Gene therapy for liver diseases: methods, challenges and opportunities

Abstract

Gene therapy has emerged as a promising and pivotal strategy for addressing numerous genetic disorders that currently lack effective therapeutic options. For many other presently incurable conditions, including the majority of inherited metabolic liver diseases, gene therapy offers a realistic and transformative treatment modality. This review highlights recent advancements in gene delivery vectors targeting the liver—specifically hepatocytes—with a focus on strategies involving gene supplementation and gene editing, as well as notable progress achieved through RNA-based therapeutic agents. Several early trials utilizing lipid nanoparticle (LNP)-based and recombinant adeno-associated virus (rAAV)-mediated approaches for gene supplementation and editing in liver diseases have yielded encouraging results. For example, rAAV-mediated FIX supplementation (2 × 10¹³ genome copies/kg) achieves sustained therapeutic activity in a subset of HOPE-B trial participants with hemophilia B, and LNP-CRISPR editing of ANGPTL3 (0.5 mg/kg mRNA) produces marked reductions in LDL-C and triglycerides. However, several translational challenges have been identified, including unintended activation of the innate immune system and severe hepatotoxicity following high-dose vector administration. These concerns necessitate rigorous and ongoing safety monitoring. In this review, we summarize major advancements, current challenges, and future perspectives in the field of liver-directed gene therapy. Additionally, we highlight key limitations inherent to existing approaches and discuss potential strategies to overcome these barriers and improve therapeutic outcomes.

Graphical Abstract

Introduction

According to the updated 2019 definition by the American Society for Gene and Cell Therapy (ASGCT), gene therapy refers to a medical method of treating or curing diseases by introducing, deleting, or altering an individual’s genetic material [1]. This definition encompasses not only traditional gene supplementation or addition strategies but also gene editing techniques that modify, repair, or integrate DNA sequences into the cellular genome, as well as methods for silencing gene expression using RNA interference or targeted nucleases (Fig. 1) [2]. In contrast, cell therapy involves the administration of living cells—either unmodified or genetically engineered—to achieve therapeutic effects. While gene therapy focuses on the genetic modification itself, cell therapy emphasizes the cellular product as the therapeutic agent. The distinction becomes particularly relevant when considering vitro approaches: genetic modification of cells outside the body followed by infusion constitutes gene therapy-mediated cell therapy, such as CAR-T cells, whereas the ASGCT definition classifies this under gene therapy when the genetic alteration is the primary therapeutic mechanism.

Fig. 1.

Current nucleic acid-based therapeutics for liver diseases. Gene therapy strategies primarily encompass involve RNA and DNA therapeutics, as well as gene editing. These strategies can be categorized by therapeutic modality into gene silencing (using ASOs and siRNAs), gene supplementation (using mRNA and viral vectors), and gene repair (employing CRISPR-Cas9 with sgRNA)

Gene therapy emerged in the early 1990s and is considered a potentially effective method for treating various genetic disorders, particularly monogenic diseases [3–5]. However, the field has faced significant challenges. In 1999, an 18-year-old patient named Jesse Gelsinger died during a clinical trial evaluating adenovirus vector therapy for ornithine transcarbamylase deficiency (OTCD). His death was caused by a massive systemic inflammatory response triggered by a cytokine storm, which led to multi-organ failure and disseminated intravascular coagulation [6]. Gelsinger received an infusion of 6 × 10¹¹ particles/kg of a recombinant adenovirus type 5 vector expressing the ornithine transcarbamylase (OTC) transgene via the right hepatic artery. Within 18 h, he developed systemic inflammatory response syndrome, acute respiratory distress syndrome, and disseminated intravascular coagulation, leading to death 98 h post-injection. This tragic event, extensively documented in the scientific literature and popular media, underscored the critical importance of rigorous preclinical safety assessment, adequate informed consent, and robust regulatory oversight in gene therapy trials [7]. Additionally, some patients with severe combined immunodeficiency who received retroviral vector-modified hematopoietic stem cell therapy subsequently developed leukemia due to insertional mutagenesis. In this process, the integrating gammaretroviral vector preferentially inserted near transcription start sites of proto-oncogenes and activated their expression through the vector’s strong enhancer elements located in the long terminal repeat (LTR), driving uncontrolled clonal T cell proliferation [8]. These adverse events have prompted researchers to explore novel carrier technologies, innovative gene therapy strategies, and rigorous preclinical safety assessments, thereby advancing the potential for successful clinical applications in the future. Over the past five years, several gene therapy products with diverse mechanisms and characteristics have been approved. These include antisense oligonucleotides (ASOs) [9], RNA interference therapies [10], autologous CD34⁺cell therapies with the human ADA gene introduced via lentiviral vectors [11], and recombinant adeno-associated virus (rAAV) vector therapies (Table 1) [12]. This article focuses on reviewing the latest advancements in gene therapy for liver diseases.

Table 1.

Liver-directed gene therapy approval schedule for the past five years

Year	Modality	Product	Target	Diseases	Liver Relevance	Stage	Administration	Dose and frequency	Ref.
2019	RNAi	Givlaari	ALAS1	Acute hepatic porphyria	Yes	Approved	Subcutaneous injection	2.5 mg/kg; once monthly	[41]
2020	RNAi	Oxlumo (lumasiran)	HAO1	Primary hyperoxaluria type 1	Yes	Approved	Subcutaneous injection	3 mg/kg monthly ×3 (loading), then quarterly maintenance	[42]
2021	RNAi	Leqvio (inclisiran)	PCSK9	Hypercholesterolemia	Yes	Approved	Subcutaneous injection	284 mg on Day 1, Day 90, Day 270, Day 450	[43]
2022	rAAV	Hemgenix (etranacogene dezaparvovec)	FIX	Hemophilia B	Yes	Approved	Intravenous infusion	2 × 10¹³ gc/kg; single dose	[44]
2023	RNAi	Rivfloza (nedosiran)	LDH	Primary hyperoxaluria type 1	Yes	Approved	Subcutaneous injection	≥ 50 kg: 170 mg; <50 kg: 136 mg; once monthly	[45]
2023	rAAV	Roctavian (valoctocogene roxaparvovec)	FVIII	Hemophilia A	Yes	Approved	Intravenous infusion	6 × 10¹³ vg/kg; single dose	[46]
2024	rAAV	Beqvez (fidanacogene elaparvovec)	FIX	Hemophilia B	Yes	Approved	Intravenous infusion	5 × 1011 vg/kg; single dose	[47]
2025	rAAV	BBM-H901 (dalnacogene ponparvovec)	FIX	Hemophilia B	Yes	Approved (China)	Intravenous infusion	5 × 1012 vg/kg; single dose	[48]

ALAS1, aminolevulinate synthase 1; HAO1, hydroxyacid oxidase 1; LDH, Lactate dehydrogenase; PCSK9, proprotein convertase subtilisin/kexin type 9; rAAV, recombinant adeno-associated virus; RNAi, RNA interference

Gene therapy can be classified into two types based on the administration method: in ex vivo and in vivo [13]. In ex vivo gene therapy, the patient’s cells are first isolated, cultured, and expanded, then genetically modified using a specific vector, and finally, the modified cells are reintroduced into the patient’s body [14]. In contrast, in vivo gene therapy involves the direct delivery of therapeutic agents to the patient’s body via systemic or local routes. Due to the limited capacity for ex vivo manipulation and expansion of liver cells, in vivo gene therapy is more practical for treating liver diseases [15]. Specifically, primary hepatocytes rapidly undergo dedifferentiation in culture, losing liver-specific functions such as albumin production and cytochrome P450 activity, while acquiring a fibroblast-like morphology within days [16]. Moreover, adult hepatocytes are largely quiescent cells with limited proliferative capacity under standard culture conditions, making sufficient expansion for ex vivo gene therapy challenging [17]. Importantly, gene therapy strategies can also be categorized by their therapeutic mechanism into gene addition (supplementation) therapy and gene correction (editing) therapy. Gene addition therapy involves introducing a functional copy of a defective gene to compensate for the loss of function, whereas gene correction therapy aims to repair the disease-causing mutation directly at the genomic level using gene editing tools such as CRISPR-Cas9, base editors, or prime editors. These two approaches have distinct regulatory pathways and safety considerations: gene addition typically results in episomal persistence of the therapeutic transgene with potential for immune responses against vector and transgene products, while gene correction carries risks associated with off-target editing, chromosomal rearrangements, and permanent heritable changes [18].

As the central organ for most metabolic processes, the liver is also the site where various hereditary metabolic liver diseases manifest. These diseases are typically caused by abnormal protein synthesis or dysfunction in key metabolic pathways [19]. Collectively, inherited metabolic liver diseases impose a substantial global health burden. Alpha-1 antitrypsin deficiency affects approximately 1 in 2,000–6,000 individuals globally, with an estimated 3.4 million people harboring severe deficiency alleles and over 116 million carriers worldwide [20, 21]. Notably, in the United States, 70,000–100,000 individuals have severe alpha-1 antitrypsin deficiency, yet less than 10% are identified, and diagnosed patients exhibit 3.5 to 6.2 times higher mortality rates than the general population [22, 23]. Other significant disorders include glycogen storage disease type I (1 in 100,000 live births) [24, 25], Wilson disease (WD) (prevalence of 21–38 per million in recent nationwide studies) [26, 27], urea cycle disorders (1 in 35,000 live births) [28, 29], and hereditary tyrosinemia type 1 (HT1) (1 in 100,000 live births, or 1 in 1,846 in Quebec) [30]. Crigler-Najjar syndrome (CNS), though exceptionally rare (0.6–1 per million newborns), carries devastating neurological consequences requiring lifelong phototherapy or liver transplantation [31]. Currently, traditional enzyme replacement therapy (ERT) is effective for only a few types of hereditary metabolic liver diseases, and for most patients, the sole treatment option remains orthotopic liver transplantation [32]. However, this approach is constrained by the limited availability of donor organs and necessitates long-term immunosuppressive therapy. The fact that liver transplantation can achieve a cure further supports the theoretical foundation for restoring missing or defective protein function in the liver through gene therapy.

In addition, liver functions include synthesizing and secreting most of the key proteins found in the blood. Due to this characteristic, the liver is considered an ideal target organ for gene therapy, which can be used to produce therapeutic proteins required for the circulatory system [33]. This strategy has been successfully applied in gene therapy research for hereditary hemorrhagic diseases such as hemophilia A and B and has also been explored for treating lysosomal storage diseases amenable to enzyme replacement therapy [34, 35]. While hepatocytes have traditionally been the primary targets for protein replacement therapies, the liver’s therapeutic potential extends to various non-parenchymal cells, which constitute approximately 40% of its cellular composition. These include Kupffer cells, hepatic stellate cells (HSCs), and liver sinusoidal endothelial cells (LSECs), each offering distinct therapeutic opportunities. Specifically, Kupffer cells—central to hepatic immunity—have been targeted for functional reprogramming in liver cancer [36] and chronic hepatitis B [37]. HSCs, the primary effectors of fibrosis, represent promising targets for antifibrotic gene therapies aimed at deactivating these collagen-producing cells [15, 38]. Additionally, LSECs, known for their potent tolerogenic properties, facilitate both immune tolerance induction and therapeutic protein production [39, 40].

Given that genetic material typically has a high molecular weight and a negative charge, which hinder its ability to freely penetrate cell membranes, it is necessary to use carrier systems to achieve effective targeted delivery. This article reviews the latest developments and clinical applications of liver-targeted gene therapy strategies.

Current nucleic acid therapeutics for liver diseases

Gene addition therapy

Gene addition therapy represents the most clinically advanced approach for liver-directed gene therapy, involving the delivery of exogenous functional transgenes to compensate for genetic deficiencies without modifying the endogenous genome. This strategy utilizes viral vectors—predominantly recombinant adeno-associated virus (rAAV)—to introduce therapeutic cDNA expression cassettes, or non-viral systems such as lipid nanoparticle (LNP)-messenger RNA (mRNA) for transient protein replacement [49].

Most non-viral vectors consist of polymer- or lipid-based nanoparticles that encapsulate and protect the genetic cargo they carry, thereby facilitating cellular uptake (Fig. 2) [50]. Compared to viral vectors, non-viral delivery systems offer several advantages, including ease of large-scale production, extended shelf life, and the potential to accommodate larger genetic payloads than viral vectors. However, this capacity is subject to practical formulation constraints. LNP encapsulation efficiency decreases with increasing nucleic acid size due to thermodynamic packing limitations, structural instability of large mRNA molecules, and the physical dimensions of the aqueous core [51]. While the absence of viral components eliminates risks associated with viral replication, insertional mutagenesis, and pre-existing anti-viral immunity, non-viral systems—particularly LNPs—present their own safety challenges. These include complement activation-related pseudoallergy, which can trigger infusion reactions; robust innate immune responses to ionizable lipids; and the development of anti-polyethylene glycol (PEG) antibodies upon repeated administration, which may accelerate drug clearance and increase the risk of hypersensitivity [52, 53].

Fig. 2.

Different vectors used most frequently in liver-directed gene therapy. In vivo gene therapy involves the direct administration of gene delivery vectors or genetic material into the organism. This approach can utilize various vector systems, including both non-viral and viral vectors, with viral vectors being the most commonly used in liver-targeted gene therapy. Each vector type possesses unique characteristics, along with distinct advantages and limitations, making the careful selection of vectors a critical factor in the success of diverse gene therapy applications

A major limitation of non-viral vectors is their relatively low efficiency in nuclear delivery and limited capacity for long-term transgene expression. However, this limitation has minimal impact on RNA therapeutics, as RNA functions directly in the cytoplasm without requiring nuclear entry. Consequently, non-viral vectors have become the preferred delivery platform for RNA-based interventions. The recent success of LNP-mRNA vaccines and RNA interference therapies underscores this potential [54]. A critical distinction in RNA therapeutics lies between chemically modified small interfering RNA (siRNA) platforms and unmodified RNA constructs. Unmodified siRNAs are rapidly degraded by serum nucleases and trigger potent innate immune responses via toll-like receptor (TLR)3/7/8 and RIG-I activation, resulting in excessive type I interferon production [55]. In contrast, 2’-O-methyl (2’-OMe) and 2’-fluoro (2’-F) modifications confer substantial advantages: 2’-OMe increases nuclease resistance and abrogates TLR recognition, while 2’-F enhances binding affinity to Argonaute 2 and reduces cytokine production [56, 57]. Current clinically validated siRNA therapeutics, such as patisiran and givosiran, utilize optimized modification patterns combining 2’-OMe, 2’-F, and phosphorothioate backbone linkages to achieve sustained gene silencing lasting up to 680 days in hepatocytes, whereas unmodified siRNAs lose activity within days [58].

Beyond serving as templates for protein synthesis, RNA molecules exhibit diverse functional properties—including catalytic activity, scaffold formation, and gene expression regulation—primarily mediated by complementary base pairing [59]. These characteristics enable RNA therapeutics to target “undruggable” molecules, such as non-coding RNAs and challenging mRNA targets, which are inaccessible to conventional small-molecule drugs and biologics [60]. Moreover, mRNA’s capacity to direct de novo protein synthesis in situ allows for the production of therapeutic proteins within patient cells, opening new avenues for precision medicine [61]. Based on their mechanisms of action, RNA-based therapeutics can be categorized into three classes: (1) agents that modulate cellular RNA function, (2) RNA-guided targeted genomic editing, and (3) mRNA-directed therapeutic protein expression [62].

