KIR3DL1 and KIR3DL2 are NK cell receptors for polymorphic HLA-B and -A determinants. The proportion of NK cells that bind anti-KIR3DL1-specific Ab DX9 and their level of binding vary between individuals. To determine whether these differences are due to KIR polymorphism, we assessed KIR3D gene diversity in unrelated individuals and families. Both KIR3DL1 and KIR3DL2 are highly polymorphic genes, with KIR3DS1 segregating like an allele of KIR3DL1. A KIR haplotype lacking KIR3DL1 and KIR3DS1 was defined. The two KIR3DL1 alleles of a heterozygous donor were expressed by different, but overlapping, subsets of NK cell clones. Sequence variation in KIR3DL1 and KIR3DL2 appear distinct; recombination is more evident in KIR3DL1, and point mutation is more evident in KIR3DL2. The KIR3DL1 genotype correlates well with levels of DX9 binding by NK cells, but not with the frequency of DX9-binding cells. Different KIR3DL1 alleles determine high, low, and no binding of DX9 Ab. Consequently, heterozygotes for high and low binding KIR3DL1 alleles have distinct subpopulations of NK cells that bind DX9 at high and low levels, giving characteristic bimodal distributions in flow cytometry. The Z27 Ab gave binding patterns similar to those of DX9. Four KIR3DL1 alleles producing high DX9 binding phenotypes were distinguished from four alleles producing low or no binding phenotypes by substitution at one or more of four positions in the encoded protein: 182 and 283 in the extracellular Ig-like domains, 320 in the transmembrane region, and 373 in the cytoplasmic tail.
Natural killer cells are lymphocytes of innate immunity that secrete immunomodulatory cytokines and kill cells compromised by viral infection or malignant transformation (reviewed in Ref. 1). These functions are controlled by NK cell surface receptors that recognize MHC class I on prospective target cells (reviewed in Ref. 2). Two distinct types of molecule provide human NK cells with MHC class I receptor function. The CD94:NKG2 receptors recognize complexes of HLA-E and peptides derived from HLA-A, -B, -C, or -G leader sequences (3, 4); their extracellular domains resemble the carbohydrate recognition domains of C-type lectins (5). In contrast, the killer cell Ig-like receptors (KIR)4 are members of the Ig superfamily (6), and some of them directly recognize polymorphic HLA-A, -B, or -C determinants (7). Certain KIR2D are specific for HLA-C epitopes defined by polymorphism at positions 77 and 80 of the HLA-C heavy chain (8), KIR3DL1 recognizes the Bw4 determinant defined by sequence motifs at positions 77–83 of the HLA-B heavy chain (9, 10), and KIR3DL2 has affinity for HLA-A (11, 12). Both the CD94:NKG2 and KIR families contain inhibitory and activating receptors that differ in the lengths of their cytoplasmic tails and in the signaling motifs they contain (2).
Selective expression of receptors in the CD94:NKG2 and KIR families gives clonal diversity to a person’s NK cell population (13, 14). Variation in the number of KIR genes present in a haplotype produces further diversity at the level of human populations (15, 16). In comparison with the KIR gene family, a part of the leukocyte receptor complex on chromosome 19 (17), the CD94 and NKG2 genes of the NK complex on chromosome 12 are relatively conserved (18).
