A Novel System of Polymorphic and Diverse NK Cell Receptors in Primates

There are two main classes of natural killer (NK) cell receptors in mammals, the killer cell immunoglobulin-like receptors (KIR) and the structurally unrelated killer cell lectin-like receptors (KLR). While KIR represent the most diverse group of NK receptors in all primates studied to date, including humans, apes, and Old and New World monkeys, KLR represent the functional equivalent in rodents. Here, we report a first digression from this rule in lemurs, where the KLR (CD94/NKG2) rather than KIR constitute the most diverse group of NK cell receptors. We demonstrate that natural selection contributed to such diversification in lemurs and particularly targeted KLR residues interacting with the peptide presented by MHC class I ligands. We further show that lemurs lack a strict ortholog or functional equivalent of MHC-E, the ligands of non-polymorphic KLR in “higher” primates. Our data support the existence of a hitherto unknown system of polymorphic and diverse NK cell receptors in primates and of combinatorial diversity as a novel mechanism to increase NK cell receptor repertoire.


Introduction
Natural killer (NK) cells are bone marrow-derived lymphocytes that form an essential part of the immune response against pathogens and are involved in the elimination of tumour cells. Equipped with a diverse array of germline-encoded receptors that are able to mediate inhibitory or activating signals [1], NK cells scan other cells for the presence of ligands of these receptors [2]. Activation of NK cells is typically achieved by discontinuation of inhibitory signalling and involvement of activating receptors, resulting in cytokine release or killing of target cells [2][3][4]. Most NK cell receptors interact with members of the MHC class I protein family and either belong to the killer cell lectin-like receptors (KLR) of the C-type lectin-like family such as CD94, NKG2, or Ly49, or the killer cell immunoglobulin-like (KIR) receptors, which are encoded in the natural killer complex (NKC) and leukocyte receptor complex (LRC), respectively [1]. Both NKC and LRC contain inhibitory and activating NK cell receptors. Inhibitory receptors are characterised by the presence of immunoreceptor tyrosine-based inhibitory motifs (ITIM) in the cytoplasmic tail, whereas activating receptors lack ITIMs and instead contain a positively charged amino acid (arginine or lysine) in their transmembrane region, thereby associating with signalling adaptor molecules DAP10, DAP12 or FccR [2].
The polymorphic NK cell receptors are represented by KIR in humans, apes, Old World and New World monkeys [5][6][7] and KLR (Ly49 molecules) in rodents [8]. These two receptor systems are not structurally related, but have similar functions: interaction with MHC class I molecules and regulation of NK cell activity [2]. Due to these functional constraints, KIR and Ly49 genes have to keep pace with the rapidly evolving and polymorphic class I genes [9] and it has been shown that combinations of this highly complex and polymorphic genetic system of NK cell receptors and their MHC class I ligands can significantly influence susceptibility and resistance to infectious and malignant diseases, autoimmune disorders, and reproduction [10]. In contrast to KIR and Ly49, another genetic system of MHC class I-binding NK cell receptors has been kept conserved in primates (hominoids, Old World and New World monkeys) [11] and rodents [12], the CD94 and NKG2 molecules, which also belong to the KLR family. The CD94 molecule can either pair with the inhibitory NKG2A or the activating NKG2C and NKG2E molecules and these heterodimeric NK cell receptors specifically recognise conserved nonclassical class I molecules, HLA-E in humans [13] and H2-Qa1 in mice [14]. These nonclassical class I molecules bind peptides derived from signal sequences of certain class I molecules [15], thereby monitoring putative downregulation of the corresponding class I molecules mediated by pathogens as part of their immune evasion strategy.
Various NK cell receptors (CD94/NKG2, KIR, Ly49L) have been identified in different nonhuman primates, and rapid evolution in particular of the KIR genes was reported [16]. Yet, two questions remain open: how far can we trace back KIR3DL diversification and what was the NK cell receptor content in the ancestor of all primates? We report here that strepsirrhine primates such as lemurs have evolved a 'third way' of a diverse NK cell receptor system and exhibit highly diversified, positively selected CD94/NKG2 receptors. This NK cell receptor system is further characterised by combinatorial diversity.

Results
In the absence of a formal taxonomic designation, we refer to strepsirrhines (lemurs, galagos, lorises) and tarsier as 'lower' primates and to humans, apes, Old and New World monkeys as 'higher' primates throughout the manuscript.
