The Role of TLR4 896 A>G and 1196 C>T in Susceptibility to Infections: A Review and Meta-Analysis of Genetic Association Studies

Background Toll-like receptor 4 plays a role in pathogen recognition, and common polymorphisms may alter host susceptibility to infectious diseases. Purpose To review the association of two common polymorphisms (TLR4 896A>G and TLR4 1196C>T) with infectious diseases. Data Sources We searched PubMed and EMBASE up to March 2013 for pertinent literature in English, and complemented search with references lists of eligible studies. Study Selection We included all studies that: reported an infectious outcome; had a case-control design and reported the TLR4 896A>G and/or TLR4 1196C>T genotype frequencies; 59 studies fulfilled these criteria and were analyzed. Data Extraction Two authors independently extracted study data. Data Synthesis The generalized odds ratio metric (ORG) was used to quantify the impact of TLR4 variants on disease susceptibility. A meta-analysis was undertaken for outcomes reported in >1 study. Eleven of 37 distinct outcomes were significant. TLR4 896 A>G increased risk for all parasitic infections (ORG 1.59; 95%CI 1.05-2.42), malaria (1.31; 95%CI 1.04-1.66), brucellosis (2.66; 95%CI 1.66-4.27), cutaneous leishmaniasis (7.22; 95%CI 1.91-27.29), neurocysticercosis (4.39; 95%CI 2.53-7.61), Streptococcus pyogenes tonsillar disease (2.93; 95%CI 1.24-6.93) , typhoid fever (2.51; 95%CI 1.18-5.34) and adult urinary tract infections (1.98; 95%CI 1.04-3.98), but was protective for leprosy (0.36; 95%CI 0.22-0.60). TLR4 1196 C>T effects were similar to TLR4 896 A>G for brucellosis, cutaneous leishmaniasis, leprosy, typhoid fever and S. pyogenes tonsillar disease, and was protective for bacterial vaginosis in pregnancy (0.55; 95%CI 0.31-0.98) and Haemophilus influenzae tonsillar disease (0.42; 95%CI 0.17-1.00). The majority of significant associations were among predominantly Asian populations and significant associations were rare among European populations. Conclusions Depending on the type of infection and population, TLR4 polymorphisms are associated with increased, decreased or no difference in infectious disease. This may be due to differential functional expression of TLR4, the co-segregation of TLR4 variants or a favorable inflammatory response.


Introduction
Toll-like receptors (TLRs) are a class of highly conserved membrane bound pattern recognition receptors (PRRs) that play an integral role in the regulation of the immune system through the recognition of pathogen-associated molecular patterns (PAMPs) and the activation of immune response genes [1,2]. Toll-like receptor 4 (TLR4), is a well-studied TLR, specifically recognizing lipopolysaccharide from Gram-negative bacteria [3,4] and initiating intracellular signal cascades, that involve the adaptor protein encoded by the myeloid differentiation primary response gene 88 (MyD88), which ultimately activates nuclear factor kappa B [5] and leads to interferon production [6]. TLR4 has also been shown to recognize mannans of fungal pathogens [7], Mycobacterium tuberculosis [8], and the fusion protein of respiratory syncytial virus [9].
Two single nucleotide polymorphisms (SNPs), TLR4 896 A>G (corresponding to an Asp299Gly substitution mutation ; SNP ID: rs 4986790) and TLR4 1196 C>T (corresponding to a Thr399Ile substitution mutation; SNP ID: rs 4986791), have been shown to be associated with LPS hyporesponsiveness [10,11]. In whites, the two SNPs are in linkage disequilibrium (D=1 and r2=0.791, HapMap accessible at: http:// hapmap.ncbi.nlm.nih.gov/). Structurally, these mutations are found outside of the ligand binding domain of TLR4 and crystal structures have shown that these mutations have no effect on LPS binding. Instead, they do cause local conformational changes around the area of the mutation that may affect folding efficiency, cell surface expression, protein stability, as well as interaction with downstream messenger proteins [12]. At the molecular level, it has been shown that the TLR4 896 A>G mutation interferes with TLR4 interaction with MyD88 and other downstream messengers [13]. These mutations also appear to affect the levels of functional TLR4 expression, leading to a 2fold reduction [14]. This reduction is further amplified to 10-fold in the absence of myeloid differentiation factor 2 (MD-2) which forms a complex with TLR4 and LPS [14,15].
There has been great interest regarding the association of the TLR4 SNPs TLR4 896 A>G and TLR4 1196 C>T to susceptibility for infection and other non-infectious disease states. Clinical studies associating these SNPs to infectious disease susceptibility have produced mixed results [16][17][18][19]. The present study aims to reassess the association of TLR4 896 A>G and TLR4 1196 C>T with infectious disease susceptibility using the Generalized Odds Ratio (OR G ), which can elucidate the magnitude and association of individual genotypes with susceptibility to disease [20].

