Chronic inflammation deters natural killer cell fitness and cytotoxicity in myeloid leukemia

Immune dysregulation is thought to account for relaxed tumor control in myeloid malignancies.^1-3 Interferon (IFN) alfa, a cytokine with immunomodulating properties, was a preferred treatment for patients with chronic myeloid leukemia (CML) until tyrosine kinase inhibitors (TKIs) became a frontline therapy.^3-5 TKIs effectively target BCR::ABL1 oncogene and induce a major molecular response in >90% of patients with CML.^6,7 Despite that, primary resistance to TKIs, insufficient rates of deep molecular response (DMR), and securing treatment-free remission (TFR) remain significant challenges in the CML field.^8-14 

Prolonged TFR in CML includes the duration of DMR, depth of treatment response, and the absence of resistance to first-generation TKIs.^15-19 The most critical immunological parameter predicting TFR was mature natural killer (NK) cell counts at TKI withdrawal, suggesting that these cells play a role in clearing drug-resistant leukemic cells.^20-26 NK cells elicit cytotoxicity against malignant and virally infected cells via an intricate interplay between signals from inhibitory (KIRs and NKG2A) and activating receptors (NKG2C, NKG2D, NKp46, and DNAM-1).^27-29 The maturation status of NK cells is associated with their effector function. The main subsets in humans include immature CD56^brCD16^lowCD57^– NK cells with a high potential to produce cytokines, and mature cytotoxic CD56^dimCD16^hiCD57^+/− cells that lyse target cells by releasing cytotoxic granules.^30,31 

NK cells are compromised in patients with CML, reflecting in decreased numbers, lower expression of activating receptors NKp46 and DNAM1, impaired degranulation, and poor cytokine production, with the rejuvenation of these after successful TKI treatment.^20-22,32-35 Further supporting NK cells as a prognostic marker is the higher NKG2C (KLRC2) gene expression that predicts optimal response, whereas NKG2A (KLRC1) expression is associated with resistance to TKI treatment and early relapse.^36,37 

Despite NK cell antileukemic potential, no mouse models have been examined, to our knowledge, to explore the effect of CML on BCR::ABL1^– NK cells, hindering the progress of NK cell immunomodulation in patients. Here, we characterize NK cells in a preclinical chimeric CML mouse model generated by transplanting SCL-tTA/BCR-ABL double-transgenic cells into congenic mice.^38,39 This model recapitulates human CML and demonstrates suboptimal responses to TKIs, highlighting its suitability for studying refractory disease.^40-42 Moreover, establishing chimeras allows for investigating the effect of CML on nonmutant hematopoietic cells,³⁹ which is critical considering that the vast majority of patients’ NK cells lack BCR::ABL1.^43,44 In this study, we confirm the feasibility of this mouse model for studying NK cells and identify inflammatory signaling pathways as candidate targets to potentiate NK cells within CML. Tumor necrosis factor α (TNFα) emerged as the key cytokine predicted to induce this CML-specific gene signature. Furthermore, NK cell cytokine checkpoint Cish (encodes cytokine-inducible SH2-containing protein, Cis), which can be targeted to potentiate NK cell functionality, was enriched across all NK cell maturation stages in CML.^45,46 Validating these findings, we observed the enrichment of proinflammatory signatures in NK cells of newly diagnosed patients with CML who further demonstrated TKI resistance. The induction of inflammation-induced genes in NK cells in CML could serve as a barometer of dysregulated cytokine signaling to inform prospective therapies.

C57BL/6 and B6.SJL-Ptprca Pepcb/BoyJ mice were maintained under specific-pathogene-free conditions at The University of Alabama at Birmingham, and all procedures were performed under the Institutional Animal Care and Use Committee-approved protocol. Double-transgenic CD45.2⁺ BCR-ABL×SCL-tTA or single-transgenic SCL-tTA cells were used for transplanting to establish CML or control chimeras, respectively.^38,39 Males and females were represented equally.

Bone marrow transplant, tetracycline, and TKI treatment are described in the supplemental Methods.

Flow cytometry and cell sorting were performed as described in the supplemental Methods.

RNA isolation, reverse transcription polymerase chain reaction (RT-PCR), bulk RNA sequencing (RNAseq), 10× single-cell RNAseq (scRNAseq), and analysis were performed as described in the supplemental Methods.