Unusual organisation of the LRC, NKC, and MHC in ''lower'' primates Clones containing the LRC, NKC, and MHC genomic regions were isolated from a BAC library of the grey mouse lemur (Microcebus murinus). Nineteen BAC clones from the LRC region were identified and three clones covering the KIR-LILR subregion were sequenced: one clone contains the NCR1, FCAR, and LILR genes in addition to a KIR3DP pseudogene mapping between FCAR and the LILR gene cluster, and the other two clones both include LILR genes and a single copy of the KIR3DX1 [17] gene ( Figure 1A). Similar to all other primate KIR3DX1 genes, the two lemur copies, of which only one is intact, also lack the MLTD1 repetitive elements in intron 3. The intact copy is transcribed in lemur PBMC and alternatively spliced ( Figure S1); the other locus (KIR3DX1P) is a fragmented pseudogene. In contrast, KIR3DP is characterised by the presence of KIR3DL-typical repetitive elements such as MLT1D/LTR33A, MSTB1, MER70B [7] and is thus the only representative of the KIR3DL lineage in the grey mouse lemur. The repetitive elements flanking KIR3DP are found in the 'outer' human KIR genes KIR3DL3 and KIR3DL2 (data not shown), suggesting that the lemur KIR3DP is equivalent to all 'higher' primate KIR genes. Thus, the only functional KIR in the mouse lemur genome is represented by KIR3DX1, a KIR gene of unknown function [17].
Unlike the KIR region, the CD94-Ly49L genomic interval of the grey mouse lemur NKC showed amplifications of the CD94 and NKG2 genes ( Figure 1B), which accounted for its 1.5 times increased size compared to humans. Only two genes in this region represent one-to-one orthologs of 'higher' primate genes: NKG2D and Ly49L. Further investigations of the three mouse lemur CD94 genes indicate they encode typical CD94 receptor structures but their amino acid sequences show considerable diversity, differing by 23-24% among each other ( Figure S2). Such level of divergence is significantly higher than in 'higher' primates, as human and common marmoset monkey CD94 sequences differ only by 13% for example [11]. Furthermore, eight NKG2-related genes were identified ( Figure 1B), of which five are functional genes ( Figure S3) and three are pseudogenes. The NKG2 molecules show different combinations of functionally relevant motifs: ITIM, positively charged residue in the transmembrane region, and the YxxM motif, which is a recognition site for the p85 subunit of phosphatidylinositol 3-kinase (PI3K). Allelic substitutions affect the presence of ITIM and YxxM motifs in NKG2-2 and NKG2-3 ( Figure S3). These findings suggest that the mouse lemur NKG2 receptors are functionally complex, with inhibitory and activating properties.
BAC clones from the MHC class I and class II regions could be mapped to the short arm of mouse lemur chromosome 6 by fluorescence in-situ hybridisation (FISH) (Figure 2). Whereas the class II region is conserved (Averdam et al., manuscript in preparation), the linked class I gene-containing regions of the mouse lemur MHC lack any functional class I genes as the four identified class I genes and the single MIC gene all represent pseudogenes ( Figure 1C). Remarkably, even MHC-E and MHC-F, the only two conserved MHC class I genes of 'higher' primates, are missing in the mouse lemur MHC. Interestingly, our screening of the mouse lemur BAC library for MHC class I genes also identified an unlinked genomic region that includes nine MHC class I genes. Complete sequencing of this region revealed six genes encoding functional MHC class I proteins, including putative alleles of the previously described classical class I genes of the mouse lemur, Mimu-W01 and Mimu-W04 [18] (Figure 1C). This class I gene cluster maps to another chromosome, the long arm of mouse lemur chromosome 26 (Figure 2), thus providing an additional example of a mammalian species where MHC class I and class II genes are not linked. A similar chromosomal splitting of class I genes was found for a further lemur species, Coquerel's giant mouse lemur (Mirza coquereli) ( Figure 2). The CD94 and NKG2 genes could be mapped to chromosome 7 in the giant mouse

Author Summary
Most receptors of natural killer (NK) cells interact with highly polymorphic major histocompatibility complex (MHC) class I molecules and thereby regulate the activity of NK cells against infected or malignant target cells. Whereas humans, apes, and Old and New World monkeys use the family of killer cell immunoglobulin-like receptors (KIR) as highly diverse NK cell receptors, this function is performed in rodents by the diverse family of lectin-like receptors Ly49. When did this functional separation occur in evolution? We followed this by investigating lemurs, primates that are distantly related to humans. We show here that lemurs employ the CD94/NKG2 family as their highly diversified NK cell receptors. The CD94/NKG2 receptors also belong to the lectin-like receptor family, but are rather conserved in ''higher'' primates and rodents. We could further demonstrate that lemurs have a single Ly49 gene like other primates but lack functional KIR genes of the KIR3DL lineage and show major deviations in their MHC class I genomic organisation. Thus, lemurs have evolved a ''third way'' of polymorphic and diverse NK cell receptors. In addition, the multiplied lemur CD94/NKG2 receptors can be freely combined, thereby forming diverse receptors. This is, therefore, the first description of some combinatorial diversity of NK cell receptors.   [43] The same split is shown for Mirza coquereli: ZNRD1 and MOG hybridise to chromosome 6 and ANKRD11, Mimu-W01, and Mimu-W04 to chromosome 26. All tested MHC class II region genes (TAP2, DQ, DR, BTNL2) map to chromosome 6 in Microcebus murinus and Mirza coquereli, respectively. Note that in both species secondary signals appear scattered along the X chromosome. In Mirza coquereli, KIR-including BAC CD257-37C10 hybridises to a small acrocentric chromosome not identical with chromosome 26 and to the X chromosome, and NKG2-including BAC CH257-96C7 maps to chromosome 7, which is orthologous to human chromosome 12. Note that further signals appear on the chromosome X. doi:10.1371/journal.pgen.1000688.g002 lemur, which corresponds to human chromosome 12, indicating conserved synteny of the NKC region. The KIR3DX1P genecontaining BAC was mapped to a small acrocentric chromosome not identical with chromosome 26 (Figure 2).
