Current and Emerging Issues in Adeno-Associated Virus Vector-Mediated Liver-Directed Gene Therapy

Neutralizing anti-AAV antibodies

Pre-existing anti-AAV neutralizing antibodies (NAb) arise from exposure to the wild-type virus, which increases with age and varies by geographic region. ^17,18 Humans are first exposed to AAVs during childhood and adolescence, with the prevalence of anti-AAV immunity reaching detectable antibody titers in up to 60% of adults for some serotypes. ^{19 –21} Anti-AAV1 NAb are the most prevalent whereas NAb directed against AAV5 are the least prevalent. ^{19,21 –23} Antibodies that arise after exposure to one AAV serotype may cross-react with serotypes that are evolutionarily similar and are less likely to react with phylogenetically distinct serotypes. ²⁴ The prevalence of anti-AAV NAbs is higher in infants, presumably due to transplacental transfer of maternal antibodies during fetal life. ¹⁷ Anti-AAV NAb can also originate from AAV vector dosing. Either natural or post-dosing, NAbs impair gene transfer to hepatocytes. ^12,25 While pre-existing antibodies to AAV affect the eligibility of patients for enrollment in gene therapy clinical trials, postdosing humoral immunity prevents vector re-administration if the expression of the therapeutic gene is lost over time. ²⁶ However, clinical trial data from hemophilia B individuals who received AAV5-based gene therapies showed sustained factor IX activity irrespective of preexisting neutralizing antibody status. This observation supports that AAV5 can be effective in most patients irrespective of preexisting AAV5 neutralizing antibodies. ^27,28 Nevertheless, the presence of anti-AAV antibodies also raises safety concerns because NAb or non-neutralizing antibodies can form immune complexes with circulating vectors and can activate the complement system via the classical pathway, leading to serious clinical complications, as discussed in the next section.

Anti-AAV antibodies can be measured by a neutralization assay, which evaluates inhibition of transduction of a cell line by a reporter AAV vector or by a binding assay, which measures antibodies binding to the AAV capsid. Binding and NAb titers do not necessarily correlate, with subjects scoring seronegative in one assay and positive in the other. ²⁹ Moreover, NAb titers may vary significantly from laboratory to laboratory depending on the assay used, leading to inconsistencies in the eligibility criteria related to NAb exclusion across clinical trials.

Various strategies have been investigated to overcome pre-existing anti-AAV antibodies, including plasmapheresis, ^{30 –33} immunoadsorption, ³⁴ and the anti-B cell monoclonal antibody rituximab in combination with cyclosporine A. ³⁵ Among these, recent preclinical work in mice and non-human primates showed that Imlifidase, a bacterial endopeptidase, efficiently cleaves circulating IgG, thus enabling liver gene transfer even in the presence of anti-AAV NAb. ³⁶ Alternatively, sequential administrations of different AAV capsids carrying the therapeutic gene might be considered once the expression from the previous administration declines. ³⁷ However, there are challenges for the high costs related to the production of multiple AAV vectors carrying the same therapeutic gene. Co-administration of nanoparticles encapsulating rapamycin together with AAV vectors has also been proposed as a strategy to enable repeated AAV administration in mice and non-human primates. ³⁸ The potential of these approaches awaits clinical confirmation.

Inflammatory and innate immunity responses

Systemic high-dose AAV vector administration in nonhuman primates resulted in acute hepatic toxicity, sinusoidal injury, and liver platelet sequestration. ³⁹ Consistent with these findings, some patients with SMA or DMD who received high AAV vector doses (>5 × 10¹³ vector genome [vg]/kg) developed inflammatory complications such as hepatotoxicity, thrombotic microangiopathy (TMA) and atypical hemolytic uremic syndrome (aHUS) with complement activation. ^40,41 A rare and lethal innate immune response has also been reported in a DMD patient who developed a cytokine-mediated capillary leak syndrome resulting in lethal cardiac and pulmonary toxicities associated with AAV9-mediated innate signaling, which were complicated by the advanced stage of the DMD. ⁴²

