Main

Gene therapy offers tremendous promise for the future of cancer treatment. This technology, more than any other, takes direct advantage of our new understanding of cancer at the molecular level and has been exploited to develop new strategies for killing cells selectively or arresting their growth. However, despite the promise of safe and rational treatment, many researchers have serious doubts as to whether this is a viable approach. Is this view accurate? Or does gene therapy have a future in mainstream clinical oncology?

The field of cancer gene therapy embraces a range of ideas and technologies — from direct attack on tumour cells to harnessing the immune response to tumour antigens (Fig. 1). Here, we will restrict our discussion to direct attack on tumour cells, which requires an understanding of the intracellular signalling pathways that have gone awry in tumour cells; the success of immunotherapy depends on understanding the complex interactions between tumour cells and the immune system — a distinct intellectual challenge that is beyond the scope of this review. There are three types of weapon in this direct offensive, and we will consider each of these in turn.

Figure 1: Cancer gene therapy and immunotherapy trials currently listed as open by the US Recombinant Advisory Committee.
figure 1

Over half of all gene-therapy-based protocols in the United States (113 currently open) are aimed at boosting the immune response to tumour antigens. Trials in melanoma alone account for 54% of immunotherapy trials. Delivery of the tumour-suppressor gene TP53 accounts for the next largest group, followed by suicide gene delivery, in which viral vectors deliver enzymes that activate prodrugs to toxic products that kill tumour cells and their neighbours. Most of these use herpes simplex virus thymidine kinase (HSV-tk), which activates the prodrug ganciclovir. Chemoprotection is an indirect approach in which bone marrow cells are infected with viruses that protect them from the toxic effects of chemotherapy, by expressing drug-resistance genes.

Full size image

Tumour suppressors and oncogenes

In the most direct application of cancer gene therapy, tumour-suppressor genes are expressed in cancer cells in which these genes are defective, resulting in cell death or growth arrest (Table 1; Fig. 2). This concept is based on two assumptions, neither of which should be taken for granted. The first is that restoration of a single genetic defect will be effective in inhibiting tumour cells that have many additional defects. Perhaps surprisingly, for each of the principal known tumour suppressors — APC 1, RB 2,3, INK4A (Refs 4,5), PTEN 6,7,8,9, ARF10 and TP53 (Refs 11,12), — it turns out to be true: expression of each of these genes in tumour cells in vitro causes an acute change in cell physiology and gene expression, resulting in cell-cycle arrest or death (Table 1). In addition to validating the concept of this form of gene therapy, these experiments clearly illustrate the selective advantage of losing tumour-suppressor gene expression in tumour development.

Table 1 Effects of expressing tumour suppressors in tissue culture and in mouse models
Full size table
Figure 2: Cancer gene therapy by delivery of tumour-suppressor genes or inhibition of oncogene expression.
figure 2

a | Tumour-suppressor gene (TSG) delivery. Vectors encoding the tumour suppressor of choice are assumed to infect normal cells and tumour cells. In tumour cells they induce either growth arrest or apoptosis, whereas in normal cells they are assumed not to have any detrimental effects. Some tumour suppressors might also exert unexpected bystander effects. For example, p53 blocks angiogenesis by downregulating the production of vascular endothelial growth factor (VEGF) and by upregulating two anti-angigogenic molecules, thrombospondin and insulin-like growth factor 1 binding protein (IGF1BP). b | Delivery of agents that block oncogene (Onc) expression. These include genes that encode antisense oligonucleotides, which block oncogene expression, and ribozymes, which cleave oncogene transcripts. Again, they are expected to have no detrimental effects on normal cells, which don't express oncogenes. By contrast, they should cause cancer cells to arrest or undergo apoptosis. In some cases, they also sensitize radio- or chemo-resistant tumour cells to radiotherapy or chemotherapy. No bystander effects have been reported for anti-oncogenic gene-therapy agents.

