Research ArticleGene Therapy

A universal system to select gene-modified hepatocytes in vivo

See allHide authors and affiliations

Science Translational Medicine  08 Jun 2016:
Vol. 8, Issue 342, pp. 342ra79
DOI: 10.1126/scitranslmed.aad8166

Gene therapy gets selective

In gene therapy, most quip that the top three challenges are delivery, delivery, and delivery, but selectively expanding the pool of gene-edited cells is a major challenge, too, to ensure that genes reach therapeutic levels. Nygaard et al. came up with a clever platform technology that selects for gene-edited cells in vivo without the hassle, time, and special facilities required for in vitro expansion and selection via cell culture. Alongside the therapeutic transgene, the authors inserted into hepatocytes a short hairpin RNA targeting an enzyme that, when knocked down, made the cells resistant to a drug called CEHPOBA. Healthy animals received liver-specific vectors to express a model gene, human factor 9, and then were given CEHPOBA or saline for several weeks. The animals receiving saline control saw no change in gene expression in hepatocytes, whereas animals receiving the drug CEHPOBA saw an order of magnitude increase in factor 9, indicating that the gene-corrected cells were pharmacologically selected in a living animal. This powerful approach can be used for genetic diseases like hemophilia B and metabolic liver diseases or extended to any tissue that proliferates after injury, including the bone marrow and skin.


Many genetic and acquired liver disorders are amenable to gene and/or cell therapy. However, the efficiencies of cell engraftment and stable genetic modification are low and often subtherapeutic. In particular, targeted gene modifications from homologous recombination are rare events. These obstacles could be overcome if hepatocytes that have undergone genetic modification were to be selectively amplified or expanded. We describe a universally applicable system for in vivo selection and expansion of gene-modified hepatocytes in any genetic background. In this system, the therapeutic transgene is coexpressed with a short hairpin RNA (shRNA) that confers modified hepatocytes with resistance to drug-induced toxicity. An shRNA against the tyrosine catabolic enzyme 4-OH-phenylpyruvate dioxygenase protected hepatocytes from 4-[(2-carboxyethyl)-hydroxyphosphinyl]-3-oxobutyrate, a small-molecule inhibitor of fumarylacetoacetate hydrolase. To select for specific gene targeting events, the protective shRNA was embedded in a microRNA and inserted into a recombinant adeno-associated viral vector designed to integrate site-specifically into the highly active albumin locus. After selection of the gene-targeted cells, transgene expression increased 10- to 1000-fold, reaching supraphysiological levels of human factor 9 protein (50,000 ng/ml) in mice. This drug resistance system can be used to achieve therapeutically relevant transgene levels in hepatocytes in any setting.


Liver-directed gene therapy has the potential to produce permanent cures for many hepatic disorders, in particular inborn errors of metabolism (1). Recently, some clinical success has been achieved with hemophilia B, using recombinant adeno-associated viral (rAAV) vectors (2). Despite this success, there are limitations to this method. rAAV is an episomal vector, and transgene expression is rapidly lost during cell division, as is the case in the growing liver of young children (3). High vector doses are required and can result in anticapsid immune responses (46). In addition, concerns about insertional mutagenesis and cancer remain (79). To achieve both transgene persistence with replication and to minimize insertional mutagenesis, several groups, including ours, have reported methods to achieve site-specific gene targeting in hepatocytes (1012). Proof-of-principle experiments have been performed; however, very high vector doses and/or targeted nucleases (zinc fingers, TALENs, and CRISPR/Cas9) are needed to achieve reasonable efficiencies of gene repair (1315). The clinical potential of targeted nucleases is currently unknown, and it is unlikely that the reported rAAV vector doses could be tolerated in humans.

Many of these obstacles could be overcome if it was possible to amplify and selectively expand hepatocytes that have already undergone the desirable genetic modification. Lower vector doses could be used, and it may be possible to achieve therapeutic levels of gene targeting without nucleases. It is known that hepatocytes can be massively selected in vivo in both mice and humans in settings where the host liver has a genetic disadvantage (16). Gene-corrected hepatocytes have a selective advantage in Wilson’s disease (17), α1-antitrypsin deficiency (18), progressive familial intrahepatic cholestasis (19), and hereditary tyrosinemia (16, 20). These rare examples highlight the inherent regenerative capacity of mature hepatocytes. However, there is no selective advantage for transplanted or genetically corrected hepatocytes in most liver diseases, including phenylketonuria, hemophilias, or urea cycle disorders.

