Research ArticleIMMUNOGENETICS

Inhibitory killer cell immunoglobulin-like receptors strengthen CD8+ T cell–mediated control of HIV-1, HCV, and HTLV-1

See allHide authors and affiliations

Science Immunology  09 Nov 2018:
Vol. 3, Issue 29, eaao2892
DOI: 10.1126/sciimmunol.aao2892

Fine-tuning CD8+ T cell responses

Human cytotoxic CD8+ T cells are important for defense against viral infections. Boelen et al. investigated whether inhibitory killer cell immunoglobulin-like receptors (iKIRs) carried by patients with chronic viral infections affected the efficacy of their CD8+ T cell responses. Possession of an iKIR gene along with a gene encoding a KIR ligand enhanced protective and detrimental HLA class I associations for HIV-1, HCV, and HTLV-1. Analysis of virus dynamics, in vitro survival assays, and mathematical modeling suggested that iKIR ligation enhances HLA associations by increasing T cell survival. In contrast to many reported iKIR-disease associations, these observations applied to all iKIRs and the three viral infections studied. This analysis reveals a consistent ability of iKIRs to influence clinical outcomes in human chronic viral infection.

Abstract

Killer cell immunoglobulin-like receptors (KIRs) are expressed predominantly on natural killer cells, where they play a key role in the regulation of innate immune responses. Recent studies show that inhibitory KIRs can also affect adaptive T cell–mediated immunity. In mice and in human T cells in vitro, inhibitory KIR ligation enhanced CD8+ T cell survival. To investigate the clinical relevance of these observations, we conducted an extensive immunogenetic analysis of multiple independent cohorts of HIV-1–, hepatitis C virus (HCV)–, and human T cell leukemia virus type 1 (HTLV-1)–infected individuals in conjunction with in vitro assays of T cell survival, analysis of ex vivo KIR expression, and mathematical modeling of host-virus dynamics. Our data suggest that functional engagement of inhibitory KIRs enhances the CD8+ T cell response against HIV-1, HCV, and HTLV-1 and is a significant determinant of clinical outcome in all three viral infections.

INTRODUCTION

Killer cell immunoglobulin-like receptors (KIRs) are a family of transmembrane glycoproteins with activating and inhibitory isoforms; their ligands include the human leukocyte antigen (HLA) class I molecules (1, 2). KIRs are expressed predominantly on natural killer (NK) cells and are primarily associated with innate NK cell–mediated immunity. However, recent studies have shown that KIRs can also affect adaptive T cell–mediated immunity.

Two distinct routes whereby inhibitory KIRs (iKIRs) affect T cell responses have been described (3, 4). First, iKIR expression on CD8+ T cells directly affects their longevity and function. KIRs are expressed on T cells with an effector memory TEM(RA) phenotype (CD28CD45RA+/−CD45RO−/+CCR7); the frequency of KIR+ T cells increases with age and some persistent viral infections (59). In mice, inhibitory Ly49 receptors (the functional homolog of iKIRs) enhance CD8+ T cell survival (10). Similarly, in KIR-transgenic mice, ligation of iKIRs on T cells reduces activation-induced death (11, 12). Furthermore, iKIR expression on human T cells is associated with higher levels of the survival molecule Bcl-2, reduced cell death, and impaired functionality (1315). Second, KIRs indirectly modify the CD8+ T cell response via their expression on NK cells. NK cells regulate T cell immunity by cytokine secretion and killing activated CD4+ and CD8+ T cells; this regulation may be modified by the expression of inhibitory receptors on NK cells (4, 16, 17). In mice, lymphocytic choriomeningitis virus–specific CD8+ T cell responses are modulated via NK cell–mediated killing of activated CD4+ T cells (18), and in murine cytomegalovirus infection, interleukin-10 production by NK cells impairs the CD8+ T cell response (19). In addition, human NK cells can lyse hepatitis B virus–specific CD8+ T cells in vitro (20).

The clinical relevance of an interaction between iKIRs and CD8+ T cells is difficult to infer from these earlier studies, because it is impossible to determine whether changes in CD8+ T cell lifespan in mice or in vitro alter the course of human disease in vivo. To investigate whether iKIRs have a clinically significant impact on the CD8+ T cell response, our starting point was the well-documented association between certain HLA class I alleles and disease outcome, because this represents compelling evidence that CD8+ T cells are clinically relevant. We reasoned that if KIRs significantly affect the CD8+ T cell response, then KIRs will affect HLA class I disease associations. In a previous study, we validated this approach by showing that KIR2DL2 exaggerates both protective and detrimental HLA class I associations with disease progression in individuals infected with hepatitis C virus (HCV) or human T cell leukemia virus type 1 (HTLV-1) (21). However, it remains unclear whether this constitutes a generalizable phenomenon that extends to other viruses and iKIRs. Furthermore, the mechanism underlying the effect was not investigated.

In this study, we used epidemiological data from multiple independent cohorts of HIV-1–, HCV-, and HTLV-1–infected individuals; in vitro T cell survival assays; ex vivo analyses of KIR expression; and mathematical modeling of host-virus dynamics. Our data indicate that both two-domain and three-domain iKIRs, together with their HLA class I ligands, enhance the CD8+ T cell response to HIV-1, HTLV-1, and HCV by prolonging CD8+ T cell survival and are a significant determinant of clinical outcome in all three viral infections.

RESULTS

We studied a well-characterized cohort of HIV-1 seroconverters from sub-Saharan Africa who were identified when seronegative and followed under protocol C of the International AIDS Vaccine Initiative (IAVI) (2224). First, we sought to identify KIR-HLA associations that could potentially confound our study. We define a “functional KIR gene” as a KIR gene together with a gene encoding its HLA class I ligand. Both functional KIR3DS1 (KIR3DS1 with the gene encoding its putative ligand HLA-Bw4-80I) and functional KIR3DL1 (KIR3DL1 with the gene for its ligand HLA-Bw4) have previously been associated with good NK cell–mediated control of HIV-1 infection (25). In the IAVI cohort, the protective effect of functional KIR3DS1, but not functional KIR3DL1, was replicated (table S1). To prevent confounding effects arising from strong linkage disequilibrium between the KIR genes, we therefore excluded all individuals with functional KIR3DS1. As expected, HLA-B*57 was significantly associated with low early set-point viral load (coefficient = −0.42, P = 0.004 and slow progression to low CD4 count (hazard ratio = 0.44, P = 0.02). The coefficient (Coeff) is the change in log10[early viral load set point] associated with possession of HLA-B*57, and the hazard (HR) is the relative risk of progression to low CD4 count associated with possession of HLA-B*57.

iKIR genes are associated with an enhancement of the protective effect of HLA-B*57

To determine whether HLA-B*57 was more protective in the presence of functional KIR2DL2, we stratified the cohort into individuals with and without functional KIR2DL2 and analyzed the protective effect of HLA-B*57 in each stratum. Two definitions of functional KIR2DL2 were considered: (i) strong functional KIR2DL2 (KIR2DL2 with genes encoding its strong HLA-C1 ligand) and (ii) weak functional KIR2DL2 (KIR2DL2 with genes encoding its weaker HLA-C2 ligand). For both early set-point viral load and time to low CD4 count, HLA-B*57 was significantly protective in the stratum with functional KIR2DL2 but weaker and nonsignificant in its absence; additionally, the magnitude of the protection was greater with the stronger ligand (Fig. 1A and Table 1). These differences were not attributable to cohort size because there were fewer HLA-B*57+ individuals in the stratum with functional KIR2DL2 than in the stratum without functional KIR2DL2. We then extended the analysis to other iKIRs with known ligands (KIR2DL1 with HLA-C2 and KIR2DL3 with HLA-C1). Functional KIR3DL1 could not be analyzed because there were no individuals with HLA-B*57 who lacked functional KIR3DL1. iKIRs with disputed ligands were excluded from the analysis. Both functional KIR2DL1 and functional KIR2DL3 enhanced the protective effect of HLA-B*57 on both early set-point viral load and time to low CD4 count (Fig. 1A and Table 1). The odds of seeing this enhancement by chance are P = 2 × 10−6 (permutation test).

