Natural killer (NK) cells are an emerging cancer cellular therapy and potent mediators of antitumor immunity. Cytokine-induced memory-like (ML) NK cellular therapy is safe and induces remissions in patients with acute myeloid leukemia (AML). However, the dynamic changes in phenotype that occur after NK-cell transfer that affect patient outcomes remain unclear. Here, we report comprehensive multidimensional correlates from ML NK cell–treated patients with AML using mass cytometry. These data identify a unique in vivo differentiated ML NK–cell phenotype distinct from conventional NK cells. Moreover, the inhibitory receptor NKG2A is a dominant, transcriptionally induced checkpoint important for ML, but not conventional NK-cell responses to cancer. The frequency of CD8α+ donor NK cells is negatively associated with AML patient outcomes after ML NK therapy. Thus, elucidating the multidimensional dynamics of donor ML NK cells in vivo revealed critical factors important for clinical response, and new avenues to enhance NK-cell therapeutics.
Mass cytometry reveals an in vivo memory-like NK-cell phenotype, where NKG2A is a dominant checkpoint, and CD8α is associated with treatment failure after ML NK–cell therapy. These findings identify multiple avenues for optimizing ML NK–cell immunotherapy for cancer and define mechanisms important for ML NK–cell function.
This article is highlighted in the In This Issue feature, p. 1775
Natural killer (NK) cells are cytotoxic innate lymphocytes that are important for mediating antiviral host defense and responding to malignantly transformed cells (1). NK-cell activation is determined by the balance of signals received through germline DNA encoded activating, inhibitory, and cytokine receptors, which differs from T cells that rely on rearrangement of the T-cell receptor genes (2). Thus, NK cells are equipped to respond to a variety of malignant cells and have been investigated as a cellular immunotherapy for acute myeloid leukemia (AML), a clinically challenging blood cancer where the primary curative therapy is hematopoietic cell transplantation (HCT; refs. 3–5).
NK cellular immunotherapies are a nascent, promising, and safe alternative to T cells for cellular cancer immunotherapy (6, 7). Several types of NK-cell therapy have been shown to mediate antitumor responses in patients with AML without cytokine release syndrome (CRS) or immune cell–associated neurotoxicity syndrome (ICANS), which are frequent complications after chimeric antigen receptor T-cell immunotherapy approaches (4, 5, 7–9). However, the in vivo dynamic changes in donor NK cells that occur after transfer have not been extensively investigated, and both donor NK cell–intrinsic and host factors that contribute to treatment response and resistance are poorly understood.
Our group and others have identified memory-like (ML) properties of NK cells after brief activation with the cytokines IL12, IL15, and IL18 followed by differentiation in vitro or in vivo in murine models and NSG mice (5, 10–12). In vitro differentiated ML NK cells have increased activating receptors, can ignore the rules of KIR–KIR ligand interactions, exhibit prolonged survival in NSG xenograft models, and have improved effector functions against a wide array of targets (5, 13, 14). We previously reported the first-in-human clinical trial demonstrating that donor ML NK cells were safe, were detectable for several weeks after transfer, and induced complete remissions in patients with high-risk relapsed/refractory (rel/ref) AML. However, not all patients responded, and the median duration of response was only a few months (5). The phenotypic changes in ML NK cells that occur during in vivo differentiation, and factors contributing to therapeutic response and resistance, were not explored and remain important questions in the field.
Here, we utilized mass cytometry to understand the dynamic changes that occur in ML NK cells during in vivo differentiation within patients with AML. We discovered that ML NK cells are clearly distinct from conventional and activated NK cells and have a unique, consistent, well-defined multidimensional signature. This multidimensional analysis was integrated with clinical results and identified NKG2A as the predominant checkpoint on ML NK cells, as well as unexpected characteristics of baseline donor NK cells that predict treatment failure.
Mass Cytometry Distinguishes In Vivo Differentiated ML NK Cells
Our initial report describing the dose-escalation cohort of the first-in-human trial using cytokine (IL12, IL15, and IL18) induced NK cells to treat patients with rel/ref AML demonstrated that donor ML NK cells expand and proliferate in vivo in patients with AML and result in complete remissions (Fig. 1A; ref. 5). The now-complete results of the phase I study are described in the Supplementary Methods, with ML NK–cell therapy being well tolerated without CRS, graft-versus-host disease (GVHD), or neurotoxicity (Supplementary Tables S1–S3). Among the 15 evaluable patients, 7 achieved complete response (CR; n = 3) or complete response with incomplete count recovery (CRi; n = 4) and 3 achieved a best response of morphologic leukemia-free state (MLFS) at day 14 by the IWG response criteria (15), for an overall International Working Group (IWG) response rate of 67% and a CR/CRi rate of 47% (Supplementary Table S1).
Donor NK cells were detected in both patient peripheral blood (PB) and bone marrow (BM) by flow cytometry, with peak expansion occurring 7 to 14 days post–NK cell infusion for patients at all dose levels (5). We hypothesized that in vivo differentiated ML NK cells are distinct from baseline NK cells and NK cells acutely activated with cytokines. To test this, patient PB and BM were analyzed using a 37-marker NK-cell mass cytometry panel (Supplementary Table S4; Supplementary Fig. S1A and S1B) and major immune cell subsets were identified using FlowSOM (ref. 16; Fig. 1B; Supplementary Fig. S1C), including NK cells. Donor NK cells in recipient peripheral blood mononuclear cells (PBMC) were identified using donor- and recipient-specific HLA mAbs (Supplementary Fig. S1B). Donor NK cells were compared at the time of initial isolation (baseline), following 12- to 16-hour cytokine activation (immediately prior to infusion), and within patient PB or BM mononuclear cells (when available) 7 days following NK-cell infusion using t-SNE–based analysis (viSNE). On the basis of 25 markers, baseline (BL), cytokine-activated (ACT), and ML NK cells are distinct (Fig. 1C and D; Supplementary Table S4) as indicated by discrete islands within the viSNE maps. These distinctions are consistent across the 11 available patients assessed by mass cytometry at this time point (Supplementary Table S1; Fig. 1E; P > 0.001 as determined by two-way ANOVA, see Methods). For a majority of the patients, donor NK cells are the main lymphocyte subset by frequency and total numbers (Fig. 1F and G), confirming initial flow cytometry results on a subset of patients (5). Within the dose level 3 cohort (2–10 × 106 ML NK cells/kg), a significant association between NK-cell frequency or absolute cell numbers in PB at day 7 and clinical response was not detected, although this study was not powered for this correlative endpoint (Supplementary Fig. S1D). Similarly, we did not detect an association between regulatory T (Treg) cell numbers or frequency and patient responses, different from other types of NK-cell therapy (17). Total circulating CD34+ cells (expressed on most AML) were significantly negatively associated with response, as expected (Supplementary Fig. S1D).
