Structural dynamics of PDZ allostery

If Michele Vendruscolo were trying to get me to blog about one of his papers, he could hardly have assembled a more perfect lure than his upcoming paper in JACS. It brings together all sorts of things I've been talking about on this webpage: NMR dynamics, MD simulations, and dynamics-driven allostery (in the PDZ domain, no less). Previous investigations of this PDZ domain indicated the existence of a network of residues that had a dynamic response to ligand binding. Dhuselia et al. extend this work using molecular dynamics simulations constrained by the existing dynamics results. This leads them to discover not one, but two networks in the PDZ domain, with different properties.

NMR experiments have enormous power to sensitively detect changes in dynamics resulting from a perturbation, but they are also quite limited. Because of the models we use, the parameters we can fit out of relaxation data only give us information about the magnitude and timescale of fluctuations. Chemical shift overlap and interference caused by nearby dipoles limit the number of probes. Moreover, because NMR can only measure an ensemble, it is practically impossible to extract anything other than the most general information about correlated motions. MD has answers to all of these problems, but as a general rule has done poorly at reproducing NMR data about side-chain motions, calling the validity of the conclusions into question. Vendruscolo has taken some interesting strides in this regard by employing the limited experimental dynamics data as a component of the energy function. By constraining the simulation to mimic the known dynamics, we can hopefully learn more about the sites to which we are blind, as well as what kinds of motions the experiment is sensing and how they are linked.

In this instance, the authors make use of the PDZ domain previously studied by Ernesto Fuentes in Drew Lee's lab (there was also some hack working there at the time). Ernie's research followed on previous evolutionary studies indicating a network of communication in PDZ domains (local summary here), and Ernie found, by comparing the dynamics of the free and ligand-bound states, that changes in motions propagated away from the binding site to two distal surfaces. The pathways of communication compared pretty well with the evolutionary results. Dhulesia et al. aim to extend these results by determining which motions are correlated and identifying the mechanisms by which energy is transmitted. They accomplished this by running multiple parallel simulations of the free and ligand-bound states of the PDZ domain constrained by Ernie's dynamics results, as well as NOE and ³J data.

They find that two regions of the protein have correlated motions internally and move in an anticorrelated fashion relative to each other (Figure 3A). One of these regions consists of part of the binding site and all of distal surface 2 (DS2), while the other includes the other half of the binding site and all of distal surface 1 (DS1). When the ligand binds, something interesting happens. The motions of DS2 become more tightly correlated to the motion of an area around V30. The tight correlation between the motions of DS1 and α2 (an element of the binding cleft) switch to a slight anticorrelation.

When a ligand binds to a protein we expect a broad increase in rigidity of the complex so that the proper orientations of bonding pairs are maintained. For the most part, the simulations affirm this expectation, but not for all regions. For the binding site and DS2, the backbone mobility decreases, as expected, but the backbone mobility of DS1 increases (I am going off the text and Table 3 here, rather than Figure 3). The side chains have a similar response. This agrees with other studies indicating that the change in conformational entropy upon binding a ligand need not be homogeneous. What is more interesting is that these results imply that opposite coherent responses can be induced in a small domain by a single stimulus.

Although (as far as I know) this PDZ domain has no allosteric behavior in vivo, one can imagine that the binding of a ligand at the cleft could alter the binding of other modules to this domain. The entropic penalty for binding to DS2 would be lower in this case, while the penalty for binding to DS1 would be higher. The opposed nature of the dynamic responses may be related to the broad regional anticorrelation of free-state motions; disruption of this mode (by linking the motion of β2 and α2) may shunt that energy into DS1.

The authors also find, using a series of structural parameters, that a set of residues have clear structural changes. Some of them appear to be associated with coupled changes in rotameric states; the authors map out one pathway in Figure 5. Because it is a rotameric pathway, it should be possible to test whether it is essential to communication experimentally—mutation of the intermediary residues should abolish the linkage. The authors also carry out a network analysis to identify the most connected residues, a prediction that may also be testable by mutagenesis. These "structural network" residues overlap only slightly with the dynamic network, and indeed do not generally intersect with the evolutionary network either. In the absence of identified allosteric behaviors or clear energetic connectivities it's difficult to say what this disjunction means. However, the residues undergoing structural changes surround most of the residues undergoing dynamic changes. It is possible that these changes in structure provide the context that allows the changes in dynamics (or vice-versa); the two properties are inextricably linked.

