Main

The first indication that tumours were immunogenic came from animal models, in which tumours were rejected. There are, however, marked differences in the degree of immunogenicity between different tumours. Some are easily rejected, whereas others escape destruction by the immune system. So, can we use this knowledge to increase tumour immunogenicity as a therapeutic strategy? A large number of pre-clinical and clinical trials in humans have demonstrated the feasibility of antitumour vaccinations. Vaccination is usually well tolerated, which highlights the safety of this approach.

Tumour vaccines make use of tumour antigens — which are, in general, derived from proteins that are produced by the tumour — and aim to trigger a cellular immune response that is executed by T cells. Tumour antigens can be either tumour specific, so are expressed exclusively in tumour tissue, or tumour associated, so are highly overexpressed in tumours but can also be found in normal tissue; differentiation antigens are overexpressed in tumours and are found only in certain types of normal tissue (see below, and Table 1). Peptides from tumour antigens, like those from every other cellular protein, are released from their source proteins during antigen processing, mainly by the proteasome, and are presented by human leukocyte antigen (HLA) class I molecules on the cell surface. Such HLA–peptide complexes are under the surveillance of the cellular immune system. The cellular immune response consists of a cytotoxic effector function, which is carried out by cytotoxic T cells that are positive for the CD8 surface marker and recognize peptides that are presented by HLA class I molecules, and a helper function, which is executed by CD4-positive T cells that are restricted by major histocompatibility complex (MHC) CLASS II MOLECULES (see Box 1). The type of immune reaction that is initiated therefore depends on which MHC molecule presents a peptide from the antigen. During the 1990s, when tumour antigens and their peptide fragments — epitopes, which are presented by MHC molecules — were extensively characterized, much of the work focused on CYTOTOXIC T LYMPHOCYTES (CTLs) and the MHC CLASS I antigen-processing system. In fact, owing to experimental limitations, our knowledge of MHC class-II-restricted T-helper epitopes is still rather limited.

Table 1 A selection of tumour antigens and their T-cell epitopes
Full size table

Many epitopes have been reported, some of which are derived from the classical tumour antigens that were identified a long time ago — such as p53, RAS and carcinoembryonic antigen — but others are from recently discovered tumour antigens. More than 70 tumour antigens that are recognized by (mainly CD8+) T cells were listed in a recent review1. In principle, a short peptide that is derived from the sequence of a tumour antigen and presented by an HLA class I molecule represents a potential tumour vaccine. We do not know exactly how many times an individual overcomes tumours during their lifetime. The risk of immunosuppressed individuals developing certain virus-associated tumours — such as non-Hodgkin's lymphoma and Kaposi's sarcoma — is rather high, but, for most kinds of tumour, impaired functions of the immune system do not seem to be important for suppressing tumour formation. Nevertheless, we have to realize that the immune system has failed when tumours develop. This might be because the tumour orginates from 'self' and is therefore not immunogenic. It could also be that the very low number of antigens that are present in the early stages of tumour development does not induce an immune response, as is the case with low doses of pathogens.

The wide variety of antitumour vaccines that have already been applied show that the perfect vaccine — with respect to the selection and number of antigens, and the application strategy — has still to be designed. One of the most favoured applications uses the most efficient professional antigen-presenting cells, DENDRITIC CELLS (DCs), pulsed with peptides from tumour antigens. The number of cells and the number of different — in part, 'engineered' — peptides administered, and the application of adjuvants or enhancers differs among all studies. Although a complete tumour regression has been reported several times2,3, reports of a poor clinical outcome unfortunately also exist4,5, and the first critical opinions have already been voiced; these speak of disappointing trials6. There are several possible reasons for the failure of antitumour vaccination, one of which is loss of the antigen from the tumour, which leads to tumour-escape variants7. But the loss or downregulation of specific molecules in the antigen-processing pathway has also been reported8,9. In addition, immune responses occasionally even fail if antigen, MHC and specific T cells are present, which indicates that tumours can also use other mechanisms to exert a certain form of immunosuppression.

