A novel immature natural killer cell subpopulation predicts relapse after cord blood transplantation

Umbilical cord blood transplantation (CBT) has become an accepted alternative treatment of patients with hematologic cancers or other disorders.¹  Many of the disadvantages of CBT, including limited numbers of total nucleated cells, have been dealt with in significant ways, leading to marked reductions in the time to hematopoietic cell recovery.^2-4  Still unclear, however, are the short- and long-term consequences of immune reconstitution from CB grafts after transplantation. Although the kinetics of T- and B-cell subset recovery after CBT are well-described,^5,6  much less is known about the recovery of CB-derived natural killer (NK) cells in the posttransplant setting. This is not surprising, as NK cells have only begun to emerge as one of the most diverse compartments of the human immune repertoire.^7-9  Indeed, NK cells express an array of germline-encoded activating and inhibitory receptors that precisely regulate their activation status, leading to a highly varied cell population with marked phenotypic and functional diversity at the single-cell level.⁸ 

An improved understanding of NK cell diversity after CBT would not only help clarify the division of labor among NK cell subsets but also may suggest strategies to better exploit CB-derived NK cells in the clinic. This prediction gains impetus in view of recent studies showing improved clinical outcomes in patients with higher NK cell counts and more mature cell phenotypes after hematopoietic stem cell transplantation (HSCT).^10-14  Thus, we have used high-dimensional mass cytometry coupled to an unsupervised analytical approach, as well as improved quantitative measures, to analyze the recovering NK cell compartment in patients with high-risk hematologic malignancies who had undergone CBT at this center. Our findings identify cytomegalovirus (CMV) reactivation and interleukin 15 (IL-15) as key drivers of posttransplant NK cell diversity, and demonstrate a close correlation between increased NK cell diversity and a lower risk for relapse. One previously unrecognized subset of NK cells, characterized by a very low diversity index, immature phenotype, and inferior effector function in vitro, was associated with a particularly poor outcome. These insights should help to identify CBT recipients with a higher risk for relapse who may benefit from adjunctive strategies, such as adoptive NK cell therapy or administration of IL-15.

The central aim of this study was to provide a comprehensive analysis of the entire spectrum of NK cell subpopulations in CB and peripheral blood (PB) from healthy donors and from patients undergoing CBT for hematologic cancers. In particular, we sought to relate changes within the NK cell compartment and NK cell diversity at the single-cell level. The summary of the different approaches used in this study is included in supplemental Methods.

All analyses were performed at MD Anderson Cancer Center, with approval by the local institutional review board. PB mononuclear cells from healthy adults and CBT recipients and CB mononuclear cells were processed as previously described.¹⁵ 

A panel comprising 40 metal-tagged antibodies was used for the detailed characterization of NK cells (see supplemental Methods for details).

Mass cytometry data were normalized on the basis of EQ 4 element signal shift over time, using Fluidigm normalization software2. Data processing was performed using FlowJo version 10.2. Calibration beads were gated out and singlets chosen based on iridium 193 staining and event length, as described.¹⁶  The diversity analysis and markers used in these studies are shown in supplemental Methods.

To assess NK cell cytotoxicity, we cocultured NK cells with ⁵¹Cr-labeled K562 targets at multiple E:T ratios (20:1, 10:1, 5:1, 1:1); cytotoxicity was measured by target release of ⁵¹Cr, as previously described.¹⁷ 

Total RNA was collected from 5 million NK cells per donor and treated as described. SmartSeq2 RNA-seq libraries were prepared as described.¹⁸  Pooled libraries were sequenced (50-bp paired-end reads generated with the Illumina HiSeq 2500).¹⁹ 

