Landscape of Tumor Antigens in T-Cell Immunotherapy

Abstract

Cancer immunotherapy is a rapidly evolving field that exploits T-cell responses to tumor-associated antigens to induce tumor rejection. Molecular identification of tumor rejection antigens has helped define several classes of antigens, including tissue-differentiation and tumor-germline antigens. The ability to genetically engineer antigen-specific receptors into T cells provides an opportunity to translate these findings into therapies. New immunotherapy agents, notably checkpoint inhibitors, have demonstrated unprecedented efficacy in certain cancers. Yet the nature of the antigens driving those beneficial immune responses remains unclear. New evidence suggests that tumors express immunogenic, tumor-specific epitopes generated from the same mutations that drive cancer development. Correlations between cancer types responding to immunotherapies and the frequency of somatic mutations may clarify what drives natural anti-tumor immune responses. This fusion of tumor immunology and genetics is leading to new ways to target this class of ideal tumor-specific antigens and could allow the application of immunotherapy to many cancers.

Adoptive T-cell transfer can mediate in vivo tumor regression as demonstrated by murine tumor models and clinical responses in trials using autologous tumor-infiltrating lymphocytes (TIL) derived from human melanomas (1). Further work has shown that the native T-cell response in TIL can recognize a variety of tumor-associated antigens (TAA) expressed in the context of MHC. However, difficulties in finding reactive T-cell populations for transfer for most other cancer types prompted studies to understand the molecular nature of the antigens recognized by clinically effective TIL. In parallel with these studies, highly efficient methods for genetically modifying human T-cells have been developed that have allowed the transduction of tumor-reactive T-cell receptors into lymphocytes for use in clinical trials. This review will also summarize the development of methods for identifying tumor associated antigens in the context of T-cell immunotherapy and present the prominent findings that have shaped our understanding of T-cell based immunological tumor rejection and pointed the way to new treatment approaches in cellular immunotherapy.

The molecular identification of human TAA began with the demonstration of tumor-reactive T-cells in cancer patients. These could be acquired either from peripheral blood lymphocytes (stimulated in vitro) (2) or from culturing tumor-infiltrating lymphocytes resident in many melanomas (3). The major factors that determine whether an antigen is a good target are 1) tumor-specificity 2) immunogenicity (ability to generate a T-cell response, often measured by tumor recognition, cytokine release, and/or cytolysis) 3) prevalence and level of expression on tumor cells. There are several strategies that can be employed to identify the antigens mediating T-cell recognition of tumor and selecting one approach versus another can lead to finding very different types of antigens.

Background

The earliest efforts to molecularly define the precise antigens on tumors responsible for T-cell recognition were based on the seminal work by the group of Boon (4). First working with murine tumors and then with human melanoma, they showed that genes encoding antigens could be identified by screening tumor DNA libraries by transfecting them into cell lines that expressed the proper MHC allele for presentation, but were not recognized. Then by adding the tumor-reactive T-cell clone from peripheral blood, the gene conferring recognition could be identified when it triggered TNF release from the T-cells. Although very labor intensive, this method was the initial way that a large number of such antigens were discovered by multiple investigators. It required starting with a tumor reactive T-cell, so most such antigens were associated with human melanoma. The observation that TIL from melanoma, when expanded in vitro with IL-2, frequently displayed autologous tumor recognition, made it a ready source of T-cells to be used in antigen cloning. In addition, the fact that human melanomas could frequently be established as stable tumor lines in vitro (unlike other common human epithelial cancers) facilitated the demonstration that T-cells could indeed be tumor reactive. Van der Bruggen was the first to identify MAGE-1, a cancer-germline antigen, as a melanoma-associated antigen recognized by T-cells (5). Tumor-germline antigens are characterized by their expression in the testis and on tumors. Epitopes from these antigens are not recognized in the testis, as those cells do not express Class I MHC, thereby making them attractive targets for T-cell immunotherapy (Table I). This was then followed with the discoveries that a series of proteins associated with pigment production in melanomas and melanocytes were frequently targets for T-cells. This latter class of antigens, often referred to as tissue differentiation antigens (normal proteins expressed as a consequence of a specific function of the target tissue), constituted the majority of TAA initially discovered. This was because in melanomas and melanocytes, they were very highly expressed and therefore were over-represented in cDNA libraries from these tissues. In addition, to avoid the burdensome task of making many cDNA libraries, if a tumor was cross-recognized by other allogeneic T-cells (via a common MHC allele), its cDNA library was often screened rather than make a new cDNA library from each patient's tumor. Kawakami et al have shown that tumor antigens can be found by screening cDNA libraries from allogeneic tumors (6). With the development of techniques for genetically engineering new T-cell specificities into PBL to target the tissue differentiation antigens, it was quickly realized that even tiny amounts of these antigens expressed by important normal tissues could cause unacceptable toxicities (Table I) (7). The melanoma-melanocyte family of antigens mediated severe skin rash and de-pigmentation as well as eye and ear inflammation (due to normal melanocytes at all these locations), while CEA targeting led to life-threatening colitis (7, 8). So even though anti-tumor activity was achieved, it was at a prohibitive cost. Subsequently, targeting this family of ubiquitous antigens has lost much of its initial appeal.