Of 13 expressed human KIR defined to date, three are of the type with three extracellular domains: KIR3DL1, KIR3DL2, and KIR3DS1. Whereas KIR3DL1 has been associated with HLA-B specificity (9) and KIR3DL2 with HLA-A specificity (11, 12), KIR3DS1 is of unknown specificity. The KIR3DL1 receptor for HLA-Bw4 was first identified and characterized using the KIR3DL1-specific mAb DX9 (19). The Ab was subsequently used in flow cytometry to compare cell surface expression of KIR3DL1 by NK cells and T cells in a sample population of some 200 individuals (20). Within this population an extensive heterogeneity was observed both in the frequency of the NK cells and T cells that bound DX9 and in the level of their binding (20). Individual donors could be characterized by whether their cells bound a high, low, or no amount of DX9. Particularly intriguing was that some individuals had distinct subpopulations of cells, binding high and low amounts of DX9. For any individual, the pattern of DX9 binding was stable over time, and the levels of binding to DX9+ NK cells and DX9+ T cells were the same. Moreover, twin comparisons and family studies suggested that the differences in frequency and levels of DX9 binding were genetically controlled, but not by the presence or the absence of genes encoding the receptor’s HLA-Bw4 ligands (20). Such differences in the pattern of DX9 binding could be due to variation in the structure of KIR3DL1 or in other components that control the transcription of the gene or the cell surface expression of its protein product. To begin distinguishing among these possibilities we have investigated the structure of KIR3DL1 in individuals whose NK cells exhibit different patterns of DX9 binding. The approach we took also facilitated assessment and comparison of variation within the three human KIR3D groups: KIR3DL1, KIR3DS1, and KIR3DL2.
Materials and Methods
In naming KIR3DL1 and KIR3DL2 alleles the recommended KIR nomenclature (21) was extended using principles established in the naming of HLA alleles (22) (Table I⇓). Following the gene designation is an asterisk and then three digits used to distinguish alleles that differ by nonsynonymous substitutions. The aim is for the numbers to correspond to the order in which alleles are reported. Thus, NKAT3 (GenBank accession no. L41269) (23) is KIR3DL1*001, and NKB1 (accession no. U31416) (24) is KIR3DL1*002. Fourth and fifth digits are used to distinguish alleles that differ only by synonymous substitution, and they need only be used in contexts where silent substitutions are of relevance. When the context is clear, the letters KIR can also be dropped from the allele designation; thus KIR3DL1*001 can be shortened to 3DL1*001. An allele and its protein product (allotype) are given the same name, but are distinguished by the use of italics for the allele name. This method of nomenclature has applicability to all KIR genes.
Cloning and sequencing of KIR3D
14). The 3DL1*004 leader fragment was amplified from cDNA of donor 2 using the following primers: forward, 5′-CTGCACCGGCAGCACCATGT-3′; and reverse, 5′TGGGAATGTGGATTCC-3′.
Sequences were aligned, and pairwise comparisons were performed using the GCG Wisconsin Package (version 10.0, Genetics Computer Group, Madison, WI). Protein structure was predicted using the Swiss-Prot package (25) (http://www.expasy.ch/sprot/).
KIR3DL1 and KIR3DL2 subtyping
Generic KIR3DL1, KIR3DS1, and KIR3DL2 typing of genomic DNA was performed as previously described, with primers based in conserved regions specific to each KIR family (15). For KIR3DL1 and KIR3DL2 subtyping, primers designed to discriminate allele-specific polymorphisms were paired with KIR3DL1 or KIR3DL2 locus-specific primers. All primers are listed in Table II⇓. Fig. 1⇓ shows the alleles amplified by each primer set and describes primer combinations and PCR conditions. Amplification of genomic DNA was performed in 25 μl reactions using 100 ng of DNA, 0.625 U of AmpliTaq polymerase, 2.5 μl of 10× buffer, and 0.2 mM dNTPs. Primers were used at a concentration of 0.5 μM.
PBMC were stained with the following Abs: CD3 (Leu 4) PerCP, and CD56 (Leu 19) FITC (Becton Dickinson, Mountain View, CA). The anti-KIR3DL1 Ab DX9 was PE-conjugated as previously described (19). PE-conjugated Z27 (anti-KIRp70) was purchased from Beckman Coulter (Miami, FL). Analysis was performed on a FACScan flow cytometer (Becton Dickinson) using the CellQuest software package.
