Peripheral blood mononuclear cells (PBMCs) were isolated as previously described (15). All blood donors were recruited at the Blood Transfusion Center (EFS, Nantes, France), and informed consent was obtained from all individuals. Authorization for the preparation and conservation of biocollections (AC-2021-4397) was provided by the French Research Minister. Cells were cultured in RPMI 1640 medium (Gibco, Paisley, Scotland, UK) containing glutamine (Gibco) and penicillin-streptomycin (Gibco) and supplemented with 10% fetal bovine serum (Gibco).

HLA class I allele assignment was performed for all blood (n=200) and registered volunteer bone marrow donors (n=12,546) as previously described (27). Genetic typing of KIR was performed on all blood donors using the KIR multiplex PCR-SSP method, as previously described (28), with primers provided by Dr. Ketevan Gendzekhadze. The presence or absence of KIR2DL1, 2DL2, 2DL3, 2DL5, 3DL1, 2DS1, 2DS2, 2DS3, 2DS4/1D, 2DS5, and 3DS1 was determined. Donor KIR genotypes were determined based on the presence or absence of activating KIR. Thus, a KIR AA genotype was defined by the presence of only KIR2DS4 as an activating KIR gene, while a KIR B⁺ genotype was defined by the presence of several activating KIR genes (29). KIR ligands, such as HLA-A3/A11, Bw4 (HLA-A and/or HLA-B), C1, and C2, were identified in all blood donors based on allelic HLA-A, -B, and -C typing. HLA-E binds to a conserved peptide cleaved from the leader sequences of HLA-A, -B, or -C. While HLA-A and HLA-C supply a peptide leading to HLA-E binding, methionine/threonine dimorphism at position -21 of the leader sequence divides HLA-B allotypes into two groups: the first group with -21T, which does not supply HLA-E-binding peptides, and the second group with -21M, which does. Thus, we distributed all blood donors into three groups: -21T, -21M, and -21T/M.

To assign KIR alleles in all volunteer blood donors, KIR genes were captured by long-range PCR and subjected to sequencing on a MiSeq sequencer (Illumina, San Diego, CA, USA) after library preparation, as previously reported (30). KIR allele assignment was performed using the Profiler software version 2.24, developed by M. Alizadeh (Research Laboratory, Blood Bank, Rennes, France). An updated KIR allele library available in the IPD-KIR 2.8 database was implemented in Profiler.

The NK cell phenotype from healthy blood donors was determined by seven-color multiparameter flow cytometry on PBMCs using the following mouse anti-human mAbs: anti-NKp46-PE (9E2, Biolegend, San Diego, CA, USA), anti-KIR2DL1/S1-FITC (VPB6, Miltenyi Biotec, Paris, France), anti-KIR2DL2/3/S2-PerCP-Cy5.5 (GL183, Beckman Coulter, Immunotech, Marseilles, France), anti-KIR3DL1-AF700 (DX9, Biolegend, San Diego, CA, USA), anti-NKG2A-PE-Cy7 (Z199, Beckman Coulter, Immunotech, Marseilles, France), anti-CD57-PB (HNK-1, Biolegend, San Diego, CA, USA), and anti-NKG2C-APC (REA205, Miltenyi Biotec, Paris, France). CD94-specific mAb was not included, as it was consistently co-expressed on NKG2A⁺ NK cells (20). Spillover and co-expression were taken into consideration. The experimental controls included unstained, single-stained, and fluorescence minus one controls. A lock template was created to acquire stained samples. All flow cytometry data were acquired using a unique FACSCanto II cytometer at our ISO 9001-certified research laboratory. BD FACSDiva CS&T Research Beads (BD Biosciences) were used prior to acquisition.

