Evidence for Epistasis Between Hemoglobin C and Immune Genes in Human P

ORIGINAL ARTICLE Evidence for epistasis between hemoglobin C and immune genes in human P. falciparum malaria: a family study in Burkina Faso

A Atkinson1,2, M Barbier3, S Afridi1,2, F Fumoux2,4 and P Rihet1,2 1INSERM, UMR928-TAGC, Marseille, France; 2Aix-Marseille Universite´, INSERM U928-TAGC, Universite´ de la Me´diterrane´e, Marseille, France; 3INSERM, UMR745, Paris, France and 4UMR-MD3, IFR 48, Marseille, France

Hemoglobin C (HbC) has been recently associated with protection against Plasmodium falciparum malaria. It is thought that HbC influences the development of immune responses against malaria, suggesting that the variation at the HbC locus (rs33930165) may interact with polymorphic sites in immune genes. We investigated, in 198 individuals belonging to 34 families living in Burkina Faso, statistical interactions between HbC and 11 polymorphisms within interleukin-4 (IL4), IL12B, NCR3, tumor necrosis factor (TNF) and lymphotoxin-a (LTA), which have been previously associated with malaria-related phenotypes. We searched for multilocus interactions by using the pedigree-based generalized multifactor dimensionality reduction approach. We detected 29 multilocus interactions for mild malaria, maximum parasitemia or asymptomatic parasitemia after correcting for multiple tests. All the single-nucleotide polymorphisms studied are included in several multilocus models. Nevertheless, most of the significant multilocus models included IL12B 30 untranslated region, IL12Bpro or LTA þ 80, suggesting that those polymorphisms play a particular role in the interactions detected. Moreover, we identified six multilocus models involving NCR3 that encodes the activating natural killer (NK) receptor NKp30, suggesting an interaction between HbC and genes involved in the activation of NK cells. More generally, our findings suggest an interaction between HbC and genes influencing the activation of effector cells for phenotypes related to mild malaria. Genes and Immunity (2011) 12, 481–489; doi:10.1038/gene.2011.19; published online 31 March 2011

Keywords: hemoglobin C; Plasmodium falciparum; mild malaria; parasitemia; genetic interaction; PGMDR

Introduction family-based association tests.7 However, a large study conducted in Ghana failed to detect an association of Hemoglobin (Hb) variants have been shown to protect HbC with mild malaria in children enrolled at 3 months.8 from clinical malaria. In particular, the protective In all, several studies indicated a protective effect of effect of HbS (b6Glu4Val, rs334) is now well esta- HbC, whereas other studies indicated a lack of asso- blished.1 More recently, the HbC mutation (b6Glu4Lys, ciation. These conflicting results may be explained by rs33930165) has been associated with protection from (i) different genetic backgrounds in the studied popula- severe malaria in Mali, Ghana and Burkina Faso,2–4 tions, (ii) different age groups studied, (iii) different although initial studies indicated lack of protection in malaria transmission intensities in different endemic Nigeria and in Mali.5,6 Modiano et al.4 detected a 29% areas, (iv) a lack of power in some studies because of the reduction in risk of clinical malaria in HbAC individuals, modest effect of HbC at the heterozygous stage (AC) and a 93% reduction in risk of clinical malaria in HbCC and/or an insufficient sample size, (v) the existence individuals in a large population living in Burkina of genetic interactions that were not taken into account. faso; this study provided evidence of protective effect In addition, the conflicting results may be mainly due to against both mild and severe malaria. In the same way, gene–age, gene–environment or gene–gene interactions. we found that HbC carriers had less frequent mild In spite of some conflicting results, it should be malaria attacks than AA individuals in the same age stressed that several association studies support the group in a longitudinal study of individuals living in protective effect of HbC in human malaria. This is Burkina Faso, and that HbC was negatively associated further supported by evidence of a positive selection,9,10 with mild malaria and maximum parasitemia by using and by in vitro studies with human cells.11–14 Although the molecular mechanisms behind the protection remain to be elucidated, several mechanisms have been pro- Correspondence: Professor P Rihet, Aix-Marseille Universite´, posed: (i) a low capacity of the parasite to replicate in red INSERM U928-TAGC, Universite´ de la Me´diterrane´e, 163 Av de blood cells; (ii) an altered expression of parasite antigens Luminy, 13288 Marseille Cedex 9, France. E-mail: [email protected] on the surface of red blood cells leading to a diminished Received 3 November 2010; revised 8 February 2011; accepted 15 cytoadhesion and/or an altered immune response; February 2011; published online 31 March 2011 (iii) and an increased phagocytosis of the infected red HbC and immune genes in human malaria A Atkinson et al 482 blood cells. Several studies have demonstrated in vitro We detected gene–gene interactive effects for interleukin that the growth of the parasite was inhibited in CC 4(IL4), IL12B, tumor necrosis factor (TNF), NCR3 and cells,12–14 whereas the parasite grows normally in AC LTA by using the pedigree-based generalized multifactor cells.12 Recently, Fairhurst et al. showed that AC and dimensionality reduction (PGMDR) method.25 CC cells have a reduced capacity of adhesion to endo- thelial monolayers expressing CD36 and ICAM-1, roset- ting interactions with non-infected red blood cells, and Results agglutination in the presence of pooled sera from malaria-immune adults, and an altered expression of Table 1 summarizes the association results of the studied PfEMP-1 on the membrane. These findings suggest that single-nucleotide polymorphisms (SNPs) with mild HbC may protect against cerebral malaria by inhibiting malaria, maximum parasitemia and asymptomatic para- PfEMP-1-mediated sequestration in cerebral microves- sitemia, and also reports previously published associa- sels. The abnormal display of PfEMP-1 on AC and CC tion studies. In addition, we assessed the family-based red blood cells also lead to a reduced CD36 phagocytosis association of NCR3 and TNF polymorphisms with of trophozoite-infected cells by blood monocytes, asymptomatic parasitemia, of LTA and IL12B polymor- suggesting that HbC does not protect against malaria phisms with mild malaria and maximum parasitemia, by enhancing phagocytosis.15 However, HbC carriers and of IL4-590 with asymptomatic parasitemia, maxi- have been found to produce increased levels of immuno- mum parasitemia or mild malaria. We used marker globulin-G directed against several plasmodial anti- information from previous studies for HBB, IL12B, gens,16,17 suggesting that HbC may increase antibody- TNF, NCR3 and LTA (Table 1), while we genotyped dependent phagocytosis or cytotoxicity, may accelerate IL4-590 polymorphism. In all, we found an association naturally acquired immunity, and may interact with of LTA þ 80 with maximum parasitemia (Z ¼ 1.94; genes involved in immune responses. Besides, natural P ¼ 0.048), and an association of TNF-238 with asympto- killer (NK) cells have been found to recognize PfEMP-1 matic parasitemia (Z ¼ 2.10; P ¼ 0.035). We did not detect at the surface of infected red blood cells via NKp30,18 other additional associations (Table 1). suggesting that the abnormal display of PfEMP-1 on AC We assessed gene–gene interaction between rs33930165 and CC red blood cells may alter the activation of NK (HBB A4C) on the one hand and 11 SNPs within several cells. One might suggest that malaria-related phenotypes immune genes (Table 1) on the other hand by using are influenced by the interaction between HbC and genes PGMDR with adjustment for age. We searched for two- encoding proteins produced by or acting on monocytes, locus models that may explain variation in asymptomatic NK cells, T and B lymphocytes. Genes that are located parasitemia, maximum parasitemia or mild malaria. within chromosomal regions linked to mild malaria or Table 2 summarizes classification parameters calculated parasitemia19–21 and those that have been associated with with the training set (odds ratio and P-value), the cross- mild malaria, parasitemia or severe malaria22–24 are of validation consistency, the prediction accuracy for the particular interest for gene–gene interaction analyses. testing set and the sign test P-value for each significant In this report, we evaluated the statistical inter- two-locus model. We found cross-validation consistency action between HbC and immune genes, which were across all significant models; prediction accuracy ranged previously associated with malaria-related phenotypes. from 0.56 to 0.62 for these models. Seven, five and two

