The SLC4A1 gene is under differential selective pressure in primates infected by Plasmodium falciparum and related parasites

Abstract

Malaria is a disease caused by Plasmodium parasites and is responsible for high mortality in humans. This disease is caused by four different species of Plasmodium though the main source of mortality is Plasmodium falciparum. Humans have a number of genetic adaptations that act to combat Plasmodium. One adaptation is a deletion in the SLC4A1 gene that leads to Southeast Asian ovalocytosis (SAO). There is evidence that SAO erythrocytes are resistant to multiple Plasmodium species. Here we analyze SLC4A1 in 23 primates and mammals to test for differential selective pressures among different primate lineages. Because primates are infected with both human Plasmodium parasites and their relatives, this analysis can be used to test which human Plasmodium parasite is the likely target of SAO. A significantly different pattern of molecular evolution was found in humans and African apes, species that are infected by P. falciparum and its relatives. This effect was restricted to the cytosolic domain of the SLC4A1 gene. The evidence is consistent with a different selective regime operating on this gene domain in humans and African apes, when compared to other primates and mammals. Alternatively, this pattern is consistent with a relaxation of selection or weak adaptive evolution operating on a small number of amino acids. The adaptive interpretation of the results is consistent with the SAO allele of the SLC4A1 gene interacting with P. falciparum in humans, rather than other Plasmodium parasites. However, additional investigation of the relationship between SLC4A1 variants and Plasmodium in humans and African apes is required to test whether the different selective regime in humans and African apes is due to natural selection or relaxed constraint.

1. Introduction

Malaria is a parasitic disease responsible for approximately 1 million human deaths annually (WHO, 2008). This disease is caused by four different species of Plasmodium parasite: P. falciparum, P. vivax, P. malariae, and P. ovale. The main source of human mortality is the P. falciparum parasite, though P. vivax infections are also responsible for considerable levels of morbidity (Mendis et al., 2001). A number of genetic traits have evolved by natural selection to mitigate this strong selective agent in humans (Kwiatkowski, 2005). One malarial adaptation is apparently related to Southeast Asian Ovalocytosis (SAO) (Baer et al., 1976). SAO is a condition caused by a mutation in the solute carrier, anion exchanger, member 1 gene, SLC4A1, which encodes the band 3 protein, a major component of erythrocytes (Liu et al., 1990). Band 3 has two main functional domains (Fig. 1). The N-terminal intracellular region plays a role in the structure of the erythrocyte (Jay, 1996) and the C-terminal transmembrane region plays a role in anion exchange (Tanner, 1993). The SAO-causing mutation is a 27 base pair deletion in SLC4A1 that removes amino acids 400–408, a region at the junction of the functional domains of the protein. This deletion causes an increased rigidity of erythrocytes (Jarolim et al., 1991; Mohandas et al., 1992). The prevalence of the SAO allele can be as high as 35% in malarial areas (Mgone et al., 1996), suggesting that there is an extremely high selective benefit for the heterozygote carriers of the deletion allele, given that homozygosity is apparently lethal (Liu et al., 1994). The evidence is clear that SAO chiefly protects individuals against severe cerebral malaria (Allen et al., 1999; Genton et al., 1995), a form of the disease caused almost exclusively by P. falciparum. Although it was originally thought that SAO worked via a reduction in cytoadherence (Allen et al., 1999), Cortés et al. (2005) suggested that the protective mechanism of SAO is actually an increased cytoadherence of infected erythrocytes through the CD36 protein.

Fig. 1.

Schematic organization of the human SLC4A1 gene, showing all translated exons (2–20). The cytosolic and transmembrane domains are shown. The external residues of the transmembrane are indicated with the letter “E.” To enhance legibility, these “E” notations approximate the codon lengths of the external residues but are not exact.

Aside from a link between SAO and malaria, there is also evidence that band 3 has a more general role in erythrocyte invasion by Plasmodium. Though the major erythrocyte receptor for P. falciparum invasion is glycophorin A (GYPA) (Pasvol et al., 1982; Sim et al., 1994), the parasite is not solely dependent on this protein (Okoyeh et al., 1999). Indeed, P. falciparum uses a suite of surface receptors for erythrocyte invasion, some of which are yet to be fully characterized (reviewed in Baum et al., 2005). The band 3 protein has been forwarded as an alternate receptor for Plasmodium that is sialic acid-independent (Goel et al., 2003; Kariuki et al., 2005; Li et al., 2009, 2008, 2004; Okoye and Bennett, 1985). Finally, aside from its role in the erythrocyte, a shorter isoform of SLC4A1 is expressed in the kidney where it plays a role in acid secretion; mutations in the gene can lead to distal renal tubular acidosis (reviewed in Williamson and Toye, 2008).

