Current efforts to engineer CAR-M have found that basic CAR design principles from the T cell field hold true for macrophage biology. Traditional CARs are modular transmembrane proteins consisting of an extracellular antigen-recognition domain, a hinge domain, and one or more cytoplasmic signaling domains (1, 66). We have demonstrated that CARs comprising an scFv against broadly representative targets CD19, HER2, and mesothelin, a CD8 hinge and transmembrane domain, and the CD3ζ intracellular domain efficiently redirect macrophages, guiding antigen dependent phagocytosis, cytokine release, and anti-tumor activity (67). Macrophages expressing CARs with CD3ζ, but not with CD3ζ deletions/tyrosine mutations, killed and phagocytosed tumor cells in an antigen-specific manner. Although CD3ζ is canonically used in CARs due to its role in T cell activation, its cytosolic domain bears significant homology with the macrophage-native Fc receptor common gamma chain (FcRγ) that drives ADCP, though with 3 ITAM domains. We confirmed that CAR-M constructed with either the CD3ζ or FcRγ activating domain were functionally similar in phagocytosis assays, conversely complementing earlier findings showing that CD3ζ- and FcRγ-based chimeric receptors were comparably capable of activating T cells (66).

Indeed, the choice of signaling domain is of particular interest when designing CAR-M, and several groups have explored alternative domains. Morrissey et al. designed CAR-M by screening cytoplasmic domains from murine phagocytic receptors including multiple EGF-like-domains protein 10 (Megf10), FcRγ, adhesion G protein-coupled receptor B1 (Bai1) and tyrosine-protein kinase Mer (MerTK) (68). Primary murine macrophages expressing the FcRγ- or Megf10-based CAR exhibited antigen-specific phagocytic capabilities. Niu et al. designed anti-C-C chemokine receptor type 7 (CCR7) CAR-M to target a newly identified LD^hiCCR7^hi immunosuppressive cell population (69). Their design utilized CCL19, the natural ligand of CCR7, as the receptor’s antigen-recognition domain, rather than an scFv. For the intracellular domain, they evaluated activation domains from MerTK, toll-like receptor 2 (TLR2), TLR4, TLR6 and the CAR-T second-generation 4-1BB-CD3ζ. When screened in the RAW264.7 cell line, CAR-M bearing the MerTK activation domain exhibited the greatest tumor cell toxicity. Interestingly, while anti-CCR7 MerTK-based CAR-M performed well in this context, Morrissey et al’s anti-CD19 CAR bearing the same cytosolic domain was unable to bind antigen-functionalized beads, despite expression at the cell surface (68). Such discrepancies hint that optimization and careful functional evaluation is necessary when generating new CAR-M architectures. In a final example, Zhang et al. designed a CAR bearing the activation domain from CD147 (CAR-147), a protein that regulates matrix metalloproteinase (MMP) expression and ECM remodeling (70). Instead of triggering phagocytosis, CAR-147 targeted the tumor ECM by upregulating expression of MMPs upon antigen recognition. While CD147 itself is not macrophage-specific, the utilization of this CAR design allowed for CAR-dependent secretion of MMPs within the tumor. These studies collectively illustrate that the modular CAR template can customize how macrophages respond to target antigens. Future efforts to engineer CAR-M will likely tap into the plethora of sophisticated CAR designs that have been developed for T cells, incorporating tandem activation domains (71), multi-antigen logic gates (72, 73), or drug-sensitive modules (74–76).

