In vivo CAR cell therapy: from bench to bedside

Abstract

Chimeric antigen receptor T cell (CAR-T) therapy has emerged as a highly efficacious treatment for refractory and relapsed hematological malignancies in recent years. However, the complex manufacturing procedures, stringent logistical requirements, and protracted production timelines associated with autologous ex vivo CAR-T cells render them costly and inaccessible to many patients. In contrast, in vivo CAR-T therapy directly delivers CAR-encoding transgenes to endogenous T cells, reprogramming them in situ. This approach obviates the need for apheresis, ex vivo cell manufacturing, and lymphodepleting chemotherapy inherent in conventional CAR-T therapy. Consequently, in vivo CAR-T represents a more efficient and economical paradigm, transforming CAR-T from individualized cellular products towards truly “ready-to-use” therapeutics. This review summarizes the latest research progress in in vivo CAR cell therapies, spanning from bench to bedside, to provide insights for advancing their clinical translation.

Introduction

Chimeric antigen receptor-T cell (CAR-T) therapy has demonstrated remarkable efficacy in treating hematological malignancies, significantly improving survival of patients with relapsed and refractory disease. In recent years, its application has been extended to solid tumors and autoimmune disorders, reflecting a new paradigm in cellular immunotherapy. To date, a total of 14 commercial CAR-T products have been approved, all of which are personalized autologous cellular therapies manufactured ex vivo. The prohibitive cost, complex and lengthy production process, and risk of manufacturing failures thus severely constrain patient accessibility. While allogeneic CAR-T therapies aim to overcome these hurdles, they have encountered challenges such as host-versus-graft rejection and graft-versus-host disease [1]. Similarly, the clinical translation of CAR-NK cells is hindered by suboptimal efficacy and short persistence [2].

In vivo CAR-T is an innovative strategy that directly reprograms endogenous T cells into CAR-Ts in situ using gene delivery vectors. This strategy replaces the cumbersome ex vivo manufacturing process with a single-step infusion, eliminating the need for apheresis and lymphodepletion. Thereby, it substantially reduces costs and shortens the time from screening to infusion (Fig. 1). Given these advantages, in vivo CAR-T holds great potential for future clinical use. This review discusses the current landscape of in vivo CAR technology for engineering T cells and other immune cells, highlighting key preclinical breakthroughs that pave the way for safer and more effective clinical translation.

Fig. 1.

The conventional ex vivo CAR-T manufacturing process involves multiple steps: leukapheresis, T-cell sorting and activation, lentiviral or retroviral transduction, CAR-T cell expansion, release testing, product transportation, and final infusion. This vein-to-vein time typically ranges from 7 to 22 days. In contrast, in vivo CAR-T therapy streamlines the process to a single step, requiring only 1 to 3 days for preparation

Gene delivery vectors for in vivo CAR cell therapies

In vivo CAR-T therapies have gained significant momentum over the last decade (Fig. 2), with a variety of gene delivery vectors having been developed for in vivo CAR engineering. These vectors can be broadly categorized as viral and non-viral systems. Viral vectors primarily comprise lentiviral vectors (LVs) and adeno-associated viral vectors (AAVs). Non-viral vectors encompass polymer nanoparticles (NPs), lipid nanoparticles (LNPs), bioinspired delivery vehicles, and macroencapsulation devices. Due to their fundamentally different compositions, these vector classes display distinct biological characteristics and utilize varying mechanisms to deliver CAR transgenes in vivo [3–6] (Table 1).

Fig. 2.

Timeline of key events for in vivo CAR cell therapies from bench to bedside. Since the proof-of-concept of in vivo CAR-T was established in 2017, preclinical studies have focused on optimizing both viral and non-viral vector platforms. The clinical pipeline for in vivo CAR-T therapy is rapidly advancing, marked by the publication of the first-in-human studies for both viral and non-viral vectors in 2025

Table 1.

Characteristics of different gene delivery vectors for in vivo CAR-T generation

Vectors	Lentivirus	Adeno-associated virus	Polymer nanoparticles	Lipid nanoparticles	Bioinspired delivery vehicles	Macroencapsulation devices
Maximum loading capacity	10 kb3	4.7 kb3	Flexible	6 kb4	Flexible	Flexible
Cargo format	ss RNA	ss DNA	mRNA or pDNA	mRNA/CircRNA or pDNA	mRNA or pDNA	Flexible (usually viral vectors)
Transduction efficiency	High	Medium	Low	Medium	Flexible	High in local areas surrounding the devices
Vector diameter	80 ~ 120 nm5	20 ~ 25 nm6	100 ~ 150 nm4	50 ~ 150 nm4	Nanoscale to cellular scale	Macroscopic scale
Genome integration	Yes	No	No	No	No	Depending on delivery vectors
Immunogenicity	High	High	Medium	Low	Very low	Flexible
Repeat dosing probability	No	No	Yes	Yes	Yes	Yes, implantable
Production difficulty	High	High	Low	Low	High	Medium
Primary Applications	Systemic, long-term immune reconstitution	Localized or site-specific disease	Systemic, flexible and controllable treatment	Systemic, flexible and controllable treatment	Diseases with complex microenvironments or formidable biological barriers	Local/in situ immune-programming

ss single-stranded; mRNA messenger ribonucleic acid; CircRNA circle RNA; DNA deoxyribonucleic acid; pDNA plasmid DNA

The established use of viral vectors lays a solid foundation for in vivo CAR-T generation by leveraging their exceptional transduction efficiency. The well-characterized tropism and genetic engineering potential of viral envelopes and capsids enable precise targeting of specific immune cell subsets. Furthermore, decades of refinement in viral vector production results in simplified processes for generating high-titer and clinical-grade materials. This is supported by comprehensive safety profiles derived from extensive clinical experience [7].

Non-viral vectors present a transformative alternative to traditional viral vectors for in vivo CAR-T engineering, effectively addressing critical challenges such as immunogenicity and insertional mutagenesis [8]. Their intrinsic biocompatibility and biodegradability remarkably enhance the safety profile, a feature further reinforced by the transient expression of delivered transgenes. Crucially, the streamlined synthesis of these materials bypasses the complexities inherent in viral vector production, such as reliance on fragile cellular substrates and labor-intensive purification process [9], thereby enabling a large-scale manufacturing with reduced batch variability and costs.

Viral delivery vectors

Lentiviral vectors

LVs are the predominant platform for both ex vivo and in vivo CAR-T production. A critical factor modulating LV behavior is pseudotyping—the substitution of envelope glycoproteins, which determines host cell tropism. As the most widely used envelope protein for pseudotyping, the vesicular stomatitis virus glycoprotein (VSV-G) confers broad tropism to mammalian cells by binding the ubiquitously expressed low-density lipoprotein receptor (LDLR) [10]. However, this poses a significant challenge for in vivo CAR-T generation. Crucially, VSV-G mediates not only ligand-receptor attachment but also the essential fusion of viral and endosomal membranes [11, 12]. Therefore, a preferable engineering strategy involves mutating, rather than knocking out, VSV-G to abolish its native tropism while preserving its fusogenic function. For instance, Nikolic J et al. [13] identified K47 and R354 in VSV-G as key residues for LDLR binding with minimal.

impact on membrane fusion. Other feasible mutation sites include H8 [14], I182 [15], and I331 [16]. Notably, introducing opposing charges at specific residues help destabilize the VSV-G–LDLR interaction [15]. Moreover, mutations like the T214N/T352A in VSV-G can confer resistance to serum inactivation [17].

Beyond VSV-G, LVs can be pseudotyped with paramyxovirus envelope proteins—including those from Nipah virus (NiV) and Measles virus (MV)—with truncated cytoplasmic tails. Unlike the endosomal pH-dependent VSV-LVs, the fusion proteins of paramyxoviruses exhibit reduced off-target transduction in phagocytic cells and minimized CAR-mediated transduction in B cells [18]. Their functionally segregated glycoproteins offer a pivotal advantage for engineering: attachment and fusion are mediated by distinct proteins. NiV utilizes the G protein for attachment and the F protein for fusion. Mutating the G protein abrogates its binding to native receptors ephrin-B2/B3 receptors on endothelial cells and neurons [19–21]. Furthermore, the absence of NiV-based vaccines and low levels of pre-existing neutralizing antibodies in humans confer NiV-pseudotyped LVs with inherent resistance to serum inactivation. [22]. These properties have spurred the development of CD8-[23, 24] and CD3-targted [25] NiV-LVs for in vivo CAR-T generation. However, the preference in NiV glycoproteins for membrane-proximal cell attachment hinders its application for some targets (e.g., CD4) [20]. Therefore, attempts are ongoing to exploit the structural elasticity of glycoproteins from other paramyxoviruses like MV. MV relies on hemagglutinin (H) to bind CD46 and SLAMF1 and fusion protein (F) to mediate the membrane fusion. Truncation of cytoplasmic tails in H and F are also necessary to pseudotyped LVs. Similarly, mutating key residues in H protein can paralyze its native attachment function and reduce the susceptibility of neutralizing antibodies [26–28]. Building on these mechanisms, CD4-targeted MV-LVs have achieved in vivo transduction of up to 60% of CD4 + lymphocytes [27, 29–32].

The envelope glycoproteins of Sindbis virus (SINV), an alphavirus, also serve as a candidate for pseudotyping. The SINV envelope comprises E1 and E2 heterodimers, responsible for membrane fusion and receptor binding, respectively, with E3 acting as a signal peptide for E2. Specific combinations of mutations in these proteins can effectively ablate native tropism while preserving the infectivity of SINV-pseudotyped LVs [33, 34]. However, this approach showed a relative low delivery efficiency.

Following the ablation of inherent viral tropism, identifying an ideal T-cell target becomes critical for achieving precise CAR transduction. CD3, a central component of the T-cell receptor complex, is a commonly targeted antigen across various vector platforms due to its pan-T cell expression [35]. CD7, a marker associated with less-differentiated T-cell states, has also been employed in LVs for engineering T cells and NK cells [36]. CD5, which is constitutively expressed on T cells and certain B-cell subsets , raises concerns for on-tumor transduction in the context of hematological malignancies during in vivo CAR-T therapy [37, 38]. Subset-specific targets such as CD4 and CD8 also represent viable options. Notably, animal studies have demonstrated that CAR-Ts generated in vivo by CD4-targeted MV-LVs exhibited superior anti-tumor activity than those generated by CD8-targeted NiV-LVs. This difference may stem from the greater exhaustion propensity of CD8 + T cells [30].

The choice of binding moieties, such as single-chain variable fragments (scFvs), variable domain of heavy-chain antibodies (VHHs), or designed ankyrin repeat proteins (DARPins), can influence the transduction efficiency [39]. These targeting elements can be expressed independently on the viral envelope, as in VSV-LVs, or fused to receptor-blind glycoproteins in the case of NiV- and MV- LVs (Fig. 3). Studies comparing different moiety constructs for T-cell targeting reveal target- and pseudotype-dependent outcomes. In VSV-LVs, a CD4-targeting DARPin underperformed in transduction efficiency compared to scFv- or VHH-based targeting [15]. In contrast, MV-LVs bearing a CD4-targeting DARPin achieved near-exclusive transduction of CD4 + lymphocytes[29, 30]. NiV-LVs decorated with a CD8-targeting DARPin demonstrated specificity comparable to its scFv counterpart while enabling higher viral titers [40]. Interestingly, another study reported that a CD8-targeting DARPin in NiV-LVs performed similarly to a VHH but significantly better than an scFv [41]. Beyond LVs, AAVs employing a CD8-targeting DARPin have also achieved >90% transduction specificity in murine CD8 + T cells in vivo [42]. Finally, bispecific adaptors offer a flexible targeting strategy for certain platforms. For instance, Sindbis virus (SINV)-pseudotyped LVs can be redirected to T cells using a bispecific anti-CD3/Sindbis E2 tandem antigen-binding fragment in a dose-dependent manner [34].

