Common NFKBIL2 polymorphisms and susceptibility to pneumococcal disease: a genetic association study

Respiratory infection is the single largest contributor to the global burden of disease and the leading cause of death in children worldwide [1, 2]. Streptococcus pneumoniae (the pneumococcus) remains the most common cause of community-acquired pneumonia in Europe and the United States [3]. In addition to pneumonia, pneumococcal infection may also manifest as invasive disease, defined by the isolation of S. pneumoniae from a normally sterile site such as blood (bacteraemia) or cerebrospinal fluid (meningitis). Although asymptomatic colonisation of the nasopharynx by the pneumococcus is widespread in the population, invasive pneumococcal disease (IPD) occurs in only a minority of individuals [4, 5]. The factors that influence development of invasive disease remain poorly understood, although increasing evidence points towards a role for genetic variation in the host's immune response [5]. In particular, recent insights from the study of animal models, rare human primary immunodeficiency states, and population-based genetic epidemiology have focused attention on the control of the proinflammatory transcription factor NF-κB in the development of IPD [5‐10].

NF-κB plays a key regulatory role in a diverse array of cellular processes, including innate and adaptive immune responses [11, 12]. In unstimulated cells, NF-κB transcription factor subunits are prevented from binding DNA through associations with the inhibitor of NF-κB (IκB) protein family. Stimulation of a variety of immune receptors (including Toll-like receptors, T-cell and B-cell antigen receptors and members of the IL-1 and TNF receptor superfamilies) leads to phosphorylation and degradation of the IκB inhibitors and release of NF-κB, which induces transcription of proinflammatory target genes [11, 12]. Genes that are activated by NF-κB include those encoding cytokines (for example, IL-1, IL-2, IL-6, TNFα) and chemokines (for example, IL-8, RANTES), as well as acute phase response proteins, adhesion molecules, antimicrobial peptides and inducible enzymes [11, 12].

Members of the IκB family are characterised by multiple ankyrin repeats and can be subdivided into the so-called classical IκBs (IκB-α, IκB-β, IκB-ε), unusual IκBs (IκB-R, IκB-ζ, IκB-L, Bcl-3), and NF-κB precursors [12, 13]. Of these, the least well studied is IκB-R (IκB-related), encoded by the gene NFKBIL2. The gene was first cloned in 1995 from human lung alveolar epithelial cells, and a modified sequence was published in 2000 [14, 15]. The gene contains only three ankyrin-repeat motifs, fewer than other IκB members, and its exons have a more complicated structure than that seen in other IκBs; overall there is only weak homology between IκB-R and other IκB proteins, leading to the suggestion that IκB-R may in fact not be a member of the IκB family [15]. There is evidence, however, to support an interaction of IκB-R with NF-κB. IκB-R was first shown to inhibit DNA binding by NF-κB in electrophoretic mobility shift assays [14], and overexpression of NFKBIL2 in lung alveolar epithelial cells was subsequently reported to significantly upregulate the production of RANTES (now renamed chemokine C-C motif ligand 5 (CCL5)) protein following stimulation with TNFα or IL-1α, although it had no effect on other NF-κB-responsive chemokines such as IL-8 [16].

Increasing evidence supports a central role for the control of NF-κB in susceptibility to severe infectious disease in humans. A mutation in the gene NFKBIA encoding the classical inhibitor IκB-α has been described in two patients with primary immunodeficiency [8]. In addition, population-based case-control studies of IPD have reported associations with polymorphisms in the IκB-encoding genes NFKBIA and NFKBIZ [9, 10]. These findings raise the possibility that variation in additional IκBs such as IκB-R may also contribute to IPD susceptibility. No functional or disease-associated polymorphisms have previously been reported in NFKBIL2, however. To investigate this further we studied the frequencies of NFKBIL2 polymorphisms in individuals with IPD and healthy controls of both European and African descent.

