In vivo CAR-M therapy: advancing precision delivery and programmable immune remodeling

Abstract

In recent years, chimeric antigen receptor-macrophages (CAR-Ms) have emerged as a pivotal branch of adoptive cell therapy. They are distinguished by their innate tumor infiltration capacity, phagocytic activity, and immunomodulatory functions, attributes that render them particularly promising for the treatment of solid tumors and non-malignant diseases. Nevertheless, the widespread clinical translation of CAR-M therapy has been constrained by the complexity, high cost, and potential genotoxicity inherent to traditional ex vivo manufacturing processes. As a transformative paradigm, in vivo CAR-M therapy bypasses these limitations by directly delivering CAR-encoding sequences to myeloid cells using advanced platforms such as viral vectors, lipid nanoparticles, extracellular vesicles, or biomaterials. This strategy enables the in situ genetic programming and functional remodeling of macrophages, thereby streamlining therapeutic workflows, reducing production costs, and broadening patient accessibility. This review systematically summarizes recent advances in in vivo CAR-M therapy, focusing on its applications across diverse solid tumors and non-malignant conditions. We compare in vivo engineering strategies with conventional adoptive CAR-M approaches from both mechanistic and translational perspectives and critically examine key challenges, including delivery specificity, immunological safety, and controllability of transgene expression. Collectively, in vivo CAR-M represents a paradigm shift from ex vivo manufacturing toward in vivo cellular programming, offering a more accessible and scalable next-generation immunotherapeutic platform for solid tumors and inflammation-related diseases.

Introduction

Adoptive cell therapy (ACT) enhances tumor-specific immune responses through the ex vivo isolation, genetic engineering, and expansion of autologous or donor-derived immune cells, followed by their reinfusion into the patient. In recent years, ACT, particularly chimeric antigen receptor T-cell (CAR-T) therapy, has progressed rapidly and is now widely recognized as the fourth major pillar of cancer treatment, alongside surgery, radiotherapy, and chemotherapy. The adoptive transfer of receptor-engineered T cells has demonstrated substantial clinical efficacy in B-cell leukemia and lymphoma [1]. In particular, several CAR-T cell products targeting CD19 or B-cell maturation antigen (BCMA) have achieved breakthrough outcomes in patients with relapsed or refractory B-cell malignancies and multiple myeloma [2]. The clinical success of CAR-T therapy in hematologic malignancies has provided a strong biological rationale and momentum for extending this strategy to solid tumors and other malignancies [3].

However, the clinical application of ACT in solid tumors has advanced more slowly than that in hematologic malignancies. The dense extracellular matrix and elevated interstitial fluid pressure characteristic of solid tumors significantly hinder CAR-T cell trafficking and penetration into the tumor core [4]. In parallel, the tumor microenvironment (TME) is typically dominated by a profoundly immunosuppressive milieu, enriched with suppressive immune cell subsets, and characterized by the accumulation of cytokines and metabolic byproducts such as IL-6 and IL-10, which collectively impair T-cell effector function [5]. Moreover, the marked heterogeneity of tumor-associated antigen expression in solid tumors facilitates immune evasion by antigen-negative or low-antigen–expressing subclones, thereby limiting the durability of therapeutic responses and contributing to disease relapse. Beyond these efficacy-related barriers, safety concerns intrinsic to CAR-T therapy, together with its complex and costly manufacturing process further restrict the widespread implementation and clinical translation of ACT in solid malignancies.

Compared with T cells, macrophages possess intrinsic chemotactic properties and superior tissue-penetrating capacity, enabling them to infiltrate dense extracellular matrix barriers and retain functional competence within hypoxic and nutrient-deprived TME [6]. Beyond their migratory advantages, macrophages display remarkable functional plasticity, dynamically polarizing between proinflammatory (M1) and immunosuppressive (M2) phenotypes in response to microenvironmental signals. M1 macrophages secrete pro-inflammatory cytokines, including IL-12 and TNF-α, directly phagocytose tumor cells, and enhance antigen presentation, thereby amplifying T cell–mediated anti-tumor immunity. In contrast, M2 macrophages promote angiogenesis, extracellular matrix remodeling, and immune evasion, ultimately facilitating tumor progression [7]. These opposing functional states create an opportunity to therapeutically reprogram macrophages toward a sustained M1-like anti-tumor phenotype through genetic engineering or targeted modulation of key signaling pathways. The TME represents a highly complex cellular ecosystem composed of heterogeneous immune populations, endothelial cells, stromal elements, and tumor-associated macrophages (TAMs), with TAMs often constituting more than 50% of the tumor-infiltrating immune compartment [8]. Their abundance, coupled with their intrinsic plasticity and adaptability, provides both a conceptual and practical rationale for macrophage-based engineering strategies. In the context of rapid advances in synthetic biology and gene-editing technologies, chimeric antigen receptor macrophages (CAR-Ms) have emerged as a promising therapeutic modality. By leveraging their inherent phagocytic machinery and antigen-presenting capacity, CAR-M therapy has demonstrated encouraging activity in preclinical and early-phase clinical studies and is increasingly regarded as a candidate with substantial translational potential.

Conventional adoptive CAR cell therapy remains heavily dependent on complex, highly individualized ex vivo manufacturing processes, which typically involve immune cell isolation, genetic modification, in vitro expansion, and subsequent reinfusion into patients [9]. This multi-step production pipeline often requires several weeks to complete, incurs substantial costs, and is significantly affected by inter-patient variability in immune cell quality. Consequently, treatment initiation may be delayed, patient access restricted, and considerable obstacles encountered in process standardization and large-scale manufacturing. In recent years, in vivo CAR cell therapy has emerged as an alternative strategy aimed at overcoming these logistical and biological limitations. This approach delivers CAR-encoding constructs directly to target immune cells in situ using platforms such as viral vectors, nanoparticles, or lipid-based delivery systems, thereby enabling endogenous immune cells to undergo genetic modification and functional reprogramming within the patient (Fig. 1). Compared with conventional ex vivo paradigms, in vivo CAR therapy offers potential advantages in manufacturing efficiency, scalability, and cost control, while reducing the risks associated with extensive cell manipulation outside the body [10].

Fig. 1.

Fundamental technical pathways of ACT and in vivo CAR-M therapy. Conventional ACT relies on a complex, multi-step ex vivo manufacturing process. This involves leukapheresis to isolate immune cells, genetic engineering to introduce a CAR, extensive in vitro expansion, and subsequent reinfusion into the patient. Although this approach has proven highly effective against hematologic malignancies, its application in solid tumors is significantly limited by challenges such as poor tumor infiltration, an immunosuppressive TME, and antigen heterogeneity. Furthermore, the elaborate production workflow is time-consuming, costly, and difficult to standardize. In contrast, in vivo CAR-M therapy offers a transformative alternative by delivering CAR-encoding constructs directly to patients using viral vectors, lipid nanoparticles, or other delivery systems. This strategy enables in situ reprogramming of endogenous macrophages, including TAMs. The resulting CAR-Ms leverage their innate tumor-homing and phagocytic capabilities to directly eliminate tumor cells. Concurrently, they enhance antigen presentation and promote CTL activation, thereby reshaping the TME from immunosuppressive to immunostimulatory. By circumventing the need for extensive ex vivo manipulation, this approach presents significant advantages in terms of efficiency, scalability, and potential for clinical translation

In the context of rapidly evolving in vivo CAR engineering strategies, the intrinsic biological differences among immune cell types dictate their suitability for in vivo programming and therapeutic potential, giving rise to divergent technical routes, most notably those represented by in vivo CAR-T, Chimeric Antigen Receptor Natural Killer cells (CAR-NK), and CAR-M [11]. Current in vivo CAR-T approaches primarily employ engineered viral vectors or lipid nanoparticles to deliver CAR transgenes, selectively transducing endogenous T cells to generate antigen-specific effector T cells directly within the body. This circumvents conventional ex vivo expansion and quality-control procedures, thereby improving treatment accessibility and manufacturing consistency [12]. In a seminal study, Pfeiffer et al. constructed a CD8-targeted lentiviral system (CD8-LV) by fusing a high-affinity DARPin to the Nipah virus glycoprotein, achieving T cell-specific reprogramming in vivo and demonstrating efficient clearance of systemic tumor burden in a humanized mouse model [13]. Li et al. reported a CD7-targeted “off-the-shelf” CAR-T strategy that showed robust expansion and antitumor activity in patients with T-cell acute lymphoblastic leukemia, indicating that CAR-T cells can sustain potent proliferative and killing functions even in an in vivo setting [14]. However, CAR-T therapy in solid tumors remains severely constrained by the tumor microenvironment, including dense stromal barriers, aberrant interstitial pressure, and T-cell functional exhaustion driven by immunosuppressive factors, in addition to safety concerns such as cytokine release syndrome (CRS), neurotoxicity (ICANS), and immune effector cell-associated hemophagocytic syndrome (IEC-HS) [12, 15].

In contrast, CAR-NK cells, owing to their innate immune origin, exhibit lower levels of pro-inflammatory cytokine release and a comparatively favorable safety profile, suggesting a reduced risk of CRS and neurotoxicity [16]. NK cells exert antitumor effects through natural cytotoxicity and antibody-dependent cellular cytotoxicity (ADCC) [17]. Christodoulou et al. demonstrated that transgenic expression of an anti-CD123-2B4-CD3 CAR together with secretory IL-15 enhanced antitumor efficacy against AML and improved persistence both in vitro and in vivo [18]. Nevertheless, CAR-NK cells are short-lived with limited in vivo persistence; expansion protocols that rely on cytokine stimulation (e.g., IL-12, IL-15, IL-18) are time-intensive, often requiring several weeks to generate therapeutic cell doses [19]. Moreover, like CAR-T cells, CAR-NK therapy has yet to achieve satisfactory efficacy in solid tumors.

Unlike CAR strategies that rely on lymphoid-lineage cells, CAR-M represents a myeloid-cell-based paradigm shift in engineered immunotherapy. Macrophages intrinsically possess high tissue-infiltrating and chemotactic migration capacities, enabling them to efficiently penetrate the tumor core within the dense, high-interstitial-pressure microenvironment of solid tumors [20]. Furthermore, their unique biological functions extend beyond phagocytic clearance of tumor cells to encompass antigen processing and presentation, thereby engaging the adaptive immune response. In contrast to CAR-T and CAR-NK cells, which primarily depend on direct cytotoxicity, CAR-M not only exerts direct tumoricidal activity but also remodels the macrophage niche and the tumor microenvironment, promotes T cell infiltration, and improves responses to other immunotherapies, thus amplifying the overall efficacy of subsequent immunotherapy [21].To date, in vivo CAR-M editing has been investigated in both oncological and non-oncological disease contexts and has demonstrated robust in vivo activity and an acceptable safety profile. Collectively, these advances may facilitate the broader clinical translation and implementation of CAR-M–based therapeutic approaches (Table 1).

Table 1.

Comparison of ex vivo and in vivo CAR-macrophage engineering strategies

Dimension	Ex Vivo CAR-M	In Vivo CAR-M Engineering
Basic Strategy	Isolation of monocytes/macrophages ex vivo, followed by genetic modification and expansion	Direct in vivo delivery of CAR-encoding systems
Cell Source	Autologous PBMC-derived monocytes; iPSC-derived macrophages	Circulating or tissue-resident monocytes/macrophages in vivo
CAR Expression Format	Stable integration or transient expression	Primarily non-integrating, transient, or inducible expression
CAR expression durability	Relatively long (stable genomic integration + ex vivo expansion support)	Short (transient expression with rapid decay)
CAR expression controllability	High (precisely optimized in vitro)	Limited (dependent on in vivo environmental regulation)
Product Consistency	Batch-to-batch variability	Higher potential for consistency
Scalability	Low (patient-specific manufacturing)	High (amenable to large-scale production)
Microenvironment adaptability	Requires post-infusion adaptation and competition with endogenous TAMs	Direct in situ reprogramming of resident macrophages within the TME
Manufacturing Timeline	Long (weeks to months)	Short (no ex vivo preparation; theoretically enables rapid administration)
Production Complexity	High (requires GMP-grade cell culture, expansion, and quality control)	Low to moderate (no ex vivo cell manipulation, but depends on delivery system optimization)
Cost	High (cell culture, viral vectors, GMP production)	Potentially lower
Safety Risks	Risk of insertional mutagenesis from viral vectors	Low integration risk; potential for off-target delivery and immunotoxicity
Key Advantages	Controllable cell phenotype, mature functionality, clinically validated	Rapid, low-cost, scalable, no complex reinfusion procedures
Key Limitations	High cost, long timeline, significant inter-individual variability	Delivery efficiency, targeting specificity, and in vivo control still require optimization

This review provides a comprehensive overview of the evolution of CAR-M therapy, with an emphasis on recent advances in in vivo CAR-M editing for both solid tumors and non-malignant diseases. It further evaluates the potential advantages of in vivo CAR-M strategies over conventional adoptive cell therapies. Additionally, the principal limitations and translational challenges associated with this approach are critically examined to inform future research directions and clinical developments.

