Co-evolution of KIR2DL3 with HLA-C in a human population retaining minimal essential diversity of KIR and HLA class I ligands

Yucpa Have the Highest Frequency Worldwide of HLA-C*07, a Strong NK Cell Educator.

The Yucpa exhibit a typical Amerindian HLA class I distribution, in which a fraction of the major allele groups worldwide is represented (11, 20). One HLA-A and three HLA-B alleles are “new” Amerindian-specific variants, the other 15 alleles, including all six HLA-C alleles, are shared with Asian populations and represent “founder” alleles brought by the Asian migrants who populated the Americas (Fig. 1A). This set of HLA class I allotypes retains the C1, C2, and Bw4 epitopes, the major ligands for inhibitory KIR (7), but lacks the A3/A11 epitope recognized by KIR3DL2 (21). During NK cell development, interactions between C1, C2, and Bw4 and cognate KIR determine the strength with which mature NK cells respond to cells whose HLA class I expression is perturbed by disease, a developmental process termed NK cell education (22). The A3/A11 epitope contributes little to NK cell education (23, 24), a distinction that correlates with the exceptionally high sensitivity of KIR3DL2 to the peptide bound by HLA-A*03 or A*11 (21).

Distinguishing the Yucpa is the frequency of HLA-C*07, the highest observed worldwide and an outlier from the distribution formed by other populations. In contrast, the Yucpa frequency of HLA-A*02, as common and widespread as HLA-C*07, is well within the distribution (Fig. 1B). Furthermore, the neighboring but linguistically separated Bari Amerindians have a C*07 frequency within the range of other populations. Some 97% of Yucpa carry C*0702 and 57% are homozygotes. HLA-C*07 carries the C1 epitope, the ligand for the inhibitory KIR2DL3 receptor. Comparison of the capacity of different C1-bearing HLA-C allotypes to educate KIR2DL3-expressing NK cells, has shown that HLA-C*07 is the most potent (23).

Natural Selection Has Maintained a Balance of KIR Haplotypes and Alleles in the Yucpa.

Although only three (KIR2DL4, 3DL2, and 3DL3) of the 14 KIR genes are fixed, all but KIR2DS3 are retained in the Yucpa (25). To define the underlying polymorphism in these genes, we determined complete allele-level KIR haplotypes and their frequencies (Fig. 2). Five A haplotypes are evenly distributed with a cumulative frequency of 46%. In contrast there is a dominant B haplotype, B1, with a frequency of 47.5%, and three low-frequency haplotypes, all representing recombinants between B1 and one of the A haplotypes (Fig. 2A). The even balance between A and B haplotypes was preserved in each of three villages contributing samples to this study (Table S1). Striking is that B1 shares no KIR allele with any A haplotype; even the 3DP1 pseudogene is represented by a distinctive allele. A1/B1 heterozygotes (21% of the Yucpa population) have 19 different KIR, thus they possess two-thirds of the 29 KIR variants present in the population.

Simulations show that maintenance of gene-content diversity of the magnitude observed between Yucpa KIR haplotypes, which is close to the maximum possible, was highly improbable under neutral evolution or under positive or negative selection, which would lead to loss or fixation of haplotypes, respectively (Table 1). Thus, the most likely cause of the haplotype diversity of the Yucpa is balancing selection. To further assess this possibility, we analyzed the haplotype sequences for Tajima's D in a sliding window. Statistically significant evidence for balancing selection was obtained for all of the polymorphic KIR except the 3DP1 pseudogene (Fig. 2B). Supporting evidence for this mode of selection on KIR is the numerous functional differences caused by the polymorphism (8, 26–28). For example, the four Yucpa KIR3DL1/S1 alleles represent the breadth of functional variation defined at this well-studied locus: the three main lineages (29), both the activating and inhibitory receptors, and inhibitory receptors having high and low levels of expression on NK cell surfaces (30). The polymorphic HLA class I genes provide a classical example of balancing selection seen in all populations (31). For the Yucpa, the evidence for balancing selection on the KIR is as strong as that on their HLA class I ligands (Table S2). This contrasts with the ABO system of blood group antigens, which has a strong signature of balancing selection in many populations (32) but no variation at all in the Yucpa (18). Furthermore the dominance of group O in Amerindians existed before European contact as it is observed in 2,000-year-old skeletal remains (33). Such dominance of O is not typical of Asians and Siberians, who retain the ABO polymorphism (34) and were likely the source of Amerindian founders (35).

KIR2DL3 Variants with Lower Avidity for C1 Have Largely Replaced the 2DL3*001 Founder.