Full size image

For genes that primarily affect genome integrity, such as BRCA1 and BRCA2 , this assumption is less likely to be valid. These genes have their selective effects during initiation of the disease, and would be expected to have little, if any, effect at later stages. But surprisingly, retroviral expression of BRCA1 causes growth arrest or apoptosis in breast cancer and ovarian cancer cells, even though these cells already express the wild-type BRCA1 gene13,14. The mechanism of these effects is not understood, and, unfortunately, has not borne fruit in the clinic: Phase I and II clinical trials of a BRCA1-expressing vector in women with ovarian cancer yielded no clinical responses, although this might have been because the vector was not stably expressed15.

Some tumour-suppressor genes affect genome stability as well as cancer-cell survival and growth. Loss of TP53 increases the rates of spontaneous cancers tremendously16. Likewise, APC, which is mutated in most colorectal cancers, might be involved in maintaining genome stability17,18. Does this have any bearing on their effects when expressed in tumour cells? Expression of either TP53 or APC in tumour cells causes rapid growth arrest or apoptosis. Their potentially therapeutic value seems to depend on these acute effects rather than on their status as guardians of genome integrity.

A second assumption about the mechanism of direct attack by tumour suppressors is that collateral delivery of these genes to normal tissue will have little effect because these genes are expressed in normal cells anyway and are appropriately regulated in these cells. This issue has not been addressed rigorously in model systems. One reason for this is that it is difficult to predict which normal tissue would be exposed to the transgene. The normal counterpart of the tumour itself would seem a logical choice for experimental investigation, but the toxic effects of the gene are more likely to be seen in cells that are exposed to the highest levels of the vector, such as the liver or vascular endothelial cells. Nevertheless, delivery of TP53 to normal bronchial epithelial cells showed no effects on cell growth, with a 2–3 log THERAPEUTIC WINDOW relative to tumour cells19.

These results go some way towards validating the notion that delivery of a tumour-suppressor gene could have therapeutic value, but the effects were expected to be cell autonomous, with little hope of any effect on surrounding, uninfected tumour cells. This meant that every cell in a tumour would need to be infected — an enormous technical hurdle, particularly for disseminated cancers. But surprisingly, such an effect (defined as a 'bystander effect') might well account for some of the efficacy of an adenovirus that expresses TP53 in solid tumours. One interesting explanation is that p53 is anti-angiogenic: it downregulates expression of vascular endothelial growth factor (VEGF)20,21, and upregulates expression of thrombospondin — a potent inhibitor of angiogenesis22. Forced expression of TP53 could therefore lead to toxic effects on uninfected neighbours by starving them of oxygen and nutrients. p53 regulates other gene products that are capable of a bystander effect. One of these is insulin-like growth factor 1 binding protein (IGF1BP)23, which neutralizes the anti-apoptotic effects of IGF-1 and could kill uninfected tumour cells that depend on IGF-1 for survival. Yet another potential means of invoking a bystander effect is local inflammation triggered by proteins in the viral vector. These might stimulate production of cytokines, such as tumour necrosis factor, which might be able to kill uninfected tumour cells selectively24. These interpretations are hypothetical, but they illustrate the tremendous biological complexity encountered in vivo and show that gene-therapy agents can have profound biological effects that cannot be explained on the basis of their expected direct mechanism of action alone.

How does this approach fare in the clinic? The first clinical effects of delivering a tumour-suppressor gene — TP53 — were published in 1996 (Ref 11): RETROVIRAL VECTORS that express TP53 from the actin promoter were injected directly into non-small-cell lung tumours. Clear signs of apoptosis were detected in injected tumours, and three out of nine patients showed regression of these tumours, as well as evidence of a bystander effect. More recent trials have used ADENOVIRAL VECTORS to deliver TP53 (Adp53), as these are easier to grow to high titres, and infect cells regardless of whether they are in the cell cycle. Adp53 is well tolerated and expressed in most patients. Furthermore, antitumour activity has been observed in a subset of patients who were treated for non-small-cell lung cancer25 or squamous cell carcinoma of the head and neck26. Adp53 is now being tested in a Phase II and III clinical trial, as well as in combination with chemotherapeutic agents27 or radiation.

