Survival of the fittest: in vivo selection and stem cell gene therapy

It took more than 15 years of research from the first description of successful gene transfer to murine hematopoietic stem cells to the first unambiguously successful clinical trials, in patients born with X-linked severe combined immunodeficiency (SCID) and adenosine deaminase (ADA) deficiency.^1-4  Low gene transfer to human stem cells has long been the most prominent obstacle to widespread clinical application of stem cell gene therapy. A solution to this problem has been in vivo selection. In vivo selection increases the proportion of circulating gene-corrected cells by conferring a selective growth and/or survival advantage to the corrected cell population. While improvements in gene transfer technology did contribute, it is generally agreed upon that in vivo selection was the key determinant of the clinical success in these studies.

In vivo selection can be constitutive, as in X-linked SCID, where the introduction of a therapeutic transgene confers a continuous proliferation and survival advantage to the transduced cell population. In diseases where overexpression of the therapeutic gene does not confer a survival advantage, a second, selectable gene could be incorporated into a vector. Ideally, this selectable gene would be placed under some form of pharmacologic (small-molecule mediated) regulation to make selection conditional. Two major classes of selectable genes have been evaluated for conditional selection: those that confer a drug-dependent proliferative advantage to the transduced cell population (so-called “cell growth switches”), and those that confer resistance to selective pressure (usually a cytotoxic drug) that eliminates the nontransduced cell population.

After the first successful studies of stem cell gene transfer in mice, early clinical trials demonstrated that gene transfer to multilineage long-term repopulating stem cells with the available retroviral vector systems was significantly more efficient in mice than in humans. In contrast, large-animal models, mainly dogs and nonhuman primates, have closely mirrored the difficulties encountered in clinical trials of stem cell gene therapy. Large-animal studies have been instrumental in devising improved transduction protocols, resulting in marking levels at, and more recently above, the threshold of anticipated clinical efficacy. Gene transfer in large animals is now consistently more than 1% and is often more than 10%,^5-8  a range predicted to be therapeutic in a number of diseases. The field of in vivo selection has undergone a similar development: It has been easier to achieve in vivo selection in murine models than in large-animal models or human patients. Here, we review how the field of in vivo selection coevolved with stem cell gene therapy, a journey that led from disappointing results^9-11  to ultimate success, of varying degrees, in large-animal models and patients.^1-4,12-14 

In a number of inherited diseases, the genetic defect imparts a survival disadvantage to the affected cell population. In these diseases, a corrected cell population is anticipated to undergo spontaneous in vivo selection in the absence of any exogenously applied selective pressure. For some diseases, natural in vivo selection has clearly been demonstrated in patients or relevant animal models. In other diseases, successful in vivo selection has not been demonstrated but can be predicted based on theoretic considerations.

SCID was the first genetic disease to be treated by stem cell gene therapy^11,15  and the first one to be treated successfully.^1-4  In the French X-SCID trial, engraftment of gene-corrected stem cells is estimated to be in the range of 0.1% to 1% as measured by production of peripheral blood granulocytes, which are not expected to undergo any positive or negative selection in this disease. This low gene transfer level leads to substantial correction of T cells and moderate (but, interestingly, not complete) correction of absolute natural killer (NK)-cell numbers. Virtually 100% of these 2 lineages contains vector, indicating substantial in vivo selection mediated by expression of the common gamma-chain. There is only moderate positive selection within the B-cell compartment, with 1% to 5% of B cells containing the vector. Data from earlier clinical trials had provided evidence that genetically corrected ADA-deficient lymphocytes may also undergo natural in vivo selection.¹¹  The interpretation of early gene therapy attempts for ADA deficiency is complicated by the fact that pegylated enzyme is available for therapeutic purposes. This treatment blunts the selective survival advantage for transduced cells. However, with-holding established treatment was initially considered unethical in these earlier trials, and was attempted only late after transplantation, with ambiguous results.¹¹  In a more recent study, ADA patients for whom ADA therapy was not available were treated with gene therapy.¹  Nonmyeloablative conditioning with busulfan was administered before the infusion of gene-modified cells. The protocol used in this study appears to have been highly successful. The study also raises interesting questions. One of the patients had a higher level of stem cell engraftment, and this correlated with a higher proportion of corrected T cells, NK cells, and B cells. The second patient had a stem cell engraftment of about 0.1% as estimated by CD34 cells and granulocytes, which are not selected for or against in ADA deficiency and should, therefore, reflect stem cell engraftment. Immune reconstitution in this second patient was slower than in the first patient, suggesting that engraftment of more than 1% gene-modified stem cells may be necessary for a full therapeutic effect in ADA deficiency. In this second patient, marking was higher in the T-cell compartment than in the NK- and B-cell compartment, even though all 3 lineages are affected in untreated patients with ADA deficiency. An explanation for this finding may lie in the fact that there is transcorrection of nontransduced cells mediated by soluble enzyme that might be released from corrected cells. Future studies of additional patients will likely lead to a better understanding of the pathophysiology and the selection process in ADA deficiency.