Study Selection
We conducted searches on Pubmed and EMBASE up to March, 2013 (last access on March 3, 2013). The search terms included: "(toll AND like AND receptor AND 4 AND polymorphism) OR (TLR4 AND polymorphism) OR Asp299Gly OR D299G OR Thr399Ile OR T399I" for PubMED; "('tlr4'/exp OR 'tlr4') AND ('receptor'/exp OR 'receptor') AND (polymorphism OR asp299gly OR d299g OR thr399ile OR t399i)" for EMBASE. The titles and abstracts of the studies were reviewed; titles that included TLR4 polymorphisms and risk for infectious disease were included for more detailed evaluation. Studies that reviewed TLR4 polymorphisms and their association with non-infectious disease were excluded, as were studies that were not published in English. An eligible study fulfilled all of the following three criteria: (i) the study reported an infectious disease outcome, (ii) the study was performed using a case-control design, where "cases" refer to subjects with a disease outcome and controls refer to a healthy population (without the disease outcome), and, (iii) the study reported genotype frequencies for TLR4 896 A>C, TLR4 1196C>T, or both.

Data Extraction
Two authors (PDZ and MLP) independently extracted data from the final included articles. Any discrepancies were reviewed and resolved by consensus. The information extracted included name of first author, origin of population being studied, number of cases and controls being studied subdivided by genotype frequencies (homozygous wild-type, heterozygous, and homozygous mutant), the disease being studied, and the conclusions reportedly drawn from each study.

Data Synthesis
We used the generalized odds ratio (OR G ) along with its 95% Confidence Interval (95% CI) to address the association of TLR4 896 A>C and TLR4 1196 C>T polymorphisms with outcomes of interest (disease susceptibility). The OR G provides a model-free approach of estimating the genetic risk in genetic association studies (GAS) and meta-analysis of GAS, depending on the mutational load [20]. The OR G is defined as follows: for any two subjects, one diseased (case) and one non-diseased (control), the OR G estimates the odds of being diseased relative to the odds of being non-diseased when the diseased subject has higher mutational load than the nondiseased subject, i.e. the risk of disease is proportional to the increased genetic exposure. Alternatively, the OR G shows how many diseased-healthy pairs exist in the study for which the diseased have the larger mutational load, relative to the number of pairs for which the non-diseased have the larger mutational load [20] [21]. The OR G estimates the overall genetic risk effect by utilizing the complete genotype distribution whereas the OR of conventional genetic models (additive, dominant, recessive, co-dominant) is calculated by merging genotypes. In addition, the conventional genetic models are not independent and thus, the interpretation of results is difficult when more than one model is significant [22]. In the metaanalysis of GAS, heterogeneity was quantified using the Cochran's Q and I 2 metric [20]. The existence of the differential magnitude of effect in large versus small studies was checked using the Harbord's test [23] for meta-analysis involving at least four studies. Also, the Hardy-Weinberg equilibrium (HWE) was used as a quality criterion for control populations. HWE deviations may result in biased estimations as they can influence type-I error in single study effects, and may alter statistical significance in meta-analysis of gene-disease TLR4 Polymorphisms and Infectious Diseases PLOS ONE | www.plosone.org associations [24,25]. The HWE deviations amongst the control populations were screened using the chi-square test [26]. For single studies deviating from HWE, a sensitivity analysis was performed after correction of control group with the expected genotype frequencies [22,27]. ORG was calculated using the ORGGASMA application available at http://biomath.med.uth.gr [20]. This study complies with the PRISMA guidelines for reporting reviews and meta-analyses (Checklist S1) [28].

