Two complement receptor one alleles have opposing associations with cerebral malaria and interact with α+thalassaemia.

Malaria has been a major driving force in the evolution of the human genome. In sub-Saharan African populations, two neighbouring polymorphisms in the Complement Receptor One (CR1) gene, named Sl2 and McCb, occur at high frequencies, consistent with selection by malaria. Previous studies have been inconclusive. Using a large case-control study of severe malaria in Kenyan children and statistical models adjusted for confounders, we estimate the relationship between Sl2 and McCb and malaria phenotypes, and find they have opposing associations. The Sl2 polymorphism is associated with markedly reduced odds of cerebral malaria and death, while the McCb polymorphism is associated with increased odds of cerebral malaria. We also identify an apparent interaction between Sl2 and α+thalassaemia, with the protective association of Sl2 greatest in children with normal α-globin. The complex relationship between these three mutations may explain previous conflicting findings, highlighting the importance of considering genetic interactions in disease-association studies.

Introduction

Complement Receptor One (CR1) plays a key role in the control of complement activation and the immune clearance of C3b/C4b-coated immune complexes (Krych-Goldberg and Atkinson, 2001). CR1 is expressed on a range of cells including red blood cells (RBCs), leucocytes and glomerular podocytes (Krych-Goldberg and Atkinson, 2001). A number of CR1 polymorphisms have been described, including four molecular weight variants and variation in the number of CR1 molecules expressed on the surface of RBCs (reviewed by [Krych-Goldberg et al., 2002; Schmidt et al., 2015]). Missense mutations of CR1 form the basis of the Knops blood group system of antigens, that includes the antithetical antigen pairs of Swain-Langley 1 and 2 (Sl1 and Sl2) and McCoy a and b (McCa and McCb) (Moulds, 2010). The non-synonymous single nucleotide polymorphisms (SNPs) A4828G (rs17047661) and A4795G (rs17047660) within exon 29 of the CR1 gene give rise to the Sl1/Sl2 and McC/McC alleles, encoding R1601G and K1590E, respectively (Moulds et al., 2001) (Figure 1).

Figure 1.

Diagram of the most common Complement Receptor 1 size variant (CR1*1).

Adapted from Schmidt et al. (2015) and Krych-Goldberg et al. (2002). The ectodomain of CR1 is composed of 30 Complement Control Protein (CCP) domains which are organized into four ‘Long Homologous Repeats’ (LHR). The single-nucleotide polymorphisms determining the Sl and McC antigens of the Knops blood group system are found in CCP 25 in LHR-D (red). Various functions have been mapped to different regions of CR1, including Site 1 (decay accelerating activity for C3 convertases; binding of the complement component C4b and the P. falciparum invasion ligand PfRH4), and Site 2 (cofactor activity for Factor I; binding of C3b and C4b and P. falciparum rosetting). LHR-D is thought to bind C1q and Mannose Binding lectin (MBL), but the specific binding sites have not been mapped. TM, transmembrane region; CYT, cytoplasmic tail.

CR1 has been implicated in the pathogenesis of multiple diseases, with epidemiological and in vitro data suggesting a role in malaria (Schmidt et al., 2015). The Sl2 and McC alleles occur at high frequencies only in populations of African origin (Figure 2) (Thathy et al., 2005; Zimmerman et al., 2003; Moulds et al., 2004; Noumsi et al., 2011; Fitness et al., 2004; Covas et al., 2007; Gandhi et al., 2009; Yoon et al., 2013; Hansson et al., 2013; Kariuki et al., 2013; Eid et al., 2010), which, given the historical prevalence of the malaria-causing parasite Plasmodium falciparum in sub-Saharan Africa, might suggest a possible survival advantage against malaria (Rowe et al., 1997; Rowe et al., 2000). CR1 is a receptor for the invasion of RBCs by Plasmodium falciparum merozoites (Spadafora et al., 2010; Tham et al., 2010) and for the formation of clusters of P. falciparum-infected RBCs (iRBCs) and uninfected RBCs, known as rosettes (Rowe et al., 1997). The rosetting phenotype is associated with severe malaria in sub-Saharan Africa (Doumbo et al., 2009), with pathological effects likely due to the obstruction of microcirculatory blood flow (Kaul et al., 1991). RBCs from donors with the high-frequency African CR1 Knops mutations bind poorly to the parasite ligand P. falciparum erythrocyte membrane protein-1 (PfEMP1) that mediates rosetting by iRBCs, potentially protecting against severe malaria by reducing rosetting (Rowe et al., 1997). Nevertheless, epidemiological data supporting this possibility are contradictory, with some studies showing an association between Sl and McC genotypes and severe malaria (Thathy et al., 2005; Kariuki et al., 2013; Tettey et al., 2015) and others finding none (Zimmerman et al., 2003; Hansson et al., 2013; Jallow et al., 2009; Manjurano et al., 2012; Toure et al., 2012; Rockett et al., 2014). Some previous studies have not considered Sl and McC genotypes together in the same statistical model, despite their physical adjacency in the CR1 molecule, nor taken into account potential interactions with other malaria resistance genes. Given the important biological role of CR1 in malaria host-parasite interactions, we aimed to clarify the relationship between the Sl and McC alleles and severe malaria in a case-control study of Kenyan children. These investigations were supplemented with a separate longitudinal cohort study of Kenyan children, examining the associations of these alleles with uncomplicated malaria and other common childhood illnesses. Finally, we also investigated the influence of these alleles on the formation of P. falciparum rosettes, as a potential functional explanation for these results through ex vivo laboratory studies.

Figure 2.

Global distribution of the CR1 Knops Sl and McC alleles

(A) Shows the global frequencies of the Sl alleles. Sl1 is represented in blue and Sl2 in red. (B) Shows the global frequencies of the McC alleles. McC is represented in green and McC in purple. The two samples in North and South America showing high frequencies of Sl2 and McC alleles are both derived from populations with African heritage. Numbers in parentheses indicate the studies from which the Sl and McC allele frequencies were derived, with * indicating data derived from this study. [1] Thathy et al., 2005; [2] Zimmerman et al., 2003; [3] Moulds et al., 2004; [4] Noumsi et al., 2011; [5] Fitness et al., 2004; [6] Covas et al., 2007; [7] Gandhi et al., 2009; [8] Yoon et al., 2013; [9] Hansson et al., 2013; [10] Kariuki et al., 2013; [11] Eid et al., 2010.

Results

The Sl2/Sl2 genotype is associated with protection against cerebral malaria and death in the Kenyan case-control study

Data were obtained from 5545 children enrolled in a case-control study of severe malaria (Figure 3). The general characteristics of the cases and controls are shown in Supplementary file 1A, and the characteristics of the dataset by Sl and McC genotype are shown in Supplementary file 1B. The Sl2 and McC allele frequencies (0.68 and 0.16 respectively) were comparable to other African populations (Figure 2). There was no significant deviation from Hardy-Weinberg equilibrium for the Sl or McC genotypes among controls (Supplementary file 1C).

Figure 3.

Patient inclusion flow chart for the Kenyan case-control study.

Using a simple logistic regression model containing only Sl and McC genotypes (referred to as the unadjusted analysis below), we found a non-significant association between the Sl2 allele and severe malaria overall, with the Sl2/Sl2 genotype being associated with an OR for severe malaria of 0.90 (95% CI 0.79–1.01; p=0.07) (Supplementary file 1D). We attempted to refine this signal by fitting a more complete model to the data, including the potential confounding factors of ethnicity, location, sickle cell trait, ABO blood group and α+thalassaemia genotype, as well as considering possible first-order interactions between terms (referred to as the full adjusted analysis below). A significant protective association was observed for Sl2 in the recessive form (adjusted Odds Ratio (aOR) 0.78; 95% CI 0.64–0.95; p=0.011), which was most marked for cerebral malaria (aOR 0.67; 0.52–0.87; p=0.006) (Figure 4 and Table 1). The Sl2/Sl2 genotype was also associated with significant protection against death from severe malaria (aOR 0.50; 0.30–0.80; p=0.002), and death among children admitted with a specific diagnosis of cerebral malaria in the full adjusted analysis (aOR 0.44; 0.23–0.78; p=0.007) (Figure 4 and Table 1). Unexpectedly, we observed a significant interaction between Sl2 and α+thalassaemia genotype, such that the protective associations of Sl2 were only seen in individuals of normal α-globin genotype (Figure 5). We found no evidence for an association between Sl2 and any other clinical form of severe malaria (Table 1), or with P. falciparum parasite density (Figure 6).