KIR3DL1 and KIR3DL2 are highly polymorphic genes
Peripheral blood NK cells from 19 donors were analyzed for binding the anti-KIR3DL1 mAb, DX9, using flow cytometry. Within this donor panel were found the four reaction patterns previously described (20): a unimodal subset of cells with high binding, a unimodal subset of cells with low binding, a combination of cells with high and low binding (bimodal pattern), and no binding at all. These patterns are not unique to the DX9 Ab, as they were also obtained with the Z27 anti-KIR3DL1 Ab (Fig. 2⇓). Cells with low DX9 binding had a mean fluorescent intensity (MFI) of 38–208, whereas cells with the high DX9 binding phenotype had an MFI of 421–739. Fourteen of the donors were selected to represent the four patterns of DX9 binding, and RNA was made from their PBMC. cDNA clones encoding KIR3DL1 were obtained by RT-PCR and sequenced. As well as KIR3DL1, this approach yielded clones encoding KIR3DS1 and KIR3DL2, thus permitting analysis and comparison of polymorphism in all three groups of human KIR3D (Fig. 3⇓).
cDNA sequences corresponding to 10 different KIR3DL1 alleles were defined; three of them were previously described, and seven were new (Table I⇑). Seven KIR3DL2 alleles were defined, of which four were novel. When combined with other previously described sequences, nine KIR3DL2 alleles were defined. Only one form of KIR3DS1 was found, corresponding to the previously described KIR3DS1v (26). The KIR3DL1 and KIR3DL2 alleles were numbered using a system based upon that used for HLA alleles (22) (Table I⇑).
From the cDNA sequences, methods were developed for subtyping genomic DNA for KIR3DL1 and KIR3DL2 alleles. These were used first to confirm the assignments made from sequencing and second to determine genotypes for the members of the donor panel for whom sequences had not been determined (Fig. 3⇑). Ten panel members were heterozygous for KIR3DL1 variants; nine panel members expressed just one KIR3DL1 variant. In the latter group were the four individuals who had KIR3DS1. From these data KIR3DS1 appears to segregate as though it were an allele of the KIR3DL1 locus (16). Fourteen of the panel members were heterozygous at the KIR3DL2 locus. Three panel members (donors 8, 9, and 10) appeared homozygous at both KIR3DL1 and KIR3DL2 loci. Assignments of homozygosity should not be considered definitive because of the possibility that certain alleles failed to be detected by the methods we used. Apparent from the number of alleles and the extent of heterozygosity in the donor panel studied is that both KIR3DL1 and KIR3DL2 are highly polymorphic genes.
High, low, and no binding of DX9 to NK cells are properties of different KIR3DL1/3DS1 alleles
Within the donor panel, levels of DX9 binding to NK cells were correlated with KIR3DL1/3DS1 genotype (Fig. 3⇑). NK cells from donor 1 did not bind DX9, showing that both the 3DL1*00401 and 3DS1 genes of this donor gave a null phenotype. This assignment for 3DL1*00401 was confirmed by the data from donor 2, who only had 3DL1*00401 and whose NK cells did not bind DX9.
NK cells from donors 3, 4, and 5 bound DX9 at high levels with unimodal distribution and had 3DS1 in combination with either 3DL1*001 or 3DL1*003. As 3DS1 gives a null phenotype, both the 3DL1*001 and 3DL1*003 genes must independently produce a high binding, unimodal phenotype. Similarly, NK cells from donors 6 and 7 had unimodal high binding phenotypes produced by genotypes consisting of a 3DL1*004 allele in combination with either a 3DL1*001 or a 3DL1*002 allele. As 3DL1*004 gives a null phenotype, 3DL1*001 was responsible for the unimodal high binding phenotype of donor 6, and likewise, 3DL1*002 was responsible for the high binding phenotype of donor 7. Donors 8–13 all had unimodal high binding phenotypes and were typed as having either one or both of the high binding alleles, 3DL1*001 and 3DL1*002.
Donor 14 had NK cells that bound DX9 at low levels with unimodal distribution. As 3DL1*005 was the only gene of the 3DL1/3DS1 group found in this donor, these data indicated that this allele gives a unimodal low binding phenotype. Consistent with this assignment was that donor 15, who also had NK cells with a unimodal low binding phenotype, had the 3DL1*005 allele. Donor 15 also had the 3DL1*007 allele, which could have contributed either a low binding phenotype or, alternatively, a no binding phenotype.