Results
A total of 962 studies from PubMed and 1615 from EMBASE were initially retrieved, comprising a total of 2,197 nonduplicate studies (Table S1-Flow diagram). After reading the title and the abstract, 117 studies were found to be suitable for further evaluation. Of the 117 articles reviewed in detail, 58 studies were excluded (18 studies did not publish genotypic frequencies, 13 had no healthy controls in their experimental design, 5 focused on in vitro functional studies, 8 did not study the desired polymorphisms, 6 were either reviews or a metaanalysis, and 8 studies had non extractable data for other reasons). A total of 59 case-control studies  were included in the analysis, reporting 37 different disease outcomes (Tables 1 and 2). The origin of studies was in descending order Europe (28 studies), Asia (12 studies), South America (7 studies), Africa (6 studies), North America (5 studies), Australia (1 studies).
Of note, all these effects were derived from Asian studies. Increased risk for tonsillar infection due to Streptococcus pyogenes (OR G 2.93; 95% CI 1.24-6.93) [46] was noted in the Greek pediatric population, as was an increased risk for urinary tract infections in adults (OR G 1.98; 95% CI 1.04-3.98) in a Chinese population [48]. Interestingly, not all outcomes were negative and the TLR4 896 A>G polymorphism was associated with significant protection against leprosy (OR G 0.36; 95% CI 0.22-0.60) in East Africa [38].
The use of the OR G metric resulted in more conservative estimates of associations, as two reportedly significant associations (1 reporting increased risk for Gram-negative osteomyelitis [41] and 1 reporting a protective effect for Streptococcus pneumoniae in children [45]) were downgraded to non-significant. Six control populations deviated for HWE equilibrium [30,35,39,53,58,62], and associations of TLR4 variants with disease were readdressed after correcting genotypes with their expected frequencies. These effects did not change (they appear in brackets in Tables 1,2). Specifically, the association of TLR4 896 A>G and brucellosis [30] remained significant after HWE correction (OR G 2.69; 95% CI 1.67-4.33).

TLR4 1196 C>T and disease susceptibility
A meta-analysis of GAS was performed for malaria (2 studies) [55,56] Table 1. Genotypic frequencies reported for the TLR4 896 A>G SNP and association with disease outcome; significant effects are in bold; outcomes that have been studied more than once have been grouped together in the table, with the overall effect described in the shaded area † genotypic frequencies of controls that did not satisfy Hardy Weinberg Equilibrium, [effects in brackets after correction of HWE deviations].   Table 2. Genotypic frequencies reported for the TLR4 1196 C>TSNP and association with disease outcome; significant effects are in bold; outcomes that have been studied more than once have been grouped together in the table, with the overall effect described in the shaded area.

Name Population C/C C/T T/T C/C C/T T/T Disease Outcome Conclusion Reported OR G (95% CI)
Goepfert et al [86] United African Americans comprised 78% of the cases) [86], leprosy in East Africa (OR G 0.23; 95% CI 0.10-0.55) [38], and Haemophilus influenzae tonsillar disease in a Greek pediatric population (OR G 0.42; 95% CI 0.17-1.00) [46]. The significant results are summarized in Table 4. Only 1 control population deviated from HWE equilibrium that assessed the risk of meliodosis [39], a risk that did not change after correction with the expected genotype frequencies ( Table 2). Two reportedly significant associations for Gram-negative osteomyelitis (increased risk) and S. pneumoniae (protection) were not confirmed in this analysis with the use of the OR G metric. The significant effects were unidirectional and similar in magnitude when both TLR4 896 A>G and 1196 C>T were examined (

