New data generated in this study were obtained from frozen peripheral blood mononuclear cells (PBMCs) of patients (N=38) enrolled in the HEMATOBIO cohort (NCT02320656). Blood collection, PBMCs extraction, and storage were performed by the Paoli-Calmettes Tumor Bank, which operates under authorization #AC-2007-33 granted by the French Ministry of Research. Patients with newly diagnosed AML, except acute promyelocytic leukemia, were eligible for study participation. All patients had received standard chemotherapy. The cohort was divided according to the frequency of CD56^neg CD16⁺ NK cells with a threshold of 10% as previously defined (8). The clinical information of all patients is summarized in Table 1 . Healthy volunteers (HV, N = 16) were recruited from the French Blood Agency (EFS). Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki. The study was reviewed and approved by the Institutional Review Boards of Paoli-Calmettes Institute.

Cryopreserved PBMCs from N=4 AML patients and N=5 age-matched HV were thawed in a 37°C water bath, resuspended in RPMI-1640 (Gibco, Grand Island, NY, USA) supplemented with 10% of heat inactivated fetal bovine serum (FBS) (Gibco), centrifugated for 5 min at 1,500 rpm and incubated in a 37°C, 5% CO2 incubator for 30 min. After resting, cells were filtered on a 30 µm pre-separation filter (Miltenyi), centrifugated and incubated in 1X PBS (Gibco) with the viability marker Live/Dead Fixable Aqua (Thermo Fisher Scientific, Waltham, MA, USA) for 25 min on ice in the dark. After washing in 1X PBS and centrifugation for 5 min at 1,500 rpm twice, cells were incubated with BV605 anti-CD56 (clone NCAM16.2), BV711 anti-CD16 (clone 3G8), BV785 anti-CD45 (cloneHI30), FITC anti-CD13 (clone SJ1D1), FITC anti-CD33 (clone HIM3-4), FITC anti-CD34 (clone 581), PE-CF594 anti-CD3 (clone UCHT1) in Brilliant Stain Buffer (BD Biosciences, Franklin Lakes, NJ, USA) for 30 min on ice in the dark. Before acquisition, cells were passed through a 35 µm cell strainer. Cell sorting of CD56^bright, CD56^dim CD16^-, CD56^dim CD16⁺, and CD56^neg CD16⁺ NK cell subsets was performed on a FACS Aria III (BD Biosciences) following the gating strategy outlined in Supplementary Figure S1 . All used antibodies are listed in Supplementary Table S1 . After cell sorting, total RNA was immediately isolated with a RNeasy Plus Micro Kit (Qiagen, Hilden, Germany) according to manufacturer’s instructions. High-quality RNA was eluted in 14 µl RNase-free water.

Samples from patients and HV were respectively pooled, at the same total RNA concentration, according to NK cell subtype prior to sequencing. RNA integrity was assessed on a Fragment Analyzer (Agilent, Santa Clara, CA, USA). Library preparation, RNA Capture Sequencing and were performed by IntegraGen (Evry, France). Briefly, libraries were prepared using NEBNext^® Ultra™ II Directional RNA Library Prep Kit for Illumina (NEB, Ipswich, MA, USA) according to supplier recommendations. The RNA was capture with the Twist Human Core Exome (Twist Bioscience, San Francisco, CA, USA) + IntegraGen Custom V2 Enrichment Systems. Sequencing was performed on an Illumina NovaSeq platform and 2 × 100 bp paired-end reads were generated. RNA and sequencing quality control metrics are summarized in Supplementary Table S2 .

Spectral flow cytometry, Cell cloning and functional assays, RNA-sequencing analysis, Mass cytometry analysis and Statistical Analysis can be found in Supplementary Materials and Methods .

Our team made the first description of CD56^neg CD16⁺ NK cells expansion in AML patients (8). Therefore, we sought to confirm that CD3^- CD56^neg CD16⁺ cells identified through mass cytometry belong to NK cell. To this end, we ran Cytosplore^+HSNE (16) on our derivation cohort data (8) to perform automatic clustering. We embedded 872 796 CD45⁺ cells following the gating strategy outlined in Supplementary Figure S2 and isolated from peripheral blood of N=10 HV and N=22 AML patients. Cytosplore^+HSNE identified 6 distinct clusters among which NK cells and myeloid/leukemic blasts clusters. Importantly, 88% of CD3^- CD56^neg CD16⁺ NK cells clustered among NK cell cluster while 12% clustered among leukemic blasts cluster indicating a strong similarity between CD56^neg CD16⁺ NK cells and conventional NK cells, i.e. CD56^dim CD16⁺,CD56^dim CD16^- and CD56^bright NK cells (p<0.0001) ( Figures 1A, B ).

