Retroviral vector integration occurs in preferred genomic targets of human bone marrow–repopulating cells

Stem cells, the natural units of tissue regeneration, hold big promise as a treatment for a multitude of diseases.1-4 Their extensive ability to proliferate and self-renew make them an attractive target for gene therapy.5-7 Peripheral blood progenitor cells (PBPCs) are a clinically relevant stem cell source and can be obtained from patients in ample quantity.8,9 Their transduction with retroviral vectors has been optimized so that high levels of gene transfer can be achieved.10,11 Recently, insertional mutagenesis following retrovirus-mediated gene transfer to mouse and human hematopoietic cells has been reported.12 13 

To assess the mutagenic risk of retroviral gene therapy, it is important to characterize vector integration sites in individual stem cells and their progeny. The experimental analysis of proviral integration sites in human hematopoietic stem cells is challenging. Stem cell tracking techniques based on detecting the genomic DNA flanking the provirus and using this as a unique tag were traditionally used in isologous mouse studies.14,15 However, these techniques are not easily applied to xenogeneic transplantation models because of lower contents of engrafted cells carrying the proviral tag in this latter setting. Retroviral integration patterns in transduced human cord blood cells transplanted into immune-deficient mice were detected by Southern blotting if engraftment and transduction efficiency were high.16,17 Inverse polymerase chain reaction (PCR), an alternative technique used by Nolta et al,18 is more sensitive but requires clonal preparations as starting material. This approach cannot reliably detect multiple integration sites in one reaction19 nor can PCR with arbitrary primers.20 An oligo-cassette-mediated polymerase chain reaction technique described by Rosenthal and Jones21 and modified by Schmidt et al22 23 is a promising approach to detect different integration sites simultaneously. We have optimized such a ligation-mediated PCR (LM-PCR), validated the results by fluorescence in situ hybridization (FISH) for retroviral integrants, and can now demonstrate that multiple transduced human PBPC clones mediated engraftment in the bone marrow (BM) of immune-deficient mice. DNA analysis also allowed a novel glimpse on preferred retroviral vector integration sites—among them coding sequences—in the genome of human marrow-repopulating cells.

Individual SF1m-transduced HT1080 cells were selected by single-cell deposition on a Becton Dickinson FACS-Vantage cell sorter (Becton Dickinson, Heidelberg, Germany) after identification of transgene expressing cells by Rh-123 dye exclusion. Three cell line clones (N2, N3, N4) were used here. DNA was prepared by using the QiaAmp protocol (Qiagen, Hilden, Germany).

In this study material from 3 healthy donors was used. They have given informed consent before CD34⁺ cell collection. CD34⁺ cells were prepared as described.24Briefly, CD34⁺ cells were isolated from frozen PBPC samples by magnetic microbead selection using the CliniMACS system (Miltenyi Biotech, Bergisch Gladbach, Germany) according to the manufacturer's description.

Retroviral vector stocks were produced and stored as described.25 Retroviral transduction was performed as described.24 In brief, CD34⁺ cells were prestimulated for 16 to 20 hours at a density of 1 × 10⁶cells/mL X-VIVO-10 medium, supplemented with interleukin 3 (IL-3; 20 ng/mL), IL-6 (10 ng/mL), stem cell factor (SCF; 50 ng/mL), Flt-3 ligand (FL; 100 ng/mL), (CellSystems, St Katharinen, Germany), and thrombopoietin (TPO; 20 ng/mL) (R & D Systems, Wiesbaden, Germany). Following prestimulation, cells were exposed to retroviral supernatant, containing the hybrid vector SF1m, which is based on the Friend mink cell focus-forming/murine embryonic stem cell virus and carries the human multidrug resistance 1 (MDR1) gene,26over 3 consecutive days. Twenty-four hours after the last infection period, cells were harvested.

Female nonobese diabetic (NOD)/LtSz severe combined immunodeficient (scid)/scid (NOD/SCID) mice were conditioned by sublethal irradiation (3 Gy) and received transplants of 4 × 10⁶ or 1 × 10⁷ SF1m or mock-transduced human CD34⁺ cells per mouse as described.24,27 Mice (n = 13) were killed by cervical dislocation 8 weeks after transplantation. Engraftment of human cells isolated from mouse BM, as well as presence and expression of the MDR1 transgene were evaluated as described.24 27 Engraftment was measured by quantifying human CD45 antigen-expressing cells.

