Ethics Statement

Written informed consent was acquired from all study participants enrolled in the Sinikithemba cohort, and the study protocol was approved by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal (approval number E028/99). The study of primary NK cell function in nine healthy HIV-1-negative individuals recruited at Massachusetts General Hospital (Boston, MA) and at Brigham and Women’s Hospital (Boston, MA; IGTB cohort) was approved by the Partners Human Research Committee (protocol #2012P002121), and each participant gave written informed consent for participation.

Study Participants

Study participants were recruited from the Sinikithemba cohort in Durban, South Africa, and included 406 individuals chronically infected with HIV-1 clade C. The study participants were antiretroviral-naïve, KIR typed, and HLA typed to four-digit resolution by molecular methods [29]. Viral load and CD4 T cell measurements at study entry were obtained using the Roche Amplicor assay (version 1.5) and TruCount technology, respectively. At baseline the median age of the cohort was 31 y (interquartile range [IQR] 27–36), median viral load was 58,300 RNA copies/ml (IQR 3,100–185,500), and median CD4 T cell absolute count was 340 cells/mm³ (IQR 238–478). The cohort was predominantly female (79% female and 21% male). For studies of primary NK cell function, nine healthy HIV-1-negative individuals were recruited at Massachusetts General Hospital (Boston, MA) and at Brigham and Women’s Hospital (Boston, MA; IGTB cohort). HLA typing was performed by 454 sequencing, and KIR typing by real-time PCR.

Sequencing of HIV-1 Gag Gene in Individuals Infected with HIV-1 Clade C

RNA was extracted from the blood plasma of patients infected with HIV-1 clade C, and the region p24 of the Gag gene was amplified by nested reverse transcription PCR as part of the gag-protease region of the virus. The entire PCR amplicon (including the p24 Gag region) was sequenced using BigDye Terminator Ready Reaction Mix V3 (Applied Biosystems) with primers described in detail elsewhere [30]. HIV-1 Gag sequences were obtained for 392 of the 406 studied individuals, and KIR genotyping was inconclusive for two of these patients. Our analysis is therefore based on the 390 study participants with full Gag sequencing and KIR/HLA genotyping. The underlying consensus clade C sequence was based on the HIV-1 consensus sequence of the 392 study participants. Gag-protease sequences obtained in this study are available in the GenBank database under nucleotide sequence accession numbers HM593106 to HM593510.

Identification of KIR/HLA-Associated Sequence Polymorphisms within HIV-1 p24 Gag

A statistical approach that yielded a rank-ordered list of the most likely instances in which HLA/KIR combinations interacted to drive selection of specific viral amino acid residues was developed to identify combinations (1) in which specific HLA/KIR combinations were strongly associated with (or against) a particular viral amino acid and (2) where this association was stronger among individuals expressing both the HLA and KIR than among those expressing only the HLA or only the KIR. To this end, two separate tests were performed: (1) a phylogenetically corrected method to identify strong HLA+/KIR+ associations and (2) a logistic-regression-based interaction test to identify instances in which there was evidence of an interaction. The phylogenetically corrected method was run as previously described [31,32], using HLA/KIR, HLA, and KIR as independent variables and using forward selection with p < 0.05 as the addition criterion. The interaction test was performed using standard logistic regression, with significance determined by a likelihood ratio test for a model that included the interaction term versus one that did not. These two sets of tests were each independently applied to each viral amino acid observed at each site (see below for details of the binary encoding).

To combine these tests, we defined a new test statistic t_i, which was simply the result of multiplying the p-values of the two tests for HLA-KIR-AA triple i. The significance (in terms of both p-values and false discovery rate) as estimated by q-values [33] was then estimated by comparing to null data, which were generated by repeating this approach on ten independent datasets. Each dataset was generated by randomizing the assignment of individuals to KIR haplotypes (thus preserving the HLA and KIR haplotype structures, as well as the strong HLA-escape-driven signals). The distribution of q-values included three associations with q ≤ 0.13, with the remainder with q > 0.35. As q-values are simultaneously conservative [34], and therefore allow post hoc selection of a threshold, we chose the top three associations to pursue further and report the q-values in Table 1. By combining these tests, we aimed to identify candidate polymorphisms selected only by the particular combination of HLA and KIR for further experimental analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. KIR footprints in HIV-1 clade C sequence.

https://doi.org/10.1371/journal.pmed.1001900.t001

To reduce the number of tests, we applied a number of filters to the data as a predefined pre-processing step. Starting with KIR/HLA combinations for which the KIR molecule was known to bind to the HLA molecule [35], we further limited the tests to HLA-KIR-AA combinations that exhibited enough variation to yield meaningful tests and to reduce the computational burden. To this end, we arbitrarily chose ten as a minimum-count threshold for the HLA and KIR alone variables. We limited our tests to HLA-AA and KIR-AA combinations for which there were at least ten individuals in each of the following groups: HLA+, HLA−, KIR+, KIR−, AA+, AA−. For the combined HLA/KIR interaction test, we applied a stronger minimum count of six individuals in the following groups: HLA+/KIR+, AA+, AA−. Although arbitrary, these minimum-count thresholds were prespecified, and we did not explore other thresholds, so as to preserve the fidelity of the statistical analyses.

