NK cells obtained by exposing mouse fetal thymocytes to appropriate combinations of IL-4, IL-2, and PMA are phenotypically indistinguishable from cultured adult splenic NK cells with the exception that they generally lack measurable expression of all of the inhibitory Ly49 molecules that can currently be detected with Abs (Ly49A, -C, -G, and -I) and of the activating molecule Ly49D. Despite this deficiency, fetal NK cells have a similar specificity to Ly49-expressing adult splenic NK cells. Individual fetal NK cell clones display an essentially invariant and broad specificity similar to that of polyclonal populations of fetal or adult NK cells, although significant differences in the fine specificity of clones can occasionally be detected. Most remarkably, cloned fetal NK cell lines display heterogeneous expression of a restricted set of surface molecules that includes 10A7, Ly6C, 3C2, CD8, certain isoforms of CD45, and also, occasionally, Ly49 molecules. This heterogeneity is not related to the cell cycle or activation status of the cells, and micromanipulation recloning demonstrates unambiguously that it is not due to a lack of a single cell origin. Diversity is generated rapidly and the capacity for diversification appears to persist indefinitely in vitro. The expression of individual variable Ags is independent and stochastic, resulting in fetal NK “clones” being potentially composed of hundreds of phenotypically distinct cells. We hypothesize that fetal NK cells behave as progenitor cells that are undergoing a process of rapid, extensive, and continuous diversification and that are individually capable of generating and regenerating a complex NK cell repertoire.

Although the existence of NK cells has been known for more than 20 years, many of the basic issues concerning these cells, such as their developmental pathway, their mechanisms of recognition, and their physiologic function, are still poorly understood. Recently, a significant advance in our understanding of NK cell recognition has come from the realization that MHC class I molecules can protect target cells from lysis by NK cells (1). This protection is mediated at least in part by receptors on NK cells that bind to class I molecules and deliver inhibitory signals (2, 3). In the mouse, the inhibitory receptors that have been identified belong to the Ly49 family of C-type lectin proteins encoded by genes in the NK complex (4). Different members of the Ly49 family bind to different class I molecules and are expressed independently on overlapping subpopulations of NK cells (5). Both the frequency of cells expressing a given Ly49 molecule and the level of expression are influenced by host MHC genes (6, 7, 8). The mechanisms involved in the creation of this Ly49 repertoire are unclear, but one explanation is that at some stage during development NK cells randomly acquire Ly49 molecules, creating a diverse repertoire that is then acted upon by negative and positive selection events to create a self-tolerant population of mature NK cells whose Ly49 expression is calibrated to be optimally sensitive to changes in target cell class I expression (9, 10). A problem with this concept, and with negative signaling in general, is that since the level of class I expression varies substantially on different cells in the body there would appear to be a continuous need to adjust the level at which self-nonself discrimination is set in individual NK cells.

Even less is understood about the mechanisms of positive recognition used by NK cells. Although most Ly49 molecules deliver inhibitory signals as judged by the presence of ITIMs3 in their cytoplasmic domains, at least two members of the Ly49 family, Ly49D and Ly49H, lack ITIMs (11). Ly49D is expressed on a subpopulation of NK cells and may convey activating signals (12). In the rat, there is good evidence that another member of the C-type lectin family, NKRP1, is involved in positive recognition (13, 14). In the mouse, the NK1.1 Ag, which is a homologue of NKRP1 (15), is expressed on all NK cells in appropriate strains (16), is encoded by genes in the NK complex (4), and may be involved in recognition as judged by the ability of Ab to NK1.1 to induce or enhance killing by NK cells (17). Recent studies have indicated that the 10A7 Ab, which reacts with most but not all mouse NK cells (18), binds to one or more members of the mouse NKRP1 family (J. C. Ryan and W. E. Seaman, personal communication) and may therefore delineate a subpopulation of NK cells with distinct recognition capabilities. An alternative view is that NK cell recognition may be mediated, at least in part, by intercellular recognition molecules expressed more widely in the immune system (19). Candidates include CD2 (20), CD28 (21, 22), and CD40 ligand (23). Another possible candidate is CD45, which exists in multiple isoforms having differential cellular expression (24) and has frequently been implicated in NK cell recognition (25, 26, 27, 28, 29, 30, 31, 32).

The understanding of NK cell recognition in the human has been greatly facilitated by studies at the clonal level. By contrast, for unknown reasons, it has proven extremely difficult to clone mouse NK cells. However, some while ago we found that by using appropriate combinations of IL-4, IL-2, and PMA it was possible to generate long-lived lines of NK cells from the thymus of embryonic mice, permitting for the first time the cloning of mouse NK cells (33). We report here a detailed analysis of the phenotype and specificity of such clones, the results of which would appear to have important implications for our understanding of NK cell recognition and development.

Spleen cells obtained from C57BL/6 mice were treated with anti-CD4 + anti-CD8 and C′, washed, and adjusted to approximately 2 × 106/ml in DMEM containing 2× nonessential amino acids and 10% FBS (D10F). Cells were cultured with IL-2 at 104 U/ml in 2-ml aliquots in 24-well plates and were harvested after 3 to 14 days.

These were generated and maintained as described previously (33). Briefly, thymocytes were prepared from day 14 embryos of timed-mated mice, cultured for 1 to 3 days in D10F containing 10 U/ml IL-4 and 10 ng/ml PMA, then transferred to D10F containing 104 U/ml IL-2. Clones were derived at this stage by cloning by limiting dilution or micromanipulation in D10F containing 104 U/ml IL-2, 1 U/ml IL-4, and 10 ng/ml PMA, followed by subsequent maintenance in D10F containing 104 U/ml IL-2 alone or 104 U/ml IL-2 + 10 ng/ml PMA.

