Current target decisions: PRAME, CTAG2 and TBXT
Reviewed 2026-10-27, against pinned HPA v23 measurements and the current complete-reference funnel. That funnel selects 343 protein-equivalence groups / 358 gene loci, with 17,824 Unsafe and 1,696 Neither among 19,863 groups. Previous CTA labels are comparison annotations, not admission criteria or the candidate universe.
Here Unsafe means an expression-policy exclusion, not demonstrated clinical toxicity. A gene-level expression screen and a peptide/HLA/receptor-specific clinical judgment answer different questions. The latter must retain its scope.
Measurements responsible for PRAME and CTAG2 flags
The first boundary is the inclusive candidate ceiling; reaching twice that boundary supplies the stronger expression exclusion. RNA units below are native HPA nTPM. Supported cluster observations require at least 20 reads.
| Gene | Normal context | RNA | Reads | Candidate / stronger boundary | Current contribution |
|---|---|---|---|---|---|
| PRAME | Heart muscle fibroblasts | 12.0 | 72 | 2 / 4 | Strong RNA exclusion |
| PRAME | Colon undifferentiated cells | 20.3 | 96 | 4 / 8 | Strong RNA exclusion |
| PRAME | Adrenal bulk | 3.6 | — | 2 / 4 | Candidate hold only |
| PRAME | Vascular fibroblasts | 2.8 | 22 | 2 / 4 | Candidate hold only |
| PRAME | Heart bulk | 0.3 | — | 16 / 32 | Pass |
| CTAG2 / LAGE-1 | Bronchial basal respiratory cells | 24.9 | 58 | 2 / 4 | Strong RNA exclusion |
| CTAG2 / LAGE-1 | Bronchial club cells | 17.0 | 197 | 2 / 4 | Strong RNA exclusion |
| CTAG2 / LAGE-1 | Bronchial ciliated cells | 8.0 | 39 | 2 / 4 | Strong RNA exclusion |
| CTAG2 / LAGE-1 | Kidney bulk | 3.4 | — | 2 / 4 | Candidate hold only |
| CTAG2 / LAGE-1 | Adipose macrophages | 4.6 | 66 | 4 / 8 | Candidate hold only |
| CTAG2 / LAGE-1 | Adipose NK cells | 5.4 | 22 | 4 / 8 | Candidate hold only |
| CTAG2 / LAGE-1 | Heart bulk | 5.1 | — | 16 / 32 | Pass |
PRAME has Enhanced-negative heart IHC for cardiomyocytes, not fibroblasts. Negative colon glandular-cell staining does not establish a negative in the undifferentiated cluster. CTAG2 has Enhanced-negative broad respiratory-epithelial IHC, but the matcher does not certify coverage of each basal/club/ciliated subset. The high heart-bulk ceiling already reflects matched negative cardiomyocyte IHC. CTAG2's heart measurement does not cause its current exclusion.
PRAME also has an imported Accepted normal-protein finding for adrenal cortex. The underlying primary abstract reports low protein expression. Its stored review explicitly lacks full methods and donor-level assessment; it is not evidence of treatment injury. This older acceptance path is less structured than the new external-IHC validation criteria and needs a uniform quality review across all accepted findings before being treated as an adjudicated protein threshold. Lezcano et al., 2021.
How IMA203 changes the interpretation
The published phase 1 trial included 40 infused patients. Its specific PRAME–HLA-A*02:01 target was unquantifiable in 41 normal tissue types. Normal-cell testing found a very weak renal epithelial response in 1/2 donors; the paper reported no specific renal toxicity at that cutoff. These are substantial context-specific data, not a guarantee for every PRAME-directed therapy. Wermke et al., 2025.
The June 2026 sponsor update describes 33 melanoma patients at the recommended dose with manageable tolerability. It is a subset/follow-up report, not 33 independent additional patients; CRS, neurotoxicity and conditioning toxicities still occurred. ASCO 2026 sponsor report.
The appropriate interpretation is PRAME: clinically supported targeting in the IMA203 context, with normal-expression flags. The expression-policy label alone understates that evidence. Clinical evidence is stored categorically and numerically in clinical-context-evidence.csv, separately from the expression decision; unreviewed products/genes are not coded as negative.
TBXT and the corrected antibody rule
TBXT now passes the uniform strict-RNA fallback. An uncertain antibody disables negative-IHC threshold increases but no longer adds a second gene-wide veto. Removing that veto admits 52 groups, including TBXT; reliable/context-qualified positive protein findings and the RNA exclusions still apply to every gene.
The 54-1 study specifically challenges the HPA macrophage staining and reports negative normal lung. This supports a reagent-specific explanation, not universal adult absence. Independent adult intervertebral-disc findings remain a review flag: the 2017 series used RNA (60 specimens), IHC (56) and flow cytometry (5). Hamilton et al., 2015, Richardson et al., 2017. The full antibody comparison retains reagent identity, compartment disagreement and missing coverage. No external antibody has automatically been assigned an HPA Enhanced-negative grade.
Which rule changes are justified
- Applied: eliminate the redundant gene-wide antibody veto; retain the lowest RNA ceilings when IHC is unusable. Keep review flags in selected rows.
- Required before protein adjudication: use the same explicit reagent, specimen, donor, cellular localization and specificity criteria for older accepted findings and newly collected IHC. A negative in one cell type cannot erase a finding in another cell type of the same organ.
- Clinical interpretation: record peptide, HLA, product, exposure, follow-up, tissue assay coverage and attributable organ toxicities. Favorable evidence can qualify that target context without declaring the entire protein safe. This audit adds evidence, not a new per-gene admission override.
For scale only: PRAME's largest RNA-to-ceiling ratio is 6. Multiplying the entire linked grid by six would clear its RNA gates, but would leave the adrenal protein veto. The next power-of-two grid starts at vital/missing 16 nTPM (8× the current 2), and still leaves CTAG2's 24.9-nTPM bronchial observation above the candidate ceiling. Neither choice is clinically validated, and neither has been substituted into the current funnel merely to admit PRAME. A numeric threshold fitted to an outcome cannot itself establish that outcome's safety.
Reproducible evidence
Run python scripts/audit_cta_target_evidence.py after the current funnel builder.
Raw HPA clusters,
raw curated HPA IHC,
assessed cluster contexts,
assessed bulk contexts,
all assay peaks,
current decisions, and receipts
retain the underlying values and the rule that consumed them. The source review
does not change released classifications or the forbidden sequence reference.