In Vivo Differentiated ML NK Cells Are Phenotypically Distinct
On the basis of in vitro studies, we hypothesized that ML NK cells could be distinguishable from conventional NK cells by examining a large number of cell surface and intracellular markers. Using the median expression of markers for each BL, ACT, and ML NK–cell subset defined on the basis of t-SNE analysis (Fig. 1), we identified the markers significantly associated with each NK-cell type. ACT NK cells were defined by significantly decreased CD56 and increased CD25, CD69, and CD137, which are well-defined markers of acute NK-cell activation, and consistent with our in vitro reports (Fig. 2A and B; Supplementary Fig. S2A–S2B; refs. 5, 13). ML NK cells were defined by significantly increased CD56, Ki-67, NKG2A, and activating receptors NKG2D, NKp30, and NKp44 (Fig. 2A and B; Supplementary Fig. S2A). In addition, modest decreases in the median expression of CD16 and CD11b were observed (Fig. 2A and B). ML NK cells expressed CD16 following in vivo differentiation (median percent positive 69 ± 16%), consistent with prior studies demonstrating ML NK cells have enhanced antibody-dependent cellular cytotoxicity (14). Increased frequency of TRAIL, CD69, CD62L, NKG2A, and NKp30-positive NK cells were observed in ML NK cells compared with both ACT and BL, whereas the frequencies of CD27+ and CD127+ NK cells were reduced (Supplementary Fig. S2C). Finally, unlike in vitro differentiated ML NK cells, in vivo differentiated ML NK cells did not express CD25 (IL2Rα; Fig. 2A and B; Supplementary Fig. S2C; ref. 5). We postulate this may be due to in vivo ligation by low-dose IL2 used to support ML NK cells.
In one case, a cytomegalovirus seropositive donor's NK cells were predominantly CD57+NKG2C+, which are presumably comprising adaptive NK cells (18, 19). Using mass cytometry, we were able to determine that adaptive NK cells could also differentiate into ML NK cells in vivo, as the cells exhibited an ML NK–cell signature, including increased CD56 and activating receptor expression (CIML020; Supplementary Figs. S2B and S3A and S3B). However, the fraction of CD57+NKG2C+ cells remained constant at BL, ACT, and following in vivo ML NK–cell differentiation, suggesting that the presence of adaptive markers and biology did not affect ML NK–cell differentiation. NKG2C expression was modest on the remaining donor NK cells and was not altered by in vivo ML NK–cell differentiation (Supplementary Fig. S2C).
Because this patient had sufficient donor and recipient NK cells for advanced analysis, PB NK cells at D7 were examined (Supplementary Fig. S3C–S3E). Using the same analysis approach (Fig. 2A and B), donor and recipient NK cells were compared using viSNE and represent distinct populations (Supplementary Fig. S3C–S3E). The separation of these populations is confirmed by HLA-A2 staining (Supplementary Fig. S3D). Finally, the donor ML NK cells demonstrate the consistent ML phenotype, whereas the recipient NK cells do not have increased NKG2A, are CD11b+, and have lower activating receptor expression compared with donor ML NK cells (Supplementary Fig. S3E). These analyses were not possible for additional patients due to a paucity of recipient NK cells present at D7, but support that ML NK cells are phenotypically distinct from baseline NK cells.
ML NK Cells Are Similar between Patient BM and Blood
Because AML routinely involves the BM as a unique AML microenvironment, patient BM mononuclear cells (BMMC) were also examined and compared with PB ML NK cells using mass cytometry at day 7 or 8 post-infusion. Consistent with our previous report, donor NK cells trafficked to the BM and represented the major population observed in this tissue for most patients assessed (Fig. 3A; Supplementary Table S4; ref. 5). Using t-SNE analysis (Figs. 1 and 2), PB and BM donor NK cells had a similar multidimensional phenotype when compared to each other, but were again distinct from BL NK cells (Fig. 3B and C). Although median expression of most markers assessed was similar between BM and PB NK cells, BM donor–positive NK cells were significantly reduced in NKp46 (PB 18.66 ± 2.14 SEM v BM 7.66 ± 2.56, P = 0.006), potentially indicating downregulation after interaction with AML blasts. KIR expression and KIR diversity on in vitro differentiated ML NK cells did not vary (5). To understand how KIR repertoire was altered by in vivo donor ML differentiation, we compared KIR diversity on donor BL and in vivo differentiated ML NK cells (5, 20). If a particular subset of KIR-expressing cells had a proliferative advantage in vivo, we would expect KIR diversity to decrease. However, here we see KIR diversity modestly increased, without significant changes in any single KIR (Supplementary Fig. S4A and S4B).
Donor ML NK Cells Differentiated in Patients Are Polyfunctional Ex Vivo
For a subset of patients with adequate cell numbers, ex vivo functional responses against K562 leukemia targets were examined using mass cytometry. Freshly isolated PBMCs were coincubated with K562 cells for 6 hours; degranulation (CD107a), cytokine production (IFNγ, TNF) and chemokine production (MIP1α) were measured using the functional mass cytometry panel (Fig. 4A; Supplementary Table S4). When cocultured with leukemia targets, donor ML NK cells produced significantly increased IFNγ and MIP1α, compared with unstimulated NK cells (Fig. 4B). When we assessed polyfunctional responses, we observed 46% to 99% of donor NK cells are producing at least 1 cytokine/chemokine in response to tumor triggering (Fig. 4C). Previous work reported that ML differentiation improved effector functions of unlicensed NK cells (14). To investigate whether in vivo ML differentiation affects unlicensed NK-cell functionality, we examined the effector functions of single KIR+ donor NK cells in response to K562. NK cells that were unlicensed in the donor would be expected to produce fewer effector molecules compared with licensed NK cells. In most cases, the unlicensed KIR-expressing ML NK–cell subsets produced IFNγ, TNF, and MIP1α, and expressed CD107a similarly to the licensed KIR-expressing ML NK–cell subsets (Fig. 4D), consistent with the idea that unlicensed donor ML NK cells have enhanced function following ML differentiation in vivo. However, interpreting these in vivo data is complicated by the fact that each KIR is predicted to be licensed in the patient (Fig. 4D), leaving open the possibility that a licensing event also occurred after in vivo transfer.
