Xenogeneic Liver Models for Gene Therapy

I n Greek mythology, Prometheus stole fire from Zeus and gave it to the mortals. Zeus then punished him for his crime by having him bound to a rock while a great eagle ate his liver every day, only to have it grow back to be eaten again the next day. This myth serves as an appropriate metaphor to underscore the remarkable regenerative capacity of human liver. Under normal physiologic conditions, the turnover of hepatocytes is very slow. However, when the liver tissue is damaged, the hepatocytes start to proliferate in an attempt to restore normal liver function. Typically, liver damage can be inflicted after exposure to toxins, chemicals, or viruses or after surgical hepatectomy. The ability to regenerate a normal liver from resident hepatocytes has been exploited to establish mouse models with “humanized” livers. Typically, to generate these xenogeneic liver mouse models, immunodeficient mice are transplanted with human hepatocytes after inflicting a liver injury that selectively damages the mouse liver. Under these circumstances, the human hepatocytes have a selective growth and survival advantage compared with the mouse hepatocytes. Consequently, the human hepatocytes selectively expand and repopulate the mouse liver to generate chimeric “humanized” liver instead.

Several models have been established on the basis of this principle. One of the earliest chimeric “humanized” liver mouse models that have been described was based on an immunodeficient mouse model that was generated by cross-breeding immunodeficient RAG2^–/– mice with a transgenic mouse that overexpressed urokinase-type plasminogen activator (uPA) in the liver, under the control of a mouse albumin enhancer/promoter (Dandri et al., 2001). The hepatocyte-specific expression of the uPA transgene induced extensive liver toxicity leading to chronic hepatic insufficiency (Sandgren et al., 1991). This functional liver deficit created a supportive niche for liver regeneration by the transplanted human hepatocytes. Nevertheless, the number of human hepatocytes present in the mouse liver was relatively low (up to 15%) compared with the transplantation efficiencies previously reported with rat or woodchuck hepatocytes. This was probably due to the use of heterozygous recipients. Instead, transplantation of primary human hepatocytes in homozygous uPA-SCID mice (severe combined immunodeficiency mice transgenic for uPA) resulted in a more robust engraftment, achieving almost complete replacement of the diseased mouse tissue with human hepatocytes (Mercer et al., 2001; Meuleman et al., 2005). This robust repopulation efficiency in the uPA-SCID mice was consistent with their ability to support replication of the hepatitis C virus (HCV), which has an exquisite tropism for human hepatocytes, whereas murine hepatocytes are resistant to HCV. Efficient engraftment of human hepatocytes in the uPA-SCID model likely resulted in production of human complement factors by the engrafted human hepatocytes that caused significant morbidity and mortality in the recipient mice. To overcome this limitation C5/C3 convertase inhibitors were administered, resulting in a significant increase in repopulation efficiency of human hepatocytes and a successful rescue of the uPA-SCID mice (Tateno et al., 2004). Although the uPA-SCID mouse model is permissive to xenotransplantation of hepatocytes it has some specific drawbacks. In particular, high mortality due to bleeding is a major disadvantage and consequently limits surgery needed for transplantation. Moreover, transplantation needs to be performed within a narrow timeframe (5–12 days after birth), and the somatic reversion of uPA-expressing to uPA-defective mouse hepatocytes can mask a successful engraftment. Furthermore, the mice breed poorly and remain unhealthy even after repopulation with human hepatocytes. To overcome these limitations a new alternative “humanized” liver model has been developed. This model is based on the immunodeficient fumarylacetoacetate hydrolase-deficient (FAH^–/–) knockout mouse that was bred with an immunodeficient knockout mouse. FAH^–/– knockout mice suffer from hereditary tyrosinemia that is characterized by toxic accumulation of tyrosine catabolites within hepatocytes (Grompe et al., 1993; Overturf et al., 1996). This mouse is considered to be a superior repopulation model compared with uPA transgenic mice because spontaneous reversion does not occur and repopulation with congenic FAH^+/+ hepatocytes is relatively efficient. Transplantation of this FAH^–/–SCID model with FAH^+/+ human hepatocytes resulted in a repopulation index of 20% (Bissig et al., 2007). In other studies, FAH^–/–SCID mice were subjected to hepatic transduction with an adenoviral vector encoding uPA, yielding robust (90%) repopulation efficiencies with human hepatocytes (Azuma et al., 2007; Shafritz, 2007).

