Evidence for Prescribed NK Cell Ly-49 Developmental Pathways in Mice

Natural killer cells are innate lymphocytes that function in immune cell surveillance by the recognition and elimination of cellular targets. Unlike their T and B lymphocyte counterparts, which acquire Ag-specific diversity through genetic recombination events, NK cells generate germline-encoded receptors that can recognize and lyse cellular targets by releasing perforin and granzymes and secrete regulatory and proinflammatory cytokines. Activating and inhibitory Ly-49 receptors employ NK cell signaling pathways, which dictate NK cell effector functions through cytoplasmic ITIMs and ITAM (1–4). Ly-49 receptors are thought to be successively acquired during the first weeks of birth such that a diverse repertoire is displayed during adulthood in mice (5, 6). The diverse NK cell subsets use a rheostat process in which NK cell subsets are tuned quantitatively by self–MHC class I (MHC-I) ligands corresponding to specific Ly-49 inhibitory receptors, which in turn provides a diversity of responsiveness toward cellular targets (7, 8). Through this mechanism, NK cells distinguish healthy from unhealthy cells. However, what regulates the acquisition of specific NK cell Ly-49 receptors during NK cell maturation is still an unanswered and complex question. In addition, how the biology of Ly-49 inhibitory receptors pertains to the acquisition of Ly-49 activating receptors is unresolved.

NK cell subsets that express single as well as overlapping Ly-49 activating and inhibitory receptors exist, which may reflect the complexity required to ensure host immunity while maintaining self-tolerance (9). Ly-49 receptors are encoded by the polymorphic and polygenic Klra genes located on mouse chromosome 6, and gene expression is often described as stochastic (10–12). Studies have shown that the NK cell inhibitory Ly-49 genes can be expressed in a stochastic monoallelic manner (13), whereas the activating Ly-49h and Ly-49d genes bias toward a biallelic expression (14). However, the process of Ly-49 surface receptor acquisition may not be entirely random, as it has been shown that NK cell inhibitory Ly-49 receptors can be regulated by self–MHC-I expression and controlled by the Ly-49 bidirectional transcriptional regulation of Pro1 and Pro2 and additional transcriptional factors (15–19). Mathematical methods to test the interdependence of surface expression of individual Ly-49 inhibitory receptors on the expression of other Ly-49 members in MHC-I–deficient and MHC-I–sufficient (wild type [WT]) mice support that Ly-49 inhibitory receptor expression may not be independently distributed and thus may not be entirely random (10, 20–23).

In this study, we tested the hypothesis that surface expression of Ly-49 activating receptors is not random and is influenced by expression of Ly-49 inhibitory receptors throughout NK cell maturation. To test this hypothesis, we used a combination of statistical, in vitro, and in vivo approaches. We provide evidence for prescribed (directed) pathways of NK cluster transitions in vitro and in vivo, which suggests that Ly-49 activating receptor acquisition is directed and influenced by NK cell maturation. Even though no known MHC-I ligand for Ly-49G2, Ly-49H, and Ly-49D is known in B6 mice, NK cell cluster distribution is altered in MHC-I–deficient mice. Taken together, our data support the idea that NK cell clusters develop in a directed manner and provide additional evidence that the regulatory system that controls the expression of both Ly-49 activating and inhibitory receptors is nonrandom. Our findings lead to an expanded model of NK cell receptor acquisition during NK maturation.

C57B6/J, B6.SJL-Ptprca Pepcb/BoyJ, and B6.129P2-B2mtm1Unc/J β-2 microglobulin (β₂m^−/−) knockout mice were obtained from The Jackson Laboratory and bred at the University of California, Merced. Mice of both sexes between the ages of 10 and 28 wk were used for each experiment. We observed no significant differences between mice of both sexes or of different ages, except for one specific study on cell proliferation (please see 10Results). Mice were housed in specific pathogen-free cages with autoclaved feed. Mice were euthanized by carbon dioxide asphyxiation, followed by cervical dislocation. All animal procedures were approved by the University of California, Merced Institutional Animal Care and Use Committee.

Splenic cells were harvested, processed, and stained for flow cytometric analysis (FACS) as described (18). Cells were stained with the following Abs, purchased from eBioscience, BioLegend, Miltenyi Biotec, and BD Biosciences: PE/Cy5-CD3 (145-2C11), PE/Cy5-CD4 (RM4-5), PE/Cy5-CD8a (53-67), PE/Cy5-Gr1 (RB6-8C5), PE/Cy5-CD19 (6D5), PE- or BV650-CD27 (LG3A10), BV421-NK 1.1 (PK136), BV510-CD11b (M1/70), biotinylated or APC–Ly-49D (4E5), APC-Cy7–Ly-49G2 (4D11), FITC– or APC–Ly-49H (3D10), APC-VID770–Ly-49I (YLI-90), and APC-Ki67 (SolA15), CD45.1 (A20), CD45.2 (104), PE-NKG2A^B6 (16A11), and FITC–Ly-49C/I (5E6). BUV395–streptavidin was used to develop biotinylated Abs. Staining of all cells included a preincubation step with unconjugated anti-CD16/32 (clone 2.4G2 or clone 93) mAb to prevent nonspecific binding of mAbs to FcγR. Extracellular staining was performed as described (18) or with the Ab mixture diluted in Brilliant Stain Buffer (BD Horizon) using the manufacturer’s instructions. Intracellular staining was performed using the True-Nuclear Transcription Factor Buffer Set (BioLegend) per the manufacturer’s instructions. Single-color stains or CompBeads were used for setting compensations, and gates were determined by historical data in addition to fluorescent-minus-one control stains. Flow cytometric data were acquired on the BD LSR II flow analyzer, FACS Aria II flow sorter (Becton Dickinson), or ZE5 flow analyzer (Bio-Rad Laboratories). The experimental data were analyzed using FlowJo Software version 10.6.0 and 10.6.3 (Becton Dickinson).