Table 1 SNPs included in the analysis and associated with malaria-related traits in the studied population and other populations

Gene (localization) SNP ID Alternative ID Allelesa Traits associated in the Traits associated in References (MAF) studied population other populations

Mild Maximum Asymptomatic malaria parasitemia parasitemia

HBB (exon 1) rs33930165 HbC A/C (0.138) Yesb Yesb Noc Severe malaria 2–4, 7, 46 NCR3 (promoter) rs27362191 NCR3-412 G/C (0.243) Yesb Nob Noc 23 TNF (promoter) rs1799964 TNF-1031 T/C (0.101) Yesb Nob Noc Severe malaria 24, 27, 55, 56 TNF (promoter) rs1800630 TNF-863 C/A (0.084) Nob Nob Noc Severe malaria 24, 27, 55, 56 TNF (promoter) rs1799724 TNF-857 C/T (0.018) Nob Nob Noc Severe malaria 24, 27, 55, 56 TNF (promoter) rs1800629 TNF-308 G/A (0.105) Yesb Nob Noc Severe malaria 24, 53–56 TNF (promoter) rs361525 TNF-238 G/A (0.025) Nob Yesb Yesc Severe malaria 24, 54–56 TNF (intron 3) rs3093664 TNF 1304 A/G (0.079) Yesb Yesb Noc 24 LTA (intron 1) rs2239704 LTA+80 C/A (0.358) Noc Yesc Yesb 22 IL4 (promoter) rs2243250 IL4-590 C/T (0.180) Noc Noc Noc IgG production 31, 32, 57 in malaria IL12B (promoter) rs17860508 IL12Bpro TTAGAG/GC (0.321) Noc Noc Nob Severe malaria 26, 28–30 IL12B (30UTR) rs3212227 IL12B 3’UTR A/C (0.340) Noc Noc Nob Severe malaria 26, 28–30

Abbreviations: HbC, hemoglobin C; IgG, immunoglobulin G; IL4, interleukin 4; LTA, lymphotoxin-a; MAF, minor allele frequency; SNP, single-nucleotide polymorphism; TNF, tumor necrosis factor; UTR, untranslated region. aWild allele/variant allele (minor allele frequency). bAssociation study previously reported. cAssociation study in the present report.

Genes and Immunity HbC and immune genes in human malaria A Atkinson et al 483 Table 2 Two-locus interaction models for SNPs in HBB and immune genes

Phenotype Genes and SNPs included in the model OR (95% CI)a P-valueb Cross-validation TBA STPc

Mild malaria HBB/TNF-1031 2.42 (1.11–5.25) 0.0245 10/10 0.60 0.0107 Mild malaria HBB/TNF-863 1.94 (0.85–4.43) 0.1150 10/10 0.56 0.0107 Mild malaria HBB/TNF-857 2.24 (0.99–5.07) 0.0504 10/10 0.58 0.0107 Mild malaria HBB/TNF-308 2.15 (1.01–4.57) 0.0448 10/10 0.59 0.0010d Mild malaria HBB/TNF-238 2.10 (0.94–4.68) 0.0672 10/10 0.58 0.0107 Mild malaria HBB/LTA+80 2.05 (0.94–4.46) 0.0683 10/10 0.57 0.0107 Mild malaria HBB/IL12Bpro 2.29 (1.07–4.90) 0.0308 10/10 0.60 0.0010d Maximum parasitemia HBB/TNF-1031 3.38 (1.65–6.93) 0.0007d 10/10 0.62 0.0107 Maximum parasitemia HBB/TNF-238 3.22 (1.52–6.79) 0.0018d 10/10 0.62 0.0107 Maximum parasitemia HBB/TNF1304 2.54 (1.25–5.17) 0.0097d 10/10 0.60 0.0107 Maximum parasitemia HBB/IL12Bpro 3.05 (1.51–6.20) 0.0017d 10/10 0.60 0.0107 Maximum parasitemia HBB/IL12B 30UTR 3.73 (1.77–7.84) 0.0004d 10/10 0.59 0.0107 Asymptomatic parasitemia HBB/LTA+80 2.69 (1.00–7.26) 0.0479 10/10 0.64 0.0010d Asymptomatic parasitemia HBB/TNF-308 1.98 (0.75–5.25) 0.1674 10/10 0.56 0.0107