Table I.

Targeting of Tumor-Associated Antigens in T-cell Immunotherapy

Class	Advantages	Concerns	Examples
Tissue differentiation antigens	Shared antigens “Off the shelf” treatments can be developed	Expression on normal tissues Potential for on-target, off-tumor toxicity	MART-1 gp100 CEA CD19
Tumor germline (“tumor-testis”) antigens	Shared antigens “Off the shelf” treatments can be developed Potentially tumor-specific	Potential for on-target, off-tumor toxicity May be expressed in a low frequency of cancers	NY-ESO1 MAGE-A3
Normal proteins overexpressed by cancer cells	Shared antigens “Off the shelf” treatments can be developed	On-target, off-tumor toxicity	hTERT EGFR mesothelin
Viral proteins*	Shared antigens “Off the shelf” treatments can be developed Tumor specific, thus minimal risk of on-target, off-tumor toxicity	Low frequency of virus-associated cancers	HPV EBV MCC
Tumor-specific mutated antigens	Tumor specific, thus minimal risk of on-target, off-tumor toxicity Shared driver/hot-spot mutations can potentially be targeted	Currently requires surgical resection for next-generation sequencing Most immunogenic mutations identified so far are patient-specific Extended time to develop personalized treatment targeting mutations	Mum-1 B-catenin CDK4 ERBB2IP

^*Not included in this review

Mass spectrometry is another tool that several groups have used to successfully identify epitopes of tumor antigens that may be recognized by CTLs. Slingluff & Hunt's group demonstrated that tandem mass spectrometry with chromatography separation could be used to identify and sequence an MHC-bound peptide on a tumor line (YLEPGPVTA, gp100_280-288) that was recognized by several CTL lines (9). Utilizing this approach, they have also identified several predominant epitopes from MART-1 and gp100 (10). Kruger at al reported that they were able to identify many novel tumor antigens associated with renal cell carcinoma using peptide elution coupled with mass spectrometry (11).

In parallel with the use of expression cloning to identify tumor antigens that triggered T-cell responses, there was an effort to find TAA's recognized by B-cells (12). NY-ESO-1, a tumor-germline antigen, was identified using this approach (13). Total RNA was extracted from cell lines, normal and tumor tissues. A cDNA library was constructed from the tumor of a patient with esophageal squamous cell carcinoma. This library was expressed in E coli lysate and then screened for immunoglobulin binding from the patient's autologous serum. The reactive clones were then found to be from eight different genes, designated NY-ESO-1 – 8. Of these, NY-ESO-1 appeared to only be expressed in the testis and the tumor mRNA, and not on normal tissue by RT-PCR. NY-ESO-1 has been reported to be expressed by approximately 80% of synovial cell sarcomas as well as 10-50% of metastatic melanomas, breast, ovarian and lung cancers (14-16). The presence of a high affinity IgG serum response implies that there are at least CD4+ T-cell responses in these patients, and both Class I and Class II responses were subsequently delineated and cloned. A clinical trial using cultured autologous T lymphocytes transduced with a retrovirus to express a high affinity anti-NY-ESO-1 TCR, in patients with tumors expressing NY-ESO-1 was conducted (17). Eleven of eighteen treated patients with synovial cell sarcoma achieved objective responses with one sustained complete response. Eleven of twenty patients with melanoma had objective clinical responses; three were complete responses ongoing at ≥ 40 months.