Of the 29 KIR in the Yucpa, 25 are founder alleles and four are variants not seen in other populations. Of these variants (gray-shaded boxes in Fig. 2A), the substitutions in the 3DP1 pseudogene and the synonymous substitution that distinguishes 3DL3*01002 are unlikely to have had any functional effect. However, the single non-synonymous substitutions that distinguish the 2DL3*008N and 2DL3*009 variants from 2DL3*001, the founder allele for this inhibitory C1 receptor, are likely to have altered its function.

KIR2DL3*008N differs from 2DL3*001 by deletion of one nucleotide from codon 86, causing premature termination at codon 124 (Fig. 3A). As the encoded protein lacks half the ligand-binding site and the signaling domain, 2DL3*008N is almost certainly non-functional. KIR2DL3*009 differs from 2DL3*001 by substitution of proline for arginine at position 148 in domain D2 (Fig. 3A). Previous comparison of 2DL3 to 2DL2, showed that substitution of arginine 148 for cysteine in 2DL2 (Fig. 3A) increased its avidity for C1 and its cross-reactivity with C2 (28). Further pointing to the functional importance of variation at position 148, phylogenetic analysis showed position 148 was a site for positive natural selection during hominoid evolution (29). Although this residue does not directly contact bound HLA-C (36), it is proposed to modulate binding avidity by altering the angle of the hinge between D1 and D2 (28).

To determine the effect of the proline 148 substitution, we made Fc-fusion proteins from 2DL3*001 and 2DL3*009 and compared their binding to beads individually coated with one of 95 different HLA-A, -B, and -C allotypes (Fig. 3B). Although 2DL3*001-Fc and 2DL3*009-Fc exhibited similar selectivity for HLA-C, and two exceptional HLA-B allotypes (B*4601 and B*7301), that carry the C1 epitope (28), 2DL3*001-Fc consistently bound to higher level than 2DL3*009 (Fig. 3B). This was not because of difference in quality of the fusion proteins, both bound equivalently to the conformation-dependent anti-KIR2DL2/3 antibody, DX27 (Fig. S1). Moreover, the difference was reproduced in titrations against HLA class I-coated beads, as illustrated for HLA-C*0702 and C*0302 that account for 92% and 5% of Yucpa C1 epitopes, respectively (Fig. 3C). One possible cause of the difference is that 2DL3*009 and 2DL3*001 have different affinities for C1, another is that they bind to different subsets of the target HLA-C molecules. Distinguishing such subsets are the variable peptides that form an integral component of MHC class I, and which are known to influence HLA-C interaction with KIR2D (36). Thus, 2DL3*009 could bind to a subset of the HLA-C*0702 molecules bound by 2DL3*001. Whichever interpretation is correct, the net effect is the same, namely 2DL3*009 has less avidity for C1 than 2DL3*001. For 2DL3*008N, which is nonfunctional, this trend to lower avidity is taken to the limit.

When KIR2DL2 and KIR2DL3 were first identified, they were considered (and named) as separate genes encoding C1 receptors because of the extent of their sequence differences (9), but subsequent population and family analyses showed they segregate as alleles (37). In general, KIR2DL3 is fixed on A haplotypes and KIR2DL2 is present only on B haplotypes. In the Yucpa, the haplotype segregation is particularly strong: 92% of the B haplotypes having 2DL2, only 8% 2DL3 (Fig. 2A). Of note, KIR2DL2 is represented only by the 2DL2*003 founder, and the only form of KIR2DL3 on B haplotypes is also the founder, 2DL3*001. In contrast, KIR2DL3 on the A haplotypes is represented by three forms: the founder, 2DL3*001, representing 16.4% of total Yucpa 2DL3; the low avidity 2DL3*009, representing 68.8%; and the non-functional 2DL3*008N representing 14.8% (Fig. 3A). Thus, the KIR2DL3 founder has been largely replaced by the two new variants. We could not find 2DL3*008N or 2DL3*009 in other populations, including the neighboring but linguistically separated Bari and the more distant Warao Amerindians (n = 41), and Venezuelan Mestizos (n = 21). The new variants appear to have restricted distribution, or to be at very low frequencies in other populations, and it is possible that 2DL3*008N and 2DL3*009 are Yucpa-specific, having evolved in that population by point mutation from 2DL3*001. Because 2DL3*009 is present on haplotypes A1, A2, and A5, at least two of them acquired the new variant by recombination (Fig. 2A). The substantial changes in the structure and function of KIR2DL3, but in none of the other functional KIR in the Yucpa, points to these changes being a consequence of natural selection, as does the sliding-window analysis shown in Fig. 2B.

The Yucpa Retain the Worldwide Range of KIR Haplotypes, Genes, and Alleles.