In contrast to non-viral systems, viral vectors used for therapeutic applications are typically engineered by deleting viral replication and pathogenicity genes and replacing them with therapeutic transgenes [63]. For liver-targeted gene therapy, commonly used viral vectors include AAV, adenovirus, retrovirus, and lentivirus, each exhibiting distinct characteristics in transduction efficiency, duration of gene expression, and immunogenicity [64].

A critical comparison of viral and non-viral vectors reveals distinct differences in intracellular trafficking and therapeutic applicability (Table 2). Viral vectors exploit endosomal acidification to induce capsid conformational changes, facilitating efficient cytosolic cargo release [65]. In contrast, non-viral vectors face significant challenges in endosomal escape and employ strategies such as the proton sponge effect, fusogenic peptides, and ionizable lipids to overcome this barrier [66, 67]. However, they generally exhibit lower escape efficiency than their viral counterparts [68]. Quantitative studies demonstrate that only approximately 1–2% of internalized LNP-encapsulated nucleic acids successfully escape from endosomes into the cytoplasm in hepatocytes, representing a major rate-limiting step for therapeutic efficacy [69]. This low escape efficiency stems from the spontaneous and transient nature of endosomal membrane breaches (< 10 nm), which create a substantial bottleneck for RISC loading and target mRNA silencing [70]. More recent analyses using the SNAP_Switch/SNAP-tag assay have confirmed that FDA-approved LNP formulations (SM-102, DLin-MC3-DMA, ALC-0315) exhibit similarly low endosomal escape efficiencies of approximately 4–10% across various cell lines in vitro, with SM-102 showing the highest escape efficiency at ~ 10%, while DLin-MC3-DMA and ALC-0315 demonstrated only ~ 5% and ~ 4%, respectively [71]. Regarding cytosolic stability and translation kinetics, AAV-mediated expression peaks within weeks but persists for months, whereas adenoviral vectors provide robust expression for 2–3 weeks [66]. Non-viral plasmid DNA delivery yields transient expression lasting days to weeks due to nuclease susceptibility [68], though mRNA-LNPs enable rapid protein expression within hours without the risk of genomic integration, making them suitable for applications requiring transient protein production [72]. Importantly, viral vectors elicit robust immune responses—including neutralizing antibodies (nAbs) and cytotoxic T lymphocytes—that preclude repeat dosing [73, 74]. In contrast, non-viral vectors exhibit markedly lower immunogenicity, enabling chronic administration, as demonstrated by patisiran and COVID-19 mRNA vaccines [75, 76].

Table 2.

The key comparative features between viral and non-viral vectors

Features	Viral Vectors	Non-Viral Vectors
Endosomal escape efficiency	High—evolved sophisticated mechanisms exploiting endosomal acidification	Low to Moderate—requires engineering (proton sponge, fusogenic peptides, ionizable lipids)
Cytosolic stability	High—viral capsid protection and nuclear targeting mechanisms	Moderate—susceptible to nuclease degradation; DNA remains non-integrated
Translation kinetics	Delayed onset (days to weeks) but sustained expression (weeks to months)	Rapid onset (hours) but transient expression (days to weeks)
Repeat dosing feasibility	Limited—high immunogenicity triggers nAbs and cellular immunity	High—low immunogenicity enables chronic administration
Payload capacity	Limited (~ 4.7 kb for AAV)	Large (> 10 kb)
immunogenicity	High—pre-existing immunity common; limits efficacy and redosing	Low—minimal immune activation
Manufacturing scalability	Complex, cell-based production	Streamlined, cell-free synthesis

However, the efficiency of these delivery systems is fundamentally constrained by physiological barriers encountered prior to reaching target hepatocytes. In vivo, vectors face a series of anatomical and immunological checkpoints that operate sequentially to filter circulating nanoparticles and viral particles, ultimately determining the fraction of administered dose available for hepatocyte transduction. Following systemic administration, gene delivery vectors encounter sequential anatomical barriers that critically determine hepatic bioavailability. The hepatic sinusoidal endothelium, characterized by fenestrations ranging from 100 to 200 nm in diameter, serves as the first physical filtration barrier [77]. Recent studies demonstrate that liver fibrosis severely impairs this filtration process: in fibrotic livers, defenestration of LSECs reduces vector uptake by hepatocytes and delays blood clearance, leading to altered biodistribution with increased accumulation in extrahepatic organs such as spleen and lung [78]. The size and density of fenestrations vary not only by disease state but also by zonal location within the hepatic lobule and across species, with human fenestration diameters averaging 107 nm—similar to non-human primates but distinct from rodent models [77]. These structural variations significantly influence vector penetration and hepatocyte transduction efficiency, highlighting the need for disease-specific and species-specific considerations in vector design. However, even for vectors that successfully permeate the fenestrated endothelium, a second and equally formidable barrier awaits within the sinusoidal lumen.

Kupffer cell-mediated clearance remains a major bottleneck for efficient hepatocyte targeting. Recent evidence indicates that liver fibrosis exacerbates this challenge by promoting vector uptake by immunocompetent and antigen-presenting cells in both fibrotic liver and spleen, potentially modifying the immune profile and compromising therapeutic efficacy [78]. Genetic studies further reveal that host genetic variation significantly influences hepatic vector transduction and biodistribution patterns, with strain-specific differences observed in vector copy number distribution between liver and spleen [79]. Given the substantial loss of therapeutic payload to these phagocytic sentinels, substantial efforts have been directed toward engineering strategies that evade or modulate Kupffer cell surveillance. Current strategies to evade Kupffer cell capture include CD47-mediated signaling, which has shown enhanced transduction efficiency in preclinical models, and transient macrophage depletion protocols [80]. Additionally, isolated hepatic perfusion has emerged as a promising localized delivery strategy that circumvents systemic Kupffer cell exposure, achieving enhanced and zonally biased AAV transduction in both rodent and non-human primate models [77].

Gene correction therapy

Gene correction therapy aims to permanently repair disease-causing mutations through direct modification of the endogenous genome, offering the theoretical advantage of mutation-specific repair under natural regulatory control. This approach employs CRISPR-Cas nucleases, zinc-finger nuclease (ZFN)s, transcription activator-like effector nucleases, or precision editing tools including cytosine base editors (CBEs), adenine base editors (ABEs), and prime editors. However, CRISPR-Cas9-mediated gene editing introduces DNA double-strand breaks (DSBs) that activate the DNA damage response pathway, including ataxia telangiectasia mutated and DNA-dependent protein kinase signaling, leading to p53-mediated cell cycle arrest or apoptosis in hepatocytes [81, 82]. Elevated p53 activity can compromise editing efficiency by eliminating successfully edited cells and potentially selecting for p53-deficient clones with malignant potential. Recent studies indicate that transient p53 inhibition or the use of DSB-free base editors and prime editors may mitigate these cellular stress responses, improving both editing efficiency and safety profiles [83, 84].

A critical safety consideration for gene correction therapies is the potential for off-target editing at unintended genomic sites. Comprehensive detection and quantification of off-target effects are mandated by regulatory agencies prior to clinical approval. Current methodologies include unbiased cell-based assays such as GUIDE-seq (genome-wide, unbiased identification of DSBs enabled by sequencing), which captures double-strand break sites through integration of oligodeoxynucleotides; CIRCLE-seq (circularization for ex vivo reporting of cleavage effects by sequencing), an ex vivo method that circularizes genomic DNA to identify nuclease cleavage sites genome-wide; and site-specific amplification methods such as targeted amplicon sequencing and digital PCR [85–87]. Additionally, long-read sequencing technologies and computational prediction algorithms (e.g., Computational Fluid Dynamics‌ score, Mismatch Intolerance Tool score, Elevation) complement experimental validation. These technologies enable sensitive detection of off-target indels, structural variants, and chromosomal translocations, informing single-guide RNA (sgRNA) design optimization and risk-benefit assessments for clinical translation.

Beyond these DSB-related concerns, gene editing introduces additional safety considerations including off-target mutagenesis, unintended on-target modifications (indels), chromosomal translocations, and the potential for mosaicism. Regulatory oversight for gene editing therapies requires comprehensive assessment of editing specificity, long-term genomic stability, and potential germline transmission risks [88]. The transient nature of mRNA-encoded editors delivered via LNPs may mitigate some safety concerns compared to persistent viral vector-mediated editing systems [89].

Regulatory and safety distinctions between gene addition and gene correction

There are essential differences in regulatory frameworks and safety considerations between gene addition therapy and gene correction therapy.

Gene addition therapy compensates for genetic defects by introducing exogenous functional gene copies without altering the patient’s original genome; Its regulatory pathway is relatively mature, mainly evaluated as a biological product, with a focus on vector production quality control, transgenic expression persistence, immunogenicity monitoring, and insertion mutation risk assessment (for integrated vectors). Due to the fact that the therapeutic effect can be diluted or attenuated with cell division, the safety risks are relatively controllable, but long-term monitoring of immune response and expression stability is required [90].

Gene correction therapy directly repairs endogenous pathogenic mutations, resulting in permanent and heritable genomic changes; Regulatory requirements are more stringent, requiring a comprehensive assessment of editing specificity (off target and off target effects), chromosome integrity, mosaic phenomena, and potential reproductive transmission risks. Preclinical evaluation must incorporate validated off-target detection methodologies, with GUIDE-seq and CIRCLE-seq representing current standards for unbiased genome-wide assessment of nuclease cleavage sites [75, 76]. Regulatory submissions typically require demonstration that off-target editing frequencies fall below established safety thresholds (generally < 0.1% at sites with > 1% on-target activity), accompanied by comprehensive bioinformatic analysis and long-term follow-up plans to monitor for clonal expansion of off-target events. Additionally, the activation of DNA damage response pathways and p53-mediated cellular stress by DSB-inducing nucleases necessitates careful evaluation of hepatocyte viability, editing efficiency trade-offs, and potential clonal selection pressures in preclinical models. Due to its irreversibility, whole genome off target analysis and long-term cloning dynamic tracking need to be conducted before clinical trials, and lifelong monitoring needs to be implemented after marketing to screen for delayed adverse events (such as tumor occurrence). In short, the regulatory focus of gene addition is on product consistency and immune safety, while gene correction requires additional confirmation of the lifelong safety of precise genomic modifications, including mitigation strategies for DSB-induced cellular stress [91].

Polycationic and LNP applicable for liver diseases

There are two primary types of non-viral vectors that have demonstrated significant efficiency in delivering genetic material to the liver: cationic polymers, also known as polycations, and liposomal formulations (Table 3).

Table 3.

