Genetic Deficiency of Chemokine Receptor CCR5 Is a Strong Risk Factor for Symptomatic West Nile Virus Infection: A Meta-Analysis of 4 Cohorts in the US Epidemic

West Nile virus (WNV) is a reemerging mosquito-borne flavivirus that infects birds as the natural host and humans and other vertebrates as dead-end hosts. First detected in the Western Hemisphere in 1999 [1], there have been 23,967 US laboratory—confirmed symptomatic cases reported to the Centers for Disease Control and Prevention (CDC) through 3 April 2007, with 959 (4.0%) WNV-related deaths (http://www.cdc.gov).

In the United States, ∼20% of WNV-infected individuals develop symptoms ranging from West Nile fever (WNF;∼60%) to WNV-induced neuroinvasive disease (WNND; ∼40%), which includes acute-flaccid paralysis, meningitis, and/or encephalitis [2]. Risk factors for WNV disease include increased age, immunosuppression, and host genetics. In mice, WNV susceptibility is dependent on 2′-5′'-oligoadenylate synthetase [3], Toll-like receptor 3 [4], CXCL10 [5], and the chemokine receptor/HIV-1 coreceptor CCR5 [6]. CCR5 also appears to be important in controlling WNV disease in humans, given that we found a strong epidemiologic association in symptomatic WNV-seropositive cohorts from Arizona and Colorado with homozygosity for CCR5Δ32 [7], a complete loss-of-function mutation found primarily in white persons [8]. CCR5^−/− mice uniformly die from fatal WNV encephalitis via a mechanism that appears to involve defective leukocyte recruitment into the WNV-infected central nervous system (CNS). In the present study, we have analyzed symptomatic WNV-seropositive cohorts from California and Illinois to further test the hypothesis that CCR5 is a WNV restriction gene in the US epidemic.

Methods. The study was approved by the Office of Human Subjects Research of the US National Institutes of Health (NIH). Three patient groups were defined: a control group of random, healthy, North American white blood donors from the NIH Department of Transfusion Medicine (n = 1318), previously established under an Investigational Review Board—approved protocol, and 2 symptomatic WNV-seropositive cohorts. The first was from the 2005 epidemic in California. Serum samples were tested for WNV at the Viral and Rickettsial Disease Laboratory by both IgM and IgG ELISAs. The case patients (n = 125) represented all forwhomat least 500 μL of serum were available and for whom the following information was available: age, sex, self-reported racial group, and CDC-defined clinical presentation (WNF, WNND [meningitis or encephalitis], or death [table 1]). The second was from Illinois during the 2005 and 2006 mosquito seasons. Patient serum samples were tested by the Illinois Department of Public Health by a WNV-specific IgM ELISA in 2005 and a WNV/St. Louis encephalitis (SLE) virus duplex microsphere immunofluorescence assay in 2006. Eligibility criteria were seropositivity for WNV, seronegativity for SLE virus, and availability of 500 μL serum. One hundred symptomatic case patients meeting these criteria were chosen at random; 25 were from the 2005 mosquito season, and 75 were from 2006. The following information was provided for each case patient if available: age, sex, and self-reported racial group (table 1). No information was provided regarding disease manifestations. Investigators were blinded to unique patient identifiers.

For DNA extraction, genomic DNA was extracted from 200 μL of serum by use of the QiaAmp 96 DNA Blood Kit, in accordance with the manufacturer's instructions (Qiagen). DNA from random blood donors was isolated from peripheral blood leukocytes as described elsewhere [9].

Genotyping was performed by standard methods described elsewhere [10]. Briefly, patient DNA was amplified using primers flanking the 32-bp deletion in CCR5Δ32: 5′-GTCTTCATTACACCTGCAGCTCTC-3′ and 5′-GTCCAACCTGTTAGAGCTACTGC-3′. Each sample was tested by independent polymerase chain reactions, and results were concordant, as determined by 2 independent investigators.

Odds ratios (ORs) and 95% confidence limits (CIs) were calculated using a recessive genetic model (i.e., CCR5Δ32 homozygotes vs. wild-type CCR5 plus CCR5Δ32 heterozygotes) by cross-tabulation. Significance was determined by the χ² test using a 2-sided P value, and 95% CIs were estimated using the approximation of Woolf as implemented in Prism (version 4.0b; GraphPad Software). The Hardy-Weinberg equation was used to calculate the expected frequencies of each of the 3 genotypes, judged using a χ² test and 2 df.