If reactivating tumour suppressors is effective, what about blocking hyperactive oncogenes? Viral vectors have been used to deliver ANTISENSE OLIGONUCLEOTIDES or RIBOZYMES to block oncogene expression. The effectiveness of these approaches depends on similar assumptions to those of tumour-suppressor gene therapy, and when tested these assumptions have been fulfilled (Table 2): tumour cells with many genetic defects still depend on oncogene expression for growth or survival. For example, tumours that are driven by oncogenic HRAS are completely destroyed when HRAS is eliminated, using either genetic or pharmacological approaches28,29, and chronic myelogenous leukaemias undergo dramatic remissions when the oncoprotein p210 BCR–ABL is blocked by the ABL inhibitor STI-571 (Gleevec), despite the presence of several genetic alterations in these cells30. Furthermore, ribozymes that target mutant RAS selectively should have no effect on normal cells. However, the viability of this approach might be hindered by lack of a bystander effect when tumour cells are treated in this way, and by the increasing number of alternative approaches to oncogene-directed therapy. Recent successes with small molecules, particularly inhibitors of kinases in the RAS pathway31, p210 BCR–ABL and the epidermal growth factor receptor (EGFR)32, indicate that more attention will be paid to this technology platform for oncogene-based therapy. By contrast, it is more difficult to develop small molecules that reactivate mutated tumour suppressors.

Table 2 Effects of blocking oncogenes in tissue culture and in mouse models
Full size table

Exploiting bystander effects

The second broad approach to direct targeting of cancer cells with gene therapy is the delivery of suicide genes to cancer cells (Fig. 3). These are enzyme-encoding genes that, once expressed, allow the cancer cell to metabolize a harmless prodrug, administered separately, into a potent cytotoxin that can diffuse into neighbouring cells, so creating a bystander effect. Several enzyme–prodrug combinations are being evaluated (Table 3), but herpes simplex virus thymidine kinase (HSV-tk) has been most widely used and has progressed farthest into the clinic: many Phase I/II trials of this approach are ongoing in the United States, and one Phase III trial has been completed. HSV-tk converts ganciclovir to the phosphorylated form that becomes incorporated into DNA, thereby blocking DNA synthesis33.

Figure 3: Suicide gene delivery.
figure 3

The vector delivers a gene that encodes a prodrug-converting enzyme, such as herpes simplex virus thymidine kinase (HSV-tk), to tumour and normal cells alike. Local delivery of either the prodrug (in this case, ganciclovir) or the vector to the tumour provides specificity. The prodrug is converted to the active, cytotoxic metabolite in the tumour cell, and diffusion to neighbouring cells confers a potent bystander effect.

Full size image
Table 3 Enzyme-prodrug combinations for suicide gene therapy*
Full size table

Selective expression of suicide genes could be achieved in several ways. Local injection of non-selective agents is an obvious solution. Early studies used retroviral vectors to deliver suicide genes to glioblastoma multiforme — a lethal form of brain tumour; these viruses are produced locally from injected PRODUCER CELLS. Retroviruses integrate only into dividing cells, so tumour cells growing in the brain express the transgene selectively. Local infusion or injection of nonspecific vectors (such as adenoviral vectors) directly into tumours also affords a degree of selectivity34.

However, a more general strategy to kill tumour cells with suicide genes exploits tumour-specific expression elements. Many of these have been tested in cell systems and in animal models, and a fairly comprehensive repertoire of selective expression elements is now available. Table 4 summarizes these elements and the principles that underlie their selectivity.

Table 4 Regulated gene expression in tumour cells
Full size table

Bystander effects can also be achieved in other ways. Several strategies to suppress angiogenesis have been tested in animal models. For example, adenoviral expression of a soluble form of VEGF receptor was recently shown to suppress tumour growth in mouse models35. Angiostatin and endostatin expressed from plasmid DNA complexed with liposomes inhibited growth of breast cancer in mice36. An adenovirus expressing secretable endostatin showed activity in vitro and in mouse models37, and a combination of viruses expressing three anti-angiogenic proteins led to complete tumour rejection in mouse models38. Another approach is to inhibit matrix metalloproteinases: the metalloproteinase inhibitor TIMP-2 expressed in an adenoviral vector caused significant reduction in metastatic cell growth39. These examples reflect the application of many years of discovery research in tumour biology to address clinical issues, and illustrate the translational opportunities that this field presents.

