Dynamics and interactions of ADP/ATP transporter AAC3 in DPC detergent are not functionally relevant

Vilius Kurauskas, Audrey Hessel, François Dehez, Chris Chipot, Beate Bersch, Paul Schanda

Preprint posted on May 10, 2018


Major concerns with the integrity of the mitochondrial ADP/ATP carrier in dodecyl-phosphocholine used for solution NMR studies

Martin S. King, Paul G. Crichton, Jonathan J. Ruprecht, Edmund R.S. Kunji

Preprint posted on May 25, 2018

Two preprints question the structural details of an ADP/ATP transporter's mechanism and call for more thorough controls to verify membrane protein functionality

Selected by Reid Alderson


Membrane proteins (MPs) comprise 15-25% of the human proteome, yet collectively sum to less than 1% of the deposited three-dimensional structures in the Protein DataBank (PDB). Such a paucity of structural information manifests from difficulties that can accompany the biochemical preparation of purified MPs, ranging from complications with faithful abstraction from the native membrane to misfolding after extraction from insoluble inclusion bodies. Moreover, despite the common view of MPs existing in one of two end states (i.e. “open” or “closed” transporters, “active” or “inactive” G-protein coupled receptors), MPs can populate a wide range of conformations that dynamically exchange on various timescales, spanning pico-nanoseconds, micro-milliseconds, or far longer than hours. Such dynamics can often hinder crystallization attempts for structural analysis by X-ray crystallography. Despite the current lack of structural knowledge pertaining to MPs, many FDA-approved drugs target them. Thus, a holistic understanding of membrane protein structures, dynamics, and interactions in native membrane environments, or reliable membrane mimetics, proves critical.

A significant factor in structural studies of MPs is the membrane environment. As pointed out in a recent review1, nearly 80% of MP structures determined to date have used detergents, which form small micelles that present a very different environment to lipids. However, recent work has revealed that many MPs embedded in such micelles, for instance of the commonly used detergent dodecyl phosphocholine (DPC, also known as Foscholine-12), are no longer functional or folded correctly1.

Thus, when choosing a membrane mimetic system, how does one ensure that a reconstituted MP is functionally relevant? The most rigorous demonstration is an assay that directly monitors biochemical activity: for example, the transport of a substrate or other reporter molecule. In the absence of such an assay, researchers typically default to the demonstration of the MP binding to a relevant ligand, generally by measuring a dissociation constant (Kd) and comparing this to values obtained by other groups who have used native membranes. In the preprints discussed below, two reproduction studies are conducted pointing out significant flaws in this practice, and show that the presumed functionality of MPs should be interpreted with caution when using detergents.


The preprints

Two recent preprints conducted reproducibility studies on the structural and functional integrity of an AAC transporter (AAC3) in DPC2,3. AAC3 functions to exchange ADP for ATP across the inner mitochondrial membrane, an essential process for the cell. This membrane protein had previously been crystallized in the presence of a strong inhibitor (CATR) and was shown to adopt a locked conformation, termed the “c-state”4,5. As part of its transport mechanism, AAC3 populates a second conformation known as the “m-state”, yet structural evidence for this state has remained elusive.

In 2015, a paper by Chou and colleagues6 in Nature Structural & Molecular Biology reported using NMR spectroscopy that, in DPC, the yeast version of AAC3 (yAAC3) could be captured in a dynamic equilibrium involving two states: the c-state and a transiently populated conformation, which was inferred to be the m-state (or similar). Critically, their results hinged on yAAC3 remaining correctly folded and fully functional while embedded in the detergent. The integrity and correct folding of the yAAC3 sample was determined by binding of the inhibitor (CATR) and substrate (ADP) as measured by isothermal titration calorimetry (ITC) or NMR6.

Preprints from Schanda and colleagues2 at the Institut de Biologie Structurale in Grenoble (France) and Kunji and coworkers3 at the Mitochondrial Biology Unit in Cambridge (UK) challenged the notion that yAAC3 is correctly folded and functional in DPC. Rather, the authors convincingly demonstrate that yAAC3 is misfolded and not functional in the presence of DPC.