Other tumor-germline antigens have also been identified and targeted in a similar fashion. After Boon and colleagues identified the MAGE-A1 gene in 1991, over twenty-five members have been identified in the MAGE family. MAGE-A is a multigene family that has twelve homologous genes (18). MAGE-A3 is one of the more frequently expressed TAA's in a variety of tumors, including melanoma (19). A high-avidity HLA-A*0201 restricted TCR against MAGE-A3 was generated by immunizing HLA-A2 transgenic mice with the MAGE A3 peptide: aa112-120 (KVAELVHFL). This murine TCR, which readily triggers human T-cell activation through association with human CD3, was then used in a clinical trial to treat HLA-A*0201 positive patients with metastatic cancer that expressed MAGE A3 (7). Nine patients were treated with a preparative chemotherapy regimen, infusion of anti-MAGE A3 TCR-modified PBL, followed by high-dose IL-2. Of these nine patients, five had objective clinical regression of their cancer. However, three patients experienced severe neurological toxicities. Further investigation showed that the anti-MAGE A3 TCR cross-recognized the homologous epitope on MAGE A12, also presented by HLA-A2. Further study showed that although MAGE-A3 could not be found in the human brain, MAGE A12 is expressed by a subset of neurons in the human brain. In summary, although the germline-tumor family of antigens remains one class of TAA to target with genetically engineered T-cells, their range of expression on normal tissues must be assessed for each candidate antigen and as with all normal-sequence self-proteins there is the potential for significant toxicities.

Targeting normal proteins over-expressed by cancers

A third class of normal self-proteins that have been targeted by T-cell therapies are proteins that are minimally expressed by healthy, normal tissues, but are constitutively over expressed by tumors as part of their malignant phenotype (Table I). These include EGFR, hTERT, p53, carbonic anhydrase IX and similar proteins. Because they are major contributors to the malignant phenotype, tumors cannot simply downregulate them and escape immune recognition and this makes them attractive targets. Yet they have normal functions in some tissues either at low levels or under select conditions. The successful targeting of this class of TAA relies on the concept of a “window” between tumor and normal tissue expression that might allow a safe and effective immune attack on cancers. Yet it is difficult to measure expression of any protein by every body tissue under every physiological condition and protocols targeting these molecules have demonstrated the potential hazards. This class of antigens has largely been targeted using novel chimeric receptors. Chimeric antigen receptors (CARs) are fusion proteins that incorporate antibody-derived antigen-binding regions and intracellular T-cell activation domains. As CARs recognize intact cell-surface proteins, CARs are not HLA-restricted and can be used to treat patients without regard to HLA haplotypes. CARs have shown good clinical activity when used to target CD19, a target that is only expressed on normal and malignant B cells. Treatment with a variety of anti-CD 19 CARs has shown a high rate of tumor regression, including complete regressions in chemorefractory patients with B-cell lymphomas and leukemias (20-22). In this case, the normal tissue expressing the antigen (i.e. the B-cell) is not essential to survival as patients can receive immunoglobulin and can be treated for infections.

This CAR approach was also used to target carbonic anhydrase IX, which is frequently over-expressed on clear cell renal cell carcinoma. A single-chain chimeric antigen receptor was constructed and then transduced into peripheral blood lymphocytes using a retroviral vector (23). Three patients were initially treated with sequential dosing of this cell therapy. All three patients developed hepatic toxicity evidenced by transaminitis and hyperbilirubinemia after four to five infusions, which resulted in cessation of the treatment. To further understand the liver toxicity associated with these cells a liver biopsy from the first patient showed cholangitis with T-cell infiltration around the bile ducts. Carbonic anhydrase IX was found to be expressed on bile duct epithelial cells (24). This represents another example of limiting “on-target, off-tumor” toxicity from targeting antigens that are minimally expressed in normal tissues. In another tragic example, a CAR targeting ERBB2 (HER-2/neu, derived from trastuzumab) caused fatal lung injury, thought to be triggered by low HER-2/neu expression in the lungs (25).

Identifying and targeting tumor-specific mutated proteins

A fourth class of TAA antigens that can induce an immune response are mutated proteins encoded by genes specifically mutated in tumors (Table I). These can be so-called “driver” mutations driving the malignant process or “bystander” mutations arising at random from either carcinogen exposure or tumor genetic instability. Although driver mutations are much more attractive as targets, tumor cells can be effectively killed by targeting any mutation, regardless of its oncologic significance. The first such naturally occurring T-cell responses to tumor-associated mutated epitopes were described twenty years ago.

