Research ArticleNK CELLS

Clonal expansion and compartmentalized maintenance of rhesus macaque NK cell subsets

See allHide authors and affiliations

Science Immunology  02 Nov 2018:
Vol. 3, Issue 29, eaat9781
DOI: 10.1126/sciimmunol.aat9781

An enduring NK cell clan

Unlike other immune cells, the processes by which different NK cell subsets expand and are maintained are not well understood. Here, Wu et al. transplanted barcoded hematopoietic progenitor cells into rhesus macaques to follow the production, expansion, and maintenance of NK cells in the peripheral circulation for more than 4 years. They identified oligoclonal CD56CD16+ NK cell populations that segregated into unique clones, expanding or contracting over time and expressing specific surface KIR molecules. Offering a compartmentalized view of NK cell maintenance, distinct mature CD56CD16+ NK cell subpopulations were maintained through self-renewal, independent of differentiation from hematopoietic progenitors or less mature NK cells, even upon challenges representing moderate proliferative stress.


Natural killer (NK) cells recognize and eliminate infected and malignant cells. Their life histories are poorly understood, particularly in humans, due to lack of informative models and endogenous clonal markers. Here, we apply transplantation of barcoded rhesus macaque hematopoietic cells to interrogate the landscape of NK cell production, expansion, and life histories at a clonal level long term and after proliferative challenge. We identify oligoclonal populations of rhesus CD56CD16+ NK cells that are characterized by marked expansions and contractions over time yet remained long-term clonally uncoupled from other hematopoietic lineages, including CD56+CD16 NK cells. Individual or groups of CD56CD16+ expanded clones segregated with surface expression of specific killer immunoglobulin-like receptors. These clonally distinct NK cell subpopulation patterns persisted for more than 4 years, including after transient in vivo anti-CD16–mediated depletion and subsequent regeneration. Profound and sustained interleukin-15–mediated depletion was required to generate new oligoclonal CD56CD16+ NK cells. Together, our results indicate that linear NK cell production from multipotent hematopoietic progenitors or less mature CD56+CD16 cells is negligible during homeostasis and moderate proliferative stress. In such settings, peripheral compartmentalized self-renewal can maintain the composition of distinct, differentiated NK cell subpopulations.


Natural killer (NK) cells are defined as lymphocytes capable of target cell killing and cytokine production independent of somatic recombination-activating gene–mediated antigen receptor recombination. Thereby, distinguished from adaptive T and B cells, this fact precludes clonal tracking of NK cells based on receptor gene structure. In humans, NK cells have been phenotypically defined by the expression of CD56 and/or CD16, in the absence of T, B, and myeloid markers (1). More recently, a range of markers have been used to define functionally distinct human NK cell subsets (24). Models of human NK cell ontogeny have been proposed on the basis of comparative phenotyping of NK cells in the bone marrow (BM), lymph nodes (LNs), and blood; kinetics of recovery after transplantation; and in vitro culture (5, 6). Precursor cells derived from multipotent hematopoietic stem and progenitor cells (HSPCs) have been hypothesized to migrate from the BM to the LNs, generating CD56brightCD16 NK cells, followed by continuous maturation and release to maintain the predominant circulating CD56dimCD16+ NK cell subset. Deuterium labeling and Ki-67 staining studies have estimated a half-life of 14 days for circulating human NK cells and a proliferation rate of 4 to 5% per day, interpreted as evidence for ongoing release of immature progenitors. Nonetheless, almost instantaneous appearance of labeled CD56dim NK cells in the blood, consistent with proliferation of circulating cells, was also noted (7, 8).

Murine NK cells were shown to respond to a specific viral or hapten exposure, conferring protective immunity upon adoptive transfer and rechallenge, uncovering unexpected memory capabilities of NK cells similar to cytotoxic effector T cells (9). In addition, cytokine exposure has been shown to result in persistence of NK cell responses to viruses or tumors, even after adoptive transfer (10). In humans, subsets of CD56dimCD16+ NK cells expand upon cytomegalovirus (CMV) infection and may represent human analogs of murine adaptive NK cells (1114). In contrast to CD56bright and canonical CD56dim NK cells, human “adaptive” NK cells variegatedly lack expression of several signaling proteins and generally express killer immunoglobulin-like receptors (KIRs) and NKG2C (13, 14). Epigenetically, adaptive NK cells approximate CD8+ effector T cells (14), with decreased methylation at the IFNG CNS1 locus (15), suggesting shared developmental pathways. Epigenetically imprinted “clonal” expansions in humans have been inferred by the presence of NK cells expressing distinct combinations of KIRs after CMV infection or reactivation (11, 14). Mechanisms propagating functional NK cell diversity and memory in the absence of somatic receptor rearrangements have not been elucidated.

There are marked differences between the phenotype and function of mouse versus human NK cells (16, 17). Human NK cells are scantly produced in immunodeficient mice, and robust in vitro clonal assays are lacking; thus, processes maintaining their homeostasis and possible memory are poorly characterized in humans. Relative to rodents, rhesus macaque (RM) NK cells are well studied, are evolutionarily close to humans, and share phenotypic and functional characteristics with human NK cells: a dominant blood CD56CD16+ NK cell population, functionally resembling human CD56dimCD16+ NK cells, and a CD56+CD16 NK cell population that is dominant in LNs but is less frequent in blood, resembling human CD56brightCD16 NK cells (18). Functional evidence for antigen-specific NK cell memory in RM after simian immunodeficiency virus (SIV)/HIV vaccination has been reported (19). Whereas human and RM KIRs are evolutionarily divergent, RM Mamu-KIR3D receptors are extremely diverse, bind major histocompatibility complex (MHC) class I molecules analogous to human lineage II KIRs, and show variegated distribution demonstrated via staining with the limited number of available anti-KIR monoclonal antibodies (mAbs) and/or MHC class I tetramer binding (2022).