Clonal expression of the two KIR3DL1 alleles in a KIR3DL1 heterozygote
A panel of 44 NK cell clones was established from donor 7, who is heterozygous for alleles KIR3DL1*002 and KIR3DL1*004, the first of which is associated with high DX9 binding, and the second with no DX9 binding. Each NK cell clone was assessed for its expression of the two KIR3DL1 alleles by RT-PCR typing and for binding the DX9 mAb (Table III⇓). Seventeen clones bound the DX9 Ab, and they were the same 17 clones that typed for the KIR3DL1*002 allele. Seven of the 17 clones also typed for expression of KIR3DL1*004. Four additional clones typed for KIR3DL1*004 expression in the absence of KIR3DL1*002 expression, and these clones did not bind the DX9 Ab. The remaining 23 NK clones expressed neither KIR3DL1 allele. This result demonstrates that the two KIR3DL1 alleles are differentially expressed in the NK cells of donor 7 and are thus independently expressed in a clonal fashion. Consequently, these data also provide a direct demonstration that the different cell surface phenotypes detected by the DX9 Ab are due to expression of different alleles. Although the two KIR3DL1 alleles can be expressed independently of each other, the observed frequency of NK cell clones expressing both KIR3DL1 alleles (16.0%) is greater than that expected from the product of the individual frequencies of expression from the two alleles (9.6%). It is therefore possible that initiation of transcription at one KIR3DL1 allele in a cell increases the probability of transcription at the second allele. In the clones that expressed both KIR3DL1 alleles, the mean level of DX9 binding was the same as that for clones expressing only KIR3DL1*002. Thus, the level of expression of one allele was not affected by expression of the other.
Bimodal patterns of DX9 binding to NK cells are the property of certain combinations of KIR3DL1 alleles
NK cells from both donors 16 and 17 had bimodal patterns of DX9 binding and were heterozygotes with one KIR3DL1 allele that gives a high binding phenotype (3DL1*002) and a second allele that gives a low binding phenotype (3DL1*005; Fig. 3⇑). In these two individuals, the bimodal pattern of DX9 binding is due to the clonal expression of these two alleles. Thus, we can infer that the NK cells binding DX9 at low level were those expressing only 3DL1*005, whereas those binding DX9 at high level comprised cells expressing only 3DL1*002 plus the smaller subpopulation of cells expressing both 3DL1*002 and 3DL1*005.
Two other members of the panel, donors 18 and 19, also had NK cells with a bimodal pattern of DX9 binding. These data also correlated with heterozygosity for KIR3DL1 alleles determining low and high DX9 binding. Donor 18 has the 3DL1*007 allele, which from the analysis of donor 15 is known to confer either a low binding phenotype or a no binding phenotype (see above). Thus, in combination, the data from donors 15 and 18 show that 3DL1*007 confers a low binding phenotype and is responsible for the NK cells in donor 18 that bind DX9 at low levels. Consequently, expression of 3DL1*008, donor 18’s other KIR3DL1 allele, was responsible for producing the subset of NK cells with the high DX9 binding phenotype. Among the panel members, donor 18 was the only individual to have the 3DL1*008 allele. In contrast, donor 19 had 3DL1*001, a common allele that produces a high DX9 binding phenotype, in combination with 3DL1*006, an allele found only in this member of the panel. Thus, expression of 3DL1*006 is probably responsible for the subset of donor 19’s NK cells that bind DX9 at a low level.
This analysis shows that different KIR3DL1 alleles and their combinations are responsible for the variety of cell surface phenotypes detected with the KIR3DL1-specific mAb DX9. We find that alleles 3DL1*001, 002, 003, and 008 determine high binding, alleles 3DL1*005, 006, and 007; low binding and the 3DL1*004 alleles determine no binding.