The availability of these different xenogeneic mouse models with “humanized” livers has important implications for gene therapy. In particular, these xenogeneic liver models may more reliably predict vector performance in patient livers. Typically, preclinical studies of liver-directed gene therapy are conducted in normal mice or rodent models that mimic the cognate human disease. Alternatively, large animal models, and in particular canine and nonhuman primates, have been used to assess the consequences of liver-directed gene delivery. Each of these models provides valuable and complementary insights into the molecular and cellular mechanisms of vector–host interactions and allows for preclinical assessment of vector safety and efficacy. However, there may be species-specific idiosyncrasies that make it particularly challenging to predict the performance of a given vector for liver-directed gene delivery in human subjects.

In particular, the use of adeno-associated viral (AAV) vectors derived from alternative serotypes has been proposed as a means to increase hepatic transduction efficiencies and transgene expression levels in human subjects, on the basis of their superior hepatic transduction efficiencies in mice (Gao et al., 2002; Thomas et al., 2004; VandenDriessche et al., 2007). More specifically, AAV8 and AAV9 vectors result in a robust increase in hepatic gene delivery and expression in mouse models compared with the quintessential AAV2 vector. However, no transduction advantage of using AAV8 over AAV2 or AAV5 vectors was apparent in nonhuman primates or dogs (Davidoff et al., 2005; Jiang et al., 2006; Nathwani et al., 2006). Thus, some caution is warranted in translating results on AAV-directed gene delivery from rodent species to larger animals and ultimately to human subjects. Moreover, in vitro transduction of human hepatocytes with AAV vectors is likely not predictive of their performance in vivo. The availability of the various “humanized” liver mouse models outlined previously may help to address some of these outstanding issues. Moreover, these xenogeneic models are valuable alternatives to better assess the impact of AAV serotype switching, AAV retargeting, or AAV capsid modifications (Choi et al., 2005; Grimm et al., 2008; Kühnel and Kubicka, 2008) on hepatic gene delivery in human hepatocytes. Comprehensive studies are needed to compare the relative hepatic transduction efficiencies of these different AAV vector types in these “humanized” liver models.

Lentiviral (LV) vectors pseudotyped with different viral envelopes have been used to transduce mouse liver, resulting in sustained expression of therapeutically relevant proteins and phenotypic correction of gene defects in hepatocytes (Follenzi et al., 2004; Kang et al., 2005; VandenDriessche et al., 2002, 2007; Matrai et al., 2009). Hepatic transduction with LV vectors compared favorably with that obtained with AAV2 vectors. However, AAV2 requires portal vein injection, whereas LV vectors can achieve comparable levels of gene transfer by systemic administration. We demonstrated the superior hepatic transduction efficiency with AAV8 or AAV9 vectors compared with LV vectors, at least in mice (VandenDriessche et al., 2007). However, this may not necessarily translate into a proportionate difference in transduction efficiencies in large animal models or ultimately in human subjects. Moreover, LV vectors can efficiently transduce human hepatocytes in vitro, in contrast to mouse hepatocytes, which are relatively resistant to LV vector transduction in vitro (Nguyen et al., 2002). The “humanized” liver mouse models are therefore particularly well suited to resolve this controversy and to compare hepatic transduction efficiencies of LV vectors pseudotyped with various envelopes. In particular, this model enables assessment of in vivo transduction with LV vectors pseudotyped with hepatotropic envelopes derived from human hepatitis B and C viruses (HBV, HCV) (Bartosch et al., 2003; Hsu et al., 2003). Similarly, the performance of hepatotropic nanoparticles displaying HBV envelopes on their surface can ideally be assessed in this xenogeneic model (Russell, 2003; Yamada et al., 2003).