We developed a robust system of genetic “barcoding” to interrogate in vivo hematopoiesis after autologous RM HSPC transplantation, allowing tracking of clonal output from thousands of HSPCs and their progeny. In our initial report of hematopoiesis early after engraftment, we were surprised to find that most of the blood NK cells were clonally distinct from T, B, and myeloid cells derived from multipotent HSPCs (23). The minor CD56+CD16 NK cell blood population was clonally related to the other lineages; however, at these early time points, the major blood population of CD56CD16+ NK cells accounted for the divergence in clonal ontogeny. We interpreted these surprising results to be consistent with a distinct developmental pathway for CD56CD16+ NK cells.

Here, we have dissected clonal patterns in NK cells for more than 4 years, allowing a detailed long-term kinetic analysis of NK cell clonal populations in steady state and upon proliferative challenge. Our results provide direct evidence for long-term clonal expansions, persistence, and self-renewal of mature NK cells independent of progenitor cells, highlighting a previously unappreciated compartmentalization among NK cell subsets and forming the basis for a new model of NK cell dynamics.


Low diversity of CD56CD16+ NK cells and distinct clonal architecture by long-term clonal mapping

We tracked clonal contributions to NK cells for periods of more than 4 years after transplantation of RM with genetically barcoded autologous CD34+ HSPCs (fig. S1A), asking whether highly lineage-biased NK cells observed during the early post-engraftment period persisted (23) and dissecting the characteristics and origins of NK cells in vivo at a clonal level. Each unique barcode marks a single HSPC and its clonal progeny. The majority of circulating RM peripheral blood (PB) NK cells are CD56CD16+ (77.3 ± 3.2%, SEM), sharing properties with human CD56dimCD16+ NK cells. A smaller and individually variable subset of RM CD16+NK cells express low levels of CD56. The minor PB NK cell population in RM is CD56+CD16 (5.8 ± 1.1%), corresponding to human CD56brightCD16 cells (Fig. 1A and fig. S2, A to C) (18).

Fig. 1 Clonal composition of PB NK cells.

(A) FACS analysis of CD3CD20CD14 PBMCs for ZH33 at 12 months and ZG66 at 17 months after transplantation, with sorting gates for CD56+CD16 and CD56CD16+ NK cell subpopulations shown. (B) Pairwise Pearson correlation coefficients between fractional abundances of all barcodes for CD56+CD16 and CD56CD16+ NK cells from 6 months through the most recent follow-up [r values, P values, and 95% confidence intervals (CIs) in table S1]. (C) Shannon diversity indices for all barcode contributions to T, B, Gr, CD56+CD16 NK, and CD56CD16+ NK cells over time. Table S3 gives the number of barcodes above threshold for each sample. † indicates that the 30-month ZH33 CD56CD16+ sample was additionally sorted to be CD8α+NKG2+.

The overall degree of relatedness between clonal contributions can be demonstrated via Pearson correlations between two samples for all barcodes. There was minimal overall correlation between contributions from all clones to CD56CD16+ versus CD56+CD16 NK cells at all time points for the three animals [Fig. 1B (r values/P values) and table S1]. We also noted markedly lower clonal diversity over time for CD56CD16+ NK cells compared with other lineages, including CD56+CD16 NK cells, as assessed by the Shannon diversity index, taking into account both overall clone number and skewing of contributions (Fig. 1C). Diversity remained relatively stable and higher in other cell types for more than 3 years [Fig. 1C and (24)].

A focus on high-contributing clones offers information on those most responsible for hematopoiesis, reducing the impact of sampling bias. Contributions from the 10 highest-contributing clones in each lineage at each time point are shown in a heat map for each animal, clustering clones by overall similarity of contributions across all samples. In all three animals, very high contributing CD56CD16+ clones showed distinct contribution patterns in contrast to multipotent clones producing CD56+CD16 NK cells, granulocytes (Grs), T cells, and B cells, even as late as 49 months after transplant (Fig. 2A and table S2). This pattern was most notable in ZJ31, with very high contributing CD56CD16+ clones virtually undetectable in the other lineages. Similar highly expanded T cell clones appeared in animal ZH33 much later after transplantation than expanded NK cell clones.

Fig. 2 Clonal bias of CD56CD16+ NK cell subpopulations.

(A) Heat maps of the natural log fractional abundance of the highest-contributing clones defined as all barcodes present as a top 10 highest-contributing barcode in at least one of the samples, mapped over all samples for each animal. Corresponding percentage contributions in table S2. Each row corresponds to one unique barcode; * indicates that the barcode is one of the top 10 for that sample. Heat maps are organized by unsupervised hierarchical clustering of Euclidean distances between barcodes’ log fractional abundances, with relative contribution shown via color gradient. (B) Highly biased CD56CD16+ NK cell clone contributions over time. A barcode is defined as highly biased at a time point if it meets the following conditions: (i) >1% contribution to CD56CD16+ cells and (ii) >10-fold abundance in contribution to CD56CD16+ cells compared with T, B, Gr, and CD56+CD16 cells. y axes show the fraction of all barcode reads in CD56CD16+ cells contributed by the “highly biased” clones. Each color represents the contribution of a single clone; the same color at different time points indicates the same clone. The 30-month ZH33 CD56CD16+ sample was additionally sorted to be CD8α+NKG2+.

