCMV causes life-threatening disease in immunologically immature and immunocompromised people, including neonates, AIDS patients, and allogenic transplant recipients (65).

Recent studies of global host gene expression using DNA microarrays show that HCMV infection dramatically changes the gene expression profile of the host cell (7, 77, 79, 100). HCMV infection alters the expression of numerous host cell genes, including genes that regulate cell cycle progression, cellular proliferation, cell adhesion, and genes encoding transcription factors (7, 79, 100). Human cells respond to HCMV infection by altering transcription in an attempt to antagonize viral replication and spread (7, 61).

To go from raw gene expression data to meaningful results generally involves normalization, filtering, and analysis to identify patterns in expression level data (26). In Affymetrix GeneChip experiments, the raw data consist of probe pair intensities. The purpose of normalization is to identify and remove systematic sources of variation in intensity values to allow between array comparisons. Filtering involves reducing the data by removing uninformative genes whose expression levels did not change or were below a user-defined threshold. The gene expression level is typically computed using a statistic that captures the response characteristic of a specific probe set. Many different commercial and free software packages can perform normalization and expression analysis of oligonucleotide arrays. Although there are too many packages to list here, a few examples include the DNA-Chip Analyzer (dChip; Ref. 43), the Affymetrix GeneChip (1, 48, 75), GeneSpring (http://www.silicongenetics.com), Cluster, and TreeView (17). Each of the packages has different capabilities and limitations.

In this study, we compared the group of human genes that responded to HCMV infection in a previous study using GeneChip and a fold change approach (7), to two clusters of coexpressed genes that we identified using dChip and singular value decomposition (SVD) analysis. The first cluster contained some of the genes identified in the previous study (7), whereas nearly all genes in the second cluster were not identified previously.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
We analyzed probe-level time course data (from Affymetrix CEL files; NCBI Gene Expression Omnibus accession GSE675; http://www.ncbi.nlm.nih.gov/geo/) pertaining to HCMV infection of human fibroblasts (7), using dChip (43, 44) to normalize the intensities of the array data and estimate the expression levels. SVD (32, 91) was employed to identify and visualize the two-dimensional subspace that captured most of the variance in the expression data. In this subspace, two clusters of coexpressed genes were identified. We annotated the genes comprising these clusters and grouped them into functional categories.

Normalization of probe-level data.
To normalize microarray data, the GeneChip software calculates composite scores for intensity (or ratio) for each gene. To correct for nonspecific hybridization and background and generate an absolute intensity measure, the GeneChip algorithm subtracts the mismatch (MM) from the perfect match (PM) intensity for each probe pair, then averages these differences to obtain the composite intensity for a gene after truncation of the largest and smallest values of the data (48, 75).

The normalization technique used in dChip is more complex, accounting for nonlinear nonbiological effects (44). This method is based on the assumption that a probe corresponding to a nondifferentially expressed gene should have a similar intensity rank in all of the arrays (44). To normalize a set of arrays, dChip (43, 44) uses an iterative process to identify an invariant set of probes representing nondifferentially expressed genes. During each iteration, the intensities of the baseline array and the array to be normalized are compared (44), the proportion rank difference is calculated for each point, and if it is below a certain threshold, the point is kept in the set for the next iteration. The iterative process continues until the number of points in the invariant set does not decrease. The normalization curve is a linear median line, calculated from the invariant set. The y-coordinates of this line are the normalized values of the array. After normalization, the baseline array and normalized array have similar overall signal intensities. Following normalization, both GeneChip and dChip compute estimates of the gene expression levels from the normalized data.

Expression level estimation.
One key issue in expression level estimation is the way that probe-specific effects are handled. GeneChip uses the average difference as an expression index for the target gene. However, even using MM intensities as controls, the expression levels of the different probe pairs in a probe set are still highly variable (43). dChip accounts for probe-specific effects in the computation of expression levels by using a probe-sensitivity index to capture the response characteristic of a specific probe pair and by calculating model-based expression indices (43).

Expression level data analysis.
SVD is a standard technique for dimensionality reduction and interpretation of data (32, 91). Several studies have applied SVD to gene expression data (3, 30, 90, 91, 97). When applied to a gene expression matrix consisting of the expression levels of m genes measured at n time points (assays), SVD can be viewed as a linear transformation of the expression data from an m x n space to a number of characteristic modes that describe the temporal patterns of gene expression. As SVD is a standard function provided in many statistical packages and linear solvers, we refer interested readers to Refs. 28 and 91 for a more detailed mathematical description.

We performed SVD analysis of the dChip normalized data in the following way. The expression levels calculated by dChip were saved as a tab-delimited file, which was imported into R (http://www.r-project.org). Prior to performing SVD, each row (transcriptional response vector) of the input matrix was centered by subtracting its mean. We performed the SVD using the R function "svd," which provides an interface to the LINPACK routine DSVDC. Following SVD analysis, we calculated the correlations of the transcriptional response vectors with the modes, then visualized the correlations in a scatter plot (91) (Fig. 3) to show the projection of the transcriptional responses onto modes 1 and 2. This plot indicated a distinct cluster of genes correlated with each mode. We identified the set of genes in each cluster by visual inspection of the correlation plot and by manually drawing a boundary around the cluster.



View larger version (43K):
[in this window]
[in a new window]
 
Fig. 3. Visualization of clusters in a correlation plot of the gene transcriptional responses with modes 1 and 2 (A). A similar visualization of data originally analyzed by Browne et al. (7) is shown in (B). Comparison of the clusters identified by plotting the correlation of the first 2 modes (A) shows that our dChip and SVD analysis identified a second cluster (top left) missed by the GeneChip analysis. Most of the genes identified by Browne et al. (B) are either highly correlated (right side of plot) or uncorrelated (left side of plot) with the first mode that we identified (right side of A).

 
The genes in our newly identified clusters were ranked by the magnitude of their expression variance and exported as a list to a file, where each gene was identified by its Affymetrix probe set ID. Since all genes in each cluster had similarly shaped transcriptional responses, the variance is a measure of the amplitude (or magnitude) of the transcriptional responses characterizing each cluster.

Annotation protocol.
We functionally annotated the list of genes extracted from each cluster by creating a tab-delimited file containing the GenBank accession numbers of the genes. We uploaded this file to Stanford University’s "sourceBatchSearch" (http://genome-www5.stanford.edu/cgi-bin/SMD/source//sourceBatchSearch) to obtain annotation information for each GenBank accession number.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
Normalization.
Figure 1 shows the plots of the residuals vs. the fitted intensity values for the statistical models used by the Affymetrix GeneChip and dChip applied to the data for the probe set "AFFX-Bio-C-3_at." The GeneChip model is additive and models the expression level of a given gene by the sum of the probe effect and gene effect plus a stochastic component, which gives the measurement error. The dChip model is multiplicative, modeling the expression level of a given gene as the product of the probe effect and gene effect plus a stochastic component (measurement error). In fitting each model, we have assumed that the data at least approximately satisfy the model. The fitted values give estimates of the expression levels (probe effect plus gene effect for the additive model and probe effect times gene effect for the multiplicative model). We calculated the residuals as the observed values minus the fitted values; the residuals represent elements of variation that remain unexplained by the fitted model.



View larger version (27K):
[in this window]
[in a new window]
 
Fig. 1. Statistical models used by GeneChip (A) and DNA-Chip Analyzer (dChip; B), applied to probe set AFFX-Bio-C-3_at. The additive model used by GeneChip shows a systematic pattern indicating lack of fit, whereas the multiplicative model used by dChip does not show any pattern.

 
Figure 1 shows the relative variation (spread) in the residuals (experimental errors) and fitted values for each model. If a model fits relatively well, then we expect the spread in the residuals to be roughly constant as a function of the fitted values and symmetric about the x-axis. The GeneChip model (Fig. 1A) shows a strong, nonlinear dependence of the residuals on the intensities, indicating that this model is a bad fit to the data. The residuals in the dChip model (Fig. 1B) were smaller (note the different scale on the plots), with a more constant and symmetric spread as a function of the fitted values, and far less dependence on the intensity values; this indicated a better fit of this model with the data.

Estimation of expression.
For most gene expression data, SVD analysis shows a decreasing singular value spectrum, with a leveling off after the first 3–5 modes. As the ordering of the modes is determined by high-to-low sorting of the corresponding singular values (91), the first few modes account for most of the patterns in the data, while the rest mainly represent noise. This can aid in identifying the most prevalent signals in the data. The reduction of dimensionality provided by SVD analysis facilitates data visualization, and clusters of genes with similar transcriptional responses can be easily identified.