While SAO clearly targets P. falciparum-caused severe malaria (Allen et al., 1999; Genton et al., 1995), there is some evidence that the SAO allele may protect against other Plasmodium parasites. SAO erythrocytes were shown to be relatively resistant to parasitemia or invasion by multiple Plasmodium parasites (Cattani et al., 1987; Foo et al., 1992; Hadley et al., 1983; Kidson et al., 1981; Serjeantson et al., 1977). However, these findings may have been artifactual due to methodological concerns in the handling of SAO cells (Bruce et al., 1999; Dluzewski et al., 1992). It also is possible that SAO may only be effective against erythrocyte invasion in particular strains of P. falciparum (Cortés et al., 2004), suggesting the efficacy of SAO is restricted even within this species. However, Allen et al. (1999) found that SAO did not effect erythrocyte invasion by P. vivax or P. falciparum, but instead caused resistance to cerebral malaria. Though cerebral malaria is most closely associated with P. falciparum, P. vivax can cause cerebral and other forms of severe malaria (Kochar et al., 2005; Rogerson and Carter, 2008). Given this, it is possible that SAO may be an adaptation to cerebral malaria, which can be caused by different Plasmodium species. A general relationship between SLC4A1 and Plasmodium is supported by the conservation of the extracellular portion of the Plasmodium PfSPP protein, which may interact with band 3 (Li et al., 2009).

Here we examine the pattern of molecular evolution at SLC4A1 in humans, primates, and other mammals to determine whether this gene is evolving adaptively in primates that harbor P. falciparum related parasites or primates that harbor Plasmodium parasites generally. A pattern of differential selection at the SLC4A1 gene in primates with particular Plasmodium infections will help to ascertain which malarial parasites are being targeted by this locus. Primates are useful models for testing between these hypotheses because they are infected with many different Plasmodium species (Table 1). Critically, it is now well established that wild African apes (chimpanzees, bonobos, and gorillas) are naturally infected with the human parasite P. falciparum and a series of its close phylogenetic relatives (Krief et al., 2010; Liu et al., 2010; Ollomo et al., 2009; Prugnolle et al., 2010; Rich et al., 2009). Phylogenetic relatives of the other human parasites (P. vivax, P. malariae, and P. ovale), on the other hand, infect a larger and more diverse set of primates (Hayakawa et al., 2008; Perkins and Schall, 2002). For this study, it is especially noteworthy that the P. falciparum group infects a relatively small set of primates and these parasites are only distantly related to the other human Plasmodium parasites (Hayakawa et al., 2008).

Table 1.

Species examined, sample information, accession numbers, and Plasmodium infection status. (See below-mentioned references for further information.)

Host Group	Host Species	Common Name	Coriell Sample ID	Accession Number	Genome Project	Genome Sequence Coordinates	Infections from P. falciparum and similar spp.	Other infecting Plasmodium spp.
Humans	Homo sapiens	Human			NG_007498.1 †		P. falciparum	P. vivax, P. malariae, P. ovale
African Apes	Pan paniscus	Bonobo	NG05253	HM065568			P. falciparum (Krief et al. 2010)	Spp. related to P. malariae (Krief et al. 2010)
	Pan troglodytes	Chimpanzee	NG06939	HM065581	panTro2	Chr 17:13296154-13309268	P. reichenowi, P. gaboni, unnamed parasite spp. (Krief et al. 2010; Ollomo et al. 2009; Rich et al. 2009; Prugnolle et al. 2010)	P. rodhaini, P. schwetzi, unnamed parasite spp. (similar to P. malariae, P. vivax and/or P. ovale, Coatney 1971, Duval et al. 2009, Krief et al. 2010, Hayakawa et al 2009)
	Gorilla gorilla	Gorilla	NG05251	HM065569			P. falciparum, unnamed parasite spp. (Prugnolle et al. 2010)	P. rodhaini, P. schwetzi (similar to P. malariae, P. vivax and/or P. ovale, Coatney 1971)
Asian Apes	Pongo pygmaeus	Orangutan	NA04272	HM065570			none	P. silvaticum, P. pitheci (similar to P. vivax, Wolfe 1999)
	Hylobates gabriellae	Yellow Cheeked Gibbon	PR00381	HM065571			none	P. hylobati (similar to P. vivax, Perkins and Schall 2002, Hayakawa et al. 2008)
Old World Monkeys	Macaca arctoides	Stump Tailed Macaque	GM03443	HM065572			none	Multiple spp. (similar to P. vivax, Perkins and Schall 2002, Hayakawa et al. 2008)
	Macaca mulatta	Rhesus Macaque	NG07109	HM065580	rheMac2	Chr 16:54355363-54341551	none	Multiple spp. (similar to P. vivax, Perkins and Schall 2002, Hayakawa et al. 2008)
	Colobus guereza	Black and White Colobus	PR00980	HM065573			none	none
	Papio anubis	Baboon	PR00036	HM065574			none	Hepatocystis (silimar to P. ovale, Perkins and Schall 2002)
New World Monkeys	Callithrix jacchus	Marmoset			calJac1	Contig103:955511- 939806	none	none
	Callicebus moloch	Dusky Titi Monkey	NG06115	HM065575			none	P. brasilanum (similar to P. malariae, Escalante et al. 1995)
	Lagothrix lagothricha	Wooly Monkey	NG05356	HM065576			none	”
	Saguinus labiatus	White Lipped Tamarin	NG05308	HM065577			none	”
Strepsirhines	Otolemur garnettii	Small Eared Galago	PR00049	HM065578			none	none
	Lemur catta	Ring Tailed Lemur	ID#6351*	HM065579			none	none
	Microcebus murinus	Gray Mouse Lemur			ensembl	Genescaffold1055:63426-52166	none	none
Rodents	Mus musculus	Mouse			mm9	Chr 11:102222709 102211601	none	Murinae infected with Plasdmodium that is outgroup to a a larger P. vivax, P. malariae, P. ovale, Hepatocystis etc. group, Perkins and Schall 2002, Hayakawa et al. 2008)
	Rattus norvegicus	Rat			rn4	Chr 10:91467178-91456310	none	”
Other Mammals	Bos taurus	Cow			bosTau4	Chr 19:45496100-45485259	none	none
	Canis familiaris	Dog			canFam2	Chr 9:22463842-22475102	none	none
	Equus caballus	Horse			equCab2	Chr 11:19271365-19283645	none	none
	Monodelphis domestica	Opossum			monDom5	Chr 2:204285378-204264654	none	none