Clinical applications of gene therapy in liver diseases

Disease	Target gene	Payload	Dose	Frequency	Phase	Major Safety Findings	Administration	Therapeutic effects	Ref.
Hepatorenal tyrosinemia	FAH	LNP	0.5 mg/kg	Every 2 days for a total of three consecutive doses	Preclinical	Well-tolerated; minimized hepatotoxicity	Intravenous injection	Normalizing tyrosine, phenylalanine, and biochemical markers of liver injury	[309]
Alpha-1 antitrypsin deficiency	FAH	shRNA	10% of the body weight	Single dose	Preclinical	N/A	Intravenous injection	Addressing liver toxicity	[310]
Fabry disease	α-GLA	LNP	N/A	Monthly	Preclinical	Well-tolerated; transient plasma GL-3 reduction	Intravenous injection	Altering the level of metabolites in the amino acid and energy metabolic pathways	[311]
Acute intermittent porphyria	PBGD	AAV	5 × 1011 to 1.8 × 1013 gc/kg	Single dose	I	Well-tolerated	Intravenous injection	The demand for heme treatment has declined, but the levels of ALA and PBG have remained unchanged	[221]
Acute hepatic porphyria	ALAS1	siRNA	2.5 mg/kg	Monthly	Approved	Generally well-tolerated; hepatic and renal monitoring recommended	Subcutaneous injection	Preventing acute porphyric attacks	[312]
Methylmalonic acidemia	MUT	LNP	0.5 mg/kg	2 weeks	Preclinical	Well-tolerated	Intravenous injection	Resulting in greater and more sustained reductions in plasma methylmalonic acid	[124]
Hypercholesterolemia	PCSK9	LNP	5 mg/kg	3 weeks	Preclinical	Well-tolerance; and not increased ALT and AST levels	Intravenous injection	Reducing PCSK9 mRNA by 50–70%, and plasma total cholesterol by 30–60%; significantly reduce PCSK9 protein and LDL cholesterol in non-human primates	[313]
	PCSK9and ANGPTL3	LNP-CRISPR-Cas9	500 µg pDNA/ mouse	Single dose	Preclinical	Well-tolerance; no significant change in serum ALT level	Intraduodenal injection	Significantly reducing serum LDL cholesterol levels	[105]
Transthyretin amyloidosis	TTR	LNP	0.3 mg/kg	3 weeks	Approved	Well-tolerability; mild to moderate infusion related reactions	Intravenous injection	Lowering the average level of serum TTR, significantly preserving the patient’s functional ability (6-minute walking distance) and quality of life (KCCQ-OS score)	[314]
Phenylketonuria	PAH	AAV-LNP and mRNA-LNP	2 mg/kg	Single dose	Preclinical	Effective instantaneous prime editor expression strategy safety	Intravenous injection	Reducing the level of L-phenylalanine in the blood	[315]
Tyrosinemia type 1	FAH	AAV-CRISPR/Cas9	3 × 1012	Single dose	Preclinical	Well-tolerance, with no significant off-target effects; restoring normal serum biochemical indicators	Intravenous injection	Recovering liver function; decreasing liver and kidney damage	[316]
Hypercholesterolemia	PCSK9	siRNA-GalNAc	284 mg	1 day, 90 day, and every 6 monthly	Approved	Acceptable safety profile; mild injection site reaction	Subcutaneous injection	Lowering LDL cholesterol levels	[317]
Primary hyperoxaluria type 1	siRNA-GalNAc	GalNAc conjugation	Weight-based: <10 kg: 6 mg/kg loading then 3 mg/kg monthly; 10–20 kg: 6 mg/kg loading then 6 mg/kg quarterly; ≥20 kg: 3 mg/kg loading then 3 mg/kg quarterly	Loading: 0, 30, 60; Maintenance: 90, 180, 270. (< 10 kg: 30-day intervals; ≥10 kg: 90-day intervals)	Approved	Acceptable safety profile; sustained oxalate lowering	Subcutaneous injection	Inhibiting the synthesis of oxalate	[42]
	LDH	siRNA	Weight-based: 160 mg in patients aged ≥ 9 years weighing ≥ 50 kg, 128 mg in those aged ≥ 12 years weighing < 50 kg and 3.3 mg/kg in children aged 9–11 years weighing < 50 kg	Monthly	Approved	Acceptable safety profile	Subcutaneous injection	Reducing oxalate overproduction	[45]
	HAO1	AAV-CRISPR/Cas9 nickase	5 × 1012 vg/kg; 1.5 × 1013 vg/kg; 5 × 1013 vg/kg	Single dose	I/II	Acceptable safety profile; the safety of the effective Cas9 cleavage enzyme strategy	Intravenous injection	Permanent disruption of HAO1 gene; Significantly reducing the level of urinary oxalate, and increasing the level of urinary acetaldehyde	[241]
Metabolic dysfunction-associated steatohepatitis	PNPLA3	ASO	25 mg, 50 mg and 80 mg	Single dose	I	Acceptable safety and tolerability profile	Intravenous injection	Reducing hepatic steatosis; dose-dependent increasing polyunsaturated fatty acids in serum triglycerides; and decreasing in high-sensitivity C-reactive protein and interleukin 6	[318]
	TTC39B	ASO	2 × 1011 GC	Single dose	Preclinical	Acceptable safety and tolerability profile	Intravenous injection	Significantly reducing the expression of genes involved in fat production (particularly Pnpla3); decreasing the deposition of triglycerides in the liver	[319]
	HSD17B13	ASO	10, 25, and 50 mg/kg	Weekly	Preclinical	Well-tolerance; dose-dependent significantly increasing the serum AST and ALT levels in higher doses (25 and 50 mg/kg)	Intravenous injection	Effectively inhibiting the expression of the Hsd17b13 gene; significantly improving liver steatosis; but having no effect on liver fibrosis	[320]
	DUSP22	AAV	1.5 × 1012 GC	Single dose	Preclinical	Well-tolerance	Intravenous injection	Inhibiting MASH-related phenotypes and HCC development	[321]
	SLC27A4	AAV	1.0 × 109 GC	Single dose	Preclinical	Well-tolerance	Intravenous injection	Inducing SLC27A4’s ectopic overexpression in the liver, and subsequently causing transport of excessive hepatic phosphatidylcholine from peripheral tissues to liver	[322]
Hemophilia B	FIX	rAAV	5 × 1011- 2 × 1013 vg/kg	Single-dose	Approved	Well-tolerance; hepatotoxicity at high doses	Intravenous injection	Increasing FIX activity	[47]
			5 × 1011 vg/kg	Single-dose	Approved	No or only low-grade adverse effects	Intravenous infusion	Increasing FIX activity	[47]
Hemophilia A	BDD-FVIII	AAV	0.5 × 1013 -2 × 1013 vg/kg	Single-dose	Approved	Well-tolerance	Intravenous injection	Maintaining FVIII expression that normalizes hemostasis	[323]
	FVIII	AAV	6 × 10¹³ vg/kg	Single-dose	Approved	Well-tolerance	Intravenous injection	Increasing FVIII activity	[46]
Ornithine transcarbamylase deficiency	OTC	AAV	N/A	Single-dose	I/II	Ammonia detoxification issues at high doses; transient liver enzyme elevations	Intravenous injection	Restoring the physiological ammonia detoxification liver function	[324]
Phenylketonuria	PAH	AAV	2.1 × 1011 vg	Single-dose	Preclinical	Generally well-tolerated; immune responses	Intravenous injection	Yielding partial restoration of liver PAH activity, and substantial reduction of blood phenylalanine	[325]
Acute intermittent porphyria	PBGD	AAV	Up to 1.8 × 1013 GC/kg	Single-dose	Preclinical	Insufficient liver transduction at tested doses; overexpression safe in preclinical models	Intravenous injection	Increasing the activity of the deficient enzyme in the liver and restoring the physiological regulation of the pathway	[326]
Methylmalonic acidemia	MMUT and CoA	nanoparticles plus AAV	2.5 × 1012 or 5.0 × 1012 vg/kg	Initial administration and 56th day administration	Preclinical	Well-tolerated	Intravenous injection	Significantly increasing the expression level of MMUT mRNA, and almost completely inhibiting the formation of anti-Anc80 IgG antibodies	[327]
Fabry	α-GLA	AAV	6 × 1012 vg/kg or 3 × 1013 vg/kg	Single-dose	I/II	Generally well-tolerated; immune responses to capsid	Intravenous injection	Resulting in clearance of globotriaosylceramide and globotriaosylsphingosine in plasma, urine, kidney, and heart	[328–330]
MPS type VI	ARSB	AAV	6 × 1012 gc/kg	Single-dose	I/II	Well-tolerated	Intravenous injection	Resulting in sustained ARSB expression and a modest increase in urinary GAGs	[227]
Gangliosidosis GM1	GLB1	AAV	1.5 × 1013 gc/kg	Single-dose	I/II	Well-tolerated	Intravenous injection	Decreasing H3N2b expression level; correlating with the increase in β-galactosidase activity and the improvement in clinical outcomes	[331]
Danon disease	LAMP2B	AAV	6.7 × 1013 GC/kg, and 1.1 × 1014 GC/kg	Single-dose	I/II	Generally well-tolerated; cardiac monitoring required	Intravenous injection	Reducing the left ventricular mass index, preserving left ventricular ejection fraction, and reducing stabilization of the levels of cardiac troponin I and N-terminal pro-B-type natriuretic peptide	[332, 333]
GSD1a Von Gierke	G6PC	AAV	2 × 1012 or 6 × 1012 vp/kg	Single-dose	I/II/III	Well-tolerated	Intravenous injection	Maintaining glucose homeostasis, and inhibiting HCA/HCC	[334]
Wilson’s disease	ATP7B	CRISPR/base editing without nucleases	N/A	Single-dose	I/II	Liver enzyme elevations; copper metabolism fluctuations	Intravenous injection	Resulting in extensive liver repopulation and ameliorating liver injury and copper metabolism	[335, 336]
Crigler-Najjar	UGT1A	AAV	1.0 × 1011 vg/kg; 1.0 × 1012 vg/kg or 5 × 1012 vg/kg	Twice; 0 week and 3 to 5 weeks	Preclinical	Bilirubin fluctuations; liver enzyme elevations	Intravenous injection	Decreasing in plasma bilirubin	[337]

Polycationic applicable for liver therapies

Cationic polymers form nanoparticles through electrostatic interactions with nucleic acids, facilitating cellular uptake and intracellular delivery. Among the most widely used cationic polymers is polyethylenimine (PEI), renowned for its high transfection efficiency. PEI-based nanoparticles and their modified derivatives have been extensively investigated in preclinical studies for delivering various therapeutic nucleic acids, including small interfering RNA (siRNA) and microRNAs (miRNAs), particularly in the context of liver malignancies [92]. However, the clinical translation of PEI has been severely limited by concentration-dependent cytotoxicity. Quantitative studies demonstrate that branched PEI exhibits higher toxicity than linear PEI, with IC₅₀ values of 37 µg/mL and 74 µg/mL, respectively, in A431 human epithelial cells [93]. This cytotoxicity arises from PEI’s high cationic charge density, which induces mitochondrial membrane destabilization and plasma membrane disruption. Branched PEI architectures cause greater impairment of mitochondrial oxidative phosphorylation and inhibition of cytochrome c oxidase than linear PEI, leading to ATP depletion, disruption of redox homeostasis via NADPH and glutathione depletion, and ultimately apoptotic and necrotic cell death [93, 94]. High-molecular-weight PEI derivatives, such as 25 kDa branched PEI, exemplify this “PEI dilemma”: enhanced transfection efficiency is invariably accompanied by proportionally increased cytotoxicity, posing an insurmountable barrier to clinical application [95].

To overcome this fundamental limitation, two primary mitigation strategies have been developed. PEGylation reduces the surface cationic charge density, thereby decreasing plasma protein binding, immunogenicity, and cytotoxicity—resulting in viability increases of 10–20% compared to non-modified particles—while improving pharmacokinetics [96]. However, excessive PEGylation compromises transfection efficiency by attenuating cellular internalization and endosomal escape due to steric shielding effects [97]. Alternatively, biodegradable PEI derivatives incorporating cleavable linkages (ester, disulfide, amide) enable intracellular degradation into low-molecular-weight, non-toxic fragments. Disulfide cross-linking is particularly effective, exploiting the approximately 1,000-fold higher glutathione concentration in intracellular versus extracellular environments to trigger rapid reduction and nucleic acid release while maintaining extracellular stability. Such biodegradable modifications achieve substantial toxicity reduction without impairing—or even enhancing—transfection efficacy [98]. Notably, galactosylated derivatives of PEI target the asialoglycoprotein receptor (ASGPR), which exhibits extraordinary recycling capacity (~ 15 min per cycle, up to 200 cycles per receptor lifetime) and rapid ligand dissociation in acidic endosomes (pH ~ 5.0) [99]. However, the therapeutic efficiency of ASGPR-targeted delivery depends critically on ligand multivalency and density. Recent studies demonstrate that trivalent N-acetylgalactosamine (GalNAc) ligands with optimized spacing exhibit the highest binding affinity and endocytic efficiency, whereas excessive ligand density on nanoparticles drives clathrin-mediated lysosomal degradation, and low-density formulations shift to caveolae-mediated pathways with altered intracellular trafficking [100]. While galactosylated PEI demonstrates hepatocyte-specific transduction ex vivo—such as 10⁴–10⁵-fold higher efficiency in HepG2 cells compared to ASGPR-negative fibroblasts—its ligand substitution remains variable (3.5%–31% of amino groups) without systematic optimization of valency or spacing comparable to tri-GalNAc standards [101, 102].

In contrast, GalNAc-siRNA conjugates and GalNAc-LNP systems have achieved clinical validation after decades of optimization. Recent clinical and preclinical data demonstrate that optimized GalNAc-siRNA conjugates achieve potent gene silencing with IC₅₀ values in the picomolar range—for example, 1 pM for targeting the DNAJB1-PRKACA fusion in fibrolamellar hepatocellular carcinoma—and result in a 60% reduction in tumor growth in patient-derived xenograft models [103]. Novel structural modifications, such as ribofuranose-incorporated GalNAc clusters (G5), further enhance liver specificity and therapeutic durability compared to approved triantennary GalNAc constructs (L96), achieving 95.6% suppression of PCSK9 protein and superior liver-to-kidney distribution ratios [104]. GalNAc-LNP systems have similarly advanced, with recent formulations achieving over 90% liver targeting efficiency and enabling CRISPR-Cas9 gene editing with up to 89% target protein reduction in non-human primates [105]. Unlike these clinically validated platforms, galactosylated PEI retains the immunogenicity and cytotoxicity inherent to high-molecular-weight PEI scaffolds and lacks quantitative benchmarking against modern GalNAc-conjugate standards. Therefore, the clinical translation of ASGPR-targeted PEI requires rigorous optimization of ligand multivalency, density-controlled endocytic routing, and direct efficacy comparisons against validated GalNAc-siRNA and GalNAc-LNP platforms.

Despite these advances, PEI-based formulations—including those delivering siRNA and miRNA for liver malignancies—have not yet advanced to clinical trials. This underscores the persistent challenge of balancing transfection efficiency with systemic safety in human therapeutic applications.

Beyond PEI-based systems, poly(amidoamine) (PAMAM) dendrimers represent a prominent class of cationic polymeric nanocarriers with substantial preclinical evidence supporting their application in liver malignancy gene therapy. These highly branched, monodisperse macromolecules possess repetitive amidoamine branches radiating from an ethylenediamine core, with surface primary amines enabling efficient nucleic acid condensation and cellular internalization [106]. In hepatocellular carcinoma (HCC) models, lactobionic acid-decorated PAMAM dendrimers have demonstrated specific targeting of the ASGPR, achieving marked inhibition of orthotopic HCC xenograft growth when delivering siRNA against astrocyte elevated gene-1, particularly in combination with all-trans retinoic acid [107]. Furthermore, antibody-functionalized PAMAM dendrimers conjugated with epidermal growth factor, human serum albumin, and the humanized monoclonal antibody h-R3 have exhibited enhanced endosomal escape and tumor-targeted delivery capabilities, resulting in efficient polo-like kinase-1 silencing and significant inhibition of HepG2 cell proliferation, migration, and invasion. Recent investigations have also explored supramolecular self-assembly nanoparticles composed of adamantane-grafted PAMAM and cyclodextrin-grafted PEI for chimeric antigen receptor T-cell engineering against hepatocarcinoma, demonstrating high transient transfection efficiency with low toxicity [107].

Poly(lactic-co-glycolic acid) (PLGA) nanoparticles constitute another extensively investigated biodegradable platform for nucleic acid delivery to hepatic malignancies. This FDA-approved copolymer offers excellent biocompatibility, controlled degradation kinetics, and versatile surface modification capabilities [108]. PEGylated PLGA nanoparticles co-encapsulating antisense-miRNA-21 and gemcitabine have demonstrated superior therapeutic efficacy in HCC cells compared to single-agent formulations, achieving dose-dependent reduction in Hep3B and HepG2 cell viability through cell cycle arrest at S-phase and upregulation of PTEN expression [108]. The versatility of PLGA systems extends to targeted delivery approaches: asialofetuin-conjugated cationic PLGA/DOTAP nanoparticles have achieved receptor-mediated gene delivery to liver cancer cells with enhanced transfection efficiency even in the presence of 60% fetal bovine serum, demonstrating the potential for in vivo applications [109]. Additionally, liver-targeting peptide-conjugated PLGA nanoparticles have shown high specificity for hepatocytes through binding to heparan sulfate proteoglycans, achieving 1.8- to 2.0-fold greater hepatic accumulation compared to non-targeted formulations [110].

Despite these promising preclinical advances, neither PAMAM dendrimers nor PLGA-based formulations delivering nucleic acids for liver malignancies have progressed to clinical trials, underscoring the persistent challenges of translating polymeric nanoparticle systems from bench to bedside. Key limitations include the inherent cytotoxicity of higher-generation cationic dendrimers, the complexity of scalable manufacturing, and the need for rigorous optimization of ligand density and surface charge to balance transfection efficiency with systemic safety profiles [107].

LNP applicable for liver therapies

LNP structural foundation, self-assembly principle, and core advantages

LNPs are formulated based on principles that closely mirror the compositional blueprint of natural cell membranes, specifically through the use of amphiphilic lipids. When introduced into an aqueous environment, these lipid molecules spontaneously self-assemble into hollow spherical structures that precisely encapsulate a hydrophilic core. This process is governed by the amphiphilic packing parameter (P), defined as P = v / (a₀ · l_c), where v represents the hydrophobic tail volume, a₀ is the optimal headgroup area, and l_c is the critical tail length [111]. For typical LNP ionizable lipids with conical geometry (P < 1/3), this thermodynamic drive favors the formation of highly curved bilayer vesicles or micellar structures, enabling the precise encapsulation of hydrophilic nucleic acid cargo within the aqueous core, while the hydrophobic lipid tails segregate to minimize unfavorable interactions with the aqueous phase [112].

The translation of these thermodynamic principles into clinically viable formulations requires precise control over the self-assembly kinetics, which is achieved through microfluidic mixing techniques that have emerged as the predominant approach for scalable LNP manufacturing. Unlike conventional bulk mixing methods that produce heterogeneous populations with variable encapsulation efficiency, microfluidic platforms enable rapid, controlled mixing of lipid and aqueous phases at the millisecond scale. T-junction and Y-junction mixers facilitate turbulent mixing at high flow rates (40–60 mL/min), while staggered herringbone micromixers employ asymmetric herringbone groove patterns to create chaotic advection, achieving controlled mixing in < 10 ms [113]. These designs facilitate the production of LNPs with narrow size distributions (typically 50–100 nm), low polydispersity indices (< 0.2), and high encapsulation efficiencies (> 90%) that are essential for clinical translation [114–116]. By finely tuning process parameters such as the aqueous-to-organic flow rate ratio and total flow rate, researchers can systematically modulate particle size and physicochemical properties. Recent advances have addressed the scalability challenge inherent to single-channel microfluidic devices through parallelized architectures; for instance, a 128-channel parallelized microfluidic device incorporating staggered herringbone mixers has achieved production rates of 18.4 L/h, representing a > 100-fold increase over conventional single-channel systems while maintaining precise control over LNP quality [117]. These continuous-flow methods offer significant advantages over traditional bulk mixing approaches, including improved reproducibility, reduced reagent consumption, and straightforward scale-up from laboratory to industrial production [118].

