Precision off-the-shelf natural killer cell therapies for oncology with logic-gated gene circuits

Article Precision off-the-shelf natural killer cell therapies for oncology with logic-gated gene circuits Graphical abstract Highlights d A three-input logic gate with activating and inhibitory CARs is optimized in NK cells d Logic-gated NK cells kill leukemic stem cells and blasts, not healthy stem cells d Selective killing of tumor cells is validated in vivo in a mixed target cell model Authors Nicholas W. Frankel, Han Deng, Gozde Yucel, ..., Russell Gordley, Timothy K. Lu, Brian S. Garrison Correspondence [email protected] (N.W.F.), [email protected] (T.K.L.), [email protected] (B.S.G.) In brief Treating acute myeloid leukemia with conventional, single-target CAR cell therapies is challenging due to tumor heterogeneity and because tumor antigens are found on hematopoietic stem cells. Frankel et al. design logic- gated CAR-natural killer cells that simultaneously detect three antigens on prospective target cells to safely decide whether to kill them. CAR-NK Cell with an OR-NOT Logic Gate for AML model cancer cells mixed target model FLT3 FLT3 CD33 CD33 FLT3 EMCN EMCN CD33 model healthy cells unmodified NK cells OR-NOT gated CAR-NK cells gg EMCN (protective antigen) Frequency 0 104 105 10610 0 20 40 60 80 100 0 104 105 10610 0 20 40 60 80 100 cancer healthycancer healthy precise killing NOT LSC Healthy HSC Blast tumor hetero- geneity challenge in AML logic gate target cell outcome gene circuit target antigen on-target/ off-tumor toxicity OR In Vivo Validation of NOT Gate Protection bivalent activating CAR bivalent activating CAR inhibitory CAR inhibitory CAR >90% Frankel et al., 2024, Cell Reports 43, 114145 May 28, 2024 ª 2024 The Authors. Published by Elsevier Inc. https://doi.org/10.1016/j.celrep.2024.114145 ll Article Precision off-the-shelf natural killer cell therapies for oncology with logic-gated gene circuits Nicholas W. Frankel, 1,4,* Han Deng,1,4 Gozde Yucel,1 Marcus Gainer,1 Nelia Leemans,1 Alice Lam,1 Yongshuai Li,1 Michelle Hung,1 Derrick Lee,1 Chen-Ting Lee,1 Andrew Banicki,1 Mengxi Tian,1 Niran Almudhfar,1 Lawrence Naitmazi,1 Assen Roguev,1 Seunghee Lee,2 Wilson Wong,2 Russell Gordley,1 Timothy K. Lu, 1,3,* and Brian S. Garrison 1,5,* 1Senti Biosciences, Inc., South San Francisco, CA 94080, USA 2Boston University, Boston, MA 02215, USA 3Massachusetts Institute of Technology, Cambridge, MA 02139, USA 4These authors contributed equally 5Lead contact *Correspondence: [email protected] (N.W.F.), [email protected] (T.K.L.), [email protected] (B.S.G.) https://doi.org/10.1016/j.celrep.2024.114145 SUMMARY Acute myeloid leukemia (AML) is an aggressive disease with a poor prognosis (5-year survival rate of 30.5% in the United States). Designing cell therapies to target AML is challenging because no single tumor-associated antigen (TAA) is highly expressed on all cancer subpopulations. Furthermore, TAAs are also expressed on healthy cells, leading to toxicity risk. To address these targeting challenges, we engineer natural killer (NK) cells with a multi-input gene circuit consisting of chimeric antigen receptors (CARs) controlled by OR and NOT logic gates. The OR gate kills a range of AML cells from leukemic stem cells to blasts using a bivalent CAR targeting FLT3 and/or CD33. The NOT gate protects healthy hematopoietic stem cells (HSCs) using an inhibitory CAR targeting endomucin, a protective antigen unique to healthy HSCs. NK cells with the combined OR-NOT gene circuit kill multiple AML subtypes and protect primary HSCs, and the circuit also works in vivo. INTRODUCTION Chimeric antigen receptors (CARs) are synthetic transmembrane receptors that recognize tumor-associated antigens (TAAs) on a target cell surface and redirect the cytotoxic capabilities of im- mune cells such as T cells or natural killer (NK) cells against these cells. CAR-T and CAR-NK products have shown promise as cancer therapies for select blood cancers and are being clinically investigated for a wide variety of liquid and solid tumor indications. NK cells present several possible advantages over T cells, including their innate anti-tumor activities, reduced pro- pensity to cause systemic toxicity, and low alloreactivity, which may facilitate the manufacture of lower-cost, off-the-shelf products.1 Existing single-target modalities, such as antibodies, anti- body-drug conjugates, and conventional CAR-based cell thera- pies, rely on a single TAA to distinguish between cancer and healthy cells. When such TAAs are also found on normal tissues, however, there is a risk of on-target/off-tumor toxicity that funda- mentally limits the therapeutic window between efficacy and safety.2 As a result, many cancers that lack a single ‘‘clean’’ TAA, such as AML, still constitute a major untreated disease burden. AML is an acute leukemia characterized by an accumulation of malignant immature white blood cells due to differentiation blockade. It is the most common type of acute leukemia in adults, constituting 80%–85% of cases, 3,4 and is the second most common—as well as the deadliest—in children. 5 Conven- tional therapy consists of remission induction therapy followed by consolidation, in which multiple courses or high doses of chemotherapy are followed by hematopoietic stem cell (HSC) transplantation.6 Even so, /C2470% of patients relapse within 3 years.7,8 Unfortunately, the successes of CAR-based cell therapies have not yet been translated to treatment of AML, likely due to the lack of a single suitable lineage-restricted TAA that is: (1) ex- pressed on both AML blast and leukemic stem cell (LSC) popu- lations; and (2) not expressed on healthy cells, such as HSCs (refer to Table S4 for immunophenotypes of these cellular sub- populations). CD33 (SIGLEC-3) and FLT3 (CD135) are validated therapeutic targets for AML. CD33 is a myeloid differentiation marker9 expressed on AML and currently targeted in the clinic using the Food and Drug Administration (FDA)-approved drug, gemtuzumab ozogamicin (Mylotarg). 10 FLT3 is a potential CAR target for AML blasts and LSCs,11–14 which is promising because recent studies suggest that AML relapse is associated in part with the LSC population. Targeting LSCs has been challenging, however, due to the risk of on-target/off-tumor toxicity. 11 For example, while FLT3 is a promising CAR target for AML, FLT3 CAR cell therapies have Cell Reports 43, 114145, May 28, 2024 ª 2024 The Authors. Published by Elsevier Inc. 1 This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/). ll OPEN ACCESS remained unavailable in the United States and Europe due to concerns over toxicity against healthy FLT3 + HSCs and early progenitors cells (HSPCs), 15,16 as has already been observed in FLT3 CAR-T cell and bispecific T cell engager preclinical studies.17–20 While B cell aplasia that results from anti-CD19 CAR-T therapy can be clinically managed with immunoglobulin replacement therapy, 21 anti-AML T cell-based therapies with off-tumor toxicity toward healthy HSPC populations have the po- tential to cause significant and life-threatening treatment compli- cations, including anemia, thrombocytopenia, or even bone marrow failure.22 Therefore, safe and effective cell therapies for AML will require tools to shield healthy cells from the deleterious effects of anti-AML CARs and must target both AML blasts and LSCs despite the fact they often have different TAAs. The lack of clean TAAs could be overcome if therapeutic mo- dalities could instead detect multiple antigens and control cyto- toxic activity based on which combination of antigens are absent or present. To measure cancer-specific signatures 23 rather than just individual TAAs, we designed CAR-based logic gates in NK cells that detect and respond to multiple inputs. Specifically, we created OR logic-gated CARs to target multiple tumor subpopu- lations and to mitigate the risk of TAA expression loss. 24,25 Furthermore, we designed a NOT logic gate to enable CAR-NK cells to detect the presence of a protective antigen (PA) on the surface of healthy cells and thus protect them from being killed due to on-target/off-tumor expression of dirty TAAs. We validated these logic gates individually and in concert, both in vitro and in vivo , and confirmed the ability of OR-NOT gate CAR-NK cells to target TAAs of multiple AML subpopula- tions while sparing primary healthy human HSCs from on- target/off-tumor toxicity. This implementation of OR-NOT logic gating in CAR-NK cells against clinically relevant cancer antigens has the potential to enhance efficacy and precision for the treat- ment of AML and is being advanced toward first-in-class clinical trials in patients. RESULTS Most TAAs are not restricted solely to cancer cells, so T or NK cells expressing a TAA-specific CAR may recognize and kill TAA-expressing healthy cells ( Figure 1 A). A protective mecha- nism is needed to prevent this on-target/off-tumor toxicity and thus maximize treatment efficacy and precision. NOT gates—components that invert signals—can be used to build circuits that shut down cellular activity in response to the pres- ence of an ‘‘off’’ signal (Figure 1 B). To create such a circuit in the context of CAR-NK cells, we designed an inhibitory CAR (iCAR) to recognize a PA on healthy target cells and inhibit acti- vating CAR (aCAR)-mediated killing of those healthy cells (Figure 1 C). FLT3 is a clinically validated TAA in AML that is also expressed on healthy HSCs ( Figure 2 A). There are multiple FDA-approved small-molecule drugs that distinguish between FLT3 + cancer cells and FLT3 + healthy cells by targeting cancer-specific muta- tions in the FLT3 intracellular domain. Targeting FLT3 with anti- body-based or CAR-based therapies has been challenging, however, because there are no such distinguishing mutations in the extracellular domain. Therefore, to distinguish these two cell types, we screened for ‘‘negative’’ targets marking healthy cells rather than conventional ‘‘positive’’ targets marking cancer cells (Figure 2B). Understanding that a curative therapeutic approach for AML likely must target the LSC population in order to prevent relapse of disease in patients, our approach was to combine (1) an anti- FLT3 aCAR targeting AML with (2) an iCAR that detects and protects healthy HSCs, enhancing the therapeutic window. To identify PAs expressed by HSCs but not AML, we used a bioin- formatics pipeline ( Figure 2 B) that filtered for membrane pro- teins, expression on HSCs, and low RNA counts on AML, which nominated several candidates which we then profiled by flow cytometry. Our manual curation process of the resulting hits inhibitory CAR activating CAR anti-tumor cytokines cytotoxic granules activationactivation protectionprotection killingkilling inhibitioninhibition cancer cell immune cell healthy cell killingkilling toxicitytoxicity AND KILL NOT TAA PA PA PA TAA effector cell output target cell PA –– no response –+ no response +– target killing ++ target protection On-target off-tumor toxicity limits CAR efficacy NOT gate shuts down killing in response to PA NOT gated CAR-NK cells using inhibitory CARs to mitigate on-target off-tumor toxicity AB C CAR TAA TAA NOT gate components Figure 1. NOT gates based on inhibitory CARs can be used to make CAR-NK cells with enhanced precision and wider therapeutic windows (A) Therapeutic immune cells with activating CARs (aCARs) targeting tumor-associated antigens (TAAs) can kill healthy cells that express the TAA (o n-target/off- tumor toxicity). We have discovered cell-surface proteins that are selectively expressed on healthy cells, not but cancer cells, that can be recognized as protective antigens (PAs). (B) NOT gates invert signals, allowing detection of a PA to shut down TAA-targeted killing of healthy cells. (C) We implemented NOT gates in NK cells by designing inhibitory CARs (iCARs) that recognize PAs and shut down aCAR-triggered killing in response to TA As, sparing healthy cells. 