Eomes and T-bet transcription factors are crucial to sustain NK cell identity and function (17). Using spectral flow cytometry, we assessed the frequencies of Eomes⁺ and T-bet⁺ cells among CD3⁺ cells, CD56^neg CD16⁺ and conventional NK cells from PBMCs of HV (N=16) and AML patients (N=7) with CD56^neg CD16⁺ NK cells expansion at diagnosis (Expanded group). In the HV group, we observed similar frequency of Eomes⁺ cells among CD56^neg CD16⁺, CD56^dim CD16^- and CD56^bright NK cells whereas CD3⁺ cells and CD56^dim CD16⁺ NK cells showed lower expression of Eomes compared to CD56^neg CD16⁺, respectively p<0.0001 and p=0.0007. In the Expanded group, frequency of Eomes⁺ CD56^neg CD16⁺ cells did not differ from the frequencies of conventional NK cells and was significantly increased compared to CD3⁺ cells, p=0.0004 ( Figure 1C , left panel). Besides, frequencies of T-bet⁺ CD56^neg CD16⁺ NK cells from both HV and Expanded groups matched the frequencies of conventional NK cells and were significantly increased compared to CD3⁺ cells, respectively p<0.0001 and p=0.0007 ( Figure 1C , right panel).

Next, we investigated whether CD56^neg CD16⁺ NK cells express NK cell markers and maintained NK cell killing abilities in vitro. We performed limiting dilution NK cell cloning from whole blood of HV. PBMCs were isolated using density gradient centrifugation and CD56^neg CD16⁺ and CD56^dim CD16⁺ NK cells were FACS-sorted before being plated at a concentration of 1, 10 or 100 cells/well in RPMI supplemented with 10% of heat inactivated FBS. Irradiated K562 cells were used as feeders. IL-2 (100U/mL), IL-15 (10ng/mL) and IL-21 (5ng/mL) were added every two days and medium was changed. After 24 days of culture, cells were harvested, phenotyped and functionality was assessed ( Figure 1D ). We observed that expression levels of CD56 and NKG2A were significantly increased in CD56^neg CD16⁺ NK cells after 24 days of culture, respectively p=0.0024 and p=0.0003. Importantly, CD56^neg CD16⁺ NK cells and CD56^dim CD16⁺ NK cells expressed similar levels of CD56 and NKG2A after 24 days in culture ( Figure 1E ). In addition, CD56^neg CD16⁺ and CD56^dim CD16⁺ NK cells exhibited similar production capacities of IFN-γ and TNF-α after 24 days of culture ( Figure 1F , left and middle panels). Finally, CD56^neg CD16⁺ NK cells expressed higher levels of CD107a/b, p=0.0057, ( Figure 1F , right panel) and showed similar specific lysis against K562 cells than CD56^dimCD16⁺ NK cells after 24 days of culture ( Figure 1G ). Together these data support that CD3^- CD56^neg CD16⁺ cells represent a distinct NK cell subset able to recover CD56 expression in vitro where they display unaltered NK cell functions.

We demonstrated that CD56^neg CD16⁺ NK cells expansion at diagnosis was associated with adverse clinical outcome in AML (8). To validate our previous results, we included N=16 HV, N=31 AML patients without CD56^negCD16⁺ NK cells expansion at diagnosis (Non-Expanded group) and N=7 patients in the Expanded group within our validation cohort. Patients’ characteristics are summarized in Table 1 . PBMCs at diagnosis were analyzed using spectral flow cytometry and NK cells subsets were gated following the gating strategy outlined in Supplementary Figure S3 . We previously defined a threshold of 10% of CD56^neg CD16⁺ NK cells among total NK cells to discriminate between patients. As expected, the threshold could optimally classify samples from the validation cohort in the Non-Expanded or Expanded groups ( Figure 2A ). Indeed, the Expanded group displayed significant CD56^neg CD16⁺ NK cells expansion compared to the HV and Non-Expanded groups, respectively p=0.0292 and p<0.001 ( Figure 2B , left panel). Frequencies of CD56^neg CD16⁺ NK cells in the Expanded group ranged from 11.8% to 75.6% ( Supplementary Table S3 ). Notably, the Expanded group did not show any increase of total NK cells ( Figure 2C , left panel). Besides, the Expanded group had significantly less CD56^dim CD16⁺, CD56^dim CD16^- and CD56^bright NK cells than the Non-Expanded (p=0.0150, p=0.0007 and p=0.0074 respectively) and HV (p=0.0237, p=0.0432, p=0.0005 respectively) groups ( Figure 2C ). Importantly, absolute count of CD56^neg CD16⁺ NK cells was significantly increased in the Expanded group compared to the Non-Expanded group (p<0.0001) ( Figure 2B , right panel). Besides, absolute counts of total NK cells and CD56^dim CD16⁺ NK cells were not altered (Figure D, left panel) while absolute counts of CD56^bright and CD56^dim CD16^- NK cells were decreased in the Expanded group compared to the HV group (Figure D, right panel). However, the paucity of these two subsets supports the hypothesis of an expansion from the CD56^neg CD16⁺ NK cells rather than a decrease of the other subsets. On the other hand, we investigated the T cell compartment in the HV, Non-Expanded and Expanded groups and found no difference in the frequencies of CD8⁺ T cells, CD4⁺ T cells, TCRVδ2⁺ T cells and regulatory T cells between groups ( Supplementary Figure S4 ).