Fig. 1 iKIR genes enhance the protective effect of HLA-B*57 on early set-point viral load and time to low CD4 count.

(A) The IAVI cohort (n = 461) was stratified into individuals with or without the functional iKIR of interest, into individuals with different inhibitory scores, and into individuals with a low or high count of functional iKIRs. In each stratum, the protective effect of HLA-B*57 on early set-point viral load was analyzed using multivariate linear regression. The coefficients (change in early set-point viral load associated with HLA-B*57) are plotted. It can be seen that the protective effect of HLA-B*57 was enhanced by all three iKIRs, by high inhibitory score, and by high count of iKIRs with a dose effect with inhibitory score. Definitions are as follows: Func 2DL1+: KIR2DL1+C2+. Func 2DL2++: KIR2DL2+C1+. Func 2DL2+: KIR2DL2+C2+. Func 2DL3+: KIR2DL3+C1+. Func 2DL1: KIR2DL1 or KIR2DL1+C2. Func 2DL2: KIR2DL2. Func 2DL3: KIR2DL3 or KIR2DL3+C1. Inhibitory scores are as follows: +++: >2.5. ++: >2.0. +: ≤2.5. −: ≤2.0. (B) Cox survival curves are plotted for HLA-B*57 individuals (black dashed line), HLA-B*57+ individuals with a low inhibitory score (red dotted line), and HLA-B*57+ individuals with a high inhibitory score (green solid line). It can be seen that individuals with HLA-B*57 and a high inhibitory score (>2.5) progressed to low CD4 count (<350 cells per mm3) at the slowest rate (n = 491). (C) The cohort (n = 461) was stratified into individuals with or without HLA-B*57, and the relationship between early set-point viral load and the count of functional iKIRs was analyzed. Early viral load is shown in gray boxes, and the corresponding linear regression is shown as a line. The number of functional iKIRs was significantly associated with decreasing viral load in HLA-B*57+ individuals (P = 0.018) but not in HLA-B*57 individuals (P = 0.59). This indicates that iKIRs enhance the HLA-B*57–associated reduction in early viral load in a dose-dependent manner and that iKIRs in the absence of HLA-B*57 are not in themselves significantly protective (see also table S2).

Table 1 Functional iKIRs enhance the protective effect of HLA-B*57 on early set-point viral load and time to low CD4 count.

The IAVI cohort was stratified into individuals with or without the functional iKIR of interest, with or without a high inhibitory score (two cutoffs considered), and with or without a high count of functional iKIRs (only one cutoff possible). In each stratum, the protective effect of HLA-B*57 on early set-point viral load (A) and time to low CD4 (B) was analyzed using multivariate linear regression and the Cox proportional hazards model, respectively. The coefficients, HRs, P values, and cohort sizes are reported. Coeff is the change in log10 viral load associated with possession of the HLA allele (a Coeff > 0 represents a detrimental effect). HR is the relative risk of progression associated with possession of the HLA allele (HR > 1 represents a detrimental effect). It can be seen that HLA-B*57 was consistently and significantly protective in the presence, but not in the absence, of functional iKIRs. Definitions are as follows: Functional 2DL1+: KIR2DL1+C2+. Functional 2DL2+ (strong): KIR2DL2+C1+. Functional KIR2DL2+ (weak): KIR2DL2+C2+. Functional 2DL3+: KIR2DL3+C1+. Functional 2DL1: KIR2DL1 or KIR2DL1+C2. Functional 2DL2: KIR2DL2. Functional 2DL3: KIR2DL3 or KIR2DL3+C1. Inhibitory score (cutoff = 2.0) indicates that the threshold inhibitory score for stratifying the cohort was 2.0. Inhibitory score (cutoff = 2.5) indicates that the threshold inhibitory score for stratifying the cohort was 2.5. Count (cutoff = 2.0) indicates that the threshold count of functional iKIRs for stratifying the cohort was 2.0. The column headings “KIR+” and “KIR” refer to the high and low strata, respectively. For example, for Functional 2DL1, “KIR+” denotes the Functional KIR2DL1+ strata and “KIR” denotes the Functional KIR2DL1 strata; for inhibitory score (cutoff = 2.0), “KIR+” denotes the strata of individuals with inhibitory score > 2.0 and “KIR” denotes the strata of individuals with inhibitory score ≤ 2.0, etc. Note that the stratification provided by the count of iKIRs is the same as one of the stratifications provided by inhibitory score. We report both because the count benefits from being a more objective measure (independent of the choice of weight), whereas the inhibitory score benefits from a greater diversity of values, typically allowing for stratification at more than one cutoff. All cutoffs that gave a sufficiently balanced analysis were considered. P < 0.001 ***; P < 0.01 **; P < 0.05 *; P < 0.1 ∙; P values are two tailed.

View this table:

Having investigated the effect of the presence or absence of individual functional iKIR, we next investigated the effect of the number of functional iKIRs. For each individual in the cohort, we assigned two numbers: (i) the “count of functional iKIR” (the count of the number of iKIR-ligand pairs) and (ii) an “inhibitory score” (similar to the count but weighted to reflect the strength of iKIR signaling; Supplementary methods). We first stratified the cohort based on the inhibitory score (all cutoffs that gave balanced strata were considered) and analyzed the strength of HLA-B*57–mediated protection in each stratum. Again, we found that the protective effect of HLA-B*57 on both early set-point viral load and progression to low CD4 count was enhanced in the high inhibitory score stratum and that the size of the protective effect increased with the magnitude of the score (Fig. 1, A and B, and Table 1). Repeating the analysis using the count of functional iKIRs gave similar results, indicating that our findings were not dependent on the choice of scoring system. Within HLA-B*57+ individuals (but not HLA-B*57 individuals), both the inhibitory score and the count of functional iKIRs were significantly associated with a decrease in early set-point viral load (Fig. 1C). This result indicates that iKIRs enhance the HLA-B*57 protective effect in a gene dose–dependent manner and shows that functional iKIRs alone (i.e., not in the context of HLA-B*57) are not significantly protective; this latter conclusion was confirmed by multivariate analysis (table S2).

The iKIR-associated enhancement of HLA-B*57–mediated protection was wholly dependent on the presence of the KIR ligands. If we counted iKIRs in the absence of ligand, then we found no enhancement of HLA-B*57–mediated protection (fig. S1).

Repeating the analysis of the impact of iKIRs on HLA-B*57 using a different approach (introducing a covariate with different levels for HLA-B*57, HLA-B*57+KIR, and HLA-B*57+KIR+ and performing multivariate regression and Cox survival analysis on the whole unstratified cohort) yielded identical conclusions: The iKIRs KIR2DL1, KIR2DL2, and KIR2DL3 with genes encoding their HLA ligands, a high inhibitory score, and a high count of functional iKIRs all enhanced the protective effect of HLA-B*57 on both outcome measures—early viral load and time to low CD4 count (table S3).

To check that the iKIR-associated enhancement we observed was not driven by leukocyte immunoglobulin-like receptor subfamily B member 2 (LILRB2) binding (26) (which was significantly protective in our cohort; table S4) or by functional KIR3DL1 [which was not protective in our cohort (table S1) but has been reported elsewhere (25)], we included these terms as covariates in our model. In both cases, the conclusions were unchanged (table S5), indicating that functional iKIR enhancement was not attributable to these potentially confounding factors. The conclusions were also unchanged when we included KIR3DL2 with genes for its classical HLA ligands (HLA-A3, HLA-A11, and HLA-B27) or genes for its classical and nonclassical HLA ligands [HLA-A3, HLA-A11, HLA-B27, and HLA-F (27)] in the count of functional iKIRs and the inhibitory score.

