Introduction

The Gram-negative bacteria cell-wall component endotoxin is a potent instigator of the inflammatory response. Lipopolysaccharide binding protein (LBP) forms complexes with the endotoxin, and these complexes are detected by the endotoxin receptor complex consisting of Toll-like receptor 4 (TLR4) and the proteins CD14 and myeloid differentiation 2 (MD2). Functional TLR4 has been described on many human cell types, including monocytes/macrophages, endothelium and platelets [1–3]. Monocyte TLR4 ligation initiates a cascade of intracellular signalling events that lead to gene transcription and the ensuing production of inflammatory mediators [Electronic Supplementary Material (ESM) Fig. E1A]. Platelet endotoxin binding is less well understood, but the mechanism includes soluble CD14 and CD62P (P-Selectin) in addition to TLR4 (ESM Fig. E1B) and contributes to thrombocytopenia [4] and neutrophil bactericidal action [5]. TLR4 has also been implicated in an endotoxin-independent response to injury, ligating fibrinogen and heat shock proteins [6].

A reduced sensitivity of the innate immune system’s capacity to recognise and respond to pathogen-associated molecular patterns has been associated with an increased frequency or severity of systemic inflammation in some studies of critically ill patients [7–9].

Two common single nucleotide polymorphisms (SNPs) exist in the human TLR4 gene: D299G and T399I. Both lie in the extra-cellular domain and are thought to reduce the effectiveness of endotoxin binding and, consequently, result in a decreased sensitivity to endotoxin [10]. The functionality of these SNPs has been observed in human studies [11] and also contested [12].

To date, only a few clinical studies have obtained convincing data that the TLR4 SNPs D299G and T399I influence the course of critical illness in general [13]. The early clinical studies that did show a difference in clinical outcome were hampered by small sample sizes (n = 77 [14], n = 159 [15], n = 91 [16]) and questionable choices in statistics [16]. Recently, more specific associations have been found between TLR4 polymorphisms and an increased vulnerability to bacteraemia in critically ill patients [17] and aspergillosis in stem cell transplantation patients [18]. The observed inconsistencies may simply reflect a lack of functional relevance of these SNPs and/or deficiencies in study design; alternatively, they may indicate that any signal from altered endotoxin recognition is drowned in the noise of other elements of host variability and environmental determinants in the acute inflammatory response. Further, the limitations of systemic inflammatory response syndrome (SIRS) as an endpoint have long-been recognised in adults [19]. This concern is compounded in the paediatric age group by the variability in the age-related normal values of the components that define SIRS (heart rate, respiratory rate and white cell count). We therefore chose a second marker of severe inflammation, namely, platelet count on admission.

Platelets actively contribute to systemic inflammation [20]. In the clinical setting, platelet count has been recognised as a sensitive and near-continuous marker of severity of critical illness in both adults [21] and children [22, 23]. Furthermore, endotoxin has been shown to directly reduce platelet count [24] to which TLR4 activity may be relevant, as shown in both mice [4] and humans [5].

We have studied 913 critically ill children to determine the relationship between altered endotoxin recognition, as identified by TLR4 SNPs, and the development of SIRS as a clinically relevant endpoint related to monocyte/macrophage activation, and platelet count on admission as an alternative marker of the acute inflammatory response. Potential confounders were assessed, including major environmental factors and other relevant SNPs in the endotoxin recognition complex and innate immune response. To validate our findings, we subsequently assessed the relationship between the TLR4 polymorphisms and platelet count on admission in a cohort of 1,170 adults with atherosclerosis. This is a well-described previously genotyped cohort with a severe inflammatory insult [25].

Methods

The Great Ormond Street Hospital for Children NHS Trust/Institute of Child Health and Southampton & South West Hampshire Local Research Ethics Committees approved this study. Parental or subject informed consent was obtained at all sites according to the Declaration of Helsinki guidelines. Clinical parameters were collected while the investigators were blinded to the genotype data and vice versa.

