Attempts to prevent illness have been sought from time immemorial. Vaccination has since proved its great worth in the prevention of once fatal illnesses caused by bacteria and viruses1. In developed countries, however, cancer is now outstripping wounds and infectious diseases as a cause of death, and there is an urgent need for effective ways of preventing it2.

Many cancers are caused by exogenous agents. Exposure to industrial carcinogens is the subject of legislation, and the recreational contact with carcinogens such as tobacco smoke, alcohol and sunlight is being discouraged through educational information. The significance of these preventive measures is great, as they are markedly reducing the incidence of some common and lethal types of tumours, namely lung cancer, squamous cell carcinomas of the head and neck, and melanoma3. As microorganisms are the cause of 10–20% of all human tumours2, vaccines that reduce infection with viruses that cause cancer are of the utmost importance in primary cancer prevention. Vaccination against hepatitis B virus, for example, has reduced the incidence of hepatocellular carcinoma4 whereas vaccines against human papilloma viruses are expected to greatly reduce the incidence of cervical carcinoma5,6.

New ways to control the endogenous causes of cancer are needed, as these causes cannot be avoided. One example is cancer promotion caused by physiological oestrogen stimulation of the mammary gland during fertile life. The duration of fertile life — which could be increased by, for example, an early menarche — is a significant risk factor for breast cancer in women7. One form of control of endogenous causes is provided by the chronic administration of drugs that target components of the signalling pathways — an approach called chemoprevention8. For instance, selective oestrogen-receptor modulators such as tamoxifen can be chronically administered to women to reduce the risk of breast cancer9. However, in its present form, chemoprevention is only partially effective and can also be accompanied by serious side effects. For instance, although tamoxifen halves the risk of invasive breast cancer in high-risk women, it also increases the risk of endometrial cancer, deep-vein thrombosis and cerebral vascular incidents9. The ideal preventive agent should obviously cause no side effects, as it will be given to healthy people.

As vaccines are minimally associated with side effects and invasive procedures, it is time to consider whether they can also be used to prevent tumour development. However, three questions must first be unequivocally answered: why should vaccines be particularly effective in cancer prevention, which antigens should be targeted, and who should be vaccinated?

Therapeutic versus preventive vaccines

Anti-tumour vaccines are effective in preventing a subsequent tumour challenge in animals — this is a well-substantiated observation established through countless tumour-challenge experiments performed in immunized animals (immunization-tumour challenge experiments) using many different fast-growing and aggressive mouse tumours10. In these experiments, immunization against a tumour antigen is followed by a subcutaneous, intramuscular or orthotopic challenge with a lethal dose of a transplantable tumour. Vaccines being tested in these models range from those consisting of live, irradiated or genetically modified tumour cells, dendritic cells, proteins, peptides or naked DNA. Each of these vaccine preparations can be given alone or combined with cytokines and co-stimulatory factors. In a few cases, vaccines are given in association with the removal of regulatory T cell (Treg) suppressor activity11,12. In mice, effective immunity is often elicited and a successful pre-immunization against almost any kind of tumour seems to be feasible.

The essential distinction between prevention and cure, however, has been forgotten in the transition from experimental models to clinical practice, as vaccines are used for therapy rather than for prevention of human tumours. Transfer of these experimental results to the clinical setting and the prospect of curing cancer with therapeutic vaccines are therefore seen as feasible goals. In 2005, more than 200 clinical trials were in progress (see the ClinicalTrials.Gov Database of the National Cancer Institute and the European Organization for the Research and Treatment of Cancer Protocols Database web sites). The results achieved so far, however, have been poor; partial responses are rare and complete responses extremely rare. Only in a few patients has the progression of previously growing tumours been halted and prolonged survivals observed. As benefits that are due to vaccination have been substantiated in no more than a handful of cases12, new strategies are being explored11,12, including vaccines based on engineered viral vectors, various approaches with dendritic cells, and strategies that are aimed at inhibiting immunosuppressive cells of lymphoid or myeloid origin. Perhaps vaccination alone is not the solution for treating existing tumours, and evidence is emerging that shows that combining immunotherapy with chemotherapy, radiotherapy, anti-angiogenic therapy and other approaches could yield synergistic or additive results13,14.