Figure 4.

The Sl2 and McC alleles have opposing associations with cerebral malaria (CM) and death.

Forest plot showing the associations between Sl and McC polymorphisms and severe malaria in Kilifi, Kenya. Filled boxes: adjusted Odds Ratios (aOR) for the Sl2 genotype in the recessive form (i.e. Sl2/Sl2 vs all other Sl genotypes). Open boxes: McC in the additive form (i.e. change in odds ratio with each additional McC allele). Sl and McC genotype were included together in a statistical model to examine their associations with malaria susceptibility. aORs displayed are adjusted for ethnicity, location of residence, sickle cell genotype, α+thalassaemia genotype and ABO blood group. An interaction term between Sl genotype and α+thalassaemia is included in the model. Model outputs following 2000 bootstrapped iterations are shown.

Table 1.

Adjusted Odds Ratios (aOR) for severe malaria by Sl2 (recessive) and McC (additive) genotype in Kenya.

Clinical outcome	Sl2 aORs (95% CI)*	P value	McCb aORs (95% CI)	P value
All severe malaria† (n = 1716)	0.78 (0.64–0.95)	0.011	1.10 (0.97–1.25)	0.108
CM§ (n = 943)	0.67 (0.52–0.87)‡	0.006	1.19 (1.02–1.38)	0.025
Severe without CM (n = 674)	1.00 (0.76–1.30)	0.994	0.98 (0.82–1.17)	0.811
Died (n = 180)†	0.50 (0.30–0.80)‡	0.002	1.31 (0.95–1.72)	0.086
Died with CM (n = 131)	0.44 (0.23–0.78)‡	0.007	1.34 (0.94–1.88)	0.104
Died without CM (n = 42)	0.73 (0.18–2.30)	0.636	1.00 (0.48–1.94)	0.940
SMA# (n = 483)	0.76 (0.55–1.05)	0.099	0.96 (0.78–1.17)	0.688
SMA without CM (n = 223)	0.82 (0.51–1.26)	0.366	0.91 (0.67–1.20)	0.553
Died with SMA¶ (n = 56)	0.65 (0.21–1.67)	0.374	1.35 (0.77–2.20)	0.229
RD** (n = 522)	0.81 (0.59–1.10)	0.181	1.12 (0.92–1.35)	0.225
RD without CM (n = 192)	1.06 (0.66–1.68)	0.805	1.07 (0.80–1.43)	0.615
Died with RD†† (n = 73)	0.39 (0.14–0.88)‡	0.027	1.01 (0.59–1.61)	0.948

*Adjusted Odds Ratios (aOR) and 95% Confidence Intervals (CI) are presented for the Sl2 genotype in the recessive form (i.e. Sl2/Sl2 vs all other Sl genotypes) and McC genotype in the additive form (i.e. change in aOR with each additional McC allele). Sl and McC genotype were included together in a statistical model to examine their associations with malaria susceptibility. aORs displayed are adjusted for ethnicity, location of residence, sickle cell genotype, α+thalassaemia genotype and ABO blood group. An interaction term between Sl genotype and α+thalassaemia was included in the model. Model outputs following 2000 bootstrapped iterations are shown.

†99 children (7 of whom died) were severe malaria cases whose CM status was not recorded, hence these children are included in the numbers for ‘All severe malaria’ and ‘Died’ but not in ‘with CM’ or ‘without CM’ categories.

‡Models that showed significant evidence of interaction between Sl2 and α+thalassaemia.

§CM, cerebral malaria (P. falciparum infection with a Blantyre coma score of < 3).

#SMA, severe malarial anaemia (P. falciparum infection with Hb < 5 g/dl).

¶34/56 cases who died with SMA also had CM.

**RD, respiratory distress (P. falciparum infection with abnormally deep breathing).

††56/73 cases who died with RD also had CM.

Figure 5.

The protective association of Sl2 with cerebral malaria and death is only evident in children with normal α-globin.

Interaction plots showing the interaction between Sl (recessive) and α+thalassaemia for the proportion of children suffering (A) cerebral malaria and (B) death. For α+thalassaemia status, 0 = wild type α-globin; 1 = heterozygote or homozygote for α+thalassaemia. For Sl (recessive) status, 0 = Sl1/Sl1 or Sl1/Sl2 genotype; 1 = Sl2/Sl2 genotype.

Figure 6.

Parasite densities by Sl and McC genotypes.

Geometric mean parasite densities in the Kenyan case-control study (severe malaria) and longitudinal disease cohort study (uncomplicated malaria and asymptomatic parasitaemia) by A) Sl genotypes and B) McC genotypes. The data on severe malaria includes 1695 children: (Sl1/Sl1 (175), Sl1/Sl2 (793), Sl2/Sl2 (727) and McC (1167), McC (478) and McC/McC (50). The data on uncomplicated malaria includes 162 children: (Sl1/Sl1 (16) , Sl1/Sl2 (75), Sl2/Sl2 (71) contributing 124, 488 and 461 episodes respectively and McC (107), McC (49) and McC/McC (6) contributing 699, 349 and 25 episodes, respectively. The data on asymptomatic parasitaemia includes 57 children: (Sl1/Sl1 (5), Sl1/Sl2 (26), Sl2/Sl2 (26) contributing 6, 35 and 35 episodes, respectively, and McC (34), McC (20) and McC/McC (3) contributing 47, 25 and 4 episodes, respectively. Differences in parasite densities by genotype were tested by linear regression analysis with adjustment for HbAS, age as a continuous variable and ABO blood group in the severe malaria cases, HbAS and season (defined into 3-monthly blocks) in the uncomplicated malaria samples and HbAS and ABO blood group in the asymptomatic parasitaemia samples. Data were adjusted for within-person-clustering of events in the uncomplicated malaria and asymptomatic parasitaemia studies. Bars represent 95% confidence intervals.

The McCallele is associated with increased susceptibility to cerebral malaria and death in the Kenyan case-control study

The unadjusted analysis showed a borderline significant association between McC and increased susceptibility to severe malaria overall (OR 1.17; 1.00–1.25; p=0.056, Supplementary file 1D), and significant associations with increased risk of cerebral malaria (OR 1.21; 1.05–1.39; p=0.008) and death (OR 1.34; 1.00–1.77; p=0.046, Supplementary file 1D). Similar associations were seen in the full adjusted analysis, although this only reached statistical significance for cerebral malaria (aOR 1.19; 1.10–1.38; p=0.025 (additive model), Figure 4 and Table 1). We found no association between McC and any other clinical form of severe malaria (Table 1 and Supplementary file 1D) or with P. falciparum parasite density (Figure 6).

Analysis of haplotypic effects and genotype combinations

We considered whether the observed results for Sl and McC could be consistent with the effect of a single haplotype spanning Sl and McC, or with the effect of a specific genotype combination. Sl and McC are 33 bp apart and are in linkage disequilibrium, with only three of four possible haplotypes observed in our data. We therefore reanalyzed the data under a haplotype model in which the per-individual count of each of the three observed haplotypes was included as a predictor along with the potential confounding factors, as well as under a genotypic model in which the count of each of the six possible Sl/McC genotype combinations was included as a predictor (Appendix 2). These analyses suggest an additive protective association with the Sl2/McC haplotype (aOR = 0.85; 0.75–0.96; p=0.007), with broadly consistent results observed for analysis of genotype combinations (Supplementary file 1E and 1F). Thus, the opposing effects of Sl2 and McC observed above could plausibly result from the protective association of a single haplotype at the locus, although this is difficult to distinguish from the individuals SNPs acting independently and additively based on the statistical evidence alone.