Genetic hepatocyte selection is strongest in fumarylacetoacetate hydrolase (FAH) deficiency and is known to occur in the human liver (16). We therefore set out to develop an in vivo selection system that takes advantage of the tyrosine catabolic pathway, which converts tyrosine to fumarylacetoacetate (FAA), to be hydrolyzed by the enzyme FAH into fumarate and acetoacetate (Fig. 1A). Hepatotoxicity in Fah deficiency is caused by accumulation of FAA. We previously developed small-molecule inhibitors of FAH and showed that treatment with these drugs faithfully mimics human genetic FAH deficiency in mice (2123). In addition, it has been shown that mutations in the tyrosine catabolic pathway upstream of FAH can prevent the accumulation of FAA, completely protect Fah−/− hepatocytes, and prevent liver disease in mice (24, 25). Pharmacological blockage of 4-OH-phenylpyruvate dioxygenase (HPD) with the small-molecule 2-[2-nitro-4-(trifluoromethyl)benzoyl] cyclohexane-1,3-dione (NTBC) is the standard treatment for human FAH deficiency (Fig. 1A) (26, 27).

Fig. 1. Identification of an shRNA that rescues Fah deficiency.

(A) The tyrosine catabolic pathway. Genetic deficiency of Fah causes hereditary tyrosinemia type 1 (HT1) due to accumulation of FAA in hepatocytes. The disease can be treated pharmacologically (NTBC) or by shRNA knockdown of the genes required for making FAA. CEHPOBA (4-[(2-carboxyethyl)-hydroxyphosphinyl]-3-oxobutyrate) inhibits FAH and causes accumulation of FAA. TAT, tyrosine aminotransferase; MAI, maleylacetoacetate isomerase. (B) Lentiviral construct. shRNAs targeting Hpd, Hgd, or Tat were expressed from a U6 promoter. Vector expresses a green fluorescent protein (GFP) reporter. LTR, long terminal repeat; UbC, ubiquitin promoter. (C) Experimental timeline. Neonatal Fah−/− mice were injected with shTat, shHpd, and shHgd lentiviruses or a vector devoid of any shRNA (control) and kept on NTBC until weaning. NTBC was then withdrawn to permit liver injury and selection of resistant hepatocytes. (D) Mouse weights during selection starting at 5 weeks of age. Gray bars represent periods of intermittent NTBC therapy. Only the control and shHgd and shTat cohorts were given NTBC after week 6. Data are means − SD (downward tick) (n = 4 to 6). (E) Mice were injected with a nonselectable vector (control) or shHpd. Liver tissues were stained for the reporter GFP and for α-fetoprotein (AFP), which is highly expressed in mutant Fah−/− hepatocytes. Yellow arrow denotes the absence of AFP within a selected nodule compared with AFP-positive surrounding tissue (black arrows). Scale bars, 100 μm. (F) Polymerase chain reaction (PCR) amplification of genomic liver DNA with primers flanking the lentiviral-shRNA sequence (two primer sets).

Taking advantage of this principle, we recently demonstrated that hepatocytes genetically deficient in homogentisic acid dioxygenase (HGD) (Fig. 1A) could be strongly selected in wild-type mice treated with the FAH inhibitor CEHPOBA (23). On the basis of this finding, we reasoned that short hairpin RNA (shRNA)–mediated knockdown of enzymes upstream of FAH would make hepatocytes resistant to CEHPOBA and achieve their in vivo selection (Fig. 1A). Here, we report the development of a versatile system that provides potent hepatocyte selection in vivo in mice, independent of genetic background, and can be used to amplify therapeutic cells in multiple settings.


Selection of a protective shRNA

To determine whether knockdown of the gene encoding Tat, Hpd, or Hgd would result in hepatocyte selection (Fig. 1A), a small lentiviral library was tested in Fah-deficient mice. Four shRNAs (table S1) were designed against each candidate gene and cloned into a lentiviral backbone under a U6 promoter (Fig. 1B). Neonatal Fah−/− mice were injected with shTat, shHpd, or shHgd lentiviruses via the facial vein and kept on NTBC until weaning (Fig. 1C). NTBC was then stopped to permit liver injury and selection of resistant hepatocytes. Only mice injected with the Hpd shRNA library gained weight after complete NTBC withdrawal, indicating the emergence of FAA-resistant hepatocytes (Fig. 1D). Animals injected with an Hgd shRNA, a Tat shRNA, or a control lentivirus devoid of an shRNA required reintroduction of intermittent NTBC therapy to maintain weight (interrupted gray bars in Fig. 1D).

After several weeks of selection, the livers were harvested and analyzed histologically. Mice injected with the shHpd library showed clear evidence of regenerative nodules (Fig. 1E). These nodules consisted of healthy-appearing hepatocytes staining positive for the GFP transgene and were negative for the damage marker AFP. Next, the shRNA sequences were rescued by PCR and sequenced (Fig. 1F). Only a single shHpd sequence, 5′-CCGGGCCTCAGAATGGTACCTGAAACTCGAGTTTCAGGTACCATTCTGAGGCTTTTTG-3′, was retrieved from multiple weight-stabilized Fah−/− mice and used for all future experiments.