Delivering CARs and other transgenes to macrophages can present a challenge for researchers, but recent advances in gene delivery have enabled several viral and non-viral strategies for doing so. Myeloid cells are proficient at detecting and responding to foreign nucleic acids (77), making macrophages and monocytes resistant to genetic manipulation. Bobadilla et al. created novel HIV-1-derived lentiviral particles capable of infecting myeloid cells by leveraging the viral accessory protein Vpx (78). Upon infection, Vpx mediates degradation of SAMHD1, a myeloid-specific HIV-1 restriction factor that inhibits lentiviral transduction by limiting the deoxynucleotide pool and preventing efficient reverse transcription (79). The group demonstrated that modified lentiviral virions containing Vpx can efficiently deliver transgenes to myeloid cells. The Vpx platform can accommodate any pre-existing HIV-based lentiviral vector and thus provides an accessible strategy for modifying myeloid cells (47, 64, 78). Given that macrophages have limited proliferative capacity, we hypothesized that non-integrating, replication deficient adenoviral vectors may allow for efficient and long-term transduction. However, human myeloid cells do not express the Coxackie-adenovirus receptor, which serves as the primary docking site for traditional Ad5 vectors. Monocytes and macrophages highly express CD46, which mediates docking of group B adenoviruses such as Ad35 (80, 81). We thus evaluated the replication-incompetent chimeric adenoviral vector Ad5f35 and demonstrated that Ad5f35 exhibited robust transduction of primary human macrophages and monocytes – with CAR% and viability routinely >80% (67, 82). Ad5f35-transduced macrophages maintained CAR expression for at least 1 month in vitro and at least 62 days in vivo, as measured by co-expression of CAR-P2A-luciferase. Notably, Ad5f35 activated the macrophage inflammasome and provided a beneficial proinflammatory priming signal, which synergized with CAR activity and rendered the CAR-M locked into an M1 phenotype (83). Such results highlight the prospect of leveraging, rather than evading, the inflammatory response that can occur when delivering genetic material.

Several non-viral strategies have also been developed for engineering monocytes and macrophages. The bacterial origin of plasmid DNA can contribute to inflammation and gene silencing. Plasmids devoid of unmethylated cytosine-phospho-guanine (CpG) dinucleotides – a signature of bacterial DNA – were shown to evade detection by TLR9 and exhibit prolonged gene expression in RAW 264.7 macrophages and primary murine BMDMs (49). Other work has optimized the transient delivery of mRNA to monocytes and macrophages, carefully selecting mRNA modifications and transfection reagents to minimize transfection-induced macrophage toxicity or activation (84, 85). Lastly, transposon systems, which enable non-viral integration into the host genome, have been explored in porcine aortic macrophages (86).

Macrophages may be sourced through several production pipelines. While proof-of-concept studies can be performed in model cell lines such as THP1 and Raw 264.7 or with primary/immortalized murine BMDM, clinical translation necessitates a scalable source of primary human cells. For autologous cell therapies, 2-3x10⁹ peripheral blood monocytes can be obtained by leukapheresis (87), and mobilization with filgrastrim or sargramostim further increases the number of available monocytes by approximately threefold (88). Our CAR-M therapy is manufactured over 1 week using filgrastrim-mobilized CD14⁺ monocytes (67). Monocytes are cultured and differentiated in the presence of granulocyte-macrophage colony-stimulating factor (GM-CSF), which is associated with a pro-inflammatory differentiated phenotype (67, 89). Cells are then transduced with CAR-encoding Ad5f35, which further cements a pro-inflammatory phenotype. To further accelerate manufacturing time, a rapid, same day CAR monocyte process has been developed which yields CAR+ CD14+ monocytes with the capacity to differentiate into M1 CAR-M or CAR-expressing dendritic cells (CAR-DC) (82). Macrophages may be attractive as allogeneic cell therapies since there is no risk of graft versus host disease. Immune cells derived from induced pluripotent stem cells (iPSCs) hold potential as a renewable, allogeneic source for CAR-M therapies. Zhang et al. generated iPSC-derived CAR-Macrophages (CAR-iMac) by reprogramming PBMC’s into iPSC’s over the course of several weeks, transducing with CAR-encoding lentivirus, then differentiating into macrophages following a 4-week differentiation process (90). CAR-iMacs were capable of antigen-dependent macrophage functions, such as cytokine secretion and phagocytosis in vitro. However, CAR-iMacs differentiated with the current protocol had a lingering anti-inflammatory phenotype, and efficacy was limited when tested in murine models. Additionally in oncology applications, a significant consideration with iPSC-derived CAR-M is MHC-matching; antigen cross-presentation is likely an important component of CAR-M activity downstream of TAA engagement, thus careful study is required to determine whether CAR-M derived from MHC knockout iPSCs can potentiate a sufficient anti-tumor T cell response. Furthermore, continued optimization of the iPSC-to-macrophage differentiation protocol, method of transduction, method of phenotype control, and GMP scale-up are necessary to translate these early findings into the clinic. Provided the process is appropriately scaled, there is theoretically no limit to the number of macrophages that can be expanded from iPSCs or differentiation intermediates, though current optimized protocols yield 2-6×10⁷ macrophage progenitors per harvest (91). Benchmarking iPSC-derived macrophage phenotype against bona fide macrophages will be critical for advancing this approach to CAR-M production.

