The Yin and Yang of Stem Cell Gene Therapy: Insights into Hematopoiesis, Leukemogenesis, and Gene Therapy Safety

It has been more than two decades since replication-incompetent retrovirus vectors were first used to genetically modify murine hematopoietic stem cells (HSCs), and more than 15 years since initiation of first-generation human clinical trials using this strategy. These trials demonstrated little evidence of successful modification of HSCs. Investigators, with guidance from a 1996 National Institutes of Health (NIH) Blue Ribbon panel and other funding agencies, focused efforts on the development of better vectors, a more detailed understanding of target diseases, and improved collection and ex vivo culture of target HSCs and progenitor cells. The use of models such as nonhuman primates, dogs, and murine immunodeficient xenografts instead of murine models allowed development of methodologies and vectors that would directly translate to success in human clinical trials. Important factors have been recently reviewed, and include the use of a fibronectin matrix to concentrate vectors adjacent to HSCs while preventing their differentiation, inclusion of cytokines such as flt3 ligand, thrombopoietin and stem cell factor in transduction cultures, and use of nontoxic conditioning regimens prior to infusion of large numbers of transduced CD34⁺ cells.¹ These efforts culminated in a second wave of clinical gene therapy trials beginning in the late 1990s and early 2000s. For the first time, clear clinical benefit was achieved, specifically in patients with serious genetic immunodeficiency diseases.^2,–4 

However, in 2003 the first serious adverse event linked to genetic modification of HSCs or primitive precursor cells with integrating retrovirus vectors was reported.^5,6 This review will summarize the encouraging results of these clinical trials, explain what is known regarding the adverse events linked to genotoxicity of vector integration, discuss the likely factors predisposing to these genotoxic events, and propose strategies for the future that can retain or improve the clinical efficacy of gene therapy directed at hematopoietic targets, while greatly decreasing the risk of leukemia and other adverse events resulting from vector insertions into the genome.

In 1999, investigators led by immunologists Alain Fischer and Maria Cavazzano-Calvo at the Necker Hospital in Paris began entering boys with X-linked severe combined immunodeficiency (X-SCID) into a gene therapy trial. Patients with X-SCID have profound immune dysfunction resulting from mutations in the common γ (cγ) component of a cytokine receptor subunit required for responses to inter-leukin (IL) 2, IL4, IL7, IL9, and other cytokines, required for proliferation, maturation, and function of cellular components of the immune system. Patients with X-SCID have complete absence of T and natural killer (NK) cells, and profound B-cell dysfunction, and invariably die in early childhood unless allogeneic stem cell transplantation from a matched sibling can be performed. For patients without a matched sibling or closely matched unrelated donor, haplo-transplantation from a parent can be performed, but outcomes are poor.

Following extensive preclinical investigations and regulatory agency and institutional review board reviews, autologous bone marrow CD34⁺ cells were collected from boys with X-SCID, cultured in the presence of a standard retrovirus vector expressing the cγ transgene, and reinfused without marrow ablative conditioning (Figure 1; see Color Figures, page 519). Of 11 patients enrolled, 10 had rapid and robust reconstitution of T and NK cell numbers and B-cell function.^7,8 All T cells contained and expressed the cγ vector transgene, in contrast to a a much lower frequency of B cells and myeloid lineage cells. This was not surprising, since patients were not given any myeloablative conditioning prior to transplantation, and there is no selective advantage for genetically modified cells in B-cell and myeloid lineages, in contrast to the T lineage. Within several months, these boys stopped antimicrobials, exited protective inpatient hospital units, and began to respond to vaccines. They had highly polyclonal T cells, as indicated by both T-cell receptor (TCR) diversity and the presence of multiple vector insertion sites. A second X-SCID trial was initiated several years later in the United Kingdom at the Great Ormond Street Hospital and also reported very encouraging results, using a similar trial design in terms of patient selection, use of marrow CD34⁺ target cells, and gene transfer vector structure, but with subtle differences in transduction culture conditions.⁴ Investigators and patients began to consider that X-SCID had been cured by gene therapy.