Which iKIR is responsible for the enhancement of HLA-B*57–mediated protection?

The KIR genes are in strong linkage disequilibrium, so the enhancements of HLA-B*57–mediated protection by functional KIR2DL1, functional KIR2DL2, and functional KIR2DL3 are not independent. Establishing which iKIR gene(s) was associated with the enhancement was difficult because the cohort size limited the power of further substratification. Across the cohort, functional KIR2DL1 and functional KIR2DL2 were both negatively associated with functional KIR2DL3, whereas functional KIR2DL1 and functional KIR2DL2 were unassociated. Given that all three functional iKIRs enhanced HLA-B*57 and none of them were positively associated suggests that more than one functional iKIR was enhancing HLA-B*57–mediated protection (the requirement for ligand makes enhancement via “absence of a KIR” unlikely). Furthermore, among HLA-B*57+ individuals, both the count of functional iKIRs and the inhibitory score were better predictors of early viral load than any of the individual functional iKIRs, again suggesting that more than one iKIR was enhancing the protective effect of HLA-B*57. Henceforth, because of the difficulty in determining which iKIR(s) was responsible for the enhancement of HLA-B*57 together with the evidence that more than one iKIR was implicated, we focus on the count of iKIRs and the inhibitory score rather than individual iKIR.

Maintenance of the protective effect of HLA-B*57 depends on iKIRs

Next, we investigated longitudinal viral load in the IAVI cohort (a total of 5197 person-visits between 0 and 2000 days after infection). The cohort was stratified in three ways: (i) functional iKIR negative and iKIR positive, (ii) low and high functional iKIR count, or (iii) low and high inhibitory score. In each case, the local nonparametric regression lines were calculated (Fig. 2A and fig. S2). A consistent picture emerged for each of the three functional iKIRs and for both the iKIR count and inhibitory score. In the presence of a strong inhibitory signal (functional iKIR positive or a high inhibitory score/count), the protective effect of HLA-B*57 on viral load was sustained over time. Conversely, in the absence of a strong inhibitory signal (functional iKIR negative or a low inhibitory score/count), the protective effect of HLA-B*57 on viral load was weak. In the absence of a strong inhibitory signal (Fig. 2A, left panel), HLA-B*57 appeared to protect at very early time points after infection, but this protective effect was eroded with time. Bootstrap estimation of the correlation between HLA-B*57–mediated protection and time confirmed this observation: The protective effect of HLA-B*57 decreased significantly more rapidly in people with a low inhibitory score (Fig. 2B). These findings are consistent with our hypothesis that a strong inhibitory signal enhances T cell survival.

Fig. 2 The HLA-B*57 protective effect on viral load is maintained in the presence of a strong iKIR signal but is lost in its absence.

(A) The IAVI cohort was stratified into individuals with a low inhibitory score (≤2.5, left) and individuals with a high inhibitory score (>2.5, right). In each stratum, the local nonparametric regression lines (loess) were calculated from the longitudinal viral load measurements (5197 person-visits). Pink circles denote viral load measurements in HLA-B*57 individuals, summarized with the pink loess line. Blue circles denote viral load measurements in HLA-B*57+ individuals, summarized with the blue loess line. Dashed lines are the 95% confidence intervals (CI) around the loess lines. The vertical lines denote the time at which the number of individuals in either arm fell below 10 (loess lines to the right of this time point, i.e., within the gray shaded area, are unreliable and should be ignored). Viral load is residualized for gender; plots of raw viral load are indistinguishable. (B) Bootstrap estimation of the correlation between the HLA-B*57 protective effect and time in the low (≤2.5) and high (>2.5) inhibitory score strata. Analysis was restricted to time points <500 days to ensure large cohort size. The protective effect of HLA-B*57 decreased significantly more rapidly in people with a low inhibitory score (P = 0.00002, Wilcoxon-Mann-Whitney two-tailed test).

What is the impact of iKIRs on the detrimental effect of HLA-B*35Px?

In the context of HTLV-1 infection, KIR2DL2 has also been shown to enhance the detrimental effects of the susceptibility allele HLA-B*54 (21). We therefore investigated the impact of iKIRs on the well-established detrimental association between HLA-B*35Px and the outcome of HIV-1 infection (28). In the IAVI cohort, HLA-B*35Px was associated with an increase in set-point viral load (Coeff = +0.35, P = 0.0007), but this effect was not consistently enhanced by iKIRs (table S6). However, when we studied longitudinal viral dynamics, it could again be seen that the detrimental impact of HLA-B*35Px on viral load was progressively eroded in individuals with a low inhibitory score and maintained in individuals with a high inhibitory score (fig. S3). This observation explains why iKIR enhancement of the HLA-B*35Px effect was not visible in the original outcome measure, early set-point viral load, which focuses on the first 3 to 9 months after infection. Furthermore, bootstrap estimates of the correlation between time and the detrimental effect of HLA-B*35Px showed that the detrimental effect weakened in individuals with a low inhibitory score but was maintained in individuals with a high inhibitory score (P = 6 × 10−7, P = 5 × 10−9, P = 9 × 10−18 for inhibitory score cutoffs of 2.0, 2.5, and 2.75, respectively; fig. S3). HLA-B*35Px did not influence the time to low CD4 count in the IAVI cohort (HR = 1.04, P = 0.8), so we did not investigate the interplay between iKIRs and HLA-B*35Px for this outcome measure.

Results are replicated in independent cohorts

We considered two independent replication cohorts: IAVI-partners and US.

IAVI-partners

IAVI-partners is a cohort of 315 HIV-1+ individuals from sub-Saharan Africa (the partners of the incident cases in the main IAVI cohort). In this cohort, HLA-B*57 showed a trend to reduce median viral load (Coeff = −0.33, P = 0.06), and HLA-B*35Px was nonsignificant. Pooling IAVI and IAVI-partners provided a much larger cohort (n = 776). In this pooled cohort, HLA-B*57 was strongly protective (Coeff = −0.43, P = 0.00003), and HLA-B*35Px showed a trend to be detrimental (Coeff = +0.16, P = 0.06). In both the unpooled and the pooled cohorts, the effects of HLA-B*57 and HLA-B*35Px were enhanced in people with a high inhibitory score (Table 2). HLA-B*35Px, which only showed a trend for susceptibility in the pooled cohort and nothing in the unpooled cohort, was highly significantly detrimental in people with a high inhibitory score.

Table 2 Functional iKIRs enhance the protective effect of HLA-B*57 and the detrimental effect of HLA-B*35Px in two independent replication cohorts.

Two independent replication cohorts were considered (IAVI-partners and US). In a separate analysis, IAVI and IAVI-partners were pooled to increase cohort size. Cohorts were stratified into individuals with a low or high inhibitory score (two cutoffs considered) and into individuals with a low or high count of functional iKIR (only one cutoff possible). In each stratum, the effects of HLA-B*57 and HLA-B*35Px were analyzed. HLA-B*35Px was not analyzed in IAVI-partners as the HLA gene alone had no detectable impact (Coeff = +0.04, P = 0.8). Definitions as in Table 1. In all cohorts, all variables that had a significant impact on outcome were included in the model as covariates: (A) linear regression, covariate: gender; (B and C) linear regression, covariates: gender and cohort; (D and F) Cox regression, covariates: age and subcohort; (E and G) Cox regression, covariates: KIR3DL1+Bw4-80I+, age, ethnicity, and subcohort. In all cohorts where functional KIR3DS1 (KIR3DS1+Bw4-80I+) was protective, individuals with functional KIR3DS1 were removed before analysis to give a clean background [cohorts reported in (B), (C), (E), and (G)]. Note that the only balanced stratification provided by the “count of functional iKIRs” (namely, ≤2 versus >2) is the same as the stratification provided by the inhibitory score with a cutoff of 2.0. The inhibitory score has more diversity than the count, and so, typically, more stratifications by inhibitory score than by count are possible. All cutoffs that gave a sufficiently balanced analysis were considered. P < 0.001 ***; P < 0.01 **; P < 0.05 *; P < 0.1 ·; P values are two-tailed.