Coulie et al first described in 1985, a tumor-specific mutation in MUM-1 that was recognized by CTL generated by a mixed autologous lymphocyte-melanoma culture (26). The mutation, which substituted isoleucine for the native serine residue at a TCR-contact residue in the epitope, actually occurred just beyond the end of exon 2 in an intronic region with the antigen being translated from an incompletely spliced RNA message. Other mutated gene products provoking T-cell responses have been identified including a CDK4 gene in a melanoma that resulted in a change from an arginine to a cysteine at amino acid 24 (27). This mutation created an epitope with an HLA-A2.1 binding motif not present in the native protein and allowed recognition by autologous CTL. Robbins et al. found a mutated B-catenin gene that encoded for a melanoma tumor-specific antigen (28). Screening the autologous tumor cDNA library identified a cDNA clone encoding B-catenin, with a single point mutation changing a serine to a phenylalanine at a TCR contact residue within an HLA-A2.1 binding 9-mer peptide. The patient's normal tissue did not express this antigen and the T-cell did not recognize wild-type B-catenin. Twelve allogeneic melanoma samples were also tested and this mutation was not identified, suggesting that this immunogenic mutation was both patient and tumor-specific. The difficulties in identifying such patient-specific reagents and translating them into therapy for each patient were daunting at the time and attention turned to shared, non-mutated tumor target antigens as described above. Yet as presciently noted by Coulie et al in their 1985 paper, “...antigens resulting from point mutations ought to be absolutely specific for the tumor, and technical progress may make the identification of such antigens so easy that treatment of patients bearing tumors with such individually specific tumor antigens will become feasible.”

Tumor-specific mutated proteins represent the ideal antigens for cellular immunotherapy because they are not only highly tumor-specific but could be very immunogenic because they are seen as “foreign” (i.e. not subjected to negative thymic selection, like normal self-epitopes). Advances in next generation DNA sequencing have accelerated studies on the immune response to this class of TAAs (29). Recently, melanoma TIL that recognized their autologous tumor with a known HLA-restriction were examined for the prevalence of T-cell reactivity to mutated epitopes. Whole exomic sequencing of tumor and normal DNA was used to identify nonsynonymous tumor-specific mutations. Then, for every mutated protein, every possible nine or ten amino acid epitope that could contain the variant amino acid (nineteen candidate epitopes for each point mutation) were vetted for potential binding to the presenting HLA allele using binding prediction algorithms (30). All potential epitopes were then ranked in order of predicted HLA affinity and only the top 25-40 were synthesized. T-cell recognition was assessed against these top binders by pulsing them onto HLA-appropriate antigen presenting cells and adding the tumor-reactive TIL. In all three of the melanoma TIL initially examined, anti-mutation reactivity was found against this highly select set of candidate-mutated epitopes (31). This provided a possible explanation for why successful melanoma TIL therapy did not induce any apparent autoimmunity and could work in some patients, despite not finding reactivity against the usual array of self-antigens. In total seven mutated epitopes in three patients were shown to be T cell targets for three TIL infusion products. Melanoma proved to be the most highly mutated human cancer type, presumably due to UV mutagenesis (32).

A parallel discovery of T-cells recognizing a mutated melanoma antigen in a patient with tumor regression in response to CTLA4 blockade has also been reported (33). Using whole exomic sequencing and HLA binding algorithms, T cell responses against two patient-specific mutant epitopes were found in this patient. Tetramer analysis of peripheral blood samples from pre-treatment and during treatment showed that after 5 weeks of ipilimumab the dominant T-cell response against an epitope encoded by mutated ATR (with a change from a serine to leucine at position eight and presented by HLA-A*03:01) had increased in frequency five-fold. A more recent review of mutated epitopes in patients receiving ipilimumab showed a significant difference in the number of tumor-associated mutations in patients with and without clinical benefit (34). Pembrolizumab, an antibody targeting the checkpoint receptor PD-1, has been associated with higher objective response rates and progression-free survival in patients with tumors that had a higher burden of nonsynonymous mutations. In one patient they were able to detect CD8+ T cells in the peripheral blood against a tumor-specific neoantigen and found that the frequency of this T cell population paralleled that of the tumor response (35).

