True or False? »« Lunch

How To Read A Retraction

A retraction was published this week in Structure in relation to a crystal structure of a complex of a G protein-coupled receptor and a G protein:

In this paper, a cocrystal structure was described of the G-protein heterodimer Gb1g2 bound to a C-terminal peptide from the parathyroid hormone receptor-1 (PTH1R) at 3.0A ° resolution (PDB id 2QNS). A subsequent refinement was later deposited in the Protein Data Bank (PDB id 3KJ5). While this structure represents a new crystal form of the Gb1g2 heterodimer, because of the lack of clear and continuous electron density for the receptor peptide in the complex structure, the paper is being retracted. We apologize for any confusion this may have caused.

What does “lack of clear and continuous electron density” mean? And why is this only coming to light now? Was this an honest mistake, or faking data?

Comments

  1. fred5 says

    I’m not sure what any of that means either but I cannot help but be amused that Elsevier actually thinks I should pay $31.50 for the priviledge to access the full retraction on their website. (And that’s for only 24 hour access to boot.)

    I should have gotten into the publishing scam when I had the chance.

  2. kougaro says

    I’m no expert in crystallography either, but I can try to explain :
    the X-ray experiment gives you a diffraction pattern,
    which can be turned into an electron density map using mathemagic (group symmetry, fourier transform, that kind of things).

    This electron density map is usually visualized as some kind of shell,
    in which you place the atoms of your molecules, as can be seen on this page : http://www.pdb.org/pdb/101/static101.do?p=education_discussion/Looking-at-Structures/methods.html

    Here, the data is high-quality, with a resolution of 1.7 A, and the fit is quite precise.
    In the case of the retracted article, the resolution is about 3 A, which is a lot less precise(and quite common for membrane-bound proteins, which are hard to crystallize), so the shell would be much larger : you would not have a precise position for each and every atom (see for example the second figure at that page : http://www.proteinstructures.com/Experimental/Experimental/electron-density.html)

    Since the data is not as precise, you have to interpret it. Usually this is done using programs that helps you place molecules in your electron density map, based on stereochemical informations, i.e torsion angle distributions, the geometry of the coordination shell of ligands, and so on.

    Also, the resolution is just one number, which is some kind of mean on the whole structure. Some parts of it may be very well resolved, while some others may be completely unresolved.

    So, “lack of clear and continuous electron density” probably means that that part of the map was particularly ambiguous, they interpreted it as being their receptor peptide. And apparently someone somewhere realized this was not well supported by the data.

    Also, it should be noted that it does not mean that they were wrong in their interpretation, just that they are not able to back it up using this data.

  3. Phillip IV says

    Can’t really say what it means, either – perhaps the “electron density” bit means they aren’t sure that a proton transfer isn’t taking place, so their presumed cocrystal might only be a crystalline salt?

    I can’t really imagine somebody faking something like this, though – it would be bound to come out sooner or later, and it’s not exactly an earth-shaking discovery to begin with. Probably just an honest mistake.

  4. Reginald Selkirk says

    What does “lack of clear and continuous electron density” mean?

    The way crystallography works is that you grow a crystal of something, then diffract X-rays through it. The part of the something that diffracts X-rays is the electrons; photons do not interact with nuclear particles. You collect a complete set of diffracted data, and find a source of phase information (X-rays, like other waves, have an amplitude and a phase. The phase value is lost because the detectors record the intensity, or amplitude squared.).

    So, once you have a fairly complete set of amplitudes and phases, you churn through a Fourier transform and get a three dimensional map of the electron density in the crystal unit cell. The diffraction pattern is the Fourier inverse of the electron density; the electron density is the Fourier inverse of the diffraction pattern. If you have good data, and good technique, then the electron density map should look like whatever your crystal is made of. In this case, it should look like G-protein heterodimer bound to a peptide. It should look enough like that that you can fit a model, and then refine the model, which improves the phases and gives you a better map.

    The original phases came from a technique called molecular replacement, which consists of starting with the structure of the G-protein and searching through various orientations until the theoretically calculated diffraction aligns well with the actual diffraction. So it’s no surprise that the G-protein fits reasonably well. G-protein structures are not news, it’s the complexes that are of current interest.

    The refinement process is iterative: refining the model improves the phases, and improving the phases improves the electron density map, which allows you to improve the model, and so on. A shortcoming of this is that the process is subject to model bias. If you stick something in there, it influences the data in that direction. In this case, they probably saw some noise in the map, figured it must be the peptide, and set out to refine it into the peptide. There are some safeguards to guide against such over-refinement, which I will not go into in detail (Free R factor, stats on quality of the geometry), other than to say that the stats will be dominated by the G-proteins, not by the relatively small peptide.

    So what you get to look at is a three dimensional contour map showing where the electrons were in the crystal unit cell. This is the blue ‘chicken wire’ visible in Figure 5, the only figure in which the actual map is displayed. It’s hard to make out in a 2D figure, its best to be able to turn it over in 3D in a viewing program. Apparently in 2008 the authors were convinced that the blob of density looked sufficiently like the peptide in question, and today they no longer think that. In 2008 they were able to convince journal editors and peer reviewers of their result. I have no idea whether the change of mind is due to honest error or deliberate fraud. All four authors of the original paper are listed on the retraction.

    After a quick spin with their data, 15 cycles of refinement with a common refinement program, the gap I see between R and Rfree(5.6) is more than twice that reported in the paper (2.3). In a viewing program, the electron density for the alleged peptide looks absolutely terrible. I don’t see how anyone, author or reviewer, could have been convinced of the success of the model. If it’s not deliberate fraud, they must have screwed up terribly and been to incompetent to notice the errors.

    Perhaps this has something to do with the recent G-protein complex published in Nature (2011) doi:10.1038/nature10361 SGF Rasmussen, et al. Perhaps something in the Nature paper result called into question the results of the earlier Structure paper.

  5. Reginald Selkirk says

    Also, it should be noted that it does not mean that they were wrong in their interpretation, just that they are not able to back it up using this data.

    But only two lines of data were put forward to back up that interpretation: the crystal structure, and surface plasminogen. The latter would tell you nothing more than that the peptide was bound, with no details of where it was bound or what its orientation was. So with the crystal structure retracted, they’ve got nothing left to stand on.

  6. Reginald Selkirk says

    Well, at least it appears that they did not simply fabricate the X-ray data. If they had done that, the map would have looked much better. This could be done pretty easily by constructing a completely fictional model, running it through the inverse transform, and adding noise. There would be telltale signs if anyone bothered to look closely enough.

    So I would guess the data is real, and its just the model fitting that was bad. But boy, was it bad.

  7. says

    Weird. Kougaro is likely correct inre the explanation given, but it’s odd that the paper would be retracted for a rickety electron density map. Sometimes you get tight structures, a lot of the time you don’t. And everybody recognises that there’s a certain amount of voodoo and circular reasoning applied in the analysis of these structures, particularly when the properties of the protein/protein crystal complicate generation of a solid unambiguous data.

    For retraction, I’d have thought there would have to be evidence of a straightforward cock-up somewhere along the line.

    At the end of the day, the test of a crystal structure is whether it gels with functional data, and vice versa. Being on the functional side, and studying ion channels, any crystallographic data is better than none, imho. Just leave it in the literature; if it’s bollocks we’ll find out soon enough.

Leave a Reply

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

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>