Fig. 3.

Strategies for viral vectors to achieve in vivo CAR-T generation. The viral vectors, primarily including lentiviral vectors and adeno-associated viral vectors, need to ablate the natural tropism by mutating attachment glycoproteins, and retarget T cells by incorporating T-cell specific moieties

Collectively, LVs enable efficient transduction of both dividing and nondividing cells, facilitate delivery of large genetic payloads, and sustain stable, long-term transgene expression. Nonetheless, their application is hampered by challenges including off-target transduction, immunogenicity, complex manufacturing processes, and the risk of insertional mutagenesis. Consequently, current research efforts focus on overcoming these hurdles to fully leverage the potential of LVs in treating diseases like hematological malignancies, where sustained CAR expression is necessary for effective long-term immune surveillance.

Adeno-associated virus vectors

AAVs have emerged as a versatile tool with unprecedented potential for treating various genetic disease, such as hemophilia, muscular atrophies/dystrophies, inherited forms of blindness [43, 44]. Structurally, AAVs are non-enveloped parvoviruses with an icosahedral capsid surrounding a single-stranded DNA genome flanked by inverted terminal repeats [45]. The genome contains the rep and cap genes: rep encodes non-structural proteins essential for replication and packaging, while cap encodes the structural proteins (VP1, VP2, VP3) that assemble into the capsid and determine the tissue tropism of different AAV serotypes [46]. Despite the broad tissue selectivity offered by 13 natural serotypes and further expanded by recombinant engineering, few AAVs efficiently transduce T lymphocytes [44]. Although intrathymic AAV injection can directly access T cells and has shown effective transduction [47–49], constraining its tissue distribution remains challenging. Therefore, a two-step strategy involving native tropism ablation and T-cell targeting (Fig. 3) is also being explored for in vivo CAR-T generation using AAVs [50].

AAV2 and AAV6 are the most commonly used serotypes. AAV2 displays broad tissue tropism, whereas AAV6 mediates potent transduction of both human and murine hematopoietic stem cells and T cells, and importantly, has lower seroprevalence of neutralizing antibodies than AAV2 [42]. Mutations in key residues of the capsid can disrupt their attachment to native receptors [50–53]. Furthermore, the surface-exposed GH2/GH3 loop of the capsid protein (Fig. 3) is an ideal site for inserting T-cell targeting moieties while preserving infectivity [54, 55]. Insertion of CD8-targeting DARPins here enabled AAV2mut to achieve ~ 40% transduction efficiency in PBMC-humanized mice by 3 days, and similarly allowed AAV6mut to target murine CD8 + T cells with near-absolute specificity in immunocompetent mice [42]. The pioneering study using AAV to generate in vivo CAR-Ts employed the recombinant serotype AAV-DJ as the vector [56]. The wild AAV-DJ was furnished with a third generation anti-CD4 CAR to generate CAR-Ts in vivo for treating CD4 + T-cell leukemia. However, the prominent accumulation of engineered T cells in the liver and off-target transduction beyond CD4 + T cells remain important hurdles.

The transgenes carried by AAV usually enter the nucleus as circular double-stranded episomes and do not integrate into the genome, which means these episomes will lose expression with the divisions of host cells. However, this transient transduction mode can be changed. Nyberg et al. [57] generated an AAV capsid library via VP3 saturation mutagenesis in a chimeric AAV2-rep/AAV6-cap backbone. Through structure-guided evolution, they identified AAV-Ark313, which highly efficiently transduces murine T cells without requiring any targeting moieties. In T cells isolated from Cas9-expressing mice, Ark313 delivered a T cell receptor alpha constant (TRAC)-targeting sgRNA to integrate a CAR cassette with > 40% efficiency and minimal impact on viability in vitro. In addition, Ark313 achieved in vivo gene editing in both tissue-resident and splenic T cells, and proves effective in suppressing tumor growth [58]. Leveraging an AAV vector that co-encapsulated a CAR transgene and a TRAC-directed ARCUS nuclease, a separate study also achieved precise in vivo generation of site-specific CAR-T cells [59].

Compared to LVs, AAVs have a reputation for enhanced safety due to their non-integrating nature. Particularly, the clinical approval of several AAV-based gene therapies has accumulated extensive pharmacokinetic and safety data. While combined with gene-editing technologies, they can also achieve precise genome integration. Nonetheless, AAVs still face challenges such as pre-existing immunity, high immunogenicity, limited packaging capacity, and difficulties in scalable manufacturing. Given these considerations, AAVs are more suitable for treating localized diseases that can be controlled or cured with a single dose.

The preclinical explorations of in vivo CAR-Ts generated with viral vectors are summarized in Table 2.

Table 2.

Overview of viral vectors-based in vivo CAR cell therapies in preclinical studies

Vector	Mutations	Targeting moiety	Location modea	Gene cargo	Dosing and routeb	Strain	Humanization method	Target cells	Reference
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	1hCD8 scFv (Okt8opt)	Tandem	hCD19 (FMC63)- CD28ζ-CAR	2 × 106 TU, i.p.	NSG	1 × 107 activated PBMCs i.p. (D-1)	5 × 105 Raji cells i.p. (D-6)	23
					2 × 106 TU, i.v.; hIL-7, i.v. (D-5, D-1)		hCD34 + cord blood cells	Endogenous B cells
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	hCD8 scFv (Okt8opt)	Tandem	hCD19 (FMC63)- CD28ζ-CAR	2.5 × 1011 particles, i.p.	NSG	5 × 106 activated PBMCs i.p. (D-1)	1 × 105 Nalm6 cells, i.v. (D-6)	24
CD4-LV: MV-LV CD8-LV: NiV-LV	CD4-LVs: Y481A, R533A, S548L, F549S (H protein) CD8-LVs: E501A, W504A, Q530A, E533A (G protein)	hCD4 DARPin8.9	Tandem	hCD19 (FMC63)- CD28ζ-CAR	4 × 1010 particles (2 × 106 TU), i.p.	NSG	1 × 107 activated PBMCs i.p. (D-1)	Endogenous B cells	30
		hCD4 DARPin8.9 hCD8 scFv (Okt8opt)			CD4-LVs: 1.0 × 1011 particles; CD8-LVs: 2.5 × 1011 particles; Mix: 5.0 × 1010 particles CD4-LV + 1.25 × 1011 particles CD8-LV	NSG	5 × 106 activated PBMCs i.v. (D-1)	1 × 105 Nalm6 cells i.v. (D-7)
					CD4-LV: 4 × 1010 particles (2 × 106 TU) i.v.; Mix: 2 × 1010 particles (1 × 106 TU) CD4-LV + 3.5 × 1011 (1.45 × 106 TU) CD8-LV, i.v.; all with hIL-7, i.v. (D-5, D-1)	NSG	hCD34 + hematopoietic stem cells	Endogenous B cells
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	hCD3 scFv (TR66.opt)	Tandem	hCD19 (FMC63)- CD28ζ-CAR	1.3 × 1012 particles, i.p.	NSG	5 × 106 activated PBMCs i.p. (D-1)	Endogenous B cells	25
					2 × 1011 particles i.v.; hIL-7, i.v., (D-4, D-1)		hCD34 + hematopoietic stem cells	Endogenous B cells
SINV-LV	Residues 61–64 deletion (SINV E3), KE159-160AA, SLKQ68-71AAAA (SINV E2)	hCD3 × SINV E2 tandem Fab	Bispecific binder	hCD19-CD28ζ- CAR	2.5 × 107 IUs (5 × 1010 particles) + 5ug tFab, i.v.	NSG	5 × 106 activated PBMCs i.v. (−30 min)	5 × 105 BV-173 cells, i.v. (D-5)	34
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	hCD4 DARPin29.2, hCD8 scFv (Okt8opt)	Tandem	hCD19-CD28ζ- CAR	CD4-LV: 2.4 × 1011 particles, i.v.; CD8-LV: 1.2 or 2 × 1011 particles, i.v.; Mix: 1.6 × 1011 particles CD4-LV + 7.6 × 1010 particles CD8-LV, i.v.; all with hIL-7 i.v. or s.c. (D-4, D-1, D1, D3)	SGM3	hCD34 + cord blood cells	Endogenous B cells	32
				hCD19-CD28ζ- CAR	CD4-LV/CD4-LVsh: 8.7 × 1010 particles;CD8-LV/CD8-LVsh: 1.4 × 1011 particles
Cocal-LV	W/O	hCD3 scFv	Independently expression	hCD19 (FMC63)- 4-1BBζ-CAR, IL2Rγ signaling domain (controlled by Rapamycin) FRB domain	0.4 × 106 TU, 2 × 106 TU, 10 × 106 TU	NSG	hCD34 + hematopoietic stem cells	Endogenous B cells	139
Cocal-LV	W/O	hCD80 + hCD52 + hCD3 scFv multi-domain fusion ligand	Independently expression	hCD19-CAR	High dose 5 × 107 TU i.p.; Low dose 1 × 107 TU, i.p.	NSG (MHC I/II KO)	2 × 107 PBMCs, i.p. (D-1)	2.5 × 105 Nalm6 cells, i.v. (D-4)	35
				NHP/human cross-reactive CD20-CAR	0.63 ~ 3.8 × 109 TU Intranodal injection (inguinal and axillary)	Macaca nemestrina	W/O	Endogenous B cells
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	2mCD8 DARPin MSE10	Tandem	mCD4-CAR- 4-1BBζ-CAR	2 × 106 TU, i.v.; IL-15 and IL-7 i.v. or s.c. (D-4, D-1)	NSG	W/O	1 ~ 2 × 107 splenic lymphoma cells from mAITL mice, i.v. (−3 months)	158
VSV-LV	I182E, T214N, T352A (VSV-G protein)	hCD7 binder	Independently expression	hCD20-4-1BBζ-CAR	Low dose: 8.0 × 106 TU, i.v.; High dose: 2.5 × 107 TU, i.v.	NSG	hCD34 + hematopoietic stem cells	Endogenous B cells	15
					1 × 107 TU, i.v.	Hu-IL-15 NSG	hCD34 + hematopoietic stem cells	Endogenous B cells
					Low dose: 6.0 × 105 TU, i.v.; High dose: 1.0 × 107 TU, i.v.	NSG (MHC I/II KO)	1.0 × 107 activated PBMCs, i.v. (D-1)	1 × 106 Raji cells i.v. (D-6)
					2 × 109 TU, i.v.	Cynomolgus macaques	W/O	Endogenous B cells
NiV-LV	E501A, W504A, Q530A, E533A (G protein)	hCD8 DARPin 53F6	Tandem 75% retargeted G protein: 25% non-targeted G protein	hCD19 (FMC63) -4-1BBζ- CAR	Dose not mentioned i.v.; hIL-7 s.c. (D-4, D-1)	NCG	hCD34 + hematopoietic stem cells	Endogenous B cells	41
AAV-DJ	W/O	W/O	-	hCD4-CD28-4-1BBζ- CAR	1.0 × 1011 particles i.p.	NCG	1.5 × 107 activated PBMCs, i.p. (−3 Weeks)	2.5 × 106 MT-2 cells i.v. (D-4)	56
AAV2	R585A, R588A	hCD8 DARPin 63A4	DARPins are inserting in the GH2/3 loop	GFP	1.7 × 1011 vg, i.v.	NSG	1.0 × 107 activated PBMCs, i.v. (D-1)	W/O	42
					1 × 1012 vg, i.v.± OKT3 i.v.	NSG (MHC I/II KO)	2.0 × 107 inactivated PBMCs, i.v. (D-2) ± hIL-2, i.p. (D-1)	W/O
AAV6	V473D, K531E	mCD8 DARPin MSE10		tdT	2 × 1011 vg, i.v.; anti-mCD3 AND mIL-2 complex	Ai9	-	W/O

