Hypothesis

Variants in the oxytocin receptor gene shape social preferences as revealed by incentivized dictator and trust games.

What failed

In 684 Swedish twins, none of the nine single nucleotide polymorphisms was significantly associated with behaviour in either game once multiple testing was corrected for, and the most significant association previously reported, between the amount donated in a dictator game and an OXTR variant, did not replicate.

Broken assumption

Common OXTR variants have effects on social preferences large enough to detect in economic games.

Why

The abstract offers no mechanism beyond the correction for multiple hypothesis testing.

From the published paper: Apicella et al. PLoS ONE, 2010. ↗CC BY

Hypothesis

With resources heterogeneously distributed, a species mixture should maximise productivity across a metacommunity and outperform every monoculture, while a balanced local resource supply should favour coexistence and higher evenness.

What failed

In an experiment with five freshwater algal species, mixtures did not outperform all monocultures at the metacommunity scale, so no general transgressive overyielding was observed; and contrary to the local-scale prediction, balanced resource supply did not generally produce higher evenness, though the lowest evenness was confined to the most imbalanced patches.

Broken assumption

Heterogeneous resource supply guarantees that diverse mixtures maximise productivity across a metacommunity.

Why

The authors note resource-use efficiency was nonetheless always higher in mixtures than monocultures predicted, suggesting resource partitioning or facilitation at high supply, and describe their support for the theory as mixed.

From the published paper: Gamfeldt et al. PLoS ONE, 2011. ↗CC BY

Hypothesis

A single experimental task could measure both the error-related negativity in EEG and pain-related avoidance behaviour, letting the two research fields be joined to test whether larger ERN amplitudes predispose people to avoid pain.

What failed

Across 12 participants and six successive task versions, no version elicited both the needed error-related negativity and avoidance behaviour; the manuscript documents each version's results and shortcomings rather than a working paradigm.

Broken assumption

The ERN and pain-avoidance paradigms can be merged into one task without each losing what makes it work.

Why

The authors describe the difficulty of guiding participants into the intended behaviour without their responding to the guidance itself, so that performance is natural yet tightly constrained, and the problems specific to each version.

From the published paper: Traxler et al. Journal of Trial and Error, 2020. ↗CC BY

Hypothesis

Pridopidine would slow functional decline in early-stage manifest Huntington's disease, measured by total functional capacity and the composite UHDRS at week 65.

What failed

Neither the primary nor the key secondary endpoint was met in the overall population: total functional capacity difference versus placebo was −0.18 (95% CI −0.49 to 0.14, P = 0.26) and the composite UHDRS difference was −0.11 (95% CI −0.40 to 0.18, P = 0.45).

Broken assumption

Sigma-1 receptor agonism slows functional decline in early manifest Huntington's disease.

Why

The paper reports a consistent pattern favouring pridopidine in a sensitivity analysis of participants never taking antidopaminergic medication, which it treats as a hypothesis, and notes the 45 mg twice-daily dose was well tolerated.

From the published paper: Reilmann et al. Nature Medicine, 2025. ↗CC BY

Hypothesis

Because genetic ablation of the CB2 receptor improved mouse survival in cerebral malaria, pharmacologically blocking CB2 with the antagonist SR144528 would attenuate the disease.

What failed

In P. berghei ANKA-infected C57BL/6 mice treated daily with SR144528 (confirmed CB2 binding, Ki 2.34 nM), 30% showed partial recovery of symptoms and survival rose by 20%, but the mice ultimately succumbed to hyperparasitemia and severe anemia.

Broken assumption

Pharmacological CB2 blockade reproduces the survival benefit that genetic CB2 ablation gives in cerebral malaria.

Why

The authors conclude that although part of the disease course may be modulated by blocking CB2, other elements drive the lethal outcome, so the hypothesis could not be fully validated in this model.

From the published paper: Escartín et al. Journal of Trial and Error, 2024. ↗CC BY

Hypothesis

Weekly transcorneal electrical stimulation would slow visual field loss in retinitis pigmentosa over a year, as an earlier exploratory study suggested.

What failed

In 52 patients randomized to sham, 150% or 200% of individual phosphene threshold and treated for 52 consecutive weeks, the primary outcome, visual field area, showed only a non-significant trend towards preservation in the 200% arm (p = 0.24); apart from light-adapted single-flash b-wave amplitudes, the other measures did not reach significance, and dry eye symptoms occurred in 31 of 52 patients.

Broken assumption

Electrically stimulating the degenerating retina preserves measurable visual field.

Why

The authors do not explain the primary-outcome null, resting their case for potential benefit on the electroretinography changes and on demonstrated safety and tolerability.

From the published paper: Schatz et al. Investigative Ophthalmology & Visual Science, 2017. ↗CC BY

Hypothesis

A light dark photon from an extra U(1) symmetry, with mass between roughly 0.5 and 1 GeV, would show up as a resonance in e+e- to U-gamma events with U decaying to two pions.

What failed

Searching 28 million e+e- to pi+pi-gamma events from KLOE at DAFNE found no signal; the result is a 90% CL upper limit on the photon-dark photon mixing strength for masses of 527-987 MeV, tighter than earlier limits above 700 MeV.

Broken assumption

A dark photon in the 527-987 MeV range couples strongly enough to be seen in the hadronic channel at KLOE.

Why

No signal was seen; the abstract does not interpret the absence beyond setting a limit.

From the published paper: Anastasi et al. Physics Letters B, 2016. ↗CC BY

Hypothesis

Treating predominantly central sleep apnea with adaptive servo-ventilation would improve outcomes in heart failure with reduced ejection fraction.

What failed

In 1,325 patients the primary endpoint did not differ (54.1% vs 50.8%, HR 1.13, P=0.10), and all-cause mortality was significantly higher with servo-ventilation (HR 1.28, P=0.01), as was cardiovascular death (HR 1.34, P=0.006).

Broken assumption

Suppressing central sleep apnea improves outcomes in heart failure with reduced ejection fraction.

Target engaged

Yes

Why

The device did its job on the target — apnea-hypopnea index fell to 6.6 events an hour — but suppressing central apnea did not help these patients and was associated with more deaths.

From the published paper: Cowie MR, Woehrle H, Wegscheider K, et al. (SERVE-HF). N Engl J Med, 2015. NCT00733343 ↗

Hypothesis

An LVS mutant that stimulates human antigen-presenting cells better than wild type would make a better tularemia vaccine, so in vitro screening of antigen-presenting-cell interaction could rationally pre-select candidates before animal work.

What failed

Strain 13B47 raised proinflammatory cytokine secretion from dendritic cells and macrophages, increased costimulatory molecule expression and activated T cells more efficiently than wild-type LVS, and a clean deletion allele reproduced those in vitro properties, yet vaccination with it did not improve survival after challenge with a virulent type A Francisella strain; the mutant was attenuated for growth in vivo and failed to stimulate comparable T cell responses in the lung.

Broken assumption

Better stimulation of antigen-presenting cells in vitro predicts vaccine efficacy in vivo.

Why

The authors conclude that improved antigen-presenting-cell stimulation in vitro did not correlate with protection against challenge in this model system.

From the published paper: Schmitt et al. PLoS ONE, 2012. ↗CC BY