The cytokine array was performed as described in the supplemental Methods.

2′-deoxy-2-fluoro-D-arabinonucleic acid antisense oligonucleotides targeting mouse and human Cish messenger RNA (mRNA) were designed and synthesized by AUM BioTech, LLC (Philadelphia, PA). Cells were treated with 2μM of silencing oligonucleotide for 24 hours before performing assays. mRNA depletion was confirmed by RT-PCR using Cish-targeting primers.

All statistical analyses were performed with GraphPad Prism (10.2.0), and P ≤ .05 was considered significant. The applied statistical tests can be found in the respective figure legends.

We comprehensively studied host NK cells’ phenotypic and functional features in a chimeric BCR::ABL1⁺ murine model of CML (Figure 1A; supplemental Figure 1A).^38,39 Briefly, sublethally irradiated CD45.1⁺ B6.SJL-Ptprca Pepcb/BoyJ (host) mice were transplanted with CD45.2⁺ SCL-tTa/BCR-ABL cells or control CD45.2⁺ cells lacking the oncogene. Experiments were performed 8 to 10 weeks after, when animals developed leukemia with pronounced myeloid expansion, splenomegaly (supplemental Figure 1B), and anemia. Consistent with existing data on newly diagnosed patients,²² NK cell frequencies and counts were reduced in CML mice, including the blood, spleen, and bone marrow (Figure 1B; supplemental Figure 1C). Moreover, NK cell abundance in the spleen negatively correlated with CML burden defined by the percentage of leukemic myeloblasts (Figure 1C). Due to the scarcity of blood NK cells in CML, further analyses focused on the marrow and spleen.^47,48 Given the association of mature CD56^dim NK cell counts with prolonged TFR in patients with CML,^20-22 we defined the maturation status of CML-exposed NK cells. Mouse NK cell maturation subsets are distinguished by differential expression of CD27 and CD11b, with the progression from immature (CD27⁺CD11b^–, “Imm”) to intermediate (CD27⁺CD11b⁺, “M1”) to mature (CD27^–CD11b⁺, “M2”) populations (supplemental Figure 1A).⁴⁹ We found that NK cells exhibit a higher frequency of immature phenotype during leukemia (Figure 1D) whereas all the absolute counts tend to decrease across all stages (Figure 1E). To validate this phenotype, we measured a marker of terminal NK cells maturation KLRG1.⁵⁰ Frequencies of KLRG1⁺ NK cells were significantly lower in CML mice than in controls (Figure 1F), confirming the hypomature NK cell state (Figure 1G) seen in patients predisposed to relapse upon TKI discontinuation. Furthermore, CML-exposed and control NK cells did not significantly differ in annexin-V binding with no bias across maturation stages (supplemental Figure 1D), suggesting that the altered subset distribution reflects a maturation block rather than an increased turnover of mature cells. Surprisingly, we observed a drastic increase in the nuclear marker of proliferation Ki-67 (Figure 1H), suggesting that NK cells in CML have an activated cell cycle despite the reduced numbers.

Next, we profiled early NK cell precursors in the marrow and revealed a decrease in common lymphoid progenitors and NK cell progenitors (Figure 1I), suggesting that NK cell scarcity arises from perturbed lymphopoiesis.⁵¹ So far, basic quantitative NK cell metrics in this chimeric mouse model align with those in untreated or high-risk patients with CML, warranting a more in-depth NK cell characterization.

NK cells recognize targets by activating and inhibitory surface receptors.²⁷ The expression of the activating receptor NKp46 is known to decrease in NK cells of patients with CML compared with healthy donors.²⁰ We observed a similar phenotype in our CML chimeras (Figure 1J) with all maturation subsets showing lower NKp46 expression (supplemental Figure 1E). Moreover, we observed a trend toward decreased expression of an activating receptor Ly49D (supplemental Figure 1F), whereas frequencies of another activating receptor DNAM-1, with described roles in myelodysplastic syndrome and acute myeloid leukemia immunosurveillance,^52,53 were slightly increased without significance (supplemental Figure 1F).