2 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS consisted of consideration of the HSC literature, which narrowed our choice to endomucin (EMCN) ( Figure 2C) due to the fact that it was previously demonstrated to be a marker that is robustly expressed on highly functional HSCs possessing long-term multi-lineage reconstitution potential. 26 We confirmed that the healthy HSC compartment in primary human bone marrow cells expressed EMCN at a frequency of up to 70%, while patient- derived AML LSC samples demonstrated no appreciable expression (Figures 2D vs. 2E). Next, we designed an aCAR/iCAR receptor pair to kill FLT3 + AML cells while protecting EMCN+ healthy cells from CAR-medi- ated toxicity ( Figure 3 A, left). We created an FLT3 aCAR with a CD28-derived co-stimulatory domain and CD3z stimulatory domain (aFLT3-28z). Like the aCAR, the iCAR had extracellular antigen-recognition and hinge domains and a transmembrane domain. Unlike the aCAR’s intracellular domains (ICDs), which are derived from T cell receptor signaling components with im- munoreceptor tyrosine-based activation motifs (ITAMs), the Cell type mRNA levels EMCN AML blastsAML LSCs Healthy HSCNeutrophils 2 3 4 5 6 7 8 Cell type log2 microarray expression AM L blasts AML LSCsHealthy HSCNeutrophils 4 6 9 6 8 10 12 12 14log2 microarray expression log2 microarray expression FLT3 expression EMCN is specifically expressed on HSCs, not AMLIdentification of HSC protective antigens Hematopoetic stem cells express FLT3, an antigen associated with AML blasts and LSCsA B CD E 20,216 protein coding genes in NCBI 7,680 genes 115 genes 15 genes 1 gene membrane proteins high HSC/AML expression ratio confirm low AML expression manual curation surface protein EMCN SSC primary AML LSCs 0% EMCN+ primary HSCs 70% EMCN+ AML blast AML LSC HSC MPP LPMP CMP GMP MEP Early PM Late PM PM MY MM Band cells MWN Monocytes Neutrophils Eosinophils DC Erythroblasts Pro-B cells Pre-B cells Naive B cells B cells Memory B cells CD4, naive CD4, central CD4, effector CD4 T cells CD8, naive CD8, central CD8, effector CD8 T cells iNKT cells NK cells NK cells, BM NK cells, thymus Figure 2. Discovery of EMCN as a PA for healthy HSCs during CAR-mediated therapy against AML (A) FLT3 expression was compared between AML and healthy hematopoietic cell subpopulations (left; for additional subpopulation details, see Table S4), with focused comparative analysis between AML blasts (n = 825), AML patient LSCs (n = 28), and healthy donor HSCs (n = 22), with healthy donor neutrophils (n =2 )a s a control (right). (B) Bioinformatics pipeline summary for identifying HSC-specific PAs not expressed on AML. (C) EMCN expression was compared between the same groups as (A, right), showing high expression in healthy HSCs compared to AML blasts or LSCs. (D and E) EMCN protein expression was assayed by flow cytometry in patient-derived AML LSCs (D) and primary HSCs from healthy donors (E), validating selective expression of EMCN on HSCs. SSC, side scatter. In this figure, box plots follow Tukey’s convention, with outliers omitted from (A) for visual clarity. Statistical significance determined using ANO VA (****p < 0.001). Cell Reports 43, 114145, May 28, 2024 3 Article ll OPEN ACCESS iCAR had ICDs adapted from immune checkpoint receptors with immunoreceptor tyrosine-based inhibitory motifs (ITIMs). This design takes advantage of the rapid, protein-based signaling of ITIMs, which recruit phosphatases SHP-1, SHP-2, and SHIP-1 to the immune synapse, blocking phosphorylation cas- cades arising from activated ITAMs on the intracellular tails of activating receptors (e.g., DAP12, CD3z, and FceRIg). 27 We packaged the aFLT3-28z and aEMCN iCAR constructs into viral vectors and co-transduced them into NK cells to create NOT gate CAR-NK cell populations. We also constructed a control NOT gate, which was targeted against a dummy antigen, HER2, not expressed on any target cells. Antibiotic selection for the iCAR construct was used to ensure that aCAR + NK cells were also iCAR + (Figure S2A). NK cells expressing NOT gates with these various iCARs were then screened in parallel (Figure 3A, upper right) for their ability to kill ‘‘on-tumor’’ target cells (a leukemia cell line, SEM, endoge- nously expressing FLT3) while sparing ‘‘off-tumor’’ target cells (the same cell line engineered to express EMCN). We compared NOT gate performance by calculating the reduction in killing be- tween on-tumor and off-tumor targets ( Figure 3 B, lower right). Higher killing reduction signified more potent inhibitory signaling by the iCAR intracellular domain. The control NOT gate with a non-targeting binder (gray bars in Figures 3B and 3C) resulted in equal killing of on- and off-tumor cells, thereby exhibiting no killing reduction, signifying no NOT gate function. NOT gates with functional iCARs ( Figure 3 B, blue bars) had positive killing reduction (i.e., lower killing of protective antigen positive target cells protective antigen positive target cells parental target cells H TM CD28CD3zanti-FLT3 scFv activating CAR perform killing assays quantify NOT gate performance Killing reduction on-tumor killing= – off-tumor killing On- tumor Off- tumor HT Minhib. domainanti-EMCN scFv inhibitory CAR native inhibitory receptors native binding domains ITIMs native TM and inhib. domains iCAR domain screening and optimization in NK cells LIR1 iCAR maximizes killing reduction NOT gate protects safety antigen-positive target cells from both CAR-mediated killing and cytotoxic cytokine secretion C BA control NOT gate anti-PA NOT gate 0 10 20 30 40CAR-mediated killing (%)SA: control NOT gate anti-PA NOT gate CAR-mediated killing (%)SA: control NOT gate anti-PA NOT gate SA: 0 2 4 6 8 10 12 14 p<10-6 TNFα secretion (pg/mL) NOT gate is still functional when on-tumor and off-tumor cells are mixed D 0 10 20 30 40 parental target cells inhibitory CAR intracellular domain optimal control sub-optimal LIR1 KIR2DL1 LAIR1 LIR3 LIR2 SIGLEC-10 SIGLEC-2 LIR5 NKG2A KIR3DL1 BTLA KLRG1 control iCAR 0 5 10 15 20 25 30 Killing reduction (% pts.) Figure 3. Design and optimization of ‘‘FLT3 NOT EMCN’’ logic gate in CAR-NK cells (A) We developed a systematic approach to design the NOT gate by screening different transmembrane and intracellular domains (TM-ICDs) taken from na tive inhibitory proteins in the context of an anti-EMCN iCAR, which were co-transduced into NK cells with an aFLT3 aCAR. Constructs were evaluated based on the difference between on-tumor (FLT3+EMCN–) killing and off-tumor (FLT3+EMCN+) killing (or ‘‘killing reduction’’). scFv, single-chain Fv (variable fragment); H, hinge; TM, transmembrane domain; ITIM, immunoreceptor tyrosine inhibitory motif. (B) Comparison of NOT gates using aFLT3 aCAR combined with aEMCN iCARs with different TM-ICDs, demonstrating optimality of the LIR1-derived TM-ICD (purple) by its high killing reduction. Control: aFLT3 aCAR with LIR1-based iCAR targeting an irrelevant antigen. (C) CAR-mediated killing (bars) and TNF- a secretion (circles) of NK cells in (B) in response to on- or off-tumor cells. Anti-PA NOT gate (aFLT3-28z and aEMCN- LIR1) significantly reduced aCAR response in a PA-dependent manner. (D) Mixed target challenge. Same cell and antigen system as (C) except that both target cell types were mixed in the same container, showing the same lev el of protection. In this figure, values represent the mean of three technical replicates, and error bars represent ±SE of mean. Welch’s test was used to discern significant dif- ferences. 4 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS no response target killing KILLFLT3 FLT3 CD33 CD33 effector cell output target cell type – – + + – + – + OR + OR gate CAR uses a bivalent scFv to to achieve multiple antigen targeting with a single receptor OR gate design principles OR gate CAR is comparable to mono-CARs against matched targetsIn vivo model for OR logic OR gate CAR-NK cells increase survivalOR gate CAR-NK cells clear leukemia cells vivo Tumor associated antigens on leukemic stem cells vs. blasts OR gated CAR-NK cells or x3.5e7 time (days) IV injection bioluminescence imaging 04 7 1 1 VH anti-FLT3 VH anti-CD33VL VL anti-FLT3 VH anti-CD33 VH CD8 SS myc tag CD8 hinge CD28 TMDlinker linker linker CD28 costim CD3zanti-CD33 VL anti-FLT3 VL A D E F GH B OR gate targets blasts and LSCsC -4 -3 -2 -1 0 log10 fold change vs. PBS control CD33hi, FLT3lo leukemia cell line FLT3hi, CD33lo leukemia cell line FLT3hi CD33lo FLT3lo CD33hi Reduction in leukemia cell burden in vivo Untransduced PBS No NK cells (PBS) 30 days post-implantation p<0.005 for all comparisons Treatment group Untransduced NK cells Untransduced NK cells NK cell type anti-FLT3 CAR anti-CD33 CAR FLT3 OR CD33 CAR OR gate CAR-NK cellsOR gate CAR-NK cells -1.5 -1.0 -0.5 0.0 02 0 4 0 6 0 8 0 0 20 40 60 80 100 Time (days) Survival (%) bivalent scFv 1e9 tumor burden (photons/s) 5e6 leukemic progenitor cell LPC leukemic stem cell LSC FLT3 leukemic blast cell Blast CD33 FLT3 expression CD33 expression blast blast, LSC LSC CD33 FLT3 FLT3 OR gate CAR mono-CARs CD33 blast LSC Figure 4. OR-gated CARs enable NK cells to target multiple AML antigens (A) In AML, less-differentiated LSCs tend to express more FLT3, while more-differentiated blasts tend to express more CD33. (B) OR gates are circuit components that activate in the presence of either or both inputs. (C) CARs that target only FLT3 or only CD33 cannot target all AML blasts or LSCs, respectively, while OR gate CARs targeting both antigens may recognize multiple AML subpopulations. (D) OR gate CAR structure based on loop bivalent scFv with both FLT3 and CD33 recognition domains. VH and VL, variable heavy and light domains. (E) In vivo validation of OR gate CAR-NK cell killing in an endogenously FLT3 +CD33+ MV4-11 leukemia xenograft model expressing luciferase. Bioluminescence imaging shows CAR-mediated tumor clearance (right) in comparison with untransduced NK cells (middle). Left: PBS treatment control. (legend continued on next page) Cell Reports 43, 114145, May 28, 2024 5 Article ll OPEN ACCESS off-target than on-target cells), signifying varying levels of NOT gate function. The NOT gate using an iCAR derived from leuko- cyte immunoglobulin-like receptor subfamily B member 1, or LIR1 ( Figure 3 B, purple; Figures 3 C and 3D, red) demonstrated the greatest level of killing reduction and therefore the highest level of NOT gate function. The NOT gate based on the LIR1 iCAR showed minimal tonic inhibition of aCAR function in the absence of EMCN expression on targets relative to the control NOT gate (Figure 3C). Moreover, it was able to suppress the secretion of tumor necrosis factor a (TNF-a)( Figures 3C [right] and S3A) and interferon-g (Figure S2C) in an EMCN-dependent fashion. The high performance of the LIR1 iCAR was not dictated by high expression level alone, as other iCARs with similar expression performed substantially worse ( Figure S3 B). Statistically significant EMCN-dependent suppression of toxicity was robust across multiple biological replicates (Figure S2D). EMCN-dependent reduction in killing was also observed when EMCN+ and EMCN– target cells were premixed and co-cultured with NOT gate CAR-NK cells with the aEMCN-LIR1 iCAR (Figures 3 D and S2B). These data suggest that the ‘‘FLT3 NOT EMCN’’ circuit enables