Figure 2 shows the singular value spectrum and first two modes of the dChip modeled data. The modes represent unique orthogonal (or uncorrelated) gene expression patterns (64); the singular value squared indicates the variance captured by a mode. The expression data consisted of 12 time points (12 arrays), representing 0.5, 1, 4, 6, 10, 12, 14, 16, 18, 20, 24, and 48 h after HCMV infection.



View larger version (11K):
[in this window]
[in a new window]
 
Fig. 2. The singular value spectrum obtained by singular value decomposition (SVD) analysis of dChip modeled data. The plot of relative singular values shows the expression level corresponding to each mode (A). Graphs of expression level vs. time for modes 1 (B) and 2 (C) show the different expression profiles corresponding to each mode. It is clear from A that the first 2 modes contain most of the expression data. hpi, hours postinfection.

 
The first two modes captured most of the signal. Figure 2B shows a significant change in the pattern of the first mode over time. This mode contained most of the variance in the data. At 1 h after HCMV infection, the expression pattern of the first mode increased sharply, up to the level at 24 h, and then decreased slightly over the next 24 h. (Note that the decrease at 48 h was based on 1 data point; we did not have data points between 24 and 48 h, so an artifact in the 48 h array could have affected the results.) All genes that were highly correlated with the first mode showed a similar transcriptional response to the pattern of mode 1.

Orthogonal to the first mode, the second mode (Fig. 2C) captured most of the remaining variance in the data. The pattern comprising the second mode decreased until about 12 h after infection, then generally increased and was somewhat higher at 48 h than for early time points. Genes highly correlated with this mode showed similar transcriptional responses (Fig. 2C). Note that whereas cluster 1 genes seemed to be upregulated initially after HCMV infection, cluster 2 genes were downregulated. This suggests that genes in cluster 1 were activated by the host’s immune response, whereas genes in cluster 2 were downregulated by the viral proteins in an attempt to evade the host’s immune response.

The third mode captured less than 10% of the variance in the data (Fig. 2A). There were very few genes highly correlated with the third mode; therefore, it probably contained mostly noise.

Data visualization.
Figure 3A shows a correlation plot of the gene transcriptional responses with modes 1 and 2. The closer the genes mapped to the periphery of the circle with radius 1, the more their transcriptional responses correlated with the first two modes. We identified two regions where the genes clustered more densely; both regions were close to the perimeter of the plot. One cluster was highly correlated with the first mode (right side of the plot in Fig. 3A); genes in that cluster showed similar transcriptional responses to the pattern of this mode (see Fig. 2). The second, smaller cluster (top left of Fig. 3A) was highly correlated with the second mode and much less (anti-)correlated with the first mode.

Figure 3B shows projections onto the first two modes of the HCMV-responsive mRNAs identified previously (7). Analysis by the Affymetrix GeneChip software showed that these genes changed at two sequential time points after infection by a factor of 3 or more (7). Many of the genes in this group were highly correlated (right side of plot in Fig. 3B) or anticorrelated (left side of plot in Fig. 3B) with the first mode. Note that the center of the plot in Fig. 3B is sparsely populated, indicating that the transcriptional responses of most of the genes identified in the previous study were also highly correlated (or anticorrelated) with the two modes.

The first mode represented the most dominant pattern in the data, with the most change in expression. Therefore it is not surprising that 377 (or 26%) of these genes were also selected by the fold change approach (7). Figure 4A shows the green region of the plot, drawn manually to select genes in the first cluster. Figure 4B shows the transcriptional response pattern common to the 1,747 genes in this cluster. This pattern was similar to the pattern of the first mode (Fig. 2B).



View larger version (37K):
[in this window]
[in a new window]
 
Fig. 4. Identification of 1,747 genes in cluster 1 (A) and their transcriptional response pattern (B). The 1,747 genes that were correlated with mode 1, which we termed cluster 1, were selected by drawing a box around the region of increased density (green area in A). The transcriptional response of the genes in this cluster shows a steady increase in expression starting at 6 hpi until 24 hpi. From 24 to 48 hpi, the expression level decreases but remains above 0.

 
Genes in the second cluster, highly correlated with the second mode, were not well represented by the group of previously identified genes (7). Only 15 genes (or 1%) in the second cluster were selected by the original analysis. Figure 5A shows the red box identifying the second cluster of 462 genes, while Fig. 5B shows the transcriptional response pattern common to the 462 genes in the cluster. Note how the transcriptional response pattern was similar to the pattern of the second mode (Fig. 2C).



View larger version (38K):
[in this window]
[in a new window]
 
Fig. 5. Identification of the 462 genes in cluster 2 (A) and their transcriptional response pattern (B). The 462 genes comprising this cluster were identified by drawing a box around the region of increased density (red area in A). We also drew 3 control boxes of the same size in different regions of the plot (not shown) and found that the number of genes in the cluster was significantly higher than in the control regions. The expression profile of the genes in cluster 2 shows a decrease until about 12–16 h postinfection, followed by an increase beginning at about 18 h postinfection. By 48 h postinfection, the expression levels of genes in this cluster are higher than at earlier time points.

 
To assess the statistical significance of the identified clusters, we randomly rotated the location of the cluster boundaries in the two-dimensional space. If the rotated cluster boundary did not overlap with the two identified clusters, then we counted the number genes inside the rotated cluster boundary. The mean and standard deviation of the number of genes inside the boundary was calculated from 100 samples. For the cluster 1 boundary, the mean was 337 genes with standard deviation of 98 genes; the number of genes that we identified in cluster 1 (1,747) was more than 5 times higher than the mean of the randomly rotated samples. Using the boundary of cluster 2, we found a mean of 76 genes with a standard deviation of 21; the number of genes identified in cluster 2 (462) was 6 times higher.

Biological functions of genes in clusters 1 and 2.
We manually analyzed the tab-delimited SourceSearch files, which contained the annotated genes comprising each cluster, looking for genes that participate in biological processes relevant to the host cell response to HCMV infection. These processes included signal transduction, immune system regulation, apoptosis, cell cycle regulation, oncogenesis, cell adhesion, and transcription. These categories were obtained from the Gene Ontology Consortium biological process ontology (12) (see http://www.geneontology.org).

The annotations of the 1,747 genes in the first cluster showed 82 genes involved in immune system regulation (Supplemental Table S1), 73 genes involved in apoptosis (Supplemental Table S2), 27 genes involved in cell adhesion (Supplemental Table S3), 277 genes involved in transcription regulation (Supplemental Table S4), 155 genes involved in oncogenesis and cell cycle regulation (Supplemental Table S5), and 128 genes involved in signal transduction (Supplemental Table S6). (Supplemental Tables S1–S6 are available at the Physiological Genomics web site.)1

Of the 462 genes in cluster 2, a search of the annotated gene list by biological process revealed 40 genes involved in immune system regulation (Supplemental Table S1), 17 genes involved in apoptosis (Supplemental Table S2), 20 genes involved in cell adhesion (Supplemental Table S3), 45 genes involved in transcription regulation (Supplemental Table S4), 20 genes involved in oncogenesis and cell cycle regulation (Supplemental Table S5), and 61 genes involved in signal transduction (Supplemental Table S6).

Some differences between the two clusters can be seen by comparing the proportion of genes in each category to the total number of biologically relevant genes in each cluster. Comparing these numbers between cluster 1 and cluster 2 (Table 1) revealed a noticeably greater percentage of genes in cluster 1 in the categories of transcription (37.3%) and oncogenesis/cell cycle regulation (20.9%) than in cluster 2 (22.2% and 9.9%). Cluster 2 contained a higher percentage of genes involved in signal transduction (30.0%), immune system regulation (19.7%), and cell adhesion (9.9%) compared with cluster 1 (17.3%, 11.1%, and 3.6%).


View this table:
[in this window]
[in a new window]
 
Table 1. Proportion of genes in each category

 

    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
Biological relevance of genes.
To investigate the biological significance of the clusters identified by dChip and SVD, and to discover differences in the biological functions of genes comprising these clusters, we annotated the genes in each cluster. We found genes that grouped similarly to those previously investigated (7) into classes representing immune system regulation, cell cycle regulation and oncogenesis, and apoptosis. However, within these broad categories, many of the genes in our newly identified clusters, and in many cases the networks to which they belonged, differed from those in the previous study. In addition, we identified genes in each cluster that fit into the category of cell adhesion, which was not reported previously.