The presence of MDR1 proviral sequences was quantified by real-time PCR (data not shown). The difference of the threshold cycles between the proviral MDR1 gene and the human erythropoietin receptor (EPO-R) gene was used to quantify the percentage ofMDR1-transduced human cells with reference to a standard curve over a 5-log range.24 Rhodamine-123 efflux analysis served to detect expression of the MDR1transgene28 in engrafting human CD45⁺ cells.

For detection of retroviral integration sites, DNA was extracted from NOD/SCID mouse chimeric BM preparations (QiaAmp Blood Kit; Qiagen). For the validation experiments, DNA from retrovirally transduced HT1080 cell line clones was isolated and mixed with up to 2.5 μg genomic mouse DNA (Promega, Mannheim, Germany).

First, DNA was digested with the restriction enzymes BsmAI (New England Biolabs, Frankfurt, Germany), PvuII, orEcoRV (both Roche Diagnostics, Mannheim, Germany) to create integration site–specific restriction fragment-length polymorphisms (RFLPs). Reactions were purified by using the PCR Purification Kit (Qiagen). Long-terminal repeat (LTR)-genomic flanking DNA junctions were marked and enriched. A biotinylated primer (GSP1-bio, 5′biotin-TGGCCCAACGTTAGCTATTTTCATGTA-3′) was annealed to DNA fragments containing vector LTRs and extended by using 2.5 U Pwopolymerase (Roche Diagnostics) in a one-step amplification reaction (94°C, 2 minutes; 65°C, 2 minutes; 72°C, 10 minutes; 2 cycles) to yield biotin-marked DNA fragments. Two types of marked fragments are obtained in that way, one of constant length resulting from binding of the specific primer to the 3′ LTR and extension to the next restriction site in the vector, the other with variable length depending on the next restriction site in the host genome adjacent to the 5′ LTR (integration site specific).

DNA was purified and enriched for biotin-marked fragments with streptavidin-coated paramagnetic beads (Kilobase binder kit; Dynal, Oslo, Norway). Next, an adapter oligo-cassette (Universal Genome Walker Kit; Clontech, Palo Alto, CA) was ligated blunt-end to the LTR-distant portion of enriched fragments while attached to the beads to create binding sites for forward primers (LM-PCR). Beads were resuspended with ligation mix (0.5 U T4-Ligase [Life Technologies, Karlsruhe, Germany], 5 × Ligation Buffer [Life Technologies], 5 pmol Genome Walker Adapter) and incubated at 16°C for 16 hours under moderate agitation. Then the ligase was inactivated for 5 minutes at 70°C. PCR amplification of the purified fragments was performed by using the Expand High Fidelity PCR System (Roche Diagnostics) with standard Mg²⁺ and deoxy-nucleotide triphosphate (dNTP) concentrations together with 15 pmol LTR-specific primer (GSP1, 5′-TGGCCCAACGTTAGCTATTTTCATGTA-3′) and 15 pmol adapter-specific primer (AP1, 5′-GTAATACGACTCACTATAGGGC-3′) by using the following PCR-program: 94°C, 15 seconds; 65°C, 30 seconds; 68°C, 3 minutes; 31 cycles. Nested PCR was done with internal primers (GSP5, 5′-CCTTGATCTGAACTTCTCTATTCTTGGTTTG-3′; AP2, 5′-ACTATAGGGCACGCGTGGT-3′) by using the same PCR conditions as before with reduced cycle number (n = 29) and reduced extension temperature (60°C) (Figure 1). All PCR reactions were performed in a PTC-200 thermal cycler (MJ Research, Watertown, MA). PCR products were analyzed on agarose gels, and individual bands were excised and purified by the Gel Extraction Kit (Qiagen).