In vivo selection of an integrating rAAV vector

We have previously shown that AAV vectors harboring homology to ribosomal DNA (rDNA) have increased integration frequency in hepatocytes (28, 29), but their absolute efficiency of chromosomal integration is still very low. We therefore constructed rDNA vectors containing a human factor 9 (hF9) transgene as well as the selectable Hpd shRNA (Fig. 2A). Twenty-five–day–old post-weaning Fah−/− mice were injected with 1 × 1011 vector genomes (vg) each of the vector, kept on NTBC for 2 weeks after injection, and then subjected to selective pressure (Fig. 2B). All mice injected with the rDNA-F9-shHpd vector gained weight after complete NTBC withdrawal, whereas control vector–injected animals required continued NTBC therapy to maintain their weight (Fig. 2C). Similarly, hF9 levels rose significantly and continuously in response to NTBC withdrawal (Fig. 2D), indicating expansion of FAA-resistant transgene-expressing hepatocytes. This result demonstrates that a transgene linked in cis to the selectable shRNA was amplified, leading to therapeutic levels of transgene expression unachievable without selection.

Fig. 2. Selection of integrating rAAV vectors.

(A) Ribosomal rAAV constructs capable of chromosomal integration. Top: Standard vector control. Bottom: Selectable Hpd shRNA vector. Both vectors contained rDNA homology arms to enhance chromosomal integration, and both expressed hF9. ITR, inverted terminal repeat. (B) Experimental timeline. Selection: Periods of NTBC withdrawal were different between cohorts and are indicated in (C). (C) Weights and NTBC administration in animals receiving control or shHpd vector. Black rectangles indicate periods of NTBC treatment in control mice, whereas gray bar is NTBC for mice receiving shHpd. Data are means ± SD (n = 3 to 5). (D) Plasma hF9 measured by enzyme-linked immunosorbent assay (ELISA). Data are means ± SD (n = 4 to 5). *P < 0.05, **P = 0.01 versus controls, by Student’s two-tailed t test assuming equal variance.

In vivo selection of gene-targeted hepatocytes

Recently, rAAV-mediated targeted homologous recombination into the highly expressed albumin gene was used to achieve therapeutic levels of transgene expression in the liver (30). These “GeneRide” constructs are promoterless and hence are predicted to have reduced cancer risk from insertional mutagenesis upon random integration (8, 31). However, the efficiency of targeted integration was <1% even when high vector doses were used in neonatal animals. To determine whether these rare gene-targeting events could be amplified in vivo, we tested two different vector designs. First, we inserted a cassette containing the selectable shHpd driven by a U6 promoter into an intron of the albumin gene (Fig. 3A). Second, a promoterless construct was generated. Here, the shHpd was embedded within a microRNA (miRNA) and inserted in an albumin intron downstream of the therapeutic F9 cassette (Fig. 3A). The intent was to have the shRNA expression driven by the chromosomal albumin promoter and hence induce Hpd gene knockdown only after homologous recombination.

Fig. 3. Selection of targeted integrations in the albumin locus.

(A) GeneRide rAAV constructs designed for chromosomal integration. Top: The selectable shHpd construct is driven by the pol3 U6 promoter. Below: The shHpd is embedded within a miRNA and controlled by the endogenous albumin promoter. Both albumin-targeted GeneRide vectors encoded hF9 complementary DNA (cDNA) flanked by mouse albumin homology arms. The structures of the wild-type and gene-targeted albumin locus are also shown. Homologous recombination led to generation of a fused mRNA transcript. RNA processing liberated the shHpd and ribosome skipping at the 2A peptide coding sequence generated separate mouse albumin (Alb) and hF9 proteins. (B) Experimental timeline. Selection cycles starting at week 5: off NTBC for 3 weeks, then on for 5 days, until week 20. (C) Plasma hF9 measured by ELISA in mice treated with the selectable or control GeneRide rAAV. The 5 and 100% levels of normal hF9 blood levels are shown with a dashed line. Data are means ± SD (n = 4). **P < 0.01, ***P < 0.001 versus controls, by Student’s two-tailed t test assuming equal variance. (D) hF9 liver immunohistochemistry showing representative nodules from mice with plasma F9 levels of 38,000 and 27,000 ng/ml (high), 800 ng/ml (low), and a control (no selection). Scale bars, 100 μm. (E) Plasma hF9 levels in mice treated with the selectable GeneRide rAAV and subjected to NTBC withdrawal from 6 to 9 weeks of age followed by reintroduction of NTBC thereafter. The dashed line indicates the therapeutic level of hF9 (250 ng/ml or 5%). Data are single measurements from individual mice (n = 3).

Neonatal Fah−/− mice were injected with 1 × 1011 vg each, kept on NTBC until weaning, and then subjected to selection by withdrawing NTBC (Fig. 3B). hF9 levels were monitored during NTBC withdrawal. Increases in F9 level were only observed with the promoterless vector design in which the endogenous albumin promoter drives the shHpd after gene targeting. Starting from a low baseline, selective pressure resulted in greater than 100-fold increase in hF9 levels and produced very high, even supraphysiologic levels of hF9, exceeding 40,000 ng/ml (Fig. 3C). This result indicated that gene-targeted hepatocytes could be selectively expanded using an miR-embedded protective shRNA.

