Gastrointestinal tumours account for a large share of global cancer incidence and mortality, and immunotherapy has arrived unevenly across them. Checkpoint inhibitors transformed outcomes in mismatch-repair-deficient colorectal cancer and in a minority of gastric and oesophageal tumours, yet overall response rates across the group stay low, and most patients still receive chemotherapy with modest survival gain. The limitation is not a shortage of targets but a shortage of targets the immune system reads as foreign. Tumour-associated antigens are largely self-derived and tolerised, so the repertoire that remains available for vaccination is narrow and dominated by molecules also expressed, at lower level, in normal tissue.
Neoantigens sit outside that constraint. They arise from somatic mutations, frameshift indels, aberrant splicing, intron retention and viral open reading frames, and they differ from self in ways that permit T-cell recognition without central tolerance and without the thymic deletion that removes most autoreactive clones. Identifying them is a pipeline rather than a discovery. Tumour and normal sequencing produce variant calls; epitope prediction estimates peptide-HLA binding and proteasomal processing; expression and HLA typing filter the candidate list; and immunopeptidomics or tetramer validation confirms that the predicted peptide is actually presented. Each step loses candidates and introduces error, which is why prediction accuracy remains the field's central technical problem, and why the gap between predicted and validated epitopes determines whether a vaccine is well aimed. Platforms handle the resulting list differently: synthetic long peptides are simple and scalable but depend on adjuvants and on uptake by antigen-presenting cells, dendritic cell vaccines present the full antigenic repertoire at the cost of manufacturing complexity, messenger RNA and DNA constructs encode multiple epitopes and fold into delivery vehicles readily, and viral vectors trade repeat dosing for strong innate stimulation.
Clinical experience is encouraging without yet being decisive. Trials in pancreatic, gastric and colorectal cancer have shown that vaccines generate antigen-specific T-cell responses, that combination with checkpoint inhibitors can deepen those responses, and that some patients achieve prolonged progression-free survival. A personalized peptide vaccine targeting KRAS mutations in pancreatic and colorectal cancer illustrates both the promise and the scale: most participants mounted mutation-specific T-cell responses, median recurrence-free survival reached 16.3 months, and patients with strong T-cell responses had not reached that endpoint while low responders averaged about four months. Such numbers come from small trials and invite over-reading, since patient selection, adjuvant choice and prior treatment differ between studies, but they do establish that the immunological mechanism works in human gastrointestinal disease and that response magnitude predicts outcome.
Three obstacles stand between mechanism and practice. Prediction remains imprecise, so some vaccines target epitopes that are never presented while missing ones that are. Tumour heterogeneity means a vaccine built from a single biopsy can miss clones elsewhere or later, which argues for multi-site sampling and multi-epitope designs covering trunk mutations shared across subclones and, increasingly, for dark antigens translated from non-canonical reading frames. Timing matters as much as composition: late-stage disease pairs a vaccine with an exhausted immune system and an immunosuppressive microenvironment, whereas adjuvant or neoadjuvant settings offer relatively intact immunity and a smaller residual burden. Cost and logistics constrain access further, because personalised manufacturing is slow and expensive, and a product that takes months to release fits poorly with a disease that recurs in weeks. Universal vaccines directed at shared neoantigens and conserved neojunctions, artificial intelligence-assisted multi-omics screening, optimised delivery systems and combinations with radiotherapy, adjuvants or microbiota modulation are the proposed responses, together with a consensus framework running from antigen selection to clinical practice. Standardising endpoint definitions across trials would help as much as any single platform improvement. Access, rather than immunogenicity, may prove the harder problem in the end, since a personalised product competes with treatments that are cheaper and simpler to deliver.
DOI:10.1007/s11684-026-1266-3