Amino acids were treated as binary variables to reduce the number of parameters required in the models, and separate models were created for all observed amino acids (subject to the minimum-count filter). Sites for which a mixture of amino acids was observed were treated as non-consensus. For example, for a site in which X, Y, and Z were amino acids that passed the minimum-count filter and in which X matched the cohort consensus, the mixture {X, Y} would yield the encoding [X = 0, Y = 1, Z = 0], while {Y, Z} would yield the encoding [X = 0, Y = 1, Z = 1]. Overall n = 13,826 KIR-HLA-AA triples were tested, with n_r = 182,734 tests in the randomized KIR runs (n_r ≠ 10n because of the HLA+/KIR+ ≥ 6 filter).

Cell Lines and Primary Natural Killer Cell Populations Used

The HLA class I—deficient human B cell line 721.221 stably transduced with ICP47 was kindly provided by Emmanuel J. H. J. Wiertz (Department of Medical Microbiology, University Medical Center, Utrecht, The Netherlands). ICP47 is an early protein of HSV-1, which blocks loading of endogenous peptide onto HLA class I at the peptide binding site of TAP [36,37]. For the generation of a 721.221-ICP47 cell line stably expressing HLA-C*03:04, 293T cells (ATCC) were transfected with VSV-G, PCG (kindly provided by Thomas Pertel, Dana-Farber Cancer Institute), and pMIP, a retroviral transfer vector containing the HLA-C*03:04 gene (kindly provided by Todd Suscovich, Ragon Institute of MGH, MIT and Harvard). Retroviral supernatant was harvested at day three after transfection and further used to transduce 721.221-ICP47 cells to obtain stable expression of HLA-C*03:04. Transduced cells were selected in 1.5 μg/μl puromycine and maintained in RPMI medium 1640 (Sigma-Aldrich) supplemented with 10% heat-inactivated fetal calf serum (Sigma-Aldrich), 2,500 U/ml penicillin, 2,500 μg/ml streptomycin, and 100 mM L-glutamine (Cellgro) at 37°C under 5% CO₂.

HLA-C*03:04 Stabilization and KIR2DL3-Fc Binding Assays

721.221-ICP47-C*03:04 cells were washed twice to remove any peptides remaining from FCS-supplemented medium and were subsequently pulsed with 100 μM of the indicated HIV-1 peptides in non-supplemented RPMI for 20 h at 26°C. Cells cultured with previously described HLA-C*03:04-stabilizing self-peptides (GAVDPLLAL [GAL] and GAVDPLLKL [GKL]) [9] were used as positive controls, while cells cultured with an influenza peptide (ILRGSVAHK) and in the absence of peptides were used as negative controls. After staining with the anti-pan-HLA antibody W6/32 (BioLegend), cells were fixed in 4% paraformaldehyde and analyzed by flow cytometry to quantify HLA class I expression. Titration experiments were performed using 100 μM, 10 μM, 1 μM, and 0.1 μM peptide, as indicated. KIR binding assays with the candidate peptides and controls were performed after 20 h of HLA class I stabilization. Cells were stained using 2.5 μg of KIR2DL3-Fc (recombinant human KIR2DL3-IgG-Fc chimera; R&D Systems, KIR2DL3*001 subtype) for 1 h on ice, followed by a secondary staining with anti-human-Fc antibody (PE-conjugated IgG goat anti-human polyclonal antibody, Invitrogen), as described previously [3]. Following fixation of cells in 4% (w/v in PBS) paraformaldehyde (Affymetrix), flow cytometric analysis was performed using a BD LSR II.

Assessment of Primary Natural Killer Cell Activation

To determine activation of primary NK cells following exposure to target cells pulsed with different peptide variants, NK cell degranulation assays measuring CD107a expression were performed using freshly isolated and purified NK cells from nine healthy KIR2DL3+ individuals [38]. Primary NK cells were isolated by incubating whole blood with RosetteSep Human NK Cell Enrichment Cocktail (Stemcell Technologies) for 20 min at room temperature, followed by Histopaque-1077 (Sigma) density gradient centrifugation. NK cells were rested overnight in RPMI medium 1640 (Sigma-Aldrich) supplemented with 10% heat-inactivated fetal calf serum (Sigma-Aldrich), 2,500 U/ml penicillin, 2,500 μg/ml streptomycin, 100 mM L-glutamine (Cellgro), and 1.0 ng/ml IL-15 (Cellgro). Subsequently, NK cells (1 × 10⁵) were co-incubated with peptide-pulsed 721.221-ICP47-C*03:04 cells (5 × 10⁵) at an effector:target ratio of 1:5 in RPMI containing anti-human CD107a-PE-Cy7 (12.5 μl/ml) and monensin (1.5 μl/ml, BD GolgiStop). Cells were incubated for 6 h at 26°C in 5% CO₂. Cells were stained with 7AAD, anti-CD3-PB, anti-CD16-BV785, anti-CD56-BV605, anti-CD14/19-BV510, anti-KIR2DL2/3-PE, and anti-KIR2DL3-APC, washed, fixed with 4% (w/v in PBS) paraformaldehyde (Affymetrix), and analyzed by multiparameter flow cytometry using a BD LSR II.