Results. The subject characteristics are given in table 1. CCR5 genotypes were successfully obtained for all 125 of the subjects from California and for 99 of the 100 subjects from Illinois. The genotype distribution observed for the combined California and Illinois cohorts deviated significantly from Hardy-Weinberg equilibrium (P = .05), mainly because of an increased frequency of CCR5Δ32 homozygotes. Eight (3.6%) of the 224 informative case patients unstratified by race were CCR5Δ32 homozygotes (table 1). To quantitate the significance of this genotype-phenotype association, we used a reference group of 1318 random, healthy, white US blood donors who donated before 1999, of which 13 (1.0%) were CCR5Δ32 homozygotes. Compared with this value, which falls within the range reported by other groups [8], the frequency of CCR5Δ32 homozygotes in each cohort was increased (table 1). The OR was 4.2 for the California cohort (95% CI, 1.5–11.9; P = .004) and 3.1 for the Illinois cohort (95% CI, 0.9–11.2; P = .06). When the California and Illinois data were combined, the OR was 3.7 (95% CI, 1.5–9.1; P = .002). This is most likely an underestimate, because the cohorts were not censored for nonwhite racial groups, whose CCR5Δ32 allele frequency is known to be very low compared with that of white persons [8].

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 1

Meta-analysis of the association between CCR5Δ32 homozygosity and the risk of symptomatic West Nile virus disease in the United States. Cohorts from Colorado (n = 148), Arizona (n = 247), Illinois (n = 99), and California (n = 125), unstratified for race, were analyzed separately or combined (n = 619) for the frequency of homozygous CCR5Δ32 genotype and compared with a reference cohort of random, healthy, white US blood donors (n = 1318) in which 1% were CCR5Δ32 homozygotes. For each cohort, odds ratios (black ticks) and 95% confidence intervals (bars) are shown. Data are plotted on a log base 2 scale. The dotted line indicates an odds ratio of 1.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1

Characteristics and CCR5 genotypes of study subjects.

Consistent with this, when the analysis was limited to selfreporting white persons, the frequency of CCR5Δ32 homozygotes increased to 4.0% in the 2 cohorts combined (n = 126; OR, 4.1 [95% CI, 1.5–11.8]; P = .004). When self-reporting white persons were analyzed from each cohort separately, an increase in CCR5Δ32 homozygosity was observed in the California cohort (OR, 4.8 [95% CI, 1.5–14.9]; P = .003). Only 38 of the 99 case patients from the Illinois cohort were self-reporting white persons, and 1 was a CCR5Δ32 homozygote. This still represents an increase relative to the control population (2.6% vs. 1%); however, the sample size was underpowered to demonstrate a statistically significant association (OR, 2.7 [95% CI, 0.34–21.2]; P = .32).

Clinical-outcome data were available only for the California cohort. Of the 125 race-unstratified patients from California, 76 (60.8%) presented with fever, 38 (30.4%) with neurological disease, and 6 (4.8%) died. This distribution is similar to the outcome distribution reported nationwide for WNV disease in the United States from 1 January to 31 December 2006 by the CDC (61.3% for WNF, 34.2% for WNND, and 4.1% for death). The 5 CCR5Δ32 homozygotes in the California cohort were all given a diagnosis of WNF and represented 6.6% of the subjects in that category (P = .0021, compared with random blood donors). None of the 38 subjects given a diagnosis of WNND in the California cohort was a CCR5Δ32 homozygote, unlike in the Arizona and Colorado cohorts we previously reported [7], in which the frequency was ∼4%. This does not mean that CCR5Δ32 homozygosity is not associated with WNND in the California portion of the epidemic, however, since a sample size of 38 is underpowered to identify an association between 4% CCR5∼32 homozygosity and WNND; only 1.5 homozygotes would have been expected among these 38 subjects, on the basis of the Arizona and Colorado data. The Arizona and Colorado WNND sample sizes were much larger (n = 125 and n = 112, respectively). Similarly, the California cohort was underpowered to evaluate the association between CCR5Δ32 homozygosity and death, since only 6 subjects died.

We next performed a meta-analysis of the association between CCR5Δ32 homozygosity and symptomatic WNV disease in all 4 cohorts studied to date. Figure 1 shows data for all subjects from each cohort, analyzed separately and together. The ORs ranged between 3 and 5 for the individual cohorts. CCR5Δ32 homozygotes comprised 4.0% of the 619 subjects from all cohorts (OR, 4.2 [95% CI, 2.1–8.3]; P < .0001).