Direct and indirect immunofluorescence staining was performed by standard methods. The following mAbs were purchased from PharMingen, San Diego, CA: phycoerythrin (PE)-53.6.7 anti-CD8α, PE-53.5.8 anti-CD8β, and FITC-CT1. The following mAbs or hybridomas were kindly donated by colleagues: KT3 anti-CD3ε (K. Tomonari, Harrow, U.K.), PK136 anti-NK1.1 (G. Koo, Merck, Rahway, NJ), A1 anti-Ly49A (W. Seaman, Veterans Administration Medical Center, San Francisco, CA), 5E6 anti-Ly49C/I and 10A7 (V. Kumar, Southwestern Medical Center, Dallas, TX), 4D11 anti-Ly49G (L. Mason, National Institutes of Health, Frederick, MD), 4E5 anti-Ly49D (J. Ortaldo, National Institutes of Health, Frederick, MD), HK1.4 anti-Ly6C (F. Fitch, University of Chicago, Chicago, IL), 3C2 (S. Pollack, University of Washington, Seattle, WA), M1/93 pan-CD45 (I. Trowbridge, Scripps Clinic, San Diego, CA), and RA3–2C2 anti-CD45RA and CT2 (L. Lefrancois, University of Connecticut, Storrs, CT). The following hybridomas were obtained from the American Type Culture Collection (ATCC), Rockville, MD: MB23G2 anti-CD45RB and I24-D6 anti-CD45RC. Staining was analyzed on a FACScan (Becton Dickinson, San Jose, CA), using forward and side scatter to gate on single viable cells.

Cytotoxicity assays were performed using a standard 4-h chromium release assay, with 5000 target cells/well. Effector cells were titrated over a wide range, and lytic activity was calculated as LU/103 effector cells from the slopes of the regression lines fitted to the linear portions of the dose-response curves, as described elsewhere (34). All of the adult and fetal NK cell lines and clones used in this study displayed very high levels of cytotoxicity, similar to those reported previously (33, 35). All assays included the Moloney virus-induced T cell lymphoma of A/Sn mice, YAC-1, as an internal control. Killing of this target invariably exceeded 25% specific lysis at the highest effector cell dose used. To simplifiy data and allow the averaging and statistical analysis of data from repeat experiments, the lytic activity on other targets was expressed as a percentage of that on YAC-1. The other target cells used were: C1498, a spontaneous lymphoma of C57 mice; MBL2, a B cell lymphoma of C57 mice; EL4, a benzpyrene-induced thymoma of C57 mice; RMA, a subline of the RBL5 Rauscher virus-induced T cell lymphoma of C57 mice; A20, a spontaneous B cell lymphoma of BALB/c mice; WR19L, an A-MuLV-induced lymphoma of BALB/c mice; J774, a spontaneous myeloid tumor of BALB/c mice; IC2, a NK-sensitive subline, and av, an NK-resistant subline of the L5178Y methylcholanthrene-induced thymoma of DBA/2 mice; L1210, a methylcholanthrene-induced leukemia of DBA/2 mice; LBRM, a radiation-induced T cell lymphoma of B10.BR mice; R1.1, a spontaneous thymoma of C58 mice; SL8, a spontaneous thymoma of AKR mice; and Y3, a myeloma of Lou/C rats. These lines were kindly provided by colleagues or were obtained from the ATCC.

Exposure of fetal thymocytes to appropriate combinations of IL-4, IL-2, and PMA results in the vigorous outgrowth of cells that are phenotypically and functionally indistinguishable from adult NK cells with the principal exception that they normally lack detectable expression of Ly49A, -C, -G, and -I (33, 35). The data presented in Figure 1 confirm and extend these findings, showing that C57BL/6 fetal NK cells that are uniformly negative for the expression of CD3 and uniformly positive for the expression of NK1.1 not only lack detectable expression of the above-mentioned inhibitory Ly49 molecules but also of the activating Ly49 molecule, Ly49D. This was true for both polyclonal C57BL/6 fetal NK cells and for a range of cloned C57BL/6 fetal NK cell lines that were examined.

FIGURE 1.

Expression of CD3, NK1.1, and Ly49 molecules on polyclonal and monoclonal populations of C57BL/6 NK cells.

FIGURE 1.

Expression of CD3, NK1.1, and Ly49 molecules on polyclonal and monoclonal populations of C57BL/6 NK cells.

Close modal

Ly49 molecules through their recognition of class I molecules and the subsequent delivery of inhibitory and perhaps also activating signals are thought to play an important role in determining the lytic specificity of NK cells. Since C57BL/6 fetal NK cells lack expression of all of the Ly49 molecules that can be detected by currently available mAbs, it was of interest to compare their lytic specificity with that of Ly49-expressing adult NK cells. To do so, we used a quantitiative cytotoxicity assay in which cytolytic activity was calculated as LU/103 effector cells from the linear portion of the dose-response curve (34). This ensures that killing is not being measured under “saturating” conditions and gives the most sensitive and accurate assessment possible of the differences in susceptibility of different target cells. The results summarized in Figure 2 show that the specificity of polyclonal fetal NK cells is quite similar to that of polyclonal adult NK cells. There was perhaps some tendency for fetal NK cells to have a broader specificity than adult NK cells, as shown by their higher relative killing of MBL2, J774, IC2, av, R1.1, and SL8. However, none of these differences between adult and fetal NK cells was statistically significant. Somewhat surprisingly, therefore, it would appear that the absence of several inhibitory and at least one activating Ly49 molecule has little impact on the specificity of NK lysis, at least under the assay conditions used.

FIGURE 2.

Comparative specificity of C57BL/6 polyclonal adult and fetal NK cells. The killing of each target is expressed as a percentage of that of YAC-1 cells. The results are the mean values of between 2 and 11 experiments with each target/effector combination. The target cells are arranged according to their MHC haplotype and species origin.

FIGURE 2.

Comparative specificity of C57BL/6 polyclonal adult and fetal NK cells. The killing of each target is expressed as a percentage of that of YAC-1 cells. The results are the mean values of between 2 and 11 experiments with each target/effector combination. The target cells are arranged according to their MHC haplotype and species origin.

Close modal

Two explanations have generally been put forward for the broad lytic specificity of NK cell populations: that individual NK cells themselves have a broad and essentially invariant specificity; or that the specificity of polyclonal populations of NK cells represents the summation of that of individual cells having restricted but clonally variable specificity. The availability of clones of fetal NK cells would seem to provide an opportunity to test these alternative theories. We first examined the ability of a large number of C57BL/6 clones to kill three relatively NK sensitive targets, YAC-1, A20, and P815. The important result, illustrated in Figure 3, is that all 14 fetal NK clones and subclones examined killed all three targets. Some differences in fine specificity were observed. For example, clone 922 gave higher killing of A20 than clone 773.2 (p = 2 × 10−4). However, it is clear that, at least on this set of target cells, individual fetal NK clones did not differ radically in specificity.