Analysis of Binding Affinities via Surface Plasmon Resonance

Surface plasmon resonance (SPR) measurements were made using a Biacore 3000 system; the experimental buffer used was HBS-EP (Biocore). To assess binding of the viral variants to KIR2DL3, biotinylated HLA-C*03:04 monomers (42 kDA) refolded with the peptides YVDRFFKTL, YVDRFFKAL, and YVDRFFKVL (Immudex) were immobilized onto a Streptavidin sensor chip (GE Healthcare) to approximately 500 response units for optimal responses. A blank flow cell with no immobilized ligand was used as a reference flow cell. Injections of 60 μl of KIR2DL3-Fc (recombinant human KIR2DL3-IgG- Fc chimera; R&D Systems, 102.4 kDA) diluted in PBS to a concentration of 10 μg/ml were performed at a flow rate of 20 μl/min, with a subsequent 10-min run of buffer to allow sufficient dissociation. Due to described instability after acid treatment of the HLA-C/peptide complex, acidic regeneration of the chip was not performed [39]. As a last step, the amount of each HLA monomer refolded with the respective epitope variant immobilized on the chip was determined by HC10 antibody staining. Raw sensograms were corrected by double referencing (subtracting from the reference flow cell response and from the PBS injection response). Given the dimeric nature of the KIR2DL3 analytes, a biphasic model was used to obtain the fit; binding constants are given as mean ± standard error of the mean (SEM). Experiments were performed at 25°C.

Computational Modeling

All-atom, explicit solvent molecular dynamic simulations were used to obtain a detailed atomistic understanding of the structural and dynamic properties that modulate the specificity of binding of the three-way interaction between KIR2DL3, HLA-C*03:04, and peptide. An experimentally resolved crystal structure (Protein Data Bank ID 1EFX) [9] of KIR2DL2, in complex with its class I ligand HLA-C*03 and a self-peptide (GAL), was used as a structural template for all the studies herein and was adapted to fit KIR2DL3 modeling. A free energy perturbation (FEP) procedure was followed to obtain differences in the binding free energy (ΔΔG) between MHC with wild-type or mutant peptide (pMHC) and the KIR molecule. Further simulation details can be found in S1 Text, describing the modeling of the three-way interaction between KIR2DL3/HLA-C*03:04/peptide. S1 Fig shows the stability of the 1EFX structure (two KIR in complex with HLA-C*03 presenting GAVDPLLAL. S2 Fig illustrates the differences in the free energy of binding upon mutations of the self-peptide GAVDPLLAL. In S3 Fig we show the stability of the T_Gag303 (YTL)–loaded HLA complex with KIR2DL3 binding. S4 Fig displays the insensitivity of the peptide binding groove to the identity of the peptide. S1 Text describes the details of the computational modeling.

Data Acquisition and Analysis

Flow cytometry data were analyzed using FlowJo software version 10.0.6 (Tree Star), and statistical analysis was performed using GraphPad Prism 6 (GraphPad Software). HLA stabilization values are shown as mean relative fluorescence intensity; error bars represent standard deviation (SD). Relative fluorescence intensity was calculated by dividing the geometric mean fluorescence intensity (gMFI) of HLA expression in the presence of the tested peptide by the gMFI of stained target cells without peptide added in each respective assay. Binding of KIR2DL3-Fc is represented as percent of KIR2DL3+ cells ± SD. NK cell degranulation values are expressed as relative degranulation, by dividing the percentage of CD107a+ NK cells in a sample by the percentage of CD107a+ NK cells in the negative control sample from the same individual (where the non-binding GKL peptide was added). A repeated-measures one-way ANOVA was subsequently performed to compare the responses between the wild-type group and the groups of the two variant peptides, for the KIR2DL3+ and KIR2DL3− NK cells, respectively. This was followed by a Tukey’s post hoc test. Each individual was sampled one time.

Identification of KIR/HLA Class I—Associated Sequence Polymorphisms within the HIV-1 Clade C p24 Gag Sequence

Previous studies have demonstrated that HIV-1 Gag represents an important target for cell-mediated immunity against HIV-1, including in individuals infected with HIV-1 clade C [40], resulting in viral escape from CD8+ T cell—mediated immune pressure by selection of sequence polymorphisms [27,30,41–46]. To determine the mechanism by which KIR+ NK cells might also impose immune pressure on HIV-1, we sequenced autologous p24 Gag genes from 392 individuals from Durban, South Africa, chronically infected with untreated HIV-1 clade C [30], for whom high-resolution HLA class I and KIR type information was available. Given the size of the cohort, we sought to identify sequence polymorphisms associated with the presence of specific KIR/HLA class I combinations in this population, rather than with the presence of the respective KIR allele or HLA class I allele alone [28], to guide further functional experiments. We therefore developed an exploratory statistical approach to identify sequence polymorphisms overrepresented in individuals with both the respective KIR and HLA class I alleles, but not in individuals with either the HLA class I allele alone or the KIR allele alone (or neither of them). This approach of combining the phylogenetically corrected method and the logistic-regression-based interaction test was chosen because, by themselves, the phylogenetically corrected method yielded a number of associations that were driven by the HLA allele alone, matching well-characterized T cell escape mutations, and the interaction test alone yielded no results that were significant after correcting for multiple tests (q = 1 for all tests).

