Introduction Around 70% of pediatric patients and 30% of adult patients with acute myeloid leukemia (AML) will achieve a cure, but the prognosis for patients who relapse remains dismal, with only 16-43% of pediatric patients and less than 20% of adult patients achieving long term survival. Memory-like natural killer (ML NK) cells are NK cells which upon stimulation with IL 12/15/18 undergo a differentiation process such that they exhibit increased cytotoxic function when encountering a tumor target. A phase I trial of adoptive transfer of ML NK cells in both adult and pediatric patients was promising with several patients achieving a complete response. However, most of these responses were not durable, suggesting the need for new strategies to improve ML NK cell ability to recognize and kill AML. We explore two strategies to enhance ML NK cell recognition of AML while enhancing their functional activity and tumor killing capacity. Objective ML NK cells target CD33+CD123+ primary AML blasts via a CD16-IL15-CD33 tri-specific killer engager (CD33 TriKE) and a CD123 chimeric antigen receptor (CAR) expressed on their surface (CD123 CAR ML). (Figure 1) Methods ML NK function is assessed using a 6-hour tumor co-culture assay (CD123 CAR ML NK or ML NK + CD33 TriKE vs. Primary AML) to detect cytokine production (IFNγ). Specific Killing is assessed during a similar co-culture of effectors and AML over 4-hours. Long-term in vitro tumor control is assessed using live cell imaging (Incucyte) over 5 days. NSG-IL-15tg mice are used for a THP-1 xenograft model. NSG-S mice will be used for a patient-derived xenograft model. Results Both CD123 CAR ML NK cells and CD33 TriKE demonstrated antigen-specific enhanced IFNg production not only against AML cell lines but also against primary AML blasts, chosen based on their CD123 and CD33 expression (Primary AML #21- CD123 CAR ML NK: IFNg +6.9% p=.0078, CD33 TriKE: IFNg +15.4% p=.007; Primary AML #34- CD123 CAR ML NK: IFNg +9.1% p=0.0092, CD33 TriKE: IFNg +16.6% p=.0118). Specific Killing of primary AML was also improved at several effector to target ratios by both CD123 CAR ML NK cells and ML NK cells treated with CD33 TriKE (Primary AML #21- ML vs CD123 CAR ML: p=.0026, ML vs ML+CD33 TriKE: p=.0036; Primary AML #34- ML vs CD123 CAR ML: p=.0006, ML vs ML+CD33 TriKE: p=.0026) (figure 2). Both strategies significantly improved tumor control in our THP-1 xenograft model. Thus, we hypothesize that both will also improve tumor control and survival in a PDX model. Conclusion Both CD123 CAR and CD33 TriKE improve recognition of CD123 and CD33 expressing AML, including patient-derived AML. Both strategies improve tumor killing in vitro and tumor control in a THP-1 xenograft model. Thus, we hypothesize that both strategies will also improve control in an AML PDX model. If both strategies continue to demonstrate efficacy, both the CD123 CAR and CD33 TriKE are highly translatable to the clinical setting.

PSMA TriKE testing against patient-derived xenograft (PDX).

Figure S2. Comparisons Between Canonical and Non-Canonical NK Cell Subsets in Tumors and Circulation. Related to Figure 2.