Phylogenetic analyses of primate CD94 and NKG2 sequences encoding the C-type lectin-like domain indicate that for both gene families 'lower' and 'higher' primate sequences form their own groups, indicating that diversification of CD94 and NKG2 occurred in 'lower' primates after their separation with 'higher' primates ( Figures 3A and 3B). 'Lower' primate NKG2 sequences form three groups, each with sequences from both lemurs, pointing to duplications that preceded speciation of both lemurs ( Figure 3B). Similarly, in the largest of the three 'lower' primate NKG2 groups, some gene sequences show species-specific clustering indicating gene duplications following speciation of both lemurs. Such patterns demonstrate that NKG2 gene duplications in 'lower' primates occurred both before and after the separation of the two lemur species, a process similar to that seen for KIR in 'higher' primates and Ly49 in rodents [20]. In contrast to the C-type lectinlike domain, NKG2 sequences encoding the stem, cytoplasmic, and transmembrane part fall into two branches: one contains 'lower' and 'higher' primate inhibitory NKG2, the other one containing 'lower' and 'higher' primate activating NKG2, indicating that these two types of sequences separated before the speciation between 'higher' and 'lower' primates ( Figure 3C).
To determine whether amplification of CD94 genes is restricted to strepsirrhine primates living in Madagascar (lemurs) or is common to other 'lower' primates, we investigated the CD94 sequences of an African and an Asian primate, the potto (Perodicticus potto) and the tarsier (Tarsius syrichta). Whereas the potto belongs to the primate suborder Strepsirrhini, the tarsier is a member of the other primate suborder, the Haplorrhini (tarsiers, New and Old World monkeys, apes and human). Using all available primate sequences as reference, generic primers were constructed to amplify the region including exon 4, intron 4 and exon 5 of CD94 and we characterised seven and nine different CD94 sequences in potto and tarsier, respectively ( Figure S4). These findings thus indicate that CD94 gene amplification is not restricted to lemurs, but can be found in other strepsirrhine primates and even in a primate more closely related to humans than to lemurs [21]. Insertions of repetitive elements of the Alu family in intron 4 revealed additional information: whereas none of the three mouse lemur CD94 genes contains any Alu element, four of the seven potto sequences include an AluJo and all nine tarsier sequences have an AluJb element ( Figure S4). However, the integration sites of these Alu elements differ between potto and tarsier, indicating that integration events occurred after these lineages diverged from each other. Hence, the amplification of CD94 sequences occurred repeatedly and independently in 'lower' primates. The intron 4 sequences of human, rhesus macaque and common marmoset CD94 share an AluSx integration (data not shown), which is not present in 'lower' primates, further supporting the monophyletic origin of 'higher' primate CD94.
While lemurs (and possibly other 'lower' primates) experienced a diversification of the CD94/NKG2 receptor system, they appear to lack MHC-E, the ligand of CD94/NKG2 in 'higher' primates, as their MHC region lacks a class I gene where the MHC-E gene of 'higher' primates is located ( Figure 1C). To investigate if any of the grey mouse lemur MHC class I genes represents an ortholog or functional equivalent of 'higher' primate MHC-E, we performed phylogenetic analyses of primate MHC class I sequences. Analyses of the full-length sequences and of the peptide-binding-domain (PBD) alone, with or without the peptide binding residues (PBR) indicate that 'lower' and 'higher' primate sequences form their own groups ( Figure 4A-4C, Figure S5). This demonstrates that MHC class I diversification in 'lower' and 'higher' primates took place in each taxonomic lineage after their separation and further indicates that the duplication that gave rise to the MHC class I genes on chromosome 26q occurred in the 'lower' primate lineage. Similarly, analysis of the PBR revealed that the 'higher' primate MHC-E sequences are more closely related to the rodent Qa1 group than to any of the 'lower' primate sequences ( Figure 4D). Thus, these findings indicate that the grey mouse lemur genome neither encodes a strict orthologue (with a one-to-one relationship) nor a functional homologue of MHC-E and confirm earlier data of our group that 'higher' and 'lower' primate MHC class I genes lack strict orthology [18].

