Hypothesis

Children who report more stress and depressive symptoms would have poorer blood vessel lining (endothelial) function by adolescence, as found in adults.

What failed

Following 203 children from age 7 to 16, the authors found no cross-sectional or long-term link between stress or depressive symptoms and endothelial function (all p > .05).

Broken assumption

Psychological stress in childhood already harms blood vessel function by adolescence.

Why

All relationships pointed in the expected direction, but none was significant in this healthy group without symptoms.

From the published paper: OLIVE et al. PLoS ONE, 2018. ↗CC BY

Hypothesis

Two large cancer cell line screens, GDSC and CCLE, would report consistent drug responses for the same drugs in the same cell lines, so either could be used to find biomarkers.

What failed

Gene expression agreed well between the two screens, but drug sensitivity was inconsistent for 8 of the 15 drugs tested in both, whichever statistic was used. Only 5 drugs showed moderate to good consistency.

Broken assumption

Large drug screens of the same cell lines give the same drug-response results.

Why

The drug-response measurements themselves differ between the two projects. The authors argue that drug-response assays need standardisation and validation.

From the published paper: Safikhani et al. F1000Research, 2017. ↗CC BY

Hypothesis

Low-dose IL2 raises regulatory T cells, which reverse protein loss in the urine in animal models of nephrotic syndrome, so it would lower protein loss in children whose nephrotic syndrome resists all other drugs.

What failed

In 5 boys given six monthly cycles, IL2 raised regulatory T cells in some, but protein in the urine and kidney function did not change at any point. One child had an asthma attack after the first dose.

Broken assumption

Raising regulatory T cells reduces protein loss in human nephrotic syndrome, as it does in rodent models.

Why

The effect of regulatory T cells seen in rats and mice did not carry over to people. The authors say this reduces interest in IL2 for this disease.

From the published paper: Bonanni et al. PLoS ONE, 2015. ↗CC BY

Hypothesis

Mitochondrial DNA released by dying cells would rise in the blood of people with HIV and could mark the cell death that drives chronic immune activation.

What failed

A new, accurate PCR test found no significant link between HIV disease status and mitochondrial DNA in plasma in acute or chronic infection.

Broken assumption

Mitochondrial DNA in the blood reflects the cell death that drives immune activation in HIV.

Why

The abstract does not give a cause. Mitochondrial DNA fell with age in people with well-controlled HIV.

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

Hypothesis

Ketamine's fast antidepressant effect works through mTOR signalling, so ketamine would change the mTOR markers reported by Li et al. (2010).

What failed

More than a dozen experiments by four independent company labs (AbbVie, Servier, Pfizer and Alkermes) failed to detect robust effects of ketamine on those markers.

Broken assumption

Ketamine's antidepressant action works through a robust activation of mTOR.

Why

The authors conclude that the effect seems to need special conditions that are hard to identify, and they stress publishing detailed methods.

From the published paper: Popp et al. F1000Research, 2016. ↗CC BY

Hypothesis

MicroRNAs in urine and blood taken at prostate surgery would predict which cancers are high-risk.

What failed

Three urine microRNAs (miR-16, miR-21, miR-222) predicted high-risk disease with AUC 0.75 in the first cohort, but the signature failed to validate in an independent cohort of 36 patients measured on a different PCR platform.

Broken assumption

A microRNA signature found in one patient cohort will predict outcomes in another.

Why

The first result did not hold in new patients with a different measurement method. The authors stress the need for independent validation cohorts.

From the published paper: Sapre et al. PLoS ONE, 2014. ↗CC BY

Hypothesis

Gene expression data from human and mouse tissues cluster by species rather than by tissue, as the Mouse ENCODE Consortium reported.

What failed

The original design put human and mouse samples on different sequencing runs, so sequencing batch was mixed up with species. After correcting for batch, the samples clustered by tissue, not by species.

Broken assumption

Human and mouse tissues differ more by species than by tissue type in their gene expression.

Why

The study design confounded sequencing batch with species, so the reported species clustering was an artefact.

From the published paper: Gilad et al. F1000Research, 2015. ↗CC BY

Hypothesis

Dexrazoxane, which binds iron and calcium inside cells, would reduce the heart damage that happens when blood flow returns after a heart attack.

What failed

In pigs given dexrazoxane or saline before reperfusion (5 per group), dead heart tissue was the same (77.2% and 76.4% of the area at risk), and troponin and markers of oxidative damage and inflammation did not differ.

Broken assumption

Blocking iron and calcium overload reduces reperfusion injury after a heart attack.

Why

The abstract does not give a cause. Metal chelation did not protect the heart in this early reperfusion model.

From the published paper: Kamat et al. PLoS ONE, 2016. ↗CC BY