NK cell clusters (cluster [C]1–C16) were determined by t-distributed stochastic neighbor embedding (tSNE) using FlowJo Software version 10.6.0 and analyzed with 3000 iterations, 50 perplexity, 2800 learning rate (η), and vantage point tree k-nearest neighbors algorithm (24, 25). The cluster frequencies were determined by manual gating on live, lineage-negative NK1.1⁺ cells and further divided by gating on populations based on Ly-49I, Ly-49G2, Ly-49H, and Ly-49D expression (please see Fig. 1E). Similar analysis was performed in subsequent studies to add Ly-49C and NKG2A (see Supplemental Figs. 3–5).

Ly-49I, Ly-49G2, Ly-49H, and Ly-49D receptors were assessed for independent expression by the product rule. The product rule predicts the frequencies of NK cells that express single or a combination of Ly-49 receptors. Assuming a model of independent expression of Ly-49 receptors, the frequencies of NK cells expressing individual Ly-49 receptors were used to calculate expected frequencies for NK cells expressing zero, one, and more receptors (10, 22, 23). For example, the expected frequency for NK cells that coexpress Ly-49H and Ly-49D, but do not express Ly-49G2, equals the probability (P) of expressing Ly-49H (e.g., frequency of NK cells expressing Ly-49H) multiplied by the probability of expressing Ly-49D, multiplied by the difference of 1 minus the probability of expressing Ly-49G2 (to account for its exclusion), such that P(H,D) = P(H)*P(D)*(1 − P(G2)) (please see Fig. 2A). We calculated the product rule error between the observed and predicted frequencies as the log₂(observed frequencies/predicted frequencies). If the observed frequencies were greater than the predicted frequencies (product rule error > 0), this indicated the model underestimated the observed frequencies. Alternatively, if the observed frequencies were less than predicted frequencies (product rule error < 0), this indicated the model overestimated the observed frequencies. If the error equaled 0, then we concluded that the expression of the pattern of receptors were independent. To calculate the total error, we summed the absolute value of the product rule errors among clusters.

Splenic NK cells from C57B6/J mice were harvested and processed to a single-cell suspension in media (RPMI 1640 media supplemented with 10% FBS, 0.09 mM nonessential amino acids, 2 mM l-glutamine, 1 mM sodium pyruvate, 100 U/ml penicillin, 100 mg of streptomycin, 0.025 mM 2-ME, and 0.01 M HEPES), and cell counts were determined using a hemocytometer. Splenic cells were enriched for NK cells as described in the EasySep Positive Selection Kit (STEMCELL Technologies) manufacturer’s suggested protocol by selecting with an anti-lineage mixture (anti-CD3, -CD4, -CD8, -CD19, -Gr1, and -Ter119). NK cell clusters were then sorted on the FACSAria II by gating on propidium iodide–negative (live), lineage-negative, and NK1.1-positive cells. A total of 90–95% postsort purity was achieved, as measured by FACS. A total of ∼20,000 cells per NK cell cluster were cultured in 96-well flat-bottom tissue culture plates with the addition of 100,000 lethally irradiated (30.0 Gy) splenic feeder cells and recombinant mouse (rm)IL-15 (75 ng/ml). NK cell cluster frequencies were quantified by FACS after 4 d of culture.

B6.SJL mice were used as a source of donor splenic NK cells. Spleens were aseptically harvested and processed to a single-cell suspension. NK cells were enriched as described above and then sorted for each of the 16 NK cell clusters. B6 recipient mice were sublethally irradiated (5.5 Gy) with a cesium irradiator 4 h before receiving a range of 50,000–200,000 donor NK cluster cells by retro-orbital i.v. injection. Donor-derived CD45.1⁺ NK cell cluster frequencies were analyzed by FACS 4 d after transfer.

Student t test with a two-tailed distribution and with two-sample equal variance (homoscedastic test) was used to determine differences in means between groups using GraphPad Prism software version 8.4.2. A p value of 0.05 was considered to be statistically significant. Asterisks indicate statistically significant differences: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Diverse frequencies and combinations of Ly-49 activating and inhibitory surface receptors on NK cells exist, but few studies have examined the relationship between expression of the activating receptor repertoire to that of the inhibitory receptor repertoire and whether the activating receptors are subject to similar stochasticity in surface protein expression as previously described for inhibitory receptor genes (10, 22). We decided to focus on inhibitory receptors Ly-49I and Ly-49G2 and activating receptors Ly-49H and Ly-49D on splenic NK cells in C57B6/J (B6) mice, which express MHC-I ligands H-2K^b and H-2D^b. Ly-49I’s inhibitory ligand is H-2K^b, Ly-49H binds to the m157 viral Ag, and Ly-49G2 and Ly-49D’s known ligand is H-2D^d (not expressed in B6 mice) (11). Previous work (15, 19), in which analyses of Ly-49H and Ly-49D were performed, provided a source of historical controls to which our data could be compared. Although other Ly-49 family members exist, we were initially limited to these four receptors because of the availability of fluorochrome-conjugated specific Abs and the number of fluorochromes available on our flow cytometers. Expression of each Ly-49 receptor was observed (Fig. 1A, 1B). We defined 16 NK cell clusters by the number and type of activating receptors and inhibitory receptors expressed. That is, C1 expresses both Ly-49H and Ly-49D activating receptors and neither Ly-49I nor Ly-49G2 inhibitory receptors, whereas C16 expresses both Ly-49I and Ly-49G2 inhibitory receptors but neither Ly-49H nor Ly-49D activating receptors (Fig. 1C). By using these 4 receptors, 16 possible combinations of Ly-49 activating and inhibitory receptor expression were identified by tSNE analysis of flow cytometric data (Fig. 1F), and further examination of the expression of individual Ly-49 receptors allowed us to confirm each cluster phenotype (Fig. 1D) (24). Furthermore, we developed an NK cell cluster gating strategy that identified the 16 unique subpopulations (clusters) and their frequencies (Fig. 1E). Similar to previous studies that analyzed inhibitory Ly-49 receptors only (10, 22), we observed C8, which expresses none of the four receptors, to be one of the most prevalent populations (Fig. 1G). By analyzing Ly-49 activating receptor frequency, we observed that clusters coexpressing Ly-49H and Ly-49D activating receptors were most represented (i.e., C1 [I⁻G2⁻H⁺D⁺], C4 [I⁻G2⁺H⁺D⁺], C5 [I⁺G2⁻H⁺D⁺], and C9 [I⁺G2⁺H⁺D⁺]) (Fig. 1G). Conversely, frequencies were lowest for clusters that express both Ly-49 inhibitory receptors, such as C16 (I⁺G2⁺H⁻D⁻), C13 (I⁺G2⁺H⁻D⁺), and C12 (I⁺G2⁺H⁺D⁻) (Fig. 1G). These data suggest a possible selective (nonrandom) pressure controlling the frequencies of these clusters in mice.