SCID due to heritable deficiencies of Jak3,^16,17  Rag2,¹⁸  and Zap70¹⁹  should theoretically be amenable to gene therapy, and the feasibility of gene therapy for these genetic defects has been demonstrated in rodent models. For other candidate SCID forms, such as deficiency of IL-7R-alpha and Rag-1, reports describing gene replacement in murine models have not yet been published to our knowledge.

In Wiskott-Aldrich syndrome, a disease characterized by congenital thrombocytopenia, immune dysregulation, and a propensity toward lymphoma development later in life, selection at the level of mature lymphocytes could be demonstrated both in a murine model of gene therapy^20,21  and in patients with spontaneous reversion.²²  Hematopoietic stem cell migration/engraftment appears to be defective in WASP knock-out cells in a murine model.²³  Whether this might translate into a selective advantage for vector-corrected stem cells in human WAS patients is unknown.

The best-characterized congenital defect of the hematopoietic stem cell is Fanconi anemia. At least 12 complementation groups have been described, and 11 single-gene defects (FANCA, FANCB, FANCC, FANCD1/BCRA2, FANCD2, FANCE, FANCF, FANCG, FANCJ, FANCL, FANCM) have been molecularly defined in Fanconi anemia, with FANCJ and FANCM being the most recent addition.^24,25  Our understanding of how the FA pathway controls the cell's genetic integrity and DNA-repair has expanded dramatically over the last few years (reviewed in D'Andrea and Grompe²⁶ ). Mouse models for several subforms exist,^27-30  and some of them have been corrected by stem cell gene therapy.^31-33  Hypersensitivity to bifunctional alkylating drugs such as cyclophosphamide is a hallmark of FA, and it has been shown that in mouse models, cyclophosphamide can be used to accentuate selection of genetically corrected cells in the hematopoietic system.³²  It has also been demonstrated in the murine model that genetic correction can stabilize the corrected cell population.³⁴  In human patients, the spontaneous reversion of the defective gene has been described in lymphocytes,³⁵  and interestingly, in a single published patient, reversion has been documented at the level of the hematopoietic stem cell.³⁶  In this patient, partial correction of the phenotype has been noted. The fact that correction was only partial in this patient raises concerns about the efficacy of gene therapy strategies for Fanconi anemia. Another concern is that genetic correction might occur in a stem cell that has already sustained genetic damage. Such a stem cell would lose the hypersensitivity to bifunctional alkylating drugs, which could potentially result in the development of drug-resistant leukemia. However, it is also possible that genetic correction in FA prevents the emergence, or outgrowth, of a malignant clone. Clinical evaluation of gene therapy and the associated risks will need to be measured against the risks associated with the only curative treatment, namely allogeneic stem cell transplantation. While matched related donor stem cell transplantation usually results in high cure rates with minimal toxicity, grafts from an unrelated or partially matched family donor for the treatment of Fanconi anemia are still associated with substantial morbidity and mortality.^37,38  We therefore suggest that gene therapy should, at least initially, be reserved for patients without a matched sibling stem cell donor. Theoretic considerations imply that gene therapy, or at least the harvest of the target cell population for genetic correction, should be offered before damage to the hematopoietic stem cell has progressed to include cytogenetic aberrations and/or myelodysplasia (ie, as soon as possible after diagnosis).

Other inherited stem cell defects that are potential candidates for gene therapy include amegakaryocytic thrombocytopenia, dyskeratosis congenita, and Shwachman-Diamond syndrome. Little experimental work has been published regarding gene therapy for these latter diseases.

The hereditary hemolytic anemias, thalassemia, and sickle cell disease are the most common group of inherited diseases world-wide and, thus, are an extremely important group of candidate diseases for stem cell gene therapy.

