Full statement of Homme Hellinga

A reader asked to see the full statement of Homme Hellinga about the ongoing debate over his study in which he designed proteins.

The entire statement is below. It is followed by a response to some questions from Looger, a coauthor on the disputed papers.

The manuscript submitted to PNAS by Schreier et al. reports studies on mutant periplasmic binding proteins (PBPs) that were constructed in my laboratory. In our original findings we reported that computationally designed mutations introduced into the binding sites of these proteins altered their ligand-binding specificity radically. These findings were based on changes in fluorescence intensities of single, environmentally sensitive fluorophores attached to mutant cysteine positions.
Discrepancies between the Schreier et al. observations and our work could have several origins. First, Schreier et al. need to make their observations at high protein concentrations (micromolarmillimolar), whereas the fluorescence assays use low concentrations (nanomolar). Second, their observations were made in the absence of conjugated fluorophore. We know that the intrinsic equilibrium between the open and closed state of PBPs can have a significant effect on apparent binding affinities, through thermodynamic linkage relationships. We have reported that this equilibrium can be manipulated experimentally by making mutations and attaching non-natural bulky groups, thereby significantly affecting binding affinities. In maltose-binding protein (MBP) this intrinsic equilibrium is very unfavorable to forming the closed state (measurements made by other groups). We now also know that this equilibrium is “tuned” by specific domains in many PBPs – we are finishing a manuscript describing these observations. Therefore, attachment of a fluorophore potentially can have significant effects on binding in the designed proteins.
We are investigating these issues for the designs in the Schreier paper, using techniques that can determine the free energies of protein stability and ligand binding using very low concentrations of protein. These techniques detect ligand binding by taking advantage of the thermodynamic linkage between protein stability and binding. This work is currently on-going, and will require some time to be completed, not least because some of the fluorophores are not available commercially and need to be resynthesized.
A second possibility (postulated by Schreier et al.) is that a subtle artifact in the fluorescence
observations gives the appearance that the changes in intensity are due to binding, but are in fact caused by another phenomenon. If we fail to observe binding in the studies outlined above, then we will draw the same conclusions. If that is the case, we will try and identify the physical phenomenon that gives rise to false positives in ligand-mediated changes of fluorescence intensities, so that it can be avoided in future work.
Schreier et al. make the important point that it is necessary to establish a set of independent
methods that allow observations to be corroborated. We have in fact carried out such a study on a different design: the design of an ibuprofen-binding site into MBP. We have completed the experimental observations for this work, and are finishing a manuscript describing it. It turns out to be a complex case study, because of some of the unanticipated properties of MBP and ibuprofen, which we had not appreciated when we started this work. Briefly, we observe interactions between ibuprofen and the mutant proteins by a) fluorescence of a conjugated fluorophore, b) isothermal titration calorimetry, c) magnetization transfer from the mutant MBPs to ibuprofen in NMR experiments, d) ligand-mediated changes in protein stability using the methods mentioned above. The apparent affinities are modest (200-400 micromolar). The ligand-mediated changes in protein stability are more complex than usual, because ibuprofen apparently acts as a mild protein denaturant at high concentrations. An X-ray structure reveals that the designed ecMBPs do not adopt the closed form, and that binding therefore appears to occur to a half-site. Our interpretation is that i) observation of ibuprofen binding in the designs is corroborated by very different techniques, ii) the binding mode is not as predicted, because the large intrinsic unfavorable equilibrium between open and closed states in MBP (see above) presents a significant barrier. These
techniques will be used in future design studies.
Schreier et al. also report that the side-chain positions in the designs are essentially as predicted in the original designs. This observation helps pinpoint the way forward in future design work: the problem is not that side-chains are packed incorrectly, but that we need to focus on the modeling of the energetics of protein-ligand interactions.
We greatly appreciate the extensive effort undertaken by Schreier et al. to characterize further these designed proteins. More studies are under way to understand the discrepancies between the sets of observations. If these studies also show that our original interpretations are in error, we deeply regret that our reports of these designed receptors do not live up to closer scrutiny. In that case, we offer our sincere apologies to researchers whose work was negatively impacted by these reports. We remain optimistic that the computational design of ligand-binding sites will become a reliable technique in the future.

