Bright side of the dark genome: antigens for next-gen cancer vaccines

Phase I: whole tumor cell vaccines and their derivatives

The first generation of modern cancer vaccines emerged in the 1960s and centered around whole tumor cell approaches. Early studies involved transferring tumor fragments between patients to prime lymphocytes, which were then adoptively transferred back to the host.¹ Other strategies used whole-cell vaccines alone or combined with adjuvants or purified antigens such as GM2 gangliosides, with signs of immunogenicity and occasional clinical effects that commonly coincided with immune evidence.2,6

A major breakthrough came with the demonstration that tumor cells engineered to secrete granulocyte-macrophage colony-stimulating factor (GM-CSF) significantly enhanced antitumor immunity in murine models.⁷ This led to the development of autologous tumor cell vaccines expressing GM-CSF, collectively termed GVAX. In a phase I non-small cell lung cancer (NSCLC) trial, GVAX alone achieved a 9% complete response (CR) rate.⁸ Further analysis suggested that effective transduction—defined as secretion of ≥40 ng GM-CSF per 10⁶ cells—correlated with improved survival. However, the complexity and cost of autologous manufacturing prompted a shift toward allogeneic cell lines engineered to secrete GM-CSF, which were admixed with irradiated autologous tumor cells. In a follow-up study to the autologous NSCLC trial, where GM-CSF secretion was substantially higher, these formulations failed to improve clinical outcomes,⁹ and efforts to optimize GVAX manufacture continued with fully allogeneic GM-CSF-secreting tumor cell lines. A phase II trial using allogeneic prostate cancer GVAX showed limited benefit,¹⁰ but the subsequent phase III trial was halted for futility, and long-term outcomes were not assessed, preventing evaluation of potential benefit in subsets. A second phase III trial (VITAL-2) comparing GVAX plus docetaxel versus docetaxel/prednisone was also terminated early due to an imbalance in deaths in the GVAX arm, after initial analysis indicated lack of survival advantage.¹¹ Similarly, a double-blind phase II trial of autologous myeloblast GVAX post-allogeneic stem cell transplant in myelodysplastic syndrome (MDS)/acute myeloid leukemia (AML) did not improve survival.¹² Although GM-CSF was known to prime effector T cell responses, its optimal dosing and potential to expand regulatory T cells were underappreciated and may have contributed to failed clinical trials.¹³

Despite these setbacks, GVAX remains under investigation, particularly in pancreatic cancer. Here, vaccination was shown to induce T-cell responses against mesothelin, an antigen expressed by pancreatic cancer.¹⁴ While a phase IIb study of GVAX and a listeria mesothelin vaccine failed to improve outcomes,¹⁵ a neoadjuvant/adjuvant trio of GVAX, anti-programmed death-1 (PD-1), and anti-4-1BB (CD137) provided encouraging increases in median disease-free and overall survival.^{16 17}

Given the central role of dendritic cells (DCs) in orchestrating adaptive immunity, vaccines that pulsed DCs with tumor lysates, cancer peptides, or messenger RNA (mRNA) encoding tumor antigens have been explored.^{18 19} Although a small study reported immune activation and rare tumor regressions,²⁰ broader clinical success has remained elusive. The most notable success is sipuleucel-T (Provenge), an autologous ‘DC-like’ vaccine of white blood cells activated ex vivo with a fusion protein of prostatic acid phosphatase and GM-CSF. In a pivotal phase III trial, sipuleucel-T demonstrated a significant improvement in overall survival for metastatic castration-resistant prostate cancer, leading to its U.S. Food and Drug Administration (FDA) approval in 2010.²¹ Epitope spreading assessed by humoral immunity to non-targeted tumor-associated antigens (TAAs) was associated with improved clinical outcomes.²² Neoadjuvant treatment was shown to increase tumor-infiltrating lymphocytes (TILs) in resected prostate compared with the pre-treatment biopsy and also increased T cell receptor (TCR) diversity within resected prostate tissue compared with controls.^{23 24} While controversial, this represented a historic milestone as the first FDA-approved therapeutic cancer vaccine.^{19 25}

A related strategy centered on heat shock proteins (HSPs) as chaperones to ferry antigens to DCs. The autologous HSP-peptide complex vaccine vitespen (HSPPC-96), derived from patient tumors, was designed to use the receptor-mediated uptake of gp96 by DCs to drive major histocompatability complex-1 (MHC-I) cross-presentation.²⁶ Although a phase III trial in metastatic melanoma did not meet its primary endpoint,²⁷ another phase III study in renal cell carcinoma suggested a possible recurrence-free survival benefit in earlier-stage disease.²⁸ Other whole-cell strategies incorporated genetic modifications to enhance immunogenicity, including GM-CSF or CD80 expression, bacterial adjuvants, TLR agonists, and chemical modifications such as haptenization.829,32 Despite occasional signs of immunologic activity, most trials failed to produce clinical benefit.