The combined approach was subsequently applied to all 13,826 KIR/HLA polymorphism combinations in HIV-1 p24 Gag, and yielded three associations at two sites (Gag₃₀₃ and Gag₃₄₀) that were considered significant given the number of hypotheses tested (q ≤ 0.13; indicating a 13% false discovery rate) and were chosen because the next most significant association had a q-value of 0.35, providing a natural threshold. A combined genotype encoding KIR2DL3 and HLA-C*03:04 was associated with a valine at position Gag₃₀₃ and an alanine at position Gag₃₄₀, while not having the combined genotype was associated with having the wild-type threonine at position Gag₃₀₃. A sequence change to having alanine at Gag₃₀₃ was furthermore significantly related to having the HLA-C*03:04 allele (p < 0.001), but upon combining with a KIR allele, significance was lost, as it was selected only in an HLA-C*03:04-related manner. In contrast, the sequence polymorphism encoding valine at Gag₃₀₃ was solely significant for the combined KIR/HLA genotype, whereas no effect for this variant was observed among individuals possessing the HLA-C*03:04 allele alone.

More than half of the viruses from individuals with the KIR2DL3/HLA-C*03:04 genotype (18/33) had a sequence polymorphism in amino acid position 303 of p24 Gag, with a T_Gag303V mutation in 39.5% of individuals (10/33) and a T_Gag303A variant in 15% of individuals (5/33) (Fig 1A). Mixed variant and wild-type residues within the same patients were treated as escape in this position (3/33), as mixed residues suggested ongoing but not yet fully fixed escape. In contrast, 90% of individuals (312/348) not possessing this combined KIR/HLA genotype had virus with the T_Gag303 wild-type sequence (p < 0.001, Fisher’s exact test; Fig 1A), and mixed residues were seen in virus from only two out of 348 patients. At amino acid position 340 of p24 Gag, 40% of the infected individuals with the KIR2DL3/HLA-C*03:04 genotype had virus with a G_Gag340A variant (14/35), compared to only 9% of KIR2DL3−/HLA-C*03:04− participants (33/353) (Fig 1B; p < 0.001, Fisher’s exact test), with the percentage of mixed residues at this position changing from 1% (4/353) to 6% (2/35) in the KIR2DL3/HLA-C*03:04 genotype. Taken together, these statistical analyses using different models revealed polymorphisms within the HIV-1 clade C p24 Gag sequence associated with the presence of a specific combined KIR/HLA genotype in the studied population, but not with the presence of the respective KIR or HLA class I allele alone.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of frequencies of wild-type and variant Gag sequences in viruses from KIR2DL3+/HLA-C*03:04+ individuals compared to individuals not possessing the combined KIR2DL3/HLA-C*03:04 genotype.

(A) A valine (V) at amino acid position 303 of HIV-1 Gag was significantly (p < 0.001, Fisher`s exact test) overrepresented in KIR2DL3+/HLA-C*03:04+ individuals (n = 33) compared to individuals not possessing this combined genotype (non-KIR2DL3 and HLA-C*03:04, n = 348), 90% of whom had virus with the consensus sequence wild-type threonine (T) in that position (n = 312). Furthermore, an alanine (A) variant at position Gag₃₀₃ was present in five of the 33 individuals in the KIR2DL3+/HLA-C*03:04+ group (15%). The percentage of individuals whose virus showed mixed residues at position Gag₃₀₃ was 0% in KIR2DL3−/HLA-C*03:04− individuals and 9% in KIR2DL3+/HLA-C*03:04+ individuals. (B) An alanine (A) at position 340 of HIV-1 Gag was significantly (p < 0.001, Fisher’s exact test) overrepresented in KIR2DL3+/HLA-C*03:04+ individuals (n = 35) compared to individuals not possessing this combined genotype (non-KIR2DL3 and HLA-C*03:04, n = 353), 90% of whom had virus with the consensus sequence wild-type glycine (G) in that position (n = 316). Two patients in the KIR2DL3−/HLA-C*03:04− subset had neither the wild-type nor the G_Gag340A variant and are not shown in this graph. The percentage of individuals whose virus showed mixed residues at position Gag₃₄₀ was 1% in KIR2DL3−/HLA-C*03:04− individuals and 6% in KIR2DL3+/HLA-C*03:04+ individuals. Conclusive KIR2DL3 typing was not available for two out of the 392 study participants; these two participants were excluded from further analysis.

https://doi.org/10.1371/journal.pmed.1001900.g001

The T_Gag303 Variants Are Contained within an HLA-C*03:04-Presented Epitope and Are Equally Well Presented by HLA-C*03:04

To determine whether the two identified KIR/HLA-associated sequence polymorphisms within HIV-1 p24 Gag were contained within epitopes presented by the respective HLA class I molecules, we initially used HLA binding prediction models to identify potential epitopes. The amino acid polymorphism at position 303 of Gag was located within a previously described HLA-C*03:04-restricted CD8+ T cell epitope [47,48], and the wild-type epitope sequence YVDRFFKTL was predicted as a binder of HLA-C*03:04 using the NetMHCpan 2.8 predictor tool [49]. Notably, the sequence polymorphisms T_Gag303V and T_Gag303A were located at position P8 of the nonamer peptide, a position within the peptide for which the crystal structure of HLA-C*03:04 and KIR2DL2—an allele closely related to KIR2DL3—indicates an important role for KIR binding [9]. The epitope containing this footprint is highly conserved in HIV-1 clade C sequences published in the Los Alamos HIV Sequence Database (http://www.hiv.lanl.gov), with variation occurring almost exclusively at position P8 of the nonamer. Also, the second KIR2DL3/HLA-C*03:04-associated sequence polymorphism at amino acid position 340 of Gag was contained within a nonamer peptide (RALGPGATL) that showed a high binding affinity to HLA-C*03:04 using the NetMHCpan 2.8 restriction tool. As it had not been previously described as an HLA-C*03:04-restricted HIV-1 epitope, we tested 8- to 11-amino-acid-long peptides containing the G_Gag340 wild-type or the G_Gag340A variant sequence, and were able to confirm that the nonamer RALGPGATL was the optimal minimal epitope binding to HLA-C*03:04 at the lowest peptide concentration (S5A and S5B Fig). However, the sequence polymorphism at position 340 of Gag was located in position P6 of the HLA-C*03:04-presented epitope, a position not classically reported to be relevant for KIR binding to HLA-C*03:04.