Next, to explore the possibility of independent (random) expression of our Ly-49 receptors of interest, we used the product rule model for independent expression assuming the nonrandom nature of Ly-49 receptor surface expression. The model states that if Ly-49I, Ly-49G2, Ly-49H, and Ly-49D expression are independent of one another, then the observed frequencies of NK cells that express zero, one, two or more receptors can be predicted by the measurement of individual Ly-49 receptor frequencies (10, 23) (Fig. 2A). Thus, if the predicted frequencies match the observed biological frequencies, then we conclude no interdependencies between expression of those Ly-49 receptors. Alternatively, if the observed frequencies do not match the predicted frequencies, the expression of these receptors are interdependent (nonrandom) for reasons including, but not limited to, linked receptor expression, gene regulation, and biased receptor selection (10, 15, 26). Our comparison of the observed versus predicted NK cluster frequencies demonstrate independent expression in some clusters (C1, C5, C6, C9, C11, C12, C15, and C16) but dependencies in others (C2, C3, C4, C7, C8, C10, C13, and C14) (Fig. 2B). We next calculated the product rule error [e.g., log₂(observed/predicted)] for each cluster to identify if the model underestimated (observed > predicted), overestimated (observed < predicted), or matched the observed frequencies (observed = predicted). An example of the product rule errors in clusters arranged by their patterns of Ly-49H and Ly-49D expression is shown in Fig. 2C. We show that NK clusters C14, C8 (which are H⁻D⁻), and C4 (H⁺D⁺) frequencies were underestimated by the model, whereas clusters C10, C3 (both H⁺D⁻) and C7, C13, and C2 (H⁻D⁺) were overestimated by the model (Fig. 2C). We noted that the product rule underestimated the frequencies of five out of the eight clusters that express only one of the Ly-49 activating receptors (C10, C3, C7, C13, and C2). These data suggest interdependencies between Ly-49H and Ly-49D expression.

We hypothesized that a source of NK cell Ly-49 expression dependencies was the stage of maturation. To test our hypothesis, we used the NK cell maturation markers CD27 and CD11b to examine the frequencies of each cluster among immature NK (iNK; CD27⁺CD11b⁻), transitional NK (tNK; CD27⁺CD11b⁺), and mature NK (mNK; CD27⁻CD11b⁺) cells (1, 18, 27, 28) (Fig. 2D). We calculated the total error by summing the absolute values of the product rule errors for each NK cell maturation stage (Fig. 2E, 2H–J). The total error was significantly higher at the iNK cell stage relative to the tNK and mNK cell stages, suggesting more Ly-49 receptor dependencies at the iNK stage (Fig. 2E). iNK total error was significantly higher in clusters with at least one activating receptor (Fig. 2F). Examination of the inhibitory receptors showed very similar patterns (Fig. 2G). The exceptions were H⁻D⁻ clusters, which displayed similar error between iNK, tNK, and mNKs, and I⁻G2⁺ clusters, in which error was high in the iNK and mNK cells (Fig. 2F, 2G). Moreover, C8, C14, C15, and C16, all of which are H⁻D⁻, showed more independent expression (lower total error) in the iNK (1.99 ± 0.21) and tNK (2.20 ± 0.18) cells relative to I⁻G2⁻ C1, C2, C3, and C8 in the iNK (5.48 ± 0.45) and tNK (3.1 ± 0.10) cells (Fig. 2F, 2G). No difference between the total error for tNK and mNK cells within the H⁻D⁻ and I⁻G2⁻ groups were observed (Fig. 2F, 2G). Additionally, we observed the majority of clusters were either overestimated or underestimated consistently throughout NK cell maturation (Fig. 2H–J). There were some exceptions: C14 (I⁺G2⁻H⁻D⁻) was overestimated by the model at the iNK stage and then underestimated at the tNK and mNK maturation stage, and C6 (I⁻G2⁺H⁻D⁺) was overestimated at the iNK stage and then slightly underestimated at the mNK cell stage (Fig. 2H–J). Taken together, the product rule model provided evidence for a tightly regulated expression system for Ly-49 receptors, especially at the iNK maturation stage, with the highest interdependencies in clusters that express at least one Ly-49H and Ly-49D activating receptor.

Further analysis revealed Ly-49 cluster frequencies were unique at each NK cell maturation stage (Fig. 3A–D). We identified Ly-49I receptor expression frequency to significantly increase from iNK to tNK to mNK stages (Fig. 3E). Ly-49G2, Ly-49H, and Ly-49D frequencies significantly increased between iNK to tNK stages and then decreased (Ly-49G2 and Ly-49D) or stabilized (Ly-49H), moving from the tNK to mNK stages (Fig. 3E). The frequencies of clusters with zero or one receptor were found to decrease; conversely, clusters expressing two, three, or four receptors increased from iNK to tNK stages (Fig. 3F).