Subjects

Between 2000 and 2006 children were recruited consecutively in three recruiting time-periods from three tertiary paediatric intensive care units. Inclusion and exclusion criteria have been documented previously [8].

Subjects were divided by primary diagnosis into ‘Bypass’, ‘Infection’ and ‘Non-infection’ groups. Briefly, children who were admitted for heart surgery on cardiopulmonary bypass were assigned to ‘Bypass’; those children who were diagnosed with an infectious process by the admitting physician were assigned ‘Infection’; the category ‘Non-infection’ was used for all other admission diagnoses, mainly trauma and elective surgery (see ESM for details).

Identification of polymorphisms

A literature search performed to identify functional common SNPs in the endotoxin recognition complex resulted in the identification of TLR4 D299G and TLR4 T399I and the CD14 promoter −159C > T. LBP P97P was included as a common SNP of unknown functionality that may influence outcome from sepsis. No common SNPs were identified in MD2.

Possible confounding SNPs in the following genes and/or promoters are listed in Table 1: mannose binding lectin (MBL-2), cholinergic receptor, nicotinic, alpha 7 (CHRNA7), interleukin (IL)-6, IL-10, tumor necrosis factor alpha (TNFα) and human plasminogen activator inhibitor-1 (PAI-1). These SNPs have either been shown to have an effect on the acute inflammatory response or are thought to have modifier effects on TLR4 activity (see ESM for details on rationale and genotyping methods).

Table 1 Main features of the single nucleotide polymorphisms in the endotoxin recognition complex
Full size table

Outcome measures

The main outcome measures were the dichotomous measure ‘development of SIRS in the first 3 days of intensive care stay’ and the near-continuous variable ‘platelet count on admission’. SIRS was defined according to the 1992 American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) Consensus Conference [26], with adjustments as described previously [8, 9].

The duration of ventilation, length of stay and development of nosocomial infection in the first week of intensive care stay were secondary outcome measures.

Validation cohort

A cohort of 1,170 Caucasian patients, angiographically diagnosed to have coronary artery stenosis (>50% stenosis in ≥1 major epicardial coronary artery) in the Southampton Atherosclerosis Study, was recruited and genotyped for the TLR4 polymorphism D299G. Genotype and phenotype are described in detail elsewhere [25]. A subset of patients (n = 777) was also genotyped by Taqman analysis for both the D299G and T399I polymorphisms. Given the substantial time and financial constraints associated with recruiting a large equivalent cohort, we chose this cohort as a readily available relevant dataset.

Based on our observations in critically ill children, we hypothesised that TLR4 variant carriers with a history of recent myocardial infarction as a severity threshold for inflammation would show lower platelet counts on admission.

Statistical analysis

Hardy–Weinberg equilibrium was assessed by the χ2 test. Univariate analysis was performed using the χ2, Mann–Whitney U, Student’s t test or one-way analysis of variance (ANOVA) as appropriate. Genotype parameters were included in a binomial regression analysis in addition to clinically relevant parameters. A two-tailed p value <0.05 was considered to be statistically significant. All statistical analyses were performed using SPSS ver. 13.0 for Mac OS X (SPSS, Chicago, IL).

Results

Genotyping

In total, we successfully genotyped 906 (99.2%) critically ill children.

All SNPs were in Hardy–Weinberg equilibrium according to ethnic background, and allele frequencies were in accordance with the literature [27].

TLR4 genotype distribution differed between ethnic groups. The children of Asian descent (n = 26) exhibited only the wild-type TLR4 genotype, whereas amongst those of African descent (n = 78) the TLR4 D299G polymorphism was common (n = 10, 12%) and T399I was observed only once (1.2%). In contrast, both variant alleles were present [D299G, n = 72 (9.6%); T399I, n = 74 (9.9%)] in the Caucasian group (n = 747) and they usually occurred concurrently, i.e., showed cosegregation (ESM Fig. E2).