First, both King et al.3 and Kurauskas et al.2 noted that the Kd values for CATR binding to yAAC3 reported by Chou and colleagues were nearly three-to-four orders of magnitude larger than the commonly accepted literature values (5–300 nM). Second, Schanda and colleagues examined the raw data used to determine that the substrate (ADP) binds specifically, and found residues all over yAAC3, including those more than 20 Å distant from the ADP binding site, were impacted by ADP binding2. While this could reflect allosteric structural changes, a recent publication noted that AAC3 embedded in DPC bound identically to ATP and GTP, despite preference for the former in the native membrane5. Such a loss of specificity implies an incorrect tertiary structure. Finally, King et al.3 measured melting temperatures of yAAC3 in DPC and other detergent systems (Figure 1A). In the native mitochondrial membrane, yAAC3 unfolded near 40°C; however, the same experiment performed in DPC revealed a large fraction of unfolded protein already present at ambient temperature, which exhibited no unfolding cooperativity3. Similar results were obtained by NMR3. Taken together, these data suggest that yAAC3 solubilized in DPC is incorrectly folded and binds to CATR and ADP non-specifically2,3. King et al. note that a non-specific interaction could be electrostatic in nature, as yAAC3 has an isoelectric point of 9.82, and therefore contains excess positive charge, whereas ADP and CATR are both negatively charged at neutral pH3.

Having established that ADP and CATR binding in fact reflects non-specific interactions, Schanda and colleagues2 re-examined the previously published NMR data used to measure conformational exchange in yAAC3 between the c-state and putative m-state (Figure 1B). The experiment used for such analysis is called Carr-Purcell-Meiboom-Gill relaxtion dispersion (CPMG RD, reviewed here), and involves the measurement of an NMR parameter (transverse relaxation rate, termed the effective R2) as a function of the number of radiofrequency pulses applied during a fixed delay. In the presence of conformational exchange between two or more states on the micro-to-millisecond timescale, CPMG RD experiments will reveal a dependence of R2 on the number of applied radiofrequency pulses (Figure 1B). Subsequently fitting these data to the appropriate equations yields the rate of conformational exchange, the relative populations, and (qualitative) residue-specific information about structural differences between the two forms. For example, such experiments have been deployed to determine accurate solution structures of otherwise invisible protein folding intermediates that exchange with a folded conformation on the millisecond timescale7.



In the original paper, Chou and colleagues6 found via CPMG RD that the rate of conformational exchange varied by roughly 20-fold between the inhibitor- and substrate-bound forms of yAAC3, and reported large structural differences between the two states (for NMR spectroscopists, 15N |delta omega| > 5 ppm). However, Kurauskas et al. re-fit the raw CPMG RD data in free state and the inhibitor- and substrate-bound forms of yAAC3 (Figure 1B), finding that no difference existed in the rate of conformational exchange (1500 s-1 for all forms), with significantly smaller structural differences than initially reported (15N |delta omega| values around 2 ppm). Finally, similar millisecond exchange dynamics that were independent of added substrate/inhibitor were obtained when analyzing conformational exchange in mutants of AAC3 that are incapable of ADP/ATP exchange, indicating that there is no link between dynamics and transport8.

In conclusion, these two new preprints from the Schanda and Kunji groups demonstrate that the yAAC3 membrane protein is not functional and misfolded when embedded in DPC. These studies highlight the difficulty and importance of choosing the right membrane mimetic (when not using native membranes). In addition, these works provide a platform for diagnosing future issues that arise from misfolded MPs (e.g. non-specific ligand binding, lower Kd than expected, pervasive millisecond motions), in particular those analyzed by NMR spectroscopy.


My opinion

These two new preprints highlight the importance of independent replication studies that seek to verify, challenge, or question biological implications obtained from biophysical experiments. While a few examples exist in the published literature, there are certainly not enough of such studies. Encouragingly, a replication preprint was recently highlighted on preLights. The bioRxiv will help to remove the stigma associated with publishing replication studies, and will provide a convenient platform for the publication of replication studies in lieu of publication in traditional journals. Likewise, PLOS Biology has instituted an initiative to publish papers from groups who have recently been “scooped”, which should also increase the number of papers that report data on the same system.

While the two preprints discussed here present negative results, their findings will together advance the field of AAC structural biology, and hopefully will deter future researchers from using DPC as a reconstitution system. More generally, the preprints from the Schanda and Kunji groups highlight the importance of selecting the appropriate membrane mimetic system and then verifying that the embedded MP retains functional and structural integrity.

As a practicing NMR spectroscopist who enjoys dynamical measurements, the experiment that I most enjoyed was actually a data analysis method to obtain insight into micro-to-millisecond motions. Schanda and colleagues noted that the CPMG RD data from Chou and coworkers could be described by a single set of parameters for all three conditions (apo-AAC3, CATR-bound AAC3, ADP-bound AAC3), and showed that these data could indeed be fitted globally. This means that the AAC3 exchange process on the micro-to-millisecond timescale, which is probed in the NMR experiment, is insensitive to added substrate or inhibitor, and is therefore not a transport filter.