To determine if specific clusters were predominantly grouped at the iNK, tNK, or mNK cell stages of maturation, we quantified the cluster frequencies at each NK maturation stage and visualized the similarities between clusters with principal component analysis (PCA) (29). NK cell cluster percentages were quantified by normalizing each cluster frequency relative to each NK maturation stage such that a cluster’s frequency was divided by the sum of that cluster’s frequencies found at each maturation stage and then multiplied by 100 (Supplemental Fig. 1). We computed the PCA from the average normalized frequencies of clusters at each NK stage and determined similarities between clusters with respect to NK maturation (Fig. 3G–J). We grouped clusters found predominantly at the iNK, tNK, and mNK cell stage in our PCA, which were confirmed by observing the plotted percentages (Fig. 3H–J). That is, we identified C8, C15, C3, and C10 to predominate the iNK cell stage (Fig. 3H); C1, C2, C4, and C6 to predominate at the tNK cell stage (Fig. 3I); and C9, C5, C13, C16, C7, C11, and C14 to predominate at the mNK cell stage (Fig. 3J). C10, C1, and C2 assembled close together in the PCA, but we decided to group C10 into the iNK-predominant group because the changes in the percentage of C10 at the iNK, tNK, and mNK were more similar to C8, C3, and C15 (Supplemental Fig. 1). Notably, the predominant clusters within each stage were represented at different proportions as maturation progressed. For example, C8, which expresses none of the four receptors, decreased in a sequential manner throughout NK cell maturation from 80.7% ± 1.8 in the iNK stage to 6.6% ± 1.5 at the mNK cell stage (Fig. 3H, 3J). In contrast, tNK-predominant clusters were lower at the iNK and mNK stages, and mNK-predominant clusters were lowest at the iNK and tNK stages (Fig. 3H–J). Furthermore, all mNK-predominant clusters expressed the inhibitory self–Ly-49I receptor, but iNK- and tNK-predominant clusters did not (Figs. 1C, 3H–J) Thus, these findings suggest that the frequency and phenotype of the NK cell clusters are regulated throughout NK cell maturation to increase the types of receptors expressed and positively select for self–Ly-49I at the mNK cell maturation stage.

After completing our initial studies with Ly-49D, Ly-49G2, Ly-49H, and Ly-49I, we gained access to a new flow cytometer that permitted us to expand our analysis to include Ly-49C and NKG2A. Ly-49C is the major self-inhibitory receptor in B6 mice, and NKG2A binds Qa-1^b in B6 mice (3, 11, 30). We costained NK cells with Ab clone 5E6, which cross-reacts with both Ly-49C and Ly-49I, and with clone YLI-90, which solely binds to Ly-49I. This combination of 5E6 and YL1-90 identified Ly-49C⁺Ly-49I⁻ NK cells, but could not distinguish between Ly-49I⁺ and Ly-49I⁺ Ly-49C⁺ NK cells because of 5E6’s known cross-reactivity (Supplemental Fig. 3A, 3B). Despite this, we identified 48 distinct NK cell populations with different NK cell receptor surface expression patterns (Supplemental Fig. 3A, 3C) and distributions (Supplemental Fig. 3D). Similar to our initial analysis (Fig. 1G), we observed that populations coexpressing Ly-49H and Ly-49D activating receptors were highly represented (i.e., populations [Pops.] 2, 3, 5, 8, 9, 11, and 12) (Supplemental Fig. 3D). However, Pop. 44, Pop. 45, and Pop. 47, which do not express either Ly-49D or -H, were found in a similar range. Coexpression of NKG2A with Ly-49D and/or Ly-49H did not appear to influence population frequencies (Supplemental Fig. 3D).

To determine the interdependencies of NK cell receptors when NKG2A and Ly-49C were included, we used the product rule to evaluate the observed versus predicted frequencies (Supplemental Fig. 3E–J). We found that there were no significant dependencies among the receptors in Pops. 4, 5, 10, 13, 16, 19, 25, 28, 31, 37, and 40, as determined by no difference between the observed and predicted frequencies (Supplemental Fig. 3G, 3H). Interestingly, Ly-49C was expressed in 9 of these 11 populations. In contrast, interdependencies were found for the other 37 populations (Supplemental Fig. 3G, 3H). Next, we focused our investigation on the influences of NKG2A and Ly-49C on the other receptors. We grouped the populations that express NKG2A and found that the majority of the observed frequencies were less than the predicted values provided by the product rule, and these populations did not express Ly-49C, with the exception of Pop. 1 (Supplemental Fig. 3I). In contrast, grouping all the populations that express Ly-49C showed that observed frequencies were greater than the predicted values (i.e., observed > predicted), except for when Ly-49C is coexpressed only with Ly-49H (Pop. 34), in which observed values were less than predicted values (Supplemental Fig. 3J). Furthermore, Pop. 46 (only Ly-49C⁺) and Pop. 43 (Ly-49C⁺Ly-49G2⁺) demonstrated the highest positive errors, which may suggest that Ly-49C⁺Ly-49G2⁺ coexpression may have strong interdependencies and that Ly-49C expression alone may be favored in WT mice. These results suggest that NK cells regulate NKG2A and Ly-49C expression in different manners relative to each other.

We next examined the patterns of NKG2A and Ly-49C expression as a function of NK cell maturation stage in WT mice (Supplemental Figs. 4, 5F, 5G). NKG2A⁺ NK cell frequencies gradually decreased between iNK and mNK cell stages (Supplemental Fig. 5F). Conversely, Ly-49C⁺ NK cell frequencies increased in the tNK and mNK cell stages compared with iNK cells in WT mice (Supplemental Fig. 5G). Next, we analyzed the 48 populations in iNK, tNK, and mNK cells (Supplemental Fig. 4). Using the same analysis tools from Fig. 3G–J, we grouped the 48 populations by PCA. Distinct groupings of predominant iNK (i.e., Pops. 42, 48, 39, and 45), predominant tNK (i.e., Pops. 9, 12, 3, 6, 30, 27, 15, 21, 33, 18, 24, 40, 7, 36, 10, 4, 1, 28, 25, 16, 13, 19, 31, 37, and 34), and predominant mNK populations (i.e., Pops. 11, 8, 29, 5, 41, 2, 26, 35, 32, 23, 46, 38, 20, 17, 14, 22, 47, 44, and 43) were identified (Supplemental Fig. 4A, 4B). To determine if the previous 16 clusters (Fig. 1) mapped to the same predominant stages of maturation within the 48 populations, we matched the clusters’ expression profiles to the corresponding populations (i.e., C1: H⁺D⁺ and Pop. 12: H⁺D⁺) and overlaid the populations on the existing PCA plot (Supplemental Fig. 4C). We observed that most of the clusters grouped in a similar manner to our previous observations, with the exception of C3 and C10, which previously grouped in the iNK stage but now grouped in the tNK stage (Fig. 3G, Supplemental Fig. 4C). Furthermore, the predominant mNK populations continue to express the self–Ly-49I receptor in both analyses (Supplemental Fig. 4A–C, 4F). The Ly-49C–expressing populations were mainly grouped in the tNK cell stage of maturation (Supplemental Fig. 4F). Additionally, we identified novel populations that express NKG2A alone (Pop. 42), both NKG2A and Ly-49G2 (Pop. 39), and Ly-49G2 alone (Pop. 45) predominantly at the iNK stage. This suggests that expression of the NKG2A and Ly-49G2 receptors occurs first during NK cell development (Supplemental Fig. 4D). These studies further support prescribed pathways for Ly-49 and NKG2A receptors throughout NK cell maturation and that NK cells favor self–Ly-49I receptor expression at the mNK cell stage.

