One sleuth, 54 more papers, and an antibody that was never going to work

Sholto David, a UK-based independent molecular biologist, says he has found at least 54 papers on cellular senescence whose methods sections list an antibody raised against beta-galactosidase from *Escherichia coli*, used to detect the mammalian enzyme of the same name ^[1][2][3]. That is his third reagent-level disclosure in a short span, and it locates the weak joint in a lot of published biology not in the statistics but in the bottle on the bench ^[6][7][8].

The sequence matters more than any single paper. David posted the senescence findings on the research-integrity blog For Better Science on 21 July ^[6]. Nature reports that came two months after he flagged hundreds of studies with a similar error, and three months after he and another researcher identified problems with antibody-validation images in the catalogue of Thermo Fisher Scientific, one of the largest suppliers, headquartered in Waltham, Massachusetts ^[7][8]. All three fall inside roughly a three-month window ^[15].

The biology is simple enough to check. Senescent cells stop dividing but stay alive, accumulate in tissue, drive inflammation and secrete proteins that damage neighbours ^[9]. Elevated beta-galactosidase activity is treated as a hallmark, so researchers use an antibody against the mammalian enzyme and read it out by immunostaining or western blot ^[10]. Whether that assay reliably identifies senescence is itself disputed, though some labs still use it ^[16]. Pointing an anti-*E. coli* antibody at mammalian cells, David says, "is not going to work"; he calls it "a big blunder", and argues that cross-kingdom reactivity is "far-fetched, and not supported by the theoretical or experimental evidence" ^[3][4].

Catalogue language gives cover. Jennifer Byrne, a cancer researcher at the University of Sydney, says some suppliers of *E. coli* beta-galactosidase antibodies note possible cross-reactivity with other species, but those disclaimers are often generic and most companies will not have experimentally confirmed cross-reactivity ^[5]. Whether the substitutions were deliberate or accidental is unclear ^[11]. At least one team's paper concedes the results were not what they expected ^[11]. Dan Liebermann, a retired cancer geneticist formerly at Temple University in Philadelphia, told Nature he used the *E. coli* antibodies by mistake in experiments reported in a 2010 *Cancer Research* paper ^[12].

Nobody has established that the flaw invalidates the 54 papers; David says it might in some cases ^[1]. The structural point stands regardless. Antibodies are the workhorse of protein experiments and a persistent source of failure, including labs unable to reproduce results with nominally identical antibodies and antibodies that bind proteins other than the advertised target ^[13]. Aled Edwards, a biochemist at the Structural Genome Consortium in Toronto, says reagents are routinely used poorly or inadequately characterised, and that validating an antibody before an experiment is time-consuming and expensive ^[14].

Edwards adds the consequence that should concern anyone building on the literature: as AI models are used to mine papers and nominate protein targets for drug programmes, wrong reagent identifiers in methods sections propagate into the predictions ^[17].