We next investigated whether sequence variants in these epitopes binding to HLA-C*03:04 modulated the level of HLA expression on HLA-C*03:04+ cells. Using 721.221 cells, an HLA class I—devoid cell line, transduced with a single HLA class I gene (HLA-C*03:04) and ICP47 (an early HSV-1 protein known to block TAP [50]), allowed us to determine the ability of exogenously added peptides to bind to and stabilize HLA-C*03:04. As positive controls, the endogenous self-peptides GAVDPLLAL (GAL) and GAVDPLLKL (GKL), previously known to bind to HLA-C*03:04 [9], were used. 721.221-ICP47-C*03:04 cells were pulsed with the wild-type epitope YVDRFFKTL and its variants YVDRFFKVL and YVDRFFKAL and tested for HLA-C*03:04 stabilization by staining with the anti-pan—HLA class I antibody W6/32 (Fig 2). Pulsing with wild-type and variant peptides induced up to 4-fold higher expression of HLA-C*03:04 on the cell surface as compared to non-peptide-pulsed 721.221-ICP47-C*03:04 cells or 721.221-ICP47-C*03:04 cells loaded with the negative control peptide. More importantly, no differences in HLA surface expression were observed between the YVDRFFKTL, YVDRFFKVL, and YVDRFFKAL peptides, suggesting a stable and similar presentation of these three epitope variants by HLA-C*03:04 (T [mean 3.24 ± 0.64 SD] to A [mean 3.17 ± 0.53 SD], p = 0.996; T to V [mean 3.40 ± 0.74 SD], p = 0.681; A to V, p = 0.907). Titration experiments for HLA-C*03:04 stabilization using serial peptide dilutions confirmed equal HLA stabilization induced by wild-type and variant peptides containing the Gag₃₀₃ polymorphism at non-saturating levels (T [mean 2.60 ± 1.45 SD] to A [mean 1.84 ± 0.79 SD], p = 0.732; T to V [mean 2.17 ± 1.36 SD], p = 0.229; A to V, p = 0.946; ANOVA/Tuckey’s post hoc test), indicating unaltered binding affinity despite the amino acid mutation in position P8 of the peptide. In contrast, RALGPGATL, the second HLA-C*03:04-restricted epitope containing position Gag₃₄₀, bound as wild-type sequence with significantly higher affinity to HLA-C*03:04 than the epitope containing the sequence variant, RALGPAATL, when using non-saturating peptide amounts (G [mean 4.24 ± 0.46 SD] to A [mean 2.72 ± 0.81 SD], p = 0.006; paired, two-tailed t-test with 1 μM peptide; S5C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Equal HLA-C*03:04 stabilization on TAP-blocked 721.221-ICP47-C*03:04 cell line with HIV-1 p24 Gag T_Gag303 wild-type and variant peptides.

(A) Representative histograms of HLA-C*03:04 stabilization with T_Gag303 wild-type and T_Gag303A and T_Gag303V variant peptides compared to addition of no peptide (−). HLA-C*03:04 surface levels were determined by flow cytometry using an anti-pan—HLA class I antibody (clone W6/32). Peptides were added at a saturating concentration of 100 μM. (B) Quantification of HLA-C*03:04 stabilization in the presence of T_Gag303 wild-type and T_Gag303A and T_Gag303V variant peptides, as well as positive endogenous control peptides for HLA-C*03:04 stabilization (GAVDPLLAL and GAVDPLLKL) and a non-HLA-C*03:04-stabilizing influenza-derived control peptide (Flu, ILRGSVAHK). Data represent mean of five experiments with error bars indicating the SD. Relative fluorescence intensity (RFI) was calculated as the gMFI of the sample divided by the gMFI of 721.221-ICP47-C*03:04 cells stained in the absence of peptide.

https://doi.org/10.1371/journal.pmed.1001900.g002

Taken together, these data demonstrate that the statistical approaches used identified two KIR2DL3/HLA-C*03:04-associated sequence polymorphisms within HIV-1 p24 Gag that were indeed contained within HLA-C*03:04-presented HIV-1 epitopes, providing experimental data that validate the statistical approach. While the KIR2DL3/HLA-C*03:04-associated T_Gag303V and T_Gag303A sequence polymorphisms had no impact on binding to and expression of HLA-C*03:04, the G_Gag340A mutation reduced the ability of the epitope to bind to HLA-C*03:04. As the subsequent experimental approach used to assess the consequences of peptide variations on KIR binding relied on equal presentation of variant and wild-type peptides by HLA-C*03:04 for conclusive results, assessment of the consequences of the G_Gag340A mutation for KIR2DL3 binding was not feasible. We therefore selected the T_Gag303 variants to determine the consequences of the KIR2DL3/HLA-C*03:04-associated sequence polymorphisms for KIR/HLA interactions and NK cell function.