Immune checkpoints are the well-established mechanism of immune evasion in cancer. We found that NK cells retain the expression of the inhibitory receptor NKG2A in CML (Figure 1K) at all maturation subsets (supplemental Figure 1G). This phenotype has been associated with failure to achieve major molecular response in patients with CML.³⁶ In addition, NK cells acquired the expression of the T-cell immunoreceptor with immunoglobulin and tyrosine-based inhibitory motif domains (TIGIT) and lymphocyte activation gene-3 (Lag-3) implicated in NK cell suppression in leukemia (Figure 1K).^33,37 

Further investigation revealed profound alterations in NK cell functionality during CML. First, we observed a significantly lower Yac-1-specific ex vivo degranulation rate in NK cells sorted from CML chimeras (Figure 1L), suggesting compromised cytotoxicity. This effect could only partially be explained by skewed maturation (supplemental Figure 1H) underscoring the severity of the functional NK cell perturbations. Second, phorbol 12-myristate 13-acetate (PMA)/ionomycin-activated NK cells from CML mice produced less IFN-γ and contained significantly lower amounts of granzyme B (Figure 1L). These results further correlate with the dysfunctional NK cell biology seen in patients with CML with TKI resistance, highlighting the urgency to understand and counteract these perturbations.

Individuals who respond to TKI treatment show a restoration of NK cell numbers, surface receptor expression, and cytotoxicity.^20-22 Therefore, we asked whether CML mice replicate this during BCR::ABL1 inhibition. BCR::ABL1 expression in our chimeras is regulated by the Tet-Off system in which dietary tetracycline deactivates oncogene expression,³⁸ mimicking a successful treatment with TKIs (Figure 2A). We observed a gradual decrease in chimerism and myeloid skewing of transgenic CD45.2⁺ cells after BCR::ABL1 repression; conversely, NK cell abundance restored over time (Figure 2B). NK cell frequencies, counts, maturation, and other features were all partially restored after 6 weeks of BCR-ABL reversion, reaching a baseline level in 11 weeks (Figure 2C-H; supplemental Figure 1I-K). Notably, there were no signs of myeloid skewing and aberrant expansion of CD45.2⁺ transplanted cells in the marrow, spleen, and blood at the 6-week point of BCR::ABL1 reversion (supplemental Figure 1L), whereas the NK cell phenotype was not fully restored at that time.

Interestingly, treatment with TKIs fails to achieve complete remission in this mouse model, making it suitable for studying leukemic resistance.⁴² This agrees with the correlation of phenotypic traits of mouse CML-NK with NK cells from patients with TKI resistance or patients who fail to sustain remission after TKI cessation. Although we did observe the partial restoration of several parameters including NK cell frequencies and maturation after a 3-week TKI treatment course (Figure 2I; supplemental Figure 1M), the expression of inhibitory receptors was not reverted (Figure 2J). This may suggest a suboptimal effect of TKI treatment in this model, or that a prolonged treatment may be required to restore NK cell phenotype.