DNA of each excised band was cloned into the pCR4 plasmid vector (TOPO TA Cloning Kit; Invitrogen, Groningen, The Netherlands) according to the manufacturer's instructions. Two to 10 colonies of each cloning reaction were screened for insert length and orientation by PCR. Screening was done with miniprep plasmid DNA (Qiagen Plasmid Miniprep Kit) or by direct screening of bacterial colonies. The insert-specific primer AP2 or GSP5 was used in combination with T3 and T7 standard primers, respectively, to check insert orientation.

Cycle sequencing of the plasmids containing LM-PCR amplicon inserts was performed by using an ABI Prism Genetic Analyzer 310 (PE Applied Biosystems, Weiterstadt, Germany) according to the manufacturer's instructions. Alignment programs using the Smith-Waterman algorithm version 3.3t01 (http://ebv.oncology.wisc.edu/molbio/align.html) were applied. Standard National Center for Biotechnology Information (NCBI) blast searches (http://www.ncbi.nlm.nih.gov/BLAST/) were done to identify cloned human DNA homologues to flanking sequences. Chromosomal localizations of mapped integration patterns were determined by using the^projectEnsembl ContigViewer (http://www.ensembl.org/).

Metaphase spreads from SF1m-transduced HT1080 cell line clones were hybridized by using the retroviral vector probe SF1m-MDR29 followed by whole chromosome painting (WCP) probes for 24-color FISH.30 For each cell line 10 to 15 metaphase spreads were acquired by using a Leica DM RXA RF8 epifluorescence microscope (Leica Microsysteme, Bensheim, Germany) equipped with a Sensys CCD camera (Photometrics, Tucson, AZ) and controlled by Leica Q-FISH software (Leica Microsystems Imaging Solutions, Cambridge, United Kingdom). Subsequently, the ReFISH protocol31 was performed, and metaphase chromosomes were hybridized by using the multicolor FISH (M-FISH) protocol30 with minor modifications. Briefly, 5 pools of WCP probes (kindly provided by Ferguson-Smith, Cambridge, United Kingdom) were amplified and labeled by degenerate oligonucleotide primed-PCR (DOP-PCR)32 with the use of 5 spectrally distinguishable fluorochromes (fluorescein isothiocyanate [FITC], Cy3, Cy3.5, Cy5, Cy5.5). Each probe (100 ng) was hybridized in the presence of Cot-1 DNA for 48 hours. For evaluation, metaphase spreads were acquired by using highly specific filter sets (Chroma Technology, Brattleboro, VT), and images were processed using the Leica Multicolor Karyotyping (MCK) software.

To test whether the number of integrations is equally distributed along the chromosomes, we used a chi-square goodness-of-fit test to analyze whether the observed number of integrations (oi) arose from a multinomial distribution with specified expected integrations (ei) for the 24 chromosomes (22 autosomes as well as the sex chromosomes X and Y). Expected integration counts were computed assuming a discrete uniform distribution but also correcting for the chromosome size distribution and the percentage of finished human genome sequences entered into databases at the time of preparation of this manuscript (European Molecular Biology Laboratory [EMBL] genome monitoring table, http://www.ebi.ac.uk/genomes/mot/). We used α = 0.05 as significance level of the test. For the detection of preferred genomic integration sites, a cutoff of (oi − ei)²/ei ≥ 3 was set. For descriptive purposes mean values and standard deviations (SDs) are given unless otherwise stated.

Our aim was to analyze the genomic integration sites of retroviral vectors within the DNA of human hematopoietic cells repopulating the bone marrow of NOD/SCID mice (SCID-repopulating cells, SRCs). To this end we used a combined LM-PCR-cloning method (Figure 1) that allowed us to simultaneously amplify multiple retroviral integration sites from human cell samples containing significant amounts of nontransduced human or mouse DNA background. The specificity of the LM-PCR results was confirmed by FISH for proviral inserts and subsequent 24-color FISH karyotyping method.

At first the LM-PCR method was established on transduced cell line clones obtained by single cell sorting. Therefore, 2 cell line clones with one integration site each and one cell line clone with 2 integration sites were analyzed. These cell line clones were mixed together to test for the ability to detect multiple clones in one reaction. Four different integration sites and one internal band were readily coamplified (Figure 2A). The addition of 1 to 2.5 μg genomic mouse DNA (as background DNA) to the sample did not change the results, and negative controls with mock-mouse DNA produced no bands on agarose gel (data not shown).