The flow cytometry standard files were analyzed using Flowjo™ 10.6 software (LLC, Ashland, OR). They were then transferred to OMIQ data analysis software (OMIQ, Inc., Santa Clara, CA, USA). Dimensionality reductions were performed using the OMIQ software. After gating, subsampling of 50,000 NK cells per file was performed. Automated data cleaning was performed using PEACOQC to remove events with marginal values. The UMAP settings included random seeds. Subsequently, the FlowSOM settings included a run elbow meta-clustering on all 200 blood donors and seven features (KIR2DL1, NKp46, KIR2DL2/3, NKG2A, NKG2C, KIR3DL1, and CD57), resulting in 10 clusters. Metaclusters from the elbow method were visualized on a tree (data not shown).

Hierarchical clustering of the eight NK cell subsets of interest was performed following ward linkage using Genesis (31). The software is available at www.genome.tugraz.at.

Categorical data were analyzed using the chi-square test, and univariate comparisons were performed using Student’s t-test. Comparisons between multiple groups were performed using one-way ANOVA. All statistical analyses were performed using the GraphPad Prism version 9.3.1 software (San Diego, CA). Statistical significance was set at P < 0.05.

We established a cohort of 200 blood donors to investigate the effects of intrinsic and extrinsic markers on the phenotypic structure of the NK cell repertoire. We defined the three blood donor groups as C1C1 (n=79), C1C2 (n=59), and C2C2 (n=62). Thirty-three percent of blood donors were CMV seropositive. Moreover, this cohort included 44% of HLA-C homozygous donors and was thus not representative of Europeans. We first depicted this cohort after stratification along the C1, C2, A3/A11, and Bw4 KIR ligands. As a result, 12 groups were defined ( Figure 1A ). We also observed large disparities in the distribution of blood donors. Blood donors who did not express Bw4 (A3/A11^- or A3/A11⁺) were significantly more represented in C1C1 than in C1C2 or C2C2 (p<0.0001) ( Figure 1B ). The absence of Bw4 in A3/A11⁺ blood donors was significantly higher in C1C1 than in C1C2 and C2C2 blood donors (p<0.01) ( Figure 1C ). In contrast, C2⁺ (C1C2 and C2C2) blood donors frequently co-expressed Bw4 ligands (p<0.001) ( Figure 1D ). These observations were confirmed in a broad cohort of registered volunteer bone marrow donors (n=12546), representative of Europeans ( Figures 1E–G ). In the cohort of 200 blood donors, we observed more C2C2 females and C1C2 males in association with the Bw4 and A3/A11 ligands respectively ( Supplementary Figure 1A ). This association was not observed in the broader cohort of volunteer bone marrow donors. However, the distribution of Bw4⁺ HLA-A3/A11^- donors with seropositive CMV status was significantly different in females and males following treatment with HLA-C KIR ligands (p=0.01) ( Supplementary Figure 1A ), whereas the representation of C1C1, C1C2, and C2C2 was similar in both sexes. In the absence of Bw4, the donor distribution was similar according to sex or CMV status. In Bw4⁺ HLA-A3/A11⁺ donors, all C1C1 (C2^-) males were CMV⁺, whereas 71% of C2⁺ were CMV^- ( Supplementary Figure 1B ). Taken together, these observations indicate a strong association between KIR ligands.