View this table:

US

The US is a cohort of 548 HIV-1+ individuals with known seroconversion date: 57.8% of individuals were white, 38.7% were African American, and 3.5% were Hispanic/“Other”; we therefore stratified by ethnicity. Among whites, but not African Americans, both functional KIR3DS1 (KIR3DS1+Bw4-80I+) and functional KIR3DL1 (KIR3DL1+Bw4+) were protective. We therefore excluded functional KIR3DS1+ individuals and included functional KIR3DL1 as a covariate in the white stratum (exclusion of functional KIR3DL1+ individuals was not possible because of their high frequency). Although neither of these steps were necessary in the African American stratum, we repeated the analysis with both steps to check that the weak, insignificant effects of functional KIR3DS1 or functional KIR3DL1 were not responsible for our results.

Among whites, 15 HLA-B*57+ and 16 HLA-B*35Px+ individuals remained after exclusion of functional KIR3DS1+ individuals. Stratification of these groups left at most eight people in each arm, so analysis was not possible. In contrast, among African Americans, there were 31 HLA-B*57+ and 45 HLA-B*35Px+ individuals. Analysis showed that both the protective effect of HLA-B*57 and the detrimental effect of HLA-B*35Px on the time to CD4 count <200 cells/mm3 were consistently enhanced by functional iKIRs (Table 2). On pooling white, Hispanic/Other, and African Americans (with ethnicity as a covariate), the protective effect of HLA-B*57 and the detrimental effect of HLA-B*35Px were still enhanced, although the enhancement of HLA-B*57 appeared weaker (Table 2), suggesting that the iKIR effect on HLA-B*57 was largely driven by the African American population. These findings were replicated when high-expressing KIR3DL1 alleles were included as a covariate (table S7).

Results across the cohorts, for the different alleles and for longitudinal viral load were consistent. We conclude that in HIV-1 infection in three independent cohorts, the effects of both protective and detrimental HLA alleles on viral load and disease progression are stronger in the presence of genes coding for iKIRs with their ligands.

iKIRs enhance HLA associations with asymptomatic status in HTLV-1 infection

We previously reported that KIR2DL2 enhances HLA class I associations in HTLV-1 infection but found little evidence for enhancement by other iKIRs (21). In contrast, the findings reported here show that all iKIRs amenable to analysis enhanced HLA-B*57 and HLA-B*35Px disease associations in HIV-1 infection. We therefore reexamined the HTLV-1 data.

We confirmed that, in the HTLV-1–infected Kagoshima cohort, functional KIR2DL2 strongly enhanced the protective effect of HLA-C*08 and the detrimental effect of HLA-B*54 on the odds of the inflammatory disease HTLV-1–associated myelopathy/tropical spastic paraparesis (HAM/TSP). Previously, we reported that, unexpectedly, functional KIR2DL2 did not enhance the protective effect of HLA-A*02 (21). However, we recently found that only HLA-A*02:07, and not the more frequent HLA-A*02:01 and HLA-A*02:06, associates with protection in the Kagoshima cohort (table S8). We therefore repeated the analysis at the level of four-digit HLA resolution. Although individual numbers were low, we found evidence that KIR2DL2 did enhance the protective effect of HLA-A*02:07 (halving the risk of developing HAM/TSP). Investigating the other iKIRs, we found that the low frequency of HLA-C2 precluded a meaningful analysis of functional KIR2DL1 (too few functional KIR2DL1+ individuals) and functional KIR2DL3 (too few functional KIR2DL3 individuals), explaining why we had not previously seen an effect for these other iKIRs. However, for functional KIR3DL1, both the positive and negative strata were of sufficient size to permit analysis, and there we found good evidence that functional KIR3DL1 enhanced both the detrimental effect of HLA-B*54 and the protective effects of HLA-A*02:07 and HLA-C*08. Furthermore, the effects of HLA-B*54, HLA-A*02:07, and HLA-C*08 were, in each case, strongest in individuals with a high inhibitory score and were weak and nonsignificant in individuals with a low inhibitory score (Table 3).

Table 3 Functional iKIRs enhance HLA class I associations in HTLV-1 infection.

The Kagoshima cohort was stratified into individuals with and without the functional iKIR of interest. In each stratum, the effect of (A) HLA-B*54, (B) HLA-A*02:07, and (C) HLA-C*08 on HAM/TSP status was analyzed using multivariate logistic regression (with covariates age and gender). The odds ratio (OR), P values, and cohort sizes are reported. An OR < 1 represents a protective effect (reduced odds of HAM/TSP). The detrimental effect of HLA-B*54 and the protective effects of HLA-A*02:07 and HLA-C*08 on HAM/TSP status were enhanced in people with strong functional KIR2DL2, functional KIR3DL1, and a high inhibitory score. There was an insufficient number of people with HLA-C2 to analyze functional KIR2DL1, weak functional KIR2DL2, and functional KIR2DL3. Similar results were obtained if we use factors instead of stratification. The ligands of KIR3DL1 are contentious, with one paper reporting that KIR3DL1 binds HLA-A*23, HLA-A*24, and HLA-A*32, as well as HLA-Bw4 molecules (60). If we amended our definition of functional KIR3DL1 to include these HLA-A alleles, then the conclusions were unchanged with the exception of HLA-C*08, where our conclusion was strengthened (magnitude of enhancement of protective effect of HLA-C*08 in the functional KIR3DL1-positive stratum is greater). All cutoffs that gave a sufficiently balanced analysis were considered. The definitions of “low” and “high” iKIR counts and scores were reduced compared with other cohorts in this study to compensate for lower overall iKIR counts and scores in the Japanese cohort. P < 0.001 ***; P < 0.01 **; P < 0.05 *; P < 0.1 ·; P values are two-tailed.

View this table:

The high inhibitory score stratum was heavily enriched for people with strong functional KIR2DL2 [P < 10−14 (cutoff = 1.5), P < 10−15 (cutoff = 1.75), χ2 test], so the KIR2DL2 enhancement and high iKIR score enhancement of HLA-B*54, HLA-A*02:07, and HLA-C*08 are not independent. However, functional KIR3DL1 is not a surrogate for functional KIR2DL2 (P = 0.86, χ2 test), so the enhancement due to KIR2DL2 is independent of the enhancement due to KIR3DL1. We conclude that, although most iKIRs cannot be analyzed (due to the low frequency of HLA-C2 alleles in Japan), there is evidence that, independent of KIR2DL2, another iKIR, KIR3DL1 and a high inhibitory score enhance HLA class I disease associations in HTLV-1 infection.

iKIRs enhance HLA associations with spontaneous clearance of HCV

We previously reported that functional KIR2DL2 enhances the protective effect of HLA-B*57 on spontaneous clearance of HCV. Again, the earlier analysis found no evidence that other iKIRs showed similar behavior (21). Reanalysis of this cohort confirmed the KIR2DL2 enhancement. It was not possible to analyze any other individual iKIR [insufficient numbers of HLA-B57+ functional KIR2DL1 and HLA-B57+ functional KIR3DL1+ individuals, and functional KIR2DL3 alone is protective (29); table S2]. We also found a strong enhancement of the HLA-B*57 effect in people with a high inhibitory score (table S9). However, the high inhibitory score stratum was heavily enriched for people with strong functional KIR2DL2 for all three cutoffs considered (P < 2.10−16 in each case, χ2 test), and so the functional KIR2DL2 enhancement and strong iKIR enhancement of HLA-B*57 are not independent. We conclude that functional KIR2DL2 and a high inhibitory score enhance the protective effect of HLA-B*57 in HCV infection, but it is not possible to test whether other iKIRs show the same behavior.