Among patients in the Expanded group, 42.8% achieved sustained complete remission after first induction therapy versus 77.4% in the Non-Expanded group. Besides, 28.6% of the Expanded group relapsed after first complete remission versus 12.9% in the Non-Expanded group. Importantly, patients from the Expanded group had significantly poorer overall survival (HR[CI95]=3.3[0.75-14.7], p=0.0251) and relapse-free survival (HR[CI95]=13.1[1.9-87.5], p=0.0079) after 36 months follow-up ( Figures 2D, E ). Results from the validation cohort were in line with our previous study (8) and demonstrated that CD56^neg CD16⁺ NK cells expansion at diagnosis results in adverse clinical outcome in AML patients.

Considering the clinical implications of CD56^neg CD16⁺ NK cells expansion in AML, we aimed to further characterize this subset. To this end, we performed bulk RNA-seq profiling of CD56^neg CD16⁺, CD56^dim CD16⁺, CD56^dim CD16^- and CD56^bright NK cells isolated from peripheral blood in N=5 HV and N=4 AML patients at diagnosis. NK cell subsets from HV and AML patients were respectively pooled prior to analysis. The purity of CD56neg CD16+ NK cells in our dataset was assessed based on cell type-specific transcriptomic signatures for myeloid DC, non-classical monocytes, B cells and T cells adapted from the human protein atlas (18) (https://www.proteinatlas.org/) and low-density neutrophils signature adapted from (19). We set our minimal abundance threshold log10(TPM+1) to 1 to remove the background noise, as suggested in (20). Sorted CD56neg CD16+ NK cells did not express 10/13 and poorly expressed 3/13 transcripts from the non-classical monocytes signature, and did not express transcripts identifying myeloid DC, B cells, T cells or low-density neutrophils ( Supplementary Figure B ). In light of these results, we considered that the purity of CD56neg CD16+ NK cells was satisfactory for downstream analysis. CD56^neg CD16⁺ NK cells from HV and AML patients expressed a unique transcriptomic profile with 177 down-regulated and 161 up-regulated transcripts in AML samples and 87 down-regulated and 252 up-regulated transcripts in HV samples ( Supplementary Tables S4 , S5 ). As expected, NCAM1, encoding for CD56, was down-regulated in CD56^neg CD16⁺ NK cells from both AML patients and HV.

To assess the impact of AML on CD56^neg CD16⁺ NK cells transcriptomic profile, we focused on differentially expressed transcripts unique to the AML group. Venn diagrams identified 120 down-regulated and 53 up-regulated AML-specific transcripts ( Figure 3A , top panel). We first investigated the maturation status of CD56^neg CD16⁺ NK cells from AML patients. This subset down-regulated MYC, involved in NK cell development (21) but up-regulated NFIL3, an essential nuclear factor for NK cell maturation (22). During maturation, NK cells lose CD62L (SELL) and NKG2A (KLRC1) expression and acquire KIRs (23). Consistently, CD56^neg CD16⁺ NK cells down-regulated SELL and KLRC1 and up-regulated KIR2DL1, KIR3DL1, KIR3DL2 and KIR2DS4. Besides, DLL1, which inhibition is involved in Notch-mediated KIRs expression (24), was also down-regulated. Therefore, CD56^neg CD16⁺ NK cells from AML patients seemed to have reached later maturation stage. CD56^neg CD16⁺ NK cells expansion could be a consequence of impaired chemokine or cytokine signaling. We observed that CD56^neg CD16⁺ NK cells from AML patients down-regulated CCR1 and CCR5, involved in NK cell recruitment in inflamed tissues, and up-regulated CCL3 and CCL4, known as CCR5 ligands. Notably, THBS1, shown to take part in late NK cell expansion upon TGF-β stimulation (25), was up-regulated along with TNFRSF1B, involved in TNFα-mediated NK cell proliferation. Besides, KLF4, playing a role in murine NK cell survival and maintenance (26), was also up-regulated. Given the adverse clinical outcome of CD56^neg CD16⁺ NK cells expansion in AML, we also explored relevant transcripts related to NK cell cytotoxicity. Among them, TLR2 and GZMH were up-regulated. However, we also found transcripts that could inhibit NK cell-mediated cytotoxicity. CD56^neg CD16⁺ NK cells down-regulated INF2 described as a key factor of T cells immune synapse (27). In addition, ATF3, a negative regulator of IFN-γ genes in NK cells (28), was up-regulated along with CST7. This molecule is known to be an inhibitor of cathepsins C and H and could promote NK cell split anergy (29) ( Figure 3A , bottom panel). Enrichment analysis on AML-specific transcripts mostly unveiled that up-regulated transcripts were enriched in cytokine-related transcripts (TNF, IL-12, and type II IFN) whereas down-regulated transcripts were enriched in cell activation-related transcripts ( Figure 3B ).