Development of SIRS in the first 3 days of intensive care stay

SIRS occurred in 594/913 (65%) patients. The prevalence of early SIRS varied between the different diagnostic groups, with significantly more children in the Infection group developing SIRS (143/179, 79.8%) than in the Non-infection (126/199, 63.3%) or Bypass groups (325/535, 60.7%) (Table 2).

Table 2 Clinical characteristics for the total paediatric cohort stratified according to early development of SIRS
Full size table

No significant difference was observed in the development of SIRS between subjects with wild-type or variant genotypes for either of the TLR4 variant alleles alone or in combination; 66% (68/103) of the subjects with a TLR4 mutation developed SIRS versus 65% (522/803) with the wild-type genotype (Fig. 1a).

Fig. 1
figure 1

a Univariate analysis for the Toll-like receptor 4 (TLR4) polymorphism variant alleles as a risk factor for systemic inflammatory response syndrome (SIRS) in the total paediatric cohort. The p value is derived from the χ2 test. b Platelet count for the total paediatric cohort stratified by TLR4 genotype—wild-type versus variants. p value is derived from Student’s t test. c One-way analysis of variance (ANOVA) for platelet count per TLR4 variant allele (0,1 or ≥2) for the total paediatric cohort. d Thrombocytopenia (platelet count < 150 × 109) stratified by TLR4 genotype for the total paediatric cohort [odds ratio (OR) 1.7, 95% confidence interval (CI) 1.1–2.6; χ2 test)]. SEM Standard error of the mean

Full size image

In the Bypass or Infection groups (ESM Figs. E3A, E4A), the incidence of SIRS did not differ when stratified according to TLR4 genotypes. In the Non-infection group, however, we noted a trend for carriers of the TLR4 variant alleles to be at an increased risk for developing SIRS [odds ratio (OR) 2.4, 95% confidence interval (CI) 0.84–6.6; p = 0.09] (ESM Fig. E5A). After multivariate analysis adjusting for basic demographics and other polymorphisms in the endotoxin recognition complex, such as the CHRNA7, PAI-1, MBL-2 and genes encoding cytokines IL-6, TNFa and IL-10 this just reached statistical significance (p = 0.03) with OR 5.4 (95% CI 1.1–26.5).

Platelet count

In the total paediatric cohort there was a marked difference in admission platelet count between patients with wild-type and those carrying the 299G and/or 399I TLR4 variant [mean ± standard error of the mean (SEM) 175 ± 4 vs. 143 ± 7 × 109/L, respectively; p = 0.0001] (Fig. 1b). This effect was additive (Fig. 1c). Consequently, the proportion of children with actual thrombocytopenia (platelet count <150 × 109/L) was significantly higher amongst those carrying one or more variant TLR4 alleles compared to those with the wild-type genotype (OR 1.7, 95% CI 1.1–2.7; p = 0.01) (Fig. 1d; Table 3).

Table 3 Independent variable analyses for thrombocytopenia (<150 × 109/L) in the total paediatric cohort
Full size table

Interestingly, the effect that TLR4 polymorphisms exerted differed between the precipitating event groups. No difference in platelet count could be seen in the Bypass group (ESM Fig. E3B), while in the Non-infection group, children with variant TLR4 alleles had lower admission platelet counts than those with the wild-type genotype (ESM Fig. E5B). A trend in the same direction was also evident in the smaller Infection group (ESM Fig. E4B).

The risk for thrombocytopenia on admission in the whole population associated with TLR4 variants remained after multiple regression analysis. It was independent of any combination of primary diagnosis, demographic characteristics, neutrophil count, SIRS, PIM score and other endotoxin complex and PAI-1, MBL-2, CHRNA7, IL-6, IL-10 or TNFα genotypes (OR 2.2, 95% CI 1.2–3.9; p = 0.01).

Polymorphisms in LBP or CD14 did not alter the risk for the development of SIRS or platelet count (ESM Table E3).