  1. Chipot C, et al. Perturbations of native membrane protein structure in alkyl phosphocholine detergents: a critical assessment of NMR and biophysical studies. (2018) Chem. Rev. 118: 3559-607.
  2. Kurauskas V, Hessel A, Dehez F, Chipot C, Bersch B, Schanda P. Dynamics and interactions of ADP/ATP transporter AAC3 in DPC detergent are not functionally relevant. (2018) bioRxiv
  3. King MS, Crichton PG, Ruprecht JR, Kunji ERS. Major concerns with the integrity of the mitochondrial ADP/ATP carrier in dodecyl-phosphocholine used for solution NMR studies. (2018) bioRxiv
  4. Pebay-Peyroula E, Dahout-Gonzalez C, Kahn R, Trezeguet V, Languin GJ, Brandolin G. Structure of mitochondrial ADP/ATP carrier in complex with carboxyatractyloside. (2003) Nature 426: 39-44.
  5. Ruprecht JJ, Hellawell AM, Harding M, Crichton PG, McCoy AJ, Kunji ERS. Structures of yeast mitochondrial ADP/ATP carriers support a domain-based alternating-access transport mechanism. (2014) Proc. Natl. Acad. Sci. USA 111: E426-34.
  6. Bruschweiler S, Yang Q, Run C, Chou JJ. Substrate-modulated ADP/ATP-transporter dynamics revealed by NMR relaxation dispersion. (2015) Nat. Struct. Mol. Biol. 22: 636-41.
  7. Hansen DF, Vallurupalli P, Kay LE. Using relaxation dispersion NMR spectroscopy to determine structures of invisible, excited protein states. (2008) J. Biomol. NMR. 41: 113-20.
  8. Kurauskas V, Hessel A, Ma P, Lunetti P, Weinhaupl K, Imbert L, Brutscher B, King MS, Sounier R, Dolce V, Kunji ERS, Capobianco L, Chipot C, Dehez F, Bersch B, Schanda P. How detergent impacts membrane proteins: atomic-level views of mitochondrial carriers in dodecylphosphocholine. (2018) J. Phys. Chem. Lett. 9: 933-8.

Tags: atp transporter, membrane protein, nmr, structural biology

Posted on: 23rd July 2018 , updated on: 30th July 2018

(No Ratings Yet)

  • Author's responses

    Paul Schanda and Edmund Kunji shared about Major concerns with the integrity of the mitochondrial ADP/ATP carrier in dodecyl-phosphocholine used for solution NMR studies

    1. Why did you pursue a replication study?

    • Paul Schanda (PS):  We have been working on mitochondrial carriers since 2011, using the same experimental setup as the Chou group. Our initial results were indeed very promising — for example, we saw millisecond dynamics and substrate interactions already about 5 years ago — and we were very excited to gain biologically relevant insight into these difficult and (presumably) highly dynamic membrane proteins. As in any serious study, we started doing extensive control experiments, to verify that the dynamics and interactions that we saw were indeed relevant. Control experiments are particularly important as membrane proteins tend to be rather sensitive to their environment, and it was not clear to us whether the behavior we saw in detergent was relevant. To our great disappointment, the more control experiments we did, the more we convinced ourselves that we had been seeing artefacts, i.e., the effects of partly denatured protein, rather than biologically relevant behavior. We were surprised when we then saw the 2015 NSMB paper appearing. In that paper, everything seemed to work much better than in our study, yet we had essentially done the same things. This situation motivated us to look more closely into the NSMB paper, and indeed we discovered that their protein samples were just as denatured as ours, but they had not done the important controls, and they have not shown all data.


    • Edmund Kunji (EK):  We have a lot of experience with this protein, as we solved its structure by electron and x-ray crystallography. Nothing described in this paper (the 2015 Nat Struct Mol Biol paper) made any sense with respect to our own observations or those of others. We then tested these claims using purified AAC3 from mitochondria, the same material we used to obtain the structure, and we found that DPC inactivates the protein instantly and irreversibly, as all known properties are lost.

    2. Why did you two and your research groups decide to work on the same system and address the same 2015 NSMB publication? 

    • PS:  About at the time when we started figuring out that we had been working for years on protein samples that were likely not biologically relevant, we got in close contact with leading experts in the field, namely E. Kunji and E. Pebay-Peyroula (with whom we had already close interactions from the start). We decided to have a close look together at all the publications of J. J. Chou on mitochondrial carriers. They had the expertise on mitochondrial carriers for decades, we came from the NMR side, so we thought that together we could understand what is really going on better than each of us by himself. The two Correspondences by Kunji’s group and our group focus on different aspects of the 2015 NSMB paper.