Then, we sought to identify a specific transcriptomic signature for CD56^neg CD16⁺ NK cells based on the overlapping transcripts between AML and HV samples. A set of 57 down-regulated and 108 up-regulated transcripts were found ( Figure 3C , top panel). We observed the differential expression of cytotoxicity-related transcripts. CD56^neg CD16⁺ NK cells up-regulated LYZ, the activating receptor CD86(30) and the inhibitory receptor CSF3R (31). Besides, transcripts from the LILR family (LILRA1, LILRA2 and LILRB2) were also up-regulated. On the other hand, CD56^neg CD16⁺ NK cells differentially expressed transcripts that could hamper NK cell-mediated killing. Namely, TMEM163 implied in NK cell degranulation (32) was down-regulated whereas SIGLEC10, which is thought to decrease NK cell cytotoxicity in hepatocellular carcinoma (33), was up-regulated. Interestingly, complement transcripts C3, C3AR1, which was shown to inhibit NK cell cytotoxicity, as well as C5AR1 with immunoregulatory properties in mice (34) were up-regulated. Next, we investigated transcripts involved in NK cell maturation. CD56^neg CD16⁺ NK cells down-regulated transcripts expressed in immature CD56^bright NK cells such as GZMK(35) or KIT(36). They also down-regulated RUNX2, repressed during canonical maturation, along with its target genes ZEB1, BACH2 and PRDM8(37). On the other hand, ZNF683, known to regulated NK cell development (38) but also involved in NK cell exhaustion in multiple myeloma (39), was up-regulated. Then, we assessed the homing abilities of CD56^neg CD16⁺ NK cells. ITGA1, a specific marker of NK cell tissue residency (40), was down-regulated along with CCR7(41) and CXCR3(42). Besides, CXCR2 and its ligands CXCL8 and CXCL16(43), IL3RA and CD4, increasing cytokine production and cell migration in NK cells (44), were up-regulated ( Figure 3C , bottom panel). Furthermore, cell activation, cytokine pathways, proliferation and endocytosis were overrepresented among over-expressed transcripts while cell activation, chemotaxis, actin organization and Wnt signaling were overrepresented among under-expressed transcripts as revealed by gene set enrichment analysis from subset-specific transcripts ( Figure 3D ). Together these results strongly suggested that CD56^neg CD16⁺ NK cells represent a unique subset of mature NK cells with functional capacities relying on different cytotoxic pathways than conventional NK cells. Notably, some transcripts were related to impaired cytotoxicity or exhaustion, which could partly explain the adverse outcome in AML patients with CD56^neg CD16⁺ NK cells.

After RNA-seq profiling, we attempted to confirm our findings with spectral flow cytometry and investigated whether CD56^neg CD16⁺ NK cells had a distinct protein expression pattern than conventional NK cells in HV, Non-Expanded and Expanded groups.