FIGURE 3.

Lysis of YAC-1, A20, and P815 targets by polyclonal adult NK cells, two polyclonal fetal NK lines (1412 and 773), and 14 clones and subclones of fetal NK cells, all of C57BL/6 origin. The killing of each target is expressed as a percentage of that of YAC-1 cells. The number of experiments (n) performed with each line and clone is shown.

FIGURE 3.

Lysis of YAC-1, A20, and P815 targets by polyclonal adult NK cells, two polyclonal fetal NK lines (1412 and 773), and 14 clones and subclones of fetal NK cells, all of C57BL/6 origin. The killing of each target is expressed as a percentage of that of YAC-1 cells. The number of experiments (n) performed with each line and clone is shown.

Close modal

To examine the specificity of individual NK clones in more detail, a further series of experiments was performed in which three clones (two of which, 773.2 and 923.3, had been derived by micromanipulation) were tested on a larger panel of targets. As shown in Figure 4, the three clones showed a striking similarity in specificity, not only to each other but to both polyclonal fetal NK cells and polyclonal adult NK cells (see Fig. 2). As in the earlier series, some differences in fine specificity were noted. In particular, clone 773.2 showed markedly lower killing of A20 compared with either clone 923.3 (p = 2 × 10−5) or clone Cc5H10 (p = 5 × 10−8) or compared with polyclonal NK cells (p = 4 × 10−9 for adult NK cells, p = 3 × 10−9 for fetal NK cells). However, overall, the data show that the specificity of individual fetal NK clones is as broad as that of polyclonal NK cell populations.

FIGURE 4.

Comparative specificity of three C57BL/6 fetal NK clones. The killing of each target is expressed as a percentage of that of YAC-1 cells. The results are the mean values of between 3 and 10 experiments with each target/effector combination. The target cells are arranged according to their MHC haplotype and species origin.

FIGURE 4.

Comparative specificity of three C57BL/6 fetal NK clones. The killing of each target is expressed as a percentage of that of YAC-1 cells. The results are the mean values of between 3 and 10 experiments with each target/effector combination. The target cells are arranged according to their MHC haplotype and species origin.

Close modal

The simplest interpretation of these results would be that most or perhaps all individual NK cells share the same broad lytic specificity. However, this conclusion is dependent on the assumption that all cells within a clonal population are identical. Detailed phenotypic analysis of fetal NK clones revealed that this assumption was not correct. Thus, whereas the cells present in cloned populations (and polyclonal populations) of fetal NK cells were uniformly positive or negative for most cell surface molecules (such as NK1.1, CD3, and Ly49; see Fig. 1), they varied in their expression of a number of molecules, most of which also vary on adult splenic NK cells and may therefore define functionally distinct subpopulations. Among these are 10A7 (18), a highly NK cell-specific molecule, which is probably a member of the NKRP1 family (J. C. Ryan and W. E. Seaman, personal communication), Ly6C, a glycophosphatidylinositol (GPI)-linked membrane-signaling molecule also expressed by subpopulations of T cells and by myeloid cells (36), and 3C2, a molecule of unknown nature that is present on a subpopulation of splenic NK cells (37). Figure 5 shows examples of the staining patterns obtained with cloned and uncloned populations of NK cells. Uncloned fetal NK cells displayed a similar pattern of heterogeneous expression of 10A7, Ly6C, and 3C2 to that shown by adult splenic NK cells. Each of the three NK cell clones showed a distinct pattern of staining. Unexpectedly, they frequently contained clearly defined subpopulations of cells differing in their expression of these molecules. Adult and fetal NK cells also showed heterogeneous expression of the CD45RA and RB isoforms and of the CD45 glycosylation variants CT1 and CT2 (38) (only data for CT2 are shown in Fig. 5). Again, the same was often true for cloned fetal NK cell lines. We also found heterogeneous expression of the B220 isoform of CD45 on cloned NK cells (data not shown). By contrast, both cloned and uncloned NK cells displayed a uniformly high level of expression of CD45 molecules in total, as revealed by staining with a pan-CD45 mAb (Fig. 5), and also of the CD45RC isoform (not shown).

FIGURE 5.

Expression of various surface molecules on polyclonal adult NK cells, polyclonal fetal NK cells, and three fetal NK cell clones.

FIGURE 5.

Expression of various surface molecules on polyclonal adult NK cells, polyclonal fetal NK cells, and three fetal NK cell clones.

Close modal

Most of the NK clones we have studied were derived by limiting dilution. As determined by Poisson analysis, their probability of monoclonality was >95%. To formally exclude any possibility of a lack of monoclonality, several of the fetal NK clones were recloned by micromanipulation. The results of one such experiment are shown in Figure 6. Clone 923 showed heterogeneous staining for 10A7, Ly6C, and 3C2. In addition, although the vast majority of cells present in adult and fetal NK cell preparations are CD8, clone 923 was one of a small number of clones that expressed CD8. CD8α but not CD8β was expressed, and only ∼25% of the cells in this clone were CD8+. All of the subclones obtained by micromanipulation recloning of clone 923 showed heterogeneous staining with at least one mAb (typical results for 5 subclones are shown). Two further important observations were made in this experiment. First, the complexity of the staining patterns in the subclones was lower than in the parental clone, several of the subclones being either uniformly positive or uniformly negative for the variable Ags. In particular, subclone 5 displayed heterogeneity only for Ly6C. Second, the pattern of staining of each subclone differed not only from that of the parent clone but also from that of other subclones. These results suggest that each cell in the highly diversified clone 923 parental population was itself capable of undergoing diversification and that the diversification process was both random and time dependent. Furthermore, since all of the subclones were maintained under the same continuous growth conditions and grew at approximately the same rate, these results collectively eliminate the possibility that the differences in Ag expression seen on individual cells within fetal NK clones are related to cell cycle or activation events.

FIGURE 6.

Expression of various surface molecules on the parental clone 923 and five subclones derived by micromanipulation. Analysis was performed on day 35 after subcloning.

FIGURE 6.

Expression of various surface molecules on the parental clone 923 and five subclones derived by micromanipulation. Analysis was performed on day 35 after subcloning.