NKG2A Is a Dominant Inhibitory Checkpoint on ML NK Cells
To determine whether any markers were associated with treatment failure (TF), we assessed the evaluable dose level of 3 patients with available cytometry by time of flight data [3 TF, 5 responders (R)] using Citrus (21). Citrus identified that NKG2A median expression on donor ML NK cells in the PB at D7 was significantly associated with TF (SAM, FDR < 0.01). Indeed, NKG2A median expression was significantly increased on donor NK cells transferred into patients with subsequent TF compared with those who achieved an IWG response (Fig. 5A and B). In contrast, NKG2A expression on baseline donor NK cells in this study was 8% to 76% with a median of 38% expression (Supplementary Fig. S2C), and was not associated with clinical outcomes. NKG2A is an inhibitory receptor that interacts with the nonclassic MHC-I molecule HLA-E (22, 23). To determine whether NKG2A inhibits ML NK–cell responses, control or ML NK cells from normal donor PBMCs were generated in vitro (Supplementary Fig. S5A) and stimulated with HLA-Elo or HLA-E+ primary AML, and analyzed for IFNγ production (Supplementary Fig. S5B and S5C). ML NK cells responding to HLA-E+ primary AML produce less IFNγ on a per-cell basis than ML NK cells triggered with HLA-Elo tumor targets, consistent with the in vivo association with TF. Next, K562 were generated that overexpress HLA-E to trigger control or ML NK cells (Fig. 5C and D). In these assays, ML NK cells produced more IFNγ than control NK cells in response to K562, as expected. However, ML NK cells, but not control NK cells, demonstrated reduced IFNγ production when stimulated with HLA-E+ K562 targets compared with HLA-E–negative targets (Fig. 5D). Indeed, the enhanced functionality typically observed in ML NK cells was completely abrogated when HLA-E was present on the targets (Fig. 5D). To determine whether NKG2A interactions with HLA-E also inhibited target killing, ML NK cells were incubated with HLA-E+ or HLA-E− K562 targets, and specific killing was measured in a flow-based killing assay (13). ML NK cells demonstrated a significantly reduced ability to kill HLA-E+ K562 targets compared with HLA-E− K562 (Fig. 5E).
Patient BM samples obtained prior to treatment on study were examined by mass cytometry, and unbiased FlowSOM was used to define cell populations within the tumor microenvironment (Supplementary Fig. S6A; Supplementary Table S4). Using this approach, HLA-E expression on these subsets was compared between responders and TF (Supplementary Fig. S6B–S6D). Although HLA-E expression on AML blasts was not associated with clinical outcomes, treatment failure was associated with increased HLA-E expression on mononuclear cells within the bone marrow (Supplementary Fig. S6B–S6D). These data suggest that increased NKG2A expression and HLA-E expression in the bone marrow negatively affected ML NK–cell responses in vivo.
NKG2A Is Transcriptionally Induced in ML NK Cells
To understand the mechanisms underlying this increased NKG2A expression by ML NK cells, qRT-PCR was performed for KLRC1 (the gene that encodes NKG2A) on in vitro control or differentiated ML NK cells over time (Fig. 5F). ML NK cells, but not control treated NK cells, induced KLRC1 mRNA. To determine whether NKG2A expression was occurring de novo, we sorted CD56dim CD16+ NKG2A+ and NKG2A− cells and examined NKG2A and Ki-67 expression on control and ML NK cells after 7 days in vitro (Fig. 5G). Here, NKG2A-negative cells induced NKG2A expression after ML NK–cell differentiation, but not control incubation. In addition, Ki-67 is increased in both NKG2A+ and NKG2A− NK ML NK–cell subsets, but to a greater extent in NKG2A+ ML NK cells (Fig. 5G). These data suggest that both an expansion of NKG2A+ NK cells and de novo NKG2A upregulation are responsible for increased NKG2A during ML NK–cell differentiation. Previous reports have implicated GATA3 as a transcription factor that regulates NKG2A expression (24). Indeed, the frequency of GATA3+ NK cells is specifically increased in ML NK cells, compared with control NK cells (Fig. 5H). In addition to GATA3, the transcription factor EOMES was increased in ML NK cells compared with control (Fig. 5I). Furthermore, EOMES and GATA3 coexpression corresponded with the NKG2Ahi cells, suggesting these transcription factors are important for the NKG2A upregulation within ML NK cells (Fig. 5J). Finally, GATA3 and EOMES are increased in both CD56bright and CD56dim subsets in response to ML differentiation (Supplementary Fig. S7A and S7B). Gene set enrichment analysis (GSEA) comparing expressed genes in control and ML NK cells revealed ML NK cells were significantly enriched in GATA3 target genes compared with control NK cells (Supplementary Fig. S7C). In similar assays, E4BP4, TCF7, TBET, BLIMP1, RUNX2, and RUNX3 median expression were similar between control and ML NK cells (Supplementary Fig. S7D and S7E). BACH2 mRNA expression was also similar between control and ML NK cells (Supplementary Fig. S7F). Together, these data support our previous findings that CD56bright and CD56dim NK cells both have the ability to differentiate into ML NK cells and demonstrate GATA3 and EOMES as specifically regulated by ML NK–cell differentiation (5, 10).
Eomes Regulates GATA3 and Promotes ML NK Cell–Enhanced Responses to Leukemia Targets
Because EOMES has a well-defined role in promoting T-cell memory (25), we hypothesized that it would be involved in memory formation in cytokine-activated NK cells. We used CRISPR/Cas9 to delete EOMES prior to ML differentiation (Fig. 5K–O). EOMES was reduced in ΔEOMES ML NK cells compared with wild-type (WT) control and WT ML NK cells (Fig 5L and M). The increase in GATA3 frequency during ML NK–cell differentiation was abrogated by loss of EOMES (Fig. 5L–M). Finally, increased IFNγ responses by ML NK cells compared with control NK cells was also partially abrogated by EOMES deletion (Fig. 5N and O), implicating EOMES as a critical transcription factor for ML NK–cell differentiation.