CD56^neg CD16⁺ NK cells from all three groups had significantly decreased expression of NK cell triggering receptors NKp30 and NKp46 compared to CD56^bright NK cells. However, CD56^neg CD16⁺ NK cells from the Expanded group expressed similar levels of NKp30 and NKp46 compared to CD56^dim NK cells. CD56^neg CD16⁺ NK cells from the HV group had decreased NKp46 expression compared to CD56^dim CD16^- NK cells whereas CD56^neg CD16⁺ NK cells from the Non-Expanded group had decreased NKp46 expression compared to CD56^dim CD16⁺ NK cells. Regarding direct cytotoxicity, CD56^neg CD16⁺ NK cells from the HV and Expanded groups significantly expressed higher levels of granzyme B and perforin compared to CD56^dim CD16^- NK cells. CD56^neg CD16⁺ NK cells from the HV and Non-Expanded groups significantly expressed higher levels of granzyme B and perforin compared to CD56^bright NK cells. CD56^neg CD16⁺ NK cells from the Non-Expanded group had decreased expression of granzyme B and perforin compared to CD56^dim CD16⁺ NK cells. On the other hand, CD56^neg CD16⁺ NK cells from the Expanded group had similar expression of granzyme B and perforin than CD56^dim CD16⁺ NK cells. Thus, our results are in line with both our transcriptomic and in vitro data: CD56^neg CD16⁺ NK cells displayed a cytotoxic profile in AML. On the other hand, our results confirmed that CD56^neg CD16⁺ NK cells had reached a mature stage as CD56^neg CD16⁺ NK cells from all three groups expressed significantly lower levels of NKG2A and higher levels of CD158a/b than CD56^bright NK cells. Moreover, CD56^neg CD16⁺ NK cells from the HV and Non-Expanded groups had decreased expression of NKG2A compared to CD56^dim CD16^- NK cells ( Figure 4A ).

Next, we investigated whether CD56^neg CD16⁺ NK cells from the Expanded group displayed an altered phenotype compared to the HV and Non-Expanded groups. We performed UMAP analysis of CD56^neg CD16⁺ NK cells and automatic clustering using FlowSOM, resulting in 17 clusters, in HV, Non-Expanded and Expanded groups ( Figure 4B ). Clusters were identified by the co-expression of 16 markers ( Figure 4C ; Supplementary Figure S6 ). We plotted CD56^neg CD16⁺ NK cells density for each group and observed significant changes in CD56^neg CD16⁺ NK cells from the Expanded group compared to the HV and Non-Expanded groups ( Figure 4D ). More CD56^neg CD16⁺ NK cells from the Expanded group were found in mature cytotoxic clusters compared to the HV and Non-Expanded groups. Indeed, cluster 1 (NKp46⁺ NKG2D⁺ DNAM-1⁺ CD96⁺ Perforin⁺ Granzyme B⁺ CD57⁺ CD158a/b⁺ TIGIT⁺) and cluster 4 (NKG2D⁺ DNAM-1⁺ Perforin⁺ Granzyme B⁺ CD57⁺ KIR⁺ Siglec-7⁺) were significantly increased in the Expanded group compared to the HV and Non-Expanded groups ( Figure 4E ). Notably, TIGIT was shown to be associated with NK cell exhaustion (45) and TIGIT⁺ NK cells displayed decreased effector function. Besides, CD56^neg CD16⁺ NK cells from cluster 3 (NKG2D⁺ DNAM-1⁺ Perforin⁺ Granzyme B⁺ CD57⁺ KIR⁺) were more abundant in the Expanded group than in the HV group ( Figure 4E ). Interestingly, the frequency of terminally mature CD56^neg CD16⁺ NK cells expressing NKG2C present in clusters 2 and 5 was not different between the three groups ( Figure 4C ). On the other hand, Siglec-7 expressing clusters, cluster 8 (NKp30⁺ NKp46⁺ NKG2D⁺ DNAM-1⁺ Granzyme B⁺ NKG2A⁺ Siglec-7⁺), cluster 9 (NKp30⁺ DNAM-1⁺ Perforin⁺ Granzyme B⁺ Siglec-7⁺), cluster 11 (NKp30⁺ NKp46⁺ DNAM-1⁺ Perforin⁺ Granzyme B⁺ TIGIT⁺ Siglec-7⁺) and cluster 13 (NKp30⁺ NKp46⁺ NKG2D⁺ DNAM-1⁺ Perforin⁺ Granzyme B⁺ KIR⁺ TIGIT⁺ Siglec-7⁺) were significantly decreased in the Expanded group compared to the HV group. Interestingly, Siglec-7 expression on NK cells were shown to mark fully functional NK cells ligand (46). Therefore, the loss of Siglec-7⁺ CD56^neg CD16⁺ NK cells could contribute to adverse clinical outcome of patients with CD56^neg CD16⁺ NK cells expansion. Together, our results showed an altered phenotype of CD56^neg CD16⁺ NK cells from the Expanded group, with increased mature cytotoxic NK cell expressing inhibitory receptors such as CD158a/b and TIGIT and decreased Siglec-7⁺ NK cells.