The potential of using these “humanized” liver mouse models for gene therapy studies is also highlighted by Kubo and colleagues in this issue of Human Gene Therapy. Although helper-dependent adenoviral (HD-AdV) vectors can efficiently deliver genes into hepatocytes, transgene expression is not stable because hepatocytes slowly divide, resulting in the gradual loss of the nonintegrated extrachromosomal HD-AdV genomes (Chuah et al., 2003). Treatment of hereditary diseases may require more stable, long-term transgene expression, which can be achieved only through stable genomic integration or episomal replication of vector genomes. To overcome this limitation, Kubo and colleagues therefore developed a hybrid helper-dependent adenoviral–lentiviral vector (designated as HL) that combines the ability of HD-AdV vectors to efficiently transduce hepatocytes with the integrating properties of LV vectors. To achieve this, LV vector and helper sequences were packaged into the HD-AdV vectors, resulting in the in situ production of LV particles and subsequent transduction of hepatocytes. However, to support production of LV vectors, human cells are preferred because in rodent cells there are multiple blocks to HIV replication at the level of cellular entry, gene expression, and particle assembly. The “humanized” liver uPA-SCID mouse model was therefore well suited to assess the performance of the HL hybrid vector. In this study, human hepatocyte repopulation indices ranging from 60 to 90% could be obtained after pretreatment of the recipient mice with a complement inhibitor. Whereas the parental HD-AdV vector yielded only transient expression of the reporter genes in this model, expression with the HL hybrid vector was relatively stable, resulting in a stable hepatic transduction efficiency of about 30%. This is consistent with stable genomic integration of the LV genomes in the human hepatocytes that had repopulated the livers of the uPA-SCID mice. It cannot be excluded that the transplanted human hepatocytes in this xenogeneic setting exhibit some differences from normal human hepatocytes in their natural environment. Interestingly, when uPA-SCID mice with “humanized” livers were injected with HD-AdV vectors, reporter gene expression declined more rapidly than was initially anticipated. This could possibly be ascribed to a higher residual hepatocyte proliferation rate than in normal livers, even at 6–8 weeks posttransplantation when a repopulation index of 60–90% had been achieved (Emoto et al., 2005). Moreover, it is possible that the de novo expression of LV antigens in the HL-transduced hepatocytes may have induced some hepatotoxicity that consequently contributed to increased hepatocyte proliferation.

Although these results further underscore the potential of using “humanized” liver mouse models for liver-directed gene therapy, they have some intrinsic limitations that impact on the preclinical assessment of vector safety and efficacy. In particular, these various xenogeneic liver models are derived from immunodeficient mice. Although the lack of a functional immune system is a prerequisite to enable efficient human hepatocyte engraftment, this obviously precludes analysis of the immune consequences of gene transfer. One of the main challenges of in vivo liver-directed gene delivery pertains to the induction of immune responses that could potentially eliminate the transduced cells and/or the therapeutic proteins. In particular, in a liver-directed gene therapy trial for hemophilia, administration of AAV-2 vectors encoding clotting factor IX (FIX) resulted in a dose-dependent T cell response that cleared the transduced hepatocytes expressing FIX, consistent with the decline in FIX expression levels and transient elevation of liver enzymes in the plasma (Manno et al., 2006). This T cell response was directed at the AAV capsid-derived antigenic peptides that were presented in association with MHC-I on the surface of the transduced human hepatocytes, possibly via “cross-presentation” (Pien et al., 2009). This response was not predicted on the basis of preclinical mouse, dog, or nonhuman primate studies, even when these animals were immunized with AAV capsid antigens (Pierce et al., 2007). To more closely mimic the clinical setting and to assess immune responses in the context of AAV-transduced human hepatocytes, one could adoptively transfer specific T cells into these “humanized” liver mouse models. Similar strategies have been employed previously in human peripheral blood lymphocyte (huPBL)–nonobese diabetic/severe combined immunodeficiency (NOD-SCID) models to assess human immune response to viral antigens (Cao et al., 2001).

In conclusion, these xenogeneic “humanized” liver models have implications for the gene therapy of hereditary diseases. In addition, they can be used to develop new approaches to inhibit HBV or HCV replication by RNA interference (Andino, 2003; McCaffrey et al., 2003). Finally, they are suitable in vivo models to assess the therapeutic potential of induced pluripotent stem (iPS) cells for the treatment of diseases affecting liver function or resulting from a defective gene that is normally expressed in the liver (Xu et al., 2009). Although these “humanized” liver models have their own shortcomings, they allow us to better understand the consequences of liver-directed gene transfer in an experimental setting that may more closely mimic the human condition. This may facilitate the development and preclinical validation of improved liver-directed gene therapy approaches and ultimately pave the way toward gene therapy clinical trials in patients suffering from liver disease and genetic disorders.