Although communication between the binding site and distal surfaces is proven in this PDZ domain, and appears to be a general feature of the fold, the absence of a known function for the propagation in this instance makes it tough to assess the quality of these results. However, the findings of Dhulesia et al. make it clear that this approach can produce testable predictions and explanations. Hopefully this approach will be employed in the near future to study PDZ domains known to possess allosteric properties.

1. Dhulesia, A., Gsponer, J., Vendruscolo, M. (2008). Mapping of Two Networks of Residues That Exhibit Structural and Dynamical Changes upon Binding in a PDZ Domain Protein. Journal of the American Chemical Society DOI: 10.1021/ja0752080

Read the rest...

PDZ domains allosterically regulate the bacterial envelope protease DegP

Anyone who reads this blog regularly knows that I have a great deal of interest in the allosteric potential of the PDZ domain, a small protein binding domain that can be found in every branch of the tree of life. In previous posts I've discussed the evidence for allosteric communication within the PDZ domain, as well as evidence for long-range energetic interactions. An upcoming paper emerging from a collaboration of several European groups provides yet another example of the PDZ domain's allosteric potential bearing fruit, in this case in the regulation of a protease. The article is open-access, so go ahead and open it up in another window to follow along.

Krojer et al. are investigating the heat response in bacteria. Just as your body responds to ambient warmth (by sweating, etc.), bacteria have several systems to help them cope with high temperatures. These include systems for folding proteins that have lost their shape due to the heat, and also systems that degrade proteins when the refolding system can't keep up. A protein that breaks down other proteins is called a protease, and enzymes of this type are widely used in nature (blood clotting, viral maturation, digestion, etc.). Because of their destructive potential proteases are often tightly regulated, as is the case with the protease in this article, DegP. In fact, DegP can also serve as a refolding protein (or chaperone)—some regulatory mechanism causes a switch between these functions. DegP is an E. coli enzyme, but has a similar architecture to some human enzymes linked to diseases that involve protein misfolding and aggregation.

Krojer et al. performed experiments to characterize the proteolytic activity of DegP, mostly summarized in Figure 1. These results indicate that DegP is processive, i.e. that it makes many cuts on a target rather than just one. Using mass spectrometry the researchers determined that the targets got cut up into chunks 8-22 residues long, with the most common lengths being 12 and 17 residues. The protease preferred to cut proteins after a valine, alanine, isoleucine, or threonine: these are very common residues in proteins and interestingly most are β-branched. However, when they made short peptides that matched the apparent cleavage pattern, they did not observe any proteolysis.

From this the authors concluded that binding to the PDZ1 domain of DegP was necessary for the proteins to get cut. They performed a series of experiments with longer peptides that showed that a proper binding site 13-17 residues away from the cleavage site was necessary to get proteolysis. Also, the C-terminal residues preferred by PDZ1 are the same as the ones where the protease cleaves. Modeling an unstructured peptide substrate into the known structure of the DegP complex (Figure 4) indicates that a peptide chain ~16 residues long is needed to reach from the PDZ1 domain of one DegP to its own protease domain, and a chain ~12 residues long is needed to stretch to the protease site of an adjacent DegP molecule.

Well, this suggests a tidy little model. The C-terminus of an unfolded protein binds at the PDZ domain and is cut by the protease about 12 or 16 residues down the line. The cut produces a new C-terminus, which binds at the PDZ domain, and the process repeats. In this way the DegP complex processively degrades unfolded proteins and the bacteria are saved from toxic aggregates. But this is not the whole story. You see, it turns out that DegP is pretty efficient at cutting peptides even if the PDZ substrate and the cleavage site are not attached to each other.

