Association of CD14 -260 (-159) C>T and asthma: a systematic review and meta-analysis

Background Asthma is a phenotypically diverse disease with genetic susceptibility. A single nucleotide polymorphism (SNP) in the CD14 gene at position -260 (also known as -159) C>T has been inconsistently associated with asthma. The aim of this study was to estimate the combined likelihood of developing asthma given the CD14 -260C>T genotype. Methods Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, a systematic search and meta-analysis of the literature was conducted to estimate the association between this SNP and asthma. Planned subgroup analyses were performed to detect potential sources of heterogeneity from selected study characteristics. Post-hoc sensitivity analysis was performed to identify studies exerting excessive influence on among-study heterogeneity and combined effects. Results Meta-analysis of 23 studies yielded a non-significant overall association with high heterogeneity across studies. After restricting analysis to studies using atopic asthma and non-atopic non-asthma case-control phenotypes and excluding studies influencing heterogeneity, the genotype-specific odds ratios (ORs) suggested a codominant model. Carriers of the TT and CT genotypes were about 33% less likely (OR = 0.67, 95% CI: 0.54-0.84) and about 20% less likely (OR = 0.80, 95% CI: 0.66-0.95), respectively, to have atopic asthma compared to carriers of the CC genotype. Among-study heterogeneity may be explained by overly broad asthma phenotype definitions, gene-environment interactions, and gene-gene interactions. Conclusions A protective dose-response relationship between the CD14 -260T allele and atopic asthma susceptibility was observed. These results demonstrate the importance of precisely specified case-control groups as well as the need to assess interactions in the investigation of complex diseases such as asthma.


Background
Asthma is a common, complex, chronic medical condition characterized by lung inflammation, reversible airflow obstruction, and enhanced airway responsiveness to a variety of environmental stimuli. Epidemiological evidence suggests increased asthma prevalence in recent decades with reduced international differences in asthma prevalence [1]. The most common asthma phenotype is atopic asthma, accounting for 56% of asthma cases in the United States [2]. Atopic asthma is an immunoglobulin E mediated hypersensitivity reaction triggered by environmental allergens, such as endotoxin and aeroallergens [3]. Although environmental factors are important determinants of asthma, numerous studies have revealed that asthma has a strong genetic component. Susceptibility genes have been identified from linkage, candidate gene association, and genome-wide association studies. As of 2010, over 250 different genes have been associated with asthma, including cluster of differentiation 14 (CD14) [4,5].
A well studied common single nucleotide polymorphism (SNP) in the promoter region of CD14, -260C>T (rs2569190; also reported as CD14 -159), is the focus of this review. CD14 encodes a receptor protein that binds to lipopolysaccharide (LPS), its primary ligand, and interacts with co-receptors toll-like receptor 4 (TLR4) and lymphocyte antigen 96 (LY96). CD14 is expressed on the surface of monocytes, macrophages, and neutrophils as membrane CD14 and in the serum as soluble CD14 and its expression may be partially regulated at the genetic level [6]. LPS, a principle component of endotoxin, induces lung inflammation and originates from the outer membrane of Gram-negative bacteria. Ligand binding activates innate immune system pathways that may trigger atopic asthma [7]. Atopic asthmatic subjects are more sensitive to respirable endotoxin than non-asthmatic subjects [8] and also show increased expression of CD14 after acute allergen provocation [9] and LPS inhalation [10].
Two earlier meta-analyses found an overall null association between the CD14 -260C>T polymorphism and asthma, where no association was reported in some studies and the risk variant identified as either the T or C allele in others [11,12]. Unfortunately, these meta-analyses lacked adequate reporting of methodology and included studies examining non-asthma phenotypes. A more recent meta-analysis found a significant decreased atopic asthma risk for the TT and CT genotypes compared with the CC genotype when analysis was restricted to studies of Asian populations and children [13]. However, that review had several significant errors regarding study inclusion, data abstraction, and analyses.
Due to the inconsistency of past meta-analyses, an updated review was conducted to estimate the metaodds of developing asthma given the -260C>T genotype in CD14. Subgroup analyses were planned in order to explore potential sources of among-study heterogeneity by examining the effect of selected study characteristics on the combined effect estimate. Methodological issues in the literature studying this association are discussed.