After optimization of the LM-PCR method on transduced cell line clones, we could reliably amplify the integration site, starting with DNA from 1000 transduced cells in a background of more than 200 000 nontransduced cells.

The proviral inserts were directly visualized by FISH with the use of vector plasmid DNA as probe and subsequently performed M-FISH with whole chromosome painting probes (Figure3). The chromosomal localization of the proviral inserts detected by FISH and mapping of the flanking DNA obtained from the LM-PCR product to the human genome showed identical integration sites in the HT1080 cell line clones (Table1). Because the sequencing of the human genome is not completed yet, in a single case the FISH assay was informative when the blast search for flanking sequences was not.

In the next step the progeny of MDR1-transduced human CD34⁺ PBPCs repopulating the BM of NOD/SCID mice were analyzed. One chimeric mouse BM DNA was digested with one of the different enzymes, with EcoRV, PvuII, orBsmAI, to find out the ideal enzyme. Subsequently, LM-PCR was performed. An integration site library was constructed by cloning the resulting PCR products. The cloning procedure allowed sensitive detection of fragments. Subsequently, the cloned PCR products were sequenced to prove the presence of LTR-genomic DNA junctions, and 32 different clones were identified. Because BsmAI has a 5-bp recognition sequence and EcoRV and PvuII recognize a 6-bp motif, it is expected that in a side-by-side comparison, the highest number of clones were identified when usingBsmAI (16 clones for BsmAI, 10 clones forEcoRV, 3 clones for PvuII, same BM sample; all clones had a unique sequence).

To investigate the specificity of the method, the LM-PCR was performed on BM DNA from 3 mice that had received transplants of untransduced human CD34⁺ cells (mock transduction). The LM-PCR product was concentrated and cloned as a whole, omitting procedures in which PCR fragments could get lost. Not a single LTR-flanking DNA junction could be identified in 3 mock mice that had received untransduced human PBPCs from different transplantation experiments. Thus, this LM-PCR is highly specific, and the number of detected retroviral integration sites can be increased by cloning of LM-PCR products.

By the optimized LM-PCR/BsmAI/cloning protocol, we studied the clonality of human hematopoiesis in chimeric NOD/SCID BMs after transplantation of retrovirally transduced PBPCs (Table 2).

Sequencing identified up to 32 different integration sites per chimeric mouse BM following repeated LM-PCR reactions (Table 2). Repeated LM-PCR reactions on one sample continued to identify new clones with only 10% overlap between analyses, pointing to the polyclonality of human hematopoiesis in our mice. It is noteworthy that the 5′ end of the nested LTR-specific primer was located 101 bp away from the junction, which was routinely sequenced to confirm the presence of a junction. The LTR sequences we obtained matched completely with the anticipated sequence of our vector LTR in all cases.

Figure 2B shows the LM-PCR bands that were cloned and sequenced from 5 chimeric BMs of NOD/SCID mice that received transplants of PBPCs from donor no. 1. As an example, all flanking sequences cloned from mouse E3M7 (no. 7) are shown (Figure 2B). Several integration sites not visible as a band in the gel were identified by cloning, and fragments with different sequences but almost identical length could be found.

A total of 156 SF1m proviral integration sites were identified by restriction enzyme digestion, LM-PCR, and sequencing. Of those, 141 sequences were found in the database (> 95% identity over ≥ 146 bp) and could unambiguously be mapped to a specific human chromosome. The remaining 15 fragments had a length of 146 to 331 bp, and one had a fragment length of 1000 bp and could not be mapped. Eighty-eight, 14, and 39 mapped proviral integration sites were derived from cells of donors no. 1, no. 2, and no. 3, respectively.

The chromosomal distribution of the 141 unambiguously mapped proviral inserts is shown (Figure 4). Integrations were found in all chromosomes (22 autosomes as well as the sex chromosomes X and Y), indicating that many target sites all over the genome are accessible for SF1m retroviral vector integration in hematopoietic stem cells. Of note, integrations were not equally distributed over the human chromosomes (P = .000 37).