Results (Supplemental Tables S1–S6) indicated that HCMV infection altered the expression of many components of the immune response. Active HCMV infection results in the expansion and activation of immune effector cells in the blood. These cells include cytotoxic cells, including CD8+ T lymphocytes and natural killer (NK) cells, monocytes/macrophages, antigen-presenting cells, and polymorphonuclear cells (neutrophils, basophils, eosinophils). In addition to their ability to kill infected cells, cytotoxic cells produce cytokines that stimulate cell differentiation, proliferation, and migration of other cytotoxic cells (25) and B cells, which produce antibodies to the virus.

HCMV employs a number of strategies to evade the host immune system and establish a life-long latent residence in the host. To accomplish this, the virus interferes with host defenses by escaping recognition by effector cells (6, 29, 49, 96), which include CD4+ and CD8+ T cells and NK cells. Specific evasion tactics include disrupting antigen processing and presentation by host cells, affecting cytokine balance, effector cell maturation, and cellular immune responses (6), and blocking apoptosis of infected host cells (101). HCMV infection also alters cell cycle regulation (27, 53, 54), cell adhesion molecule (76) and transcription factor expression and activities (36), and influences genes that can be involved in oncogenesis (7). Each of the two clusters of host genes that responded to HCMV infection in the present study included genes encoding proteins that participate in each of these processes, which we discuss below.

Antigen processing and presentation.
We expected the host response to HCMV infection to include an increase in the expression of genes encoding proteins involved in antigen processing and presentation. In fact, the genes that we identified in the first cluster did show increased expression over the first 24 h after infection; this cluster contained many genes whose products are integral to antigen processing and presentation, including 18 proteasome subunits, 4 MHC molecules, TAP1, itch homolog, CD83, and RFX5. Of these genes, several are specific to MHC class I antigen processing and presentation, while others are involved in MHC class II functions.

MHC class I proteins on the surface of infected host cells present virus-derived antigens to T cells, which create an antiviral response and enlist other participants in the immune response. The proteasome is a protease responsible for degrading proteins to produce antigenic peptides, which are presented by MHC class I molecules on the cell surface (86, 98). Itch homolog is an E3 ubiquitin protein ligase, which is responsible for promoting ubiquitin ligation of proteins, which marks the proteins for recognition and consequent degradation by the 26S proteasome (68). TAP1 is a component of the transporter associated with antigen processing, which brings the peptide antigens processed by the proteasome from the cytoplasm into the endoplasmic reticulum, where they are complexed with MHC class I molecules. Cluster 1 also included the gene encoding MHC class I molecule HLA-C, which is expressed on all nucleated cells and functions in presentation of peptide antigens that have been processed by the proteasome to effector cells as part of the cellular immune response (39).

Cluster 1 also contained genes encoding proteins involved in MHC class II antigen processing and presentation; this group included the genes for legumain, an asparagine endopeptidase involved in MHC class II antigen processing (95); CD83, expressed on the surface of mature dendritic cells and stimulated T lymphocytes; and MHC class II molecules HLA-DPB1, HLA-DPA1, and HLA-DRA, expressed on immune system cells where they function to present peptide antigens to helper T cells, eliciting their activation and proliferation as part of the cellular immune response. In addition, RFX5 is essential for activation of MHC-II promoters (87). An increase in the expression of these genes, whose protein products are instrumental in processing and presentation of viral antigens by infected cells, is not surprising.

HCMV itself produces gene products that disrupt antigen processing and presentation by class I and class II molecules of the MHC. These effects of HCMV include inhibition of MHC class I molecule expression by infected cells, production of a MHC class I homolog, and impaired MHC class II antigen presentation by macrophages. In cluster 2, we identified genes involved in antigen processing and presentation whose transcriptional responses were characterized by decreased expression over the first 24 h. These genes encoded endothelial protein C receptor and LIR-2. Endothelial protein C is homologous to MHC class I/CD1 family (46) and plays a role in the host anti-inflammatory response to infection (21). LIR-2 is an MHC class I binding protein that inhibits monocyte activation signals (18). Interfering with the functions of endothelial protein C receptor via a decrease in expression may help reduce the host anti-inflammatory response to infection and also help HCMV avoid early immune system recognition by T cells. Decreased expression of LIR-2 may release host immune cells from inhibition so they contribute to the immune response.

Although the genes in cluster 1 showed increased expression over the first 24 h as part of the host response to HCMV infection, our finding that the expression of the protein C receptor gene decreased over the same time period indicates that HCMV infection also impairs the host response, in this case by impairing one aspect of antigen presentation.

Cytokine balance.
We expected expression of host genes for components of the cytokine system to increase as part of the response to HCMV infection. These include genes encoding cytokines and cytokine receptors, genes whose products are part of cytokine-induced signal transduction cascades, and transcription factors that either regulate the expression of cytokine and cytokine receptor genes or perpetuate the downstream changes in gene expression as a result of cytokine signaling. Both of our clusters contained genes whose products are involved in host cytokine balance.

In cluster 1, we found genes for several interferon response genes, transcription factors, and transcriptional regulators that influence cytokine gene expression in T cells, components of TNF signaling pathways, IL-15, and the class I cytokine receptor. Genes in cluster 1 whose products are part of interferon response included RI58 and G1P2, which are induced by IFN-{alpha}; MX1, induced by IFN-ß; PRKRA, a protein kinase involved in mediating the antiviral actions of IFN; and IRF4, which binds to the interferon-stimulated response element (isre) of the MHC class I promoter.

Transcription factor genes whose products influence T cell expression of cytokines included ILF2, ILF3, NFATC3, and NFATC1. We also found several genes encoding regulators of NF-{kappa}B function; NF-{kappa}B stimulates the transcription of genes involved in immune and proinflammatory cytokine responses, as well as cell adhesion, apoptosis, differentiation, and growth (23). These genes encode IKBA, which inhibits NF-{kappa}B by complexing with and trapping it in the cytoplasm; CHUK/IKK1, whose product phosphorylates inhibitors of NF-{kappa}B (like IKBA), thus leading to the dissociation of the inhibitor/NF-{kappa}B complex and activation of NF-{kappa}B; and IKBKAP, which can bind NF-{kappa}B-inducing kinase (NIK) and I{kappa}B kinases (IKKs) through separate domains and assemble them into an active kinase complex.

Cluster 1 genes involved in TNF signaling included those for TNFSF9 cytokine and TRAF6; TRAF proteins are associated with, and mediate signal transduction from, members of the TNF receptor superfamily. TRAF6 also functions as a signal transducer in the NF-{kappa}B pathway that activates IKK in response to proinflammatory cytokines. Last, IL-15 regulates T and NK cell activation and proliferation, whereas class I cytokine receptor (WSX-1) has homology to the IL-12 receptors, is expressed mainly in T cells, and is necessary for initiation of T helper responses (99).

HCMV modulates the host immune response by altering the functions of cytokines and their receptors (38, 49, 83). The ability of HCMV to slow cytokine production and interfere with cytokine functions is crucial to the survival of virus-infected cells. Our results for cluster 2 do indeed show a decrease in the expression levels of genes for cytokines, cytokine receptors, transcription factors, and an important member of the Jak/STAT signaling pathway that mediates cellular responses to cytokine stimulation. The genes in this group encode the chemokine MIP-1ß, IL-12b, IL-8 receptor-{alpha}, IL-1 receptor-like 1, Burkitt lymphoma receptor 1 (BLR), lymphotoxin-{alpha} (TNF-ß), TNF receptor member 6, and STAT5B.

MIP-1ß, IL-12b, IL-8 receptor-{alpha}, IL-1 receptor-like 1, BLR, TNF-ß, and TNF receptor 6 are important stimulators of cellular immune responses (13, 15, 22, 33, 37, 40, 45, 50, 51, 57, 69, 92). STAT5B is a component of the Jak/STAT pathway, which is activated by cytokines, including IFN-{alpha}, -ß, and -{gamma}. Cytokine binding to cell surface receptors that are coupled to this pathway results in rapid (within minutes) activation of signal transducers and activators of transcription (STATs), which are already present in the cytosol; upon activation, STATs are phosphorylated by Janus family tyrosine kinases (Jaks). Phosphorylated STATs dimerize and translocate to the nucleus, where they modulate the expression of target genes (41). Decreased expression of STAT5B, coupled with the HCMV-induced decrease in cytokine and cytokine receptor gene expression that we observed, may result in a reduction in the expression of target genes that are crucial to the immune response.