The second congenital immunodeficiency syndrome successfully treated with HSC gene therapy is adenosine deaminase (ADA) deficiency. In this disorder, T-, B-, and NK-cell numbers and function are impaired due to lack of an enzyme (ADA) that normally prevents accumulation of metabolites toxic to lymphoid cells. These patients also have a variety of nonimmune metabolic abnormalities that impair growth, development, and quality of life. Several gene therapy trials in the early 1990s gave patients autologous lymphocytes or CD34⁺ cells transduced with retroviral vectors expressing the ADA gene, but insufficient transgene expression, minimal engraftment of gene-modified cells, and continuation of patients on exogenous ADA enzyme replacement therapy resulted in lack of demonstrable clinical benefits.^9,–11 In contrast to X-SCID, in the absence of myelosuppression prior to transplantation, transduced CD34⁺ cells did not engraft at levels resulting in clinical benefit. Investigators from Milan reported that treatment with moderate-dose busulfan prior to infusion of retrovirus-transduced autologous CD34⁺ cells was well tolerated and resulted in sustained expression of the ADA gene, with restoration of T-cell numbers and normalized immune function.³ These patients with ADA did not receive concurrent PEG-ADA enzyme replacement therapy, favoring the survival and expansion of corrected T lineage cells.

Optimism regarding these encouraging clinical trial results was abruptly dashed in late 2002 when 2 boys enrolled in the initial X-SCID trial developed aggressive T-cell leukemias, 3 years after transplantation of gene-modified CD34⁺ cells.⁶ In both patients, expanded T cells were clonal and contained the vector. In the first child the tumor cells contained a vector integration within the first intron of the Lmo2 gene, and in the second, just upstream of the same locus, which encodes the transcription factor Rhombotin-2, known to be required for lymphoid development and activated by chromosomal translocations in some spontaneous human T-cell acute lymphoblastic leukemias (TALLs).¹² The Rhombotin-2 mRNA was expressed at very high levels in the leukemic cells, expressed from the Lmo2 allele containing the vector insertion, while the cγ transgene was unmutated, expressed at normal levels, and the IL2R complex did not appear to be abnormally activated in the tumor cells. Subsequently, two additional patients enrolled in the original trial have been reported by the investigators to have developed T-cell proliferation more than 3 years following transplantation (personal communication, Alain Fischer and Marina Cavazzana-Calvo), although molecular details have not yet been published. All patients have responded to chemotherapy, but 1 patient died after a matched-unrelated allogeneic stem cell transplantation performed to treat recurring disease.

The occurrence of vector-related leukemia only in the patients with X-SCID, and not in patients with other types of immune deficiencies receiving retrovirus vector–transduced CD34⁺ cells, has resulted in investigation of the role of X-SCID–specific factors in the adverse event etiology. Are all patients receiving autologous CD34⁺ cells or other cells transduced with integrating retrovirus vectors at risk, or only those with X-SCID? The answer has important implications for further clinical development of gene therapy with integrating vectors. The finding that both the cγ gene and the Lmo2 gene were activated concurrently in a murine T-cell lymphoma arising following infection of mice with replication-competent retrovirus suggested a unique cooperativity between these two genes when activated aberrantly by viral enhancers.¹³ Overexpression of cγ in murine bone marrow cells resulted in a high rate of T-cell leukemia in one study; however, these results were controversial, and other investigators did not find that common γ expression was inherently oncogenic.^14,15 The level of cγ expression and signaling in mature T cells seemed to be in the normal range in the patients with X-SCID treated with gene therapy; however, it seems possible that constitutive expression of the transgene earlier in T-cell development could have contributed to the later aberrant behavior of these cells. A second important factor increasing the risk of further mutational events and eventual leukemia in X-SCID may have been the very rapid T-cell expansion into a completely empty T-cell compartment, following transduction of T-cell progenitors arrested at a susceptible stage of development.

Virologists and gene therapy researchers have had concerns regarding the potential genotoxicity and risk of leukemogenesis associated with integrating retroviruses since their initial development as gene transfer vectors. Figure 2 (see Color Figures, page 519) summarizes several mechanisms by which the integrated proviral DNA of a retrovirus can be oncogenic, either by upregulation of expression of adjacent protooncogenes by strong viral enhancers and promoters, or conversely by inactivation of tumor repressor genes via disruption of an exon or positive regulatory elements.

Standard retrovirus vectors were originally derived from the Moloney murine leukemia virus (MLV). The wild-type replication-competent strain of this virus induces thymic lymphomas in newborn mice of susceptible strains, despite the lack of an oncogene encoded by the viral genome. Tumors only arise in mice infected with wild-type replicating virus during active thymopoiesis. Each tumor has a large number of viral insertions, and investigators hypothesized that repeated integration of MLV proviruses into the genome of T-cell progenitors in the thymus eventually resulted in activation of protooncogenes. Certain proto-oncogenes such as pim and myc were found to be recurrent viral insertion sites in these lymphomas.