To assess NK cell diversity after posttransplant engraftment of CB, we first sought to establish baseline measurements for both CB and PB samples from healthy donors. Thus, we used cytometry by time-of-flight and a panel of 40 antibodies (supplemental Table 1) against inhibitory and activating receptors, as well as differentiation, homing, and activation markers, to gain fresh insights into CB-NK cell diversity. In keeping with a previous report,⁸  we found that CB NK cells (n = 10) had a significantly lower diversity compared with NK cells from adult PB donors (n = 10; median diversity index, 537 [range, 311-891] vs median, 2858 [range, 1228-5630]; P < .0001). Using t-SNE, a t-distributed stochastic neighbor embedding algorithm, we next analyzed and displayed high-dimensional data on 2-dimensional maps.²⁰  This approach yielded considerable segregation between CB and PB-derived NK cells, with a predominantly immature phenotype (increased expression of CD56^bright, NKG2A, CD94, c-kit, Trail, CD62L, and CD27) dominating the CB-NK cell compartment. In contrast, PB-NK cells were characterized by a mature phenotypic signature consistent with previous antigen experience (increased expression of CD57, KLRG1, CD16, KIRs, NKG2C, Siglec-7, CD158b, TIGIT, and DNAM-1; Figure 1A). Similarly, transcription factors and other proteins associated with NK cell maturity and function, including T-bet, eomesodermin, perforin, granzyme A (GrA), and granzyme B (GrB), were expressed at higher levels by PB-NK compared with CB-NK cells (Figure 1A).

To obtain a different perspective on the diversity of CB- and PB-derived NK cell populations, we used SPADE (spanning-tree progression analysis of density-normalized events), an analytical tool designed for single-cell cytometry data analysis and visualization that segregates single cells into clusters and then projects them as nodes into a tree (Figure 1B). The PB-NK cell compartment was characterized by a larger number of nodes and larger node size, reflecting a more heterogeneous composition compared with that of the CB-NK cell compartment. When we divided the spade tree into 2 zones, the dominant zone within the CB NK compartment was NKG2A⁺, with a paucity of nodes expressing maturation markers such as CD57 or CD16, consistent with their less mature phenotype (Figure 1B). To assess the effect of these phenotypic findings on NK cell function, we tested the response of NK cells selected from CB or PB directly ex vivo without prior activation against K562 targets. PB-NK cells showed greater degranulation (CD107A), produced significantly higher amounts of interferon (IFN)-γ and tumor necrosis factor (TNF)-α, and exerted significantly greater cytotoxicity against K562 targets compared with results for CB-NK cells (Figure 1C-D).

To gain a more comprehensive understanding of NK cellular transcriptome complexity, we performed RNA sequencing in purified NK cells from 3 CB and 6 healthy PB donors. To functionally profile the set of genes that are over- or underexpressed, we performed gene ontology analysis on the differentially expressed genes between CB- and PB-NK cells, using the DAVID functional annotation tool.²¹  The results revealed very distinct signatures between CB and PB-NK cells (Figure 1E). Similarly, a Volcano plot confirmed unique gene expression profiles for CB- vs PB-NK cells (Figure 1F). Although genes involved in activation and adhesion were enhanced in PB-NK cells, those with functions in the cell cycle and replication were highly enriched in CB NK cells (Figure 1G). To identify possible transcriptional network differences between CB- and PB-derived NK cells, and to define groups of coordinately expressed genes, we next generated 2 coexpression networks using Pearson’s correlation coefficient^22,23  (Figure 1H). Within these networks, interconnected modules were identified on the basis of topological overlap²⁴  (supplemental Figure 1). We did not identify common hub genes between CB and PB-NK cells, indicating that each cell category possesses highly distinct transcriptional profiles (Figure 1I). Moreover, when we performed a gene functional annotation analysis (DAVID) to better understand the function of each module, only 2 of 17 modules were shared between CB- and PB-NK cells (supplemental Table 2). Together, these data reveal that PB- and CB-NK cells differ not only in the expression of their individual genes but also in the gene networks that regulate their function.

NK cells are the first lymphoid population to emerge after allogeneic HSCT; however, the early posttransplant dynamics of recovering CB-NK cell diversity remain unclear, largely because earlier studies incorporated only a limited number of phenotypic parameters. Thus, using a 40-parameter NK-focused mass cytometry panel (supplemental Table 1), we analyzed samples that were collected prospectively at multiple points from 43 patients who had undergone CBT at this center. The NK cell diversity index in these patients increased progressively over time (Figure 2A). By 3 months posttransplant, the median diversity of the CB-NK cells had more than doubled and was similar to that of patients undergoing PB HSCT measured at the same point (day 100; median, 2546; range, 1360-3874; n = 3).

