A team of researchers led by Daniel St. Johnston, director of the Gurdon Institute at Cambridge and a prominent developmental biologist in the UK, has lost a pair of articles after finding that their data were unreliable. But rather than “correct” the record with subsequent papers, they’ve withdrawn the problematic work.
To our mind, this is a poster case of doing the right thing by science. We think the notices — as provided by the authors and reported by the journals — pretty much say it all, so we’ll let them speak for themselves, followed by some details St. Johnston shared with us.
The first article, “LKB1 and AMPK maintain epithelial cell polarity under energetic stress,” appeared in 2007 in the Journal of Cell Biology and has been cited 108 times, according to Thomson Scientific’s Web of Knowledge.
According to the retraction notice:
The editors of The Journal of Cell Biology have been notified by Dr. Daniel St Johnston and Dr. Jay E. Brenman that they and the other authors of the paper referenced above retract the paper. As a result of this retraction, no data in this paper should be cited in the scientific literature.
The authors state:
Although the identification of lethal point mutations in the Drosophila ampkα gene and the originally identified neuronal phenotype (Fig. 1) remain valid, in follow-up experiments we have discovered that the defects in polarity marker localization, in actin distribution, and in epithelial integrity reported for lkb1 and ampk mutant clones in the ovary follicular epithelium under starvation conditions (Fig. 2, A–C, and Fig. 4) result from an artefact. In short, follicular cell clones that should contain GFP can become damaged and create “false clones” of GFP-negative cells with the above-described defects. Bona fide lkb1 and ampk mutant clones marked using a different technique do not show these phenotypes. As a result, our previous conclusion that LKB1 and AMPK function are required to maintain epithelial polarity and epithelial integrity under conditions of energetic stress is incorrect. Our conclusion that expression of AMPKα-T184D can rescue these defects in lkb1 mutant clones (Fig. 5) therefore also is incorrect. We have further described this damage-induced artefact in a separate publication (Haack et al., 2013).
Not affected by this artefact are the following results reported in Mirouse et al. (2007): the identification and neuronal characterization of ampkα mutant alleles (Fig. 1); the absence of polarity defects in tend mutant clones (Fig. 2 D) and in ampkα3 mutant clones from flies grown on a glucose-only diet (Fig. 2 E); the localization of Cherry-AMPKα and GFP-LKB1 in wild-type follicle cells (Fig. 3 A); and the characterization of the anti-PhosphoT184-AMPKα antibody and its use to localize activated AMPKα in wild-type cells (Fig. 3 B). The ampkα transgenic Drosophila animals and ampkα alleles generated in the study remain suitable for use, and their description is now presented in a separate publication (Swick et al, 2013).
We apologize for any inconveniences that these erroneous conclusions may have caused.
The second paper, “Dystroglycan and Perlecan Provide a Basal Cue Required for Epithelial Polarity during Energetic Stress,” came out about two years later in Developmental Cell and has been cited 37 times.
As its retraction notice states:
In following up the experiments reported in this paper, we have discovered that the polarity phenotypes of Dystroglycan and perlecan mutant clones under starvation conditions shown in Figures 4, 5, 6F, 6H, and 6J are the result of an artifact. We now believe that these panels do not represent real clones with a polarity phenotype, but rather false clones caused by a damage-induced artifact that creates patches of GFP-negative cells that mimic the appearance of a mutant clone with apical-basal polarity defects. Bona fide Dystroglycan and perlecan mutant clones marked using a different system do not show this phenotype, and thus our conclusion that they are required for the apical-basal polarity of the follicle cells under conditions of energetic stress is incorrect. We have described this damage-induced artifact in an article in Biology Open (Haack, T., Bergstralh, D., and St Johnston, D. (2013). Damage to the Drosophila follicle cell epithelium produces “false clones” with apparent polarity phenotypes. http://dx.doi.org/10.1242/bio.20134671). We apologize for any inconvenience that this erroneous conclusion may have caused. The results pertaining to the characterization of null Dg alleles and their role in basal planar cell polarity and egg shape remain valid. In light of our new data, the results from the paper lead to the conclusion that neither Dg nor Dys has an essential role in follicle cell polarity.