To confirm that selection had occurred, histology was performed in animals with plasma hF9 levels ranging from high (38,000 ng/ml) to low (800 ng/ml) (Fig. 3D). As expected, a high percentage of their liver mass (up to 50%, estimated by eye) had been replaced by hF9-producing hepatocytes, whereas less than 0.1% of cells were positive in the controls. In addition to histology, molecular analysis was performed to detect properly targeted albumin alleles. PCR amplification demonstrated the amplification products expected from homologous integration, which were confirmed by sequencing (fig. S1). In aggregate, these data prove that gene-targeted hepatocytes expanded more than 100-fold in response to FAA-mediated selection producing very high levels of therapeutic transgene expression.

High levels of shRNA expression can be toxic in the liver (32). To test whether continuous expression of the protective shRNA was toxic to gene-targeted hepatocytes, animals underwent NTBC withdrawal until therapeutic levels of hF9 (250 ng/ml or 5%) were reached (dashed line in Fig. 3E) and were then put back on NTBC to halt selective pressure. We hypothesized that the hF9 plasma concentration would drop with time if gene-targeted hepatocytes were damaged by the shRNA expression. hF9 levels remained stable on NTBC, indicating that selective shHpd did not cause hepatocyte death (Fig. 3E). To further demonstrate that the expanded hF9-expressing nodules consisted of healthy cells, markers characteristic of normal hepatic zonation and function were evaluated and found to be normal (Fig. 4A). Additionally, hepatocyte morphology and cell cycle status were normal in selected nodules (Fig. 4A).

Fig. 4. Histology of shHpd selected hF9-positive nodules.

(A) Serial sections of representative fields from three separate Fah−/− mice after multiple cycles of selection (NTBC withdrawal). Mice were given either the shHpd vector or saline control. The adjacent sections were stained for expression of hF9; glutamine synthase, a marker for hepatocytes adjacent to the central vein of the hepatic lobule (zone 3); hematoxylin and eosin (H&E); and the proliferation marker Ki67. The “x” and “o” symbols mark vessels as landmarks for lining up the serial sections. For shHpd mouse 1 (shHpd 1), serial sections are the same area in low magnification (×62); for shHpd 2 and the control mouse, serial sections are in high magnification (×150). Black arrows indicate nodules expressing hF9. Scale bars, 100 μm. (B) In vivo selection after gene transfer in adults. Four adult male Fah−/− mice treated from birth with NTBC were injected with 8 × 1011 vg of the GeneRide vector (Fig. 3A) at day 52. NTBC therapy was stopped 10 days later. hF9 levels were measured. Data are averages ± SD (n = 4). Statistical differences between hF9 levels at different time points were evaluated by Student’s two-tailed t test assuming equal variance.

To establish that this approach could also work not only after neonatal injection but also in adult animals, four mice 8 weeks of age were injected with 8 × 1011 vg of the selectable GeneRide vector each. After NTBC withdrawal, hF9 levels increased more than 20-fold (Fig. 4B).

Pharmacologic selection in wild-type animals

Transgene selection in Fah−/− mice was only used as proof of principle to develop a gene therapy module that could be selected by resistance to high levels of FAA. To apply this system in any genetic liver disease, selection must work in animals that are not deficient in FAH. We previously described a small-molecule inhibitor of FAH that was capable of inducing FAA accumulation and exerting selective pressure in vivo in mice (22, 23). To demonstrate that the Hpd shRNA could be used in any genetic background, neonatal wild-type C57Bl6 mice were injected with 2 × 1011 vg of the promoterless F9 GeneRide construct. Starting at 4 weeks of age, three treated animals were injected daily with CEHPOBA; littermate controls that had also been given vector were given saline instead (Fig. 5A).

Fig. 5. Selection of gene-targeted hepatocytes in wild-type mice using pharmacologic FAH inhibition.

(A) Experimental timeline. Beginning at 4 weeks of age, the mice were given daily intraperitoneal injections of CEHPOBA (1 μmol/g) or saline until 8 weeks of age. (B) Plasma hF9 measured by ELISA in mice treated with CEHPOBA or saline. Data are individual mice (n = 3). Dashed line denotes 5% of normal F9 levels, which is considered therapeutic. The gray rectangle indicates CEHPOBA treatment periods. A second period of CEHPOBA treatment was administered to only one CEHPOBA mouse (inverted triangle). (C) hF9 immunohistochemistry showing representative liver nodules from two separate CEHPOBA-treated mice and two saline-treated controls. Arrows denote hF9-positive hepatocytes. Scale bars, 100 μm. (D) H&E staining of liver and kidney tissues from a representative CEHPOBA-treated animal and saline-treated control. Scale bars, 100 μm.

CEHPOBA administration was well tolerated, and hF9 levels in plasma rose steadily during the injection period (Fig. 5B), reaching therapeutic levels of >250 ng/ml in all cases. Histological analysis demonstrated clusters of hepatocytes expressing hF9, as expected from clonal expansion of gene-targeted hepatocytes in CEHPOBA-treated animals (Fig. 5C). In contrast, only rare, single hF9-positive hepatocytes were seen in saline controls. After 4 weeks of injections and at therapeutic levels of hF9, selection was stopped and hF9 levels were monitored. hF9 levels were stable in two animals but declined in the third (Fig. 5B). A second round of CEHPOBA injections was initiated in this animal, resulting in an increase of transgene expression back into the therapeutic range.