The KIR2DL3/HLA-C*03:04-Associated T_Gag303V Mutation Results in Enhanced Binding of the Inhibitory KIR2DL3 Receptor

To assess the consequences of the T_Gag303V polymorphism on KIR2DL3 binding to HLA-C*03:04, we stained with a fusion chimera of KIR2DL3 attached to and dimerized by the Fc portion of hIgG1 (KIR2DL3-Fc) after labeling 721.221-ICP47-C*03.04 cells with the three different peptide variants (YVDRFFKTL, YVDRFFKVL, and YVDRFFKAL), and quantified KIR binding using a secondary PE-conjugated anti-human-IgG antibody. The positive control peptide GAVDPLLAL (GAL), known to allow for binding to KIR2DL2, which has binding patterns similar to those of KIR2DL3 [3], was used as an internal control for the engagement of KIR2DL3. The corresponding HLA-C*03:04-binding variant GAVDPLLKL (GKL) did not allow for KIR2DL3 binding, while still stabilizing HLA-C*03:04 at similar levels, and thus was used as a negative control. The results of repeated (n = 6) KIR2DL3-Fc binding assays are summarized in Fig 3A and 3B. While the wild-type peptide YVDRFFKTL and the variant peptide YVDRFFKAL allowed for a certain degree of KIR2DL3 binding, the YVDRFFKVL variant peptide preferentially selected in KIR2DL3+/HLA-C*03:04+ individuals resulted in significantly stronger binding of KIR2DL3-Fc (T [mean 18.01 ± 10.45 SD] and V [mean 44.67 ± 14.42 SD], p = 0.002; Fig 3A and 3B). Binding of KIR2DL3-Fc to HLA-C*03:04 in complex with the T_Gag303V variant even exceeded binding to HLA-C*03:04 in complex with the positive control peptide GAL. The T_Gag303A variant, previously reported to mediate escape from cytotoxic T lymphocyte (CTL) recognition [47], did not enhance binding to KIR2DL3-Fc compared to the wild-type sequence (T to A [mean 15.46 ± 7.24 SD], p = 0.811). This is in line with our findings showing no statistical association of this variant with the KIR2DL3+/HLA-C*03:04+ combined genotype, but rather only with HLA-C*03:04 alone, which might hint at CD8 T cell—related escape. Taken together, these data demonstrate that the T_Gag303V polymorphism selected in KIR2DL3+/HLA-C*03:04+ individuals led to a considerable increase in binding of the inhibitory KIR2DL3 receptor to HLA-C*03:04 as compared to the wild-type sequence.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Binding of KIR2DL3-Fc and inhibition of primary KIR2DL3+ NK cells is significantly stronger when target cells are pulsed with the T_Gag303V variant peptide.

(A) Representative dot plots of KIR2DL3-Fc staining of 721.221-ICP47-C*03:04 cells pulsed with positive control peptide GAL, negative control peptide GKL, T_Gag303 wild-type peptide, or T_Gag303V variant peptide. Staining was measured by flow cytometry after addition of a secondary PE-conjugated anti-IgG antibody. All cells express GFP, indicating successful transduction of ICP47 and thus optimal blockade of TAP. (B) Quantification of KIR2DL3 binding expressed as percent of cells binding KIR2DL3-Fc. Means of six independent experiments, with error bars representing SD, are shown. Binding of KIR2DL3-Fc was significantly stronger to 721.221-ICP47-C*03:04 cells pulsed with the variant T_Gag303V (YVDRFFKVL) peptide than to those pulsed with wild-type (YVDRFFKTL; p = 0.002) or T_Gag303A (YVDRFFKAL; p = 0.002) peptide. (C and D) Degranulation of primary KIR2DL3+ (C) and KIR2DL3− (D) NK cells measured as CD107a expression after incubation with 721.221-ICP47-C*03:04 cells pulsed with T_Gag303 wild-type and T_Gag303A and T_Gag303V variant peptides. The percentage of CD107a+ NK cells in response to the respective peptide is normalized to the percentage of CD107a+ NK cells for KIR2DL3− and KIR2DL3+ NK cell subsets after co-incubation with target cells pulsed with GKL (GAVDPLLKL) control peptide. 721.221-ICP47-C*03:04 target cells presenting YVDRFFKVL significantly inhibit degranulation of KIR2DL3+ NK cells compared to targets pulsed with YVDRFFKTL (p = 0.0121) or YVDRFFKAL (p = 0.0019). The different peptide variants had no detectable effect on CD107a expression in KIR2DL3− NK cells. All p-values stated are adjusted for multiplicity of testing. Primary NK cells from nine different healthy KIR2DL3+ participants were tested. (E) Sensogram of KIR2DL3 dimeric analyte binding to biotinylated HLA-C*03:04/YVDRFFKTL, HLA-C*03:04/YVDRFFKAL, and HLA-C*03:04/YVDRFFKVL monomers on a Streptavidin chip; an empty well served as negative control. The sensogram data are normalized to the amount of respective HLA monomer immobilized on the chip by HC10 antibody. Each fit was obtained using double reference subtraction and is shown as a black line. Affinity of the KIR2DL3/HLA-C*03:04/YVDRFFKVL protein-protein interaction was highest. *p < 0.05; **p < 0.01.

https://doi.org/10.1371/journal.pmed.1001900.g003

The KIR2DL3/HLA-C*03:04-Associated T_Gag303V Mutation Inhibits KIR2DL3+ NK Cell Degranulation