CAR-M therapies are able to clear tumor cells in vitro and in preclinical in vivo models. In vitro, human CAR-M exhibit antigen-specific phagocytosis, cytokine/chemokine secretion, and killing of target antigen expressing targets (67). In two immunodeficient NSGS xenograft models, a single dose of anti-HER2 CAR-M reduced tumor burden and prolonged overall survival against HER2+ SKOV3 tumors. Furthermore, IV-administered CAR-M localized to tumors in several xenograft models and persisted in tumor-free mice (primarily within the liver) for at least 62 days, detected by whole-body bioluminescent imaging of CAR-P2A-luciferase. RNA sequencing revealed that Ad5f35 transduction induced a proinflammatory profile resembling that of classically activated M1 macrophages, which resisted polarization by M2-inducing cytokines in vitro. Furthermore, supernatant from CAR-M was sufficient to induce a proinflammatory phenotype in cultured M2 macrophages. These phenotypic results held true in a humanized immune system (HIS) solid tumor xenograft model, where adoptively transferred CAR-M maintained a durable M1 phenotype and induced pro-inflammatory gene expression in host macrophages. In vitro analysis further showed that CAR-M could coordinate an antitumor T cell response by recruiting T cells and cross-presenting antigens from phagocytosed cells. Recently, our group established an immunocompetent, syngeneic CAR-M model and demonstrated that murine CAR-M increased intratumoral T cell infiltration, NK cell infiltration, dendritic cell infiltration/activation, and TIL activation (92). We found that CAR-M locally administered in HER2⁺ tumors simultaneously controlled growth of contralateral HER2^-negative tumors and prevented antigen-negative relapse upon HER2-negative tumor rechallenge, indicating epitope spreading and induction of long-term immune memory. Notably, this work also demonstrated for the first time that CAR-M synergize with PD1 blockade in PD1-monotherapy resistant solid tumor models (92).

Tumor killing by CAR-M was similarly achieved by Niu et al. using CCR7-targeting CAR-M in the RAW264.7 cell line (69). These CAR-M, which exhibited antigen-specific cytotoxicity in vitro, prolonged survival and prevented metastasis to distal tissues in a 4T1 breast cancer model. CAR-M recruited CD3⁺ T cells and decreased PD-L1⁺ cells in the tumor site, confirming that engineered macrophages themselves are not the sole driver of the antitumor response. Adoptive macrophage therapy also increased levels of pro-inflammatory cytokines IL1-β, IL-6, and TNF-α in the serum, indicative of a systemic immune response (69).

CAR-M’s ability to facilitate an immune response was underscored by the CAR-147 technology, which targeted the tumor ECM rather than tumor cells directly (70). Zhang and colleagues hypothesized that degrading the dense tumor ECM would improve immune cell infiltration and thereby trigger antitumor activity. CAR-M engineered with a CD147 cytosolic domain upregulated MMP expression in an antigen-specific manner in vitro, but exhibited no changes in phagocytosis, killing, or cytokine release. In a HER2⁺ 4T1 breast cancer model, CAR-M slowed tumor growth by reducing its collagen content, enhancing the presence of T cells, and increasing IL-12 and IFN-γ signaling. Taken together, these pioneering studies showcase the ability of CAR-M to infiltrate the tumor niche and initiate a broad anti-tumor response by the host immune system.