SNPs were found to interact with HBB for mild malaria, in 18, 14 and 13 of the multilocus models (n ¼ 29), maximum parasitemia and asymptomatic parasitemia respectively. IL4-590 and TNF-1031 were in 10 out of the at the 0.05 nominal level, respectively. These included 29 models, whereas NCR3-412 and SNPs within the TNF mostly several SNPs associated with one of these pheno- promoter were in more than 6 models. In all, 26 out of types. IL12Bpro, IL12B 30 untranslated region (30UTR), the multilocus models included IL12B 30UTR or IL12Bpro, TNF-863 and TNF-857 that were not found to be asso- suggesting that they have a central role in gene–gene ciated with mild malaria or maximum parasitemia interactions. In all, 24 out of the 29 SNP combinations were interacted with rs33930165 for these phenotypes. Interest- identified for the three phenotypes on the basis of the ingly, HBB:IL12Bpro, HBB:TNF-1031 and HBB:TNF-238 training set statistics with a FDR of 5%, whereas there was were significant two-locus models for mild malaria no SNP combination associated with the three phenotypes and maximum parasitemia, whereas HBB:LTA þ 80 and on the basis of the testing set statistics with a FDR of 5% HBB:TNF-308 were significant for mild malaria and (Table 3). At this significance level, all the SNP combina- asymptomatic parasitemia. tions were significant for mild malaria and maximum We further evaluated three-, four- and five-locus parasitemia (n ¼ 18), maximum parasitemia and asympto- models containing rs33930165 (Table 3). Table 3 shows matic parasitemia (n ¼ 9), or mild malaria and asympto- the best models identified on the basis of the classi- matic parasitemia (n ¼ 2) on the basis of the testing set fication parameters with the training set and the cross- statistics. Nevertheless, we identified 16 out of the 29 SNP validation analysis with the testing set (prediction combinations for the three phenotypes based on the testing accuracy and sign test P-value) for mild malaria, maxi- set statistics after applying a FDR of 10% (Table 3). mum parasitemia and asymptomatic parasitemia after applying a false discovery rate (FDR) of 5%. There was cross-validation consistency across all significant models. Discussion The statistical parameters of the multilocus models were better than those of the two-locus models. Although The objective of this paper was to assess statistical some two-locus P-values remained significant for either epistasis between HbC (rs33930165) and genetic variants the training set statistics or the testing set statistics after that are located in immune genes, and that may influence applying a FDR of 5%, the two-locus models did not the level of gene expression. We applied PGMDR that reach the significance threshold for both odds ratio allows adjustment for qualitative and quantitative and sign test P-values: HBB:TNF-308 and HBB:IL12Bpro covariates, and is applicable to dichotomous and quanti- were significant for mild malaria on the basis of the sign tative traits in family-based study designs. To our best test, but were not significant on the basis of the odds knowledge, this is the first comprehensive study ratio statistics (Table 2). In contrast, the multilocus of gene–gene interactions that provided evidence of models that showed a significant sign test P-value for epistatic interactions in human malaria in a familial study. the cross-validation analysis had a significant P-value We evaluated two-, three-, four and five-locus inter- for the odds ratio calculated with the training set, except actions involving HbC. We initially detected several for two SNP combinations (HBB:TNF-857:TNF-238: significant two-locus models at the 0.05 nominal level, IL12Bpro and HBB:TNF1304:IL12B 30UTR:IL12Bpro) when suggesting that HbC may interact with NCR3, TNF, LTA analyzing asymptomatic parasitemia (Table 3). Each SNP and IL12B polymorphisms. However, those were no was included in several significant multilocus models more models that reached the significance threshold on (nX4). IL12B 30UTR, IL12Bpro and LTA þ 80 were included two different statistics after applying a FDR of 5%; the

Genes and Immunity 484 ee n Immunity and Genes

Table 3 Significant three-, four- and five-locus interaction models for SNPs in HBB and immune genes

Genes and SNPs Mild malaria Maximum parasitemia Asymptomatic parasitemia

OR (95% CI) a P-value b TBA STP c OR (95% CI) a P-value b TBA STP d OR (95% CI) a P-value b TBA c STP c