To address the conflicting evidence as to whether SAO is an adaptation to P. falciparum specifically or to Plasmodium generally, we probe the relationship between band 3 and Plasmodium by analyzing the coding region of SLC4A1 in 23 mammals. The following hypotheses are tested. First, we test whether SLC4A1 has been evolving adaptively across all 23 species. Second, we test whether patterns of molecular evolution at SLC4A1 differ between primate species harboring P. falciparum related parasites (humans and African apes) and those harboring other Plasmodium parasites (Table 1; Fig. 2). Third, we test whether specific structural and functional domains of SLC4A1 are evolving differently from one another.

Fig. 2.

Phylogeny of the species examined in the present study. To the left of the species names, the phylogenetic tree highlights the lineages inferred to have harbored P. falciparum and related parasites (see Table 1). To the right of the species names, the phylogenetic tree highlights the lineages inferred to have harbored any species of Plasmodium, except for rodents (see Section 2 and Table 1). These bold branches are those used by PAML to enable the branch tests and branch site tests as described in the Methods.

2. Methods

2.1. Samples

The primate samples outlined in Table 1 were obtained from the Coriell Institute and the Duke Lemur Center. These were targeted for PCR and sequencing of the entire coding region of the SLC4A1 gene as described in the following sections. These sequences were added to existing DNA sequences from other primates and non-primate mammals from the UCSC Genome Center (Kent et al., 2002), Genbank (Benson et al., 2009), and Ensembl (Hubbard et al., 2009) (Table 1).

2.2. PCR and sequencing

The region targeted for sequencing in these primates was the region encompassed by the human SLC4A1 gene (NG_007498.1). The entire coding region was sequenced, including all translated exons and intervening introns. In general, large overlapping regions of between 1 and 5 kb were targeted by PCR, amplicons were gel purified, these amplicons were subsequently cloned, clones were screened for the presence of the desired insert, and positive clones were sequenced in both directions. Occasionally, small fragments were sequenced directly from purified PCR products. PCRs were done using the Eppendorf Triplemaster and 5 Prime PCR Extender kits using a large number of primer sets and PCR parameters that were often species specific and fragment specific (protocol details and primer sequences available upon request). Both kits utilize a proofreading enzyme. Gel purification was done using the Eppendorf Perfectprep Gel Cleanup kit and TOPO XL Gel Purification kit. PCR purification was done using Millipore Montage Columns. Amplicons were cloned via Invitrogen TA and XL cloning kits. Plasmid DNA from multiple colonies was miniprepped using the Eppendorf FastPlasmid kit and screened for the insert via clone-test PCR and/or EcoRI restriction digestion. Plasmids were sequenced with the M13R and T7 sequencing primers, as well as PCR primers. DNA sequencing was performed by the Dana-Farber/Harvard Cancer Center (DF/HCC) High-Throughput DNA Sequencing Facility, Macrogen Sequencing Services, and Genewiz. Clones positive for the desired insert were sequenced in both directions for the entire sequence using multiple primers (sequencing primer sequences available upon request).