Abstract Background: Memory-like natural killer (ML NK) cells are NK cells which upon stimulation with IL12/15/18 undergo a differentiation process to develop improved metabolic fitness and enhanced killing capabilities of different types of malignant cells. They are a promising cellular therapeutic against acute myeloid leukemia (AML), for which the prognosis remains dismal for both pediatric and adult patients who experience relapse of their AML. Phase I studies of adoptive ML NK cell therapy in adult and pediatric patients showed promising results with several patients achieving complete responses, but many go on to relapse after a few months. We investigate two strategies to improve recognition of AML by memory like NK cells by targeting them to AML surface antigens CD33 and CD123. Objective: We target ML NK cells to CD123 by engineering ML NK cells to express a chimeric antigen receptor (CD123 CAR ML NK). We target ML NK cells to CD33 via a trispecific killer engager (CD33 TriKE) that enhances the immune synapse by engaging CD16 on the ML NK cell and CD33 on AML. The CD33 TriKE also contains IL-15 to enhance ML NK activation and survival. We also explore whether both technologies function synergistically to prevent immune escape. Methods: Treatment of AML with ML NK cells vs CD123 CAR ML NK cells, ML NK cells plus CD33 TriKE, or the combination, was compared in series of short and long-term assays. AML targets include MOLM13, THP-1, and Primary AML blasts. Interferon gamma production was assessed in a 6-hour co-incubation. In-vitro killing was assessed at 4-hours using a flow-based killing assay, and at 24 hours using a luciferase-based assay. Tumor control was assessed over 5 days using Incucyte (live cell imaging). Tumor control and survival in vivo was assessed using an THP-1 Xenograft model in NSG mice. Primary AML xenografts using NSG-S mice is under development. Results: Both targeting strategies demonstrate an antigen-specific killing response. Both strategies significantly increase interferon gamma production compared to ML NK cells (CAR vs MOLM13: +16.57%, p=0.0002, Primary AML #34: IFNg +9.1% p=0.0092; CD33 TriKE vs MOLM13: +24.42%, p=.0015, AML34: +9.1% p=0.0092). Both strategies also demonstrated increased AML killing at various E:T ratios (CAR vs MOLM13: p=.0011; AML34: p=.0006; CD33 TriKE vs MOLM13: p<.0001; AML34 p=.0026). Both CD123 CAR and CD33 TriKE improved THP-1 control in vivo (bioluminescence at 23 days-CD123 CAR: p=0.004, CD33 TriKE: p<.0001). Both therapies improved median survival by about 3 weeks (CD123 CAR: p<.0001, CD33 TriKE: p=.0001). Conclusion: CD123 CAR ML NK cells and ML NK cells treated with CD33 TriKE both enhance the abilities of ML NK cells to kill AML in vitro and in vivo, and prolong time of survival of THP-1 bearing mice. Treatment of CD123 CAR ML NK cells with CD33 TriKE may prevent antigen escape. Either of these targeting strategies are highly translatable to the clinical setting. Citation Format: Emily Phillips, Lyra Morina, Lynne Marsala, Wilbur Song, Michelle Becker-Hapak, Yoojin Ahn, Sushanth Pureti, Elizabeth Juarez Diaz, Sadia Afrin, Pamela Wong, Jennifer Tran, Joseph Rueve, Allison Burdi, Nancy Marin, Melissa Marie Berrien-Elliott, Martin Felices, Jeffrey S. Miller, Todd A. Fehniger. Memory-like NK cell targeting to CD123 and CD33 improves killing of acute myeloid leukemia [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 135.

Calculations of equifunctional dose of IL15 compared to PSMA TriKE.

Figure S3. Divergent Patterns of Stress and Exhaustion Related to NK Cell Identity and Setting. Related to Figure 3.

BACKGROUND:N-803, an IL-15 superagonist, is currently being studied in clinical trials as a treatment to reverse HIV latency. However, its effects on the gut microbiome are not well understood. METHODS:In this exploratory longitudinal metagenomic study, we analyzed fecal microbiomes from 10 ART-suppressed people with HIV at four different timepoints before, during, and after N-803 treatment. RESULTS:Overall taxonomic and functional diversity did not change significantly, yet beneficial microbial taxa and pathways were nominally enriched after N-803. Specifically, the relative abundance of Faecalibacterium prausnitzii showed a nominal increase after N-803, whereas histidine degradation pathways, often associated with pro-inflammatory mucosal state, decreased. A higher baseline microbial diversity correlated with stronger CD8+ and natural killer (NK) cells activation and reduced frequency of rectal HIV RNA+ cells. MaAsLin2 analyses further identified potentially important associations between short-chain fatty acid (SCFA)-producing taxa and pathways with increased immune activation markers. CONCLUSIONS:These findings in a limited Phase 1B clinical study suggest that gut microbiome diversity prior to immunotherapy may influence host response. These results provide a basis for further investigation into microbiome-based strategies to improve efforts to cure HIV.

Immunotherapies putatively require tumor-specific T cells. Here we show how T cells can eliminate tumors without tumor specificity via paracrine signaling. Activating unexhausted bystander non-tumor-specific T cells within tumors resulted in tumor elimination without conventional recognition-dependent mechanisms and in the absence of any tumor-specific T cell receptor (TCR)αβ+ T cells. Robust T cell activation recruited immune cells, used innate leukocytes and triggered a tumoricidal combination of effector molecules and panoptotic pathways. Mechanistically, interferon-γ, tumor necrosis factor and nitric oxide induced caspase-dependent death, recapitulating melanoma clearance in mice or human melanoma cell death in vitro. Gene expression signatures associated with this response in mice were predictive of survival among human patients with melanoma. Thus, triggering productive T cell activation within tumors can be sufficient for immunotherapy, without needing to induce or rescue cancer-specific responses. Here the authors show that T cell activation within tumors triggers tumor clearance that is independent of tumor antigen recognition and occurs via a paracrine mechanism involving effector cytokines, nitric oxide, leukocyte recruitment and innate immune cells.

Figure S1. Unbiased Clustering and Corresponding Phenotypic Annotations for NK Cells Across Cancer. Related to Figure 1.