Inheritance of patterns of DX9 binding in families
To test further the correlations between KIR3DL1 genotype and cell surface phenotype made from analysis of the donor panel, we re-examined families in which the DX9 binding patterns had been previously described but could not be explained (20). When family members were typed for KIR3DL1 alleles, a precise correlation between genotype and DX9 binding pattern was observed. Data from two representative families are shown in Fig. 4⇓. In both families it can be seen that children whose NK cells gave a bimodal pattern of DX9 binding have inherited a high binding KIR3DL1 allele from one parent and a low binding KIR3DL1 allele from the other parent. Members of family A have either unimodal high or bimodal DX9 binding patterns that arise from segregation of the high binding KIR3DL1*001 and 3DL1*002 alleles and the low binding 3DL1*005 allele. In family F the segregation of KIR3DL1 alleles that determine high (3DL1*002), low (3DL1*005), and no (3DL1*004, 3DS1) binding of the DX9 Ab results in siblings with NK cells exhibiting three different phenotypic patterns of DX9 binding: no binding, unimodal low binding, and bimodal binding. These data clearly show how the pattern of DX9 binding by human NK cells is determined by the combination of KIR3DL1/3DS1 alleles a person inherits.
Definition of a KIR haplotype lacking KIR3DL1 and KIR3DS1 in a family
Analysis of the African-American family C was particularly informative and revealed a KIR haplotype that has neither the KIR3DL1 nor the KIR3DS1 gene (Fig. 5⇓). Two types of KIR haplotype have been distinguished from the correlation of Southern blotting patterns with KIR genotype (15): A haplotypes are characterized by having KIR2DS4 as the only gene encoding a short-tailed KIR with the D1 + D2 configuration of Ig-like domains, and B haplotypes are characterized by KIR2DL5 (27), additional short-tailed KIR, and a large 24-kb HindIII fragment. Five KIR haplotypes were shown to segregate in family C, two of which are A haplotypes (A1 and A2) and three of which are B haplotypes (B1, B2, and B3). Each haplotype could be distinguished by the KIR3DL1 or KIR3DS1 genes it does or does not carry. The A1, A2, and B2 haplotypes carry the 3DL1*004, 3DL1*002, and 3DL1*001 alleles, respectively. In contrast, haplotype B1 has KIR3DS1, and haplotype B3 has neither a KIR3DL1 gene nor KIR3DS1. Further supporting this conclusion were sequence analyses of KIR3DL cDNA from all available family members and of all KIR cDNA from family member C5. The latter study showed that the B3 haplotype is also distinguished by a novel recombinant form of the KIR2DL2 gene (KIR2DL2v1 GenBank accession no. AF285433) (28).
Patterns of variation in KIR3DL1 and KIR3DL2 are distinct
Comparable numbers of KIR3DL1 and KIR3DL2 alleles were defined in this study. However, by several criteria there are striking differences in the polymorphism of the two genes. In nucleotide sequence KIR3DL1 alleles are more divergent from each other than are KIR3DL2 alleles, as measured either by simple pairwise comparisons (Fig. 6, A and B) or by the numbers of polymorphic nucleotide positions: 36 for KIR3DL1 and 15 for KIR3DL2. Consequently, the average genetic distance between KIR3DL1 alleles (1.1% difference in nucleotide sequence) is >5-fold greater than for KIR3DL2 alleles (0.2% difference). These differences between the genes would have been even greater if KIR3DS1 had been included in the KIR3DL1 analysis.
Although KIR3DL2 has less nucleotide diversity than KIR3DL1, a comparable difference is not seen in pairwise comparisons of the amino acid sequences. Almost all nucleotide substitutions in KIR3DL2 alleles are nonsynonymous, whereas KIR3DL1 alleles differ by a mixture of synonymous and nonsynonymous substitutions (Fig. 6⇓, C and D). A further difference is that the KIR3DL1 alleles have a greater proportion of shared nucleotide substitutions (74%) than do KIR3DL2 alleles (23%). Thus, there is greater evidence for recombination in the generation of KIR3DL1 diversity, a mechanism that is also likely to have produced KIR3DS1. In contrast, point mutation is more apparent in the pattern of KIR3DL2 polymorphism. More than 80% of the substitutions in the extracellular domains of KIR3DL1 are in loops of the Ig-like domains, whereas for KIR3DL2 >80% of the substitutions are in strands of the β-pleated sheets (Fig. 7⇓). These striking differences all point to the combined forces of mutation, selection, and drift having operated in very different ways upon the human KIR3DL1 and KIR3DL2 genes.