All mice were followed for another month and then harvested to assess renal and hepatic function and histology (Table 1 and Fig. 5D). No differences in kidney or liver function in treated and normal C57Bl6 mice were detected, demonstrating that the selective regimen had no permanent negative effects on these organs. Hepatocytes of CEHPOBA-treated livers displayed minimal variation in size of the cell and the nuclei (Fig. 6A). The patterns of the CEHPOBA-treated animal were completely normal with rare Ki67+ hepatocytes (no proliferation), normal homogeneous hepatocyte nuclear and cellular size, normal Hnf4α staining (all hepatocyte nuclei positive), and normal zone 3 distribution of CYP2E1 (brown cytoplasmic staining next to vessels). Other cell types, including bile ducts, endothelial cells, Kupffer cells, and stellate cells, were also unremarkable. In contrast, Fah-deficient animals have severe liver damage. H&E staining showed hepatocyte macrocytosis, nuclear ballooning, and pleiocytosis (Fig. 6A).

Table 1. Liver function tests after CEHPOBA selection.

Liver function parameters 1 month after the end of CEHPOBA or saline selection. Data are means ± SD (n = 3 animals per group). P values were determined by Student’s two-tailed t test assuming equal variance. ALT, alanine aminotransferase; AST, aspartate aminotransferase.

View this table:
Fig. 6. Histology of CEHPOBA-treated livers.

(A) Liver histology of a representative CEHPOBA-treated animal and a saline control 1 month after stopping the drug. An untreated age-matched Fah−/− off NTBC (a positive control from a separate experiment) is shown at the right side for comparison. Serial sections were stained with H&E, followed by antibody stains for Ki67, HNF4α, and CYPE1. Black arrows show portal bile ducts. The “*” marks central veins, and “o” delineates portal veins. Pleiocytosis is marked by yellow arrows. (B) Relative abundance of fused transcripts/mouse albumin transcripts from saline-treated C57Bl6 mice, CEHPOBA-treated C57Bl6 mice (8 weeks after completing selection), and NTBC-cycled Fah−/− mice injected with either control vector or the GeneRide vector harboring the shHpd selection cassette (20 weeks of age; see Fig. 3C). Data are individual animals (n = 3) with means ± SD.

Finally, molecular analysis was performed using quantitative reverse transcription PCR to detect and measure mRNA molecules that were the product of gene targeting. The relative abundance of mouse albumin mRNA from untargeted chromosomes was compared to that of hybrid mRNAs. PCR primer pairs specific to the hybrid mouse albumin–hF9 transcripts were used. As expected from the hF9 levels in blood, CEHPOBA selection increased the hybrid mRNA levels by ~1000-fold compared to vector-injected wild-type animals that received no selection (Fig. 6B). Similarly, we found high levels of hybrid albumin mRNA derived from targeted hepatocytes after genetic selection in vector-injected Fah−/− mice (Fig. 6B). Together, these results demonstrate that CEHPOBA treatment can be used to amplify gene-targeted hepatocytes and significantly enhance the therapeutic transgene expression regardless of genetic background.


The need for cytotoxic conditioning is necessary in hematopoietic stem cell gene therapy and bone marrow transplantation to achieve therapeutic levels of healthy stem cells (33). Typically, the conditioning regimen is applied before transplantation of stem cells, such as the enzyme O6-methylguanine-DNA methyltransferase (MGMT), which renders the cells resistant to the chemotherapy drug 1,3-bis(2-chloroethyl)-1-nitrosourea (BCNU) (34). Clinical trials using this combination of cytotoxic drug and resistance gene have been successful in protecting the bone marrow during intense chemotherapy of glioblastoma (35). Although hepatocytes are known to harbor stem cell–like regenerative capacity and can be extensively selected in certain genetic conditions, aclinically relevant drug/resistance gene combination for the liver—analogous to BCNU/MTMG—has not been developed to date. Overexpression of the oncogene Bcl2 can protect hepatocytes from Fas-mediated apoptosis (36, 37) and produce extensive hepatocyte selection. However, this approach has little clinical potential because the protective gene renders cells permanently apoptosis-resistant and, hence, cancer-prone.

We have previously shown that transplanted hepatocytes can be selected with CEHPOBA in vivo if they carry a genetic mutation in a tyrosine catabolic enzyme (23). However, patients with such mutations are rare, leaving most clinically relevant metabolic liver diseases unamenable to this powerful selection. Here, we demonstrated that shRNA-mediated knockdown of Hpd can protect hepatocytes of any genotype against the effects of the potent FAH inhibitor CEHPOBA. Hepatocytes expressing this shRNA were strongly selected in mice, resulting in a 10- to 30-fold expansion in only 4 weeks. This method worked in wild-type mice with no genetic liver disease. Any transgene connected in cis to the protective shRNA may be coselected, making this system applicable to any liver-directed gene therapy producing stably modified cells. Potential target diseases include hemophilia B and metabolic liver diseases with neonatal onset, such as urea cycle disorders. Normal liver growth in infants leads to rapid loss of episomal transgene expression (3), but the ability to select for chromosomally integrated genes may overcome this problem. Although the expansion achieved with the pharmacological agent CEHPOBA was at maximum ~30-fold over 4 weeks, the experiments performed in mice with genetic Fah−/− deficiency demonstrated that expansion of >100-fold is feasible with longer selection periods.