The above model suggests two predictions. First, a peptide that binds to the PDZ1 domain (ALE peptide) but cannot be cleaved by the protease should inhibit the proteolysis of an unfolded protein. Second, the activity of DegP towards a substrate that is too short should not be affected by adding an ALE peptide. However, when the researchers in this case performed these experiments the results were quite different than expected. The ALE peptide slightly activated the degradation of an unfolded protein, and enhanced the cleavage of the short peptide 50-fold. Further experiments indicated that the binding of a peptide to the PDZ1 domain of one DegP molecule activated the protease of that molecule and one neighboring molecule of the complex.

So now the model is more complex, but also much more interesting. Binding of an unfolded protein's C-terminus to the PDZ1 domain does help generate the processivity and molecular ruler effects mentioned previously. But this binding also allosterically increases the intrinsic rate of proteolysis in nearby protease subunits. The effect is a sort of positive feedback mechanism that keeps the protease running until the whole target protein is degraded. This adaptation means that the protease can activate rapidly when there are a lot of free termini floating around, but won't go chopping up every flexible loop it comes across.

The precise mechanism by which binding at the PDZ site activates the protease is beyond the scope of the present study. It may be that the networks suggested by previous research do not come into play in this instance. Looking at the biological complexes predicted from the crystal structure (explore it at the PDB) I would guess that binding-dependent remodeling of the strand linking the protease domain to PDZ1 may be responsible, rather than the dynamic networks. Future experiments will probably address this allosteric mechanism.

1. Krojer, T., Pangerl, K., Kurt, J., Sawa, J., Stingl, C., Mechtler, K., Huber, R., Ehrmann, M., Clausen, T. (2008). Interplay of PDZ and protease domain of DegP ensures efficient elimination of misfolded proteins. Proceedings of the National Academy of Sciences 105(22), 7702-7707 DOI: 10.1073/pnas.0803392105 OPEN ACCESS

Read the rest...

Long-range energetic coupling in a PDZ domain

Despite its relative ease and proven utility, mutagenesis is a frustrating way to dissect the energetics of a protein domain. No matter how carefully you choose the position to be altered or the residue to swap in, unexpected results or unusable protein are frequent. Even when you get relatively clean data, the interpretation of the results is usually difficult because mutations are classic violators of ceteris paribus assumptions. Proteins are in general quite flexible, and this mobility means that the effects of a mutation can often be moderated by compensating deformations. Thus it is difficult to pin down energetic coupling by double mutant cycle analysis. However, in the course of an attempt to assess the usefulness of a computational method, a group from Sweden has shown an energetic coupling that correlates with some NMR dynamics results (1).

So, what exactly are Chi et al. assessing? Well, as I've discussed before on this blog, almost a decade ago Rama Ranganathan published an algorithm that used co-conservation of residues in a protein sequence to identify an energetic pathway in PDZ domains (2). The linkage between residues was defined in terms of a "statistical coupling energy" or ΔΔG^stat. As the authors of the present study note, several subsequent papers have criticized the computational approach employed by Lockless and Ranganathan. Chi et al. aim to re-examine this network prediction in the context of the PSD95 PDZ3 domain, which was used by Lockless and Ranganathan in some of their non-computational work. Chi et al. performed double mutant cycle analysis involving several residues in PSD95 PDZ3, some of them predicted to be coupled by Ranganathan's procedure and some of them not.

I should note that when these groups look at identical cycles (K380A, A376V) they generally get different results. Just judging off the raw data shown I am inclined to trust Chi et al., but a detailed inspection of all the raw data would be necessary to judge which approach was more accurate (Lockless and Ranganathan used FRET while Chi et al. used fluorescent emission).

No project of this kind can definitively address the quality of Ranganathan's algorithm. Although the statistical coupling is expressed as an energy, it is best interpreted as a probability. That is, the prediction of Ranganathan's algorithm is not that any protein will display precisely the calculated ΔΔG^stat, even assuming we knew what kind of observable to measure that reflected the observed statistical coupling (ligand binding? folding? mechanical disruption?). Rather, ΔΔG^stat reflects a predicted probability over the set of all PDZ domains that two sites will be energetically coupled. Thus, finding or not finding energetic coupling in a particular PDZ domain is not evidence for or against the accuracy of the algorithm, just as rolling 6 ones in a row is not evidence that the probability of rolling one on a fair die is anything other than 1/6. Nor is dissimilarity between the energetic network in any PDZ domain (or even in all PDZ domains) and the algorithmic prediction evidence that the algorithm is wrong. To find out if the algorithm is wrong using mutant cycle experiments, you must determine coupling energies in a number of PDZ domains and compare those results back to the prediction.