To investigate possible mechanisms underlying these above-observed alterations, we characterized the transcriptomic signature of CML-exposed NK cells. We sequenced fluorescence-activated cell sorted host single NK cells (CD45.1⁺Lin^–CD122⁺NK1.1⁺) from the spleen and marrow of control and CML mice (Figure 3A; supplemental Figure 2A-B). We filtered Il12rb⁻ and Nkg7^low cells to exclude contaminating lineages. CML-exposed NK cells clustered apart from control (Figure 3B), and immature cells were well distinguishable from mature by Cd27 and Itgam expression (Figure 3B). K-means clustering analysis identified 3 clusters that we classified as Mature (K1), Immature (K2), and Cycling (K3) based on gene markers (Figure 3B; supplemental Figure 2C; supplemental Table 1). More CML-exposed cells belonged to Immature and Cycling clusters (Figure 3C); furthermore, the Cycling cluster was predominantly composed of CML-exposed cells (Figure 3C). The abundance of cell cycle–related transcripts (Mki67, Cks1b, Top2a) caused “Cell cycle” to be the second from the top-enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway in CML-exposed NK cells (supplemental Figure 2D; supplemental Table 2). To identify therapeutic targets, we narrowed further analyses to canonical NK cells represented by K1 and K2 clusters. In alignment with flow data, immune checkpoint genes Tigit and Lag3 were among differentially expressed genes (DEGs) significantly upregulated in CML along with other inhibitory molecules including CD200r1 and Lilr4a/b (Figure 3D), whereas CD27 and the NKG2A-coding gene Klrc1 were only moderately elevated (fc = 1.44; P = 8.04e−8 and fc = 1.27; P = 4.88E−7, respectively). NK cells from healthy mice were enriched in maturation markers such as S1pr5, Gzma, and transcription factors Zeb2 and Klf2 (Figure 3D). Among 48 KEGG pathways significantly enriched in K1 + K2 NK cells in CML (supplemental Table 3), many were involved in cytokine signaling and inflammatory processes (Figure 3E). The genes involved in these pathways and significantly enriched in CML-exposed NK cells included cytokine receptors Il2ra, Il2rb, and Il7r; cytokine-encoding genes Csf2, Tnf, Tnfsf14, and Tnfsf8; soluble mediators of inflammation S100A8 and S100A9; chemoattractants Lta/b and Xcl1, and suppressor of cytokine signaling (SOCS)-family cytokine checkpoint genes Cish and Socs1/2 (Figure 3D-E; supplemental Figure 2E). To avoid bias caused by the source tissue or impaired NK cell maturation in CML, we compared gene expression of control vs CML for K1 and K2 clusters in marrow- and spleen-derived NK cells independently, which identified 13 genes that were upregulated in CML (fold change ≥2; P < .05; Figure 3F) including Cish, a known NK cell cytokine checkpoint that limits NK cell persistence and function (supplemental Figure 3A).^46,54-56 We validated this increase by RT-qqPCR on sorted immature (CD11b^–) and M1+M2 (CD11b⁺) cells (Figure 3G). Enhanced expression of Cish and SOCS genes is an expected feature in immune cells subjected to inflammation. Notably, among genes expressed at a greater level in CML, we observed genes for cytokine receptors capable of inducing Cish expression via STAT5 activation (Il2rb and Il7r; Figure 3D-E; supplemental Figure 3B).

Next, we asked whether the decrease in target-specific NK cell degranulation observed in CML could be identified at the transcriptomic level. We constructed a degranulation gene signature by bulk RNAseq from CD107a⁺ and CD107a^– NK cells cocultured with Yac-1 targets (supplemental Figure 3C). We applied this “degranulation” gene list to our scRNAseq data set and identified 63 genes differentially expressed between control and CML-exposed K1+K2 NK cells. Surprisingly, only 3 genes (Vegfa, Fos, and Snhd4) were upregulated in control, and the remaining upregulated in CML (supplemental Figure 3D). Furthermore, CML-exposed NK cells had a significantly higher degranulation score calculated for the entire gene list (Figure 3H), suggesting that NK cells display chronic degranulation activity in CML, possibly contributing to their lower target-specific degranulation (Figure 1L).

Given that cytokine signaling pathways were among the top enriched in CML-exposed NK cells, we asked whether the leukemic cytokine milieu affects NK cell function. We treated healthy NK cells with equal concentrations of peripheral blood serum from control or CML mice for 24 hours. NK cells exposed to CML serum displayed decreased degranulation and killing capacity in cocultures with Yac-1 cells (Figure 4A; supplemental Figure 4A) whereas the baseline (no target) degranulation level tended to increase (supplemental Figure 4B), supporting earlier suggested chronic NK cell activation in CML. The distribution of NK cell maturation subsets was unchanged upon the treatments (supplemental Figure 4C), emphasizing that the CML is suppressive for NK cell cytotoxicity rather than depleting mature cells. We also observed no changes in the expression of surface markers KLRG1, NKp46, or TIGIT (supplemental Figure 4C), suggesting a 24-hour exposure to CML soluble factors insufficient to acquire a profoundly altered phenotype seen in diseased animals. Furthermore, Cish mRNA levels were significantly higher in healthy NK cells treated with CML serum than in controls (Figure 4B). To explore the role of Cish in restraining NK cell cytotoxicity in CML, we targeted its transcript using 2′-deoxy-2-fluoro-D-arabinonucleic acid antisense oligomers (Cish^KD). Cish^KD or scrambled oligo-treated NK cells were then exposed to CML or control serum for 24 hours followed by a degranulation assay. We observed a partial restoration of NK degranulation capacity in CML serum-treated Cish^KD NK cells (Figure 4C).