(a) The relation between T-cell specific moiety and mutated attachment protein (fusogen); (b) The day of vector infusion is regarded as day 0;

¹. hCD8, the “h” in this case refers to human, and the same as follows. ². mCD8, the “m” in this case refers to mouse, and the same as follows

NiV Nipah virus; LV lentiviral vector; MV measles virus; SINV Sindbis virus; AAV adeno-associated viral vector; W/O without; VSV-G vesicular stomatitis virus glycoprotein; scFv single-chain variable fragment; DARPin designed ankyrin repeat protein; CAR chimeric antigen receptor; FRB FKBP12–rapamycin-binding protein; NHP nonhuman primate; GFP green fluorescent protein; tdT tdTomato; i.v. intravenous injection; i.p. intraperitoneal injection; s.c. subcutaneous injection; vg vector genome; D day; NSG NOD.Cg.PrkdcscidIL2rgtmWjl/SzJ; NCG NOD/ShiLtJGpt-Prkdcem26Cd52Il2rgem26Cd22/Gpt; MHC I/II KO major histocompatibility complex class I/II knockout; Ai 9 B6.Cg-Gt(ROSA)26Sor tm9(CAG-tdTomato)Hze/J; PBMC peripheral blood mononuclear cell

Non-viral delivery vectors

Polymer NPs

Polymer NPs, including cationic polymer, liposome, and peptide-based NPs, as delivery vehicles represent a highly promising platform for enabling in vivo generation of CAR-Ts, an emerging frontier in immunotherapy.

Cationic polymers-based NPs

Cationic polymers are key non-viral carriers carrying positive charges to deliver acid drug, including poly(β-amino ester) (PBAE), polyethyleneimine (PEI), and etc [60, 61]. They bind tightly to negatively charged nucleic acids through electrostatic interaction and condense into nanocomplexes [60, 62]. The unique “proton sponge” effect then facilitates endosomal escape [63, 64].

PBAEs, containing protonatable tertiary amines and biodegradable ester linkages, represent a well-balanced design combining strong nucleic acid binding with favorable degradability [65, 66]. A pioneering study from the Fred Hutchinson Cancer Center established the proof-of-concept for in vivo CAR-T generation. Using PBAE-based DNA nanoparticles to deliver a CD19 CAR transposon system, this approach achieved effective leukemia clearance in animal models [67]. Employing a similar strategy, PBAE-447 NPs functionalized with PGA-modified anti-CD8 antibodies achieved transient in vivo T-cell reprogramming by loading a CAR or TCR mRNA [68]. In recent years, CAR technology has expanded to engineering macrophage (MΦ) [69]. Gao et al. utilized the PBAE polymer C32-122-based NPs to deliver ErbB2 CAR plasmids into tumor-associated macrophages, reprogramming M2-like cells into M1-like antitumor phenotypes [70].

PEI, a well-known cationic polymer with a high amine density [71], is often used in vivo through copolymerization or functional modification to enhance delivery efficiency. By targeting CD206, Kang et al. employed mannose-functionalized PEI (MPEI) to condense piggyBac transposon plasmid DNA (pDNA) encoding both a CAR and IFN-γ, forming a targeted delivery system [72]. The CAR-IFN-γ transgene also facilitated in situ reprogramming of CAR-MΦs into a proinflammatory M1 phenotype. In a subcutaneous murine model of anaplastic lymphoma kinase positive neuroblastoma, MPEI effectively inhibited tumor progression [72].

Collectively, polymer-based nanoparticles constitute a versatile and tunable non-viral platform for in vivo CAR generation, enabling targeted immune cell reprogramming (Fig. 4). Their modular architectures allow precise regulation of delivery efficiency and cellular specificity. Nonetheless, challenges such as cytotoxicity, limited stability, and suboptimal transfection efficiency continue to hinder clinical translation. Advancing this field will require a deeper integration of rational chemical design with immunological principles to realize safe, precise, and clinically viable in situ CAR engineering.

Fig. 4.

Schematic illustration of cationic polymer and liposome-based NPs for in vivo CAR therapies. (A) A classic nanoparticle delivery carrier based on engineered cationic polymers requires four essential components, including the cationic polymer, targeting ligand, auxiliary element, and the delivered CAR gene cargo. (B) The liposomalized nanoporter was produced using a coextrusion method. The liposomalized nanoporter prepared could induced in situ generation of CAR-MΦ in the TME of the pancreatic cancer after intravenous injection, and CAR-MΦ could mediate tumor killing

Liposome-based NPs

Liposomes are artificial vesicles consisting of a phospholipid bilayer [73, 74]. Liu et al. synthesized positively charged mannosylated chitosan oligosaccharide-arginine for MΦ-targeted gene delivery [75]. Using a thin-film hydration-ultrasonication method, they prepared liposome-encapsulated nanotransporters to co-deliver a CD47 inhibitor and transposon pDNA encoding a CAR targeting mucin 1, enabling the in situ genetic engineering of tumor-associated macrophages into CAR-M1Φ cells within pancreatic adenocarcinoma lesion [75].

Peptide-based NPs

Peptide-based NPs (PEP-NPs) have demonstrated unique advantages, primarily due to their programmable molecular structures and excellent biocompatibility. Two seminal studies have demonstrated the feasibility of in vivo CAR therapies via these NPs.

Aanastra Inc. has developed PEP-NP for delivering anti-CD19 CAR mRNA to T cells in vivo, with 15 ~ 18% GFP-positive T cells detected in both blood and spleen following a single injection in mouse models [76]. Regarding this aspect of CAR-MΦ, Li and co-workers pioneered a study [77]. By designing peptide-stearic acid (SA) monomers containing both MΦ-targeting sequences and nuclear localization signals peptide, amphiphilic peptide-SA monomers were successfully synthesized. These monomers could self-assemble and encapsulate pDNA containing anti-SasA CAR and cysteinyl aspartate-specific protease 11 (CASP11) short hairpin RNA (shRNA), forming PNPs [77]. When applied as a nanoscale coating anchored onto titanium implants, the PNPs locally induced the generation of super CAR-MΦs at the bone-implant interface. These engineered CAR-MΦs specifically engulfed SasA + staphylococcus aureus (S. aureus), while the shRNA-mediated knockdown of CASP11 significantly enhanced the phagolysosomal bactericidal activity of the SasA-specific CAR-MΦs. The study demonstrated that these super CAR-MΦs effectively eradicated S. aureus in prosthetic joint infection models [77].

LNPs

LNPs are a type of non-vacuolated nanostructured particles formed through self-assembly, and represent one of the most advanced and clinically successful non-viral delivery platforms in the field of nucleic acid therapeutics, with particularly promising applications in mRNA vaccines and gene therapy [78, 79]. Recent years have witnessed LNP-mediated in vivo generation of CAR-engineered immune cells for the treatment of neoplastic diseases and non-oncological indications have emerged in an explosive manner [80].

Targeted LNPs

Targeted LNPs (tLNPs) refer to lipid-based nanocarriers engineered through surface functionalization to specifically bind cell or tissue biomarkers, which aims to reduce the inherent hepatic tropism [81]. Malignancies are the main focus of LNP-mediated in vivo CAR-immune cell generation. IL-6-low CAR-T cells, generated in vivo by CD3-targeted LNPs delivering a transposon-based pDNA encoding CD19 CAR and IL6 shRNA, effectively eradicated leukemia while mitigating CRS [82]. Another team developed anti-CD3 tLNPs loaded with mRNA encoding a tyrosinase-related protein 1 (TRP1) CAR to generate CAR-Ts, specifically recognized and eliminated TRP1 + B16F10 cells [83]. Most recently, Capstan Therapeutics reported in Science that proprietary tLNPs using a novel ionizable lipid L829 and conjugated with anti-CD8 antibodies successfully delivered CAR mRNA to CD8 + T cells. In humanized mice and primates, in vivo CAR-T cells showed strong efficacy and safety for cancer and autoimmune diseases [84]. In cynomolgus monkeys, CD8-L829-tLNP-CD20 mediated robust depletion of CD20 + memory B cells and reconstituted the B-cell compartment primarily with naïve cells, demonstrating a state of immune reset [84]. These results highlight a promising in vivo CAR-T strategy achieving tumor clearance and durable immune reprogramming.

Transposon systems enable stable expression but risk random integration, while mRNA offers transient expression requiring repeat dosing. To address these limits, nucleic acids or gene-editing tools enabling site-specific integration or prolonged expression are needed. At ASGCT2025, several biotech firms showcased tLNP-based delivery of advanced gene-editing tools for in situ CAR-T therapy (Table 3).

Table 3.

Overview of non-viral vectors-based in vivo CAR cell therapies in preclinical and clinical studies