Therapeutic vaccines are also poorly effective in mouse models of cancer, so the lack of benefit seen in clinical trials is perhaps not surprising. We15,16,17 and others18 have shown that only a minority of tumour-bearing mice are cured by therapeutic vaccination, and even this limited efficacy is only obtained when the vaccine is given soon after the challenge. At present, monoclonal antibodies (mAb), based on B-cell immune responses, and not therapeutic vaccines to activate T-cell immune responses, have passed the acid test of clinical trials. A stark incongruity is evident between the clinical efficacy of B-cell-derived mAbs — such as trastuzumab for breast cancers that overexpress the protein product of ERBB2 (also known as HER2/NEU)19 or rituximab for lymphomas that express CD20 (Ref. 20) — and less successful attempts to selectively activate a T-cell response, primarily cytotoxic T lymphocytes (CTLs), by vaccination12.

Vaccines prevent cancer: preclinical models

The CTL chauvinism that dictates vaccine strategies21 stems from immunization-tumour challenge experiments showing that CTL reactivity is the only way to block a fast-growing tumour challenge. Conventional mouse models, however, do not provide an accurate assessment of prevention, as they measure acute rejection of a tumour challenge performed when mice are optimally immunized (Box 1).

More accurate models to investigate cancer prevention are those in which the entire natural history of the tumour in a single mouse is investigated. The conventional system is to subject mice to carcinogens22, whereas it is now more convenient to study genetically engineered mice (GEMs) that harbour activated oncogenes and/or inactivated tumour-suppressor genes that predispose them to develop cancer23. As assessment of the ability of a vaccine to inhibit the onset of an autochthonous tumour and control the progression of precancerous lesions in GEMs requires long experiments (typically 1 year or longer), relatively few studies have been completed. They do, however, provide a coherent picture of the great potential of vaccines to induce sustained prevention of cancer and demonstrate that a vaccine-alerted immune system effectively blocks aggressive carcinogenesis driven by the overexpression of distinct oncogenes (Table 1a,b).

Table 1a Vaccines in the prevention of autochthonous tumours in various GEM models
Table 1b Vaccines in the prevention of autochthonous tumours in Erbb2 GEM models

The degree of protection varies from one model to another and within the same model, ranging from a mere delay in tumour latency to complete protection from tumour onset for 1 year or even to near the end of the natural lifespan. The most successful vaccines have been combinations of the antigen with other immunological signals such as microbial CpG sequences24, cytokines25,26,27,28, allogeneic glycoproteins of the major histocompatibility complex (MHC)27,28,29 and/or co-stimulatory molecules30. Analysis of the variables involved in vaccine design indicates that different formulations (for example, cells, proteins, peptides or DNA) are equally effective, provided that the immune system is exposed to high concentrations of the antigen combined with potent adjuvants.

The data also endorse the concept that for vaccination to be effective it must be started at an early stage of tumour formation. Its ability to protect decreases or disappears when the precancerous lesion reaches a more advanced stage26,31,32. It is clear, therefore, that a vaccine capable of preventing cancer development cannot be expected to cure an established tumour. Vaccine-induced block of tumour onset that is macroscopically evident in the organs at risk is mirrored by microscopic findings showing that very few cells, if any, still express the transgene. In the protected mice, pathological observations coupled with gene-expression-profile analysis show that a vaccine can delay the progression of advanced pre-cancerous lesions29,33 for long periods. Various GEMs show true cancer syndromes with the synchronous or metachronous occurrence of neoplasms in several organs34,35,36 and vaccines have proved highly effective in preventing these tumours of diverse origin34,36. However, in some models, specific tumour types escape surveillance — in particular, sarcomas with low expression of the target antigens34. This is an important issue as the tumour histotype cannot always be predicted in humans.