Next, we asked whether the CML microenvironment is enriched in proinflammatory factors that could cause the observed perturbations in NK cells. We measured concentrations of 200 soluble molecules in blood serum and marrow using a mouse cytokine array Q4000 (RayBiotech). Factors upregulated in CML included proinflammatory cytokines interleukin-1β (IL-1β), IL-6, IL-3, IL-2, IL-15, TNFα, granulocyte-macrophage colony-stimulating factor, and IFN-γ, which reflects previous findings in patients with CML (Figure 4D; supplemental Figure 4D).^58-62 Cytokines TNFα, IL-15, IL-10, IL-2, and IL-6 were present at higher concentrations in both tissues (Figure 4D).

Next, we referred to the single-cell data to evaluate the expression of receptors for identified soluble factors. The putative list of binding partners for CML-specific soluble molecules was obtained using CellphoneDB.⁶³ Among 21 genes differentially expressed between CML-exposed and control K1+K2 NK cells, 18 were enriched in CML (supplemental Table 4). Among these, we found genes encoding cytokine receptors Tnfrsf1b, IL-2ra, IL-2rb, IL-3ra, and IL-6st (supplemental Figure 4E), suggesting that NK cells are equipped to sense CML inflammation.

Because of the complexity of the CML cytokine network, we attempted to predict the most pronounced cytokine-induced signaling pathway in CML-exposed NK cells based on the scRNAseq dictionary of immune response to cytokines.⁵⁷ Using the input of a complete list of DEGs enriched in K1+K2 NK in CML, we identified TNFα (key candidate), IFN-α/β, IL-18, IL-15, and IL-1α/β as the top predicted triggers of global transcriptional changes (supplemental Figure 4F; Figure 4E-F). Among these, TNFα, IL-15, and IL-1α/β were found elevated in our CML model (Figure 4D,G; supplemental Figure 4D), suggesting they are candidate factors that directly or indirectly induce CML-specific gene signature in NK cells. Among corresponding binding partners, TNFα receptor TNFR2 and its gene Tnfrsf1b (but not TNFR1 or corresponding gene Tnfrsf1a) were significantly upregulated in CML-exposed NK cells (Figure 4H; supplemental Figure 4E,G; and data not shown). Supporting the notion of TNFα as the top factor, we observed a partial restoration of degranulation capacity in CML serum–exposed NK cells treated with TNFα-blocking agent etanercept (Figure 4I).

According to the dictionary, IL-15, IL-15 + IL-2, IL-2, and, surprisingly, TNFα were the top predicted cytokines to induce Cish expression in CML (Figure 4J).⁵⁷ Despite that IL-15 and IL-2 are important for NK cell expansion, maintenance, and activation, the chronic exposure of NK cells to these cytokines is known to perturb NK cell fitness and function.^64-68 Furthermore, the interplay with other CML cytokines can shape NK cell responses to IL-15, potentially exacerbating exhaustion.^69-73 In agreement, we observed that treatment with etanercept abrogated the robust induction of Cish upon exposure of NK cells to CML serum (Figure 4K).

Next, we used a scRNAseq data set on bone marrow mononuclear cells derived from newly diagnosed patients with CML with known clinical outcomes to interrogate the relevance of our findings.³⁶ Patients were stratified into 3 groups based on the response to imatinib: group A and group B with optimal and suboptimal responses to the drug, respectively, and group C patients who subsequently displayed TKI resistance.³⁶ Using DEGs associated with inflammatory pathways and enriched in CML-exposed NK cells (Figure 3I; supplemental Table 5), we generated a corresponding human gene set (iNK-CML). A substantial number of genes from the iNK-CML list were upregulated in newly diagnosed high-risk patients with CML from group C compared with other groups (Figure 5A). Notably, the iNK-CML module score was significantly higher in group C NK cells than healthy individuals, but not in NK cells from group A and group B patients (Figure 5B). CISH was 1 of the genes enriched in group C NK cells (Figure 5C). These findings suggest a correlation of transcriptomic signatures between the mouse CML-NK cells and NK cells from patients resistant to treatment. In addition to the alignment with mouse-derived iNK-CML signature, an unbiased DGE analysis of human NK cells revealed the enrichment of inflammation-associated pathways in group C patients when compared with healthy donors (Figure 5D). We speculate that this transcriptomic signature of an aberrant cytokine activation reflects the inflammation-mediated perturbation in immune cells before TKI treatment and correlates with inferior response to therapy.