2.3. Sequence alignment and analysis

The codon sequences were aligned by eye to the human exonic sequences. This approach was satisfactory due to the relatively constrained exonic structure across the mammals studied. However, a limited number of exons were aligned using backtranslation from amino acid sequences (Bininda-Emonds, 2005). Codon sequences were analyzed for the action of positive natural selection by maximum likelihood using the codeml program of the PAML package (v. 4) (Yang, 2007). In each test, the F3x4 codon model was used, κ was estimated, and a user tree was supplied. The user tree (Fig. 2) was based on a generally accepted mammalian and primate phylogeny (Goodman et al., 1998; Murphy et al., 2004; Wildman et al., 2009). This was used to generate likelihood values under a number of different models. The convergence of the likelihood values was assessed by running each analysis multiple times. These likelihoods were compared using a series of nested likelihood ratio tests (LRTs) (Yang, 2006). In a LRT, a significant difference between models is assessed by the test statistic 2ΔlnL, which follows a χ² distribution with degrees of freedom equal to the difference between the numbers of parameters in the competing models. When a significant difference is detected, the model with the higher likelihood is favored. If a significant difference is not detected, the simpler model is preferred, i.e. the one with fewer parameters. Three types of analyses (Yang, 2006) were conducted to determine that pattern of adaptive molecular evolution at the SLC4A1 gene, ‘branch tests’, ‘site tests’, and ‘branch site tests’.

In the branch tests, a LRT is used to determine whether sequences across the entire tree were evolving neutrally by comparing the likelihood of a model where d_n/d_s (or ω) is fixed at 1 (the neutral model) against a model where the most likely value for ω is estimated. Subsequently, a model where ω was estimated was compared to two alternative models (Fig. 2). In both models, ω values are estimated for two sets of branches. In the P. falciparum model, the two sets of branches are (1) those primates infected with P. falciparum and P. falciparum-related parasites and (2) the remaining branches. In the Plasmodium model, the two sets of branches are (1) those primates infected with any Plasmodium species and (2) the remaining branches (not infected with any malarial parasite). In these models, we did not include stem branches among those being infected because some of these branches are exceptionally long in the primate phylogeny. It is not clear when along these long lineages that Plasmodium transferred into them. Therefore, it is more conservative not to include these lineages. Also, for the Plasmodium model, we infer three origins of Plasmodium infection in primates, one each in apes, Old World monkeys, and New World monkeys. An alternative would be to infer that primates are ubiquitously infected with Plasmodium, with infection being lost in Callithrix, Colobus, and lemurs and lorises. However, there is no evidence to suggest that Plasmodium is common in all of the main primate groups. This is especially true in lemurs, where infections have only been detected in Eulemur (Ellegren, 2008) and in a single specimen of Propithecus, despite sampling 55 samples from 6 lemur genera (Duval et al., 2010). Further information on the Plasmodium infections of the species sampled is presented in Table 1. In this analysis, Papio was included due to its infection with Hepatocystis, a relative of Plasmodium and known selective force (Tung et al., 2009). While rodents of the subfamily Murinae are infected by Plasmodium, Mus musculus and Rattus norvegicus were not included because these two species themselves are not infected.

Because multiple branch tests are being done (two alternative models vs. a null model) using the same data, a correction for multiple tests is necessary (Anisimova and Yang, 2007). There are a number of correction methods, which all essentially adjust the critical value (α) to reduce the type I error rate. Anisimova and Yang (2007) examined a number of such tests regarding the branch tests used here and determined that many methods were able to control error rates. Here, we used the Bonferroni correction, which is very conservative. The Bonferroni correction adjusts α by the number of tests. In the present case we are using two tests, so the adjusted α value is 0.025. These branch tests were conducted on four different functionally relevant partitions of exons. These are the entire gene (911 AA), the cytosolic domain (403 AA in humans), the transmembrane domain (508 AA), and the external sites, a subset of transmembrane amino acids that occur on the cell surface (122 AA) (from Bruce, 2006). The inferred nine amino acid insertion found in African apes was not included in the selection tests. In all cases, the number of the codon given corresponds to that from the human SLC4A1 gene. Here, the tests of the different partitions are a refinement used to determine which regions of the gene are under selection. In this regard, the tests across these different partitions is considered a test of the robustness of the findings are do not require a correction for multiple tests.

Site tests determine if particular amino acid sites have experienced positive selection across the entire phylogeny (Yang, 2006). In these cases, three site models are compared. One is a neutral model with one ω estimated for all sites, the second is a nearly neutral model (with two ω classes, one evolving neutrally and a second evolving under purifying selection), and a third is a model with three ω classes (one neutral (ω = 1), one purifying (ω < 1), and one for sites evolving under positive selection (ω > 1)). (These are models M0, M1, and M2 of PAML.) A related comparison fits 10 site classes for ω between 0 and 1, using a beta distribution (M7) and compares this to a model that also includes a site class with ω > 1 (M8). LRTs that accept models allowing for a class of selected sites indicate the action of positive selection; Bayes empricial Bayes (BEB) (Yang et al., 2005) posterior probabilities are then used to determine the probabilities of those sites. These tests were conducted on the 4 different functionally relevant partitions of exons, as described above. In this regard, the tests across these different partitions is considered a test of the robustness of the findings are do not require a correction for multiple tests.