Phase II: cancer vaccines dominated by CTAs, TAAs, and viral antigens

The discovery that MHC class I molecules present short intracellular peptides catalyzed a surge in efforts to identify cytotoxic T lymphocyte targets within cancer cells. This led to the identification of TAAs such as glycoprotein 100 (gp100), melanoma antigen recognized by T cells (MART-1/MelanA), human epidermal growth factor receptor 2 (Her2/neu), prostate-specific antigen (PSA), prostate specific membrane antigen (PSMA), mucin 1 (MUC-1), and cancer-testis antigens (CTAs), such as the melanoma antigen gene (MAGE) family, New York esophageal squamous cell carcinoma 1 (NY-ESO-1), Preferentially expressed antigen in melanoma (PRAME), Kita-Kyushu lung cancer antigen-1 (KK-LC-1 or CT83), and Synovial Sarcoma, X Breakpoint 2 (SSX2)—that were capable of eliciting CD8+ T cell responses.33,35 These findings triggered a wave of clinical trials focused on translating these antigens into peptide-based vaccine strategies, including those prioritized by an National Cancer Institute (NCI) pilot project listing 75 cancer antigens deemed ready for clinical development.³³ Enthusiasm for TAA vaccines declined after a phase III trial of a gp100 peptide vaccine in incomplete Freund’s adjuvant combined with ipilimumab showed no survival benefit over ipilimumab alone and a higher risk of progression (p=0.04). Although the combination improved median overall survival vs gp100 alone (10.0 vs 6.4 months), overall survival was no better than ipilimumab monotherapy (HR ≈1.04), indicating minimal added benefit from the vaccine.³⁶ Notably, the two gp100 peptides used in this trial were modified (gp100:209-217(210M) and gp100:280-288(288V)) to enhance HLA-A2 binding; however, such altered peptides bear the risk of generating T cells with suboptimal tumor recognition, as only a subset of the expanded T cells may cross-react with the native epitope on melanoma cells.^{37 38} Exploring a possible mechanism for these observations, preclinical studies revealed that peptide vaccines in incomplete Freund’s adjuvant (IFA) could misdirect vaccine-induced T cells to the injection site, promoting T-cell dysfunction and deletion rather than effective tumor trafficking and killing.³⁹ However, the clinical use of IFA in vaccine strategies has been buttressed by positive clinical trial data. For example, positive results from a phase III trial of gp100 peptide, in IFA, combined with interleukin (IL)-2 reported a significantly increased response rate (p=0.03) and progression-free survival (PFS) (p=0.008) over IL-2 alone.⁴⁰ While a positive trial, this approach was not advanced, likely the result of IL-2-associated adverse events and clinical complexity associated with IL-2 administration. However, many groups have continued to explore TAA vaccine strategies. A subsequent seven-arm clinical trial evaluated immune responses in patients receiving vaccines containing long peptides for seven different TAAs and several TLR agonists, administered with or without IFA, and demonstrated stronger immune responses in patients who received IFA.⁴¹ Here, the addition of TLR agonists was associated with an improved response, and there are other agents with adjuvant properties that, in preclinical studies, can improve vaccine efficacy that remain to be studied.^{39 42}

OSE2101 is another TAA vaccine with nine human leukocyte antigen (HLA)-A2-presented peptides covering HER-2/neu, carcinoembryonic antigen (CEA), MAGE 2, MAGE 3, and p53 epitopes plus one pan-DR peptide. In 219 patients randomized to OSE2101⁴³ or standard of care chemotherapy,⁴⁴ overall survival (OS) numerically favored OSE2101 but was not significant (HR 0.86, p=0.36). In patients with secondary resistance to immune checkpoint blockade (ICB) (n=118), OSE2101 significantly improved median OS (11.1 monthsvs 7.5 months; HR 0.59, p=0.017), postprogression survival, performance status deterioration, and global quality of life. Disease control and PFS were similar. Grade ≥3 adverse events were lower with OSE2101 (11.4% vs 35.1%, p=0.002).⁴⁵ Several trials are listed as active on clinicaltrials.gov, including one in combination with anti-PD-1.

Cancer antigens delivered via vaccinia and fowlpox viral vectors were a heterologous prime-boost tactic to stimulate immunity in the context of an anti-viral response. In one setting (TRICOM), the delivery vector, based on preclinical studies, was engineered to express CD80, ICAM-1, and LFA-3. An early study delivered TRICOM with PSA to men with prostate cancer and suggested a potential clinical benefit.⁴⁶ However, a phase III trial failed to show a survival benefit.⁴⁷ Similarly, the VANCE study, which evaluated a regimen using chimpanzee adenovirus and Modified Vaccinia Ankara targeting the oncofetal antigen 5T4 in early prostate cancer, demonstrated safety and immunogenicity, though clinical efficacy remains to be established in larger trials.⁴⁸ The same TRICOM strategy combined with two TAAs, CEA and epithelial mucin (MUC-1), combined with BCG, recently failed to provide clinical benefit in patients with bladder cancer.⁴⁹