In 1992, T-cell lymphoma was diagnosed in 3 rhesus macaques that received transplants of autologous CD34⁺ cells transduced with an MLV vector preparation contaminated with replication-competent recombinant viral particles.¹⁶ These tumors contained large numbers of proviral insertions. However, no tumors were linked to replication-incompetent vectors for the first 15 years of their use in transduction of HSCs from mice, non-human primates, or patients. It was thus assumed that ongoing viremia with replication-competent virus was a prerequisite for leukemogenesis, and that use of replication-incompetent viral vectors would result in only a single or very few random insertions per cell. The size of the mammalian genome, the multi-hit hypothesis of cancer initiation, and early studies completed prior to availability of the sequence of the full human and murine genome suggesting random MLV integration were all additional factors thought to predict a very low risk of leukemogenesis when using replication-defective retrovirus vectors.¹⁷ Regulatory agencies and investigators focused almost completely on prevention and detection of replication-competent virus contaminating any vector preparation to be used for clinical trials.

Availability of the complete murine and human genome sequences and the development of powerful PCR-based methodologies for identifying large numbers of proviral genomic integration sites stimulated a reassessment of retrovirus integration patterns. Human immunodeficiency virus (HIV) and HIV vectors preferentially integrated into actively expressed genes in T-cell lines, and MLV vectors preferentially integrated near the 5′ transcription start sites of genes in fibroblast cell lines.^18,19 These patterns would be predicted to increase the risk of perturbation of gene expression by vector insertion, as compared to random integration.

Our group performed a large-scale analysis of integration sites in granulocytes and lymphoid cells produced by retrovirus- or lentivirus-transduced CD34⁺ cells engrafted months to years previously in rhesus macaques.²⁰ Both MLV and simian immunodeficiency virus (SIV) vectors non-randomly inserted in or near genes, with SIV insertions distributed throughout the entire gene length and MLV insertions clustered near transcription start sites. Recent large integration analyses of X-SCID and ADA deficiency gene therapy trials reported similar findings for MLV vectors.^21,–23 Table 1  summarizes current knowledge regarding integration patterns of these and other gene transfer vectors. These accumulating data suggest that each parental virus evolved to use different cellular cofactors for accessing and integrating within chromosomal DNA.

Large integration site datasets from mice, non-human primates, and humans enrolled in clinical trials have also revealed convincing evidence for the impact of specific integrations on the survival, engraftment and/or proliferation of transduced primitive hematopoietic cells, even in the absence of leukemic transformation. Based on the size of the genome and the number of integration sites identified, the occurrence of more than one independent integration within a certain gene or located within 30 kb to 100 kb of each other has been called a “common integration site” (CIS) and, when found to occur within independent tumors in replication-competent oncogenesis models, linked causally to the tumors.²⁴ 

Our analysis of integrations in granulocytes following autologous transplantation of MLV-transduced rhesus macaque CD34⁺ cells identified the Mds1/Evi1 genomic locus as highly over-represented, accounting for 2% of all sites mapped.²⁵ These clones were found only in the myeloid lineage and to date have not been associated with progression to leukemia. However, it is worrisome that this locus was also identified as a vector insertion site in the first mouse reported to have a replication-incompetent vector-associated myeloid tumor.²⁶ 

In a recent clinical trial for chronic granulomatous disease (CGD), 2 patients initially demonstrated very encouraging levels of genetically corrected granulocytes and cleared serious chronic infections.²⁷ However, several months following transplantation, the level of corrected cells gradually increased from 5% to 10% up to greater than 50%, and this expansion was found to be due to clonal dominance of cells with MLV vector insertions in the Mds1/ Evi1. One patient lost expression of the corrective transgene and died of an infection, and the second patient has remained well, without progression to abnormal hematopoiesis or leukemia. Primary murine bone marrow cells transduced with MLV vectors were cultured in vitro, and several weeks later immortalized myeloid cell lines emerged. All had vector insertions activating the Mds1/Evi1 gene locus, a closely related homologous locus, or an upstream pathway that also turned on expression of Evi1.²⁸ 