Close modal

If diversity were occurring continuously, it would be predicted that if an individual clone were studied over a period of time, new subpopulations of NK cells would appear and the “complexity” of the population would increase. To test this hypothesis, the relatively uniform subclone 923.5 was stained at various time points; the results are shown in Figure 7. At day 40, 5 days after the data for Figure 6 had been collected, one change in the staining pattern had already occurred, namely the acquisition of a small subpopulation of 3C2low cells. Even more convincingly, by day 103 subpopulations of both 10A7+ cells and CD8+ cells had appeared, resulting in the 923.5 “clonal” population now being as complex as the parental 923 clone. Over the next 170 days, although the relative balance between positive and negative subpopulations for each Ag fluctuated, there were no further major changes. Importantly, the newly arising variants did not out-grow and displace the original cell population. It therefore appears that once a highly diverse population has been created, there is some mechanism for maintaining it in a state of semiequilibrium. A partial but incomplete explanation for this is that the capacity of fetal NK cells to diversify is long lived. Thus, 6 mo after the recloning of clone 923, two of the daughter subclones, 923.3 and 923.5, were themselves recloned by micromanipulation. The second generation of daughter subclones showed a similar initial reduction in complexity, followed by gradually increasing diversification as described above for the first generation of subclones (data not shown).

FIGURE 7.

Expression of various surface molecules on clone 923.5 at various times after derivation.

FIGURE 7.

Expression of various surface molecules on clone 923.5 at various times after derivation.

Close modal

The finding that daughter subclones isolated from a complex parental clone each has a different subpopulation composition, and that the proportion of cells positive for one Ag can fluctuate independently of the proportion of cells positive for other Ags, indicates that the expression of variable Ags is independent. To test this directly, two-color fluorescence experiments were performed. The results illustrated in Figure 8 show that for a given pair of variable Ags all four possible subpopulations (double negative, double positive, and each of the single positives) existed. More importantly, the proportion of double positives was close to that expected for independent expression, i.e., the product of the proportions of cells positive for expression of each individual Ag. For example, clone 923.5 contained 59.6% Ly6C+ cells and 15.6% CD8+ cells, giving an expected proportion of double positive cells of 9.2%; the observed figure was 7.8%. Similarly, in clone Cc5H10 there were 48.2% CT1+ cells and 60.3% 3C2+ cells, giving an expected proportion of double positive cells of 29.1%; the observed figure was 30.4%.

FIGURE 8.

Variable Ags are expressed independently on cloned NK cells.

FIGURE 8.

Variable Ags are expressed independently on cloned NK cells.

Close modal

Although fetal NK cells derived from C57BL/6 mice are uniformly negative for expression of all Ly49 molecules for which Abs are currently available, we have found that fetal NK cell populations generated from certain other strains of mice occasionally show limited expression of Ly49 molecules. Some examples are shown in Figure 9. The second column shows a population of fetal NK cells derived from CBA mice, stained after 20 days in culture, in which 3% of cells expressed molecules recognized by mAb 5E6 (presumably Ly49C and/or -I) and about 1% expressed molecules recognized by mAb 4D11 (presumably Ly49G). When clones derived from such populations were examined, most were uniformly negative. However, occasional positive clones were found. For example, one clone (see third column of Fig. 9), derived by limiting dilution from CBA mice, expressed Ly49C/I but not -G. Another (fourth column), derived by micromanipulation from (C57BL/6 × BALB/c) F1 mice, expressed Ly49G but not -A (at least the C57 allelic form), -C, or -I. Importantly, of the four fetal NK cell clones that we have so far found to express Ly49 molecules, in each case only a proportion of the cells within the clones was positive, and only one of the three anti-Ly49 mAbs that we have used detected positive cells within a given clone.

FIGURE 9.

Expression of Ly49 molecules on polyclonal adult C57BL/6 NK cells, polyclonal CBA fetal NK cells, CBA fetal NK clone F4, and (C57BL/6 × BALB/c)F1 clone Hm5.

FIGURE 9.

Expression of Ly49 molecules on polyclonal adult C57BL/6 NK cells, polyclonal CBA fetal NK cells, CBA fetal NK clone F4, and (C57BL/6 × BALB/c)F1 clone Hm5.

Close modal

Our previous studies have shown that NK cells derived from fetal mice are phenotypically and functionally indistinguishable from adult splenic NK cells, with the exception that they lack expression of a number of Ly49 molecules. In the present study, we have extended this observation by showing that: 1) cloned fetal NK cell lines derived from C57BL/6 mice lack measurable expression of all of the inhibitory Ly49 molecules that can currently be detected with Abs, namely Ly49A, -C, -G, and -I; 2) they also fail to express the activating Ly49 molecule Ly49D; and 3) fetal NK cells derived from certain other strains of mice sometimes show a limited expression of Ly49 molecules.

Despite the general failure of C57BL/6 fetal NK cells to express any of the Ly49 molecules currently detectable with Abs, polyclonal populations of these cells displayed essentially the same specificity as Ly49-expressing adult NK cells when tested on a large panel of tumor target cells, even though all the targets (with the exception of LBRM) expressed MHC class I molecules. The similarity in specificity even extended to target cells expressing syngeneic (H-2b) class I molecules. It would therefore appear that under the conditions that these experiments were performed, recognition and signaling by Ly49 molecules played a negligible role in determining the lytic specificity of NK cells. The reason for this is unclear, since in both the original description of the function of Ly49 molecules (39) and in several subsequent studies (for example, see Refs. 40 and 41) similar assays and target cells were used, yet the influence of Ly49 molecules could be clearly seen. Although in one study (42) allorecognition by mouse NK cells was found to be more pronounced on blast cell targets than on tumor cell targets, this was the opposite of what was found in another study (40), and in our own system adult NK cells showed strong preferential killing of mutant class I-deficient tumor cells (our unpublished observations). Two explanations can therefore be offered for the similar specificity of Ly49-deficient fetal NK cells and Ly49-sufficient adult NK cells: either negative signaling following class I recognition in this system is mediated predominantly by molecules other than Ly49A, -C, -G, or -I; or most of the NK sensitive tumor cells used in this study have positive recognition structures that override negative signaling induced by class I. Intriguingly, we have recently found that Ly49-deficient fetal NK cells can indeed discriminate between matched pairs of class I-sufficient and class I-deficient tumor and blast cell targets, albeit to a lesser extent than adult NK cells (61).