The Sl2/Sl2 genotype was associated with protection against uncomplicated malaria in the Kenyan longitudinal cohort study

We next examined the association between Sl2 and McC alleles and uncomplicated malaria in a longitudinal prospective study of 208 Kenyan children. General characteristics of the cohort study population by Sl and McC genotypes are shown in Supplementary file 1G. After adjusting for variables known to influence malaria susceptibility, the Sl2 allele was associated with a >50% reduction in the incidence of uncomplicated malaria (additive model) (Table 2; the number of episodes, incidence and unadjusted Incidence Rate Ratios for the diseases studied in the longitudinal cohort are shown in Supplementary file 1H, I and J). Once again, a significant interaction was seen with α+thalassaemia, such that the protective association of Sl2 was only demonstrated in children of normal α-globin genotype (Table 3). We found no significant association between the McC allele and uncomplicated malaria (Table 2).

Table 2.

Adjusted Incidence Rate Ratios (aIRR) for uncomplicated malaria and non-malarial diseases in Kenya by Sl and McC genotype*.

Clinical Outcomes	Sl2 aIRRs† (95% CI)	P value	McCb aIRRs (95% CI)	P value
Uncomplicated malaria	0.49 (0.34–0.72)‡	<0.001 4	1.24 (0.90–1.70)	0.184 1
All non-malaria clinical visits	1.13 (0.96–1.32)	0.140 1	0.76 (0.61–0.96)‡	0.020 4
LRTI§	1.09 (0.81–1.47)	0.561 1	0.39 (0.16–0.96)	0.040 1
URTI#	1.21 (0.98–1.50)	0.073 1	0.79 (0.63–0.99)	0.047 3
Gastroenteritis	0.66 (0.43–1.03)	0.066 2	0.55 (0.31–0.97)‡	0.038 2
Skin infection	1.33 (0.79–2.26)	0.285 2	0.42 (0.16–1.13)	0.086 1
Helminth infection	1.98 (0.83–4.71)	0.122 2	0.68 (0.43–1.07)	0.094 4
Malaria negative fever	0.83 (0.58–1.18)	0.293 2	1.03 (0.80–1.33)	0.828 3

*Data were collected from 22 Sl1/Sl1, 94 Sl1/Sl2 and 92 Sl2/Sl2 individuals during 49.4, 213.8 and 188.8 cyfu (child-years of follow-up), respectively, and 137 McC, 63 McC/McC and 8 McC individuals during 294.5, 143.2 and 14.3 cyfu, respectively. Both Sl2 and McC alleles were tested for their association with the disease outcomes of interest using Poisson regression in the recessive, dominant, heterozygous and additive models. The best fitting models as examined using the Akaike information criterion (AIC) were used in the final analysis that included adjustment for McC genotype (for Sl analyses), Sl genotype (for McC analyses) α+thalassaemia and sickle cell genotype, ABO blood group, season (divided into 3 monthly blocks), ethnicity, age as a continuous variable and within-person clustering of events.

†aIRRs: adjusted Incidence Rate Ratios.

‡Models that showed significant evidence of interaction between either Sl2 or McC and α+thalassaemia.

§LRTI: Lower Respiratory Tract Infection.

#URTI: Upper Respiratory Tract Infection.

Table 3.

Incidence of uncomplicated malaria by Sl genotype and α+thalassaemia status in the Kenyan longitudinal cohort study.

	Sl1/Sl1		Sl1/Sl2		Sl2/Sl2
	Number of episodes	Incidence	Number of episodes	Incidence	Number of episodes	Incidence
All samples	124	2.51	493	2.31	461	2.44
Normal α globin	73	4.18	238	2.87	77	1.64
Heterozygous α+thalassaemia	32	1.58	209	1.92	302	2.88
Homozygous α+thalassaemia	19	1.63	46	2.09	82	2.20

Incidence = number of episodes per child-year of follow up (cyfu). Data were collected from 22 Sl1/Sl1, 94 Sl1/Sl2 and 92 Sl2/Sl2 individuals during 49.4, 213.8 and 188.8 child-years of follow-up, respectively.

The McC allele was associated with protection from common non-malarial childhood diseases in the Kenyan longitudinal cohort study

The data shown above are incompatible with malaria being the selective pressure for McC in the Kenyan population, and suggest that other life-threatening childhood diseases may have been responsible for selection of McC. We therefore used the same longitudinal cohort study to investigate whether the McC and Sl2 alleles influence the risk of other childhood diseases. McC was associated with borderline significant protection against several common infectious diseases including LRTIs, URTIs and gastroenteritis (Table 2). Sl2 was associated with a borderline reduced incidence of gastroenteritis (Table 2). The association of McC with gastroenteritis was predominantly seen in children of normal α-globin genotype, echoing the interaction seen with Sl2 and malaria.

The Sl2 allele was associated with reduced ex vivo rosette frequency in P. falciparum clinical isolates from Mali

A previous in vitro study based on a culture-adapted P. falciparum parasite line suggested that RBC from Sl2 genotype donors had a reduced ability to form rosettes, providing a possible mechanism for protection against severe malaria (Rowe et al., 1997). P. falciparum clinical isolates were not available from the Kenyan case-control study to investigate this potential mechanism in that population. However, the association of Sl and McC genotypes with ex vivo P. falciparum rosette frequency could be examined using 167 parasite isolates from a case-control study of children with clinical malaria in Mali (Doumbo et al., 2009). Analysis of this small case-control study suggested a protective association between the Sl2/Sl2 genotype and cerebral malaria (aOR 0.35, 95% CI 0.12–0.89, p=0.024) and the Sl2/Sl2-McC genotype combination was associated with protection against cerebral malaria (aOR 0.14, 95% CI 0.02–0.84, p=0.031, Appendix 1). As such, we considered samples from this population to be appropriate for testing rosetting as a potential mechanism of action. The median rosette frequency (percentage of iRBC that form rosettes) was significantly lower in P. falciparum isolates from malaria patients with one or more Sl2 alleles than in isolates from Sl1/Sl1 donors (Figure 7), whereas McC genotype had no significant associations with P. falciparum rosette frequency (Figure 7).

Figure 7.

The Sl2 allele is associated with reduced ex vivo rosette frequency of P. falciparum clinical isolates.

Parasite isolates were collected from 167 malaria patients in Mali and matured in culture for 18–36 hr before assessment of rosette frequency (percentage of infected erythrocytes forming rosettes with two or more uninfected erythrocytes). Red bars show the median rosette frequency and interquartile range (IQR) for each genotype. (A) Rosetting by patient Sl genotype. Sl1/Sl1 (n = 22, median 20.0, IQR 8.3–36.5), Sl1/Sl2 (n = 82, median 4.0, IQR 0–20.3), Sl2/Sl2 (n = 63, median 5.0, IQR 0–17.0); *p<0.05, Kruskal Wallis with Dunn’s multiple comparison test; (B) Rosetting by McC genotype. McC (n = 81, median 9.0, IQR 0–22.0), McC (n = 73, median 6.0, IQR 0–20.0), McC (n = 13, median 4.0, IQR 1–14.5); not significant, Kruskal Wallis with Dunn’s multiple comparison test.

Discussion

The data presented here provide epidemiological evidence supporting a role for CR1 in the pathogenesis of cerebral malaria. Two neighboring CR1 polymorphisms belonging to the Knops blood group system of antigens had opposing associations on risk of cerebral malaria. The Sl2/Sl2 genotype was associated with protection against cerebral malaria and death, while the McC allele was associated with increased susceptibility (Figure 4 and Table 1). The Sl2 allele was also associated with significant protection against uncomplicated malaria, whereas the McC allele was associated with borderline protection against several common infections in Kenyan children (Table 2). The protective association of Sl2 against cerebral malaria, death and uncomplicated malaria was influenced by α+thalassaemia, being most evident in children of normal α-globin genotype.