AAV can trigger the activation of the complement system, an innate defensive mechanism that senses the surface of the invading pathogens inducing their rapid destruction. Complement activation plays a major role in the proinflammatory response induced by high AAV doses. Although activation by direct interaction of the C3 protein through the alternative pathway has been reported, ⁴³ complement activation by AAV is primarily antibody dependent (classical pathway) and is triggered by anti-capsid IgM and IgG antibodies that cause complement-mediated cell damage. ⁴⁴ Complement activation pathways converge in the formation of the C5 convertase that produces C5a, a potent mediator of inflammation, and C5b, which initiates the formation of the membrane attack complex (MAC) mediating cell lysis and cell death. Both C5a and MAC can cause acute hepatic and myocardial injury. ⁴⁵

Hepatotoxicity and complement-dependent severe acute reactions including TMA and aHUS have been observed in patients with various diseases after systemic AAV vector infusions employing different capsid serotypes, expression cassettes, and production protocols. Therefore, overall, the AAV vector doses appear to be the main driver of the inflammatory responses, whereas other factors do not appear to play a significant role. In addition to nausea, fever, and vomiting likely due to cytokine release, patients receiving high doses of AAV vector developed TMA, acute kidney injury due to complement activation, thrombocytopenia, and immune-mediated myocardial injury. Moreover, among the few thousand SMA patients receiving gene therapy, a few cases developed acquired hemophagocytic lymphohistiocytosis and an uncontrolled, self-perpetuating activation of cytotoxic lymphocytes and macrophages, further supporting the severe consequences of systemic high doses of AAV on immune activation. ⁴⁶ Life-threatening complement activation has been managed with hemodialysis, platelet transfusion, and eculizumab, a monoclonal antibody that binds and inhibits C5. ^41,46

In summary, toxicity is a concern when high AAV vector doses are injected intravenously in patients. ^40,42 Nevertheless, the AAV vector doses needed to achieve sufficient hepatocyte gene transfer are relatively low compared to the doses used for neuromuscular disorders (SMA and DMD) and are not expected to result in significant toxicity. However, there are concerns that acute inflammatory responses might develop at lower doses in rare subgroups of genetically susceptible patients or in patients with underlying disorders, such as hepatic fibrosis. ⁴⁷ A trial of gene therapy in X-linked myotubular myopathy suggested that pre-existing liver disease was responsible for liver failure and death in patients who received a high AAV vector dose. ⁴⁸

Because complement activation is mediated by antibodies, these observations also highlight the importance of dosing both NAb and total anti-AAV antibodies prior to AAV gene therapies because NAb titer provides only partial information about the total amount of complement-activating antibodies. Moreover, they support the importance of immunosurveillance, including serial measurements of complement, D-dimer, and other markers of endothelial activation, especially in the first 30 days after AAV dosing.

Adaptive immune response

Starting from the first gene therapy clinical trial for hemophilia, one of the most recurrent adverse events correlated with AAV is liver inflammation. This is characterized by a subclinical increase in serum aminotransferases, mostly alanine aminotransferase (ALT), often accompanied by an increase in aspartate aminotransferase (AST), without alteration of the synthetic hepatic functions. ¹ The hepatitis is variably associated with a loss of transgene expression, compromising the efficacy of the gene therapy. A T-cell response to the AAV capsid occurring several weeks after the intravenous injections of AAV vectors is responsible for the hepatitis. ⁴⁹ However, the time of resolution of the hepatitis is uncertain and later transaminase elevations have also been observed. ^{7,50 –52} These later responses were associated with expansion of capsid-specific T-cells. Transient treatments with glucocorticosteroids or other immunosuppressants have been effective in blunting the T-cell immune response and to control the increase in liver transaminases, allowing long-term expression of the therapeutic gene. ⁵² Although corticosteroids were the most widely used, a variety of immunosuppressive treatments have been used, including different corticosteroids, tacrolimus, mycophenolate mofetil, cyclosporine, sirolimus, and rituximab. ⁵³ The drawbacks of corticosteroids or other immunosuppressant drugs are the side effects related to the non-specific immunosuppression mandating careful clinical monitoring and appropriate prophylaxis to minimize the risk of infection. Despite the importance of immunosuppression for sustained phenotypic correction, neither a consensus nor specific guidelines are available to recommend the most effective regimen.

