Engineered Salmonella kills tumours. A salt-triggered gel keeps it from poisoning the mice

The abdominal cavity is salty enough to set a gel. On that trick rests a delivery system reported in The Innovation: a small molecule that stays liquid in the syringe and, meeting sodium in the peritoneum, gels within a minute. Packed with attenuated Salmonella, it shrank melanoma tumours as well as free bacteria did, while cutting bacteria-induced weight loss by 82% and reversing the enlarged, damaged livers and spleens they normally leave behind.

VNP20009 has been the field's best candidate for two decades without delivering: a 1999 trial found it safe but weak, and strains armed with nanobodies against PD-1 or TNF-α raised potency and toxicity together. The bacteria colonise liver and spleen as well as tumour, and the response that kills cancer cells injures the organs filtering the infection.

The gelator, isosteviol-1,6-diester sodium sulfonate (1,6-DAS), is made from isosteviol, a diterpenoid from the sweetener stevioside. Its acid precursor stays liquid when cooled; sodium ions swap in and trigger ribbon-like fibres about 50 nm across that trap water. Gelation took under 60 s at half saline's salt concentration, and 100 μL held a billion colony-forming units — a thousand times the therapeutic dose.

Release is mechanical. Left still, the gel gave up almost nothing for 72 h; under gentle agitation mimicking an animal moving, about 80% of the bacteria came out. Tumour titres at 24 h ran below free bacteria and caught up by 72 h, while spleen titres fell and splenomegaly eased: target kept, off-target load cut. Two injections 24 h apart at the same dose reproduced the protection, so timing does the work.

The second mechanism is a coat. Released bacteria carried residual 1,6-DAS and a lower surface charge, and simulations put binding to flagellin at −7.96 kcal/mol for the D0 domain and −7.83 for D1 — burying the protein the immune system reads as Salmonella. Liver and spleen TNF-α and IL-6 fell to control levels.

The harder test was VNP-TNF-α nb, a strain carrying an anti-TNF-α nanobody that shrank tumours better and damaged spleens worse. Encapsulated, it cut weight loss by 78.8% (5.47 g to 1.16 g), restored spleen pathology to control levels and kept its efficacy, while dendritic cells and CD8+ T cells rose in tumour and blood. In a two-tumour model, four of five mice cleared the untreated side, against one of five given the free strain.

A meta-analysis explains the detour: across five trials, anti-TNF-α agents produced a partial response in only 10% of patients, with severe adverse events the strongest negative predictor. The gaps are acknowledged — mice, intraperitoneal delivery, no tracer to image the gel — so the claim that 1,6-DAS reaches dendritic cells in tumours rests on simulation. What it establishes is narrower: the ceiling on bacterial cancer therapy may lie less in the bacterium than in how it is delivered.