Dec 112009
ResearchBlogging.orgA few weeks ago, I wrote that the goal of a structural biology research program ought to be to “characterize the conformation and energy of key, functionally-relevant members of the protein’s structural ensemble and identify the pathways between them.” The Nature paper last week, among other examples I mentioned in the preceding post, described functionally significant minor members of the native-state ensemble, and this is certainly an area where structural studies are making a lot of progress. But what about the other part of that statement, the transition pathways? How are we to study them, and what can we learn about them? Experiments alone are unlikely to tell us everything we want to know about the intermediates between different native structures. We can, however, use simulations validated by experiments to investigate the mechanisms of structural change. Today in Cell, research primarily performed by my coworkers Alexandra Gardino and Janice Velos demonstrates that the bacterial signaling protein NtrC  rapidly samples its active conformation even when it is not phosphorylated. Moreover, they confirm predictions that the intermediates between these two states are stabilized by hydrogen bonds not present in either one.

Phosphorylation, or the covalent addition of a phosphate group to a protein, frequently appears as a way of transferring information within a cell. Often this chemical modification is described as a “switch” that flips a protein from an “off” state to an “on” state. The receiver domain of the bacterial protein NtrC (part of the nitrogen-fixing pathway) gets phosphorylated on  aspartate 54 in response to environmental stimuli, causing a change in conformation.  This causes NtrC to switch from a dimer to a hexamer, with the result that it binds to DNA, and eventually activates transcription of target genes. As you can see from the figure on the right, the change from inactive (red, PDB: 1DC7) to active (green, PDB: 1DC8) phosphorylation causes significant changes in the ’3445 face’ of the protein (dark colors), involving changes in not only the position of the helices, but also their length. That means numerous hydrogen bonds are broken and formed during the process, which would naturally lead one to suspect that going from one form to the other takes a while and is difficult to do. Neither is true

That NtrC converts rapidly from its active to its inactive state has been known for some time. Volkman et al. showed in 2001 that the 3445 face experiences some kind of chemical exchange process (2). Moreover, mutations that cause NtrC to become active in the absence of phosphate do not cause it to adopt the active conformation. As shown in Fig. 2e in (1) and Fig. 4 in (2), the NMR spectra for these mutants show peaks that lie partway between the active and inactive chemical shifts. This indicates that the exchange process reflects conversion between the active and inactive conformations in the absence of phosphorylation, and that this process is fast on the NMR timescale. The activating mutations merely shift the relative populations of the active and inactive states. As Janice’s folding experiments in the current paper show, these mutations operate by destabilizing the inactive state, i.e. they make it less energetically favorable relative to the active conformation. By contrast, phosphorylation of D54 significantly stabilizes the active state. However, conformational exchange is still observed in the phosphorylated protein, indicating that the protein samples the inactive state even when it is presumably activated. NtrC is never “locked” into one state or the other.

This description obviously does not jive with the typical language describing phosphorylation as a “switch” that turns signaling pathways “on”. One might well wonder why phosphorylation matters if the unphosphorylated protein can sample the active conformation. Of course, the NtrC receiver domain might not behave the same in isolation as it does when it’s integrated into a whole protein. NtrC signaling in vivo involves communication from the receiver domain to the DNA binding domain, as well as changes in oligomeric state, neither of which are addressed by examining the receiver domain alone in solution. Given that D86N is functionally activating, it’s likely that these results translate to the whole protein, but phosphorylation may still be considered an effective “switch” because of the way it extends the lifetime of the active state. That is, the receiver domain samples the active state occasionally in solution but doesn’t stay there long enough for the full sequence of steps that are required to result in transcription. In this model, the effect of the phosphorylation is simply to hold the protein in its active conformation long enough for the full transcriptional activation to occur.