In general, tumor histologies with the highest mutations frequency (melanoma, non-small cell lung cancer, bladder cancer) seem to be responsive to checkpoint inhibition with anti-PD-1 or anti-PDL-1 (36-38). In fact, smokers with NSCLC seem to respond better than non-smokers and have much higher mutation frequencies due to cigarette carcinogens (39). A recent phase I study evaluating the use of pembrolizumab in patients with NSCLC showed an objective response rate of 19.4% with a median duration of response of 12.5 months (40). The efficacy of pembrolizumab was also evaluated in patients with progressive metastatic colorectal cancer with and without mismatch-repair instability (MSI), which is associated with higher rates of genetic mutations. They an objective response rate of 40% in the cohort of patients with MSI and 0% in patients without MSI, which again supports the hypothesis that higher number of mutation-associated neoantigens may be driving responses to checkpoint therapy (41). However, one exception seems to be the high rates of response of clear cell renal carcinoma to IL-2 (42), anti-CTLA-4 (43) or anti-PD-1 (44) despite a low frequency of tumor-specific mutations. Overall, these findings have reinvigorated cancer vaccine efforts by incorporating tumor DNA sequencing into the selection of target epitopes (45).

Recently, a new method has been developed to identify immunogenic tumor-associated mutations uses minigenes to display mutated epitopes to T cells. In this approach, whole exomic sequencing of the tumor and normal cell DNA is completed to identify nonsynonymous mutations. The mutated amino acid residues and flanking twelve amino acids on both sides are encoded by a “minigene” of 75 nucleotides. Eight to fifteen different minigenes are then concatenated into longer tandem minigene constructs (TMG). These TMGs are cloned into an expression vector, transcribed as RNA and electroporated into the patient's own dendritic cells to display the 8-15 candidate epitopes in the context of all of the patient's MHC loci. T-cells (typically TIL) are then added and reactivity screened for. Using this technique Lu et al have been able to rediscover T-cells in melanoma TIL known to recognize tumor specific neoantigens as well as discover new mutated antigens not identified by standard expression cloning (46).

Significantly, this technique can be extended to tumors that were previously thought to be less immunogenic, such as gastrointestinal cancers. In a recent demonstration of the potential of this technology, mutation-specific CD4+ T cells were identified, isolated and administered to a patient with metastatic cholangiocarcinoma (47). A metastatic lung lesion was excised from this patient and used to generate TIL. Whole exomic sequencing was completed on the tumor as well as PBL, which identified twenty-six non-synonymous tumor-associated mutations. Minigenes encoding all 26 variant amino acids and the twelve flanking amino acids from the normal protein sequence on each side were created for each mutation and assembled into three TMG. Reactivity of the TIL against these mutations was detected by co-culturing the TIL with TMG-transfected autologous antigen-presenting cells and measuring IFN-gamma release. Reactivity was identified against one of the tandem minigenes and further evaluation showed that the TIL recognized a mutated epitope in ERBB2IP (Erb-B2-interacting protein) presented by MHC Class II. Prior to these discoveries, the patient had been treated with her bulk-expanded TIL without selection. In retrospect, approximately 25% of this TIL infusion of 40 billion cells were CD4+ cells reactive with mutERBB2IP. The patient had a minor response lasting 7 months before developing progressive disease. Three lung biopsies confirmed that the ERBB2IP mutation was still present in her relapsing lesions. Subsequently, a TIL product containing 120 billion T-cells with 95% of these being the same CD4 population reactive with mutERBB2IP was prepared and transferred. The patient had a much more dramatic partial response by RECIST criteria and 20 months afterwards is showing continued regression following this second treatment.

Conclusions

In summary, new evidence has suggested that cancers responsive to checkpoint inhibitors may be those with an abundance of potential tumor specific target antigens generated by a high frequency of mutations. Yet most cancers do not seem to respond to these agents, perhaps due to an inadequate repertoire of tumor-reactive T-cells. The administration of autologous T-cells derived from melanomas has shown that adoptive T-cell transfer can be a potent therapeutic modality capable of inducing complete and durable regressions of metastatic disease. Some non- mutated tumor-associated self-antigens have been effectively targeted with autologous lymphocytes gene-engineered with TCRs or CARs but there is a paucity of safe targets for those off-the-shelf reagents. The advent of methods to rapidly identify T-cells reactive with tumor specific mutations may allow selection or selective expansion of native T-cells for administration to change the repertoire of available tumor reactive T-cells. This latter possibility could open the door to safe and effective cellular-based immunotherapies for the majority of human tumors. Adoptive therapy protocols are now underway to test this possibility by identifying and administering mutation-reactive T-cells to a wide variety of patients with advanced cancer. Lastly, it is likely that the addition of reagents targeting the immunosuppressive tumor microenvironment could greatly augment outcomes from adoptive T-cell therapy and these combination regimens will be actively investigated in the near future.