Covering your Tracks
is a good idea if you think people are on to you. Eric points us to the communication in Nature that brought into question PDB entry 2HR0 (background posts: here and here). The reply was also published in Nature with the authors standing by their data. The 30-40 Angstrom gaps are explained by citing a personal communication with W. A. Hendrickson as well as the presence of protein fragments that serve as disordered lattice contacts.
The possible location of the protein fragment is shown in the reply.
Below is a display of the protein packing without the ‘protein fragments’ that served as the lattice contacts.
The figure below, shows additional symmetry related proteins, gaps occur vertically (in the c-direction) and are noted with a red circle.
Artem notes that they could have done a better job fabricating the data. We all Most of us try to learn from our mistakes and correct them, if possible.
Learning from your mistakes:
A glaring problem with the 2HR0 structure was the existence of these gaps. You can come to your conclusions on whether you believe there are disordered lattice contacts. However, what would greatly undermine their credibility would be if they had deposited another structure that contained these unusual gaps.
This brings us to the structure 2OU1, but wait. This structure doesn’t have very large gaps…
They learned.
If you take a close look at the PDB entry you will notice that this structure was updated (see Deposition Summary: right side, under the picture of the molecule). The structure that was initially deposited with the PDB entry 1L6K.
Here is the textual comparison:
The major change is shown with the red arrow above noting that the c-axis was nearly cut in half.
Why update?
Covering their Tracks:
The crystal packing of 1L6K:
The gaps… ~30 Angstroms, sound familiar?
The crystal packing of 2OU1:
The gaps have been significantly reduced.
They learned from their mistake in the 2HR0 entry. The large reduction in the length of the c-axis results in more reasonable crystal packing.
Time line:
The PDB entry 2HR0 was initially released at the end of October in 2006. The initial correspondence questioning the structure and reply were published in August 2007.
The original structure PDB entry 1L6K was deposited in 2002. The update of this structure was in February of 2007. It would be interesting to know when the authors were contacted about 2HR0.
One explanation is that once they were contacted about the 2HR0 structure, they realized there was a similar issue with 1L6K and replaced it with 2OU1.
As with the hypothesis about the entry 1BEF, I do not have any proof that this is what is going on, but definitely thought it was worth mentioning.
What do you think? Sound reasonable?
Sili
December 14th, 2009 at 8:41 AM #
But why make such an incompetent packing in the first place? And doesn’t ‘fixing’ it just bring more attention to the implausibility of the original? Admittedly a doubling of the UC is not an unreasonable mistake to make, but not without doubling the content too.
Nat
December 14th, 2009 at 8:49 AM #
The letter challenging 2hr0 says “Received 22 December 2006″, so your hypothesis would appear to be correct.
Note that they were only able to get away with this (temporarily, anyway) because they didn’t deposit structure factors for 1l6k – if they had, and the structure factors were indeed Fcalc, the molecule in its phony lattice would still have had acceptable R-factors. In comparison, structure 1y8e (which appears to be related to 2hr0) has smaller but still noticeable gaps between layers of molecules, and this is one of the five for which data is available. It would have been interesting to hear their excuse for this – I do not think you could fit the supposed contaminant from 2hr0 in the gaps there.
Nat
December 14th, 2009 at 8:54 AM #
Sili: the amazing thing about all of these structures is how badly they were faked. They (or he, since we don’t know if any of the co-authors were complicit) didn’t even bother to idealize the geometry, let alone crystal contacts, prior to extracting Fcalc. The result: impossible geometry with a perfect fit to density.
Crystal-man
December 14th, 2009 at 11:57 AM #
http://www.proteinscience.org/details/journalArticle/110975/RosettaHoles_Rapid_assessment_of_protein_core_packing_for_structure_prediction_r.html
This publication discusses “gaps” in modeled versus crystal structures. All of the Pdb codes in question are outliers!! using this criteria.
XtalDave
December 15th, 2009 at 3:46 AM #
Strange that the apparently “new” crystal form in 2OU1 has EXACTLY the same R-factor after refinement and resolution.
IF one wanted to obsolete a structure and replace it with a better structure (rather than just add an additional pdb entry), you’d imagine that it would have superior resolution or better refinement statistics.
Devils advocate
December 15th, 2009 at 5:56 AM #
Sili: Cheating is worth it only if it’s so much easier than the real thing. C3 wasn’t straightforward at all, but making a plausible model, including crystal contacts, packing, ramachandran statistics etc. and proper noise is a PhD’s amount of work in itself. And being such an unrewarding task, I wouldn’t bother to do it right either.
Jim
December 15th, 2009 at 2:29 PM #
Many red flags call into question these structures from Dr. Murthy….. not just large gaps between molecules. For example, the Rosetta team’s analysis of poor atomic packing is significant.
I want to point out from recent experience that gaps in electron density due to one-dimensional disorder, as discussed in that Trame and McKay manuscript that Murthy cites in their C3b reply letter, is seemingly a real phenomenon. In one of our as yet unpublished projects, ~1/3 the a.u. along the c-axis is disordered and the molecular model lattice built (to decent Rfactor/Rfree) is not contiguous along this axis. After much work, the cell and symmetry are correct. Multiple crystallographers have had this diffraction data in hand and no better explanations have been forthcoming.
Sean
December 15th, 2009 at 8:38 PM #
Hi Jim and welcome!
Good reminder there are certainly other inconsistent features other than just the gaps.
Best wishes with that data set
Sean
December 15th, 2009 at 8:41 PM #
Hi XtalDave,
Exactly, strange indeed
Nico
December 16th, 2009 at 10:57 PM #
Sorry Devils advocate, but the communication in Nature pinpoints different unusual features in this model, other than just packing and correct geometry. One figure shows a solvent exposed part of the molecule which perfectly fits the 2fofc map at 2.5 sigmaa !!! A solvent exposed Trp side chain !!!! at 2.5 sigmaa !!!
Sean
December 17th, 2009 at 12:02 AM #
Hi Nico,
No need to be a Devils advocate
My goal with this post was to show the connection between the unusual packing feature between these two structures and not to repeat the Nature communication.
Aaron
December 17th, 2009 at 6:33 PM #
Sean, I think he was referring to a previous poster…
Eric Williams
January 6th, 2010 at 4:00 PM #
Boy, it sure would be nice if the PDB reported why an author chose to obsolete a model.