Immune mechanisms of tumour prevention

As carcinogenesis is a lengthy process, the protection afforded by a preventive vaccine acquires significance only as time progresses. This requirement for long-term effectiveness of the protective immune response is the source of a significant difference between cancer prevention and therapy. Cancer therapy is typically given over a few months (around 1% of the patient's lifespan), and the most prolonged adjuvant therapies, such as tamoxifen or aromatase inhibitors that are used to treat breast cancer, are typically given for 5 years (5–10% of the patient's lifespan)37,38. Acute management with cytotoxic drugs, as well as treatment with CTL and natural killer (NK) cells, seeks to rapidly destroy a tumour. By contrast, cancer prevention necessitates treatments that must be protracted throughout the individual's life to counteract the unpredictable and unremitting generation of neoplastic cells (Table 2).

Table 2 Key differences between anti-tumour vaccines in cancer therapy and prevention

This extended time-frame for tumour prevention must be considered in the development of vaccines: the vaccine must be able to elicit immune memory mechanisms that persist for very long periods. Results obtained by different laboratories with GEMs that are transgenic for the rat Erbb2 gene show that several vaccines afford almost lifelong protection against a genetic predisposition to mammary cancer (Table 1). In all cases, a long-lasting protective memory relies on repeated booster vaccinations. Identification of immune signals that could reduce this need for frequent boosting is a key issue in preventive vaccination research. However, unlike viruses (in which the requirement for memory maintenance has been thoroughly investigated39) induction of a prolonged immune memory to tumours is a more difficult task. In both humans and GEMs, tumour antigens are mostly self-antigens that are overexpressed by the tumour. This implies that a vaccine should be able to reverse a tolerant state40,41 and activate low-avidity immune reaction mechanisms that escaped central tolerance42,43. The requirements for maintaining a sustained low-avidity immune response are not clear.

Because rat Erbb2-transgenic mice mirror several features of the stepwise mammary carcinogenesis in women44, these mice have been the most thoroughly investigated (see the BALB-neuT web page on the National Cancer Institute's Cancer Models web site). In BALB/c mice that are transgenic for activated rat Erbb2 (BALB-neuT GEMs), tolerance to the transgene Erbb2 impedes the activation of a high-avidity CD8+ CTL response42,45 to the dominant epitope of ERBB2 protein, which is recognized by wild-type BALB/c mice46. Despite the lack of such a crucial component of immune reactivity, we and others have shown that many different vaccines delay the onset of these mammary carcinomas and significantly extend the tumour-free survival25,27,28,30,32,42,47,48. The study of BALB-neuT GEMs that are rendered deficient in various immune components, as well as adoptive transfer experiments, have made it clear that the protection afforded rests on the activation of CD4+ T cells. This activation results in the release of interferon-γ (IFNγ) and the induction of anti-ERBB2 antibodies (Fig. 1). As most immunization-tumour challenge studies that were performed with ERBB2+ tumours emphasized the role of CTLs in tumour inhibition, it was somewhat surprising to find that anti-ERBB2 antibodies are both necessary and sufficient for protecting BALB-neuT GEMs, and that the IgG2a isotype is the most effective42,47,49,50. CD4+ T-helper (TH) cells that release IFNγ have a central role in controlling the switch of antibody isotype to IgG2a51. However, it seems that this is not their only activity. In protected mice, T cells infiltrate precancerous lesions and recruit granulocytes, orchestrating complex reactions of delayed-type hypersensitivity and inhibiting angiogenesis42,45. Adoptive-transfer experiments show that better protection is transferred when the recipient mice receive both antibodies and T cells from the immunized donor mice42, indicating that effective vaccines should be designed to elicit both humoral and cellular immunity. Anti-ERBB2 antibodies and IFNγ-producing T cells are easily monitored. The antibody response in young vaccinated mice predicts freedom from tumour onset at 1 year of age34. These parameters might therefore provide a way to monitor the vaccine efficacy in GEMs (and possibly in humans) and to design and optimize vaccination schedules and strategies.