cell-by-cell decision making with no loss of function in a well-mixed environment, implying that the circuit works on a fast timescale, as would be expected from the ITIM-mediated mechanism of action based on protein recruitment and modification. This is particularly important for clinical applications in oncology, where healthy cells and cancer cells are often in close proximity to each other. For example, in AML, cancer cells intermingle continuously with healthy hemato- poietic cells. Since no single TAA can be used to wholly target all AML sub- populations, we developed a bioinformatics pipeline (Figure S4A) to identify target antigen pairs which together possess the po- tential to concurrently eliminate both LSCs and blasts. FLT3 is often more associated with less-differentiated hematopoietic cells, including LSCs, while CD33 is often more highly expressed within more-differentiated myeloid cells, including blasts, 28 and inconsistently expressed among LSCs29 (Figure 4A). Concurrent targeting of both FLT3 and CD33 should: (1) produce an inclusive targeting approach covering all AML subpopulations; (2) help prevent tumor antigen escape; and (3) increase total killing of the many AML cells that are double positive for FLT3 and CD33 expression. Thus, this OR gate strategy, which activates when either of two inputs (or both) is detected (Figure 4B), should have advantages over conventional ‘‘mono-CARs’’ that only target one subpopulation ( Figure 4C). We created an ‘‘FLT3 OR CD33’’ circuit using a bivalent aCAR design containing two binders, one for CD33 and one for FLT3, combined in a loop configuration 25 (Figure 4 D). As with the aFLT3 aCAR above, the intracellular domain of the OR gate CAR consisted of CD28 co-stimulatory and CD3z stimulatory domains. We tested in vivo efficacy of OR gate CAR-NK cells within an AML xenograft immunocompromised mouse model generated with patient-derived MV4-11 AML cells, which endog- enously express FLT3 and CD33. Target and effector cells were both injected on day 0. This approach has been shown to be clinically translatable in the field of CAR-NK cell therapy, with re- sults having been successfully translated from bench 30 to bedside31,32 and back to bench, using the same approach again with patient samples. 33 Other preclinical studies 34,35 in the field have taken this approach as well. Bioluminescence imaging (BLI) of tumor-expressed luciferase (Figures 4 E and S4B) revealed the innate cytotoxic behavior of untransduced NK cells compared to a saline (PBS) control (mid- dle vs. left groups), but OR gate CAR-NK cells exhibited substan- tially higher suppression of tumor burden than untransduced NK cells alone (right vs. middle group). As a result of the OR-gated CAR-mediated tumor killing, mouse survival was also signifi- cantly extended compared to the other treatment groups (Figure 4F). After confirming the cytotoxic activity of the OR-gated CAR in vivo, we sought to verify its logic gating performance in com- parison to mono-CARs that solely bind only FLT3 or CD33. We created an in vivo OR gate vs. mono-CAR challenge by injecting two different groups of immunocompromised mice with either SEM leukemia cells, which endogenously express high levels of FLT3 and low levels of CD33, or MOLM-13 leukemia cells, which endogenously express low levels of FLT3 and high levels of CD33. Both patient-derived cell lines expressed luciferase, al- lowing us to track tumor burden and progression, as well as CAR-NK cell anti-tumor activity compared to a PBS control (Figure 4G). As expected, based on antigen expression levels, anti-FLT3 mono-CAR-NK cells statistically significantly reduced tumor burden in the FLT3 hi SEM high tumor model but not in the FLT3lo MOLM-13 tumor model ( Figure 4 H, blue box in left vs. right plot; Figure S5 ). Anti-CD33 mono-CAR-NK cells, on the other hand, showed the expectedly reversed trend by specif- ically killing the CD33 hi tumor model but not the CD33 lo tumor model ( Figure 4 H, green box in left vs. right plot). Importantly, OR-gated CAR-NK cells with both FLT3 and CD33 binders reduced tumor burden in both models, to a similar extent as each mono-CAR in its own ‘‘specialized’’ tumor model ( Fig- ure 4H, red box in both left and right plots). Thus, the OR-gated CAR results in an increased breadth of targeting across different heterogeneous leukemia models with varying expression levels of AML antigens FLT3 and CD33. We combined OR and NOT gate components to make a ‘‘(FLT3 OR CD33) NOT EMCN’’ circuit that targets both AML LSC and blast antigens while also protecting vulnerable (F) Kaplan-Meier curve of mouse study in (E), showing survival extension with OR gate CAR-NK cell treatment (red) relative to other groups. All groups are significantly different from each other ( p < 0.005, log-rank test, 4–5 mice per group). (G) In vivo OR gate challenge: NK cells were injected into two groups of mice, each engrafted with either SEM cells (FLT3hi, CD33lo) or MOLM13 cells (FLT3lo, CD33hi). (H) Change in tumor burden (measured by BLI) relative to PBS treatment control in each of the two challenge groups, comparing mono-CARs (blue, green) t o the OR gate CAR (red; day 11 post injection). Mono-CARs each target one tumor type, while OR gate CAR (red) targets both. Markers represent mean of at least fi ve mice; error bars represent ±SE of mean. Significant differences assessed with ANOVA (* p < 0.05, *** p < 0.005). 6 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS EMCN+ HSCs (Figure 5A). To do so, we designed a construct en- coding an anti-FLT3/anti-CD33 bivalent aCAR, an anti-EMCN iCAR, and a modified form of interleukin-15 (IL-15), a cytokine that can enhance NK cell persistence,36 all on a single tricistronic retroviral payload (Figure 5B), and confirmed high levels of aCAR and iCAR co-expression in NK cells ( Figure 5C). NK cells transduced with the tricistronic construct showed a majority of cells expressing all three payloads, demonstrating AND KILL KILL NOT FLT3 CD33 EMCN ORFLT3 CD33 OR CD8 SS anti-EMCN scFv V5 tag CD8 hinge LIR1 TMD LIR1 ICD CD8 SS myc tag CD8 hinge CD28 TMD anti-FLT3, anti-CD33 bivalent scFv 2A 2A CD28 costimIL-15 CD3z Tricistronic construct encoding NK cell OR-NOT gene circuit with IL-15 cytokine support Targeting AML LSCs and blasts while sparing HSCs using a 3-input OR-NOT gated CAR Coexpression of OR-NOT gate circuit components OR-NOT gate protects primary EMCN+ HSCs A B C OR-NOT gate CAR-NK cells kill both AML blasts and LSCs DE OR gate aCAR 0 10 20 30 40 50 60 leukemia cell line EMCN+ HSCs aCAR-mediated killing (%) NKcell type OR gate CAR-NK OR-NOT gate CAR-NK FLT3 CD33 EMCN Potential HSC toxicity FLT3 CD33 EMCN HSCs protected OR gate OR-NOT gate blast 1 LSC blast 2 LSC blast 3 LSC 0 20 40 60 80 100 AML patient sample Cytotoxicity (%) NKcell type Untransduced NK cells OR-NOT CAR-NK cellsOR gate CAR OR-NOT gate CAR-NK cell iCAR 73.5% 0 -103 103 104 105 106 iCAR expression 0-104 104 105 106 Figure 5. Combining OR and NOT gates to protect primary healthy human hematopoietic stem cells (A) Applying NOT gate architecture to the OR-gated CAR yields an OR-NOT gate circuit that targets multiple AML subpopulations while protecting HSCs. Three- dimensional antigen space reveals how 3-input logic gating is required to achieve this clinical goal by targeting cells defined by specific combinatio ns of the three antigens. (B) Construct design incorporating OR gate aCAR, aEMCN iCAR, and a modified form of IL-15 (to enhance NK cell persistence) into a single tricistronic pa yload. (C) OR gate aCAR and iCAR from construct design in (B) were co-expressed at high levels. Gate was set based on an untransduced NK cell control. (D) Killing of blasts and LSCs from multiple AML patient samples by OR-NOT gate NK cells (green) is enhanced compared to untransduced NK cells (black). (E) OR-NOT gate CAR-NK cells (medium green) reduce killing of EMCN+ HSCs from freshly thawed primary human CD34-enriched bone marrow cells, compared to OR gate CAR-NK cells without the NOT gate (light green). No statistically significant reduction of killing occurs in response to SEM leukemia cells. Values represent the mean of three technical replicates, and error bars represent ±SE of mean. Welch’s test was used to discern significant differences (** p < 0.01). Cell Reports 43, 114145, May 28, 2024 7 Article ll OPEN ACCESS the ability to express multiple logic gates in NK cells along with another clinically important effector ( Figure 5 C). OR-NOT gate CAR-NK cells showed enhanced killing of both AML blasts and LSCs from multiple patient samples ( Figure 5 D). Significant NOT gate performance was confirmed across a large number of NK cell donors (ten of ten tested, Figure S8 A), highlighting the robustness of the gene circuit, even in the context of an aCAR targeting multiple AML antigens and co-expression of a clinically useful activator, IL-15. While we found that aCAR and iCAR responses did scale with antigen density on target cells, we confirmed that NOT gate function was robust across a range of TAA or PA expression levels. The EMCN iCAR maintains a comparable level of protection (20 to 25 percentage points of killing reduction) even when challenged with target cells with 10-fold lower FLT3 expression level ( Figure S8 B). While EMCN iCAR function does depend modestly on target EMCN expres- sion levels, changing EMCN expression level on the target cells by 6-fold results in a change of protection by only 1.3-fold, signi- fying a relatively insensitive relationship ( Figure S8C). We have shown that the NOT gate protects target cells in a PA- dependent manner using an experimental system in which healthy cells are modeled by cell lines overexpressing the PA, but how well does the circuit perform in protecting actual primary EMCN+ HSCs? We compared OR gate CAR-NK cells expressing the ‘‘FLT3 OR CD33’’ gene circuit to OR-NOT gate CAR-NK cells expressing the full ‘‘(FLT3 OR CD33) NOT EMCN’’ gene circuit in a co-culture assay with either SEM leukemia cells or EMCN + HSCs from primary human CD34-enriched bone marrow. We confirmed that CD34-enriched bone marrow samples used in this study included HSCs that expressed both EMCN ( Fig- ure S6A) and FLT3 ( Figure S6B). Both OR gate and OR-NOT gate CAR-NK cells killed cancer cells equally, reinforcing earlier results that, when not engaged, the iCAR does not tonically inhibit cytotoxicity ( Figure 5 E, left group, light green vs. medium green). Both OR gate and OR- NOT gate CAR-NK cells killed cancer cells slightly more than EMCN+ HSCs, which is unsurprising due to NK cells’ innately enhanced cytotoxicity against tumor cells (left vs. right group), which is one of the attributes NK cells possess that make them promising cell types for cell therapies. Crucially, however, the OR-NOT gate CAR-NK cells were significantly less toxic than OR gate CAR-NK cells toward primary EMCN + HSCs (Figure 5E, right group, light green vs. medium green; Figure S7 ), showing that the NOT gate can protect vulnerable primary healthy cells, potentially mitigating the risk of CAR-mediated, on-target/off-tu- mor toxicity. Within the context of well-established clinical proto- cols for hematopoietic cell transplantation, only a portion of the total normal HSC population is transplanted into conditioned transplant recipients, which is sufficient to provide long-term multi-lineage repopulation of an entire healthy hematopoietic sys- tem.37,38 Consequently, it is likely that this level of NOT gate- mediated protection has the potential to be clinically impactful. HSCs are the least differentiated hematopoietic cells and are capable