The Hpd shRNA system is versatile and not limited to any particular delivery vector. We showed equally potent selection with both lentiviral vectors and integrating rAAV vectors. Any gene delivery system that produces chromosomal integration of a transgene cassette can be selected. Therefore, potential gene delivery platforms include nonviral systems, such as Sleeping Beauty and PiggyBac transposons, as well as typical viral vectors, such as oncoretroviral, lentiviral, and rAAV vectors, capable of integration. Lentiviruses in particular may be of interest because of their large packaging limit and could therefore be used for transgenes too big to fit into rAAV vectors. To apply our approach to cell therapy, the donor hepatocytes would have to be genetically modified to express Hpd shRNA before transplantation. This could be achieved with a variety of ex vivo gene delivery systems.

Although our initial constructs expressed the protective shRNA from the strong, nonspecific U6 promoter, we also showed that selection could be achieved when a tissue-specific promoter drove expression. To make this possible, the Hpd shRNA was embedded in a miRNA such that it could be processed from a Pol2 transcript. We were able to incorporate this approach into a promoterless rAAV construct designed to produce transgene expression only after homologous recombination into the hepatocyte-specific albumin locus (38). Although the initial targeting frequency with this system has been estimated to be only <1%, gene-targeted transgene-expressing hepatocytes constituted ~50% of the liver after selection. Our vector design permitted in vivo pharmacological selection of a specific gene-targeting event in hepatocytes, which has not been possible before.

Given the recent concerns about the oncogenic potential of random integration of rAAV vectors containing liver-specific promoters (79, 31, 39), the ability to use promoterless constructs provides an important safety feature. Although the frequency of random rAAV integrations is not affected by concurrent gene-targeting (11), randomly integrated vectors without promoters cannot trigger oncogene activation. Efficient gene therapy using specific genome modifications as opposed to random gene addition should now be feasible, even without the use of site-specific nucleases. No adverse side effects of CEHPOBA were seen in our gene-selected mice several weeks after the drug was stopped. Liver function reverted back to normal within 4 weeks of stopping selection. However, CEHPOBA is not a U.S. Food and Drug Administration (FDA)–approved drug, and no clinical experience with transient FAH blockage exists in humans. Conversely, much is known about the natural history of FAH deficiency in humans, which is informative regarding the predicted risks of transient FAH inhibition in humans. Several studies have shown that the effects of FAH deficiency in humans are completely reversible as long as NTBC therapy is started before 6 months of age (27, 40). Most patients treated before 2 years of age have excellent long-term outcomes. These studies suggest that transient inhibition of FAH is likely to be tolerated in humans and not lead to long-term sequelae, including liver cancer. Given the high efficiency of our selection system shown herein, large animal experiments to further evaluate the safety and clinical potential of this approach now appear warranted.

The impact of our results is not narrowly confined to only the tyrosine catabolic pathway or the liver. There are many drugs that are specifically hepatotoxic, and the toxin/protective shRNA selection approach described herein could apply to many of these. Some drugs have to be metabolically activated to be hepatotoxic (41); others require specific transporters to enter the hepatocyte (42). Each of these activation-dependent liver toxins, many of which are FDA-approved medicines, represents an opportunity to develop a clinically applicable hepatocyte selection system. Furthermore, the overall approach should apply to any tissue or cell type capable of proliferation after injury, including the bone marrow, intestine, renal tubules, or skin.


Study design

Our study involved comparisons between cohorts of mice. In some experiments, different gene therapy vectors were compared to each other. All mice were exposed to selective pressure but received vectors to determine which vector design permitted in vivo selection. Presence or absence of selection was determined by measurement of blood level of hF9 or measurement of vector copy number in liver DNA. Laboratory personnel performing the hF9 ELISAs were blinded to the treatment regimen. In other experiments, mice were treated with the same vector but exposed to different selection regimens. Again, both hF9 measurements and vector copy number were measured to determine whether selection had occurred. All measurements were performed at least in duplicate, and all in vivo experiments were repeated multiple times. The precise number of experimental animals and replicates is given in the figure legends. The number of experimental animals was chosen to permit the detection of at least 10-fold differences in either hF9 levels or DNA copy number with a P value of <0.05 determined by an unpaired two-tailed t test. Animals were assigned to experimental cohorts randomly.