Branch site tests determine whether particular sites are evolving adaptively on particular lineages (Yang, 2006). Here, partitions of branches are the same as in the branch tests (Fig. 2). For the branch site test, the model estimates parameters for these two sets of branches. For the non-Plasmodium affected branches, two site classes are estimated, those under neutral evolution (ω = 1) and purifying selection (ω < 1). The lineages harboring Plasmodium parasites have these two site classes, and importantly also have an additional site class of positively selected sites. The LRT is between a model where this final site class is neutral (ω = 1) and one under positive selection (ω≥1). These tests were conducted on the 4 different functionally relevant partitions of exons, as described above. In this regard, the tests across these different partitions is considered a test of the robustness of the findings are do not require a correction for multiple tests. However, the tests between the two sets of branches do require a correction for multiple tests, implemented as described above.

3. Results

3.1. Alignment

We collected SLC4A1 DNA sequences from 12 primates (Table 1). These sequences were added to publicly available DNA sequences from a range of primate and non-primate mammals. For two primates, we collected additional DNA sequences to cover unknown regions in the available genome sequences. In total, the alignment of SLC4A1 exons comprised 23 species and was 2898 basepairs in length. One noteworthy aspect of the alignment was a potentially exonic 27 basepair insertion in the African apes, but not in humans (Fig. 3). This insertion is in frame and it is likely to be coding due to its pattern of intron–exon splice sites. The African ape insertion obliterates one splice site but has added another. It is also unusual to find an insertion with this pattern because it suggests that this insertion was gained in the African ape ancestor and then subsequently lost in humans because of the phylogenetic relationships of humans, chimpanzees, and gorillas. Furthermore, the insertion and subsequent deletion was at the same position and length; also very unlikely. A second interpretation is that insertion occurred once in the ancestor of humans, chimpanzees, and gorillas. Subsequently, due to ancestral polymorphism and incomplete lineage sorting the derived condition is found in chimpanzees and gorillas, while the ancestral condition is found in humans. Because incomplete lineage sorting is a known problem within humans and African apes (Ruvolo, 1997), we find the second interpretation to be more likely for this short region. When the entire alignment is analyzed phylogenetically, the accepted phylogeny for apes is obtained, suggesting that the incomplete lineage sorting only affects this small region. Therefore, we used the accepted primate phylogeny in subsequent analyses requiring a user-generated tree. The alignment is available upon request.

Fig. 3.

Detail of the insertion at the 3′ end of exon 3 that is reconstructed to have occurred in African Apes. There is clear homology among the sites in exon 3, the insertion, and following intron. The canonical splice sites are highlighted in gray; the obliterated splice site is highlighted in black. The in frame inferred amino acid sequence of the insertion is also given.

3.2. Analyses of adaptive evolution across mammals

Before examining patterns of evolution specifically within primates harboring different Plasmodium species, the pattern of adaptive evolution over the entire mammalian phylogeny was examined. Multiple data partitions were analyzed: the entire gene (911 AA), the cytosolic domain (403 AA in humans), the transmembrane domain (508 AA), and the external sites, a subset of transmembrane amino acids that occur on the cell surface (122 AA). As expected for a functional gene, none of the partitions were evolving neutrally (d_n/d_s = ω = 1); all had signatures of evolution consistent with purifying selection (Table 2) (ω ranged from 0.12 to 0.28). This level of purifying selection is consistent with average genome-wide estimates for ω in mammals (Ellegren, 2008). Comparing a model with a uniform ω across sites to a ‘nearly neutral model’ with two ω values for different sites (1 and <1) recovered a significant difference for all data partitions, showing that the amino acid sites are not evolving under a uniform neutral selective regime (‘Site tests’; Table 3). This ‘nearly neutral model’ was tested against a model that incorporates an additional class of sites under positive selection, and thus is able to detect particular amino acids under selection. The model incorporating positive selection was a significantly better fit for both the entire gene and the transmembrane partition. In these analyses, nearly 1% of the amino acids were found to be evolving with a ω of approximately 3, suggesting that a limited number of sites are evolving positively in SLC4A1. Two amino acids were identified as evolving under statistically significant positive selection (367,658). A related comparison, where ω is modeled using a beta distribution, found evidence for positively selected sites in multiple comparisons, including codon 658 and additional sites (Table 4). Overall, SLC4A1 is evolving in a manner consistent with a functional gene: the gene is under purifying selection and a limited number of sites are under positive selection.