NKC-encoded genes of ''lower'' primates are genetically and functionally diverse
Our genomic analysis in one mouse lemur individual and cDNA sequencing in three individuals (two mouse lemurs and one blackand-white ruffed lemur) revealed that 'lower' primates experienced a diversification of both CD94 and NKG2. Such diversification is in contrast with the situation in 'higher' primates where CD94 is a non-polymorphic, single-copy gene and where the NKG2 gene family experienced limited diversification with moderate polymorphism. To investigate if such difference is the result of natural selection favouring functional diversity in 'lower' but not in 'higher' primates, we compared the non-synonymous to synonymous substitution ratio rate (dN/dS) of CD94 and NKG2 sequences of 'lower' and 'higher' primates (Table 1). This analysis shows that 'lower' primate CD94 sequences display highly significant evidence of positive diversifying selection (a = 0.001) with 15 positions positively selected (posterior probability (PP).0.95). In contrast, their 'higher' primate counterparts do not show any evidence of positive selection (Table 1). A similar, yet not as marked, trend is seen for NKG2 as 'lower' primate sequences show highly significant evidence of positive diversifying selection (a = 0.001), with six positions being positively selected (PP.0.95) while in 'higher' primates the evidence was not as strong (a = 0.01-0.05) and limited to only one site ( Table 1). Mapping of the 21 positively selected sites in 'lower' primates on the three-dimensional model of the human CD94/NKG2A heterodimer in contact with HLA-E [22,23] suggests that many of these positions represent functionally relevant sites ( Figure 5A). Indeed, closer inspection revealed that the distribution of these 21 positively selected sites in the CD94/ NKG2 lectin-like domains shows a significant bias (a = 0.05) toward the 54 sites involved in ligand-binding and/or dimer formation (Table 2, Figure S6). Further dissection of this bias shows it is particularly marked for the 28 ligand-binding sites (a = 0.01), but not for the 30 sites involved in dimer formation. Within the ligandbinding sites a significant bias toward the 7 sites contacting the MHC class I peptide (a = 0.003) was observed, but for the sites contacting the MHC-class I heavy chain the bias was marginal (a,0.06) (Table 2, Figure S6). Consistent with this, the four CD94/NKG2 positively selected positions that contact the MHC class I peptide account for 75% (9 out of 12) of all the contacts with the peptide ( Figure 5B). This analysis thus shows that the genomic diversification of CD94/NKG2 in 'lower' primates was accompa-nied by positive diversifying selection that particularly targeted the sites contacting the MHC class I peptide. In 'higher' primates the limited (NKG2) or lack of (CD94) genomic diversification was accompanied by limited (NKG2) or lack of (CD94) sequence diversification through positive diversifying selection.
To fully investigate the extent of allelic diversity of 'lower' primate NKC genes we specifically amplified the exons encoding the lectinlike domain of mouse lemur CD94, NKG2, and Ly49L genes in a cohort of 46 free-living unrelated animals derived from the Kirindy region in Madagascar [24]. Twelve animals were analysed for all  , peptide binding domain excluding peptide binding residues (C) and peptide binding residues only (D). A phylogenetic analysis of MHC class I genes with complete coding sequences excluding the peptide binding residues was also conducted and the results are presented in Figure S5. Analysis and display are as described for Figure 3. Numbers as gene names for MHC class I are the same as in Figure 1C. Non-primate sequences included are mouse (H2) and rat (RT1). Peptide binding residues were defined according to Bjorkman et al. [44] genes and further 34 animals were additionally analysed for the three CD94 genes. No presence/absence polymorphisms of grey mouse lemur CD94 or NKG2 genes were observed in our cohort, indicating that copy number variation similar to what is known for human KIR and mouse/rat Ly49 haplotypes is not evident in grey mouse lemur CD94/NKG2 haplotypes. Single nucleotide polymorphisms (SNPs) were detected for all genes and with the exception of CD94-3 and NKG2D, all the NKC genes show an excess of nonsynonymous substitutions (Table S1). While 30% (71/236) of all the codons forming the lectin-like domains of CD94 and NKG2 have non-synonymous allelic polymorphisms in at least one of the 'lower' primate CD94 or NKG2 genes, 66% (47/71) of these codons are neither sites involved in ligand binding or dimer formation nor positions we detected as positively selected in our gene analysis.  Table 1. Natural selection diversified CD94 and NKG2 residues in ''lower'' primates, but not in ''higher'' primates.  Because the differences between the genes largely overshadowed allelic differences in our analysis for positive selection, such an observation suggests that following gene duplication natural selection first diversified functional sites and subsequently favoured in each species polymorphisms at sites often not directly involved in function, presumably to further 'tweak' functions. Unlike the ligand binding sites, the positions involved in dimer formation are not significantly enriched in positively selected sites (Table 2, Figure S6). Because of this apparent lack of major diversification of the dimer formation function, we hypothesised that the three CD94 and five NKG2 molecules of the grey mouse lemur can be freely combined to form various CD94/NKG2 heterodimers. To test this hypothesis, we checked all possible combinations with (extracellular) V5-tagged NKG2 and (intracellular) GFP-tagged CD94 molecules. Expression constructs were transiently transfected into 293T cells and CD94/NKG2 heterodimer formation was assayed by co-immunoprecipitation using tag-specific monoclonal antibodies. Heterodimers were found