KIR3D genes encode inhibitory receptors that engage HLA-A and -B ligands, products of the most polymorphic HLA class I genes. Our study demonstrates that the KIR3DL1 and 3DL2 genes are also highly polymorphic, and for KIR3DL1 the polymorphism has been directly correlated with differences in cell surface phenotype as detected with the KIR3DL1-specific mAb, DX9 (19, 20). This analysis also shows for the first time that in heterozygotes the two alleles of a KIR gene (KIR3DL1) are expressed differentially in clonal fashion, giving rise to four subpopulations of NK cells: positive for one allele, positive for the second allele, positive for both alleles, and negative for both alleles.
From sequence analysis of just 14 donors, eight different inhibitory KIR3DL1 allotypes were defined. In addition, the noninhibitory KIR3DS1 is particularly homologous to KIR3DL1 in the extracellular domains, and in this donor panel its gene distributed as though it were an allele of KIR3DL1. The eight KIR3DL1 allotypes subdivide into three groups according to how NK or T cells expressing them bind the DX9 Ab: with high level binding, low level binding, or no binding. These different levels appear to depend solely upon the KIR3DL1 allele expressed, not upon other genetic or environmental factors including HLA class I type (20). In heterozygotes who have a high and a low binding KIR3DL1 allele, their independent and clonal expression produces a characteristic bimodal pattern of DX9 binding in flow cytometry. In these patterns the small proportion of KIR3DL1-expressing cells that express both alleles probably group with the cells expressing just the high binding allele, a conclusion supported by the analysis of NK cell clones. The strength of our conclusion is the ability to predict the cell surface phenotype of family members based solely upon their KIR genotype.
Through high resolution analysis of an informative family we have been able to demonstrate the presence and segregation of a KIR haplotype that lacks both KIR3DL1 and KIR3DS1. Thus, null haplotypes also represent an element in the population diversity of KIR3DL1. We estimate this type of haplotype may be present in human populations at frequencies of about 7% (data not shown), suggesting that individuals lacking KIR3DL1 and KIR3DS1 altogether could be present at frequencies of up to 0.5%. In this regard analysis of an Australian panel of 147 donors revealed three who typed for neither KIR3DL1 nor KIR3DS1 (29). Such individuals could lack the genes or, alternatively, have novel KIR3DL1 alleles that did not anneal to the typing primers.
Comparison of KIR3DL1 allotypes identifies four positions within the amino acid sequence (182, 283, 320, and 373) where polymorphic substitutions present in more than one allotype correlate with DX9 binding phenotype. Thus, all four alleles (3DL1*001, 002, 003, and 008) associated with high levels of DX9 binding encode proline 182, tryptophan 283, isoleucine 320, and glutamate 373. By contrast, the four alleles (3DL1*004, 005, 006, and 007) associated with the low or no binding DX9 phenotypes have one or more of the following substitutions: serine 182, leucine 283, valine 320, and glutamine 373 (Fig. 7⇑A). The 3DL1*004 allele that gives no binding has all four substitutions, whereas the three alleles that give low binding have either one or two of them. In addition, substitutions that are unique to individual allotypes might also contribute to the DX9 binding phenotype. For example, the unique cysteine residue at position 277 of 3DL1*006 might form disulfide-bonded homo- or heterodimers in which the DX9 epitope becomes obscured.
Two of the shared positions of substitution are in the extracellular Ig-like domains and could directly affect the structure of the epitope recognized by DX9. These are residue 182 in the D1 domain and residue 283 in the D2 domain. A model in which the D1 and D2 domains form the site of the DX9 epitope is consistent with the results reported by Rojo et al. (30), who from analysis of KIR3DL1 mutants lacking individual Ig-like domains concluded that the D1 and D2 domains were essential for expression of the DX9 epitope. In further support of this model is the observation that DX9-binding chimpanzee KIR3DL has the same residues at positions 182 and 283 as high binding KIR3DL1 allotypes (31).