So, Chi et al. cannot establish whether the algorithm is right, but they can at least tell us whether the predictions of the algorithm are accurate in this particular case. They assert that the predictions are not very good in this instance, and judging on a linear correlation plot they provide this seems to be true. Before we draw any conclusions, however, let's take a look at the results in the context of the structure.

The figure I've made for us on the right is pretty busy, so you may find it helpful to open it in a new window. You can also explore this structure at the PDB. The key residue for this study, H372, is in red, and the peptide ligand is in purple. The backbone and side chains are colored for other residues that were mutated in this study. Residues with blue side chains were not predicted to be part of the energetic network by Lockless and Ranganathan (green side-chains were). Note that G329 was also predicted to be on the network, but of course it has no side chain for me to color. If the detected coupling energy for a mutation was larger than the error (based on Chi et al. Fig. 3A), I painted the backbone ribbon gold; otherwise it is blue. You will note that every mutation producing a coupling energy larger than the error in the measurement lies on the network predicted by Lockless and Ranganathan.

To be fair, an A376V mutation did not produce a coupling while A376G did, so the results are equivocal at this residue. Because G has a very low helix propensity it may be significantly altering local secondary structure (ceteris paribus violation). Therefore, the A376V result may be more representative. This highlights a significant weakness of mutational experiments. Because of glycine's quirks, there is no such thing as a conservative mutation of an alanine. Also, because glycine and proline have unusual properties with respect to secondary structure, mutations that have a G or P on either end are virtually impossible to interpret without careful structural studies.

The Ranganathan method gave a substantial number of false positives in this study. It bears mentioning that Lockless and Ranganathan saw substantial correlation between experimental couplings and ΔΔG^stat in this domain, though they used different mutations, which if I recall correctly were chosen based on the second most common residue for a given site. However, even in these results, the Ranganathan algorithm did accurately identify a distal residue (V362) which displays a significant coupling energy even though the β carbons of it and H372 are 14.0 Å apart. It is very odd for the authors to say in light of this that the domain has no coupling other than a straightforward distance relationship. From the data they show it is evident that distance from H372 would not be an accurate predictor of coupling. The absence of a coupling at both V428 and A376 (for the V mutation) indicates that very close residues may have no coupling at all. Also, given the A376V results, it seems premature to designate V428A as an outlier.

Of further interest, previous NMR experiments regularly identified a homologous residue (V61) of hPTP1e PDZ2 as having a dynamic response to peptide binding (3). A shamelessly stolen figure to the right shows some of these results. Ligand binding induced a decrease in the S² and an increase in τ_e on the side chain of V61. While mutational studies did not indicate a substantial effect of a V61A mutation on ligand binding (4), it would be interesting to check whether non-additivity between H71Y and V61A mutations is observed in hPTP1e. Similarly, it would be of great interest to examine the dynamic response of PSD95 PDZ3 to ligand binding.

The data in the Chi et al. paper indicate that there is, in fact, at least one long-range energetic coupling in the PSD95 PDZ3 domain. This coupling, between H372 and V362, was predicted by Ranganathan's algorithm. Moreover, NMR experiments have shown that the homologous residue in hPTP1e experiences dynamic changes in response to ligand binding, further strengthening the case for a functional connection. Clearly, Ranganathan's algorithm produces a substantial number of false positives in the case of this particular domain, supporting the author's contention that it poorly predicts the energetic behavior of any single domain.