Regardless of the role of negative signaling receptors, the close similarity in specificity of polyclonal fetal NK cells and adult NK cells strongly suggests that fetal and adult NK cells share the same positive recognition receptors and that these receptors have a similar clonal distribution in terms of both frequency and level of expression on these two types of NK cell. Three general theories have been put forward to explain the broad specificity of recognition by polyclonal populations of NK cells: 1) each NK cell has a unique and highly specific receptor, the broad specificity of polyclonal NK cell populations representing the summation of the specificities of individual NK cells; 2) all NK cells share a single receptor that recognizes broadly expressed target cell structure(s); 3) there are multiple receptors, all of which, or only some of which, may be shared between individual NK cells. Although the nature of the activating receptor(s) on NK cells is still unknown, it is clear from the present study that the single activating molecule envisaged in models 1 and 2 cannot be Ly49D.

In principle, theory 1 (as stated above) can be distinguished from theories 2 and 3 by examining the specificity of cloned NK cells. In the present case, such an examination revealed that individual fetal NK clones shared the same broad recognition capability both with each other and with polyclonal populations of fetal and adult NK cells, although occasional minor yet highly significant differences existed between individual clones and between clones and polyclonal NK cell populations in the lysis of particular targets. These findings argue strongly against model 1. However, because fetal NK cell clones violate the normal principles of clonal homogeneity and invariance it could be argued that each cell within an NK cell clone expresses a unique receptor. This seems unlikely, however, especially as it would require that the (other) surface molecules that vary within NK clonal populations have no influence on specificity. Model 2 cannot be ruled out, provided the differences in fine specificity were due to the influence of negative signaling receptors. Overall, however, we believe that the cytotoxicity and phenotypic data are most compatible with a version of model 3 in which NK cell diversity is created by variable expression of a limited set of cellular interaction molecules resulting in individual NK cells sharing some but not all receptors and polyclonal populations (including established “clones”) having a similar broad but not necessarily identical specificity. Our finding that fetal NK cell clones all have a similar broad specificity is different from that reported in a study of NK cell clones derived from adult p53-deficient mice (43). A possible explanation would be that adult NK cells do not undergo diversification during culture in vitro.

The surface molecules that we have found to display variation on fetal NK clones are 10A7, Ly6C, Ly49, 3C2, CD8, and certain isoforms of CD45. Variation in expression of these molecules is clearly not an artifact of long-term culture nor a peculiarity of fetal NK cells, because all of these molecules, with the exception of CD8, also show variable expression on short-term cultures of both fetal and adult NK cells. The finding of CD8 expression on some fetal NK clones was unexpected, as mouse NK cells are generally considered to be CD8. By contrast, CD8 is expressed on rat and human NK cells, but in both cases shows two parallels with the expression we have found on mouse fetal NK clones: 1) CD8 α- but not CD8 β-chains are expressed (44, 45); and 2) it is only expressed on a proportion of NK cells (46, 47). Its unexpected appearance on some mouse fetal NK clones suggests that the fetal/neonatal NK repertoire may differ somewhat from the adult repertoire, that CD8-expressing NK cells are strongly selected against in polyclonal populations, or that CD8-expressing NK cells become sequestered at sites other than the spleen.

Variation in Ag expression on fetal NK cells is a highly restricted and selective phenomenon, since most surface molecules show no variation either on polyclonal populations of NK cells or on individual fetal NK clones (Refs. 33 and 35; our unpublished observations). The selectivity of antigenic variation on NK cells is illustrated not only by the differential expression of CD8α and CD8β discussed above, but by the variable expression of one NKRP1 family member (10A7) and the invariant expression of another (NK1.1), and most strikingly within the CD45 family, where the total amount of CD45 expression on NK cells is remarkably uniform even on polyclonal NK cell populations, but the RA and RB isoforms, together with the glycosylation-dependent isoforms CT1 and CT2, are variable. Yet, although variation is restricted to a subset of surface molecules, the diversity created is potentially enormous. If each of the variable CD45R isoforms is counted separately, the number of variable molecules that we have identified so far, excluding Ly49, is eight, and presumably this is a fraction of the actual number. Two-color flow cytometric analysis revealed that the expression of individual variable Ags occurred independently. Thus, within a few weeks of growth, an individual fetal NK cell is potentially capable of generating a repertoire composed of at least 28 (>200) members.

The pattern of expression of the variable Ags on fetal NK cells rules out several trivial explanations. The possibility that variation is cell cycle dependent is eliminated by the observations that: 1) for most of the variable Ags, clones or subclones have been found in which expression over a period of days or weeks is either 0 or 100%; 2) there appears to be no correlation between the expression of individual variable Ags; and 3) there can be spontaneous gain of Ag expression. Furthermore, we have found that if a clone is enriched by cell sorting into Ag+ and Ag fractions, the proportions of Ag+ and Ag cells in the two enriched subpopulations changes very little, if at all, over a period of several days active growth (our unpublished data). These same considerations, coupled with the fact that the fetal NK cell lines and clones grow in continuous culture, without the need for feeder cells or periodic restimulation, rules out the notion that these molecules are “activation Ags.” (Note that although PMA was used to promote the growth of some clones, the use of this reagent was not the cause of heterogeneity, since its withdrawal did not affect the pattern of Ag expression, and several of the clones studied, in particular clone 923 and its subclones, were maintained throughout in the absence of PMA.) Lack of monoclonality was ruled out by serial micromanipulation recloning. Conventional mutation can also be ruled out for many reasons; in particular, the acquisition of variation is far too rapid and far too common (most variable Ags show variation on most clones). In addition, there is a clear predominance of Ag gain over Ag loss, which, along with the speed of variation and its existence among the cells present in short term cultures of fetal or adult NK cells, distinguishes this phenomenon from cellular senescence, a process in which long-term cell lines show a progressive loss of Ag expression. It is also important to note that although Ly6C and CD45 isoforms often show differential expression on different types of T cells, their expression on individual long-term T cell clones is both uniform and constant (see for example, Refs 38 and 48–52), as is CD8.