d d d d d HBB/NCR3-412/LTA+80/TNF-1031/TNF-857 4.00 (1.84–8.68) 0.0003 0.6011 0.0010 9.21 (4.22–20.09) 2.62EÀ10 0.68245 0.0010 5.41 (1.89–15.50) 0.0013 0.57095 0.0547 malaria human in genes immune and HbC HBB/NCR3-412/LTA+80/TNF-1031/IL12Bproe 5.01 (2.27–11.08) 4.30EÀ05d 0.57515 0.0107 16.19 (6.95–37.68) 2.55EÀ12d 0.69335 0.0010d 10.20 (3.29–31.64) 4.30EÀ05d 0.60335 0.0010d HBB/NCR3-412/TNF-308/IL4-590/IL12B 30UTR 5.62 (2.52–12.54) 1.35EÀ05d 0.60875 0.0010d 12.06 (5.35–27.21) 1.52EÀ10d 0.6342 0.0010d 5.93 (2.05–17.15) 0.0007d 0.5211 0.1719 HBB/NCR3-412/TNF-238/IL12Bpro 2.88 (1.35–6.16) 0.0057d 0.57985 0.0010d 6.05 (2.88–12.71) 9.22EÀ07d 0.68895 0.0010d 3.29 (1.13–9.60) 0.0270 0.51345 0.3770 HBB/NCR3-412/TNF-238/IL4-590/IL12B 30UTR 5.36 (2.41–11.91) 2.19EÀ05d 0.6052 0.0010d 8.40 (3.88–18.21) 2.19EÀ05d 0.609 0.0010d 5.06 (1.76–14.55) 0.0020d 0.5104 0.3770 HBB/NCR3-412/TNF-238/IL12B 30UTR/IL12Bpro 4.37 (2.00–9.55) 0.0002d 0.6305 0.0010d 9.13 (4.19–19.89) 1.57EÀ04d 0.6914 0.0010d 4.48 (1.55–12.95) 0.0046d 0.5439 0.0547 HBB/LTA+80/IL12Bproe 2.86 (1.34–6.09) 0.0060d 0.5912 0.0010d 5.80 (2.74–12.29) 2.08EÀ06d 0.6516 0.0010d 4.05 (1.45–11.31) 0.0064d 0.6092 0.0107 HBB/LTA+80/TNF-1031/IL12B 30UTRe 3.73 (1.70–8.19) 0.0008d 0.57715 0.0107 10.24 (4.63–22.64) 1.23EÀ09d 0.689 0.0010d 7.07 (2.36–21.20) 0.0003d 0.63275 0.0010d HBB/LTA+80/TNF-1031/IL12Bproe 3.50 (1.63–7.53) 0.0011d 0.5899 0.0010d 10.77 (4.83–24.00) 6.64EÀ10d 0.6744 0.0010d 6.83 (2.26–20.63) 0.0004d 0.6288 0.0107 HBB/LTA+80/TNF-863/IL12Bproe 3.37 (1.56–7.27) 0.0016d 0.5572 0.0107 7.73 (3.57–16.73) 5.87EÀ08d 0.67435 0.0010d 5.72 (1.97–16.63) 0.0009d 0.6104 0.0010d HBB/LTA+80/TNF-857/TNF-308/TNF 1304e 4.22 (1.83–9.72) 0.0005d 0.58265 0.0010d 6.13 (2.91–12.91) 8.12EÀ07d 0.64695 0.0107 5.03 (1.74–14.58) 0.0022d 0.6038 0.0010d HBB/LTA+80/TNF-857/IL12B 30UTR 3.09 (1.41–6.73) 0.0040d 0.55335 0.0547 7.63 (3.37–17.29) 3.78EÀ07d 0.62845 0.0010d 5.18 (1.80–14.92) 0.0017d 0.60405 0.0010d HBB/LTA+80/TNF-857/IL12Bproe 3.17 (1.48–6.80) 0.0026d 0.5795 0.0107 7.16 (3.30–15.53) 1.91EÀ07d 0.66755 0.0010d 4.88 (1.71–13.90) 0.0023d 0.62495 0.0010d Atkinson A HBB/LTA+80/TNF-308/TNF 1304/IL12B 30UTR 5.01 (2.24–11.23) 5.58EÀ05d 0.6058 0.0010d 11.82 (5.28–26.50) 1.79EÀ10d 0.66145 0.0010d 6.36 (2.12–19.06) 0.0007d 0.53565 0.1719 HBB/LTA+80/TNF-238/IL4-590/IL12B 30UTRe 5.28 (2.39–11.69) 2.41EÀ05d 0.6084 0.0010d 12.20 (5.41–27.55) 1.31EÀ10d 0.66875 0.0010d 7.95 (2.66–23.78) 0.0001d 0.6148 0.0107 HBB/LTA+80/TNF-238/IL12B 30UTR/IL12Bpro 3.73 (1.71–8.14) 0.0008d 0.5524 0.0547 9.11 (4.17–19.93) 5.64EÀ09d 0.6597 0.0010d 6.37 (2.18–18.57) 0.0004d 0.5978 0.0010d HBB/LTA+80/TNF 1304/IL4-590/IL12B 30UTRe 6.60 (2.92–14.91) 2.48EÀ06d 0.6234 0.0010d 15.61 (6.69–36.43) 6.62EÀ12d 0.69955 0.0010d 9.59 (3.12–29.45) 6.62EÀ12d 0.6304 0.0107 HBB/TNF-1031/TNF-863/TNF-308/IL12B 30UTRe 4.19 (1.84–9.57) 0.0005d 0.59625 0.0010d 8.94 (4.11–19.45) 7.13EÀ09d 0.69645 0.0010d 5.28 (1.84–15.14) 0.0014d 0.5919 0.0107 al et HBB/TNF-1031/TNF-863/IL4-590/IL12Bpro 4.75 (2.16–10.45) 7.19EÀ05d 0.5937 0.0010d 9.56 (4.08–22.41) 2.53EÀ08d 0.64535 0.0010d 5.73 (1.99–16.52) 0.0009d 0.49995 0.3770 HBB/TNF-1031/TNF-857/IL4-590/IL12B 30UTRe 4.98 (2.26–10.98) 4.41EÀ05d 0.6183 0.0010d 10.76 (4.83–23.95) 6.74EÀ10d 0.6704 0.0010d 5.58 (1.94–16.04) 0.0010d 0.5881 0.0107 HBB/TNF-1031/TNF-308/IL12B 30UTR 2.97 (1.34–6.60) 0.0066d 0.51855 0.1719 6.87 (3.19–14.79) 2.80EÀ07d 0.63825 0.0010d 5.49 (1.75–17.21) 0.0025d 0.57815 0.0010d HBB/TNF-1031/TNF-238/IL4-590/IL12B 30UTR 4.58 (2.08–10.06) 0.0001d 0.62445 0.0010d 8.07 (3.71–17.57) 3.76EÀ08d 0.64815 0.0010d 5.00 (1.73–14.42) 0.0022d 0.54135 0.1719 HBB/TNF-1031/IL4-590/IL12B 30UTR/IL12Bproe 5.51 (2.48–12.24) 1.57EÀ05d 0.6196 0.0010d 12.83 (5.50–29.94) 1.78EÀ10d 0.6628 0.0010d 7.05 (2.39–20.80) 0.0002d 0.5779 0.0107 HBB/TNF-863/TNF-857/TNF-308/TNF-238e 3.03 (1.41–6.48) 0.0039d 0.61545 0.0010d 5.12 (2.35–11.19) 2.14EÀ05d 0.6365 0.0010d 2.90 (1.07–7.84) 0.0339 0.55715 0.0107 HBB/TNF-863/TNF-857/IL4-590/IL12B 30UTR 4.30 (1.98–9.37) 0.0002d 0.6103 0.0010d 7.41 (3.38–16.23) 1.65EÀ07d 0.61155 0.0010d 5.26 (1.73–16.05) 0.0026d 0.5496 0.1719 HBB/TNF-863/IL12B 30UTR/IL12Bproe 2.68 (1.26–5.73) 0.0099 0.5608 0.0107 5.84 (2.79–12.24) 1.37EÀ06d 0.6784 0.0010d 4.88 (1.54–15.42) 0.0051d 0.60105 0.0010d HBB/TNF-857/TNF-238/IL12Bproe 2.90 (1.33–6.32) 0.0066d 0.60915 0.0010d 4.70 (2.23–9.89) 2.95EÀ05d 0.6033 0.0107 3.50 (1.07–11.47) 0.0336 0.5812 0.0010d HBB/TNF-238/IL4-590/IL12B 30UTR/IL12Bproe 4.19 (1.91–9.17) 0.0003d 0.5808 0.0010d 9.03 (4.07–20.05) 1.26EÀ08d 0.622 0.0010d 5.99 (2.05–17.49) 0.0007d 0.5841 0.0107 HBB/TNF 1304/IL12B 30UTR/IL12Bpro 3.01 (1.39–6.54) 0.0047d 0.56465 0.0547 6.27 (2.96–13.26) 6.81EÀ07d 0.6644 0.0010d 4.00 (1.34–11.92) 0.0109 0.6104 0.0010d