We further characterized highly biased NK cell clones, defined as a clone size >1% and contributing to CD56CD16+ NK cells at a ratio >10:1 compared with fractional contribution to any other cell type, including CD56+CD16 NK, T, B, or Gr cells. Figure S3 provides histograms validating a ratio >10:1 as clearly delineating a separate population of highly biased CD56CD16+ clones. When using the same criteria for comparisons with other lineages, only the T cell compartment in ZH33 showed appreciable numbers of high-contributing biased clones, developing late after transplant (Fig. 2A and fig. S3). The highly biased CD56CD16+ NK cell clones were few in number; a median of 8 (range, 4 to 15) per (a total of) 667 (range, 471 to 1503) clones was detected in this cell type per time point (table S3). However, these rare clones comprised a major fraction of blood CD56CD16+ NK cells in all three animals (Fig. 2B), indicating very marked expansion of these highly biased clones, with single clones contributing up to 21%.

Both CD8α and NKG2 represent additional markers expressed on virtually all CD56+CD16 as well as CD56CD16+ RM NK cells (18). Antibodies that react only with human NKG2A bind to both NKG2A and NKG2C isoforms on RM NK cells (18, 2527) and are thus referred to as NKG2. Sorting of CD3CD20CD14CD8α+NKG2+ cells for CD56+ and CD16+ confirmed highly biased clones as contributing to distinct subsets of bona fide CD56+CD16 or CD56CD16+ NK cells, respectively (fig. S2, D and E).

Overall, these biased clonal patterns are difficult to explain via continuous ongoing production of terminally differentiated short-lived RM CD56CD16+ NK cells from HSPCs via maturation through CD56+CD16 NK cells. Rather, NK cell clonal patterns could result from ongoing production via an independent pool of long-lived NK cell progenitors for these highly biased populations or homotypic clonal expansions of the distinct CD56CD16+ NK cells themselves, analogous to memory T cells (28).

NK cell clonal dynamics over time

To study the clonal dynamics of NK cell subpopulations, clones were grouped hierarchically into clusters within each cell type based on overall similarities of contributions over time, and the mean fractional contributions of each cluster over time were plotted. In ZH33 Grs, there were two clusters: one representing short-term engrafting progenitors that disappeared by 2 months, and the second representing very stable ongoing production from long-term clones beginning at 2 months and persisting to 49 months (Fig. 3A, left). Each clone contributed an average of less than 1% (Fig. 3A, right). In CD56+CD16 cells, the major cluster shows very stable output from clones contributing an average of less than 1 to 2% per clone (Fig. 3B). These clones were primarily multilineage, also producing T, B, and myeloid cells (Fig. 2A). Similar patterns emerged in additional animals (fig. S4).

Fig. 3 Cluster tracking of barcodes with similar kinetic behavior over time.

Heat maps (left) show the log fractional contribution to hematopoiesis of the highest-contributing clones (top 10 in each sample, as defined in Fig. 2A) organized by unsupervised hierarchical clustering in each cell type over time for animal ZH33. The hierarchical tree in each cell type was cut at a level to retrieve clear clusters via visual inspection of the dendrograms and heat maps. The clusters so defined are designated by color bars on the left of the heat maps. The fractional abundances of each clone in these clusters are averaged for all clones in the cluster and plotted over time (right), with the shaded ribbons around each line representing the SEM of the average fractional abundance. Line colors match the colors of the clusters in the corresponding heat map. (A) Gr, (B) CD56+CD16 NK cells, (C) CD56CD16+ NK cells, and (D) T cells.

The CD56CD16+ NK cell patterns were notably different in all three animals, both in kinetics and in magnitude of individual contributions (Fig. 3C and fig. S4). There were clear expansions and contractions of groups of clones over time, with some clusters composed of clones with average contributions of 6 to 8%. Plots of the changes over time for ZH33 clones shown individually are shown in fig. S5. A similar pattern of expanding and contracting clusters was observed in T cells later after transplant in animal ZH33 (Fig. 3D).

NK cell clonal segregation on the basis of specific cell surface receptor phenotypes

KIRs are diverse and rapidly evolving receptors mediating activating and inhibitory interactions with target cells. NK cell expansions occurring coincident with CMV reactivation after transplantation in humans are frequently KIR+, with clonal-like variegated patterns of specific KIRs expressed on subsets of expanding NK cells (11, 12). Expression of highly polymorphic and polygenic KIR genes has been described as stochastic and is presumed to be maintained epigenetically. Investigation of NK cell responses in vivo at a clonal level has been previously impossible, with no ability to link different KIR repertories to specific clonal expansions.

Many specific antibodies have been developed to detect diverse human KIRs. However, almost all human KIR antibodies lack cross-reactivity with RM KIRs, given the highly diverse and rapidly evolving nature of these loci in primates, and few RM-specific KIR antibodies have been developed (20, 29). An anti-human KIR2D (clone NKVFS1) antibody has been shown to recognize RM KIR3DL01 (30, 31). An alternative approach to identifying expression of specific RM KIRs has been staining with tetrameric peptide–MHC class I (pMHC-I) complexes shown to bind to KIRs and segregate RM NK cell populations with specific patterns of KIR expression (22, 32).