It has long been known that there is a survival advantage for corrected red blood cells in patients with thalassemia at a relatively mature stage of erythropoiesis. Analysis of thalassemic patients with mixed bone marrow donor-host chimerism after fully myeloablative conditioning and allogeneic stem cell transplantation demonstrated in vivo selection of genetically normal mature donor red cells.^39,40  In vivo selection of mature red cells has also been shown in a murine model of beta-thalassemia⁴¹  and has recently been characterized systematically in the mouse model.⁴²  It appears that a bone marrow chimerism of approximately 20% increases the hemoglobin content by about 20 g/L, while reducing reticulocytosis from 20% to 10% in a murine model of thalassemia intermedia. At this level of bone marrow chimerism, about 70% of circulating red cells were donor derived, indicating in vivo selection of mature red cells.

However, an improvement in hematocrit may not be sufficient to define a cure for the thalassemias and sickle cell disease. In a mouse model of sickle cell disease, 100% donor-derived red cells were required to completely correct organ pathology mediated by vascular occlusion.⁴³  In the thalassemias, iron overload is a central therapeutic problem before and also after allogeneic bone marrow transplantation. The most efficient treatment for iron overload is phlebotomy; however, this treatment is not routinely offered to patients with mixed hematopoietic chimerism after allogeneic stem cell transplantation.⁴⁴  Clinical data regarding the iron status of thalassemic patients with mixed hematopoietic chimerism after stem cell transplantation have not been published. However, a canine model of nonmyeloablative allogeneic stem cell transplantation for another hemolytic anemia, pyruvate kinase deficiency, suggests that persistent recipient hematopoiesis may sustain iron loading, even if substantial donor chimerism leads to improvement of hematologic parameters. Furthermore, iron-loading has been described in transfusion-independent patients with β-thalassemia intermedia.^45,46  The demonstration of suppressed hepcidin levels in mice⁴⁷  and humans⁴⁸  with thalassemia intermedia provides a mechanistic explanation for this phenomenon.

In summary, these findings suggest that the natural in vivo selection of corrected red cells may not be sufficient to allow for aggressive lowering of total body iron in patients with low allogeneic mixed chimerism or a low level of genetically corrected cells. Incorporation of a separate, selectable marker may be desirable in the design of gene therapy vectors for the treatment of hemolytic anemias. The feasibility of using selectable 2-gene vectors for thalassemias and hemoglobinopathies has recently been demonstrated in a very encouraging pilot study in a murine model of β-thalassemia.⁴⁹  This group of diseases constitutes an example of how conditional and constitutive in vivo selection could potentially be combined to maximize the therapeutic effect.

For genetic defects that result in cellular dysfunction but not diminished growth or survival (eg, chronic granulomatous disease), transfer of the therapeutic gene alone is not expected to result in in vivo selection. In these diseases, a second selectable gene can be incorporated into the vector to allow for conditional in vivo selection (Table 1). An important concept has recently been appreciated and confirmed in preclinical models: For the selection to be most durable, it should ideally occur at the level of a long-lived cell with high proliferative potential (Figure 1).

In recent years, several systems have been developed to allow for pharmacologically regulated, reversible forced protein-protein interaction in vitro and in vivo, using so called “chemical inducers of dimerization” (CIDs). The use of these systems in gene therapy has recently been reviewed⁵⁰  and will be mentioned only briefly here. Dimerizer systems rely on 2 components. The first is a ligand or drug, and the second is a fusion protein that combines a ligand-binding domain and an effector domain (usually the intracellular proportion of a growth factor receptor). The effector domain is activated by drug binding, with subsequent dimerization. General feasibility has recently been demonstrated in large-animal models.^51,56  The large-animal experiments show 2 limitations of dimerizer-based in vivo selection that will need to be addressed before this system can be successfully translated to the clinic: (1) Selection occurred in short-lived cells, creating a need for continuous or repeated drug selection to maintain therapeutic marking levels. (2) In one study, besides the desired expansion of the favored population (erythroid cells), unexpected pro-B-cell expansion was observed.⁵¹  If the signaling domain could be placed under the control of a tissue-restricted regulatory element, continuous or repeated dimerizer-mediated stimulation of the genetically corrected cell population (maybe analogous to the use of G-CSF in Kostmann syndrome) could be envisioned. This approach, however, would make the patient dependent on life-long dimerizer administration. The efficacy and safety of chronic dimerizer administration should be further investigated in relevant large-animal models before proceeding to clinical trials. Alternatively, the identification of a dimerizable signaling domain with the potential for stem cell expansion would theoretically require a finite number of dimerizer administrations and is therefore more likely to help translate this strategy into clinical applications (Figure 1). No signaling domains are currently known that upon dimerization mediate expansion of human or large-animal stem cells. In the mouse, constitutive Stat3 activation suffices to maintain immaturity of embryonic stem cells in vitro,⁵⁷  and preliminary data suggest that overexpression of a constitutively active Stat3 mutant can expand murine hematopoietic stem cells in vivo.⁵⁸  Therefore, Stat3 may be a reasonable candidate for stem cell expansion using dimerizer systems.