Abbreviations: CI, confidence interval; IL12Bpro, interleukin 12Bpro; LTA, lymphotoxin-a; OR, odds ratio; SNP, single-nucleotide polymorphism; STP, sign test P-value; TBA, testing-balanced prediction accuracy; TNF, tumor necrosis factor; UTR, untranslated region. a Training OR, CI. b P-value from training w2 . c Computed using the non-parametric sign test. d Significant P-value after applying a false discovery rate of 5%. e Significant SNP combination for the three phenotypes after applying a false discovery rate of 10%. HbC and immune genes in human malaria A Atkinson et al 485 first statistics was based on odds ratio calculated with reduce the inhibition effect. TNF and LTA that are the training set, whereas the second statistics was based produced by activated NK cells activate the effector on a cross-validation sign test calculated with the testing functions of monocytes; TNF and LTA can be viewed set. Furthermore, we focused on the models that showed as activation markers for NK cells,40–42 and as activa- an association with at least two malaria phenotypes, ting molecules for monocytes.41 In addition, TNF-1031, because we found strong correlations between mild TNF-857, TNF-308, TNF-238 and LTA þ 80, which are malaria and maximum parasitemia on the one hand, and cis-regulatory polymorphisms, may influence both the between maximum parasitemia and asymptomatic para- production of TNF and LTA in activated NK cells and sitemia on the other hand; the correlation between mild the ability of monocytes to eliminate the parasite. It malaria and asymptomatic parasitemia was also signifi- should be stressed that TNF is also produced by cant. On this basis, we identified significant 29 three-, monocytes and T lymphocytes, whereas LTA is produced four- or five-locus models, which contained all the by Th1 lymphocytes.43 This suggests that TNF and LTA SNPs studied. Most of the multilocus models included polymorphisms may also act on malaria phenotypes IL12B 30UTR, IL12Bpro or LTA þ 80, suggesting that independently of NK cells. In this way, we provided those polymorphisms have a particular role in the inter- evidence of significant multilocus models that did not actions detected. Interestingly, significant multilocus contain NCR3-412. In particular, we identified 11 multi- models involved SNPs that were not associated with locus combinations that contained LTA þ 80, all the TNF mild malaria, maximum parasitemia or asymptomatic SNPs studied, IL4-590, IL12Bpro and IL12B 30UTR, and parasitemia in the population studied: these are TNF-863 12 multilocus combinations that contained the TNF SNPs and TNF-857 within TNF,24 IL12Bpro and IL12B 30UTR studied, IL4-590, IL12Bpro and IL12B 30UTR. Although within IL12B,26 and IL4-590. Nevertheless, TNF-863, genetic variation in TNF and LTA likely influences the TNF-857, IL12Bpro and IL12B 30UTR have been asso- activation of monocytes, the molecular basis of HbC– ciated with severe malaria,27–30 and IL4-590 has been TNF and HbC–LTA interaction effects on malaria pheno- associated with antibody levels in malaria-infected types remains, nevertheless, unclear. In contrast, HbC individuals.31,32 has been shown to influence the production of immuno- The statistical epistatic interactions that we detected globulin-G directed against malarial antigens,16,17 sug- may correspond to interactions on the biological level. gesting an interaction between HbC and genes involved Here we propose that the statistical interactions reflect in the activation of B lymphocytes and/or the Th1/Th2 the effect of allelic combinations on the activation of balance, such as IL4 and IL12B. Taken together, our effector cells, such as NK cells, monocytes or B lympho- findings support the hypothesis that cis-regulatory cytes. We detected six multilocus models containing polymorphisms within immune genes influence the NCR3-412 located within the promoter of NCR3, which activation of effector cells and phenotypes related to encodes the activating NK receptor NKp30. This is in line mild malaria, and that their effects could be altered by with the accumulating evidence of an important role of genetic variation in HBB locus. NK cells in human malaria.33,34 NK cells are among the The HBB locus was recently shown to be a major first blood cells to produce interferon-g in response to locus associated with severe malaria on the basis of a P. falciparum-infected red blood cells, and this production genome-wide one-locus association study in Africa.44 requires contact between NK cells and P. falciparum- The authors detected a strong association signal at the infected red blood cells.35,36 The NK receptors and the HbS variant, and did not detect any significant associa- parasite ligands that are involved in the NK activation tions at other locus, suggesting that the effect of other are poorly known. Recently, Mavoungou et al.18 reported genes was very small. Our findings suggest that the that NKp30 binds to PfEMP-1 at the surface of infected effect of other genes on severe malaria may depend red blood cells, and that this interaction mediates on the genetic variation at the HBB locus. Interestingly, cytolysis of P. falciparum-infected red blood cells. Baratin HbC, the frequency of which reaches high values in et al.36 found that PfEMP-1 also binds to ICAM-1, and West Africa, is close to HbS; HbC and HbS mutations that this interaction was mandatory for NK cell are located within the same codon. Although the issue is interferon-g response. In contrast, D’Ombrain et al.37 still debated, it is thought that HbC and HbS may share found that PfEMP-1 suppresses interferon-g production some effects, such as enhanced immune responses.17,45–47 by peripheral blood mononuclear cells, and that the loss Further studies on genetic interaction between HbC and of PfEMP-1 fails to impact cytolosis-associated LAMP-1 HbS on the one hand and immune genes on the other surface translocation. These conflicting results imply that hand may help to unravel the genetic control of severe the molecular basis of NK activation needs further malaria. investigation. They are, however, consistent with the In conclusion, our main findings are that HbC- hypothesis that the variation in HBB influences both immune genes interactions influences mild malaria, the display of PfEMP-1 and the activation of NK cells. maximum parasitemia and asymptomatic parasitemia, The multilocus models involving HbC and NCR3-412 and that multilocus interactions better explain the included a combination of the following polymorphisms: variation in the phenotypes. These statistical interactions TNF-1031, TNF-857, TNF-308, TNF-238, LTA þ 80, IL4- suggest the existence of biological interactions involving 590, IL12Bpro and IL12B 30UTR. As IL12 is a potent HbC. Our findings support the hypothesis that HbC activator cytokine for NK cells,38 IL12Bpro and IL12B influences the protective immune response in indivi- 30UTR that are thought to influence the production of duals living in malaria endemic areas. Symmetrically, the IL12 in monocytes may influence the activation of NK protective effect of HbC may depend on the genotypes at cells. In contrast, IL4 inhibits both the cytotoxicity and other locus. Therefore, our findings may partly explain the cytokine production in NL cells,39 and IL4-590 that the conflicting results concerning the protective effect of alters the production of IL4 in T lymphocytes may also HbC in human malaria.