Concept and manufacturing pathway of CAR-M

CAR-M therapy is based on genetic engineering of macrophages to express CARs that selectively recognize tumor-associated surface antigens. This engineering strategy directs and augments the intrinsic phagocytic, antigen-presenting, and immunoregulatory capacities of macrophages, thereby enabling targeted elimination of tumor cells. A canonical CAR construct consists of three primary functional domains: an extracellular antigen-recognition domain, a transmembrane domain, and an intracellular signaling domain [4, 22].

The extracellular antigen-recognition domain of a CAR is typically composed of a single-chain variable fragment (scFv) derived from monoclonal antibodies. In this structure, the variable heavy (VH) and variable light (VL) chains are connected via a flexible peptide linker that facilitates specific recognition of the target antigen [1, 23]. scFvs are frequently derived from either murine or human monoclonal antibodies and can achieve stable expression while preserving the antigen-binding specificity of the parent IgG. Importantly, the orientation of VH-linker-VL versus VL-linker-VH can markedly affect the scFv’s three-dimensional conformation, affinity, and specificity, thereby influencing the overall functional performance of the CAR [23].

The hinge region links the antigen-recognition domain to the transmembrane domain, functioning as a structural bridge between the target antigen and the cytoplasmic membrane. This region is typically derived from IgG subclasses (IgG1 and IgG4), IgD, or CD8 molecules, with the IgG1-derived hinge being the most commonly employed [24]. Optimal hinge length and flexibility facilitate improved antigen accessibility for the scFv, thereby enhancing CAR-mediated effector functions [25].

The transmembrane domain (TMD) functions as a critical bridge linking the extracellular antigen-recognition domain of a CAR to its intracellular signaling domain. Through the TMD, antigen-binding signals are efficiently transmitted to co-stimulatory and cytoplasmic activation modules, such as CD3ζ or Fc receptor γ subunit (FcRγ) [26]. Accumulating evidence indicates that the TMD is not merely a structural support element; its sequence characteristics and molecular origin can substantially influence CAR signaling efficiency and activation threshold. For instance, Majzner et al. demonstrated that substituting a CD28-derived hinge/transmembrane domain (HD/TMD) for a CD8-derived HD/TMD significantly lowered the activation threshold of a CD19 CAR without modifying the intracellular signaling domain (ICD), suggesting that the TMD itself can modulate CAR function by altering receptor conformation and promoting signal microcluster formation [27].

The ICD of a CAR constitutes the core module responsible for driving effector cell functions, and its design has undergone continuous evolution. While CARs share a fundamental structural design across immune cell types, their functional output is ultimately dictated by the intrinsic biology of the effector cells. In CAR-M research, first-generation CARs typically incorporate only the CD3ζ signaling domain [28]. CD3ζ is a canonical ITAM-dependent activation module originally derived from the T-cell receptor complex. In CAR-T, triggering downstream signaling through recruitment of the tandem SH2 (tSH2) domain of ZAP70 induces cytotoxic response and cytokine release [29]. However, macrophages lack a typical TCR signaling machinery. Instead, they primarily rely on Syk-dependent signaling pathways, which contain the tSH2 domain, downstream of ITAMs, as well as on innate immune signaling pathways that engage and activate CD3ζ. This effectively drives phagocytosis and inflammation-related signaling cascades [30].

Although this adaptation preserves the functionality of CD3ζ in CAR-M, a variety of natural phagocytic signaling molecules derived from the signaling modules of the innate immune system have been found to better align with the biological characteristics of macrophages and are therefore more suitable for the design of the intracellular domain in CAR-M. Morrissey et al. demonstrated that both the FcRγ and multiple epidermal growth factor-like domains protein 10 (Megf10) can robustly induce antigen-specific phagocytic responses even when detached from their native extracellular receptors, with efficacy comparable to that of CD3ζ [31]. FcRγ acts as a central mediator of antibody-dependent cellular phagocytosis (ADCP), whereas Megf10 primarily facilitates the clearance of apoptotic cells, playing a critical role in tissue homeostasis [32]. In addition to direct phagocytosis, CAR-M engineered with non-canonical signaling molecules can also modulate the tumor microenvironment. For example, Zhang et al. engineered HER2-targeted CAR-CD147 macrophages that upregulated matrix metalloproteinases (MMPs), leading to degradation of the tumor extracellular matrix, enhanced T-cell infiltration, and suppressed tumor growth [33]. In this process, scFv-mediated antigen recognition is transmitted via the transmembrane domain to the intracellular signaling domain, activating downstream pathways and ultimately eliciting effector functions [34].

Similar to CAR-NK, the structural design of the intracellular domain of CAR-M has followed the T-cell-derived co-stimulatory domains used in CAR-T (e.g., CD28 or 4-1BB combined with CD3ζ) [30]. However, unlike the strictly ordered activation program in T cells, macrophages lack a well-defined counterpart co-stimulatory signaling axis. Consequently, the mechanisms by which such domains function in CAR-M remain unclear. Their regulatory effects on phagocytic activity or polarization status may be relatively limited and may even depend, to some extent, on the basal activation signals mediated by CD3ζ [35]. Notably, recent studies have begun exploring co-stimulatory strategies more suited to myeloid cell characteristics. For instance, Ibrahim and colleagues constructed an intracellular domain containing FcRγ and the co-stimulatory molecule CD86, which exhibited stronger M1-polarizing properties. This design more effectively activated cytotoxic T lymphocytes and promoted the secretion of inflammatory cytokines such as IFN-γ and TNF-α, providing new insights into the design of CAR-M based on co-stimulatory domain activity [36].

CAR-T and CAR-NK primarily mediate target cell lysis through the release of perforin and granzymes, whereas CAR-M exerts its main effector mechanism through antigen-dependent phagocytosis, accompanied by inflammatory cytokine secretion and tumor microenvironment remodeling [37]. Based on these differences, the design philosophy of CAR-M is gradually shifting from the simple transplantation of CAR-T structures toward the integration of phagocytic receptor signaling and innate immune activation pathways, aiming to more precisely match the functional characteristics of macrophages. For example, intracellular adaptor elements derived from TLR or CD40 signaling pathways (such as TIR (intracellular Toll/IL-1 receptor) or MyD88) have been employed to enhance inflammatory polarization, survival capacity, and metabolic reprogramming of CAR-M [38]. Lei et al. constructed a CD3ζ-TIR tandem domain containing the TIR domain of Toll-like receptor 4 (TLR4), which activates NF-κB signaling, thereby enhancing antigen-dependent phagocytosis and inducing a stable M1-like phenotype while conferring some resistance to the M2-like state [39]. Furthermore, Wu et al. reported that MYD88-fused CAR-M exhibited favorable antitumor effects in a brain metastasis model, mediating not only antigen-specific phagocytosis but also bystander effects on surrounding antigen-negative tumor cells through TNF production [40]. Collectively, these studies mark a shift in CAR design toward alignment with macrophage-specific properties.

In addition to the optimization of intracellular signaling, the clinical translation of CAR-M also faces key challenges, including manufacturing scalability, product consistency, and safety. The growing demand for enhanced expansion capacity, precise dosage control, and product consistency in CAR-M therapy has highlighted the substantial potential of induced pluripotent stem cells (iPSCs) [41]. As a renewable and scalable source of human macrophages, iPSC-derived CAR-M cells exhibit superior functional stability and phenotypic homogeneity, positioning them as ideal candidates for standardized clinical-grade production [42]. Zhang et al. engineered induced pluripotent stem cell-derived macrophages (iMACs) from healthy donor PBMCs to generate a second-generation CAR-iMAC incorporating the intracellular TIR domain of TLR4. This design conferred both orthogonal phagocytic capacity and pro-inflammatory polarization capability, resulting in significantly enhanced anti-solid tumor efficacy compared to first-generation CAR-M counterparts [43]. However, the prolonged production timeline of iPSC-CAR-M (typically 4–6 weeks) and its heavy dependence on a GMP-compliant differentiation system remain critical hurdles for clinical translation.

Clinically, multiple CAR-M clinical trials have entered the early clinical validation phase, with the majority being Phase I trials (Table 2). CT-0508, developed by Klichinsky et al., is a pioneering autologous, monocyte-derived, anti-HER2 CAR-M product. Its construct comprises a HER2-specific scFv linked to intracellular signaling domains, including CD28 and CD3ζ. Gene transfer was performed using an Ad5f35 adenoviral vector targeting patient peripheral blood CD14⁺ monocytes. CT-0508 has received FDA authorization to proceed to a first-in-human, open-label, multicenter Phase I clinical trial (NCT04660929) [44]. CT-0525 represents a next-generation HER2-directed monocyte programming approach (NCT06254807) and has demonstrated superior antitumor activity and tumor tissue infiltration compared to CT-0508 in preclinical studies. MCY-M11 (NCT03608618) utilizes non-viral MaxCyte electroporation to transiently express a mesothelin-targeted CAR in autologous peripheral blood monocytes. This approach has been evaluated in patients with mesothelin-positive solid tumors, including ovarian cancer and peritoneal malignancies. Furthermore, additional clinical trials based on CAR-M or myeloid cell engineering technologies are currently ongoing. For example, CT-1119 (NCT05007379) employs patient-derived organoid models to assess the phagocytic activity of CAR-M cells against breast cancer with varying HER2 expression levels, while SY001 (NCT06562647) targets MSLN in patients with advanced solid tumors to evaluate safety, tolerability, and preliminary efficacy.

Table 2.

Clinical trials of CAR-M–based therapies

ID	Phase	Product Name	Indication	Target Antigen	Delivery Route	CAR-M Source
NCT04660929	Phase I	CT-0508	HER2+ solid tumors	HER2	Intravenous / Intraperitoneal	Autologous monocytes
NCT06254807	Phase I	CT-0525	HER2+ solid tumors	HER2	Intravenous	Autologous monocytes
NCT05138458	Phase I/II	MT-101	CD5+ relapsed/refractory T-cell lymphoma	CD5	Intravenous	Autologous monocytes, mRNA-engineered CAR expression
NCT06224738	Phase I	Human HER2-targeted CAR-M	HER2+ gastric cancer with peritoneal metastasis	HER2	Intraperitoneal	Autologous monocytes
NCT06562647	Phase I	SY001	Solid tumors	MSLN	Intravenous	Autologous monocytes
NCT03608618	Phase I	MCY-M11	Ovarian cancer / peritoneal mesothelioma	MSLN	Intraperitoneal	Autologous monocytes
NCT05007379	Phase I	CT-1119	Patient-derived organoids from breast cancer	HER2	Intravenous / Intraperitoneal	Autologous monocytes
ChiCTR2400082776 / 2,400,080,078	Phase I	HER2-CAR-M	Relapsed/refractory OC with high HER 2 expression	HER2	Intravenous / Intraperitoneal	Autologous monocytes

Currently, the ex vivo manufacturing of CAR-M cells remains heavily dependent on viral vector-mediated gene transfer, whose inherent propensity for random genomic integration poses the risk of insertional mutagenesis and potentially severe adverse effects [45]. This safety concern, together with high costs and complex production workflows, constitutes a critical bottleneck limiting the widespread clinical adoption of CAR-M therapy. In recent years, the convergence of gene-editing technologies and novel delivery systems has emerged as a vibrant research focus. The intersection of RNA therapeutics and CAR immune cell therapy has opened new avenues for constructing non-integrating, controllably expressed CAR-M platforms [10]. For instance, Xu et al. employed optimized lipid nanoparticles (LNPs) to deliver CAR mRNA and successfully generated functional CAR-M and CAR-T cells ex vivo, which exhibited potent anti-tumor activity in a B-cell lymphoma model [46]. Furthermore, MT-101, an mRNA-engineered CAR-M therapy developed by Myeloid Therapeutics, has received FDA Fast Track designation for the treatment of relapsed or refractory CD5⁺ peripheral T-cell lymphoma and is currently undergoing Phase I/II clinical trials. Although nonviral and mRNA engineering strategies offer certain advantages in terms of safety and expression controllability, they still fundamentally depend on ex vivo cell manipulation and reinfusion procedures, presenting inherent limitations in manufacturing complexity, cost, and scalability [47]. Building on this foundation, strategies centered on in vivo delivery and in situ engineering of myeloid cells have progressively been proposed, demonstrating encouraging prospects in tumors such as glioma and renal cell carcinoma, as well as in non-malignant diseases (e.g., MRSA infection and cardiac fibrosis) (Fig. 2). Indeed, as early as 2006, Biglari et al. established the feasibility of such strategies for targeting CEA-positive tumors both in vitro and in vivo by inducing human monocytes to express a CEA-specific chimeric CD64 receptor, thereby laying the experimental groundwork for the subsequent conceptualization of in vivo CAR-M therapy [48].