Recent reports have raised interest in HOXB4, Notch, and the Wnt/beta-catenin pathway for expansion of hematopoietic stem cells ex vivo and in vivo. Overexpression of Hoxb4 has been shown to expand murine stem cells in vivo⁵⁹  and ex vivo.⁶⁰ Hoxb4 has also been shown to expand human immature hematopoietic cells.^61,62  However, in contrast to cytokine receptors, Jaks, and Stats, Hoxb4 activity is not controlled by simple homodimerization, which makes the generation of conditional mutants technically more challenging. Similar deliberations apply to notch⁶³  and beta-catenin (which, among other effects, up-regulates Hoxb4).⁶⁴  The demonstration that these genes play a clear role in stem cell self-renewal makes them attractive candidates for use in gene therapy protocols. The use of transcriptional regulation, or alternative methods for the creation of conditional proteins such as the use of estrogen receptor-fusion proteins controlled by tamoxifen (reviewed in Neff and Blau⁵⁰ ), may allow for the use of HoxB4,⁶⁶  Notch, or beta-catenin for in vivo selection.

These systems rely on the use of a selectable gene that confers cellular resistance to a drug that kills the target cell population (reviewed in Sorrentino⁶⁵ ). For practical purposes, the agents used in this context have been cytotoxic drugs used in the treatment of neoplastic disease. Ideally, the system would allow for selection of hematopoietic stem cells, so that selection is long lived or possibly even permanent. This implies that the drug used for selection should be a stem cell toxin. Also, the drug should have minimal toxicity outside of the hematopoietic system. A desirable feature of the selectable gene would be a short coding sequence that could be easily incorporated into retroviral vectors. This becomes particularly important in applications where complex regulatory elements are required to ensure optimal transgene expression, such as in the case of the thalassemias and sickle cell disease.^67,68 

Drug selection systems have 2 potential clinical applications that are related but not identical: in vivo selection and marrow protection. In vivo selection leads to an increase in the proportion of genetically corrected cells, whereas marrow protection prevents the myelosuppression normally associated with cytotoxic drug administration. Thus, drug-resistance gene therapy may be useful for the treatment of nonmalignant diseases by allowing in vivo selection, and also may allow for more dose-intense and/or more dose-dense chemotherapy regimens in the treatment of malignancy by providing marrow protection. Of note, the clinical protocols for the use of drug-resistance genes in the context of genetic diseases would likely differ from the protocols used for marrow protection and dose-intensified chemotherapy in the treatment of cancer (Figure 2). Especially the use of drug-resistance genes in the context of genetic diseases requires a selection strategy with acceptable nonhematologic toxicity to minimize the overall toxicity of the gene therapy protocol.

A wide variety of genes has been shown to confer cellular resistance to cytotoxic drugs when overexpressed. Some of these have been shown to confer marrow protection in murine models of stem cell gene therapy^69,70  but have not been tested in large-animal models. We will focus here on the most thoroughly characterized selectable genes for in vivo selection, namely mutant forms of dihydrofolate reductase (DHFR), multiple drug-resistance gene 1 (MDR1), and mutant forms of methylguanine methyltransferase (MGMT).

Mutant forms of DHFR confer resistance to folate analog drugs and had been proposed for gene therapy as early as 1987. This system allows for marrow protection but not in vivo selection in mice.⁹  First attempts at using this system in a canine model were hampered by low gene-transfer rates and resulted in death of the animals without evidence of marrow protection or in vivo selection.¹⁰  As pointed out, selection should ideally occur at the level of the stem cell or at least a very primitive progenitor cell. Antifolate drugs exert their hematologic toxicity on relatively differentiated cells within the hematopoietic hierarchy.⁷¹  Sensitizing cells by pharmacologic prevention of folate rescue has allowed the ablation of true stem cells with the antifolate drug trimetrexate (TMTX) in the mouse,⁷²  and has allowed for in vivo selection of true stem cells in the murine model.⁷³  Unfortunately, these results have not translated into successful stem cell selection in a more clinically predictive large-animal model. Specifically, the selection achieved using this system in a nonhuman primate model was short lived and extrahematopoietic toxicity was clinically unacceptable.¹³  This somewhat unexpected negative result underscores the importance of testing stem cell gene therapy strategies in large-animal models before proceeding to clinical trials.