The observed distribution of clusters within the iNK, tNK, and mNK stages (Fig. 3, Supplemental Fig. 1) led us to hypothesize that the iNK-predominant clusters may be precursors to clusters that predominate at the tNK to mNK stages. To test this, we sorted each of the 16 NK cell clusters and cultured them with 75 ng/ml rmIL-15 and lethally irradiated splenic feeder cells for 4 d (Fig. 4A). We observed that iNK-predominant clusters C8, C15, C3, and C10 and tNK-predominant clusters C1, C2, C4, and C6 all upregulated Ly-49I receptor after culture (Fig. 4B–D), which would recategorize them into clusters found predominantly in the mNK cell stage (C5, C7, C9, and C13, respectively; Figs. 1C, 3J). Cultures initiated with mNK-predominant clusters maintained their Ly-49 receptor phenotypes (data not shown). To verify our findings in vivo, we adoptively transferred sorted C8, C15, C3, C10, C1, C2, and C14 cells from B6.SJL (CD45.1⁺) mice into sublethally irradiated B6 (CD45.2⁺) hosts and analyzed their differentiation after 4 d (Fig. 4E). Similar to our in vitro results, we observed C8, C15, C3, C10, C1, and C2 to upregulate Ly-49I (Fig. 4F–I) and that Ly-49H and Ly-49D maintained stable expression (data not shown). However, in vivo C8, C1, and C2 upregulated a small frequency of inhibitory Ly-49G2 receptor after 4 d (Fig. 4F, 4G). Ly-49 receptor expression on C14, an mNK-predominant cluster, was unchanged after adoptive transfer (Fig. 4H). Altogether, these results strongly support the existence of prescribed pathways of NK cell maturation from precursor NK cell clusters and that Ly-49I surface expression is a key step for mNK cells (Fig. 4J, 4K).

Given our discovery of prescribed pathways of NK cluster development, we further analyzed the clusters for differences in their proliferative state. We hypothesized that the proliferation rates of specific NK cell clusters would be distinct. Furthermore, we expected proliferation rates to correlate with maturation stage. To test our hypotheses, we sorted and cultured bulk iNKs, tNKs, and mNKs (Fig. 5A) on lethally irradiated splenic feeder cells in media containing 75 ng/ml of rmIL-15 and measured cellularity at day 2 and day 6 postculture. We observed the highest fold change in cellularity for cultures initiated with iNK cells, as compared with cultures initiated with tNK or mNK cells (Fig. 5B). We disaggregated the fold changes in proliferation between iNK-predominant (C8, C15, C3, and C10), tNK-predominant (C1, C2, C4, and C6), and mNK-predominant (C9, C5, C16, C12, C13, C14, C11, and C7) clusters within each sorted population (Fig. 5C–E). Our data show that iNK-initiated cultures increased the fold change in cellularity of tNK-predominant clusters relative to iNK- and mNK-predominant clusters (Fig. 5C). In contrast, mNK-initiated cultures showed reduced proliferation in tNK-predominant clusters relative to iNK- and mNK-predominant clusters (Fig. 5E). Additionally, the frequency of C8, which expresses none of the four Ly-49 receptors and is the most prevalent in the iNK stage, dramatically decreased during the culture period (Fig. 5F), which is consistent with our observation of decreased frequency of C8 at the tNK and mNK stages in vivo (Fig. 3I, 3J).

To confirm these proliferation patterns, we stained NK cells for Ki-67 expression postculture. After culture, we noticed that CD11b expression was downregulated universally on NK cells, preventing us from using CD11b to distinguish iNK, tNK, and mNKs (data not shown), likely as an artifact of in vitro culture (31). However, CD27 expression was still binary, allowing us to distinguish CD27⁺ (presumably iNK and tNKs) from CD27-mNKs. We found that NK1.1⁺CD27⁺ cells expressed higher Ki-67 levels compared with NK1.1⁺CD27⁻ cells (Fig. 5G, 5H), regardless of Ly-49 receptor expression (Fig. 5I). This pattern persisted when the Ki-67 expression was examined in specific NK cell clusters, with the exception of cluster C15 (which only expresses Ly-49G2) (Fig. 5J). These data show that cluster designation (and hence Ly-49 receptor expression) does not dictate NK cell proliferation. Rather, proliferative potential is a general characteristic of NK cell maturation stage, with highest proliferation in the iNK cells and lowest proliferation in mNK cells.

We next investigated the effects of MHC-I on NK cell cluster heterogeneity and NK cell maturation in β₂m^−/− (MHC-I^−/−) mice. In MHC-I^−/− spleens, the frequencies of NK cells expressing inhibitory Ly-49I and Ly-49G2 receptors increased, whereas the frequencies of NK cells expressing activating Ly-49H and Ly-49D receptors decreased (Fig. 6A). These differences in Ly-49 frequencies in MHC-I^−/− mice did not appear to influence NK cell maturation (Fig. 6B). MHC-I^−/− mice displayed significantly lower frequencies of I^–G2^– NK cells (Fig. 6C, 6D) and higher frequencies of H^–D^– NK cells (Fig. 6E, 6F). Thus, NK cells from MHC-I^−/− mice have decreased frequencies of activating receptors and increased frequencies of inhibitory receptors. The increased frequency of inhibitory receptor–expressing NK cells is due to an increase in dual inhibitory receptor–expressing I⁺G2⁺ NK cells, as lower frequencies of I^–G2⁺ NK cells were observed in MHC-I^−/− mice (Fig. 6C, 6D). Additionally, our activating receptor analysis showed that H⁺D⁻ NK cells increased, whereas H⁺D⁺ cell frequencies decreased in MHC-I^−/− mice (Fig. 6E, 6F). Further investigation of individual Ly-49 receptor frequencies as a function of iNK, tNK, and mNK maturation stages revealed different behaviors of the inhibitory and activating Ly-49 receptor behaviors in MHC-I^−/− mice (Fig. 6G, 6H). The frequencies of Ly-49I⁺, Ly-49G2⁺, and Ly-49C⁺ NK cells were significantly higher in MHC-I^−/− mice compared with MHC-I^+/+ (WT) controls at each stage of NK maturation (Fig. 6G, Supplemental Fig. 5C, 5D, 5G). In contrast, the frequencies of Ly-49H⁺ and Ly-49D⁺ NK cells were significantly decreased only at the tNK and mNK stages, but similar at the iNK stage (Fig. 6H). Moreover, NKG2A⁺ cells were significantly higher in MHC-I^−/− mice than WT mice only at the iNK cell stage (Supplemental Fig. 5F). Overall, these data suggest MHC-I molecules play a differential role in the expression of inhibitory and activating Ly-49 receptors and that MHC-I is not essential for progression from iNK, tNK, and mNK stages during maturation.