We next hypothesized that CMV infection, which appears to drive NK cell maturation,^25-27  might explain the speed with which the NK cell repertoire acquires diversity after CBT. Indeed, CMV reactivation has been shown to induce upregulation of maturation markers including NKG2C, KIR, and CD57 on the reconstituting NK cells after CBT.²⁸ 

We first measured NK cell diversity in the PB of healthy adult CMV seropositive and CMV seronegative PB donors and found significantly greater diversity among NK cells from CMV seropositive healthy donors (Figure 2B). CMV reactivation was defined as the number of CMV antigen-positive cells per 1 × 10⁶ leukocytes or by polymerase chain reaction in the first 100 days after CBT. Preemptive therapy was initiated for more than 3 pp65 Ag cells/million white blood cells or CMV DNA copy numbers higher than 137. We further hypothesized that the effect of CMV infection on NK cell maturation and diversity would be most pronounced within the KIR⁺ compartment, as CMV reactivation stimulates the rapid maturation of NKG2A-KIR⁺ NK cells.²⁰  To test this idea, we used a serial gating strategy to select and define KIR⁺ and KIR⁻ populations for each sample (supplemental Figure 2). KIR expressing NK cells in CMV seropositive samples displayed a more mature phenotype, characterized by higher expression of maturation markers such as NKG2C, CD57, CD16, T-bet, perforin, GrA and GrB, and lower expression of CD94, CD62L, and NKG2A compared with results for their KIR⁻ counterparts (Figure 2C), whereas the maturational phenotype did not differ between these 2 populations in healthy CMV seronegative donors (supplemental Figure 3). NK cell diversity was significantly greater within the KIR⁺ vs the KIR⁻ population in CMV seropositive samples (Figure 2D), but not in the absence of CMV infection (Figure 2D). These results support CMV infection as an important contributor to the maturation and diversification of KIR⁺ NK cells.

We next tested the influence of CMV reactivation on NK diversity posttransplant. All patients with CMV reactivation after CBT were CMV seropositive before transplant. Notably, the patients with CMV reactivation had significantly higher NK cell diversity indices over the first 100 days after CBT than did those who lacked reactivation (Figure 3A). This relationship was even more pronounced within the KIR⁺ compartment, whereas neither the KIR⁺ nor the KIR⁻ NK cell subpopulation showed a change in diversity in patients without CMV reactivation (Figure 3B). Finally, in the cohort with CMV reactivation, KIR⁺ NK cells expressed increased levels of maturation markers (CD16, CD57), and cytolytic enzymes (GrA, GrB, and perforin), as well as transcription factors and adaptor proteins (Figure 3C), but there were no significant differences in the maturational profiles of KIR⁺ and KIR⁻ populations among patients without CMV reactivation (supplemental Figure 4). Thus, CMV infection appears to enhance NK cell maturation and diversity during the early posttransplant stages of immune reconstitution, especially among KIR⁺ NK cells.

We next determined the effect of CB-NK cell diversity on clinical outcome. Because NK cells recover early after CBT (supplemental Figure 5A-B), we measured NK cell diversity in PB samples collected from 43 patients within the first 100 days posttransplant (all patients were in complete remission at the time of sample collection) and correlated the results with the 2-year risk for relapse. Receiver operating characteristic (ROC) analysis identified a diversity of less than 805 as the cutoff predicting best for a higher relapse rate (sensitivity, 0.67; specificity, 0.88; AUC, 0.75; P = .02). In samples collected 100 days after CBT (n = 43), an NK diversity of less than 805 was associated with a significantly higher 2-year cumulative incidence of relapse risk at 60.0% (95% confidence interval [CI], 25.3%-82.7%) compared with 9.6% (95% CI, 2.4%-22.8%) in those with a diversity of at least 805 (Table 1). Although the absolute NK cell count early after CBT has been reported to influence clinical outcome,²⁹  we did not find a statistically significant correlation between this variable and NK cell repertoire diversity (Figure 4A). Thus, of these 2 variables, NK cell diversity appears to be a more specific indicator of NK cell maturational state, as well as a better predictor of relapse risk.