Notably, 10 and 9 integrations resulting from all 3 donors were found on chromosome 17 and chromosome 19, respectively, whereas 3 were expected when chromosome size and finished human genome sequences entered into the database (EMBL genome monitoring table,http://www.ebi.ac.uk/genomes/mot/) at the time of preparation of this report were taken into account. These differences are highly significant as evidenced by (oi − ei)²/ei values of 21 and 12, respectively (cutoff ≥ 3). In case of nonpreferential integration a value of 0 would be expected. With 15 observed versus 9 expected integrations, chromosome 6 was also overrepresented, whereas chromosome 4 was underrepresented (observed 4, expected 10; [oi − ei]²/ei values 3.2 and 3.2, respectively) (Figure 4).

Furthermore, although as a whole chromosome 13 was not targeted more frequently than expected, we observed 5 integrations isolated from different donors into the small subchromosomal regions 13q13/13q14 (Table 3). Additionally, 4 integrations from different donors were found in the region 16p12/16p13/16q13 (Table3). On chromosomal region 19p13 a similar clustering was found (Table3).

Repetitive elements were not excluded from the blast search analysis. We did not find preferential integrations in repetitive elements such as long interspersed nuclear element (LINE) or short interspersed nuclear element (SINE) sequences. Mean guanine cytosine (GC) content was 46% for sequenced flanking DNA.

In this study we found preferred genomic targets for retroviral vector integration in human hematopoietic cells with marrow-repopulating potential.

We analyzed mobilized human PBPCs that had been transduced with the hybrid vector SF1m that is based on the Friend mink cell focus-forming/murine embryonic stem cell virus and carries the humanMDR1 gene. A high level of engraftment of these cells in the BM of NOD/SCID mice (Table 2) and gene transfer levels equivalent to recent clinical10 or myeloprotective experimental trials33 with this transgene were found. Our data suggest that approximately one vector copy integration had occurred perMDR1-transduced SRC, which can be estimated from the gene transfer frequency (mean, 17.6%, Table 2), the transgene expression (mean, 8.5%) and an additional proportion of nonfunctional splice variants of the wild-type MDR1 transgene used here.34 35 

Several groups have used PCR-based techniques to analyze retroviral vector integration into genomic DNA of human cells.18,36 We used an LM-PCR approach with which we were able to simultaneously amplify multiple integration sites from transduced bulk cell populations. We have optimized the described detection protocols21,22 for specificity and sensitivity and have adapted them to our vector and the chimeric mouse setting. Our protocol differs in terms of restriction enzymes, adapters, polymerases, template type (double-stranded fragments attached to beads), and an additional PCR product cloning step from recently published LM-PCR protocols.23,37 We were able for the first time to validate LM-PCR results by multicolor FISH for retroviral inserts (Figure 3). When we used different restriction enzymes for LM-PCR on human SRCs, different clones were found. This finding may be explained by the fact that different restriction enzymes cut the DNA into fragments that differ in amplification efficiency in the subsequent PCR or/and that there is a stochastic component in the amplification of less abundant fragments, as has been described for conventional LM-PCR methods without solid-phase fragment capture.38 

We were able to identify the highest number of repopulating human cell clones in chimeric mouse BMs reported so far16 39 (Table2). Considering that the gene-marked cells accounted for less than 10% of human leukocytes (Table 2), the multiclonality of human hematopoiesis in our small animal model becomes obvious.

Our experimental data are in line with clonality studies after transplantation of retrovirally marked autologous PBPCs in primates19 and after allogeneic transplantation in humans,40,41 supporting the conclusion that the NOD/SCID mouse assay is a valid model for human hematopoiesis.42,43On the basis of reports by Larochelle et al,42Cashman et al,44 Gothot et al,45 it was originally proposed that only pluripotent long-term repopulating cells would be readout in these human-mouse models. Given reports in mice46 and human NOD/SCID mouse studies,17 47there is growing evidence—not surprising—that primitive, lineage-restricted progenitors exist. These populations with short-term (lymphoid and myeloid restricted) and long-term (pluripotent) self-renewal capacity are all present in the SRC analyzed here.