Not only is Ag gain more frequent than Ag loss, but in fact in our extensive studies of fetal NK clones and lines we have not observed a single instance of the complete extinction of an Ag-defined subpopulation once acquired. These findings pose intriguing questions regarding the cellular and molecular mechanisms involved in the generation and maintenance of diversification within NK clones. It bears a superficial resemblance to phase variation in bacteria, which in at least some cases is mediated by random inversion of DNA in the promoter region of the relevant gene (53). It is also reminiscent of position effect variegation in Drosophila (54) and the possibly related phenomenon of genomic imprinting in mammals (55). Control of gene expression in these circumstances is thought to be mediated by alterations in local chromatin structure and/or DNA methylation. A striking example of the random expression of a human CD2 transgene in mouse T lymphocytes caused by position effect variegation has recently been reported (56). Another possibility is that the regulatory factors that control the expression of the variable genes in fetal NK cells are at limiting concentration and due to dilution or asymmetric partitioning at division fall below a critical level. In the case of CD45, the variant forms of Ag are generated at the post-transcriptional level via differential exon splicing (24) and by changes in glycosylation (38); but the mechanisms involved are poorly understood and may involve changes in splicing factors and glycosylation enzymes that are ultimately controlled at the transcriptional level.

The function of most of the molecules that we have found to vary on NK cell clones is unclear. However, with the exception of 3C2, about which virtually nothing is known, there is indirect evidence that they may play some role in NK cell recognition and signaling. As already mentioned, 10A7 is an NK-specific molecule that probably belongs to the NKRP1 family. Ly6C is more widely expressed in the immune system, but recent studies indicate that it may have an important role in intercellular recognition. Thus, whereas alloreactive Ly6C+ and Ly6C CTL clones were able to lyse lymphoid target cells, only Ly6C+ clones could lyse nonlymphoid target cells, and both the killing and the binding of Ly6C+ cells to nonlymphoid target cells was inhibited by an anti-Ly6C mAb (48). Likewise, Abs to CD45 have frequently been reported to interfere with NK cell killing (25, 26, 27, 28, 29), and in some cases inhibition has been selective for certain types of target cell (28, 29). It has been suggested that sugar residues on CD45 molecules are involved in the binding of NK cells to targets (30). An alternative possibility is that CD45 itself is not directly involved in NK cell recognition but is noncovalently associated with molecules that are involved (57). Different CD45 isoforms may associate with different surface molecules (58). It is interesting to note that the invariant intracellular domain of CD45 is a protein tyrosine phosphatase (59). In view of the recent finding that negative signaling by p58/p70 killer cell inhibitory receptors in the human involves the binding of protein tyrosine phosphatase SHP-1 to ITIMs in the cytoplasmic domains of these molecules (60), it is conceivable that CD45 isoforms may act as conveyors of inhibitory signals in NK cells.

Of particular interest is that Ly49 molecules are also affected by the variation mechanism. Although fetal NK cell lines and clones generally fail to express any of the Ly49 molecules that can currently be detected with Abs, we have found some occasional exceptions to this rule, most commonly in fetal NK cells from non-C57 strains. Most remarkably, even in clonal populations of these cells only a proportion of the cells were positive for Ly49. Furthermore, within the limits of the range and specificity of anti-Ly49 Abs available to us, it appeared that a given clone expressed only a single Ly49 species. These observations suggest that both stochastic and nonstochastic mechanisms control the expression of Ly49 molecules in fetal NK cell clones. Thus, as with the other variable Ags discussed above, a stochastic element leads to only a proportion of cells within an NK cell clone expressing a given Ly49 molecule. However, each of the fetal clones that express Ly49 molecules seems committed to expressing only one (or a limited number) of all Ly49 molecules. These intriguing findings suggest that further investigation of this system may provide important insights into the cellular and molecular mechanisms that control the development of the NK cell recognition repertoire in vivo.

We thank Ms. F. Gays and Ms. B. Rossiter for excellent technical assistance and Drs. Austin Diamond, Jack Heinemann, and Zosia Chrzanowska-Lightowlers for helpful discussions.

1

This work was supported by grants from the Medical Research Council, U.K., and the University of Newcastle.

3

Abbreviations used in this paper: ITIM, immunoreceptor tyrosine-based inhibitory motif; D10F, DMEM containing 2× nonessential amino acids and 10% FBS; LBRM, a radiation-induced T cell lymphoma of B10.BR mice.