We stained CD3NKG2+ NK cells from nine RMs with anti-human KIR2D and the RM pMHC-I tetramer SIV-Gag GY9/Mamu-A1*002. Positive and negative subpopulations after staining with these reagents were distinct and varied between animals (Fig. 4, A and B), reminiscent of observed variegated, persistent KIR expression patterns on human NK cells (11). We sorted subpopulations of CD3CD14CD20CD16+NKG2+ NK cells from four barcoded monkeys for KIR2D and pMHC-I tetramer staining and performed barcode analysis on each distinct subpopulation of sufficient size for barcode retrieval (Fig. 4B). The highly biased and expanded CD16+NKG2+ NK cell clones showed clear segregation between sorted subpopulations (Fig. 4, C and D). For instance, in monkeys ZJ31 and ZG66, different sets of expanded clones completely segregated into the same KIR3DL01+ or KIR3DL01 subpopulations. The clones present in the KIR3DL01 fraction likely expressed other KIRs not reactive with available reagents. In ZH19, three major NK cell subpopulations were sorted on the basis of both KIR3DL01 and GY9/Mamu-A1*002 staining, and each population contained a set of specific clones. Within the KIR3DL01 NK cell subset, specific GY9+ clones were enriched and contributed major fractions to this subpopulation, although, within the bulk population, these clones were relatively small contributors, for instance, nos. 8, 9, 10, and 11 at 48.5 months and no. 44 at 52.5 months (Fig. 4C). ZK22 had no detectable KIR3DL01+ NK cells but had clear populations of GY9/Mamu-A1*002+cells showing barcode segregation. In all four animals, clonal segregation linked to KIR expression was stable long term, documented from 9 months to 2 years (Fig. 4D).

Fig. 4 Clonal segregation of CD16+NKG2+ NK cells based on KIR3DL01 expression and/or SIV-Gag GY9/Mamu-A1*002 tetramer staining.

(A) NK cell surface receptor subpopulations of CD3NKG2+ NK cells from RM PB (n = 9) detected by SIV-Gag GY9/Mamu-A1*002 tetramer and anti-KIR2D (NKVFS1) staining. (B) FACS plots of SIV-Gag GY9/Mamu-A1*002 tetramer and anti-KIR2D staining of PB CD3CD16+NKG2+ NK cells from barcoded RMs. (C) Top 10 barcoded clone heat maps for bulk CD16+NKG2+ NK cells and sorted NK cell subpopulations based on anti-KIR2D and tetramer staining for the time points and animals shown in (B). Distinct clonal distribution within different NK cell subpopulations distinguished by the presence and/or absence of staining is shown. (D) Pearson correlation coefficients comparing pairwise fractional contributions between NK KIR subpopulations from four monkeys same as in (C). The color scale for r values is shown on the right (r values, P values, and 95% CIs in table S1).

RMs in this study were rhesus CMV (rhCMV)+ at the time of transplantation. We documented reactivation of latent rhCMV immediately after transplant (fig. S6A). Although antibodies to markers for PLZF, FcεRγ, EAT-2, or SYK distinctly stained RM lymphocyte subsets, down-regulation of these signaling molecules, previously linked to human adaptive NK cells in the context of CMV infection (14), was not observed in CD16+CD56 NK cells of CMV+ compared with CMV RMs (fig. S6B). We also examined methylation of the CNS1 locus upstream of the IFNG promoter (fig. S6, C and D). Methylation at this locus strongly inversely correlates with human adaptive NK cell expansion after CMV infection (15). Although methylation at CNS1 was decreased in CD56CD16+ as compared with CD56+CD16 NK cells and B cells (fig. S6, E and F), the methylation levels were consistent with CNS1 methylation in human CD56bright versus canonical CD56dim NK cells, respectively. Thus, in our RMs, we did not identify NK cells directly analogous to human adaptive NK cells.

Altogether, these results are consistent with the expansion and maintenance of distinct CD16+ NK cell clones identified by KIR expression patterns. The relationship between distinct KIR expression patterns and specific barcoded clones persisted for up to 2 years, consistent with a model of peripheral homotypic expansion and maintenance of phenotypically and epigenetically distinct NK cell proliferating clones.

Regeneration of biased CD56CD16+ NK cell clones after short term in vivo depletion with anti-CD16 Ab

To gain insights into peripheral NK cell homeostasis, we asked whether regeneration of circulating CD56CD16+ NK cells after in vivo depletion would be supported via proliferation of pre-existing biased dominant clones, appearance of a new group of dominant clones, or maturation of a polyclonal population from multipotent HSPC clones via CD56+CD16 NK cells. To this end, we administered the anti-CD16 antibody 3G8 to RM ZH33 32 months after transplant, previously reported to deplete greater than 90% of RM NK cells in vivo with a single dose (33). By day 1, 99% of CD16bright NK cells disappeared from blood, with negligible impact on circulating CD56+CD16 NK cells (Fig. 5A); however, a population of CD16dim NK cells persisted. There was no impact on the numbers or barcode composition of T, B, or myeloid cells (fig. S7, A to C). To confirm true depletion of CD56CD16+ cells, versus trogocytosis or CD16 epitope blockade, we demonstrated nearly equivalent depletion of CD3CD20CD14NKG2+ cells (Fig. 5A). Highly biased CD56CD16+ clones (nos. 25 to 29, 49 to 55, 57, and 59) were profoundly depleted from peripheral blood mononuclear cells (PBMCs) on days 4 and 11 (Fig. 5C). The number and fraction of CD56CD16dim NK cells increased by day 1 and peaked at day 17. CD56CD16bright NK cells began to reappear by day 17 and normalized in number by day 36.