Discussion
We performed a systematic literature review to address the potential association of 2 common TLR4 single nucleotide polymorphisms (TLR4 896 A>G, TLR4 1196 C>T) with infectious diseases. An increased risk was documented for all parasitic infections combined, malaria [54][55][56], brucellosis [30], cutaneous leishmaniasis [36], typhoid fever [47], neurocysticercosis [40] and adult urinary tract infections [48]. Interestingly, all these effects were reported in populations of Asian descent, with the exception of parasitic infections and malaria where the effect was a combined effect from Asian,

Control Genotype
Case Genotype African and South American populations. This finding is more striking when we consider that European populations comprised the majority of GAS data (28 out of 59 studies, 48%) and a significant risk was found only for TLR4 polymorphisms and S. pyogenes tonsillitis among Greek children [46]. Another notable finding is that, for some infections, these single nucleotide polymorphisms were associated with lower infection rates. Overall, these effects sum to a total of 11 significant SNPs-disease associations that represent almost one third (30%) of all outcomes addressed in the eligible studies and there was consistency of effects (risk or protection) between 896 A>G and 1196 C>T variants when both associations were studied.

Name Population C/C C/T T/T C/C C/T T/T Disease Outcome
In this study we utilized the generalized odds ratio (OR G ) metric to quantify the magnitude of associations. This metric provides a straightforward interpretation of the relative risk effect, based solely on genotype distribution [20]. The generalized odds ratio overcomes this problem by directly quantifying the magnitude of association of a gene with disease [20]. Implementing the OR G obviates the need for selecting, estimating and interpreting individual genotype contrasts (dominant, recessive and co-dominant) and their effect. OR G can also be used in meta-analysis of GAS to summarize effects and produce robust results, avoiding the shortcomings of multiple model testing, namely the lack of biologic justification and non-independency of effects [20,88,89]. For example, for TLR4 896 A>G association with malaria, the combined OR G showed that the probability of having malaria might be 31% higher for subjects having higher mutational load relative to those with lower mutational load (subjects who are homozygous for G allele have the highest mutational load, those homozygous for A allele have the lowest, and heterozygous have an intermediate level). The application of the OR G metric also resulted in a more conservative estimate of associations, given that associations for infections such as osteomyelitis (39) and S. pneumonia (43) were downgraded to insignificant. The associations derived from tuberculosis data were insignificant similar to those reported [90].
In our analysis, TLR4 polymorphisms were associated with susceptibility to a diverse spectrum of infections including Gram-negative, Gram-positive bacteria as well as parasitic infections, such as cutaneous leishmaniasis and neurocysticercosis. This wide spectrum of associations correlates with the spectrum of recognition molecules by TLR4. Indeed, TLR4 is involved in induction of cell-mediated immunity to Brucella abortus in mice [7] and TLR4 signaling also upregulates macrophage anti-leishmanial activity [91]. Similarly, binding of the Salmonella typhi porin OmpS1 to TLR4 leads to overexpression of MHCII and CD40 molecules and activation of dendritic cells [92]. TLR4 can recognize LPS of Gram-negative bacteria [3,4], glycans of the helminth Taenia solium [93] as well as the fusion protein of respiratory syncytial virus [9].
Interestingly, our analysis also confirmed that these polymorphisms are also protective for certain types of infection, such as leprosy. It is not clear why such polymorphisms confer increased susceptibility to some infection, but protect from others. It could be speculated that in some infections the immune response leads to an inflammatory response that is protective, whereas in others such response may be essential in the pathogenesis of the infectious process. An example is Mycobacterium leprae where the TLR4-mediated immune response to the pathogen may modulate inflammatory processes that influence disease manifestations but are not attributable to direct stimulation by M. leprae. Indeed, Bochud et al [38] found that the stimulation of monocytes with M. leprae inhibited their subsequent response to TLR4 stimulation with LPS. Among Indo-European populations, 6-14% of the individuals are double heterozygous for both polymorphisms [94]. It is suggested that the double heterozygous TLR4 896 A>G/TLR4 1196 C>T haplotype does not functionally differ from wild type TLR4. Therefore, co-segregation may result in a functionally neutral phenotype and, as seen in European populations, lead to the lack of significant associations. Conversely, TLR4 896 A>G was frequently found (10-18%) among African populations, with only 2% having TLR4 1196 C>T cosegregation. Two studies (on typhoid fever and leprosy) indicated weak linkage disequilibrium in Malaysian [47] and East African populations [38]. These differences between Europeans (co-segregation) compared to Asian and African population (lack of co-segregation) may explain why the majority of significant associations were noted for endemic diseases of Asia and Africa.