Nonetheless, the demonstration of a long-range energetic coupling between H372 and V362, when most other core residues tested showed no coupling, strongly suggests the existence of some sparse energetic network within PSD95 PDZ3, consistent with the findings of Fuentes et al. and the predictions of Lockless and Ranganathan. It is to be hoped that Jemth's group will undertake further studies, perhaps guided by the existing structural and dynamic results, to identify the precise pathway by which energy is transmitted from V362 to H372.

1. Chi, C.N., Elfstrom, L., Shi, Y., Snall, T., Engstrom, A., Jemth, P. (2008). Reassessing a sparse energetic network within a single protein domain. Proceedings of the National Academy of Sciences, 105(12), 4679-4684. DOI: 10.1073/pnas.0711732105

2. Lockless, S.W., and Ranganathan, R. (1999). Evolutionarily Conserved Pathways of Energetic Connectivity in Protein Families. Science, 286(5438), 295-299. DOI: 10.1126/science.286.5438.295

3. Fuentes, E.J., Der, C.J., and Lee, A.L. (2004). Ligand-dependent Dynamics and Intramolecular Signaling in a PDZ Domain. Journal of Molecular Biology, 335(4), 1105-1115. DOI: 10.1016/j.jmb.2003.11.010

4. Fuentes, E.J., Gilmore, S.A., Mauldin, R.V., and Lee, A.L. (2006). Evaluation of Energetic and Dynamic Coupling Networks in a PDZ Domain Protein. Journal of Molecular Biology, 364(3), 337-351. DOI: 10.1016/j.jmb.2006.08.076

Read the rest...

Allosteric interactions of PDZ domains

Another article entering the preprint stage last week also has some history behind it, although it doesn't have quite the wayback machine mojo of the last articles I discussed. A collaboration between researchers at Radboud University Nijmegen and the University of Pisa produced a really interesting story about the protein tyrosine phosphatase PTP-BL (citation 1 below). Like many phosphatases, PTP-BL is a large, multi-domain protein; it is typical to view these domains as independent functional units and see their conjunction in a protein as just a form of scaffolding. In the case of PTP-BL, however, it appears that the domains interact with each other, and that this has consequences for their binding specificity. In addition, their results agree with previous work identifying allosteric networks in the PDZ domain.

However, the interesting part of this story starts almost a decade ago with a bioinformatics experiment carried out by Lockless and Ranganathan published in Science (citation 2 below). Standard bioinformatic analysis of homologous proteins involves aligning their sequences based on similarity and looking for residues which are conserved across evolutionary time and space. It is generally believed that invariant residues are essential to either the structural integrity of a given protein fold or its function. Lockless and Ranganathan decided to take this a step further and ask which residues were co-conserved with a particular amino acid position, that is, whether amino acid changes at some position X are correlated with changes at some other position Y. They performed this experiment using the PDZ domain, a very common ligand-binding domain that appears in multiple proteins (and often in multiple copies within the same protein) in nearly all eukaryotes as well as bacteria. As the target of their co-conservation analysis, they chose a histidine in the binding cleft of PDZ.

Naturally, one would expect other residues within the binding site to show co-conservation, and this is indeed the case. The surprise, however, comes from the fact that in addition to these expected residues, an additional patch of residues on the opposite side of the domain also appeared to be co-conserved with the histidine, as shown in their figure at right. The histidine in question is residue 76, and co-conserved residues are rendered with pinkish molecular surfaces. The peptide bound by this particular PDZ domain is shown as yellow sticks. What you can see here is that there appears to be some linkage in an evolutionary sense between residues in the active site and residues in the β-strand structure on the other side of the protein. Lockless and Ranganathan did some binding studies that seemed to support their findings, and attributed this apparent pattern to structural perturbations.

A few years later, NMR virtuoso Ernesto Fuentes performed an NMR dynamics study on the second PDZ domain of human protein tyrosine phosphatase 1e (citation 3 below), in which he compared side-chain dynamics of the free (isolated) domain to those of the protein in a ligand-bound state. The results of that study are remarkably similar to those of the Lockless and Ranganathan work, though not identical. The ligand is shown in green, while side-chains of dynamically-responding residues are shown in red, yellow, and blue. As expected, most of the dynamic changes upon ligand binding occur right next to the binding site. However, two distal surfaces of the domain also appear to feel dynamic effects from the binding of the peptide. This gives even more direct evidence of some kind of allosteric interaction between the binding site of a PDZ domain and parts of the protein that are further away.