mRNA is also being explored to deliver TAAs. Phase II data presented at European Society for Medical Oncology (ESMO) meeting in 2025 showed that FixVac (BNT111), an mRNA vaccine targeting four TAAs (NY-ESO-1, MAGE-A3, Tyrosinase, transmembrane phosphatase with tensin homology (TPTE), was safe and immunogenic, and when combined with cemiplimab, achieved a statistically significant overall response rate (ORR) of 18.1% in heavily pretreated programmed cell death protein 1 (PD-(L)1)-relapsed/refractory melanoma, exceeding the historical 10% benchmark.^{50 51} The combination produced a disease control rate of 55.3%, with median PFS of 3.1 months and median OS of 20.7 months. BioNTech has reportedly opted not to pursue further trials of BNT111. An alternative to targeting TAAs is to target immunosuppressive mechanisms. IO102-IO103 is a peptide vaccine targeting indoleamine 2,3-dioxygenase (IDO1) and PD-L1 that was evaluated with anti-PD-1 therapy in a randomized phase 3 trial of 407 treatment-naïve patients with advanced melanoma. Median PFS was 19.4 months with IO102-IO103 plus anti-PD-1 versus 11.0 months with anti-PD-1 alone (HR 0.77; p=0.0558), narrowly missing the prespecified significance threshold. PFS favored the combination across subgroups, including PD-L1–negative and high-LDH patients; in PD-L1–negative tumors, median PFS was 16.6 versus 3.0 months (HR 0.54). IDO1- and PD-L1-specific T cells were significantly expanded in the vaccine arm.⁵²

In parallel, viral antigens—particularly those from oncogenic viruses such as human papillomavirus (HPV) and Epstein-Barr virus (EBV)—have proven valuable as vaccine targets. The success of prophylactic HPV vaccination in preventing cervical and other HPV-related cancers demonstrates the potential of viral antigen-targeted immunization.53,55 It is notable that prophylactic vaccines like Gardasil and Cervarix target HPV’s capsid protein, induce B and T cell responses, but rely primarily, or exclusively, on neutralizing antibodies to prevent viral entry, whereas therapeutic vaccines must induce potent T cell responses to eliminate established, virus-transformed cancer cells.⁵⁶ Therapeutic vaccines for HPV-positive malignancies have primarily targeted the viral oncoproteins E6 and/or E7, using a range of platforms including peptide-based, mRNA-based, and DNA-based vaccines, as well as bacterial and viral vector delivery systems.^{57 58} Clinical trials in advanced cancer now combine therapeutic vaccination with ICB. However, randomized phase II trials of the synthetic long peptide (SLP) HPV vaccine plus anti-PD-1 therapy have not demonstrated added benefit over anti-PD-1 alone in either recurrent cervical cancer⁵⁹ or recurrent/metastatic HPV-positive head and neck squamous cell carcinoma.⁶⁰ Notably, these data also highlight how limited our current understanding remains and underscore the need for improved immune monitoring: vaccinated patients with combined positive score (CPS) <20 had worse outcomes than those treated with PD-1 blockade alone, whereas patients with CPS >20 appeared to benefit.60,62 Similar approaches targeting EBV-associated malignancies have also shown promise. Trials using autologous DCs loaded with EBV antigens (latent membrane proteins (LMP)1/2) or EBV-transformed cell lysates (eg, KSD-101 vaccine) demonstrated immune activation and encouraging clinical outcomes.^{63 64} Notably, CRs were observed in a subset of patients with EBV+ hematologic malignancies. These studies underscore the potential of viral antigens as immunotherapeutic targets.

Phase III: the shift to neoantigen vaccines and the de-emphasis of TAAs

The advent of next-generation sequencing (NGS) enabled comprehensive identification of somatic mutations in tumor cells, catalyzing a major shift in cancer vaccine development. Neoantigens—derived from tumor-specific mutations not present in normal tissue—emerged as attractive targets due to their lack of central tolerance and lower risk of on-target, off-tumor toxicity. These advantages, coupled with clinical observations that tumors with high tumor mutational burden (TMB) responded better to ICB, reinforced neoantigens as a preferred class of vaccine targets.⁶⁵ This paradigm was further supported by evidence that adoptive transfer of neoantigen-specific T cells could mediate tumor regression, along with early clinical vaccine studies.^{66 67} In one of the first studies of personalized neoantigen vaccines, six melanoma patients treated in the adjuvant setting demonstrated robust CD4⁺ and CD8⁺ T cell responses, with associated clinical benefit.⁶⁸ A subsequent phase II trial⁶⁹ in patients with resected stage IIb–IV melanoma compared adjuvant therapy with a personalized mRNA neoantigen vaccine plus anti-PD-1 versus anti-PD-1 alone and demonstrated a favorable trend toward improved recurrence-free survival (p=0.053), which led to the initiation of a phase III trial.⁶⁹ Vaccination both amplified pre-existing neoantigen-specific T-cell responses and elicited de novo responses to vaccine antigens, with expansion of both CD4⁺ and CD8⁺ T-cell populations underscoring the potential of mRNA platforms to immunize against modified self-antigens.⁷⁰ Additional studies showed that personalized vaccines—delivered as mRNA or synthetic long peptides and administered in combination with checkpoint blockade—could elicit durable CD8⁺ T cell responses in patients with pancreatic⁷¹ and renal cell carcinoma,⁷² correlating with prolonged recurrence-free survival in vaccine responders.