Our analysis on the impact of these polymorphisms in periodontitis illustrates the different impact of polymorphisms based on the population. More specifically, despite the bulk of studies on aggressive and chronic periodontitis, TLR4 variants did not show any significant association, even though TLR4 has been shown to be overexpressed in gingival epithelial cells and gingival fibroblasts [95][96][97] in association with periodontal inflammation involving pathogens related to periodontitis, such as Porphyromonas gingivalis, Fusobacterium nucleatum and Aggregatibacter actinomycetemcomitans [98][99][100][101]. One possible explanation is that this finding was because all relevant studies were almost exclusively confined to European ancestry populations and the lack of susceptibility may be related to the strong linkage disequilibrium, that is the nonrandom association between 896 A>G and 1196 C>T in Europeans [94]. Importantly, our analysis highlights the need to evaluate the impact of these polymorphisms in different populations and various clinical conditions. Moreover, the absence of significant associations in meta-analysis data for periodontitis, tuberculosis, meningococcal disease and sepsis, signifies that the functional alterations related to polymorphic TLR4 variants may not be critical to produce the clinical phenotype. Lack of reproducibility stands as a barrier for conclusive evidence, and design, sample size and environmental and genetic heterogeneity between populations may affect results. Finally, the presence of a significant effect may rely on the magnitude of functional expression of TLR4. Protection or risk may be moderated by the level of TLR4 functional expression, which is modulated by TLR4 polymorphism and MD-2 presence [14,15]. Therefore, it is essential to explore whether MD-2 is important in the response to some infections, but not others, or that levels of TLR4 vary in one infection compared to another.
The heterogeneity of the populations studied along with multiple endpoints should also be considered as potential study limitation that may influence statistical power. Moreover, different populations mount diverse immunologic responses and the clinical relevance of polymorphisms is not always straightforward. The lack of association for a disease phenotype highlights that gene-to-gene interactions and geneenvironment interactions may be influential parameters of disease association. Case-control design of individual GAS precludes adjusted analysis for gene-gene-environment interactions and may have reduced the efficiency of genetic risk estimates, though it is unlikely to inflate false-positive results [89].
Despite these limitations, genetic markers of immune response such as TLR4 variants, are valuable not only to classify high-risk patients based on disease susceptibility but also to predict disease severity and other sequelae. The associations of TLR4 896A>G with hearing loss in survivors of bacterial meningitis [102] and the increased risk of tympanostomy among toddlers with history of bronchiolitis [103] are indicative examples.
In conclusion, our analysis highlights the complex effect of TLR variants in susceptibility to infectious disease. Some of the effects, such as in malaria, are validated in a variety of studies, whereas single case-control studies should be cautiously interpreted until more information on the specific outcomes is added. Taken in their totality, our results indicate that depending on the infection and the population studied, the same polymorphism may be associated with risk, protection or have no effect. In this context, our analysis provides the rationale for understanding the protective or adverse effect of TLR4 Polymorphisms and Infectious Diseases PLOS ONE | www.plosone.org TLR4 polymorphisms and may provide a basis to explain the maintenance of these polymorphisms.

Author Contributions
Analyzed the data: PDZ MLP. Contributed reagents/materials/ analysis tools: EZ. Wrote the manuscript: PDZ MLP JK EZ EM. Search strategy, data selection, data extraction: PDZ MLP.