In clinical gene therapy trials for X-linked severe combined immunodeficiency, the development of leukemia has come up as a severe adverse effect. In all five cases, T-cell acute lymphoblastic leukemia (T-ALL) occurred as a direct consequence of insertional mutagenesis by the retrovirus used to deliver the therapeutic gene. Here, we review the mechanisms of insertional mutagenesis, the fuction of the Il2RG gene and the future developments in the field. New lentiviral and γ retroviral vectors can significantly improve the safety profile of the tools used but still carry the risk of insertional mutagenesis, as shown in this issue of Leukemia. Finally, the unfortunate side effects of gene therapy have given more insight into the development of human T-ALL.
The development of leukemia in five children participating in the French (n=4 serious adverse effects) and British (n=1 serious adverse effects) trials treating X-linked severe combined immunodeficiency (X-SCID) has significantly altered the field of gene therapy.1, 2, 3, 4 However, recent progress in safer vector design, insertional mutagenesis and transgene toxicity has put the field back on track. This has led to the notion that gene therapy is still experimental therapy, but can be highly successful, and that the therapeutic window between too low a level of vector transduction leading to incomplete immune restoration and too high transduction levels increasing risks of unwanted side effects still needs to be worked out (reviewed by Baum et al.,5 Cavazzana-Calvo and Fischer,6 Gaspar and Thrasher,7 Pike-Overzet et al.8 and Baum et al.9). Although the five participants in the gene therapy trials for X-SCID have developed leukemia secondary to insertional mutagenesis, two of the patients are in remission following chemotherapy and two other patients are under active chemotherapy treatment and responding well.
In a recent issue of Leukemia, a manuscript from Baum's group deals with two problems at once.10 First, the potential contribution of deregulated expression of the therapeutic gene in X-SCID, the interleukin-2 receptor common γ chain (IL2Rγ), to the development of the observed T-cell leukemias is dealt with. Second, the risk of development of leukemias using different vector designs is discussed. Both issues have been the subject of intense study by various laboratories, and we will briefly discuss these topics here, whereas for specific details, the reader will also be referred to other recent reviews.5, 6, 7, 8
Is the IL2Rγ an oncogene?
Retroviral (both γ-retroviral and lentiviral) vectors are attractive for gene therapy of genetic diseases because of efficient and stable transgene insertion with a controlled copy number11 However, in both mice and men, leukemias have been reported in which insertional activation of a cellular proto-oncogene by integrated retroviral vectors represented the initiating event (Figure 1). In the X-SCID trials, the culprit in most cases was the LMO2 gene, a well-known oncogene involved in T-cell acute lymphoblastic leukemia (T-ALL).1, 2, 11 The question then arose whether deregulated LMO2 expression in the T-cell lineage would suffice as initiating event for leukemogenesis, or whether a function of the IL2Rγ chain as a contributing factor would exist (when overexpressed as transgene). We have reported previously that in human stem cells cultured under conditions that drive them to develop into thymocytes, overexpression of IL2Rγ had no effect, whereas LMO2 overexpression blocked human T-cell development in an apparent preleukemic stage.12 Modlich et al.10 now report results from animal models with secondary transplantations to confirm that ectopic expression of IL2Rγ was insufficient to trigger leukemia.
This controversy started with two short reports employing murine models of which the results were interpreted in such a way that IL2Rγ itself could contribute to leukemic transformation. In the first report, Copeland and co-workers13 showed that insertions near Lmo2 as well as Il2rg were detected in a single murine T-cell leukemia present in the Mouse Retroviral Tagged Cancer Gene Database, which has led to the notion that these genes may act as cooperating oncogenes. However, it should be noted that several other oncogenes were targeted in the same leukemic clone as well. In the second report, Woods et al.14 claimed the first direct experimental evidence to suggest that the expression of IL2Rγ through a lentiviral vector affected the incidence of leukemias in a mouse model system. Mice transplanted with these transduced hematopoietic stem cells (HSCs) expressed very high levels of IL2Rγ certainly when compared with the slightly lower than normal levels observed in X-SCID trials. Thus, the expression level might have had an important function in the development of T-cell lymphomas in this setting. Insertional mutagenesis may also have been a significant, if not the most important, contributory factor to lymphoma development. It can be noted that the phenotype of the murine tumors (B220+CD3+) is very different from that of the T-ALL-like tumors that occurred in the patients enrolled in the X-SCID trial. The tumors found in the X-SCID trial express T-cell-specific molecules without coexpressing B-cell markers. Unfortunately, no further follow-up of these mice has been reported after the initial publication.