for all CD94/NKG2 combinations by co-immunoprecipitation ( Figure 6). Cell surface expression of all CD94/NKG2 receptor combinations was examined with (both extracellular) FLAG-and V5-tagged CD94 and NKG2 molecules, respectively. All combinations of NKG2-1, NKG2-3, and NKG2-8 with the three CD94 molecules were found on the cell surface ( Figure 7A). In contrast, cell surface expression of both NKG2-2 and NKG2-5 receptors in combination with CD94 molecules was either low (CD94-1/NKG2-5) or lacking ( Figure 7A), suggesting that these receptors depend on the DAP12 adaptor molecule. An expression construct of mouse lemur DAP12 with a c-myc-tag was established and expression was controlled in parallel experiments by intracellular staining (data not shown). Co-transfection with cmyc-tagged mouse lemur DAP12 significantly increased cell surface expression of all combinations formed by CD94/NKG2-2 and CD94/NKG2-5 heterodimers ( Figure 7B). This finding identifies NKG2-2 and NKG2-5 as classical activating receptors whose cell surface expression is dependent on interaction with DAP12. CD94 and NKG2 expression constructs were also transfected in 293T cells alone. Similar to human CD94, all three  mouse lemur CD94 proteins can be found at the cell surface in the absence of any NKG2 molecule, but unlike human NKG2 molecules, also mouse lemur NKG2-2 (with DAP12) and NKG2-3, and to a lower extent NKG2-5 (with DAP12) and NKG2-8, are expressed at the cell surface in the absence of CD94 ( Figure 7A and 7B). Taken together, our data show that three CD94 molecules can be combined with five NKG2 molecules to form 15 different NK cell receptors that display significant diversity at sites of interaction with MHC class I ligands and their bound peptide. This indicates that besides gene amplification and positive diversifying selection combinatorial diversity is also a mechanism that significantly contributes to the increase of NK cell receptor diversity in 'lower' primates.

Discussion
The polymorphic NK cell receptors of humans and mice are functionally similar but structurally unrelated. Therefore, we traced back NK cell receptors and MHC class I ligands to the base of the primate evolutionary tree by analysis of the LRC, NKC, and MHC genomic regions in primates distantly related to humans. We demonstrate that 'lower' primates deviate from 'higher' primates in the usage of polymorphic NK cell receptors. Except for the KIR3DX1 gene, lemurs and possibly other 'lower' primates have neither functional nor highly polymorphic KIR genes like their relatives, the catarrhine (Old World monkeys, apes, and humans) and platyrrhine (New World monkeys) primates, nor do they show an expansion of Ly49 genes as in rodents. Instead, lemurs have considerably amplified and diversified their C-type lectin-like CD94 and NKG2 genes. We conclude from our findings that the NK cell receptor repertoire of 'lower' primates is at least as diverse as in 'higher' primates or rodents. Thus, in addition to the KIR and Ly49 genes of 'higher' primates and rodents, the CD94/NKG2 heterodimers of lemurs represent a third system of polymorphic and diverse NK cell receptors. Compatible with such a system is that the duplicated lemur CD94 and NKG2 genes show sequence diversifications and strong signs of positive diversifying selection. In accordance with these characteristics, lemur NKG2 genes do not show signs of gene homogenisation as opposed to 'higher' primate NKG2 sequences [11]. Such homogenisation may serve to keep NKG2 amino acid sequences conserved for interaction with the invariable MHC-E ligand, a situation that is not observed for the polymorphic CD94/NKG2 receptors in 'lower' primates.
CD94 gene duplications are not restricted to 'lower' primates from Madagascar as they were found in an African strepsirrhine primate (Perodicticus potto) and in the Asian tarsier (Tarsius syrichta). The latter is particularly interesting, as the tarsier is more closely related to 'higher' primates than to 'lower' primates [21]. Analysis of repetitive elements in CD94 intron 4 sequences revealed that duplications had occurred repeatedly and independently in 'lower' primates. Thus, the polymorphic CD94/NKG2 system is likely present in many if not all 'lower' primates. This is in sharp contrast to the situation in 'higher' primates where CD94 is a single copy, non-polymorphic and highly conserved gene.
Three KIR genes were detected in the lemur LRC region, a functional and a pseudogene copy of KIR3DX1 and a KIR3DP pseudogene. According to its characteristics, the KIR3DP gene may represent the 'Ur-KIR' gene of all 'higher' primate KIRs. This gene already contains the repetitive elements MLTD1, MER70B, MSTB1 in its introns, which are assumed to be integrated about 60-100 mya [25], a time that is compatible with the splitting of lemurs and human of about 65-90 mya [26]. Thus, we postulate that KIR and CD94/NKG2 receptors evolved differently in primates: while in the lineage leading to 'lower' primates CD94 and NKG2 but not KIR genes expanded, the opposite happened in the lineage leading to 'higher' primates where KIR genes expanded and CD94 and NKG2 co-evolved with the non-classical MHC-E molecule to become a conserved receptor/ligand system (Figure 8). The finding that 'lower' primates did not amplify and diversify KIR or Ly49 genes but, instead, evolved a polymorphic CD94/NKG2 system, strengthens previous assumptions that mammals only utilise a single class of polymorphic NK cell receptors, despite their obvious ability to develop multiple classes [27,28]. Most likely, this development is influenced by the pathogenic threat these organisms are subjected to and can involve different receptor types such as monomeric KIR, heterodimeric CD94/NKG2, and homodimeric Ly49. Nevertheless, mammalian species are equipped with all types of receptor genes [1], which gives some flexibility for adaptation according to NK cell receptor and ligand requirements.