Primary HLA class I associations are unlikely to be attributable to NK cells

In theory, an HLA class I disease association could be attributable to the role of these molecules as recognition targets for either CD8+ T cells (via somatically rearranged T cell antigen receptors) or NK cells (via germline-encoded KIRs). If a given HLA association is attributable to NK cells rather than CD8+ T cells, then other HLA class I molecules with similar KIR-binding properties should be similarly protective/detrimental (in the context of the relevant KIRs). This pretext forms the basis of classical KIR-HLA analyses (3033). We investigated this possibility for each of the HLA class I alleles included in this study (HLA-B*57 and HLA-B*35Px in the HIV-1–infected (Functional KIR3DS1-) IAVI cohort; HLA-A*02:07, HLA-C*08, and HLA-B*54 in the HTLV-1–infected Kagoshima cohort; and HLA-B*57 in the HCV-infected cohort). The results are detailed in the Supplementary Materials and table S10. We found no evidence to suggest that any of the HLA class I associations could be explained by a ligand preference for certain KIRs. This is consistent with existing evidence that these HLA class I associations are attributable to CD8+ T cells (3439). Furthermore, simply considering iKIR-ligand pairs (without also considering the protective or detrimental HLA class I alleles) finds no significant protective or detrimental effect of any individual iKIR-ligand or of the count of functional iKIR or of the inhibitory score in HIV-1, HTLV-1, or HCV (table S2).

iKIR ligation enhances CD8+ T cell survival in vitro

Studies in mice and in vitro have shown that inhibitory NK receptors on T cells reduce activation-induced cell death (1012). An iKIR-mediated enhancement of CD8+ T cell survival may therefore explain our immunogenetic findings. However, whereas our immunogenetic findings show that the presence of the HLA ligand is crucial (fig. S1), the role of ligand in the in vitro survival studies is less clear. Two studies reported that iKIRs are associated with CD8+ T cell survival independent of ligand (13, 40), whereas a third study indicated that KIR ligation is required for the increased survival (15). To clarify and extend this work, we investigated the impact of blocking iKIR ligation on CD8+ T cell survival in vitro. Three iKIR-expressing CD8+ T cell lines were established: two expressing KIR2DL3 and one expressing KIR3DL1 (fig. S4). The lines were cocultured with the HLA class I–deficient B cell line 721:221 (henceforth 221) that was untransfected, transfected to express a noncognate HLA molecule (HLA-B57:01 or HLA-C04:01 in the case of KIR2DL3 and HLA-C03:04 or HLA-C04:01 in the case of KIR3DL1), or transfected to express a cognate HLA molecule (HLA-C03:04 in the case of KIR2DL3 and HLA-B57:01 in the case of KIR3DL1). Activation was induced with staphylococcal enterotoxin E (SEE). T cell survival was quantified in the presence of antibodies that block KIR (GL183 and Dx9) (41), antibodies that block HLA class I (Dx17) (42), or an isotype control. The gating strategy is illustrated in fig. S5. We found that blocking the iKIR-HLA interaction either by blocking iKIR or by blocking HLA class I significantly decreased the count of live T cells compared with isotype control (median decrease, 40 and 48%, respectively; Fig. 3, A and B). Survival was similarly impaired in the absence of ligand (221 transfected with noncognate HLA or untransfected; median decrease, 57 and 53%, respectively; Fig. 3, A and B).

Fig. 3 Impact of iKIR-HLA interaction on CD8+ T cell survival in vitro and iKIR expression ex vivo.

(A) CD8+ T cell lines (plots from left: ED, KIR2DL3+; VTE1, KIR2DL3+; and SK1, KIR3DL1+) were cocultured with superantigen without adding ligand for KIR [i.e., with 221 cells or with 221 cells transfected to express irrelevant, noncognate HLA (open bars)], with ligand [i.e., 221 transfected to express cognate HLA with or without isotype control antibody (black bars)], and with KIR-HLA blocking [i.e., 221 transfected to express cognate HLA with addition of antibody to block KIR (GL183 and Dx9) or to block HLA (Dx17) (red bars)]. The count of live CD8+ T cells after 5 days is shown (mean ± SE). ED, five experiments each consisting of three replicates; VTE1, one experiment, four replicates; SK1, one experiment, two replicates. It is consistently seen that there are greater numbers of viable T cells with isotype control than when the KIR-HLA interaction is blocked or when cognate ligand is absent. IgG1, immunoglobulin G1. (B) Meta-analysis of survival of the three KIR-expressing CD8+ T cell lines. Block HLA: Coculture with 221 cells transfected to express cognate HLA with isotype control antibody compared with coculture with 221 cells transfected to express cognate HLA with anti-HLA antibody. Median difference, 48%; P = 0.001. Block KIR: Coculture with 221 cells transfected to express cognate HLA with isotype control antibody compared with coculture with 221 cells transfected to express cognate HLA with anti-KIR antibody. Median difference, 40%; P = 0.001. Irrelevant HLA: Coculture with 221 cells transfected to express cognate HLA compared with coculture with 221 cells transfected to express irrelevant, noncognate HLA. Median difference, 57%; P = 1.5 × 10−5. 221 only: Coculture with 221 cells transfected to express cognate HLA compared with coculture with 221 cells. Median difference, 53%; P = 0.002. Survival of VTE1 plotted on the left-hand y axis (for all conditions); ED and SK1 plotted on the right-hand y axis (for all conditions). We consistently found that the absence of KIR-HLA ligation (by blocking HLA, blocking KIR, adding a nonligating HLA, or not adding HLA) resulted in significantly decreased CD8+ T cell survival. (C) PBMCs from three HLA-B27+ individuals with ankylosing spondylitis were left untreated (open bars), cultured with superantigen (SEB) with or without isotype control antibody (black bars), or cultured with SEB with antibodies that block KIR3DL2 or homodimeric HLA-B27 (red bars). The count of live KIR3DL2+ CD4 T cells after 5 days is shown (mean ± SE). Blocking the iKIR-HLA interaction with specific antibodies decreased the count of live KIR3DL2+ T cells compared with isotype control (block KIR: median difference, 63%; block HLA: median difference, 50%). (D) The KIR2DL3-expressing CD8+ T cell line VTE1 was cultured as in (A). The number of live CD8+ CD107a+ T cells was quantified by flow cytometry at day 5. (E) The fraction of pentamer+ CD8+ T cells expressing iKIRs (KIR2DL1, KIR2DL2/L3, KIR3DL1, or KIR3DL2) directly ex vivo are plotted for HLA-A2+ individuals infected with HIV-1 (n = 16) or HTLV-1 (n = 9). Horizontal lines are at the mean.

Next, we extended this work to primary CD8+ T cells. Peripheral blood mononuclear cells (PBMCs) from three HLA-B*27+ individuals with ankylosing spondylitis, who have a high proportion of PBMCs expressing KIR3DL2 (6), were stimulated with staphylococcal enterotoxin B (SEB). KIR3DL2 binds HLA-B27 homodimers (6). Survival of KIR3DL2+ CD8+ T cells was quantified in the presence of antibodies that block KIR3DL2 (Dx31), antibodies that block HLA-B27 homodimers [HD6 (43)], or an isotype control (fig. S6). We found that blocking the iKIR-HLA interaction reduced the count of live KIR3DL2+ CD8+ T cells compared with isotype control (block KIR: median difference, 63%; block HLA: median difference, 50%; Fig. 3C).