Genes and Immunity HbC and immune genes in human malaria A Atkinson et al 486 Materials and methods malaria and asymptomatic parasitemia (P ¼ 0.013). A linear regression analysis that took into account the effect Subjects and phenotype determination of age showed a positive correlation between maximum The study population consisted of 198 individuals parasitemia and asymptomatic parasitemia (Po0.0001). belonging to 34 families living a urban district of Bobo Dioulasso in Burkina Faso, in which infected mosquitoes Genotyping were detected only during August, September and Blood samples were taken by venipuncture. The Hb October; the numbers of infective bites per person and genotypes were identified by electrophoresis of red per year was 30. Phenotypes and DNA were available for blood cell lysates on acetate membrane at an alkaline all the individuals. The mean age of sibs was 12.1 þ 6.2 pH. Acetate sheets were stained with ponceau red. This years (range 1–34 years). The study population and the yielded discrimination of Hbs A, S and C. area of parasite exposure have been described.48,49 The DNA was extracted from mononuclear cells separated Medical Authority of Burkina Faso approved the study by the Ficoll–Hypaque density gradient as described.21 protocol. Samples were first subject to prior whole-genome Febrile episodes were extensively recorded by active amplification by primer extension preamplification.50 case detection during 24 months. For patients with fever, For NCR3, TNF, LTA and IL12B polymorphisms, we a thick blood film was prepared by the standard used the data sets previously reported.22–24,26 procedures. Diagnosis of mild malaria attack was based IL4-590 C/T genotypes was determined by using PCR on P. falciparum parasitemia, fever (axillary temperature and allele-specific restriction enzyme digestion methods more than 37.5 1C) and clinical symptoms (headache, as previously described.51 The PCR products were aching, vomiting or diarrhea in the children); in that case digested with the restriction enzymes BsmFI (New no threshold of parasitemia was used. In the absence of England Biolabs Inc., Ipswich, MA, USA), resulting in classical symptoms of malaria, and once others pathol- two fragments of 192 and 60 bp for the 590C allele and an ogies could not be eliminated, only children (ageo15 intact (252 bp) product for the 590 T allele. Digestion years) with more than 5000 parasites per ml and older products were analyzed by electrophoresis on a 3% subjects with more than 2000 parasites per ml were agarose gel stained by ethidium bromide. In addition, considered as having had a malaria attack. Each episode sequencing was performed to control the genotype of 10 of illness was treated according to the recommendation individuals. Briefly, the PCR products were purified with of the CNRFP (Centre National de Recherche et Forma- the Qiagen QIAquick PCR purification kit (Qiagen, tion sur le Paludisme) of Burkina Faso. Parasitemia was Courtaboeuf, France) and quantified by 2% agarose checked at the end of the treatment. Subjects who gel electrophoresis. Sequencing reaction was perfor- presented at least one mild malaria attack during the med with the CEQ 8000 kit and a CEQ 8000 auto- survey were considered in the analysis affected, whereas mated fluorescent sequencer (Beckman Coulter, Roissy the others were considered unaffected. CDG, France). All the genotypes were confirmed by Determination of parasitemia was described in our sequencing. previous study.48 Briefly, each family was visited 20 times during the 24 months of the study and parasitemia was Family-based association and interaction analyses measured. In addition, parasitemia was measured dur- We carried out family-based association analyses ing febrile episodes. The mean number of parasitemia using the family-based association test approach. This measurements per subject was 15.2±5.1 (range 1–24). approach avoids biases because of the population Fingerprint peripheral blood samples were taken from stratification, population heterogeneity or population all family members present, and thick and thin blood admixture.52 The family-based association test statis- films were stained with Giemsa. The parasite determina- tics, which uses data from sibships in nuclear families, tion and numeration were established blindly from two takes into account sibling correlations. The default independent readings. Only P. falciparum asexual forms null hypothesis is no linkage and no association. The were retained to determine parasitemia. Parasitemia was statistics under this hypothesis calculated the distri- defined as the number of parasitized erythrocytes bution of offspring genotypes that are conditional on observed per ml in thin blood films. parental genotypes and on trait values. We calculated a Maximum parasitemia was based on a logarithmic Z-score and a two-side P-value based on a normal transformation of the highest parasitemia that was distribution, and also empirical P-values based on measured in each individual during the survey. Mean 100 000 permutations. of adjusted asymptomatic parasitemia was a logarithmic Gene–gene interactions between HbC and other SNPs transformation of the parasitemia adjusted for seasonal located within NCR3, TNF, LTA, IL4 and IL12B were transmission48 after excluding parasitemia during febrile analyzed by using PGMDR.25 Table 1 presents the SNPs episodes. To take into account the seasonality of the that were selected on the basis of their association with transmission, the influence of the date of the visits on malaria phenotypes.22–24,27–32,46,53–57 ln(1 þ parasitemia) (LP) was evaluated by one way The PGMDR is a score-based multifactor dimension- analysis of variance. The mean LP observed during each ality reduction method that uses the same data reduction visit was calculated. The individual LP was then strategy as does the original multifactor dimensionality corrected for the visit effect by subtracting from each reduction method to detect nonlinear genetic inter- individual LP the mean LP of the corresponding visit. actions,58 that is to classify multilocus genotype combi- Logistic regression analyses that took into account nations as high risk or low risk ones. The original the effect of age showed a strong positive correlation multifactor dimensionality reduction method uses the between mild malaria and maximum parasitemia ratio of cases and controls to identify the combinations of (Po0.0001) and a positive correlation between mild genotypes that show the strongest association with the

Genes and Immunity HbC and immune genes in human malaria A Atkinson et al 487 phenotype. The generalized multifactor dimensionality 3 Mockenhaupt FP, Ehrhardt S, Cramer JP, Otchwemah RN, reduction approach uses statistical scores that replace the Anemana SD, Goltz K et al. Hemoglobin C and resistance to ratio of cases and controls to discriminate between high severe malaria in Ghanaian children. J Infect Dis 2004; 190: risk and low risk genotype combinations, allowing to 1006–1009. analyze both qualitative and quantitative phenotypes, 4 Modiano D, Luoni G, Sirima BS, Simpore J, Verra F, Konate A and to take into account covariates.59 The PGMDR et al. Haemoglobin C protects against clinical Plasmodium method extends the generalized multifactor dimension- falciparum malaria. Nature 2001; 414: 305–308. ality reduction one, and offers handling different family 5 Gilles HM, Fletcher KA, Hendrickse RG, Lindner R, Reddy S, 25 Allan N. Glucose-6-phosphate-dehydrogenase deficiency, types and sizes as well as patterns of missing data. sickling, and malaria in African children in South Western Briefly, we used a 10-fold cross-validation procedure. Nigeria. Lancet 1967; 1: 138–140. The informative sibs were randomly divided into nearly 6 Guinet F, Diallo DA, Minta D, Dicko A, Sissoko MS, Keita MM 10 nearly equal subsets, and the cross-validation was et al. A comparison of the incidence of severe malaria in repeated 10 times. Each time, nine subsets were used as Malian children with normal and C-trait hemoglobin profiles. the training set, whereas the last subset was considered Acta Trop 1997; 68: 175–182. the testing set. We used a linear regression model and a 7 Rihet P, Flori L, Tall F, Traore AS, Fumoux F. Hemoglobin C is logistic regression model for the quantitative traits associated with reduced Plasmodium falciparum parasitemia (maximum parasitemia and asymptomatic parasitemia) and low risk of mild malaria attack. Hum Mol Genet 2004; 13: and the dichotomous trait mild malaria, respectively; this 1–6. 8 Kreuels B, Kreuzberg C, Kobbe R, Ayim-Akonor M, allowed us to take into account age as a covariate to Apiah-Thompson P, Thompson B et al. Differing effects of calculate the scores. Statistical scores were calculated for HbS and HbC traits on uncomplicated falciparum malaria, each genotype combination on the basis of the training anemia, and child growth. Blood 2010; 115: 4551–4558. set, leading to label high risk and low risk genotype 9 Modiano D, Bancone G, Ciminelli BM, Pompei F, Blot I, combinations; high risk and low risk genotype combina- Simpore J et al. Haemoglobin S and haemoglobin C: ‘quick tions were further pooled into separate groups, creating a but costly’ versus ‘slow but gratis’ genetic adaptations to binary model. The classification accuracy was calculated Plasmodium falciparum malaria. Hum Mol Genet 2008; 17: for the training set, and the odds ratio, the corresponding 789–799. 95% confidence interval and the P-value were calculated. 10 Wood ET, Stover DA, Slatkin M, Nachman MW, Hammer MF. The model that had the maximum classification accuracy The beta-globin recombinational hotspot reduces the effects was chosen as the best. The consistency of the model was of strong selection around HbC, a recently arisen mutation providing resistance to malaria. Am J Hum Genet 2005; 77: tested 10 times; it corresponded to the number of times 637–642. that a given attribute combination was selected as the 11 Fairhurst RM, Baruch DI, Brittain NJ, Ostera GR, Wallach JS, best model. The testing set was used to estimate the Hoang HL et al. Abnormal display of PfEMP-1 on erythrocytes prediction accuracy of the best model, which is the ratio carrying haemoglobin C may protect against malaria. Nature of correct classifications to the total number of instances 2005; 435: 1117–1121. classified; because cross-validation was used, this value 12 Fairhurst RM, Fujioka H, Hayton K, Collins KF, Wellems TE. was estimated from the average of all accuracies across Aberrant development of Plasmodium falciparum in hemo- the 10 subsets. We used the non-parametric sign test to globin CC red cells: implications for the malaria protective evaluate the significance of the prediction accuracy. effect of the homozygous state. Blood 2003; 101: 3309–3315. Multiple test corrections were further performed using 13 Friedman MJ, Roth EF, Nagel RL, Trager W. The role of 60 hemoglobins C, S, and Nbalt in the inhibition of malaria the FDR method for odds ratio and sign test P-values; parasite development in vitro. Am J Trop Med Hyg 1979; 28: we applied a FDR of 5 or 10%. 777–780. 14 Olson JA, Nagel RL. Synchronized cultures of P falciparum in abnormal red cells: the mechanism of the inhibition of growth Conflict of interest in HbCC cells. Blood 1986; 67: 997–1001. 15 Cholera R, Brittain NJ, Gillrie MR, Lopera-Mesa TM, Diakite The authors declare no conflict of interest. SA, Arie T et al. Impaired cytoadherence of Plasmodium falciparum-infected erythrocytes containing sickle hemoglobin. Proc Natl Acad Sci USA 2008; 105: 991–996. 16 Storey J, Fleming AF, Cornille-Brogger R, Molineaux L, Acknowledgements Matsushima T, Kagan I. Abnormal haemoglobins in the Sudan savanna of Nigeria. IV. Malaria, immunoglobulins We thank all volunteer families of Bobo Dioulasso. AA and antimalarial antibodies in haemoglobin AC individuals. and MB were supported by a studentship from the Ann Trop Med Parasitol 1979; 73: 311–315. French Ministry of Research and Technology. This work 17 Verra F, Simpore J, Warimwe GM, Tetteh KK, Howard T, was supported by INSERM. Osier FH et al. Haemoglobin C and S role in acquired immunity against Plasmodium falciparum malaria. PLoS One 2007; 2: e978. 18 Mavoungou E, Held J, Mewono L, Kremsner PG. A Duffy References binding-like domain is involved in the NKp30-mediated recognition of Plasmodium falciparum-parasitized erythro- 1 Allison AC. Genetic control of resistance to human malaria. cytes by natural killer cells. J Infect Dis 2007; 195: 1521–1531. Curr Opin Immunol 2009; 21: 499–505. 19 Flori L, Kumulungui B, Aucan C, Esnault C, Traore AS, 2 Agarwal A, Guindo A, Cissoko Y, Taylor JG, Coulibaly D, Fumoux F et al. Linkage and association between Plasmodium Kone A et al. Hemoglobin C associated with protection from falciparum blood infection levels and chromosome 5q31-q33. severe malaria in the Dogon of Mali, a West African Genes Immun 2003; 4: 265–268. population with a low prevalence of hemoglobin S. Blood 20 Flori L, Sawadogo S, Esnault C, Delahaye NF, Fumoux F, Rihet 2000; 96: 2358–2363. P. Linkage of mild malaria to the major histocompatibility