Conversely, reports from our laboratory and those of Drs Thrasher, Baum, Fisher and Cavazzana-Calvo presented data arguing against a direct oncogenic function of IL2Rγ.12, 15, 16 We argued that retrovirus-mediated expression of IL2RG in hematopoietic precursors is unlikely to represent a preleukemic event, as this chain is also expressed at substantial levels in normal hematopoietic CD34+ precursors as well as throughout T-cell development. Overexpression of IL2RG did not hamper T-cell development. Accordingly, as reported in this issue of Leukemia,10 no adverse effects of the transgene were observed with retroviral vectors used to express IL2RG in bone marrow cells of C57Bl6/J mice (including the very same vectors (MFG) as used in X-SCID patients). However, some of the transplanted mice developed leukemia in conjunction with insertion near a proto-oncogene such as Evi1 or Prdm16. Evi1 is the most frequently targeted oncogene in murine retroviral insertion experiments. It can be noted that these leukemias could be detected only after secondary transplantation, which puts additional mutagenic stress on HSCs. Taken together, these observations strongly argue against IL2Rγ acting as a cooperative oncogene in the human or murine gene therapy setting.8
This controversy has formed the impetus for the development of two important experimental approaches to make gene therapy safer (and also more effective). First, the development of assays and model systems that accurately and quantitatively address the issue of vector toxicity. Second, the continuous efforts to improve retroviral and lentiviral vectors tailored to specific diseases.
Novel mouse models
Consequently, new mouse model systems that may enhance sensitivity in the detection of leukemogenic events have been developed by Shou et al.17 and Montini et al.18 In the first model, both the Arf tumor-suppressor gene and the Il2rg gene were ablated. Retroviral transduction of IL2RG into the HSCs of these mice also leads to integration-dependent T-cell tumors as were found in the X-SCID trials, although with higher frequency. The deletion of Arf leads to a more rapid development of T-cell lymphomas, allowing for a much easier readout of oncogenic events.17 Naldini and coworkers18 have developed a different tumor-prone mouse model. They have used Cdkn2a−/− mice to demonstrate that the lentiviral integration profile, especially when combined with a safer vector architecture, has a more favorable safety profile when compared with standard Moloney-Murine Leukemia Virus (Mo-MLV) vectors. In addition, in vitro screening of vectors to provide quantitative data on vector design-related insertional mutagenesis have led to promising results.19 Whether any one of these approaches will provide the necessary sensitivity to accurately measure vector and/or transgene-related toxicities remains to be seen. As shown in this issue by Modlich et al.10 and earlier by the same group, serial transplantations remarkably increase the sensitivity of detecting genotoxic effects of viral vectors.20, 21 In this report, none of the 60 primary transplantations, but in 12 out of 46 secondary recipients, developed vector-related leukemias.10 Combining tumor-prone mice with early retransplantation of transduced HSC therefore appears to provide a quicker way to detect insertional mutagenesis events.
