Gastroenterology, Hepatology and Endocrinology, Hannover Medical School, Germany
Twincore, Centre for Experimental and Clinical Infection Research, Hannover, Germany
Hannover Medical School, Hannover, Germany, or Ute Modlich, Twincore, Centre for Experimental and Clinical Infection Research, Feodor-Lynen-Str. 7-9, Department of Gastroenterology, Hepatology and Endocrinology, Feodor-Lynen-Str. 7, Hannover, Lower Saxony, Germany 30625===
Potential conflict of interest: Nothing to report.
Supported by grants of the Deutsche Forschungsgemeinschaft (SFB 738 and the Excellence Cluster REBIRTH).
Lentiviral (LV) vectors are promising tools for long-term genetic correction of hereditary diseases. In hematopoietic stem cell gene therapies adverse events in patients due to vector integration-associated genotoxicity have been observed. Only a few studies have explored the potential risks of LV gene therapy targeting the liver. To analyze hepatic genotoxicity in vivo, we transferred the fumarylacetoacetate hydrolase (FAH) gene by LV vectors into FAH(-/-) mice (n = 97) and performed serial hepatocyte transplantations (four generations). The integration profile (4,349 mapped insertions) of the LV vectors was assessed by ligation-mediated polymerase chain reaction and deep sequencing. We tested whether the polyclonality of vector insertions was maintained in serially transplanted mice, linked the integration sites to global hepatocyte gene expression, and investigated the effects of LV liver gene therapy on the survival of the animals. The lifespan of in vivo gene-corrected mice was increased compared to 2-(2-nitro-4-trifluoromethylbenzoyl)-1,3-cyclohexanedione (NTBC) control animals and unchanged in serially transplanted animals. The integration profile (4,349 mapped insertions) remained polyclonal through all mouse generations with only mild clonal expansion. Genes close to the integration sites of expanding clones may be associated with enhanced hepatocyte proliferation capacity. Conclusion: We did not find evidence for vector-induced tumors. LV hepatic gene therapy showed a favorable risk profile for stable and long-term therapeutic gene expression. Polyclonality of hepatocyte regeneration was maintained even in an environment of enforced proliferation. (HEPATOLOGY 2013)
Stable gene transfer into hepatocytes with viral vectors offers a cure for many hereditary liver diseases. Clinical examples include hemophilia, lysosomal storage disorders, urea cycle defects, and α1-antitrypsin deficiency.1 The first ex vivo retroviral gene transfer into hepatocytes was attempted in patients with homozygous familial hypercholesterolemia, but low transduction rates have limited the therapeutic benefit in these patients.2, 3 Lentiviral (LV) vectors, however, are able to transduce noncycling cells, opening new opportunities for hepatic gene therapy.4, 5 Efficiency of LV vector gene transfer has been demonstrated in several murine models of hereditary metabolic liver diseases.6-8
The transforming potential of retroviral vectors due to insertional mutagenesis is a major concern in hematopoietic gene therapy.9 Transduction of hematopoietic stem cells with gamma retroviral vectors (GV) led to leukemias in animal models10-12 and in clinical trials.13, 14 The oncogenic potential of retroviral vectors originates from a combination of their preference to integrate into promoter regions and CpG islands15 and the capacity of the viral enhancer/promoter sequences to activate cellular genes close to the integration site. LV and GV vectors in a self-inactivating architecture improved their safety profile.16, 17 The characteristic insertional pattern of LV vectors in gene coding regions rather than promoters18-20 also reduces the risk of insertional up-regulation of proto-oncogenes; however, interference with natural splicing of the host messenger RNAs (mRNAs) cannot be excluded.21, 22
High proliferative capacity, such as in hematopoiesis, is also believed to foster transformation compared to postmitotic tissues. Delivery of nonprimate LV vectors into the fetal liver, which is characterized by massive hepatoblast proliferation, induced liver tumors in offspring mice.23 In contrast, tumor induction by LV gene transfer to adult mouse or rat livers has not been reported, most likely due to the small cell turnover of parenchymal cells in postnatal livers of healthy mammals.24 Analysis of the tumorigenic potential of postnatal LV hepatic gene transfer, however, needs to consider the extensive proliferation capacity of parenchymal liver cells in response to acute or chronic injuries as an independent risk factor for liver tumor development.