A critical determinant of LNP functionality is the pKa of the ionizable lipid, which must be carefully optimized to balance two competing requirements: at physiological pH (~ 7.4), the lipid must remain approximately neutral to minimize systemic toxicity and prevent rapid clearance, while in the acidic endosomal environment (pH ~ 5.0–6.0), it must become positively charged to facilitate endosomal escape through membrane destabilization. Empirical studies have established that ionizable lipids with pKa values between 6.0 and 6.5 achieve optimal transfection efficiency, as this range ensures adequate protonation in endosomes while maintaining sufficient neutrality in circulation [119, 120].

LNPs possess four fundamental advantages: excellent biocompatibility, biodegradability, generally favorable but formulation-dependent safety profiles, and tunable physicochemical properties [121]. Additionally, their structural versatility allows for facile functionalization and scalable, reproducible manufacturing via established industrial processes. These attributes collectively render LNPs the most promising and widely adopted delivery platform for nucleic acid therapeutics [122]. In recent years, LNPs have experienced a resurgence in gene therapy and now serve as the principal delivery vehicles—or molecular couriers—for both siRNA and mRNA-based therapies (Fig. 3) [123].

Fig. 3.

Mechanism of delivery and action of therapeutic mRNAs. LNPs represent the most effective nanocarrier system for liver-targeted therapy following systemic administration. A typical LNP formulation comprises ionizable cationic lipids containing amine groups, cholesterol, PEG-lipid conjugates, helper phospholipids such as phosphatidylcholine, and the active mRNA molecule. Due to the negatively charged nature of mRNA, its encapsulation within LNPs is primarily achieved through electrostatic interactions with the positively charged ionizable lipids. After systemic administration, LNPs undergo a process known as corona formation, during which surface-associated proteins are exchanged with those present in the bloodstream. This results in the incorporation of ApoE, a key determinant for hepatocyte uptake. The presence of ApoE facilitates hepatic uptake of LNPs via interactions with the liver-enriched LDLR. Additionally, LNPs modified with GalNAc-lipid conjugates enable liver-specific delivery through the ASGPR, bypassing the LDLR-dependent pathway. Upon cellular internalization, LNPs are trafficked into acidic endosomal compartments. The ionizable lipids within the LNP become protonated in this low-pH environment, leading to endosomal membrane destabilization and subsequent endosomal escape. This critical step enables the release of the encapsulated mRNA into the cytoplasm, where it is translated by the host translational machinery into the desired therapeutic protein. Depending on the specific design and targeting signals of the mRNA construct, the resulting protein can remain localized within specific subcellular compartments of hepatocytes or be secreted into systemic circulation, thereby exerting therapeutic effects in extrahepatic tissues or organs

Liver targeting strategy and application in liver diseases

One of the major challenges in translating LNP-based therapies into clinical applications is the development of tissue-specific delivery systems that are both highly efficient and precisely targeted. Traditional strategies typically involve the physical or chemical conjugation of ligands specific to certain receptors onto the nanoparticle surface. For example, a targeting approach has been developed by binding a multivalent GalNAc cluster to the LNP. These nanoparticles have demonstrated the ability to deliver mRNA molecules to hepatocytes with high efficiency, offering potential applications for genetic diseases such as methylmalonic acidemia (MMA) [124], primary hyperoxaluria type 1 (PH1) [125], glycogen storage disease 1a (GSD1a) [126], citrin deficiency [127], acute intermittent porphyria (AIP) [128], maple syrup urine disease [129], arginase deficiency [130], OTCD [131], and progressive familial intrahepatic cholestasis type 3 (PFIC3) [132]. However, detailed preclinical efficacy data have only been published for MMA, demonstrating dose-dependent biomarker rescue (75–90% reduction in plasma methylmalonic acid) and improved survival with repeated dosing over 12 weeks. The duration of therapeutic protein expression from a single dose of mRNA-LNP lasted 2–3 weeks, necessitating repeat administration to sustain a curative effect [124, 133, 134].

A classic example of liver-targeted LNPs is Onpattro (patisiran), which utilizes a precisely engineered nanocapsule composed of an ionizable cationic lipid, phospholipids, cholesterol, and a PEGylated lipid (Fig. 3). This formulation encapsulates a siRNA specifically targeting transthyretin (TTR) and is indicated for the treatment of TTR-mediated amyloidosis [135]. Upon intravenous administration, the LNP is opsonized in vivo by endogenous apolipoprotein E (ApoE), acquiring a lipoprotein-like surface coating. This facilitates recognition by the low-density lipoprotein receptor (LDLR), enabling receptor-mediated endocytosis into hepatocytes [136].

However, this ApoE-mediated targeting mechanism faces significant competition from endogenous lipoproteins, which circulate at high concentrations (1–2 mg/mL for LDL) and bind LDLR with affinities comparable to or exceeding those of LNP-ApoE complexes. This competition can substantially reduce hepatocyte uptake efficiency, particularly in hyperlipidemic patients where elevated circulating lipoprotein levels may saturate available LDLR binding sites. Quantitative studies indicate that only a fraction of administered LNP dose successfully engages hepatocyte LDLR, with the majority being cleared by the reticuloendothelial system or remaining in circulation [72, 137]. To address these delivery limitations, several strategies have been developed to enhance hepatocyte uptake. These include transient reduction of circulating lipoproteins through fasting or lipid-lowering agents, or engineering LNP formulations with enhanced LDLR-binding affinity through optimized ApoE presentation or alternative ligand incorporation [138].

The clinical success of this liver-targeted approach is exemplified by patisiran’s regulatory approval and therapeutic application. By suppressing the production of mutant TTR and thereby reducing amyloid fibril deposition, patisiran was approved by the FDA in 2018 as the first siRNA therapeutic, specifically for hereditary transthyretin-mediated amyloidotic polyneuropathy. However, its clinical administration requires rigorous premedication protocols to mitigate infusion-related reactions, including corticosteroids, acetaminophen, and antihistamines prior to each infusion [139]. Despite these precautions, infusion-related adverse events—such as flushing, back pain, nausea, and headache—occur in a subset of patients, necessitating prolonged observation periods and occasionally dose modifications. Long-term follow-up data from the APOLLO extension study indicate generally favorable tolerability profiles over multiple years, with sustained TTR suppression and manageable safety signals, although concerns regarding peripheral edema and hepatic enzyme elevations warrant ongoing monitoring [140].

Building upon the success of patisiran, this liver-targeted delivery strategy is now being adapted for other hepatocyte-associated diseases, particularly hypercholesterolemia. siRNA-loaded LNPs are currently under clinical investigation for silencing PCSK9 and ApoE—two key regulators of low-density lipoprotein (LDL) cholesterol metabolism—with the aim of lowering circulating LDL cholesterol levels and managing hypercholesterolemia [135].

Nevertheless, translating these LNP-based therapies from rodent models to humans presents substantial challenges due to significant species differences in ApoE corona formation and LDLR biology. Wild-type mice express a single ApoE isoform that is structurally distinct from human ApoE isoforms, differing by approximately one-third of its nearly 300 amino acids and functionally resembling human ApoE3 rather than ApoE2 or ApoE4 [141]. More critically, the human population exhibits three distinct ApoE isoforms with markedly divergent LDLR-binding properties, humans exhibit three isoforms (ApoE2, ApoE3, ApoE4) with markedly different LDLR-binding affinities: ApoE2 binds to LDLR with 50- to 100-fold weaker affinity than ApoE3 or ApoE4, potentially impairing LNP uptake in ApoE2 carriers [142]. Furthermore, ApoE4 shows preferential association with VLDL particles and altered endosomal recycling compared to ApoE3, which may affect LNP trafficking and endosomal escape [143]. These interspecies and interindividual differences necessitate sophisticated preclinical models. Specifically, humanized APOE knock-in models (APOE2-TR, APOE3-TR, APOE4-TR) are required to accurately predict LNP efficacy across human genotypes [144].

LNPs carrying siRNAs are being investigated for the treatment of HCC [145] and chronic hepatitis B virus (HBV) infection [146]. In oncology, siRNAs targeting oncogenic drivers such as MYC and polo-like kinase-1 (PLK1) have entered phase I/II clinical trials, demonstrating acceptable tolerability but limited antitumor efficacy to date [147]. This limited efficacy arises from multiple technical barriers inherent to delivery within solid tumors. First, tumor penetration barriers exist, including a dense stromal extracellular matrix composed of fibroblasts, growth factors, and adhesion molecules that physically impede LNP diffusion, alongside vascular abnormalities and hypoxic tumor microenvironments that reduce bioavailability [148]. The second challenge is the immunosuppressive microenvironment, where elevated myeloid-derived suppressor cells and regulatory T cells in the HCC tumor microenvironment consume L-arginine and secrete TGF-β and IL-10, creating cascading immune suppression that diminishes therapeutic responses [149].

Additionally, dose limitations arise due to clearance by the reticuloendothelial system in the liver and spleen, causing rapid systemic clearance and necessitating repeated high-dose administrations that risk innate immune activation and interferon signaling via TLR activation [150]. Beyond TLR signaling, LNP-siRNA formulations activate multiple innate immune pathways. PEGylated components trigger complement activation-related pseudoallergy via anti-PEG antibodies, causing anaphylatoxin release (C3a/C5a) and acute inflammatory responses [151]. Furthermore, cytosolic nucleic acid delivery activates the cGAS–STING pathway. Ionizable lipids and dsDNA impurities trigger cGAS-dependent cGAMP production, inducing type I interferon and pro-inflammatory cytokines. This creates a dual-edged effect: dose-limiting toxicities in chronic viral infections versus potential antitumor immunity in oncology [152, 153]. These convergent challenges necessitate strategies such as optimized lipid compositions, STING antagonists, and immunosuppressive co-therapies to improve therapeutic indices. Another significant limitation is endosomal escape, as only approximately 1% of internalized RNA therapeutics successfully escape from endosomes into the cytoplasm. The spontaneous and transient nature of endosomal membrane breaches (< 10 nm) creates a major rate-limiting step for RISC loading and target mRNA silencing [70]. These convergent challenges necessitate advanced formulation strategies—such as ultra-small LNPs (< 60 nm) for stromal penetration, pH-responsive systems for tumor microenvironment-specific release, and combination therapies with immune checkpoint inhibitors—to overcome the physical and biological barriers limiting siRNA efficacy in HCC [103].

For chronic HBV infection, an additional critical consideration is vector genome dilution during hepatocyte proliferation. Unlike integrated viral DNA or gene therapy vectors that permanently modify the host genome, siRNA-LNP formulations do not integrate and therefore do not replicate with dividing hepatocytes. The adult liver is largely quiescent under normal conditions, with hepatocyte turnover occurring slowly (average lifespan of 200–300 days) and only 1–2% of hepatocytes dividing at any given time [154]. However, during chronic HBV infection, hepatocyte turnover can be significantly accelerated due to ongoing necroinflammation, compensatory regeneration, and fibrotic remodeling. This heightened proliferative activity creates a “dilution effect” where non-integrated therapeutic agents are progressively lost as transduced hepatocytes divide, potentially diminishing the durability of RNAi-mediated viral suppression over time [155]. This phenomenon is particularly relevant for pediatric applications or patients with advanced fibrosis/cirrhosis where hepatocyte proliferation is elevated, and may necessitate repeated dosing regimens or the development of sustained-release formulations to maintain therapeutic efficacy [156].

In the context of HBV, a multi-siRNA LNP formulation containing three chemically modified siRNAs targeting distinct viral transcripts has completed Phase II clinical evaluation. This approach aims to suppress viral antigen expression and interrupt the HBV replication cycle, offering a potential RNAi-based therapy for chronic hepatitis B [157].

Viral vectors applicable for liver disease

During the preparation of viral vectors for gene therapy, genetic elements essential for viral replication and pathogenicity are typically deleted, and therapeutic genetic sequences are subsequently inserted into the viral genome. However, in the context of oncolytic viruses—where the goal is to enable active infection and subsequent lysis of tumor cells—partial viral replication capacity is deliberately retained to achieve the desired therapeutic effect [12]. The most commonly used viral vector platforms for liver-directed gene therapy include AAV, adenovirus, retrovirus, and lentivirus. The selection of a specific vector system depends on the therapeutic objectives, target cell tropism, transgene expression duration, and safety considerations (Table 3).

AAVs applicable for liver therapies

AAV genome structure and vector design principles

The genome of AAV consists of a single-stranded DNA molecule approximately 4.7 kilobases in length, encoding two primary genes: rep, which produces proteins involved in viral replication, and cap, responsible for synthesizing viral capsid proteins [158]. Flanking the genome are inverted terminal repeat (ITR)s contain fold into hairpin secondary structures. These ITRs harbor essential cis-acting elements, including origins of replication, termination signals, and packaging signals required for viral genome encapsidation [159]. AAV possesses several characteristics that make it an attractive vector for gene therapy applications: it is replication-deficient, not associated with any known human pathology, and requires co-infection with a helper virus—such as adenovirus—for productive replication and virion assembly. In wild-type AAV, viral genomes can integrate site-specifically into the host genome at a locus known as AAVS1, where they typically remain latent. However, this integration strictly depends on the presence of the viral rep protein [160]. To generate rAAV vectors, the viral rep and cap genes are removed from the vector genome and replaced with a transgene expression cassette flanked by the ITRs. During rAAV production, the rep and cap genes are supplied in trans to enable vector particle formation. Crucially, the absence of rep sequences in the final rAAV vector abrogates site-specific integration at AAVS1. Instead, rAAV genomes predominantly persist as episomal concatemers in the nucleus, with rare integration events occurring randomly throughout the genome at extremely low frequencies (estimated at < 0.1% of transduced cells), thereby significantly reducing—but not completely eliminating—the risk of insertional mutagenesis [161]. Moreover, by employing different cap genes—either naturally occurring or engineered—diverse AAV capsid variants can be generated, each displaying distinct serological properties and tissue tropisms (Fig. 4) [162]. Importantly, in preclinical animal models, most AAV serotypes elicit minimal immunogenicity, allowing efficient and long-term transgene expression with limited host immune recognition. rAAVs demonstrate high transduction efficiency in non-dividing cells, such as hepatocytes. However, clinical studies have shown that systemic administration of rAAV vectors can trigger T-cell-mediated immune responses directed against capsid proteins, necessitating the use of immunosuppressive agents, such as corticosteroids, to prevent clearance of transduced cells and sustain therapeutic efficacy [163].

Fig. 4.