The Yucpa descend from Asian migrants who arrived in Alaska approximately 15,000 years ago and peopled the Americas through population expansion and southward migration (14). During the approximately 750 generations following the initial bottleneck (13), Yucpa ancestors likely experienced cycles of population contraction and expansion caused by epidemics of infectious disease, most recently an epidemic of combined hepatitis B and D (19). We have defined the system of KIR and HLA class I ligands in the surviving Yucpa population.

Despite the population bottlenecks, we find that with 29 forms of KIR and 19 forms of HLA-A, -B, and -C, the Yucpa retain almost all major elements of the KIR-HLA class I system present worldwide, including all components for which immunological functions are well defined. The one KIR gene missing from the Yucpa, KIR2DS3 is of questionable function, because its protein product is not cell-surface expressed (27). And in the Yucpa, and other Amerindian groups, the absence of KIR2DS3 is compensated by increased frequency of the closely related KIR2DS5 (25, 38–40). The polymorphic KIR genes of the Yucpa retain the breadth of the polymorphism worldwide with three to four alleles, and similar numbers of HLA class I allotypes provide the three major epitopes recognized by KIR: C2 carried by HLA-C, C1 carried by HLA-C and HLA-B, and Bw4 carried by HLA-A and HLA-B. Absent from the Yucpa, and generally rare in Amerindians, is the A3/A11 epitope recognized by KIR3DL2.

The Yucpa maintained high KIR diversity through population bottlenecks as a consequence of balancing selection. At the time of sampling in 1993, the population had an even frequency of A and B haplotypes that are maximally divergent in gene content and share no single KIR allele. The B haplotype is essentially invariant, whereas five A haplotypes, give breadth to the polymorphism of KIR2DL3, 2DL4, 2DS4, 3DL1, 3DL2, and 3DL3. Contrasting with their diverse KIR, the Yucpa retain only one of three to five allele lineages of the ABO blood group locus, which in other populations have been subject to balancing selection (41). Because of the intensity of selection and small population size, the Yucpa system of six KIR haplotypes and three HLA class I epitopes has become streamlined, and emerges as a candidate for having the minimal essential diversity needed for long-term survival of a human population.

Selective Evolution of the Interaction between KIR2DL3 and C1 in the Yucpa.

All Yucpa HLA class I alleles encoding KIR ligands, and 25 of the 29 Yucpa KIR, are founder alleles; they came from Asia with the original migrants and have remained unchanged since then. In this general context of stability, the functional changes in KIR2DL3 and its cognate C1 ligand are striking. The founder allotype, 2DL3*001 has severely decreased in frequency, being replaced by two variants, mainly by 2DL3*009, that has reduced avidity C1, but also by the non-functional 2DL3*008N. Such changes are evidence for selection to reduce the strength of the KIR2DL3-C1 interaction. These changes in KIR2DL3 are associated with an unusually high frequency of HLA-C*07, the dominant C1-bearing allotype and the one most potent at educating NK cells to attack and kill cells deficient in MHC class I (23). In the Japanese, who share many KIR factors with the Yucpa, homozygosity of C1 results in the education of fewer 2DL3*001-expressing NK cells than C1 heterozygosity (23), suggesting that increasing the avidity of 2DL3-C1 interactions beyond a certain point is disadvantageous. Emergence of weaker KIR2DL3 variants in the Yucpa can thus be interpreted as a compensatory response to the elevated frequencies of C*0702 and of C*0702 homozygotes, one that reduces the avidity of the 2DL3-C1 interaction and increases the abundance of functional NK cells expressing KIR2DL3.

A further distinguishing characteristic of C*0702 is that its leader peptide forms complexes with HLA-E that prevent interaction with the conserved inhibitory receptor CD94:NKG2A (5). Thus, C*0702 not only favors NK cell regulation through interaction with KIR2DL3, but it may also act to disfavor education and regulation through interaction of CD94:NKG2A with HLA-E. Contrasting with C*0702, the second most frequent C1-bearing allotype in the Yucpa, C*0302, combines with HLA-E to form a high avidity ligand for CD94:NKG2A (5).

Homozygosity for C1 and KIR2DL3 has been correlated with successful termination of acute hepatitis C virus (HCV) infections in U.K./U.S. Caucasians and African Americans (42). The observed co-evolution of C1 and KIR2DL3 in the Yucpa, suggests that the combination of KIR2DL3 with C1 could also be beneficial for terminating other types of infection, including the combined hepatitis B and D that affected the Yucpa in 1979–1981. Although KIR2DL2 is allelic to KIR2DL3, it was not associated with resistance to HCV (42) and unlike KIR2DL3, it has undergone no change in the Yucpa. KIR2DL3 is a fixed locus of the A haplotype and the 2DL3*008N and 2DL3*009 variants are only present on A haplotypes, whereas KIR2DL2 is only present on B haplotypes. These qualitative differences point to the group A and B KIR haplotypes having been subject to different types of selection pressure.