of both self-renewal and multi-lineage repopulation. HSCs give rise to multi-potent progenitors (MPPs), which in turn give rise to all other lineages. Because MPPs express lower levels of EMCN than HSCs, we wanted to determine whether the EMCN iCAR still protected them, although any potential loss of MPPs would be reconstituted by HSCs, which are significantly protected by the iCAR (Figure 5E). We determined that OR-NOT-gated CAR- NK cells also significantly protected MPPs (Figure S9). Finally, we aimed to confirm that the NOT gate can reduce on- target/off-tumor toxicity in vivo , using the full OR-NOT circuit with IL-15 support. To create an in vivo model for assessing on-target/off-tumor toxicity, we employed ‘‘on-tumor’’ and ‘‘off-tumor’’ cell lines that were FLT3 +CD33+EMCN– and FLT3+CD33+EMCN+, respectively. Both cell lines were based on endogenously FLT3 + SEM cells modified to highly express CD33, but only ‘‘off-tumor’’ cells also exogenously expressed EMCN. Before engraftment, these two target cell subpopulations were mixed 1:1. In the following weeks, tumor burden was tracked via BLI of luciferase expressed by target cells. (Because both target cell subpopulations expressed the same luciferase cassette, BLI served only as a measurement of bulk killing.) To determine NOT gate function, peripheral blood was analyzed by flow cytometry to assess the relative abundance of each target subpopulation ( Figure 6 A). A functional OR-NOT circuit should result in an increase in the PA expression among target cells and a specific reduction in on-tumor cells. The initial 50% PA expression among target cells was not appreciably changed by treating with untransduced NK cells (Figure 6 B [representative examples], top row, right plot). When we injected control CAR-NK cells expressing the OR gate with a non-functional iCAR, we observed a drop in abun- dance of tumor cells relative to untransduced NK cells (Figure 6B, middle row, middle plot); however, the lack of a NOT gate in this circuit resulted in roughly equal proportions of on-tumor and off- tumor target cells (Figure 6B, middle row, right plot) due to indis- criminate cytotoxicity against both target populations. Notably, NK cells expressing the complete OR-NOT gate circuit resulted in precise ablation of the on-tumor target cell subpopulation, en- riching the frequency of off-tumor target cells to over 90% of target cells (Figure 6B, third row, middle and right plots). To sum- marize, the OR-NOT gate resulted in a very significant enrich- ment of off-tumor target cells in vivo , while none of the other treatments resulted in any change from baseline ( Figure 6 C [group-averaged results], red box vs. others). In terms of target cell abundance, OR-NOT gate CAR-NK cells significantly killed on-tumor cells beyond the level of innate killing of untransduced NK cells ( Figure 6 D, left). OR-NOT gate CAR- NK cells did not, however, exhibit CAR-mediated killing of off-tu- mor cells, while OR gate CAR-NK cells with a non-functional NOT gate exhibited substantial CAR-mediated on-target/off-tu- mor toxicity (Figure 6D, right). In other words, the addition of the NOT gate abolished CAR-mediated off-tumor cell killing in vivo. The NOT gate effect was also reflected in BLI ( Figure S10A) and at other time points in the study ( Figures S10 B and S10C). The target cell PA expression and target cell abundance results, taken together, demonstrate that the OR-NOT gate can kill tumor cells while minimizing on-target/off-tumor toxicity in vivo. DISCUSSION Synthetic gene circuits afford the ability to engineer cells with complex, customizable phenotypes and input-output re- sponses.39–41 The field of logic-gated CARs is continuously 8 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS off-tumor (healthy) target cells OR-NOT gated CAR-NK cells on-tumor (cancer) target cells off-tumor (healthy) target cells on-tumor (cancer) target cells x1.5e6 x1.5e6 x2.5e7 FLT3+ CD33+ FLT3+ EMCN+ CD33+ ted ells 7 Untransduced NK cells OR gate only CAR-NK cells time (days) peripheral blood flow cytometry IV injection bioluminescence imaging 01 362 0 2 7 Protection of off-tumor cells In vivo model for NOT logic gating OR-NOT gate circuit results in precise killing of on-tumor cells Cytotoxicity with mitigation of off-tumor toxicity target cells 93.5% human cells 4.2% mouse cells 89.1% 0 104 105 106107 human CD45 0 -103 103 104 105 106 107 mouse CD45 NK cells 6.5% 0 104 105 106107 CD56 0 -103 103 104 105 106 107 EMCN off-tumor 51.8% on-tumor 48.2% 0 104 105 10610 EMCN 0 20 40 60 80 100% of mode target cells 77.4% human cells 1.6% mouse cells 94.5% off-tumoron-tumor human CD45 mouse CD45 CD56 EMCN EMCN % of mode 0 104 105 106107 0 -103 103 104 105 106 107 NK cells 22.6% 0 104 105 106107 0 -103 103 104 105 106 107 92.1%7.9% 0 104 105 10610 0 20 40 60 80 100 fff target cells 25.5% human cells 0.1% mouse cells 96.5% off-tumoron-tumor human CD45 mouse CD45 CD56 EMCN EMCN % of mode 0 104 105 106107 0 -103 103 104 105 106 107 NK cells 74.5% 0 104 105 106107 0 -103 103 104 105 106 107 58.3%41.7% 0 104 105 10610 0 20 40 60 80 100 fff OR-NOT gated OR NOT t d CAR-NK cells % off-tumor cells of target cells on-tumor cells (rel. frequency) off-tumor cells (rel. frequency) non-functional intracellular domain precise killing A C D B PBS UntransducedOR gate onlyOR/NOT gate 0 10 20 30 40 50 60 70 80 90 100 Treatment group PBS UntransducedOR gate onlyOR/NOT gate 0.00 0.01 0.02 0.03 Treatment group PBS UntransducedOR gate onlyOR/NOT gate 0.00 0.01 0.02 0.03 0.04 0.05 Treatment group Figure 6. Triple-input OR-NOT gate CAR-NK cells avoid on-target/off-tumor toxicity in vivo with precise killing (A) In vivo mixed target study design. ‘‘On-tumor’’ target cells (FLT3+CD33+EMCN– cancer model) and ‘‘off-tumor’’ target cells (FLT3+CD33+EMCN+ healthy model) were mixed with NK cells and injected into mice. (B) Peripheral blood was collected for flow-cytometry analysis of circulating target and NK cells. Representative data from each group illustrating identification of cell populations. Rows from top to bottom: untransduced NK cells; control NK cells expressing the OR gate CAR with a ‘‘dummy’’ iCAR containing a non- functional ICD; NK cells expressing full OR-NOT gate circuit. Columns from left to right: mouse vs. human cells; target vs. NK cells; on-tumor (gray) v s. off-tumor (purple) target cells. The OR-NOT gate potentiated precise killing of on-tumor cells (red arrow). (C) Percentage of off-tumor target cells (of all target cells) in peripheral blood was specifically enriched by NK cells expressing the full OR-NOT gat e circuit (red), day 20 post implantation. Box: interquartile range with line at median; whiskers: range of data. (D) Target cell abundance (relative frequency among all CD45 + cells) at day 27 post implantation. Left: OR-NOT gate CAR-NK cells (red) result in reduced fre- quency of on-tumor cells in peripheral blood via CAR-mediated killing. Right: without the NOT gate, OR gate CAR-NK cells (dark blue) kill off-tumor ce lls, but the full OR-NOT circuit (red) abolishes this toxicity. Values represent the mean of at least five mice, and error bars represent ±SE of mean. In this figure, significance was tested using ANOVA (* p < 0.05, **p < 0.01, ***p < 0.005, **** p < 0.001). Cell Reports 43, 114145, May 28, 2024 9 Article ll OPEN ACCESS expanding, and many individual ‘‘components’’ have already been described. Advances in the field today focus on combining these components into more complex, clinically oriented cir- cuits. For example, inhibitory CARs have been designed in T cells and NK cells to create NOT gates, but they have not been combined with other logic-gated components to create larger circuits as we have here. Here, we demonstrate a syn- thetic three-input logic gate in NK cells, but it is also noteworthy as a therapeutic gene circuit in any cell type to target three clin- ically relevant antigens in a simultaneous, rather than sequen- tial,42 fashion. By protecting primary EMCN + HSCs, this thera- peutic strategy has the potential to enable a wider therapeutic window for the treatment of AML and better post-treatment reconstitution of a healthy hematopoietic system, which would be medically meaningful if confirmed in clinical trials. It has been proposed that the development of sophisticated cell therapies that sense, integrate, and respond to multiple ligand inputs will become increasingly common. 43,44 In devel- oping such circuits, however, it can be problematic to measure their response only to individual target cell types presented separately. The experimental approach of artificially providing one type of target cell in isolation in vitro may be somewhat appropriate for modeling the physiological setting of some particularly homogeneous solid tumors, provided that cancerous and healthy tissues are sufficiently physically separate in vivo.45 However, in heterogeneous tumors, which include liquid tumors (in which circulating cancer cells are invariably mixed with vulnerable healthy cells) and many solid tumors, the precision of killing will depend on fast-timescale decisions made by thera- peutic cells. Switchable circuits based on protein-protein or protein-small molecule interactions 35,46–51 can signal faster than those based on gene expression 52–54 (for further discus- sion, see also Gordley et al. 55). Our NOT gate distinguishes PA+ from PA – target cells in a mixture potentially because ITIM signaling proceeds through protein recruitment and post-trans- lational modification. 27 These data are also consistent with a model in which the iCAR works primarily in cis, rather than in trans, which is to say by responding to a PA on the same target cell that is already being targeted via an interaction between the aCAR and TAA. Since ITIM-based inhibitory receptors (including our iCARs) act by spatially recruiting phosphatases to the im- mune synapse, whereupon they locally dephosphorylate kinases associated with immune activation, 56 it follows that iCARs should primarily act at the same cell-cell interface as the aCAR’s target. The bivalent CAR demonstrated here can recognize two TAAs with one receptor and responds to target cells that are CD33 +, FLT3+, or both. This is important in AML due to the heteroge- neous nature of the disease, characterized by the presence of LSC and blast subpopulations with differing expression levels of these two antigens. Targeted therapies against CD33 have shown successful killing of AML blasts, but cases of relapse have occurred, likely due to residual LSCs that are low or nega- tive for CD33. By targeting both FLT3 and CD33, we have shown that the OR gate described here can kill both primary blasts and LSCs, increasing its potential to control AML disease. Interestingly, the extent of iCAR protection was revealed to be much higher in our in vivo experiments than initially suggested by overnight co-culture experiments. This might be due to the short timescale (1 day) and supraphysiological cell density in this stan- dard in vitro assay format (i.e., high cell counts seeded in U-bottom plates that tend to accumulate cells at the bottom). In such conditions, activation of CAR-NK cells may be hypersti- mulated, making it artificially difficult for the NOT gate to sup- press cytotoxicity. The fact that NK cells exhibit a strong innate proclivity toward killing cancer cells may account for the super- ficially higher NOT gate protective effect in T cells in similarly arti- ficial short-term in vitro assays using cancer cells to model ‘‘healthy cells.’’57 In the in vitro HSC protection experiments described herein, we chose a high density