NKG2A Checkpoint Blockade or Elimination Restores ML NK–Cell Responses to AML
Because NKG2A interactions with HLA-E are inhibitory for ML NK cells, we hypothesized that abrogating this interaction would restore antileukemia responses (Fig. 6). Indeed, ML NK cell IFNγ production in response to HLA-E+ K562 was significantly increased by blocking with anti-NKG2A mAb (Fig. 6A and B), returning to similar levels as ML NK cells triggered with K562. HLA-E+ K562 killing by ML NK cells was also significantly increased in the presence of NKG2A checkpoint blockade compared with isotype mAb (Fig. 6C). NKG2A checkpoint blockade also enhanced ML NK–cell responses, but not control NK–cell responses, to multiple primary AML (Fig. 6D and E). To provide an orthogonal loss-of-function approach, we also used CRISPR/Cas9 to disrupt the NKG2A-encoding gene KLRC1 prior to control or ML NK–cell differentiation (Fig. 6F–J). After electroporation with KLRC1-targeting guide RNA (gRNA) and Cas9 mRNA, cells were rested in vitro for 24 hours and then control (IL15) treated or ML-cytokine (IL12, IL15, and IL18) activated. Cells were allowed to differentiate for 4 to 7 days in IL15, and NKG2A expression was assessed by flow cytometry. Using this approach, NKG2A expression on both control and ML NK cells was significantly reduced (Fig. 6G and H). WT or ΔNKG2A control or ML NK cells were stimulated with HLA-E− K562 or HLA-E+ K562 and IFNγ measured by flow cytometry. Control NK–cell responses were similar in response to K562 with or without HLA-E expression (Fig. 6J), whereas ML NK–cell IFNγ responses were reduced when triggered with HLA-E+ K562 (Fig. 6J). NKG2A deletion did not affect the enhanced ML NK–cell responses to K562 (HLA-E−), with WT and ΔNKG2A ML NK cells producing similar levels of IFNγ as expected. However, ΔNKG2A ML NK–cell responses were significantly increased compared with WT ML NK–cell responses against HLA-E+ K562 (Fig. 6J). To determine whether NKG2C interactions with HLA-E were driving this enhanced response, ΔNKG2A ML NK cells were stimulated with HLA-E+ K562 in the presence of α-NKG2C–blocking antibody, or isotype control (Supplementary Fig. S8). Blocking NKG2C on WT or ΔNKG2A ML NK cells had no impact on IFNγ production in response to HLA-E+ K562 leukemia targets. Overall, these data reveal that NKG2A is a critical inhibitor of ML NK–cell responses, but not control NK–cell responses, to AML targets.
CD8+NKG2A+ NK Cells Predict Treatment Failure and CD8+ NK Cells Do Not Proliferate in Response to IL12, IL15, and IL18
In addition to NKG2A, Citrus unexpectedly identified that CD8α expression on D7 in vivo differentiated ML NK cells was negatively associated with treatment outcomes (SAM, FDR < 0.01). No other markers were associated by Citrus with clinical outcomes. Although there was not a significant difference in CD8+ ML NK cells in vivo at day 7 (TF: Mean 1322 cells/mL ± 1,158 cells/mL SD; R: Mean 618.9 cells/mL ± 701.3 cells/mL SD; unpaired t test P = 0.31), median CD8α expression was significantly increased on donor NK cells in the TF patients compared with the responding patients (Fig. 7A and B). Individually, NKG2A or CD8α expression at BL was not associated with clinical responses. To determine whether NKG2A and CD8 coexpression at BL was associated with patient outcomes, we examined the frequency of NKG2A+CD8+ NK cells in purified NK-cell products (Fig. 7C–E). The frequency of NKG2A+CD8+ NK cells in the product was significantly associated with response to treatment (Fig. 7D and E), with increased frequencies of NKG2A+CD8+ NK cells occurring with treatment failure. Consistent with another study (26), CD8α was expressed on approximately 23% of CD56bright and approximately 35% on CD56dim NK cells with a high interindividual variability (Supplementary Fig. S9A and S9B). CD8 was not specifically induced in vivo in response to ML NK–cell differentiation (Supplementary Fig. S2C), but was increased in vitro in both control and ML NK cells. This implicates IL15 signaling in regulating CD8 upregulation in vitro (Supplementary Fig. S9C). The majority of CD8+ NK cells are CD8αα+, with a minor subset expressing CD8αβ (Supplementary Fig. S9D). ML differentiation does not alter these frequencies, relative to control or baseline (Supplementary Fig. S9E). Finally, CD8+ NK cells do not express CD3 or other T-cell receptors (TCR) and represent a subpopulation of NK cells which are distinct from T cells, including iNKT cells (Supplementary Fig. S9F and S9G).
To explain the negative association of CD8 with patient outcomes, we hypothesized that CD8α+ NK cells were not optimally responding to IL12, IL15, and IL18 activation. To test signaling competency, freshly isolated NK cells were stimulated with IL12, IL15, and IL18 for 0 to 120 minutes, and phosphorylation of proximal cytokine signaling molecules STAT4, ERK, STAT5, p38, and p65 was measured (27, 28). For both CD8α+ and CD8α− NK cells, similar phosphorylation was observed relative to the unstimulated condition (P > 0.05; one sample t test, test value = 1; Fig. 7F). No differences in cytokine receptor signaling were observed between CD8+ and CD8− NK cells (Fig. 7F). Next, we compared the ability of CD8α+ and CD8α− NK cells to proliferate in response to ML-cytokine activation. Sorted CD8α+ and CD8α− NK cells were cell trace violet (CTV)–labeled, activated with IL12, IL15, and IL18, washed after 16 hours, and allowed to differentiate. Proliferation was assessed after 6 days. CD8α+ NK cells divided less compared with CD8− NK cells, and expression of Ki-67 was reduced, both indicating significantly inferior proliferation (Fig. 7G and H). We hypothesized that the larger number of CD8α+ donor NK cells infused into TF patients were not expanding to the same extent as the predominantly CD8α− donor NK cells in responding patients. Consistent with this, the amount of Ki-67 in donor NK cells was significantly associated with treatment outcomes (Fig. 7I). However, median Ki-67 expression between NKG2A+ and NKG2A− ML NK cells in TF and responders was not significantly different (Supplementary Fig. S10A and S10B). Patients with donor ML NK cells with lower Ki-67 expression failed treatment. However, in this small sample size, total donor NK–cell numbers in the PB at a single time point (7 days) measured post-infusion did not directly correlate with response (Supplementary Fig. S1D). Although these data provide insight into the mechanisms underlying treatment failure, they include a single time point, and further studies are needed. These in vivo data are consistent with the in vitro observations that CD8α+ ML NK cells do not proliferate as strongly as CD8α− ML NK cells, and may explain the inferior clinical responses.