Transduction mechanism of recombinant AAV vectors in hepatocytes. The process of AAV transduction begins with the binding of the viral capsid to primary glycan receptors on the cell surface, followed by interaction with the AAVR, which is utilized by most AAV serotypes. The AAV virion is then internalized via endocytosis and trafficked to acidified endosomes. Within this compartment, the phospholipase domain of the VP1 protein facilitates membrane penetration, enabling the virus to escape from the endosomal lumen into the cytoplasm. AAV particles that evade proteasomal degradation are subsequently transported to the nucleus. Inside the nucleus, capsid uncoating occurs, releasing the single-stranded DNA genome. The AAV genome is converted into a double-stranded, circular intermediate through host cell DNA repair mechanisms. This double-stranded form serves as a transcriptionally active episomal template, supporting the expression of the therapeutic transgene and the subsequent synthesis of the target protein

Liver tropism differences and cross species transduction

Following intravenous administration, nearly all AAV serotypes exhibit a natural hepatotropic tendency, preferentially accumulating in the liver—a phenomenon commonly referred to as hepatic tropism [164]. However, the extent of viral particle accumulation within the liver does not necessarily correlate with the level of transgene expression achieved. This discrepancy arises because productive transduction requires the sequential completion of three rate-limiting steps: receptor binding, intracellular trafficking, and nuclear uncoating/genome conversion—each of which exhibits species-specific variability. Receptor binding establishes tropism. While AAV2 utilizes heparan sulfate proteoglycans, AAV8 and AAV9 bind the laminin receptor and require adeno-associated virus receptor (AAVR) (KIAA0319L) for cellular entry [165]. AAV3B preferentially engages the human hepatocyte growth factor receptor (c-MET), explaining its superior transduction efficiency in humanized livers [166]. Intracellular trafficking determines the post-entry fate of the virus. Murine hepatocytes route AAV through Rab5/Rab7-positive late endosomes toward degradation, whereas human and NHP cells favor Rab11-positive recycling endosomes, facilitating viral escape [167]. The viral protein 1 (VP1) phospholipase A2 domain mediates endosomal membrane disruption upon acidification [168]. Nuclear events constitute the final bottleneck. Capsid uncoating requires ubiquitin-proteasome activity and nuclear proteases [169], while genome conversion depends on overcoming FKBP52-mediated inhibition. Phosphorylated FKBP52 binds AAV ITR D-sequences, blocking second-strand synthesis. Elevated levels of phosphorylated FKBP52 in murine hepatocytes delay transcription [170].

Moreover, the degree of hepatotropism varies among AAV serotypes and is also influenced by the host species. Transduction efficiencies observed in mice often do not reliably translate reliably to other species [171]. Significant differences in transduction efficiency have been observed among various AAV serotypes when comparing human or primate hepatocytes to murine hepatocytes, particularly in non-human primate (NHP) models and in mice engrafted with humanized livers [172]. NHPs share highly conserved liver cell surface receptors—such as AAVR, HSPG, and EGFR—with humans, which directly affect the transduction efficiency of AAV capsids. For example, the transduction efficiency of AAV8 in crab-eating macaques and human primary hepatocytes is significantly higher than in mouse models, as mice lack corresponding receptor homologs. NHPs can accurately simulate the human immune clearance mechanisms against AAV vectors, including complement activation thresholds, Kupffer cell phagocytic activity, and the kinetics of nAb production. Studies have shown that the nAb titer induced by AAV-LK03 in rhesus monkeys closely matches clinical patient data, whereas the rodent model underestimates immunogenicity risk [173, 174]. The size of fenestrations in hepatic sinusoidal endothelial cells, hemodynamics, and extracellular matrix components in NHPs are highly similar to those in humans, directly influencing the delivery efficiency of AAV particles to liver parenchymal cells. For example, due to the size limitation of fenestrations in rhesus monkeys, AAV5 primarily accumulates in endothelial cells, a phenomenon consistent with autopsy data from humans [175]. The NHP model demonstrates excellent clinical translational value in the dose-response relationship of AAV liver-targeted therapy. The clinical starting dose of Zolgensma^® (AAV9-SMN1) is directly based on the PD/PK data of rhesus monkeys, with a prediction error rate of less than 30%, far superior to the over 200% deviation observed in rodent models [176].

However, NHP models possess important limitations regarding immune repertoire diversity that may affect their predictive validity. Rhesus macaques exhibit substantially greater immunoglobulin diversity than humans: their IGHD locus contains 39 genes (vs. 23 in humans), and the IGHV3 family is significantly expanded with 38 genes and 112 alleles identified in just 10 macaques. This increased germline diversity—several-fold higher than human IMGT database entries—can lead to the generation of antibody responses against AAV epitopes that may not be representative of the human response [177, 178]. Additionally, rhesus macaques possess expanded MHC class I and II regions with multiple gene copies per haplotype (up to three MHC-A genes), creating more complex antigen presentation landscapes than humans. These immunological differences may result in NHPs mounting qualitatively distinct adaptive immune responses—both cellular and humoral—to AAV vectors, potentially overestimating or altering the epitope specificity of neutralizing antibody responses compared to human patients [179]. Consequently, while NHPs remain indispensable for assessing vector biodistribution and acute toxicity, the divergence in immune repertoire diversity necessitates cautious interpretation of immunogenicity data and supports the development of humanized mouse models with human immune system engraftment as complementary preclinical platforms [180].

Shell engineering and optimization of liver specific carriers

To generate more liver-specific AAV vectors, researchers employ diverse engineering strategies, including directed evolution, barcoded capsid libraries, and computational capsid design approaches [181]. Directed evolution involves constructing randomly mutated capsid libraries through two primary methods: inserting random peptide sequences into defined surface-exposed capsid regions and performing DNA shuffling across capsid genes from different serotypes. Barcoded capsid libraries enable high-throughput screening of thousands of variants simultaneously, allowing rapid identification of capsids with desired tropism profiles. Computational design approaches, including machine learning algorithms and structure-guided engineering, further expand the rational design of liver-specific vectors by predicting capsid-receptor interactions and immunogenicity [74]. Another important vector engineering strategy involves the use of self-complementary AAV (scAAV) vectors. Unlike conventional single-stranded AAV (ssAAV), scAAV vectors package a dimeric inverted repeat genome that self-anneals into double-stranded DNA upon cellular entry, bypassing the rate-limiting step of second-strand synthesis. This modification results in 5- to 140-fold higher transduction efficiency and more rapid transgene expression compared to ssAAV. However, the packaging capacity of scAAV is limited to approximately 2.3–2.5 kb—half that of ssAAV (4.7 kb)—restricting its use to smaller transgenes. The scAAV genome is generated by mutating or deleting the terminal resolution site in one ITR, which prevents Rep-mediated cleavage and forces packaging of the dimeric genome. scAAV vectors have demonstrated particular utility in liver-directed gene therapy for hemophilia B, where a mini-hFIX cassette (1.6 kb) packaged in scAAV8 achieved therapeutic expression levels at significantly lower doses than ssAAV counterparts in both mice and non-human primates [182, 183].

AAV manufacturing faces significant scalability challenges. Yield variability from transient transfection—due to variable plasmid uptake and cellular stress—creates batch-to-batch inconsistencies that complicate process validation [184]. Capsid heterogeneity presents another hurdle: up to 70–90% of assembled capsids remain empty or contain fragmented DNA, reducing effective dose and increasing immunogenicity risk [185]. Separating empty from full capsids is technically demanding due to similar biophysical properties, with current methods achieving only 60–80% full capsid enrichment. Emerging analytical technologies such as mass photometry enable rapid capsid ratio monitoring, but standardization remains incomplete [186, 187]. These constraints limit dose scalability and increase cost of goods for liver-directed therapies.

Although humanized mouse models cannot fully recapitulate the complexity of the human immune response to AAV, they remain the most relevant preclinical platform for optimizing liver-targeted vectors. This iterative selection process led to the development of AAV-LK03, a capsid variant with significantly enhanced tropism for human hepatocytes. AAV-LK03 has since advanced into clinical trials for the treatment of Hemophilia A (HemA). Notably, its capsid sequence exhibits 98.9% homology to AAV3B, a naturally occurring serotype previously deemed suboptimal due to modest transduction efficiency in murine models [33]. This high sequence similarity underscores that few amino acid changes—specifically seven surface-exposed residue substitutions concentrated in the VP3 variable region—are sufficient to dramatically shift tropism from murine to human hepatocytes [188]. These substitutions likely enhance binding affinity for human-specific receptors such as c-MET, rather than altering the fundamental entry mechanism. Importantly, the near-identical capsid sequence raises considerations regarding immune cross-reactivity: pre-existing nAbs against AAV3B in human populations may similarly recognize AAV-LK03, potentially limiting transduction efficiency in seropositive individuals [173].

To overcome pre-existing immunity, two complementary approaches have been developed: capsid engineering to evade antibody recognition, and pharmacological interventions to transiently remove or suppress neutralizing antibodies. Capsid engineering strategies include epitope masking, directed evolution under immune pressure, and structure-guided rational design. Epitope masking modifies immunodominant variable regions (VR-IV, VR-V, VR-VIII) via site-directed mutagenesis or PEGylation to shield antibody-binding sites [189, 190]. Directed evolution under immune pressure selects AAV libraries in the presence of neutralizing antibodies, such as intravenous immunoglobulins (IVIG), to isolate variants with mutations at antigenic sites. For example, AAV-DJ was generated by DNA shuffling of AAV2, 4, 5, 8, 9 and three non-human serotypes under IVIG selection, demonstrating enhanced liver transduction in antibody-positive mice [191, 192]. AAV-S3 incorporates additional capsid modifications to improve manufacturability, reduce immunogenicity, and further enhance human hepatocyte transduction, representing the next generation of liver-directed AAV gene therapy vectors. Structure-guided rational design yields variants such as AAV3B-V04 and hAEV6-Q588R, which incorporate strategic substitutions at antibody-binding epitopes to reduce seroreactivity by 5- to 20-fold while maintaining hepatocyte tropism [151, 166, 193]. Recent efforts to address this limitation have led to the development of AAV-S3, an engineered derivative of AAV3B currently undergoing Phase I/II clinical trials for Hemophilia B (HemB) and Fabry disease [194]. Additional strategies include the use of empty capsids as decoys to saturate neutralizing antibodies, FcRn inhibitors to reduce IgG half-life, and tolerogenic rapamycin nanoparticles to suppress adaptive immune responses [195].

Pharmacological strategies include plasmapheresis, which non-specifically removes circulating antibodies and improves transduction in non-human primates [196]. More specific approaches include immunoadsorption with AAV-specific columns, which selectively removes anti-AAV antibodies while preserving other immunoglobulins. IgG-degrading enzymes such as imlifidase (IdeS) and IdeZ, derived from Streptococcus pyogenes and Streptococcus equi respectively, cleave IgG at the hinge region, generating F(ab’)2 and Fc fragments. These enzymes have demonstrated efficacy in rescuing AAV transduction in seropositive mice and non-human primates by transiently degrading neutralizing antibodies, with optimal dosing regimens achieving up to 77% reduction in nAb levels [197, 198].

AAV clinical translation in liver diseases

With multiple advantageous attributes, AAV has emerged as the preferred vector for liver-directed gene therapy. In the clinical pipeline, the most prominent indications remain HemA and HemB, where the liver is engineered to function as a biofactory capable of sustained secretion of the deficient clotting factors [199]. Since the earliest trials employing AAV2 to deliver factor IX, researchers have successively evaluated a range of natural and engineered AAV serotypes—including AAV5, AAV8, AAVrh10, AAVS3, and AAV-Spark100/200—with mixed outcomes [200]. Critical analysis reveals that these divergent results stem not from serotype selection alone but from the complex interplay of four key determinants of efficacy: promoter architecture and strength, transgene codon optimization, capsid dose and vector genome configuration, and disease-specific pathobiology. Promoter selection fundamentally determines hepatic expression levels and durability. Early HemB trials using AAV2 vectors with the CMV promoter achieved only transient factor IX expression due to promoter silencing and immune responses [201]. In contrast, subsequent studies employing liver-specific promoters—such as the thyroxine-binding globulin (TBG) promoter or hybrid liver-specific promoters—demonstrated sustained expression. The TBG promoter, in particular, enables hepatocyte-restricted transcription, minimizing off-target expression and immune activation while maintaining therapeutic factor levels for years [202]. Codon optimization significantly enhances translational efficiency and protein secretion. A recent meta-analysis showed that codon-optimized B-domain-deleted factor VIII (BDD-FVIII) constructs achieved superior outcomes compared to non-optimized versions, with a pooled standardized mean difference of 1.83 in bleeding rate reduction [203]. Capsid dose and vector genome configuration influence both efficacy and safety margins. High-dose AAV administration (> 2 × 10¹³ vector genomes/kg) is associated with an increased risk of hepatotoxicity, complement activation, and thrombotic microangiopathy, particularly with certain serotypes such as AAV5 [204]. Disease-specific pathobiology represents a fourth critical determinant. In hemophilia, the non-diseased liver parenchyma supports prolonged persistence of AAV episomes, whereas in metabolic liver diseases characterized by ongoing hepatocyte injury or regeneration, vector genomes may be lost during cell division. Baseline inflammation, as observed in advanced liver fibrosis or concomitant infections, impairs AAV intracellular trafficking and nuclear entry through altered endosomal processing and cytosolic innate immune sensing [12, 205].

Although durable transgene expression has been achieved in some cases, other studies have reported suboptimal efficacy or safety concerns. Importantly, these inconsistencies are increasingly attributed not to inter-individual patient variation but to fundamental differences in disease-specific pathobiology—such as hepatocyte turnover rate, baseline inflammatory milieu, intracellular trafficking limitations, or target-gene expression requirements—that differentially affect vector transduction efficiency, promoter activity, and transgene stability [206, 207]. Despite these challenges, robust evidence demonstrates that AAV vectors can achieve prolonged hepatic residence and stable transgene expression in appropriately matched disease contexts, providing a strong rationale for their continued development in inherited metabolic liver diseases.

The first application of AAV vectors in the treatment of inherited metabolic liver diseases is for hypercholesterolemia [208]. In these studies, AAV vectors carrying the very low-density lipoprotein receptor (VLDLR) gene are directly administered into the portal circulation of hypercholesterolemic mice. This intervention resulted in an approximate 40% in both serum cholesterol and triglyceride levels, with the therapeutic effect remaining stable throughout the observation period [209]. Following this pioneering work, research in this area has expanded rapidly. Proof-of-concept studies have since been reported in animal models of various inherited metabolic liver disorders, including acute intermittent porphyria (AIP) [210], primary hyperoxaluria type 1 (PH1) [211], Wilson’s disease (WD) [212], MMA [213], propionic acidemia (PA) [214], phenylketonuria (PKU) [215], tyrosinemia type I (HT1) [216], progressive familial intrahepatic cholestasis type 3 (PFIC3) [217], CNS [218], glycogen storage disease type Ia (GSDIa) [219], and UCDs [220]. Although these studies utilize different AAV serotypes, promoters, and administration routes, they consistently demonstrate sustained therapeutic efficacy without any significant safety concerns.

AAV makes its initial debut in the clinical arena in the treatment of AIP. Phase I trials demonstrate an acceptable safety profile. However, the therapeutic efficacy is modest at best [221]. Currently, numerous AAV-based programs targeting inherited metabolic liver diseases are either underway or in the pipeline, encompassing indications such as PKU [222], OTCD [223], CNS [224], familial hypercholesterolemia [225], Fabry disease [226], mucopolysaccharidosis type VI [227], Pompe disease [228], GSDIa [219], and WD [229]. For example, clinical translation has since advanced significantly, with published Phase 1–2 trial results for CNS showing that AAV8 vector GNT0003 (5 × 10¹² vg/kg) enabled patients to discontinue phototherapy for at least 78 weeks with substantial bilirubin reduction and no serious adverse events [224]. For WD, Vivet Therapeutics’ VTX-801 (AAV5-ATP7B) is currently in Phase 1/2 trials, supported by preclinical data demonstrating normalization of serum ceruloplasmin and reversal of hepatic copper accumulation in knockin mouse models [229]. These developments, alongside established clinical validation from hemophilia A gene therapy (valoctocogene roxaparvovec, showing 84.5% reduction in bleeding rates at 2 years) [230].