However, translating this promise into effective therapy for advanced cancers has proven far more difficult. In a single-arm study of advanced solid tumors, a personalized vaccine (NEO-PV-01) containing up to 20 unique neoepitopes, administered with polyinosinic-polycytidylic acid-poly-L-lysine carboxymethylcellulose (poly-ICLC) and anti-PD-1, induced peripheral T cell responses but failed to improve objective response rates beyond what would be expected from anti-PD-1 alone.⁷³ Similarly, a vaccine targeting 20 shared neoantigens, given with local anti-cytotoxic T-lymphocyte associated protein 4 (CTLA-4) and systemic anti-PD-1, yielded no responses in 19 patients with advanced disease.⁷⁴ Another personalized neoantigen vaccine (iNeo-Vac-P01), though immunogenic, achieved only modest efficacy, with a median progression-free survival of 4.6 months in a cohort of patients with diverse advanced tumors.⁷⁵ Furthermore, the concept of shared neoantigens—recurrent mutations like Kirsten rat sarcoma viral oncogene homolog (KRAS) or tumor protein p53 (TP53) hotspots found across patients—has spurred interest in off-the-shelf vaccines, a strategy highlighted by recent reviews.^{76 77} Vaccines targeting these shared drivers aim to induce epitope spreading, where the initial T-cell attack on the target antigen leads to the release of secondary antigens, broadening the immune response to subclones.⁷⁷ However, critical questions remain: why do patients generate neoantigen-reactive T cells in the periphery, yet still experience tumor progression? Are these T cells failing to infiltrate the tumor, or are other immune suppressive mechanisms—such as exhaustion or antigen loss—limiting their function?

Beginning in the late 20th century, molecular, biochemical, and mass-spectrometry-based studies established that MHC molecules on tumor cells present a restricted, reproducible repertoire of peptides, some of which are shared across tumors and can serve as targets for tumor-reactive T cells.78,80 While these studies focused on non-mutated shared antigens, they rapidly led to the recognition that neoantigens, derived from tumor-specific mutations and therefore exempt from central tolerance, might represent particularly attractive and tumor-specific immunotherapy targets. Consistent with this concept, preclinical studies established that a tumor-expressed neoantigen can function as a unique and effective target for T-cell-mediated tumor rejection.⁸¹

Building on these foundational principles, an integrated strategy combining genomic sequencing, MHC-binding prediction algorithms, and mass spectrometry was developed to directly identify and validate naturally presented neoantigenic peptides.⁸² Despite this early framework for experimental validation, the rise of large-scale cancer genomics fostered the widespread assumption that most somatic mutations predicted to bind HLA would be presented on tumor cells. Subsequent tumor immunopeptidomics studies, however, demonstrated that only 1–3% of computationally predicted neoantigens are naturally processed and displayed by MHC class I molecules.^{44 83 84}

Although a number of manuscripts continued to suggest that the majority of naturally presented tumor antigens were non-mutated,85,87 a majority of the field had continued to focus on neoantigens as primary targets for anti-cancer immunity. Recently, two independent studies provided definitive evidence that, across both high-TMB malignancies such as melanoma and NSCLC, as well as low-mutational burden cancers such as pancreatic ductal adenocarcinoma, only a very small fraction—and in some cases potentially none—of predicted neoantigens are detectably presented on the tumor cell surface in the context of HLA class I.^{88 89} Even after accounting for potential limitations in mass spectrometry sensitivity, this striking disconnect between mutational load and the expressed immunopeptidome reinforces the now well-recognized reality that current neoantigen-prediction pipelines markedly overestimate the number of clinically relevant targets. These findings underscore a critical bottleneck in translating tumor genomic data into actionable immunotherapy targets and re-emphasize the enduring importance of direct experimental validation of HLA-presented peptides rather than reliance on in silico prediction alone. This disconnect also provides a mechanistic explanation for the long-standing clinical paradox in which robust peripheral neoantigen-specific T-cell responses frequently fail to translate into meaningful tumor control.

If neoantigens are rarely presented by HLA of cancer cells, what explains the success of anti-PD-1 in patients with TMB-high cancers? Does the cellular state that drives TMB affect presentation of other cancer antigens? Why have neoantigen vaccines combined with checkpoint blockade shown improved survival in the adjuvant setting? Could RNA vaccines provide broad immune activation that amplifies weak pre-existing responses to neoepitopes or TAAs? For example, a recent report suggested that patients with NSCLC who received a checkpoint inhibitor (CPI) and a SARS-CoV-2 mRNA vaccine had longer progression-free and overall survival than unvaccinated patients.⁹⁰ While this finding was not replicated by another study,⁹¹ those patients reportedly received either an mRNA or a subunit vaccine, potentially explaining the variance. In a report at ESMO 2024, and in a recent publication, Grippin and colleagues reported an increase in tumor PD-L1 expression post SARS-CoV-2 mRNA vaccination and a significant increase in OS across large retrospective cohorts of patients with NSCLC (p<0.001) or melanoma (p<0.003) who received the vaccine within 100 days of initiating CPI.⁹² Preclinical studies demonstrated that SARS-CoV-2 vaccination induces potent type I interferon signaling, promoting innate immune activation and CD8⁺ T-cell priming against TAAs and providing a possible mechanism of action for the clinical findings.⁹² Although retrospective and potentially controversial, these observations from large cohorts raise important considerations for interpreting the clinical effectiveness of mRNA-encoded cancer vaccines, including the roles of antigen selection and dose of mRNA administered. As COVID-19 vaccination appears safe in patients receiving CPI,⁹³ this remains an active area of investigation with substantial relevance to the field.