Mouse strains and animal husbandry

All animal experiments were performed according to the guidelines for animal care at Oregon Health & Science University. Male and female Fah−/− mice were of the FahΔexon5 strain on the 129s4 background as described previously (43). Fah−/− mice were maintained with drinking water containing NTBC (Yecuris Corporation) at a concentration of 8 mg/liter (44).

Plasmid vectors and rAAV production

Lentiviral vectors encoding U6-driven shRNAs against Tat, Hgd, and Hpd were produced using the backbone plasmid pLKO.1-hPGK-Puro-CMV-tGFP (Sigma) (table S1). The selected U6-HPD shRNA cassette was subcloned into an AAV vector with 500–base pair (bp) rDNA homology arms (27). From the pLKO.1 template, a 579-bp PCR product was amplified using the following primers: 5′-AGTCGAATTCGCCTATTTCCCATGATTCCTTC-3′ (forward) and 5′-GACTACTTTTCCTCGCCTGTTC-3′ (reverse). The PCR product was cut with Eco RI and ligated to the recipient rDNA AAV vector linearized with Eco RI. The selectable GeneRide vector was constructed by first incorporating the HPD shRNA into an miRNA. An optimized miR-30, “miR-E,” was chosen as a template (45). The HPD-targeted miR-E sequence was TCGACTTCTTAACCCAACAGAAGGCTCGAGAAGGTATATTGCTGTTGACAGTGAGCGCGCCTCAGAATGGTACCTGAAATAGTGAAGCCACAGATGTATTTCAGGTACCATTCTGAGGCATGCCTACTGCCTCGGACTTCAAGGGGCTAGAATTCGA (target and guide sequence are in bold). To incorporate this HPD shRNAmir sequence into the GeneRide vector, pAB269 was cut with PflMI to excise an 810-bp fragment spanning the hF9 transgene and 298 bp of an albumin intron sequence. We generated a synthetic sequence identical to this 810-bp fragment but replaced 157 bp of albumin intron sequence with the HPD shRNAmir. The HPD shRNA was inserted 55 bp downstream of the hF9 stop codon, with the following flanking albumin intron sequence: 5′-AACCGGT-HPD shRNAmir-CCAACTT-3′. This synthetic fragment was cut with PflMI and ligated to the backbone of pAB269, cut with the same enzyme.

Standard transfection and viral isolation protocols were described previously (46). AAV titering was performed by quantitative PCR. Lentiviral vector titers were determined using puromycin resistance in 3T3 cells.

Vector administration, NTBC cycling, and bleeding

Adult Fah−/− mice (4 to 8 weeks of age) were administered rAAV (in 100-μl volume) by intravenous tail vein injection. NTBC treatment was continued for 2 weeks after vector administration. NTBC cycling consisted of 2- to 3-week withdrawal periods interspersed with 4 days of NTBC administration. For neonatal AAV administration, P1 wild-type or Fah−/− neonates were injected with 1 to 2 × 1011 vg per mouse (in 10-μl volume) of rAAV8 or 4 × 103 transducing units of lentivirus by intravenous injection into the superficial temporal vein. NTBC administration was continued for 5 weeks after vector administration, and then cycling was performed as described above. For vectors expressing hF9, blood samples were collected by saphenous vein puncture at the onset of NTBC cycling and every week thereafter.

CEHPOBA administration

Wild-type P1 neonates were injected with 2 × 1011 vg per mouse of rAAV8. At 3 weeks of age, blood samples were collected by saphenous vein puncture and analyzed for hF9 levels. Two cohorts of three mice each with similar baseline hF9 levels were chosen. Beginning at 4 weeks of age, each cohort was given daily injections of saline or CEHPOBA (1 μmol/g) as described previously (23). During CEHPOBA/saline treatment, the mice received drinking water supplemented with l-phenylalanine (20 mg/ml) (Acros) and dextrose (30 mg/ml). After 30 days of CEHPOBA/saline treatment, injections were terminated, and the mice were given regular drinking water for another 30 days then sacrificed for tissue and blood collection. Blood samples were drawn weekly throughout the experiment to monitor hF9 levels.

Liver immunohistochemistry

Immunohistochemical fixation and staining protocols for FAH and H&E were completed as described (47). For analysis of turboGFP (tGFP), methods for FAH staining were used except the primary antibody was rabbit anti-tGFP (Axxora) used at 1:500 overnight at 4°C. Fixation and staining methods for hF9 immunohistochemistry were as described (30).

For serial sections, the primary antibodies used were in order: glutamine synthetase (1:200), Ki67 (1:200), CYP2E1 (1:100; Abcam, ab28146), HNF4α (1:150; Santa Cruz Biotechnology, sc-8987), and hF9 (1:400). Vector ImmPRESS conjugated to a horseradish peroxidase detection kit was used as the secondary antibody. Positive staining was visualized with a 3-amino-9-ethylcarbazole and a 3,3-diaminobenzidine substrate. Microscopy was performed on a DM IL LED microscope (Leica) using Leica LAS Image Analysis Software.