The secondary clinical outcome measures, i.e. length of stay, length of ventilation or nosocomial infection, did not show any stratification according to genotype (ESM Table E4).

Validation cohort

Both phenotypic data [age 63.3 ± 0.3 years (mean ± SEM), sex ratio (M/F) 3.2, 888/276] and TLR4 D299G genotype (in Hardy–Weinberg equilibrium) were available for 1,065 patients. Of these patients, 573 (54%) had experienced a recent (within 3 months) myocardial infarction. In this latter group, those subjects with TLR4 variant genotypes had lower platelet counts than those with the wild-type genotype (mean ± SEM: 217 ± 7.0 vs. 237 ± 2.8 × 109/L; p = 0.021; Fig. 2). Because of the high prevalence of allele cosegregation in this Caucasian cohort, this study was not powered to assess a ‘gene dose effect’. There was no significant difference in platelet count in the group that had not suffered a recent myocardial infarction (variant vs. wild-type genotype: 244 ± 10 vs. 229 ± 3.5 × 109/L; p = 0.14).

Fig. 2
figure 2

Platelet count for adults with a history of myocardial infarction, stratified by TLR4 genotype (wild-type vs. variant). p value is derived from Student’s t test

Full size image

Discussion

In this cohort of more than 900 critically ill children we found little impact of common functional polymorphisms in the endotoxin receptor on the development of SIRS. Conversely, we were able to show a clear and unequivocal effect of TLR4 genotype on platelet count. Polymorphisms associated with reduced TLR4 function were found to exert a dose-dependent effect, with the lowest platelet counts found in those haplotypes associated with the lowest TLR4 function. The stepwise association between TLR4 polymorphisms and lower platelet counts may imply that these two SNPs reduce TLR4 function by addition or interaction.

We have also confirmed that common co-segregation for the D299G and T399I polymorphisms is found only in Caucasians. This result indicates that studies analysing either of these SNPs in isolation will be suboptimal in multiethnic cohorts [28]. This factor may underlie some of the variability in the results reported in published studies.

What does a clinically small (175 vs. 143 × 109/L) but statistically highly significant difference in platelet count associated with TLR4 haplotype mean given the lack of effect on SIRS? These observations appear to be contradictory, but perhaps they offer some insight into acute inflammatory processes and the limitations of the current definitions of systemic inflammation employed for clinical research.

The apparent conflict arises from the view that platelets are an integral part of the acute inflammatory response [29] and that there is an important relationship between thrombocytopenia and severity of illness across all age ranges [21, 22, 30, 31]. Numerous observations suggest that platelets actively contribute to the intensity of the acute inflammatory response rather than being bystanders that are consumed as a ‘para-phenomenon’ [32, 33]. The suggested mechanisms include an acceleration of the recruitment of neutrophils and monocytes through the formation of heterotypic complexes as well as the production of inflammatory mediators, including CD40 ligand, tissue factor, RANTES and matrix metalloproteinases [34].

If platelet activation is inseparable from the acute inflammatory response, why might reduced endotoxin responsiveness affect platelet count but have no detectable effect on clinical markers of systemic inflammation? The simplest explanation would be a type II error. SIRS may be an insufficiently precise measure and, therefore, despite the numbers of cases reported here, this study may be inadequately powered. The limitations of a clinical diagnosis of SIRS are well documented [19]. In part this is due to therapy-induced changes on the parameters temperature, heart rate and respiratory rate that constitute clinical SIRS. The label SIRS may thus poorly reflect the underlying state of immune activation. Also, we have only studied children with a precipitating insult of sufficient severity to require admission to intensive care for organ support. Any variability in the development of SIRS that may be attributable to TLR4 polymorphisms may be swamped by the severity of this level of insult.

The fact that SIRS is a binary outcome measure compounds these limitations. In contrast, platelet count on admission is a near continuous measure that is unlikely to be confounded dramatically by therapy. Alternatively, it may be that platelet count is just a more sensitive measure of the degree of inflammation than the crude indices used to define SIRS.