The development of capsid-specific CD8+ T-cells is triggered by hepatocytes that function as antigen-presenting cells by expressing the capsid proteins on their surface via MHC class 1 receptors. ^54,55 Although not typically performed in AAV gene therapy trials, the limited availability of liver biopsies showing inflammatory infiltrates supports the proposed mechanism of CD8+ T-cell activation. ⁴⁰ However, transaminitis without cytotoxic T-cell responses, as detected by ELISpot assays, suggests either limited sensitivity of the assay or alternative/additional mechanisms responsible for hepatitis. For example, activation of Toll-like receptor 9 (TLR9) by unmethylated cytidine-phosphate-guanosine (CpG) sequences of the AAV vector genome has been implicated in triggering immune responses against transduced hepatocytes. ⁵⁶

Studies in some hemophilia A patients showed sustained abnormalities of liver enzymes. It is unknown whether the cause of the persistent transaminitis is independent of the capsid-specific CD8+ T-cell response or caused by cellular stress induced by ectopic expression of FVIII. ⁵⁷ Additional putative mechanisms for immune mediated hepatitis include AAV vector preparation purity and host-dependent factors, such as age and the prior presence of inflammation. ^57,58

Liver biopsies can accurately evaluate hepatocyte injury. They can be used to measure expression of the therapeutic gene delivered to hepatocytes, T-cell cytotoxicity, and inflammation. Analyses of a limited number of liver biopsy samples collected from hemophilia A patients who several years earlier received AAV vector infusion found no evidence of inflammation and only noted sinusoidal infiltrates of uncertain significance. ⁵⁹ Although these data are overall encouraging, recent epidemiological findings in children with acute severe hepatitis have raised concerns. ^60,61 In late 2021 and early 2022, after the relieve of social distancing due to COVID-19, a rise in cases of acute severe hepatitis was detected among 3-to-4-year-old children, initially in the United Kingdom and then, worldwide. ^62,63 Surprisingly, high amounts of AAV2 were detected on explanted livers from acute severe hepatitis cases, suggesting a causal involvement of the AAV2 infection in the immune-mediated process resulting in hepatitis. This immune-mediated process is reminiscent of the hepatitis occurring after AAV gene therapy. Although causality between AAV2 and acute hepatitis needs to be proven, these findings raise the hypothesis that infections by AAV2 (and potentially other AAV serotypes) might be pathogenic in a subset of susceptible individuals and that overlapping genetic factors might play a role in the acute severe hepatitis induced by wild-type AAV or in the T-cell immune response induced by AAV vectors in the context of gene therapy.

Liver fibrosis

Liver fibrosis is characterized by excessive accumulation of extracellular matrix (ECM), primarily composed of collagens in the hepatic parenchyma. The buildup of the ECM leads to the formation of scar tissue that disrupts the normal liver architecture and functions. Liver fibrosis results from acquired chronic injury due to various causes such as viral hepatitis, alcohol, metabolic dysfunction-associated steatotic liver disease, and autoimmune diseases. However, liver fibrosis can also result from genetic disorders, including cystic fibrosis, α−1 antitrypsin deficiency, Wilson disease, and others. ⁶⁴ Moreover, liver fibrosis is increasingly being observed in patients with various inherited metabolic disorders previously not linked to liver damage, as improved diagnosis and management are increasing their survival. ^65,66 In addition in genetic disorders, the pathogenesis of liver fibrosis may not be directly linked to the underlying genetic defect, such as chronic hepatitis C infection in hemophilia patients. ⁶⁷ The development and progression of liver fibrosis is often subtle and can take years or even decades to develop. This slow process is difficult to detect by standard changes in serum liver function tests and more specific and sensitive diagnostic tools are needed. ⁶⁸