Genes and Immunity HbC and immune genes in human malaria A Atkinson et al 488 complex in families living in Burkina Faso. Hum Mol Genet membrane protein-1 specifically suppresses early production 2003; 12: 375–378. of host interferon-gamma. Cell Host Microbe 2007; 2: 130–138. 21 Rihet P, Traore Y, Abel L, Aucan C, Traore-Leroux T, Fumoux 38 Agaugue S, Marcenaro E, Ferranti B, Moretta L, Moretta A. F. Malaria in humans: Plasmodium falciparum blood infection Human natural killer cells exposed to IL-2, IL-12, IL-18, or IL-4 levels are linked to chromosome 5q31-q33. Am J Hum Genet differently modulate priming of naive T cells by monocyte- 1998; 63: 498–505. derived dendritic cells. Blood 2008; 112: 1776–1783. 22 Barbier M, Delahaye NF, Fumoux F, Rihet P. Family-based 39 Marcenaro E, Dondero A, Moretta A. Multi-directional cross- association of a low producing lymphotoxin-alpha allele with regulation of NK cell function during innate immune reduced Plasmodium falciparum parasitemia. Microbes Infect responses. Transpl Immunol 2006; 17: 16–19. 2008; 10: 673–679. 40 Cooper MA, Fehniger TA, Turner SC, Chen KS, Ghaheri BA, 23 Delahaye NF, Barbier M, Fumoux F, Rihet P. Association Ghayur T et al. Human natural killer cells: a unique innate analyses of NCR3 polymorphisms with P. falciparum mild immunoregulatory role for the CD56 (bright) subset. Blood malaria. Microbes Infect 2007; 9: 160–166. 2001; 97: 3146–3151. 24 Flori L, Delahaye NF, Iraqi FA, Hernandez-Valladares M, 41 Iversen AC, Norris PS, Ware CF, Benedict CA. Human NK Fumoux F, Rihet P. TNF as a malaria candidate gene: cells inhibit cytomegalovirus replication through a noncyto- polymorphism-screening and family-based association analy- lytic mechanism involving lymphotoxin-dependent induction sis of mild malaria attack and parasitemia in Burkina Faso. of IFN-beta. J Immunol 2005; 175: 7568–7574. Genes Immun 2005; 6: 472–480. 42 Marcenaro E, Della Chiesa M, Bellora F, Parolini S, Millo R, 25 Lou XY, Chen GB, Yan L, Ma JZ, Mangold JE, Zhu J et al. A Moretta L et al. IL-12 or IL-4 prime human NK cells to mediate combinatorial approach to detecting gene-gene and gene- functionally divergent interactions with dendritic cells or environment interactions in family studies. Am J Hum Genet tumors. J Immunol 2005; 174: 3992–3998. 2008; 83: 457–467. 43 Hunt NH, Grau GE. Cytokines: accelerators and brakes in the 26 Barbier M, Atkinson A, Fumoux F, Rihet P. IL12B polymor- pathogenesis of cerebral malaria. Trends Immunol 2003; 24: phisms are linked but not associated with Plasmodium 491–499. falciparum parasitemia: a familial study in Burkina Faso. 44 Jallow M, Teo YY, Small KS, Rockett KA, Deloukas P, Clark TG Genes Immun 2008; 9: 405–411. et al. Genome-wide and fine-resolution association analysis of 27 Hananantachai H, Patarapotikul J, Ohashi J, Naka I, Krudsood malaria in West Africa. Nat Genet 2009; 41: 657–665. S, Looareesuwan S et al. Significant association between TNF- 45 Cabrera G, Cot M, Migot-Nabias F, Kremsner PG, Deloron P, alpha (TNF) promoter allele (À1031C, À863C, and À857C) Luty AJ. The sickle cell trait is associated with enhanced and cerebral malaria in Thailand. Tissue Antigens 2007; 69: immunoglobulin G antibody responses to Plasmodium 277–280. falciparum variant surface antigens. J Infect Dis 2005; 191: 28 Marquet S, Doumbo O, Cabantous S, Poudiougou B, Argiro L, 1631–1638. Safeukui I et al. A functional promoter variant in IL12B 46 Ntoumi F, Flori L, Mayengue PI, Matondo Maya DW, Issifou S, predisposes to cerebral malaria. Hum Mol Genet 2008; 17: Deloron P et al. Influence of carriage of hemoglobin AS and the 2190–2195. Fc gamma receptor IIa-R131 allele on levels of immunoglobulin 29 Morahan G, Boutlis CS, Huang D, Pain A, Saunders JR, Hobbs G2 antibodies to Plasmodium falciparum merozoite antigens in MR et al. A promoter polymorphism in the gene encoding Gabonese children. JInfectDis2005; 192: 1975–1980. interleukin-12 p40 (IL12B) is associated with mortality from 47 Williams TN, Mwangi TW, Roberts DJ, Alexander ND, cerebral malaria and with reduced nitric oxide production. Weatherall DJ, Wambua S et al. An immune basis for malaria Genes Immun 2002; 3: 414–418. protection by the sickle cell trait. PLoS Med 2005; 2: e128. 30 Phawong C, Ouma C, Tangteerawatana P, Thongshoob J, 48 Rihet P, Abel L, Traore Y, Traore-Leroux T, Aucan C, Fumoux Were T, Mahakunkijcharoen Y et al. Haplotypes of IL12B F. Human malaria: segregation analysis of blood infection promoter polymorphisms condition susceptibility to severe levels in a suburban area and a rural area in Burkina Faso. malaria and functional changes in cytokine levels in Thai Genet Epidemiol 1998; 15: 435–450. adults. Immunogenetics 2010; 62: 345–356. 49 Traore Y, Rihet P, Traore-Leroux T, Aucan C, Gazin P, 31 Tangteerawatana P, Perlmann H, Hayano M, Kalambaheti T, Coosemans M et al. Analysis of the genetic factors controlling Troye-Blomberg M, Khusmith S. IL4 gene polymorphism and malarial infection in man. Sante 1999; 9: 53–59. previous malaria experiences manipulate anti-Plasmodium 50 Zhang L, Cui X, Schmitt K, Hubert R, Navidi W, Arnheim N. falciparum antibody isotype profiles in complicated and Whole genome amplification from a single cell: implications uncomplicated malaria. Malar J 2009; 8: 286. for genetic analysis. Proc Natl Acad Sci USA 1992; 89: 32 Tangteerawatana P, Pichyangkul S, Hayano M, Kalambaheti T, 5847–5851. Looareesuwan S, Troye-Blomberg M et al. Relative levels of 51 Walley AJ, Cookson WO. Investigation of an interleukin-4 IL4 and IFN-gamma in complicated malaria: association with promoter polymorphism for associations with asthma and IL4 polymorphism and peripheral parasitemia. Acta Trop 2007; atopy. J Med Genet 1996; 33: 689–692. 101: 258–265. 52 Laird NM, Horvath S, Xu X. Implementing a unified approach 33 Roetynck S, Baratin M, Johansson S, Lemmers C, Vivier E, to family-based tests of association. Genet Epidemiol 2000; Ugolini S. Natural killer cells and malaria. Immunol Rev 2006; 19(Suppl 1): S36–S42. 214: 251–263. 53 McGuire W, Hill AV, Allsopp CE, Greenwood BM, Kwiat- 34 Korbel DS, Finney OC, Riley EM. Natural killer cells and kowski D. Variation in the TNF-alpha promoter region innate immunity to protozoan pathogens. Int J Parasitol 2004; associated with susceptibility to cerebral malaria. Nature 34: 1517–1528. 1994; 371: 508–510. 35 Artavanis-Tsakonas K, Eleme K, McQueen KL, Cheng NW, 54 McGuire W, Knight JC, Hill AV, Allsopp CE, Greenwood BM, Parham P, Davis DM et al. Activation of a subset of human NK Kwiatkowski D. Severe malarial anemia and cerebral malaria cells upon contact with Plasmodium falciparum-infected are associated with different tumor necrosis factor promoter erythrocytes. J Immunol 2003; 171: 5396–5405. alleles. J Infect Dis 1999; 179: 287–290. 36 Baratin M, Roetynck S, Pouvelle B, Lemmers C, Viebig NK, 55 Sinha S, Mishra SK, Sharma S, Patibandla PK, Mallick PK, Johansson S et al. Dissection of the role of PfEMP1 and ICAM- Sharma SK et al. Polymorphisms of TNF-enhancer and gene for 1 in the sensing of Plasmodium-falciparum-infected erythro- FcgammaRIIa correlate with the severity of falciparum malaria cytes by natural killer cells. PLoS One 2007; 2: e228. in the ethnically diverse Indian population. Malar J 2008; 7:13. 37 D’Ombrain MC, Voss TS, Maier AG, Pearce JA, Hansen DS, 56 Ubalee R, Suzuki F, Kikuchi M, Tasanor O, Wattanagoon Y, Cowman AF et al. Plasmodium falciparum erythrocyte Ruangweerayut R et al. Strong association of a tumor necrosis