In our present study we performed LV gene transfer in the fumarylacetoacetate hydrolase (Fah)(-/-) mouse model, which resembles human hereditary liver disease tyrosinemia type I.25 In both patients and mice lacking Fah protein expression, the drug 2-(2-nitro-4-trifluoromethylbenzoyl)-1,3-cyclohexanedione (NTBC) can prevent liver failure.26, 27 Despite partial pharmacological protection late-onset tumors of the liver still occur in a considerable number of mice.28 Confounding lentiviral genotoxicity could thus result in earlier onset or increased numbers of liver tumors and increased mortality. Fah gene transfer provides a selective advantage and favors the expansion of gene-corrected hepatocytes, which could trigger LV-associated tumor formation due to insertional mutagenesis. To maximize proliferative stress, we serially transplanted the cells into three subsequent recipient adult mouse generations.
We produced VSVg pseudotyped lentiviral vectors expressing the murine Fah complementary DNA (cDNA) from an internal SFFV promoter. The enhanced green fluorescence protein (eGFP) was coexpressed either by an internal ribosomal entry site (IRES) or by a 2A proteolytic cleveage site (RRL.PPT.SFFV.eGFP.pre*, RRL.PPT.SFFV.Fah.ires.eGFP.pre*, RRL.PPT.SFFV.Fah.2a.eGFP.pre*).29 The vector supernatants were produced by 293T cells, concentrated with Centricon Plus-70 filter units (Millipore, Schwalbach, Germany) and titrated on HepA1.6 cells.30 Viral titers were in the range of 107 to 108 IU/mL.
Hepatocyte Isolation and Transduction.
Hepatocytes were isolated from 3 to 6-month-old mice as described.6 For further enrichment of hepatocytes we used discontinuous Percoll (GE Healthcare) gradients. A 25% phase prevented dead cells and debris from entering the gradient. A lower 50% phase facilitated the enrichment of small, highly viable, and robust hepatocytes. For depletion of nonparenchymal liver cells, we used PE (Phycoerythrin)-labeled anti-CD45 and anti-CD31 antibodies (BD Pharmingen), anti-PE MicroBeads on an automated MACS Separator (Miltenyi Biotech, Bergisch Gladbach). Analysis of eGFP expression was performed on day 5 after in vitro transduction of primary murine hepatocytes or directly after hepatocyte isolation during serial transplantations by FACS. For in vitro experiments, hepatocytes were cultured on Primaria dishes (1.1 × 106 cells/35 cm2) in HCM medium (Lonza). Fah(-/-) hepatocytes isolated from C57Bl/6-Fahtm1Mgo were cultured in the presence of NTBC (9.6 ng/mL) (Swedish Orphan). Transduction (multiplicity of infection [MOI] 10) was performed as previously described by our group.6
Ligation-Mediated Polymerase Chain Reaction (LM-PCR), 454-Pyrosequencing, and Insertion Site Analysis.
LM-PCR was performed and analyzed as described.31, 32 (See also Supporting Material and Methods.)
C57Bl/6-Fahtm1Mgo (Fah(-/-)) mice are defective for the Fah gene, which encodes the enzyme fumaryacetoacetate hydrolase (Fah).25 Mice were maintained by supplementation of drinking water with NTBC (4 mg mL-1, Swedish Orphan International).26, 27 To allow engraftment of hepatocytes the medication was discontinued
LV Gene Transfer and (Serial) Hepatocyte Transplantations.
For in vivo gene transfer, lentiviral particles were injected in 250 μL phosphate-buffered saline (PBS) intrasplenically (∼1 infectious particle / parenchymal liver cell). For ex vivo gene therapy 1 × 106 cultured Fah(-/-) hepatocytes were transduced overnight (MOI 10). For serial hepatocyte transplantation the livers from gene-corrected mice were perfused with collagenase. Aliquots of 0.5 × 106 liver cells were injected in 250 μL PBS into the next generation of Fah(-/-) mice. NTBC drug was withdrawn 2 days after surgery and 50% of the vital NTBC dose was supplied, if body weight decreased to ∼75% during progression of liver failure. Animals fully recovered within 6-8 weeks and gene-corrected hepatocytes were reisolated after 100 days. Successful repopulation of recipient livers was documented by flow cytometry (eGFP) and by Fah-immunohistochemistry. Subcohorts from serially transplanted mice independent from NTBC treatment were observed for their full life span and sacrificed close to the timepoint of death. Mice that died from insufficient repopulation within the first 50 days after cell transplantation were excluded from survival analysis.