Signed,
H.W. Hellinga
Department of Biochemistry
Duke University Medical Center
Durham, North Carolina
October 9, 2009

Looren Looger responds to questions from The Scientist:
1. Did Hocker send you a copy of this manuscript prior to publication? Were you
surprised by the results, or did you already have a suspicion that these proteins
didn't work quite as intended? Do you dispute anything in the results or the
interpretation?
Yes, she did, on July 21st, after it was submitted. I am not terribly surprised by the
results; I had become suspicious that the designed proteins did not work very well, due to
several lines of data. While at Duke, many in the lab noticed that the heavily-mutated
proteins seemed to suffer from poor thermodynamic stability: a fluorescence titration
result, although robust over several repeats at a given time, would not infrequently drop
off over the course of days or weeks, suggesting that something about the protein-dye
conjugate was changing, likely due to aggregation, unfolding, or precipitation. Second, in
my post-doc lab, we made the mutations to genetically-encoded FRET glucose and ribose
reporters that should have conveyed lactate binding (we wished to have a genetically
encoded lactate sensor). The resulting proteins showed fluorescence spectra consistent
with a poorly-folded sensor. It is conceivable that the fusion with the two fluorescent
proteins placed such a thermodynamic stability demand on the redesigned proteins that it
dramatically altered their folding state; but I began to suspect that the redesigned proteins
were sitting “on the edge” of being unfolded, most likely as molten globules. I do not
dispute anything in Dr. Hoecker’s work; in fact, I strongly encouraged her to pursue this
line of research.

2. Have you previously heard of researchers having problem with the fluorescence
assay? Do you know of any groups outside of Duke that have succeeded with it?
Did Hocker ever try it when she was still at Duke?
Other labs have used the small molecule dye fluorescence as a reporter for binding, but to
my knowledge, this has been confined to work with wild-type proteins. Several
individuals in the Hellinga lab, not on the 2 publications, were able to reproduce the
fluorescence titrations from the 2 papers. I hear that labs outside of Duke were able to
similarly repeat the fluorescence titrations, but encountered problems with aggregation
and variability over time. The growing consensus is that for stable, well-folded (e.g. wildtype)
proteins, the fluorescence reporter assay can work fine. But the current work, as
well as (Marvin & Hellinga PNAS 2001; Telmer & Shilton J. Molec. Biol. 2005), shows
that heavily-redesigned proteins can have their open-to-closed equilibria sufficiently
altered that: 1. Fluorescence response does not equate with binding, and 2. Binding does
not equate with formation of the expected bound state. While at Duke, Dr. Hoecker
worked with dye-labeled wild-type proteins, where the signaling mechanism seemed to
work well.

3. It should work, of course. So, do you have any idea why there is a discrepancy
between that method and the structural studies?
I think that the robust biophysical characterization given to the proteins in the PNAS
paper is correct. I think that the small molecule dye-based allosteric signaling mechanism
is prone to artifacts, particularly for destabilized proteins in which the open-to-closed
conformational equilibrium may be greatly perturbed. The small molecule dyes in
question are exquisitely sensitive to small environmental perturbations; that was the
rationale for choosing them in the first place. The discrepancy serves to highlight the risk
in using the small molecule dyes as a proxy for ligand binding, without robust
“orthogonal” characterization by a non-fluorescent method.

4. I understand you mostly worked on the software side of things, right? The
authors note that your computational model seems to work for the side chains. Why
do you think it is only partly predictive? Does this mean you could make a few
tweaks -- or is this a big issue that will take a lot of work to resolve?
That is a correct characterization of my involvement, although I did do some
experimentation for the Nature 2003 paper (see #5). I have data (including a highresolution
crystal structure of a designed protein-ligand interface) showing that a
computational methodology similar in its spirit to that described in Nature 2003 can be
extraordinarily useful for prioritizing potential protein-ligand complexes. This designed
protein-ligand interface was based initially on a computational model, and the first round
of mutants tested showed pretty good activity. But this was refined by many rounds of
directed evolution; it would have been very difficult to achieve the desired result without
this. I believe that the most important take-home message is that it is important to
combine rational & computational modeling with non-biased screening & selection
methods.

Grazie mille.