MDR1 encodes p-glycoprotein, a membrane pump that increases cellular drug efflux. Overexpression of p-glycoprotein confers resistance to a variety of xenobiotic chemotherapeutic agents including vinblastin, colchicine, doxorubicin, and paclitaxel. In vivo selection of murine hematopoietic cells transduced with MDR1 has been demonstrated.^74,75  In these early reports published more than 10 years ago, it was not studied if selection occurred at the stem cell level or the level of a more committed cell. More recently, successful selection of MDR1-transduced human NOD/SCID-repopulating progenitors has been shown.⁷⁶  In contrast, large-animal studies and clinical trials using MDR1 as a selectable marker demonstrated overall disappointing results. Among the problems observed were excessive toxicity,¹⁴  low initial marking, and moderate and/or transient selection after drug administration.^52-54,77-79  As with DHFR, studies in the mouse revealed that the drugs used in conjunction with MDR1 appeared to be toxic mainly to more committed hematopoietic cells, with only moderate effects on immature progenitors,⁸⁰  giving a potential explanation for the only transient selection observed with this system (Figure 1). Exceptions are the anthracyclines, which exert pronounced toxicity, at least at the level of clonogenic progenitors. However, extrahematopoietic (mainly cardiac) toxicity would limit the aggressive clinical use of this class of drugs for in vivo selection. The cDNA for MDR1 is more than 4 kb, making its incorporation into 2-gene vectors cumbersome. Furthermore, it has been reported that overexpression of MDR1 is associated with transformation of mouse hematopoietic stem cells.⁸¹  Others have argued that the malignant transformation was caused by the high copy number observed with one particular producer clone rather than specifically by the MDR1 transgene, and is thus a result of insertional mutagenesis and causally unrelated to MDR1.^82,83  Further studies will be necessary to evaluate the usefulness of MDR1 for in vivo selection. If sufficient preselection marking levels can be attained (by perhaps combining MDR1 with a second drug-resistance gene), the gene may be useful for marrow protection in the context of dose-escalating chemotherapy.

MGMT is a DNA repair enzyme that removes DNA adducts in the O⁶-position of guanine.⁸⁴  High levels of MGMT in gliomas are associated with resistance to BCNU and a shortened survival.⁸⁵  Likewise, methylation-mediated down-regulation of the MGMT gene promoter is associated with a favorable prognosis.^86,87  Forced overexpression of the cDNA for MGMT confers cellular drug resistance,^88,89  proving this association to be a causal one. Cellular toxicity to alkylating agents (ie, nitrosoureas such as BCNU) that create DNA lesions repaired by MGMT can be enhanced by the active site inhibitor O⁶-benzylguanine (O⁶BG),⁹⁰  a pseudosubstrate of MGMT. Mutant forms of MGMT have been developed that are insensitive to the inhibitory effect of O⁶BG.^91,92  MGMT-mediated in vivo selection at the level of the stem cell has been demonstrated in murine models using O⁶BG in combination with either BCNU or the DNA-methylating agent, temozolomide.^93,94,109  In vivo selection of MGMT-transduced human NOD/SCID-repopulating cells has also been described.^95,96  Mutant MGMT fulfills all the theoretic requirements for an in vivo selection transgene: the cDNA is short (624 bp), the drugs used are considered stem cell toxins, and the toxicity of the drug combinations of O⁶BG with either BCNU or temozolomide appears to be mainly hematopoietic. Consequently, we have recently demonstrated sustained multilineage selection of MGMT(P140K)-transduced hematopoietic cells in a clinically relevant canine model using O⁶BG and BCNU^12,55  (Figure 3). Several important points regarding this study need to be raised: (1) The prolonged rise in gene marking after drug administration implies selection at the level of the hematopoietic stem cell (Figures 1 and 3). (2) Extrahematopoietic toxicity was minimal. (3) Hematopoietic toxicity depends on pretreatment marking levels: If marking is more than 50% before drug administration, the marrow is protected by the MGMT transgene, and hematopoietic toxicity is averted. (4) Multiple relatively moderate doses of O⁶BG and BCNU allow for in vivo selection in the complete absence of hematopoietic toxicity, so that myelosuppression can be avoided if an intraindividual dose escalation strategy is used. (5) We have demonstrated MGMT-mediated in vivo selection in a model of allogeneic MHC-identical sibling stem cell transplantation (the DLA-identical littermate dog). In one animal, initial donor chimerism was increased from 50% to 100% after in vivo selection. These findings provide proof of principle for the concept of using drug-resistance gene therapy to enhance allogeneic stem cell transplantation for 2 potential purposes: early posttransplantation chemointensification and modulation of mixed chimerism.