The protective association between Sl2 and cerebral malaria was first reported in a small case-control study from western Kenya (Thathy et al., 2005), but has remained controversial, especially as most prior studies have been underpowered. Hence, our study is the first adequately powered independent sample set that replicates the protective association between Sl2 and cerebral malaria. Other studies found no consistent significant associations between Sl genotypes and severe malaria (Zimmerman et al., 2003; Hansson et al., 2013; Jallow et al., 2009; Manjurano et al., 2012; Toure et al., 2012; Rockett et al., 2014), including a recent multi-centre candidate gene study that included the sample set analysed here (Rockett et al., 2014). A weak association between McC and an increased odds ratio for cerebral malaria was shown in the multi-centre study (Rockett et al., 2014).

The complex interactions between Sl2, McC and α+thalassaemia revealed by our study provide possible reasons for the previous inconsistent findings. Although Sl2 was associated with protection against cerebral malaria in our study, McC and α+thalassaemia both counteracted this effect. The protective association of Sl2 was observed most clearly when both McC and α+thalassaemia genotypes were included in the statistical model, something that has not been considered in previous studies. It is possible that some of the other discrepant genetic associations with severe malaria (Rockett et al., 2014) might result from interactions between multiple loci that vary across populations and may not be revealed by standard analyses. Biologically, it makes sense to account for McC genotype when investigating associations with Sl2 and vice versa, as the two polymorphisms encode changes only 11 amino acids apart in the CR1 molecule (Figure 1). The possibility that the observed association might be due to a haplotype rather than independent effects of Sl and Mc cannot be discounted.

The interaction we describe here between Sl2 and α+thalassaemia is reminiscent of the epistatic interactions that have been observed between α+thalassaemia and other malaria-protective polymorphisms including sickle cell trait (HbAS) (Williams et al., 2005a) and haptoglobin (Atkinson et al., 2014). It is possible, therefore, that α+thalassaemia has a broad effect on multiple malaria-protective polymorphisms, influencing their restricted global frequencies (Penman et al., 2009), and contributing to the discrepant outcomes of previous association studies. Recent large genetic association studies on malaria do not include data on α+thalassaemia, because the causal deletions are not typed on automated platforms (Rockett et al., 2014), instead requiring manual genotyping using labour-intensive PCR-based methods (Chong et al., 2000). Replication of the Sl2/α+thalassaemia interaction will be required, and we suggest that α+thalassaemia genotype should be included as an important confounding variable in future malaria epidemiological studies and that efforts should continue to discover the mechanism of protection afforded by α+thalassaemia, which remains controversial (Carlson et al., 1994; Fowkes et al., 2008; Krause et al., 2012; Opi et al., 2014; Opi et al., 2016).

We examined one possible biological mechanism by which the Sl2 allele might influence cerebral malaria by studying P. falciparum rosetting, a parasite virulence factor associated with severe malaria in African children (Doumbo et al., 2009). Previous in vitro experiments showed that CR1 is a receptor for P. falciparum rosetting on uninfected RBCs, and that RBCs serologically typed as negative for the Sl1 antigen (likely to be from donors with Sl1/Sl2 or Sl2/Sl2 genotypes) (Moulds et al., 2001) show reduced binding to the parasite rosetting ligand PfEMP1 (Rowe et al., 1997) . In this study, we found a significantly lower median rosette frequency in P. falciparum parasite isolates from Malian patients with Sl2 genotypes compared to Sl1/Sl1 controls (Figure 4). Therefore, similar to HbC (Fairhurst et al., 2005), blood group O (Rowe et al., 2007) and RBC CR1 deficiency (Cockburn et al., 2004), it is possible that reduced rosetting and subsequent reduced microvascular obstruction (Kaul et al., 1991) may in part explain the protective association of Sl2 against cerebral malaria. However, given the protective association of Sl2 with uncomplicated malaria, and the possible associations of Sl2 and McC with other common childhood infections, it seems likely that the Knops polymorphisms may be associated with broader effects, for example on the complement regulatory functions of CR1. Previously, we have shown that neither cofactor activity for the breakdown of C3b and C4b nor binding to C1q are influenced by the Sl2 and McC mutations (Tetteh-Quarcoo et al., 2012). In addition, we can find no association between Knops genotype and CR1 clustering on erythrocytes (Paccaud et al., 1988; Swann et al., 2017). However, other potential effects such as altered immune complex binding and processing or activation of the complement lectin pathway via mannose-binding lectin (Ghiran et al., 2000) have not yet been investigated.

Our studies have several limitations: McC homozygotes are relatively infrequent in Kenya, which limited our power to detect associations with McC in the homozygous state. Our longitudinal cohort study generated several values of borderline statistical significance for the McC allele which are inconclusive. Studies with larger sample sizes will be needed to examine the specific associations of McC on assorted childhood diseases. Another limitation is that our functional (Mali) and epidemiological (Kenya) studies were conducted in different populations. The mechanisms of rosetting and associations with malaria severity are thought to be similar across sub-Saharan Africa (Rowe et al., 2009), suggesting that data collected in either location are likely to be comparable. Furthermore, examination of a small set of cerebral malaria cases and controls from Mali suggests a protective association between Sl2/Sl2 genotype and cerebral malaria also occurs in this setting (Appendix 1). Ideally, future epidemiological and functional studies of specific polymorphisms on malaria should be conducted within a single population, although this remains logistically challenging.

In conclusion, we show that two high frequency CR1 polymorphisms have opposing associations with cerebral malaria and death in Kenyan children. While the Sl2 allele may have reached high frequency in African populations by conferring a protective advantage against cerebral malaria, our data suggest that McC arose due to a survival advantage afforded against other non-malarial infections (Noumsi et al., 2011; Fitness et al., 2004). Sl2 may in part protect against cerebral malaria by reducing rosetting, but additional effects seem likely. Further work is needed to examine both the epidemiological effects of the Knops polymorphisms on diverse childhood diseases, and the biological effects of the Sl2 and McC polymorphisms on CR1 function. Future epidemiological studies should account for the effect of α+thalassaemia on the associations between Sl2 and McC on malaria and other infectious diseases.

Materials and methods

Datasets studied

This study uses data from a Kenyan case-control study of severe malaria, with samples collected between 2001 and 2010, a Kenyan longitudinal cohort study, with samples collected between 1998 and 2001 and a Malian case-control study performed between July 2000 and December 2001. Historic datasets (i.e. >10 years old) are widely used in genetic epidemiological studies of malaria due to the logistical challenges of sample collection in malaria endemic countries and the changing epidemiological patterns of disease.

The Kenyan study area

All epidemiological and clinical studies in Kenya were carried out in the area defined by the Kilifi Health and Demographic Surveillance System (KHDSS), with Kilifi County Hospital (KCH) serving as the primary point of care (Scott et al., 2012). Malaria transmission is seasonal in this region following the long and short rains. An Entomological Inoculation Rate (EIR) of up to 50 infective bites per person per year was measured in the late 1990s (Mbogo et al., 2003), but transmission has since declined (O'Meara et al., 2008).