We then took advantage of banked samples from patients with CML to follow-up on the altered CD27/CD11b NK cell phenotype we found in CML mice. These markers have been associated with different maturation stages in human NK cells.⁷⁴ First, we profiled patients’ NK cells to confirm the CML-NK cell phenotype previously described (Figure 5E; supplemental Figure 5A-B).²⁰ Similar to murine CML, we observed a higher percentage of CD27⁺ NK cells in patients (Figure 5F). Further work will be needed to investigate the feasibility of these receptors as markers of NK dysfunction.

To test the impact of the CML milieu on human NK cells, we used plasma samples from newly diagnosed patients with CML who subseqently failed to achieve major molecular response (supplemental Table 6). CML plasma samples imposed a suppressive effect on healthy NK cell ex vivo degranulation and cytotoxicity (Figure 5G-H) and induced a slight upregulation of CISH (supplemental Figure 5C). We targeted CISH in healthy donor NK cells with antisense oligomers and exposed CISH^KD and scrambled oligo-treated NK cells to CP-CML plasma samples. CISH knockdown potentiated the degranulation capacity of NK cells exposed to CML plasma (Figure 5I).

Finally, we referred to the human scRNAseq data set to identify candidate sources of soluble mediators sensed by NK cells in CML. Using the list of molecules upregulated in murine CML, we compiled a CML-ligand gene set (supplemental Table 7) and calculated the expression scores in group C patients’ marrow myeloid cells, plasmacytoid dendritic cells, and progenitor cells. Among various cell types, classical monocytes displayed the highest CML-ligand score (supplemental Figure 5D-E). We thus performed NicheNet analysis using classical monocytes as sender cells and NK cells as receiver cells to predict signaling circuits establishing iNK-CML signature.⁷⁵ Although the receptor gene set (supplemental Table 8) score did not differ between healthy and group C patients’ NK cells, the score tended to decrease in NK cells from group A patients with CML who effectively responded to imatinib (supplemental Figure 5F). NicheNet analysis again revealed TNFα, IL-15, and IL-1β as candidates to trigger inflammatory gene expression within group C patients’ NK cells when secreted by classical monocytes (supplemental Figure 5G), aligning with findings in mice. Similarly, we observed a modest restoration of degranulation capacity in CML plasma-exposed healthy NK cells upon TNFα blockade (Figure 5J). Altogether, our findings indicate a cytokine-activated, proinflammatory signature in NK cells during CML with the activation of an NK cell–intrinsic brake system including CISH, resulting in chronic NK cell stimulation and anergy.

NK cells became significant in CML due to the robust correlation of NK cell biology with patient response to TKIs and prolonged TFR.^{20-22,24,25,36} Urgent questions arise about what mechanisms underlie the relaxed NK cell immunosurveillance and how they can be targeted. In this study, we aimed to evaluate the contribution of the CML microenvironment to NK cell dysfunction. We characterized NK cells in chimeric BCR::ABL1⁺ mice to confirm the feasibility of this model. We found that many metrics including lower NK cell numbers, decreased NKp46 expression, and poor degranulation and IFN-γ production replicate NK cell biology in untreated patients with CML.^20-22,32-35 Consistent with these data, a previously published scRNAseq study showed decreased mRNA levels of cytotoxic marker granzyme B in NK cells from newly diagnosed patients compared with individuals treated with dasatinib.⁷⁶ Furthermore, NK cells from patients with an optimal response to dasatinib had a higher granzyme B content and enhanced degranulation and cytokine production than patients who subsequently failed to achieve major molecular response.⁷⁶ We demonstrated that NK cells gain inhibitory receptors in CML mice; similarly, immune checkpoints such as T-cell immunoglobulin and mucin domain-containing protein 3 (TIM3), TIGIT, and programmed cell death protein 1 were found overexpressed by NK cells in patients with CML.^22,33,34,37 Higher NKG2A mRNA expression by NK cells at diagnosis was predictive of refractory disease with >80% accuracy³⁶; furthermore, this transcript was enriched in individuals who relapsed after TKI cessation compared with those with prolonged TFR.³⁷ 