While recent studies suggest that a majority of neoantigens are not detectable on the surface of cancer cells, rare neoantigens are detected and can still serve as targets for anti-cancer immune responses. One study identified 1800 potential neoantigens from 5 patients and found 30 that were immunogenic and where the TCR could recognize the autologous tumor cells. However, other immunogenic neoantigen-reactive TCRs could not recognize the autologous tumor cells, raising the issue of whether the neoantigen was presented.⁹⁴ For example, a phase 1 study of a driver neoantigen vaccine (VAC85135) targeting calreticulin and Janus kinase 2 (JAK2) mutations combined with ipilimumab in myeloproliferative neoplasms failed to elicit strong immune responses or clinical benefit, illustrating the challenges of overcoming tolerance.⁹⁵ An alternative possibility that aligns with the clinical observations is that clonal neoantigens are edited by the immune system, preserved in limited numbers in early-stage tumors, but induce T-cell tolerance in advanced cancer. A further question is whether cultured tumor cells have a higher level of neoepitope presentation than tumor in vivo or the primary explant. There may be hints of this in the recent report of Apavaloaei et al,⁸⁸ but to our knowledge, a direct comparison of HLA-bound neoepitopes by the same fresh and cultured tumor cells has not been reported. However, the concept that cultured tumor cells are easier for tumor-reactive immune cells to recognize has been known for more than 40 years.^{96 97} On this basis, lysis of ‘fresh’—enzymatically digested and cryopreserved—human tumor cells was the gold standard in the Rosenberg Lab of the 1980s, when lymphokine-activated killer cells and TIL were being characterized and moved into clinical trials.^{96 98} While cultured autologous cancer cells are frequently the only available source of a patient’s cancer, it may be that studies exploring antigen expression on these cultured cells overestimate the repertoire of epitopes presented by cancer in vivo.

Phase IV: dark non-canonical antigens and TAAs in combination immunotherapy strategies

The fourth phase, which we are just entering, is built on the spectrum of recently discovered non-canonical tumor antigens and our appreciation for the importance of ICBs. Early efforts, including Johnson & Johnson’s systematic search in prostate cancer, mapped RNA-derived antigenic sources such as alternative splicing events, out-of-frame fusions, intron retention and related transcriptional alterations that generate measurable, immunogenic non-canonical transcripts.⁹⁹ As this landscape broadens, the tumor immunopeptidome now encompasses aberrantly expressed micropeptides and, importantly, non-canonical translation products that constitute a new class of shared antigens.^{87 100 101} These non-canonical antigens arise from diverse molecular events including translation of previously considered non-coding RNAs (eg, 5’untranslated region (UTR), 3’UTR, long non-coding RNA (lncRNAs), and circular RNAs (circRNAs)), out-of-frame translation (eg, from unconventional open reading frames (ORFs) or frameshifts), alternative splicing, and intragenic rearrangements (IGRs).87102,105 These non-canonical antigens are now emerging as promising cancer vaccine candidates that can be combined with adjuvants studied previously and combined with ICB. Concurrently, TAAs are being reinvigorated as clinically relevant targets, with recent evidence highlighting their value as predictive markers for ICB response,^{106 107} their success in combined vaccine-ICB strategies,^{108 109} their rediscovery as key targets for effective TIL therapy,^{110 111} and advanced characterization showing that certain TAAs, despite arising from canonical sequences, can exhibit tumor-exclusive expression.^{112 113}

New proteogenomic methodologies uncovered the dark antigen landscape

Mass spectrometry has been used to characterize the immunopeptidome, peptides presented by MHC, for several decades; however, the databases used to identify the sequences from the mass spectrometry readouts were primarily the sequences of canonical proteins. A comprehensive identification of the dark antigen repertoire required the application of advanced proteogenomic methodologies.¹¹⁴ Several groups introduced an integrative framework that combined whole exome, transcriptome, ribosome profiling, and high-resolution mass spectrometry to identify non-canonical peptides presented on HLA molecules of cancer cells.^{86 115} Another approach, Peptide PRISM, efficiently searches the mass spectrometry spectra database using a keyword data structure, ‘trie’, and the Aho-Corasick algorithm to identify peptide sequences from genomic and transcriptomic translations.¹¹⁶

In parallel, recent studies have underscored the critical role of next-generation computational tools in decoding non-canonical ORFs and analyzing extensive immunopeptidomic datasets.¹¹⁷ Deutsch et al harnessed extensive human proteomics datasets to decode non-canonical ORFs.¹¹⁷ Impressively, they found that at least 25% of non-canonical ORFs give rise to translated gene products. The establishment of the Human Immunopeptidome Project has enabled immunopeptidome-wide association studies, providing valuable insights into how antigenic variations influence responsiveness to vaccines and immunotherapies.¹¹⁸ It is notable that many short-lived non-canonical dark matter proteins, although structurally disordered and inherently unstable with rapid intracellular degradation, were five times more likely to generate a peptide presented by HLA class I than a canonical peptide.¹⁰⁴

While much is yet to be discovered, immunopeptidomic databases, including SysteMHC Atlas V.2.0, caAtlas, IEAtlas, HLA Ligand Atlas, and Peptides for Cancer Immunotherapy Database (PCI-DB) are cataloging tumor-specific antigens (TSAs) and non-canonical dark antigens, providing essential references for comparison with benign tissues.112119,122 Table 1 provides a summary of these key resources available to researchers for identifying potential vaccine targets.

Why are we only now discovering CD8 T cell reactivity to non-canonical dark matter antigens?