Table 2.

Branch tests of adaptive evolution.

Model	N parameters	All sites		Cytosolic sites		Transmembrane sites		External sites
		lnL	ω	lnL	ω	lnL	ω	lnL	ω
A
(A) ω = 1	44	−18873.419	1.00	−9687.625	1.00	−9010.701	1.00	−2581.484	1.00
(B) Estimate one ω	45	−18039.469	0.19	−9412.785	0.28	−8413.696	0.12	−2462.885	0.20
(C) P. falciparum branches vs. others	46	−18036.433	0.37, 0.19	−9407.341	0.86, 0.27	−8413.530	0.09, 0.12	−2462.808	0.14, 0.209
(D) Plasmodium branches vs. others	46	−18039.340	0.21, 0.19	−9410.411	0.38, 0.26	−8412.150	0.09, 0.13	−2462.599	0.15, 0.20

Likelihood ratio tests	All sites		Cytosolic sites		Transmembrane sites		External sites
	2ΔlnL	P value	2ΔlnL	P value	2ΔlnL	P value	2ΔlnL	P value
B
Neutral evolution (A vs B (df = 1))	1667.9	0.000	549.7	0.000	1194.0	0.000	237.2	0.000
P. falciparum branches different (C vs B (df = 1))a	6.07	0.014b	10.9	0.001b	0.3	0.565	0.2	0.694
Plasmodium branches different (D vs B (df = 1))a	0.3	0.611	4.7	0.029c	3.1	0.079	0.6	0.449

A: parameters and likelihoods estimated for the different models.

B: likelihood ratio tests described in Methods and Results. Bold indicates P < 0.05.

^aA correction for multiple tests is used for C vs. B and D vs. B tests.

^bSignificant after Bonferroni correction for multiple tests.

^cNot significant after correction for multiple tests.

Table 3.

Site tests of adaptive evolution, fixed class model.

Model	N parameters	All sites		Cytosolic sites		Transmembrane sites		External sites
		ln L	ω Values	lnL	ω Values	lnL	ω Values	lnL	ω Values
A
(A) ‘M0’, one ω class	45	−18039.469	0.19	−9412.785	0.28	−8413.696	0.12	−2462.885	0.20
(B) ‘M1a’, two ω classes (<1,1)	46	−17475.971	0.07, 1.00	−9178.578	0.10, 1.00	−8172.866	0.05, 1.00	−2367.630	0.07, 1.00
(C) ‘M2a’, three ω classes (<1, 1,>1)	48	−17469.195	0.07, 1.00, 2.86	−9178.578	0.10, 1.00, 1.00	−8166.515	0.05, 1.00, 3.12	−2364.868	0.07, 1.00, 2.88

Likelihood ratio tests	All sites			Cytosolic sites			Transmembrane sites			External sites
	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites
B
Two ω classes vs. One ω class (A vs. B (df = 1))	1127	0	n/a	468.4	0	n/a	481.7	0	n/a	190.5	0	n/a
Two ω classes vs. Three ω classes (B vs. C (df = 2))	13.6	0.001	367-1a, 658	0	1	n/a	12.7	0.002	658	5.5	0.063	n/a

A: parameters and likelihoods estimated for the different models.

B: likelihood ratio tests described in Sections 2 and 3. Bold indicates P < 0.05.

^aThe position under selection is a gap in the human alignment corresponding to the codon before the 367th codon in human.

Table 4.

Site tests of adaptive evolution, beta distribution model.

Model	N parameters	All sites		Cytosolic sites		Transmembrane sites		External sites
		lnL	ω of selected class	lnL	ω of selected class	lnL	ω of selected class	lnL	ω of selected class
A
(A) ‘M7’, ω beta distribution	46	−17452.735		−9173.066		−8159.716		−2358.899
(B) ‘M8’, ω beta distribution with ω class < 1	48	−17424.343	1.59	−9164.167	1.25	−8134.548	2.04	−2352.06	1.79

Likelihood ratio tests	All sites			Cytosolic sites			Transmembrane sites			External sites
	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites	2ΔlnL	P value	Selected sites
B
Beta dist. For ω vs. Beta with a site class with ω < 1 (A vs. B (df = 2))	56.8	0	658	17.8	0	None	50.3	0	558, 560, 656, 658	13.7	0	658

These site tests are based on a beta distribution for site classes for ω values for each site. A: parameters and likelihoods estimated for the different models. B: likelihood ratio tests described in Sections 2 and 3. Bold indicates P < 0.05.