Talking about lifetimes is all well and good, but that’s a pretty nebulous discussion unless you have some idea of the rates at which these forms are interconverting. The dispersion traces in Fig. 2 show that the rate is fast, so fast, in fact, that a refocusing field of 1000 Hz has a negligible ability to suppress the effects of conformational exchange in the unphosphorylated protein. In order to fit these curves, Alex and Janice had to use an alternative approach to determine the intrinsic relaxation rate. Once they did that, they found that the rate of interconversion exceeded 104 /s for the unphosphorylated protein, but was only around 2000 for the phosphorylated form. How can such a complex process, requiring so many bonds to be broken, occur so quickly? The only way to know is to examine the pathway of transition between the inactive and active forms.

Identifying transition pathways between structural states is a tough problem, though, because the conformations of intermediates and transition states are poorly populated in the solution ensemble. Obtaining a high-resolution structure of one of these states through experiment is essentially impossible, which suggests that our best hope for getting detailed information about the transition pathway is by simulating the conformational changes in the protein. This also presents a problem, however, because these events occur on the microsecond to millisecond timescale, which means that a simulation would need to be on the order of a second long to really sample the relevant fluctuations. Even using hugely parallel computer nodes, however, simulating a protein at equilibrium for as little as a microsecond takes several months. You would need to run such a simulation for years to adequately sample even this rapid of a transition. To get around this time barrier, researchers perform various tricks with their simulations, simplifying the representation of molecules and forces or biasing them so that transitions happen more frequently.

The latter approach was used by Ming Lei in his targeted molecular dynamics simulation of this transition (3) — he applied an external force to the protein in that simulation that pushed it from one state to another. Here some rotation students put in some good work. Aleksandr (you may have noticed that the Kern lab attracts a lot of people named Alex) carefully examined Ming’s simulation to identify unusual interactions and noticed that a number of short-lived hydrogen bonds appeared to be forming, specifically hydrogen bonds that were not present in the active or inactive conformations. Of course, when you include a fictional force pushing a  protein one way or the other you may end up with artifacts in the simulation, so Janice, along with rotating students Ce Feng and Phillip, tested the simulation predictions by generating mutants that were incapable of forming these hydrogen bonds.

As you can see from Fig. 4, these mutations dramatically slowed the conformational fluctuation, confirming the importance of these bonds in lowering the energy barrier of the structural transition. Even though the bonds exist for only a few nanoseconds in the simulation, they appear to play a significant role in lowering the energy barrier. Note, however, that the effect of these mutations is not necssarily additive — the double S85D/Y101F mutant is no slower than either S85D or Y101F alone. This suggests that these hydrogen bonds stabilize the transition pathway at different points, so that different non-native bonds are responsible for lowering the energy barriers of multiple independent steps. This finding has significant implications for efforts to model structural transitions using simplified potentials based entirely on native-state contacts; it is possible, perhaps even likely, that such simulations will miss critical interactions that stabilize intermediates along these transition pathways.

Our ability to successfully design functional proteins will require that we consider not only the structural heterogeneity of the native state, but also how the movement of the protein between different conformational substates can be tuned by raising or lowering the energy barrier between them. Here, NMR experiments have shown how mutations and phosphorylation alter the energy landscape of NtrC and alter the balance between its inactive and active conformations. Moreover, this study validates the prediction from a targeted molecular dynamics simulation that NtrC uses short-lived, non-native hydrogen bonds to facilitate the transition between these two conformational states. Beyond expanding our knowledge of NtrC’s conformational energy landscape, these findings suggest possibilities for other proteins that are activated by phosphorylation, one of nature’s most pervasive signaling methods.

1) Gardino, A., Villali, J., Kivenson, A., Lei, M., Liu, C., Steindel, P., Eisenmesser, E., Labeikovsky, W., Wolf-Watz, M., Clarkson, M.W., & Kern, D. (2009). Transient Non-native Hydrogen Bonds Promote Activation of a Signaling Protein Cell, 139 (6), 1109-1118 DOI: 10.1016/j.cell.2009.11.022

2) Volkman, B.F., Lipson, D., Wemmer, D.E., and Kern, D. (2001) Two-State Allosteric Behavior in a Single-Domain Signaling Protein. Science, 291 (5512), 2429-2433. DOI: 10.1126/science.291.5512.2429