Discussion. We have shown that, compared with that in a random US blood donor population, the frequency of CCR5Δ32 homozygosity is increased among symptomatic WNV-infected subjects from both the 2005 WNV epidemic in California and the 2005/2006 WNV epidemic in Illinois. The associations are consistent in magnitude with those defined previously for the 2003 Colorado and 2004 Arizona WNV epidemics [7]. Demonstrating this in 4 of 4 independent cohorts and in a total of 619 subjects significantly strengthens the hypothesis that CCR5 is an important regulator of host defense to WNV infection.

This confirmation is important because human gene association studies, for reasons of inadequate sample size and/or selection bias, often cannot be reproduced [11, 12] and because neither the Arizona nor the Colorado cohort previously studied was large. The sample-size problem is well-illustrated in the Illinois cohort, in which increased CCR5Δ32 homozygosity among symptomatic WNV-seropositive subjects did not quite reach statistical significance (P = .06), probably because of the cohort's small size (n = 99). This is, nevertheless, a very strong trend and is fully consistent with the strong statistical significance found in the other 3 cohorts; the weaker association in Illinois may relate not only to sample size but also to possible differences in racial makeup. Race information, available for only 40% of the Illinois cohort (compared with all of the California cohort), is especially relevant for CCR5Δ32, given that it is found primarily in white persons. The more subjects in a cohort who are not white, the lower the apparent strength of the genotype-phenotype association will be if no subjects are censored. Conversely, analyzing only the self-reporting white subset should result in an increase in the frequency of CCR5Δ32 homozygotes compared with that in the complete cohort, as long as all of the homozygotes report race.

With regard to CCR5 heterozygosity and symptomatic WNV disease, our data show no evidence for a gene-dosage effect, since the presence of 1 copy of CCR5Δ32 was found at a frequency of 13.5% (table 1), which is similar to the frequency of heterozygotes in our control population (15.4%) and is also in agreement with the findings of our previous study [7]. These data imply that an intermediate amount of CCR5 is sufficient for defense against WNV disease in humans and suggest that other functional variants of CCR5 (such as promoter polymorphisms) that modulate, rather than ablate, CCR5 expression [13, 14] are unlikely to affect genetic susceptibility of disease in WNV-infected individuals.

There are 3 important limitations to the analysis of the WNV epidemic we have performed. The first relates to the association between CCR5Δ32 homozygosity and disease outcome. Although we have access to 619 WNV-seropositive subjects, all were analyzed retrospectively and only 19 are CCR5Δ32 homozygotes with defined disease outcomes. This does not provide sufficient statistical power to determine the relative risk of outcomes once infected. Second, none of the WNV-seropositive subjects we have analyzed was asymptomatic. Therefore, we are unable to determine on the basis of the clinical material currently available to us whether CCR5Δ32 homozygosity affects the risk of initial infection, the risk of clinical disease, or both. Third, despite the strength and reproducibility of the association between CCR5Δ32 and symptomatic WNV disease, these analyses do not address the precise mechanism of CCR5 action in human disease and leave as an open question what genetic and environmental risk factors account for symptomatic WNV disease in the 96% of individuals studied who were not CCR5Δ32 homozygotes. Our previous study of a mouse model of WNV infection suggested that CCR5 plays an important role in the protection of the CNS by mediating leukocyte accumulation [6].

The role played by CCR5 in WNV and HIV pathogenesis is diametrically opposed [15], with antimicrobial function in the context of WNV pathogenesis and promicrobial action in HIV pathogenesis. CCR5Δ32 homozygosity is protective in cohorts of HIV exposed but uninfected individuals with a magnitude of CCR5Δ32 enrichment (4.5%; OR, 6.04 [95% CI, 1.4–25.7]; P = .02) [9] that is very similar to what was observed in the WNV cohorts we have analyzed. Since CCR5Δ32 homozygotes develop normally and appear healthy, pharmacologic blockade of CCR5, which is intended to imitate the CCR5Δ32 genetic defect, may carry the cost of increased risk of symptomatic WNV disease in infected individuals. Prospective studies will be needed to test this hypothesis.

• Received May 17, 2007.
• Accepted July 6, 2007.

• © 2008 by the Infectious Diseases Society of America