To evaluate the effects of KIR and HLA genetic polymorphisms on the phenotypic structure of the NK cell repertoire, we determined, by flow cytometry, the frequency of KIR2DL1⁺, KIR2DL2/3⁺, KIR3DL1⁺, and four other subsets co-expressing these KIR and KIR^- NK subsets in blood donors (n=200). We further clustered these eight NK cell subsets from the cohort of 200 blood donors using unsupervised ward linkage clustering (Genesis^®) ( Figure 2A ). Six clusters of blood donors with specific phenotypic profiles were identified ( Figure 2B ). The investigated NK cell subsets are illustrated according to their mean frequencies in all clusters of blood donors ( Figure 2C ). We observed broad diversity in the representation of these NK cell subsets, with dominance of the KIR^- NK cell subset. Genetic and allelic KIR typing was performed on all blood donors. The cohort comprised 37 blood donors with non-expressed KIR3DL1 (18.5%). For KIR3DL1, we deduced high (H), low (L), high/low (H/L), and null (null/-) expression of KIR3DL1 from both KIR3DL1 allelic typing and KIR3DL1 mean fluorescent intensity. Specific KIR genetic characteristics (i.e., KIR2DL1, 2DL2, 2DL3, 2DL5, 3DL1^H, 3DL1^L, 3DL1^H/L, 3DL1^null/-, 2DS4, 3DS1, and 2DS1 genes) were compiled for each cluster of blood donors to analyze their potential impact on the structure of the NK cell repertoire ( Figure 2D ). Interestingly, clusters 1 (n=42) and 2 (n=34) harbored the highest frequencies of the KIR2DS4 gene, whereas cluster 6 (n=55) harbored the lowest frequency of this gene, and its presence gradually decreased from cluster to cluster. This result was in accordance with the representation of KIR genotypes in all blood donor clusters ( Supplementary Figure 2A ). KIR3DL1 expression was high in clusters 1 and 5, in which blood donors mainly harbored KIR3DL1^H alleles ( Supplementary Figure 2B ). Cluster 1 had the highest frequency of the KIR3DL1⁺ KIR2DL2/3/S2⁺ NK cell subset, whereas cluster 5 (n=18) presented the highest frequency of the KIR3DL1⁺ KIR2DL1/S1⁺ KIR2DL2/3/S2⁺ NK cell subset. In contrast, the clusters 4 (n=24) and 6 (n=55) were characterized by a high frequency of KIR3DL1^null/-, explaining a poor KIR3DL1 NK cell repertoire. In parallel with the decreased expression of KIR3DL1 from cluster to cluster, the frequency of KIR2DL5 and KIR2DS1 gradually increased from cluster 1 to cluster 4 and from cluster 5 to cluster 6 ( Figure 2E ). The presence of the KIR2DS1 gene was logically accompanied by increased expression of KIR2DL1/S1⁺ NK cells, as observed in clusters 4 and 6. Interestingly, clusters 1 and 5, which shared numerous genetic KIR characteristics, harbored different NK cell repertoire structures. Cluster 1 expressed a high frequency of KIR2DL2/3/S2⁺ NK cells, in contrast to cluster 5, in which NK cells co-expressing numerous KIR were well represented. The main difference between these two clusters can be explained by the differential KIR2DS4 and KIR2DL2 representation. Allelic KIR typing confirmed that the six donor clusters differed, particularly in the predominant KIR2DL1, KIR2DL2, KIR3DL1, and KIR2DS4 allele distributions ( Supplementary Figure 2B ). The HLA characteristics of each cluster in terms of KIR ligands (A3/A11, C1, C2, and Bw4) did not explain the main differences in the phenotypic structure of the KIR NK cell repertoires ( Supplementary Figure 2C ). Finally, we investigated the potential impact of age (<50 or >50 years), sex (male or female), and CMV status ( Supplementary Figure 2D ) on the phenotypic structure of the NK cell repertoire. No features seemed to be involved in driving the phenotypic KIR NK cell structure. Altogether, our observations show the importance of the level of expression of the KIR3DL1 allele-encoded product in the phenotypic structure of the KIR NK cell repertoire.

Using unsupervised clustering, we determined six main clusters among blood donors. Sub-clustering, particularly for cluster 6, which was marked by the absence of KIR3DL1 expression and an important number of blood donors (n=55), was further conducted to determine the secondary influence of particular KIR genes and/or ligands. Thus, four sub-clusters of cluster 6 (a, b, c, and d) were identified and compared ( Figure 3A ). Interestingly, the main difference was in the frequency of KIR2DL1/S1⁺ and KIR2DL2/3/S2⁺ NK cell subsets, with a high frequency of KIR2DL1/S1⁺ NK cells in cluster 6a, which declined from cluster to cluster until cluster 6c. This decreased representation was directly linked to KIR2DL2, whose frequency increased from cluster 6a to cluster 6c ( Figure 3B ). Moreover, the C2 ligands were highly represented in cluster 6a, and this representation declined in cluster 6c ( Figure 3C ). Radar charts of genetic KIR characteristics were similar for clusters 6c and 6d; however, cluster 6d blood donors showed significant expression of C2 ligands. Altogether, KIR2DL2 and C1/C2 KIR ligands were subtly interlinked to drive the phenotypic structure of the KIR2D⁺ NK cell repertoire.