Delivery vectors	Target moiety	Delivery payloads	Cell type	Dosing route			Target		Indication	Reference
Polymer NPs
PBAE-447 NPs	1mCD3ε f(ab’)₂	iPB7-mCD19-4-1BBζ-CAR pDNA	CAR-T	i.v.			mCD19		B cell leukemia	9
PBAE-447 NPs	mCD3ε f(ab’)₂	mCD19-28ζ-CAR mRNA	CAR-T	i.v.			mCD19		B cell leukemia	60
	2hCD8 f(ab’)₂	hCD19-28ζ-CAR mRNA	CAR-T	i.v.			hCD19		B cell lymphoma
	mCD8 f(ab’)₂	h2ROR1-4-1BBζ-CAR mRNA	CAR-T	i.v.			hROR1		Prostate carcinoma
PBAE-C32-122 NPs	RP-182 peptide	piggyBac-mErbB2-CD3ζ CAR pDNA	CAR-MΦ	i.t.			mErbB2		Brainstem glioma	62
PEI NPs	Man	piggyBac-mALK-CD28ζ-IFN-γ-CAR pDNA	CAR-MΦ	i.v.			mALK		Neuroblastoma	66
Poly(aspartamide) polymer-based NPs	Unknown	mCD19 CAR pDNA	CAR-T	i.v.			mCD19		Solid tumor	ASGCT 2025 #699
MCA-liposome	Man	piggyBac-mMUC1-FcRγ1ζ-CAR pDNA	CAR-MΦ	i.v.			mMUC1		Pancreatic adenocarcinoma	69
PEP-NPs	Unknown	hCD19-CAR mRNA	CAR-T	i.v.			hCD19		B cell leukemia/ Autoimmune disease	ASGCT 2025 #1283
pPNP	MTS peptide	piggyBac-SasA-CD3ζ-CAR + CASP11 shRNA pDNA	CAR-MΦ	i.v.			SasA		S. aureus in PJI	70
LNPs
tLNPs
aCD3-LNP/CAR19 + shIL6	CD3 Ab	iPB7-hCD19-4-1BBζ-CAR + IL6 shRNA pDNA	CAR-T		i.v.			hCD19	B cell lymphoma	82
Novel tLNPs	Unknown	Transposon pDNA encoding hCD19 CAR + transposase mRNA	CAR-T		i.v.			hCD19	B cell leukemia	ASGCT 2025 #213
aCD3-LNP	CD3 Ab	mTRP1-CD28ζ-CAR + IL-7 mRNA	CAR-T		i.v.			mTRP-1	Melanoma	83
L829-tLNP	CD5/8 Ab	hCD19/CD20 CD28ζ-CAR mRNA	CAR-T		i.v.			hCD19/CD20	B cell leukemia, Autoimmune disease	84
aCD5-LNP-mPMSA/mCAR19	CD5 Ab	mCD19-CD28ζ-CAR + mPSMA mRNA	CAR-T		i.v.			mCD19	B cell lymphoma	162
EnC-LNPs	Unknown	mCD19/CD20 CAR mRNA	CAR-T		i.v.			mCD19/CD20	SLE	ASGCT 2025 #794
tLNPs (Aera Therapeutics)	Unknown	hCD19 CAR mRNA	CAR-T		i.v.			hCD19	Autoimmune disease	ASGCT 2025 #770
tLNPs (Kernal Biologics)	Unknown	CAR mRNA	CAR-T		i.v.			Unknown	Unknown	ASGCT 2025 #1753
tLNPs (Tessera Therapeutics)	Unknown	hBCMA/CD19/CD20-RNA Gene Writer	CAR-T			i.v.		hBCMA/CD19/CD20	Unknown	ASGCT 2025 #153
aCD3-LNP (Stylus Medicine)	hCD3	hCD19 CAR DNA + Serine recombinase mRNA	CAR-T			i.v.		hCD19	B cell leukemia	ASGCT 2025 #797
tLNPs (Sail Biomedicines)	Unknown	hCD19 CAR circular endless RNA	CAR-T			i.v.		hCD19	autoimmune disease	ASGCT 2025 #774
tLNPs (RiboX Therapeutics)	Unknown	hCD19 CAR circRNA	CAR-T			i.v.		hCD19	B cell leukemia/ autoimmune disease	ASGCT 2025 #1762
Mal-LNP	mCD3/5/7 Fab	mCD19-CD28ζ-CAR mRNA	CAR-T			i.v.		mCD19	None	86
tLNPs (Sanofi Biotech)	CD8 VHH	CD19 CAR mRNA	CAR-T			i.v.		CD19	B cell leukemia/ autoimmune disease	ASGCT 2025 #260
ctLNPs (Nitto Denko Corporation)	CD8 VHH	hCD19 CAR mRNA	CAR-T			i.v.		CD19	B cell leukemia	ASGCT 2025 #665
tLNP-CD7/CD3	CD7 VHH + CD3 scFv	hCD19-4-1BBζ-CAR mcDNA + SB100x transposase mRNA	CAR-T			i.v.		hCD19	B cell leukemia	87
		hCD19/CD22-4-1BBζ-CAR mcDNA + SB100x transposase mRNA	CAR-T			i.v.		hCD19/CD22	B cell leukemia
CLAMP-tLNPs	Unknown	hCD19 CAR mRNA	CAR-T			i.v.		CD19	B cell leukemia	ASGCT 2025 #1282
Grabs-tLNPs	CD3/5 Grab Ab	hCD19 CAR mRNA	CAR-T			i.v.		CD19	B cell leukemia	ASGCT 2025 #796
aCD5-LNP	CD5 Ab	mFAP-CD28ζ-CAR mRNA	CAR-T			i.v.		mFAP	Cardiac injury	91
MPEP/S2P-LNP	MPEP/S2P	mFAP-CD3ζ-CAR + Lgmn mRNA	CAR-MΦ			i.v.		mFAP	Cardiac fibrosis after MI	92
CD5/βLNP	CD5 Ab	mFAP-4-1BBζ-CAR mRNA	CAR-T			i.v.		mFAP	Pulmonary fibrosis	93
aCD5-LNP	CD5 Ab	mFAP-CD28ζ-CAR mRNA	CAR-T			i.v.		mFAP	MASH	94
MLNP	Man	mFAP-CD3ζ-CAR mRNA	CAR-MΦ			i.v.		mFAP	PDAC	98
None	None	mFAP-CD3ζ-CAR-△TGFβRII+ mGPC3-CD3ζ-CAR-Super IL2 mRNA	CAR-MΦ			i.v.		mFAP/mGPC3	HCC	99
CRV/LNP-RNAs	CRV peptide	SasA-CAR mRNA + CASP11 siRNA	CAR-MΦ			i.v.		SasA	MRSA in sepsis	100
LNP (Myeloid Therapeutics)	Unknown	hHER2-CAR-CD89/NKp44 mRNA or hgp75-CAR CD89/NKp44 mRNA	CAR-monocytes			i.v.		hHER2/gp75	Colon cancer or melanoma	ASGCT 2025 #1727
Spleen SORT LNPs	18:1 PA	mCD19-4-1BB/CD28ζ-CAR mRNA	CAR-T			i.v.		mCD19	B cell lymphoma	104
SORT-LNP-circRNACAR	18:1 PA	mHER2-4-1BBζ-CAR circRNA	CAR-T/NK/MΦ			i.v./i.t.		mHER2	HER2 + solid tumor	105
CAMP-LNP	PL 40	muPAR-CD28ζ-CAR circRNA	CAR-T			i.v.		muPAR	Liver fibrosis and RA	109
CAR&Siglec-GΔITIMs LNP	PPZ-A10	mGPC3-CD3ζ-CAR mRNA; Siglec-GΔITIMs mRNA	CAR-MΦ			i.v.		mGPC3	HCC	110
PSβ-LNPs	Phosphatidylserine	mHER2-CD3ζ-TLR4-CAR mRNA	CAR-MΦ			i.p.		mHER2	Peritoneal metastatic carcinoma	112
oLNPs	Oxidized lipids	hCD19-CAR mRNA	CAR-monocytes			i.v.		hCD19	None	113
CD3-PMPP	mCD3 Ab	mCD123-4-1BBζ-CAR Cas9 pDNA	CAR-T			i.v.		mCD123	T cell lymphoma	120
mRNA-LNP-Ery	None	hHER2-CAR mRNA	CAR-myeloid cells			i.v.		hHER2	Colon cancer	ASGCT 2025 #1284
αCD3-AEV	mCD3 Ab	AEV6-mCD19-CAR	CAR-T			i.v.		mCD19	B cell lymphoma	124
eMSC/pCAR	None	piggyBac-mCD133-CD3ζ-TIR-CAR	CAR-MΦ			i.v.		mCD133	Glioblastoma	127
Virus-mimetic fusogenic NVs	mCD3 scFv	mCD19-CD28ζ-CAR protein	CAR-T			i.v.		mCD19	B cell lymphoma	128
T-FVLP	gp160	hCD19-CD28ζ-CAR mRNA	CAR-T			i.v.		hCD19	B cell lymphoma	39
MASTER	None	hCD19-28ζ-CAR γ-retrovirus	CAR-T			s.c.		hCD19	B cell lymphoma	132
CD3-PPP/pCAR@α-	hCD3ε F(ab’)2	hFR-CD28-CD137ζ-CAR pDNA	CAR-T			i.t.		hFR	Ovarian cancer	134
Nanoporter hydrogel superstructure	CA-dextran	iPB7-mCD68p-mCD133 -CD3ζ-CAR pDNA	CAR-MΦ			Intracavity injection		mCD133	Glioma	137
CSR-mLNP + CAR-mLNP/IL-2@gel	Man	IL-2R–TLR4 CSR + CA9-CD3ζ CAR mRNA	CAR-MΦ			i.t.		CA9	Renal cell carcinoma	138

¹. mCD3, the “m” in this case refers to mouse, and the same as follows. ². hCD8, the “h” in this case refers to human, and the same as follows

NP nanoparticle; PBAE poly(β-amino ester)s; PEI polyethyleneimine; MCA mannosylated chitosan oligosaccharide-arginine; PEP peptide-based; LNP lipid nanoparticle; tLNP targeted lipid nanoparticle; PSMA prostate-specific membrane antigen; Mal maleimide-engineered; ctLNP cell-targeted lipid nanoparticle; MLNP mRNA-LNP; SORT selective organ-targeting; circRNA circular RNA; CAMP cardiolipin-mimic phosphoamide lipids; Siglec-GΔITIMs sialic acid-binding ig-type lectin-g δ immunoreceptor tyrosine-based inhibitory motifs; PSβ phosphatidylserine/β-sitosterol; Ery erythrocyte; AEV artificial enveloped adeno-associated virus; eMSC enucleated mesenchymal stem cells; T-FVLP T cell-specific fusogenic virus-like particle; MASTER multifunctional alginate scaffold for T cell engineering and release; PPP poly(ethyleneglycol)-b-poly(-caprolactone)-b-poly(ethyleneimine); CSR chimeric signaling receptor; m anti-mouse; h anti-human; Man mannose; MTS macrophage-targeting sequences; Ab antibody; VHH variable domain of heavy chain of heavy chain antibodies; scFv single-chain variable fragment; 18:1 PA 1,2-dioleoyl-sn-glycero-3-phosphate; PL 40 pegylated lipid 40; gp160 glycoprotein 160; CA-dextran cyclodextran-dextran conjugate; pDNA plasmid DNA; ROR1 receptor tyrosine kinase-like orphan receptor 1; ErbB2 erb-b2 receptor tyrosine kinase 2; ALK anaplastic lymphoma kinase; MUC1 mucin 1; SasA staphylococcus aureus surface protein A; CASP11 cysteinyl aspartate-specific protease 11; shRNA short hairpin RNA; TRP1 tyrosinase-related protein 1; FAP fibroblast activation protein; GPC3 glypican-3; HER2 human epidermal growth factor receptor 2; NKp44 natural killer p44-activating receptor; hgp75 human glycoprotein 75; TLR4 toll-like receptor 4; CA carbonic anhydrase; TIR transferrin receptor; FR folate receptor; CAR chimeric antigen receptor; i.v. intravenous injection; i.t. intratumoral injection; i.p. intraperitoneal injection; s.c. subcutaneous injection; PJI prosthetic joint infection; SLE systemic lupus erythematosus; MI myocardial infarction; MASH metabolic dysfunction-associated steatohepatitis; PDAC pancreatic ductal adenocarcinoma; HCC hepatocellular carcinoma; MRSA methicillin-resistant S. aureus; ASGCT American Society for Gene and Cell Therapy

While antibody-modified tLNPs are widely used for T-cell targeting, full-length IgG presents challenges such as big size, low efficiency, high immunogenicity and serum clearance [76, 77]. To address this, smaller fragments or binders (e.g., Fab, scFv, VHH, DARpins) are employed to enhance targeting precision and safety [85–87] (Fig. 5). For instance, three companies showcased VHH-based tLNPs for in vivo CAR-T generation. Acuitas Therapeutics also reported a CD8-targeted DARPin-LNP system that effectively bypasses hepatic tropism for efficient in vivo T-cell engineering [88]. Emerging technologies are replacing random covalent conjugation for attaching targeting moieties to tLNPs. One approach employs optimized linker and site-specific antibody residues (e.g., GRIT’s CLAMP technology) to improve efficiency [89]. Another biomimetic strategy utilizes “Grab antibodies”—ApoA1-fused ligands that self-assemble onto LNPs via lipid affinity, bypassing chemical conjugation altogether. Both methods have demonstrated success in generating in vivo CAR-T cells, facilitating scalable manufacturing [90].

