Gene Expression Profiles in Hepatitis C Virus (HCV) and HIV Coinfection: Class Prediction Analyses before Treatment Predict the Outcome of Anti-HCV Therapy among HIV-Coinfected Persons

Coinfection with hepatitis C virus (HCV) is seen in 15%–30% of all HIV-infected persons in the United States [1]. With the marked reduction in AIDS-associated opportunistic infections that has resulted from the judicious use of antiretroviral therapy, liver disease is becoming a leading cause of death in this group [2]. Treatment of HCV infection with pegylated (peg) interferon (IFN) and ribavirin is associated with significant toxicities, with rates of sustained virological response (SVR) of <30% [3–6]. Coinfection with HIV also increases the rate of relapses after therapy among those who have achieved an end-of-treatment response (ETR) to combination therapy [3, 4, 6]. Several adverse events associated with combination therapy are magnified among HIV-coinfected persons, making anti-HCV therapy less tolerable for the HIV-coinfected population. The current standard of care for the treatment of HCV infection suggests that treatment for 12 weeks is required before making a determination of the likelihood of success of therapy and whether to continue therapy [3, 4, 6]. Predicting the chances of attaining SVR before initiation of therapy would limit the exposure of persons with a low likelihood of therapeutic success to the toxicities of the treatment. In this study, we used class prediction analysis of gene expression patterns in peripheral blood mononuclear cells (PBMCs) before therapy to predict the response of patients with HCV infection to combination therapy. We also used class prediction analysis of gene expression patterns in PBMCs after therapy to predict SVR among patients who have had an early virological response

MethodsTwenty-nine HIV-infected patients with chronic HCV infection and stable HIV disease with or without antiretroviral therapy were enrolled in studies at the National Institute of Allergy and Infectious Diseases and the University of Cincinnati for treatment of HCV infection among HIV-coinfected persons with peg-IFN-α2b (1.5 μg/kg/week) and ribavirin 1000–1200 mg/day (table 1). All patients gave institutional review board–approved protocol informed consent. Patients’ responses were defined as no response (NR), ETR (but not SVR; relapse), or SVR

PBMCs were collected before treatment and at the end of treatment (48 weeks, unless treatment was stopped earlier because of adverse events). Of the patients who had HCV loads below the limit of detection at the end of treatment (n=19), 9 went on to achieve SVR, and 10 had a relapse of HCV load. We did not have end-of-treatment samples for microarray analysis for 2 patients

PBMCs were collected by Ficoll-Hypaque separation for DNA microarray analysis. Total RNA and labeled cRNA synthesis and hybridization to the Affymetrix U133A human microarray were performed according to the manufacturer’s recommended protocol. Gene expression values (log₂) were determined by use of the GC-Robust Multi-Array (GC-RMA) algorithm (Z. J. Wu and R. Irizarray; available at: http://www.bioconductor.org/) followed by Loess normalization using an R package (available at: http://www.elwood9.net/spike). Genes that showed low variability or had expression levels for the majority of samples that were below the level of detection were eliminated from analysis, by removing those that had GC-RMA values with an interquartile range of <0.263 or a 75th percentile of <5, resulting in retention of 7527 of the initial ∼22,000 genes for subsequent analysis

Total RNA was isolated from 10 million PBMCs, using an RNAEasy Total RNA Isolation Kit (Qiagen), and 1 μg of total RNA was reverse-transcribed with random hexamer primers, using the TaqMan Reverse Transcription Kit System (Applied Biosystems). For quantitative real-time polymerase chain reaction (PCR), 1/100 of the first-strand cDNA synthesis reaction was used as template for detection of each gene. GIP2, IFITM2, MX1, OAS1 and B2M levels were detected using the Assay-on-Demand System (Applied Biosystems) and TaqMan Universal Master Mix (Applied Biosystems). The fluorescence signals were measured in real time during the extension step, using the iCycler iQ Real Time Detection System (Bio-Rad). The ratio of change in target genes relative to the B2M internal reference control gene was determined within each sample by the 2^−ΔΔCT method (an assessment of the log₂ fold change relative to B2M)

A mixed-model analysis of variance (PartekPro) was used, with patients as a random effect, treatment time (before or after) and response group (NR, ETR, SVR) as main effects, and treatment time by response group as the interaction. Of the genes, 1352 significantly modulated genes were identified as having a main effect or interaction comparison P value of <.05 and an absolute fold-change difference of >1.3. Differentially expressed genes were grouped using K-means clustering (Spotfire), functional category enrichment was performed using DAVID (version 2.1; available at: http://david.abcc.ncifcrf.gov/), and class prediction was performed using Prediction Analysis for Microarrays (PAM) software (version 2.0; Stanford University), which is a variant of nearest-centroid classification with an automated gene selection step integrated into the algorithm [7]