Fig. 2.

Manufacturing strategies, structural design, and therapeutic functions of CAR-Ms and emerging in vivo delivery concepts. CAR-M therapy involves genetically engineering macrophages to express CARs that specifically recognize tumor-associated antigens. This engineering enhances the macrophages’ intrinsic phagocytic, antigen-presenting, and immunomodulatory functions. CAR-Ms can be generated from various sources, including autologous monocytes, iPSC-derived macrophages, or through transient expression systems based on nanoparticles or mRNA. A typical CAR construct consists of an extracellular scFv for antigen recognition, a hinge and transmembrane region, and an intracellular signaling domain. This signaling domain may comprise CD3ζ alone or in combination with innate phagocytic or co-stimulatory modules (e.g., FcRγ, Megf10, CD28, 4-1BB), which collectively determine the strength of activation and functional polarization of the CAR-M. Functionally, CAR-Ms mediate direct tumor cell phagocytosis, enhance antigen presentation to T cells, promote CTL activation, and remodel the tumor microenvironment toward a pro-inflammatory, anti-tumor state. Although current CAR-M production predominantly relies on ex vivo genetic modification followed by reinfusion, emerging in vivo delivery strategies aim to introduce CAR constructs directly into myeloid cells within the patient. This approach enables in situ engineering of macrophages and offers potential advantages in manufacturing efficiency, scalability, and clinical translation

Advances in in vivo CAR-M programming for solid tumors

Renal cell carcinoma

Renal cell carcinoma (RCC) is the predominant histological subtype of kidney cancer and is widely recognized as one of the most immunogenic solid tumors [49]. TME in RCC represents a highly dynamic and multifaceted ecosystem characterized by extensive infiltration of immunosuppressive cells and a profoundly immunosuppressive milieu. Among these, TAMs within the TME play a critical role, accompanied by CD8+ T cell exhaustion, which promotes immune infiltration [50]. Conventional chemotherapy and radiotherapy have limited efficacy in RCC [51]. In recent years, immunotherapy for RCC has begun to emerge [52]. IL-2 is a key component of immune system regulation, essential for T cell proliferation and survival as well as for the generation of effector and memory cells. As the first immunotherapeutic agent approved for cancer treatment [53], IL-2 has been clinically applied to metastatic RCC [54] and melanoma [55], but its use is limited by severe systemic toxicities, including capillary leak syndrome and multiorgan dysfunction [56]. Additionally, cellular therapies such as CAR-T and TCR-T face challenges in solid tumors, including insufficient infiltration and limited persistence within the immunosuppressive microenvironment [57, 58]. Jing et al. described an in situ CAR-M engineering strategy that enhanced macrophage anti-tumor activity by engineering a chimeric IL-2 receptor system (CSR) [59]. In this study, LNPs were used to deliver circular RNA (circRNA) encoding an IL-2R–TLR4 chimeric receptor and an anti-carbonic anhydrase IX (CA9) CAR. These LNPs were embedded in an injectable hydrogel, enabling the in situ generation of CAR-Ms expressing a chimeric IL-2 receptor. Upon binding IL-2, the CSR activates the TLR4 signaling pathway, thereby maintaining immune activation while avoiding systemic toxicity. This drives macrophage polarization toward a pro-inflammatory M1 phenotype and enhances the secretion of inflammatory cytokines. Functionally, these engineered CAR-Ms achieved a phagocytosis rate of over 25% against hCA9⁺ Renca cells (an approximately 16.1-fold increase compared to the control group) and significantly extended survival in a mouse model. Furthermore, CAR-Ms sustained an M1 phenotype in vivo, accompanied by a marked increase in CD4⁺ and CD8⁺ T cell infiltration. Overall, this study establishes a critical proof-of-concept for in vivo CAR-M engineering based on a circRNA-LNP platform, which holds significant promise for translational applications.

Hepatocellular carcinoma

Hepatocellular carcinoma (HCC), the most prevalent form of primary liver cancer, is characterized by high aggressiveness and immune evasion [60]. TME is a key driver of HCC progression, enriched with M2-like TAMs, Tregs, and immunosuppressive cytokines, which significantly restrict the infiltration and function of effector T cells [60]. Early-stage HCC lacks specific symptoms, and most patients are diagnosed at an advanced stage, resulting in generally limited efficacy of conventional treatments [61]. In contrast, immunotherapy has emerged as a transformative strategy, leveraging advances in immunobiology to enhance antitumor immune responses. For example, CAR-T cell therapies targeting antigens such as GPC3 have entered clinical evaluation; however, their therapeutic efficacy is constrained by insufficient trafficking and infiltration into solid tumors, susceptibility to TME-induced dysfunction, and safety concerns including cytokine release syndrome [62]. In vivo CAR-M therapy offers a potential alternative. Unlike CAR-T cells, which rely on T cell migration and expansion, macrophages are inherently enriched in HCC, possessing natural tumor-homing capabilities and tissue infiltration advantages [63]. Yang et al. described an in vivo CAR-M engineering strategy leveraging an mRNA–LNP platform. In this study, LNPs formulated with the ionizable lipid PPZ-A10 were used to preferentially co-deliver GPC3-specific CAR mRNA and mRNA encoding Siglec-G with deleted ITIM domains (Siglec-GΔITIMs) to liver macrophages, enabling transient and controllable CAR expression [64]. GPC3, a therapeutic target highly expressed in HCC, has been extensively evaluated in numerous clinical trials [65]. Meanwhile, the “don’t eat me” signal mediated by the CD24-Siglec-G-ITIM axis potently suppresses macrophage phagocytic function [66]. By introducing Siglec-GΔITIMs to relieve phagocytic inhibition, the results showed that 30% of cells in the liver co-expressed the CAR 24 h after engineering, and the phagocytic efficiency was increased by 10-fold, thereby enhancing the anti-tumor immune effect of GPC3-specific CAR macrophages against HCC. Carisma Therapeutics, in collaboration with Moderna, has reported that the intravenous delivery of mRNA-LNPs encoding GPC3-specific CARs can directly induce CAR expression in macrophages in vivo, leading to a marked reduction in tumor burden in mouse models of solid liver tumors. In another preprint study, Zhang et al. leveraged the distinct cellular components of the TME and adopted a dual-targeting, immune-enhancing CAR design strategy to develop an mRNA–LNP delivery platform. This platform achieves an in vitro editing efficiency of over 80% and an in vivo efficiency exceeding 20%. The researchers then used this platform to construct and co-encapsulate GPC3-CAR–SuperIL-2 and FAP-CAR-4TGFβRII [67]. This system acts through the local secretion of a modified IL-2 variant (SuperIL-2), which enhances the activation of CD8⁺ T cells and NK cells while limiting Treg expansion. Concurrently, FAP targeting engages cancer-associated fibroblasts (CAFs), and expression of a TGFβRII decoy receptor sequesters immunosuppressive TGF-β within the TME, thereby disrupting the stromal barrier and restoring immune cell infiltration. Notably, the researchers also observed CAR-M-mediated antigen spreading, which induced endogenous T cell responses against antigen-negative tumor populations, potentially preventing immune evasion by heterogeneous tumor cells. These findings highlight the therapeutic potential of the mRNA–LNP platform for in vivo CAR-M engineering with multi-targeting capabilities in HCC.

Pancreatic cancer

Pancreatic ductal adenocarcinoma (PDAC) is a highly malignant and profoundly immunosuppressive solid tumor. One of its prominent pathological features is extensive stromal fibrosis, with over 90% of the tumor tissue composed of non-neoplastic components, including CAFs, TAMs, and dense extracellular collagen [68]. These components form a compact barrier that surrounds and infiltrates pancreatic cancer cells, impeding drug penetration and immune cell infiltration, thereby leading to therapeutic resistance. Activated CAFs can suppress T cell proliferation and upregulate immune checkpoints (e.g., TIM-3 and PD-1), further compromising immunotherapy. Although several immunotherapeutic strategies for PDAC have emerged, such as angiotensin-related agents and anti-fibrotic small molecules that attenuate fibrosis by inhibiting the TGF-β pathway and CAR-T cells targeting fibroblasts have shown therapeutic potential in cardiac fibrosis, challenges including physical barriers and insufficient infiltration remain in solid tumor fibrosis, limiting overall efficacy [69, 70]. In vivo CAR-Ms, owing to their inherent tissue-infiltrating capacity and potential to remodel the tumor microenvironment, represent a promising approach to overcome the fibrotic barrier in PDAC. For instance, Wang et al. constructed FAP-targeting CAR-Ms, which significantly reduced FAP expression, collagen volume fraction, and type I collagen levels in an orthotopic PDAC model. At an effector-to-target ratio of 1:1, phagocytic efficiency was increased by more than two-fold, thereby eliminating the stromal barrier, enhancing intratumoral penetration of gemcitabine and immune cells, and substantially improving the sensitivity of PDAC to chemotherapy and immunotherapy [71]. By simultaneously targeting tumor cells and the fibrotic stroma, this strategy can effectively remodel the immunosuppressive microenvironment of PDAC, demonstrating unique therapeutic advantages in highly fibrotic solid tumors. Furthermore, in vivo CAR-M can be engineered with multi-modular cooperative designs to simultaneously relieve immunosuppressive signals. Liu et al. engineered an MUC1-specific CAR driven by a macrophage-specific CD68 promoter using plasmid DNA, restricting transcription to macrophages for the recognition of pancreatic cancer cells. They co-delivered the hMUC1-CAR plasmid along with a CD47 inhibitor to the tumor site via liposomal nanocarriers, thereby blocking the CD47-SIRPα signaling axis and efficiently generating MUC1-CAR macrophages (MUC1-CAR-M). This approach markedly enhanced specific phagocytosis and antigen presentation again PDAC cells. This nanosystem can be cleaved by legumain, an enzyme overexpressed in pancreatic cancer, enabling tumor-specific release and thereby efficient in situ generation of CAR macrophages. At 24 h post-transduction, approximately 31% of macrophages expressed the CAR, and the phagocytic efficiency exceeded 30%, leading to sustained activation of antitumor responses [72].