Previously it was shown that Fab and F(ab′)2 of the DX9 Ab dissociated with similar kinetics from high and low binding NK cells from the same donor, and that similar titration curves were given by the two types of Ab fragment in binding to both high and low binding cells (20). These data favor a model in which the cell surface KIR3DL1 molecules on the high and low binding NK cells bind the Ab with similar affinity, but the number of epitopes accessible to Ab differ on the two types of cell. Knowing now that the cells express different KIR3DL1 alleles suggests that the 3DL1*005 allele, conferring a low binding phenotype, gives rise either to fewer protein molecules on the cell surface than the high binding 3DL1*002 allele or to molecules for which a substantial proportion do not bind the DX9 Ab. That 3DL1*005 is only distinguished from the high binding 3DL1*001 allele at positions 182 and 283 in the extracellular domains, argues against these substitutions acting to reduce the affinity of binding to DX9.
The remaining two positions of substitution (320 and 373) that distinguish low and no binding KIR3DL1 allotypes from high binding ones are found in the transmembrane region (position 320) and the cytoplasmic domain (position 373). Their importance is highlighted by the low binding allele 3DL1*007 for which these are the only substitutions that distinguish it from the group of high binding alleles. These substitutions cannot directly affect the structure of the extracellular epitope recognized by the DX9 Ab, so their effects must be indirect in nature. It is possible that they influence some aspect of the folding of the protein, its translocation to the cell surface or interactions there with other molecules. Alternatively, nucleotide polymorphisms in regulatory, noncoding regions that are in linkage disequilibrium with the polymorphisms in codons 320 and 373 could reduce or eliminate transcription or translation of the alleles conferring low or no binding phenotypes.
Although polymorphism in the KIR3DL1 gene is sufficient to explain differences in the level with which the DX9 Ab binds to NK cells and T cells, it alone cannot account for differences in the frequency of DX9-binding NK cells that distinguish individuals within the population. Thus, donors 16 and 17 have the identical combination of 3DL1 alleles, but donor 16 has twice as many DX9 binding NK cells as donor 17 (Fig. 3⇑). On the other hand, possible trends can be seen in our data, such as 3DL1*001 being generally expressed by a higher proportion of NK cells than 3DL1*002. Polymorphism in the KIR3DL1 promoter, which need not be in complete linkage disequilibrium with the structural polymorphisms defined here, could contribute to differences in the frequency of KIR3DL1-expressing cells. In this regard, polymorphisms in the promoter of the KIR2DL5 gene may set a precedent (32). Also of likely importance are the identities of the other KIR genes in an individual’s genotype and the strengths of their promoters.
The results we present here, the genomic analysis of Wilson et al. (16), and other population studies of KIR genotype (29, 33) paint a consistent picture in which KIR3DL2 appears to be an invariant component of human KIR haplotypes, whereas KIR3DL1 and KIR3DS1 are only present on some haplotypes. In the donor panel we studied KIR3DS1 was represented by a single sequence that was distributed as though it were an allele of the same locus as KIR3DL1. From sequence analysis of one KIR haplotype having the 3DS1 gene and another having the 3DL1 gene Wilson et al. reached the identical conclusion; they found KIR3DL1 and KIR3DS1 to be closely related in sequence and to occupy orthologous positions on the 3′ side of the invariant KIR2DL4 gene. Analysis of partial sequences for the KIR3DL1 and KIR3DS1 genes in B cell lines derived from American Indian populations is also consistent with them being alleles (C. Vilches and M. Pando, unpublished observations),
Other results, however, indicate that the situation in some KIR haplotypes may be more complicated. In KIR genotyping an Irish panel, Crum et al. (33) identified one person, from 90 studied, who typed for two variants of KIR3DL1 as well as KIR3DS1. In ongoing studies of patients undergoing bone marrow transplantation we have also identified an individual with a similar genotype (H. Shilling, unpublished observations). Thus, in these individuals one KIR haplotype must contain either KIR3DL1 and KIR3DS1 (the interpretation proposed by Crum et al.) or, alternatively, two variants of KIR3DL1. Evidence for the latter type of haplotype is the report by Vyas et al. (31), who found four different KIR3DL1 variants to be expressed by one donor. The close juxtaposition of the KIR genes and their separation by highly homologous segments are properties that could favor unequal recombination and the production of haplotypes with duplicated genes (16). In this regard a haplotype containing two copies of KIR2DL5 has been recently described (32). The low frequency of individuals with KIR3DS1 and two KIR3DL1 variants suggests that haplotypes containing such duplications are relatively rare and probably represent recent derivatives of haplotypes carrying just one KIR3DL1/3DS1 gene. If that is the case the existence of haplotypes carrying two KIR3DL1/3DS1 should not be seen as a refutation of the model that KIR3DL1 and KIR3DS1 have an allelic relationship, as has been proposed (33, 34), but as an exception to this general rule.