Identification of eligible studies
Complete details of study methods are in Additional file 1. The review process followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14]. A PubMed, EMBASE, and Scopus search was conducted on April 29, 2011 using a sensitive strategy to identify relevant articles. The HuGE Literature Finder database was consulted for its listing of articles under the asthma phenotype and CD14. An article in press at time of search was added to the review [4]. Reference lists of articles retained for review and past meta-analyses were inspected for relevant publications. No publication date or language restrictions were imposed.
Article titles and abstracts of studies identified from the searches were screened and excluded from further analysis for the following reasons: ineligible phenotype, ineligible SNP, review article, basic science research, or animal research. The full-text of studies passing initial screening was reviewed and excluded based on the aforementioned and following criteria: not case-control or nested case-control study design, unreported genotype frequencies, or subjects included in another study.
Studies must have an asthma outcome definition that followed accepted diagnostic guidelines, was physician diagnosed, or used a combination of questionnaire and clinical ascertainment.
For multiple publications based on related data sets, the study with the greatest number of subjects was included. Reviewers extracted study information independently and disagreements were resolved by discussion and consensus.

Statistical analysis
The general approach to meta-analysis has been described previously [15,16]. The pooled frequency of the putative risk allele (-260T) was estimated in various ethnic groups using the inverse variance method. Heterogeneity of studies was assessed using the I 2 statistic [17] separately for the genotype-specific odds ratios (ORs) across studies: TT versus CC (OR 1 ), CT versus CC (OR 2 ), and TT versus CT (OR 3 ). If no or low heterogeneity existed (I 2 < 25%), the inverse variance method was used to estimate the pooled OR and 95% confidence interval (CI), assuming a fixed effects model. Otherwise, a random effects model was used. Comparisons of OR 1 , OR 2 , and OR 3 indicated the most appropriate genetic model for the -260T allele [16].
Subgroup analyses were planned when sufficient information was reported in at least four studies in each subgroup. The effect of having more homogeneous case and control phenotype definitions (atopic asthma versus non-atopic non-asthma), ethnicity, age, publication year, or study size on the association was examined to identify potential sources of heterogeneity. Post-hoc sensitivity analysis using the sequential algorithm [18] with an I 2 threshold of 25% was conducted in the presence of significant among-study heterogeneity to evaluate studies responsible for the heterogeneity. Influence analysis was conducted to allow identification of studies excessively perturbing the summary estimate. Publication bias was assessed visually using a funnel plot of the standard error of the logarithm of the effect estimate against the effect estimate of each study.
Review Manager Version 5.1.1 (Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Sweden) was used to conduct the meta-analysis, sequential analysis, and publication bias assessment. MetaAnalyst Version Beta 3.13 (Tufts Medical Center, Boston, MA) was used to estimate the pooled -260T allele frequency and conduct the influence analysis.

Study inclusion and characteristics
The literature search identified 204 potentially relevant articles. Initial screening of titles and abstracts excluded 159 studies which did not meet the eligibility criteria.
T and asthma risk",1,0,1,0,0pc,0pc,0pc,0pc>Association between CD14 -260C>T and asthma risk The pooled ORs for each pair-wise genotype comparison and corresponding I 2 statistics are summarized in Table 2. For all studies, heterogeneity ranged from moderate to high for the non-significant genotype-specific ORs, suggesting no association between the polymorphism and asthma risk. Subgroup analyses (data not shown) did not show significant gene effects when studies were subset by ethnicity (European or East Asian), age range of cases and controls (adults or children), year of study publication (2006-2010 or 2001-2005), and genotyped study sample size (≥ 100 cases and ≥ 100 controls or < 100 cases or < 100 controls). Low to moderate among-study heterogeneity was present in all subgroups for OR 2 and moderate to high heterogeneity for OR 1 and OR 3 . Sensitivity analysis excluding studies that appeared to account for appreciable heterogeneity and influence did not meaningfully change the results for overall and subgroup meta-analyses (data not shown). Relatively symmetrical funnel plots indicated the absence of publication bias for the genotype-specific ORs (see Figures S1-S3, Additional files 3, 4 and 5).