1
Ljunggren, H.-G., K. Kärre.
1990
. In search of the “missing self”: MHC molecules and NK recognition.
Immunol. Today
11
:
237
2
Karhofer, F. M., R. K. Ribando, W. M. Yokoyama.
1992
. MHC-class I alloantigen specificity of Ly49+ IL2-activated natural killer cells.
Nature
358
:
66
3
Moretta, A., C. Bottino, M. Vitale, D. Pende, R. Biassoni, M. C. Mingari, L. Moretta.
1996
. Receptors for HLA class I molecules in human natural killer cells.
Annu. Rev. Immunol.
14
:
67
4
Brown, M. G., A. A. Scalzo, K. Matsumoto, W. M. Yokoyama.
1997
. The natural killer gene complex: a genetic basis for understanding natural killer cell function and innate immunity.
Immunol. Rev.
155
:
53
5
Takei, F., J. Brennan, D. L. Mager.
1997
. The Ly49 family: genes, proteins, and recognition of class I MHC.
Immunol. Rev.
155
:
67
6
Karlhofer, F. M., R. Hunziker, A. Reichlin, D. H. Margulies, W. M. Yokoyama.
1994
. Host MHC class I molecules modulate in vivo expression of a NK cell receptor.
J. Immunol.
153
:
2407
7
Olsson, M. Y., K. Kärre, C. L. Sentman.
1995
. Altered phenotype and function of natural killer cells expressing the major histocompatibility complex receptor Ly-49 in mice transgenic for its ligand.
Proc. Natl. Acad. Sci. USA
92
:
1649
8
Held, W., J. R. Dorfman, M.-F. Wu, D. H. Raulet.
1996
. Major histocompatibility complex class I-dependent skewing of the natural killer cell Ly49 repertoire.
Eur. J. Immunol.
26
:
2286
9
Raulet, D. H., W. Held, I. Correa, J. Dorfman, M.-F. Wu, L. Corral.
1997
. Specificity, tolerance, and developmental regulation of natural killer cells defined by expression of class I-specific Ly49 receptors.
Immunol. Rev.
155
:
41
10
Höglund, P., J. Sunbäck, M. Y. Olsson-Alheim, M. Johansson, M. Salcedo, H.-G. C. Öhlén, C. L. Sentman Ljunggren, K. Kärre.
1997
.
Immunol. Rev.
155
:
11
11
Ryan, J. C., W. E. Seaman.
1997
. Divergent functions of lectin-like receptors on NK cells.
Immunol. Rev.
155
:
79
12
Mason, L. H., S. K. Anderson, W. M. Yokoyama, H. R. C. Smith, R. Winkler-Pickett, J. R. Ortaldo.
1996
. The Ly49D receptor activates murine natural killer cells.
J. Exp. Med.
184
:
2119
13
Chambers, W. H., N. L. Vujanovic, A. B. DeLeo, M. W. Olszowy, R. B. Herberman, J. C. Hiserodt.
1989
. Monoclonal antibody to a triggering structure expressed on rat natural killer cells and adherent lymphokine-activated killer cells.
J. Exp. Med.
169
:
1373
14
Bezouska, K., C.-T. Yuen, J. O’Brien, R. A. Childs, W. Chai, A. M. Lawson, K. Drbal, A. Fiserova, M. Pospisil, T. Feizi.
1994
. Oligosaccharide ligands for NKR-P1 protein activated NK cells and cytotoxicity.
Nature
372
:
150
15
Ryan, J. C., J. Truck, E. C. Niemi, W. M. Yokoyama, W. E. Seaman.
1992
. Molecular cloning of the NK1.1 antigen, a member of the NKR-P1 family of natural killer cell activation molecules.
J. Immunol.
149
:
1631
16
Hackett, J., M. Tutt, M. Lipscomb, M. Bennett, G. Koo, V. Kumar.
1986
. Origin and differentiation of natural killer cells. II. Functional and morphologic studies of purified NK1.1+ cells.
J. Immunol.
136
:
3124
17
Karlhofer, F. M., W. M. Yokoyama.
1991
. Stimulation of murine natural killer (NK) cells by a monoclonal antibody specific for the NK1.1 antigen: IL-2 activated NK cells possess additional specific stimulation pathways.
J. Immunol.
146
:
3662
18
Sentman, C. L., J. Hackett, T. A. Moore, M. M. Tutt, M. Bennett, V. Kumar.
1989
. Pan natural killer cell monoclonal antibodies and their relationship to the NK1.1 antigen.
Hybridoma
8
:
605
19
Thiele, D. L., P. E. Lipsky.
1989
. The role of cell surface recognition structures in the initiation of MHC-unrestricted “promiscuous” killing by T cells.
Immunol. Today
10
:
375
20
Siliciano, R. F., J. C. Pratt, R. E. Schmidt, J. Ritz, E. L. Reinherz.
1985
. Activation of cytolyic T lymphocyte and natural killer cell function through the T11 sheep erythrocyte binding protein.
Nature
317
:
428
21
Azuma, M., M. Cayabyab, C. Buck, J. H. Phillips, L. L. Lanier.
1992
. Involvement of CD28 in major histocompatibility complex-unrestricted cytotoxicity mediated by a human NK leukemia cell line.
J. Immunol.
149
:
1115
22
Chambers, B. J., M. Salcedo, H.-G. Ljunggren.
1996
. Triggering of natural killer cells by costimulatory molecule CD80 (B7.1).
Immunity
5
:
311
23
Carbone, E., G. Ruggiero, G. Terrazzano, C. Palomba, C. Manzo, S. Fontana, H. Spits, K. Kärre, S. Zappacosta.
1997
. A new mechanism of NK cell cytotoxicity activation: the CD40-CD40 ligand interaction.
J. Exp. Med.
185
:
2053
24
Thomas, M. L..
1989
. The leukocyte common antigen family.
Annu. Rev. Immunol.
7
:
339
25
Kasai, M., J. C. LeClerc, F.-W. Shen, H. Cantor.
1979
. Identification of Ly5 on the surface of “natural killer” cells in normal and athymic inbred mouse strains.
Immunogenetics
8
:
153
26
Seaman, W. E., N. Talal, L. A. Herzenberg, L. A. Herzenberg, J. A. Ledbetter.
1981
. Surface antigens on mouse natural killer cells: use of monoclonal antibodies to inhibit or enrich cytotoxicity activity.
J. Immunol.
127
:
982
27
Sparrow, R. L., I. F. C. McKenzie.
1983
. A function for human T200 in natural killer cytolysis.
Transplantation
36
:
166
28
Newman, W..
1982
. Selective blockade of human natural killer cells by a monoclonal antibody.
Proc. Natl. Acad. Sci. USA
79
:
3858
29
Newman, W., L. D. Fast, L. M. Rose.
1983
. Blockade of NK cell lysis is a property of monoclonal antibodies that bind to distinct regions of T-200.
J. Immunol.
131
:
1742
30
Gilbert, C. W., M. H. Zarouhian, W. J. Esselman.
1988
. Poly-N-acetyllactosamine structures on murine cell surface T200 glycoprotein participate in natural killer cell binding to YAC-1 targets.
J. Immunol.
140
:
2821
31
Bajpai, A., Z. Brahmi.
1994
. Target cell-induced inactivation of cytolytic lymphocytes. Role and regulation of CD45 and calcyculin A-inhibited phosphatase in response to interleukin-2.
J. Biol. Chem.
269
:
18864
32
Xu, X., A. S.-F. Chong.
1995