The clinical translation of AAV-mediated liver gene therapy raises significant ethical concerns regarding permanent genomic modification, germline transmission risk, and off-target effects. While AAV vectors predominantly persist as episomes, rare random integration events (< 0.1%) and inadvertent germline homing mechanisms (3.7% extrahepatic genomic editing risk) blur the boundary between somatic and germline editing [231]. Off-target effects and mosaicism—where multiple cell populations with different genetic makeups coexist—may lead to unpredictable hepatocyte function and clonal expansion. Preclinical studies have demonstrated that AAV integration can induce hepatocellular carcinoma through insertional mutagenesis near oncogenes such as CCNA2, CCNE1, or TERT, necessitating rigorous long-term monitoring of patients, particularly pediatric recipients [88, 232]. These convergent ethical challenges underscore the imperative for comprehensive informed consent, lifetime patient surveillance, and robust regulatory frameworks [233].

Integrated and editorial strategies of AAV

While conventional AAV-mediated transduction typically represents gene addition—introducing functional transgenes as episomal or randomly integrated elements without correcting the underlying disease allele—recent developments have enabled true gene correction through targeted genomic editing and precision engineering strategies. These integrated and editorial approaches bridge the gap between temporary gene supplementation and permanent genetic cure, each with distinct mechanisms, efficacy profiles, and regulatory considerations.

An alternative strategy involves leveraging intrinsic homologous recombination capabilities without the use of exogenous nucleases. Recently, researchers isolate a novel AAV variant, designate AAV-HSC, from human CD34⁺ hematopoietic stem cells. This vector has been reported to mediate efficient homology-directed repair (HDR) without requiring co-delivery of exogenous nucleases, although the underlying mechanism remains incompletely understood, and these findings have been challenged by independent studies. The editing process is guided by homology arms complementary to the genomic target site and appears to operate exclusively through homologous recombination, as evidenced by the requirement for BRCA2 and the absence of non-homologous end joining (NHEJ)-associated indels or viral ITR integration at the target site [234]. However, these results could not be replicated without the use of targeted nucleases, suggesting that the phenomenon may be context-dependent or require specific experimental conditions [235]. In a murine model of phenylketonuria, AAV-HSC15-mediated insertion of a functional phenylalanine hydroxylase (PAH) transgene achieved sustained phenotypic correction, leading to the initiation of a phase I/II clinical trial evaluating AAV-HSC15 for the treatment of late-onset phenylketonuria in humans [236].

Recently, a significant area of research has focused on using AAV vectors to co-deliver gene-editing nucleases and donor templates, enabling precise genomic corrections through HDR. HDR-based correction of single-nucleotide mutations has already been successfully demonstrated in a humanized mouse model of OTCD [237]. However, the efficiency and precision of this approach are limited by fundamental biological constraints. In mammalian cells, DNA double-strand breaks are primarily repaired through NHEJ, an error-prone pathway that generates unpredictable insertions or deletions, rather than through the desired HDR pathway [238]. This challenge is further compounded by the inherent inefficiency of HDR in non-dividing cells, as it relies on the availability of sister chromatids as repair templates during the S and G2 phases of the cell cycle. Since adult hepatocytes are predominantly quiescent, the therapeutic application of HDR-based strategies in liver diseases is significantly hindered, necessitating the development of methods to induce hepatocyte proliferation or otherwise enhance HDR efficiency in non-dividing cells [239]. Furthermore, AAV-mediated gene editing has also been employed to reduce the expression of specific proteins through strategies such as metabolic pathway redirection (MPR) or substrate reduction therapy. These approaches have been successfully applied in animal models of hypercholesterolemia and primary hyperoxaluria type 1, where AAV vectors effectively downregulate the expression of PCSK9 and HAO1, respectively [240, 241]. More recently, AAV vectors carrying the PiggyBac transposon system have been developed as an alternative strategy for achieving long-term genetic correction [242, 243]. In murine models of progressive PFIC3 and UCDs, the AAV-PiggyBac system has demonstrated sustained gene expression and therapeutic efficacy [244]. Additionally, in a diabetic mouse model, this system has been used to deliver the insulin gene, resulting in normalization of blood glucose levels and restoration of glucose tolerance [245].

Emerging precision genome editing technologies—namely base editing and prime editing—offer significant safety advantages for liver-directed in vivo correction by operating independently of DNA double-strand break formation. CBEs and ABEs enable direct C·G-to-T·A or A·T-to-G·C conversions, respectively, without requiring homology-directed repair or generating indels. In a phenylketonuria mouse model, dual AAV-mediated delivery of an intein-split Staphylococcus aureus KKH-CBE3 achieved approximately 23% on-target correction of the disease-causing Pahenu2 mutation, reversing the disease phenotype without detectable transcriptome-wide or genome-wide off-target mutations [246, 247]. Similarly, ABE-mediated correction of the same mutation has demonstrated therapeutic efficacy. Notably, LNP delivery of base editor mRNA has emerged as a promising alternative to viral vectors, enabling transient editor expression and redosing capability while minimizing immunogenicity and off-target risks. In non-human primates, LNP-mediated delivery of an adenine base editor targeting PCSK9 achieved over 60% hepatic editing, resulting in approximately 90% reduction in plasma PCSK9 levels and a 60% reduction in LDL cholesterol [248, 249]. Prime editing, which employs a Cas9 nickase fused to reverse transcriptase along with a prime editing guide RNA, enables the direct installation of all possible base transitions as well as small insertions and deletions without inducing double-strand breaks. This approach has shown promise in liver-directed applications. For example, AAV-mediated delivery of a size-reduced prime editor achieved 15% editing efficiency at the Dnmt1 locus in neonatal mouse liver. Similarly, adenoviral delivery of the same system corrected the Pahenu2 mutation in a phenylketonuria mouse model with 11–17% efficiency, resulting in a therapeutic reduction of blood phenylalanine levels without detectable off-target mutations or prolonged liver inflammation [250]. These precision editing modalities represent significant advances over conventional CRISPR-Cas9 nucleases, particularly in post-mitotic hepatocytes where HDR is inherently inefficient. However, challenges remain regarding delivery efficiency, immune responses, and the optimization of dosing regimens to achieve therapeutic editing levels.

AAV vector limitations

Despite its clinical promise, AAV vectors have notable limitations. In most cases, the vector persists in an episomal state, which reduces the risk of insertional mutagenesis but also renders the therapeutic effect transient and susceptible to dilution—particularly in pediatric patients undergoing continuous growth or in individuals experiencing liver regeneration. Quantitative studies demonstrate that episomal AAV genomes are progressively diluted with each hepatocyte division: following a single cell division, vector copy numbers decrease by approximately 50%, leading to a logarithmic decline in transgene expression over time in rapidly proliferating tissues. This dilution effect is especially pronounced in pediatric patients, where liver mass increases 10- to 15-fold from birth to adulthood, necessitating substantial hepatocyte proliferation that progressively partitions episomal vector genomes into daughter cells until therapeutic expression falls below efficacious thresholds [251, 252]. By contrast, integrating lentiviral vectors achieve permanent genomic insertion, ensuring stable transgene inheritance through cell division and sustained expression regardless of hepatocyte proliferative status [253, 254]. However, this integration advantage comes at the cost of insertional mutagenesis risk, requiring careful balance between durability and safety [255–257].

The fundamental dichotomy between episomal persistence and genomic integration dictates distinct clinical applications: AAV-mediated gene addition is preferred for quiescent or slowly dividing hepatocytes (e.g., hemophilia, where therapeutic levels can be maintained for years in adult livers), whereas lentiviral integration may be necessary for diseases requiring lifelong expression in proliferative contexts (e.g., pediatric metabolic liver diseases) or where precise copy number control is critical [164, 234]. Consequently, integrating strategies are being explored specifically for pediatric patients to overcome this proliferation-dependent vector loss, ensuring durable transgene expression despite ongoing liver growth. To address this challenge, researchers are actively investigating several approaches aimed at achieving more sustained transgene expression. These include engineering rAAV vectors containing homology arms to facilitate targeted integration into the host genome; utilizing naturally integrating AAV variants such as AAV-Rep78; incorporating transposon systems like Sleeping Beauty or piggyBac into AAV vectors to mediate stable genomic insertion; and employing AAV as a delivery vehicle for gene-editing tools such as CRISPR-Cas9 to enable precise integration of the therapeutic sequence directly into the chromosome [80, 258].

The most straightforward approach involves using AAV to deliver a minimal promoter along with the therapeutic gene to the albumin locus. The therapeutic gene is flanked by homologous arms derived from the albumin gene, enabling precise integration upstream of the albumin stop codon. This results in the transcription of a fusion mRNA, which is subsequently translated into a chimeric protein composed of albumin fused to the therapeutic protein [259, 260]. This strategy has demonstrated efficacy in animal models of hemophilia, CNS, alpha-1 antitrypsin deficiency, and MMA, and has progressed to clinical trials for MMA [261, 262]. However, this approach faces several critical limitations. Initial studies indicate that only about 1% of hepatocytes are successfully edited, meaning therapeutic benefits are reliably achieved only in cases where a low transduction rate suffices for clinical effect or when the edited hepatocytes possess a proliferative advantage that allows them to gradually repopulate the liver—conditions that are not typically met in most inherited metabolic liver diseases [261, 262]. Furthermore, the albumin fusion protein strategy presents substantial challenges related to post-translational processing; the chimeric protein may exhibit altered folding, impaired secretion efficiency, reduced functional activity, or aberrant intracellular trafficking compared to the native therapeutic protein, potentially compromising therapeutic efficacy [263]. Moreover, expression variability arises from position effects and integration site heterogeneity, leading to inconsistent therapeutic protein levels across edited hepatocytes and unpredictable dosing [264]. Additionally, there is a significant immunogenicity risk, as the fusion protein may elicit immune responses against both the therapeutic component and the albumin carrier, potentially triggering nAbs, clearance of the therapeutic protein, or even autoimmune reactions against endogenous albumin, thereby compromising long-term efficacy and safety [195]. To overcome this limitation, the use of AAV vectors encoding CRISPR-Cas9 to introduce a site-specific double-strand break at the albumin locus has markedly enhanced the efficiency of homology-directed repair, offering a promising avenue for improving the therapeutic potential of this approach [265].

Adenoviral vectors applicable for liver therapies

Adenoviruses (Ads) are structurally more complex than AAVs, possessing a linear, double-stranded DNA genome of approximately 36 kb that encodes multiple structural and non-structural viral proteins. These viruses are non-enveloped, with the most commonly utilized serotypes being human adenovirus serotype 2 and serotype 5 [266]. However, in recent years, Ads isolated from different species have been employed to circumvent pre-existing humoral immune responses against these human serotypes. In immunocompetent individuals, human Ads typically cause mild, self-limiting illnesses [267].

Ads possess several advantageous characteristics for liver-directed gene delivery, including natural hepatotropism, robust transgene expression across multiple cell types, ease of production with high functional titers, large cargo capacity, and relatively low genotoxic potential [268]. This hepatotropism is primarily mediated by the high-affinity interaction between coagulation factor X (FX) and the adenoviral hexon protein, which occurs immediately upon intravascular delivery. FX binds to the hypervariable regions of the hexon protein, forming a bridge that facilitates adenovirus attachment to heparan sulfate proteoglycans (HSPGs) on hepatocyte surfaces, thereby promoting efficient liver transduction [269]. However, the same FX-hexon interaction also contributes to the vector’s immunogenicity and toxicity profile. The formation of the FX-adenovirus complex may influence immunogenicity through multiple mechanisms, including complement regulation and innate immune activation [270, 271]. Additionally, it promotes sequestration of the vector in liver sinusoids and Kupffer cells, which can lead to hepatotoxicity and dose-limiting toxicities at high vector doses [272, 273]. First-generation recombinant adenoviral vectors lack essential viral replication genes but retain the majority of the viral genome, allowing the incorporation of up to approximately 8.5 kb of foreign genetic material. These vectors demonstrate high efficiency in hepatocyte transduction [267]. However, deletion of replication-essential genes does not prevent the expression of certain structural proteins that elicit significant inflammatory and immune responses. This host immune reaction leads to rapid clearance of transduced cells, resulting in only transient therapeutic effects [274].

To overcome these limitation, third-generation adenoviral vectors, known as helper-dependent adenoviral vectors (HDAds), have been developed. HDAds retain only the terminal inverted repeats and packaging signal sequences, with all other viral coding sequences deleted [275]. This extensive genomic deletion allows the insertion of large genetic payloads (up to 36 kb) and facilitates long-term transgene expression, as demonstrated in both murine and NHP models. For example, HDAds expressing the LDLR have been shown to improve lipid profiles and reduce aortic atherosclerosis in rodent and NHP models of hypercholesterolemia [208, 276]. Moreover, the therapeutic potential of HDAds has been clearly demonstrated in animal models of various inherited metabolic liver diseases, including PH1, CNS, AIP, GSD-Ia, and OTCD [277–279]. However, clinical translation of HDAds has been hindered by manufacturing challenges, primarily the reliance on helper virus co-infection and the difficulty of scalable, clinical-grade production free from contamination. State-of-the-art systems employ Cre/loxP recombination in 293-derived producer cells, where Cre excises the helper virus packaging signal to prevent its packaging while enabling HDAd trans-complementation [280]. Recent advances include suspension-adapted cell lines, such as 293Cre66S, and evolved FLPe recombinase, which achieves approximately 100% excision efficiency compared to about 80% with Cre. These improvements have enabled scalable production with helper virus contamination reduced to 0.1–0.4% without purification and 0.01–0.02% post-purification, yielding over 10¹³ particles from 3-liter cultures within two weeks [281]. CAP cells—human amniocyte-derived, E1-complementing lines without adenoviral sequence overlap—also enable replication-competent adenovirus-free production when engineered with Cre recombinase. Persistent bottlenecks include the increased complexity and risk of genome rearrangement caused by co-infection with auxiliary viruses, the lower virus particle yield in suspension cultures (3,000 to 8,000 particles per cell) compared to adherent systems (10,000 to 20,000 particles per cell), and the limited scalability and regulatory compliance of density gradient ultracentrifugation used for auxiliary virus isolation [282]. Emerging approaches, such as helper virus-free systems that use stable producer cell lines expressing all necessary viral components, along with chromatography-based purification methods, are actively being developed to overcome these limitations and enhance the commercial viability of HDAd-based therapies.

Despite the numerous advantages of Ad vectors, their inherent immunogenicity remains a significant barrier, severely limiting their safe and effective clinical application. To address these challenges, two main categories of immune evasion strategies have been developed: administration of immunosuppressive agents to the host and modification of the viral capsid itself. Capsid engineering can be further divided into three distinct approaches: chemical or physical shielding of the capsid using polymers; surface display of immunomodulatory or immune-evasive proteins; and direct genetic modification of capsid proteins.