Figure 1: Vaccine-activated immune reactions against membrane oncoantigens.
figure 1

Administration of antigen and adjuvants attracts polymorphonuclear cells (PMNs) and macrophages, and leads to the maturation of dendritic cells, which capture the antigen, present it to T-helper cells (TH) and release interleukin-12 (IL-12). Antigen presentation might occur at the site of vaccination if the ongoing reaction attracts T cells, or in the draining lymph node where the antigen presented by dendritic cells activates TH cells. This activation causes TH cells to release interferon-γ (IFNγ) and contribute to both cell-mediated lysis through cytotoxic T lymphocyte (CTL) activation and the differentiation of B cells into plasma cells, which produce antibodies. TH cells and CTLs migrate from the lymph node, infiltrate the precancerous lesion, recruit and activate polymorphonuclear cells and macrophages, and release tumour-necrosis factor-α (TNFα) and IFNγ42,45. TNFα and IFNγ are cytostatic for tumour cells and increase major histocompatibility complex class I (MHC-I) expression on normal and tumour cells — moreover, IFNγ blocks the tumour-driven angiogenic switch and inhibits tumour invasion42,45,50. When the antigen is expressed on the cell membrane, antibodies might activate complement-mediated lysis, or antibody-dependent cellular cytotoxicity, cause antigen internalization and degradation, inhibit cell signalling and block cell proliferation47,91. The immune response ultimately leads to the disappearance of oncoantigen-positive pre-neoplastic cells and normalization of the molecular phenotype and morphology of tissues at risk of cancer. The long-term maintenance of a protective immune response guarantees surveillance and the continuing elimination of oncoantigen-positive cells as soon as they emerge.

Modulation of Treg cells combined with vaccine administration seems a promising way to improve the array of the immune mechanisms that are activated by a vaccine and improve their persistence52. Recent work by Ercolini et al.41 shows that latent pools of high-avidity CD8+ T cells are recruited by the removal of CD4+ CD25+ Treg cells during anti-ERBB2 protein vaccination in FVB-neuN GEMs. These findings fit in well with the burgeoning data on the role of Treg cells in controlling the reaction against self antigens53,54. Ongoing experiments on BALB-neuT GEMs show that the removal of Treg cells combined with DNA vaccination activates CD8+ CTLs against ERBB2 protein and extends the protective memory to almost the lifespan of immunized mice (E. Ambrosino and F.C., unpublished observations).

Therapeutic versus preventive mechanisms

There are many reasons why vaccination is poorly effective as a tumour therapy. The kinetics of tumour growth might just surpass the destructive potential of the immune system55. Tumours also orchestrate a large array of immunosuppressive activities56 and build a micro-environment that does not allow leukocyte extravasation and local recruitment of immune cells57. Proliferating tumours resist immune attack not only because of their cell number, but also because they actively secrete cytokines. These cytokines — for example, interleukin-10 (IL-10)58— can downmodulate the immune response. Other cytokines — for example, colony stimulating factors (CSFs) and vascular-endothelial growth factor (VEGF)59—elicit the expansion of immunosuppressive immature myeloid cells that express the Gr-1 (also known as Ly6G), MAC1 (also known as CD11b) and platelet/endothelial-cell-adhesion molecule (PECAM1, also known as CD31) surface markers.

In addition, following their initial genetic alteration, tumour cells become genomically unstable. Fuelled both by the crisis that precedes telomerase activation and by mutations in caretaker genes, this instability increases as the tumour progresses60. Therefore, an established tumour can be seen as an aggregate of millions of genetically and epigenetically different and unstable cells. The selective pressure exerted by the immune system, whether activated spontaneously or by vaccination, leads to the selection of tumour clones that either no longer express the antigen targeted by the immune attack or that can pre-empt immune-effector mechanisms61. Through this 'immunoediting', an immune response that is not strong enough to lead to tumour eradication results in the selection of tumour variants that are less antigenic and less sensitive to immune attack; a situation that is sometimes observed in clinical trials. Preventive vaccines that resulted in such an adverse effect would be totally unacceptable. Their chances of success must therefore be magnified to the utmost by selecting target antigens that are essential for tumour growth and progression (oncoantigens) (Box 2).