The Kenyan case-control study

Between January 2001 and January 2008, children aged <14 years who were admitted to KCH with severe malaria were recruited as cases, as described previously (Rockett et al., 2014), except that children who were resident outside the KHDSS were excluded (Figure 3). Severe malaria was defined as the presence of blood-film positive P. falciparum infection complicated by one or more of the following features: cerebral malaria (CM) (a Blantyre coma score (BCS) of <3) n = 943; severe malarial anaemia (SMA) (hemoglobin concentration of <5 g/dl) n = 483; respiratory distress (RD) (abnormally deep breathing) n = 522 or ‘other severe malaria’ (no CM, SMA or RD but other features including prostration (BCS 3 or 4), hypoglycemia and hyperparasitemia) n = 318. Controls (n = 3829) consisted of children 3–12 months of age who were born consecutively within the KHDSS study area between August 2006 and September 2010 and were recruited to an ongoing genetic cohort study (Williams et al., 2009). As such, controls were representative of the general population in terms of ethnicity and residence but not of age. The use of controls who are considerably younger than cases differs from the classical structure of a case-control study. However, this method (using cord blood or infant samples as controls) has been widely used in African genetic association studies (e.g. [Band et al., 2013; Busby et al., 2016; Clarke et al., 2017]) and is the most logistically feasible way of collecting sufficiently large numbers of control samples in many sub-Saharan African settings.

Sample processing and quality control for the Kenyan case-control study

The Sl and McC polymorphisms were originally typed as part of a larger study by Rockett et al., 2014, which included case-control data from 12 global sites. In Kenya, 0.5 ml blood samples were collected into EDTA tubes and DNA extracted using Qiagen DNeasy blood kits (Qiagen, Crawley, UK). DNA was stored at −20°C and shipped frozen to Oxford. Sample processing is described in detail in the supplementary methods of Rockett et al., 2014. Briefly, samples underwent a whole-genome amplification step using Primer-Extension Pre-Amplification. Genotyping was performed using SEQUENOM iPLEX Gold with 384 samples processed per chip. In Rockett et al.’s study, samples were typed for 73 SNPs; 55 of these SNPs were chosen on the basis of a known association with severe malaria, 3 SNPs were used to confirm gender and the remaining 15 SNPs to aid quality control. Samples were excluded if they did not have clinical data for gender or if genotypic gender of the sample did not match clinical gender. Samples were included if they were successfully genotyped for more than 90% of 65 ‘analysis’ SNPs. The Kenyan samples studied by Rockett et al. originally comprised 2741 cases of severe malaria and 4183 controls. After the quality control of both phenotypic and genotypic data described above, 2268 cases and 3949 controls were analysed by Rockett et al., 2014.

Comparison between this study and Rockett et al., 2014.

The 2268 Kenyan cases and 3949 controls that were analyzed by Rockett et al., 2014 were the starting point for our study. Children living outside the KHDSS were excluded, because this allowed us to use ‘location’ as a random effect in the final statistical model, which greatly improved model fit. Children with missing genotypes (Sl, McC, sickle cell, α+thalassaemia or ABO blood group) were also excluded (Figure 3). After applying these exclusion criteria, 1716 severe malaria cases and 3829 community controls were available for analysis.

Hence, the number of severe malaria cases differs between our study and Rockett et al., 2014 due to differing exclusion criteria. The inclusion of the severe malaria cases who lived outside the KHDSS into our statistical models did not alter the findings of our analysis (Supplementary file 1K). In both our study and Rockett et al., 2014, the control samples were identical and all came from within the KHDSS. Our study has 120 fewer controls than Rockett et al., 2014 due to missing genotypes, because we only used controls for whom full Sl, McC, sickle cell genotype, α+thalassaemia genotype and ABO blood group data were available.

Our analytical methods differed from Rockett et al., 2014, in that we included both Sl and McC in the same statistical model and adjusted for confounders, whereas Rockett et al. examined each SNP independently.

The Kenyan longitudinal cohort study

This study has been described in detail previously (Nyakeriga et al., 2004). Briefly, this study was established with the aim of investigating the immuno-epidemiology of uncomplicated clinical malaria and other common childhood diseases in the northern part of the KHDSS study area, approximately 15 km from KCH (Williams et al., 2005b). The study was carried out between August 1998 and August 2001 involving children aged 0–10 years recruited either at the start of the study or at birth when born into study households during the study period. They were actively followed up on a once-weekly basis for both malaria and non-malaria related clinical events. In addition, on presentation with illnesses, cohort members were referred to a dedicated outpatient clinic for more detailed diagnostic tests. The cohort was monitored for the prevalence of asymptomatic P. falciparum infection through four cross-sectional surveys carried out in March, July and October 2000 and June 2001. Exclusion criteria included migration from the study area for more than 2 months, the withdrawal of consent and death. Uncomplicated clinical malaria was defined as fever (axillary temperature of > 37.5°C) in association with a P. falciparum positive slide at any density. The most common non-malaria-related clinical events reported during the study period included upper respiratory tract infections (URTIs), lower respiratory tract infections (LRTIs), gastroenteritis, helminth infections and skin infections, as defined in detail previously (Williams et al., 2005b). Malaria negative fever was defined as an axillary temperature of > 37.5°C in association with a slide negative for P. falciparum. This analysis includes 208 children aged < 10 years for whom full Sl, McC, sickle cell genotype, α+thalassaemia genotype and ABO blood group data were available.

The Malian case-control study

This study has been described in detail previously (Lyke et al., 2003). Briefly, between July 2000 and December 2001, children ranging from 1 month to 14 years of age were recruited into a case-control study in the Bandiagara region in East Central Mali, an area of intense and seasonal P. falciparum malaria infection. In order to address the specific question of whether the Sl2/Sl2 genotype is associated with protection against cerebral malaria in Mali, only the subset of children suffering strictly defined cerebral malaria (a BCS of <3, with other obvious causes of coma excluded, n = 34) or uncomplicated malaria (n = 184, symptomatic children with P. falciparum parasitemia and an axillary temperature ≥37.5°C, in the absence of other clear cause of fever), and for whom Sl and McC genotyping was available were analyzed.

Ex vivo rosetting

The rosette frequency (percentage of mature infected erythrocytes forming rosettes with two or more uninfected erythrocytes) of P. falciparum isolates from patients recruited into the Mali case-control study was determined by microscopy after short term culture (18–36 hr), as described in detail previously (Doumbo et al., 2009). Of the 209 isolates studied previously (Doumbo et al., 2009), 167 were successfully genotyped for the Sl and McC alleles and are analysed here. The rosetting assays were performed before we genotyped the study participants, excluding observer bias. The rosette frequency of parasites from hosts with differing Sl and McC genotypes were compared by a Kruskal-Wallis test with Dunn’s multiple comparisons (Prism v6.0, Graphpad Inc, San Diego, CA).

Laboratory procedures

DNA was extracted either from fresh or frozen whole blood by proprietary methods using either the semi-automated ABI PRISM 6100 Nucleic acid prep station (Applied Biosystems, Foster City, CA) or using QIAamp DNA Blood Mini Kits (Qiagen, West Sussex, UK). SNPs giving rise to the Sl and McC alleles were genotyped using either the SEQUENOM iPLEX Gold multiplex system (Agena Biosciences, Hamburg, Germany) (Kenyan study) (Rockett et al., 2014) or by an established PCR-RFLP method as described previously (Malian study) (Moulds et al., 2004). Genotyping for sickle cell trait (HbAS) and the common African α+thalassaemia variant caused by a 3.7 kb deletion in the HBA gene were performed by PCR as described in detail elsewhere (Chong et al., 2000; Waterfall and Cobb, 2001).

Statistical analysis

The effects of the Sl and McC alleles were examined in genotypic, dominant, recessive and additive models of inheritance, with the best fitting model selected based on Akaike information criterion (AIC). Analyses for the Kilifi case-control study were performed in R (R Foundation for Statistical Computing, Vienna, Austria) (R Development Core Team, 2010) using the ‘ggplot2’, ‘lme4’, and ‘HardyWeinberg’ packages (Wickham, 2009; Bates et al., 2015; Graffelman and Camarena, 2008), while analyses for the longitudinal study were performed in Stata v11.2 (StataCorp, Texas, USA). In both studies, a p value of < 0.05 was considered statistically significant. Graphs were generated using R or Prism v6.0 (Graphpad Inc, San Diego, CA).