Fig. 5.

Schematic illustration of tLNPs and SORT-LNPs mediating in vivo CAR cell therapies. The redirection strategies of tLNPs targeting specific immune cells and the process of encapsulating CAR gene cargo in tLNPs, which mediates the in vivo reprogramming of circulating T lymphocytes into CAR-T cells and in situ generation of CAR-MΦs in the TME, respectively

tLNP-based in vivo CAR therapies are emerging for fibrosis by targeting fibroblast activation protein (FAP). A University of Pennsylvania team developed CD5 tLNPs delivering FAP CAR mRNA directly to T cells, reversing cardiac fibrosis and restoring function in mice [91]. Another group created dual-targeting LNPs modified with MΦ-targeting S2P and MI-targeting MPEP peptides, delivering mRNA for Legumain and FAP CAR to reprogram MΦs in situ. The engineered CAR-MΦ showed enhanced efferocytosis and FAP specificity, reducing infarct size, attenuating fibrosis, and improving cardiac function in myocardial infarction mice [92]. Yan et al. optimized CD5 tLNPs by replacing cholesterol with β-sitosterol, boosting mRNA expression; in bleomycin-induced mice, FAP CAR-Ts cleared fibroblasts, restored ECM mechanics, promoted alveolar repair, and prolonged survival [93]. Yashaswini et al. similarly used CD5 tLNPs delivering FAP CAR mRNA to combat hepatic fibrosis in metabolic dysfunction-associated steatohepatitis [94].

Beyond inducing organ damage, TME fibrosis from hyperactivated fibroblasts creates a barrier blocking drug delivery and immune cell infiltration [95–97]. Wang et al. engineered mannose-modified mRNA-LNP (MLNP) to directly generate FAP-targeting CAR-MΦs in vivo [98]. These MLNP-induced CAR-MΦs efficiently cleared activated cancer-associated fibroblasts in pancreatic cancer, consequently reducing tumor fibrotic barriers and potentiating the therapeutic efficacy of both the chemotherapeutic agent gemcitabine and anti-PD-1 immune checkpoint inhibitors [98]. Another research team employed in vivo CAR-MΦ to disrupt the CAF-constructed physical barrier, unleashing immune-mediated antitumor capacity. Zhang et al. used mRNA-LNPs to generate tetravalent CAR-MΦs combining GPC3-CAR-Super IL-2 and FAP-CAR-ΔTGFβRII. The ΔTGFβRII blocked TGF-β signaling, driving M1 polarization, while Glypican-3 (GPC3) CAR-MΦs secreted Super IL-2 to boost CD8 + T-cell activation. This approach dismantled CAF barriers, enhanced CD8 + infiltration, and achieved complete regression of orthotopic hepatocellular carcinoma (HCC) [99].

In addition, in situ CAR therapies also appears to be a promising strategy for combating infectious diseases. Tang et al. developed YHS-12, an ionizable lipid from fumarate, to create CRV/LNP-RNAs co-delivering CASP11 siRNA and SasA CAR mRNA. Conjugated with a MΦ-targeting CRV peptide, these tLNPs selectively reprogrammed MΦs after systemic injection, inducing SasA-CAR expression and CASP11 knockdown [100]. The engineered CAR-MΦs effectively engulfed methicillin-resistant S. aureus (MRSA) and overcame immune evasion, achieving efficient bacterial clearance.

SORT-LNP

A novel approach known as SORT molecules can reprogram LNPs biodistribution, which help diminish the strong liver tropism and increase the bioavailability [101–103]. SORT-LNPs enable organ-specific targeting by introducing a fifth functional component, SORT molecules, into the standard four-part LNP formulation, dynamically modulating the adsorbed biomolecular corona on the surface. Their core mechanism hinges on the charge properties of splenic SORT molecules directly dictating organotropism: anionic SORT lipids reroute LNPs to the spleen by enriching ApoE adsorption [103, 104], thus enabling precise extrahepatic mRNA delivery.

Álvarez-Benedicto and colleagues employed a 10% 18:1 PA five-component splenic SORT formulation to deliver CAR mRNA via SORT-mRNA-LNPs [104]. Upon intravenous administration, these NPs efficiently transfected splenic T cells in vivo, enabling the generation of functional in situ CAR-Ts without active targeting ligands [104]. Wang et al. combined circRNA-based in situ CAR-T with vaccines to generate SORT-LNP-circRNA^CAR, demonstrating potent HER2 + CT26 tumor suppression and enhanced survival in colorectal cancer models [105].

In recent years, CAR-based therapies targeting the selective elimination of senescent cells have emerged as a rapidly growing area of research for the treatment of age-related diseases [106–108]. Zhang et al. designed SORT-LNPs containing cardiolipin-mimic phosphoamide lipids (CAMP), which increase rigidity and induce phase separation to promote preferential uptake by T cells, significantly enhancing the splenic transfection efficiency of CAR-T in vivo [109]. The study utilized CAMP-containing LNPs encapsulating anti-uPAR CAR circRNA, demonstrating that in situ generated CAR-Ts effectively cleared inflam-aging cells in liver fibrosis and RA models [109].

Extending SORT-LNPs-based in vivo CAR-T to other CAR-engineered immune cell types shows promise. Yang et al. designed PPZ-A10-formulated SORT-LNPs to co-deliver GPC3-specific CAR mRNA and Siglec-GΔITIMs mRNA encoding an ITIM-deficient Siglec-G variant into liver MΦs [110]. Peritoneal metastases in cancer patients generate ascites rich in immune cells, with MΦs constituting ~ 45% of this population [111], presenting an opportunity for in situ CAR-MΦ therapy. MΦ-targeting phosphatidylserine/β-sitosterol LNPs (PSβ-LNPs), incorporating phosphatidylserine (PS) to enhance MΦ tropism, further enabled CAR-MΦ programming in solid tumor models, synergizing with anti-PD-1/L1 therapy [112]. Pioneering work by the Mitchell Lab has established in situ CAR-monocytes using a novel oxidized LNP (C14-O2-based) identified via high-throughput screening. This platform achieved specific CAR mRNA delivery to monocytes (~ 2% transduction), leading to a 45% reduction in B cells without off-target expression in T or B cells [113].

LNPs have become the most used non-viral vectors for in vivo CAR-T generation, due to their self-assembly and scalability. Advances in tLNPs and SORT-LNPs have successfully expanded tissue tropism beyond the liver, allowing precise reprogramming of T cells and macrophages for cancer, fibrosis, and infectious diseases. However, they still face notable challenges, including limited delivery efficiency to extrahepatic tissues, transient expression requiring repeated administration, and potential immunogenicity of targeting ligands. Manufacturing complexity and the need for consistent conjugation further constrain clinical translation. Diseases that need systemic, flexible and controllable treatment are good indications for LNP-based in vivo CAR-T therapies, such as autoimmune disorders.

Bioinspired delivery vehicles

Bioinspired delivery vehicles are drug delivery platforms designed by mimicking natural biological systems such as cell membranes, or exosomes [114–116]. They leverage biological camouflage to prolong circulation time, intelligent targeting to enhance lesion accumulation [117, 118], thereby improving drug precision and efficacy in tumor therapy and gene delivery.

Platelet-mimicking NPs cloak loaded drug with platelet membrane components [119], replicating platelets’ the homing behavior and selectively adhering to damaged endothelium or pathological cells. Building on this strategy, Wu et al. [120] constructed a CD123-specific CAR pDNA/PBAE-447 nanocomplex coated with platelet membranes and conjugated anti-CD3ε f(ab’)₂ fragments, enabling T-cell targeting and generation of functional CAR-Ts within the TME in an EL-4 lymphoma model.

Erythrocyte-inspired delivery system harnesses the innate long-circulating capability (120-day span) and spleen homing ability of red blood cells (RBCs), enabling ultra-prolonged circulation cycles and precise lymphocyte editing [121–123]. A group from Westlake University developed a “plug-and-play” erythrocyte-mediated mRNA delivery system (mRNA-LNP-Ery) that covalently links mRNA-LNPs to RBCs, significantly enhancing spleen delivery to immune cells—particularly CD11b + myeloid cells (>90% expression) via efficient phagocytosis. Crucially, mRNA-LNP-Ery enables in vivo generation of anti-HER2 CAR-myeloid cells, which directly kill cancer cells, reprogram the TME, and enhance lymphocyte infiltration (ASGCT2025 #1284). Recently, Pan et al. [124] designed a novel biomimetic enveloped AAV (AEV) vector by encapsulating AAV with RBC membranes, thereby deterring the antibodies-mediated neutralization and immune cell-mediated phagocytosis. Furthermore, by modifying the AEV surface with anti-CD3 antibodies, they effectively generated in vivo CAR-Ts in a mouse B-cell lymphoma model.

Enucleated mesenchymal stem cells (eMSCs) retained this inherent lesion-tropic migratory potential of MSCs following nuclear removal [125, 126]. Given their inability to sustain long-term viability post-enucleation, Zhou et al. revealed that eMSCs initiated a programmed apoptotic cascade approximately 60 h post-preparation [127], culminating in phosphatidylserine externalization and surface exposure. This process generated specific “engulfment signals”, thereby facilitating efficient recognition and phagocytosis by MΦs. This team further engineered eMSCs to deliver pDNA encoding a CD133-specific CAR, specifically reprogramming glioma-associated microglia/MΦs (GAMs) into CAR-MΦs within the tumor, effectively suppressing glioblastoma (GBM) progression compared to LNP-based delivery [127].

Composed of viral structural proteins, Virus-like particles (VLPs) retain their parent viruses’ topological and antigenic features, serving as promising alternatives to conventional nanocarriers and LVs due to their biocompatibility, structural homogeneity, stability, scalability, and versatility. To overcome native tropism, Gui Z et al. [128] developed virus-mimetic fusogenic nanovesicles (FuNVs) co-expressing both T-cell-specific fusogen (p14-based T-cell fusogen/p14TF or mutant measles virus-based T-cell fusogen MVTF) and anti-CD19 CAR on their surface. Upon binding to T cells, FuNVs transferred the CAR protein to the T-cell surface through membrane fusion and generated ∼2% CAR + cells among CD3 + T cells in blood/spleen within 24 h, effectively suppressing tumor growth. The same team also devised a T cell-specific fusogenic virus-like particle (T-FVLP) that mimics human immunodeficiency virus to deliver CAR mRNA,, producing ~ 3% CAR + T cells in humanized mice. For stable genomic integration, Hamilton JR et al. [39] designed a VSV-G-mutated, LV-derived VLP decorated with anti-CD3/CD4/CD28 scFvs. These VLPs packaged Cas9 RNPs and an anti-CD19 CAR, enabling successful production TRAC locus-integrated in vivo CAR-Ts in PBMC-humanized mice, albeit at lower efficiency than modified LVs. However, in contrast to LVs, this VLP platform failed to efficiently deplete the B-cell compartment—a functional limitation reflecting its reduced potency in vivo.