Recognition of antigens by CTLs rests on the expression of the antigenic protein itself and correct functioning of the machinery that leads to the membrane translocation of class I MHC glycoproteins (MHC-I) presenting a fragment (peptide) of the antigenic protein62. Large masses of genetically unstable tumour cells escape CTL recognition through downmodulation of the antigenic protein, MHC-I or peptide-processing machinery. Between 70% and 95% of all human tumours downregulate MHC-I expression63,64, an impressive percentage that is higher than the percentage of tumours that have p53 mutations and is on a par with that of telomerase activation. As complete loss of MHC-I activates the NK-cell reaction, tumour cells that express a low level of MHC-I simultaneously minimize recognition by both T- and NK-cells65. Although several other molecular pathways through which tumours evade a CTL attack have been described66, appraisal of their frequency in human tumours is still lacking.

The main reason why vaccines that are poorly effective in tumour therapy are effective in tumour prevention is that the target of the prevention is an oncoantigen that has not yet initiated tumorigenesis (or at least not at a level that we can detect) or is only a precancerous lesion. This implies that most of the difficulties of using vaccines in the therapy of malignant tumours simply do not apply to prevention (Fig. 2). Although recent data on cancer progression have shifted the onset of genetic instability to the early phases of the natural history of tumours, the accumulation of genetic 'hits' that leads to cancer progression is a function of time60. The low proliferation rate of indolent pre-neoplastic lesions cuts down the likelihood of selecting immune-resistant clones, whereas the vaccine-alerted immune response might inhibit the lesion before complete neoplastic transformation takes place. Usually, precancerous lesions are not yet sheltered from the immune attack by the fibroblastic stroma, nor do they show either the tissue organization or the immunosuppressive abilities of large, proliferating tumour masses67.

Figure 2: As a tumour progresses, a vaccine-induced reaction becomes less and less able to cope with tumour evasion mechanisms.
figure 2

The data on cancer-prone genetically engineered mice (GEMs) show that a vaccine-alerted immune system efficiently checks the onset of neoplastic lesions27,35,36,48,100,101,102. Tumour eradication can be elicited when mice already harbour an initial, precancerous lesion45,47. At later stages (advanced or multifocal precancerous lesions) fewer mice are protected, and less efficient induction of immunity and a lower efficacy of effector mechanisms are evident if vaccination begins when advanced multifocal lesions are present. Pathological studies coupled with gene-expression-profile analysis show that a vaccine halts the progression of advanced precancerous lesions instead of leading to their eradication29,42. In addition, vaccines that provide a sustained inhibition of precancerous lesions fail to control invasive cancer26,42. Expansion and accumulation at the tumour site of Treg cells103 and the onset of immature myeloid cells104 impede the induction of immunity, whereas tumour proliferation, clonal diversification and inaccessibility to immune cells and molecules limit its efficacy. A limited protection might still stem from the combination of the vaccine with other immunological manoeuvres such as Treg depletion41,105 and anti-angiogeneic molecules26. No data on vaccine cure of clinically evident tumours are available for the GEM models that are more closely relevant to human cancer. Antigenically diverse tumour clones are shown in different shades of blue. CTL, cytotoxic T lymphocyte; TH, T-helper cells; Treg, regulatory T cells.