Subgroup analysis by case-control phenotype definitions
Initial subgroup analysis of studies that had defined case-control phenotypes as atopic asthma and non- The number of successfully genotyped cases and controls may be less than the total number of cases and controls in the study (i.e. SNP call rate < 100%). Genotype frequencies presented as reported, otherwise calculated from reported genotype percent frequencies. Abbreviations: AA, atopic asthma; CC, casecontrol; HWE, Hardy-Weinberg equilibrium; MA, mixed asthma; N, genotyped sample size; NA, non-asthma; NAA, non-atopic asthma; NANA, non-atopic nonasthma; NCC, nested case-control. a Genotype frequencies estimated based on allele frequencies assuming HWE among cases and controls. b Genotype distribution for AA cases shown (genotype distribution for NAA cases not shown). c Genotype distribution for NA controls shown (genotype distribution for NANA controls not shown). d Genotype frequency information from this data set for atopic asthma cases and corresponding controls (not shown) abstracted from Leung et al. [21].
atopic non-asthma showed non-significant gene effects plus high among-study heterogeneity ( Table 2). Further subgrouping of studies comparing atopic asthmatics and non-atopic non-asthmatics by ethnicity (European only), age (children only), year of study publication, and genotyped study sample size did not meaningfully change the results. Post-hoc sensitivity analysis identified three studies that may be responsible for the significant amongstudy heterogeneity: Kedda et al. [11], de Faria et al. [27], and Lachheb et al. [42]. While reported study characteristics for these three studies were not atypical compared to other studies, the controls for the studies by de Faria et al. and Lachheb et al. deviated significantly from HWE (p < 0.01). Influence analysis found moderate influence on the combined effects exerted by these three studies.
The genotype-specific ORs for the subgroup of studies with atopic asthma versus non-atopic non-asthma casecontrol groups, excluding the three studies identified by the post-hoc sequential analysis, implied a codominant model ( Table 2). Compared to subjects with the CC genotype, the pooled ORs suggested that subjects with the TT genotype were some 33% less likely to have atopic asthma (OR 1 = 0.67, 95% CI: 0.54-0.84, I 2 = 10%) ( Figure 2) and subjects with the CT genotype were about 20% less likely to have atopic asthma (OR 2 = 0.80, 95% CI: 0.66-0.95, I 2 = 0%) (Figure 3), showing a dose-response relationship for the T allele. No substantial heterogeneity was detected and publication bias was not evident in the funnel plots (see Figures S4 and S5, Additional files 6 and 7). Exclusion of any one particular study in the influence analysis did not meaningfully change the results (data not shown).

Discussion
The present meta-analysis found a non-significant association between the CD14 -260C>T polymorphism and overall asthma. There was also high among-study heterogeneity in the meta-analysis, possibly accounting for the inconsistently reported findings between this SNP and asthma [43]. Subgroup analysis of selected study characteristics did not reveal any significant associations or substantial decreases in the I 2 estimate of heterogeneity. When restricting analysis to studies that used atopic asthma versus non-atopic non-asthma case-control phenotypes and excluding studies influencing heterogeneity, the genotype-specific ORs suggested a codominant model.
A sequential analysis revealed three studies that appeared to account for the high among-study heterogeneity (see Additional file 1 for methodology). Two had controls that departed from HWE, which may represent possible sources of bias. The exploratory nature of posthoc sequential analysis may present a weakness, but Abbreviations: AA, atopic asthma; NAA, non-atopic asthma; NANA, non-atopic non-asthma; NS, non-significant; OR, odds ratio. a Guideline for interpretation of the I 2 statistic: I 2 = 0% no heterogeneity, I 2 = 25% low heterogeneity, I 2 = 50% moderate heterogeneity, and I 2 = 75% high heterogeneity [17]. b Excluding studies with excessive contribution to among-study heterogeneity identified by post-hoc sequential analysis [11,27,42]. c Including the in press article at time of search by Murk et al. [4]. d Numbers include only genotyped cases and controls.
advantages include its objective approach and the fact that specific study characteristics that may contribute to heterogeneity are not always known or recorded. The latter is important: if various methodological nuances are not reported, subsequent meta-analysis would not account for these factors and the ability to assess sources of heterogeneity would be hampered. For example, reported study characteristics in the article by Kedda et al. [11], one of the studies identified to incur a large amount of heterogeneity, did not reveal any particular characteristic that deviated from other studies. Stronger associations and significant relationships were found when analysis was restricted to studies with more homogeneously defined case-control phenotypes Figure 2 Forest plot of CD14 -260 (-159) TT versus CC genotypes for studies with precisely defined phenotypes. The forest plot displays the meta-analysis results of studies included in the review that used atopic asthma versus non-atopic non-asthma case-control phenotypes, excluding heterogeneous studies identified by sequential analysis [11,27,42]. Meta-analysis was conducted using an inverse variance (IV), fixed effects model. For each study in the forest plot, the area of the black square is proportional to study weight and the horizontal bar represents the 95% confidence interval (CI). Atopic asthma and non-atopic non-asthma are abbreviated as AA and NANA, respectively.

