Glutamine-gated VCP inhibition in RB1-deficient osteosarcoma

What the work claims

This is a primary preclinical result, not a method or a position piece. The authors, a Shanghai General Hospital and Shanghai Jiao Tong University orthopaedic-oncology group, argue that loss of the RB1 tumor suppressor creates a druggable dependency on VCP (valosin-containing protein, also called p97), an ATPase that extracts ubiquitinated proteins for degradation and is a central node in endoplasmic reticulum (ER) proteostasis. They claim the dependency is gated by metabolism: VCP function depends on acetylation at lysine 615 (K615), glutamine feeding supplies the acetyl groups, and the VCP inhibitor NMS-873 works in part by suppressing that acetylation.¹

The bold move is to nominate a non-cell-cycle synthetic-lethal target for RB1 loss. Almost all prior RB1 synthetic-lethality work, such as Aurora kinase inhibition in RB1-deficient small-cell lung cancer, targets the cell-cycle machinery that RB1 directly restrains, at the cost of narrow therapeutic windows. A proteostasis dependency would be a different therapeutic class entirely.

How it works

The group first established clinical relevance. In their institutional osteosarcoma cohort (SGH-OS) RB1 deficiency occurred in about 45% of cases, and in the TARGET-OS cohort about 35%. In a retrospective cohort of 60 patients, 30 RB1-deficient and 30 non-deficient, the deficient group showed only progressive disease or stable disease after neoadjuvant chemotherapy, with no partial or complete responses, and worse pathological necrosis.¹

The screen itself was compact: 2,097 compounds at a single 2 micromolar dose, four days of treatment, comparing RB1-knockdown (shRB1) against control (shNC) cells in two osteosarcoma lines, HOS and MG63. NMS-873, a selective allosteric VCP inhibitor, was the only compound in the top ten of both lines. Validation ran through clonogenic assays, live-cell imaging, 3D spheroids, patient-derived organoids (PDOs), and a tibial orthotopic xenograft model in nude mice, where 20 mg/kg NMS-873 shrank RB1-deficient tumors without body-weight loss or organ toxicity. Three chemically distinct VCP inhibitors (CB-5083, DBeQ, UPCDC-30245) and siRNA against VCP reproduced the selectivity.¹

The mechanistic chain has three links. First, NMS-873 does not change VCP RNA or protein levels; instead it sits near K615 in the D2 ATPase domain, predicted by AlphaFold3 docking to interfere with acetyl-CoA binding at that lysine. Second, mutagenesis showed K615 is the dominant acetylation site: the acetylation-mimetic K615Q mutant reduced drug sensitivity, while the acetylation-deficient K615R mutant enhanced it, and in xenografts NMS-873 inhibited growth only in K615R tumors. Third, the acetyl groups come from glutamine. Untargeted metabolomics flagged glutamine metabolism; glutamine restriction phenocopied NMS-873 by reducing VCP acetylation and killing RB1-deficient cells, glucose restriction did neither, and glutamine supplementation partially rescued the drug effect. Downstream, treated cells accumulated aggresomes, induced the unfolded protein response (IRE1, XBP1, PERK, EIF2alpha, CHOP, ATF6, HSPA5), and died; two ER-stress-alleviating compounds partially rescued colony formation. Single-cell transcriptomics of 77,998 tumor cells found ER-stress and protein-folding pathways the most over-represented of 38 pathways consistently upregulated in RB1-deficient disease across four datasets. Combining NMS-873 with the glutamine-transport inhibitor V-9302 was additive in vivo.¹

Where a skeptic should push

The load-bearing assumption is that shRNA knockdown in RB1-wild-type cell lines models RB1-mutant osteosarcoma. It does not fully. Acute RB1 depletion in HOS and MG63 captures the proliferative phenotype but not the years of clonal evolution, genomic scar, and rewired chromatin of a real RB1-null tumor. The gold-standard control, an isogenic pair of endogenous RB1-mutant and restored lines, is absent. The clinical cohorts are retrospective and single-institution for the response analysis.

Second, the acetylation mechanism rests on inference more than structure. The NMS-873-to-K615 interaction is a docking prediction, not a co-crystal structure, and the paper never directly shows that K615 acetylation changes VCP ATPase activity. The direction of the logic is also unusual: loss of an acetyl mark inactivating the enzyme deserves direct biochemical confirmation. Third, the screen was single-dose and viability-only, so selectivity margins at clinically reachable exposures are unknown; the authors themselves note that first-generation VCP inhibitors failed on dose-limiting toxicity and frame RB1 stratification as a way to lower exposure, which remains untested pharmacokinetically. Fourth, PDO numbers are not stated in the main text, so the organoid validation is effectively anecdotal in scale.¹

Organoid screens and the nutrient confound

For organoid-based drug discovery the uncomfortable implication is that the result is real precisely where the field is careless. Standard organoid culture media are glutamine-rich, and this paper shows glutamine abundance directly suppresses the very vulnerability being screened for. A PDO panel run in off-the-shelf medium could under-call NMS-873 sensitivity in a patient whose tumor would respond, or, worse, report sensitivity that evaporates in the patient's actual metabolic state. Any organoid screen aimed at metabolism-linked targets now has to treat the medium itself as a dosing variable: glutamine concentration, glutaminase activity, and the timing of medium exchange become part of the assay's metadata, not housekeeping.

The opportunity cuts the other way. If a drug's mechanism is nutrient-gated, the organoid gives something a xenograft cannot: controlled access to the gating variable. Pairing a VCP inhibitor panel with graded glutamine restriction inside the same PDO converts a confound into a biomarker; the response surface across nutrient conditions is a richer patient-stratification signal than any single IC50. The V-9302 combination data suggest co-targeting the transporter, which is exactly the kind of combination hypothesis PDOs are built to triage cheaply.

The threat is generalization failure in the opposite direction. This is one tumor type, two engineered cell lines, and a handful of unquantified PDOs. VCP dependence of RB1-deficient disease is a hypothesis about a tumor state, not a property of RB1 loss in general, and extending it to small-cell lung cancer or retinoblastoma models on this evidence would be exactly the one-line, one-model overreach a reviewer should reject.

The bottom line

Established: VCP inhibition preferentially kills RB1-depleted osteosarcoma cells across 2D, 3D, PDO, and in vivo models, with concordant genetic evidence. Plausible but unproven: that K615 deacetylation is the causal switch, that the effect holds in endogenous RB1-mutant tumors, and that a biomarker-stratified exposure can sidestep VCP inhibitor toxicity. What would confirm it: isogenic endogenous RB1-mutant models, direct enzymology of K615 acetylation, and PDO response data with reported n and medium composition. What would break it: failure of endogenous RB1-mutant organoids to show the selectivity, or loss of the glutamine coupling under controlled rescue experiments.