Effector cell maturation and cellular immune response.
Many cluster 1 genes encode components of signaling pathways important for T cell activation. Two signaling events are required for T cell activation (20). First, antigen binding to the T cell receptor results in the activation of intracellular protein tyrosine kinases, which activate three signaling pathways: the p21ras/MAP kinase, calcium/calcineurin, and protein kinase C pathways. Second, a costimulatory signal is transduced by one of CD28, CD2, LFA-1, CD5, or interleukin receptors. This signal is necessary for T cells to proliferate and produce cytokines IL-2, TNF-{alpha}, IFN-{gamma}, GM-CSF, and IL-3. Of these, IL-2 is critical for T cell-dependent immune response. IL-2 transcription is mediated by the AP-1 transcription factor, which is a c-Fos and c-Jun heterodimer. We found components of each of these signaling pathways in cluster 1, indicating that T cell IL-2 production is a prominent part of the host response to HCMV infection.

Genes in cluster 1 whose products are part of the p21ras/MAP kinase pathway included Raf-1, MAP2K1, ELK1, MAP2K4, MAPK14/p38, and several members of the NFAT transcription factor family. Another member of cluster 1, CD58/LFA3, binds to the T cell CD2 molecule, which transduces the second signal required for IL-2 production. So, in cluster 1, we found genes encoding components of the first (MAP kinase signaling pathway) and second (CD2 stimulation) signals that lead to AP-1-mediated IL-2 transcription in T cells. The IL-2 promoter is also regulated by members of the NFAT transcription factor family; we found genes for several of these, as well as proteins that regulate NFAT, in cluster 1. These genes included NFATC1, NFATC3, ILF2, and ILF3 heterodimers that comprise NFAT, XPO1, which regulates NFAT and AP-1, and ILF1, which binds to NFAT-like motifs in the IL-2 promoter.

The calcium/calcineurin and protein kinase C pathways are also involved in transducing the first T cell stimulation signal, and we found genes encoding components of these pathways in cluster 1. These included PPP3CB, FK506 binding protein 5, and DCSR1 (calcium/calcineurin pathway); protein kinase C-{delta}, inositol hexaphosphate kinase 1 (IHPK1), the regulatory subunit for phosphoinositide 3-kinase (PIK3R3), diacylglycerol kinase-{zeta} (DGKZ), phosphoinositide 3-kinase class 3 (PIK3C3), a PI3-kinase related kinase (SMG1), and phosphatidylinositol 4-kinase type II (protein kinase C pathway).

Increased expression of genes involved in effector cell functions, as our results for cluster 1 show, indicates a generalized host immune response to HCMV infection. In contrast, decreased expression of effector cell-related genes in cluster 2 in response to HCMV infection, as our results also indicate, would severely compromise the host immune response. These findings reflect a complex interplay between the host cellular immune response and viral immune subversion.

Apoptosis.
Apoptosis, or programmed cell death, is a mechanism that eliminates cells that have outlived their useful life span or that have sustained irreparable injuries. Apoptosis of virus-infected cells is a critical process in the immune system, because it serves to limit the spread of intracellular pathogens (reviewed in Ref. 83). Increased expression of genes involved in apoptosis, as we found in cluster 1, indicates that the host response to HCMV infection includes an increase in the killing of virus-infected cells. The genes in this category included those encoding transcription factors, components of tumor necrosis factor (TNF) signaling pathways, proteins involved in other signal transduction pathways, and other proteins with roles in various aspects of apoptosis.

Many genes in cluster 1 encode proteins involved in TNF/Fas signaling pathways and caspase activation. The TNF/Fas system regulates immune defense and response to infection (92) through cellular phosphorylation regulation by kinases and phosphatases. T cells, B cells, and NK cells all undergo apoptosis in response to Fas receptor ligation (62, 94). All three known MAPK cascades (SAPK1/JNK, SAPK2/p38, and ERK/MAPK) are activated by both TNF and Fas systems. Genes in cluster 1 that are part of these systems include TNFSF9, BIRC2, TRIAD3, cytochrome c, caspase 7, caspase 3, RAD21 homolog, apoptotic activating factor 1 (APAF1), BH3 interacting domain death agonist (BID), PAK2, and modulator of apoptosis 1. Genes encoding other apoptosis-related proteins included PPAR binding protein and the magnesium-dependent phosphatases PPM1A and PPM1D, whose protein products regulate p53 activity.

Cluster 1 also contained genes encoding other signaling molecules that act in apoptosis-mediating pathways; these genes included MX1, Raf-1, growth factor receptor-bound protein 2, BAG1, BCL2-antagonist/killer 1, MAPK14 (p38), MAP3K7, dual specificity phosphatase 6, nonreceptor protein tyrosine kinase PTK2, Axin 1, estrogen receptor binding site associated antigen 9, CED-6, and IL-15. Other proteins that induce apoptosis, whose genes were present in cluster 1, include TIAL1, programmed cell death 2 and 10, BCL2/adenovirus E1B interacting protein 3, adenylate cyclase 2, DED, FEM-1 homolog b, and magnesium-dependent phosphatase PPM1B. Genes encoding proteins that suppress apoptosis included apoptosis inhibitor 5 and macrophage erythroblast attacher.

Not surprisingly, many viruses, including HCMV, try to subvert the apoptotic process to create a favorable environment for viral replication (83). While our results for cluster 1 show increased expression of many genes involved in apoptosis, results for cluster 2 fit with a subversive action of HCMV, in that many of the genes that we identified as part of this cluster, and whose expression decreased after HCMV infection, play a role in apoptosis. These genes included TNF and Fas family members TNF-ß, TNF receptor member 11/RANK, TNF ligand member 8/CD30L; MLK2, a component of TNF-activated SAPK1/JNK signaling cascade; Jak2, a tyrosine kinase that binds to the TNF receptor CD120a and activates STAT5; granzyme B, caspases 2, 8, and 9; and genes encoding NF-{kappa}B regulators.

Cell adhesion.
Cell adhesion molecules are key to several functions of the immune response, including T cell-antigen-presenting cell interactions, T cell-B cell interactions, cytotoxic T cell-NK cell interactions with infected target cells (60), lymphocyte recirculation, leukocyte migration, and interactions between lymphocytes and endothelial cells. All of these are essential components in the generation of effective inflammatory responses and the development of rapid immune responses (80). In contrast, HCMV induces changes in adhesion molecules that may influence inflammation, induce morphological alterations of infected cells, and reduce their ability to adhere to extracellular matrix molecules (93).

Cluster 1 contained genes that fit into the categories of focal adhesion components, integrin-related proteins, and other proteins involved in various aspects of cell adhesion. Focal adhesion components included the protein tyrosine kinase substrate villin 2/ezrin, protein tyrosine kinase 2, nonreceptor protein tyrosine phosphatase 12, vasodilator-stimulated phosphoprotein, and PPFIA1. The integrin-related proteins in cluster 1 included integrin-{alpha}6, EED, and {alpha}-integrin binding protein 63. CD58/LFA3 is a cell surface molecule, expressed by many cell types, involved in T cell activation (88). Increased expression of proteins involved in cell adhesion, as we found in cluster 1, might aid lymphocyte migration to sites of infection during the inflammatory response.

Cluster 2 also contained genes encoding cell adhesion molecules and proteins that interact with cell adhesion molecules. This group included the integrin ß1- and ß4-subunits, ß3-endonexin, and ICAM2. Cluster 2 also included genes for ELAM/selectin E and CECAM 1. Our results show that the expression levels of all of these genes decreased over the first 24 h postinfection. This observation fits with many previous studies showing that HCMV infection affects host cell processes that involve cell adhesion molecules; these cell-based functions include lymphocyte adhesion, migration, proliferation, differentiation, and activation and cell cycle progression. By negatively impacting these cellular functions in the immune system, decreased expression of adhesion molecules in response to HCMV infection would severely compromise the host response to the infection. Several studies support this idea; for example, HCMV infection of human endothelial cells (76) and human fibroblasts (93) altered the expression of the integrin ß1-subunit. Another report showed that integrin activation of cyclin-dependent kinases 2 and 6 was necessary for cells to progress through G1 of the cell cycle and also for the expression of cyclin D1 (24). This suggests that a decrease in integrin expression may negatively impact the cell cycle.