of cells in the assay to promote a high level of killing to facilitate measurement of the effect of the iCAR. This experimental design allowed us to successfully show that the iCAR significantly reduces toxicity toward HSCs, but, in a more physiological setting, this magnitude of killing would not occur. With this therapeutic strategy, critical HSC loss in vivo is unlikely because of NK cells’ natural tendency to avoid killing normal cells (especially stem cells) under physiolog- ical conditions, HSCs’ substantial capacity for self-renewal, and NK cell therapies’ short-term persistence in vivo. Given the clin- ical success of current hematopoietic cell transplantation proto- cols,37,38 the efficacy and protection aspects of our engineered OR-NOT CAR-NK cell circuit are promising for application to- ward AML. We chose a short-term format for our HSC protection experi- ment because HSCs are not suitable for long-term experiments. This is because they tend to rapidly differentiate when cultured ex vivo , which drastically complicates the interpretability of the experiment. Long-term in vivo experiments using mouse models engrafted with human HSCs have similar pitfalls, including a paucity of bona fide, self-renewing HSCs after engraftment. For studying the system long term, here we instead leveraged an in vivo model using model cell lines, which confirmed that the NOT gate effect is even stronger over the longer term than observed in short-term in vitro experiments. In vivo studies with model cell lines are broadly used in the cell therapy and syn- thetic immunology fields, especially in logic gating studies where complex antigen combinations must be systematically tested in vivo.42,45,47,52,58,59 The longer time frame and circulating envi- ronment of the in vivo experiments are likely to better reflect physiological settings and is where we observed the largest magnitude of effect. Another interesting observation was that the OR-NOT gate CAR-NK cells exhibit stronger cytotoxicity to- ward on-tumor cells compared to OR gate CAR-NK cells in vivo, possibly due to the latter cell type having to tackle both target cell types (PA + and PA –), effectively reducing the effector-to- target (E/T) ratio. In our experiments, NOT-gated CAR-NK cells in the peripheral blood were low in frequency after several weeks, at which time the proportion of model healthy cells was >90% ( Figure 6 C) and comparable in abundance to our PBS treatment control ( Figure 6 D). These long-term data suggest that the kinetics of the system would preclude a collapse of the healthy cell population; instead, the margin of protected healthy cells widens over time. Despite the artificial overexpression of EMCN in the model target cells, our data show that the level of protection is 10 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS comparable to primary HSCs with endogenous expression of EMCN, making it an informative model. The functional impact of comparatively ‘‘high’’ or ‘‘low’’ EMCN expression levels can only be understood in the context of the lineage and phenotype of the target cell. Our model cells are derived from tumor cells, which NK cells naturally tend to kill, whereas HSCs are primary stem cells that NK cells naturally tend to avoid. It follows that an artificially higher level of EMCN expression on tumor cells would be required to elicit a level of protection similar to that in endogenously EMCN+ HSCs. An alternative mechanism for mitigating CAR toxicity is to kill the therapeutic cells through drug- 32 or antigen-inducible 60 expression of apoptotic effectors. Either strategy is undesirable with NK cells, due to challenges maintainingin vivo persistence.61 The antigen-mediated kill switch is especially problematic if the death antigen is expressed in tissues that NK cells frequently visit after administration, such as the liver or lungs, in which case ther- apeutic cells may die altogether before reaching their true target. This strategy may further suffer if the timescale of CAR activation is faster than the kill switch, as is the case with transcription- based circuit components. 60 Because it is challenging to exter- nally control cell behavior specifically in the vicinity of tumor cells, the drug-regulated version is more relevant to systemic toxicity, but, since NK cells, unlike T cells, rarely cause cytokine release syndrome or neurotoxicity, this approach may be of limited appli- cability. Compared to either approach, a key advantage of our iCAR-based NOT gate is the ITIM mechanism of action to directly inhibit the activating CAR through phosphatase intermediates rather than to compromise cell viability. EMCN is broadly expressed on endothelial cells in addition to hematopoietic cells, meaning that the EMCN iCAR will likely be frequently engaged in vivo in different contexts. Unlike T cells, NK cells are part of innate immunity and use a ‘‘missing self’’ sys- tem for choosing targets. They express a large array of different inhibitory receptors that are continuously probing diverse li- gands on various types of normal cells. Although they recognize a broad distribution of ubiquitous ligands (such as human leuko- cyte antigens, collagen, sialic acid, and lectins), many of these receptors share very similar ITIM-containing intracellular do- mains to LIR1 (e.g., KIRs, SIRPa, LAIR-1, SIGLECs, CEACAM- 1, and other LIRs), meaning that NK cells normally receive ongoing signaling through these pathways (i.e., SHP-1, SHP-2, and SHIP-1). Indeed, it is central to NK cell maturation and func- tion to be continually inhibited by healthy cells throughout all of the tissues that they traverse. Therefore, frequent stimulation of the LIR1 iCAR described here would not be expected to cause any abnormal outcomes in NK biology. There are other logic gating strategies that can theoretically be employed to improve safety of CAR products. AND-gated CARs are theoretically only activated when two ligands are simulta- neously detected on a target cell. One approach involves co- expression of two weak CARs recognizing different ligands: each engaged alone potentiates negligible activation, but ligating both creates additive or synergistic signaling that drives an above-threshold cytotoxic response. 46,47 Alternatively, a transcriptional switch responding to one antigen can be used to drive expression of a CAR recognizing a second, in which case killing only occurs when both antigens are present (although not simultaneously; see above regarding problems with slow kinetics being less amenable to discrimination). 59 In either case, the AND gate strategy is potentially more vulnerable to antigen escape, whereby cancer cells lose or downregulate expression of a TAA. 61 The two-antigen requirement of the AND gate gives cancer twice the opportunity to evade detection. NOT gates, on the other hand, rely on consistent PA expression on normal cells, which could potentially remain more stable than TAA expression on cancer cells, which is known to fluctuate, both over time and across tumor cells, especially under thera- peutically exerted negative selection. Stability of PA expression on normal cells could hypothetically arise either from compara- tively higher genetic stability of normal vs. cancer cells, from positive selection by effector cells, or from both. Having demonstrated how this technology can be success- fully applied within a liquid AML tumor setting, we are developing this circuit into a clinical candidate, SENTI-202, for patients with hematologic malignancies, including AML (clinical trial ID NCT06325748). Additionally, applying the OR-NOT gate circuit toward the treatment of solid tumors holds equally transforma- tive potential. For example, anti-CEACAM5 engineered T cell ap- proaches have demonstrated therapeutic potential in the clinic for the treatment of colorectal cancer; however, trials were halted due to off-tumor toxicity against the gastrointestinal tract62 and possibly lung63 tissues. We believe that such toxicity could be significantly reduced through the application of the NOT gate technology described in this work, potentially increasing the therapeutic window. Additionally, this circuit uti- lizes a form of IL-15, which could potentially be beneficial in the context of solid tumors by increasing the persistence and anti-tumor activity of NK (and other immune) cells within the immunosuppressive tumor microenvironment. Limitations of the study There remain avenues to further refine and understand NOT gate gene circuits in NK cells. Although we screened for high-per- forming, naturally occurring iCAR ICDs, we have not shown here any optimization of the ‘‘structural’’ domains of the iCAR, namely the hinge or transmembrane regions. Similarly, we focused solely on the 28z ICD for the mono- and OR-gated aCAR and did not elucidate here the effect of LIR1 iCARs on different aCAR co-stimulatory domains and their respective dif- ferential responses. Optimizing these parameters could be use- ful for refining both the function of these circuits and our under- standing of them. To further characterize the circuit in vivo, other animal models could also be employed. For example, NK cell injection could be delayed, allowing tumor cells to engraft further, which could make the sequence of events more closely resemble the clinical scenario of treating a patient. Although the extent of NOT gate function is likely to be similar, we anticipate that such a model may test the killing efficacy of the gene circuit more stringently, analogous to lowering the E/T ratio in an in vitro experiment. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: Cell Reports 43, 114145, May 28, 2024 11 Article ll OPEN ACCESS d KEY RESOURCES TABLE d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS B NK cell engineering B Cell lines B Primary cells B In vivo models d METHOD DETAILS B Bioinformatic pipelines B NK cell engineering B In vitro cytotoxicity assay setup B In vivo studies d QUANTIFICATION AND STATISTICAL ANALYSIS B Visualization and statistics B In vitro cytotoxicity quantification from flow cytometry-based cell counting SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j. celrep.2024.114145. ACKNOWLEDGMENTS The authors would like to thank Wes Gorman and his team for their viral pro- duction support, Carmina Blanco for activating primary human NK cells, Tiffany Pan for additional characterization of EMCN expression, and the staff at Senti Biosciences for their support of the research and helpful review of the manuscript. This project has been funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health, Depart- ment of Health and Human Services, under contract no. 75N91021C00026. AUTHOR CONTRIBUTIONS N.W.F.: conceptualization, formal analysis, supervision, visualization, writing – original draft, writing – review & editing, methodology, and project administra- tion. H.D.: conceptualization, methodology, investigation, supervision, formal analysis, visualization, and project administration. G.Y.: investigation, supervi- sion, formal analysis, and visualization. M.G.: investigation, formal analysis, and visualization. N.L.: investigation and visualization. A.L.: investigation and visualization. Y.L.: investigation and visualization. M.H.: conceptualization. D.L.: methodology. C.-T.L: investigation, supervision, and methodology. A.B.: investigation, formal analysis, and visualization. M.T.: investigation and methodology. N.A.: investigation. L.N.: investigation. A.R.: software. S.L.: conceptualization. W.W.: conceptualization. R.G.: supervision. T.K.L.: conceptualization and resources. B.G.: conceptualization, supervision, and project administration. DECLARATION OF INTERESTS N.W.F., H.D., G.Y., M.G., N.L., A.L., Y.L., M.H., D.L., C.