We next compared NK cell cluster Ly-49 receptor expression dependencies among WT and MHC-I^−/− mice using the product rule. Any deviations between the two conditions (WT and MHC-I^−/−) would indicate dependencies that are influenced by MHC-I expression. We compared the product rule errors for each cluster in WT and MHC-I^−/− mice and observed significantly different errors in 9 of the 16 clusters in MHC-I^−/− mice (Fig. 6I). C1, C2, C5, C6, and C9 were found to increased error (e.g., more dependencies) in MHC-I^−/− mice, and these clusters commonly express the activating Ly-49D receptor (Fig. 6I). In contrast, C10 (I⁻G2⁺H⁺D⁻), C12 (I⁺G2⁺H⁺D⁻), and C14 (I⁺G2⁻H⁻D⁻) in MHC-I^−/− mice decreased dependencies (error) relative to WT and do not express Ly-49D. Furthermore, C15 (I⁻G2⁺H⁻D⁻) was the only cluster that flipped directionality of the error from being overestimated in WT to underestimated in MHC-I^−/− mice. This suggest that Ly-49G2 is negatively regulated in MHC-I–sufficient microenvironments. Addition of NKG2A and Ly-49C into the analysis of MHC-I^−/− mice revealed that the populations with observed frequencies greater than predicted frequencies further increased in error for Pops. 13, 19, 25, 40, 41, and 42 in MHC-I^−/− mice relative to WT (Supplemental Fig. 5H). Five out the six populations express NKG2A, four of the six express Ly-49C, and three out of six coexpress NKG2A and Ly-49C (Supplemental Fig. 5H). These results suggest NKG2A and Ly-49C interdependencies exist in the MHC-I^−/− mice.

Next, we wanted to determine MHC-I’s influence on NK cell cluster distributions. First, we examined NK cell frequencies based on the number of Ly-49 receptors expressed, regardless of receptor type. Collectively, clusters that expressed only one receptor were increased at the iNK cell stage (Supplemental Fig. 2A), but this increase was attributed to an increase in C15 only (Fig. 6J). The frequencies of clusters that express two receptors were collectively lower at all stages of NK cell maturation in MHC-I^−/− mice (Supplemental Fig. 2B), which was attributed to decreases in C1, C6, C7, and C11 clusters. C10 and C16, which also expressed two receptors, were increased in MHC-I^−/− mice (Fig. 6J). The overall increase in NK cells that express three receptors in MHC-I^−/− mice was attributed to C12 (Fig. 6J, Supplemental Fig. 2C). All of the clusters that were increased (C15, C10, C16, and C12) express Ly-49G2 (Fig. 6J). Furthermore, C3, C1, C2, C5, C7, and C11, which lack Ly-49G2 expression, were significantly decreased in NK cell frequencies (Fig. 6J). These data suggest that MHC-I is a major regulator of Ly-49G2 expression despite the fact that no known MHC-I ligand for Ly-49G2 in B6 mice has been described (3). In addition, Ly-49G2–positive clusters C15, C10, C16, and C12 all lack Ly-49D expression and were increased. In contrast, the Ly-49G2–negative clusters C1, C2, C5, and C7, which express Ly-49D, were decreased (Fig. 6J). This suggests that there is a reciprocal dependency between Ly-49D and Ly-49G2 expression and MHC-I in the distribution of cluster frequencies. No common relationship between MHC-I deficiency and Ly-49I or Ly-49H on cluster distributions could be determined with our cluster frequency dataset (Fig. 6J).

We continued to investigate the influence of MHC-I on NK cell cluster frequencies found predominantly within the iNK, tNK, and mNK cell stage of maturation. Similar to the previous analysis in which we determined the predominant clusters within NK maturation stages (Fig. 3), we compared MHC-I^−/− and WT (MHC-I^+/+) clusters using PCA. We observed that most clusters in MHC-I^−/− mice matched the same trends in cluster maturation found in WT mice (Supplemental Fig. 2D); however, C1, C2, and C16 maintained steady frequencies throughout maturation in MHC-I^−/− mice (Supplemental Fig. 2F, 2G). In WT mice, C15, C10, and C3 iNK-predominant clusters transition into C16, C12, and C11 (Fig. 4). In MHC-I^−/− mice, we observed increased C15 and C10 frequencies, matching the observed frequencies for C16 and C12. Furthermore, we observed decreased C3 and C11 frequencies (Supplemental Fig. 2E, 2G). Additionally, tNK-predominant clusters C2 and C1 were both decreased in MHC-I^−/− mice, and their subsequent mNK clusters C7 and C5 were also both decreased (Fig. 6J). Although tNK C4 was decreased, its subsequent cluster, C9, was found in normal frequencies at the mNK stage. C6 to C13 frequencies were unaffected. Notably, C4, C9, C6, and C13 all coexpress Ly-49G2 and Ly-49D. Overall, these data lend further support to the prescribed pathways of NK cluster differentiation and suggest that MHC-I influences the cluster frequencies via regulation of Ly-49G2 and Ly-49D expression.