Other cellular behaviors that involve adhesion molecules include cell invasion, metastasis, and tumor growth. Downregulation of integrin-{alpha}3ß1 has been linked to malignancy in carcinoma cells, resulting in increased invasion, metastasis, and tumor cell growth (85). The ß4 integrin subunit is part of integrin-{alpha}6ß4, which shows decreased expression in breast and prostate adenocarcinomas (34, 35, 63); this evidence suggests that tumor growth may be facilitated by the downregulation of integrin-{alpha}6ß4. Since our results show decreased expression of both the ß1 and ß4 integrin subunits over the first 24 h after infection, they indicate that HCMV infection may contribute to the cellular changes involved in cancer.

Oncogenesis and cell cycle regulation.
Both clusters 1 and 2 contained many genes encoding proteins implicated in oncogenesis and cell cycle regulation. Cluster 1 genes included many oncogenes, genes encoding tumor-associated proteins, kinases, cell cycle regulators, transcription factors, and transcriptional regulators, components of the Wnt signaling pathway, and other proteins involved in oncogenesis and/or cell cycle regulation.

Oncogenes in cluster 1 included RAB5A, REL, RELA, RAB1A, KRAS2, MEL, RAB14, RAB21, YES1, LYN, PIM2, RAB27A, BRAF, ERBB3, SKI, RAP1A, Raf-1, and CRKL. Most of these oncogenes encode kinases or related components of signaling pathways. Tumor-associated proteins encoded by genes in cluster 1 included ARMET, MAPRE1, Ewing sarcoma breakpoint region 1, PTTG1, HYOU1, and PPM1D. In addition to the oncogenes in cluster 1, other kinases and phosphatase genes of interest in cluster 1 were PTPN12, PTPN21, PTPN3, PTP4A1, and BCR. Protein-tyrosine phosphatase (PTPs) are signaling molecules that regulate a variety of cellular processes including cell growth, differentiation, mitotic cycle, and oncogenic transformation.

Many genes in cluster 1 encode cell cycle regulators. Cyclins and related proteins included cyclin-dependent kinase 2, CDC25A, CKS2, cyclin C, cyclin D2, cyclin D3, cyclin E1, cyclin E2, CDKN1B, CKS1B, CCRK, cyclin G associated kinase, and PPM1B. Other cell cycle regulators in cluster 1 were serine threonine kinase 6, nucleoporin 214 kDa, mortality factor 4 like 2, BTG3, RAD21 homolog, PPM1G, cell nuclear antigen, TERF1, retinoblastoma binding protein 8, MAPK6, DLEU1, CDC7L1, RAN binding protein 1, SMC1L1, MCM3, AKAP11, NOL1, PTPN3, and TERF2. Transcription factors and transcriptional regulators involved in oncogenesis and cell cycle regulation included TFE3, CBFP, promyelocytic leukemia, SKI interacting protein, Ret finger protein, zinc finger protein 151, MLLT2, ELK1 and ELK4, ATF5, E2F3, E2F4, E2F5, E2F6, RB1, CTCF transcriptional repressor, NOLC1, and BTRC. Components of the Wnt signaling pathway were casein kinase 2{alpha}1, axin 1, and frizzled homolog 9 receptor. The Wnt signaling pathway functions in cell fate determination, cell polarity, cell adhesion, apoptosis, and tumorigenesis.

In terms of cell cycle regulation and oncogenesis, cluster 2 contained genes for several members of the Wnt signaling pathway, oncoproteins, kinases, cyclins, cyclin-dependent kinases, other cell cycle regulators, transcription factors, and other proteins involved in either oncogenesis or the cell cycle. Members of the Wnt pathway in cluster 2 included ICAT, wingless-type MMTV integration site family member 4 (Wnt-4), and secreted frizzled-related protein FRZB. ß-Catenin interacts with the TCF/LEF family of transcription factors and activates transcription of Wnt target genes, which play an important role in development and tumorigenesis. ICAT inhibits the interaction of ß-catenin with TCF and antagonizes Wnt signaling (81). By reducing ICAT expression, HCMV infection may diminish the negative regulation of ß-catenin, allowing it to form a transcriptional activation complex with LEF; this complex may promote tumor formation (4). Furthermore, as mentioned above, the decrease in expression of the transcriptional regulator ICAT that we found may result in inappropriate activation of the Wnt pathway, which plays an important role in human cancers (66, 78). Transcriptional targets of the Wnt pathway include the cellular oncogene c-myc and cyclin D1.

The genes for cyclins and cyclin-dependent kinases in cluster 2 included cyclin A2 and cyclin G-associated kinase. Related is S-phase kinase-associated protein 2 (p45), which associates with cyclin A and CDK2 in transformed cells (100). A decrease in the expression of these cell cycle regulators would most likely alter cell cycle progression. In fact, HCMV infection results in stimulation of cellular DNA synthesis (2), implying overriding of normal cell cycle control. HCMV-mediated perturbations result in inhibition of cell cycle progression at multiple points (53), including the G1 to S transition (27, 53, 54) and G2/M (31).

Summary: A new method of analysis lead to different insights.
The previous study by Browne et al. (7) using Affymetrix software to analyze GeneChip data found that the levels of 1,425 cellular mRNAs changed by threefold or greater in at least two consecutive time points during HCMV infection. The classes of genes affected included genes involved in immune system regulation, particularly interferon-responsive genes, genes involved in cell cycle regulation and oncogenesis, and genes whose protein products promote or inhibit apoptosis.

Our dChip and SVD analysis of the same expression data resulted in two separate clusters of coexpressed genes responding differently to HCMV infection. The original analysis (7) used GeneChip to preprocess and normalize the data and obtain expression values for the probe sets. In the analysis presented here, we used dChip to preprocess and normalize the data and obtain expression values for the probe sets. We found that dChip’s multiplicative model for calculation of expression values led to lower residuals and less dependence of the residuals on the magnitude of the expression values. We then used SVD to analyze the data obtained with dChip. The SVD analysis produced two significant modes, which captured over 75% of the variance in the data. The correlation plot in Fig. 3A shows two statistically significant higher density regions (clusters) of coexpressed genes that were highly correlated with mode 1 and mode 2, respectively. Twenty-six percent of the genes selected by Browne et al. (7) fold change filtering were present in the first cluster, but only 1% were present in cluster 2. That cluster 2 was most likely due to a real biological response is supported by the fact that over 22% of the variance in the data was captured by mode 2. In addition, the density of genes in cluster 2 was significantly higher than in other regions of the space spanned by mode 1 and mode 2. The transcriptional response pattern of cluster 2 was very different from that of cluster 1. Cluster 2 genes showed a transient expression, first decreasing and then increasing again. This suggests that cluster 2 genes might be affected by the immune evasion strategies of the virus. This suggestion is supported by an observation by Browne et al. (7) that found many fewer downregulated genes in cells infected with an inactivated virus.

Our results indicate that the choice of analysis methodology for gene expression data is important. Although one method may work well for detecting one type of pattern in the data, it may miss another pattern altogether.

Discussion of why cluster 2 genes were missed by fold-change filter.
As the previous study included only those genes that had a fold change of at least ±3 at two consecutive time points, but did not have a difference call of "NC" (no change), we investigated each of these criteria as the reason why our second cluster was missed by the previous study.

Our initial hypothesis was that cluster 2 genes exhibited a more transient expression profile and were therefore less likely to exhibit consecutive threefold or greater changes in expression. A more detailed study of the original GeneChip data (available at http://www.molbio.princeton.edu/labs/shenk/browneetal2001/HCMVTimeCourse.txt) revealed that this is not the case. Rather, the cause is the NC difference calls made by the GeneChip software; these expression values were not included in the original study by Browne et al. (7). When we ignored the difference calls and considered the GeneChip fold change values for all genes, we found that 184 of the 462 genes in cluster 2 (~40%) exhibited consecutive changes of at least threefold. This is an even higher percentage than we found in cluster 1: 467 genes of the total 1,747 (~27%). Clearly it is the NC calls that eliminated many large fold change values from the original study that we found in cluster 2.

As a portion of the original data set has been validated both by Northern blot analysis and by comparison with other published results (7, 100), we are confident in the reliability of the data. Therefore, the fundamental difference between the two methodologies is that in the original study (7) a difference call was made for each gene one time point at a time, and values that were called NC by GeneChip were not used further. In contrast, the SVD analysis considered the complete dChip-normalized two-dimensional data matrix (gene expression value vs. time for all arrays), focusing on extracting significant patterns from the data. With this approach we identified an additional statistically and biologically significant cluster of coexpressed genes.