Importantly, no influence of MBL-2 genotype on platelet count was demonstrated, while others and our group have previously described an increased risk of SIRS in children carrying MBL-2 genotypes associated with reduced plasma levels of the pattern recognition molecule MBL. Therefore, another explanation for the discordant results between the impact of TLR4 polymorphisms on the development of SIRS and platelet count requires consideration. Rather than these results reflect the degree of complexity and specificity of the two end-points themselves, it may be that they reflect a difference in the level of complexity of the inflammatory pathways relevant to the two end-points. The development of SIRS is a complex process with multiple pathways and, therefore, a high level of redundancy. As such, the inherited dysfunction or therapeutic inhibition of single elements are unlikely to have a large impact on overall outcome [35, 36]. Specifically, the intra-cellular signalling response to TLR ligation has been described in great detail as an example of a biologically robust process [37]. In contrast to the main effector cells of systemic inflammation (monocyte/macrophages and endothelial cells), platelets express a very limited range of surface receptors for inflammatory mediators and have a limited response repertoire due to the absence of genomic DNA [33]. Thus, it is biologically plausible that endotoxin-TLR4 binding causes a direct, non-redundant response of platelet activation and consumption that is independent of the development of SIRS.

The demonstration of functional TLR4 expression on platelets [3, 38] and the observation that adults with TLR4 variant genotypes have significant differences in platelet function [39] provide some support for this concept.

We sought to validate this idea in an independent cohort with a relevant inflammatory insult: reduced severity of atherosclerosis has previously been associated with TLR4 variant genotypes [39, 40]. We were indeed able to show that adults with coronary artery stenosis and a recent history of myocardial infarction who carried TLR4 variant alleles had platelets counts which, on average, were 20 × 109/L lower than those of their wild-type counterparts.

Finally, clinical trial evidence further supports the possibility of the impact of this pathway on platelet count. In a double blind multi-centre randomised study of 393 children with severe meningococcal disease, the anti-endotoxin molecule recombinant bactericidal/permeability-increasing protein fragment (rBPI21) had no significant effects on outcome, but thrombocytopenia was less severe in patients receiving rBPI21 (control group 36% transfused platelets vs. 25% r-BPI21 group, p = 0.03) [41].

Our study has several limitations. The paediatric cohort is heterogeneous. Different precipitating events in nature and time course before admission may have led to our inability to determine a differentiation in SIRS. Given the inherent non-protocolised nature of this observational study, we did not attempt to describe SIRS by way of humoral biomarkers of inflammation, such as cytokines. Future strictly protocolised studies may elucidate the temporal pattern of circulating cytokines in relation to the genetic profile.

The validation cohort differed in age, inflammatory insult and ethnic mix and was, therefore, not a perfect match. However, myocardial infarction is an inflammatory process of substantial severity with a relevance to platelet count and TLR4. Despite this imperfect match, the small (although statistically significant) difference in platelet count and the fact that counts were in the normal range, this result still corroborates our findings in the paediatric cohort.

Polymorphisms in endogenous bactericidal/permeability-increasing protein (BPI) [42] and at least two components of TLR4-induced intracellular signalling, such as Mal [43] and IRAK4 [44], may change susceptibility or outcome in severe infections. We did not type for these polymorphisms, and thus our study might be labelled as incomplete. However, given the ever-evolving knowledge in this area, all similar studies are hampered by this phenomenon.

Last, we did not confirm our findings in human ex vivo experiments or in vivo knock-out mice models to determine the underlying processes. Future studies will need to focus on the specific underlying pathophysiological mechanisms.

In conclusion, we have shown that TLR4 polymorphisms are associated with lower platelet counts in severe inflammation. The reasons for this are unclear but may point to a direct effect of the TLR4 pathways on platelets or indicate that platelet counts are a more sensitive marker of systemic inflammation than SIRS criteria. These data support the view that variation in TLR4 function influences the early inflammatory response.