Central nervous system tumors

Glioblastoma (GBM) is one of the most aggressive and lethal primary brain tumors, and its treatment has long been limited by significant molecular heterogeneity and a profoundly immunosuppressive TME [73, 74]. Clinically, over 70% of GBM patients undergo surgical resection, yet nearly all patients eventually relapse after surgery [75]. Although immunotherapy has achieved progress in some solid tumors, its overall efficacy in GBM remains limited. T cell-based therapies, such as CAR-T cells, face multiple obstacles in this disease, including insufficient T cell infiltration, intratumoral hypoxia and metabolic stress, and persistent immunosuppressive signals, which compromise their long-term survival and functional maintenance within the tumor [76]. TAMs are highly abundant in GBM, constituting 30%–50% of the tumor mass, and play a critical role in sustaining immunosuppression, promoting tumor progression, and driving therapeutic resistance. In central nervous system (CNS) tumors, the in situ generation of CAR-M has emerged as a novel strategy to overcome the blood–brain barrier and the immunosuppressive tumor microenvironment. Chen et al. utilized nanoparticle–hydrogel superstructures to locally induce CD133-specific CAR-M (CD133-CAR-M) within the resection cavity of GBM, redirecting macrophages to phagocytose glioma stem cells (GSCs), achieving a more than four-fold increase in phagocytic efficiency, and subsequently activated adaptive immune responses while effectively preventing postoperative recurrence. This effect was consistently observed across multiple orthotopic intracranial models [77]. Gao et al. developed a convection-enhanced lipid–polymer hybrid nanoparticle system for the precise delivery of HER2-CAR DNA to brainstem glioma regions. Modification with RP-182 enhanced TAM targeting, and convection-enhanced delivery (CED) enabled deep parenchymal diffusion, inducing the local generation of M1-like CAR-M, enhancing phagocytosis and antigen presentation, and promoting CD8⁺ T cell infiltration. Seven days after intratumoral injection of targeted nanocarriers carrying the ErBb2-CAR plasmid, approximately 8% of macrophages at the tumor site were converted into CAR-M. This treatment significantly suppressed the growth of orthotopic brainstem glioma, and more than half of the mice exhibited extended survival beyond 100 days [78]. Zhou et al. employed enucleated mesenchymal stem cells (MSCs) as carriers to deliver CD133-CAR to the GBM region, achieving in situ generation of sufficient CAR-M, which significantly suppressed tumor growth in models and achieved near-complete tumor clearance when combined with CD47 blockade. This strategy also reinforced M1 polarization and antigen presentation through NF-κB and IFN-γ signaling [79]. Zhang et al. reported an approach utilizing engineered Escherichia coli Nissle for localized tumor colonization and sustained secretion of bispecific engagers simultaneously targeting EGFRvIII and IL-13Rα2. These engagers labeled heterogeneous glioblastoma cells in situ, thereby recruiting and activating non-virally engineered CAR-macrophages and achieving in situ immune reprogramming within the tumor microenvironment [80]. In a neuroblastoma model, Kang et al. employed mannose-modified MPEI to deliver plasmids encoding IFN-γ and an ALK-targeting CAR. This approach successfully reprogrammed M2-type macrophages in the TME into macrophages that co-expressed the CAR and exhibited an M1 phenotype. Importantly, this phenotypic transition persisted even after tumor cell phagocytosis, leading to enhanced phagocytosis and antigen presentation, while simultaneously activating naive T cells and inducing the generation of tumor-specific cytotoxic T lymphocytes (CTLs). In vivo evaluation in murine models demonstrated that, following administration of MPEI/pCAR-IFN-γ, approximately 10.3% of tumor-infiltrating macrophages expressed CAR. Notably, more than half of the treated mice achieved long-term survival exceeding 100 days [81]. Aberrant hypersialylation, frequently observed in GBM, activates immunosuppressive Siglec receptors and drives immune tolerance in TAMs [82]. To exploit this mechanism, Fu et al. engineered a Siglec-9-based CSR in which the native inhibitory intracellular domain was replaced with an activating signaling module. Upon recognition of the abundantly sialylated ligands on the tumor surface, the CSR reprograms macrophages to convert an otherwise suppressive signal into a pro-inflammatory activating signal. These CSR-engineered macrophages, designated CAR-M, displayed markedly enhanced phagocytic capacity, inflammatory cytokine secretion, and antigen-presenting function, effectively reversing the immunosuppressive GBM microenvironment and amplifying antitumor efficacy. In murine models, the engineered CAR-M significantly prolonged survival (beyond 100 days) [83].

A key advantage of in vivo CAR-M therapy for CNS tumors lies in its amenability to local delivery and its ability to exploit the immune-privileged intracranial niche, while simultaneously promoting secondary adaptive immune activation beyond direct cytotoxicity. Together, these findings underscore the unique therapeutic potential of CAR-M in highly immunosuppressive brain tumors and provide a novel immunotherapeutic strategy for solid CNS malignancies.

Lung metastatic tumors

Metastasis accounts for the majority of cancer-related deaths [84]. It occurs when cancer cells detach from primary tumors, disseminate through the bloodstream, lymphatic system, or other routes, and colonize distant tissues while adapting to new microenvironments [85]. The lungs are a common site of metastasis for various solid tumors, including breast cancer, melanoma, and hepatocellular carcinoma. A central challenge in managing metastatic disease is achieving efficient, organ-selective delivery while minimizing systemic clearance caused by hepatic and splenic sequestration. Although liposomes are widely employed for in vivo RNA delivery, their utility is hampered by off-target effects, potential immunogenicity, and limited in vivo stability. In contrast, engineered small extracellular vesicles (sEVs) derived from natural cells offer a promising alternative as native mRNA carriers. As early as 2007, Valadi et al. demonstrated that sEVs can effectively carry mRNA [86]. To enhance mRNA loading efficiency, Kojima et al. developed a strategy to achieve selective packaging of specific mRNAs into sEVs during biogenesis by fusing the RNA-binding protein L7Ae to sEV membrane proteins and incorporating C/D box tags into the target mRNAs [87]. Inspired by this approach, Xiao et al. engineered sEVs modified with anti-CD206 single-chain variable fragments for the targeted delivery of MSLN-specific CAR mRNA to macrophages within lung metastases. Conventional intravenous administration restricts the in vivo distribution of sEVs, resulting in predominant hepatic accumulation and limited enrichment in extrahepatic tissues such as the lung [88]. In contrast, inhaled delivery substantially enhances the pulmonary delivery and utilization efficiency of sEVs [89]. Employing this inhalation-based delivery system, in situ CAR-M generation can be detected in the lung within 24 h, markedly boosting local engineering efficiency. In a mouse model, over 28% of tumor-associated macrophages expressed the CAR, and stable fluorescence signals persisted for at least 48 h after administration. Furthermore, this strategy effectively circumvents the nonspecific hepatic and splenic homing that typically limits ex vivo-generated CAR-M, enabling precise lung-targeted macrophage reprogramming [90]. Although this strategy has shown preliminary success in murine models, its translation to large mammals faces challenges, including interspecies differences in respiratory physiology that affect sEV distribution and pharmacokinetics, as well as the need for comprehensive long-term safety assessments.

Melanoma, colorectal cancer, and other solid tumors

Park et al. designed a phosphatidylserine (PS)-containing LNP system to deliver anti-GP75 CAR mRNA in a mouse melanoma model. PS, a membrane lipid that becomes externalized on apoptotic cells, is recognized by phagocytes such as macrophages, thereby triggering endocytic uptake. This LNP system also co-loaded a STING agonist, reprogramming TAMs toward a pro-inflammatory M1 phenotype while depleting immunosuppressive M2-like macrophages and increasing CD8⁺ T cell infiltration, resulting in synergistic immune activation [91].

In recent years, vaccine-like strategies for in vivo enhancement have been explored to augment the function of CAR immune cells. For instance, in 2023, Ma et al. demonstrated that the administration of a vaccine in vivo could enhance CAR-T cell activity, promote dendritic cell recruitment and antigen presentation within tumors, activate endogenous T cell responses, and facilitate antigen spreading [92]. Qu et al. reported a circRNA-based in vivo panCAR platform that uses lipid nanoparticles to deliver CAR-encoding circRNAs, achieving in vivo CAR expression in macrophages, T cells, and NK cells. In murine models of melanoma and colorectal cancer, this strategy significantly inhibited tumor growth. Furthermore, when combined with an HER2-specific circRNA vaccine (circRNAVAC), the in vivo panCAR strategy enhanced anti-tumor immune responses, while circRNACAR in turn promoted vaccine-induced HER2-specific antibody production, demonstrating potent synergistic anti-tumor effects across multiple immunocompetent mouse models [93]. Gu et al. described an intraperitoneally delivered, macrophage-targeting mRNA-LNP platform that directly induces in vivo CAR-M generation in models of colorectal cancer metastasis and pancreatic cancer. They systematically compared the ICD functionality of 36 different CAR constructs and identified that CARs incorporating both CD3ζ and TLR4 intracellular domains significantly enhanced macrophage inflammatory activation and antigen presentation capabilities [94] (Table 3).

Table 3.

Preclinical studies of in vivo CAR-M engineering in solid tumors

Tumor Type	Target Antigen		In Vivo CAR-M Engineering Strategy	Delivery System	Treatment Method	Key Mechanism of Action	Preclinical Model	Reference
RCC	CA9	circRNA encoding CA9-CAR + chimeric IL-2 receptor		circRNA-LNP + injectable hydrogel	Intratumoral injection	IL-2-dependent activation of TLR4 signaling; induces M1 polarization; pro-inflammatory cytokine release	Orthotopic RCC mouse model	Jing et al. [59]
HCC	GPC3	GPC3-CAR mRNA + Siglec-GΔITIMs mRNA		mRNA-LNP (PPZ-A10 lipid)	Intravenous injection	Disruption of CD24–Siglec-G phagocytic inhibitory signal	Orthotopic HCC mouse model	Yang et al. [64]
	GPC3 + FAP	Dual-target CAR-M (GPC3-CAR–SuperIL-2 + FAP-CAR–4TGFβRII)		Co-encapsulated mRNA-LNP	Intravenous injection	IL-2 immune enhancement; CAF depletion; TGF-β sequestration	Orthotopic HCC mouse model	Zhang et al. [67]
PDAC	MUC1	MUC1-CAR plasmid DNA driven by CD68 promoter		Tumor-responsive liposomal nanocarrier	Intravenous injection	Macrophage-specific transcription; CD47–SIRPα blockade	Orthotopic PDAC mouse model	Liu et al. [72]
	FAP	FAP-targeting CAR-M		Mannose-modified mRNA-LNP	Intravenous injection	CAF depletion; reduces collagen deposition	Orthotopic PDAC mouse model	Wang et al. [71]
GBM	CD133	CAR-encoding plasmid		Enucleated MSC delivery system	Intravenous injection	Macrophage-specific endocytosis; generation of sufficient CAR-Ms	Orthotopic GBM mouse model	Zhou et al. [79]
	EGFRvIII; IL-13Rα2	EGFRvIII + IL-13Rα2 dual-targeting CAR-M		Engineered E. coli Nissle + injectable local delivery vehicle	Local injection	Dual-target CAR-M phagocytosis of heterogeneous tumor cells; TME remodeling	Orthotopic GBM mouse model	Zhang et al. [80]
	CD133	CD133-CAR plasmid		Nanoparticle–hydrogel superstructure	Local injection into post-surgical resection cavity	Targeting glioma stem cells; enhanced phagocytosis	Orthotopic GBM mouse model	Chen et al. [77]
	IL-13Rα2	CAR + SIGLEC9-based CSR circRNA		Injectable hydrogel–embedded LNPs	intratumoral injection	Siglec-9 inhibitory signaling is converted into activation, maintaining an M1 phenotype	Orthotopic GBM mouse model	Fu et al. [83]
Brainstem Glioma	HER2	HER2-CAR plasmid		Lipid–polymer hybrid nanoparticles	Convection-enhanced delivery	Deep brain tissue penetration; M1 polarization	Brainstem glioma model	Gao et al. [78]
Neuroblastoma	ALK	CAR + IFN-γ plasmid		MPEI delivery vector	Intratumoral injection	M1 polarization; enhanced phagocytosis and antitumor activity	Neuroblastoma model	Kang et al. [81]
Lung Metastases	MSLN	MSLN-CAR mRNA		Anti-CD206 modified sEV	Inhalation (nebulized delivery)	Targeted enrichment in pulmonary macrophages	Lung metastasis mouse model	Xiao et al. [90]
Melanoma	GP75	GP75-CAR mRNA + STING agonist		PS-containing LNP	Intratumoral injection	TAM reprogramming; STING activation	Melanoma model	Park et al. [91]
Melanoma / Colorectal Cancer	PanCAR + HER2	circRNA panCAR ± circRNA vaccine		circRNA-LNP	Intravenous injection	CAR expression in multiple immune cell types; vaccine-like amplification effect	Mouse disease model	Wang et al. [93]
Colon carcinoma /PDAC	CD47; HER2	Transient mRNA CAR		mRNA-LNP	Intraperitoneal injection	Enhances phagocytosis; drives CD8⁺ T cell-dependent systemic antitumor immunity	CT26 murine colon carcinoma model; PAN02 murine pancreatic cancer model	Gu et al. [94]

Abbreviations: ALK Anaplastic Lymphoma Kinase, CA9 Carbonic Anhydrase IX, CAR-M Chimeric Antigen Receptor Macrophage, CT26 Colon Carcinoma 26, EGFRvIII Epidermal Growth Factor Receptor Variant III, FAP Fibroblast Activation Protein, GBM Glioblastoma, GP75 Glycoprotein 75, GPC3 Glypican-3, HCC Hepatocellular Carcinoma, HER2 Human Epidermal Growth Factor Receptor 2, IL-13Rα2 Interleukin-13 Receptor Alpha 2, MSLN Mesothelin, MUC1 Mucin 1, PAN02 Pancreatic Adenocarcinoma 02, PDAC Pancreatic Ductal Adenocarcinoma, RCC Renal Cell Carcinoma, TME Tumor Microenvironment