Ultimately, the clinical utility of CB-NK cell diversity will depend on the identification of phenotypically and functionally defined NK cell subsets that confer an especially high risk for relapse after CBT. Thus, using our large NK-focused mass cytometry panel, we studied 43 patients (discovery cohort), 9 of whom relapsed by 1 year post-CBT at a median time of 155 days (range, 92-253 days). Comparison of the t-SNE plots based on day 100 samples from the 9 patients who relapsed vs the 34 who remained in complete remission revealed strikingly distinct NK cell clusters, indicating major differences in phenotypic profiles (Figure 4B). Day 100 NK cells from patients who remained in remission expressed higher levels of specific maturation markers (NKG2C, CD16, and CD57), transcription factors, and signaling molecules (T-bet, eomesodermin, Syk), as well as markers of functionality (perforin and granzyme; Figure 4B), than did those from patients who relapsed in the first year after CBT. Patients who relapsed in the first year posttransplant had a large population of NK cells with a unique phenotype characterized by higher expression of NKG2A, CD94, and CD62L and low expression of CD16, NKG2C, and CD57. This NK subset also expressed significantly lower levels of cytotoxic molecules and transcription factors (Figure 4B), and its frequency was negatively correlated with increasing NK cell diversity (supplemental Figure 6). The median day 100 frequency of these cells within the total PB NK cell population in the 9 patients who relapsed in the first year post-CBT was 19.3% (range, 10.4%-43.0%) compared with 4.6% (range, 0.2%-22.0%) in the 34 patients who continued in remission (P < .0001; Figure 5A). Our findings are strengthened by the cells’ inability to perform natural cytotoxicity in response to K562 targets and by their decreased production of IFN-γ and TNF-α against K562 targets compared with results for their mature counterparts (Figures 5B; supplemental Figure 5C). Thus, we designated this novel NK cell subset as CB-NK_im.

To test the predictive strength of the CB-NK_im subset further, we first analyzed the 1-year risk for relapse according to higher or lower frequencies of this subset. Using multidimensional scaling plots in an unsupervised manner to determine similarities and segregation based on frequencies of the CB-NK_im subset, we observed that patients who remained in complete remission segregated from those who relapsed within the first year (Figure 5C). ROC analysis identified CB-NK_im frequencies of at least 11.8% as the cutoff predicting best for a higher relapse rate (Figure 5D; Table 1). We validated this finding in an independent cohort of 25 CBT recipients (validation cohort), using an abbreviated flow cytometry panel that could be applied in clinical practice (Figure 5E; supplemental Figure 7; supplemental Table 3). On the basis of selected markers from our larger cytometry by time-of-flight analysis, including CD57, NKG2A, CD62L, CD16, NKG2C, perforin, and T-bet, we showed that the CB-NK_im subset identified by flow cytometry met the previously defined cutoff frequency of at least 11.8% to identify patients with a higher risk for relapse after CBT (Figure 5F-H). Indeed, the 2-year cumulative incidence of relapse among patients with a CB-NK_im frequency of at least 11.8% vs less than 11.8% were 72.7% (95% CI, 37.1%-90.3%) and 4.3% (95% CI, 0.3%-18.0%; P = .0003) in the discovery and 85.7% (95% CI, 33.4%-97.8%) and 0.0% (P = .001) in the validation cohorts, respectively (Figure 5G-H; Table 1).

We performed univariate and multivariate analysis for 2-year risk for relapse posttransplant including the variables shown in Table 1. Multivariate analysis including these variables identified CB-NK_im frequency (hazard ratio, 24.4; 95% CI, 4.8-123.5; P < .0001) and disease risk index (hazard ratio, 2.3; 95% CI, 1.16-9.2; P = .01) as the only independent predictors for relapse in the discovery cohort. Multivariate analysis confirmed CB-NK_im frequency to be the only predictor for relapse in the validation cohort (hazard ratio, 23.6; 95% CI, 2.4-232.0; P = .007). Thus, we have identified CB-NK_im cells as a significant contributor to the relapse of high-risk hematologic malignancies after CBT.