Co-administration of pharmacological immunotherapies or chemotherapy could further improve CAR-M efficacy. For example, antibody-based immunotherapies rely on macrophage phagocytosis to stimulate an immune response and could be evaluated for augmenting CAR-M efficacy (93, 94). The Fc region of antibodies binds and stimulates macrophage-expressed Fc receptors, leading to ADCP. Antibodies such as trastuzumab and rituximab thus direct macrophages to phagocytose opsonized target cells (95). Antibodies that block phagocytosis-inhibiting signals, such as CD47/SIRPα or the inhibitory Fc receptor FcγRIIB, have enhanced macrophage-mediated immunotherapies (96–98). T cell checkpoint inhibitors blocking PD1 signaling have also been shown to improve macrophage phagocytic capabilities in vivo (99). Given the impact of CAR-M on surrounding immune cells, we therefore hypothesized that CAR-M could synergize with PD1 checkpoint inhibitors. In a syngeneic CT26 model, which resists anti-PD1 monotherapy, we demonstrated that the combination of CAR-M with PD1 blockade indeed additively improved overall survival (92). Chemotherapy or radiation therapy could also synergize with CAR-M by inducing immunogenic cell death (100). The efficacy of combining radiation therapy and engineered macrophages was demonstrated by Bian et al. using SIRPα-KO macrophages (61). Furthermore, it is noteworthy that CAR expression is not mutually exclusive from other engineering manipulations described herein. Therefore, future iterations of CAR-M could likely synergize with gene editing or accessory transgene overexpression.

Clinical studies will be crucial to elucidating the toxicity profile of CAR-M in patients. The FDA-approved anti-CD19 CAR-T products tisagenlecleucel, brexucabtagene autoleucel, and axicabtagene ciloleucel carry black box warnings for CRS and neurotoxicity (101). CRS is driven by significant CAR-T expansion and secretion of pro-inflammatory cytokines for sustained periods of time in the peripheral blood. Given that CAR-M have limited expansion potential and do not persist in peripheral blood, severe CRS is not expected, and indeed was not seen in older studies of M1 polarized non-engineered macrophages (41). Engineered macrophages have been shown to persist in pre-clinical glioblastoma models without associated toxicity, indicating that CAR-M may safely interact with the central nervous system (44, 47). A particular concern that may be more relevant for CAR macrophages than CAR T cells is that the TME could subvert tumor-localized CAR-M into a tumor-supporting phenotype (102). Although preclinical models suggest the opposite – that CAR-M reprogram the TME (67) – correlative studies in patients will be necessary to understand the bidirectional dynamics. At present, the first-in-human CAR-M Phase I clinical trial is underway using Carisma Therapeutic’s lead product CT-0508 for treating HER2 overexpressing solid tumors (NCT04660929). Results from this Phase I trial and others will provide invaluable insights to guide the design of safe and effective CAR-M therapies.

Future therapies using engineered macrophages may extend beyond oncology indications. CAR-T have been shown to target fibrotic cardiac and liver tissues, and CAR-M may be even better suited for acellular pathogenic targets (103, 104). Novel therapies could also leverage macrophage tissue remodeling and anti-inflammatory capabilities, rather than their proinflammatory functions. For example, adoptive transfer of anti-inflammatory macrophages has been shown to reduce fibrotic tissue in liver injury models (105). From remodeling synapses to repairing cardiac tissue, macrophages are ubiquitous in maintaining tissue homeostasis, and their therapeutic application should be compatible with myriad tissue contexts (106, 107). In conclusion, macrophage phenotypic plasticity, when combined with synthetic biology, presents an exciting new platform for therapeutic applications to advance cellular engineering and deliver effective immunotherapies.