Fig. 5 In vivo depletion of CD16+ cells.

FACS plots and NK cell subset absolute concentrations (A) and Ki-67+ percentage of PB NK cell subsets (B) before and after anti-CD16–depleting antibody treatment (50 mg/kg) of ZH33 32 months after transplant. (C) Top 15 clones’ heat map shows the log fractional abundance of the T, B, Gr, CD16bright, CD16dim, CD56+, and total PBMC samples over time before and after antibody treatment. Depletion and regeneration of individual day −7 CD56CD16+ highly biased NK cell clones (numbered in red; defined in Fig. 2) are shown (left). The fractional contributions of these highly biased clones decreased on day 4 versus day 0 (P = 0.002, paired t test). (D) Pearson correlation between clonal contributions before and after anti-CD16 (r values, P values, and 95% CIs in table S1). (E) Stacked bar plots of relative clonal contributions to CD56CD16bright NK cells from CD16+ highly biased clones versus multilineage or clones biased toward other lineages before and after anti-CD16. NK cell–biased clones defined as >10-fold abundant in contribution to CD56CD16+ cells compared with T, B, Gr, and CD56+CD16 cells regardless of clone size. Gray bars: non-CD16bright NK cell–biased clones’ contributions; CD16bright NK cell–biased clones’ contributions are shown in colors reflecting time point of appearance, and each individual clone’s contribution is delineated by lines and stacked to create bars.

At baseline, as previously reported for human NK cell subsets (8), fewer CD56CD16+ compared with CD56+CD16 NK cells were actively cycling, as determined by Ki-67 staining (Fig. 5B). Ki-67 was expressed in 8.8 to 19.8% of the specific KIR+ subsets (fig. S7F), consistent with the hypothesis that these fully mature cells can self-renew. During regeneration, there was a marked increase in cycling CD16dim cells, peaking just before CD56CD16bright cells reappeared at day 17, consistent with regeneration of CD16bright from CD16dim NK cells. Despite a higher baseline, there was no consistent increase in cycling of CD56+CD16 cells.

In terms of cellular barcodes, overall clonal contributions to CD56CD16bright NK cells comparing baseline day −7 with day 36 were strongly correlated (r = 0.74; Fig. 5D). Most highly biased individual NK cell clones present before depletion appeared again after recovery (Fig. 5C). The overall fraction of clonal contributions to CD56CD16bright NK cells derived from biased clones compared with multipotent clones decreased somewhat after recovery but remained almost 50% (Fig. 5E). Many CD56CD16dim clones were shared with CD56CD16bright clones at baseline, and their overall clonal patterns were highly correlated (Fig. 5D). However, some biased dominant CD56CD16dim clones decreased contributions to the CD56CD16dim NK cell population by day 36, perhaps reflecting rapid proliferation and partial depletion of these clones via maturation during regeneration of the CD16brightCD56 compartment by certain clones within the CD56CD16dim compartment.

In contrast, we found only limited recruitment of multilineage, unbiased clones to regenerate CD56CD16+ NK cells after this short-term depletion. Only three large new highly biased clones appeared after recovery (nos. 6, 30, and 58; Fig. 5C). Overall clonal correlations between CD56CD16+ NK cell and other lineages, including CD56+CD16 NK cells, remained very low (Fig. 5D). Even under this proliferative stress, our data indicate that regeneration of CD56CD16+ cells from circulating CD56+CD16 precursors is unlikely. Rather, clonal patterns remained generally stable, suggesting self-renewal and high proliferative potential of CD56CD16+ NK cells. These NK cells may have regenerated from circulating CD56CD16dim clonally restricted NK cells surviving antibody depletion, given the induction of rapid cycling in this subset before refilling of the CD56CD16+ compartment.

Prolonged profound in vivo NK cell depletion via anti–interleukin-15

Next, we aimed to examine NK cell dynamics upon long-lasting in vivo depletion of NK cells. Eighteen months after full recovery from anti-CD16 antibody treatment and clonal stabilization, we administered three biweekly doses of an anti–interleukin-15 (IL-15) neutralizing antibody to RM ZH33. This treatment is reported to result in a profound and very prolonged depletion of all NK cell subsets within RM blood and tissues (Fig. 6A) (34). This antibody resulted in greater than 540-fold depletion of PB NKG2+CD56CD16+NK cells and in more than 10-fold depletion of NKG2+CD16CD56+ NK cells. In contrast to anti-CD16 antibody effects, both CD16bright and CD16dim cells were depleted by anti–IL-15. Recovery of NK cell numbers did not begin until more than 3 months after treatment, not reaching a normal range until more than 12 months later (Fig. 6, B and C). As reported previously, effector T cells were also depleted by this antibody (34) but to a lesser degree than NK cells and only short term (Fig. 6, B and C). Most of the residual CD16+ cells in PB during the NK cell depletion were NKG2CD8α cells expressing CD11c and human leukocyte antigen–antigen D, isotype R (HLA-DR): These cells likely represented atypical monocytes (fig. S7D).