To determine the origin of the CB-NK_im cells, as well as factors that could modulate them in the transplant recipient, we measured their frequency in the CB grafts before infusion (n = 10) and in PB from adult healthy donors (n = 10). CB-NK_im were present at significantly higher frequencies in CB compared with PB from healthy donors (Figure 6A), indicating that this NK subset is naturally present in CB and transferred to the recipient during transplant. To determine whether CB-NK_im cells have immunoregulatory function, we sort-purified these cells based on their expression of 5 surface markers (NKG2A⁺CD57⁻CD16⁻NKG2C⁻CD62L⁺) and performed a suppression assay by culturing them at a 5:1 ratio with autologous CB-derived CD4/CD8 T cells, as previously described.³⁰  CB-NK_im cells did not suppress T-cell proliferation (supplemental Figure 8A-B). To identify factors that could drive the maturation of CB-NK_im cells in vivo, we sort-purified CB-NK_im cells based on expression of 5 surface markers (NKG2A⁺CD57⁻CD16⁻NKG2C⁻CD62L⁺) vs mature NK cells (CD16⁺CD62L⁻) and then stimulated both populations with IL-15 or IL-2 for 7 days in vitro (Figure 6B). IL-15 induced differentiation and maturation of the CB-NK_im subset with upregulation of activating receptors, transcription factors, and cytotoxic molecules (Figure 6C; supplemental Figure 9), while enhancing IFN-γ and TNF-α production in response to K562 targets. In contrast, CB-NK_im cells cultured in the presence of IL-2 retained their immature function and phenotype (Figure 6B-C). Indeed, the frequency of CB-NK_im cells was significantly lower after 7 days of culture with IL-15 compared with IL-2 (Figure 6D). Consistent with these findings, the overall diversity of NK cells increased significantly when the CB-NK_im population was cultured in the presence of IL-15, in contrast to only marginal changes after exposure to IL-2 (Figure 6E).

Finally, a number of studies have reported higher IL-15 levels in CMV seropositive individuals³¹  and after CMV reactivation after HSCT.^32,33  Thus, we hypothesized that CMV reactivation could drive IL-15 production and consequent maturation of CB-NK_im cells after CBT. The day 100 plasma IL-15 levels were significantly higher in patients with CMV reactivation post-CBT compared with those without CMV reactivation (Figure 6F). Together, these data suggest that IL-15 drives NK cell maturation and diversity after CBT and that CMV reactivation may contribute to NK cell maturation by inducing higher levels of IL-15.

A number of research groups have shown a link between the kinetics of NK cell reconstitution and disease relapse and survival after HSCT.^29,33-35  These and other authors emphasize the prognostic contribution of the recovering absolute NK cell count early after transplantation and^34,36,37  the influence of certain combinations of donor and recipient HLA/KIR genotypes,^38-40  as well as CMV reactivation in achieving favorable outcomes after HSCT.⁴¹  However, the effect of NK cell diversity and of individual NK subpopulations on the antitumor response remains in question. Here, we used mass cytometry together with a number of innovative analytical tools to assess NK cell diversity after CBT. This comprehensive analysis of the entire NK cell repertoire, rather than particular markers or a single-cell population, provided an unbiased approach to understanding the diversity of the NK cell repertoire and its functional consequences during the first year after CBT.

NK cells in healthy CB display substantial phenotypic and functional immaturity, marked by lower expression of KIRs, higher expression of the NKG2A inhibitory receptor, and lower expression of activating receptors.⁴²  This raises an intriguing question: Is the diversity of the donor CB-NK cell repertoire sufficient to protect against early relapse after CBT, or do some functionally deficient subsets persist and dominate the NK cell compartment? Comprehensive analysis of the diversity of CB- and PB-derived NK cells from healthy donors, using a panel of cytometry by time-of-flight analytical platforms, indicated initially low diversity with strikingly immature phenotypes for CB-NK cells, contrasted with a full array of highly diverse and phenotypically mature subsets among PB-NK cells. This segregation of CB-derived from PB-derived NK cells is well supported by the results of RNA sequencing analyses showing distinct gene signatures for CB-NK vs PB-NK cells with few shared hub genes, pointing to the distinct transcriptomic, phenotypic, and functional profiles of these 2 cell populations.