. Cross-linking of CD45 on NK cells stimulates p56lck-mediated tyrosine phosphorylation and IFN-γ production.
J. Immunol.
155
:
5241
33
Brooks, C. G., A. Georgiou, R. K. Jordan.
1993
. The majority of immature fetal thymocytes can be induced to proliferate to IL-2 and differentiate into cells indistinguishable from mature natural killer cells.
J. Immunol.
151
:
6645
34
Brooks, C. G., G. R. Flannery.
1980
. Quantitative studies of natural immunity to solid tumours in rats: persistence of natural immunity throughout reproductive life, and absence of suppressor cells in infant rats.
Immunology
39
:
187
35
Manoussaka, M., A. Georgiou, B. Rossiter, S. Shrestha, J. A. Toomey, P. V. Sivakumar, M. Bennett, V. Kumar, C. G. Brooks.
1997
. Phenotypic and functional characterization of long-lived NK cell lines of different maturational status obtained from mouse fetal liver.
J. Immunol.
158
:
112
36
Shevach, E. M., P. E. Korty.
1989
. Ly-6: a multigene family in search of a function.
Immunol. Today
10
:
195
37
Linnemeyer, P. A., J. M. Tsuji, S. Gill, S. B. Pollack.
1994
. Hamster monoclonal antibodies to murine natural killer cells.
Nat. Immun.
13
:
49
38
Lefrancois, L., L. Puddington, C. E. Machamer, M. J. Bevan.
1985
. Acquisition of cytotoxic T lymphocyte-specific carbohydrate differentiation antigens.
J. Exp. Med.
162
:
1275
39
Karhofer, F. M., R. K. Ribando, W. M. Yokoyama.
1992
. MHC-class I alloantigen specificity of Ly49+ IL2-activated natural killer cells.
Nature
358
:
66
40
Correa, I., L. Corral, D. H. Raulet.
1994
. Multiple natural killer cell-activating signals are inhibited by major histocompatibility complex class I expression in target cells.
Eur. J. Immunol.
24
:
1323
41
Mason, L. H., J. R. Ortaldo, H. A. Young, V. Kumar, M. Bennett, S. K. Anderson.
1995
. Cloning and functional characteristics of murine large granular lymphocyte-1: a number of the Ly49 gene family (Ly-49G2).
J. Exp. Med.
182
:
293
42
Sentman, C. L., M. Y. Olsson, M. Salcedo, P. Höglund, V. Lendahl, K. Kärre.
1994
. H-2 allele-specific protection from NK cell lysis in vitro for lymphoblasts but not tumor targets: protection mediated by α1/α2 domains.
J. Immunol.
153
:
5482
43
Karlhofer, F. M., M. M. Orikuela, W. M. Yokoyama.
1995
. Ly-49-independent natural killer (NK) cell specificity revealed by NK cell clones derived from p53-deficient mice.
J. Exp. Med.
181
:
1785
44
Baume, D. M., M. A. Caligiuri, T. J. Manley, J. F. Daley, J. Ritz.
1990
. Differential expression of CD8α and CD8β associated with MHC-restricted and non-MHC-restricted cytolytic effector cells.
Cell. Immunol.
131
:
352
45
Torres-Nagel, N., E. Kraus, M. H. Brown, G. Tiefenthaler, R. Mitnacht, A. F. Williams, T. Hünig.
1992
. Differential thymus dependence of rat CD8 isoform expression.
Eur. J. Immunol.
22
:
2841
46
Pesussia, B., V. Fanning, G. Trinchieri.
1983
. A human NK and K cell subset shares with cytotoxic T cells expression of the antigen recognized by antibody OKT8.
J. Immunol.
131
:
223
47
Cantrell, D. A., R. A. Robins, C. G. Brooks, R. W. Baldwin.
1982
. Phenotype of rat natural killer cells defined by monoclonal antibodies marking rat lymphocyte subsets.
Immunology
45
:
97
48
Havran, W. L., D. W. Lancki, R. L. Moldwin, D. P. Dialynas, F. W. Fitch.
1988
. Characterization of an anti-Ly6 monoclonal antibody which defines and activates cytolytic T lymphocytes.
J. Immunol.
140
:
1034
49
Johnson, R., D. W. Lancki, F. W. Fitch.
1993
. Accessory molecules involved in antigen-mediated cytolysis and lymphokine production by cytotoxic T lymphocyte subsets. I. Identification of functions for the T cell surface molecules Ly-6C and Thy-1.
J. Immunol.
151
:
2986
50
Luqman, M., P. Johnson, I. S. Trowbridge, K. Bottomly.
1991
. Differential expression of the alternatively spliced exons of murine CD45 in Th1 and Th2 clones.
Eur. J. Immunol.
21
:
12
51
Rogers, P. R., S. Pilapil, K. Hayakawa, P. L. Romain, D. C. Parker.
1992
. CD45 alternative exon expression in murine and human CD4+ T cell subsets.
J. Immunol.
148
:
4054
52
Inoue, T., Y. Asano, S. Matsuoka, M. Furutani-Seiki, S. Aizawa, H. Nishimura, T. Shirai, T. Tada.
1993
. Distinction of mouse CD8+ suppressor effector T cell clones by cytokine production and CD45 isoforms.
J. Immunol.
150
:
2121
53
Simon, M., J. Zieg, M. Silverman, G. Mandel, R. Doolittle.
1980
. Phase variation: evolution of a controlling element.
Science
209
:
1370
54
Henikoff, S..
1994
. A reconsideration of the mechanism of position effect.
Genetics
138
:
1
55
Razin, A., H. Cedar.
1994
. DNA methylation and genomic imprinting.
Cell
77
:
473
56
Festenstein, R., M. Tolaini, P. Corbella, C. Mamalki, J. Parrington, M. Fox, A. Milou, M. Jones, D. Kioussis.
1996
. Locus-control region function and heterochromatin-induced position effect variegation.
Science
271
:
1123
57
Burns, G. F., J. A. Werkmeister, T. Triglia.
1994
. A novel antigenic cell surface protein associated with T200 is involved in the post-activation stage of human NK cell-mediated lysis.
J. Immunol.
133
:
1391
58
Dianzani, U., V. Redoglia, F. Malavasi, M. Bragardo, A. Pileri, C. A. Janeway, K. Bottomly.
1992
. Molecular associations on the T cell surface correlate with immunological memory.
Eur. J. Immunol.
22
:
365
59
Charbonneau, H., N. K. Tonks, K. A. Walsh, E. H. Fischer.
1988
. The leukocyte common antigen (CD45): a putative receptor-linked protein tyrosine phosphatase.
Proc. Natl. Acad. Sci. USA
85
:
7182
60
Olcese, I., P. Lang, F. Vély, A. Cambiaggi, D. Marquet, M. Blery, K. L. Hippen, R. Biassoni, A. Moretta, L. Moretta, J. C. Cambier, and E. Vivier. Human and mouse killer-cell inhibitory receptors recruit PTPIC and PTPID protein tyrosine phosphatases. J. Immunol. 156:4531.
61
Toomey, J. A., S. Shrestha, S. A. de la Rue, F. Gays, J. H. Robinson, Z. M. A. Chrzanowska-Lightowlers, and C. G. Brooks. 1998. Major histocompatibility class I expression protects target cells from lysis by Lyx49-deficient fetal natural killer cells. Eur. J. Immunol. 28. in press.