The initial genotyping approach utilised the UK Caucasian IPD case-control study group and focused on three SNPs within NFKBIL2: rs760477, rs2306384, and rs4082353. Whilst both rs760477 and rs4082353 are intronic, rs2306384 encodes a serine/glycine substitution at position 334 of the IκB-R protein. Each of these SNPs was found to be common in Europeans (minor allele frequencies approaching 50%) and to associate with susceptibility to IPD (P = 0.002 to 0.007; Table 3). Logistic regression analysis demonstrated no effect of age, comorbidity or gender on genotype. Genotyping was then extended to flanking SNPs in both directions spanning a 74 kb region across chromosome 8q24.3 to delineate the extent of LD and disease association. Twelve SNPs were found to be either nonpolymorphic or extremely rare (minor allele frequency <0.01) and could not be analysed further (Table 3). A further six SNPs were polymorphic, of which two associated with IPD susceptibility at the 0.05 significance level and one trended towards association (P = 0.035 to 0.085; Table 3). In each case the minor alleles were again found to be common, and the direction of association was one of heterozygote protection against IPD (Table 3).

The extent of LD between SNPs was next assessed. All five IPD-associated SNPs were found to be located within a 20 kb block of strong LD in this European population (Figure 1). The absence of association with SNPs outside this block suggests that the causative locus is indeed localised to this 20 kb region, which contains the entire NFKBIL2 gene as well as the neighbouring gene in a 3' direction, vacuolar protein sorting 28 (VPS28). This extensive LD presents a considerable challenge, however, in identifying the IPD-causative polymorphism. The extent of LD in African populations is typically shorter than in Europeans [24], and this can be advantageous when attempting to fine map an extensive region of disease association. With this in mind, all nine polymorphisms were then genotyped in the Kenyan bacteraemia case-control study (Tables 4 and 5). The LD was noted to be much less extensive in this African population, and no haplotype blocks were predicted by the Gabriel algorithm within the region studied (Figure 2).

Two of the NFKBIL2 SNPs genotyped were found to be significantly associated with susceptibility to Gram-positive and pneumococcal bacteraemia in Kenyan children (rs4925858 and rs760477; Table 6). In each case the direction of association was of heterozygote protection, the same genetic model as that observed in the UK Caucasian study. Logistic regression analysis demonstrated no effect of age, comorbidity, HIV infection or gender on genotype. Comparison of ORs for rs4925858 and rs760477 did not demonstrate any evidence of heterogeneity between the UK and Kenyan case-control groups for either SNP. The strongest association was with rs760477; on combining and stratifying the UK and Kenyan study groups, heterozygosity at rs760477 was associated with significant protection against invasive bacterial disease overall (IPD in the UK study and overall bacteraemia in the Kenyan study; Mantel-Haenszel 2 × 2 χ² = 18.567, P = 1.6 × 10^-5, OR = 0.66, 95% confidence interval for OR = 0.55 to 0.80) and against invasive pneumococcal disease specifically (Mantel-Haenszel 2 × 2 χ² = 11.797, P = 0.0006, OR = 0.67, 95% confidence interval for OR = 0.53 to 0.84). Heterozygosity at the neighbouring SNP rs4925858 was also found to be protective against both invasive bacterial disease overall (Mantel-Haenszel 2 × 2 χ² = 8.610, P = 0.003, OR = 0.76, 95% confidence interval for OR = 0.63 to 0.91) and IPD (Mantel-Haenszel 2 × 2 χ² = 9.104, P = 0.003, OR = 0.70, 95% confidence interval for OR = 0.55 to 0.88). None of the SNPs appeared to be associated with outcome of bacteraemia in these groups (data not shown), although the number of individuals in the poor outcome groups was small (mortality rates were 10% in the UK IPD study and 28% in the Kenyan study), resulting in a lack of power to examine possible effects of genotype on mortality.