About 92.8% of human genome sequences were finished and entered into databases at the time of preparation of this manuscript (EMBL genome monitoring table, http://www.ebi.ac.uk/genomes/mot/). Accordingly, we were able to identify 90% of the obtained sequences (n = 156) in the human genome. A study analyzing 178 different human T-cell leukemia virus-1 (HTLV-1) proviral integration sites in human blood cells 2 years ago was only able to identify the location of 47% of flanking sequences in the human genome.48 

It is well established for wild-type retroviruses that genomic integration is not a completely random process.49 It was shown by some researchers that the central core domain of the retroviral integrase plays an important role in determining the target specificity.50 The efficiency of chromosomal sites to become a preferred integration target appears to be further affected by several factors, such as transcriptional activity,51 DNAse I hypersensitivity,52 methylation,53 GC content,54 nuclear scaffold attachment,55nucleosome structure,56 and DNA structure of higher order.57 In a study with turkey embryo cellular DNA, all genomic regions contained integration targets for the wild-type avian leukosis virus DNA, with a frequency that varied from approximately 0.2 to 4 times that expected for random integration. Within regions, the frequency of use of specific sites varied considerably. Integrations in some sites occurred 280 times more frequently than expected for random integrations.58 In the mouse model ecotropic wild-type viral integration (evi) sites have been described that contribute to lymphomagenesis when located upstream of proliferation-inducing genes.59 In a recent publication it was reported that global analysis of cellular transcription indicated that active genes were preferential integration targets for lentiviral cDNA.60 

Our findings support the existence of hot spots for retroviral vector integration in human SRCs, as there was a nonrandom distribution of integrations (P = .000 37). Integrations occurred significantly more frequently into the subchromosomal regions 13q13/13q14, the small chromosomal stretch 16p12/16p13/16q13, and subchromosomal region 19p13 that were targeted in several donors and isolated from independent mice (Table 3). Other overrepresented genomic regions that were predominant in single donors include both arms of chromosome 6, and—when chromosome size and sequencing progress (EMBL genome monitoring table) were taken into account—also into chromosome 17 (Figure 4). However, the chromosome 4 was underrepresented. The GC content was 33% to 58% for flanking sequences (mean, 46%), so that DNA composition may not be the major determinant for retroviral vector integration. Other factors must be at work.49 

Mutagenesis based on the loss of tumor suppressor genes is considered a multistep process. This may be a question of statistics, with the number of integrations per cell, the total number of transduced-engrafting stem cells, and the number of patients receiving transplants as variables. Our work introduces a further factor into this equation, namely the existence of preferred integration sites of a retroviral vector in human chromosomes of marrow-repopulating cells. Progress in the human genome project by assigning function to DNA sequences will allow others to predict the consequences of retroviral vector insertions from sequencing data like that obtained in our study. Recent reports suggest that the integration site also influences the expression level of retroviral genes.51 Knowing which preferential integration sites are used in stem cells may help to understand mechanisms of integration and eventually allow vectors to be targeted to preferred sites. This knowledge may help to avoid insertional mutagenesis and reduce the genotoxicity of retroviral vector–mediated gene transfer that has recently been reported for mice12 and humans.13 

The technical assistance of Bernhard Berkus, Hans Jürgen Engel, Heidi Holtgreve-Grez, Sigrid Heil, Brigitte Schoell, and the support of the animal facility team of the German Cancer Research Center are gratefully acknowledged. We thank Dr Manfred Schmidt, Freiburg, Germany, and Dr Christoph von Kalle, now from Cincinnati, OH, for helpful discussions at the start of this project in 1998. We are grateful to Prof Christopher Baum, Hannover, Germany, for providing the retroviral vector used in this investigation. The help of Dr Andrea Schilz in cell transductions is gratefully acknowledged.

This article is dedicated to Harald zur Hausen on the occasion of his retirement as head of the German Cancer Research Center (Deutsches Krebsforschungszentrum [DKFZ], Heidelberg) with gratitude and appreciation for 20 years of leadership.