Fig. 6 In vivo depletion of NK cells with anti–IL-15.

(A) Schematic of anti–IL-15 treatment. ZH33 received anti–IL-15 mAb on day 0 (20 mg/kg) and on days 14 and 28 (10 mg/kg). (B) Top and middle rows show NK cell subsets based on NKG2, CD16, and CD56 expression in CD3CD14CD20 PBMCs; bottom row shows CD95 and CD28 expression on CD3+ T cells. (C) Absolute PB concentration of NK and T cell subpopulations. (D) Heat map of the top 50 most abundant clones in CD16+ NK, CD56+ NK, T, B, mono, and Gr samples over time. Depleted and newly arising clones of different bias types are designated on the left of the heat map (defined per criteria in Fig. 2). Newly biased T clones are defined as non–NK cell–biased clones contributing to T cells at a ratio of 10:1 compared with fractional contribution to any other cell type, including CD56+CD16 NK, CD56CD16+ NK, B, or Gr cells at day 362. CD56+CD16 NK samples from days 167 and 341 were not obtained. (E) Stacked bar plots of relative contributions to CD56CD16+ NK cells from CD56CD16+ NK cell–biased clones versus multilineage or clones biased toward other lineages before and after anti–IL-15, using the same criteria defined in Fig. 5E. Each individual clone’s contribution is delineated by lines and stacked to create the bars, with colors designating clone type. NK cell–biased clones defined as >10-fold abundant in contribution to CD56CD16+ cells compared with T, B, Gr, and CD56+CD16 NK cells regardless of clone size.

The highly expanded and biased NK cell clones present at baseline (day −22) were profoundly depleted after the anti–IL-15 antibody treatment, and in contrast to the effect of anti-CD16 NK cell depletion, these clones almost completely disappeared upon NK cell recovery (Fig. 6D, designated by red bars on the left of the heat map and in Fig. 6E). Instead, new clonal expansions arose in the CD16+ NK cell compartment at day 167 after treatment, the earliest time point at which sufficient NK cells could be collected for clonal analysis. These expanded clones were present specifically in CD56CD16+ cells and, in contrast to results after the less profound and shorter-term depletion induced by the anti-CD16 antibody, were not the same expanded clones present at baseline (compare clones labeled with red bars with those labeled with yellow, blue, and purple bars in Fig. 6, D and E). Although some of these new expanded NK cell clones were biased (Fig. 6, D and E, yellow, blue, and purple bars), most contributed at detectable but relatively lower levels to T, B, Gr, and CD56+ NK cells (Fig. 6D). In addition, expanded clones not fitting criteria for CD16 bias began contributing to this compartment after recovery (Fig. 6, D and E, gray bars). New sets of expanded T cell clones also appeared (Fig. 6D, green bar). These results suggest that after profound and lasting depletion of all NK cells, including biased and expanded CD56CD16+ clones, new NK cells could not be regenerated peripherally from long-lived self-renewing NK cells and instead had to be produced from long-term engrafted HSPCs also contributing to other cell lineages.


Maintenance of distinct cell populations is a function of survival, self-renewal, and differentiation. The prevailing human model posits that CD56bright NK cells are produced from BM precursors and continuously generate CD56dim NK cells in secondary lymphoid tissues (5, 6). A CD34+CD38+CD123CD45RA+CD7+CD10+CD127 NK cell progenitor was recently identified in human hematopoietic tissues and could differentiate to CD56bright and rarer CD56dim NK cells in vitro and in xenografted mice (35). We recently used the RM clonal tracking approach to identify sites of NK cell production (36). CD56+CD16 NK cells were produced in the marrow from HSPCs; in contrast, oligoclonally expanded CD56CD16+ NK cells were homogeneously distributed between blood, nodes, and multiple marrow locations, suggesting that clonal expansions occur outside the marrow and nodes, even if the final steps in maturation occur in nodes (37). As evidenced by Ki-67 staining, human blood CD56bright NK cells were reported to cycle more rapidly than CD56dim NK cells (8). In the present study, we also observed higher steady-state turnover of RM CD56+CD16 compared with CD56CD16+ NK cells. However, in contrast to predictions from a continuous lineage differentiation model, the barcode composition of CD56+CD16 and CD56CD16+ NK cells remained distinct under homeostatic conditions over several years. Barcodes displayed unique, longitudinally stable profiles even within small populations of differentiated KIR-expressing CD56CD16+ NK cells. Moreover, transient depletion of CD56CD16+ NK cells did not increase CD56+CD16 NK cell proliferation but rather that of CD56CD16dim NK cells, which displayed the same clonal makeup as the CD56CD16bright NK cell population. These findings do not fit the paradigm of continuous differentiation from HSPCs and CD56bright NK cells during steady-state maintenance of mature NK cells and support a model of compartmentalized NK cell homeostasis (Fig. 7).

Fig. 7 Model for NK cell homeostasis and regeneration.