Our results underscore the need for independent validation of the results of gene expression experiments and indicate that there are significant advantages to using more than one method to analyze gene expression data to understand complex biological phenomena, such as the immune system response to herpesvirus infection.


    GRANTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 
This work was supported in part by National Cancer Institute Grant CA-87661 and by Los Alamos National Laboratory LDRD-DR Grant W-7405-ENG-36.


    ACKNOWLEDGMENTS
 
We gratefully acknowledge Dr. Michael Wall for valuable discussions on the SVD method and for comments and suggestions on the manuscript.

Present address of R. Gottardo: University of Washington, Department of Statistics, Office B307, Box 354322, Seattle, WA 98195-4322.


    FOOTNOTES
 
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).

Address for reprint requests and other correspondence: J. F. Challacombe, Los Alamos National Laboratory, Bioscience Division, Mail Stop M888, Los Alamos, NM 87545 (E-mail: jchalla{at}lanl.gov).

1 The Supplementary Material for this article (Supplemental Tables S1–S6) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00155.2003/DC1. Back


    References
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 GRANTS
 References
 

  1. Affymetrix. Gene Chip Analysis Suite User Guide. Santa Clara, CA: Affymetrix, 1999.
  2. Albrecht T, Nachtigal M, Stjeor S, and Rapp F. Induction of cellular DNS-synthesis and increased mitotic-activity in Syrian hamster embryo cells abortively infected with human cytomegalovirus. J Gen Virol 30: 167–177, 1976.[Abstract]
  3. Alter O, Brown P, and Botstein D. Singular value decomposition for genome-wide expression data processing and modeling. Proc Natl Acad Sci USA 18: 10101–10106, 2000.[CrossRef]
  4. Aoki M, Hecht A, Kruse U, Kemler R, and Vogt P. Nuclear endpoint of Wnt signaling: neoplastic transformation induced by transactivating lymphoid-enhancing factor 1. Proc Natl Acad Sci USA 96: 139–144, 1999.[Abstract/Free Full Text]
  5. Beere H and Green D. Stress management: heat shock protein-70 and the regulation of apoptosis. Trends Cell Biol 11: 6–10, 2001.[CrossRef][ISI][Medline]
  6. Brander C and Walker B. Modulation of host immune responses by clinically relevant human DNA and RNA viruses. Curr Opin Microbiol 3: 379–386, 2000.[CrossRef][ISI][Medline]
  7. Browne E, Wing B, Coleman D, and Shenk T. Altered cellular mRNA levels in human cytomegalovirus-infected fibroblasts: viral block to the accumulation of antiviral mRNAs. J Virol 75: 12319–12330, 2001.[Abstract/Free Full Text]
  8. Brunet A, Bonni A, Zigmond M, Lin M, Juo P, Hu L, Anderson M, Arden K, Blenis J, and Greenberg M. Akt promotes cell survival by phosphorylating and inhibiting a forkhead transcription factor. Cell 96: 857–868, 1999.[ISI][Medline]
  9. Bui T, Zhang L, Rees M, Bicknell R, and Harris A. Expression and hormone regulation of Wnt 2, 3, 4, 5a, 7a, 7b and 10b in normal human endometrium and endometrial carcinoma. Br J Cancer 75: 1131–1136, 1997.[ISI][Medline]
  10. Bzowska A, Kulikowska E, and Shugar D. Purine nucleoside phosphorylase: properties, functions, and clinical aspects. Pharmacol Ther 88: 349–425, 2000.[CrossRef][ISI][Medline]
  11. Cohen G. Caspases: the executioners of apoptosis. Biochem J 326: 1–16, 1997.[ISI][Medline]
  12. Consortium TGO. Gene Ontology: tool for the unification of biology. Nat Genet Suppl 25: 25–29, 2000.[CrossRef]
  13. Craigen J, Yong K, Jordan N, MacCormac L, Westwick J, Akbar A, and Grundy J. Human cytomegalovirus infection up-regulates interleukin-8 gene expression and stimulates neutrophil transendothelial migration. Immunology 92: 138–145, 1997.[CrossRef][ISI][Medline]
  14. Crispe I. Death and destruction of activated T lymphocytes. Immunol Res 19: 143–157, 1999.[ISI][Medline]
  15. Doffinger R, Dupuis S, Picard C, Fieschi C, Feinberg J, Barcenas-Morales G, and Casanova JL. Inherited disorders of IL-12 and IFN-mediated immunity: a molecular genetics update. Mol Immunol 38: 903–909, 2001.[CrossRef][ISI]
  16. Duggan D, Bittner M, Chen Y, Meltzer P, and Trent J. Expression profiling using cDNA microarrays. Nat Genet Suppl 21: 10–14, 1999.[CrossRef]
  17. Eisen M, Spellman P, Brown P, and Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 95: 14863–14868, 1998.[Abstract/Free Full Text]
  18. Fanger N, Cosman D, Peterson L, Braddy S, Maliszewski C, and Borges L. The MHC class I binding proteins LIR-1 and LIR-2 inhibit Fc receptor-mediated signaling in monocytes. Eur J Immunol 28: 3423–3434, 1998.[CrossRef][ISI][Medline]
  19. Fearnhead H, Rodriguez J, Govec EE, Guo W, Kobayashi R, Hannon G, and Lazebnik Y. Oncogene-dependent apoptosis is mediated by caspase-9. Proc Natl Acad Sci USA 95: 13664–13669, 1998.[Abstract/Free Full Text]
  20. Foletta V, Segal D, and Cohen D. Transcriptional regulation in the immune system: all roads lead to AP-1. J Leukoc Biol 63: 139–152, 1998.[Abstract]
  21. Fukudome K and Esmon C. Identification, cloning, and regulation of a novel endothelial-cell protein-C activated protein-C receptor. J Biol Chem 269: 26486–26491, 1994.[Abstract/Free Full Text]
  22. Gately M, Renzetti L, Magram J, Stern A, Adorini L, Gubler U, and Presky D. The interleukin-12/interleukin-12-receptor system: Role in normal and pathologic immune responses. Annu Rev Immunol 16: 495–521, 1998.[CrossRef][ISI][Medline]
  23. Ghosh S, May M, and Kopp E. NF-kB and rel proteins: conserved mediators of immune responses. Annu Rev Immunol 16: 225–260, 1998.[CrossRef][ISI][Medline]
  24. Giancotti F and Ruoslahti E. Transduction: integrin signaling. Science 285: 1028–1032, 1999.[Abstract/Free Full Text]
  25. Guidotti L and Chisari F. Cytokine-mediated control of viral infections. Virology 273: 221–227, 2000.[CrossRef][ISI][Medline]
  26. Harrington C, Rosenow C, and Retief J. Monitoring gene expression using DNA microarrays. Curr Opin Microbiol 3: 285–291, 2000.[CrossRef][ISI][Medline]
  27. Hayashi M, Blankenship C, and Shenk T. Human cytomegalovirus UL69 protein is required for efficient accumulation of infected cells in the G1 phase of the cell cycle. Proc Natl Acad Sci USA 97: 2692–2696, 2000.[Abstract/Free Full Text]
  28. Hendler R and Shrager R. Deconvolutions based on singular value decomposition and the pseudoinverse: a guide for beginners. J Biochem Biophys Methods 28: 1–33, 1994.[CrossRef][ISI][Medline]
  29. Hengel H, Brune W, and Koszinowski U. Immune evasion by cytomegalovirus: survival strategies of a highly adapted opportunist. Trends Microbiol 6: 190–197, 1998.[CrossRef][ISI][Medline]
  30. Holter N, Mitra M, Maritan A, Cieplak M, Banavar J, and Fedoroff N. Fundamental patterns underlying gene expression profiles: simplicity from complexity. Proc Natl Acad Sci USA 97: 8409–8414, 2000.[Abstract/Free Full Text]
  31. Jault F, Jault JM, Ruchti F, Fortunato E, Clark C, Corbeil J, Richman D, and Spector D. Cytomegalovirus infection induces high levels of cyclins, phosphorylated Rb, and p53, leading to cell cycle arrest. J Virol 69: 6697–6704, 1995.[Abstract]