For the Kenyan case-control study, Sl and McC genotype were included together in a statistical model to examine their associations with malaria susceptibility. Odds Ratios (ORs) and 95% Confidence Intervals (CI) were generated using mixed effect logistic regression analysis both with and without adjustment for ethnicity and location of residence as random effects, and sickle cell genotype, α+thalassaemia genotype, and ABO blood group (O or non-O) as fixed effects (variables which have been associated with malaria susceptibility in multiple previous studies in this population) (Jallow et al., 2009; Rockett et al., 2014; Williams et al., 2005a; Atkinson et al., 2014; Rowe et al., 2007; Williams et al., 2005b; Fry et al., 2008; Malaria Genomic Epidemiology Network et al., 2015). The ethnicity variable was compressed from 28 categories to four; Giriama (n = 2728), Chonyi (n = 1800), Kauma (n = 588) and other (n = 429). Binary parameterization of the α+thalassaemia variable was used, that is, comparing those children with no α+thalassaemia alleles against those with one or more α+thalassaemia alleles. This division was chosen in accordance with a previous report showing that both heterozygous and homozygous α+thalassaemia genotypes are associated with protection against severe malaria and death in the Kilifi area (Williams et al., 2005c). 2000 bootstrapped iterations were run to give 95% CIs and p values.

For the Kenyan longitudinal cohort study, Incidence Rate Ratios (IRRs) and 95% CIs were generated using a random effects Poisson regression model that took into account within-person clustering. Data were examined with and without adjustment for confounding by McC genotype (for Sl analyses), Sl genotype (for McC analyses) sickle cell genotype, α+thalassaemia genotype, ABO blood group, ethnic group, season (defined as 3 monthly blocks), and age in months as a continuous variable.

For the Malian case-control study, ORs and 95% CIs were computed using mixed effect logistic regression analysis with adjustment for location of residence as a random effect and age, ABO blood group (O or non-O) and ethnicity (Dogon or non-Dogon) as fixed effects. α+thalassaemia genotyping was not available for the Malian study and sickle cell trait is extremely uncommon in this population, therefore neither variable was included in the model. 2000 bootstrapped iterations were run to give adjusted ORs.

Corrections for multiple comparisons were not performed, instead all adjusted odds ratios, confidence intervals and p values have been clearly reported. This approach has been repeatedly advocated, particularly when dealing with biological data (Rothman, 1990; Perneger, 1998; Nakagawa, 2004; Fiedler et al., 2012; Rothman, 2014). A detailed description of the Malian dataset is given in Appendix 1, and a detailed description of the statistical model fitting for the Kenyan studies is given in Appendix 2.

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Two complement receptor one alleles have opposing associations with cerebral malaria and interact with α+thalassaemia" for consideration by eLife. Your article has been favorably evaluated by Prabhat Jha (Senior Editor) and three reviewers, one of whom, Madhukar Pai (Reviewer #1), is a member of our Board of Reviewing Editors.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this letter to draw your attention to significant concerns that you would need to address before this work could be considered for publication.

Summary:

This is an interesting but complex study with different study designs, looking at interactions between 3 mutations and malaria. The study raises hypotheses that will need to be confirmed in future studies.

The study evaluated the relationship between 2 SNPs (rs17047661 and rs17047660) of CR1 and several malaria phenotype (cerebral malaria, death, uncomplicated malaria other severe forms of malaria) as well as non-malarial disease. The authors found that 1) the Sl2/Sl2 genotype is associated with protection against cerebral malaria and death, p=0.006 and p= 0.002 respectively, 2) the McC allele is associated with increased susceptibility to cerebral malaria and death, p=0.008 and p=0.046 respectively, 3) the Sl2/Sl2 genotype was associated with protection against uncomplicated malaria p<0.001, and 4) the McC allele was associated with protection from non-malarial disease p=0.02. Finally, the authors demonstrated that the allele Sl2 was associated with reduced ex vivo rosette frequency, whereas McC had no significant effect on P. falciparum rosette frequency. The paper is interesting and suggests the role of CR1 polymorphisms in susceptibility/resistance to malaria and non-malarial disease. However, the reviewers have major concerns on several points.

Essential revisions:

1) Given the dramatic reductions in severe malaria and deaths in the past decade, it is a bit strange that the studies were done with data collected during 2001-08 (case control study), and 1998-2001 (cohort study). So, a decade later, the authors have gone back to analyze stored samples. The implications of this are not clear and need to be clearly addressed. Are there time trends that can confound the associations? What about use of old samples and sample quality? Were the hypotheses pre-stated or generated post-hoc? (presumably the latter).

2) As regards the case-control study, I would need more convincing that cases and controls were derived from the same study base. The difference in age distribution will likely not matter for the genetic mutations, but what about implications for disease severity?

3) The authors indicated that the study included 5545 children from Kenya consisting in 1716 severe malaria cases and 3829 community controls.a) The authors performed the analysis on a subset of the samples used previously by Rockett et al. (2014). The authors should clarify why they did not use the total sample set of 2268 severe malaria cases included in Rockett et al. and how the 1716 samples were selected.

b) As indicated by the authors, 407 individuals are excluded as lived out with KHDSS (Figure 3). Why? The authors should clarify this point. DNA not available? It would be surprising because all CR1 genotype (rs17047661 and rs17047660), sickle cell genotype, a+thalassaemia genotype and ABO blood group were available in the previous study of Rockett for the Kenyan population.

c) It is not clear for me whether new genotyping has been performed for this study or if they re-used the genotypes obtained in the Rockett's study.

d) Why did the authors use controls recruited to an ongoing genetic cohort that were representative of the general population (Williams et al., 2009) as indicated in the subsection “The Kilifi case-control study” rather than used the controls of the previous study (Rockett et al.) that were also representative of the general populations? Is there a scientific reason? Are the Rockett controls not relevant? It would be interesting if the authors performed statistical analysis with the Rockett controls to confirm the results.

e) According to the sample size even if the p-value was lower than 0.05 and due to multiple-testing, the statistical results indicated only suggestive associations. As the p-value is borderline, these findings need to be replicated in an independent sample set to be convincing.

f) In the Discussion, the authors indicated that no significant association was seen with CR1 polymorphisms in the multi-centre candidate gene study (Rockett et al.), but the cases of the present study come from the Rockett's study and were compared to another control group. This information is confusing. It would be very important to confirm the association by using the Kenya controls from the Rockett's study to exclude potential selection bias in the control group. To perform this analysis, the authors should also take into account the complex interactions between Sl2, McC and α+thalassaemia detected in the present study.

4) This reviewer's primary struggle with this paper is that it does not fully address how the known interaction between α-thalassemia and Sickle may be influencing or overlapping with these results. As the raw data are not included in the supplementary for these loci, the reader is forced to interpret these results with only the models and genotype frequencies that the authors have presented. This work uses substantially larger populations with better exposure controls than past reports on this topic, and thus after addressing several concerns, this work will be a valuable contribution to the community.

5) What is the population breakdown for the combined genotype that includes α-thalassemia, Sickle, Sl, and Mc together in each patient? Ideally this would also be broken down by disease outcome as well.

6) The binary parameterization of α-thalassemia described in Appendix 1 subsection “The Sl2/Sl2 genotype is associated with protection against cerebral malaria in Mali” would appear to be miss-specified in light of Williams, 2005. It would seem more appropriate to model the double deletion vs. single or normal; or perhaps better in the early models to allow all three categories to compete for Sickle and Swain interactions. During model building the α-thalassemia-Sickle interaction may have been missed because of this, and it makes it difficult to determine whether the α-thalassemia-SL interaction is independent. Although the main effect term for Sickle is included in downstream adjusted models, there is not enough information to assess whether this factor would account for the α-thalassemia-Sickle interaction.

7) Cerebral malaria numbers are reported for the Mali cohort in Supplementary file 2, however associations with Knops are not discussed. It appears that these numbers may conflict with the trends in Kenya, and yet rosetting is still tested against these genotypes and presented as a potential mechanism of protection. Please reconcile this section or consider limiting the mechanistic connection to citing Rowe, et al.