-T.L., A.B., M.T., N.A., L.N., A.R., R.G., B.G., and T.K.L. are either current or former employees of Senti Biosciences, Inc., of which T.K.L. is also CEO and co-founder. Received: August 21, 2023 Revised: March 25, 2024 Accepted: April 9, 2024 REFERENCES 1. Marofi, F., Al-Awad, A.S., Sulaiman Rahman, H., Markov, A., Abdelbasset, W.K., Ivanovna Enina, Y., Mahmoodi, M., Hassanzadeh, A., Yazdanifar, M., Stanley Chartrand, M., and Jarahian, M. (2021). CAR-NK Cell: A New Paradigm in Tumor Immunotherapy. Front. Oncol. 11, 673276. 2. Sun, S., Hao, H., Yang, G., Zhang, Y., and Fu, Y. (2018). Immunotherapy with CAR-Modified T Cells: Toxicities and Overcoming Strategies. J. Immunol. Res. 2018, 2386187. https://doi.org/10.1155/2018/2386187. 3. De Kouchkovsky, I., and Abdul-Hay, M. (2016). Acute myeloid leukemia: a comprehensive review and 2016 update. Blood Cancer J. 6, e441. https:// doi.org/10.1038/bcj.2016.50. 4. Gocek, E., and Marcinkowska, E. (2011). Differentiation therapy of acute myeloid leukemia. Cancers 3, 2402–2420. https://doi.org/10.3390/ cancers3022402. 5. Slats, A.M., Egeler, R.M., van der Does-van den Berg, A., Korbijn, C., Ha¨ h- len, K., Kamps, W.A., Veerman, A.J.P., and Zwaan, C.M. (2005). Causes of death – other than progressive leukemia – in childhood acute lympho- blastic (ALL) and myeloid leukemia (AML): the Dutch Childhood Oncology Group experience. Leukemia 19, 537–544. https://doi.org/10.1038/sj.leu. 2403665. 6. (2022). Acute Myeloid Leukemia Treatment (PDQ /C226 )–Health Professional Version - NCI. https://www.cancer.gov/types/leukemia/hp/adult-aml- treatment-pdq. 7. Yanada, M., Garcia-Manero, G., Borthakur, G., Ravandi, F., Kantarjian, H., and Estey, E. (2007). Potential cure of acute myeloid leukemia : analysis of 1069 consecutive patients in first complete remission. Cancer 110, 2756– 2760. https://doi.org/10.1002/cncr.23112. 8. Mangan, J.K., and Luger, S.M. (2011). Salvage Therapy for Relapsed or Refractory Acute Myeloid Leukemia. Ther. Adv. Hematol. 2, 73–82. https://doi.org/10.1177/2040620711402533. 9. Sumide, K., Matsuoka, Y., Kawamura, H., Nakatsuka, R., Fujioka, T., Asano, H., Takihara, Y., and Sonoda, Y. (2018). A revised road map for the commitment of human cord blood CD34-negative hematopoietic stem cells. Nat. Commun. 9, 2202. https://doi.org/10.1038/s41467-018- 04441-z. 10. Molica, M., Perrone, S., Mazzone, C., Niscola, P., Cesini, L., Abruzzese, E., and de Fabritiis, P. (2021). CD33 Expression and Gentuzumab Ozogamicin in Acute Myeloid Leukemia: Two Sides of the Same Coin. Cancers 13, 3214. https://doi.org/10.3390/cancers13133214. 11. Valent, P., Bauer, K., Sadovnik, I., Smiljkovic, D., Ivanov, D., Herrmann, H., Filik, Y., Eisenwort, G., Sperr, W.R., and Rabitsch, W. (2020). Cell-based and antibody-mediated immunotherapies directed against leukemic stem cells in acute myeloid leukemia: Perspectives and open issues. Stem Cells Transl. Med. 9, 1331–1343. https://doi.org/10.1002/sctm. 20-0147. 12. Bonardi, F., Fusetti, F., Deelen, P., van Gosliga, D., Vellenga, E., and Schuringa, J.J. (2013). A Proteomics and Transcriptomics Approach to Identify Leukemic Stem Cell (LSC) Markers. Mol. Cell. Proteomics 12, 626–637. https://doi.org/10.1074/mcp.M112.021931. 13. Sachs, K., Sarver, A.L., Noble-Orcutt, K.E., LaRue, R.S., Antony, M.L., Chang, D., Lee, Y., Navis, C.M., Hillesheim, A.L., Nykaza, I.R., et al. (2020). Single-Cell Gene Expression Analyses Reveal Distinct Self- Renewing and Proliferating Subsets in the Leukemia Stem Cell Compart- ment in Acute Myeloid Leukemia. Cancer Res. 80, 458–470. https://doi. org/10.1158/0008-5472.CAN-18-2932. 14. Ng, S.W.K., Mitchell, A., Kennedy, J.A., Chen, W.C., McLeod, J., Ibrahi- mova, N., Arruda, A., Popescu, A., Gupta, V., Schimmer, A.D., et al. (2016). A 17-gene stemness score for rapid determination of risk in acute leukaemia. Nature 540, 433–437. https://doi.org/10.1038/nature20598. 15. Sitnicka, E., Buza-Vidas, N., Larsson, S., Nygren, J.M., Liuba, K., and Ja- cobsen, S.E.W. (2003). Human CD34+ hematopoietic stem cells capable of multilineage engrafting NOD/SCID mice express flt3: distinct flt3 and c-kit expression and response patterns on mouse and candidate human hematopoietic stem cells. Blood 102, 881–886. https://doi.org/10.1182/ blood-2002-06-1694. 12 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS 16. Kikushige, Y., Yoshimoto, G., Miyamoto, T., Iino, T., Mori, Y., Iwasaki, H., Niiro, H., Takenaka, K., Nagafuji, K., Harada, M., et al. (2008). Human Flt3 Is Expressed at the Hematopoietic Stem Cell and the Granulocyte/ Macrophage Progenitor Stages to Maintain Cell Survival. J. Immunol. 180, 7358–7367. https://doi.org/10.4049/jimmunol.180.11.7358. 17. Sommer, C., Cheng, H.-Y., Nguyen, D., Dettling, D., Yeung, Y.A., Sutton, J., Hamze, M., Valton, J., Smith, J., Djuretic, I., et al. (2020). Allogeneic FLT3 CAR T Cells with an Off-Switch Exhibit Potent Activity against AML and Can Be Depleted to Expedite Bone Marrow Recovery. Mol. Ther. 28, 2237–2251. https://doi.org/10.1016/j.ymthe.2020.06.022. 18. Karbowski, C., Goldstein, R., Frank, B., Kim, K., Li, C.-M., Homann, O., Hensley, K., Brooks, B., Wang, X., Yan, Q., et al. (2020). Nonclinical Safety Assessment of AMG 553, an Investigational Chimeric Antigen Receptor T-Cell Therapy for the Treatment of Acute Myeloid Leukemia. Toxicol. Sci. 177, 94–107. https://doi.org/10.1093/toxsci/kfaa098. 19. Yeung, Y.A., Krishnamoorthy, V., Dettling, D., Sommer, C., Poulsen, K., Ni, I., Pham, A., Chen, W., Liao-Chan, S., Lindquist, K., et al. (2020). An Opti- mized Full-Length FLT3/CD3 Bispecific Antibody Demonstrates Potent Anti-leukemia Activity and Reversible Hematological Toxicity. Mol. Ther. 28, 889–900. https://doi.org/10.1016/j.ymthe.2019.12.014. 20. Jetani, H., Garcia-Cadenas, I., Nerreter, T., Thomas, S., Rydzek, J., Mei- jide, J.B., Bonig, H., Herr, W., Sierra, J., Einsele, H., and Hudecek, M. (2018). CAR T-cells targeting FLT3 have potent activity against FLT3/C0ITD+ AML and act synergistically with the FLT3-inhibitor crenolanib. Leukemia 32, 1168–1179. https://doi.org/10.1038/s41375-018-0009-0. 21. June, C.H., O’Connor, R.S., Kawalekar, O.U., Ghassemi, S., and Milone, M.C. (2018). CAR T cell immunotherapy for human cancer. Science 359, 1361–1365. https://doi.org/10.1126/science.aar6711. 22. Pratap, S., and Zhao, Z.J. (2020). Finding new lanes: Chimeric antigen re- ceptor (CAR) T -cells for myeloid leukemia. Cancer Rep. 3, e1222. https:// doi.org/10.1002/cnr2.1222. 23. Dannenfelser, R., Allen, G.M., VanderSluis, B., Koegel, A.K., Levinson, S., Stark, S.R., Yao, V., Tadych, A., Troyanskaya, O.G., and Lim, W.A. (2020). Discriminatory Power of Combinatorial Antigen Recognition in Cancer T Cell Therapies. Cell Syst. 11, 215–228.e5. https://doi.org/10.1016/j.cels. 2020.08.002. 24. Zah, E., Nam, E., Bhuvan, V., Tran, U., Ji, B.Y., Gosliner, S.B., Wang, X., Brown, C.E., and Chen, Y.Y. (2020). Systematically optimized BCMA/ CS1 bispecific CAR-T cells robustly control heterogeneous multiple myeloma. Nat. Commun. 11, 2283. https://doi.org/10.1038/s41467-020- 16160-5. 25. Qin, H., Ramakrishna, S., Nguyen, S., Fountaine, T.J., Ponduri, A., Stetler- Stevenson, M., Yuan, C.M., Haso, W., Shern, J.F., Shah, N.N., and Fry, T.J. (2018). Preclinical Development of Bivalent Chimeric Antigen Recep- tors Targeting Both CD19 and CD22. Mol. Ther. Oncolytics 11, 127–137. https://doi.org/10.1016/j.omto.2018.10.006. 26. Reckzeh, K., Kizilkaya, H., Helbo, A.S., Alrich, M.E., Deslauriers, A.G., Grover, A., Rapin, N., Asmar, F., Grønbæk, K., Porse, B., et al. (2018). Hu- man adult HSCs can be discriminated from lineage-committed HPCs by the expression of endomucin. Blood Adv. 2, 1628–1632. https://doi.org/ 10.1182/bloodadvances.2018015743. 27. Billadeau, D.D., and Leibson, P.J. (2002). ITAMs versus ITIMs: striking a balance during cell regulation. J. Clin. Invest. 109, 161–168. https://doi. org/10.1172/JCI14843. 28. Fathi, E., Farahzadi, R., Sheervalilou, R., Sanaat, Z., and Vietor, I. (2020). A general view of CD33 + leukemic stem cells and CAR-T cells as interesting targets in acute myeloblatsic leukemia therapy. Blood Res. 55, 10–16. https://doi.org/10.5045/br.2020.55.1.10. 29. Pollyea, D.A., and Jordan, C.T. (2017). Therapeutic targeting of acute myeloid leukemia stem cells. Blood 129, 1627–1635. https://doi.org/10. 1182/blood-2016-10-696039. 30. Liu, E., Tong, Y., Dotti, G., Shaim, H., Savoldo, B., Mukherjee, M., Orange, J., Wan, X., Lu, X., Reynolds, A., et al. (2018). Cord blood NK cells engi- neered to express IL-15 and a CD19-targeted CAR show long-term persis- tence and potent antitumor activity. Leukemia 32, 520–531. https://doi. org/10.1038/leu.2017.226. 31. Marin, D., Li, Y., Basar, R., Rafei, H., Daher, M., Dou, J., Mohanty, V., Dede, M., Nieto, Y., Uprety, N., et al. (2024). Safety, efficacy and determi- nants of response of allogeneic CD19-specific CAR-NK cells in CD19+ B cell tumors: a phase 1/2 trial. Nat. Med. 30, 772–784. https://doi.org/10. 1038/s41591-023-02785-8. 32. Liu, E., Marin, D., Banerjee, P., Macapinlac, H.A., Thompson, P., Basar, R., Nassif Kerbauy, L., Overman, B., Thall, P., Kaplan, M., et al. (2020). Use of CAR-Transduced Natural Killer Cells in CD19-Positive Lymphoid Tumors. N. Engl. J. Med. 382, 545–553. https://doi.org/10.1056/NEJMoa1910607. 33. Li, L., Mohanty, V., Dou, J., Huang, Y., Banerjee, P.P., Miao, Q., Lohr, J.G., Vijaykumar, T., Frede, J., Knoechel, B., et al. (2023). Loss of metabolic fitness drives tumor resistance after CAR-NK cell therapy and can be over- come by cytokine engineering. Sci. Adv. 9, eadd6997. https://doi.org/10. 1126/sciadv.add6997. 34. Daher, M., Basar, R., Gokdemir, E., Baran, N., Uprety, N., Nunez Cortes, A.K., Mendt, M., Kerbauy, L.N., Banerjee, P.P., Shanley, M., et al. (2021). Targeting a cytokine checkpoint enhances the fitness of armored cord blood CAR-NK cells. Blood 137, 624–636. https://doi.org/10.1182/ blood.2020007748. 35. Li, Y., Basar, R., Wang, G., Liu, E., Moyes, J.S., Li, L., Kerbauy, L.N., Upr- ety, N., Fathi, M., Rezvan, A., et al. (2022). KIR-based inhibitory CARs over- come CAR-NK cell trogocytosis-mediated fratricide and tumor escape. Nat. Med. 28, 2133–2144. https://doi.org/10.1038/s41591-022-02003-x. 36. Imamura, M., Shook, D., Kamiya, T., Shimasaki, N., Chai, S.M.H., Cou- stan-Smith, E., Imai, C., and Campana, D. (2014). Autonomous growth and increased cytotoxicity of natural killer cells expressing membrane- bound interleukin-15. Blood 124, 1081–1088. https://doi.org/10.1182/ blood-2014-02-556837. 37. Konturek-Ciesla, A., and Bryder, D. (2022). Stem Cells, Hematopoiesis and Lineage Tracing: Transplantation-Centric Views and Beyond. Front. Cell Dev. Biol. 10, 903528. https://doi.org/10.3389/fcell.2022.903528. 38. Khaddour, K., Hana, C.K., and Mewawalla, P. (2023). Hematopoietic Stem Cell Transplantation. In StatPearls (StatPearls Publishing). 39. Guye, P., Li, Y., Wroblewska, L., Duportet, X., and Weiss, R. (2013). Rapid, modular and reliable construction of complex mammalian gene circuits. Nucleic Acids Res. 41, e156. https://doi.org/10.1093/nar/gkt605. 40. Hasty, J., McMillen, D., and Collins, J.J. (2002). Engineered gene circuits. Nature 420, 224–230. https://doi.org/10.1038/nature01257. 41. Mansouri, M., and Fussenegger, M. (2022). Therapeutic cell engineering: designing programmable synthetic genetic circuits in mammalian cells. Protein Cell 13, 476–489. https://doi.org/10.1007/s13238-021-00876-1. 