Liver fibrosis induces various changes that potentially can affect gene therapy: (1) loss of sinusoid fenestrations and deposition of ECM in the perisinusoidal space that hampers blood-borne viral vector interactions with hepatocytes, ⁶⁹ (2) infiltration and proliferation of immune cells that can enhance viral vector scavenging and trigger immune reactions, ^70,71 (3) inflammation associated with fibrosis that can induce priming of immune cells, ⁷² and (4) regenerative responses to damage with increased hepatocyte proliferation that can lead to rapid episomal vector dilution and loss of transgene expression. ⁷³ The risks and the efficacy of AAV vectors in fibrotic livers is poorly understood and has been neglected so far. In most ongoing AAV-mediated liver-directed gene therapy clinical trials, clinically relevant liver fibrosis is an exclusion criterion, because of concerns about reduced therapeutic efficacy and safety. However, only a few preclinical studies investigated AAV-mediated gene transfer in fibrotic livers. Unexpectedly, an early study showed that hepatocyte transduction efficiency by AAV1 was similar between rats with hepatotoxin-induced liver cirrhosis and non-cirrhotic rats. ⁷⁴ Conversely, studies in mouse models of genetic disorders with liver fibrosis showed significant reduction of hepatocyte transduction in fibrotic livers. In Abcb4^-/- mice, a model for progressive intrahepatic cholestasis type 3 (PFIC3) and biliary fibrosis, AAV8 vectors showed a marked reduction of transgene expression. ⁷⁵

Hepatocyte proliferation

Once it enters the host cell nucleus, the AAV vector genome remains largely episomal. However, during cell proliferation the AAV episomal genome cannot replicate, and is rapidly lost. ^76,77 AAV vector genome dilution has been observed in mice after partial hepatectomy ⁷³ and after neonatal AAV vector administration ^76,78 resulting in rapid decline of transgene expression and loss of therapeutic efficacy. ^{79 –81} At one and five weeks post-injection into 1-week-old rhesus macaques, hepatic AAV vector genome content and transduced cells decreased by 20- and 100-fold, respectively. ⁸² However, these preclinical results, especially in rodents, must be translated cautiously to humans because of differences in liver development, including fetal hematopoiesis, ⁸³ and growth between mice and humans: the human liver needs 6 years to reach half the size of an adult liver, ⁸⁴ while the mouse liver growth is completed by 6–8 weeks of age. ⁷⁶

Hepatocyte proliferation also occurs in response to acute or chronic liver damage to regenerate the liver and replenish the loss of hepatocytes. Hepatocyte proliferation can be arrested if the primary cause of the insult is removed. In genetic disorders, removal of the hepatic insult might be pursued by replacing the defective gene, but in most diseases a high percentage of hepatocytes needs to be transduced to reduce cell proliferation. ^{85 –87} Achieving this high level of hepatic transduction with clinically relevant AAV doses is challenging. If insufficient transduction is achieved, liver damage stimulates the regenerative response, leading to the proliferation of hepatocytes and loss of transgene expression. However, there are exceptions with a few disorders that require lower transduction efficiency, such as Wilson disease because corrected hepatocytes can reduce the burden of systemic toxic copper, thereby limiting toxicity on non-corrected hepatocytes. ^86,88