All class I genes in the MHC equivalent genomic region on grey mouse lemur chromosome 6 are pseudogenes and all functional MHC class I genes were translocated to chromosome 26. Additionally, no functional MIC gene could be identified in any of the class I gene-containing regions in the grey mouse lemur. However, this finding is not surprising, as NKG2D ligands are numerous and functionally redundant [29], absence of MIC is evident in rodents [30] and deletions of both MICA and MICB were reported in East-Asians [31]. In addition to this unusual organisation of the grey mouse lemur MHC, a striking difference to 'higher' primates is the apparent absence of a strict HLA-E orthologue or functional homologue. As the BAC library was screened exhaustively and in the light of our genome-wide approach published earlier [18], it appears rather unlikely that MHC-E-like class I genes were not detected. Thus, the putative CD94/NKG2 ligands are expected among the sequenced MHC class I genes. Nevertheless, we cannot completely rule out the possibility that a HLA-E-like gene was not detected by our approach or that a gene with HLA-E-like function is among the detected MHC class I genes. There has been some debate in the past on whether HLA-E and mouse H2-Qa1 have a common origin [32] or whether their similarity in the peptide-binding region is a consequence of convergent evolution [33]. In the light of our data, the second hypothesis appears more likely, as diversified KIR and conserved MHC-E/CD94/NKG2 emerged only in 'higher' primates and diversified Ly49 and conserved H2-Qa1/CD94/NKG2 evolved independently in rodents.
The observed positive diversifying selection in 'lower' primates is more pronounced in the three CD94 than the NKG2 genes, suggesting that the CD94 molecules have significant impact on the binding of the polymorphic MHC class I ligands and their bound peptides. Indeed, the recently determined three-dimensional structure of human CD94/NKG2A in complex with HLA-E revealed that CD94 and NKG2A contribute about 80% and 20%, respectively, to the interaction with HLA-E and its bound peptide [22,23]. Translated to the situation in 'lower' primates, duplication of CD94 genes and their strong sequence diversification by positive diversifying selection reflects the requirement to maintain binding to polymorphic (classical) MHC class I molecules and bound peptides. Finally, we demonstrate that all possible CD94/NKG2 combinations are able to form heterodimers at the cell surface, revealing important implications: exchange of the CD94 or the NKG2 subunit is likely to influence the binding specificity for MHC class I ligands and their bound peptide and the functional properties (inhibitory or stimulating) of the receptor. For example, the combination of the three CD94 and five NKG2 molecules in the grey mouse lemur or three CD94 and eight NKG2 molecules in the ruffed lemur gives rise to 15 or 24 different NK cell receptors, respectively. Thus, we conclude that the NK cell receptor repertoire in 'lower' primates is not mainly achieved by duplications, but rather by combinatorial diversity, a phenomenon that was so far unknown for any NK cell receptor. NK cells were recently shown to exhibit features of adaptive immunity, namely immunological memory [34]. Combinatorial diversity of immune receptors is a further hallmark of adaptive immunity. Although the combinatorial diversity of 'lower' primate CD94/NKG2 receptors is obviously much less than that usually seen for B or T cell receptors, our data additionally highlight the close relationship of two lymphocyte subsets: natural killer cells and cytotoxic T cells.
In summary, we have uncovered a 'third way' of polymorphic and diversified NK cell receptors in mammals. The CD94/NKG2 receptor system (and not KIR) of 'lower' primates is characterised by duplication, sequence diversification by means of positive diversifying selection and allelic diversity. Consistent with this highly dynamic CD94/NKG2 system, the MHC class I molecules as putative ligands of these receptors show strong signs of coevolution and an unusual chromosomal organisation. CD94 and NKG2 subunits constitute the main NK cell receptor repertoire in 'lower' primates and are freely combinable. This finding discloses a so far unknown mechanism of generating the NK cell receptor diversity: combinatorial diversity.

Blood and DNA samples
Grey mouse lemurs (Microcebus murinus) and black-and-white ruffed lemurs (Varecia variegata) are housed in the facilities of the UMR CNRS/MNHN and the German Primate Center, respectively. Blood was obtained during regular veterinary inspections. DNA samples from potto and tarsier were kindly obtained from Helga Schulze (University of Bochum, Germany) and Jürgen Schmitz (University of Münster, Germany).
BAC clone DNA was purified by CsCl density centrifugation and isolated DNA was sheared by sonification. DNA fragments of 1.5-3.5 kb were selected and cloned into plasmid pUC19. Inserts were amplified by PCR using insert-flanking M13 forward and reverse primers and amplificates were sequenced with Applied Biosystems BigDye terminator chemistry and analyzed in ABI3730xl sequencers (Applied Biosystems). Raw sequences were processed by Phred (www.phrap.org) and assembled into a contiguous sequence by Phrap (www.phrap.org). Both programs are available from Phil Green, University of Washington. Exons and introns of genes were identified in finished BAC clone sequences manually or by BLAST and FGENESH-2 algorithms (http://www.ncbi.nlm.nih.gov/BLAST/; http://www.softberry. com/all.htm). All BAC clone sequences have 'finished sequencing' quality (1 putative mistake/100,000 bases) and the data have Phred values of 60.