The use of T-cell epitopes in immunotherapy of tumours still presents us with the problem that individual differences between patients — both in the type of HLA molecules and in the epitopes that are presented and recognized — have not yet been finely ascertained. Patient-specific expression of tumour-associated antigens, quantitative differences in HLA-restricted presentation of epitopes and T-cell pools that vary among individuals might all have a significant role. For example, in the SYFPEITHI database10 (see online links box), which lists cancer-related T-cell epitopes, 11 different HLA-A*0201-restricted CTL epitopes from ERBB2 (HER2/neu) are listed, and we do not know which is most efficient in the struggle against an individual tumour (see Box 2). A significant limitation of many clinical trials is their use of one or a few epitopes that is presented by only one HLA restriction element, as this facilitates tumour escape because loss of only one antigen suffices. To promote the exciting option of defeating tumours by 'self', the basic strategies must now be corrected. The future of a cancer immunotherapy that exploits T-cell epitopes from tumour antigens lies in individual, patient-specific analysis and therapy, and in using several tumour antigens and epitopes that are presented by more than two HLA molecules. However, the number of tumour-specific T-cell epitopes that are known today is still too low, with the possible exception of those for malignant MELANOMA, which remains the model system for tumour vaccines. There is an urgent need for more epitopes so that several tumour antigens can be addressed with one vaccine, and to expand the applicability to a larger number of different tumours. So, which strategies will allow us to define new tumour antigens and T-cell epitopes?

Tumour antigens

Tumour immunologists have defined five different classes of antigens that are associated with tumours (Table 1) and that can be targets of antitumour T-cell responses if short peptide fragments that are derived from their sequences are presented by HLA class I molecules. However, we still do not know if most of them are also capable of delivering antigenic peptides to MHC class II molecules. So, what are the characteristics of tumour antigens that make them useful in antitumour vaccines? Most important are tumour-specific antigens: cancer-testis antigens, protein variants that are created by somatic mutations within tumour cells, and proteins from tumour viruses. These are found only in tumour tissue1, and therefore do not encompass the risk of autoimmune reactions during immunotherapy. Antigens that are highly overexpressed in tumours are referred to as tumour associated and might also serve as targets of a cellular immune response. The occurrence of the antigen within the population is also a measure of how successful a vaccine might be. Ubiquitous or shared tumour antigens are expressed by most tumour-bearing individuals and include telomerase11 and survivin12, but antigens that are expressed by only a small percentage of tumours — such as the MAGE gene family13,14,15 — might also be used. So, in present clinical studies, tumour specificity and widespread expression have to be considered to further the development of tumour vaccines. An ideal treatment would include only tumour antigens that are expressed by the respective tumour.

Identifying T-cell epitopes from tumour antigens

Two experimental strategies are currently being used. One begins with a documented T-cell reaction against tumour cells, and characterizes T-cell specificity and the nature of the antigen that is recognized. The other uses known tumour antigens as its starting point, and uses the 'reverse immunology' approach to predict in silico T-cell epitopes from the sequence of such antigens, according to allele-specific peptide motifs of HLA molecules. T cells are subsequently induced by stimulation with synthetic peptides, and peptide-specific T cells are then tested for the recognition and killing of tumour cells (Fig. 1).

Figure 1: Strategies for the characterization of T-cell epitopes.
figure 1

Depending on the starting point, which is either a T-cell reaction (a) or a known tumour antigen (b), two different strategies can be followed. a | The specificity of a T-cell receptor can be characterized using expression libraries, in order to identify the relevant antigen and its T-cell epitope. b | Potential T-cell epitopes can be predicted from the sequence of a tumour antigen. Their ability to induce peptide-specific T cells is assayed, and the tumour specificity of such T cells is verified by assaying tumour-cell recognition. HLA, human leukocyte antigen; MHC, major histocompatibility complex.

Full size image

From T-cell recognition to T-cell epitopes. The classical approach is to investigate the specificity of T cells that recognize tumour cells in order to define the epitope (Fig. 1a). The first tumour-specific T-cell epitope that was identified was from MAGE1 (Ref. 16), and was presented by HLA-A*0101. It was defined by narrowing down the complexity of antigens that produced the immune response in the target melanoma cell. Thierry Boon and colleagues expressed human cDNA libraries that were generated from melanoma cells, in COS cells that were transfected with the HLA-A*0101 gene, until the T-cell reaction focused on a single gene. Using truncation experiments, a small region of DNA was identified that harboured the relevant T-cell recognition site17. Since then, the methods for expression cloning have been substantially improved; a very efficient expression library has recently been described, which enables identification of T-cell specificity after enrichment steps18.