We further evaluated the frequency of educated KIR⁺ NK cells by considering C1C2 and Bw4 KIR ligands, following the gating strategy illustrated for all Bw4^- and Bw4⁺ profiles (C1C1, C1C2, and C2C2) ( Figure 4A ). The frequency of educated KIR-NK cells was inversely correlated with that of KIR^- NK cells ( Figure 4B ). The frequency of educated NK cells was similar between AA and B⁺ blood donors (data not shown). However, the number of KIR ligands (C1, C2, and Bw4) increased the frequency of educated NK cells ( Figure 4C ). HLA-A3/A11 does not constitute a ligand that appears to contribute to NK cell education (data not shown), as previously reported (32). Although not significant, the frequency of educated KIR NK cells was higher in AA than in B⁺ blood donors with the three KIR ligands. In B⁺ blood donors, activating KIR genes can negatively modulate NK cell education, as previously described (33). We further evaluated the effect of the well-known KIR2DS1 gene and found that it did not affect the frequency of the pool of educated NK cells in B⁺ blood donors ( Figure 4D ). Sex did not affect the number of educated NK cells (data not shown). Finally, the frequency of educated NK cells significantly increased with age ( Figure 4E ) and CMV status ( Figure 4F ). Taken together, these data show that C1-, C2-, and Bw4-rich environments contribute to the mounting of a large pool of functionally educated NK cells. Moreover, CMV infection and advanced age boost the pool of educated NK cells.

In parallel with the supervised analysis, we performed an unsupervised analysis using OMIQ software based on 200 blood donors with the NK cell phenotype, including NKp46 (NK cells), KIR2DL1/S1, KIR2DL2/3/S2, KIR3DL1, NKG2A, NKG2C, and CD57. Ten clusters of NK cell subsets were identified. The NK cell subsets were classified according to their frequency in all blood donors ( Figure 5A ). These clusters were characterized by differential expression of NKG2A, CD57, KIR2DL2/3/S2, and NKG2C. The undifferentiated NKG2A⁺ KIR^- CD57^- NK cell subset was the most abundant subset. In contrast, the memory NKG2C⁺ KIR⁺ CD57⁺ NK cells were less abundant ( Figure 5A ). Notably, the range of NK cell frequencies was particularly important for all blood donors, underlying the broad diversity of the NK cell repertoire. UMAP profiles as a function of the number of KIR ligands are presented in Figure 5B . We observed an increased distribution of cluster 4, which was characterized by the expression of CD57 only in C2⁺ donors (C1C2 and C2C2) and in blood donors with a high number of Bw4 ligands ( Figure 5C ). Interestingly, the -21T motif, mainly displayed in the Bw4 and C2 molecules, was associated with a higher frequency of NKG2A^- KIR^- CD57⁺ NK cells (cluster 4). Finally, the immature NKG2A⁺ NK cell subset (cluster 1) was more abundant in the absence of HLA-A3/A11 ligands ( Figure 5D ). These data show the main influence of KIR ligands on the maturation profile of NK cells, with a high distribution of the NKG2A^- KIR^- CD57⁺ NK cell subset in the C2 and Bw4 environments, in contrast to the high distribution of the immature NKG2A⁺ NK cell subset in the absence of HLA-A3/A11 ligands.