Early on, cDNA libraries were used to identify genes encoding antigens recognized by tumor-reactive T cells derived from human TIL or mice immunized with tumor cells.^{34 35 123} Why were these antigens not discovered then? One explanation is related to the short intracellular half-life of many of these proteins, which are rapidly degraded but stabilized in MHC class I.¹⁰⁴ While capable of being recognized by antigen-specific effector T cells, direct presentation of peptide, at best, is an inefficient mechanism to prime immunity. Efficient priming requires uptake of intact antigen by professional antigen-presenting cells that process and cross-present antigen in the context of appropriate costimulatory signals.¹²⁴ Thus, as a consequence of being short-lived proteins (SLiPs) and poorly cross-presented in vivo, it is postulated that endogenous responses are not induced. This hypothesis explains why the early whole genome screens using TIL in humans, or mice immunized with tumor cells, failed to identify these targets.¹²⁵ However, in vitro stimulation with non-canonical dark matter antigens that are SLiPs can drive robust T-cell responses that recognize cancer cells and that presumably could also be driven by vaccine strategies.^{85 101 103}

Origins of non-canonical antigens

The non-canonical antigens can derive from a large spectrum of the dark genome as well as fusions. Here, we will outline our current understanding of the origin of these novel epitopes, that in addition to serving as potential targets for immune recognition, in some cases, also have cancer-promoting activity.

Antigens from intergenic or intronic regions, or frameshift mutations

A transformative insight in recent years has been the realization that many ncRNAs, long regarded as functionally inert, can encode immunogenic peptides. An early study of B cell lymphomas revealed that 17% of the 14 498 MHC class I bound peptides were non-canonical proteins. Of these 2503 peptides, approximately 72% were termed cryptic proteins—proteins derived from non-canonical regions or frameshifts of canonical transcripts.¹⁰⁴ The remainder was novel isoforms. Peptides derived from the 5’UTR, frameshifts, and annotated non-coding transcripts were the largest contributors to the non-canonical immunopeptidome, with intergenic, 3’UTR, and intronic contributors making up the remainder. A high percentage of peptides from 5’UTR and frameshift has been seen in other studies as well as our own work.^{85 116 125 126} Interestingly, Zhang et al recently identified naturally occurring ‘complete T cell antigens’ generated by frameshift-derived polypeptides that link MHC class I and II epitopes to secure CD4+T cell help and broaden the immune response.¹²⁷ In parallel, Kina et al reported that triple-negative breast cancer (TNBC) exhibits the highest levels of aberrantly expressed TSAs (aeTSAs) among breast cancer subtypes. These aeTSAs can be derived from exons (canonical antigens), as well as introns and intergenic regions (non-canonical antigens), and are linked to enhanced leukocytic infiltration and improved patient survival, inferring that some of these antigens are recognizable by the host.¹⁰⁰

Antigens from lncRNAs, circRNAs, and pseudogenes

Barczak et al provided compelling evidence that peptides derived from lncRNAs, under the regulation of protein arginine methyltransferase 5 (PRMT5) and E2F transcription factor 1 (E2F1), are actively translated and presented on MHC‐I molecules.¹²⁸ These peptides have been shown to elicit potent CD8+ T cell responses and inhibit tumor growth in preclinical models. Furthermore, Huang et al reported that circRNAs can serve as robust templates for cryptic peptide synthesis. Their preclinical studies with circFAM53B revealed that the resulting peptides bind with high affinity to both MHC class I and II molecules, a finding corroborated by in vivo vaccination that demonstrated effective priming of antigen-specific cytotoxic T cells.¹⁰³ Importantly, microproteins derived from some supposed ncRNAs have tumor-promoting functions.129,139

Intron retention and alternative splicing

The landscape of tumor antigens is further diversified by aberrant RNA processing events, which are particularly pronounced in hematologic malignancies. In AML, for example, atypical transcripts arising from intron retention and alternative splicing have been identified as rich sources of immunogenic epitopes, and many of these may be long-lived proteins and effectively cross-presented in vivo. Ehx et al employed a rigorous proteogenomic strategy to identify 58 novel AML-specific TSAs that are widely shared, non-mutated, arise predominantly from non-coding regions, and generate highly immunogenic peptides. As increased numbers of TSAs are associated with improved patient outcomes, we would postulate that many of these TSAs are long-lived proteins, effectively cross-presented, and induce tumor-reactive T cells responsible for the improved survival.⁸⁷ Furthermore, in a seminal study, Kwok et al identified tumor-specific antigens arising from neo-junctions generated by aberrant RNA splicing events. Importantly, these antigens represent clonal, public neoantigens shared across diverse cancer types.¹⁴⁰

Transposable element

Transposable elements (TEs), including endogenous retroviruses (endogenous retroelements (ERE), human endogenous retroviruses (HERV), Long interspersed element-1 (Line-1)), and other repetitive sequences, compose about half of the genome, are normally silenced but can become derepressed in tumors due to epigenetic dysregulation.¹⁴¹ Estimates suggest that HERVs and HERV-like sequences alone may account for ~45% of the genome,⁴³ underscoring their substantial footprint. Their reactivation can lead to the expression of TE-derived transcripts and peptides that are absent in healthy tissues.¹⁴² TE-derived antigens can trigger a dual immune response: they may provoke a ‘viral mimicry’ effect through double stranded RNA (dsRNA) formation and type I interferon induction^{143 144} and provide novel peptides that are presented on MHC I, rendering them recognizable by T cells. Activation of endogenous retroviral elements has been correlated with enhanced T-cell infiltration and improved ICB responses in a murine model.¹⁴⁴