Last, iKIR ligation has been shown to inhibit CD8+ T cell killing under some conditions (44). To assess this possibility, which could mitigate the biological impact of enhanced survival, we quantified the impact of iKIR ligation on the mobilization of CD107a, a surrogate marker of degranulation. Greater numbers of live CD107a+ cells were observed in the presence of a functional KIR-HLA interaction after stimulation of the KIR2DL3-expressing CD8+ T cell line VTE1 with SEE (Fig. 3D). In summary, KIR ligation enhances the number of viable cells and the number of functional cells both for CD8+ T cell lines and for primary T cells in vitro.

Virus-specific CD8+ T cells express iKIR ex vivo

To probe the relevance of these findings, we quantified the fraction of virus-specific CD8+ T cells that expressed iKIRs (KIR2DL1, KIR2DL2/L3, KIR3DL1, and KIR3DL2) in PBMCs from HIV-1–positive and HTLV-1–positive donors. We found considerable proportions of iKIR-expressing pentamer+ CD8+ T cells in both groups (HIV-1: mean, 27%; HTLV-1: mean, 15%; Fig. 3E and figs. S7 and S8). On analyzing the differentiation stage of virus-specific KIR-expressing CD8+ T cells, we found, consistent with previous work on bulk CD8+ T cells (9), that about 40% of iKIR+ pentamer+ cells lie within the CD45RA+CD28 T effector memory revertant gate; the remainder were distributed approximately equally between CD45RACD28+ central memory T cells, CD45RACD28 effector memory T cells, and CD45RA+CD28+ stem cell memory/naïve T cells (fig. S9).

We did not investigate iKIR expression on pentamer+ CD8+ T cells from HCV-seropositive individuals because the frequency of HCV-specific CD8+ T cells in the blood is known to be extremely low (45, 46), due in part to T cell homing to the liver (47). Instead, we investigated the relationship between functional iKIR expression and cell death ex vivo using bulk CD8+ and CD4+ T cells from healthy donors and HCV-seropositive individuals. We found that iKIR+ CD8+ and iKIR+ CD4+ T cells were both less likely to stain with annexin V in individuals where the cognate HLA ligand was present (Supplementary Results and fig. S10).

Mathematical modeling can explain the enhancement of both protective and detrimental HLA associations

The hypothesis that iKIRs enhance HLA class I disease associations by enhancing T cell survival sounds reasonable, but closer inspection reveals a number of difficulties. First, iKIRs would be expected to increase the survival of all CD8+ T cells, not just those restricted by protective HLA alleles. In an immunogenetic study, a protective HLA effect reflects the difference in viral control between average and protective HLA alleles, so it is not obvious how a nonspecific increase in T cell survival would enhance protective HLA associations. Second, it is difficult to envisage how the same mechanism could also have the opposite effect and enhance detrimental HLA associations. Last, there is evidence that iKIRs can suppress the lytic activity of CD8+ T cells under some conditions (14) (although this was not replicated in our work with CD107a). This raises another question: Is it reasonable to suggest that in people with a high iKIR score, and potentially a higher level of functional inhibition, CD8+ T cell protection is actually enhanced? To address these questions, we constructed a mechanistic mathematical model of host-virus dynamics for HIV-1. In this model, CD8+ T cells restricted by protective HLA class I molecules were assumed to lyse targets more efficiently than average, whereas CD8+ T cells restricted by detrimental HLA class I molecules were assumed to lyse targets less efficiently than average. In the first instance, we assumed that the only effect of iKIRs was to enhance T cell survival; in later analyses, we assumed that iKIRs also impaired CD8+ T cell lysis. We generated 100,000 random parameter combinations by sampling from the physiological range for each parameter (table S11) and solved the system numerically (Fig. 4A). For each parameter combination, we recorded whether the protective effect (decrease in set-point viral load associated with possession of a protective HLA class I molecule compared with an average HLA class I molecule) was enhanced in people with a high number of functional iKIRs. We found that most of the runs (97%) showed iKIR-mediated enhancement of protective HLA class I associations (Fig. 4B). Next, we repeated the analysis for detrimental HLA class I molecules. This showed that detrimental HLA class I–associated effects were also enhanced across a wide range of parameter space (Fig. 4C). Comparison of the region of parameter space where protective HLA class I associations were enhanced and where detrimental HLA class I associations were enhanced showed that there was broad overlap, that is, there exist parameter regions where both protective and detrimental HLA class I associations were simultaneously enhanced. In addition, when iKIRs were assumed to inhibit T cell lysis, enhancement of protective and detrimental HLA class I effects was decreased, but most of the runs still showed enhancement (Fig. 4, D and E). Further analysis showed that these results were robust to both structural and parametric changes in the model (Supplementary Methods).

Fig. 4 Simulations of iKIR enhancement of HLA class I disease associations in HIV-1.

(A) Dynamics of HIV-1–specific CD8+ T cells, infected and uninfected CD4+ T cells, and viral load during HIV-1 infection for a representative hypothetical in silico individual with average (dashed black line) or protective (green solid line) HLA class I alleles. (B and C) Fraction of runs that result in an iKIR-mediated enhancement of the protective (B) or detrimental (C) HLA effect as a function of virus transmissibility (β) and CD8+ T cell proliferation rate (s2) (left), as a function of CD8+ T cell lysis rate (d2) and proliferation rate (s2) (middle), and as a function of CD8+ T cell death rate (μ4) and proliferation rate (s2) (right). (D and E) Distribution (D) and frequency (E) of runs with enhancement when we consider an impairment in the CD8+ T cell lysis rate due to iKIRs (percentage reduction from left to right: 0, 5, 15, and 30%). (F) Viral load as a function of CD8+ T cell lifespan (1/μ4) for a hypothetical person with detrimental (magenta dotted), average (black dashed), or protective (green solid) HLA class I alleles. It can be seen that, as the survival of CD8+ T cells increases, both the effect of protective HLA (difference in viral load between average and protective HLA) and the effect of detrimental HLA (difference in viral load between average and detrimental HLA) increases. Compare, for example, a hypothetical person with a high number of iKIR-HLA pairs (blue bar marked KIR hi) and a hypothetical person with a low number of iKIR-HLA pairs (blue bar marked KIR low).

To understand how an increase in CD8+ T cell survival can enhance both protective and detrimental HLA class I associations, we simulated viral load as a function of CD8+ T cell lifespan for three representative individuals: one with average HLA class I molecules, one with a protective HLA class I molecule, and one with a detrimental HLA class I molecule (Fig. 4F). On seeing this figure, the explanation immediately becomes obvious. In the limit, when CD8+ T cell lifespan is very short, it makes little difference whether an individual has average, protective, or detrimental alleles; everyone has a similar viral load independent of their HLA genotype. In an immunogenetic study, this would manifest as an absence of association between either protective or detrimental alleles and outcome. As CD8+ T cell lifespan is increased (e.g., by iKIRs), the “quality” of the CD8+ T cell response becomes important, and a large difference in viral load emerges between people with average, protective, and detrimental HLA class I alleles. In an immunogenetic study, this would manifest as large protective and detrimental effects; that is, as CD8+ T cell lifespan increases, both protective and detrimental effects are enhanced. Although this model does not prove the hypothesis, it does show that the hypothesis predicts our findings and provides a plausible and intuitive explanation for a seemingly contradictory observation.