42. Choe, J.H., Watchmaker, P.B., Simic, M.S., Gilbert, R.D., Li, A.W., Kras- now, N.A., Downey, K.M., Yu, W., Carrera, D.A., Celli, A., et al. (2021). Syn- Notch-CAR T cells overcome challenges of specificity, heterogeneity, and persistence in treating glioblastoma. Sci. Transl. Med. 13, eabe7378. https://doi.org/10.1126/scitranslmed.abe7378. 43. Lim, W.A., and June, C.H. (2017). The Principles of Engineering Immune Cells to Treat Cancer. Cell 168, 724–740. https://doi.org/10.1016/j.cell. 2017.01.016. 44. Lee, S., Khalil, A.S., and Wong, W.W. (2022). Recent progress of gene cir- cuit designs in immune cell therapies. Cell Syst. 13, 864–873. https://doi. org/10.1016/j.cels.2022.09.006. 45. Srivastava, S., Salter, A.I., Liggitt, D., Yechan-Gunja, S., Sarvothama, M., Cooper, K., Smythe, K.S., Dudakov, J.A., Pierce, R.H., Rader, C., and Rid- dell, S.R. (2019). Logic-gated ROR1 chimeric antigen receptor expression rescues T cell-mediated toxicity to normal tissues and enables selective tumor targeting. Cancer Cell 35, 489–503.e8. https://doi.org/10.1016/j. ccell.2019.02.003. 46. Cho, J.H., Okuma, A., Sofjan, K., Lee, S., Collins, J.J., and Wong, W.W. (2021). Engineering advanced logic and distributed computing in human Cell Reports 43, 114145, May 28, 2024 13 Article ll OPEN ACCESS CAR immune cells. Nat. Commun. 12, 792. https://doi.org/10.1038/ s41467-021-21078-7. 47. Tousley, A.M., Rotiroti, M.C., Labanieh, L., Rysavy, L.W., Rietberg, S.P., de la Serna, E.L., Dalton, G.N., Klysz, D., Weber, E.W., Kim, W.-J., et al. (2022). Coopting T cell proximal signaling molecules enables Boolean logic-gated CAR T cell control. Preprint at bioRxiv. https://doi.org/10. 1101/2022.06.17.496457. 48. Fedorov, V.D., Themeli, M., and Sadelain, M. (2013). PD-1- and CTLA-4- Based Inhibitory Chimeric Antigen Receptors (iCARs) Divert Off-Target Immunotherapy Responses. Sci. Transl. Med. 5, 215ra172. https://doi. org/10.1126/scitranslmed.3006597. 49. Ng, A.H., Nguyen, T.H., Go ´ mez-Schiavon, M., Dods, G., Langan, R.A., Boyken, S.E., Samson, J.A., Waldburger, L.M., Dueber, J.E., Baker, D., and El-Samad, H. (2019). Modular and tunable biological feedback control using a de novo protein switch. Nature 572, 265–269. https://doi.org/10. 1038/s41586-019-1425-7. 50. Giordano-Attianese, G., Gainza, P., Gray-Gaillard, E., Cribioli, E., Shui, S., Kim, S., Kwak, M.-J., Vollers, S., Corria Osorio, A.D.J., Reichenbach, P., et al. (2020). A computationally designed chimeric antigen receptor pro- vides a small-molecule safety switch for T-cell therapy. Nat. Biotechnol. 38, 426–432. https://doi.org/10.1038/s41587-019-0403-9. 51. Wu, C.-Y., Roybal, K.T., Puchner, E.M., Onuffer, J., and Lim, W.A. (2015). Remote control of therapeutic T cells through a small molecule-gated chimeric receptor. Science 350, aab4077. https://doi.org/10.1126/sci- ence.aab4077. 52. Roybal, K.T., Williams, J.Z., Morsut, L., Rupp, L.J., Kolinko, I., Choe, J.H., Walker, W.J., McNally, K.A., and Lim, W.A. (2016). Engineering T Cells with Customized Therapeutic Response Programs Using Synthetic Notch Receptors. Cell 167, 419–432.e16. https://doi.org/10.1016/j.cell.2016. 09.011. 53. Scheller, L., Strittmatter, T., Fuchs, D., Bojar, D., and Fussenegger, M. (2018). Generalized extracellular molecule sensor platform for program- ming cellular behavior. Nat. Chem. Biol. 14, 723–729. https://doi.org/10. 1038/s41589-018-0046-z. 54. Daringer, N.M., Dudek, R.M., Schwarz, K.A., and Leonard, J.N. (2014). Modular Extracellular Sensor Architecture for Engineering Mammalian Cell-based Devices. ACS Synth. Biol. 3, 892–902. https://doi.org/10. 1021/sb400128g. 55. Gordley, R.M., Williams, R.E., Bashor, C.J., Toettcher, J.E., Yan, S., and Lim, W.A. (2016). Engineering dynamical control of cell fate switching us- ing synthetic phospho-regulons. Proc. Natl. Acad. Sci. USA 113, 13528– 13533. https://doi.org/10.1073/pnas.1610973113. 56. Stebbins, C.C., Watzl, C., Billadeau, D.D., Leibson, P.J., Burshtyn, D.N., and Long, E.O. (2003). Vav1 Dephosphorylation by the Tyrosine Phospha- tase SHP-1 as a Mechanism for Inhibition of Cellular Cytotoxicity. Mol. Cell Biol. 23, 6291–6299.https://doi.org/10.1128/MCB.23.17.6291-6299.2003. 57. Hamburger, A.E., DiAndreth, B., Cui, J., Daris, M.E., Munguia, M.L., Desh- mukh, K., Mock, J.-Y., Asuelime, G.E., Lim, E.D., Kreke, M.R., et al. (2020). Engineered T cells directed at tumors with defined allelic loss. Mol. Immu- nol. 128, 298–310. https://doi.org/10.1016/j.molimm.2020.09.012. 58. Tokatlian, T., Asuelime, G.E., Mock, J.-Y., DiAndreth, B., Sharma, S., Toledo Warshaviak, D., Daris, M.E., Bolanos, K., Luna, B.L., Naradikian, M.S., et al. (2022). Mesothelin-specific CAR-T cell therapy that incorpo- rates an HLA-gated safety mechanism selectively kills tumor cells. J. Immunother. Cancer 10, e003826. https://doi.org/10.1136/jitc-2021- 003826. 59. Roybal, K.T., Rupp, L.J., Morsut, L., Walker, W.J., McNally, K.A., Park, J.S., and Lim, W.A. (2016). Precision Tumor Recognition by T Cells With Combinatorial Antigen-Sensing Circuits. Cell 164, 770–779. https://doi. org/10.1016/j.cell.2016.01.011. 60. Williams, J.Z., Allen, G.M., Shah, D., Sterin, I.S., Kim, K.H., Garcia, V.P., Shavey, G.E., Yu, W., Puig-Saus, C., Tsoi, J., et al. (2020). Precise T cell recognition programs designed by transcriptionally linking multiple recep- tors. Science 370, 1099–1104. https://doi.org/10.1126/science.abc6270. 61. Majzner, R.G., and Mackall, C.L. (2018). Tumor Antigen Escape from CAR T-cell Therapy. Cancer Discov. 8, 1219–1226. https://doi.org/10.1158/ 2159-8290.CD-18-0442. 62. Parkhurst, M.R., Yang, J.C., Langan, R.C., Dudley, M.E., Nathan, D.-A.N., Feldman, S.A., Davis, J.L., Morgan, R.A., Merino, M.J., Sherry, R.M., et al. (2011). T Cells Targeting Carcinoembryonic Antigen Can Mediate Regres- sion of Metastatic Colorectal Cancer but Induce Severe Transient Colitis. Mol. Ther. 19, 620–626. https://doi.org/10.1038/mt.2010.272. 63. Thistlethwaite, F.C., Gilham, D.E., Guest, R.D., Rothwell, D.G., Pillai, M., Burt, D.J., Byatte, A.J., Kirillova, N., Valle, J.W., Sharma, S.K., et al. (2017). The clinical efficacy of first-generation carcinoembryonic antigen (CEACAM5)-specific CAR T cells is limited by poor persistence and transient pre-conditioning-dependent respiratory toxicity. Cancer Im- munol. Immunother. 66, 1425–1436. https://doi.org/10.1007/s00262- 017-2034-7. 64. Knorr, K.L.B., Finn, L.E., Smith, B.D., Hess, A.D., Foran, J.M., Karp, J.E., and Kaufmann, S.H. (2017). Assessment of Drug Sensitivity in Hematopoi- etic Stem and Progenitor Cells from Acute Myelogenous Leukemia and Myelodysplastic Syndrome Ex Vivo. Stem Cells Transl. Med. 6, 840–850. https://doi.org/10.5966/sctm.2016-0034. 14 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS STAR+METHODS KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Mouse monoclonal anti-V5 tag Alexa Fluor 647 Life Technologies 451098 Rat monoclonal anti-FLAG Brilliant Violet 421 BioLegend 637321 Mouse monoclonal anti-Myc-tag Alexa Fluor 488 Cell Signaling Technology 2279S Mouse monoclonal Anti-Human IL-15 Biotin BD Biosciences 554713 Mouse monoclonal anti-human CD56 Brilliant Violet 650 BioLegend 362532 Mouse monoclonal anti-human CD34 FITC BioLegend 343504 Mouse anti-human lineage cocktail APC BioLegend 348803 Mouse monoclonal anti-human CD38 BUV395 BD Biosciences 563811 Mouse monoclonal anti-human CD45RA BUV737 BD Biosciences 612846 Mouse monoclonal anti-human CD33 PE BD Biosciences 555450 Mouse monoclonal Mouse monoclonal anti-human CD90 Brilliant Violet 421 BioLegend 405225 Mouse monoclonal Anti-mouse CD45-PerCP-Cy5.5 BioLegend 103132 Mouse monoclonal Anti-human CD45 PE BioLegend 304008 Mouse monoclonal Anti-human CD56 APC BioLegend 318310 Bacterial and virus strains Ready-to-Use Lentiviral Packaging Plasmid Mix Cellecta, Inc. CPCP-K2A Biological samples AML BMMC, Cryo, Blast >20% Discovery Life Sciences 100030.2 CD34+ BMMCs AllCells CD34 + PS Chemicals, peptides, and recombinant proteins Zombie UV BioLegend 423108 SYTOX Red Invitrogen S34859 PE Streptavidin BioLegend 504204 CellTrace Violet Invitrogen C34557 FuGENE Promega E2311 PEI MAX/C226 MW 40,000 PolySciences 24765 D-Luciferin Gold Biotechnology LUCK-5G RBC Lysis Buffer (10X) BioLegend 420301 DNAse I Millipore Sigma 11284932001 2-mercaptoethanol Gibco 21985023 Stem Cell Factor R&D systems 255-SC IL-3 R&D systems 203-IL GM-CSF R&D systems 7954-GM G-CSF R&D systems 214-CS Erythropoietin Millipore Sigma 11120166001 Transferrin Millipore Sigma 10652202001 Critical commercial assays Human ProcartaPlex Mix&Match 6-Plex Immunoassay Invitrogen PPX-06-MXMFYNN (Continued on next page ) Cell Reports 43, 114145, May 28, 2024 15 Article ll OPEN ACCESS RESOURCE AVAILABILITY Lead contact Requests for further information should be directed to the lead contact, Brian Garrison ( [email protected]). Continued REAGENT or RESOURCE SOURCE IDENTIFIER Deposited data Affymetrix Human Genome U133 Plus 2.0 Array datasets (various) Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) GSE13159, GSE15434, GSE17054, GSE24006, GSE28490, GSE28491, GSE42519, GSE49910, GSE63270, GSE6891, GSE93777 Human (NCBITaxon:9606) ‘‘cell surface’’ proteins Gene Ontology Resource (https://geneontology.org) GO:0009986 Human (NCBITaxon:9606) ‘‘membrane’’ proteins Gene Ontology Resource (https://geneontology.org) GO:0016020 Antibody-based annotations of protein localization Human Protein Atlas ( https:// www.proteinatlas.org/) N/A AML bulk RNAseq data https://www.cancer.gov/ccg/research/ genome-sequencing/tcga N/A The gene and gene product information of KLRG1 Uniprot ( https://www.uniprot.org/) Q96E93 The gene and gene product information of BTLA Uniprot ( https://www.uniprot.org/) Q7Z6A9 The gene and gene product information of KIR3DL1 Uniprot (https://www.uniprot.org/) P43629 The gene and gene product information of NKG2A Uniprot ( https://www.uniprot.org/) P26715 The gene and gene product information of SIGLEC-2 Uniprot (https://www.uniprot.org/) P20273 The gene and gene product information of SIGLEC-10 Uniprot (https://www.uniprot.org/) Q96LC7 The gene and gene product information of LIR-2 Uniprot ( https://www.uniprot.org/) Q8N423 The gene and gene product information of LIR-3 Uniprot ( https://www.uniprot.org/) O75022 The gene and gene product information of LAIR1 Uniprot ( https://www.uniprot.org/) Q6GTX8 The gene and gene product information of KIR2DL1 Uniprot (https://www.uniprot.org/) P43626 The gene and gene product information of LIR1 Uniprot ( https://www.uniprot.org/) Q8NHL6 The gene and gene product information of CD33 Uniprot ( https://www.uniprot.org/) P20138 The gene and gene product information of FLT3 Uniprot ( https://www.uniprot.org/) P36888 The gene and gene product information of EMCN Uniprot ( https://www.uniprot.org/) Q9ULC0 Experimental models: Cell lines Human: SEM (ALL, female) DSMZ ACC 546 Human: MOLM13 (AML, male) AddexBio C0003003 Human: MV4-11 (AML/ALL, male) ATCC CRL-9591 Human: Lenti-X 293T (transformed, female) Takara 632180 Human: GP2-293 (transformed, female) Takara 631458 Experimental models: Organisms/strains NOD.Cg-Prkdcscid Il2rgtm1Wjl Tg(IL15)1Sz/SzJ Jackson Laboratories 030890 Recombinant DNA Gene circuit constructs This paper N/A Antigen expression constructs This paper N/A Software and algorithms Prism GraphPad https://www.graphpad.com/ scientific-software/prism/ 16 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS Materials availability Requests for information regarding plasmids and cell lines generated in this study should be directed to the lead contact , Brian Garrison ([email protected]). Some materials may not be available as they are proprietary to Senti Biosciences. Data and code availability d Requests for data generated in this study should be directed to the lead contact. Senti may disclose certain analyzed data at its sole discretion and subject to Senti and the requester being able to enter into a written agreement as required by Senti. d This paper does not report original code. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact, subject to the abovementioned conditions. EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS NK cell engineering Primary NK cells were isolated from PBMCs from healthy donors and frozen in liquid nitrogen. Male and female donors were used interchangeably with no noticeable difference. For individual experiments, single vials of frozen NK cells were thawed and stimulated with irradiated feeder cells (K562 cells engineered to express membrane bound IL-21 and membrane bound IL-15). NK cells were expanded in 6-well plates in NK media (NK MACS media with 5% human AB serum with 10 ng/mL IL-15 and 100 U/mL IL-2) at a cell concentration range of 5e5 to 1e6 cells/mL. After 10 days of expansion, cultures were analyzed by flow cytometry to ensure a lack of residual feeder cells or CD3 + cells. Cell lines SEM and MOLM13 were cultured in RPMI (VWR Life Sciences) with 10% FBS (Seradigm) and 1% Penicillin-Streptomycin (Gibco) and MV4-11 cells were cultured in IMDM (Gibco) with 10% FBS and 1% Penicillin-Streptomycin. All cells were cultured at 37 /C14C and 5% CO2. To create CD33+fluc+ SEM target cells, 100 mL of lentiviral supernatant of a construct encoding constitutive expression of CD33, fLuc, and blasticidin resistance was applied to 0.5e6 SEM cells and incubated for 2–3 days. Transductants were enriched using 4 mg/mL blasticidin for 1 week. To create EMCN-expressing variants, either parental SEM cells or CD33 + SEM cells were similarly transduced with a lentiviral vector constitutively co-expressing EMCN, GFP, and a puromycin resistance cassette as above, and 2 mg/mL puromycin was used for enrichment. Lenti-X 293T or GP2-293 were cultured in DMEM (Gibco) with 1 mM Sodium Pyruvate (Gibco), 10% FBS, 1% Penicillin-Streptomycin. Primary cells Frozen bone marrow samples from AML patients (Discovery Life Sciences, Inc.) were thawed, treated with DNAse I (Millipore Sigma), washed, and cultured overnight in RPMI media supplemented with 10% fetal bovine serum, 0.055 mM 2-mercaptoethanol (Gibco), 3 ng/mL stem cell factor (R&D systems), 20 ng/mL IL-3 (R&D systems), 200 mg/ml Transferrin (Millipore Sigma), 20 ng/mL GM-CSF (R&D systems), 10 ng/mL G-CSF (R&D systems), 3 U/ml erythropoietin (Millipore Sigma) prior to being washed in assay medium, counted, and added to cytotoxicity assays. Male and female donors were used interchangeably with no apparent difference. CD34-enriched bone marrow cells from healthy donors (AllCells, LLC) were thawed and immediately washed in assay media, counted, and added to a cytotoxicity assay. Male and female donors were used interchangeably with no apparent difference. In vivo models Five-week-old female NOD.Cg-Prkdcscid Il2rgtm1Wjl Tg(IL15)1Sz/SzJ mice were purchased from The Jackson Laboratory. On arrival, these mice were socially housed in single-use polycarbonate cages. These cages were housed in individually ventilated Innovive IVC rodent racks. Sterilized ALPHA-dri bedding, sterilized water, and irradiated Teklad global soy protein-free extruded rodent diet were provided in each cage. All animal procedures were performed in strict accordance with the recommendations in the Guide for the Assessment and Accreditation of Laboratory Animal Care International. The protocol was approved by the Institutional Animal Care and Use Committee (IACUC) of Explora Biolabs. Mice were engrafted with cancer cell lines via tail vein injection on day 0. For OR gate studies, either MOLM13 cells (1e5) or SEM cells (5e6) was used. For OR-NOT gate studies, a mixture of 1.5e6 SEM CD33+ and 1.5e6 SEM CD33 +EMCN+ cells was used. At least 6 mice were used per group. METHOD DETAILS Bioinformatic pipelines Affymetrix Human Genome U133 Plus 2.0 Array datasets corresponding to cell types of interest were obtained from GEO ( https:// www.ncbi.nlm.nih.gov/geo/). The following datasets were used: GSE13159, GSE15434, GSE17054, GSE24006, GSE28490, GSE28491, GSE42519, GSE49910, GSE63270, GSE6891, GSE93777. Data was Robust Multichip Average normalized and expres- sion values were extracted. A set of 7,680 membrane and cell surface proteins was derived using GO (using Gene Ontology terms ‘‘cell surface’’ and ‘‘membrane’’) and HPA (Human protein Atlas, antibody-based annotations) from a set of 20,216 protein coding Cell Reports 43, 114145, May 28, 2024 17 Article ll OPEN ACCESS genes (NCBI). Of those, 115 displaying higher expression in hematopoietic stem cells (HSCs) compared to AML were selected using microarray data (1,128 samples and 13,744 probes representing 882 AML associated samples and 246 samples of normal blood cell types after removing outlier and low-correlating data). Statistically significant hits (t test) were ranked by fold change to isolate po- tential NOT gate targets. Of these genes, 15 were confirmed to be expressed at low levels in AML using bulk RNAseq data (TCGA) and one of them (EMCN) was selected after manual curation incorporating literature sources. NK cell engineering Constructs were first designed in silico (SnapGene), after which DNA was synthesized (Genscript) and reprepped as needed (Gen- ewiz). We used pL17d as the lentiviral backbone for smaller constructs and SINvec and retrovec as the gamma retroviral backbones for larger constructs. Constructs used in this publication were designed by combining sequence fragments exactly as shown in the main text figures, except as noted below. Among these fragments, naturally occurring sequences were acquired either from publicly available sequences (Uniprot) or cDNA clones (Genscript), and scFv binder sequences were either sourced from NCI (FLT3, CD33, HER2) or internally generated (EMCN). For iCAR validation and screening (Figure 3), aCARs and iCARs were expressed from separate lentiviral constructs. The iCAR constructs also included a puromycin resistance gene (not diagrammed), which was placed after the receptor gene separated by a 2A ribosomal skip sequence. FLAG (single target) and myc (OR gate) were used as epitope tags on aCARs, and V5 was used as an epitope tag on iCARs. To create multicistronic constructs (OR-NOT gate), receptor chains were linked together with 2A ribosomal skip sequences in one retroviral vector. DNA was transfected into Lenti-X 293T (lentivirus) or GP2-293 (gamma retrovirus) cells using either FuGENE or PEI, respectively, following manufacturer recommendations. Viral supernatant was collected from these cultures, clarified by centrifugation, and concentrated using either Lenti-X concentrator (smaller batches for in vitro experiments) or Amicon spin filters followed by MgCl 2 and benzonase treatment (larger batches for in vivo experiments), according to manufacturer recommendations and standard practices. Twelve-well plates were coated with recombinant human fibronectin fragment (RetroNectin, Cat#T100B) according to manufacturer protocols. NK cells and lentivirus or retrovirus were added to coated plates and centrifuge at 1000g for 2 h at 32 /C14C. After 3 days, media was exchanged for NK media (for iCAR screening, 2 mg/mL puromycin was added). After 4 more days, receptor expression was checked by flow cytometry (for staining reagents, see Table S1 ) and cells were harvested for use in assays. For in vivo experiments, cells were transferred to 6-well G-Rex (Wilson Wolf, CAT#MSPP-80660M) for further expansion post-transduction. In vitro cytotoxicity assay setup At day 7 post-transduction, NK cells were washed twice in assay medium (RPMI [VWR Life Sciences] with 10% FBS [Seradigm]) and plated in 96-well U-bottom plates. Co-cultures of 5e4 target cells and 2.5e4 NK cells in a total volume of 200mL of assay medium were incubated for 16–20 h, after which plates were analyzed on a Beckman CytoFLEX flow cytometer. For cell line experiments, target cells were stained with CellTrace Blue (Invitrogen) according to manufacturer protocols and resuspended in assay medium prior to starting the co-culture. After the co-culture, plates were centrifuged and 100 mL supernatant was set aside for cytokines to be quan- tified using Luminex assays according to manufacturer protocols. Co-cultures were then stained with Sytox Red (Invitrogen) and the number of live target cell events in a fixed volume was recorded. For AML patient bone marrow experiments, CellTrace and Sytox were not used; rather, co-cultures were instead stained with an antibody cocktail: non-NK cells were separated by low CD56 expres- sion, AML blast cells were identified by low CD45 expression and low side scatter, and LSCs were identified by a CD34 +CD38– phenotype ( Table S2 ). Experiments with CD34 + bone marrow were performed similarly, except that IL-15 was added to OR gate control to compensate for the IL-15 expressed by the OR-NOT gate circuit, and a different antibody staining panel was used for im- munophenotyping64 (Table S2; Figure S7). In vivo studies At day 12 post-transduction, NK cells were collected from the 6M G-rex, washed, and counted for in vivo injection. NK cells (3.5e7 for OR gate; 2.5e7 for OR-NOT gate) were injected via tail vein at day 0. For OR gate studies, either Molm13 cells (1e5) or SEM cells (5e6) was used. For OR-NOT gate studies, a mixture of 1.5e6 SEM CD33+ and 1.5e6 SEM CD33+EMCN+ cells was used. Bioluminescence imaging was performed using a Lago Imager (Spectral Instruments) twice weekly. Mice were injected with 200mL of D-luciferin (150 mg/ kg, Gold Biotechnology Inc, LUCK-5G) and imaged 10 min later. For OR-NOT gate experiments, peripheral blood was collected once a week by submental bleeding into EDTA tubes (Vacuette Minicollect 0.5 mL, Greiner, CAT#450474), of which 150–200mL of blood was processed per mouse for flow cytometry analysis using 7 mL red blood cell lysis buffer (Biolegend CAT#420302) for 5 min, followed by quenching with 7 mL FACS buffer (FBS: Avantor 89510-186; EDTA: EMD Millipore 324506; DPBS: Cytiva SH30028.02). This process was repeated and followed with a final FACS buffer wash, followed by antibody staining ( Table S3). QUANTIFICATION AND STATISTICAL ANALYSIS Visualization and statistics Data visualization and statistical analysis was performed using Prism (ver. 9). We used Brown-Forsythe ANOVA for comparisons of multiple groups and Welch’s test for two-group comparisons, with p-values annotated as * p < 0.5, ** p < 0.01, *** p < 0.005, ****p < 0.001. For visualizing summary statistics, we used the arithmetic mean and standard error of the mean. 18 Cell Reports 43, 114145, May 28, 2024 Article ll OPEN ACCESS In vitro cytotoxicity quantification from flow cytometry-based cell counting Killing described as ‘‘aCAR mediated’’ was calculated by subtracting basal killing from total killing. Total killing was defined as the percent reduction of target cell counts with NK cells relative to without. Basal killing was defined as the total killing specifically with ‘‘control NK cells’’ lacking an aCAR. When puromycin selection was used to enrich NOT gate CAR-NK cells, ‘‘control NK cells’’ were transduced with iCAR alone and selected in the same manner. In all other cases, untransduced NK cells served as ‘‘control NK cells.’’ All measurements were performed in technical triplicate. Results were robust to multiple biological replicates ( Figure S2 C) and different NK cell donors ( Figure S8A). Cell Reports 43, 114145, May 28, 2024 19 Article ll OPEN ACCESS

The release as fetched from its publisher. sentibio.com ↗