To evaluate the cell-intrinsic role for CD8α on ML NK–cell functionality, ΔCD8a ML NK cells were compared with WT ML NK cells in in vitro functional assays (Fig. 7J–M). Using this approach, CD8α expression was reduced on ΔCD8a ML NK cells compared with WT ML NK cells (Fig. 7K). K562 target killing by ΔCD8a ML NK cells was slightly, yet significantly, decreased compared with WT ML NK cells (Fig. 7L). Furthermore, in response to cytokines and tumor targets, IFNγ, TNF, and CD107a were similar between ΔCD8a ML NK cells compared with WT ML NK cells (Fig. 7M). These data suggest that CD8α does not impair ML NK–cell responses to prototypical stimuli, but further studies are warranted.
Here we used multidimensional immune correlative phenotyping by mass cytometry to identify the in vivo differentiated human ML NK–cell phenotype, which was distinct from cytokine-activated and conventional NK cells. ML NK cells were safe, expanded in vivo, and induced IWG responses in 67% (47% CR/CRi) of evaluable patients. We demonstrated that NKG2A is transcriptionally regulated in ML NK cells and represents a critical induced checkpoint for cytokine-induced ML NK–cell responses, associating with treatment failure in patients with AML treated with donor ML NK cells. Although little is known about the role of CD8α on NK cells, we identified a new association with CD8α+ NK cellular therapy and inferior patient outcomes, likely due to their inability to robustly proliferate in response to combined cytokine activation.
NKG2A is a C-type lectin inhibitory receptor that heterodimerizes with CD94 and recognizes the nonclassic class I-MHC HLA-E, resulting in ITIM-mediated NK-cell inhibition (22). NKG2A expression on baseline donor NK cells was not associated with clinical outcomes. Furthermore, the importance of NKG2A for conventional (naïve) or control (low-dose IL15 supported) NK-cell response to HLA-E+ tumor targets was not observed, suggesting that NKG2A is a minor inhibitory receptor on conventional NK cells. Here we demonstrated that NKG2A is an inducible checkpoint molecule on cytokine-induced ML NK cells and is a critical inhibitor of ML NK–cell responses against HLA-E+ tumor targets. NKG2A can be transcriptionally induced during ML differentiation. However, both enhanced proliferation of NKG2A+ NK cells and de novo NKG2A upregulation are likely operative in regulating overall NKG2A expression during ML NK–cell differentiation. Our previous report showed that ML NK cells are not inhibited through the regular rules of inhibitory KIR to KIR-ligand interactions (5). Data presented here indicate that ML NK cells are instead primarily inhibited through NKG2A, and further studies examining the role for NKG2A and immune tolerance in the setting of cytokine activation and inflammation are warranted. Moreover, the fraction of NKG2A+ NK cells does not appear as important as the per-cell NKG2A expression, because nearly all donor ML NK cells expressed NKG2A, but only those donors with supraphysiologic expression were associated with treatment failure. However, HLA-E expression was overall increased in the treatment-failure tumor microenvironment, suggesting both NKG2A supraexpression and increased HLA-E contribute to resistance to ML NK cellular therapy. Although future work will elucidate the mechanisms of supraphysiologic expression by some donors, we showed that NKG2A is transcriptionally induced after IL12, IL15, and IL18 activation and is associated with a concomitant increases in GATA3, a known regulator of NKG2A, as well as EOMES, which is important for establishing a central memory phenotype in CD8+ T cells (25). The interplay between these two transcription factors and how they establish ML NK–cell differentiation program is unclear, but this is an active area for further investigation. Translationally, blockade of NKG2A or gene editing of KLRC1 represent exciting potential strategies to improve on ML NK cellular therapy. Preclinical studies utilizing these approaches are ongoing and critical for establishing proof-of-principle needed to move this strategy into the clinic. Indeed, recent reports have demonstrated that combination anti-NKG2A and anti–PD-L1 mAb controlled tumor growth in murine models of B- and T-cell lymphoma, as well as established the safety of anti-NKG2A mAb for patients with squamous cell carcinoma of the head and neck (29), supporting the feasibility of translating our findings to the clinic.
There are limited reports of the role of CD8 on human NK cells, which is normally expressed as a CD8α homodimer, leaving CD8 receptor biology unclear in this context. Previous reports indicate that CD8+ NK cells have enhanced cytotoxicity and undergo reduced activation induced–apoptosis (26, 30). In addition, the presence of CD8+ NK cells has been associated with slower HIV-1 progression in chronically infected individuals (31). CD8α expression on NK cells was reported to contribute to KIR3DL1 signaling (32). Stronger CD8 interactions with MHC-I were hypothesized to improve licensing by enhancing KIR–KIR-ligand interactions, which is one possible mechanism for increasing NK-cell functionality (32). However, our study implies a negative role for CD8α+ NK cells in adoptive NK-cell immunotherapy. Potentially explaining this conundrum, we show that CD8α+ NK cells have reduced proliferative capacity compared with CD8α− NK cells. Indeed, there are some studies examining the role of CD8αα in limiting T-cell responses (33). CD8αβ is a well-characterized coreceptor for TCR interactions with MHC-I, and enhances TCR signaling (34). However, studies have implicated that CD8αα inhibits T-cell responses and that CD8αα may act as an inhibitory molecule in nonclassic T-cell subsets (33, 35). Although CD8α+ NK cells exhibit reduced proliferation, it remains unclear if CD8α is a marker of a differentiated, terminal phenotype with limited replicative capacity, or if CD8 is directly inhibiting proliferation in vivo. The negative association of CD8 expression with clinical outcomes following NK-cell immunotherapy identifies the importance of understanding its role on NK cells, as well as the potentially distinct biology of CD8α+ NK cells from CD8α− NK cells.