Figure 3
Forest plot of CD14 -260 (-159) CT versus CC genotypes for studies with precisely defined phenotypes. The forest plot displays the meta-analysis results of studies included in the review that used atopic asthma versus non-atopic non-asthma case-control phenotypes, excluding heterogeneous studies identified by sequential analysis [11,27,42]. Meta-analysis was conducted using an inverse variance (IV), fixed effects model. For each study in the forest plot, the area of the black square is proportional to study weight and the horizontal bar represents the 95% confidence interval (CI). Atopic asthma and non-atopic non-asthma are abbreviated as AA and NANA, respectively. and with heterogeneous studies excluded. These results indicated that the -260T allele was significantly protective under the codominant model when comparing atopic asthmatics to non-atopic non-asthmatics. Observed among-study heterogeneity may be partially explained by the employment of overly broad case-control phenotype definitions. It has been suggested in genome-wide association studies that use of homogeneous case phenotypes and precisely specified control groups-those who unambiguously do not have the case phenotypemay improve study efficiency [44]. This principle, borrowed from extreme discordant sib-pair analysis [45], naturally extends to case-control selection in candidate gene association studies.
There is possible gene-environment interaction, in which the SNP acts as a modifier of asthma risk in individuals with different degrees of environmental endotoxin exposure. Carriers of the TT genotype have been found to have higher serum levels of CD14 than carriers of the CT or CC genotypes [43]. This epidemiologic evidence is supported by functional genomic studies that showed increased transcriptional activity of the -260T allele in a monocytic cell line [46]. An antagonistic interaction has been demonstrated between CD14 and endotoxin exposure: homozygotes for the T allele appear to be protective for asthma at low levels of endotoxin exposure, but may increase asthma risk at high levels of endotoxin exposure [43]. Based on these findings, Martinez [43] hypothesized that higher CD14 expression in TT homozygotes increased sensitivity to the protective effects of low level endotoxin exposure compared to carriers of other genotypes. However, at higher levels of endotoxin exposure, induced CD14 expression could be increased in carriers of the C allele, showing a reversed protective effect. The findings of the present meta-analysis, restricted to studies using the atopic asthma versus non-atopic non-asthma case-control phenotypes, are consistent with this hypothesis at low endotoxin exposure levels. The codominant model for the -260T allele implied a dose-response relationship in CD14 expression and reduction of atopic asthma risk. This gene-environment interaction may be a source of heterogeneity among studies in the present and earlier meta-analyses [11][12][13].
In addition to the promoter, many additional regulatory elements are necessary to influence gene expression, particularly for genes like CD14, which exhibit highly complex expression patterns. Regulatory elements, such as enhancers and repressors, may reside in intronic regions or up-and down-stream of the transcriptional unit [47]. A risk variant with no obvious and no known function may regulate a gene at a considerable genomic distance from the location of the SNP. Therefore, it is important to study the influence of gene-gene interaction as well as other polymorphisms in CD14 on the effects of this locus on asthma susceptibility.