Exons 4 to 6 of the grey mouse lemur CD94, NKG2, and Ly49L genes were amplified from genomic DNA of 12 (all genes) and additional 34 (only CD94 genes) unrelated free-living grey mouse lemur individuals from the Kirindy region in Madagascar. Specific primers (Table S2) were designed to be located in exon-flanking introns. PCR products were completely sequenced on both strands and SNPs were identified.

RT-PCR and expression constructs
Total RNA was extracted from a liver sample of a grey mouse lemur and a blood sample from a further grey mouse lemur housed at UMR CNRS/MNHN in Brunoy (France) and a blood sample from a black-and-white ruffed lemur (Varecia variegata variegata) housed at the German Primate Center. Reverse transcription was performed with M-MLV reverse transcriptase (Promega). Based on the mouse lemur BAC sequences primers were designed to obtain the complete open reading frames of mouse lemur and ruffed lemur NKC genes. cDNA sequences of the ruffed lemur were completed by 59-and 39-RACE PCR with the GeneRacer Kit (Invitrogen). All cDNA sequences were cloned and sequenced. For interaction studies, grey mouse lemur CD94 and NKG2 cDNA sequences were inserted into the pcDNA3.1/NT-GFP-TOPO and the pcDNA3.1/ V5-His-TOPO vector (Invitrogen), respectively. For cell surface expression, the three CD94 cDNA sequences were expressed without GFP at the NH 2 terminus, but with a FLAG tag (DYKDDDDK) at the COOH terminus for extracellular detection. The grey mouse lemur DAP12 cDNA was isolated by RT-PCR. The encoded DAP12 was fused at the COOH terminus with the cmyc peptide epitope (EQKLISEEDL). Primer sequences and performed PCRs are listed in Table S2.
Transfection, co-immunoprecipitation, and flow cytometry CD94 and NKG2 molecules were tagged with GFP and with V5 for co-immunoprecipitation experiments. To avoid any potential unwanted interactions between the tags, GFP was fused to the amino terminus of CD94 ( = intracellular localisation) and V5 to the carboxy terminus of NKG2 molecules ( = extracellular localisation). Expression constructs were transiently transfected in the human 293T cell line using Metafectene (Biontex GmbH). 24 h after transfection, cells (1610 7 cells/ml) were lysed in NP-40 lysis buffer containing 0.1% NP-40, 50 mM Tris, pH 7.6, 150 mM NaCl, 4 mM EDTA and protease inhibitors (Roche). Lysates were incubated overnight at 4uC with monoclonal mouse anti-GFP antibody (Clontech) and then incubated with protein G Sepharose 4 Fast Flow beads (GE Healthcare) for 3 h. Sepharose beads were washed five times with lysis buffer and bound proteins were eluted with loading buffer at 95uC for 10 minutes. Samples were electrophoresed in 10% polyacrylacrylamide gels and transferred to nitrocellulose membrane. Western blotting was performed with HRP-coupled monoclonal mouse anti-V5 antibody (Invitrogen) to test for co-immunoprecipitated V5-tagged NKG2 molecules. Deglycosylation was performed with PNGase F according to supplier's recommendations (New England Biolabs).
For analysis of CD94/NKG2 receptors on the cell surface, CD94 was FLAG-tagged at the carboxy terminus ( = extracellular localisation). Respective expression constructs of CD94 together with the V5-tag NKG2 constructs (see above) were transiently transfected in 293T cells. Cell surface expression of CD94/ NKG2-2 and CD94/NKG2-5 receptors were tested by additional transient transfection with c-myc-tagged mouse lemur DAP12 expression constructs. As control, single transfections with either CD94 or NKG2 expression constructs were performed. Potential formation of heterodimers of grey mouse lemur NKG2 and human CD94 (from 293T cells) was excluded by testing with an anti-human CD94 antibody (AbD Serotec). 24 hours post transfection cells were detached from cell culture dishes and washed twice with 16 PBS. Cells were stained with APC-labelled anti-V5 (Abcam) and FITC-labelled anti-FLAG (Sigma-Aldrich) monoclonal antibodies for surface expression of NKG2 and CD94. In parallel experiments, expression of DAP12 was monitored with an anti-c-myc mouse monoclonal antibody (Sigma-Aldrich) by intracellular staining of cells, which were previously fixed with 1.5% paraformaldehyde (Merck) and permeabilized with 0.25% saponin (Roth), and binding was detected by a PE-Cy5-labelled goat anti-mouse IgG polyclonal antibody (Santa Cruz Biotechnology). After washing twice with 16 PBS cells were resuspended in 200 ml of 16 PBS and 50,000 events were measured. Living cells were gated based on forward and side scatter characteristics and analysed for APC and FITC staining. All samples were analysed in a LSRII flow cytometer (BD). Data were acquired with BD FACS Diva 5.1 software (BD) before analysis with FlowJo 7.2.9 software (TreeStar).