Several strategies have been developed to overcome the limitation of short-term expression by AAV vectors in growing or regenerating livers. One of these strategies is vector re-administration, circumventing the obstacle of anti-AAV NAb, as previously discussed. Alternatively, stable integration of the therapeutic gene into the host genome prevents AAV vector genome dilution because it allows transmission of the transgene into daughter cells when replication occurs. AAV-mediated integrative platforms can be either targeted or non-targeted. Non-targeted integrative strategies are based on transposons. A dual AAV vector system, including piggyBac transposase driving somatic integration of a donor transgene expression cassette, resulted in long-term phenotypic correction in mouse models of PFIC3 and ornithine carbamoyltransferase (OTC) deficiency after administration in newborns. ^75,89,90 However, this approach raises safety concerns, because non-random, untargeted transposase-driven integration frequently occurs within transcriptionally active chromatin regions and in close proximity to transcriptional start sites. ⁸⁹ In contrast, targeted integration of the therapeutic gene holds reduced risks of genotoxicity. Through the generation of a site-specific double-strand break (DSB), zinc-finger (ZFN) endonucleases or CRISPR/Cas9 allow integration of donor DNA at a desired genomic locus that is recognized by a single guide RNA (sgRNA). Repair of DSBs occurs primarily by two pathways: error-prone nonhomologous end joining (NHEJ) and precise homology-directed repair (HDR). HDR mediates seamless integration of a donor DNA delivered by the AAV that bears two homology arms for the target locus where the DSB is generated. ^91,92 In principle, HDR-mediated integration does not require the generation of site specific DSBs by programmable nucleases, because the targeting specificity is provided by the homology arms. However, the development of bespoke genome editing reagents for correction of several specific disease-causing mutations is challenging. Alternatively, integration of a promoterless transgene can be targeted within an ectopic transcriptionally active genomic locus, such as the albumin locus. ^{93 –97} This approach performed without the use of a nuclease exploits spontaneous HDR to integrate a transgenic cDNA preceded by a sequence encoding a porcine teschovirus-1 2A peptide (P2A) at the 3′ end of the Alb gene coding sequence. ⁹⁸ While the albumin promoter ensures robust liver-specific transgene expression, P2A-driven ribosomal skipping permits the therapeutic gene and the albumin to be expressed as separated proteins, without perturbing serum albumin concentrations and function. Being devoid of promoter sequences, the integrated promoterless vectors cannot transactivate neighborhood genes and thus, they are associated with reduced risks of genotoxicity. First-in-human phase I/II studies investigating ZFN in vivo genome editing in mucopolysaccharidosis I (MPS I), MPS II, and hemophilia B were found to be safe and showed evidence of genome editing by the detection of albumin-transgene fusion transcripts in the liver, but no evidence of sustained therapeutic enzyme expression in blood. ⁹⁹ Spontaneous HDR is inefficient resulting in on-target integration in less than 1% of the hepatocytes in healthy livers, ⁹³ which is insufficient to achieve therapeutic benefit in most inherited metabolic disorders. Nevertheless, HDR can be enhanced by CRISPR/Cas mediating DSB at the target locus. ¹⁰⁰ Moreover, in some disorders with ongoing hepatocellular damage, hepatocytes corrected by genome editing have a proliferative advantage over diseased cells and can expand to repopulate the liver. ^{95 –97}

The advantage of HDR-mediated genome editing is precision but the weaknesses of this approach include the need for large homology arms flanking the transgene and the limited efficiency in actively replicating cells. ¹⁰¹ Conversely, NHEJ operates consistently across cell cycle phases, ¹⁰² is more efficient compared to HDR, and does not require the homology arms, thus also enabling the insertion of larger transgenes. ⁹² A recently developed NHEJ-based strategy to achieve therapeutic gene knock-in in both dividing and nondividing hepatocytes entails CRISPR/Cas9-mediated homology-independent targeted integration (HITI). ^{103 –107} In HITI, the donor DNA is flanked at its ends by the same sgRNA sequence as the endogenous target site. Cas9-mediated cleavage at both the endogenous locus and the extremities of the donor DNA prompts NHEJ repair and donor DNA integration at the nuclease-induced DSBs. ^103,108

Within the rapidly evolving toolkit for genome editing, innovative platforms like base editors and prime editors are also available for targeted variant correction. These technologies enable precise modification of specific nucleotides or small regions of DNA without inducing DSBs, thus significantly reducing the risk of unintended mutations or large-scale genomic rearrangements. Both editing systems rely on a CRISPR/Cas complex to guide the machinery to the desired DNA locus, fused to effector enzymes, such as deaminases for base editors and reverse transcriptase for prime editors. The coding sequences for these large protein complexes exceed the packaging capacity of a single AAV vector, necessitating the use of dual AAV vectors for delivery. ^{109 –115}

To overcome the limitations of dual AAV vectors, small Cas orthologs (miniaturized Cas), ¹¹⁶ and streamlined base and prime editing components ^{117 –120} that can fit within a single AAV have been developed. Nevertheless, AAV-mediated delivery of genome editors still raises concerns, particularly for prolonged expression of the editing machinery, risks of off-target editing events, ^{121 –123} and immune reactions. ^124,125 Alternatively, lipid nanoparticles (LNPs) can be used to deliver CRISPR components, in the form of mRNA or ribonucleoproteins. ¹²⁶ In contrast to AAVs, LNPs offer several advantages including transient expression of the editing machinery, lower immunogenicity, and the option for repeated dosing without the risk of eliciting an immune response. ¹²⁷ Nevertheless, AAV vectors are still needed for delivering the donor DNA for targeted gene integration. Another approach to achieve long-term transgene expression entails replication of the episomal AAV genome using scaffold/matrix attachment regions (S/MAR) as anchor points to the nuclear matrix, in close proximity to the nuclear machinery required for DNA replication and gene expression. Their inclusion as cis elements in the vector genome provided with an origin of replication results in episome retention in dividing mammalian cells in vitro. ¹²⁸ However, application of currently available S/MAR sequences in gene therapy settings is hampered by their relatively large size, which surpasses the AAV cargo capacity.