DISCUSSION

In this study, we conducted an immunogenetic analysis of three independent cohorts of HIV-1–infected individuals to investigate the impact of iKIR genes on well-established HLA class I disease associations. We found that genes for iKIRs consistently enhanced the protective effect of HLA-B*57 and the detrimental effect of HLA-B*35Px on viral load and progression to low CD4 count. The result was critically dependent on the presence of the KIR ligand. A reanalysis of two other persistent viral infections, HCV and HTLV-1, found consistent results: all four HLA class I associations studied were enhanced in people with a high inhibitory score. We previously reported this effect as an enhancement associated with one particular iKIR (KIR2DL2) (21). In HCV infection, enhanced HLA class I associations were detected for individuals with KIR2DL2 and individuals with a high inhibitory score, but it was impossible to determine which effect was the driver. However, in HTLV-1 infection, new data made it possible to show that, independent of KIR2DL2, another iKIR (functional KIR3DL1) enhanced both protective and detrimental HLA class I disease associations. The probability of seeing this result by chance is P = 10−16; furthermore, the HIV-1 findings were replicated in two independent cohorts. Collectively, these data suggest that all iKIRs behave in a similar manner and that KIR2DL2 is not functionally unique [although its prevalence and distribution make its effects easier to detect in some cohorts—a hypothesis we suggested previously (21)]. The iKIRs are closely related, being the product of gene duplication (48), so it is gratifying that we find a unifying picture in which all iKIRs enhance HLA class I associations in a similar fashion.

We investigated the underlying mechanism using three approaches: (i) analysis of the duration of HLA class I–mediated effects in the HIV-1–infected cohort, (ii) in vitro survival assays combined with ex vivo KIR expression analysis, and (iii) mathematical modeling. In the HIV-1 cohort, the phenomenon of iKIR-mediated enhancement appears to be related to the longevity of the CD8+ T cell response, because both the HLA-B*57 protective effect and the HLA-B*35Px detrimental effect were eroded over time in individuals with a low inhibitory score but were maintained in individuals with a high inhibitory score; the difference in rates between these two groups was consistently highly significant. In vitro, we found that blocking the KIR-HLA interaction significantly reduced cultured and primary T cell survival and functional CD107a+ T cell survival after activation with superantigen. In addition, the presence of the KIR ligand was associated with a decrease in annexin V binding on CD4+ and CD8+ T cells expressing iKIR directly ex vivo. We also found substantial levels of iKIR expression on virus-specific CD8+ T cells among PBMCs isolated from HIV-1 and HTLV-1–seropositive individuals. Last, in a mathematical model of host-virus dynamics in HIV-1 infection, we predicted that, if iKIRs increase CD8+ T cell lifespan (either directly or indirectly), then we would see an enhancement of HLA class I associations, consistent with the immunogenetic data. Further, the model provides an explanation for how increasing CD8+ T cell lifespan can simultaneously have two seemingly contradictory effects: enhancing both protective and detrimental associations. The modeling shows that if CD8+ T cells have longer lifespans (e.g., due to the presence of iKIRs), then the quality of the CD8+ T cell response becomes more important, and so differences between protective, average, and detrimental HLA alleles become more apparent; that is, both protective and detrimental effects are enhanced.

Our conclusions are that iKIRs enhance HLA class I associations and that this can be explained by an increase in CD8+ T cell survival in the presence of functional iKIRs. A number of ways in which iKIRs can increase CD8+ T cell survival have been described. These can be broadly divided into “direct” (iKIRs on T cells affect T cell lifespan) and “indirect” (iKIRs on other cell populations, e.g., NK cells, affect T cell lifespan); these pathways are summarized in fig. S11. Our finding that, in vitro, iKIR expression on CD8+ T cells is directly associated with increased survival does not preclude a role for iKIR on NK cells in indirectly enhancing CD8+ T cell survival. Both effects may be occurring and increasing CD8+ T cell survival, and both are consistent with our immunogenetic analysis, the longitudinal study, and the mathematical modeling.

The iKIR-ligand binding groups that we use in our definition of “functional iKIR” are simplistic. Incomplete knowledge of how different alleles and different peptides affect binding and signaling precludes a more sophisticated definition. Nevertheless, these simple groupings have proved very powerful in other studies (25, 29, 31, 33, 49, 50). With the definition of inhibitory score used, we saw clear and reproducible results. This suggests that the inhibitory score is a meaningful metric (the odds of seeing all the results by chance is <10−23). The score is used only to split the cohorts in half, so second-order changes to the calculation of the score will not necessarily change the results; we found that our conclusions were robust to quite dramatic changes to the score (e.g., whether KIR3DL2 with classical class I ligands was included). This is because a person’s score can change by a large amount, and they will remain in the same strata. Whether we simply count KIR-ligand pairs or use the inhibitory score has little impact on the conclusions. Another limitation of this work is that the survival assays were only conducted in vitro. The strong associations between possession of combinations of iKIR/HLA genes and clinical outcomes such as time to low CD4 count suggest that these effects are important in vivo, but direct in vivo assays were not performed. Murine in vivo assays provide limited information because Ly49 receptors differ considerably from KIR (this includes profound differences in structure and in the pattern of expression on T cells). Human in vivo studies are limited by ethical considerations but would be an important next step in this work.

HCV, HIV-1, and HTLV-1 are all characterized by continuous high-level viral replication and chronic immune activation, which has been associated with CD8+ T cell exhaustion and activation-induced T cell death (5154). Chronic immune cell activation is also associated with increased KIR expression on T cells (59, 44, 55). iKIR “rescuing” of activated CD8+ T cells may be particularly important in this setting. Most KIR-HLA genetic associations are not replicated across viral infections, for example, KIR3DS1 with HLA-Bw4 is protective in the context of HIV-1 but not HTLV-1 (56); this is difficult to understand unless a very strong peptide dependence is postulated. In contrast, we report that all iKIRs enhance HLA class I associations in all three persistent virus infections we have studied (HIV-1, HCV, and HTLV-1), which suggests a universality that is often missing from KIR studies.

The iKIR-HLA receptor-ligand system also has similarities to the programmed death 1 (PD-1)–programmed death ligand 1/2 system. Both iKIR and PD-1 are inhibitory receptors that interfere with proximal T cell receptor signaling and are up-regulated in the context of chronic viral infection and on tumor-infiltrating lymphocytes. PD-1 and iKIR are both expressed on late-stage differentiated memory T cells, typically in a mutually exclusive manner (57). Blockade or activation of iKIR would thus affect a different population of T cells compared with PD-1 activation/blockade. These observations, combined with the demonstration here that the iKIR-HLA receptor-ligand system is clinically significant, suggest novel and synergistic targets for therapeutic immune checkpoint blockade.

MATERIALS AND METHODS

Study design

The aim of this study was to investigate whether iKIRs have a clinically significant impact on the human CD8+ T cell response. We studied the impact of iKIR genotype on six well-documented HLA class I disease associations in three viral infections (Table 4) in five independent cohorts. The work was extended and interpreted by mechanistic mathematical modeling, in vitro T cell survival assays, and ex vivo analysis of iKIR expression.

Table 4 HLA class I disease associations studied.

View this table:

Ethics statement

The immunogenetic study was approved by the National Health Service (NHS) Research Ethics Committee (13/WS/0064) and the Imperial College Research Ethics Committee (ICREC_11_1_2). Informed consent was obtained at the study sites from all individuals. For the cellular work, written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. Research was conducted under the governance of the Imperial College Healthcare NHS Trust Tissue Bank, approved by the UK National Research Ethics Service (09/H0606/106 and 15/SC/0089).

Immunogenetic cohorts

IAVI is a prospective cohort of treatment-naïve HIV-1 seroconverters from sub-Saharan Africa, identified and followed under protocol C of the IAVI (22, 58). Individuals who had both KIR3DS1 and the gene encoding its putative ligand, HLA-Bw4-80I, were excluded from the analysis to eliminate the potentially confounding effects of functional KIR3DS1. The cohort comprised individuals with time to low CD4 count information (n = 491) and individuals with early viral load information (n = 461). IAVI-partners is an HIV-1 seroprevalent cohort (n = 315). US is a cohort of HIV-1 seroconverters (n = 548). Kagoshima is a cohort of HTLV-1–infected individuals (n = 392). HCV comprises four subcohorts of HCV-seropositive individuals (n = 782).