Deletion and duplication of the KIR3DL1 gene as well as formation of the short-tailed form KIR3DS1 are all changes likely to have resulted from unequal crossing over between KIR haplotypes. Recombination between KIR3DL1 alleles is also implicated in the generation of new KIR3DL1 alleles (26). In contrast to KIR3DL1, the KIR3DL2 gene appears relatively resistant to recombination. A single KIR3DL2 gene encoding a long-tailed KIR appears a stable feature of human KIR haplotypes, and point substitutions are a more characteristic feature of KIR3DL2 polymorphism than is the case for KIR3DL1. Indeed, the higher frequency of nonsynonymous substitutions in KIR3DL2 could be due to diversifying selection in a gene in which low recombination does not facilitate hitchhiking of synonymous substitutions. Although both the KIR3DL1 and KIR3DL2 genes are highly polymorphic their patterns of diversity are strikingly different, as is also the case for their respective HLA-B and -A ligands (35). The outstanding question is the extent to which the differences in KIR3DL1 and 3DL2 polymorphism reflect structural constraints in the genomic organization of KIR haplotypes or in the selection imposed by pathogens on the immune functions of these molecules.
Comparison of human and common chimpanzee KIR shows that the lineage of KIR, which in humans comprises KIR3DL1 and KIR3DL2, is represented by a single gene (Pt-KIR3DL1/2) in common chimpanzee (31). The structure of Pt-KIR3DL1/2 is a pastiche of elements shared with either KIR3DL1 or KIR3DL2 in which the extracellular domains are more like KIR3DL1 and the intracellular domains are more like KIR3DL2. Correlating with this structure, the Pt-KIR3DL1/2 receptor has an inhibitory specificity that is different from those defined for the human KIR3DL but which includes both MHC-A and -B allotypes. This comparison between species shows that structural divergence within this lineage of KIR3D is associated with change in MHC class I specificity. It is therefore a reasonable working hypothesis that among the different KIR3DL1 and KIR3DL2 allotypes we have defined are ones with distinctive HLA class I specificity.
We thank those who have donated blood to this project, and Drs. Stewart Cooper and Salim Khakoo for their help in obtaining blood samples. We thank Drs. Marco Colonna, Lewis Lanier, Eric Long, and Charles Lutz for their comments on the draft manuscript.
↵1 This work was supported by National Institutes of Health Grants AI22039 and AI17892 (to P.P.). C.M.G. was the recipient of a Leukemia Research Foundation Fellowship. C.V. was supported by a fellowship from the Instituto de Salud Carlos III, Spain (FIS BAE 98/5105).
↵2 Current address: Servicio de Inmunología, Hopital Universitario Clínica Puerta de Hierro, San Martín de Porres 4, 28035 Madrid, Spain.
↵3 Address correspondence and reprint requests to Dr. Peter Parham, Departments of Structural Biology and Microbiology and Immunology, Sherman Fairchild Building, Stanford University School of Medicine, Stanford, CA 94305. E-mail address:
4 Abbreviations used in this paper: KIR, killer cell Ig-like receptors; MFI, mean fluorescent intensity.
- Received September 27, 2000.
- Accepted December 14, 2000.
- Copyright © 2001 by The American Association of Immunologists