Pathology, Histology, and Immunohistology.
Liver, spleen, lungs, heart, kidneys, pancreas, brain, and intestine from the observation cohorts of mice were analyzed macroscopically for the presence of abnormalities. Normal liver and tumor-like structures were separated using a scalpel. An aliquot of each liver tissue sample was immediately frozen in liquid nitrogen for the extraction of DNA. The rest of the organ was fixed with 4% formalin, embedded in paraffin, and cut into 5-μm thick slices for histological and immunohistochemical analysis.
LV Copy Number Analysis.
Vector copy numbers (VCN) were determined as described.6 A primer/probe combination specific for the wPRE element of the vector was measured and normalized to an intronic, genomic sequence of the Ptbp2 gene. Due to the different ploidies of hepatocytes, VCN is given as copies per haploid genome. All samples were analyzed in triplicate on a Roche Light Cycler 480 (LC480) system.
Locus-Specific Quantitative PCR (qPCR).
For all samples analyzed by locus-specific qPCR we used a common forward primer (lv-LTRIII: 5′-AGTAGTGTGTGCCCGTCTGT-3′) and probe (Q-probe: 5′-FAM-TCCCTCAGACCCTTTTAGTCA-TAMRA-3′) specific for the residual part of the self-inactivating (SIN) - long terminal repeat (LTR) region of the vector. The reverse primers were designed according to output of the 454-sequencing run, so that the amplicon size was between 100-160 bp. (See primer information in the Supporting Material and Methods.)
Statistical Analysis and Data Mining.
The survival analysis was performed using Kaplan-Meyer curves and a Mantel-Cox test to calculate P values. The capture-recapture analysis used the Lincoln-Peterson estimation. Statistical significance was assumed for P < 0.05.
Lentiviral Vectors Preferentially Integrate in or Near Actively Transcribed Genes of Primary Hepatocytes With Local Hotspots.
First, we analyzed lentiviral integration patterns in cultured murine hepatocytes. We depleted collagenase digested liver cells (n = 3) from CD45+ hematopoietic and CD31+ endothelial cells (Fig. 1A) and transduced the remaining cells (>98% hepatocytes) with the LV RRL.PPT.SFFV.eGFP.pre* vector (Fig. 1B-D) at an MOI of 10. After 6 days genomic DNA was isolated. Sequences flanking the lentiviral insertion sites were amplified by LM-PCR for further analysis by 454 high-throughput sequencing. The median distance of lentiviral insertions (2,775) in hepatocytes was 6.4 (± .4) kb downstream to the next transcription start site (TSS) (Fig. 1E) and thus similar to previously analyzed hematopoietic cells (8.0 (±3.0 kb)33 (Fig. 1E). This result indicates the general preference of LV vectors to insert downstream of the TSS into genes. In both cell types we observed a preference of LV to integrate inside or near genes, which were transcribed at the time of transduction (Fig. 1F). Interestingly, we found overlaps between common insertion sites in hepatocytes and lineage negative BM cells32 (Supporting Table 1) by kernel density estimations33 (Table 1). The common insertion site within and around the gene Sfi1 was detected with one of the highest densities in both datasets (Supporting Fig. 2).
Table 1. Lentiviral Common Insertion Sites in Transduced Hepatocytes
Number of Genes in 50 kb Distance
Top10 Read Counts in 454
Animals With Insertion
Kernel density estimate analysis (46) was performed using integrome data of lentivirally transduced hepatocytes. Their genomic location is given according to the mouse genome reference assembly MGSCv37.
Serial Transplantation of Gene-Corrected Hepatocytes After Hepatic Gene Therapy.
To assess potential genotoxicity in vivo, we used a self-inactivating, VSV-G pseudotyped LV expressing Fah from the spleen focus forming virus (SFFV) promoter (RRL.PPT.SFFV.Fah.ires.eGFP.pre*, Fig. 2A). This promoter showed transcriptional activity similar to the liver-specific transthyretin promoter (TTR) in hepatocytes (Supporting Fig. 1), but was active and potentially genotoxic in all liver cell types. We injected the vector at a dose of approximately one infectious particle per parenchymal liver cell by way of the spleen into Fah-deficient C57BL/6-Fahtm1Mgo mice (in vivo series). To account for differences in integration patterns of in vivo and ex vivo transduced hepatocytes, we added a second series of Fah-deficient mice that were transplanted with in vitro transduced hepatocytes (ex vivo series). The ex vivo applied vector (Fig. 2B) used a P2A protease cleavage site for brighter eGFP fluorescence compared to the IRES sequence.34 A total of 21 mice were treated by Fah gene transfer (Table 2).