Thus far, all cases of leukemogenic complications either in clinical trials or animal models involved the use of conventional retroviral vectors with long terminal repeats containing strong enhancer/promoters. Self-inactivating retroviral vectors that contain only one internal enhancer/promoter should reduce the incidence of interactions with nearby cellular genes,22 but the work of Baum10 shows that these self-inactivating vectors are still associated with the risk of insertional mutagenesis, as are lentiviral vectors. This study is too small to make quantitative risk assessments, but the authors probably conclude correctly that retroviral or lentiviral vectors with strong internal promoters should be avoided and, where possible, replaced by weaker cellular and preferentially lineage or tissue-specific promoters. Nevertheless, the lower propensity of lentiviral vectors for integration at the 5′ start side of genes may suggest a preferred safety profile for gene therapy applications of such vectors.23, 24 Indeed, many laboratories in the world are now developing clinical gene therapy applications using such self-inactivating vectors.
Integration site selection has emerged as the major factor responsible for toxicity of an integrating vector. Integration of both lentiviral and γ-retroviral vectors is known to be favored in active transcription units, but a major advantage of lentiviral vectors is that they are not attracted to gene-regulatory regions such as DNAse hypersensitive sites.23, 25, 26, 27 Three back-to-back reports in The Journal of Clinical Investigation have mapped γ-retroviral integration sites to gene expression profiles of HSCs, each report dealing with a different SCID gene therapy trial.28, 29, 30 In these reports, the combination of two powerful techniques, LAM-PCR to identify retroviral insertion sites,31, 32 and genome-wide gene expression profiling such as that of HSCs33, 34 and thymocytes35 have been used. The authors from these three reports (French X-SCID, British X-SCID and ADA-SCID) reach similar conclusions: the integrations of retroviruses in the genome certainly do not occur at random, with up to two-thirds of the insertions being in or very near to genes that are highly expressed in CD34+ cells, often clustered as common integration sites. This shows that viral integrations are correlated with the level of expression of genes in the CD34+ stem/progenitor cells that are used as target cells. As a result, retroviral vector integrations in HSCs occur preferentially near highly expressed genes with specific functions in immature HSCs.
Insights in human T-ALL development
Recent follow-up work on these The Journal of Clinical Investigation reports by the laboratory of Adrian Thrasher shows the power of careful molecular monitoring of gene therapy trials for understanding human leukemia development.3 In a detailed analysis of the one British patient in the X-SCID trial that developed T-ALL, these investigators could show that peripheral T lymphocytes with LMO2 integrations could be detected before the development of full-blown leukemia. Frank leukemia was detected only after acquisition of additional somatic mutations, most notably in NOTCH1 and deletion of the CDNK2A tumor suppressor gene. This underscores in a human setting that leukemia development is a multistep process that requires multiple, consecutive, genetic abnormalities.
Clearly, integrating gene transfer vectors have been associated with potentially dangerous genetic lesions. Multiple integrations into one cell, although attractive to reach high levels of expression of the desired genes, specifically need to be avoided, as cooperating effects between two ectopically expressed target genes may occur.13 The relationship between retroviral transduction efficiency and the actual number of integrated vector copies has been carefully investigated by Boris Fehse and coworkers.36, 37, 38 They showed a linear increase of expression levels with insertion frequency: one vector insertion per transduced cell for a gene transfer of less than 30%, 3 for 60% and approximately 9 for 90%.37 Given the risk of insertional mutagenesis, the dosage of viral vectors should be chosen to limit gene transfer to approximately 30%, thereby largely avoiding clones containing multiple insertions. Thus, these studies have defined an upper limit of viral transduction and thereby help define the therapeutic index of integrating vector systems for various gene therapy strategies.
Finally, we would argue for gene therapy constructs to be tested for safety in human cells. The non-obese diabetic (NOD)-SCID model is classically used for in vivo studies of human hematopoiesis. The disadvantage of this model is the lack of proper human T-cell development. Recently developed immunodeficient mice that are used for studying human hematopoiesis, including T-cell development, may serve as good models to study gene therapy vectors.39, 40 Although such models would mainly be used to address efficacy of gene therapy, careful analysis of human lymphoid and myeloid development in such mice could reveal partial blocks in differentiation and preleukemic conditions. This creates an extra tool for investigators attempting to develop successful gene therapy for immunodeficiencies and other monogenetic diseases.