The SNP rs760477 is located within the fourth intron of NFKBIL2. The first seven exons of NFKBIL2, which surround rs760477, were then sequenced in 48 Kenyan individuals in an attempt to identify novel and potentially functional variants. The sequencing covered a 3,100 base pair region extending in a 3' direction from 780 base pairs prior to the start of transcription in exon 1. No novel exonic polymorphisms were identified with the exception of a synonymous polymorphism encoding asparagine at position 23 of the IκB-R protein, which has subsequently been listed on databases and named rs35913924. This SNP was then genotyped in the Kenyan cases and controls: the mutant allele was found to be uncommon (allele frequency 3.6%), and no association with disease was identified (3 × 2 χ² = 0.37, P = 0.83). The sequencing also confirmed the genotyping accuracy of rs760477 (100% concordance between direct sequencing and Sequenom genotyping).

In this study we demonstrate associations between common NFKBIL2 polymorphisms and susceptibility to IPD in UK Caucasian and Kenyan individuals. Important causes of false positive associations in genetic studies are a failure to adjust significance levels when multiple polymorphisms have been analysed, and confounding by population substructure. Nine polymorphisms were analysed, and applying a Bonferroni correction results in a threshold significance level of 0.0055, rather than 0.05. With this corrected significance level, rs760477, rs2306384 and rs4082353 in the UK population remain associated with protection against IPD. The Bonferroni correction, however, assumes that markers are independent, whereas many of the polymorphisms studied here are in strong or complete LD in UK individuals and are therefore not truly independent from each other. Applying instead a correction based on the total number of LD blocks and singleton (not part of a LD block) polymorphisms, five independent tests were performed in the UK study group, suggesting a threshold P value of 0.01 for statistical significance. The extent of LD was much less in Kenyan individuals, and as a result no LD blocks were predicted (Figure 2). In this setting, none of the polymorphisms in the Kenyan study reaches the Bonferroni-corrected P value threshold of 0.0055 to declare significance. Nevertheless, given the observed association between NFKBIL2 SNPs and IPD in UK individuals, the a priori probability that such a SNP protects against IPD in the Kenyan population might be expected to be higher than for a random marker, and in this situation the Bonferroni adjustment may be overly stringent. It is also noteworthy that the SNPs rs4925858 and rs760477 trend or associate in the same direction (heterozygote protection) in the Kenyan study as that observed in the UK study group, and combined analysis of the UK and Kenyan study groups using the Mantel-Haenszel test further strengthens the association between NFKBIL2rs760477 and IPD.

Addressing the possibility of population substructure, recent analysis of an extensive dataset of over 15,000 individuals from Britain demonstrated remarkably little evidence of geographic population differentiation within British Caucasians [18], and moreover our cases and controls are from a relatively restricted geographic area (Oxfordshire). Furthermore, the observation of a trend towards heterozygote protection against IPD in a second, independent study of African individuals provides additional support for an association between NFKBIL2 polymorphisms and pneumococcal disease. The results of the Kenyan study additionally suggest that the NFKBIL2 association may be with bacteraemia overall, rather than a specific effect on pneumococcal susceptibility, although this finding requires replication.

In general, a possible disadvantage for the use of the Kenyan samples as a replication study group is their different ethnic background: a lack of replication may reflect true ethnic heterogeneity in pneumococcal disease susceptibility. On the other hand, the study of a second population with differing patterns of LD may aid fine-mapping of associations within regions of strong LD, and it has been suggested that the demonstration of genetic associations with disease susceptibility across different populations is perhaps of even more value than the identification of population-specific effects [25]. The IPD-associated polymorphisms in the UK Caucasian study span a distance of 20 kb including the genes NFKBIL2 and VPS28. On the basis of these results alone it is not possible to further localise the disease association within this region, although the associations within the Kenyan study group appear to focus the association within NFKBIL2. Despite the use of such a transethnic mapping approach, the functional variant in NFKBIL2 that is responsible for the association with IPD remains unknown. Perhaps the most probable functional variant within NFKBIL2 is the coding change rs2306384, but it is noteworthy that this association did not replicate in the Kenyan study group. The SNP rs760477 is located within intron 4 and is unlikely itself to exert a functional effect, and no disease-associated polymorphisms were identified in the surrounding exons. The polymorphism rs4925858 is located 1,650 base pairs before the transcription start, and could conceivably affect a regulatory region such as a promoter, enhancer or silencer.