3) Lei, M., Velos, J., Gardino, A., Kivenson, A., Karplus, M., and Kern, D. (2009). Segmented Transition Pathway of the Signaling Protein Nitrogen Regulatory Protein C. Journal of Molecular Biology, 392 (3), 823-836 DOI: 10.1016/j.jmb.2009.06.065

 Posted by at 3:02 PM  Tagged with:
Jan 232009
ResearchBlogging.orgNumerous and diverse biological processes depend on the functioning of an internal clock. Biological timers determine your heart rate, the frequency of cell division, and the way you feel at 3 AM, among other things. Similarly, mechanical and electronic clocks serve essential functions in many kinds of man-made devices. As we begin to develop synthetic organisms for medical and industrial purposes, it will be useful for us to be able to construct timers within these micro-organisms to control their activity. In two recent papers, scientists have created molecular systems in mammalian and bacterial cells with tunable oscillation periods.

Although the methods used to construct these oscillators and the kinds of cells they were made in differ significantly, the two systems had one key similarity. Both oscillators used both a positive and a negative feedback loop. In principle, it should be possible to construct an oscillating system using only a negative feedback loop. For instance, you could have a system in which a transcriptional activator enhances the expression of a functional protein as well as that of a transcriptional repressor. As the concentration of the repressor increases, that of the activator falls, causing levels of the functional protein and the repressor to fall, allowing the concentration of the activator to rise again. By tuning the lag in this system one could in theory produce an oscillator with a range of possible frequencies. Yet many systems seem to have evolved with a positive feedback loop as well (in which the activator enhances its own expression).

This curious feature was the subject of a series of simulations reported by Tsai et al. (1) last July in Science. Their studies indicated that a system using only a negative feedback loop would produce a periodic oscillation just as expected. However, they also found that systems relying only on negative feedback were limited in that it was difficult to adjust the frequency of the oscillation without also altering its amplitude. Introducing a positive feedback loop stabilized the system so that the oscillator could be tuned to a wider range of frequencies without altering peak amplitude.

The benefits of this approach were demonstrated in data reported by Stricker et al. last November in Nature (2). They constructed a circuit that expressed green fluorescent protein in an oscillatory manner in response to stimulation by arabinose and isopropyl β-D-thiogalactopyranoside (IPTG). They created a circuit (see their figure, right) in which every component ran off a hybrid promoter that could be activated by AraC (which binds arabinose) and inhibited by LacI (which binds IPTG). Arabinose binds to and activates AraC, while IPTG binds to and inactivates the LacI repressor, with the result that all three genes are transcribed, and the cells fluoresce due to the presence of GFP. As the concentration of LacI increases, the activating power of the IPTG decreases, leading to an eventual repression of transcription and the end of fluorescence. As the LacI proteins get degraded the IPTG concentration again becomes sufficient to activate transcription, leading to a new fluorescent phase.

Stricker et al. found that they could alter the frequency and amplitude of this oscillation by altering the growth conditions of the bacteria (temperature and nutrient availability) as well as by adjusting the concentrations of the activating reagents arabinose and IPTG. By attempting to match computer models of their oscillator to the data they collected, they found that the time needed for translation, folding, and multimerization played a critical role in establishing the existence and period of the oscillation. Stricker et al. constructed an additional circuit using only negative feedback from LacI, proving that this was possible, but they found that in this case the period was not very sensitive to IPTG concentration and the oscillations were not as regular.

A similar system was constructed in Chinese hamster ovary cells by Tigges et al., who described their results recently in Nature (3). The circuit they constructed used tetracycline (TC) and pristinamycin I (PI) as activating molecules. The tetracycline-dependent transactivator (tTA) served as the positive feedback lood, activating transcription of itself, GFP, and the pristinamycin-dependent transactivator (PIT). In this system, increased levels of PIT cause the production of antisense RNA to tTA, causing its mRNA to be destroyed prior to protein production. This, in turn, diminishes production of all proteins until the reduced levels of PIT allow tTA to again activate transcription. They found that they could control the period of oscillation by altering the gene dosage (i.e. the quantity of DNA used to transfect the cells).