Here we report the first high-dimensional characterization of in vivo differentiated ML NK cells in the context of the final phase I clinical data, demonstrating the safety and efficacy of ML NK cells to treat patients with rel/ref AML. ML NK cells are well tolerated and did not cause GVHD, CRS, or ICANS, nor ≥ grade 3 adverse events related to ML NK–cell infusion. The observed CR/CRi rate of 47% is remarkable for a population of older adults with rel/ref AML, and is consistent with our initial report. Although the duration of response was relatively short (2–6 months) for most patients, one patient, who became HCT-eligible, had a durable response that persisted after allogeneic HCT. This strategy as a “bridge to HCT” is being tested in our phase II cohort for rel/ref AML (NCT01898793). Based upon their ability to ignore inhibitory KIR ligation, we are also evaluating the effectiveness of ML NK cells against solid tumors (36), and expanding their repertoire against NK-resistant tumors using bispecific triggering (37) as well as chimeric antigen receptor engineering (38). With our extensive immune correlative studies, we have identified NKG2A as a targetable checkpoint that could be combined with ML NK–cell adoptive therapy in future trials. Moreover, a new strategy for donor NK-cell donor selection based on NKG2A+CD8+ NK cell frequency was discovered. We have reported in preliminary form multiple clinical trials at Washington University utilizing ML NK–cell adoptive immunotherapy, including as a bridge to HCT (NCT01898793), as augmentation of MHC-haploidentical HCT with same-donor ML NK cells (NCT02782546), and as therapy for relapse after allogeneic HCT (NCT03068819; refs. 39, 40). As evidenced by this study, multidimensional immune correlates will be performed to understand if NKG2A and CD8 can predict patient outcomes in other ML NK–cell clinical contexts. Thus, this study highlights the importance of multidimensional immune monitoring to identify mechanisms of response and resistance following NK-cell therapy.
Patients treated on an open-label, nonrandomized, phase I dose-escalation trial (NCT01898793) are included in this study. Prior to any study-related testing or treatment, written informed consent was obtained from all patients under a Washington University School of Medicine Institutional Review Board (IRB)–approved clinical protocol, and all studies were conducted in accordance with the Declaration of Helsinki. The initial escalation was previously reported (5). Briefly, patients were treated with fludarabine/cyclophosphamide between days -7 and -2 for immunosuppression, followed on day 0 by allogeneic donor IL12, IL15, and IL18 activated NK cells. Patients in dose level 3 received the maximum NK cells that could be generated (capped at 1 × 107 cells/kg). After donor NK-cell transfer, rhIL2 was administered subcutaneously every other day for a total of 6 doses. Donor NK cells were purified from a nonmobilized apheresis product using CD3 depletion followed by CD56+ selection (CliniMACS device). Purified NK cells were activated with IL12 (10 ng/mL), IL15 (50 ng/mL), and IL18 (50 ng/mL) for 12 hours under current GMP conditions.
Samples were obtained from the PB (day 7, 8, and 14 after infusion) and BM (screening, day 8 and 14 after infusion). Clinical responses were defined by the revised IWG criteria for AML (15). All patients provided written informed consent before participating and were treated on a Washington University IRB-approved clinical trial (Human Research Protection Office #201401085).
Reagents and Cell Lines
Anti-human mAbs were used for flow and mass cytometry (Supplementary Methods, Supplementary Table S4). Endotoxin-free, recombinant human (rh) IL12 (BioLegend), IL15 (Miltenyi Biotec), and IL18 (InVivo Gen) were used in these studies. K562 cells (ATCC, CCL-243) were obtained in 2008, viably cryopreserved, and maintained for <2 months at a time in continuous culture according to ATCC specifications. K562 cells were authenticated in 2015 using single-nucleotide polymorphism analysis and were found to be exactly matched to the K562 cells from the Japanese Collection of Research Bioresources, German Collection of Microorganisms and Cell Cultures (DSMZ), and ATCC databases (Genetic Resources Core Facility at Johns Hopkins University). HLA-E+ K562 were a gift from Dr. Deepta Bhattacharya (Washington University School of Medicine). These cells were generated using the AAVS1-EF1a donor plasmid containing the coding sequence for human HLA-E. The K562 cells were electroporated using a Bio-Rad Gene Pulse electroporation system. HLA-E+ cells were sorted to >98% purity.
NK-Cell Purification and Cell Culture
Normal donor PBMCs were obtained from anonymous healthy platelet donors. NK cells were purified using RosetteSep (StemCell Technologies; routinely >95% CD56+ CD3−). ML and control NK cells were generated as described previously (5). Cells were maintained in 1 ng/mL IL15, with media changes every 2 to 3 days. For proliferation assays, cells were labeled with 2.5 μmol/L CTV (Life Technologies) for 15 minutes at 37°C.
Patients with newly diagnosed AML provided written informed consent under the Washington University IRB–approved protocol (Human Research Protection Office #2010–11766) and were the source of primary AML blasts for in vitro stimulation experiments. Patient PBMCs and BMMCs were isolated by Ficoll-Paque PLUS (GE Health) centrifugation and immediately used in experiments. For assessing HLA-E expression in patient BM, viably frozen cells were thawed and stained immediately using mass cytometry.
Functional Assays to Assess Cytokine Production
For patient stimulation assays, PBMCs or BM cells were stimulated in a standard functional assay (5). Cells were stimulated with K562 leukemia targets (5:1 effector-to-target ratio). Functionality was measured using mass cytometry as described previously (5). For each patient sample, a normal donor sample was thawed and stimulated with K562 and used as a control for the functional assay. For in vitro differentiated NK-cell functional assays, control and ML NK cells were harvested after a rest period of 5 to 7 days to allow memory-like NK-cell differentiation to occur. Cells were incubated with K562 ± HLA-E or freshly thawed primary AML blasts. All cytokine secretion assays were performed for 6 hours in the presence of GolgiPlug/GolgiStop (BD Biosciences) for the final 5 hours. Anti-CD107a antibodies were included in the well at the beginning of the assay to measure degranulation. For antibody blocking experiments, 10 μg/mL anti-NKG2A (Z199), anti-NKG2C (134522), or isotype control (IgG) were added directly to the wells at the beginning of the assay.
Flow-Based Killing Assay
Flow-based killing assays were performed by coincubating ML or control NK cells with carboxyfluorescein succinimidyl ester (CFSE)–labeled K562 ± HLA-E for 4 hours and assaying 7-aminoactinomycin D (7AAD) uptake as described previously (13).
Cells were resuspended in TRIzol and RNA extracted using Zymo Directzol RNA microPrep according to the manufacturer's directions. cDNA was generated using Life Technologies High Capacity cDNA Reverse Transcription Kit (4368814) according to the manufacturer's instructions. Real-time qPCR was performed using ABI Master Mix with TaqMan Gene Expression Assay, Hs00970273_g1 KLRC1, Hs00935338_m1 BACH2, and Hs01060665_g1 ACTB, according to the manufacturer's instructions. Samples were analyzed on StepOnePlus Real-Time PCR system (Applied Biosystems). Relative quantification was determined by ΔΔ threshold cycle method, by normalizing KLRC1 or BACH2 to ACTB (β-actin).