ResultsThe expression patterns of the 1352 genes significantly modulated in pretherapy samples were variably associated with the 3 patient response groups (figure 1A and table 2table 2 [an unedited tab-delimited ASCII file that can be downloaded into a spreadsheet]), suggesting that pretherapy expression profiles may be predictive of therapeutic responses. Two profiles of particular interest were those found in clusters 1 and 2, which represent 540 genes up-regulated in samples from the NR group, relative to those from the ETR and SVR groups, before therapy. These 2 clusters were highly enriched in genes involved in the IFN-stimulated response, defense response, immune cell signaling pathways, and cell death, which was suggestive of an ongoing endogenous IFN/innate immune response before therapy. The 20 IFN-stimulated genes found in clusters 1 and 2 (see table 2table 2 for the identity of these genes) were found to be significantly up-regulated as a group in samples from the NR group compared with samples from the ETR or SVR groups (P<.0001, mixed model analysis of variance) (figure 1B). Additionally, these 20 IFN-stimulated genes were not significantly induced by peg-IFN-α2b and ribavirin therapy in the NR group at the end of treatment, whereas they were induced in the ETR and SVR groups (P<.0001, mixed-model analysis of variance; data not shown). The lack of induction of IFN-stimulated genes in patients not responding to therapy has been seen by others, in a study of an HCV-monoinfected cohort [8]

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 1

Association between specific peripheral blood mononuclear cell (PBMC) gene expression profiles from pretherapy samples and different responses to anti–hepatitis C virus (HCV) therapy among patients coinfected with HIV. Analysis of treatment effects, response group effects, and treatment by response group interaction using mixed-model analysis of variance identified 1352 significantly modulated genes. A Average expression levels for each pretherapy response group were transformed to a relative scale (Z score) and grouped by K-means clustering. Clusters 1 and 2 represent genes with elevated levels before therapy (relative color scale: red, high expression; black, average expression; green, low expression) in samples from patients with no response to treatment (NR), relative to samples from those with an end-of-treatment response (ETR) or a sustained virological response (SVR). Functional annotation analysis of the 540 genes found in clusters 1 and 2 by use of DAVID 2.1 found that these 2 clusters were enriched for genes involved in the interferon (IFN) response and immune activation. B The levels of 20 IFN-stimulated genes found in clusters 1 and 2 were significantly elevated, as a group, in samples from the NR group, as shown by mixed-model analysis of variance. The scale for gene expression level (Y-axis) begins at the lower limit of detection—that is, 5 log₂

Real-time PCR detection of 4 IFN-stimulated genes in a small subset of pretherapy samples confirmed the induction of these genes in the NR group (table 3). In this group of samples, there was a strong correlation between fold change detected by real-time PCR and by Affymetrix GeneChip hybridization (r = 0.73; P<.001; data not shown)

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 2

A Correct association of 25 (86%) of the 29 pretherapy peripheral blood mononuclear cell samples with the patients’ response to anti–hepatitis C virus therapy, by class prediction analysis using 79 gene expression profiles from pretherapy samples and Prediction Analysis for Microarrays software. Samples belonging to known patient response groups are indicated across the upper X-axis. The 10× cross-validation probabilities of a given sample belonging to each patient response group are indicated by the symbols on the graph. The sum of all probabilities for a given sample will equal 100%. B Class prediction analysis using 105 gene expression profiles from 17 patients with end-of-treatment response (ETR) correctly predicted 16 of 17 times whether or not a patient would achieve a sustained virological response (SVR). Samples belonging to known patient response groups are indicated across the upper X-axis. The 10× cross-validation probabilities of a given sample belonging to each patient response group are indicated by the symbols on the graph. The sum of all probabilities for a given sample equal 100%. NR, no response to treatment

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1

Demographic profile of patients who participated in a study of gene expression profiles and the response to anti–hepatitis C virus (HCV) therapy in patients coinfected with HIV

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 3

Up-regulation of 4 interferon-stimulated genes in patients with no response to treatment (NR) relative to those with sustained virological response (SVR), as determined by Affymetrix GeneChip hybridization and confirmed by real-time polymerase chain reaction (PCR)

The distinctive expression profiles associated with the different patient response groups before therapy suggested that gene expression patterns may be predictive of patients’ responses to anti-HCV therapy. PAM software was used to predict group membership on the basis of gene expression patterns [7]. Class prediction model training was conducted with the 79 gene expression profiles that gave the lowest classification error rate (table 4table 4 [an unedited tab-delimited ASCII file that can be downloaded into a spreadsheet]). After model training, 10× cross-validation was conducted to test the model’s predictive capability (figure 2A). The model assigned a percentage to the likelihood of a specific subject falling into a response category. The figure illustrates this likelihood by showing the observed response group for each patient (numbers at bottom) and the predicted probability of belonging to each of the 3 response groups (symbols). The sum of the 3 predicted group probabilities for a given sample equals 100%. The group with the highest probability is classified as the predicted response group for the sample. The model correctly classified all samples from the NR group, 8 of 10 samples from the ETR group, and 7 of 9 samples from the SVR group, resulting in an overall correct-classification rate of 86%. Most significantly, all of the subjects in the NR group were correctly predicted by the microarray analysis, suggesting that there may be a role for this technology in determining, before therapy, which patients have a high likelihood of not responding to therapy. Interestingly, 9 of the 79 genes found to be most predictive of therapy response were IFN-stimulated genes