  32. Joliffe I. Principal Component Analysis. New York: Springer-Verlag, 1986.
  33. Kim C and Broxmeyer H. Chemokines: signal lamps for trafficking of T and B cells for development and effector function. J Leukoc Biol 65: 6–15, 1999.[Abstract]
  34. Knox J, Cress A, Clark V, Manriquez L, Affinito K, Dalkin B, and Nagle R. Differential expression of extracellular matrix molecules and the alpha 6-integrins in the normal and neoplastic prostate. Am J Pathol 45: 167–174, 1994.
  35. Koukoulis G, Virtanen I, Korhonen M, Laitinen L, Quaranta V, and Gould V. Immunohistochemical localization of integrins in the normal, hyperplastic, and neoplastic breast. Correlations with their functions as receptors and cell adhesion molecules. Am J Pathol 139: 787–799, 1991.[Abstract]
  36. Kowalik T, Wing B, Haskill J, Azizkhan J, Baldwin A, and Huang E. Multiple mechanisms are implicated in the regulation of NF-kappa-B activity during human cytomegalovirus-infection. Proc Natl Acad Sci USA 90: 1107–1111, 1993.[Abstract]
  37. Kumar S, Tzimas M, Griswold D, and Young P. Expression of ST2, an interleukin-1 receptor homologue, is induced by proinflammatory stimuli. Biochem Biophys Res Commun 235: 474–478, 1997.[CrossRef][ISI][Medline]
  38. Lalani A and McFadden G. Evasion and exploitation of chemokines by viruses. Cytokine Growth Factor Rev 10: 219–233, 1999.[CrossRef][ISI][Medline]
  39. Lankat-Buttgereit B and Tampé R. The transporter associated with antigen processing: function and implications in human diseases. Physiol Rev 82: 187–204, 2002.[Abstract/Free Full Text]
  40. Legler D, Loetscher M, Roos R, Clark-Lewis I, Baggiolini M, and Moser B. B cell-attracting chemokine 1, a human CXC chemokine expressed in lymphoid tissues, selectively attracts B lymphocytes via BLR1/CXCR5. J Exp Med 187: 655–660, 1998.[Abstract/Free Full Text]
  41. Leonard WJ and O’Shea JJ. JAKS and STATS: biological implications. Annu Rev Immunol 16: 293–322, 1998.[CrossRef][ISI][Medline]
  42. Levy D. Physiological significance of STAT proteins: investigations through gene disruption in vivo. Cell Mol Life Sci 55: 1559–1567, 1999.[CrossRef][ISI][Medline]
  43. Li C and Wong W. Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci USA 98: 31–36, 2001.[Abstract/Free Full Text]
  44. Li C and Wong W. Model-based analysis of oligonucleotide arrays: model validation, design issues and standard error application. Genome Biol 2: 1–11, 2001.
  45. Li H, Tago K, Io K, Arai T, Iwahana H, Tominaga S, and Yanagisawa K. The cloning and nucleotide sequence of human ST2L cDNA. Genomics 67: 284–290, 2000.[CrossRef][ISI][Medline]
  46. Liaw P, Mather T, Oganesyan N, Ferrell G, and Esmon C. Identification of the protein C/activated protein C finding sites on the endothelial cell protein C receptor: implications for a novel mode of ligand recognition by a major histocompatibility complex class 1-type receptor. J Biol Chem 276: 8364–8370, 2001.[Abstract/Free Full Text]
  47. Lin JX and Leonard W. The role of Stat5a and Stat5b in signaling by Il-2 family cytokines. Oncogene 19: 2566–2576, 2000.[CrossRef][ISI][Medline]
  48. Lipshutz R, Fodor S, Gingeras T, and Lockhart D. High density synthetic oligonucleotide arrays. Nat Genet Suppl 21: 20–24, 1999.[CrossRef]
  49. Loenen W, Bruggemen C, and Wiertz E. Immune evasion by human cytomegalovirus: lessons in immunology and cell biology. Semin Immunol 13: 41–49, 2001.[CrossRef][ISI][Medline]
  50. Loetscher P, Seitz M, Baggiolini M, and Moser B. Interleukin-2 regulates CC chemokine receptor expression and chemotactic responsiveness in T lymphocytes. J Exp Med 184: 569–577, 1996.[Abstract]
  51. Loetscher P, Uguccioni M, Bordoli L, Baggiolini M, Moser B, Chizzolini C, and Dayer J. CCR5 is characteristic of Th1 lymphocytes. Nature 391: 344–345, 1998.[CrossRef][ISI][Medline]
  52. Lu L, Zhu J, Zheng Z, Yan M, Xu W, Sun L, Theze J, and Liu X. Jak-STAT pathway is involved in the induction of TNF- gene during stimulation by IL-2. Eur J Immunol 28: 805–810, 1998.[CrossRef][ISI][Medline]
  53. Lu M and Shenk T. Human cytomegalovirus infection inhibits cell cycle progression at multiple points, including the transition from G1 to S. J Virol 70: 8850–8857, 1996.[Abstract]
  54. Lu M and Shenk T. Human cytomegalovirus UL69 protein induces cells to accumulate in G1 phase of the cell cycle. J Virol 73: 676–683, 1999.[Abstract/Free Full Text]
  55. Luo G and Yu-Lee L. Stat5b inhibits NF-kB-mediated signaling. Mol Endocrinol 14: 114–123, 2000.[Abstract/Free Full Text]
  56. Maguire J, Santoro T, Jensen P, Siebenlist U, Yewdell J, and Kelly K. Gem: an induced, immediate early protein belonging to the ras family. Science 265: 241–242, 2003.
  57. Mahalingham S and Karupiah G. Chemokines and chemokine receptors in infectious diseases. Immunol Cell Biol 77: 469–475, 1999.[CrossRef][ISI][Medline]
  58. Medema J, Toes R, Scaffidi C, Zheng T, Flavell R, Melief C, Peter M, Offringa R, and Krammer P. Cleavage of FLICE (caspase-8) by granzyme B during cytotoxic T lymphocyte induced apoptosis. Eur J Immunol 27: 3492–3498, 1997.[ISI][Medline]
  59. Miyagi Y, Imai N, Sasatomi T, Yamada A, Mine T, Katagiri K, Nakagawa M, Muto A, Okouchi S, Isomoto H, Shirouzu K, Yamana H, and Itoh K. Induction of cellular immune responses to tumor cells and peptides in colorectal cancer patients by vaccination with SART3 peptides. Clin Cancer Res 7: 3950–3962, 2001.[Abstract/Free Full Text]
  60. Montoya M, Sancho D, Vicente-Manzanares M, and Sanchez-Madrid F. Cell adhesion and polarity during immune interactions. Immunol Rev 186: 68–82, 2002.[CrossRef][ISI][Medline]
  61. Mossman KLMP, Rozmus JJ, Goryachev AB, Edwards AM, and Smiley JR. Herpes simplex virus triggers and then disarms a host antiviral response. J Virol 75: 750–758, 2001.[Abstract/Free Full Text]
  62. Nagata S. Fas ligand-induced apoptosis. Annu Rev Genet 33: 29–55, 1999.[CrossRef][ISI][Medline]
  63. Natali P, Nicotra M, Botti C, Mottolese M, Bigotti A, and Segatto O. Changes in expression of alpha 6/beta 4 integrin heterodimer in primary and metastatic breast cancer. Br J Cancer 66: 318–322, 1992.[ISI][Medline]
  64. Nielsen TO, West RB, Linn SC, Alter O, Knowling MA, O’Connell JX, Zhu S, Fero M, Sherlock G, Pollack JR, Brown PO, Botstein D, and van de Rijn M. Molecular characterisation of soft tissue tumours: a gene expression study. Lancet 359: 1301–1307, 2002.[CrossRef][ISI][Medline]
  65. Pass R. Cytomegalovirus. In: Fields Virology, edited by Knipe D, Howley P, Griffin D, Lamb R, Martin M, Roizman B, and Strauss S. Philadelphia, PA: Lippincott-Raven, 2001, p. 2675–2705.
  66. Peifer M and Polakis P. Wnt signaling in oncogenesis and embryogenesis: a look outside the nucleus. Science 287: 1606–1609, 2000.[Abstract/Free Full Text]
  67. Pucharos C, Casas C, Nadal M, Estivill X, and de la Luna S. The human intersectin genes and their spliced variants are differentially expressed. Biochim Biophys Acta 1521: 1–11, 2001.[ISI][Medline]