Clinical studies with suicide gene therapy have shown that these agents are safe, but are not sufficiently active. A Phase III clinical trial of retrovirus-encoded HSV-tk showed no patient benefit40. Local injection into brain tumours of adenoviral vectors that express HSV-tk looks slightly more promising, with some early signs of survival benefit in a small number of patients41. Nevertheless, there is an obvious need to improve efficacy. Efforts are underway to develop this concept further, using different vectors and suicide genes that might produce stronger bystander effects, combinations of suicide genes that act synergistically42 and by pharmacological manipulation. For example, GAP JUNCTIONS can be upregulated by lovastatin and other compounds43, increasing the magnitude of the bystander effect. Another problem is that many tissue-specific promoters are comparatively weak. Sakai and co-workers44 have developed an innovative approach to solve this problem, by using a weak, but specific, promoter to drive expression of Cre recombinase, and a strong Cre-responsive promoter to drive expression of the therapeutic transgene.

These creative solutions will certainly improve the potency of suicide gene therapy, but will they be sufficient? Poor distribution of vectors within solid tumours and low infectivity of tumour cells might be the main barriers to success in this area. The next generation of protocols and vectors will need to include many innovations to improve selectivity and gene expression, but will also need to address these issues to realize their full clinical potential.

Replication-competent viruses

Exploitation of bystander effects represents one solution to the technical hurdle of infecting all the cells in a tumour. A second approach uses the ability of viruses to spread from their site of inoculation and infect neighbouring cells. In this approach, cells are killed as a consequence of virus infection, as they become factories for producing new infectious virus particles (Fig. 4). The success of this approach depends on our ability to engineer or select viruses that replicate specifically in tumour cells, and the ability of these viruses to infect tumour cells efficiently and to spread through the tumour.

Figure 4: Conditionally replicating viruses.
figure 4

a | Mechanism of action. The viruses infect both normal and tumour cells, but can only replicate in tumour cells. The progeny then go on to kill surrounding tumour cells. b | Replication of a conditionally replicating virus, ONYX-015, in a cancer cell from a patient with head and neck cancer during Phase II clinical testing. 109 infectious particles were injected over a 5-day period. After 8 days, biopsy was performed and analysed by electron microscopy. The inset on the left panel is magnified on the right. Clearly, this cell is doomed to die: presumably the new virus particles it produces will infect its neighbours.

Full size image

The possibility of using viruses to kill cancer cells selectively was proposed and tested in 1956 (Ref. 45), based on the observation that certain viruses grow efficiently in cancer cells, and on anecdotal reports of spontaneous tumour regressions in patients who harbour viral infections. Attempts to treat cervical cancer by injecting an adenovirus date back to 1956 (Ref. 45), and the mumps virus was used to treat cancer in 1974 (Ref. 45). However, lack of molecular characterization, of either the viruses or the tumours, prevented logical development of this approach. More recently, several viruses have been created that replicate selectively in cancer cells45. Indeed, this marriage of molecular virology and cancer biology provides a tremendous range of opportunities for creative engineering of selective agents. In retrospect, the field of tumour-virus biology sought to use viral oncoproteins to understand the molecular basis of cancer. Landmark discoveries in human cancer genetics, such as the functions of RB and p53, revealed the significance of the classical tumour-virus proteins — E1A and SV40 large T-antigen — in viral replication. Ironically, tumour viruses themselves are now at the forefront of cancer therapy.

Targeting dividing cells with herpesviruses. The first engineered therapeutic virus to enter clinical testing was a defective herpesvirus mutant, G207, that lacks the gene encoding ribonucleotide reductase46 and can therefore replicate efficiently only in dividing cells. G207 also lacks the virulence gene ICP34.5 , to improve safety. This virus was initially tested in mouse models by direct injection into tumours in the brain. On the basis of encouraging data from these preclinical experiments, G207 advanced to a Phase I clinical trial for malignant glial-cell tumours. G207 administered directly into these tumours was safe and well-tolerated, and early signs of clinical effects have been reported47. In parallel, another ICP34.5 mutant, HSV-1716, has undergone successful preclinical testing for metastatic melanoma48 and has recently been injected directly into subcutaneous lesions of metastatic melanoma49 and into recurrent malignant gliomas50 in Phase I clinical trials. So far, these agents seem to be safe and well tolerated.

