Protein structure retracted after investigation into “highly improbable features,” journal calls it fraud

In 2010, a group of crystallographers immunologists and allergy researchers at the University of Salzburg published a paper in the Journal of Immunology claiming to have derived the structure of a birch pollen allergen.

That structure, however, caught the attention of Bernhard Rupp, an eminent crystallographer. In January of this year, Rupp submitted a paper to Acta Crystallographica Section F pointing out problems with it, which prompted the editors of the crystallography journal to contact the authors of the original paper a month later. Those authors, it turns out, agreed with Rupp, they write in a response to his paper published in the April 2012 issue of Acta Crystallographica Section F:

This manuscript presents strong evidence that the diffraction data of Bet v 1d (PDB code 3k78 ; published in the J. Immunol. paper) are not derived from a diffraction experiment and that the model of 3k78 contains some highly improbable features.

That, in turn, prompted the University of Salzburg to ask the Austrian Agency for Research Integrity (OeAWI) to investigate whether the “highly improbable features” were due to fraud

…on the part of author Robert Schwarzenbacher, the co-author solely responsible for the Bet v 1d structure and the crystallographic section of the J. Immunol. paper.

A report of that investigation is being prepared, according to the authors, but in the meantime, Schwarzenbacher confessed:

Author Schwarzenbacher admits to the allegations of data fabrication and deeply apologizes to the co-authors and the scientific community for all the problems this has caused.

But did he? The authors add:

Note added in proof: subsequent to the acceptance of this article for publication, author Schwarzenbacher withdrew his admission of the allegations.

Fatima Ferreira, who assumed responsibility as corresponding author of the paper when former corresponding author Gernot Achatz passed away last year, tells Retraction Watch:

I really have no explanation for that. We had elaborated the response and author Schwarzenbacher agreed to the text as it has been published now. However, later on, he contacted the Workers Union and the Union sent us a letter where he withdrew his confession. This is the reason for the note added in proof.

We’ve reached out to Schwarzenebacher and others involved in the case to find out more details, and will update with anything we learn.

In the meantime, the authors asked the Protein Data Bank to retract the 3k78 entry, which happened in February. The Journal of Immunology paper — which has been cited 11 times, according to Thomson Scientific’s Web of Knowledge, has not been retracted, however:

The main body of the J. Immunol. publication concerns the immunological study and the retraction of the crystallographic section does not affect the major conclusions. The Editors of J. Immunol. have been informed about the problem with the structure 3k78 . Co-authors Zaborsky, Brunner, Wallner, Himly, Karl, Ferreira and Achatz were in no way involved in the generation of the crystallographic data.

Ferreira tells us:

I think the crystallographic data should be retracted, since it was fabricated. However, the immunological data is solid. The conclusions in the paper can still hold without the crystallographic data.

My group has produced and characterized the recombinant allergen, which was used for the immunological experiments. The same batch was given to Schwarzenbacher for the crystallization, which obviously was never done.

(This raises a related question: What’s the right place to publish critiques of papers? You’d think it would be the journal in which the original paper was published, but we often hear from scientists who tell us that journals decline to publish their critiques because they lack space. That happened to a group that submitted a critique to Science recently who ended up publishing it in PLoS Genetics, we learned on Twitter. And Keith Baggerly and Kevin Coombes, who did much of the heavy lifting to bring down the Anil Potti oeuvre, ended up publishing much of their analysis in the Annals of Applied Statistics. We’ll probably come back to this issue, so we welcome all feedback.)

For their part, in an editorial titled “Another case of fraud in structural biology,” the editors of Acta Crystallographica Section F mince no words. They bemoan the fact that this is “another instance of scientific misconduct in the literature of macromolecular crystallography,” referring to an episode that forced about a dozen retractions:

The second painful insult, disclosed in this issue, was also the act of a single individual. While it seems to be limited to one structure, one journal, one institution and fewer colleagues, and may or may not attract the same amount of attention as the first, it is no less painful, no less disappointing.

The editors are clearly frustrated:

What motivates these hoaxes? It seems clear that the pressures on scientists early in their careers are so severe that a few are compelled to risk their careers in order to further them. The dilemma is perhaps more fathomable when one considers the publication and citation metrics academic departments now use to evaluate staff, the difficulties crystallographers face in attracting funds early in their careers, and the seemingly inexorable march toward commoditization of the crystallographic product. Can this be changed any time soon?

The editors cite the Validation Task Force’s new recommendations, published recently in Structure (which also had a retraction last year):

Where scientific publication is the concern, however, their impact will only be fully effective if all relevant journals follow the path of IUCr Journals and require that validation reports as well as coordinates and structure factors be made available for peer review upon submission. It is equally important that all relevant journals include at least one expert crystallographer among the referees for all submissions that describe crystallographic structure determinations, even if those structures are but one aspect of the paper.

That will take effort, they note:

In the current case, however, validation by re-refinement and electron-density evaluation seems to have been the key. To do this on a routine basis will put an extra burden on crystallographers who serve as referees, making development of tools to ease that burden another worthwhile contribution.

Fraud will be tough to beat, they acknowledge:

It is important to note, however, that in neither of these cases was a single frame of data collected. Not one. This alone demands a redoubled effort to produce a universal system for deposition and storage of original diffraction images.