Natural killer (NK) cells are promising platforms for off-the-shelf immunotherapy, yet nonviral precision engineering remains limited by poor HDR efficiency, DNA toxicity, and manufacturing challenges. The aim of this study was to establish a high-yield, nonviral knock-in platform. Through extensive in-depth rational screens, we achieved ∼90% HDR insertion of therapeutic payloads while maintaining 100% postediting recovery. By hijacking endogenous transcriptional programs, we installed genetic circuits into defined genomic loci to tune transgene expression. To enable context-dependent therapeutic responses, we integrated a synthetic positive feedback circuit at the CISH locus, which enhanced NK cell persistence and drove strong expression of anti-CD22/19 dual CAR. A hypoxia-responsive IL-12 circuit gated by the PFKFB4 promoter restored cytotoxicity under environmental stress. Finally, we showed this platform is compatible with GMP manufacturing and supports clinical-scale expansion. These findings provide a scalable framework for programmable, nonviral editing of NK cell effector functions for therapeutic and research applications.

BACKGROUND:Glioblastoma is the most aggressive primary brain tumor, with poor prognosis and limited treatment options. Natural killer (NK) cell therapy is a promising immunotherapeutic strategy, yet its efficacy remains limited. We evaluated FT538, a clinical-grade NK product derived from induced pluripotent stem cells (iPSCs), in glioblastoma models. METHODS:FT538, engineered with a high-affinity non-cleavable CD16 Fc receptor, a membrane-bound IL-15/IL-15Rα fusion protein, and CD38 knockout, was tested against 13 patient-derived glioblastoma stem-like cells (GSCs) in vitro and in orthotopic xenograft models. Intracranial persistence and neurotoxicity were assessed in mice. Surface proteomics identified therapeutic targets, and a B7-H3-targeted tri-specific killer engager (TriKE) was evaluated with FT538 and NKG2C+ adaptive NK cells. RESULTS:GSCs were classified as sensitive (38%), moderately sensitive (38%), or resistant (23%) to FT538. Intracranial administration in mice was well tolerated, persisted for at least 35 days, and caused no neurotoxicity. A single intratumoral dose induced complete regression in sensitive xenografts. Surface profiling identified B7-H3 as a target to overcome resistance. Combination therapy with FT538 and a B7-H3 TriKE enhanced antitumor efficacy in resistant models, an effect also observed with adaptive NK cells. CONCLUSIONS:FT538 exhibits potent tumoricidal activity in 77% of GSC lines (NK-sensitive and moderately sensitive), with curative potential in sensitive models, and demonstrates favorable persistence and tolerability in vivo. B7-H3-targeted TriKE restores NK sensitivity in resistant tumors. These findings provide a strong preclinical rationale for further clinical evaluation of FT538, alone or combined with B7-H3-targeted TriKE, for glioblastoma and other solid tumors.

Secretome analysis of supernatant from co-culture of NK cells with monocytes or MDSCs.

Figure S4. Associations with TGFB1 and IFNG Expression in NK Cells. Related to Figure 4.

Representative images (10X magnification) from the IncuCyte live cell imaging assay.

B7-H3 acts as a checkpoint inhibitor, influences the tumor microenvironment, and is necessary for osteoclast (OC) development. Given its dual roles in promoting cancer survival and bone breakdown, we looked at its expression pattern in >200 multiple myeloma (MM) patients and association with cytogenetic abnormalities and outcomes. We found that most newly diagnosed patients had B7-H3+ plasma cells or stroma regardless of cytogenetic risk and expression on plasma cells increased with subsequent relapses. In marrow aspirates and core biopsies we identified B7-H3+ myeloid derived suppressor cells (MDSC) and (OC). We found that high B7-H3 expression was associated with pathologic fractures, bone lesions, and worse progression free survival (PFS). To confirm the utility of B7-H3 as a target in MM, we tested a tri-specific killer engager (TriKE) that binds CD16 on natural killer (NK) cells and B7-H3 while also delivering a recombinant IL-15 molecule to promote NK cell activity. The B7-H3 TriKE restored the ability of patient-derived NK cells to recognize and kill MM and enhanced NK cell-mediated killing of tumor, MDSC, OCs, and fibroblasts. These data suggest that targeting B7-H3 may positively impact outcomes and prevent the development of new bone lesions. In addition to its ability to directly target MM and immunosuppressive cells in the tumor microenvironment, the B7-H3 TriKE improved cytokine secretion and polyfunctionality of NK cells in a single cell assay. These findings advance our understanding of B7-H3’s role in MM pathogenesis and validate the use of the B7-H3 TriKE to improve therapeutic outcomes in MM.

Reagent list for time-of-flight mass cytometry (CyTOF).