We subsequently investigated whether the stronger KIR2DL3-Fc binding to HLA-C*03:04 induced by the T_Gag303V variant peptide altered the effector function of KIR2DL3+ NK cells. Primary NK cells from healthy KIR/HLA-typed donors were co-incubated with 721.221-ICP47-C*03:04 cells pulsed with the peptide variants YVDRFFKTL, YVDRFFKVL, and YVDRFFKAL. Degranulation was measured by CD107a expression on NK cells, and KIR2DL3+ and KIR2DL3− NK cell subsets were distinguished using an antibody specific for KIR2DL3. When evaluating the function of KIR2DL3+ NK cells, no significant differences in CD107a expression after stabilization with the wild-type (YVDRFFKTL) or YVDRFFKAL peptide were observed (T [mean 0.78 ± 0.07 SEM] to A [mean 0.89 ± 0.05 SEM], p = 0.086), reflecting the results of the KIR2DL3-Fc binding assays (Fig 3B). However, co-incubation with 721.221-ICP47-C*03:04 cells pulsed with the YVDRFFKVL variant peptide led to a significant inhibition of degranulation of KIR2DL3+ NK cells compared to the wild-type peptide (T to V [mean 0.63 ± 0.07 SEM], p = 0.0121; Fig 3C) and to variant YVDRFFKAL (A to V, p = 0.002). In contrast, the different peptide variants had no effect upon the activation of KIR2DL3− NK cells (Fig 3D), as expected (T [mean 1.06 ± 0.04 SEM] to A [mean 1.06 ± 0.04 SEM], p = 0.933; T to V [mean 1.04 ± 0.05 SEM], p = 0.828; A to V, p = 0.603). Overall, results from KIR2DL3 binding to peptide-pulsed 721.221-ICP47-C*03:04 cells and degranulation of KIR2DL3+ NK cells in response to peptide-pulsed 721.221-ICP47-C*03:04 cells were closely correlated for the different peptides tested (R = 0.76, p = 0.02). Altogether, the sequence polymorphism T_Gag303V in p24 Gag selected for in individuals infected with HIV-1 clade C who have the combined genotype KIR2DL3/HLA-C*03:04 resulted in a blunting of primary KIR2DL3+ NK cell responses against HLA-C*03:04-expressing cell lines, while effector function of KIR2DL3− NK cells remained unaffected.

Surface Plasmon Resonance Displays Enhanced Binding of HLA-C*03:04-Presented T_Gag303V to KIR2DL3 Dimers

In order to further validate our findings, we performed SPR analysis to quantitatively assess the affinity between KIR2DL3 and the HLA-C*03:04-presented wild-type and variant epitopes. Variant T_Gag303V (YVDRFFKVL) had the highest affinity to KIR2DL3 of the tested epitope variants upon presentation by HLA-C*03:04, with a K_D = 77 ± 3 nM. This was followed by the wild-type peptide T_Gag303 (YVDRFFKTL), with a K_D = 81 ± 3 nM. HLA-C*03:04 refolded with variant T_Gag303A (YVDRFFKAL) exhibited the lowest binding affinity of the three peptides, with a K_D = 93 ± 4 nM. The kinetic data indicated that the higher affinity of T_Gag303V to KIR2DL3 was driven by both faster complex formation and a lower dissociation rate. Together, these data indicate that the selected T_Gag303V variant binds with higher affinity to inhibitory KIR2DL3 than the wild-type or variant T_Gag303A epitope.

Computational Modeling Results Can Predict Binding of Variant Peptides to KIR2DL3-Fc and Functional Outcome of NK Cell Activation

The crystal structure of KIR2DL2 in conjunction with HLA-C*03:04 presenting a self-peptide (GAVDPLLAL) has been described previously [9]. Based on this structure, we used all-atom, explicit solvent molecular dynamic simulations adapted to the closely related KIR2DL3 structure, which differs by only 24 amino acid positions from KIR2DL2 (Immuno Polymorphism Database; https://www.ebi.ac.uk/ipd/kir/). We did this to obtain a detailed atomistic understanding of the structural and dynamic properties that modulate the specificity of binding between KIR2DL3 and HLA-C*03:04, with a particular interest in the dependence of the sequence of the HLA class I—presented peptide. First, to assess the validity and reach of the simulation protocol, two independent simulations were run for 50 ns each of the 1EFX structure containing an HLA-C*03:04 molecule loaded with GAL bound to a KIR2DL3 molecule. While some small inter-domain, hinge-like motions were observed in both simulations, each domain remained stable, as shown by Cα root mean square deviations (RMSDs) smaller than 0.15 nm (S1 Fig). Moreover, the binding surface between peptide-loaded HLA class I and KIR was unmodified during the length of the simulations, further supporting the thermal stability of the complex. KIR2DL3 binds orthogonally to one end of the peptide-binding groove, interacting with both the α1 and α2 helices of the HLA molecule and the C-terminus of the peptide (only residues 7 and 8), as shown in Fig 4. The complex is held together by strong charge complementarity between the KIR and corresponding HLA-C*03:04 surface, forming a dense network of salt bridges. Eight hydrogen bonds and four salt bridges formed between the negatively charged KIR surface and the positively charged HLA surface, and led to the strong recognition.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Structural details of the HLA/peptide/KIR three-way complex.