The mechanism by which IκB-R variation influences susceptibility to IPD is also unclear. One possibility is through an effect on CCL5 expression, which has been reported to be upregulated in lung epithelial cells following overexpression of IκB-R [16]. The mechanism behind this cytokine-induced upregulation appears to be sequestering of transcriptionally repressive NF-κB p50 homodimer subunits by IκB-R, thereby facilitating NF-κB-mediated gene transcription of CCL5 [16]. Both CCL5 mRNA and protein expression are stimulated following exposure to the pneumococcal proteins pneumolysin and choline-binding protein A in dendritic cells, and furthermore CCL5 blockade during pneumococcal carriage in mice is associated with an attenuated immune response and greater transition to lethal pneumonia [26, 27]. Further research is required to examine the possible role of IκB-R in regulation of CCL5 during pneumococcal disease, and indeed to identify the cellular roles of IκB-R more generally. This protein has been relatively neglected compared with the extensive literature on other IκBs, and it remains unclear for example which specific NF-κB dimers interact with IκB-R [14, 16].

The direction of association with disease is noteworthy: heterozygosity was associated with protection against IPD in each study population. The finding of heterozygote protection is unusual in genetic disease association studies, but is well described in the study of human infectious disease genetic susceptibility - examples include sickle cell trait and malaria, prion protein gene variation and spongiform encephalopathy, and human leukocyte antigen and HIV/AIDS disease progression [28‐30]. More recently, heterozygosity at loci within both the Toll-like receptor adaptor protein Mal/TIRAP and NFKBIZ have been found to associate with protection against IPD [10, 31]. Interestingly, studies in animal populations have found that increased levels of genome-wide heterozygosity correlate with overall fitness, and more specifically with resistance to infectious disease; for example, resistance to bovine tuberculosis in the Iberian wild boar [32]. These animal studies raise the possibility that heterozygote advantage against infectious disease may be a more widespread phenomenon in humans than previously considered. The biological mechanisms that underlie this remain unclear, although in the setting of inflammatory signalling pathways it has been speculated that homozygote states may lead to extremes of inflammatory response that, under certain circumstances, are detrimental to the host, whereas heterozygosity may result in intermediate signalling that leads to an optimal inflammatory response [31]. Such a model will be further modified by environmental exposures, and differing burdens of bacterial disease may in part account for the observed variation in NFKBIL2 allele frequencies between European and African populations.

Finally, it is interesting that four out of the five IκB genes studied to date show apparent associations with susceptibility to IPD [9, 10]. This further highlights the importance of the control of NF-κB in the host immune response, and suggests that the remaining members of the IκB family are likely to be promising candidates for a role in pneumococcal susceptibility. Study of the genetic basis of NF-κB inhibition may be increasingly relevant given current interest in the regulation of NF-κB activity as a therapeutic target for inflammatory disease [33]. Within the field of infectious disease, inhibition of NF-κB has been demonstrated to improve outcome in animal models of sepsis and pneumococcal meningitis [34, 35]. The anti-inflammatory activity of glucocorticoids is mediated at least in part through physical interference of the glucocorticoid receptor complex with NF-κB DNA binding and increased synthesis of IκB [36], and there is some evidence of benefit from corticosteroids in the treatment of pneumococcal meningitis and perhaps also severe community-acquired pneumonia [37, 38]. Taken together, these findings raise the intriguing possibility that anti-inflammatory treatments such as glucocorticoids may be more effective if tailored on the basis of an individual's genetic profile of NF-κB activation.