St. Johnston, who has 30 papers that have been cited more than 100 times each, responded generously to our queries with the following:
We discovered the problem because Timm Haack and Dan Bergstralh in my lab spent two frustrating years trying follow up on the results in these two papers and could sometimes see the published phenotype and sometimes couldn’t. The two retracted papers reported that patches of epithelial cells mutant for ampk, lkb1 or Dg give no obvious phenotypes under well-fed conditions, but lose their polarity under starvation conditions. We initially thought that we weren’t seeing the loss of polarity phenotype consistently because we weren’t starving our flies properly, so Timm started to test every possible variable, but nothing made any difference. To their great credit, Timm and Dan eventually overcame their boss’s confidence in the published results and proposed that the loss of polarity in starved ampk mutant clones was an artefact of damage during the dissection. Their idea was that damage could cause green fluorescent protein to leak out of a group of wild-type cells to produce the appearance of a mutant clone marked by the loss of GFP. They also proposed that damaging the cells and leaking out their cytoplasm might produce the polarity phenotype that we had observed in the supposed ampk mutant clones. Once they had persuaded me that their hypothesis was plausible, we came up with an experiment to test this idea: Timm made ampk mutant clones that were marked by the expression of GFP rather than by its loss. What he observed was that all of the GFP-expressing, mutant cells looked normal, but some of the GFP negative (and therefore normal) cells occasionally showed the loss of polarity phenotype. This proved that ampk mutant cells do not lose polarity under starvation, and subsequent experiments confirmed the idea that the loss of polarity was due to damage. I should also add that the idea that damage to one cell could give a cluster of GFP-negative cells that looked like a mutant clone did not occur to us until Lynn Cooley’s lab published a paper showing that the follicle cells retain cytoplasmic connections with their sister cells after they divide (Airoldi et al (2012) J Cell Sci. 1244077-86), so that damage to a single cell could cause the loss of GFP from the whole group of interconnected cells.
When did he come to realize that a retraction was necessary?
This is a bit of a loaded question because I don’t think that the retractions were necessary, i.e. we were not obliged to do this. We could have just published our paper in Biology Open explaining the artefact and assumed that people who cared would notice. We decided to retract the two papers because we felt it was the right thing to do. Both of the retracted papers have been reasonably well cited, and we were worried that some people would not notice our paper describing the artefact and would continue to believe (and cite) these incorrect results. We therefore felt that the only way to ensure that this information came to everyone’s attention was by somehow linking the Biology Open paper to the original incorrect publications, and retractions were the only mechanism available to do this.
I am still ambivalent about our decision to retract, because there is a lot of stigma attached to a retraction and many people assume that it must be the result of some wrongdoing on the part of the authors, which is not the case in this instance. We were just misled by a very convincing artefact that has fooled several other groups as well. On the one hand, I am convinced that this is the right thing to do, as it is the only transparent way to set the record straight. On the other hand, many other incorrect papers are never retracted and I am worried that these two retractions will affect the careers of my co-authors, particularly my collaborators, Jay Brenman and Rob Ray. Jay and Rob were senior authors on the Journal of Cell Biology and Developmental Cell paper respectively, because they initiated each project and provided the key mutants that made the work possible. In each case, they contributed important data to the paper that are still correct, as all of the results affected by the artefact came from my lab. This has highlighted a flaw in the retraction process to me, which is that the whole paper is wiped from the record once it is retracted, including the bits that are still valid and useful. Indeed, this almost derailed the retraction of the Journal of Cell Biology paper. The situation was only resolved when Biology Open agreed to republish Jay’s part of the paper, so that the isolation and characterisation of the widely-used ampk mutants can still be cited.
It is nice to see that the authors took responsibility in this way, but could Letters to the Editor at both journals have addressed the issue without retracting the entirety of each paper? Would this have been an appropriate way to handle the situation and avoid the concern of “the whole paper is wiped from the record” that St. Johnston mentions.
This should become an epochal point in genomics/proteomics. What about the, as I suspect, thousands and thousands of papers where, as I suspect, an artifact produces not a flickering effect, but a perfectly stable and repeatable effect that is taken as a molecular mechanism of something? Should the titles of the papers like “The … protein regulates … in … cells” be always continued with “A repeatable artefact?”
The first great discoveries in molecular biology were done in much more controlled conditions. Now, the experiments most often are a mixture of in vitro and in vivo. In vitro, using isolated components, we can have an approximation to a pure system. In vivo, we can have a real response. But a mixture offers not the first and not the second advantage. Not the time yet for such research, or not the time for such comment?
Bravo. Excellent way to follow up and work in the system to keep the record accurate. It would be nice if the system provided another way to update existing work, but they’ve done what they could.
As St. Johnston notes, it is all very well “dong the right thing”, but the potential for damage to what should be enhanced reputations is there, particularly since we have so many so called “corrections” where there has clearly been something more than an artefact afoot.
The main issue is how this is viewed in the long term. One would hope positively, and that there is no stigma attached, only kudos.
One way forward that might change how such things are dealt with would be if papers were replaced by micropublications. Groups of authors could then build these into an ever shifting thesis. In this case, an artefact would simply change “position” and the conclusions would be altered accordingly. However, the “paper” may be here for a good while yet, although given the momentum generated by Open Access (PLOS is but 10 years old), change may be faster.