RNA Sequencing and GSEA
Cells were stored in TRIzol at −80°C until RNA isolation using the Direct-zol RNA MicroPrep Kit (Zymo Research). NextGen RNA sequencing was performed using an Illumina HiSeq 2500 sequencer. RNA-sequencing reads were then aligned to the Ensembl release 76 primary assembly with STAR version 2.5.1a. Gene counts were derived from the number of uniquely aligned unambiguous reads by Subread:featureCount version 1.4.6-p5. Analysis of sequencing data was performed using Phantasus, a browser-based gene expression analysis software. Genes were log2 normalized and filtered to remove duplicate reads and low-expressed genes. Differential expression analysis was performed on the top 12,000 expressed genes using the LIMMA package to analyze differences between conditions. GSEA was performed using the Harmonizome database of GATA3 target genes (41).
Flow Cytometric Analysis and Sorting
Cell staining was performed as described previously (5), and data were acquired on a Gallios flow cytometer (Beckman Coulter) and analyzed using FlowJo (Tree Star) software. Dead cells were stained using Zombie reagent (BioLegend) according to the manufacturer's instructions, except for phospho-flow. eBio Fix/Perm was used for all intracellular staining except phospho-flow. For phospho-flow, cells were stimulated with IL12, IL15, and IL18 for 0, 15, 60, and 120 minutes. Cells were fixed in 1% prewarmed formalin and permeabilized using 100% ice-cold methanol. Cells were washed three times and stained overnight, as described previously (42). CD8+ and CD8− NK cells were sorted to >99% purity using FACSAria II Cell Sorter (BD Biosciences) or purified using Automacs column (Miltenyi Biotec). CD56dim CD16+ NKG2A+ and NKG2A− NK cells were sorted using FACSAria II cell sorter (BD Biosciences).
Mass cytometry was performed on freshly isolated patient PB or BM cells as described previously (5, 43), or on thawed BM samples. Cells were stained for live/dead using cisplatin, and surface staining performed at 4°C for 15 minutes. Cells were washed, and fixed with eBio Fix/Perm overnight at 4°C. Cells were stained using intracellular antibodies at 4°C for 15 minutes. Cells were washed and resuspended in PBS containing 1% paraformaldehyde and stored until all samples were stained. Once all samples were stained, they were washed and barcoded according to the manufacturer's instructions. Data were collected on a Helios mass cytometer (Fluidigm) and analyzed using Cytobank (44). Data were analyzed using previously described methods (45). KIR diversity (KIR2DL1, KIR2DL2/2DL3, KIR3DL1, KIR2DS4, KIR2DL5) was assessed at baseline and after in vivo differentiation as described previously (5). For each patient sample, a normal donor PBMC sample was thawed and stained, providing a staining control at day 0, day 7, and day 8 of patient sample staining. These were used for quality control. Staining for each marker was confirmed to be consistent across the days on which the samples were stained using the same master mix. Comparisons were also made (Student t test or Mann–Whitney) between matched normal donors stained on D0 and D7. For all markers, there were not significant differences in median expression between normal donors thawed and stained on D0 or D7. These control samples served to increase our confidence that changes we observed in our patient samples represented biologically relevant changes and did not reflect technical issues that could arise from these assays. Citrus was performed assessing median with default settings, using the same clustering channels as the viSNE (Fig. 1; Supplementary Table S4), with the addition of CD8. To define the in vivo ML differentiated phenotype (Fig. 1.) and the lymphocyte subsets, we used all patients with mass cytometry data available, including CIML026 who expired prior to day 28, post-infusion, and was thus not evaluable for response (male, aged 77, diagnosed with AML, treated at dose level 3, 2 prior therapies, 7% blasts prior to treatment). CIML025 was evaluable but was treated at a dose level 2 and was excluded from the Citrus analyses (Figs. 5 and 7). Phenotypic intracellular markers for CIML028 were not assessed. For CIML020, an inappropriate amount of Ki-67 antibody was used, making the median expression value an outlier [ROUT (Q = 1%)]; however, percent positive could still be reliably assessed. For functional assays, CD107a was omitted from the functional assay for CIML026. For patient CIML027, due to limiting cells in the PB, BM donor NK cells were used to assess licensing. CIML028 had only had cells available for a functional assay at D14.
CRISPR/Cas9 Gene Editing
NK cells were purified from normal donors and rested overnight in 1 ng/mL IL15. Cells were washed with PBS, two times to remove serum, and resuspended in MaxCyte EP buffer plus CAS9 mRNA (Trilink; ref. 46). Next, gRNA [NKG2A: AACAACUAUCGUUAACCACAG (Trilink, Synthego); EOMES: AACCAGUAUUAGGAGACUCU (IDT); or CD8A: GACUUCCGCCGAGAGAACGA (IDT); 2 × 108 cells/mL] or no gRNA (control) was added to the cells, which were then electroporated in a Maxcyte GT using the WUSTL-2 setting in an OC-100 processing assembly. Cells were removed from the OC-100 and incubated for 10 minutes at 37°C. Prewarmed media containing 3 ng/mL IL15 was added and cells rested for 24 hours. Cells were then control treated (3 ng/mL IL15) or cytokine activated (IL12/15/18) for 16 to 18 hours, as described previously (10). Cells were washed three times and maintained in complete RPMI supplemented with 10% Human AB serum and 3 ng/mL IL15. Media changes were performed every 2 to 3 days. Gene-editing efficiency was determined as described previously (Supplementary Fig. S11A and S11B; ref. 47).
Before statistical analyses, all data were tested for normal distribution (Shapiro–Wilk). If data were not normally distributed, the appropriate nonparametric tests were used (GraphPad Prism v8), with all statistical comparisons indicated in the figure legends. Uncertainty is represented in figures as SEM, except where indicated. All comparisons used a two-sided α of 0.05 for significance testing.
Data and Materials Availability
The RNA-sequencing data are accessible within Gene Expression Omnibus (GEO) under accession code GSE154694.