We investigated the influence of KIR genotypes on the phenotypic structure of the NK cell repertoire. Cluster 4 (NKG2A^- KIR^- CD57⁺) and, to a lesser extent, cluster 5 (NKG2A^- KIR^- CD57^-) were more abundant in AA (n=76) blood donors than in B⁺ (n=124) blood donors ( Figure 5E ). In contrast, clusters 6 and 8 (NKG2A⁺ KIR⁺ CD57^+/-) were more represented in B⁺ compared to AA blood donors. KIR3DL1 harbors a broad allelic polymorphism, and some KIR3DL1 allele-encoded products are not expressed on NK cell surfaces. These variants are associated with the B⁺ KIR genotypes. Blood donors expressing KIR3DL1 displayed a higher frequency of CD57⁺ NK cells (cluster 4), whereas those who did not express KIR3DL1 displayed a higher frequency of the NKG2A⁺ KIR2DL2/3/S2⁺ CD57^- NK cell subset (cluster 8) ( Figure 5F ). Altogether, our data showed that KIR B⁺ and AA genotypes respectively drive a high distribution of KIR2DL2/3/S2⁺ NKG2A⁺ and KIR2DL2/3/S2^- NKG2A^- NK cell subsets.

To complete our study, we further investigated the impact of age and sex on the phenotypic structure of the NK cell repertoire. Using an unsupervised approach (UMAP), the distribution of NK cells was presented as density scatterplots in 10-year age brackets ( Figure 6A ). Immature NKG2A⁺ KIR^- CD57^- NK cells (cluster 1) were significantly more represented in younger blood donors (<50 years) ( Figure 6B ). Moreover, mature NKG2A^- KIR2DL2/3/S2⁺ CD57⁺ NK cells (cluster 7) were significantly more abundant in older blood donors (>50 years) ( Figure 6B ). As expected, the KIR2DL2/3/S2⁺ NKG2C⁺ CD57^high NK cell subset was significantly associated with CMV seropositivity ( Figure 6C ). Interestingly, the absence of KIR2DL5 was significantly associated with better representation of this memory NK cell subset (data not shown).

In parallel with this unsupervised analysis, we performed a supervised analysis of 32 NK cell subsets determined from NKG2A, CD57, and KIR (KIR2DL1, KIR2DL2/3, and KIR3DL1). Among the more represented NK cell subsets (mean threshold > 1%) from all identified NK cell subsets (n=32), KIR^- NK cells were predominant, particularly KIR^- CD57^- NKG2A⁺ NK cells (data not shown). KIR⁺ NK cell subsets preferentially express KIR2DL2/3/S2 and KIR2DL1/S1, and to a lesser extent, KIR3DL1. Using this approach, we showed that the KIR2DL1/S1⁺ NKG2A^- CD57⁺ NK subset was more frequent in blood donors displaying C2 ligands, the AA genotype, and the Cen AA motif, characterized by the absence of KIR2DL2 ( Supplementary Figure 3A ). In contrast, KIR2DL2/3/S2⁺ NKG2A^- CD57⁺ NK cell subset was more frequent in blood donors displaying C1 ligands ( Supplementary Figure 3A ). The KIR3DL1⁺ NKG2A^- CD57⁺ NK cell subset was more frequent in blood donors displaying the KIR AA genotype and Tel AA motif, while the KIR3DL1⁺ KIR2DL2/3/S2^+/- NKG2A^- CD57⁺ NK cell subset was more frequent in blood donors displaying the KIR AA genotype ( Supplementary Figure 3A ). Similarly, the KIR^- NKG2A⁺ CD57^- NK cell subset, the most frequent subset, was more frequent in blood donors displaying KIR2DL5 and those without the HLA-A3/A11 KIR ligand ( Supplementary Figure 3B ). Immature KIR^- NKG2A⁺ CD57^- NK cell subsets were higher in younger blood donors (<50 years) than in older blood donors (>50 years), which is in accordance with our results obtained using an unsupervised strategy ( Figure 6D ). The frequency of CD57^- NK cell subsets (NKG2A^- KIR^- CD57^- and NKG2A^+/- KIR2DL2/3/S2⁺ CD57^-) was higher in females than in males ( Figure 6E ). In contrast, the frequency of CD57⁺ NK cell subsets (NKG2A^+/- KIR2DL2/3/S2^+/- CD57⁺) was higher in males than in females. Altogether, our data show that the rules that guide the phenotypic structure of the NK cell repertoire integrate not only KIR and KIR ligands but also age, sex, and CMV status.