Among these strategies, coating the capsid with non-immunogenic polymers, particularly PEG, has garnered the most attention. Experimental evidence demonstrates that PEGylation not only significantly enhances hepatic transduction efficiency but also markedly suppresses the production of inflammatory cytokines and nAbs. Similar to AAV, Ad vectors maintain their genomes in an episomal state following infection. For long-term and stable gene correction, Ad vectors can be employed to deliver either the CRISPR/Cas9 system or transposon-based elements into hepatocytes [283].

Retroviral and lentiviral vectors applicable for liver therapies

Retroviruses and lentiviruses are both enveloped, single-stranded RNA viruses that carry reverse transcriptase, enabling them to convert their RNA genomes into double-stranded DNA intermediates, which are then stably integrated into the host cell genome [253]. Their genomic structure consists of three essential genes—env, gag, and pol—flanked by LTRs. The LTRs contain promoter and enhancer elements critical for viral integration and transcriptional regulation [284]. The env and gag genes encode structural proteins of the virus, while pol encodes non-structural proteins, including reverse transcriptase [285]. Lentiviruses possess additional regulatory genes, tat and rev, which are absent in simple retroviruses [286].

In vector design, the LTRs are retained to preserve transcriptional and integration competence, while all viral coding sequences are removed to create space—approximately 8 to 9 kilobases—for the insertion of therapeutic transgenes. Retroviral vectors are limited to transducing dividing cells, making them more suitable for ex vivo gene therapy applications [287]. In contrast, lentiviral vectors can efficiently transduce both dividing and non-dividing cells, including quiescent primary cells such as terminally differentiated hepatocytes, thereby offering broader applicability [288].

Currently, both vector systems are predominantly used to correct monogenic disorders of the hematopoietic system [289]. Notably, lentiviral-mediated gene transfer into hematopoietic stem cells has demonstrated remarkable clinical efficacy in treating lysosomal storage disorders, such as adrenoleukodystrophy [290, 291]. These viral vectors are utilized for ex vivo modification in the treatment of inherited liver diseases [205]. In these studies, primary hepatocytes are first isolated from liver biopsies, genetically modified ex vivo using recombinant viral vectors, and subsequently reinfused into the liver or spleen. Research conducted in murine, rat, canine, and NHP models demonstrates that this ex vivo reimplantation approach can achieve long-term and stable transgene expression [292]. Sustained therapeutic benefits have been observed in animal models of hypercholesterolemia [276], hypertension [293], CNS [294], and hemophilia [295], ultimately leading to the initiation of early-phase clinical trials.

Lentiviral vector-mediated transduction of primary hepatocytes ex vivo has achieved transduction efficiencies approaching 90% [296]. However, the primary challenge remains the unstable engraftment and low repopulation efficiency of the transduced cells following transplantation, further complicated by significant inter-individual variability. The fundamental issue is the lack of a proliferative advantage in the implanted cells, which limits their long-term persistence and expansion within the host liver. Recent studies have demonstrated that delivering the corrected FAH gene via lentiviral vectors, which confers a selective growth advantage to the modified hepatocytes, can lead to complete correction of liver disease in the HT1 porcine model [297].

To transition retroviral or lentiviral delivery from an ex vivo manipulation and reinfusion strategy to a direct systemic administration approach, early experimental protocols typically induce robust hepatocyte proliferation—through partial hepatectomy, chemical injury, or administration of hepatocyte growth factor—thereby creating a permissive window for viral integration [298]. However, sustained transgene expression achieved under these conditions is often limited by adaptive immune responses targeting the foreign antigen, necessitating the concurrent implementation of immunosuppressive regimens [299, 300].

In murine models of CNS [301], MMA [302], and GSDIa [303], the systemic administration of recombinant lentiviral vectors has demonstrated therapeutic efficacy [303]. The risk of insertional mutagenesis historically associated with lentiviral vectors has been substantially reduced through the development of self-inactivating (SIN) vectors, in which the enhancer/promoter elements within the LTRs have been deleted. Administration of SIN lentiviral vectors to adult mice and dogs has resulted in sustained hepatic expression of factor IX, induction of immunological tolerance to the transgene product, and no observed genotoxic effects [254]. Nevertheless, despite these safety improvements, residual integration risks persist and warrant rigorous evaluation. While SIN vectors eliminate LTR-driven transcriptional activation, integration into active host gene regions can still cause promoter-mediated dysregulation of nearby oncogenes or tumor suppressor genes, particularly in the context of hepatocyte proliferation, where clonal expansion of individual integration events may occur [255]. However, the permanence of integration necessitates rigorous preclinical evaluation of insertion site profiles, clonal dynamics, and potential genotoxicity through methods such as linear amplification-mediated PCR (LAM-PCR) and high-throughput sequencing. Long-term follow-up studies in animal models have demonstrated that specific integration sites can confer subtle proliferative advantages, leading to progressive clonal dominance over months to years, underscoring the need for lifetime monitoring in clinical applications [255–257]. The long-term clonal dynamics of lentivirally transduced hepatocytes remain incompletely characterized, and serial transplantation studies in animal models have demonstrated that specific integration sites can confer subtle proliferative advantages, leading to progressive clonal dominance over months to years [256]. Current insertion site analysis methods—including LAM-PCR, non-restrictive LAM-PCR, and high-throughput sequencing—offer snapshots of integration site distributions but face technical limitations in detecting low-abundance clones and distinguishing true clonal expansion from sampling bias. Conventional LAM-PCR fails to detect 30–40% of clones even after exhaustive analysis and can distort relative clone abundance by up to 60-fold [257]. Clinical surveillance requirements for liver-directed lentiviral gene therapy remain undefined, as the latency period for insertional oncogenesis in hepatocytes is unknown. Additionally, the optimal frequency and modality of monitoring—whether through circulating cell-free DNA analysis, liver imaging, or periodic biopsy with insertion site profiling—require prospective validation [296]. These considerations underscore the need for comprehensive preclinical safety packages that incorporate clonal tracking studies and risk assessment modeling prior to clinical translation, as well as long-term follow-up protocols for treated patients to monitor delayed adverse events.

However, mild acute toxicity and suboptimal therapeutic efficacy have been observed following the administration of viral vectors, likely due to off-target transduction of liver-resident antigen-presenting cells (APCs), such as Kupffer cells and hepatic stellate cells [304, 305]. Various strategies are currently being investigated to prevent vector uptake by APCs, aiming to enhance transduction efficiency in hepatocytes and minimize systemic toxicity. Notably, incorporating the human phagocytosis inhibitor CD47 into lentiviral vector particles has been shown to significantly enhance transgene expression while reducing vector-associated toxicity in NHPs [292]. In the coming years, the development of novel lentiviral pseudotypes with improved hepatocyte tropism and the ability to evade transduction of off-target cells may expand the application of lentiviral vectors in the direct treatment of liver-targeted diseases [306].

Currently, clinical data on the direct systemic administration of retroviral or lentiviral vectors for treating liver diseases remain limited. In 2003, a study investigated the intravenous delivery of a recombinant retroviral vector expressing coagulation factor VII in 13 patients with hemophilia A. The treatment is well tolerated but failed to demonstrate significant therapeutic efficacy [307]. Although two lentivirus-based clinical trials for treating hemophilia A and B have been registered on ClinicalTrials.gov (NCT03217032 and NCT03961243), neither has initiated participant recruitment to date [308].

α-GLA, α-galactosidase A; ALAS1, 5-aminolevulinic acid synthase 1; ANGPTL3, angiopoietin-like 3; ARSB, arylsulfatase B; ATP7B, ATPase copper-transporting beta; BDD-FVIII, B-domain-deleted factor VIII; DUSP22, dual-specificity phosphatase 22; F8, coagulation factor VIII gene; F9, coagulation factor IX gene; FAH, fumarylacetoacetate hydrolase; G6PC, glucose-6-phosphatase catalytic subunit; GLB1, β-galactosidase; HAO1, hydroxyacid oxidase 1; HSD17B13, 17-β-hydroxysteroid dehydrogenase 13; IFN, interferon; LAMP2B, lysosomal-associated membrane protein 2b; LDLR, low-density lipoprotein receptor; MASH, metabolic dysfunction-associated steatohepatitis; MMUT, methylmalonyl-CoA mutase; MUT, methylmalonyl-CoA mutase; OTC, ornithine transcarbamylase; PAH, phenylalanine hydroxylase; PBGD, porphobilinogen deaminase; PCCA, propionyl-CoA carboxylase alpha; PCCB, propionyl-CoA carboxylase beta; PCSK9, proprotein convertase subtilisin/kexin type 9; PNPLA3, patatin-like phospholipase domain-containing protein 3; OTC, ornithine transcarbamylase; SLC27A4, solute carrier family 27 member 4; TTC39B, tetratricopeptide repeat domain 39B; TTR, transthyretin; UGT1A, uridine diphosphate glucuronosyltransferase 1 A.

Gene correction therapy

LNP have been developed as delivery vehicles for gene-editing components, capable of co-delivering ZFN or Cas9 mRNA along with their corresponding sgRNA in a single formulation (Fig. 3). Encapsulation of ZFN-encoding mRNA within LNPs, targeting either the TTR or PCSK9 gene results in a reduction of the respective protein expression by over 90% in mice, offering potential therapeutic applications for transthyretin amyloidosis and hypercholesterolemia, respectively [338]. To enhance editing precision and efficiency, more sophisticated strategies integrate LNPs with viral vectors. Specifically, combining LNPs with viral vectors carrying promoterless DNA templates enables concurrent delivery of ZFNs [338]. Through HDR, this strategy facilitates the efficient integration of desired sequences into the genome [339].

However, conventional CRISPR-Cas9 nucleases induce DSBs, which pose risks of indels and chromosomal rearrangements, particularly in post-mitotic hepatocytes where HDR is inherently inefficient. To circumvent these limitations, precision genome editing platforms—namely base editing and prime editing—have emerged as particularly promising modalities for correcting point mutations in metabolic liver diseases. CBEs and ABEs enable direct C·G-to-T·A or A·T-to-G·C conversions, respectively, without requiring DSB formation or HDR, thereby minimizing risks of indels and chromosomal rearrangements. The therapeutic potential of these precision tools has been validated in large animal models. LNP delivery of base editor mRNA has achieved over 60% hepatic editing efficiency in non-human primates, resulting in approximately 90% reduction in plasma PCSK9 levels and 60% reduction in LDL cholesterol [248, 249]. Parallel advances in prime editing further expand the therapeutic repertoire. Prime editing, which employs a Cas9 nickase fused to reverse transcriptase along with a prime editing guide RNA, enables installation of all possible base transitions as well as small insertions and deletions without inducing DSBs. AAV-mediated delivery of prime editors has demonstrated 11–17% correction efficiency of the disease-causing Pahenu2 mutation in a phenylketonuria mouse model, with therapeutic reduction of blood phenylalanine levels [250].

These precision editing modalities are especially advantageous for post-mitotic hepatocytes, and their transient expression kinetics via LNP delivery further enhance safety profiles for clinical translation. A critical advantage of LNP-mediated mRNA delivery over AAV-based systems lies in the transient nature of Cas9 expression. LNP-delivered Cas9 mRNA and sgRNAs are rapidly degraded in vivo (typically active for 1–2 weeks), thereby minimizing the risk of cumulative off-target mutations and insertional mutagenesis associated with persistent AAV-mediated transgene expression [89]. This temporal control stands in stark contrast to viral vector persistence. In contrast, AAV vectors persist for years in terminally differentiated cells, leading to prolonged Cas9 activity that increases the probability of off-target effects and raises safety concerns regarding AAV integration into CRISPR-induced double-strand breaks at rates up to 47% [340].

Beyond safety considerations, the non-viral nature of LNPs offers practical therapeutic advantages in clinical settings. Specifically, LNPs enable repeat dosing to enhance editing efficiency, whereas AAV vectors are limited by pre-existing immunity and cannot be effectively readministered [340]. These combined attributes—transient activity, reduced off-target risk, and dosing flexibility—position LNPs as particularly promising for applications requiring repeat administration. Such applications include the treatment of mucopolysaccharidosis types I and II mentioned previously, where sustained therapeutic effects may require multiple editing cycles.

Combination approaches

Combination of gene therapy and pharmacological treatments

For hereditary metabolic liver disease, combination therapy demonstrates unique advantages. The combination of ERT and liver targeted AAV gene therapy has shown a synergistic effect in Pompe disease. Research has shown that the combination of secreted GAA (acid alpha glucosidase) expressed in the liver and ERT achieves more significant glycogen clearance effects in refractory tissues such as skeletal muscle compared to ERT alone [341]. This joint strategy enhances the bioavailability of ERT in refractory tissues by increasing the levels of enzymes in the circulation. In multiple sclerosis, the combination of liver-directed AAV gene therapy expressing myelin oligodendrocyte glycoprotein and rapamycin demonstrates strong synergistic effects. This combination achieves near complete remission of established experimental autoimmune encephalomyelitis by day 30, whereas gene therapy alone is only effective in early disease stages. The synergy arises from rapamycin-promoted expansion of regulatory T cells that stabilize the antigen-specific tolerance induced by hepatic gene transfer [342].

The combination of autophagy enhancing drugs and gene therapy shows promising prospects in alpha-1 antitrypsin deficiency. Research has found that autophagy enhancing drugs such as rapamycin or carbamazepine can promote the degradation of mutant SERPINA1 protein, while liver targeted TFEB gene therapy promotes the clearance of abnormal proteins by upregulating the macroautophagy pathway [343]. Recent systematic reviews have analyzed immunosuppressive protocols across 38 AAV gene therapy clinical trials. Prophylactic corticosteroid regimens (prednisone 1 mg/kg with tapering over 7-133 days) are now preferred over reactive approaches, reducing ALT elevations in 74% of studies while maintaining transgene expression [344]. These joint strategies avoid the problem of long-term systemic medication while achieving tissue-specific therapeutic effects.

Combination of gene therapy and RNA therapeutics

Emerging hybrid platforms combining AAV tissue targeting with mRNA-encoded genome editing machinery and LNP-delivered guide RNAs enable transient, dose-controlled “hit-and-run” editing, minimizing long-term immunogenicity while maintaining therapeutic efficacy [315]. For chronic hepatitis B, combination therapy strategies have shown superior viral suppression effects compared to monotherapy. siRNA (JNJ-3989), the triple combination of core protein allosteric modulators and nucleoside analogs achieved significant reductions in HBsAg, HBV DNA, RNA, and HBcrAg in 12 patients. Similarly, the combination of siRNA (JNJ-3989), capsid assembly modulators, and nucleoside analogs achieved significant reductions in HBsAg and viral RNA in chronic hepatitis B patients, while base editing approaches alone have shown promise in targeting cccDNA in primary hepatocyte models [345]. In the treatment of diabetes, the dual carrier system of gene silencing (such as siRNA for gluconeogenase) and gene addition (such as insulin transgene) shows a synergistic effect on reducing fasting blood glucose [346].

Hybrid vector systems combining viral and non-viral platforms represent an emerging paradigm. Zakas et al. demonstrated that AAV vectors carrying Sleeping Beauty transposon donors combined with LNP-delivered transposase mRNA achieved stable genomic integration and long-term expression in OTC deficiency models, enabling dose reduction while maintaining efficacy [347]. More recently, dual AAV-LNP approaches for prime editing have shown promise: scAAV-delivered pegRNAs combined with LNP-encapsulated PE7 mRNA achieved 20.7% correction of the Pah^enu2 mutation in phenylketonuria mice, reducing blood phenylalanine below therapeutic thresholds [315]. These multimodal gene therapy approach achieve comprehensive metabolic correction by simultaneously targeting multiple metabolic pathways.