Quality and methodology of studies
Assessing study quality was difficult due to inadequate reporting from all studies included in the meta-analysis. Many studies reported insufficient information about recruitment methodology and study participant characteristics, particularly for controls. Genotype distributions of controls departed from HWE in three studies [27,33,42]. Deviation from HWE in controls, or healthy populations, may indicate selection bias, population stratification, or genotyping errors [48]. Even in the absence of deviation from HWE, these biases could not be assessed given the inadequately reported information. Eight studies reported implementing some form of genotyping quality control [4,20,28,30,37,38,40,41]. Only two published studies mentioned blinding of phenotype when genotyping [29,38]. Furthermore, there is a potential for publication bias, where positive rather than negative findings tend to be published [49]. The completeness of evidence is also impeded by language bias. Studies conducted in non-English speaking countries tend to publish significant results in international journals and non-significant results in local journals, many of which are not indexed [50]. Selective publication of polymorphism and disease associations may obscure their true relationships.
Results from the pooled CD14 -260T allele frequency in controls revealed differences among the broad ethnic categories: 0.457 for European populations, 0.577 for East Asian populations, and 0.621 for an Indian population. In comparison, the International HapMap Project (Phase 3) reported the -260T allele frequency among Utah residents with Northern and Western European ancestry, Han Chinese in Beijing, China, Japanese in Tokyo, Japan, and Yoruba in Ibadan, Nigeria to be 0.474, 0.500, 0.488, and 0.293, respectively. The average heterozygosity reported in Build 132 of dbSNP is 0.488 ± 0.078 [51]. Interethnic differences in the allele frequencies of the CD14 -260C>T polymorphism is of concern as some studies included in this meta-analysis have different ethnic compositions between the cases and controls. Reported associations in studies of varying ethnic composition may have been influenced by population stratification. Even among apparently homogeneous ethnic groups, population stratification may be a problem [52]. The effect of this type of stratification has been reported to be small in most situations, but a small bias may be important in studies of genetic association, which typically consider small or moderate effects [53]. Only four studies included in this review reported an assessment of population stratification.
A commonly cited solution to addressing population stratification is the use of family-based designs to study genetic associations [44,54]. However, the family-based design has its own inherent limitation to susceptibility variant discovery. It has been argued that neither common nor rare genetic variants are heritable, as they do not give rise to a substantial familial concentration of cases due to low penetrance [55]. Three family-based studies have explored the association of CD14 -260C>T and asthma with conflicting results [12,41,42]. Therefore, efforts should be made to accrue controls from the same source population as cases to avoid population stratification, particularly when ethnicity is not matched or controlled [44].

Conclusions
This meta-analysis provides a comprehensive examination of the available evidence concerning the association between the CD14 -260C>T polymorphism and asthma susceptibility. The significant association between this polymorphism and atopic asthma may be of clinical and public health importance. The genetics of asthma follow the "common disease, common variants" hypothesis, which posits that multiple genetic variants of interest are common to many individuals with the disease. These common variants typically have weak individual effects and low penetrance, but their high frequency confers a relatively large attributable risk in the population. Therefore, this common polymorphism, along with endotoxin exposure level information, has potential to be a useful and efficient predictor of atopic asthma risk. This review also emphasizes the importance of having precisely defined case-control groups to study complex diseases and demonstrates the need to incorporate gene-environment and gene-gene interaction analyses in future epidemiological investigations of asthma genetics.

Additional material
Additional file 1: Supplemental methods. Complete details of the study methodology.
Additional file 2: Table S1. Summary of abstracted characteristics of reviewed studies on CD14 -260 (-159) C>T and asthma. Complete summary of abstracted characteristics from studies included in the systematic review and meta-analysis.
Additional file 3: Figure S1. Funnel plot of CD14 -260 (-159) TT versus CC genotypes for all reviewed studies. Standard error of the logarithm of the odds ratio (SE(log[OR])) was plotted against the OR of each study.
Additional file 4: Figure S2. Funnel plot of CD14 -260 (-159) CT versus CC genotypes for all reviewed studies. Standard error of the logarithm of the odds ratio (SE(log[OR])) was plotted against the OR of each study.
Additional file 5: Figure S3. Funnel plot of CD14 -260 (-159) TT versus CT genotypes for all reviewed studies. Standard error of the logarithm of the odds ratio (SE(log[OR])) was plotted against the OR of each study.
Additional file 6: Figure S4. Funnel plot of CD14 -260 (-159) TT versus CC genotypes for studies with precisely defined phenotypes. The funnel plot displays studies included in the review that used atopic asthma cases and non-atopic non-asthmatic controls, excluding heterogeneous studies identified by sequential analysis [11,27,42]. Standard error of the logarithm of the odds ratio (SE(log[OR])) was plotted against the OR of each study. Additional file 7: Figure S5. Funnel plot of CD14 -260 (-159) CT versus CC genotypes for studies with precisely defined phenotypes. The funnel plot displays studies included in the review that used atopic asthma cases and non-atopic non-asthmatic controls, excluding heterogeneous studies identified by sequential analysis [11,27,42]. Standard error of the logarithm of the odds ratio (SE(log[OR])) was plotted against the OR of each study.