    • EK:   We worked on this completely independently pursuing the transport mechanism of these protein using very different techniques. Paul was trying to repeat and extend the NMR observations of the Chou group, but with appropriate controls, whereas we had interest to pursue other structures by x-ray crystallography. For the latter, we had carried out systematic stability trials for related proteins (see Crichton et al. below), as those are a prerequisite for crystal trials. From this comparative study it was clear to us that DPC was a very harsh detergent and we questioned whether it was also for AAC3. That is why we then looked into it and found that the properties of AAC3 are impaired in DPC. Prior to this, we and others have been very concerned about the 2011 NMR backbone structure of UCP2 (Nature), another family member, which seems to be inconsistent with their own data (e.g. PRE data) and does not resemble the structures of AAC at all.  Also here the functional assays are meaningless.

    Crichton, P. G., Lee, Y., Ruprecht, J. J., Cerson, E., Thangaratnarajah, C., King, M. S., and Kunji, E. R. S. (2015) Trends in thermostability provide information on the nature of substrate, inhibitor, and lipid interactions with mitochondrial carriersJ. Biol. Chem. 290, 8206-8217

    3. What type of assays or controls should the membrane protein field perform in order to ensure the membrane protein of interest retains structural and functional integrity?
    • PS:  Assaying functionality of transporters in solubilized form is tricky, as one cannot measure transport. Binding of inhibitors and substrates is one way, but it is critical to perform many control experiments with similar molecules which are not transported. If the effects in detergents are similar, as is the case for GTP and ATP here, even though in terms of transport these two nucleotides are very different, one should be very cautious. A very useful control experiment is the thermostability shift assay, which informs on the tertiary structure and stability. MD simulations may also tell whether structures determined by NMR (or X-ray diffraction or EM) make sense; but of course, this is only an “a posteriori” control. We have shown other control experiments on mitochondrial carriers in our recent paper in J. Phys. Chem. Lett. (Kurauskas et al, 2018).
    • EK:  It is essential for any biophysical or structural project that the folding and function of the membrane protein in detergent is checked carefully. Best criterion is a functional assay, but in its absence, thermostability assays are a good alternative. It is also important to check the functional data of any expressed protein against those of the native protein too, otherwise it is easy to suffer from self-delusion.
    (Note:  the following two questions were addressed to PS only)
    4. NMR is inherently a low sensitivity method that requires extensive sample optimization to yield high quality spectra. How should membrane protein NMR spectroscopists balance the quality of their NMR spectra and integrity of the membrane protein? There could (and likely will be) situations where membrane mimetic systems that yield correctly folded and fully functional membrane proteins will not provide workable NMR spectra.
    • PS:  This is a major issue. It is very dangerous to screen for the system that provides the best NMR spectra, as it is well known that highly flexible (unfolded) proteins yield very sharp lines. It may well be that functional states yield lousy NMR spectra, e.g. due to the presence of dynamics, or simply because of the size. There are workarounds that may work in some cases. E.g. specific methyl labeling, or selective introduction of fluorine labels (and 19F NMR) may give access to interactions and dynamics by NMR even though the spectral quality of a fully labeled sample is not “workable”. The primary criterion should be functionality of the protein, rather than the beauty of the NMR spectrum.
    5. You have stressed in various articles and platforms that NMR, in principle, can solve accurate structures of membrane proteins. The difficulties lie in the interpretation of the NMR data (which yield resonance frequencies and not precise atomic coordinates). There have been many arguments in the field over the stoichiometry, architecture, and dynamics of membrane proteins studied by NMR. How can the field begin to reconcile these differences? Do you think that a community-wide effort like CASP (Critical Assessment of protein Structure Prediction) would help outline standard protocols, critical controls, or improve structure determination of membrane proteins by NMR?
    • PS:  I think it is very important that the field comes up with such efforts. NMR is an excellent technique for providing insight into dynamics and interactions, and in some cases also structures. However, with each structure that later turns out to be likely incorrect, the whole field is damaged. One problem is the sparsity of structure restraints. In particular intermolecular distance restraints are difficult to get, and there are several cases where it has been proposed that intermolecular restraints may have been incorrectly interpreted. E.g., the structures of p7 (OuYang, Nature), gp41 (Dev, Science) and the mitochondrial calcium uniporter (MCU; Oxenoid, Nature) may all have such issues, as highlighted by several independent recent X-ray and EM structures of MCU, which are all very different from the NMR structure, even concerning the oligomerization state, as well as recent NMR work by the Bax group on gp41, which also suggests that the oligomerization state proposed in the 2016 Science paper (Dev et al) may not be correct. Given the increasing number of NMR-derived membrane protein structures that have issues, it really is crucial to act now. It may be difficult to come up with generalized standard protocols. However, even simple observations, such as a Kd to an inhibitor that is several orders of magnitude off — as was the case for MCU and ADP/ATP carrier — should ring alarm bells, for the authors and for the reviewers.

    Have your say

    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.

    Sign up to customise the site to your preferences and to receive alerts

    Register here