Assessment of F9-containing albumin mRNAs by quantitative PCR

The method was described in detail in our previous GeneRide work (30). Briefly, cDNA produced from reverse transcription with a poly-dT primer served as a template for two different TaqMan quantitative PCR (qPCR) assays. We quantified the abundance of wild-type albumin mRNA by qPCR with primers 5′-CTGACAAGGACACCTGCTTC-3′ and 5′-TGAGTCCTGAGTCTTCATGTCTT-3′, and TaqMan probe 5′-CCACAACCTTCTCAGGCTACCCTGA-3′. We quantified the abundance of fused mRNAs by a TaqMan qPCR with primers 5′-CCAAGGTGTCCAGATACGTG-3′ and 5′-TGAGTCCTGAGTCTTCATGTCTT-3′, and TaqMan probe 5′-CCACAACCTTCTCAGGCTACCCTGA-3′. For noninjected controls, no qPCR signal was detected with primers specific for the fused mRNA. The rate of F9-containing albumin mRNAs was calculated as the ratio between the abundance of template for the wild-type albumin primer pair to the abundance of template for the fusion mRNA–specific primer pair.

PCR to identify selected shRNA

To identify shRNAs that were clonally expanded during NTBC withdrawal, genomic liver DNA from weight stabilized mice was amplified using two primer sets: (i) 5′-TGGAATCACACGACCTGGATGG-3′ (forward) and 5′-CCCGTCCTAAAATGTCCTTCTGC-3′ (reverse); (ii) 5′-ATCGTTTCAGACCCACCTCCCAAC-3′ (forward) and 5′-CCCAAGTCCCGTCCTAAAATGTC-3′ (reverse). PCR products were gel-purified, subcloned using TOPO TA Cloning Kit (Life Technologies), and sequenced using M13 forward and reverse primers.

Integration site analysis after selection in gene-targeted mouse liver DNA

Primers used to amplify the five-prime GeneRide integration site were 5′-CAACGTCATGGGTGTGACTTTTG-3′ (primer 1) and 5′-TACTGCTTCCAGAACTCGGTGG-3′ (primer 2). Sequencing primers were 5′-GGTGGAACAACATTATGAGC-3′ and 5′-AGAGACATAGTGCTGTGTAGGG-3′. Primers used to amplify the three-prime GeneRide integration site were 5′-CACCAACATCTTTCTGAAGTTCGG-3′ (primer 3) and 5′-TTCCACACCTGGCTCGTTATTAG-3′ (primer 4). Sequencing primers were 5′-CTGTCTACACCAGTGAAAATCG-3′ and 5′-TGACCCTTTGACCACGCAAC-3′.


hF9 levels were measured using the Asserachrom IX:Ag ELISA kit (Stago). The manufacturer’s directions were followed with the following modifications. Blood samples (5 μl) were collected from saphenous vein puncture and diluted into 245 μl of R4 buffer from the ELISA kit supplemented with 11 mM sodium citrate as an anticoagulant. After brief centrifugation to pellet blood cells, supernatants were stored at −80°C for up to 3 weeks. Two hundred microliters of each sample was used for the hF9 ELISA. The limit of detection was about 20 ng/ml hF9 in whole blood. Specificity for human F9 was confirmed by diluting purified hF9 in mouse plasma or saline with no significant difference in measured hF9 levels.

Liver function and blood chemistry tests

Mouse serum samples were obtained by submandibular vein puncture. ALT levels were measured using ALT (SGPT) Reagent Kit (Color Endpoint) (GenWay). For mice used in CEHPOBA selection, serum was obtained by terminal cardiac puncture and submitted to IDEXX Laboratories for a comprehensive blood chemistry panel.

Statistical analysis

Statistical analyses were conducted with GraphPad Prism software version 4.0 (GraphPad). Experimental differences were evaluated by Student’s two-tailed t test assuming equal variance. P values <0.05 were considered statistically significant.


Fig. S1. Integration site analysis after selection in gene-targeted mouse liver DNA.

Table S1. shRNA sequences tested in lentiviral vectors.

Data values in tabular format


Acknowledgments: We thank L. Wakefield for assisting with the animal husbandry. Funding: This work was supported by NIH grants DK048252 (M.G.) and HL064274 (M.A.K.). Author contributions: S.N. performed the bulk of the experiments and generated figures for the manuscript; A.B. provided the GeneRide construct and the RNA abundance quantitation; A.H. did the animal husbandry and performed the F9 assays; A.M. performed the immunohistochemistry; M.F. analyzed the immunohistochemistry; M.A.K. designed experiments and provided vector constructs; and M.G. oversaw the experiments and wrote the manuscript. Competing interests: M.G. is a shareholder and consultant for Yecuris Corporation, the company that has licensed the Fah−/− mouse technology from Oregon Health & Science University. M.A.K. and A.B. are shareholders for LogicBio Therapeutics that has licensed the GeneRide technology from Stanford University. Data and materials availability: All data and materials are available from the corresponding author upon request. Plasmids containing viral vectors and Fah−/− mutant mice require the completion of a standard academic material transfer agreement.

Stay Connected to Science Translational Medicine

Navigate This Article