Update, 11:30 a.m. Eastern, 4/2/12: First line edited to clarify that most of the people in the group were immunologists and allergy researchers. Schwarzenbacher was the only crystallographer on the paper.

Please see our update including the news of Schwarzenbacher’s firing.

15 thoughts on “Protein structure retracted after investigation into “highly improbable features,” journal calls it fraud”

  1. As for the “pressures on scientists early in their careers”, the suspect is this time a tenured professor. So, one can not so easily use this example as an argument that giving people permanent and (compared to postdocs) well-paid positions does away with the problem.

  2. it would be great if we had to deposit original diffraction data along with structure factors. the only problem is that structures often have gigabytes worth of images. considering that the cost of data is getting ever cheaper, perhaps it is time to pull the trigger on this.

    1. People will just write programs to fake the original diffraction data. While I agree this would be awesome, it won’t stop the true frauds.

  3. It is really reprehensible that another scientist should spend time and effort investigating fraudulent work of somebody like this Schwarzenbacher… This also highlights the fact that outside crystallographic field there is a dearth of knowledge needed to deal with x-ray structures. Most people take the pdbs as if they were physical entities, directly observable in the crystals… Nice to see that intital clue came from a check in the PDB-REDO. I strongly urge everybody who – like me – fancies playing around with “structures” to consult this resource as well as PDBREPORT first to get al least some idea how reliable and usable a given pdb file is.

    1. I absolutely agree with you that there should be more verification mechanisms available to check a structure, and that everyone should be checking for themselves. However, you also highlighted yourself that there is a large gap in terms of people who CAN analyze this data, and know what to look for in terms of recognizing data that just doesn’t match reality. I can look all day at reports, but when it comes down to it, I just don’t know what I’m looking for. I think part of the solution is finding a way to make crystallographic/structural data more accessible to non-structural scientists, whether through education, layman reports, or what have you.

      1. The problem is that a lot of people use pdbs for downstream analysis anyways, be it as simple as designing or interpreting “mutants” in the proteins of interest, or something as time-consuming as MD simulations of reaction mechanisms, etc. In this case PDBREPORT is a good start, because it doesn’t limit the statistics of the structure to the R-factors, which most people indeed would have trouble interpreting, but gives simple enough to understand analysis of unusual bond distances and angles, fit to Ramachandran plot, etc. It should give one a pause when the “structure” s/he wants to use for modeling, sports physically-implausible backbone and mauled side chains…

  4. The editorial mentions the commoditization of crystallography. With the wonderful success of structural genomics (and the ready solution of easily crystallizable and solvable structures) the assumption has grown that all structure solution is like this. Unfortunately it isn’t. In order to meet the structure-on-demand mentality, models are being built unsupported by density and, in this case, apparently being made up.

  5. I feel stricter technical control cannot overcome the basis of the problem. More emphasis should be put on moral issues in education of scientists and less emphasis should be put on competitivity. Don’t we basically want to solve problems nature casts to us together?

    1. I agree with this point. Unfortunately, to live it, you ahve to have the mindset that you will be honest no matter what money you might end up making either as a tenured profession whose projects worked or a measley research associte whose projects did not work. Many can live this way, but family pressures may make this a hard road to walk. I know my own mother values success and status in her son rather than die hard honesty.

      1. It seems then if we paid tenured professors less and research associates more we should expect to see an improvement in the quality of science.

        No argument from me there. Society has a tendency towards greater inequality of incomes and this inevitably leads to an increase in dishonesty in the upper echelons. Minimising income inequality should lead to a more honest society (and better science as an added bonus).

  6. In the comment-sections of Austria newspaper are to be found some discussions wether Schwarzenbacher should pay back a “Marie-Curie-Stipendium” worth 1.7 Millionen Euro. The first news spoke of a stipendium, but in facto it’s a grant for the time period from 2006 to 2010: . There is no connection to the fraud.

  7. I’ve been lurking in the reading room of Retraction Watch for a while and one thing that strikes me is the actual costs involved in dealing with falsified information.
    The journals, through their staff, spend time working on this – unless their staff are volunteers that is a cost.
    If it came from academia then the schools involved spend time dealing with it and those involved – another cost.
    Over time any co-authors are going to pay unknown costs due to their connection to a fiasco. Positions and opportunities lost or not offered due to a taint.
    Similarly: unsuccessful, or ineffective, research efforts spent by others relying on false data as a basis, a beginning or a lead could also be added to the accumulated costs arising from bad data if the bad data hangs around unexposed long enough.

    As technology improves capabilities to fake, as proliferation of diploma mills and predatory journals increases the ranks of entities willing to engage and foster this behaviour the problem gets worse.
    Those financial costs, pre World Wide Web, might have been easy to write off as a generic cost of doing science publishing – to be absorbed by the wider research world.
    I can’t but imagine that, in this new world we live in, that part of the budgets of many institutions grows and grows.
    Journals’ costs go up and subscription charges will rise.
    Add all that up and it means the growing costs are a burden to Everyone.

    If only there was some way to make the brazen fakers obligated to compensate the system financially . . . instead of what appears to be a free poke in the eye of science publishing if one chooses to try it.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.