Precancerous lesions might also be ignored by the immune system as they are too small to be perceived. This avoids the induction of various suppressive and regulatory cells and peripheral tolerance triggered by inappropriate antigen presentation, as happens when tumour antigens are presented by proliferating tumour cells in the absence of co-stimulatory signals36,68. By contrast, when an ignored antigen is first recognized through optimal presentation by the vaccine, the probability of eliciting an efficacious immune response is much greater. Therefore, a person with a precancerous lesion might be expected to mount a more effective response than a patient with a tumour who has already been treated in various ways. The chances that a vaccine-alerted immune system will eradicate a precancerous lesion should be greater than when dealing with an invasive tumour67.

Target antigens for cancer prevention

The extended time-frame that characterizes tumour prevention makes immune selection of antigen-loss clones that escape the vaccine more likely. This possibility calls for the use of target antigens that are directly involved in promoting the neoplastic process, which we have called oncoantigens in this Review (Box 2). Targeting the immune response against such antigens minimizes the likelihood of selecting antigen-loss clones. When this kind of target is expressed on the cell surface (membrane oncoantigens), a vaccine's ability to activate an antibody response will also minimize the importance of the loss of MHC-I, which is required for presentation of the antigen to T cells. In addition, once a persistent immunity is elicited, it should remain effective for long periods, which will offset the appearance of genetically unstable tumour cells. Targeting the immune response to oncoantigens and inducing an antibody response are therefore key elements for the success of preventive vaccines.

The long-lasting prevention of mammary and salivary carcinomas that arise in GEMs that are transgenic for rat Erbb2 probably rests mostly on the fact that the immune response is elicited against the ERBB2 protein, which is a tyrosine-kinase membrane receptor. The ERBB2 receptor is not only the product of an oncogene that is instrumental for carcinogenesis69, but also an antigen recognized by the immune system — a combination of properties that makes it a very attractive oncoantigen. In GEMs that are transgenic for the rat Erbb2 oncogene, mammary and salivary carcinogenesis is driven by ERBB2-receptor expression, whereas downmodulation or loss of the receptor blocks tumour progression29,70, at least during the early phases. In the course of tumour progression, additional genetic hits might make ERBB2-receptor signalling redundant as other oncogenes and signalling pathways are activated71. As preventive vaccines operate during the early phases of carcinogenesis, effective inhibition of the ERBB2 receptor arrests the whole process and renders the selection of ERBB2-receptor-loss variants unlikely. The ERBB2 receptor and many other oncoantigens are ideal targets in tumour prevention. This is because their key role in the maintenance of the transformed phenotype restrains the emergence of antigen-loss variants35,72, provided that preventive vaccination is undertaken during the early carcinogenic stages before further genetic hits take place.

A second key feature of the ERBB2 receptor is its membrane localization, which preserves its antigenicity and accessibility to antibodies even in escaping tumour-cell clones that express little or no MHC-I. Although processed ERBB2-receptor peptides presented by MHC-I are recognized by CD8+ CTLs73, ERBB2-expressing tumour cells can avoid T-cell recognition by downregulating MHC-I (Refs 74, 75). One of the Achilles' heels of T-cell immunity is that immune recognition is disjointed from the function of the target antigen because T cells recognize pro-cessed peptides rather than functional proteins. By contrast, antibodies bind to target antigens that are functional and on the cell surface, independently of their processing and MHC-I presentation. In this case, the only escape would be through the selection of antigen-loss variants. Targeting the immune response to membrane oncoantigens minimizes the emergence of immunoresistant tumour variants that are due to the loss of either the antigen or MHC-I. Relatively few of the tumour antigens that are commonly targeted by therapeutic vaccines are oncoantigens or membrane oncoantigens because such properties are of little relevance for CTL-based immunotherapeutic strategies.

The risks of preventive cancer vaccines

As for all preventive medicine, an extremely low incidence of adverse effects is a prerequisite of preventive cancer vaccines. Although, in general, vaccines are among the safest products of modern medicine, they are constantly the subject of suspicion and apprehension76. In a patient with a life-threatening cancer, the side effects of treatment are, on balance, worth putting up with as long as benefit is achieved, and might be evaluated as surrogate markers of an effective immune response. For example, in patients with melanoma the appearance of vitiligo following therapeutic vaccination is taken as evidence that the immune response is destroying melanin-forming cells77. However, this sort of side effect would be unacceptable for preventive vaccines that are to be given to healthy people for long periods of their life.