IGRs: a dark source of neoantigen in TMB-low tumor entities

Although TMB has long been used as a biomarker to predict ICB response, emerging data reveal that many patients with TMB-low tumors can also derive significant clinical benefit.145,148 For instance, in TNBC, high TMB is rare,¹⁴⁹ yet a substantial subset of tumors exhibit strong T-cell inflammation and responds favorably to ICB.150,153 Similarly, in ovarian cancer, where most tumors are TMB-low, results from the IMagyn050 trial show no correlation between TMB and ICB response.¹⁴⁶ In esophageal adenocarcinoma, only 2.4% of tumors are TMB-high, and TMB fails to predict either T-cell infiltration or therapeutic response.^{147 148 154} These findings point to the existence of alternative, previously unrecognized antigen sources driving immune activity in TMB-low tumors.

Recent work by Han et al sheds light on this question, identifying IGRs—cryptic, kilobase-level structural mutations within genes, culminating in exon duplications or deletions—as a novel and overlooked source of neoantigens in TMB-low cancers.¹⁰⁵ Importantly, IGR burden correlates strongly with T-cell inflammation across multiple TMB-low tumor types, including breast, ovarian, uterine, and esophageal cancers, and is associated with improved ICB outcomes.¹⁰⁵ Meta-analyses from the MEDI4736 trial in esophageal adenocarcinoma and the IMvigor210 trial in urothelial carcinoma further validate the predictive power of IGR burden, particularly in IGR-dominant or platinum-treated tumors.¹⁰⁵ Collectively, these observations suggest that IGR burden may serve as a biomarker of response, deepening our understanding of the antigen landscape in TMB-low, immune-inflamed tumors.

RNA editing-derived neoantigens

Furthermore, RNA editing, primarily A-to-I conversion mediated by adenosine deaminases acting on RNA (ADAR) enzymes, creates novel peptide sequences absent in the DNA, yielding neoantigens that are strictly tumor-specific. These editing events alter mRNA codons such that edited proteins harbor amino acid substitutions not seen in normal cells, as demonstrated by TCGA analyses that identified numerous tumor-restricted editing sites.^{155 156} Edited peptides are naturally processed and presented on MHC class I molecules, eliciting robust CD8+ T cell responses in vitro and in patients.¹⁵⁷

The hidden antigens—proteins that are short-lived and not efficiently cross-presented

SLiPs and defective ribosomal products (DRiPs) constitute a significant, though historically underappreciated, component of tumor immunopeptidome.¹⁵⁸ These rapidly degraded proteins—either arising from canonical coding sequences or from non-canonical, ‘dark genome’ regions—are frequently processed via the ubiquitin-proteasome system and efficiently stabilized and presented on MHC-I molecules.^{104 114 115 159 160} Some of these proteins, derived from cryptic ORFs, untranslated regions, lncRNAs, or other unconventional genomic elements, appear to be shared across tumors of the same or even distinct histologies.^{89 101} Their shared expression and immunogenicity, as well as data suggesting tumor-promoting activity, make them attractive candidates for immunotherapy strategies.^{129 133} While the focus has been on HLA class I epitopes, it is unlikely that dark antigens are restricted to class I MHC. Immunopeptidome databases such as PCI-DB, which catalog MS-identified MHC ligands across malignant and benign primary tissues, are enabling systematic identification of tumor-associated class I and class II–restricted canonical peptides, and we expect membrane-associated antigens will be characterized next . ¹¹² Given advances in the field, we expect the raw mass spectrometry data used to build the PCI-DB will soon be queried using ribo-seq or other databases for the presence of non-canonical “dark matter” antigens.

One approach to generating a vaccine enriched in SLiPs and DRiPs blocked the proteasome and captured tumor-derived SLiPs and DRiPs prior to their degradation. Vaccines developed with this strategy induced robust anti-tumor T cell responses that provided the first evidence of vaccine-induced cross-protection against otherwise antigenically distinct 3-methylcholanthrene-induced sarcomas, something whole tumor vaccines failed to do.^{161 162} This approach, applied as combination immunotherapy for allogeneic mammary cancers, as well as lung and pancreatic cancers, also provided therapeutic efficacy.^{163 164} A clinical version of this strategy, DRibbles Platform vaccine-001 (DPV-001), enriched for SLiPs and DRiPs and containing both canonical and non-canonical proteins from both adenocarcinoma and squamous/mixed histology tumor cell lines, is now being evaluated in the clinic.^{165 166} In a patient receiving DPV-001 as adjuvant treatment for NSCLC, a post-vaccination T cell response was detected to a non-canonical peptide contained in the vaccine and presented by the patient’s cancer cell line. This first report of an in vivo generated CD8 T cell response to a dark antigen supports the concept that vaccines can induce immunity to these dark antigens.¹⁶⁷ DPV-001 is also being combined with ICB for the treatment of metastatic/recurrent HNSCC with a reported tripling of responses and increased PFS and OS in a cohort of 18 patients.^{50 166}

Another emerging category of non-canonical antigens are T cell epitopes associated with impaired peptide processing (TEIPP). TEIPP peptides arise from signal sequences or other intracellular sources that bypass transporter associated with antigen processing (TAP)-dependent processing, and importantly, they are selectively presented on TAP-deficient tumor cells but not on normal TAP-sufficient tissues, despite deriving from ubiquitous self-proteins ¹⁶⁸. TEIPP-based vaccines have now entered clinical testing, with the first-in-human phase I/II trial of a LRPAP1 synthetic long peptide vaccine (TEIPP24) demonstrating excellent safety and strong immunogenicity in checkpoint-resistant NSCLC patients, with vaccine-induced LRPAP1-specific CD8+ T cells detected in 83% of evaluable patients ¹⁶⁹. These findings suggest that TEIPP antigens represent a promising strategy to target immune-escaped tumors that have acquired resistance to conventional immunotherapy.