Advances in in vivo CAR-M programming for non-malignant diseases

MRSA

Periprosthetic joint infection (PJI) is one of the most severe complications following artificial joint replacement surgery. In hip arthroplasty, the all-cause mortality rate is 5% at one year and increases to 20% at five years post-operatively [95, 96]. The core pathological mechanism involves bacterial adherence to the prosthetic surface, leading to the formation of a biofilm (extracellular polysaccharide glycocalyx) that shields the bacteria from host immunity, antibiotics, and mechanical debridement. Among causative pathogens, methicillin-resistant Staphylococcus aureus (MRSA) is the most prevalent and destructive [97, 98]. At the infection site, MRSA suppresses macrophage phagocytosis and bactericidal function through immune evasion mechanisms, rendering the internalized bacteria difficult to eliminate and resulting in persistent infection and frequent recurrence. The surface protein SasA, which is highly conserved across invasive MRSA strains, contributes to host immune evasion [99]. Concurrently, MRSA inhibits inflammasome activation by blocking mitochondrial recruitment to intracellular bacterial vacuoles via CASP11, thereby reducing macrophage bactericidal efficiency [100]. Furthermore, the implant itself triggers local tissue reactions that contribute to the formation of an immunosuppressive microenvironment, further suppressing macrophage activity and function, thereby making the infection difficult to eradicate. Together, the biofilm barrier, immunosuppressive microenvironment, and inactivation of macrophage function constitute the core bottlenecks in the treatment of PJI. Unlike other immune cells, macrophages possess unique functional properties and bacterial chemotaxis that render them particularly suitable for bacterial infection scenarios. They exert phagocytic and lysosomal degradation functions at the innate immune level, rather than relying on cytotoxic killing [101]. Therefore, reprogramming macrophages to restore and enhance their antibacterial functions represents a promising therapeutic strategy. Li et al. developed a bone implant nanocoatings technology that enables in situ generation of super CAR-macrophages (SasA-CAR-M) at the local implant interface. This strategy enhances MRSA-specific phagocytosis through CAR-mediated recognition and concurrently downregulates CASP11 via shRNA to restore mitochondrial-associated antimicrobial pathways, thereby synergistically boosting intracellular bacterial killing capacity. As a result, SasA-CAR-M significantly increased the phagocytic capacity of macrophages against MRSA (approximately 5.2-fold) and improved bactericidal efficiency to 83.6%, while reducing the postoperative infection rate by 83% in a mouse model. Furthermore, this strategy promoted osteogenic repair and improved the local immune microenvironment, achieving synergistic effects of anti-infection and tissue regeneration [102].

Sepsis is defined as a life-threatening organ dysfunction resulting from a dysregulated host response to infection [103]. The disease course typically transitions from an early hyperinflammatory response to a late immunosuppressive state. During this process, immune cell exhaustion, dysfunction, and apoptosis are markedly increased. Concurrently, the accumulation of immunosuppressive factors, tissue hypoxia, and nutrient deprivation further disrupt energy metabolism, synergistically weakening the host’s anti-infection capacity [104]. Macrophages in particular, driven by aberrant epigenetic reprogramming, readily enter an immunosuppressive state, leading to a significantly elevated risk of secondary infections and organ failure [105].

Although infection with any pathogen can lead to sepsis, bacteria are the most frequently identified causative agents [106]. Among these, persistent MRSA infection and the ensuing host immunoparalysis are major contributors to sepsis-related mortality [107]. Under systemic immunosuppression, host clearance capacity is insufficient, and pathogen immune evasion leads to persistent infection. Additionally, the lack of coordinated regulation of both infected cells and immune function results in impaired immune competence. Consequently, conventional antibiotics often fail to achieve significant therapeutic efficacy [108]. Achieving efficient and specific systemic clearance of MRSA remains a substantial clinical challenge. In contrast to antibiotics, the in vivo CAR-M strategy does not rely on passive diffusion or bacterial susceptibility. Instead, it eliminates intracellular and immune-evading pathogens through active recognition and phagocytosis, offering a potential therapeutic approach. Tang et al. developed an ionizable lipid nanoparticle platform (CRV/LNP-RNAs) co-encapsulating siCASP11 and SasA-CAR mRNA, functionalized with the macrophage-targeting peptide CRV to enable systemic, macrophage-specific delivery. Engineered SasA-CAR macrophages exhibited a 6.2-fold increase in phagocytosis against MRSA. In a mouse model of sepsis, this treatment significantly improved survival (reaching 80%) and achieved long-term recovery of physiological functions [109]. However, it should be noted that the antibacterial efficacy of this therapy has currently only been validated in a single MRSA mouse model; therefore, evaluation of its efficacy against other Staphylococcus aureus strains and assessment of long-term safety are necessary prior to clinical translation.

Fibrotic diseases

Fibrosis is a pathological process in which tissue repair becomes dysregulated following chronic injury or persistent inflammatory stimulation, leading to extracellular matrix (ECM) deposition, scar formation, and tissue hardening [110]. This process is typically driven by sustained activation of myofibroblasts, accompanied by tissue remodeling and progressive loss of normal organ function [111].

Myocardial infarction (MI) leads to irreversible cardiomyocyte necrosis due to acute or persistent ischemia of the coronary arteries [112]. During subsequent cardiac repair, activated cardiac fibroblasts (CFs) secrete extracellular matrix components such as fibronectin and collagen. Excessive deposition of these components results in myocardial fibrosis, which reduces ventricular wall compliance, impairs cardiac function, and may ultimately progress to heart failure—a leading cause of mortality in post-MI patients [113, 114]. Current treatments (e.g., reperfusion, β-blockers, and RAAS inhibitors) primarily reduce cardiac load and inhibit neuroendocrine activation, yet they lack direct intervention against pathogenic fibroblasts [115]. FAP, a specific marker of activated fibroblasts, has emerged as an important therapeutic target [116]. Although various FAP-targeted strategies are under investigation, the application of cell-based therapies such as CAR-T in cardiovascular diseases remains limited by potential cardiovascular toxicity, and their safety profile warrants further evaluation [116]. Macrophages play a pivotal role in regulating local immune responses and tissue repair after myocardial infarction [117]. Recent studies have demonstrated that uPAR-targeted CAR-M alleviates liver fibrosis and that adoptive transfer of BMDM-CAR-P cells targeting FAP⁺ myofibroblasts mitigates fibrosis [118, 119]. To overcome the limitations associated with complex ex vivo CAR-M preparation and high costs, Liu et al. developed a macrophage-targeting LNP system that delivers Lgmn mRNA to induce in vivo generation of CAR-MΦs with enhanced efferocytosis. This strategy significantly improved phagocytic capacity against FAP-positive target cells, achieving a 7.2-fold increase compared to the mRNA-only group and a 2.1-fold increase compared to the CAR-LNP group, thereby enhancing the specific clearance of activated fibroblasts. Furthermore, in a mouse model, this approach markedly improved survival (approximately 2-fold increase), thus effectively enhancing anti-fibrotic efficacy [120].

In myocardial ischemia–reperfusion injury (MIRI), the restoration of blood flow paradoxically exacerbates oxidative stress, calcium overload, and inflammatory responses, promoting cardiomyocyte death and driving myocardial fibrosis, which are hallmarks of adverse cardiac remodeling [121, 122]. Du et al. developed a macrophage-targeting LNP-FAP CAR system, in which mRNA encoding a FAP-specific CAR was encapsulated into lipid nanoparticles to enable in situ generation of FAP CAR-M. This strategy increased the phagocytic efficiency against FAP-positive fibroblasts by more than five-fold and, in an ischemia-reperfusion model, significantly reduced the number of activated fibroblasts while alleviating myocardial fibrosis. Notably, the LNP-FAP CAR treatment group continued to show improved left ventricular ejection fraction and fractional shortening two months after reperfusion, with no observable toxicity, indicating durable functional benefits [123].

The core pathological feature of liver fibrosis is the abnormal deposition of ECM, a process primarily driven by the sustained activation of hepatic stellate cells (HSCs), ultimately leading to liver failure and progression to cirrhosis [124]. Current clinical interventions are limited, with no approved anti-fibrotic drugs specifically targeting fibrosis; patients with advanced disease still rely on liver transplantation, which is constrained by immune rejection and donor shortages [125]. In recent years, CAR-based cellular immunotherapies have provided new strategies for combating fibrosis. For instance, Mao et al. demonstrated that FAP-CAR and FAP-CAR-ΔZETA significantly ameliorated disease progression in a CCl₄-induced liver fibrosis model [126]. Yashaswini and colleagues, using CD5-targeted LNPs to deliver anti-FAP CAR mRNA, generated transient CAR-T cells in vivo that effectively alleviated fibrosis by depleting pro-fibrotic HSCs [127]. In the context of in vivo CAR-M strategies, Gao et al. developed an efferocytosis-based LNP system that co-delivers TRIM13 mRNA along with a FAP-specific CAR, enabling precise engineering of fibrosis-associated macrophages. This strategy employs TRIM13-mediated inhibition of the mitochondrial DNA (mtDNA)–STING signaling pathway, thereby maintaining a stable anti-inflammatory macrophage phenotype while preserving robust phagocytic capacity against FAP-positive cells. In this manner, it curbs inflammatory responses while promoting fibrosis reversal [128], offering a novel approach with both safety and functional plasticity for liver fibrosis treatment and holding significant translational potential. Nevertheless, its long-term safety and clinical controllability warrant further validation.

Intervertebral disc degeneration

Intervertebral disc degeneration (IVDD) constitutes a major pathological basis for low back pain, neck and shoulder pain, and disc herniation [129]. Its pathogenesis involves sustained inflammatory activation, diminished cellular function and metabolic activity, and alterations in cell number and phenotype [130]. Macrophages participate in the early stages of IVDD by regulating inflammation, apoptotic cell clearance, and tissue repair, thus serving as critical regulators of disease progression [131]. Although current pharmacological and surgical interventions can alleviate symptoms, their efficacy is limited, and they fail to address the core pathological mechanisms, thereby lacking the capacity to effectively modify or reverse disease progression [132]. Phagocytes recognize “eat-me” signals, specifically phosphatidylserine, on the surface of apoptotic cells and execute efferocytosis via receptors such as TIM-4, BAI1, and Stabilin-2 [133]. In IVDD, enhancing macrophage-mediated clearance of apoptotic nucleus pulposus cells can improve the local microenvironment and reduce apoptotic cell accumulation, thereby slowing degeneration [134]. Building on this concept, Zhou et al. developed a microneedle array delivery system to directly administer CAR-engineered macrophages with enhanced BAI1 expression to the intervertebral disc, thereby significantly improving local efferocytosis and modulating the degenerative disc microenvironment. These CAR eMs not only exhibited enhanced phagocytic activity but also secreted elevated levels of anti-inflammatory and tissue repair factors. Moreover, microneedle delivery enables minimally invasive penetration of the annulus fibrosus, precisely targeting CAR-M to damaged regions of the nucleus pulposus or annulus fibrosus while minimizing injury to the annulus and prolonging local cell retention. This approach has demonstrated potential as a novel immune cell therapy for IVDD [135] (Table 4 and Fig. 3).

Table 4.