Different diseases will likely require different levels of gene marking and correction. The following discussion is by no means exhaustive and serves to illustrate the complexities of the issue.

Data from the clinical gene therapy trials for X-SCID suggest that 0.1% to 1% engraftment is sufficient for sustained correction, at least with a follow-up of currently approximately 5 years. Data from allogeneic stem cell transplantation suggest that T-cell precursors can correct at least the T-cell deficiency in the complete absence of stem cell engraftment.^97,98 

The clinical gene therapy data for ADA deficiency suggest that rapid immune recovery may require slightly higher stem cell engraftment than SCID, maybe in the range of 5% to 10% corrected cells, although the patient with lower engraftment eventually also showed some degree of immune reconstitution. This is, however, only an estimate that may need to be corrected as more data become available. Data from allogeneic stem cell transplantation show engraftment patterns similar to the ones observed in X-SCID (ie, differential chimerism with preferential T-cell engraftment). The implication is that in fact no detectable stem cell engraftment is required for T-cell reconstitution. More data are necessary to reconcile these differences. Of note, for both ADA deficiency and SCID, it has been shown that somatic reversion of a single T-cell precursor can lead to substantial amelioration of phenotype.^99,100  In summary, it appears that for X-SCID and ADA deficiency minimal stem cell engraftment should suffice for a substantial therapeutic effect.

These diseases have been discussed; it appears that stem cell engraftment in the range of 10% to 20% will improve the associated anemia. Whether this level of engraftment also suffices to ameliorate associated problems such as organ pathology in sickle cell disease and iron overload in thalassemia will need to be determined.

Data from a murine gene therapy model suggest that about 30% of corrected cells may be required to protect the host from infectious complications.¹⁰¹  Data for allogeneic chimerism after stem cell transplantation are available. However, most of the CGD patients reported in the literature have a relatively high level of donor-derived chimerism, which makes it difficult to draw conclusions regarding a minimal level of chimerism that can be considered beneficial in this disease. In a recent clinical study in CGD patients, it appears that a level of 10% gene-corrected neutrophils leads to a substantial improvement in the clinical course of CGD.¹⁰² 

In summary, it appears that for the different SCID subtypes, very low chimerism levels are likely to have a beneficial effect, whereas for red cell diseases and phagocyte defects, a higher level of correction of at least 10% to 20% may be required for a therapeutic effect. Further studies will be required to provide definitive answers.

If stem cell gene therapy is to become used routinely, a more complete understanding of gene-modified hematopoiesis is desirable. Data available to date are limited, but the development of novel technologies such as ligation-mediated polymerase chain reaction (PCR) to analyze individual clones contributing to hematopoiesis has started to shed light on this important aspect of gene therapy. Three factors influence the size of the gene-marked stem cell pool and the contribution of genetically modified cells to this pool. (1) Efficiency of gene marking during the ex vivo protocol: if gene marking during transduction is insufficient, the in vivo proportion of gene-modified stem cells will be low. Unless there is a survival advantage for gene-modified cells, the proportion of circulating transduced cells will be low. Advances in ex vivo gene transfer have recently been reviewed and will not be discussed further here.^103,104  (2) Competition with endogenous stem cells: cells cultured during transduction (ie, the graft) upon reinfusion compete with a numerically advantaged nontransduced pool of endogenous stem cells. The amount of contribution from the transduced cells depends on the number and functionality of stem cells in the graft after ex vivo transduction. It also depends on the degree to which endogenous competition was reduced by pretransplantation cytoreductive conditioning. This conditioning needs to be chosen to allow for sufficient engraftment but avoid toxicity. (3) In vivo selection: a relatively small number of transduced cells can contribute to a large proportion of hematopoiesis if there is an efficient in vivo selection process in place. However, the long-term effects on hematopoiesis after selection pressure applied on a relatively small number of gene-marked stem cells require careful examination in appropriate large-animal models and ongoing human clinical trials.