Targeting a defective p53 pathway. An attenuated adenovirus — mutant dl1520 (Ref. 51) or ONYX-015 — relies on a coincidental biological property of adenoviruses and tumour cells: both need to block p53 function to replicate efficiently (Fig. 5)52. Adenoviruses eliminate p53 by producing the EARLY REGION PROTEIN E1B 55K. This protein binds p53 and, with help from another viral protein, E4 orf6, targets it for destruction. ONYX-015 lacks E1B 55K and therefore cannot destroy p53. In normal cells, p53 blocks replication, but in cancer cells that lack p53, ONYX-015 can replicate — at least in theory.

Figure 5: Alterations in the p53 pathway in adenovirus-infected cells and tumours.
figure 5

a | Adenovirus E1A binds the tumour suppressor RB, thereby preventing it from repressing E2F. E2F is then free to activate host target genes that are involved in S-phase of the cell cycle, and the adenovirus E2 region. One of the host genes that E2F activates is ARF, an inhibitor of the oncoprotein MDM2. MDM2 is an E3 ubiquitin ligase that targets the tumour suppressor p53 for destruction. Adenovirus E1B 55K and E4 ORF6 cooperate to target p53 for destruction, because in the presence of p53 viral replication cannot occur. b | In tumours, there are several mechanisms by which the RB and p53 pathways can be inactivated. RB is mutated in approximately 25% of tumours, but can also be functionally inactivated by overactivation of D-type cyclins, which phosphorylate and inactivate RB. This can occur by loss of the CDK inhibitor INK4A, increased expression of cyclin D1 or activation of CDK4. In addition, the human papillomavirus (HPV) E7 protein, similar to adenovirus E1A, can bind to and inhibit RB. p53 itself is mutated or deleted in around 60% of tumours, and can be targeted for destruction by HPV E6 protein. It can also be functionally inactivated by amplification of MDM2, or inactivation of ARF by hypermethylation, mutation or deletion.

Full size image

The specificity of ONYX-015 for p53-deficient cells remains controversial53. Confusion has arisen for two reasons: first, ONYX-015 can replicate in some tumour cells that retain wild-type p53, apparently contradicting the premise of its specificity54,55,56,57. However, most tumour cells that retain wild-type p53 have other defects in the p53 pathway (Fig. 5). Assessment of the functional status of p53 based solely on the presence or absence of mutant p53 is therefore misleading. In normal cells, MDM2, an E3 UBIQUITIN LIGASE that targets p53 for destruction, keeps the levels of p53 low. Loss of ARF, a protein that inhibits MDM2, is a common mechanism of suppressing p53. In tumour cells that lack ARF, p53 is not activated during ONYX-015 infection, and therefore cannot block replication. Expression of ARF in such cells induces high levels of p53 that inhibits replication of ONYX-015, but not wild-type adenovirus replication58.

The second source of confusion derives from the fact that E1B 55K has other functions that are distinct from blocking p53. These functions affect selective export and translation of viral mRNAs during infection (reviewed in Ref. 59). Some tumour cells (such as C33a cervical carcinoma52) complement these other functions of E1B 55K perfectly, and others (such as U20S osteosarcoma56) do not. Among tumours with a defective p53 pathway, then, variations in ONYX-015 replication are determined by the degree to which these other functions of E1B 55K are complemented59.

While these issues regarding the selectivity of ONYX-015 were being investigated, clinical development of ONYX-015 proceeded. Treatment of patients with head and neck cancer — who had failed surgery, radiation and chemotherapy — with ONYX-015 alone revealed that this agent is safe and well tolerated, but clinical responses were modest60. Responses might have been limited by several factors, but poor distribution of injected virus seems likely to be the most significant limitation. Tumours of this type, particularly those from patients treated by radiation, contain high proportions of stromal and fibrotic tissue that are likely to limit efficient spread of virus through the tumour.