Genes and Immunity HbC and immune genes in human malaria A Atkinson et al 489 factor-alpha promoter allele with cerebral malaria in Myan- high-order interactions among estrogen-metabolism genes in mar. Tissue Antigens 2001; 58: 407–410. sporadic breast cancer. Am J Hum Genet 2001; 69: 138–147. 57 Vafa M, Maiga B, Israelsson E, Dolo A, Doumbo OK, 59 Lou XY, Chen GB, Yan L, Ma JZ, Zhu J, Elston RC et al. A Troye-Blomberg M. Impact of the IL-4 -590 C/T transition generalized combinatorial approach for detecting gene-by- on the levels of Plasmodium falciparum specific IgE, IgG, gene and gene-by-environment interactions with application IgG subclasses and total IgE in two sympatric ethnic groups to nicotine dependence. Am J Hum Genet 2007; 80: 1125–1137. living in Mali. Microbes Infect 2009; 11: 779–784. 60 Benjamini Y, Drai D, Elmer G, Kafkafi N, Golani I. Controlling 58 Ritchie MD, Hahn LW, Roodi N, Bailey LR, Dupont WD, the false discovery rate in behavior genetics research. Behav Parl FF et al. Multifactor-dimensionality reduction reveals Brain Res 2001; 125: 279–284.

Genes and Immunity