3.3. Analyses of adaptive evolution in primates harboring different Plasmodium parasites

Subsequently, we examined whether the lineages reconstructed to have harbored P. falciparum related parasites were evolving differently from lineages that are not reconstructed to have harbored these parasites (‘branch tests’; Fig. 2). When the entire coding region is tested, a model incorporating different ω values for these two sets of lineages is a better fit to the data than a model fitting only one ω to all lineages (P = 0.014; significant after correction for multiple tests) (Table 2). The ω of the lineages reconstructed to harbor P. falciparum-related species was higher (0.37) than the lineages reconstructed not to harbor these parasites (0.19). Of the different data partitions, only the cytosolic domain recovered a significant difference. In the cytosolic domain, the pattern of difference in ω was more marked (ω = 0.86 in P. falciparum related lineages vs. 0.27 in other lineages) than in the entire gene (P = 0.001; significant after correction for multiple tests). A phylogeny showing the reconstructed numbers of nonsynonymous and synonymous changes in humans and African apes reveals that all branches have higher absolute numbers of amino acid changes, except the P. troglodytes branch, which had no changes (Fig. 4).

Fig. 4.

Summary of the significant branch test for the cytosolic domain. This simplified phylogeny shows the two branch classes (species harboring relatives of P. falciparum vs. all other species) and their respective estimated ω values for the cytosolic domain. The codeml program was used (with Rateancestor = 1) to infer the number of nonsynonymous (numerator) and synonymous (denominator) along each branch. These estimates are shown above each branch. Gapped regions were not included in these estimates.

A similar test was done to test whether primates harboring any type of Plasmodium species were evolving differently from lineages that were not reconstructed to have harbored these parasites (Fig. 2). When the entire coding region is tested, a model incorporating different ω values for these two sets of lineages is not a better fit to the data than a model fitting only one ω to all lineages (Table 2). In the cytosolic domain there is an elevated ω ratio in the species harboring any Plasmodium species (ω = 0.38 vs. 0.26), though this finding was not significant after correction for multiple tests. None of the other partitions yielded significant results.

Finally, we tested whether particular sites were evolving adaptively in particular lineages (i.e. the ‘branch site test’). First, positive evolution was tested at sites on the lineages reconstructed to have harbored P. falciparum and its relatives (Table 5). These analyses showed evidence for sites under positive selection in the lineages harboring P. falciparum relatives in the cytosolic region, but only using a less conservative critical value for the LRT (P < 0.05) (Yang, 2006). In this less conservative test, one site (111) was found to be evolving positively in lineages reconstructed to have harbored P. falciparum. However, correction for multiple tests would render this comparison insignificant. Second, we examined the sites on the lineages reconstructed to have harbored any Plasmodium species. None of these comparisons recovered sites under positive natural selection.

Table 5.

Branch-site tests of adaptive evolution.

	Null model		Selection model		Likelihood ratio tests
	N parameters	lnL	N parameters	lnL	P value
A
All	47	−17473.9418	48	−17472.7279	0.119
Cytosolic sites	47	−9174.2083	48	−9172.7565	0.088a
Transmembrane sites	47	−8172.8663	48	−8172.8663	1.000
External sites	47	−2367.4868	48	−2367.4868	1.000
B
All	47	−17475.2612	48	−17473.9605	0.107
Cytosolic sites	47	−9176.0386	48	−9175.5231	0.310
Transmembrane sites	47	−8172.8663	48	−8172.4822	0.381
External sites	47	−2367.6301	48	−2367.6301	1.000

A: tests on lineages harboring P. falciparum and related parasites.

B: tests on lineages harboring any plasmodium parasites.

^aSee text for additional information regarding this comparison.

4. Discussion

The pattern of molecular evolution at the SLC4A1 gene offers insight into the differences in adaptations between primate species harboring different sets of Plasmodium parasites. Differential ω values were detected in humans and African apes—species harboring P. falciparum related parasites. The differential ω was most pronounced in the cytosolic domain of SLC4A1. Although the ω values for the cytosolic domain in humans and African apes are not greater than 1, the benchmark for positive evolution, they are three times higher than the values estimated for the rest of the tree (ω = 0.88 vs 0.27) and higher than average ω values estimated from mammalian genome-wide comparisons (Ellegren, 2008). This is a significant difference, despite the conservative nature of the test employed, and it shows that a shift in selective regimes has occurred in the African apes and humans at SLC4A1. A cautious interpretation of this finding is that the higher ω value in African apes is reflective of a relaxation of constraint or decreased purifying selection in the cytosolic domain. Less cautiously, these findings can be interpreted to support a hypothesis where P. falciparum and its relatives, found exclusively in humans and African apes, are a selective force acting on the cytosolic domain of SLC4A1. This latter interpretation is tenable based on the relationship the SAO mutation in humans and their relationship to malaria (Baer et al., 1976), the idea that Plasmodium is a selective force in the evolution of primates (Fooden, 1984), the presence of P. falciparum-like parasites throughout the African hominoid radiation (e.g. Duval et al., 2010), and the apparent restriction of this effect to the cytosolic domain. It may be the case that the small number of changes are under selection, a pattern of selection that is particularly difficult to detect (Yang, 2006). Furthermore, testing among the relaxed constraint and adaptive hypotheses will require additional work, from a combination of epidemiological, population genetic, and functional studies. Particular examination of the inferred nine amino acid insertion in African apes appears promising given its unique phylogenetic pattern and properties, including an inferred alternate start codon. Interestingly, a recent case report found evidence for an altered initiation site in SLC4A1 in a severely anemic individual, a mutation that leads to P. falciparum resistance in vitro (Perrotta et al., 2005). It is worth mentioning that the case for a relationship between this putatively adaptive evolution and Plasmodium will require additional functional work as a rigorous test, as it is possible that the evolution of this gene may be related to a wholly different selective pressure than Plasmodium.