So what, if any, conditioning will be necessary and sufficient to ensure polyclonal engraftment with the capacity to ensure normal hematopoiesis for the patient's natural life span? The exact answer is unknown, but available data suggest that the level of conditioning will depend on several different variables. The fully myeloablative doses of total body irradiation (TBI), as used in allogeneic stem cell transplantation, are limited by toxicity. Data derived from mouse models have demonstrated that megadoses of transplanted cells can achieve engraftment in the range of 10% to 40% without any cytotoxic conditioning in a syngeneic setting.¹⁰⁵  The applicability of this strategy is complicated by the fact that the culture protocols used for gene transfer—at least with retroviral vectors—reduce the engraftment potential of stem cells.¹⁰⁶  Studies in mice have shown that minimal levels of TBI-based conditioning could provide a substantial engraftment advantage to the graft.¹⁰⁷  As an alternative to TBI, busulfan, a known stem cell toxin, could be used as shown in the recent ADA gene therapy trial.¹  Another alternative could be the use of melphalan. More data from clinical trials and large-animal studies will be needed to minimize conditioning and, at the same time, maintain in vivo gene marking in the therapeutic range.

In vivo selection strategies now allow for virtually 100% of gene marking in selected cell lineages, which makes rapid translation to clinical application tantalizing. An important question for future clinical trials is: What is the absolute number of gene-marked stem cells/gene-marked clones that must engraft to allow for safe in vivo selection while still maintaining a functional hematopoietic system for the life of the individual? The answer to this question is complex and depends on many variables including: (1) the age of the recipient (ie, how long will the graft need to function?); (2) the age of the donor (in an allogeneic setting); (3) the disease treated—of particular importance is the question whether only the lymphoid arm of the immune system needs to be reconstituted (as in SCID) or if all lineages will be derived from the graft; and (4) the selection strategy. These issues are not only academic but are in fact vital to the outcome of treatment as is illustrated by the recent demonstration of the failure of X-SCID gene therapy in older patients.¹⁰⁸ 

The concerns with applying proliferative pressure on a very small fraction of transduced cells in vivo are 3-fold: (1) the early loss of the transduced cell population and engraftment failure with failure of in vivo selection have been described in a murine study of MGMT-mediated selection when gene-modified cells were transplanted without prior cytoreduction⁹³ ; (2) the development of oligoclonality potential for leukemic transformation; and (3) clonal exhaustion after in vivo selection. The latter 2 complications could potentially result from placing a great proliferative stress on a very limited number of stem cells. The concern has been raised that temozolomide-mediated selection leads to loss of reconstituting ability of the marrow in a mouse model. This effect could be mitigated by overexpression of HOXB4.¹⁰⁹  How these data translate to large animals or humans is not known. In the canine model, no marrow failure was observed so far after BCNU or temozolomide-mediated selection of MGMT-transduced cells with a follow-up of 3 years after gene therapy.^12,55  Murine¹¹⁰  and feline¹¹¹  models have demonstrated that one or a few stem cells can maintain hematopoiesis for extended time periods, approaching the natural life span of the animal without obvious adverse effects. A mathematic model has been proposed that predicts that in a situation where few (< 20) stem cells engraft, there is considerable variability in the contribution of individual clones, with potential loss of some clones.¹¹²  Data from SCID patients receiving allogeneic grafts indicate that there was a trend toward oligoclonality in the T-cell compartment more than 10 years after transplantation.¹¹³  In a clinical gene therapy study, “benign monoclonality” has been described,¹¹⁴  with one clone dominating the T-cell compartment up to 8 years after gene therapy. The long-term clinical significance of these findings is not entirely understood, but the available data suggest that a limited number of clones can sustain normal hematopoiesis for extended time periods.

Whether reconstitution after no or minimal cytoreduction consistently leads to oligoclonality in a clinical gene therapy setting, and whether this oligoclonality in turn yields undesirable outcomes, needs to be studied more systematically. These studies are currently under way and will depend on careful longitudinal integration site analysis in large-animal models and clinical trials. Data from our laboratory show polyclonal hematopoiesis after temozolomide-mediated in vivo selection of MGMT-transduced cells in one dog from a baseline of approximately 1% in granulocytes to currently more than 80%, with a follow-up of up to 3 years (H.-P.K., unpublished data, September 15, 2005).⁵⁵  An important parameter that merits further study is the proportional contribution of individual gene-marked clones before and after aggressive drug selection. We have in one dog observed the emergence of clones that are fairly quiescent early after engraftment but selected for after repeated exposures to cytotoxic drugs. It is possible that the stem cell population after in vivo selection looks indeed very different from the stem cells that initially engrafted. Obviously, larger cohorts and longer follow-up will be needed. Likewise, the question of clonal exhaustion is important and can ultimately be answered only by long-term follow-up of large-animal models and human patients.