Essential revisions:

1) Given the dramatic reductions in severe malaria and deaths in the past decade, it is a bit strange that the studies were done with data collected during 2001-08 (case control study), and 1998-2001 (cohort study). So, a decade later, the authors have gone back to analyze stored samples. The implications of this are not clear and need to be clearly addressed. Are there time trends that can confound the associations? What about use of old samples and sample quality? Were the hypotheses pre-stated or generated post-hoc? (presumably the latter).

Making full use of existing datasets is of major importance in malaria genetic epidemiological studies, and there is no evidence to suggest that malaria phenotype/genotype associations vary over time. Historic datasets represent a huge investment of time and funding in establishing and collecting appropriate samples under logistically challenging conditions. Moreover, in many of the institutions with the greatest capacity for the ethical collection and processing of carefully phenotyped patient samples, the rates of severe malaria have declined in recent years. This is evident in some of the traditionally low-moderate malaria transmission areas of sub-Saharan Africa such as Kilifi in Kenya, the main focus of the current study. Unfortunately, most areas with continuing high malaria transmission and associated high mortality do not yet have the scientific infrastructure necessary to carry out large, high quality case-control studies. Hence, it is currently difficult to perform epidemiological studies on malaria using newly collected samples. We briefly describe these issues in the “Datasets studied” section of the Materials and methods which states:

“Historic datasets (i.e. >10 years old) are widely used in genetic epidemiological studies of malaria due to the logistical challenges of sample collection in malaria endemic countries and the changing epidemiological patterns of disease.”

All the samples used were of high quality and underwent stringent quality control. The “Sample processing and quality control for the Kenyan case-control study” section in the Materials and methods has been expanded to cover the quality control steps employed in sample collection and genotyping.

The hypotheses addressed in this study were generated after collection of the specimens, but prior to any exploratory analysis of the relevant datasets.

2) As regards the case-control study, I would need more convincing that cases and controls were derived from the same study base. The difference in age distribution will likely not matter for the genetic mutations, but what about implications for disease severity?

The cases and controls studied here were all derived from exactly the same area, which is the geographic region defined by the Kilifi Health and Demographic Surveillance System (KHDSS) (Scott et al., 2012). This is described in the Materials and methods, subsection “The Kenyan study area”, and a precise description of the cases and controls is described in an expanded section in the Materials and methods, subsection “The Kenyan case-control study”.

The controls were infant samples that were used to estimate the relevant genotype frequencies in the population from which the cases arose. This method has been widely used in African genetic association studies (e.g. Clarke et al., 2017; Busby et al., 2016; Band et al., 2013; Rockett et al., 2014) and is supported by general commentators (e.g. Rothman, 2014 and Vandenbroucke et al. International Journal of Epidemiology 2012). This is currently the most logistically feasible way of collecting sufficiently large numbers of control samples in many sub-Saharan African settings.

Text describing the rationale for using infant control samples has been added to the Materials and methods, subsection “The Kenyan case-control study”.

3) The authors indicated that the study included 5545 children from Kenya consisting in 1716 severe malaria cases and 3829 community controls.

a) The authors performed the analysis on a subset of the samples used previously by Rockett et al. (2014). The authors should clarify why they did not use the total sample set of 2268 severe malaria cases included in Rockett et al. and how the 1716 samples were selected.

We apologise for the confusion caused by the lack of detail on how our study design differs from Rockett et al. The differences between our study and Rockett et al. is now fully explained in the “Comparison between this study and Rockett et al.” section in the Materials and methods (also Figure 3). See also the comments below.

b) As indicated by the authors, 407 individuals are excluded as lived out with KHDSS (Figure 3). Why? The authors should clarify this point. DNA not available? It would be surprising because all CR1 genotype (rs17047661 and rs17047660), sickle cell genotype, a+thalassaemia genotype and ABO blood group were available in the previous study of Rockett for the Kenyan population.

We excluded the severe malaria cases from outside the KHDSS because all of the controls came from within the KHDSS. Using only cases and controls from exactly the same area allowed the use of location as a random effect in the final statistical model, which greatly improved model fit. This is now described in the “Comparison between this study and Rockett et al.” section in the Materials and methods.

We have repeated our analysis including the severe cases of subjects who lived outside of the KHDSS study area (Supplementary file 1K), and this gives the same overall result as that shown in the main text. This is stated in the Materials and methods subsection “Comparison between this study and Rockett et al. (Rockett et al., 2014)”.

c) It is not clear for me whether new genotyping has been performed for this study or if they re-used the genotypes obtained in the Rockett's study.

We re-used genotypes from the Rockett’s study. This is now explained in the “Sample processing and quality control for the Kenyan case-control study” section of the Materials and methods, which states “The Sl and McC polymorphisms were originally typed as part of a larger study by Rockett et al., which included case-control data from 12 global sites (Rockett et al., 2014).”

The only additional genotyping of the Kenyan samples that was done for our study was for sickle cell trait and α +thalassaemia, as described in the Materials and methods, subsection “Laboratory procedures”.

d) Why did the authors use controls recruited to an ongoing genetic cohort that were representative of the general population (Williams et al., 2009) as indicated in the subsection “The Kilifi case-control study” rather than used the controls of the previous study (Rockett et al.) that were also representative of the general populations? Is there a scientific reason? Are the Rockett controls not relevant? It would be interesting if the authors performed statistical analysis with the Rockett controls to confirm the results.

The controls in our study and the controls in Rockett et al. are the same. This is now explained in the Comparison between this study and Rockett et al.” section in the Materials and methods, which states:

“In both our study and Rockett et al. (Rockett et al., 2014), the control samples were identical and all came from within the KHDSS. Our study has 120 fewer controls than Rocket et al. (Rockett et al., 2014) due to missing genotypes, because we only used controls for whom full Sl, McC, sickle cell genotype,α+thalassaemia genotype and ABO blood group data were available.”

e) According to the sample size even if the p-value was lower than 0.05 and due to multiple-testing, the statistical results indicated only suggestive associations. As the p-value is borderline, these findings need to be replicated in an independent sample set to be convincing.

We agree that there is evidence, but not unequivocal evidence, for the suggested associations. The protective association between Sl2 and cerebral malaria was first reported in a small case- control study from western Kenya (Thathy et al., 2005), but most subsequent studies have been underpowered. Our study is thus the first adequately powered independent sample set that replicates the protective association between Sl2 and cerebral malaria. Furthermore, our study shows for the first time that McC and α+thalassaemia influence the protective association of Sl2 on cerebral malaria risk. Hence, our study contributes significantly to this debate and provides novel insights that may explain previous conflicting findings on CR1 polymorphisms and malaria.

We agree that further replication of our novel findings will be required. Unfortunately, because α+thalassaemia results from a deletion mutation, not a SNP, it cannot be detected on classical SNP-genotyping platforms, and has to be performed manually using labour intensive PCR-based methods (Chong et al., 2000), as in our study. The attempts of our collaborators in the MalariaGen consortium to impute α+thalassaemia genotypes from existing SNP genotyping data have not yet been successful. Hence, currently, to our knowledge, there are no other large sub-Saharan case-control studies that are typed for α+thalassaemia and Sl and McC polymorphisms that could be used for replication. Future work will address this need, but this is beyond the scope of the current manuscript, which is already complex, incorporating as it does results from two different epidemiological studies and functional data.

We have added text discussing the important issues of replication in the second and fourth paragraphs of the Discussion.

f) In the Discussion, the authors indicated that no significant association was seen with CR1 polymorphisms in the multi-centre candidate gene study (Rockett et al.), but the cases of the present study come from the Rockett's study and were compared to another control group. This information is confusing. It would be very important to confirm the association by using the Kenya controls from the Rockett's study to exclude potential selection bias in the control group. To perform this analysis, the authors should also take into account the complex interactions between Sl2, McCb and α+thalassaemia detected in the present study.