The oscillating systems constructed in these papers serve more as test cases and examinations of principles than as functional pieces of synthetic systems. You will not be using an E. coli alarm clock any time soon. However, it has always been true that you learn more from trying to build something than from trying to tear it apart. These attempts to construct artificial periodic oscillators have provided interesting insights into those that have evolved naturally. The knowledge gained from these experiments will help us to understand oscillatory systems like the circadian rhythm and cardiac pacemaker, in addition to illuminating design principles for synthetic biology.

(1)T. Y.-C. Tsai, Y. S. Choi, W. Ma, J. R. Pomerening, C. Tang, J. E. Ferrell (2008). Robust, Tunable Biological Oscillations from Interlinked Positive and Negative Feedback Loops Science, 321 (5885), 126-129 DOI: 10.1126/science.1156951

(2)Jesse Stricker, Scott Cookson, Matthew R. Bennett, William H. Mather, Lev S. Tsimring, Jeff Hasty (2008). A fast, robust and tunable synthetic gene oscillator Nature, 456 (7221), 516-519 DOI: 10.1038/nature07389

(3)Marcel Tigges, Tatiana T. Marquez-Lago, Jörg Stelling, Martin Fussenegger (2009). A tunable synthetic mammalian oscillator Nature, 457 (7227), 309-312 DOI: 10.1038/nature07616

 Posted by at 12:45 AM
Aug 222007

I had a conversation this week with Annette about the structural ensemble versus individual structures that I’m still trying to coalesce into a fully-formed idea. The kernel of it is this: there is a dichotomy between the way we know that proteins act and the way we talk about their action. Proteins give rise to phenomenological effects as ensembles, but we discuss their states as individuals.

Consider a signaling molecule, say a member of a MAP kinase cascade (picture at right). A given protein can exist in either an inactive (A) or active (B) form. When active, the kinase phosphorylates some downstream target, otherwise it just sits in your cytoplasm taking a nap. Typically in this kind of system the active form of the kinase is also the phosphorylated form (red B). It’s typical to say something like, “The kinase is activated by phosphorylation.” At the same time we know from some of Dorothee’s work with Dave Wemmer that certain bacterial proteins that get phosphorylated already sample their active conformations even before they are modified (blue B).

Even for systems where this kind of sampling hasn’t been directly demonstrated it’s reasonable to assume it takes place. After all, the active and inactive structures have the same amino acids to work with. Unless the phosphate group itself is a lynchpin of the new structure (perhaps by bridging two structural elements), then the active structure must be one the unmodified kinase can adopt. Naturally, we expect this structure to be higher in energy than the inactive state (so blue B is higher on our energy diagram than blue A), and that phosphorylation decreases the energy of the active state so that it is subsequently preferred (so red B is lower than blue A).

The implication of this is that, unless the unmodified active structure is much higher in energy than the inactive structure, some proportion of our kinase is active even when not phosphorylated. Perhaps this is as low as 1-2%, a fraction that’s difficult to detect directly. Still, because enzymes are so efficient, this quantity may be significant. Or, for a single protein, we could say that it adopts an active form without phosphorylation 1-2% of the time. But we tend to talk about phosphorylation and other post-translational modifications as if they were switches, with phrases like “protein X is turned on by phosphorylation”. The reality, though, is that the switch is less a matter of turning a protein on than of turning it on more.

This points to a reality far less clean and orderly than typically depicted in block schematics. Inappropriately active (i.e. active without modification) members of the various protein ensembles must give rise to a considerable amount of noise in biological information processes. The system must therefore have some way to distinguish the signal from the noise that’s more than just the binary on/off typically depicted and discussed. These filters could take several forms — for instance, the kinase of our kinase may mediate the interaction between our kinase and its target, though in this case inappropriate activation of the MAPKK could still give rise to signaling noise. Alternately, the phosphate could mediate the kinase – target interaction. Or the cell could simply have an inefficient signaling system, so that multiple nearly-simultaneous signaling events are necessary to activate a response.

Is this point important? Maybe and maybe not. Most of our experiments can only access the behavior of ensembles, so the ensemble nature of protein action is not likely to lead us astray. But as single-molecule studies become more popular it may be important to keep the ensemble perspective in mind so as not to be confused by their results. Moreover, a conceptually accurate picture of cellular signaling and regulation will require us to keep this feature in mind.