Bioinspired delivery vehicles possess ideal properties—inherent biocompatibility and lesion-specific tropism—that give them a distinct edge over synthetic nanocarriers. Consequently, they represent an excellent platform for in vivo CAR cell therapy to treat diseases featuring complex microenvironments or formidable biological barriers. Nevertheless, their path to the clinic is fraught with obstacles, from complex preparation and batch variability to limited scalability and potential immunogenicity of biological components.

Macroencapsulation devices

Macroencapsulation devices (e.g., membrane-controlled release systems, 3D scaffolds, microneedles, hydrogels) provide a supportive microenvironment for cells, with applications in diabetes, cardiovascular disease, and tumor immunotherapy [129–131]. For in vivo CAR-T generation, a macroencapsulation system typically comprises nanosized vectors loaded with CAR gene payloads combined with a macroscopic-scale substrate carrier that enables encapsulation or attachment.

Agarwalla et al. developed an alginate-based 3D scaffold for in situ CAR engineering, termed the Multifunctional Alginate Scaffold for T cell Engineering and Release (MASTER) [132]. T cells can be activated by cyclooctyne-conjugated anti-CD3, anti-CD28 antibodies and IL-2, subsequently transfected with retrovirus encoding a CD19-specific CAR within the scaffold. Following subcutaneous implantation, MASTER mediates the release of functional CAR-Ts in mice within 24 h and induce suppressed distant tumor growth [132].

Hydrogels are hydrophilic polymer-based carriers with 3D network structures that mimic the ECM, exhibiting exceptional biocompatibility and programmable physicochemical properties [133]. They represent an alternative strategy for yielding locoregionally CAR therapies. A research group engineered an injectable supramolecular hydrogel system (CD3-PPP/pCAR@α-CD) that enabled sustained in situ reprogramming of CAR-Ts through peritumoral administration in solid tumors [134]. The core architecture comprised folate receptor-alpha-CAR plasmids featuring T cell-specific promoters electrostatically complexed with cationic mPEG-PCL-PEI (PPP) copolymers to form nanocomplexes, whose surfaces were functionalized with T cell-targeting anti-CD3ε F(ab’)₂ fragments. Subsequent host-guest interactions between α-cyclodextrin (α-CD) and PPP driven spontaneous supramolecular assembly into a hydrogel network. Upon implantation, this system provides controlled release of pCAR-loaded nanocomplexes within the TME, achieving concurrent in situ genetic editing and localized expansion of T cells [134].

The abundant distribution of MΦ within the tumor provides promising sources for hydrogel-mediated locoregional delivery [135, 136]. For CAR- MΦ generation, a team prepared an injectable brain ECM-mimetic hydrogel based on self-assembling Fmoc-modified peptides, functioning as a flowable void filler for post-resection GBM cavities [137]. This system co-delivers MΦ-targeted gene-editing vectors (pCAR-NPs) and CD47 antibodies to locally reprogram perilesional MΦs into CAR-MΦ that eliminate glioma stem cells [137]. They also constructed an injectable hydrogel via dynamic crosslinking between oxidized hyaluronic acid and hydrazide-modified gelatin [138]. This platform simultaneously delivered LNPs encapsulating circRNA encoding a carbonic anhydrase IX (CA9) CAR and a chimeric IL-2 signaling receptor (CSR), along with free IL-2 cytokine, generating CSR-functionalized CAR-MΦs. The IL-2 activated synthetic CSR-mediated JAK-STAT signaling, promoting antitumor polarization. In renal carcinoma models, this delivery system effectively reversed the immunosuppressive TME and induces significant tumor regression.

The modular architecture of macroencapsulation devices enables precise spatial control, high local agent concentration, and minimized off-target effects, rendering them particularly suitable for solid tumor treatment and post-surgical applications. However, limitations remain, including limited scalability, challenges in implantation or retrieval, potential foreign-body responses, and restricted diffusion of nutrients or immune cells within dense matrices. Moreover, the complexity of combining biological and material components poses translational hurdles for clinical manufacturing. Despite these obstacles, macroencapsulation systems represent an innovative bridge between biomaterials and therapies, enabling next-generation localized CAR therapies with enhanced safety and efficacy.

In vivo CAR-T: current clinical-stage landscape

Recent advancements in techniques give new impetus to the clinical translation of in vivo CAR-T. Several in vivo CAR-T products based on LVs and mRNA-LNPs have entered clinical practice (Fig. 6), with over 30 clinical trials registered (Table 4).

Fig. 6.

The clinical landscape of in vivo CAR-T based on viral and non-viral systems. In vivo CAR-Ts generated by viral vectors are designed for hematological malignancies and autoimmune diseases, while their counterparts generated by non-viral vectors have a wider therapeutic profile, including autoimmune diseases, solid tumors and organ fibrosis

Table 4.

In vivo CAR cell therapies advancing to clinical studies

Developer	Country	Products	Delivery vector	Fusogen	T cell-specific antigen	Costimulation molecule	Targeting cell	CAR construct	Indication	ClinicalTrials.gov ID
Interius BioTherapeutics	America	INT2104	LV	VSV-G mutant	CD7	-	T cells, NK cells	CD20 scFv/4-BB	R/R B-cell malignancies	NCT06539338
Umoja Biopharma	America	UB-VV111	LV	Cocal	CD3	CD58 and CD80	T cells	CD19 scFv/4-BB	R/R large B-cell lymphoma and chronic lymphocytic leukemia	NCT06528301
Umoja Biopharma	America	UB-VV400	LV	Cocal	CD3	CD58 and CD80	T cells	CD22 scFv/4-BB	R/R large B-cell lymphoma	NCT06743503
EsoBiotec	Belgium	ESO-T01	LV	VSV-G mutant	TCR	-	T cells	BCMA VHH/4-1BB	R/R multiple myeloma	NCT06691685 NCT06791681
Kelonia Therapeutics	America	KLN-1010	LV	VSV-G mutant	CD3	NR	T cells	Fully human BCMA CAR	R/R multiple myeloma	NCT07075185
Genocury	China	JY231	LV	VSV-G	CD3	NR	T cells	CD19 scFv/4-BB	R/R B-cell malignancies, neurologic Immune disorder, and autoimmune diseases	NCT06890065 NCT06797024 NCT06689917
OriCell Therapeutics	China	OriV508	LV	VSV-G mutant	CD3、CD7	-	T cells, NK cells	Loop CD19 and BCMA CAR	R/R B-cell malignancies	NCT07101705
Legend Biotech	China	LVIVO-TaVec100	LV	Cocal mutant	CD3	-	T cells	Bicistronic CD20 and CD19 CAR	R/R B-cell malignancies	NCT07002112
Sana Biotechnology	America	SG299	LV	paramyxovirus fusogen	CD8	NR	T cells	NR	Oncology and autoimmune diseases	Undisclosed
Exuma Biotec	America	GCAR-CD19	LV	NR	CD3	NR	T cells	CD19 scFv/NR	R/R B-cell malignancies	Undisclosed
Ensoma	America	NR	VLP	NR	CD46 (multi-lineage)	NR	T cells, NK cells, macrophages	HER2 scFv/NR	Multiple solid tumors	Undisclosed
Capstan therapeutics	America	CPTX2309	LNP	-	CD8	-	T cells	CD19 scFv/CD28	R/R autoimmune diseases	NCT06917742a
Immorna	China	JCXH-213	tLCNP	-	NR	-	T cells, NK cells, macrophages	NR	R/R B-cell non-Hodgkin’s lymphoma	NCT06618313
Myeloid Therapeutics	America	MT-302	LNP	-	FcRγ	-	Myeloid cells	TROP2 scFv/CD89 (mRNA)	Advanced epithelial tumors	NCT05969041
Myeloid Therapeutics	America	MT-303	LNP	-	FcRγ	-	Myeloid cells	GPC3 scFv/CD89 (mRNA)	Advanced hepatocellular carcinoma	NCT06478693
Shenzhen MagicRNA Biotechnology	China	HN2301	LNP	-	CD8	-	T cells	CD19 scFv (mRNA)	Systemic lupus erythematosus	NCT06801119
Orna Therapeutics	America	ORN-145	LNP	-	NR	-	T cells, NK cells, macrophages	CD19 CAR (cirRNA)	R/R B-cell malignancies	NR
Orna Therapeutics	America	ORN-252	LNP	-	NR	-	T cells, NK cells, macrophages	CD19 CAR (cirRNA)	R/R autoimmune diseases	NR
Orna Therapeutics	America	ORN-328	LNP	-	NR	-	T cells, NK cells, macrophages	BCMA CAR (cirRNA)	R/R multiple myeloma	NR
Sanofi	France	NR	LNP	-	CD8	-	T cells	CD22 scFv (mRNA)	Oncology and autoimmune diseases	NR
Tessera Therapeutics	America	NR	LNP	-	NR	-	T cells	CD19, CD20, BCMA CAR using RNA gene writer	Oncology and autoimmune diseases	NR
Orbital Therapeutics	America	OTX-201	LNP and VLP	-	NR	-	T cells	CD19 CAR(cirRNA)	R/R autoimmune diseases	NR
Carisma/Moderna	America	NR	LNP	-	NR	-	macrophages	GPC3 CAR(mRNA)	Hepatocellular carcinoma	NR

a. The study is currently recruiting healthy adult participants only

LNP lipid nanoparticle, LV lentiviral vector, tLCNP targeted lipid complex nanoparticles, VSV-G vesicular stomatitis virus glycoprotein, NR Not reached, TCR T cell receptor, NR not released, TROP2 trophoblast cell surface antigen 2, GPC3 glypican-3, BCMA B-cell maturation antigen, VHH variable domain of heavy chain of heavy-chain antibody, R/R relapsed or refractory, HER2 human epidermal growth factor receptor 2

Clinical development of in vivo CAR-T based on viral platform

LVs, as a key delivery tool, are advancing the clinical translation of in vivo CAR-T therapy. INT2104 is a LV engineered to retarget CD7 + T and NK cells, encoding an anti-CD20 CAR. Preclinical data showed that a single intravenous infusion of INT2104 induced good tumor regression in mice humanized with PBMCs, and also led to the reduction of B-cell compartment by at least 75% in 15 of 16 cynomolgus macaques. Its developer, Interius BioTherapeutics, has initiated the first-in-human Phase 1 study (NCT06539338) to evaluate the safety of INT2104 in relapsed or refractory B-cell malignancies, with detailed clinical data undisclosed. A novel product, INT2106 (encoding an anti-CD19 CAR) from the same developer, has also been developed to treat autoimmune disorders. Vivovec is a surface-engineered, third-generation LV platform rationally designed to generate CAR-Ts in situ. The lead candidate product derived from Vivovec, UB-VV100 (targeting CD3 + T and carrying anti-CD19 CAR), demonstrated effective depletion of endogenous B cells in CD34-humanized mice by 18 days following intraperitoneal injection, with predominant biodistribution in the liver and spleen. Studies in canines via intranodal or intraperitoneal administration indicated that UB-VV100 primarily transduced immune cells and tissues [139]. The optimized version UB-VV111 additionally incorporates a multidomain fusion protein (CD58, CD80 molecules and CD3-targeting scFv) to augment T-cell activation, proliferation, and transduction efficiency. In canines, GFP-expressing UB-VV111 did not show quantifiable transduction in any non-lymphoid tissues and was well tolerated combined with rapamycin in CD34-humanized mice [140]. Umoja Biopharma has announced the FDA’s approval to initiate the investigational new drug trial with UB-VV111 for treating hematological malignancies. Beyond UB-VV111, a similar product encoding an anti-CD22 CAR, UB-VV400, is also entering clinical practice with or without rapamycin for treating relapsed or refractory large B-cell lymphoma. Developed by Genocury Biotech, JY231 incorporates an anti-CD3 scFv and an activation molecule on the envelope, and delivers an anti-CD19 CAR transgene. A 64-year-old male patient with refractory diffuse large B-cell lymphoma received JY231 treatment and achieved complete remission by day 35 post-infusion [141]. It is noteworthy that this approach retained apheresis and lymphodepletion chemotherapy to facilitate the subsequent co-infusion of ex vivo-sorted T cells and lentiviral vectors for CAR transduction. KLN-1010 is an in vivo BCMA CAR-T therapy developed using Kelonia Therapeutics’ iGPS™ platform. This platform employs an engineered LV with a VSV-G mutant and a T-cell-specific CD3 scFv fused into the viral membrane. Preclinical results have demonstrated KLN-1010’s efficacy and safety in both mouse and non-human primate (NHP) models [142], and a Phase 1 study is registered recently (NCT07075185).