Previously, the product rule has been used to determine the interdependencies of inhibitory Ly-49 receptors and their respective MHC-I ligand (10, 23). In this study, we extended this analysis to include the Ly-49I, Ly-49C, Ly-49G2, and NKG2A inhibitory receptors in combination with activating receptors Ly-49H and Ly-49D, as well as a careful study of Ly-49 receptor clusters based on NK cell maturation. In this study, we show for the first time, to our knowledge, that the majority of the interdependencies in Ly-49 receptor expression originate at the iNK cell maturation stage (10, 22) and that specific clusters predominate at each stage. Our results demonstrate strong interdependencies of the activating Ly-49H and Ly-49D receptors that we propose can explain the altered cluster frequencies observed in MHC-I^−/− mice. Moreover, our results further resolve a role for Ly-49I as an important selective marker of completion of NK cell development in B6 mice (27).

Ly-49I and Ly-49C known ligand is MHC-I K^b, and NKG2A binds Qa-1^b in B6 mice (3, 11, 30). Our NK cluster development studies suggest a process in which Ly-49I–negative iNK and tNK clusters develop with high proliferative potential. We propose that the surface expression of Ly-49I transitions NK cells into the mNK cell stage, and once the interactions between Ly-49I and K^b occur, proliferation decreases. Furthermore, we observed NK cell surface NKG2A to decrease and Ly-49C to increase receptor frequency from iNK to mNK cell stage of maturation (Supplemental Fig. 5F, 5G). Moreover, NK cells in the β₂m^−/− mice were observed to increase the frequencies of all NK populations that express Ly-49C receptors (Supplemental Fig. 5G). However, given that the fluorescence intensity of staining of Ly-49C/I is visibly higher in β₂m^−/− mice than WT B6 mice (Supplemental Fig. 5C, 5D), the Ly-49C frequency increase we observed in β₂m^−/− mice may be attributed to the lack of engagement to host MHC-I molecules (32), and we cannot exclude the possibility that the frequency of Ly-49C⁺I⁺ and Ly-49C⁺I⁻ NK cells in WT B6 mice is underestimated. Zhang et al. (33) have shown that the synergized role between NKG2A and inhibitory Ly-49 expression are required for NK cell education and to properly mediate missing-self and induced-self recognition. Thus, the education process between NKG2A and Ly-49C and Ly-49I may be required at alternate stages in NK cell maturation (33). Moreover, these observations are most consistent with the sequential expression model, which proposes splenic NK cells sequentially accumulate inhibitory Ly-49 receptors until receptors specific for self–MHC-I molecules are expressed (22, 34, 35). We observed that NK cells that express NKG2A⁺, NKG2A⁺Ly-49G2⁺, and Ly-49G2⁺ are expressed at the highest frequencies at the iNK cell stage of maturation (Supplemental Fig. 4D) and might suggest that NKG2A and Ly-49G2 are the first receptors expressed on NK cells during development. We also observed that Ly-49I⁺ clusters become more prevalent at the mNK cell stage, indicating that NK cells also successively accumulate activating and non–self-binding Ly-49 receptors before expressing Ly-49I. Consistent with previous findings, we observed decreased mNK cell proliferation relative to iNK and tNK cells (2), which may suggest that self-inhibitory Ly-49I completes the maturation process and maintains mNK cells in a quiescent state ready to be triggered in an immune response.

Sternberg-Simon et al. (10) show that the inhibitory receptor Ly-49C expression is overrepresented and Ly-49I is underrepresented when they analyzed 32 NK cell populations using the product rule in WT mice. Because we did not have access to the Ly-49C–specific clone 4LO331 (10, 36–38), our results using FITC–Ly-49C/I (clone 5E6) and APC-VID770–Ly-49I (clone YLI-90) were able to identify Ly-49C–positive populations, but could not distinguish between Ly-49I⁺ and Ly-49I⁺ Ly-49C⁺ NK cells because of 5E6’s known cross-reactivity (Supplemental Fig. 3A, 3B). Thus, our analysis was limited to 48 possible populations rather than the 64 possible populations. Additional studies are required to clarify this issue; however, our results using 5E6 and YLI-90 show that the majority of Ly-49C population observed frequencies were greater than predicted frequencies (i.e., overrepresented), which were similar to the findings of Sternberg-Simon et al. (10). Moreover, we found that Ly-49C was expressed during the predominant tNK and mNK maturation stages and that the Ly-49C⁺ (Pop. 46) population grouped in the predominant mNK stage with the highest frequency compared with the other Ly-49C–expressing populations (Supplemental Fig. 4F). Furthermore, our analysis of MHC^−/− mice suggest that the populations that increased the most in error in MHC^−/− mice were those that express NKG2A⁺ (Pop. 42), NKG2A⁺Ly-49I⁺ (Pop. 41), and NKG2A⁺Ly-49C⁺ (Pop. 40) (Supplemental Fig. 5H). Overall, the acquisition and regulations of self-inhibitory Ly-49s and NKG2A receptors may be attributed to an NK cell education process, in particular the expression of NKG2A, Ly-49C, and Ly-49I in C57BL/6 mice (33, 39). These studies suggest that NK cells regulate the expression of Ly-49s and NKG2A expression in a manner that is controlled by MHC-I.