The new paper by van den Berk et al. puts a kind of exclamation point on this story. Their effort began, essentially, as a fishing expedition to find what peptides exactly the various PDZ domains of PTP-BL—it has five total—bind. Among others, they found that PDZ2 would bind to peptides from APC (binding site -VTSV) and RIL (binding site -VELV). This was the case when the PDZ2 domain was tested alone. When a construct containing both the PDZ1 and PDZ2 domains was tested, however, the RIL peptide no longer bound to PDZ2. This was true whether or not the 200 amino-acid linker between them was included in the construct, indicating that a bona-fide interaction between PDZ1 and PDZ2 was responsible.

van den Berk et al. then used NMR chemical shift perturbation mapping to identify the binding site of PDZ1 on the PDZ2 domain. They found, in what should hardly be a surprise at this point, that the primary site of interaction is a distal surface of PDZ2. On the basis of modeling studies they suggest that the mobility of Ile 48 is critical to enabling the binding of the bulkier RIL peptide; they attribute the allosteric effect of PDZ1 to a restriction of PDZ2 Ile 48 so that it cannot move out of the way and allow RIL to bind. Thus we come to the model at right. Note that van den Berk et al. did not, as far as I can tell, determine what part of PDZ1 binds to PDZ2. It's possible that the normal binding cleft is used, but the binding site on PDZ2 looks like a broad hydrophobic surface rather than a narrow structure that could easily insert into a binding cleft. This suggests that the binding site on PDZ1 is still available, and thus that peptide binding to PDZ1 could fine tune the behavior of PDZ2.

The functional importance of this change in affinity is not yet clear. RIL is also bound by another PDZ domain in PTP-BL, so the PDZ1-PDZ2 interaction does not abrogate RIL binding. Additionally, even in the absence of PDZ1, APC has a higher affinity for PDZ2 than RIL, so the PDZ1-PDZ2 interaction is not really switching the target of PDZ2 or anything. However, if the local concentration of RIL is significantly higher than APC, improved specificity for APC may be necessary for kinetic reasons. Alternately, the improved APC specificity may be an incidental feature of an interaction that evolved for another reason, perhaps to bring APC into close proximity to some protein bound to PDZ1.

The possibility that allosteric communication pathways might exist in small protein modules like the PDZ domain was initially met with a great deal of resistance. The classic descriptions of allosteric and cooperative interactions all involved very large protein oligomers. Dynamics experiments like those carried out by Fuentes et al. and functional studies such as this one, however, have borne out the predictions of the bioinformatics studies. Allostery, or the potential for allostery, is a feature of small, isolated domains just as it for large protein assemblies.

Another point to keep in mind out of this paper is that it is a mistake to assume that domains within a protein are completely independent. It turns out that this is often the case, that, for instance, a binding domain and a catalytic domain exist together without really interacting, and when that happens it's not a problem to analyze each domain separately. However, domains within a protein can and do interact with one another, even when they are separated by sizable linking regions (the linker here is >200 amino acids). Often the easiest (and sometimes the only) way to investigate the structure of large proteins is to look at their domains individually. This paper, among others, is a reminder that this strategy is not always appropriate and may fail to capture important features of the domains' functions.

(1) van den Berk, LCJ, Landi, E, Walma, T, Vuister, GW, Dente, L, and Hendriks, WJAJ. "An Allosteric Intramolecular PDZ-PDZ Interaction Modulates PTP-BL PDZ2 Binding Specificity." Biochemistry ASAP (2007)

(2) Lockless, SW and Ranganathan, R. "Evolutionarily Conserved Pathways of Energetic Connectivity in Proteins." Science 286 (1999) p. 295-299.

(3) Fuentes E.J., Der C.J., and Lee, A.L. "Ligand-Dependent Dynamics and intramolecular signaling in a PDZ domain." J. Mol. Biol. 335 (2004) pp. 1105-1115.

Read the rest...