The first rigorously confirmed human data was reported by Wuhan Union Hospital and EsoBiotec, using the vector ESO-T01. T-cell specificity is achieved through mutating VSV-G and employing an envelope-anchored VHH targeting TCRαβ, and is further enhanced by using a T-cell-synthesized promoter to regulate expression of BCMA-targeted CAR. Additionally, the envelope is engineered to delete major histocompatibility complex class I and overexpress CD47 to evade complement inactivation and phagocytosis, respectively. In four participants treated with ESO-T01 at a single dose of 2.0 × 10⁸ TU, an overall response rate reached 100% (2 CR and 2 PR). The safety profile is deemed acceptable, with no adverse events exceeding grade 3, except for hematologic toxicities. ESO-T01 displayed a cellular kinetics comparable to those of commercial CAR-T products. The capacity of ESO-T01 in eliminating extramedullary lesions is particularly noteworthy, corresponding to the effective infiltration of CAR-T cells within the TME [143].

Clinical development of in vivo CAR-T based on non-viral platform

When it comes to LNPs, several pipelines have also entered clinical trials. MT-302 and MT-303 are the first two in vivo CAR cell therapies developed by Myeloid Therapeutics for the treatment of solid tumors. Both MT-302 and MT-303 are designed to engage and reprogram FcRγ(CD89)-expressing myeloid cells with mRNA-LNP. Their CAR constructs feature distinct scFvs targeting trophoblast cell surface antigen 2 (TROP2) or Glypican-3 (GPC3), combined with the shared transmembrane domain and cytoplasmic tail of CD89. Beyond inducing potent cytotoxicity against antigen-specific tumor cells, their myeloid-modified properties help stimulate T cells by presenting tumor neoantigens, thereby mediating a comprehensive anti-tumor immunity. Two ongoing Phase1, multicenter, open-label, dose-escalation studies (NCT05969041 and NCT06478693) are evaluating the safety and efficacy of MT-302 in epithelial cancers and MT-303 in HCC. Early clinical results demonstrated that MT-302 induced robust TROP2 CAR expression in circulating myeloid cells within hours post-infusion, and these CAR + myeloid cells were found co-localized with tumor cells in biopsies after infusion [144]. The systemic interferon-driven chemokine responses were observed following MT-302 treatment, with an increase in antigen presentation markers and pro-inflammatory signaling molecules. Repeat dosing of MT-303 at an interval of 14 days has proven feasible in patients, although updated efficacy data are pending [145]. CPTX2309 developed by Capstan Therapeutics has shown great potential in treating autoimmune diseases. In NHP studies, CPTX2309 (carrying a substitute anti-CD20 CAR) mediate a potent reconstitution of the B-cell compartment, characterized by the eradication of memory B cells and the recovery of the naïve phenotype [84]. A Phase 1 study (NCT06917742) is ongoing to assess the safety and tolerability of CPTX2309 in healthy adult participants. JCXH-213 is a mRNA-LNP-based in vivo CAR cell therapy that can deliver CAR to various immune cells (including T cells, NK cells, and MΦ), eliciting a synergistic anti-tumor activity. Its developer Immorna also announced that the first patient has received the treatment of JCXH-213 (NCT06618313).

Composed by propietary lonizable lipid ILB-3132, HN2301 is an engineered CD8-targeting LNP encapsulating CD19 CAR mRNA [146]. Preclinical results suggested that HN2301 can effectively generate in vivo CAR-T cells and killed B cells from both healthy donors and patients with systemic lupus erythematosus (SLE). Five patients with refractory SLE received 1 ~ 3 times of HN2301 infusion every other day at a dose of 2 or 4 mg. All patients exhibited decrease in disease activity 3 months post-infusion with no significant elevation of liver enzymes and cytopenia. In addition, no CRS exceeding grade 2 and immune effector cell-associated neurotoxicity syndrome were reported, indicating a favorable safety profile. These data indicated the great potential of in vivo CAR-T therapy in autoimmune disease.

Current limitations for in vivo CAR-T

Technical hurdles in design and manufacturing for in vivo CAR-T

In vivo CAR-T therapy holds great promise, yet both viral and non-viral gene delivery systems face considerable challenges. For viral vectors, the main limitations stem from the difficulties to control CAR expression and abrogate native tropism. In addition, the activation, expansion and persistence of in vivo generated CAR-Ts remain. Pre-existing immunity, particularly against AAVs, can lead to vector clearance and diminished transduction efficiency [147, 148]. Additionally, complement-mediated inactivation [149] and MΦ phagocytosis [150] further impair effective transduction after the infusion. Therefore, repeat dosing is often impractical due to potent humoral immune responses, which complicates long-term control of diseases. Additionally, LVs carry theoretical risks of genotoxicity from random genomic integration, potentially triggering oncogene activation or tumor suppressor disruption. AAVs are non-integrating and episomal expression wanes during T-cell proliferation [151]. While site-specific genome editing facilitates stable CAR expression, the requirement for co-delivered enzymes raises complexity and off-target risks. Moreover, AAVs’ limited cargo loading capacity (~ 4.7 kb) further constrains design flexibility [152].

Nonviral delivery platforms offer greater flexibility and lower immunogenicity but also encounter several critical barriers. Though targeting moieties improve selectivity, in vivo delivery efficiency remains low and variable, especially in humans. These systems also mainly rely on endosomal escape—a notoriously inefficient and context-dependent process. In addition, transgene expression is often transient, potentially insufficient for sustained tumor control. Although repeat dosing may partially compensate, it heightens the risk of anti-drug antibodies and increases treatment costs. Bioinspired delivery systems demonstrate attractive immune-evasive and targeting capabilities. However, their structural heterogeneity, low loading capacity, and unpredictable biodistribution profiles compromise therapeutic predictability and hinder regulatory approval. Similarly, macroencapsulation devices facilitate in situ gene delivery and immune cell programming but are confined to localized diseases and incompatible with systemic immune modulation. Last but not least, scaling up GMP manufacturing of aforementioned complex vectors—especially those incorporating functional ligands or membranes—poses substantial technical and regulatory hurdles. Ensuring batch-to-batch consistency, long-term stability, and reproducible quality at scale also remains challenging.

Translational barriers in clinical application for in vivo CAR-T

From a translational perspective, safety and controllability represent the paramount concerns for in vivo CAR-T therapy. The design of clinical studies—including selection of appropriate patient populations, determination of dosing strategies, and implementation of full-course management protocols—is of great importance yet largely undefined. Unlike ex vivo CAR-T with pre-infusion quality control, in vivo approaches lack precise regulation in transduction efficiency and CAR expression levels. Despite displaying comparable expansion kinetics to conventional ex vivo CAR-Ts, in vivo CAR-Ts fail to achieve sustained persistence, raising concerns about response durability. Strategies such as optimizing CAR constructs (e.g., co-expressing IL-7/IL-15) or adopting combination therapies (e.g., epigenetic drugs) have be proposed to address the same challenge based on ex vivo CAR-T therapies [153]. More importantly, systemic vector exposure triggers rapid-onset immune reactions within 24 hours. This is distinct from ex vivo CAR-T therapies, and necessitates proactive countermeasures like tiered dosing, antihistamine prophylaxis, and corticosteroid premedication. Non-viral platforms (e.g., tLNPs) has a lower risk of such immune reactions. Collectively, clinicians should adopt enhanced monitoring and intervention protocols that exceed conventional CAR-T standards.

Despite these translational challenges, in vivo CAR-T therapy offers unique biological advantages. Bypassing prolonged ex vivo culture, in vivo CAR-Ts may favor a less differentiated, more naïve phenotype, potentially enhancing therapeutic efficacy [154]. Furthermore, development within the intact host immune system enables in vivo CAR-Ts to synergize more effectively with endogenous immune cells, fostering cooperative anti-tumor activity. However, this also elevates the safety risks for inflammatory storm, due to increased sources of cytokine production. Moreover, the same host environment contains immunosuppressive components that can undermine CAR-T cell expansion and persistence. Notably, conventional ex vivo CAR-Ts struggle to infiltrate solid tumors and extramedullary lesions of hematologic malignancies [155, 156]. In vivo CAR-T holds great promise for targeting anatomically challenging sites. The delivery vectors used (e.g., LVs, LNPs) are typically much smaller than T cells, facilitating their penetration through the physical barriers surrounding solid masses. Once transduction occurs in vivo, the resulting favorable T cell phenotype may further improve tumor infiltration [143]. Thus, while mechanistic insights into efficacy/safety balances require deeper investigation, the approach warrants prioritized exploration in solid tumors.

Solutions to overcome preclinical and clinical challenges

Optimizing vectors for in vivo CAR-T generation

Regarding the design of LVs, researchers focus on optimizing the transduction efficiency and in vivo CAR-T functionalities. One study demonstrated that incorporation of T-cell activation and costimulation signals via a multidomain fusion protein (anti-CD3 scFv/CD80/D58) significantly augmented the antitumor functionality of in vivo CAR-Ts [35]. This highlights their dependency on T-cell activation state—a factor influenced by the use of wild cocal glycoprotein, which shares 71.5% identity to VSV-G and also binds LDLR [157]. However, as tropism-deleted viral vectors and targeted non-viral vectors bypass the VSV-G/LDLR recognition pattern, it is hard to illustrate the contributions of activation and costimulation signaling to in vivo CAR-T generation—despite both being essential for ex vivo CAR-T production. Evidence suggested that with CD3-retargeted NiV-LVs, costimulation signaling mainly enhances T-cell proliferation without affecting CAR transduction efficiency [25]. The promoter controlling CAR expression in LVs also influences transduction efficiency. The PGK promoter has been shown to drive higher CAR expression than the EF1α promoter [158]. In addition, to better regulate the growth and proliferation of in vivo CAR-Ts, Michels KR et al. [139] also employed rapamycin-activated cytokine receptor system to control additional IL-2R signaling.