Our data expand on the sequential expression model and suggest an updated working model in which MHC-I affects NK cell Ly-49 activating and inhibitory receptor expression and alters the prescribed pathways of NK cluster differentiation but via distinct mechanisms. Our model distinguishes between the differentiation of clusters expressing only inhibitory (Fig. 6K, 6L) or only activating receptors at an early maturation stage (Fig. 6M, 6N). In this working model, we focus on the Ly-49 receptor interactions at the iNK and tNK stages of maturation that result in expression of Ly-49I to complete NK cell maturation. In Fig. 6K, in the predominant iNK cluster that expresses inhibitory Ly-49G2, but no activating receptors (C15), Ly-49G2 inhibitory signals are not initiated (because there is no Ly-49G2 ligand in B6 mice). Because of this lack of inhibitory signal, developing NK cells then upregulate inhibitory receptor Ly-49I (differentiating to C16). Ly-49I binds to its H-2K^b ligand, and in turn, C16 NK cell development is completed and then sustained (Fig. 6L). The increased frequencies of C15 and C16 observed in MHC-I^−/− mice (Fig. 6I) can be explained by this model (Fig. 6L). In MHC-I^−/− mice, the C15 cluster upregulates Ly-49I in the same fashion as in WT mice. However, Ly-49I inhibitory signals are not generated (because H-2K^b is not present), and the developing C15 cluster continues to expand (Fig. 6L). Extending this working model to clusters that coexpress a single activating Ly-49 receptor with Ly-49G2 suggests that Ly-49H and Ly-49D regulate cluster frequencies differently, as C10 (G2⁺H⁺D⁻I⁻) and its mature counterpart C12 (G2⁺H⁺D⁻I⁺) are both increased in MHC-I^−/− mice, whereas frequencies of C6 (G2⁺D⁺H⁻I⁻) and its counterpart C13 (G2⁺D⁺H⁻I⁺) are unaffected. This suggests that the Ly-49H⁺ clusters are controlled by presence of MHC-I. However, C4, which expresses both activating receptors (G2⁺H⁺D⁺I⁻), was lower in frequency in MHC-I^−/− mice, but frequencies of its counterpart C9 (G2⁺H⁺D⁺I⁺) were normal. This indicates that the roles of Ly-49H and Ly-49D during NK cell development are complex, and MHC-I is not essential for some of these roles.

Our accompanying working model of NK cell development starting with clusters expressing only activating receptors at the iNK and tNK stages is shown in Fig. 6M. The known ligand for Ly-49H is m157, a mouse CMV MHC-like protein (2, 40), whereas the known ligand for Ly-49D is H-2D^d. No known self–MHC-I ligand has been identified for Ly-49H and Ly-49D in B6 mice (41), but it is possible weak binding to self–MHC-I or non–MHC-I ligands exists (42, 43). Freund et al. (15) reported that activation signals via SLP-76 upregulates inhibitory Ly-49A, Ly-49G2, and Ly-49I receptor expression. Similarly, we presume that in B6 mice, an activating ligand exists for Ly-49H and Ly-49D. NK clusters with one activating receptor (e.g., C2 and C3) will generate a signal in response to this activating ligand, which promotes its differentiation and expansion. The activating signal also results in expression of inhibitory Ly-49I at the mNK stage (Fig. 6M). We interpret the clear effects of MHC-I deficiency on Ly-49H and Ly-49D frequencies to demonstrate a relationship between MHC-I and the activating receptors, but this is not a direct ligand–receptor interaction. NK cell development and NK cell survival in MHC-I–deficient mice may be impaired by dysfunctional dendritic cell expression of IL-12 and IL-15 transpresentation (44, 45). In MHC-I^−/− mice, we assume the activating ligand is absent or signaling is impaired. In the absence of these signals, low expansion and impaired differentiation to the corresponding mNK cluster in MHC-I^−/− mice results. Lack of activating signals in MHC-I^−/− also dysregulates the expression of Ly-49I at the mNK stage (Fig. 6N). We observed an enhanced decrease in cluster frequencies in MHC-I^−/− mice when two activating receptors were expressed (C1 and C5), which we posit could result from lower expansion of dual H⁺D⁺ clusters (Fig. 6J). However, C9, which is H⁺D⁺ and coexpresses both inhibitory receptors, is unaffected in the MHC-I^−/−, suggesting a canceling out or balancing of activating and inhibitory signals in this case. This balancing is further supported by the similar frequencies of C6 and its mNK counterpart C13, which express one inhibitory and activating receptor, in B6 and MHC-I^−/− mice (Fig. 6J).

In our study, we focused on the role of Ly-49 receptors on NK cell development. However, given that NK cytotoxicity is governed by balance of signals between Ly-49 activating and inhibitory receptors (3), it is also possible that our evidence of prescribed NK cell developmental pathways can be applied to the identification of NK cell clusters with high cytotoxic potential. Transcription of cytotoxicity genes increases with NK maturation with the highest cytotoxic gene expression at the mNK stage (27, 28). NK cell education via the expression of Ly-49I and Ly-49C binding to K^b is consistent with our observations that mNK cells acquire Ly-49I and Ly-49C expression in a manner that correlates to NK cell maturation. These results further suggest that self-inhibitory Ly-49 receptors may acquire functional potential at different stages of NK cell maturation, which may be driven by NK cell education.

Although we have focused on splenic NK cells in our work, it is also noteworthy that NK cell maturation could be tissue specific and may adapt to specific perturbation, which may alter the NK cell Ly-49 cluster pathways we observed in the spleen (27, 46). Although NK cells originate in the bone marrow, they continue to mature in peripheral tissues. These tissues have been shown to express different frequencies of iNK, tNK, and mNK cells (27, 47). Additionally, Barao et al. (48) show Ly-49G2–positive NK cells as the first responders after hematopoietic stem cell transplants and infection in B6 mice. Alternatively, Williams et al. (49) show that NK1.1⁻IL-2/IL-15R β⁺ precursors can generate NK cells that first express NKG2A, NKG2C, and Ly-49B, then Ly-49G2 and Ly-49I, and finally Ly-49A, Ly-49D, Ly-49E, and Ly-49F receptors. Our results also confirm that NKG2A and Ly-49G2 expression is found early in NK cell development (Supplemental Fig. 4D). These findings, along with ours, support that NK Ly-49 receptors may be acquired successively and in receptor combination during development and may adapt during perturbation. Additional studies are necessary to determine if the prescribed pathways of NK development we observed in the spleen are conserved in other tissues and regulated during perturbations.

We thank the staff of the Department of Animal Research Services and Dr. David Gravano from the Flow Cytometry Core of the Stem Cell Instrumentation Foundry at the University of California, Merced for excellent animal care and technical support. We thank Dr. Kirk Jensen for the gift of rmIL-15 and Dr. Anna Beaudin and Dr. Lewis Lanier for sharing flow cytometry reagents. We are also grateful to Drs. Marcos E. García-Ojeda and Katrina K. Hoyer, as well as the University of California, Merced Immunology Journal Club for comments on the manuscript.

The authors have no financial conflicts of interest.