Preclinical studies of in vivo CAR-M engineering in non-malignant diseases

Disease Type	Target Antigen	In Vivo CAR-M Engineering Strategy	Delivery System		Treatment Method	Key Mechanism of Action	Preclinical Model	Reference
PJI	MRSA surface protein SasA	SasA-CAR-M + shRNA CASP11		Bone implant nanocoding	Local intervention via bone implant coating	Enhances CAR-M phagocytosis and lysosomal killing of MRSA; overcomes CASP11-mediated immune evasion	PJI mouse model	Li et al. [102]
Sepsis	MRSA surface protein SasA	SasA-CAR mRNA + siCASP11		CRV/LNP	Intravenous injection	Blocks MRSA immune evasion; restores phagocytosis and inflammasome-mediated bactericidal function	MRSA-induced sepsis mouse model	Tang et al. [109]
Myocardial Fibrosis	FAP	mRNA overexpression of Lgmn to induce in situ generation of CAR-Ms		LNP	Intravenous injection	Enhances phagocytosis and efferocytosis of activated fibroblasts; inhibits excessive ECM deposition; improves cardiac remodeling	Myocardial infarction mouse model	Liu et al. [120]
Myocardial Ischemia-Reperfusion Injury	FAP	FAP-CAR mRNA to induce in situ generation of FAP-CAR-Ms		LNP-FAP-CAR	Intravenous injection	Depletes activated fibroblasts; reduces inflammation and fibrosis; improves cardiac remodeling	Myocardial ischemia-reperfusion mouse model	Du et al. [123]
Liver fibrosis	FAP	Co-delivery of FAP-CAR mRNA and TRIM13 mRNA		LNP	Intravenous administration	TRIM13 blocks mtDNA–STING signaling, while CAR targets FAP⁺ fibroblasts	CCl₄-induced liver fibrosis mouse model	Gao et al. [128]
Intervertebral Disc Degeneration	BAI1	BAI1-enhancing CAR-M (CAR eMs)		Annular microneedle delivery system	Local minimally invasive delivery	Enhances efferocytosis; clears apoptotic nucleus pulposus cells; regulates local inflammation and promotes tissue repair	Intervertebral disc degeneration mouse model	Zhou et al. [135]

Abbreviations: BAI1 Brain-Specific Angiogenesis Inhibitor 1, CAR-M Chimeric Antigen Receptor Macrophage, CASP11 Caspase-11, ECM Extracellular Matrix, FAP Fibroblast Activation Protein, LNP Lipid Nanoparticle, MRSA Methicillin-Resistant Staphylococcus aureus, PJI Periprosthetic Joint Infection, TRIM13 Tripartite Motif Containing 13

Fig. 3.

Landscape of in vivo engineered CAR-M applications in solid tumors and non-malignant diseases. This schematic overview depicts representative pathological contexts in which in vivo CAR-M programming has been explored. In solid tumors, diverse delivery platforms enable the in situ generation of CAR-expressing macrophages, frequently combined with immune-modulatory elements to potentiate macrophage activation, phagocytosis, and remodeling of the tumor microenvironment. Beyond oncology, in vivo CAR-M strategies have been extended to infectious and fibrotic conditions, where engineered macrophages facilitate pathogen clearance and tissue repair. Collectively, these approaches underscore the expanding therapeutic scope of in vivo CAR-M engineering across diverse disease settings

Delivery platforms for in vivo CAR-M programming

Viral vectors

Viruses, as natural delivery systems, can bypass physicochemical barriers, prolong circulation time, and enhance cell penetration. Ex vivo virus-mediated gene transduction technologies, such as those using lentivirus and adenovirus, enable stable integration of exogenous genes into the genome of target cells and represent the most commonly used transduction approaches for ex vivo CAR cell therapy. Among these, the modified adenoviral vector Ad5F35 achieves a transduction efficiency of 88.5% in autologous PBMCs while also exhibiting pro-inflammatory properties [44]. Adeno-associated virus (AAV), a non-enveloped single-stranded DNA virus, has emerged as a key tool for in vivo gene therapy. However, viral vectors possess inherent tissue tropism; their in vivo distribution is typically governed by serotype or envelope proteins, often making it difficult to fully align with therapeutic needs. Additionally, random integration carries a risk of insertional mutagenesis [136]. Furthermore, AAV faces multiple limitations in in vivo applications, including a stringent packaging capacity (typically < 5 kb) and the widespread prevalence of pre-existing neutralizing antibodies in humans, which collectively restrict its applicability and delivery efficiency.

LNPs

Lipid-based nanocarriers primarily include LNPs. Liposomes can carry both hydrophobic and hydrophilic molecules, whereas LNPs are widely applied for nucleic acid delivery [137]. As the most mature non-viral delivery platform, LNPs played a critical role during the COVID-19 pandemic and have emerged as one of the most advanced nucleic acid delivery systems in clinical practice [138]. LNPs currently used for nucleic acid delivery are typically composed of four core lipid components: ionizable lipids, helper phospholipids, cholesterol, and PEGylated lipids. Ionizable lipids undergo protonation and become positively charged in acidic environments, facilitating efficient nucleic acid encapsulation and endosomal escape. Helper phospholipids and cholesterol primarily maintain particle structural integrity and promote membrane fusion. PEGylated lipids form a hydrophilic barrier on the particle surface, thereby extending circulation time and reducing non-specific clearance. The cationic form of ionizable lipids itself possesses pro-inflammatory properties and certain cytotoxicity [139]. Under physiological pH conditions, these lipids remain overall electrically neutral, which helps reduce systemic toxicity and prolong in vivo circulation. In the acidic endosomal environment (pH < 6.5), they become protonated and acquire positive charges [140], thereby destabilizing the endosomal membrane and facilitating nucleic acid release into the cytoplasm. This charge-switching process is recognized as a key mechanism for efficient delivery [141].

EVs

EVs, including sEVs, are nanoscale lipid-bilayer vesicles secreted by various cell types, ranging in diameter from approximately 30 to 2000 nm [142]. As natural intercellular communication vehicles, EVs transport a diverse array of biomolecules, including proteins, RNA, and lipids [143]. EVs have been demonstrated to cross biological barriers and exhibit intrinsic targeting properties [144]. Engineered EVs or small vesicles offer advantages such as relatively straightforward preparation, long-term storability, and convenient transport, thereby facilitating the scale-up production and cost reduction of CAR immune cell-based biologics [90]. Currently, clinical applications of sEV-based delivery systems are being explored [145]. Notably, EVs retain certain intrinsic biological signals from their parental cells, including pathogen-associated molecular patterns (PAMPs), membrane proteins, and endogenous bioactive molecules. These components can engage in complex, context-dependent interactions with the host immune system [146]. Regarding nucleic acid delivery, transfection efficiency remains a key limiting factor for EV application. To date, only a few studies have indicated that, under specific conditions, sEVs may surpass LNPs in transfection efficiency [147–149]. The delivery efficiency and tolerability of EVs are influenced by multiple factors, including administration route, dosage, source cell type, and final product composition. For instance, in a lung metastasis model, Xiao and colleagues markedly enhanced delivery efficiency through inhalation administration [90]. Nevertheless, the inherent structural and compositional complexity of EVs poses challenges for their development. Elucidating the key bioactive constituents within EVs and their mechanisms of action is critical for advancing clinical translation. In the future, artificial intelligence-driven multi-omics analyses may enable the identification of key regulatory pathways and guide the rational design of engineered EVs, thereby achieving precise modulation of macrophage function and enhancing delivery efficiency and therapeutic controllability [150]. Furthermore, combining EVs with LNPs to construct hybrid nanoparticles has emerged as a promising strategy, demonstrating potential benefits in improving stability, reducing cytotoxicity, and optimizing oral siRNA delivery [151].

Polymeric nanocarriers

Polymer‑based delivery systems constitute a key strategy for in vivo CAR immunotherapy, offering advantages including straightforward fabrication and favorable biodegradability. Poly (β‑amino esters) (PBAEs), as representative cationic polymers, exhibit low cytotoxicity and desirable degradation profiles [152]. They form nanocomplexes with nucleic acids via electrostatic interactions and facilitate endosomal escape under acidic conditions [153]. In 2017, Smith and colleagues first employed PBAE nanocarriers to reprogram circulating T cells into CAR‑T cells in vivo, thereby establishing the feasibility of in vivo CAR‑T generation [154]. Subsequent studies have further enhanced therapeutic efficacy through mRNA delivery [155]. Moreover, this strategy has demonstrated potential for CAR‑M delivery in a brainstem glioma model. Polyethyleneimine (PEI), a classic cationic polymer, promotes endosomal disruption and nucleic acid release via the “proton sponge effect” [156]. High‑molecular‑weight cationic polymers tend to accumulate in vivo, bind to plasma proteins within the complex bloodstream, and form unstable aggregates, which can damage normal tissues and induce cytotoxicity [157]. Consequently, for in vivo applications, the stability and biocompatibility of polymer nanoparticles must be carefully optimized. Current research primarily focuses on strategies such as PEGylation, surface ligand functionalization, and the development of stimuli‑responsive nanocarriers to modulate charge properties and in vivo behavior, thereby enabling more precise targeted delivery and in vivo distribution visualization [156] (Table 5).

Table 5.

Summary of in vivo CAR-M delivery systems

Delivery type	Cargo gene	Target cell specificity	Advantages	Limitations	Reference
LNP delivery system	CA9-CAR + IL-2R-TLR4 circRNA	TAMs	Avoids systemic IL-2 toxicity; induces M1 polarization; local immune activation	Limited circRNA expression duration; delivery efficiency depends on material	Jing et al. [59]
	GPC3-CAR mRNA + Siglec-GΔITIMs mRNA	TAMs	Synergistic dual mRNA enhances phagocytosis	Partial uptake by non-immune cells	Yang et al. [64]
	GPC3-CAR–Super IL-2 + FAP-CAR–ΔTGFβRII	TAMs + CAFs	> 80% in vitro editing efficiency; >20% in vivo efficiency	Complex design; safety not fully evaluated	Zhang et al. [67]
	CD68 promoter-driven MUC1-CAR plasmid DNA	TAMs	Dual mechanism with CD47 blockade enhances phagocytosis	Potential toxicity remains to be evaluated	Liu et al. [72]
	Mannose-modified FAP CAR mRNA	TAMs + CAFs	Degrades fibrotic barrier; enhances drug penetration	CAF heterogeneity	Wang et al. [71]
	GP75-CAR + STING agonist	TAMs	Synergistic activation of innate immunity and phagocytosis	Requires intratumoral injection; risk of off-target uptake	Park et al. [91]
	Multiple CAR mRNAs	TAMs	Synergistic immune activation	Long-term CAR expression remains unclear	Gu et al. [94]
	circRNA panCAR	T cells, NK cells, macrophages	Simultaneous activation of multiple immune populations; vaccine synergy	Targeting efficiency and mechanism not fully understood	Wang et al. [93]
	SasA-CAR mRNA + siCASP11	Macrophages	Systemic anti-infective effect; immune restoration	Complex pathogenesis; limited strain specificity	Tang et al. [109]
	FAP-CAR + Lgmn	Macrophages	Enhanced efferocytosis; improved cardiac function	Safety not established	Liu et al. [120]
	FAP-CAR mRNA	Macrophages	Extended therapeutic window; sustained functional improvement	Systemic toxicity; off-target effects	Du et al. [123]
	CAR + SIGLEC9-based CSR circRNA	TAMs	High stability; low production cost	Intratumoral injection limits application scope	Fu et al. [83]
	FAP-CAR mRNA + TRIM13 mRNA	Fibrosis-associated macrophages	TRIM13 maintains an anti-inflammatory state, while CAR targets FAP⁺ fibroblasts	Safety evaluation	Gao et al. [128]
Biological vector	CD133-CAR	TAMs	Safe for repeated administration	Complex vector preparation	Zhou et al. [79]
	Bispecific adaptor molecules	TAMs	Targets tumor heterogeneity; sustained expression	Interaction with engineered EcN and antibiotics unclear	Zhang et al. [80]
Nanoparticle-hydrogel system	CD133-CAR plasmid	TAMs	Local sustained release; reduced systemic toxicity	Surgery-dependent; long-term safety unclear	Chen et al. [77]
Polymeric nanoparticle system	HER2-CAR DNA	TAMs	Deep tissue delivery; improved brain distribution	Requires intratumoral injection; complex application	Gao et al. [78]
	CAR + IFN-γ plasmid	TAMs	M2→M1 repolarization; sustained immune activation	Polymer-associated cytotoxicity	Kang et al. [81]
Extracellular vesicles	MSLN-CAR mRNA	TAMs	Natural delivery system; inhalation improves lung targeting	Biodistribution and stability issues	Xiao et al. [90]
Implant coating system	SasA-CAR + shCASP11	Macrophages	Local long-term antibacterial effect; biofilm disruption	Limited to local infection applications	Li et al. [102]

Abbreviations: CA9 Carbonic Anhydrase IX, CAR Chimeric Antigen Receptor, CAR-M Chimeric Antigen Receptor Macrophage, CAFs Cancer-Associated Fibroblasts, CSR Chimeric Switch Receptor, EcN Escherichia coli Nissle, FAP Fibroblast Activation Protein, GP75 Glycoprotein 75, GPC3 Glypican-3, HER2 Human Epidermal Growth Factor Receptor 2, IFN-γ Interferon Gamma, IL-2 Interleukin-2, Lgmn Legumain, LNP Lipid Nanoparticle, MSLN Mesothelin, MUC1 Mucin 1, siCASP11 Small Interfering RNA Targeting Caspase-11, STING Stimulator of Interferon Genes, TAMs Tumor-Associated Macrophages, TGFβRII Transforming Growth Factor Beta Receptor II, TRIM13 Tripartite Motif Containing 13

Translational and mechanistic comparison of in vivo and ex vivo CAR-M therapies

As CAR-M therapy moves from the lab to the clinic, high costs, complex manufacturing, and limited scalability have become major bottlenecks. In this setting, engineering CAR-M directly in vivo offers a key addition to conventional ex vivo adoptive cell therapy and is reshaping how we design and translate CAR-based immunotherapies.