Intriguingly, we observed an impairment of NK cell maturation in CML mice, the phenotype associated with treatment-free relapse.^20-23 We speculate that the hypomature status of NK cells contributes to inferior NK cell cytotoxicity. All the above suggests that NK cells in our mouse model correlate with high-risk patients with CML, emphasizing the feasibility of this model for studying the mechanisms of immune dysregulation and testing immunotherapies.

scRNAseq revealed cytokine signaling and inflammation-associated pathways in CML-exposed NK cells. We observed that NK cells sense inflammatory signals enriched in the leukemic microenvironment, upregulate genes involved in the cell cycle, cytokine signaling, and negative regulation of cytokine response such as Cish. Cish is induced by cytokines that signal through STAT5 and is known to limit effector responses in cytotoxic cells.^46,54,77,78 In CML, Cish upregulation could indicate chronic NK cell stimulation and exhaustion. This proinflammatory immune state may contribute to leukemic progression because of nonspecific NK cell activation and failure to clear TKI-resistant cells. Our data suggest that TNFα is the top candidate affecting NK cells in mice and patients with CML. The effect of TNFα on NK cells in the leukemic microenvironment is unclear. Outside of myeloid malignancies, TNFα has been shown to impose dual effects on NK and other effector lymphocytes, from activation of NK cell cytotoxicity and cytokine production to induction of inhibitory receptor expression such as Tim3 and apoptosis upon prolonged exposure to this cytokine in the tumor microenvironment.^68,79-82 Intriguingly, TNFR2, found upregulated by CML-exposed NK cells, was proposed as a novel NK cell checkpoint in gastrointestinal tumors.⁸¹ Blocking CML-associated cytokines (IL-6, IL-1β, and TNFα) or corresponding signaling pathways to prevent leukemic cells from feeding off these signaling axes and sensitize resistant cells to TKI has been proposed.^83-87 We speculate that this approach may also restrain immune cell exhaustion aiding in the elimination of residual leukemia by cytotoxic immune effector cells. Besides, shielding immune cells from excessive inflammation by genetic engineering could improve the efficacy of adoptive therapy. CISH deletion appears to be 1 such strategy, considering the superior antitumor activity of CISH-knockout NK cells in preclinical studies.^46,54-56,78 

In conclusion, our study highlights the susceptibility of NK cells to the inflammatory environment orchestrated by CML, revealing new paths for therapeutic intervention. By uncovering the interplay between inflammation and NK cell dysfunction, we lay the groundwork for developing immunotherapies that could disrupt the immune evasion cycle.

The authors thank Christopher Klug for reviewing the original draft of the manuscript; the University of Alabama at Birmingham Flow Cytometry and Single Cell Core Facility, especially Vidya Sagar Hanumanthu and Shanrun Liu, for their assistance; and the patients and University of Alabama at Birmingham Hospital personnel involved in this study.

R.S.W. received grants from the National Institutes of Health (R01HL150078 and 1PO1HL131477) and the Mark Foundation Endeavor award.

Contribution: V. Kuznetsova and R.S.W. conceptualized the study; V. Kuznetsova, V. Krishnan, A.C., X.R., R.S.W., and R.B. were responsible for the study methodology; V. Kuznetsova, V. Krishnan, and X.R. were responsible for formal analysis; V. Kuznetsova, V. Krishnan, A.C., X.R., S.B.P., A.N.C., P.G., J.P.K., F.L., and A.M.F. performed study investigation; S.B.P., T.D.R., R.B., S.P., and R.S.W. provided resources; R.S.W. was responsible for funding acquisition and project administration; R.S.W., S.T.O., and L.L.F.-P. supervised the study; and V. Kuznetsova, A.N.C., R.S.W., V. Krishnan, and R.B. were responsible for reviewing and editing the manuscript.

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

Correspondence: Robert S. Welner, Department of Medicine, The University of Alabama at Birmingham, 1824 6th Ave S, Wallace Tumor Institute 420C, Birmingham, AL 35233; email: rwelner@uab.edu.