Phylogenetic analyses
CD94, NKG2 and MHC class I nucleotide sequences were aligned with MAFFT [35] and corrected manually. Phylogenetic analyses were conducted using three methods: neighbor-joining (NJ), parsimony and Bayesian phylogenetics. NJ analyses were performed with MEGA3.1 [36] using the Tamura-Nei method with 1,000 replicates. PAUP*4.0b10 [37] and the tree bisectionreconnection branch swapping algorithm were used for parsimony analyses with 1,000 replicates and a heuristic search. For the Bayesian analysis, the model of DNA substitution was selected using MODELTEST3.7 [38] and the Akaike information criterion. Bayesian phylogenetic analyses were conducted with MRBAYES3.1.2 [39]; sampling was performed with one cold chain and three heated chains, which were run for 10 6 generations or until average standard deviation of split frequencies was ,0.01. Trees were sampled every 200 generations and the first 2,500 trees were discarded before a consensus tree was generated; three simultaneous runs were conducted and average standard deviation of split frequencies was always ,0.01. The tree topologies obtained with the three methods were compared with PAUP*4.0b10 using the Shimodaira-Hasegawa test of alternative phylogenetic hypotheses with re-sampling estimated log-likelihood optimization, and 10,000 bootstrap replicates; in all analyses the test failed to reject any of the alternative tree topologies (a = 0.05). This comparison was made with the maximum likelihood model defined by MODELTEST. A maximum-likelihood analysis was also performed for the study of the NKG2 cytoplasmic and transmembrane sequences using RAxML7 [40] under the GTR+CAT model with 1,000 replicates (rapid bootstrapping).

Selection analyses
dN/dS (v) ratios for CD94 and NKG2 lectin-like domains were estimated using PAML v3.15 [41] with the F3 X 4 model of codon frequencies. Bayesian tree topologies were used for these analyses and three sets of likelihood ratio tests were conducted to compare null models that do not allow v.1 (M1a, M7 and M8a) with models that do (M2a and M8). Significance was assessed by comparing twice the difference in likelihood between the models (2DL) to a x2 distribution with one (M8a/M8) or two (M1a/M2 and M7/M8) degrees of freedom. Codons with v.1 were identified using the Bayes Empirical Bayes approach [42].
The distribution of the selected sites in the CD94/NKG2 lectinlike domains was studied using a binomial distribution: considering V = (0,1,2,…,n), "k M V, p = (X = k) = n C k * p k * q n2k . This indicates for example that under a random distribution it is unlikely to have more than 5 of the 21 positively selected sites in the ligand binding region that represents 28 of the 239 total sites (a = 0.05); so the fact that we observe 7 of the 21 positively selected sites in this region indicates a distribution biased toward this region (a = 0.01). Figure S1 Mouse lemur KIR3DX1 sequence. (A) cDNA, (B) deduced amino acid sequence and (C) genomic sequence. cDNA sequence was determined by RT-PCR of mouse lemur PBMC. Start codon is marked in green, stop codon in red, exons in grey, and ITIM in the protein sequence in yellow. Nucleotides and corresponding amino acid residues that are absent in an alternatively spliced product are shown in italics. Transmembrane region is underlined. The shading threshold is 55% sequence identity. N in the sequences of potto CD94-6 and -7 and of tarsier CD94-9 denotes an unknown number of thymidine nucleotides. The CD94-7 sequence of the tarsier contains a stop codon in exon 4 which is highlighted in bold and underlined. Pepo, Perodicticus potto; Tasy, Tarsius syrichta. Alu elements are underlined. Found at: doi:10.1371/journal.pgen.1000688.s004 (0.03 MB PDF) Figure S5 Phylogenetic analyses of MHC class I genes with complete coding sequences excluding the peptide binding residues. Analysis and display are as described for Figure 3. Numbers as gene names for MHC class I are the same as in Figure 1C. Non-primate sequences included are mouse (H2) and rat (RT1). Peptide binding residues were defined according to Bjorkman et al. [44]. Patr, Pan troglodytes; Gogo, Gorilla gorilla; Hyla, Hylobates lar; Saoe, Saguinus oedipus; Sasc, Saimiri sciureus; Mamu, Macaca mulatta; Mimu, Microcebus murinus; Aotr, Aotus trivirgatus; Caja, Callithrix jacchus; Pipi, Pithecia pithecia. Found at: doi:10.1371/journal.pgen.1000688.s005 (0.07 MB PDF) Figure S6 Contacts of CD94/NKG2 to bound peptide (Peptide) and to the MHC class I heavy chain (MHC) as well as residues mediating heterodimer formation (Dimer) based on the human CD94/NKG2A crystal structure [22,23]. Additionally, Peptide and MHC contacts (Ligand binding) and all known contact residues (All) are summarised. Found at: doi:10.1371/journal.pgen.1000688.s006 (4.54 MB TIF)