Engineering paradigms and manufacturing constraints

Ex vivo CAR-M engineering uses lentiviral or adenoviral vectors for stable gene integration. This approach allows researchers to fine-tune CAR expression in a controlled ex vivo setting, from choosing promoters and optimizing expression levels to designing signaling domains. As a result, it offers a high degree of engineering control. Nevertheless, conventional adoptive CAR-M therapy involves a multi-step process, including peripheral blood collection, cell isolation and activation, ex vivo gene transduction, expansion culture, quality control, and eventual reinfusion, typically spanning several weeks or longer [158]. CAR-M must be generated from patient-derived monocytes through ex vivo differentiation and genetic modification, rendering the production workflow both complex and time-consuming. Moreover, variability in differentiation status, CAR expression, and inflammatory polarization among cells from different patients compromises batch-to-batch consistency. This process is heavily reliant on GMP-grade manufacturing facilities, viral vectors, and stringent quality-control systems [159]. To obtain a clinical-grade dose of CAR-M, current ex vivo protocols typically require 21–27 days [160]. During this period, many patients with rapidly progressing solid tumors may lose the opportunity for treatment due to disease progression. In vivo CAR-M strategies use lipid nanoparticles or other non-viral delivery systems to deliver mRNA or DNA encoding sequences directly inside the body, reprogramming endogenous macrophages right where they are. The clearest clinical advantage of generating CAR-M in vivo is that it greatly shortens the time from diagnosis to treatment. In vivo CAR-M relies on systemic or local injection of therapeutic vectors; following one or a few administrations, these vectors successfully transduce target cells and drive functional CAR expression within 24–48 h [161]. This brings the timeline of CAR-M therapy closer to that of monoclonal antibodies or chemotherapeutic agents, potentially improving patient survival outcomes. One important caveat is that CAR cells generated through in vivo delivery do not survive or persist as well as their ex vivo manufactured and infused counterparts. Conventional linear mRNA is unstable inside cells, so CAR expression from in vivo delivery often drops sharply within a few days. This makes it hard to sustain long-term antitumor effects [162]. In addition, ex vivo protocols can supply high cytokine levels and optimized co-stimulatory signals to produce more stable engineered cells. The in vivo microenvironment, by contrast, seldom provides such exogenous support and also imposes suppressive pressures from tumor-related or autoimmune pathological settings [163]. Producing high-quality cell products requires precisely controlled ex vivo conditions. This constraint directly limits the long-term antitumor efficacy of the therapy.

Clinical translation: cost, scalability, and safety trade-offs

When it comes to clinical translation, how well CAR-M therapy is adopted depends not just on its biological effects but also on real-world factors like safety, scalability, and cost. In autologous cell therapy, each patient-specific batch requires dedicated GMP suites or automated closed systems, high-grade reagents (e.g., GMP-grade cytokines), and intensive manual monitoring to prevent contamination and ensure product viability [164]. In contrast, vectors carrying mRNA or plasmid DNA are not subject to patient-specific constraints and parallel vaccine production models, offering high scalability and batch consistency. These platforms are amenable to centralized, large-scale industrial manufacturing, with marginal costs decreasing substantially as production volume increases [42]. Even so, this production path does not fully remove the influence of individual differences. In some diseases, such as sepsis, the underlying mechanisms and the range of pathogens are highly variable [165]. A single, standardized CAR-M design may not work for all clinical subtypes. The key challenge for in vivo strategies, then, is how to achieve mechanism-driven precision control within a limited treatment window.

Current adoptive cell therapies typically require patients to undergo lymphodepleting chemotherapy prior to infusion to reduce competition from endogenous immune cells. However, this approach is associated with significant side effects, including pancytopenia and an increased risk of infection [166]. Moreover, such preconditioning may preclude effective leukapheresis or result in harvested cells that are difficult to expand and transduce ex vivo due to their exhausted state. In contrast, in vivo CAR-M programming directly harnesses endogenous phagocytes for conversion, eliminating the need for lymphodepleting preconditioning and thereby preserving the integrity of the patient’s immune system. This strategy extends treatment to patients who are unable to undergo apheresis or in whom monocyte mobilization has failed, substantially broadening the clinical accessibility of CAR-M therapy [167]. Still, the safety edge of in vivo strategies is not guaranteed. For one thing, delivery vehicles like LNPs do not stay where they are supposed to. They tend to build up in the liver, which can cause off-target expression and toxicity. For another, CAR expression is not subject to precise dose control or in vitro quality checks. If activation goes too far or targeting goes wrong, the result could be an unpredictable inflammatory response. Also, while short-lived expression lowers the risk of long-term toxicity to some degree, it may call for repeated doses. That in turn brings new concerns like immunogenicity or cumulative toxicity [168]. So the key to ensuring the clinical safety of in vivo CAR-M will be how to finely tune expression intensity, duration, and spatial distribution while improving delivery efficiency.

Biological constraints in the tumor microenvironment

Macrophages exhibit high plasticity and remarkable lineage heterogeneity [169]. However, as terminally differentiated cells, they have very limited ability to be passaged continuously in vitro [170], which makes large-scale expansion as achievable with T cells difficult. Upon entering the complex TME, ex vivo-derived CAR-Ms are limited in number, and their phenotypic stability is susceptible to modulation by the dominant endogenous TAM population and associated microenvironmental cues. These exogenous CAR-Ms must compete with endogenous M2-like TAMs for spatial and nutritional resources upon in vivo administration while being continuously exposed to immunosuppressive signals, including IL-10, TGF-β, and hypoxic metabolic stress [171]. This competitive and suppressive environment may drive their polarization toward an immunosuppressive phenotype, undermining the capacity to sustain CAR-mediated reprogramming. Furthermore, organ-specific immune landscapes and intratumoral spatial heterogeneity impose greater demands on the precise customization and optimization of ex vivo CAR-M strategies [172]. In contrast, in vivo CAR-M approaches utilize systemic delivery vectors to directly engineer immunosuppressive endogenous TAMs, enabling in situ acquisition of CAR or signaling-regulatory domains. This strategy thereby reshapes the functional status and transcriptional profiles of pre-existing myeloid cells within the TME, circumventing the potential risks associated with introducing exogenous cell populations while offering both safety and accessibility. Consequently, it provides a novel avenue for sustaining CAR-M function and enabling continuous immune modulation.

That said, in vivo CAR-M comes with its own set of challenges around precision and control. CAR cells generated in situ must recognize dynamically changing target antigens inside the body, including molecules that also appear on normal tissues. Unlike ex vivo engineering, this approach lacks built-in quality control and dose calibration. Poor targeting accuracy can therefore cause treatment failure or even serious side effects [168]. Thanks to recent advances in spatial transcriptomics and single-cell sequencing, we now know that tumor antigens are highly heterogeneous across both space and time [173]. Uneven antigen expression within the tumor microenvironment may prevent CAR-M from reprogramming all TAMs. In some settings, local signals can even push CAR-M toward a pro-tumor phenotype [174].

Translational barriers to in vivo CAR-M engineering

Safety and off-target risks

The foremost principle of in vivo gene therapy is delivery specificity. Despite significant progress in the development of nanocarriers and viral vectors for in vivo delivery, off-target effects remain a major safety concern. The liver and spleen function as the primary filtering organs of the circulatory system. Macrophages within the reticuloendothelial system, such as the Kupffer cells in the liver, possess a potent capacity for clearing exogenous particulates. While this inherent macrophage tropism is advantageous for myeloid cell targeting, it also results in substantial vector accumulation in the liver and spleen rather than at tumor sites [41]. After entering the bloodstream, LNPs rapidly adsorb apolipoprotein E (ApoE) onto their surface. These ApoE-decorated LNPs are subsequently recognized and internalized by low-density lipoprotein receptors (LDLR) on hepatocytes through receptor-mediated endocytosis [175]. Although this biodistribution profile may be therapeutically beneficial for liver diseases, ectopic CAR expression in normal tissue macrophages could potentially induce severe toxicity, leading to damage in healthy tissues. Activation of downstream ITAM signaling by CAR signaling domains (e.g., CD3ζ or FcRγ) in non-target cells may trigger systemic inflammatory responses or cellular dysfunction. Furthermore, macrophages derived from distinct tissues exhibit signifiscant phenotypic and functional heterogeneity, resulting in inconsistent CAR expression and signal transduction efficacy across different macrophage subsets. This variability further amplifies the functional uncertainty arising from uncontrolled in vivo biodistribution [176].

In recent years, engineering optimization of delivery systems has emerged as a critical direction in this field. Tissue distribution profiles can be improved to some extent by modifying administration routes and integrating active and passive targeting strategies [177]. For example, adjusting the ratio of ionizable lipids to helper lipids in LNPs, or incorporating selective organ-targeting lipids, can alter the overall charge characteristics and plasma protein adsorption profiles of nanoparticles, thereby enabling organ-selective delivery, reducing innate liver tropism, and redirecting nanoparticles to other tissues [177, 178]. Furthermore, conjugating antibody fragments, scFvs, nanobodies, or specific carbohydrate ligands onto nanoparticle surfaces can enhance uptake efficiency by specific immune cell subsets while minimizing deposition in non-target tissues [179]. Notably, however, in a complex in vivo environment, the formation of a protein corona may partially shield surface ligands, potentially compromising the efficacy of active targeting strategies and contributing to inter-individual variability in targeting efficiency. Future optimization of delivery systems should extend beyond material design to incorporate insights from genomics and proteomics. By tailoring LNP composition and surface modification strategies based on patient-specific molecular profiles and disease states, it may be possible to accommodate individual physiological differences, thereby enhancing delivery precision and therapeutic specificity [178].

Innate immune activation and inflammatory responses to delivery systems

In vivo, CAR-M delivery systems themselves can also activate host innate immune responses. Upon entering the body, LNP carriers and their nucleic acid cargo can be recognized by pattern recognition receptors, activating Toll-like receptors (TLRs) or cytosolic nucleic acid sensing pathways (such as cGAS-STING), and promoting the expression and release of inflammatory cytokines (e.g., IL-1β, IL-6) and inflammasome activation (e.g., NLRP3), thereby triggering inflammatory cascades [180]. This immune activation not only induces local or systemic adverse effects but may also accelerate clearance of the delivery vector, shorten the duration of CAR expression, and compromise therapeutic efficacy. Furthermore, this property may elicit adaptive immune responses upon repeated administration, thereby reducing delivery efficiency. Concurrently, repeated exposure to PEG-lipids may induce anti-PEG antibody production, leading to the “accelerated blood clearance (ABC) phenomenon” and diminishing the efficacy of subsequent doses, a disadvantage in the treatment of chronic diseases requiring long-term management, such as advanced malignancies [181, 182].

To address immune activation triggered by delivery systems, next-generation ionizable lipids have been engineered with structural modifications, such as the incorporation of biodegradable ester bonds and reduced cationic charge density, to attenuate aberrant TLR and inflammasome pathway activation while preserving endosomal escape capacity [179]. For instance, incorporating the anionic lipid DOPG into LNP formulations can modulate immune cell tropism, dampen early immunostimulatory signals, and promote immune tolerance, thereby reducing inflammatory cytokine release rather than merely inducing immune activation and ultimately enhancing the immunocompatibility of the delivery vehicle [183]. Furthermore, tuning PEG density and lipid composition on LNP surfaces can alter the protein corona architecture, improving delivery efficiency upon repeated administration and mitigating the generation of anti-PEG antibodies [178]. Nevertheless, immunological memory effects in the context of chronic inflammation may still impact the stability of long-term therapeutic efficacy. Therefore, in chronic disease settings or multicycle treatment regimens, the immune tolerance of delivery systems requires further optimization.