Table 2. Mice Included in the Study
n.d.: not determined. Number of Fah-deficient mice successfully repopulated after in vivo or ex vivo gene transfer. Mice listed under short term were used as cell donors for integrome analysis and serial transplantation and were sacrificed after 100 days of repopulation. Others were used for long-term survival analysis.
Transgene expression corrected the metabolic Fah deficiency within 100 days as documented by the survival of mice without NTBC treatment and increased body weights (Supporting Fig. 3). The Fah protein expression was confirmed by immunohistochemistry (Fig. 2C).
In addition to the long-term observation cohorts (n = 59 mice, Table 2) we induced extensive proliferation of in vivo (Fig. 2D) or ex vivo (Fig. 2E) gene-corrected hepatocytes by serial transplantations. After 100 days we isolated gene-corrected hepatocytes from first-generation founder mice (5 in vivo, 3 ex vivo) and transplanted them into secondary recipients. The transplantation procedure was repeated to generate third- and fourth-generation cohorts. Repopulation rates ranged from ∼25% (in vivo) to up to ∼73% (ex vivo) (Fig. 2F,G). We estimated the primary hepatocytes to have undergone more than 65 cell doublings (Supporting Table 3, Supporting Fig. 4) in latest-generation mice.
Late-Onset Tumors Originate From Endogenous Tissue.
Survival of the first generation in vivo long-term observation cohort (n = 12) was increased after systemic vector injection (623 days) compared to NTBC-treated controls (396 days) indicating a stable therapeutic effect (Fig. 3A). The life spans of the second (n = 19), third (n = 11), and fourth (n = 17) generations of serially transplanted mice (≥ 357 days) were similar to the NTBC treated control cohort (P ≥ 0.41) (Supporting Table 2).
At the time of necropsy 44.4%, 69.2%, 55.6%, and 36.4% of all long-term observed animals in first, second, third, and fourth generation cohorts showed nodular liver tissue parts consistent with tumor development (Fig. 3B). Normal tissue stained uniformly positive for the Fah protein (Fig. 3B, I), whereas it was undetectable in nodular areas (Fig. 3B II-V) except for some displaced tissue surrounding the tumor-like structures (Fig. 3B III). The hypothesis that LV-mediated insertional mutagenesis was not associated with tumor formation was supported by the fact that tumorous tissue had low copy numbers (0.01 ± 0.02) compared to histologically normal areas expressing the Fah protein (0.40 ± 0.04; P < 0.05) (Fig. 3C; Supporting Fig. 5).
Liver Repopulation Remains Polyclonal in Gene-Corrected Mice.
Even in the absence of hepatic tumors, LV integration could initiate clonal imbalance by activating growth promoting genes as it was demonstrated in gene therapy of the hematopoietic system.35, 36 To test for this, LV integration sites from serially transplanted hepatocytes of the in vivo (n = 25) and the ex vivo group (n = 13) were amplified by LM-PCR and 454 pyrosequencing. In a total of 296,036 sequences we identified 4,349 independent insertion sites from 38 repopulated animals, which located to 2,483 unique gene IDs (GID). Numerous insertion sites were found in all generations of serially transplanted mice with no dominant bands in agarose gel indicating a polyclonal regeneration of the recipient livers (Fig. 4A).
All vector-genome junctions located closer than 500 kb to the TSS of annotated genes were included for analysis of clonality. Using LM-PCR we aimed to identify expanded cells and clonal imbalance rather than the full repertoire of insertions. The limited input of DNA for LM-PCR (0.5%-1% of total liver cells and 10% of initial DNA for nested PCRs) and the coverage using three enzymes for genome fragmentation (76.5% as determined by capture-recapture analysis (Supporting Fig. 7) reduced the overall number of detectable insertions. Based on our calculations (Supporting Fig. 6) we expected to recover around 150-200 insertion sites per repopulated mouse liver. Averages of 109 ± 25 and 142 ± 84 unique insertion sites per liver were allocated in the in vivo and ex vivo groups, respectively (Supporting Table 4). The mean vector copy numbers of 1.70 ± 0.24 per liver were similar in all generations of serially transplanted mice (Fig. 4B).