Reinvigoration of TAAs as clinically relevant immunological and vaccine targets

Recently, the prevailing view that TAAs are not significantly associated with ICB response is also being challenged. The emergence of large-scale ICB trial datasets has spurred a reassessment of TAAs’ role in immunotherapy response, revealing their predictive value in immune-naïve contexts.¹⁰⁶ Wang et al developed a TAA burden (TAB) algorithm and demonstrated that high burdens of both known and putative TAAs¹⁷⁰ correlate with improved ICB outcomes, when the tumor immune microenvironment (TIME) exhibits low T-cell exhaustion.¹⁰⁶ The authors propose that in immunologically ‘hot’ tumors with preexisting T-cell immunity, an established antitumor response may lead to the immunoediting of TAAs, resulting in lower TAB and terminal exhaustion of TAA-reactive T cells. In contrast, ‘cold’ tumors with elevated TAB and minimal pre-existing immunity, TAA-reactive TILs could remain functional and responsive to ICB.¹⁰⁶ This association is most pronounced in tumors with low PD-L1 immune cell expression and tumors with low CD8 T cell exhaustion signature. This pattern was validated across four independent datasets, including urothelial carcinoma cohorts treated with anti-PD-1/PD-L1 and neoadjuvant anti-PD-1 trials in head and neck cancer, validating TAB’s association with ICB efficacy in non-exhausted TIME.¹⁰⁶ This observation offers potential insight into the tumor antigenic landscape that may explain why approximately 10–15% of patients with bladder cancer or head and neck squamous cell carcinoma—despite having PD-L1–negative, low–TMB profiles—still derive meaningful clinical benefit from ICB.171,173 Further supporting the immunoediting theory, pan-cancer analyses revealed that most tumor types with higher T-cell exhaustion exhibited lower TAB, suggesting that TAAs may be edited in tumors with established antitumor immunity.

Consistent with this finding, in melanoma, baseline levels of antibodies specific to MelanA/MART1, TRP1/TYRP1, TRP2/TYRP2, and NY-ESO-1, indicative of a T cell response to those antigens, were independently linked to ICB benefit.107174,176 In NSCLC, a CTA burden (CTAB) metric—summing the expression of 17 CTAs—was studied in 250 immunotherapy-treated patients.¹⁷⁷ Patients with high CTAB had significantly improved overall survival and objective response rates on first-line pembrolizumab monotherapy, even when TMB was controlled for, highlighting that CTAB appears to provide unique predictive information. Interestingly, a similar decreasing CTAB with increasing PD-L1 level was observed, consistent with the above observation from the pan-cancer analysis. It should be stressed that unlike neoantigens, TAAs are normal proteins that do not induce danger signal. For this reason, TAAs can remain unedited in cold tumors, and TAA-reactive immunity could be activated when danger signal is induced. Consistent with this idea, COVID-19 mRNA vaccine expanded TAA-reactive CD8⁺ T cells and improved survival when combined with ICI even in PD-L1-null tumors ⁹², possibly by providing the danger signal.

These biomarker findings are underpinned by growing evidence that TAAs are fundamentally relevant targets for antitumor immunity. In a study of 21 melanoma patients treated with adoptive TIL therapy, TIL recognition of gp100 was significantly associated with clinical response (p=0.024).¹¹⁰ More recently, a 2023 analysis of long-term survivors of TIL therapy revealed that their circulating T cells were predominantly reactive to shared TAAs, rather than patient-specific neoantigens.¹¹¹ These findings challenge the assumption that neoantigens are the sole drivers of effective antitumor immunity and suggest a role for TAAs. Furthermore, a peptide-based TAA vaccine tested with pembrolizumab induced responses in PD-1-naïve melanoma patients (67% ORR), although limited activity is observed in PD-1-experienced patients.¹⁰⁸ In NSCLC, the mRNA vaccine BI1361849, encoding six TAAs (MUC1, survivin, NY-ESO-1, 5T4, MAGE-C2, MAGE-C1), combined with durvalumab yielded a 29% ORR, with no added benefit from tremelimumab.¹⁰⁹ Importantly, despite concerns of possible adverse events when targeting shared antigens, toxicity from complex cell-based and TAA vaccines has been low, with only two adverse events observed per 1000 vaccine doses across 239 phase I trials.¹⁷⁸

Despite these advances, limitations in immunomonitoring strategies persist. Most studies rely on peripheral blood sampling, which inadequately reflects antigen-specific responses within the tumor. The variability in vaccine efficacy underscores the need for mechanistic studies examining how antigen selection and vaccine strategy shape intratumoral immunity. To accelerate progress, phase I clinical trials should incorporate serial tumor biopsies whenever feasible to directly assess antigen-specific immune dynamics in situ.