  68. Qiu L, Joazeiro C, Fang N, Wang HY, Elly C, Altman Y, Fang D, Hunter T, and Liu YC. Recognition and ubiquitination of Notch by Itch, a Hect-type E3 ubiquitin ligase. J Biol Chem 275: 35734–35737, 2000.[Abstract/Free Full Text]
  69. Rensing E, Hess S, Ziegler Heitbrock H, Reithmuller G, and Engelmann H. Fas/Apo-1 activated nuclear factor kappa B and induces interleukin-6 production. J Inflamm 45: 161–174, 1995.[ISI][Medline]
  70. Ridley S, Sarsfield S, Lee J, Bigg H, Cawston T, Taylor D, DeWitt D, and Saklatvala J. Actions of IL-1 are selectively controlled by p38 mitogen-activated protein kinase. Regulation of prostaglandin H synthase-2, metalloproteinases, and IL-6 at different levels. J Immunol 158: 3165–3173, 1997.[Abstract]
  71. Roitt I, Brostoff J, and Male D. Immunology. New York: Gower Medical, 1985.
  72. Saitoh A, Hansen L, Vogel J, and Udey M. Characterization of Wnt gene expression in murine skin: possible involvement of epidermis-derived Wnt-4 in cutaneous epithelial-mesenchymal interactions. Exp Cell Res 243: 150–160, 1998.[CrossRef][ISI][Medline]
  73. Sayegh C, Quong M, Agata Y, and Murre C. E-proteins directly regulate expression of activation-induced deaminase in mature B cells. Nat Immun 4: 586–593, 2003.[CrossRef][ISI]
  74. Scammell J, Hubler T, Denny W, and Valentine D. Organization of the human Fk506-binding immunophilin FKBP52 protein gene (FKBP4). Genomics 81: 640–643, 2003.[CrossRef][ISI][Medline]
  75. Schadt E, Li C, Su C, and Wong W. Analyzing high-density oligonucleotide gene expression array data. J Cell Biochem 80: 192–202, 2000.[CrossRef][ISI][Medline]
  76. Shahgasempour S, Woodroffe SB, Sullivan-Tailyour G, and Garnett HM. Alteration in the expression of endothelial cell integrin receptors alpha 5 beta 1 and alpha 2 beta 1 and alpha(6)beta(1) after in vitro infection with a clinical isolate of human cytomegalovirus. Arch Virol 142: 125–138, 1997.[CrossRef][ISI][Medline]
  77. Simmen K, Singh J, Luukkonen B, Lopper M, Bittner A, Miller N, Jackson M, Compton T, and Fruh K. Global modification of cellular transcription by human cytomegalovirus is initiated by viral glycoprotein B. Proc Natl Acad Sci USA 98: 7140–7145, 2001.[Abstract/Free Full Text]
  78. Smalley M and Dale T. Wnt signaling in mammalian development and cancer. Cancer Metastasis Rev 18: 215–230, 1999.[CrossRef][ISI][Medline]
  79. Song Y and Stinski M. Effect of the human cytomegalovirus IE86 protein on expression of E2F-responsive genes: a DNA microarray analysis. Proc Natl Acad Sci USA 99: 2836–2841, 2002.[Abstract/Free Full Text]
  80. Steeber D and Tedder T. Adhesion molecule cascades direct lymphocyte recirculation and leukocyte migration during inflammation. Immunol Res 22: 299–317, 2000.[CrossRef][ISI][Medline]
  81. Tago K, Nakamura T, Nishita M, Hyodo J, Nagai S, Murata Y, Adachi S, Ohwada S, Morishita Y, Shibuya H, and Akiyama T. Inhibition of Wnt signaling by ICAT, a novel ß-catenin-interacting protein. Genes Dev 14: 1741–1749, 2000.[Abstract/Free Full Text]
  82. Thrasher A. WASp in immune-system organization and function. Nat Rev Immunol 2: 635–646, 2002.[CrossRef][ISI][Medline]
  83. Tortorella D, Gewurz B, Furman M, Scust D, and Ploegh H. Viral subversion of the immune system. Annu Rev Immunol 18: 861–926, 2000.[CrossRef][ISI][Medline]
  84. Ugolini F, Adelaide J, Charafe-Jauffret E, Nguyen C, Jacquemier J, Jordan B, Birnbaum D, and Pebusque M. Differential expression assay of chromosome arm 8p genes identifies Frizzled-related (FRP1/FRZB) and fibroblast growth factor receptor 1 (FGFR1) as candidate breast cancer genes. Oncogene 18: 1903–1910, 1999.[CrossRef][ISI][Medline]
  85. Ura H, Denno R, Hirata K, Yamaguchi K, and Yasoshima T. Separate functions of {alpha}2ß1 and {alpha}3ß1 integrins in the metastatic process of human gastric carcinoma. Surg Today 28: 1001–1006, 1998.[CrossRef][ISI][Medline]
  86. Van den Eynde BJ and Morel S. Differential processing of class-I-restricted epitopes by the standard proteasome and the immunoproteasome. Curr Opin Immunol 13: 147–153, 2001.[CrossRef][ISI][Medline]
  87. Villard J, Peretti M, Masternak K, Barras E, Caretti G, Mantovani R, and Reith W. A functionally essential domain of RFX5 mediates activation of major histocompatibility complex class II promoters by promoting cooperative binding between RFX and NF-Y. Mol Cell Biol 20: 3364–3376, 2000.[Abstract/Free Full Text]
  88. Wakkach A, Cottrez F, and Groux H. Differentiation of regulatory T cells is induced by CD2 costimulation. J Immunol 167: 3107–3113, 2001.[Abstract/Free Full Text]
  89. Waldmann T, Dubois S, and Tagaya Y. Contrasting roles of IL-2 and IL-15 in the life and death of lymphocytes: implications for immunotherapy. Immunity 14: 105–110, 2001.[CrossRef][ISI][Medline]
  90. Wall M, Dyck P, and Brettin T. SVDMAN: singular value decomposition analysis of microarray data. Bioinformatics 17: 566–568, 2001.[Abstract/Free Full Text]
  91. Wall M, Rechtsteiner A, and Rocha L. Singular value decomposition and principal component analysis. In: A Practical Approach to Microarray Data Analysis, edited by Berrar D, Dubitzky W, and Granzow M. Boston, MA: Kluwer, 2003.
  92. Wallach D, Varfolomeev E, Malinin N, Goltsev Y, Kovalenko A, and Boldin M. Tumor necrosis factor receptor and Fas signaling mechanisms. Annu Rev Immunol 17: 331–367, 1999.[CrossRef][ISI][Medline]
  93. Warren A, Owens C, Borysiewicz L, and Patel K. Down-regulation of integrin alpha-1/beta-1 expression and association with cell rounding in human cytomegalovirus-infected fibroblasts. J Gen Virol 75: 3319–3325, 1994.[Abstract]
  94. Warren H and Smyth M. NK cells and apoptosis. Immunol Cell Biol 77: 64–75, 1999.[CrossRef][ISI][Medline]
  95. Watts C, Moss CX, Mazzeo D, West MA, Matthews SP, Li DN, and Manoury B. Creation versus destruction of T cell epitopes in the class II MHC pathway. Ann NY Acad Sci 987: 9–14, 2003.[Abstract/Free Full Text]
  96. Wiertz E, Mukherjee S, and Ploegh H. Viruses use stealth technology to escape from the host immune system. Mol Med Today 3: 116–123, 1997.[CrossRef][ISI][Medline]
  97. Yeung M, Tegner J, and Collins J. Reverse engineering gene networks using singular value decomposition and robust regression. Proc Natl Acad Sci USA 99: 6163–6168, 2002.[Abstract/Free Full Text]
  98. Yewdell J and Bennink J. Cut and trim: generating MHC class I peptide ligands. Curr Opin Immunol 13: 13–18, 2001.[CrossRef][ISI][Medline]
  99. Yoshida H, Hamano S, Senaldi G, Covey T, Faggioni R, Mu S, Xia M, Wakeham A, Nishina H, Potter J, Saris C, and Mak T. WSX-1 is required for the initiation of Th1 responses and resistance to L-major infection. Immunity 15: 569–578, 2001.[ISI][Medline]
  100. Zhu H, Cong J, Mamtora G, Gingeras T, and Shenk T. Cellular gene expression altered by human cytomegalovirus: global monitoring with oligonucleotide arrays. Proc Natl Acad Sci USA 95: 14470–14475, 1998.[Abstract/Free Full Text]
  101. Zhu H, Shen Y, and Shenk T. Human cytomegalovirus IE1 and IE2 proteins block apoptosis. J Virol 69: 7960–7970, 1995.[Abstract]