The X-SCID trials carried out in Paris and London have yielded therapeutic benefit in the majority of patients. However, these trials have raised concerns about therapy-related leukemia in 3 patients. Thus, the most pressing issue for stem cell gene therapy now is no longer efficacy, but therapy-related toxicity.

Genomic insertion of retroviruses in mouse bone marrow has been associated with leukemogenesis¹¹⁵  and, more recently, with clonal dominance in the absence of overt leukemia.¹¹⁶  In these studies, transgenes were used that are believed to be inert with respect to genomic stability, cell growth, survival, and proliferation. These data suggest that, at least in mice, vector insertion can skew or transform hematopoiesis. In nonhuman primates, lymphomas were reported in monkeys that received cells transduced with a contaminated vector preparation containing replication competent helper virus. Recently, a case of granulocytic sarcoma has been reported in a monkey receiving cells transduced with a helper virus-free DHFR-encoding vector. This case, however, has so far been the only large animal with a gene transfer-related tumor of more than 80 dogs and monkeys with marking levels more than 1%.¹¹⁷ 

In 2 of the X-SCID patients developing leukemia, dysregulation of LMO2 was demonstrated. Data from a murine model further suggest that lmo2 and the common gamma chain may cooperate to produce leukemogenesis.¹¹⁸  Recent data in the mouse also suggest that loss of tumor suppressor function may synergize with the development of tumors in a SCID background.¹¹⁹  Unfortunately, the exact causal factors are not entirely clear. It is particularly intriguing that no cases of leukemia have been observed in the X-SCID trial conducted by Thrasher et al¹⁰⁸  in London. It is hoped that additional follow-up will provide more definitive answers.

No case of leukemia has been reported in ADA deficiency. This is not surprising since the genetic correction does not provide a stimulatory signal but, rather, removes the toxic effects of aberrant metabolites. The risk for leukemogenesis should thus be low in this disease.

There is a theoretic concern that the use of cytotoxic drugs could possibly increase the risk of leukemogenesis in the setting of drug-resistance gene therapy. However, so far we have not observed any development of monoclonality or leukemia in our dog and nonhuman primate studies, even though some of the animals received multiple courses of chemotherapy.⁵⁵  Studies of drug-resistance gene therapy have now also been initiated in patients with malignant diseases, and careful and long-term follow-up analyses of both large animals and patients will be extremely important for future progress in the field.

Much progress has been made in the field of hematopoietic stem cell gene therapy in general and in the field of in vivo selection in particular. The X-SCID and ADA-deficiency trials have proved the validity of the concept of in vivo selection for gene therapy. The last few years have witnessed successful in vivo selection using dimerizer-based approaches and drug-resistant strategies in clinically relevant large-animal models. Gene marking attained in some of these studies will likely be therapeutic for most genetic diseases of the hematopoietic stem cell. Marrow protection has been demonstrated in clinically relevant large-animal models. This opens up the door for clinical trials of marrow protection drug-resistance gene therapy for diseases such as malignant glioma, where curative treatment is not yet available. The transfer of drug-resistant genes may also be useful to enhance allogeneic stem cell transplantation, especially after nonmyeloablative conditioning. Concerns remain about the long-term effects of gene therapy, in particular about the risk of leukemogenesis.

In our opinion, the field has reached a point where it appears warranted to carefully test strategies in the clinic and compare the results to the more conventional treatment options, in particular to those of allogeneic stem cell transplantation. In the setting of nonmalignant diseases of the bone marrow, many patients do not have a matched sibling donor available, and the results of haploidentical stem cell transplantation and matched unrelated donor transplantation for genetic diseases vary and demonstrate significant morbidity and mortality in a number of reports.^{38,97,120,121}  There appears to be agreement between the bone marrow transplantation community and gene therapists that gene therapy trials should proceed, albeit with caution. In the setting of solid tumors, therapeutic options for many patients are so limited that gene therapy for the purpose of marrow protection appears to be a viable option. The recent advances of in vivo selection have set the stage for efficacious gene therapy trials. Success will now depend on meticulous study design and careful analysis of the results.

The authors would like to thank Peter Kurre, Grant Trobridge, and Kathrin Bernt for critically reviewing the paper. We would like to thank Bonnie Larson and Helen Crawford for assistance with the preparation of the paper.