Inflammatory reactions that accompany the destruction of an anticancer vaccine by leukocytes are usually local and transient (Fig. 3). The release of pro-inflammatory cytokines might induce a mild, systemic flu-like syndrome that is usually transient and tapers off with repeated administrations of the vaccine. A major cause of concern is the choice of adjuvants. Whereas vaccines that are approved for the prevention of infectious diseases use alum — a bland, well-tolerated adjuvant — vaccines used in cancer therapy adopt more toxic adjuvants — for example, incomplete freund's adjuvant — because they induce more potent anti-tumour immune responses. Experimental vaccines that are tested in mice use various powerful adjuvants, ranging from complete Freund's adjuvant to cytokines such as IL-12. Vaccine components might also have an adjuvant role. Local damage associated with in vivo electroporation and the presence of CpG sequences have an adjuvant role in DNA vaccines, whereas with whole-cell vaccines many antigenic signals can have an adjuvant role78. Experimental evidence shows that almost all vaccines fail to immunize the host if the antigen is administered without adjuvants79. In the translation from preclinical studies to clinical application, the choice of adjuvant remains a crucial and poorly explored issue that might cause dramatic losses of immunogenicity.

Figure 3: Mechanisms of action and side effects of cancer vaccines.
figure 3

Vaccination against tumour antigens elicits immune responses that can also cause unwanted damage to the host. Degradation of the vaccine is accompanied by a varying degree of local inflammation. Adjuvants that are co-administered with the vaccine determine the intensity of this local response and can trigger adverse local reactions that are mediated by infiltrating leukocytes, as well as systemic consequences mediated by cytokines reaching the bloodstream. Mild adenopathy can develop in the lymph nodes draining the site of vaccination as a consequence of antigen presentation, cell migration and proliferation. Autoimmunity is the most severe unwanted side effect of vaccines against tumour antigens that are expressed also by normal cells. Autoimmune reactivity can be directed against normal components antigenically related to the antigen, but not included within the vaccine, a phenomenon called antigen spreading. CTL, cytotoxic T lymphocyte; DC, dendritic cell; IFNγ, interferon-γ; IL-12, interleukin-12; NK, natural killer cell; PMN polymorphonuclear cell; TH, T-helper cell; TNFα, tumour-necrosis factor-α.

The most important long-term concern is the induction of autoimmunity80, which depends on the kind of tumour antigen that is targeted and the response that is elicited. We found no signs of autoimmunity in the heart, kidney and liver of 1-year-old GEMs that were transgenic for rat Erbb2 and that had been protected against mammary cancer development by repeated rat Erbb2 DNA vaccinations42. In these mice, anti-ERBB2 immunity successfully controlled the progression of mammary pre-neoplastic lesions that overexpressed the ERBB2 receptor through coordinate activation of low-avidity reaction mechanisms. A low-avidity response might be crucial in discriminating between quantitative differences in the expression of the target antigen81. The absence of overt autoimmune lesions is attributable to the combination of the poor expression of ERBB2 oncoantigen by the tissues of adult mice and the inability of transgenic mice to generate a high-affinity immune response. In contrast to high-avidity mAbs, the immune response elicited by immunization in naturally tolerant mice (and possibly patients) is not expected to attack the heart or nervous system. Signs of autoimmunity were only found in the mammary gland of vaccinated mice82. This gland continues to develop in adult life after hormone stimulation when differentiating cells that overexpress the ERBB2 receptor are present, and it also maintains its normal structure83. The alteration of the histological structure of the mammary gland in long-term vaccinated mice is restricted to the cells that overexpress ERBB2, and resulted in the reduction of branching of ducts and lactation ability, an acceptable price for carcinoma prevention. Generation of tissue-restricted autoimmune responses might even be an additional approach to cancer immunotherapy that will allow treatment of cancers arising from non-vital organs such as breast, prostate, thyroid and testis. This issue is of practical importance for young women at hereditary risk of breast cancer, for whom an 'immunological mastectomy' would represent a great improvement over prophylactic surgical double mastectomy84.