As stated in response to 3d above, the controls used in our study are the same as the controls used in Rockett’s study.

As explained in response to 3a and 3b above, our study excludes some of the cases included by Rockett et al., due to their residence outside of the KHDSS. However, re-analysis including the additional cases studied by Rockett et al. (Supplementary file 1K) returns the same results as those shown in our main text. In other words, the difference in inclusion criteria does not explain why our results differ from Rockett.

The key distinctions that explain why we obtain a different result to Rockett et al. are:

i) We include Sl and McC in the same statistical model and adjust for confounding factors, including the statistically significant interaction with α+thalassaemia, whereas Rockett et al. looked at each SNP in isolation. This is now stated in the Materials and methods, subsection “Comparison between this study and Rockett et al. (Rockett et al., 2014)”, last paragraph.

The rationale for examining the effects of Sl and McC together in the same statistical model is that the two SNPs are only 33 bp apart in the CR1 gene, and each causes a change in amino acid charge (Sl: rs17047661 R1601G and McC: rs17047660 K1590E) with potential to impact upon the structure and function of the CR1 protein. Hence, it is biologically plausible that Sl might influence McC and vice versa.

ii) We are examining data from a single region in Kenya, whereas Rockett et al. analysed pooled data from 12 global sites. They state in their paper “…it is undoubtedly also the case that authentic genetic associations might be missed in multicentre studies if there is heterogeneity of effect across different study sites”.

Given that our data suggest that McC and α+thalassaemia influence the protective association of Sl2 with cerebral malaria, and that the frequencies of McC (Figure 2) and α+thalassaemia are known to vary geographically, it would be predicted that the association of Sl2 with cerebral malaria will vary by location, and may be missed by pooling data across different study sites.

We have summarised these arguments in the third and fourth paragraphs of the Discussion.

4) This reviewer's primary struggle with this paper is that it does not fully address how the known interaction between α-thalassemia and Sickle may be influencing or overlapping with these results. As the raw data are not included in the supplementary for these loci, the reader is forced to interpret these results with only the models and genotype frequencies that the authors have presented. This work uses substantially larger populations with better exposure controls than past reports on this topic, and thus after addressing several concerns, this work will be a valuable contribution to the community.

The potential effect of the α-thalassemia/sickle interaction on our results is answered in detail in a new section “Exploration of the negative epistasis between sickle trait and α+thalassaemia” in Appendix 2, and in new tables (Supplementary file 1N, 1O and 1P). See also comments below.

5) What is the population breakdown for the combined genotype that includes α-thalassemia, Sickle, Sl, and Mc together in each patient? Ideally this would also be broken down by disease outcome as well.

We have added data for the combined genotypes compromising sickle trait, α+thalassaemia and Sl for each clinical outcome in Supplementary file 1P.

6) The binary parameterization of α-thalassemia described in Appendix 1 subsection “The Sl2/Sl2 genotype is associated with protection against cerebral malaria in Mali” would appear to be miss-specified in light of Williams, 2005. It would seem more appropriate to model the double deletion vs. single or normal; or perhaps better in the early models to allow all three categories to compete for Sickle and Swain interactions. During model building the α-thalassemia-Sickle interaction may have been missed because of this, and it makes it difficult to determine whether the α-thalassemia-SL interaction is independent. Although the main effect term for Sickle is included in downstream adjusted models, there is not enough information to assess whether this factor would account for the α-thalassemia-Sickle interaction.

The parameterization of α+thalassaemia used in our model i.e. no α+thalassaemia alleles (normal) vs. one or more α+thalassaemia alleles was chosen in accordance with a previous report showing that both heterozygous (single deletion) and homozygous (double deletion) α+thalassaemia genotypes are associated with protection against severe malaria and death in the Kilifi area (Williams, Wambua, et al., 2005). A description of the parameterization of theα+thalassaemia variable can now be found in the Materials and methods, subsection “Statistical analysis”, second paragraph.

Investigation of interaction between sickle trait and α+thalassaemia is now described in the “Exploration of the negative epistasis between sickle trait and α+thalassaemia” section in Appendix 2. Analysis of our dataset confirms the existence of negative epistasis between sickle trait and α+thalassaemia in this population (Supplementary file 1N). However, of interest, this negative epistatic interaction was only seen in the severe malaria cases without cerebral malaria, whereas the α+thalassaemia/Sl interaction was specific to cerebral malaria cases. Therefore, the two interactions appear to be mutually exclusive. Therefore, we can confirm that for cerebral malaria in this dataset, no significant interaction exists between sickle trait and α+thalassaemia. This is the case no matter which α+thalassaemia parameterisation is used.

7) Cerebral malaria numbers are reported for the Mali cohort in Supplementary file 2, however associations with Knops are not discussed. It appears that these numbers may conflict with the trends in Kenya, and yet rosetting is still tested against these genotypes and presented as a potential mechanism of protection. Please reconcile this section or consider limiting the mechanistic connection to citing Rowe, et al.

There is no conflict between the Mali and Kenya data. A protective association between the Sl2/Sl2 genotype and cerebral malaria was identified on analysis of the Mali dataset (aOR 0.35, 95% CI 0.12-0.89, p=0.024) and the Sl2/Sl2-McC genotype combination was associated with protection against cerebral malaria (aOR 0.14, 95% CI 0.02-0.84, p=0.031). As such, we considered blood samples from this population to be appropriate for testing rosetting as a potential mechanism of action.

To make this clearer we have now included a sentence describing the protective association between Sl2 and cerebral malaria in the Mali case-control to the Results, subsection “The Sl2 allele was associated with reduced ex vivo rosette frequency in P. falciparum clinical isolates from Mali”.

Appendix 1—table 1.

General characteristics for cases and controls in the Mali case- control study.

	Controls (Uncomplicated malaria)	Cases (Cerebral malaria)	P value
Mali	n = 184	n = 34
Gender Males Females	90 (49 %) 94 (51 %)	17 (50 %) 17 (50 %)	0.920
Ethnicity Dogon Non-Dogon	161 (87.5 %) 23 (12.5 %)	30 (88 %) 4 (12 %)	1
Age in months* Median (IQR)	36.5 (19–56)	28 (16–41)	0.026

Comparisons performed using Pearson’s χ2 test except *Kruskal Wallis test.

Appendix 1—table 2.

Table of Characteristics for the Malian dataset by Sl and McC genotype.

		Sl1/Sl1	%	Sl1/Sl2	%	Sl2/Sl2	%	P value	McCa/McCa	%	McCa/McCb	%	McCb/McCb	%	P value
Clinical status	Uncomplicated malaria	19	10.3	78	42.4	87	47.3		103	56.0	61	33.2	20	10.9
	CM	5	14.7	19	55.9	10	29.4	0.135#	16	47.1	15	44.1	3	8.8	0.478#
Gender	M	11	10.3	48	44.9	48	44.9		57	53.3	41	38.3	9	8.4
	F	13	11.7	49	44.1	49	44.1	0.945	62	55.9	35	31.5	14	12.6	0.430
Ethnicity	Dogon	21	11.0	88	46.1	82	42.9		108	56.5	66	34.6	17	8.9
	Non-Dogon	3	11.1	9	33.3	15	55.6	0.448#	11	40.7	10	37.0	6	22.2	0.078
Blood group	O	13	10.0	63	48.5	54	41.5		68	52.3	49	37.7	13	10.0
	Non-O	11	12.5	34	38.6	43	48.9	0.355	51	58.0	27	30.7	10	11.4	0.566
Median age	Months	32	IQR 20–57.5	31	IQR 18–52	30	IQR 17–55	0.927	30	IQR 17–52.5	31	IQR 18–54.5	51	IQR 29–64.5	0.229

CM, cerebral malaria; IQR, interquartile range. Significance testing uses uncomplicated malaria as the reference group. Age: Kruskal Wallis test. #: Fisher’s exact test. All other analyses: Pearson’s χ2.