An alternative approach, which also starts from T-cell function, uses biochemical methods to purify MHC molecules that present the relevant peptide epitope. After acid elution from the MHC molecule, the highly complex mixture of MHC ligands is fractionated by chromatography (capillary high-performance liquid chromatography) and the fraction that is recognized by T cells is then analysed by mass spectrometry19. The low number of tumour-specific peptides among thousands of inconspicuous peptides, the often low affinity of T cells for tumour epitopes, and the very low number of peptides, impede this molecular characterization. Less than 10 picograms of peptide have to be analysed, which requires highly sensitive analytical techniques. This strategy is very tedious and risky, for all these reasons, and such cases remain rare exceptions in tumour-epitope characterization.

From antigens to T-cell epitopes. Immediately after publication of the first HLA-allele-specific peptide motifs20, scientists used the motifs to predict T-cell epitopes from antigen sequences21,22. This approach has been termed 'reverse immunology'23 and usually starts from a known tumour antigen, but without pre-existing T cells (Fig. 1b). Epitope prediction has now been carried out for more than ten years, and most tumour-specific T-cell epitopes known today have been defined with the help of peptide motifs. Several programmes offer free epitope predictions for a wide range of MHC class-I- and class-II-restricted epitopes from different organisms — the most commonly used are BIMAS24 and SYFPEITHI. In addition, programmes have evolved during the past two years that offer the prediction of proteasomal cleavages25,26: an overview on prediction programmes is given in Ref. 27. The experimental verification of proteasomal cleavage sites28 or the selected potential epitopes starts with peptide synthesis. Synthetic peptides might be used to immunize transgenic mice to produce a T-cell response29, or to re-stimulate precursor T cells in vitro, which are derived from either healthy blood donors or tumour patients30. The final step to establishing tumour-specific T cells that have a distinct specificity is accomplished when the raised T cells recognize tumour cells that express the respective antigen. A direct demonstration of natural processing of the predicted epitopes can be achieved by the 'predict, calibrate and detect' method31, which uses synthetic peptide candidates as calibrants for a capillary-chromatography–mass-spectrometric system, and by comparing the results with those obtained after acid elution of naturally processed peptides from HLA molecules of tumour cells.

New antigens as a source of new epitopes

With the development of screening methods to identify tumour-associated antigens, huge data sets had to be evaluated. The first screening method, SEREX, uses antibodies from sera of tumour patients to determine potential tumour antigens32: all proteins recognized by such antibodies (but not by antibodies from healthy persons) have been included in the SEREX database (see online links box), which now contains many hundreds of antigens. Two other screening methods — PROTEOMICS33,34,35 and MICROARRAY ANALYSIS36,37 — investigate the differences between healthy tissue and tumour tissue at the protein or mRNA level, respectively. With this type of differential analysis, several dozen to several hundred tumour-associated antigens can be defined within one assay, depending on the experimental thresholds. After screening a series of samples, scientists need to interpret large amounts of data in the light of published work and database contents. Results from complementary methods have to be combined to obtain reliable information. For example, the combination of differential display at the protein level, and the recognition of antigens by patients' antibodies might reveal interesting candidate antigens38, as well as the combination of expression profiling using DNA microarrays with the analysis of peptides presented by HLA molecules of the same tumour sample. After establishing a new tumour-associated antigen, the search for T-cell epitopes from this antigen might start anew by employing reverse immunology36.

Help is required

Vaccinations that are aimed at recruiting cytotoxic T cells for tumour destruction or prevention have been successful but, unfortunately, the success has been limited to just a few cases. The response rates that have been documented in many clinical studies are encouraging, but must be increased. In addition to tumour-escape mechanisms, inefficient triggering of the T-cell system might be a reason for low response rates. In order to mimic the concerted action of the immune system against pathogens, epitopes that stimulate CD4+ T-helper cells have recently been included in antitumour peptide vaccination studies39,40,41. Ideally, the epitopes that are recognized by CTLs and T-helper cells should be derived from the same tumour antigen. However, although many tumour-specific CTL epitopes are known, only a very small number of HLA class-II-presented, tumour-specific T-helper epitopes have been characterized. Increased efforts are therefore being made to identify tumour-associated epitopes that are presented by class II molecules42,43,44,45, which would activate CD4+ helper cells to induce and maintain CD8+ killer cells46. In principle, the strategies that are used for the characterization of CTL epitopes can also be applied to helper epitopes. This holds true for the classical approach in particular, which starts from T-cell recognition, although transfection of MHC class-II-expressing antigen-presenting cells with expression libraries represents a specific problem. An elegant strategy directs the antigen of interest to the lysosomal/endosomal compartment, where antigen processing leads to the generation of MHC class-II-presented epitopes47,48,49,50. Performing reverse immunology for MHC class-II-restricted antigen presentation is far more difficult, for two reasons. First, the peptide motifs of MHC class II molecules are more degenerate compared with class I motifs and therefore do not allow for similarly precise predictions. Second, it is impossible, at present, to predict the proteolytic events that take place during antigen processing within the MHC class II pathway.