Even if lengthy studies on vaccine-mediated cancer prevention in GEMs have not disclosed severe side effects, induction of autoimmunity remains the greatest theoretical concern in the use of preventive cancer vaccines aimed at inducing long-term protection in healthy individuals. Its avoidance must be sought through further, accurate preclinical investigation.

Who would be vaccinated?

The data on the ability of vaccines to prevent a tumour emerging from immunization-tumour challenge experiments and those from GEMs cannot be directly translated to humans. This is not only because of the idiosyncrasy of each experimental system, but mostly because it is almost impossible to rule out the possibility that the different patterns and timing of transgene expression in GEMs lead to an immune tolerance to oncoantigens different from that of humans. Nevertheless, the results reviewed here provide a serious preclinical rationale and indicate that vaccines are an effective prospect for the prevention of carcinogenesis. Induction of a specific immune response against the most common onco-antigens overexpressed by pre-neoplastic lesions might constitute a new scenario in cancer prevention.

The translation of this rationale into novel preventive treatments requires sequential clinical approaches, as vaccination of a healthy individual is not the same as a compassionate attempt to help a patient with advanced cancer. Given the long latency of tumour development and the many difficult issues there are to tackle, early trials should be aimed at using vaccines to prevent tumour relapse in patients at high risk after successful tumour management. Several cancer vaccines are now entering phase III clinical testing in the setting of minimal residual disease, in which they are aimed at preventing tumour recurrence and prolonging survival. Even if they fail to show clinical benefit, as the immune system of these patients is imprinted by the primary tumour and weakened by previous treatments, they will nevertheless provide preliminary information on the ability of vaccines to overcome tolerance to oncoantigens and on the risks of autoimmunity.

Vaccination against selected oncoantigens of healthy people who have a specific genetic risk of cancer, who have been exposed to an exogenous carcinogen, or who bear multifocal pre-neoplastic lesions (such as those with the basal nevous syndrome35 would provide a more appropriate scenario. One can also envisage a further application of vaccines to prevent tumours in the general population85. Because of the extensive data on ERBB2 as a model oncoantigen, clinical trials that are focused on this antigen would be reasonable. Early intervention would possibly lead to a reduced incidence of carcinomas with the worst prognosis, whereas a compensatory increase in other less-aggressive subtypes might not be precluded. However, as only about 20–30% of breast carcinomas overexpress ERBB2, a primary prevention trial would require a large number of patients and long-term follow-up to have the statistical power to show an effect. The logistical hurdles and the costs of such a trial would not be too dissimilar from those of other large cancer-prevention trials9.

Moreover, as we remain ignorant of what enables immunity to be maintained for decades86, how can 'successful' cancer vaccination be assessed before waiting for disease development? Can induction of an antibody response to oncoantigens be equated to a successful result? The positive data from the clinical trials with mAbs indicate that the vaccine elicitation of a de novo response against an oncoantigen is a significant surrogate end point. Initially, vaccine boosting schedules will be based on the induction and maintenance of a significant antibody response. How much the preventive ability of vaccine is optimized by an additional triggering of IFNγ-producing T cells and other components of cell-mediated immunity will probably only be established through the clinical use of vaccine formulations that induce a humoral or cellular response individually or in combination.

It is expected that increasingly predictive oncology will assess the individual risk of cancer as a function of sex, age, family history, genetic makeup and lifestyle (see the American Cancer Society Who is at Risk and the Your Disease Risk web sites), whereas gene expression profiles and molecular biology outline the probability that a particular oncoantigen will be expressed by the tumour for which the person is at risk. By combining this information, one can select those for whom preventive vaccination might be useful.