A major outcome of this analysis was the discovery of a previously unrecognized immature CB-NK subset (CB-NK_im) that was present at a relatively high frequency in healthy CB and in the first 100 days after CBT in samples from patients relapsing in the first year posttransplant. This subset is different from the previously described NKG2A⁺CD62L⁺KIR⁻ population after allogeneic HSCT³⁵  in terms of their origin, their kinetic of maturation, and their functional status. CB-NK_im cells do not have characteristics of exhaustion^43-46  as they proliferate, diversify, and acquire effector function in response to IL-15 (as shown in Figure 6); they are present at frequencies of around 10% in CB, where we expect the NK cell and T-cell repertoires to be naive, rather than antigen-experienced, and are only rarely found in healthy PB. Moreover, they are characterized phenotypically as CD57⁻CD16⁻CD62L⁺NKG2A⁺NKG2C⁻perforin-T-bet⁻ (all markers of immaturity).

Notably, this subset was rare in the 34 patients who remained in remission at 1-year posttransplant. We validated this finding in an independent cohort of 25 CB recipients, using an abbreviated flow cytometry panel that could be applied in clinical practice. Given the negative prognostic strength of the CB-NK_im phenotype, we suggest that it be detected as early as possible in the posttransplant course, thus allowing interventions that might restore immune protection as outlined here. Still, rapid reconstitution of the CB-NK cell repertoire within the first 100 days posttransplant in approximately three-quarters of the patients contributed to a relapse rate comparable to that in patients undergoing HSCTs.⁴⁷ 

To elucidate the mechanism or mechanisms underlying the effect of CB-NK_im cells on disease relapse, we performed a series of studies showing that although these cells can expand in response to IL-2, they fail to undergo phenotypic and functional maturation and retain a low diversity index. In contrast, culturing the cells in the presence of IL-15 induced both functional maturation and diversification. Thus, we propose that in the absence of adequate concentrations of IL-15, CB-NK_im cells can expand in the early post-CBT period while failing to mature and attain a functional phenotype. Given that higher IL-15 levels in the early post-CBT period correlate with greater NK cell diversity and lower frequencies of CB-NK_im subpopulation (supplemental Figure 6A-C), and that IL-15 is more abundant in the sera of patients with CMV reactivation post-CBT (Figure 6F), we also suggest that CMV contributes to NK cell maturation and diversity at least partly by increasing IL-15 production.

In the present study, analysis of donor CB-NK cell diversity and its effects on the recipient was limited to the first and second years post-CBT, and did not provide information on the long-term effect of this factor on clinical outcome. We would stress, however, that the vast majority of posttransplant relapses occur within the first year.⁴⁸  Although the relatively small sample size and patient heterogeneity could limit some of the conclusions from our study, it is nonetheless the first to measure NK diversity by high-dimensional single-cell analysis and assess its clinical implications after CBT. Finally, our findings suggest that high-dimensional mass cytometry and innovative analytical platforms may be useful tools in predicting the effect of this variable on treatment outcome and informing the design of future therapeutics, as well as strategies to prevent relapse after CBT.

The authors thank Duncan Mak and Jared K. Burks (Department of Leukemia, MD Anderson Cancer Center) for expert advice and technical assistance with mass cytometry analysis.

This work was supported by National Institutes of Health, National Cancer Institute grants (CA061508-21A1 and CA211044-01), the Cancer Prevention and Research Institute of Texas (RP160693), and the MD Anderson Cancer Center Institutional Research Grant Program. The Flow Cytometry and Cellular Imaging Facility is supported in part by the National Institutes of Health, National Cancer Institute through a support grant to MD Anderson Cancer Center (CA016672).

Contribution: L.L. performed experiments and interpreted and analyzed data; H.C., D.M., Y.X., and Q.M. analyzed data and commented on the manuscript; J.L., J.J., B.H., and M.K. assisted with experiments; P.P.B., H.S., M.D., R.B., N.I., R.S.M., L.N.K., M.M., G.O., E.G., M.H.S., Y.L., K.C., J.W., W.-L.Z., R.E.C., and E.J.S. provided advice on experiments and commented on the manuscript; K.R. designed and directed the study; and L.L., L.M.-F., and K.R., wrote the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Katayoun Rezvani, Department of Stem Cell Transplantation and Cellular Therapy, The University of Texas MD Anderson Cancer Center, 1515 Holcombe Blvd, Box 423, Houston, TX 77030-4009; e-mail: krezvani@mdanderson.org.