(A) Overall structure of the HLA/peptide/KIR complex. The peptide is buried in the HLA class I binding grove, while KIR interacts with both helices of HLA class I as well as the C-terminus of the peptide. (B) Comparison of the interactions around the mutated viral residue (top: in fat sticks; bottom: in spheres). The YVDRFFKTL (YTL) wild-type peptide includes a hydrogen bond between the side-chain oxygens of Thr₈ and Asn₈₀, while the YVDRFFKVL (YVL) variant improves the hydrophobic packing against Val₇₆. Interestingly, Asn₈₀ participates in a hydrogen bond expected to confer allotype specificity to KIR2Ds.

https://doi.org/10.1371/journal.pmed.1001900.g004

Once the binding mode was identified as adequate and stable under the simulation parameters employed, a rigorous FEP strategy was used to calculate the free energy changes of binding between the positive control peptide (GAVDPLLAL) and the negative control (GAVDPLLKL), along with three other peptide variants previously characterized experimentally for KIR2DL2 binding (GAVDPLLYL, GAVDPLLVL, and GAVDPLLSL) [9]. The results of the FEP calculations (S2 Fig) agree remarkably well with experimental results described by Boyington et al. [9] (S2 Fig), allowing for validation of both the current simulation parameters and the use of the FEP protocol to obtain the changes in binding free energy of the complex upon sequence variations of the self-peptide. Next, the endogenous self-peptide in the complex was replaced by the wild-type HIV-1 peptide YVDRFFKTL (T_Gag303); details of the two-step preparation procedure, equilibration, and stability assessments are provided in S3 and S4 Figs and S1 Text. The overall shape of the HLA-C*03:04 binding pocket for the peptide remained highly conserved upon substitution of the peptide chain (S4A Fig), consistent with the experimental data from the HLA-C*03:04 stabilization assays (Fig 2). The peptide backbone, however, differed only in the position P5, which twisted significantly and became more exposed to the solvent (S4B Fig). Both termini of the peptide remained completely buried in the binding pocket, with the charged C-terminus of the peptide engaging in multiple hydrogen bonds with Tyr₈₄, Tyr₁₂₃, and Thr₁₄₃ of HLA-C*03:04. The stabilized structure was then used to compute the changes in free energy (ΔΔG) upon binding of the wild-type sequence (T_Gag303) and five different variants: the in-vivo-selected HIV-1 sequence polymorphisms (T_Gag303V and T_Gag303A), a naturally occurring minor variant T_Gag303C (1.9%, Los Alamos HIV Sequence Database), and two additional experimental variants, T_Gag303F and T_Gag303G.

Compared to the wild-type sequence YVDRFFKTL, the T_Gag303V mutant showed a moderate enhancement in binding affinity (ΔΔG = −0.51 ± 0.41 kcal/mol; Table 2), which agreed well with the experimental observations (Fig 3B and 3C; Table 2). The free energy decomposition revealed that the stability of binding of this mutation was mainly due to a favorable electrostatic term, which contributed about −1.31 kcal/mol. It turned out that this contribution was largely from the much less favored free state because of the loss of hydrogen bonds with the solvent in the free state. The bound state also lost some hydrogen bonds with water, but because of the compensation from the KIR, the overall loss in total hydrogen bonds was less significant. In contrast, the T_Gag303A mutant slightly destabilized the complex (ΔΔG = 0.38 ± 0.34 kcal/mol), again in line with the experimental data. While threonine and valine are similar in size, the alanine residue is significantly smaller and induced a penalty in the overall favorable hydrophobic packing of the system. A similar situation occurred with the T_Gag303G mutant (1.10 ± 0.74 kcal/mol), with an energetic profile comparable to that of the alanine variant. The bulky aromatic group introduced by the T_Gag303F mutant resulted in a large entropic penalty due to less flexible Phe side chains. The mutations T_Gag303A, T_Gag303C, T_Gag303F, and T_Gag303G thus either decreased or completely destroyed the binding between HLA-C*03:04 and KIR2DL3 in our computational model, while the T_Gag303V mutant selected in KIR2DL3+/HLA-C*03:04+ individuals enhanced the binding to the inhibitory KIR2DL3 receptor in this model, in line with the experimental data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Experimental validation of modeling results.

https://doi.org/10.1371/journal.pmed.1001900.t002

Based on the KIR2DL3 binding predictions for the additionally tested minor variant T_Gag303C, and the two experimentally designed peptide variants (T_Gag303F and T_Gag303G), we performed subsequent experimental validations using these peptides. As shown in Table 2, a close correlation between the experimental data and modeling data for the variants T_Gag303F and T_Gag303G was observed, similarly to the previously tested T_Gag303 wild-type, T_Gag303V, and T_Gag303A peptides. Only the T_Gag303C variant gave discordant results—with the experimental data showing a more drastic decrease in KIR2DL3-Fc binding compared to wild-type (mean 0.08 ± 0.06 SD) than predicted by the change in ΔΔG (0.18 ± 0.52 kcal/mol) (Table 2). The ΔΔG values resulting from the modeling studies and the KIR2DL3-Fc binding data from the experimental studies were correlated (R = 0.5), but this correlation did not reach significance given the discordant results for the T_Gag303C variant. Excluding this variant resulted in a significant correlation between modeling results and experimental data (R = 0.8, p = 0.04). Taken together, the modeling data based on the KIR2DL2 crystal structure adapted to KIR2DL3 agreed very well with the observed experimental data, and also allowed for predictions about how peptide variants might modulate KIR/HLA class I interactions that could be experimentally validated.