(A) Long-term repopulating stem cells (LT-HSCs), short-term repopulating progenitors (ST-HPCs), or NK cell restricted progenitors may be lentivirally barcoded. Each may give rise to mature CD16+NK cells, containing barcodes corresponding to the parental progenitors (red for those derived from ST-HPCs and blue for those derived from LT-HSCs). A few of these mature NK cells expand to large clonal populations, maintaining receptor expression and barcodes of the originating NK cells and persisting for months to years. ST-HPCs disappear, explaining lack of overlap of ST-HPC–derived NK cells with other hematopoietic lineages, including CD56+CD16 immature NK cells. (B) Upon anti-CD16 antibody depletion of CD56CD16bright NK cells, CD56CD16dimNK cells proliferate and regenerate the mature NK cell pool without significant contributions from CD56+CD16NK cells; thus, the populations remain clonally distinct. (C) After prolonged depletion of all NK cells with anti–IL-15, many expanded clones disappear, to be replaced by a different set of expanded NK cell clones upon return of IL-15. Some of the new expanded NK cell clones have barcodes overlapping multiple lineages because they develop from mature NK cells that have arisen from LT-HSCs after anti–IL-15 treatment.

The persistence of differentiated, oligoclonal expanded NK cell clones, even in the setting of proliferative stress, also argues against the notion of mature RM CD16+ or human canonical CD56dim NK cells as senescent (38). This conclusion is also supported by results of clinical NK cell infusions and murine adoptive transfer studies, documenting in vivo persistence of mature NK cells for several months without dependence on regeneration from marrow HSPCs (39, 40). We recently reported evidence from two clinical syndromes suggesting that human CD56dim NK cells can be maintained independently of ongoing production from HSPCs, suggesting longevity and self-renewal (41, 42). Hypothetically, high levels of homeostatic cytokines, such as IL-7 and IL-15, or even inflammatory cytokines, may be required to break the steady-state barrier between these subsets to rejuvenate the CD56CD16+NK cell pool. Very severe and prolonged depletion of the entire NK cell pool, using IL-15–depleting reagents, was required to disrupt the clonal makeup of this CD56CD16+ NK cell compartment and to eventually recruit new, distinct NK cell clones into the CD56CD16+ NK cell compartment. The lack of overlap between initial populations of expanded and biased CD56CD16+ clones arising and persisting after transplantation and long-term multipotent HSPC clones may result from rapid initial recovery of NK cells from short-term progenitors unable to contribute to other lineages long term, with acquisition of homotypic self-renewal in a subset of these initially produced NK cells, allowing clonal persistence without clonal overlap with HSPCs or CD56+CD16 NK cells (Fig. 7). Identification of the precise engrafting CD34+ short-term multipotent progenitor or NK-committed precursor serving as the source of these cells is the subject of future barcoding studies.

Our observations raise important questions as to how dynamics among NK cell subpopulations is orchestrated. The segregation of specific expanded clones with distinct surface KIR expression in our model, with many individual clones being uniformly positive or negative for a given KIR, provides evidence that a single original cell that proliferated to generate this clone had already undergone the process of KIR acquisition and epigenetic stabilization and was likely a fully mature NK cell. This notable result provides clear evidence for the stability of specific clonal KIR expression patterns through multiple cell divisions, for time periods up to several years. Specific KIRs segregating with individual or groups of barcoded NK cell clones may be directly responsible for signaling expansion or may simply be stably coexpressed with other receptors driving expansion and persistence of a relatively small number of contributing clones. Whether response to a specific stimulus or stochastic factors are responsible for NK clonal expansion and persistence is unclear. Waxing and waning groups of expanded NK cell clones over time suggest an environmentally driven process. Recent reports that CMV infection can induce the differentiation of epigenetically unique and stable human adaptive or memory-like CD56dim NK cell populations imply that RM CMV reactivation after transplantation might drive oligoclonal expansions (13, 14) and can be tested via analysis of NK cell clonal patterns in CMV RMs after de novo CMV infection. However, we could not demonstrate the fact that RM CD56CD16+ NK cells expressed the signature of signaling molecules and transcription factors previously linked to human adaptive NK cells arising in the context of CMV infection. Regrettably, no antibodies reactive with RM CD57 or NKG2C exist, markers also associated with human adaptive or memory-like NK cell subsets after CMV and other viral infections, although a recent report used RNA probes and flow cytometry to distinguish NKG2C- versus NKG2A-expressing RM NK cells (43, 44). Therefore, it remains unclear whether the RM models human adaptive NK cell differentiation and homeostasis.

In conclusion, the tracking of individual NK cell clonal histories in RMs provides direct evidence for the peripheral expansion and stable persistence of differentiated mature NK cell clones and associates these clones with stable expression of specific KIRs, offering insights into the development and maintenance of mature circulating NK cells. Under homeostatic conditions, our findings suggest a barrier between HSPC-dependent CD56bright and the more stably maintained CD56dim NK cell populations, providing a model for understanding NK cell differentiation and homeostasis (Fig. 7). In addition, these findings have translational implications because development of NK cell therapies requires an understanding of how, where, and why specific NK cells are generated (45, 46).


Barcoded autologous transplantation model and experimental design

Animal studies were approved by the National Heart, Lung, and Blood Institute (NHLBI) Animal Care and Use Committee. RM CD34+ HSPC transductions with barcoded lentiviral libraries and autologous transplantation after 10 Gy total body irradiation were performed as previously described (23, 47), with individual animal data summarized in fig. S1A. The transplanted product contained virtually no barcoded mature cells. Thus, barcoded hematopoiesis after engraftment was generated from CD34+ HSPCs, not from residual NK cells present in the initial graft (fig. S1B).