Although still limited, epigenome editing strategies for the treatment of inherited liver diseases have also been investigated. These strategies do not generate DSBs and are based on transcriptional repressors guided to a specific genomic location by a DNA-binding domain, such as a catalytically inactive Cas9 (dCas9) or zinc-finger proteins. These editors can modify the epigenome through histone modifications or DNA methylation affecting the expression of a given target gene. ¹²⁹ Recent findings in mice showed that by a single dose, the mRNA encoding the epigenome editors delivered by LNPs resulted in nearly 50% sustained reduction of circulating PCSK9 that persisted after induced hepatocyte proliferation. ¹³⁰ The vast majority of inherited liver disorders are due to loss of function mutations and applications of epigenome editing have potential for diseases with available substrate reduction therapies, such as inhibition of liver-specific glycolate oxidase (GO) in primary hyperoxaluria type 1. ¹³¹

Genotoxicity

Although persisting as an episome in the nucleus of transduced cells, the AAV genome can also integrate into the host genome in a dose-dependent manner ^132,133 at an estimated frequency of 0.1 to 1 integration event per 100 cells in large animal models injected with doses ranging from 1 × 10¹³ vg/kg to 6 × 10¹³ vg/kg ^{134 –137} or of 0.7 to 4 integration event per 1,000 cells in patients injected with doses ranging from 5 × 10¹¹ vg/kg to 6 × 10¹² vg/kg. ¹³⁷ These integrations preferentially occur within actively transcribed loci, ¹³⁸ thus raising concerns about potential risks of AAV-mediated tumorigenesis. These concerns are supported by clonal integration of wild-type AAVs in hepatocellular carcinoma (HCC) samples from young and non-cirrhotic patients. ^139,140 The proposed mechanisms for tumor development involves integration of the AAV2 rep gene, ^141,142 and transactivation of nearby oncogenes by a liver-specific enhancer sequence lying within the 3′-ITR of the AAV2. ¹⁴³ Being devoid of both rep and 3′-ITR enhancer sequences, AAV vectors are thought to have reduced integration frequency and risks of HCC development. Consistently, AAV vector administration in adult mice and large animals has not shown an increased risks of HCC. ^138,144 Nevertheless, increased incidence of HCC has been detected in mice injected as newborn with high doses of AAV vectors. Most genomic integration events associated with murine HCC mapped at the Rian locus within the Mir341 gene, ^133,145,146 where vector-derived regulatory elements could drive aberrant miRNA expression. ¹³³ The orthologous locus in humans lacks the Mir341, thus preventing direct translation of increased murine HCC to humans. Moreover, this locus is highly expressed during embryonic development and the neonatal period, ¹⁴⁷ potentially explaining the increased prevalence of HCC in newborn animals but not in adults. ¹³⁷ Although long-term studies in dogs and cats revealed no hepatic tumors up to 8–10 years post-vector administration, ^{148 –151} some hemophilia A dogs showed a progressive increase of transgenic FVIII expression associated with clonal expansion and AAV integration in the proximity of cancer genes. ¹³² In non-human primates and patients, AAV-mediated insertional mutagenesis has not been reported so far. While studies in non-human primates are usually short and thus not very informative, long-term follow up (up to 15 years) of the first hemophilia B patients are available and have not revealed any evidence of neoplasia linked to AAV. ¹⁵² Among the five cases of cancer in patients treated with AAV reported so far, none of them was found to be related to AAV integration ¹⁵³ and they were attributed to comorbidities, such as viral hepatitis B and C and metabolic-associated fatty liver disease. ¹⁵⁴

The role of underlying liver disorders on AAV vector genotoxicity is debated. One study investigating tumor formation following systemic AAV administration in mice found increased incidence of HCC in mice with chronic liver injury induced by a high-fat diet. ¹⁵⁵ However, it remains unclear whether liver injury and inflammation directly contributed to tumorigenesis, or if the increased HCC risk was primarily due to the proliferative response of hepatocytes.