To determine potential clonal selection after serial transplantations, we calculated the number of clones contributing to 50% of all 454-reads per liver. The contribution of top 50% (T50) clones in latest-generation livers was not significantly reduced when compared to first-generation livers, either in the in vivo group (10.6 ± 1.2% and 11.6 ± 2.2%, P = 0.686) or in the ex vivo group (13.9 ± 1.4% and 12.9 ± 2.7%, P = 0.806), respectively (Fig. 4C).
A relatively high proportion of clones was detected in only one mouse due to incomplete coverage (36% of insertions <5 reads), although theoretically present in all earlier generation mice (cell donors) of the same series. The relative contribution of animal specific new insertions compared to all insertions sites revealed a slight, but not significantly reduced polyclonality in the fourth generation (61.5% ± 9.4%) compared to first-generation livers (76.03% ± 10.87%) of in vivo gene-corrected hepatocytes (P = 0.350, Fig. 4C). In the ex vivo group the percentage of new unique insertions in third-generation livers (61.0 ± 2.4%) was similar compared to first-generation livers (65.6 ± 5.9%) (P = 0.600).
Insertion Frequencies Indicate Mild Clonal Selection in Late-Generation Mice.
A mild reduction in clonality after serial transplantation became obvious by resampling the same specimen with the restriction enzyme (Tsp509I). The coverage of clones in the first generation increased from 11% to 39% in the last generation. Of the high read insertion sites, 24% were found in more than one animal. The 10 most often detected insertions based on reads (top 10 clones) of fourth-generation in vivo and third-generation ex vivo animals were analyzed for their abundance in earlier generations (Fig. 5A-G). The number of reads was considered a measure of the abundance of specific clones (for detailed information see Supporting Table 5). The qPCR analysis of selected clones confirmed the presence of expanded clones but also indicated overestimation of the abundances of such clones by the sequence read method in most cases (Brugman et al.37).
Several hepatocytes with specific insertions such as Alcam, Pms2, Factor 11, Dnase 1l3, or Adcy9 (Fig. 5H-L; Supporting Fig. 9) expanded towards the last-generation mice. Several clones listed in Supporting Table 5 were present in the oncogenomic database of hepatocellular carcinoma (OncoDB.HCC). Intriguingly, seven genes closest to the identified common insertion sites (Table 1) were also Top 10 read clones in the 454 analysis (Supporting Fig. 10). This may indicate that insertions at specific locations can become selected under proliferative stress.
Unlike several other solid organs the liver can respond to acute and chronic injuries by the proliferation of hepatocytes. For risk assessment of hepatic lentiviral gene therapy we considered the extensive regenerative capacity of the liver as a confounding factor for LV-associated tumor formation. The Fah(-/-) mouse model is ideally suited to study LV-mediated genotoxicity in hepatocyte proliferative states, since gene-corrected hepatocytes selectively repopulate the host liver. Due to limitations of the model the effect of proliferative stress could not be studied in nonparenchymal liver cells and cells of other organs.
Leukemias in mice after retroviral gene transfer into hematopoietic stem cells were mostly observed after secondary transplantation.10, 38 To mimic this experimental condition, we performed serial transplantations and analyzed four (in vivo) and three (ex vivo) subsequent generations of serially transplanted mouse cohorts. We calculated 65 hepatocyte doublings, a number, which by far exceeds the normal turnover of hepatocytes in a lifetime.
In vivo LV injection rescued the disease phenotype in immunocompetent first-generation mice. Long-term transgene expression in the liver by retroviral vectors can be problematic due to immune recognition of modified cells. This can be overcome by the use of liver-specific promoters or microRNA (miRNA) target sequences.39 Similar to results of others,40 however, we did not observe loss of modified cells, although the transgene was constitutively expressed. This is probably due to a strong selective advantage of gene-corrected cells in our model. Although we induced excessive proliferative stress to hepatocytes in targeted livers, the in vivo LV-treated mice did not show reduced long-term survival compared to NTBC-treated controls. Additionally, the number of mice with potential tumor nodules was similar in all mouse cohorts and metastatic tumor tissues were absent (n = 49). The Fah(-/-) mouse model itself is prone to spontaneous tumor development of endogenous hepatocytes. The lack of transgene expression and very low viral copy numbers excludes insertional mutagenesis as the cause of tumor formation. Lentiviral genotoxicity in the adult liver appeared to be surprisingly low.