The proteins encoded by either Evi1 or Mds1/Evi1 alternatively spliced mRNAs are transcription factors with DNA-binding activity.²⁹ Their function is poorly understood, but spontaneous human myeloid leukemias have been reported with translocations activating this locus, and overexpression of the Evi1 isoform is a poor prognostic factor in human acute myeloid leukemia (AML).³⁰ In aggregate, these findings suggest that insertions landing in or near the Mds1/Evi1 gene complex in primitive hematopoietic cells may have a profound impact on cellular behavior, immortalizing myeloid progenitor cells or otherwise affecting the survival or output from transduced clones. Serial transplantation experiments using replication-incompetent MLV vectors in mice have also identified hematopoietic clonal dominance or leukemias arising from cells containing vector common integration sites in or near Mds1-Evi1 and other proto-oncogenes.^31,32 A rhesus macaque that underwent transplantation 5 years earlier with MLV-transduced CD34⁺ cells died of AML and tumor cells had vector activation of the anti-apoptotic Bcl2A1 gene.³³ Development of MLV vectors for clinical applications targeting hematopoietic stem and progenitor cells almost completely halted by 2005, while investigators and regulators reassessed risk-benefit based on the human X-SCID trial adverse events and new knowledge of vector integration patterns and genotoxicity.

A number of modifications in vector design are being explored to decrease the risk of genotoxicity from integrating viral vectors. Patients with HIV do not appear to develop integration-related tumors, and in animal models, vectors derived from HIV have been less leukemogenic than standard MLV retrovirus vectors.³⁴ This is potentially due to a less risky integration pattern, with HIV and SIV integration sites spaced along the length of genes instead of clustered at the transcription start site, where activation of an adjacent gene may be most likely.^19,20 But HIV and related SIV vectors were designed with large deletions of enhancer elements in the LTRs at either end of integrated proviral forms. This design (called “self-inactivating” [SIN]) was originally chosen to decrease the risk of vector recombination with endogenous HIV in patients receiving transduced cells, but may have the additional benefit of being less likely to activate adjacent proto-oncogenes (Figure 2; see Color Figures, page 519). MLV vectors can also be modified to a SIN design, and one study indicated a lower risk of activation of growth-promoting genes after transduction with SIN versus standard MLV vectors.³⁵ However, strong internal promoters necessary to drive adequate transgene expression may also activate adjacent proto-oncogenes, so even vectors with a SIN design can prove genotoxic, via either gene activation or inactivation. DNA elements known as “insulators” can be added to vectors to decrease the impact of vector elements on the surrounding genome, and conversely to protect expression of the transgene from position effects related to the integration site³⁶ (Figure 2; see Color Figures, page 519). Several other retroviruses, including human foamy virus and avian sarcoma leucosis virus, have been shown to have more random integration patterns, not targeting genes, and are thus less likely to activate proto-oncogenes.^37,38 However, even a random integration pattern would clearly be associated with some genotoxic risk, particularly in situations with multiple vector insertions per target cell.

A number of strategies are in development to target integration at specific sites in the genome. The ultimate goal is correction of a defective gene via homologous recombination. Alternatively, use of an integrase that targets one or more specific sites in the genome distant from proto-oncogenes may be safer than current viral vector integration preferences. Most exciting is the ability to design zinc finger nucleases that bind to specific DNA motifs surrounding a gene sequence in need of correction, then make double-stranded DNA cuts and facilitate homology-directed repair. In 2005, scientists at Sangamo reported correction of the X-SCID defect in primary human T cells at a rate of 7%.³⁹ 

Bacteriophage ϕ31 contains an integrase that directs integration to several specific “att” sites in the bacterial genome. Calos and colleagues have reported that mammalian genomes contain pseudo-att sites, which are used as preferential integration targets for transgenes carried by plasmids also containing a ϕ31 integration motif in the presence of the ϕ31 integrase.⁴⁰ The major challenge for these and other targeted non-viral integration strategies is development of an efficient and non-toxic method to transport new genetic material into the nucleus of primary cells. Physical or chemical methods are generally very toxic to hematopoietic repopulating cells, despite high transfection efficiencies reported with Amaxa and other optimized electroporation devices for CD34⁺ cells studied in vitro. Non-integrating lentivirus vectors are being devised to transfer transgenes and integrases or nucleases into target cells. These approaches are very complex and likely years from application to clinical gene therapy of hematopoietic stem cells.

The 21st century thus far has been tumultuous for gene therapy, with triumphs such as clinical correction of several immunodeficiency disorders tempered by the realization that genotoxicity related to integrating vectors is a concrete and very serious risk. Continued investigation of vector biology; genomic, viral and cellular determinants of integration sites; and target cell characteristics; along with further innovation in the design and optimization of novel, safer vectors should allow continued progress in the field; however, current trials targeting HSCs will clearly need to be limited to patients with truly life-threatening diseases.