Higher response rates were seen in patients with recurrent head and neck cancer treated with ONYX-015 in combination with 5-fluorouracil and cisplatin61, and a Phase III trial is now in progress. In these trials, evidence has accumulated that a bystander effect contributes to the clinical outcome. Anecdotally, tumours injected with ONYX-015 show widespread necrosis in a timeframe that cannot be accounted for by viral replication alone. Virus replication triggers an acute and local inflammatory reaction that involves production of tumour necrosis factor-α and other cytokines. These agents might kill nearby tumour cells with some selectivity, augmented by the presence of chemotherapeutic agents. Ironically, this bystander effect might be more effective in tumours that retain wild-type p53. The overall clinical outcome of this treatment therefore depends on complicated interactions between the host immune system, the tumour cell and its neighbours, and the replicating virus.

Targeting the RB pathway. Adenovirus E1A proteins bind and neutralize RB and its relatives. This releases the transcription factor E2F, which is repressed by RB, with consequences for both the virus and its host. The cellular function of E2F is to drive the expression of genes involved in S-phase and chromatin synthesis, but adenovirus has also hijacked E2F to activate expression of the E2 region of the viral genome. A virus expressing a mutant form of E1A that fails to bind RB should therefore be restricted to cells that lack functional RB. Most, if not all, tumour cells are defective in the RB pathway and, as a result, should be permissive to E1A-defective viruses. Such viruses have been described62,63,64 and tested in preclinical models. The potency of one of these viruses can be increased further by engineering it to overexpress the adenovirus death protein (ADP): this increases cell lysis and virus release, making the virus more active in vivo than the E1A mutant alone65. E1A has many additional functions: it binds to p300 and it activates the E4 region of the viral genome66. Viruses that exploit these features are under development64 (L. Johnson, F. M. and A. Fattaey, unpublished observations).

Tissue or tumour-specific regulation of viral replication. Viruses that replicate in cancer cells or tissues selectively based on differential regulation of viral gene expression have also been engineered and tested in the clinic. For example, Calydon CN706 (Ref. 67) is an adenovirus in which the prostate-specific antigen (PSA) promoter drives E1A. In CN787, the rat probasin promoter (another prostate-specific regulatory element) drives E1A and the PSA-regulatory element drives E1B (Ref. 68). These viruses do not discriminate between normal and cancerous prostate because both PSA and probasin are produced by normal prostate tissue, but in this clinical setting this is not thought to be a significant issue. Both viruses are currently undergoing clinical investigation.

Iggo and co-workers69 have made a virus that depends on abnormal signalling from the β-catenin/TCF-4 pathway to replicate selectively in colon cancer cells. Virtually all colorectal cancer cells, and many other types of cancers, have a defect in this pathway. Loss of the tumour suppressor APC in up to 80% of colorectal cancers allows accumulation of high levels of β-catenin, because one function of APC is to target β-catenin for destruction. In other tumours, mutations occur in β-catenin itself that make it resistant to degradation. High levels of β-catenin bind to the transcription factor TCF-4, and binding de-represses TCF-responsive genes, such as cyclin D1, c-MYC and matrilysin. A virus containing a TCF-responsive element therefore grows selectively in cells in which this pathway is deregulated. Using a similar strategy, The MUC1 promoter has been used to drive E1A, so supporting virus replication selectively in breast cancer cells in which MUC1 is aberrantly expressed70.

In many ways, this innovative technology is an extension of suicide gene therapy, and again illustrates that cancer gene therapy provides a field day for molecular biologists who are interested in applying basic principles of cancer biology to clinical problems. Unfortunately, commercial and clinical development of the viruses and vectors using tissue-specific promoters might be complicated by a dominating patent on the use of tissue-specific promoters for gene therapy (Novartis US patent 5,998,205, Hallenbeck, Chang and Chiang71).

Second-generation replicating viruses might combine these two strategies. For example, a virus similar to ONYX-015 that expresses HSV-tk seems to be more potent than either approach on its own72,73. Likewise, a replication-competent herpesvirus vector expressing HSV-tk in combination with a cytochrome P450 enzyme that converts the prodrug cyclophosphamide to its active phosphoramide derivitive showed potent antitumour activity in mouse models74.