Multitargeted combined gene therapy

The convergence of multiple therapeutic modalities targeting distinct pathogenic mechanisms has emerged as a transformative strategy for monogenic and multifactorial diseases, offering synergistic efficacy beyond monotherapy limitations. For liver fibrosis, combined gene therapy targeting multiple pathogenic pathways has shown synergistic effects. The combined delivery of HIF-1α-siRNA and silibinin [348], as well as the combined strategy of miR-29b/miR-122 targeting collagen production, achieved better anti fibrotic effects compared to single drugs [349]. Co-delivery of miR-29b with germacrone using cyclic RGD-modified nanoparticles has further enhanced targeted delivery to activated HSCs, significantly inhibiting type I collagen production in CCl₄-induced fibrosis models [350].

In Krabbe disease, triple combination therapy—CNS-directed AAV9 gene therapy, hematopoietic stem cell transplantation, and small molecule substrate reduction therapy (L-cycloserine)—extended the median lifespan from approximately 40 days in untreated mice to approximately 400 days, compared to 269 days with dual therapy alone. This result demonstrates the powerful efficacy of therapeutic strategies with complementary mechanisms targeting enzyme replacement, neuroinflammation modulation, and toxic metabolite reduction [351]. Building on this preclinical success, two Phase I/II clinical trials (NCT04693598 and NCT05739643) are currently evaluating AAVrh10-mediated GALC gene transfer, alone or in combination with prior HSCT, in patients with infantile and late infantile Krabbe disease [352]. In chronic hepatitis B, multitargeted combination strategies have entered advanced clinical trials. The REEF-1 Phase 2b trial demonstrated that triple therapy with siRNA JNJ-3989, capsid assembly modulator JNJ-56,136,379 (bersacapavir), and nucleos(t)ide analogues achieved HBsAg reduction of 1.7 log₁₀ IU/mL with 97.5% of patients reaching ≥ 1 log₁₀ decline (NCT03982186) [345].

The clinical transformation prospects of combined therapy

The clinical translation of combination therapy strategies faces multiple challenges, including determining the optimal timing of administration, dosage ratio, and administration sequence. Sequential treatment strategies, such as using LNP mRNA for rapid protein replacement followed by AAV for long-term correction, may be particularly suitable for pediatric patients to avoid the dilution effect of AAV vector genomes in proliferating liver cells [353].

The core advantages of these joint strategies lie in achieving synergistic effects through complementary mechanisms, reducing single drug doses to minimize toxicity, and intervening in multiple pathological stages of the disease simultaneously. With the deepening understanding of disease mechanisms and the advancement of delivery technologies, rationally designed combination therapy plans will become an important development direction for gene therapy of liver diseases.

Current limitations in liver-targeted gene therapy

With mounting evidence affirming the therapeutic potential of liver-targeted gene delivery, there is an urgent need to concentrate resources on in-depth research to address the key challenges that must be overcome to translate these cutting-edge technologies into clinical practice.

A major bottleneck currently limiting the widespread application of LNPs in RNA therapeutics is the challenge of achieving targeted delivery beyond the liver. The critical challenge in LNP-based RNA therapeutics is the feasibility of repeated administration. The primary driver of the immune response against LNPs following repeated dosing is the pro-inflammatory reactogenicity, including anti-PEG antibody production and TLR4 activation by ionizable lipids, which can compromise the efficacy of subsequent doses. For example, in rats, a second dose of PEGylated LNPs administered 21 days after the first reduced serum LNP levels by more than 50%, coinciding with peak anti-PEG IgM and elevated C5a—confirming complement-mediated accelerated blood clearance [354, 355]. Moreover, LNPs containing SM-102 or ALC-0315 (without mRNA) activated TLR4-MyD88 signaling, inducing IL-1β, IL-6, CCL2, and CCL4. This response is abolished by TAK-242 [356]. While current mRNA vaccine regimens typically involve booster doses every 3–4 weeks, therapeutic applications such as protein replacement therapy or in vivo cellular engineering may require more frequent dosing—on the order of weekly or even more frequent intervals. For instance, in the context of liver cancer, PCTAIRE1-targeting siRNA-LNPs achieved optimal tumor suppression through twice-weekly administration, as the therapeutic effect of a single dose waned within 4–7 days [357]. SPTLC2-targeting siRNA-LNPs for metabolic dysfunction-associated steatohepatitis (MASH) exhibited more potent and sustained hepatic gene silencing (~ 70% knockdown) and therapeutic efficacy with twice-weekly administration compared to once-weekly dosing, which produced less pronounced effects [358]. Consequently, optimizing LNP formulations to minimize immune recognition and reduce complement activation has emerged as a key focus in the next phase of LNP development. LNPs hold great promise for revolutionizing delivery paradigms in RNA therapeutics and gene editing [359]. Nevertheless, the full realization of their clinical potential will depend on continued advances in both fundamental and translational research.

While significant progress has been made in addressing the immunogenicity challenges of non-viral delivery systems such as LNPs, viral vectors present a distinct and equally complex landscape of immune-related challenges. These challenges primarily arise from the inherent immunogenicity of viral vectors, which can activate both innate and adaptive immune responses. In addition, concerns regarding potential hepatotoxicity persist, and the scalable, clinical-grade manufacturing of these vectors remains technically challenging. Taking liver-targeted gene delivery as an example, Ad vectors, although supported by clinical experience, have been predominantly applied in the context of cancer. Their use in hereditary metabolic liver diseases is limited by a pronounced tendency to induce inflammatory responses, as well as by low vector production yields. Similarly, retroviral and lentiviral vectors have seen minimal application in liver-directed gene therapy. Given these constraints, this review focuses on current empirical insights and advancements in AAV-based vector systems.

Upon entering the body, any gene delivery vector is rapidly recognized by the immune system. Both viral and non-viral vector platforms have the potential to activate innate and adaptive immune responses, which may significantly diminish or even neutralize the intended therapeutic effect (Fig. 5). More critically, the inflammatory response elicited by recombinant vectors in diseased livers can exacerbate pre-existing hepatic damage. Once adaptive immunity is engaged, transduced hepatocytes are often eliminated, resulting in transient transgene expression—a phenomenon commonly referred to as immune clearance. The critical immunological barrier arises from two mechanistically distinct immune challenges that occur at different stages of therapy. First, pre-existing humoral nAbs generated by natural infections can immediately neutralize the vector upon administration, thereby limiting initial patient eligibility and excluding a substantial proportion of the general population from receiving gene therapy. Second, even in seronegative individuals who initially qualify for treatment, the immune response elicited by the first therapeutic dose—comprising de novo nAb production and activation of memory T cells—creates a formidable barrier to vector re-administration. This post-administration immunity can completely block hepatocyte transduction and preclude subsequent dosing when therapeutic efficacy proves insufficient or wanes over time. This limitation is especially pertinent in pediatric patients, whose expanding liver mass may effectively dilute the therapeutic effect as they grow, yet no opportunity exists for subsequent dosing due to established immune memory [360]. To counteract this immune-mediated clearance, the most straightforward clinical approach has been the administration of immunosuppressive agents. This strategy has demonstrated efficacy in both preclinical models and human trials. However, it inevitably places patients in a state of transient immunosuppression, increasing their susceptibility to infections. Another promising approach involves silencing the transgene through sequence editing to eliminate CpG motifs known to trigger TLR, thereby dampening the initial immune alert at its source [361].

Fig. 5.

rAAV gene therapy faces multiple immune barriers and corresponding response strategies. Firstly, pre-existing anti-AAV neutralizing antibodies in the body can directly bind to the viral capsid, thereby blocking transduction of liver cells. To address this, strategies such as IgG protease-mediated clearance, plasma exchange, capsid bait competition, capsid modification, or induction of immune tolerance can be employed. TLRs recognize capsid proteins and viral genomes as danger signals, activating innate immunity. This response can be attenuated by capsid design optimization and reduction of CpG sequences. Activation of TLRs within the endosome can also induce proteasomal degradation of the capsid, which can be inhibited by agents such as bortezomib. The resulting degradation products are presented to CD8⁺ and CD4⁺ T cells via MHC class I and II molecules, leading to the elimination of transduced liver cells and amplification of the humoral immune response. These effects can be mitigated by corticosteroids or broader-spectrum immunosuppressants. Additionally, transcription of nuclear double-stranded RNA can trigger RIG-I/MDA5-mediated type I interferon responses, which can be alleviated by further modification of the AAV genome, such as removal of TLR9 recognition sequences. Moreover, intracellular defense mechanisms may cause silencing or loss of the free rAAV genome, which can also be prevented through genome engineering

To address these fundamental challenges, multiple targeted strategies are currently under investigation. To circumvent pre-existing nAbs, approaches include employing non-cross-reactive serotypes, engineering capsids with reduced immunogenicity, chemically modifying vector surfaces, or utilizing plasmapheresis and immunoadsorption to remove nAbs. To enable re-administration in the context of post-therapeutic immunity, a recently reported bacterial protease capable of degrading human IgG has demonstrated successful vector re-dosing in rodent and NHP models, offering promising potential for clinical translation [362]. Additionally, rapamycin has emerged as an effective immune evasion strategy to suppress nAb generation following AAV administration, offering a targeted approach to mitigate the adaptive immune response without inducing broad immunosuppression [363].

Recently, three adolescents suffering from fatal neuromuscular diseases developed hepatic injury following systemic administration of high-dose rAAV, once again bringing vector-related hepatotoxicity into sharp focus. These patients are relatively older, heavier, and all has pre-existing hepatobiliary conditions, underscoring the necessity for heightened caution in subjects with underlying liver disease—a common clinical scenario in patients with inherited metabolic liver disorders [159]. Fortunately, to date, AAV-based clinical trials targeting AIP, OTCD, and GSD1a have not reported major adverse events.

Oncogenic risks also warrant long-term vigilance. Although certain animal studies have suggested a potential link between AAV and integrative tumorigenesis, and wild-type AAV genomes have been suspected of association with human hepatocellular carcinoma, long-term follow-up (12–15 years) of patients treated with rAAV has not revealed persistent toxicity or malignant transformation attributable to vector integration [88]. Nevertheless, regardless of the vector used, lifelong monitoring remains essential. Targeted integration of the therapeutic gene into genomic safe harbors may represent a promising compromise between durable expression and long-term safety.

Prospective

Encouragingly, ongoing clinical trials and the increasing number of liver-targeted gene therapy products receiving regulatory approval are generating positive outcomes and bolstering confidence in the field. Although significant challenges remain—such as issues related to vector delivery, immune response, and long-term safety—continuous technological advancements and accumulating clinical experience hold the potential to transform the treatment landscape for many genetic diseases. In the coming years, conditions that have historically required orthotopic liver transplantation may become amenable to routine, and potentially curative, gene-based therapies.

Abbreviations

2’-F

2’-fluoro

2’-OMe

2’-O-methyl

AAV

Adeno-associated virus

AAVR

Adeno-associated virus receptor

ABEs

Adenine base editors

Ad

Adenovirus

AIP

Acute intermittent porphyria

ALAS1

Aminolevulinic acid synthase 1

α-GLA

α-galactosidase A

ANGPTL3

Angiopoietin-like 3

APC

Antigen-presenting cell

ARSB

Arylsulfatase B

ASGPR

Asialoglycoprotein receptor

ASO

Antisense oligonucleotide

ATP7B

ATPase copper-transporting beta

BDD-FVIII

B-domain-deleted factor VIII

Cas9

CRISPR-associated protein 9

CBEs

Cytosine base editors

CD34⁺

Cluster of differentiation 34-positive

CD47

Cluster of differentiation 47

CNS

Crigler-Najjar syndrome

CpG

Cytosine-phosphate-guanine

CRISPR

Clustered regularly interspaced short palindromic repeats

DUSP22

Dual-specificity phosphatase 22

Env

Envelope

ERT

Enzyme replacement therapy

F8

Coagulation factor VIII gene

F9

Coagulation factor IX gene

FX

Factor X

FAH

Fumarylacetoacetate hydrolase

G6PC

Glucose-6-phosphatase catalytic subunit

GalNAc

N-acetylgalactosamine

GLB1

β-galactosidase

GSD1a

Glycogen storage disease type Ia

HAO1

Hydroxyacid oxidase 1

HCC

Hepatocellular carcinoma

HDR

Homology-directed repair

HIF-1α

Hypoxia-inducible factor-1α

HSPG

Heparan sulfate proteoglycans

HSC

Hepatic stellate cell

HT1

Hereditary tyrosinemia type 1

IgG

Immunoglobulin G

IL

Interleukin

ITR

Inverted terminal repeat

IVIG

Intravenous immunoglobulins

LDL

Low-density lipoprotein

LDH

Lactate dehydrogenase

LDLR

Low-density lipoprotein receptor

LNP

Lipid nanoparticle

LSEC

Liver sinusoidal endothelial cells

LTR

Long terminal repeat

LAMP2B

Lysosomal-associated membrane protein 2b

MASH

Metabolic dysfunction-associated steatohepatitis

MDA5

Melanoma differentiation-associated protein 5

MHC

Major histocompatibility complex

MMA

Methylmalonic academia

mRNA

Messenger RNA

MUT

Methylmalonyl-CoA mutase

nAbs

Neutralizing antibodies

NHEJ

Non-homologous end joining

NHP

Non-human primate

OTC

Ornithine transcarbamylase

OTCD

Ornithine transcarbamylase deficiency

PAH

Phenylalanine hydroxylase

PBGD

Porphobilinogen deaminase

PCSK9

Proprotein convertase subtilisin/kexin type 9

PEG

Polyethylene glycol

PEI

Polyethylenimine

PFIC3

Progressive familial intrahepatic cholestasis type 3

PKU

Phenylketonuria

PLGA

Poly (lactic-co-glycolic acid)

PLK1

Polo-like kinase 1

rAAV

Recombinant adeno-associated virus

RNAi

RNA interference

RIG-I

Retinoic acid-inducible gene I

RLR

RIG-I-like receptor

sgRNAL

Single-guide RNA

siRNA

Small interfering RNA

SIN

Self-inactivating (lentiviral vector)

SLC27A4

Solute carrier family 27 member 4

scAAV

Self-complementary AAV

ssAAV

Single-stranded AAV

TBG

Thyroxine-binding globulin

TGF-β

Transforming growth factor-β

TTR

Transthyretin

TLR

Toll-like receptor

UCD

Urea cycle disorder

VLDLR

Very low-density lipoprotein receptor

VP1

viral protein 1

ZFN

zinc-finger nuclease

Author contributions

Mengyao Yan performed reference searches, wrote the manuscript and created the tables and figures. Qian Xiang conceptualized, reviewed and edited the manuscript. The final version of the work has been read and approved by all authors.

Funding

This study was supported by grants from National High Level Hospital Clinical Research Fund (Peking University First Hospital Science and Technology Achievement Transformation and Incubation Guidance Fund) grant No. 2025CX34; by National Science Foundation of China grant No. 82274024.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All authors have provided consent for the manuscript to be published in Journal of Nanobiotechnology.

Competing interests

The authors declare no competing interests.