 Posted by at 9:01 PM
Aug 132007

Another feature of the protein society was a continued emphasis on trying to understand natively-disordered proteins, and by extension, the denatured state of natively ordered proteins. Because these two fields are highly related and use the same techniques, it seems to me best to lump them together for now. A couple of interesting points came up that I wanted to get down here for my own memory’s sake.

One point, and one that became a recurring theme in several talks at the symposium, was averaging bias. The first real discussion of this came from a really good talk by Michele Vendruscolo on the study of the natively-disordered 131-deletion mutant of staphylococcal nuclease. Some models that Dave Shortle had produced of the disordered state on the basis of paramagnetic relaxation enhancement had predicted ensembles that were too small with respect to the known radius of gyration. Michele pointed out that the PRE is an ensemble measurement, and many different ensembles can give rise to the same PRE. Additionally, the PRE is biased because below a certain threshold the effect is invisible. This means that the measurement ends up being biased towards closer approaches. Essentially his point was that the normal distribution cannot be assumed for the ensemble average of distance measurements in the denatured state (and it’s probably a questionable assumption in the native state as well).Kevin Plaxco gave a talk later on that really hammered this point home. He did a series of SAXS experiments to determine the radius of gyration for a ton of proteins, including several that had shown residual structure in NMR experiments. His results indicated that the experimentally determined radius of gyration matched that predicted for a random coil for all these proteins. As he pointed out, though, the Rg is totally insensitive to local structure, whereas because of anomalous averaging much of the NMR data is hypersensitive to local structure. This means that both results can be right — any given protein can have some percentage of its structure intact and as long as it’s a different piece for each protein and not too much, the ensemble can retain a random-coil-like Rg. If tertiary interactions are preserved this becomes a slightly more difficult proposition to swallow, though. Still, his work, and several other talks and posters presented during the symposium, made an excellent point. We simply cannot rely on the assumption of a normal distribution when we are analyzing NMR data from systems with so many degrees of freedom.

Another thread that showed up repeatedly was the ongoing attempt to understand exactly how disordered states interact and are regulated, especially by post-translational modifications such as phosphorylation. Most disordered regions have multiple binding partners, with affinity enhanced for a particular partner by a particular modification. In the simplest model for these interactions, the modification itself and some of the surrounding primary sequence is recognized. However, there’s an increasing amount of data, including a nice talk by a postdoc from Julie Forman-Kay’s group, that the post-translational modifications alter the structural characteristics of the disordered state itself. The Forman-Kay talk suggested that phosphorylation induced a condensation of the protein by attenuating a surplus of positive charge.

This could conceivably be taken further. Consider a bit of sequence like DKRSDKA, which could conceivably take the form of a β-strand if it weren’t for that concentration of positive charge on one side. A phosphate group on the serine could conceivably stabilize this structure and preorganize it for binding to a ligand, thus increasing affinity by reducing the energetic cost of binding.

It might even be possible to tune things more specifically. Take a sequence like GRDSSKAKSR. If you put this on a helix wheel you’ll see a huge blast of positive charge on one side, but also a pair of serines. Phosphorylate S5 and S9 and you could stabilize the helix. At the same time, this would make a β-strand conformation less likely because such a strand would have negative charges on one side and positive charges on the other. By contrast, if you phosphorylate S4 you’d do nothing to stabilize the unfavorable charge concentration on the helix, but the positive charge concentration on the strand would be attenuated (see cartoon). In this way phosphorylation might be used as a kind of conformational switch to preorganize the same sequence in different ways and thus reach different downstream effectors. We know that conformational rearrangements of the kind that lymphotactin undergoes give rise to different signals and protein behaviors. The role of differential preorganization in disordered proteins hasn’t been extensively studied yet, but may be equally important.

It’s increasingly clear that disordered regions are a major factor in cellular signaling. I’m not having much luck with the one I’m working on now, but I’m excited to see where the next few years lead this field.

 Posted by at 5:11 PM