Barcode libraries with a diversity 10 times greater than that necessary to be more than 95% certain that each individual barcode marked a single CD34+ cell were used, at a transduction efficiency resulting in a single barcode per HSPC (23, 48). DNA samples from hematopoietic lineage cells were used to amplify the barcode region via polymerase chain reaction (PCR), followed by next-generation sequencing to recover the barcodes for lineage clonal relationship analysis. Output was processed using custom Python code to identify barcoded clones contributing above sequencing error and sampling thresholds (23, 24). Each clone’s fractional contribution to a sample could then be calculated. Custom R code is available on GitHub at

Cell purification and fluorescence-activated cell sorting analysis

Blood, LN, and BM cells were isolated on a density gradient and stained with various antibody panels for fluorescence-activated cell sorting (FACS) analyses and sorting. Antibodies are listed in table S5. Cellular subsets were FACS sorted to >98% purity using gating as shown in fig. S1C. Intracellular Ki-67 staining was performed using the eBioscience Foxp3/Transcription Factor Staining Buffer. Cells were sorted on a BD FACSAriaII sorter. All FACS data were analyzed using FlowJo. GY9/Mamu-A1*002 monomers, provided by D. A. Price (Cardiff University), were tetramerized with either phycoerythrin or allophycocyanin streptavidin and were used for staining 1 million PBMCs with 0.6 μg of tetramer for 15 min at room temperature and then with other surface marker antibodies for 30 min at 4°C.

Barcode retrieval

Cell DNA was extracted with the DNeasy Kit (Qiagen). DNA (100 to 500 ng) underwent 28-cycle PCR using Phusion High-Fidelity DNA Polymerase (Thermo Fisher Scientific). A universal reverse primer and a unique forward primer (table S4) were used to multiplex sample sequencing. After gel purification, 10 to 24 multiplexed samples were then pooled for sequencing on an Illumina HiSeq 2500 or 3000. The Python code used for extraction from FASTQ files is available at

In vivo depletion of NK cells

The mouse anti-human CD16 monoclonal NK depleting antibody (3G8) and the anti-IL-15 neutralizing antibody (M111) were obtained from the Nonhuman Primate Reagent Resource( and were used for NK cell depletion, as previously described (33, 34).

Computational and statistical analyses

Data analysis, Pearson correlations, Euclidean distances, P values, and plot generation were performed using R (Foundation for Statistical Computing) and Prism (GraphPad Software). The R code used is available online at


Fig. S1. Barcoded CD34+ HSPC monkey transplantation parameters and hematopoietic lineage cells FACS sort gating strategy.

Fig. S2. Phenotype of NK cell populations in RM blood, marrow, and LNs.

Fig. S3. Bias of all clones toward T, B, Gr, CD56+CD16 NK, and CD56CD16+ NK fractions.

Fig. S4. Cluster tracking of barcodes with similar kinetic behavior over time for ZG66 and ZJ31.

Fig. S5. Individual clonal tracking of CD56CD16+-biased clones in ZH33.

Fig. S6. Epigenetic profile and functionality of NKG2+CD56CD16+ NK cells.

Fig. S7. The effect of in vivo depletion of CD16+ NK cells on other lineages and Ki-67 expression on RM PB NK cell subsets.

Material and Methods

Table S1. Pearson correlation coefficient r values, P values, and 95% confidence intervals for the correlation coefficient.

Table S2. Corresponding percentage contributions to the samples presented in Fig. 2A.

Table S3. The number of barcodes above threshold for each sample from ZH33, ZG66, and ZJ31 in Figs. 1 and 2.

Table S4. Primer sequence information.

Table S5. Antibody information.

Data file S1.


Acknowledgments: We thank N. Uchida for the χHIV plasmid, K. Keyvanfar, NHLBI FACS Core, DNA Sequencing Core, animal care staff, and the NIH Biowulf High-Performance Computing Resource. Funding: This work was supported by NHLBI and NIAID Divisions of Intramural Research and the Scientific Research Training Program for Young Talents sponsored by the Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (to D.Y.); by the European Research Council/European Union Seventh Framework Programme (FP/2007-2013, ERC 311335), Swedish Research Council, Swedish Foundation for Strategic Research, Swedish Cancer Foundation, the Wallenberg Foundations, Stockholm County Council, and Karolinska Institutet Center for Innovative Medicine (to Y.T.B. and H.S.); and by the Leibniz ScienceCampus Chronic Inflammation German Research Foundation grants SFB TRR2141(B02), R03565/4-1, and Heisenberg Program R03565/1-1 (to C.R.). Author contributions: Conceptualization: C.W., Y.T.B., and C.E.D.; analytics and statistical analyses: B.L., S.J.K., D.A.E., R.L., and L.T.; investigation: C.W., D.A.E., S.J.K., D.Y., H.S., B.A.L., J.K.D.-M., R.L., A.K., Q.H., B.L., D.A.A., and S.P.; resources: A.K., R.L., and R.E.D.; writing: C.E.D., C.W., J.K.D.-M., D.A.E., and Y.T.B.; supervision: C.E.D., C.R., R.W.C., and Y.T.B. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All necessary data related to this research are included in this publication within the text or Supplementary Materials. All custom code utilized for data extraction and analysis can be accessed at Questions can be addressed to C.E.D. and Y.T.B.

Stay Connected to Science Immunology

Navigate This Article