A Model in Which A and B KIR Haplotypes Diversified under Selection for Immunity and Reproduction, Respectively.

Although group A KIR haplotypes protect against HCV infection, they are risk factors for the pregnancy syndromes preeclampsia (43) and recurrent miscarriage (44). Preeclampsia and eclampsia are leading causes of death for women of child-bearing age, especially in undeveloped countries (45). At risk are pregnancies in which C2-expressing fetuses are carried by group A KIR homozygous mothers; conversely, maternal group B KIR haplotypes and fetal C1 homozygosity are protective (43, 44). The inverse correlation between C2 and group A KIR haplotype frequencies in human populations worldwide, argues for the importance of selection by diseases of pregnancy (43). Although the evidence is correlative and the studies are few in number, they raise the intriguing possibility that A KIR haplotypes are principally selected for their role in immune defense, whereas B KIR haplotypes are selected for their role in placental reproduction. In this model the balancing selection that has maintained A and B haplotypes in all human populations would come from distinctive pressures upon the immune and reproductive systems. Thus, an episode of viral infection is predicted to select for A KIR haplotypes, which will be enriched in the survivors, but in subsequent expansion of the surviving population there will be selection for B haplotypes and against A.

Although individuals homozygous for A or B haplotypes are numerous, healthy and able to reproduce, none of the >140 human populations examined for KIR gene content lacks either A or B haplotypes (8, 10, 46). Thus, long-term survival of human populations appears to have selected for retention of both A and B haplotypes, a corollary being that populations losing either A or B haplotypes were out-competed by those retaining both haplotype groups. Because viruses and other pathogens can evolve rapidly in response to the human immunity, the pressure on the immune functions of the KIR is likely to be more variable and changing than the pressure on the reproductive functions. Consistent with this thesis, the group A KIR haplotype genes are highly polymorphic, whereas the group B KIR haplotype genes are generally conserved (10). Although observed in all populations studied, these properties are most vividly illustrated by the Yucpa, for whom we speculate that the exigencies of selection have retained what is both minimal and essential for long term survival in the struggle against infectious disease and other human populations.

Defining Yucpa KIR Haplotypes.

To define allele-level KIR haplotypes for the panel of Yucpa donors, we first sequenced coding regions for all KIR genes from the 23 individuals who were homozygous for KIR gene-content haplotypes (Fig. S2), then extended the analysis to selected heterozygotes and finally genotyped the remaining individuals. In 15 A/A homozygotes, KIR2DL4, 2DS4, 3DL1, 3DL2, and 3DL3 were represented by two to five alleles; KIR2DL1 and the 2DP1 and 3DP1 pseudogenes were monomorphic. Seven of these individuals were homozygous for one of three common allele-level A haplotypes: A1, A2, and A3. Reasoning that the eight A/A heterozygotes each carried a common haplotype, we defined rarer haplotypes A4 and A5. All eight B/B homozygotes were homozygous for B1 (Fig. 2A). In analyzing heterozygotes suspected to harbor additional B haplotypes, we reasoned that one KIR haplotype was already defined (either A1, A2, A3, A4, A5, or B1), thus permitting identification of B2, B3, and B4. Based on the gene sequences, allele-typing methods were devised and applied to the panel (Fig. S2). Within the panel studied were 35 members of 13 families. Segregation analysis for these family members gave results consistent with the assigned haplotypes (Table S1). The KIR haplotype distribution within the panel complied with Hardy-Weinberg equilibrium, providing further evidence for the validity of the defined haplotypes.

Statistical Analysis.

Statistical significance of Tajima's D was assessed by comparing the observed values to those expected under neutral evolution. Expected values were generated by coalescence simulations using the program ms (47). ms was used to generate 10,000 independent replicate samples under user-defined models. The simulations were performed with a founder of 100 effective population size, followed by growth to 1,000 over 600 generations; similar to the model proposed by Hey (13). The forward-simulations (Table 1 and Table S3) were performed using simuPOP 0.8 (ref. 48).

KIR/HLA-Binding Assay.

Soluble Fc-fusion proteins were produced using methods described (28). The purified fusion proteins were tested for binding to a panel of 29 HLA-A, 47 HLA-B, and 16 HLA-C allotypes using the LABScreen single-antigen beads (One Lambda) and a Luminex 100 reader (Luminex Corp.). Relative fluorescence ratios were calculated using the formula (specific binding-control bead binding)/(positive binding-control bead binding). W6/32, an antibody specific for HLA class I was used as the positive control (28). The nucleotide substitution that distinguishes 2DL3*009 was introduced to 2DL3*001 cDNA using site-directed mutagenesis (Invitrogen).