Because it is restricted to humans and African apes, an adaptive interpretation supports the hypothesis that the SAO mutation of SLC4A1 is targeting P. falciparum in humans. Furthermore, it suggests that African apes and humans may share some common mechanisms of adaption to P. falciparum-related parasites. This is especially interesting because examination of other human malarial adaptations in non-human primates has found mixed support for common adaptations between humans and other primates. At the G6PD and β-globin loci of chimpanzees there is no evidence for selection (MacFie et al., 2009; Verrelli et al., 2006) and in orangutans the evidence for selection at the α-globin locus is limited (Steiper et al., 2005, 2006), though there is evidence for selection at these loci in humans (Allison, 1954; Flint et al., 1986). Tung et al. (2009) recently examined the relationship between variants of the baboon FY locus (known to be important in human malarial resistance) and infection with Hepatocystis (a relative of Plasmodium). Using multiple lines of evidence, Tung et al. clearly showed that particular FY variants conferred resistance to Hepatocystis, revealing that some mechanisms of parasitic adaptation are broadly similar between humans and non-human primates (2009). Further studies examining the population genetics of SLC4A1 in African apes and humans, including functional assessments of segregating variants and their relationships with particular Plasmodium parasites, will enable direct tests of the hypotheses forwarded here. Finally, the nine amino acid insertion inferred to have occurred in African apes, yet lost in humans, is worthy of special attention in future functional analyses of the primate SLC4A1 gene.

The results presented here also bear on our understanding of the evolutionary history of the human Wright blood group antigen. In humans, the Wr antigen is formed when the 658th amino acid at SLC4A1 links to the 61st amino acid of the GYPA protein (arginine) (Bruce et al., 1995). The human Wr^a allele is exceptionally rare (e.g. Arriaga et al., 2005). Interestingly, the 658th amino acid was identified in multiple tests as evolving positively in SLC4A1 across mammals. Furthermore, GYPA is one of the most rapidly evolving loci in the primate genome (Baum et al., 2005; Wang et al., 2003). The positive selection at this site linking SLC4A1 to GYPA suggests that these genes may be evolving adaptively and in a correlated manner. Previous evolutionary work exploring the relationship between GYPA and SLC4A1 in primates suggested that the specific amino acids bonding these proteins may differ across species (Huang et al., 1996). Because SLC4A1 amino acid 658 is evolving positively across all mammals studied, it is likely that the correlated evolution is not related to Plasmodium but is a more ancient or generalized selective force. Although a recent study of the population genetics of SLC4A1 in humans suggests the selective history of this gene may relate to anion transport (Wilder et al., 2009), the Wr antigens do not differ in this feature (Bruce et al., 1995). Future study will help to elucidate the relationship between the SLC4A1 and GYPA genes across primate evolution and also determine the selection pressure acting on these loci.

5. Conclusions

These results show that there has been a differential selective regime operating on SLC4A1 gene in humans and African apes relative to other primates and mammals. The effect is mainly found in the cytosolic domain of this protein. This finding has two interpretations. One interpretation is that the cytosolic domain of SLC4A1 is experiencing relaxed selective constraint in humans and African apes. A second interpretation is that there have been a small number of adaptive evolutionary changes within the human and African ape lineages, potentially in response to P. falciparum and its relatives. The adaptive hypothesis suggests that these African hominoid primates have been evolving to combat Plasmodium since their last common ancestor. Future studies investigating the relationship between SLC4A1 variants and Plasmodium in humans and wild African apes can further evaluate this hypothesis. Also of note is the finding that the amino acid in SLC4A1 that encodes the human Wright blood group antigen is under positive selection across mammals. The Wright blood group antigen is little studied, and additional research is required to determine the potential evolutionary, functional, and clinical significance of this finding.

6. Database ID

HM065568, HM065581, HM065569, HM065570, HM065571, HM065572, HM065580, HM065573, HM065573, HM065575, HM065576, HM065577, HM065578, HM065579 (all at Genbank, the Genetic sequence database at the National Center for Biotechnical Information (NCBI)).