Disclosure of Potential Conflicts of Interest
M.M. Berrien-Elliott reports other from Maxcyte (speaking and travel), other from Fluidigm (travel), and other from Wugen (consulting) during the conduct of the study; in addition, M.M. Berrien-Elliott has a patent for 017001-PRO1 pending and a patent for 62/963,971 pending. C.C. Cubitt reports other from Pionyr Immunotherapeutics [cash compensation received from selling of personal stocks (acquired after employment at Pionyr) after Pionyr entered into a purchase agreement] outside the submitted work. J.A. Foltz reports grants from American Association of Immunologists during the conduct of the study; grants from NHLBI (T32HL007088-44) outside the submitted work; in addition, J.A. Foltz has a patent for 62/623,682 pending, licensed, and with royalties paid from Kiadis. M.L. Cooper reports personal fees from Wugen Inc. (consulting, royalties, equity), NeoImmuneTech (royalties), and RiverVest Venture Partners (consulting) outside the submitted work. C.N. Abboud reports other from Ryvu (clinical research phase I study), other from AlloVir (clinical research in respiratory viral infections post transplant), Forty Seven Inc (clinical trial in MDS), Abbott Labs (stock ownership; 5%), AbbVie (stock ownership; 5%), BMS (stock ownership; 5%), Gilead (stock ownership; 5%), and other from Johnson & Johnson (stock ownership; 5%) outside the submitted work. G.L. Uy reports personal fees from Jazz, Genentech, Astellas, and Pfizer outside the submitted work. M.A. Jacoby reports personal fees from Takeda outside the submitted work. M.A. Schroeder reports personal fees from Amgen, Astellas, Dova Pharmaceuticals, FlatIron Inc, GSK, Gilead Sciences Inc, Incyte, Novo Nordisk, Partners Therapeutics, Pfizer, Sanofi Genzyme, Abbvie, Merck, and Takeda outside the submitted work. T.A. Fehniger reports grants from NIH/NCI (R01CA205239, P50CA171063), Leukemia and Lymphoma Society (translational research award), V Foundation for Cancer Research (translational award), American Association of Immunologists (intersect Fellowship), and Siteman Cancer Center at Washington University School of Medicine (Siteman Investment Program) during the conduct of the study; grants from Children's Discovery Institute (interdisciplinary award); personal fees and other from Nkarta (consultant), Indapta (scientific advisory board), Gamida Cell (consultant); Nektar (consultant), Kiadis (scientific advisory board), Wugen (consultant), other from Affimed (research funding); and other from Altor BioSciences (research funding), Compass Therapeutics (research funding), and HCW Biologics (research funding) outside the submitted work; in addition, T.A. Fehniger has a patent for 15/983,275 pending (Compositions, methods of using, and methods of making a NK cell-based therapy.), a patent for 62/963,971 pending "Provisional; Compositions and methods of using CD8 inhibiting agents and methods and assays for detecting CD8 in cells.", and a patent for PCT/US2019/060005 pending [chimeric antigen receptor memory-like (CARML) NK cells and methods of making and using the same]. No potential conflicts of interest were disclosed by the other authors.
M.M. Berrien-Elliott: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A.F. Cashen: Formal analysis, supervision, funding acquisition, investigation, visualization, writing-original draft, writing-review and editing. C.C. Cubitt: Formal analysis, investigation, visualization, writing-review and editing. C.C. Neal: Investigation, writing-review and editing. P. Wong: Formal analysis, investigation, visualization, writing-review and editing. J.A. Wagner: Validation, writing-review and editing. M. Foster: Investigation, writing-review and editing. T. Schappe: Formal analysis, investigation, writing-review and editing. S. Desai: Investigation, writing-review and editing. E. McClain: Investigation, writing-review and editing. M. Becker-Hapak: Supervision, investigation, writing-review and editing. J.A. Foltz: Software, formal analysis, investigation, writing-review and editing. M.L. Cooper: Resources, methodology, writing-review and editing. N. Jaeger: Investigation, writing-review and editing. S.N. Srivatsan: Software, writing-review and editing. F. Gao: Formal analysis, writing-review and editing. R. Romee: Writing-review and editing, provided clinical care to patients enrolled on trial. C.N. Abboud: Writing-review and editing, provided clinical care to patients enrolled on trial. G.L. Uy: Writing-review and editing, provided clinical care to patients enrolled on trial. P. Westervelt: Writing-review and editing, provided clinical care to patients enrolled on trial. M.A. Jacoby: Writing-review and editing, provided clinical care to patients enrolled on trial. I. Pusic: Writing-review and editing, provided clinical care to patients enrolled on trial. K.E. Stockerl-Goldstein: Writing-review and editing, provided clinical care to patients enrolled on trial. M.A. Schroeder: Writing-review and editing, provided clinical care to patients enrolled on trial. J. DiPersio: Writing-review and editing, provided clinical care to patients enrolled on trial. T.A. Fehniger: Conceptualization, resources, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.
We would like to thank our patient volunteers and the HCT/leukemia physician and nursing teams who care for them at the Washington University School of Medicine (WUSM). We acknowledge support from the Genome Engineering and iPSC Center (GEiC) at WUSM for gRNA validation services, as well as the Siteman Flow Cytometry Core (Bill Eades), Immune Monitoring Lab (Stephen Oh), and Biological Therapy Core Facility of the Siteman Cancer Center. We thank the Genome Technology Access Center in the Department of Genetics at Washington University School of Medicine for help with genomic analysis. The Center is partially supported by NCI Cancer Center Support Grant #P30 CA91842 to the Siteman Cancer Center and by ICTS/CTSA Grant# UL1TR002345 from the National Center for Research Resources (NCRR), a component of the NIH, and NIH Roadmap for Medical Research. This publication is solely the responsibility of the authors and does not necessarily represent the official view of NCRR or NIH. Figure 1A illustration is by Astrid Rodriguez Velez and Anne Robinson in association with InPrint at Washington University in St. Louis. This work was supported by NIH/NHLBI (T32HL00708843, to J. Wagner and P. Wong); NIH/NCI (F32CA200253, to M. Berrien-Elliott), K12CA167540 (to M. Berrien-Elliott), P50CA171063 (to M. Berrien-Elliott, A. Cashen, T. Fehniger) and R01CA205239 (to T. Fehniger). Additional funding was from the Leukemia and Lymphoma Society (to T. Fehniger), V Foundation for Cancer Research (to T. Fehniger), The American Association of Immunologists Intersect Fellowship Program for Computational Scientists and Immunologists (to J. Foltz and T. Fehniger), and the Children's Discovery Institute (CDI) at WUSM (to T. Fehniger).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.