A fusion-junction vaccine in extraskeletal myxoid chondrosarcoma: what can be established today, and the capabilities that would change it

Abstract

Background. Extraskeletal myxoid chondrosarcoma (EMC) is a rare soft-tissue sarcoma defined by rearrangement of NR4A3, most often to EWSR1. The fusion junction encodes a peptide sequence absent from either parent protein, and because the fusion is the truncal driver it is present in every tumour cell and cannot be lost without loss of the driver. The sponsors of an individualised neoantigen therapy have announced a positive phase 3 result in resected melanoma, which makes the platform question timely for other tumours. Purpose. This paper neither predicts that an EMC vaccine will work nor argues that it will not. It reports what current instruments and access establish about the target, separates limits of the tumour from limits of method or access, and records for each movable limit what would move it. Methods. Junctions were derived at the transcript level from Ensembl exon structure, so the acceptor exon is retained whole including its 5' untranslated region. Class I binding was predicted with MHCflurry 2.1.4, models release 2.2.0, on a ten-allele panel for the junction screen and a 34-allele panel for the coverage scan, calling a peptide strong at a presentation percentile of 0.5 or below; class II binding with MHCnuggets on 23 class II alleles across DR, DP and DQ at 100 and 1000 nM. Coverage is the union carrier frequency of the presenting alleles over Allele Frequency Net Database records; the sampling model a pooled binomial would require does not hold, so no confidence interval is placed on it and three other readings are reported in its place: the exact within-locus form of the coverage expression, distribution-free Fréchet bounds on the between-locus dependence the formula assumes away, and the empirical spread of the same quantity recomputed inside each source population. Novelty was assessed by exact-match search against the UniProt reviewed human proteome including isoforms, with the unreviewed entries of the same reference proteome searched separately and reported separately. Clinical figures come from a curated EMC registry. No wet-laboratory data were generated. Results. Of 27 declared exon pairs, 5 are in frame, yielding 174 junction-spanning peptides and 11 distinct predicted binders of which 4 are strong; there is no pan-EMC epitope. Predicted coverage is a property of the screen as much as of the junction: the commonly reported EWSR1 exon 7 to NR4A3 exon 3 junction covers 8.5% on ten alleles, presented on HLA-B*15:01 alone, and 12.3% on 34, where the same lead peptide is also strong on HLA-A*30:02; pooling every in-frame junction gives 27.4% and 30.4% on those two panels. None is a ceiling. Coverage against the acceptance threshold is computed here as a continuous function rather than sampled: it is a step function, every step is one peptide-allele call, and the four steps below the conventional cut all fall inside a window 0.0844 percentile units wide that closes before the cut is reached. Across the span of cuts this field routinely uses it runs from 0% at a conservative 0.2 through 30.4% at 0.5 to 72.6% at 2.0, so the reported figure is one point on a curve rather than an estimate with a tolerance. It also moves with the predictor: an independently trained class I model over the same peptides and panel returns 8 presenting alleles and 45.2% at its own conventional cut, sharing only three alleles with the four above. In place of the confidence interval withdrawn from earlier versions, the pooled figure is bounded three ways: the exact within-locus form raises it by 0.33 percentage points, so it cannot be too high for that reason; Fréchet bounds place it in [17.5%, 29.9%] under any linkage disequilibrium whatever; and recomputed inside each of the 112 source populations that measured the whole presenting panel it has a median of 24.5% and a range of 0% to 66.0%. That last is the width a reader should carry: it is an order of magnitude wider than the other two, and wider than the panel and the predictor axes, though not than the threshold. 170 of 174 peptides are absent from the reviewed human proteome including isoforms and all 4 strong binders survive; the 4 that do not occur in an NR4A3 isoform, belong to the four aspartate-seam junctions, and cost one predicted binder. A near-self search at one to two substitutions places the lead peptide one residue from DMPCVQAQY in that isoform and two from a paralogue peptide, neither difference at an anchor, against a chance expectation of 0.02; no binder in the screen has a near-self neighbour differing only at anchors. On a class II panel widened from three DRB1 alleles to 23 across DR, DP and DQ, 44 peptide-allele pairs bind and one is strong, on DRB1*14:01, so combined CD8 and CD4 coverage is computable for the first time and is 1.8%. The candidate construct is a 15-residue synthetic long peptide carrying both arms. The four out-of-frame junctions, screened here for the first time, give read-through tracts of 9 to 31 residues before a premature stop, and their 10 strong calls are tighter than any in-frame call — the only peptides in this work surviving a conservative cut come from junctions that cannot encode the driver. What bounds each conclusion. Ten limits are enumerated in Section 3 and graded there as bounded by the disease, by current instruments, or by access. Each movable one is paired with the advance that would move it and the observation that would show it had arrived; no date is offered for any of them. Interpretation. The most defensible present statement is neither that the route is viable nor that it is closed, but that it is instrument-limited in identifiable ways, and that several of the numbers a reader would take as bounding it are bounding the screen instead. Two findings are offered as results. Seam-proximal peptides of four of the five in-frame junctions reproduce a sequence in a normal NR4A3 isoform, which withdraws a predicted binder and is a defect in the novelty filter that will recur at any breakpoint whose seam reconstructs an isoform boundary. And the coverage figures this route has been graded on move with the panel and the threshold by more than the distance between them. The paper also observes, of this programme's own route ledger and not of the field, that several priming-directed classes were excluded there for want of antigen supply while a vaccine is an antigen supply, so the combination was never graded here as a unit; the standing objection to the vaccine is not the one that observation answers. Predicted binding is a screen and not evidence of presentation, immunogenicity or benefit, and nothing here supports use of any agent outside a clinical trial.

Keywords