Diverse clinical applications: partial success

The first vaccination that was carried out with a tumour-specific T-cell epitope was in a melanoma patient. Since then, melanoma has represented the model system for cancer immunotherapy. So far, most T-cell epitopes from tumour antigens have been derived from melanomal or MELANOSOMAL proteins. With the growing number of T-cell epitopes identified, the studies have now been expanded to other tumours. Clinical studies have been carried out in more then ten tumour entities, including cancers of the breast, cervix, gastrointestinal duct, lung, ovary, pancreas, prostate, kidney and medullary thyroid.

Although many cancer patients have been treated successfully, these represent just a small percentage of all individuals who underwent vaccine therapy (Table 2). Most patients that were included in clinical studies did not benefit from treatment with peptide epitopes that were identified from tumour antigens. We can speculate about the possible reasons that lead to such a disappointing outcome, but it is important to keep in mind that terminal patients — who have already undergone therapies in which their tumours proved to be resistant — are most frequently included in trials. It would be very interesting to compare these results with a first-line immunotherapeutic treatment of low-risk patients.

Table 2 Outcome of vaccine trials performed with different modes of application of tumour-associated T-cell epitopes
Full size table

An additional factor is the individual differences between people (Box 2), which have never been assessed in their entirety. We are also still unsure of the optimal application of tumour-associated T-cell epitopes: the many possible methods of vaccination have never been compared systematically in humans. Even if we leave out viral or other vectors, proteins and DNA, and only consider the application of synthetic peptides, each protocol that has been applied — use of PERIPHERAL-BLOOD MONONUCLEAR CELLS (PBMCs) or DCs as carriers; peptide only, intradermal or subcutanous administration; or addition of different adjuvants, cytokines or helper epitopes — has a limited, but comparable success (Table 2). It is therefore impossible to state that one thing works and another does not. Most application protocols have been shown to induce T-cell responses in patients, but tumour regression has been observed in only a few cases.

New approaches, future directions

There is much scope for the successful application of tumour-associated T-cell epitopes in cancer therapy, as the proof of principle was established long ago: immune responses are induced in most cases, and most tumour types are potential targets of immunotherapy. The problems, however, include the unsolved question of optimal clinical application, unknown escape mechanisms of tumours, and the low number of T-cell epitopes that have been identified for non-melanoma tumours. It has often been argued that HLA polymorphisms are a significant drawback in tumour immunology, but this is not as hopeless as it might seem at first glance: the most abundant HLA molecules have been investigated thoroughly, and their peptide motifs or even crystal structures have been determined. Our knowledge includes peptide motifs of HLA molecules that are expressed in more than 95% of Caucasians.

Activities must now focus on the identification of more tumour-associated antigens, and T-cell epitopes that are derived from their sequences. Fig. 2 shows how many different methods can be used to obtain complementary information about the characteristics of a given tumour. The combination of proteomics and antibody screening has been reported38, and the combination of expression profiling, HLA-ligand characterization and T-cell assays from blood has been established in our own department (Weinschenk, T. et al., manuscript in preparation). A modern anticancer vaccine should not target one or a few antigens only. It is also important to leave the model system of HLA-A*0201, and to characterize T-cell epitopes that can be presented by many different HLA class I or class II molecules. The most promising strategy seems to be an individual analysis of every tumour that is accessible, and individual treatment of every tumour patient (see also Box 2 and Fig. 2). The modern peptide vaccine should consist of tumour-associated CTL epitopes and helper epitopes from various tumour antigens that are restricted to more than two HLA class I molecules and one HLA class II molecule. Hopefully, these developments will lead to real benefits for patients with cancer.

Figure 2: Complementary information leads to the identification of useful tumour epitopes.
figure 2

As much information as possible has to be compiled about the presence of antigens or epitopes in a given tumour in order to identify new tumour antigens, and new tumour-associated T-cell epitopes that can be used in clinical studies. This strategy could lead to the characterization of an individual, patient-specific T-cell epitope pattern, which represents the prescription to treat this tumour. HLA, human leukocyte antigen.

Full size image