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 2

List of genes differentially modulated among the different pretherapy patient response groups

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 4

List of genes in pretherapy samples that are predictive of the therapeutic response to interferon-α treatment for hepatitis C virus infection

Similarly, class prediction analysis was performed with 17 available posttherapy samples from patients in the ETR group, in an attempt to predict which of these patients would have an SVR. Gene expression patterns from 105 genes (table 5table 5 [an unedited tab-delimited ASCII file that can be downloaded into a spreadsheet]) were able to predict 7 of 8 patients with SVR and 9 of 9 patients who went on to have a rebound in HCV load, with an overall correct classification rate of 94% (figure 2B). The biological processes associated with these predictive genes were involved in protein biosynthesis, mRNA maturation, and general metabolism

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 5

List of genes from end-of-treatment samples that are predictive of a sustained virological response to interferon-α treatment

DiscussionClass prediction analysis using differentially expressed genes in PBMCs from HIV-infected patients undergoing therapy for HCV infection accurately predicted the outcome of therapy. Gene expression profiling of samples from patients coinfected with HIV and HCV suggested that immune activation before anti-HCV therapy is associated with nonresponse. Increased levels of immune activation in HIV-infected patients have been associated with poor prognosis [9], resistance to highly active antiretroviral therapy, and decreased recovery after highly active antiretroviral therapy [10] and, thus, may similarly be associated with a lower rate of response to anti-HCV therapy in HIV-infected patients. Trippler et al. [8] have used class prediction analysis to study HCV-monoinfected persons undergoing therapy with peg-IFN-α2b and ribavirin at the same dosage and reached similar conclusions

The relationship between increased expression of IFN-stimulated genes in pretherapy PBMC samples (along with the inability of those genes to be induced by IFN-α) and nonresponsiveness to therapy with peg-IFN-2bα and ribavirin therapy is intriguing. The inability of IFN-stimulated genes to be further stimulated in pretherapy samples by IFN administration was also observed by Trippler et al. [8] in a study of HCV-monoinfected patients. Taken together, these results strongly suggest that patients showing an endogenously activated type I IFN induction pathway before therapy are resistant to the beneficial antiviral effects of IFN-α administration, suggesting that endogenous activation of IFN-stimulated genes in chronic disease may be a biomarker of immune dysfunction

Elevated levels of IFN-stimulated genes have been associated with active diseases, including systemic lupus erythematosus [11], tumor development [12], rheumatoid arthritis [13, 14], and HIV-1 infection (R.A.L., M.A.P., J.Y., D.W.H., and H.C.L., unpublished data), as well as simian immunodeficiency virus infection in cynomolgus macaques [15]. These results demonstrate that up-regulation of IFN-stimulated genes can be associated with progression of chronic diseases and that their expression may not always be coupled with the beneficial antiviral effect of IFN-α. When the data were reanalyzed excluding the 5 patients infected with HCV genotypes other than 1, there was no difference in the predicted response groups or the relative expression of IFN-stimulated genes, suggesting that the gene expression patterns are not merely a result of infection with a given genotype

The mechanism by which IFN-stimulated genes are induced in chronic diseases is unclear. It is unlikely that IFN-α itself is involved, at least during HIV infection, because a number of studies of HIV-infected patients have shown decreases in serum IFN-α levels; decreases in numbers of precursor dendritic cells, the major IFN-α producing cell in the body; and decreases in the IFN-α–producing capability of precursor dendritic cells. Alternative mechanisms of induction of IFN-stimulated genes may include stimulation through the retinoic acid signaling pathway; stimulation through Toll-like receptor pathways, including TLR-7 and TLR-9; or by circulating immune complexes, a hallmark of chronic immune activation. Such pathways may be induced during chronic diseases by other opportunistic infections; by metabolic dysregulation, in the case of retinoic acid signaling; or by increased apoptosis that allows exposure of Toll-like receptors to high levels of stimulating ligands, such as DNA and RNA. Whatever the reason, IFN-stimulated genes appear to be an important biomarker of disease progression in a number of chronic immune activating diseases while providing important antiviral roles in acute infections. It will be interesting to see whether IFN-stimulated genes are associated with nonresponsiveness to IFN-α therapy in patients coinfected with hepatitis B virus and HIV. Planned future studies using microarrays and targeted probes to the IFN-stimulated genes, larger numbers of samples for training, and a large number of blinded samples for model validation may serve to establish this novel approach for routine clinical use as a means to predict, in advance of treatment, which patients will respond anti-HCV therapy

• Received January 18, 2005.
• Accepted November 9, 2005.

• © 2006 by the Infectious Diseases Society of America