Regression

The impact of genotype on outcome was analyzed using multivariate linear, logistic, and Cox regression in R v3.0.1 (59). Potentially confounding covariates were identified and included in the analysis (listed in Supplementary Methods). All reported P values are two tailed.

In vitro T cell survival

CD8+ T cell lines

Three iKIR-expressing CD8+ T cell lines were established from the peripheral blood of healthy volunteers (fig. S4). The lines were cocultured with the HLA class I–deficient B cell line 221, either untransfected; transfected to express a noncognate HLA molecule (HLA-B57:01 or HLA-C04:01 in the case of KIR2DL3, and HLA-C03:04 or HLA-C04:01 in the case of KIR3DL1); or transfected to express a cognate HLA molecule (HLA-C03:04 in the case of KIR2DL3, and HLA-B57:01 in the case of KIR3DL1). Cells were activated with superantigen SEE. The count of live T cells at day 5 was enumerated using CountBright Absolute Counting Beads (Invitrogen) in the presence of antibodies to block KIR (GL183 and Dx9) (41), antibodies to block HLA class I (Dx17) (42), or an isotype control. The gating strategy is illustrated in fig. S5.

Primary T cells

PBMCs from three HLA-B*27+ individuals with ankylosing spondylitis were stimulated with SEB. Survival of KIR3DL2+ CD4 T cells was quantified in the presence of isotype controls or antibodies that block KIR3DL2 (Dx31) or homodimeric HLA-B27 (HD6) (43).

Pentamer staining

Cryopreserved PBMC samples from HLA-A*02+ HIV-1+ (n = 16) and HLA-A*02+ HTLV-1+ (n = 9) individuals were thawed and stained sequentially with a fixable amine dye to exclude dead cells, fluorochrome-labeled peptide–HLA class I pentamers (ILKEPVHGV-A02:01, SLYNTVATL-A02:01, or LTFGWCFKL-A02:01 for HIV-1 and LLFGYPVYP-A02:01 for HTLV-1), and directly conjugated monoclonal antibodies specific for CD3, CD8, CD28, CD45RA, KIR2DL1, KIR2DL2/L3, KIR3DL1, and KIR3DL2. The fraction of iKIR+ pentamer+ CD8+ T cells were enumerated within each phenotypic compartment defined by the expression of CD28 and CD45RA.

SUPPLEMENTARY MATERIALS

immunology.sciencemag.org/cgi/content/full/3/29/eaao2892/DC1

List of IAVI Protocol C Investigators

Supplementary Methods

Results

Fig. S1. Impact of ligands on iKIR enhancement of the HLA-B*57 protective effect on early viral load set point.

Fig. S2. Impact of iKIRs on the protective effect of HLA-B*57 on longitudinal viral load.

Fig. S3. Impact of iKIRs on the detrimental effect of HLA-B*35Px on longitudinal viral load.

Fig. S4. KIR expression of the CD8+ T cell lines ED, VTE1, and SK1.

Fig. S5. Gating strategy to measure activated CD8+ T cell survival in vitro.

Fig. S6. Survival of primary CD4 T cells is enhanced by iKIR ligation in vitro.

Fig. S7. iKIR expression on CD8+ T cells.

Fig. S8. iKIR expression on pentamer-positive and pentamer-negative cells.

Fig. S9. Differentiation stage of iKIR+ pentamer+ CD8+ T cells.

Fig. S10. Annexin V binding to iKIR+ CD8+ T cells is decreased in individuals carrying the KIR ligand.

Fig. S11. Schematic and predictions of possible hypotheses.

Table S1. Impact of KIR3DS1 and KIR3DL1 in the full IAVI cohort.

Table S2. Influence of individual functional iKIRs, inhibitory score, and count of functional iKIRs alone.

Table S3. Functional iKIRs enhance the protective effect of HLA-B*57, analysis by factor.

Table S4. Impact of LILRB2–HLA class I binding on early viral load set point in IAVI.

Table S5. Functional iKIRs enhance the protective effect of HLA-B*57 on early set-point viral load, even when LILRB2_B binding score and functional KIR3DL1 are included.

Table S6. Functional iKIRs have no clear impact on the detrimental effect of HLA-B*35Px on viral load set point in the IAVI cohort.

Table S7. Impact of functional iKIRs on the effect of HLA-B*57 and HLA-B*35Px in the US cohort when KIR3DL1_h/y + Bw4-80I is included as a covariate.

Table S8. The protective effect of HLA-A*02 on odds of HAM/TSP in the Kagoshima cohort is driven by HLA-A*02:07.

Table S9. The protective effect of HLA-B*57 on odds of spontaneous clearance of HCV is enhanced with increasing inhibitory scores.

Table S10. Can the HLA class I associations studied be explained by NK cell–mediated effects?

Table S11. Parameter ranges used in the mathematical model of host-pathogen dynamics.

Table S12. Most frequent HLA B-C combinations in the IAVI cohort.

Table S13. Data.

References (6192)

REFERENCES AND NOTES

Acknowledgments: We are grateful to R. Busch, A. Rowan, G. Taylor, D. Peppa, P. Borrow, and the Center for HIV/AIDS Vaccine Immunology AI067854. Funding: B.A. is a Wellcome Trust (WT) Investigator (103865Z/14/Z) and is also funded by the Medical Research Council (MRC) (J007439 and G1001052), the European Union Seventh Framework Programme (317040, QuanTI), and Leukemia and Lymphoma Research (15012). M.S. received financial support from the Brazilian National Council for Scientific Development (206435/2014-2). C.h.R. is funded by the WT Institutional Strategic Support Fund (105609/Z/14/Z). E.C.Y.W. is funded by the WT (090323/Z/09/Z) and the MRC (MR/L018373/L, MR/P001602/1). J.T. and J.A.T. are funded by the MRC, the NIH, the Cambridge Biomedical Research Centre, and the European Research Council under the European Union’s Horizon 2020 Research and Innovation Programme (695551). C.T. is supported by NIH R01 DA13324. D.A.P. is a WT senior investigator (100326Z/12/Z). The UK HCV cohort is funded by the MRC (MR/M019829/1). The Hemophilia Growth and Development Study is supported by the National Institute of Child Health and Human Development (R01-HD-41224). Data presented in this manuscript were collected by the ALIVE Study funded by the National Institute on Drug Abuse (U01-DA-036297, R01-DA-12568, and K24-AI118591). This work was partially funded by IAVI with the support of USAID and other donors; a full list of IAVI donors is available at www.iavi.org. This project has been funded in whole or in part with federal funds from the Frederick National Laboratory for Cancer Research under contract HHSN261200800001E and by the Intramural Research Program of the NIH, Frederick National Laboratory, Center for Cancer Research. The content of this publication is the responsibility of the authors and does not necessarily reflect the views or policies of the Department of Health and Human Services, USAID, or the U.S. Government, nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. Author contributions: L.B. performed the analysis, contributed to interpretation, performed the statistical analysis, and wrote the manuscript. B.D. and S.K. performed the experimental work. M.S. performed the mathematical modeling. C.h.R., J.J., J.A.T., J.T., J.M., A.S., K.T., K.L, K.L.M., J.G., and D.C.M. helped with experimental work. D.A.P. helped with experimental work and edited the manuscript. E.C.Y.W. and J.F. contributed materials. Y.Q. and M.P.M. helped with statistical analysis. I.P.C.I., C.T., J.A., G.K., S.M.D., S.B., S.I.K., and J.J.G. provided cohort data. M.C. helped with analysis, contributed to interpretation, and edited the manuscript. B.A. conceived and directed the project, obtained funding, performed the analysis, contributed to interpretation, performed the statistical analysis, and wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Request for access to this data can be made to the relevant cohort PIs. All other data is provided in the paper or in the Supplementary Materials.
View Abstract

Navigate This Article