Many viruses have a natural tropism for tumour cells, for one reason or another. RNA viruses such as Newcastle disease virus and vesicular stomatitis virus replicate in tumour cells because these cells fail to mount a protective interferon response. Parvoviruses, such as adenovirus-associated virus (AAV), replicate selectively in tumour cells because in normal cells, single-stranded AAV genomes trigger a p53-dependent DNA-damage response that blocks cells in G2 and prevents replication75 — and for other reasons that are not fully understood. Another type of parvovirus, H1, has entered clinical testing based on its tumour selectivity. The merits of these approaches, and the status of clinical testing, have been discussed elsewhere76.

Challenges

Clinical data from gene-therapy agents and replication-competent viruses administered locally indicate that some of these approaches will have clinical value, and indeed a significant number of patients could benefit from local or regional treatment. These include patients suffering from head and neck cancer, glioblastoma, ovarian cancer and metastatic colorectal cancer, which usually localizes to the liver. Two main challenges need to be overcome to convert cancer gene therapy into a strategy that is effective against disseminated metastatic disease. First, the vector or virus needs to reach the tumour efficiently. Second, it needs to avoid neutralization by the immune system. In contrast to more traditional approaches to cancer therapy, based on small-molecule inhibitors of enzyme targets, the success of this field lies in the hands of creative molecular biologists, rather than medicinal chemists and pharmacologists. These challenges might not have captured the imagination of basic scientists, but they are of paramount importance to successful evolution of the field and, again, present opportunities for state-of-the-art translational research.

Biodistribution. The magnitude of these problems is difficult to estimate accurately, but it is already clear that several issues need to be addressed. Agents that are administered systemically can interact with many cell types in the blood and with endothelial cells, and are cleared rapidly from the bloodstream by the liver77. If they survive these obstacles, they must leak from blood vessels into tumours, and spread within the heterogeneous mass of the tumour. These issues sound formidable, but, of course, they have been faced many times before, as all forms of systemic cancer therapies face similar problems. A principal distinction is that for these biological agents, biologists might have the answers. For example, the adenovirus receptor, CAR78, is an adhesion protein79 that is expressed at high levels in liver, kidney, brain, heart, pancreas, endothelium, colon and prostate, but not in peripheral lymphocytes, spleen, skeletal muscle and fibroblasts80. Some advanced cancer cell lines lose CAR expression81,82. This information helps determine which normal cells are likely to be infected during systemic administration, and which tumour cells are susceptible. Tumour types that fail to express CAR should probably be excluded from trials using the current generation of adenovirus vectors.

Of greater significance for the long-term development of these vectors are intense efforts to re-direct adenoviruses away from CAR in favour of tumour cells. This has been facilitated by the rapid solution of the structure of CAR that is bound to adenovirus fibre protein83. One approach to re-targeting adenoviruses is to engineer ligands into viral proteins, and to use tumour cell-surface proteins (the fibroblast growth-factor receptor, for example84) as new receptors. Another is to increase binding of virus particles to integrins or simply to enhance CAR-independent nonspecific attachment85. Another creative approach uses an adenovirus serotype that does not use CAR as its receptor, such as serotype 35 (Ref. 86). Binding of CAR could also be masked by polyethylene glycol treatment of virus particles77,87. Alternatively, CAR expression might be regulated in tumour cells by pharmacological intervention (M. Anders, F. M. and W.M. Korn, unpublished observations). The full potential of gene therapy using adenoviruses might ultimately depend on one of these new strategies to infect tumour cells selectively, or prevent infection of normal cells that express relatively high levels of receptor. It is therefore likely that distribution of adenoviruses between normal cells and tumours can be changed dramatically by rational and creative molecular approaches, many of which are currently under active investigation.