Challenges in gene editing efficiency and expression durability

In gene editing, viral vectors such as lentivirus and adeno-associated virus (AAV) achieve high gene delivery efficiency but pose risks of genomic integration and immunogenicity [184]. Although mRNA-LNP delivery avoids the risk of genomic integration and is theoretically better suited for short-term, controllable expression, macrophages are evolutionarily specialized for pathogen recognition and clearance of exogenous nucleic acids. The introduction of foreign DNA or RNA may trigger interferon responses that suppress transgene expression [185]. Consequently, the functional transfection efficiency of macrophages in vivo remains relatively low and expression is typically transient, often necessitating repeated administration to sustain therapeutic efficacy. Moreover, the tumor microenvironment is highly heterogeneous and dynamic. Factors such as hypoxia, acidic metabolic conditions, immunosuppressive cytokines, and phenotypic variability among distinct macrophage subsets can all influence cellular uptake of delivery vectors, nucleic acid stability, and transgene expression levels. Whether current gene editing and delivery platforms can achieve precise control over target cell specificity, timing, and expression intensity within such complex contexts remains an open question [21].

The development of responsive or intelligent nanocarriers, such as pH-sensitive, enzyme-sensitive, or TME-responsive systems, enables more efficient release of nucleic acids and enhances intracellular bioavailability within tumors and immunosuppressive niches, leading to improved functional expression and engineering efficiency [178]. Combining LNPs with novel nanocarrier platforms, including fusogenic exosomes, microvesicles, or engineered vesicle-derived carriers, offers advantages such as enhanced immunocompatibility and reduced clearance rates attributable to their natural origin [186]. Small molecule-based synergistic regulation, exemplified by transient modulation of macrophage signaling pathways using STAT3 inhibitors, can enhance LNP internalization and nucleic acid expression efficiency upon delivery. This strategy helps circumvent intrinsic clearance mechanisms and antiviral responses of macrophages, thereby improving transfection efficiency [187].

Delivery-constrained CAR expression and functional limitations

In in vivo CAR delivery strategies, the expression level and duration of the CAR molecule are key determinants of therapeutic efficacy and safety. Insufficient expression may lead to inadequate immune activation, whereas persistent or excessive expression can result in off-target toxicity and exhaustion of immune cell function.

Current non-viral delivery systems predominantly rely on nucleic acid molecules to mediate CAR expression. Conventional linear mRNA typically reaches its peak expression rapidly after in vivo delivery and declines to baseline levels within approximately 72 h, which is mainly attributed to its high susceptibility to exonucleases. To prolong expression persistence, researchers have developed various structurally optimized RNA molecules. Circular modified mRNA (cmRNA) enhances molecular stability, thereby significantly extending the translation duration; it can maintain CAR expression at > 50% of the peak level at 72 h post-administration and extend the effective therapeutic window to 7–14 days [188]. Self-amplifying RNA (saRNA) encodes a replicase that enables RNA self-amplification within cells, achieving more durable and higher-level expression at doses far lower than those required for conventional mRNA [189]. For instance, Yu et al. encapsulated saRNA targeting the p38 and TFEB promoters into a metal-organic framework-based delivery system, which significantly prolonged circulation time and enhanced tumor targeting [190]. In addition, circRNA, owing to its covalently closed topology, exhibits greater resistance to exonucleases and similarly demonstrates sustained expression potential [191].

In recent years, research has gradually shifted toward the fine-tuning of CAR expression. Inducible CAR systems enable higher precision by controlling CAR expression or activity through exogenous small molecules or environmental cues. For example, hypoxia-responsive CAR systems, which leverage features of the tumor microenvironment, offer new strategies for solid tumor therapy. The core region of solid tumors is typically hypoxic (oxygen concentration below 2%), with hypoxia-inducible factors (HIFs) playing a key regulatory role [192]. Kosti et al. engineered hypoxia-responsive CAR cells that selectively express and function under hypoxic conditions [193]. He et al. developed the HiTA-CAR system, in which the CAR gene is placed under the control of a hypoxia response element (HRE) promoter, combined with an oxygen-dependent degradation (ODD) domain. This design achieves protein degradation under normoxic conditions and specific expression under hypoxic conditions, thereby significantly improving tumor specificity [194]. Similarly, synthetic biology-based Boolean logic-gated CAR systems represent a key technology for precise regulation. For instance, the “AND-gate” design requires immune cells to recognize both a specific tumor antigen and tumor microenvironment signals (e.g., high lactate or low pH) for activation, thereby markedly reducing off-target effects [195]. ON-switch CAR systems enable reversible control of CAR signaling via split architectures or small molecule-dependent activation mechanisms, allowing activation only when needed [196]. These strategies not only improve therapeutic safety but also partially compensate for functional limitations resulting from transient CAR expression.

The structural design of the CAR molecule itself also significantly influences functional output. Unlike CAR-T cells, CAR-M relies more heavily on myeloid-specific signaling modules (e.g., FcRγ, Megf10, or TLR-related domains) to drive phagocytosis and inflammatory responses. Therefore, optimizing the intracellular signaling domain may sustain robust effector functions even at relatively low expression levels. For instance, incorporating the CD147 intracellular domain induces the secretion of matrix metalloproteinases upon CAR-M recognition of the target antigen, thereby promoting tumor stroma degradation [197].

Of note, sustained CAR signaling is not necessarily more beneficial with longer duration. Excessive or prolonged signaling can drive immune cell exhaustion, whereas overly transient expression fails to establish an effective antitumor immune response. Achieving optimally sustained signaling through modulation of expression kinetics, rather than simply prolonging expression time, may be more critical than pursuing extended expression alone. For instance, in CAR-T studies, different costimulatory domains (e.g., CD28 vs. 4-1BB) influence T-cell persistence by modulating metabolism and signaling durability [198]. Although systematic validation in CAR-M is still lacking, these findings offer important insights for translation (Fig. 4).

Fig. 4.

Advantages and limitations of in vivo CAR-M programming. This schematic outlines the key benefits and challenges associated with in vivo CAR-M engineering. In contrast to conventional ex vivo generated CAR-Ms, in vivo programming circumvents the need for lymphodepletion and complex manufacturing steps, enabling rapid induction of CAR expression in endogenous macrophages following systemic or local vector delivery. This strategy shortens the treatment timeline, reduces costs and dependency on GMP facilities, and broadens patient accessibility. Moreover, by directly reprogramming tumor-associated macrophages in situ, this approach facilitates remodeling of the tumor immune microenvironment without the infusion of large numbers of exogenous cells. Nevertheless, several limitations persist. Systemic delivery poses challenges in biodistribution control, often resulting in vector accumulation in the liver and spleen and raising the risk of off-target CAR expression. Such ectopic activation may trigger inflammatory toxicity. Furthermore, certain delivery platforms such as LNPs can elicit innate immune responses, while non-integrating vectors typically confer only transient gene expression, thereby limiting therapeutic durability and necessitating repeated administration

Conclusion and prospect

In vivo CAR-M therapy represents a paradigm shift in macrophage engineering, offering the potential to overcome the inherent limitations of traditional ex vivo cell therapies, such as manufacturing complexity, high costs, and limited scalability. By enabling the direct in vivo delivery of CAR constructs for in situ engineering of host macrophages, this strategy theoretically enhances therapeutic accessibility and flexibility. Future efforts should prioritize the optimization of delivery systems to improve cell selectivity and expression efficiency, for instance, through refined organ-targeting formulations and enhanced mRNA stability, thereby achieving more stable and controllable in vivo expression. At the same time, challenges related to expression regulation and safety warrant continued investigation. The complexity of the tumor microenvironment and the phenotypic plasticity of macrophages represent key variables that may influence CAR functionality. Incorporating microenvironment-responsive or inducible regulatory elements may offer a means to enhance therapeutic efficacy. Moreover, the balance between safety and durability remains essential. While transient expression provides reversibility, repeated administration may provoke immune responses against delivery components or CAR structures. Consequently, the development of immunocompatible material platforms and repeat-dosing regimens constitutes a critical direction for advancing long-term clinical applications.

From a translational perspective, the potential of in vivo CAR-M therapy is further highlighted by its applicability in combination regimens and the continued refinement of clinical translation pathways. By harnessing their intrinsic phagocytic activity, antigen-presenting capacity, and the ability to remodel the tumor immune microenvironment, CAR-Ms are well positioned to synergize with other therapeutic modalities, such as immune checkpoint inhibitors, CAR-T cells, or oncolytic viruses, thereby broadening the therapeutic landscape for solid tumors. At the same time, the development of standardized assays for monitoring in vivo CAR expression and the establishment of robust safety evaluation parameters will provide an essential foundation for clinical implementation. Ultimately, as delivery technologies and gene regulation strategies continue to advance, in vivo CAR-M therapy is expected to mature into a scalable and broadly applicable next-generation immunotherapeutic platform for solid tumors and other macrophage-associated diseases.

Abbreviations

ACT

Adoptive Cell Therapy

ADCP

Antibody-Dependent Cellular Phagocytosis

AAV

Adeno-Associated Virus

CAR

Chimeric Antigen Receptor

CAR-M

Chimeric Antigen Receptor Macrophage

CAR-NK

Chimeric Antigen Receptor Natural Killer Cell

CAR-T

Chimeric Antigen Receptor T Cell

CA9

Carbonic Anhydrase IX

CAF

Cancer-Associated Fibroblast

CD3ζ

CD3 zeta chain

CRS

Cytokine Release Syndrome

CTL

Cytotoxic T Lymphocyte

ECM

Extracellular Matrix

sEVs

Small Extracellular Vesicles

FcRγ

Fc Receptor Gamma Chain

FAP

Fibroblast Activation Protein

GMP

Good Manufacturing Practice

GBM

Glioblastoma

GPC3

Glypican 3

HER2

Human Epidermal Growth Factor Receptor 2

HCC

Hepatocellular Carcinoma

ICD

Intracellular Domain

iMACs

Induced Pluripotent Stem Cell-Derived Macrophages

iPSC

Induced Pluripotent Stem Cell

ITAM

Immunoreceptor Tyrosine-Based Activation Motif

ITIM

Immunoreceptor Tyrosine-Based Inhibitory Motif

LNP

Lipid Nanoparticle

Megf10

Multiple EGF-Like Domains 10

MMPs

Matrix Metalloproteinases

MI

Myocardial Infarction

MRSA

Methicillin-Resistant Staphylococcus aureus

MyD88

Myeloid Differentiation Primary Response 88

PBMC

Peripheral Blood Mononuclear Cell

PDAC

Pancreatic Ductal Adenocarcinoma

PJI

Periprosthetic Joint Infection

RCC

Renal Cell Carcinoma

scFv

Single-Chain Variable Fragment

Syk

Spleen Tyrosine Kinase

TAM

Tumor-Associated Macrophage

TGF-β

Transforming Growth Factor Beta

TIR

Toll/Interleukin-1 Receptor Domain

TLR

Toll-Like Receptor

TME

Tumor Microenvironment

TNF-α

Tumor Necrosis Factor Alpha

tSH2

Tandem Src Homology 2 Domain

VH

Variable Heavy

VL

Variable Light

ZAP70

Zeta-Chain-Associated Protein Kinase 70

IVDD

Intervertebral Disc Degeneration

Authors’ contributions

Shuai Wang: Conceptualization, Writing – original draft and Visualization. Lucheng Zhou: Writing – original draft, Data curation. Xinlei Chen: Writing – original draft, Investigation. Zhihao Xu: Literature review, Data curation. Yizhao Chen: Supervision, Writing – review & editing. Ning Lin: Supervision, Project administration, Writing – review & editing. Jiajie Tu: Conceptualization, Supervision, Writing – review & editing.

Funding

This study was supported by the National Natural Science Foundation of China (82373877, 82504836), the Scientific Research Foundation of the Education Department of Anhui Province (Grant No. 2024AH040093), the Chuzhou Science and Technology Program (Grant No. 2024YF007), the Health Research Program of Chuzhou (Grant No. CZWJ2024A001), and the Postgraduate Innovation Research and Practice Program of Anhui Medical University (Grant No. YJS20250044, YJS20250122).

Data availability

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All authors have read and approved the final manuscript and consent to its publication.

Competing interests

The authors declare no competing interests.