A target that worked, in a molecule that ran out of runway

What was tried

Amgen ran study 20190505 (NCT04631601), an open-label Phase 1/2 master protocol testing acapatamab, a half-life extended PSMA x CD3 bispecific T cell engager, in metastatic castration-resistant prostate cancer. It opened on 15 January 2021 across 14 sites and enrolled 55 participants, 54 of whom received treatment. Four subprotocols ran in parallel: acapatamab with enzalutamide, with abiraterone, with the anti-PD-1 antibody AMG 404, and as monotherapy. A fifth arm gave AMG 404 alone to 10 patients as an internal checkpoint-blockade reference.

Acapatamab was given intravenously, starting at 0.09 mg as a 72-hour extended infusion and stepping up to target doses of 0.15 mg or 0.3 mg. Primary endpoints were dose-limiting toxicities and treatment-emergent adverse events, with objective response, PSA response and circulating tumour cell conversion as secondaries. Primary completion was 23 October 2023, and the record carries the note "Amgen made a business decision to discontinue all AMG 160 clinical trials. This decision is not related to safety." Results are posted.

The biological hypothesis

FOLH1 encodes prostate-specific membrane antigen, a transmembrane glutamate carboxypeptidase whose expression rises on prostate cancer cells and climbs further after progression on androgen deprivation. Open Targets scores the FOLH1 to prostate cancer association at 0.883, clinical component 0.997 and genetic association 0.901. The target is already validated commercially: lutetium Lu-177 vipivotide tetraxetan is an approved PSMA-directed radioligand,.

The bispecific premise was that tethering PSMA on the tumour to CD3 on a T cell would produce cytotoxicity independent of any pre-existing antitumour immune response, which matters in a tumour type that responds poorly to checkpoint inhibitors. Preclinical work supported it. Acapatamab produced half-maximal lysis of PSMA-expressing lines at 6 to 42 pmol/L, drove regression of established 22Rv-1 xenografts at 0.2 mg/kg weekly, and showed enhanced activity with enzalutamide or anti-PD-1. That last observation is why this master protocol existed at all.

What actually happened

The trial was stopped by portfolio decision rather than a stopping rule, and the posted limitations statement says so: the study "was not stopped for safety reasons or lack of efficacy." Twenty-four of the 55 came off under "decision by sponsor" and 19 died on study, consistent with a heavily pretreated mCRPC population. Eight completed.

Two things are clear from the posted tables. The molecule engaged T cells in almost everyone. Cytokine release syndrome occurred in 41 of 44 participants in acapatamab-containing cohorts, was reported as serious in 13 of those 44, and occurred in none of the 10 who received AMG 404 alone.

Efficacy in the individual cohorts cannot be read as a rate, since most held three to six patients. The first-in-human study is the honest reference. In its 56-patient dose expansion at the 0.3 mg recommended dose, confirmed PSA50 responses occurred in 30.4 percent and radiographic partial responses in 7.4 percent. Median PSA progression-free survival was 3.3 months (95% CI 3.0 to 4.9) and median radiographic progression-free survival was 3.7 months (95% CI 2.0 to 5.4). Treatment-emergent anti-drug antibodies appeared in 55 percent of dose-expansion patients and reduced serum exposure in 36 percent.

openFDA FAERS holds two reports naming acapatamab, both serious, citing cytokine release syndrome, disease progression and uveitis. That is far too sparse for disproportionality analysis.

Failure mechanism, best guess

This is strategic reprioritisation, and the reason is legible in the earlier data rather than in this record. Amgen was choosing between two PSMA T cell engagers and kept the other one, AMG 340, which carries a deliberately low-affinity CD3 arm.

The design tension is straightforward. Acapatamab produced near-universal cytokine release syndrome, a step-up dosing requirement and an anti-drug antibody rate above half, in exchange for a 30 percent PSA50 rate and roughly three months of progression-free survival. High CD3 affinity buys potency and pays for it in cytokine release, exposure ceilings and immunogenicity. Published work on attenuating CD3 affinity in a PSMA bispecific showed tumour killing preserved with reduced cytokine release, which is the trade Amgen appears to have taken.

Target biology sets a second ceiling that no CD3 tuning fixes. A rapid autopsy series found no detectable PSMA in 13 of 52 cases, heterogeneous expression across metastases in 23 of 52, and at least one PSMA-negative site in 33 of 52. A separate analysis reported PSMA loss in 15 to 20 percent of men with castration-resistant disease, with lower expression in liver metastases and epigenetic silencing of the FOLH1 locus. A T cell engager needs antigen on the cell it is asked to kill.

How to prevent this next time

Endpoint-level comparative data for a formal quantitative model were never generated here, so the applicable levers are qualitative.

Choose the CD3 affinity before the clinic, not after two Phase 1 programmes. Both molecules entered patients and one was discarded. A head-to-head preclinical package benchmarking potency against cytokine release in the same models could have picked the format earlier and cheaper.

Enrich for antigen, and measure it. PSMA PET was used here for response evaluation, but no expression threshold gated entry. Given documented loss in a fifth to a quarter of patients, a baseline PSMA quantification requirement would have separated antigen-negative non-response from mechanism failure.

Treat anti-drug antibodies as a design endpoint. A 55 percent immunogenicity rate with 36 percent exposure impact caps achievable duration of response regardless of potency, and belongs in the durability model rather than in a pharmacokinetic footnote.

The single highest leverage change would have been to select CD3 affinity on a preclinical potency-versus-cytokine-release benchmark before committing two molecules to parallel clinical development, so that one PSMA engager entered the clinic in the right format instead of two competing for one portfolio slot.

What this means for similar programs

The read here is about molecular format and antigen coverage, not about PSMA. The target remains one of the best validated in oncology, with an approved radioligand built on it. What this record argues is that a T cell engager competing against a radioligand on the same antigen must clear a high durability bar, and three to four months of progression-free survival with near-universal cytokine release syndrome does not clear it.

Open questions

What fraction of non-responders in the first-in-human study were PSMA-low or PSMA-negative at baseline? Without that split, the 30 percent PSA50 rate cannot be attributed to mechanism rather than antigen coverage.

Sources

1. • ClinicalTrials.gov, NCT04631601 (Amgen 20190505), with posted results. https://clinicaltrials.gov/study/NCT04631601 - Dorff T, et al. A Phase I study of acapatamab, a PSMA-targeting bispecific T-cell engager, in mCRPC. Clin Cancer Res, 2024.https://- Deegen P, et al. AMG 160 in preclinical models of mCRPC. Clin Cancer Res, 2021.https://- Bakht MK, et al. PSMA heterogeneity and regulation in metastatic prostate cancer. Nat Cancer, 2023.https://- Sayar E, et al. Epigenetic alterations mediate PSMA expression heterogeneity in mCRPC. JCI Insight, 2023.https://- Attenuating CD3 affinity in a PSMAxCD3 bispecific. J Immunother Cancer, 2021.https://- FierceBiotech on Amgen dropping its BCMA and PSMA BiTE assets, 5 August 2022. https://www.fiercebiotech.com/biotech/amgen-takes-bites-out-pipeline-dropping-bcma-and-psma-assets-light-data-and-competition - Open Targets, FOLH1 (ENSG00000086205) and prostate cancer (MONDO_0008315). https://platform.opentargets.org/evidence/ENSG00000086205/MONDO_0008315 - ChEMBL, CHEMBL4802258 (acapatamab). https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL4802258/ - openFDA FAERS drug event endpoint, queried 27 August 2026.