Avoiding the humoral immune system. In other fields of gene therapy (cystic fibrosis or other heritable disorders, for example), persistent gene expression is often thwarted by destruction of the producer cell by both the B-cell and T-cell arms of the immune system. For cancer gene therapy, destruction of the target cell is an objective and, for this indication, suppression of the B-cell arm is the main issue (although the innate immune system can also contribute to virus neutralization88). Most adults have low levels of neutralizing antibodies against adenoviruses of the serotypes used in most contemporary vectors (Ad serotype 5), but for local injection or infusions of high doses of virus (1012 virions, for example), these levels of neutralizing antibody are probably insignificant. However, systemic treatment results in relatively low levels of circulating virus, and neutralizing antibodies are almost certain to curtail the effectiveness of these approaches, particularly as titres rise during repeated treatment89. Again, creative solutions might be at hand. Serotypes could be used that are less antigenic or less common in the population, and epitopes could be engineered out of viral particles or masked with agents such as polyethylene glycol87. Alternatively, the B-cell arm of the immune system could be suppressed by selective immune suppression, using the anti-CD20 antibody Rituxan, for example (T. Ried, personal communication). Whether these or other solutions are effective remains to be seen. A great deal depends on their success, as the future of gene therapy for systemic diseases would be transformed if this hurdle were eliminated or minimized.

Future prospects

The future of gene therapy for cancer depends on a combination of applied bioengineering and traditional clinical development. To a large extent, the problem of selectively killing cancer cells has been solved. Critics of cancer gene therapy would probably agree that the biological principles are sound, but translating these principles into reality remains a formidable — perhaps prohibitive — challenge. Most of the innovative strategies designed to circumvent these problems are still in preclinical or early clinical testing, and it is too early to predict how successful they will be. However, even the first-generation agents that have entered clinical testing have proved safe and there have been several suggestions of efficacy. Considering the small number of concepts that have been tested extensively (HSV-tk-based suicide vectors, Adp53, ONYX-015, G207 and CN706), it is too early to assess this field conclusively, particularly as only one of these approaches — HSV-tk-based suicide-gene therapy — has completed a Phase III trial.

The impact of cancer gene therapy will depend to some extent on the success of other strategies. Of these, small molecules that are directed at known molecular targets, and also therapeutic antibodies, seem to be the most promising. Small molecules are the most traditional and preferred approach for most pharmaceutical companies. However, this approach has so far been remarkably unsuccessful in oncology. Very few new small molecules have been approved for cancer treatment based on rational targets, despite tremendous efforts throughout the industry. However, the tide could be turning. With the dramatic results of STI-571 (Gleevec) in treating early stages of chronic myelogenous leukaemia, and promising data from epidermal growth-factor receptor (EGFR) inhibitors and others, a new era of optimism has arrived. But the problems of obtaining selectivity for cancer cells, and of achieving suitable biodistribution and safety, are formidable, and are frequently underestimated by investigators unfamiliar with the process of drug development. Furthermore, new agents such as Gleevec have already encountered the old problem of drug resistance that limits the potency of previous generations of small-molecule drugs90. Yet the success of Gleevec shows us that there is a successful path to a systemically active and safe anticancer drug, however tortuous this path may be. The same cannot yet be said of cancer gene therapy, even though the concepts have strong biological credentials.

Antibody agents have enjoyed a renaissance. Indeed, trastuzumab (Herceptin), which blocks the oncogenic EGFR ERBB2, was the first approved drug to be developed directly as a result of the discovery of an oncogene, and similar agents are in the pipeline. Antibodies have more predictable pharmacological and toxicological properties than small molecules, and HUMANIZED versions are not as immunogenic as viral vectors, so do not face the problem of neutralization that must be addressed for cancer gene therapy. However, their impact is limited by the small number of targets that are accessible and specific on the surfaces of tumour cells.

Ultimately, these diverse approaches will be used together, as they use different mechanisms to control or kill cancer cells and they are very likely to have distinct toxicity profiles. In the early 1990s, it was not possible to claim with any assurance that cancer cells could be killed selectively and rationally. Now, the debate has shifted to a discussion of which approach is most likely to be successful and, in the case of cancer gene therapy, whether the ingenuity of bioengineers and clinical scientists can surmount the final hurdles. We already know how to make more potent suicide genes, and how to make viruses that replicate more efficiently in tumour cells. We can re-target viruses away from their natural receptors, and we can imagine ways to keep the immune system from attacking these therapeutic agents. None of these new ideas has yet been tested in the clinic, but any one of them could help enormously. It seems unlikely to me that cancer gene therapy will fall at the final hurdle.