In contrast to our study, late-onset hepatocellular carcinomas (HCCs) have been reported by others in animals that were intrafetally or neonatally transduced with nonprimate and HIV-derived LV vectors.41, 42 The extensive proliferative state of nonadult hepatocytes had been proposed as a risk factor for tumor formation. In view of our data, other parameters such as the different gene expression state of fetal and neonatal versus adult hepatocytes, differences in the vector design, or in the regulation of DNA repair and apoptosis may have played a role.
Insertional mutagenesis was also observed after neonatal adeno-associated virus (AAV) gene therapy in mice due to integrations in the miR341 locus on chromosome 12.43, 44 Expression of miRs in the syntenic regions on human chromosome 14 has been linked to human cancer. Hence, integrations were likely to be causative for HCC induction in this study. However, other preclinical studies, including a comprehensive analysis in 80 mice and a follow-up of 18 months gave no evidence of AAV vector integration-associated transformation in the liver.45, 46
Our report provides the first systematic analysis of clonality in a liver repopulation model using lentiviral insertion sites. In a recent study insertion sites were mapped but clonality in the liver was not investigated.47 In our in vitro LV integrome analysis we mapped more than 2,000 individual insertion sites and compared the integration patterns with those of hematopoietic stem cells. We detected a partial overlap of common insertion sites in these cell types, indicating the presence of cell type-independent “hot spots” for lentiviral integration, which need to be distinguished from selection events.48
Clonal dominance in the hematopoietic system is typically accompanied by vector insertions in proto-oncogenes or genes controlling cell proliferation, such as EVI1, PRDM16, or HMGA2.21, 35, 36 To address whether a selection for dominant clones also occurs in the liver, we analyzed the integration sites in 38 mice comprising four generations of serially transplanted mice. The number of reads obtained from a specific sequence in 454 pyrosequencing provides a semiquantitative measure for the abundance of individual hepatic clones. We found that multiple insertion sites were maintained in all mice and all generations, indicating polyclonal liver repopulation (Supporting Figs. 10, 11). Polyclonal liver repopulation was also proposed in a recent study using fluorescently labeled vectors.49 One-third of all insertions were sequenced with low read counts, indicating low abundance of hepatocytes with these specific insertions. The number of insertion sites, which accounted for 50% of all reads in one liver, did not change significantly from the first to latest generation. However, insertion sites from repopulated livers showed higher read counts compared to in vitro transduced hepatocytes that did not undergo proliferation. Hence, we assessed the level of clonal selection in vivo by monitoring the presence of hepatic clones with high read counts in the last generation, which were also present in earlier generations. Locus-specific qPCR for some of the most prevalent insertions in the fourth generation confirmed the presence of clones, which showed an increase in population size through the series of serial transplantations. Some of the Top10 integrations (4.1%) were located close to genes with a potential function in HCC as the genes are listed in the OncoDB/HCC database. The total amount of genes listed in this database was, however, very similar between the preinfused in vitro sample (2.6%) and the data from the repopulated mice (2.9%). Apart from liver-specific metabolic functions, no common pathway could be associated to the Top10 integration sites. The contribution of differentially expressed genes to hepatocyte proliferation cannot be irrevocably determined in our study. A potentially deregulated gene in a respective clone is measured against a huge background of all other clones in the liver as well as the host hepatocytes. This makes the detection of altered gene expression due to vector integration virtually impossible in our model (data not shown). Further studies are currently being performed to analyze the contribution of the candidate genes to hepatocyte proliferation and HCC formation.
Taken together, our study clearly shows that despite the occurrence of mild clonal selection, the risk of liver tumor development due to insertional mutagenesis after hepatic lentiviral gene transfer is low, even under conditions of extensive proliferative stress of LV-transduced hepatocytes.
The authors thank Sabine Knoess and Johanna Krause for excellent technical assistance, Stefan Bartels for help with bioinformatic questions, and Tobias Mätzig for discussing data (all Institute of Experimental Hematology, Hannover Medical School, Hannover, Germany). We also thank Qinggong Yuan, Gastroenterology, Hepatology and Endocrinology, Hannover Medical School, Germany, for help with hepatocyte transplantations.