In order to diminish off-target transduction, T-cell specific promoters and safety switches are employed to get rid of off-target tropism [159, 160]. Beyond this, a recent study shows that integration of the 4th-generation SupA2KO-LV backbone with the TRiP system eliminates aberrant RNA splicing and suppresses transgene translation during LV production. This approach reduces incorporation of CAR protein into vector particles, preventing undesirable B-cell transduction and enhancing overall safety [41]. To mitigate the immunogenicity of viral vectors, several strategies can be employed. A primary approach involves engineering the viral capsid or envelope to evade immune recognition. This can be achieved through rational mutagenesis of known antigenic epitopes, high-throughput screening of mutant libraries for optimized variants, or pseudotyping with heterologous envelope proteins. Complementary strategies focus on the host’s immune response, such as co-injecting empty viral particles to adsorb pre-existing neutralizing antibodies, or depleting the B-cell compartment to blunt humoral immunity.

Optimization of non-viral vectors primarily focuses on their core components, such as genetic cargo and lipid. While transposon-based pDNA allows stable gene integration for durable CAR expression, mRNA-based approaches result in transient transfection, limiting CAR persistence—especially in solid tumors. Circular RNA has an extended half-life, which enables prolonged CAR expression and makes it a promising alternative to linear mRNA. In addition, the co-delivery of sequence encoding RNA-dependent RNA polymerase in self-replicating or trans-amplifying RNA systems facilitate prolonged mRNA expression, but these two systems suffer from high immunogenicity and design complexity, respectively [161]. Furthermore, codon optimization can enhance mRNA stability, increase translational efficiency, and reduce immunogenicity. For the delivery vehicle itself, how to improve endosomal escape for boosting nucleic acid bioavailability remains a key challenge. This drives efforts to developing next-generation ionizable lipids and refining lipid compositions. Beyond delivery, strategies for monitoring in vivo CAR-T cells are also critical for clinical development. For example, a Stanford team developed anti-CD5 tLNPs co-delivering CD19 CAR and prostate-specific membrane antigen (PSMA) mRNA, with PSMA serving as a positron emission tomography (PET) imaging tag [162]. ⁶⁸Ga-PSMA-617 PET confirmed CAR-T generation and tumor infiltration. ⁶⁸Ga-PSMA-617 PET confirmed generation and tumor infiltration of in situ CAR-Ts88. However, the clinical translation remains challenging.

Evolving design for in vivo CAR-T clinical trials

Determining the dosing and infusion schedule is one of the primary challenges in designing clinical trials for in vivo CAR-T therapy. Strategies can be informed by a variety of data sources, including the human equivalent dose derived from effective murine doses (scaled by body weight or surface area), the multiplicity of infection used for in vitro transduction, and data from non-human primate studies [84, 150]. A fundamental distinction between in vivo and ex vivo CAR-T therapies is the lymphocyte requirement at baseline. While lymphodepletion plays a critical role in preparing the host environment conducive to infused CAR-Ts, in vivo CAR-T relies on abundant circulating T cells as targets for systemic vectors. This aligns with the clinical findings of ESO-T01, where the patient with lowest baseline lymphocyte count exhibited suboptimal CAR-T kinetics [143].

For in vivo CAR-T therapies, sufficient emphasis ought to be placed on safety monitoring following infusion, especially for viral vectors. While intranodal administration of LVs in NHPs induces mild inflammatory symptoms, intravenous administration risks acute systematic reactions. A NHP study reported fever, elevated liver enzymes, and a surge in cytokines within 24 hours post-intravenous infusion of hepatotropic LVs [150]. Similar changes were observed in four patients receiving ESO-T01 treatment, with cytokine release syndrome-like symptoms (fever, hypoxemia, hypotension) occurring within the first day, likely due to LV-triggered immune activation. However, in five patients treated NH2301, no similar symptoms were reported, though CAR-T cells peaked 6 hours after the infusion. These phenomena suggested a disparity in administration and management of patients in clinical studies using different platforms.

For viral vectors, residual lentiviral RNA must be closely monitored in patient-derived biological samples by qPCR during the early phase. This is critical for both gathering pharmacokinetic data and excluding viral-mediated infections. Furthermore, despite built-in replication defects, the potential risk of generating replication-competent lentiviruses through genetic recombination necessitates continuous and long-term monitoring for in vivo CAR-T therapies. This concern is heightened for in vivo CAR-T due to the systemic exposure of somatic cells to the integrating viral vector, although T-cell malignancies have also been reported following ex vivo CAR-T treatment [163–166]. For non-viral vectors with an improved safety profile, liver function must be closely monitored. Ultimately, whatever the platform is, the risk of off-target transduction should get shared attention.

Conclusion and future perspectives

The evolution of in vivo CAR cell therapy represents a transformative milestone in the field of adoptive cell transfer. Leveraging viral or non-viral vectors to directly reprogram immune cells within the human body, in vivo CAR engineering offers a streamlined, scalable, and potentially more accessible modality for personalized immunotherapy. Although in vivo CAR technology is still maturing and costs vary by platform, it nevertheless substantially reduces treatment prices from several hundred thousand to tens of thousands of dollars. Over the past decade, impressive strides have been made in vector engineering, including the development of tropism-modified LV and AAV systems, a new generation of synthetic and bioinspired NPs as well as macroencapsulation devices. These efforts have collectively laid the groundwork for in vivo CAR-T development with increasing specificity, safety, and efficiency.

Looking ahead, the applicability of in vivo CAR engineering has extended beyond hematological malignancies. While targeting solid tumors with stroma-specific CARs helps overcome the inhibitory TME, employing CAR-Ts to treat autoimmune diseases holds promise for reshaping abnormal immunity. In addition, fibrosis is also a potential indication for in vivo CAR-T based on LNP platform. The in vivo approach also opens new opportunities for treating infectious diseases, including chronic viral infections and invasive fungal or parasitic pathogens [108], where immediate, localized immune reprogramming is advantageous. CAR-immune cells beyond T lymphocytes—such as CAR-NK cells, CAR-neutrophils, and CAR-MΦ cells—also represent new avenues for in vivo CAR cell therapy. Furthermore, non-conventional lymphocyte populations, including γδ T cells, invariant NKT cells, and mucosal-associated invariant T cells, can also serve as target cells. These cell types exhibit inherent tissue tropism and natural antitumor or antimicrobial activities, making them highly attractive candidates for next-generation cell therapies. To achieve this, future vector platforms must be engineered with unparalleled specificity and sensitivity, for instance, by leveraging unique surface markers, distinct endocytic pathways, or tissue-resident niches to enable selective transduction.

To accelerate the clinical translation of this technology, establishing a unique gene-cell synergistic integration paradigm is imperative. Efforts should focus on developing distinct quality control standards for in vivo CAR-T products, while also designing clinical protocols that precisely define target populations. It is essential to explore sequential or combination therapies with conventional ex vivo CAR-T or other immunotherapies, without compromising therapeutic efficacy. More importantly, large-scale real-world cohorts must be established to gather robust clinical evidence, complemented by randomized controlled trials benchmarking efficacy against traditional ex vivo CAR-T therapies. As the datasets grow, professional societies should develop clinical practice guidelines to promoting this approach from experimental exploration shift to standardized treatment. Additionally, realizing the full potential of in vivo CAR cell therapy requires cross-disciplinary collaboration. Advances in synthetic biology, nanotechnology, structural virology, and systems immunology must be integrated with regulatory science and clinical expertise to ensure safety, efficacy, and scalability. Setting new standards for vector characterization, off-target profiling, and long-term safety will be vital as these technologies advance toward clinical trials.

In summary, in vivo CAR cell therapy offers a compelling paradigm for the treatment of oncology and autoimmune disorders. As the field matures, in vivo CAR-T and other CAR-immune cell therapies are expected to ultimately reshape the global therapeutic landscape of cellular therapies. However, as we move toward this promising future, we must balance the risks and benefits to make better use of this novel treatment modality.

Abbreviations

CAR

Chimeric antigen receptor

CAR-T

Chimeric antigen receptor T-cell

CAR-MΦ

Chimeric antigen receptor-macrophage

LVs

Lentiviral vectors

AAVs

Adeno-associated virus vectors

NPs

Nanoparticles

LNPs

Lipid nanoparticles

VSV-G

Vesicular stomatitis virus glycoprotein

LDLR

Low-density lipoprotein receptor

NiV

Nipah virus

MV

Measles virus

SINV

Sindbis virus

scFv

Single-chain variable fragment

DARPin

Designed ankyrin repeat protein

VHHs

Variable domain of heavy chain of heavy chain antibodies

TCR

T-cell receptor

tFab

Tandem antigen-binding fragment

PBMCs

Peripheral blood mononuclear cells

TRAC

T cell receptor alpha constant

VLPs

Virus-like particles

NSG

NOD.Cg-Prkdcscid IL2rgtm1Wjl/SzJ

TRiP

Transgene repression in vector production

PBAE

Poly(β-amino ester)

MΦ

Macrophage

PEI

Polyethyleneimine

pDNA

Plasmid DNA

PGA

Polyglutamic acid

MTAS

Microtubule-associated sequences

PPP

Poly(ethylene glycol)-b-poly(-caprolactone)-b-poly(ethylene imine)

TME

Tumor microenvironment

MPEI

Mannose-functionalized PEI

PEP-NP

Peptide-based nanoparticle

SA

Stearic acid

CASP11

Cysteinyl aspartate-specific protease 11

shRNA

Short hairpin RNA

S. aureus

Staphylococcus aureus

tLNPs

Targeted lipid nanoparticles

TRP1

Tyrosinase-related protein 1

PSMA

Prostate-specific membrane antigen

circRNA

Circular RNA

mcDNA

Minicircle DNA

ApoA1

Apolipoprotein A1

FAP

Fibroblast activation protein

ECM

Extracellular matrix

NHP

Non-human primate

SLE

Systemic lupus erythematosus

HCC

Hepatocellular carcinoma

GPC3

Glypican-3

ApoE

Apolipoprotein E

CAMP

Cardiolipin-mimic phosphoamide lipids

PSβ-LNPs

Phosphatidylserine/β-sitosterol LNPs

PS

Phosphatidylserine

RBCs

Red blood cells

eMSCs

Enucleated mesenchymal stem cells

GAMs

Glioma-associated microglia/macrophages

GBM

Glioblastoma

FuNVs

Fusogenic nanovesicles

p14TF

p14-based T-cell fusogen

MVTF

Measles virus-based T-cell fusogen

T-FVLP

T cell-specific fusogenic virus-like particle

MASTER

Multifunctional Alginate Scaffold for T cell Engineering and Release

α-CD

α-Cyclodextrin

CSR

Chimeric signaling receptor

TROP2

Trophoblast cell surface antigen 2

iNKT

Invariant natural killer T cells

PSMA

Prostate-specific membrane antigen

PET

Positron Emission Tomography

Author contributions

JX and ZZC contributed equally to this work. HM conceived the concept. JX, ZZC and HM wrote the original draft. JX, ZZC, LMS and AQR reviewed the manuscript. Supervision was made by HM. All authors have read and approved the article.

Funding

This study was supported by the National Natural Science Foundation of China (NSFC, No. 82425003, 82350103, and 82330005)

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.