A point mutation remodels a binding interface

Decades of studies involving extensive mutagenesis of proteins and protein domains have impressed on us the idea that the folded tertiary structures of proteins are fairly resilient. While a particular mutation may abolish function by directly ablating a key chemical group, it is rare for a single mutation, or even a group of several mutations, to significantly change the overall conformation of a folded polypeptide chain. When a major change does result, it often takes the form of complete denaturation. Because of this, it may seem that protein folds occupy stable islands in sequence space, surrounded by a sea of sequences that form random coils or molten globules. However, there is some evidence that this view is mistaken, that substantially divergent structures may have very similar sequences. A paper recently published in PNAS adds to this view by describing a major change in the structure of a PAS domain resulting from 1-3 mutations.

PAS is a large family of protein-protein interaction domains contained in many signaling proteins. Although many of them have cofactors that modulate their binding, some PAS domains are constitutively active, which is the case for the domain under study here, the PAS-B domain from one of the founding members of the family, the aryl hydrocarbon receptor nuclear transporter (ARNT). The PAS-B domain was believed to dimerize with PAS domains from other proteins through one side of its β-sheet, and as a consequence Evans et al. decided to make several mutations on the outer surface of the sheet and monitor their effects on dimer formation.

One of these mutations, Y456T, had a strange effect on the domain's NMR spectrum: about 30 new peaks showed up in the ¹H-¹⁵N HSQC. Because this spectrum should show a single peak for each chemically unique proton-nitrogen pair, this result suggests that the pure proteins in the magnet exist in two distinct conformations. By lowering the temperature and repurifying the protein, Evans et al. were able to mostly separate these two conformations from each other. However, over time these conformationally purified samples became heterogeneous again, which means that the conformations can freely interconvert. This process was very slow, however — too slow to be detected using NMR relaxation techniques. From monitoring the HSQCs the authors concluded that the time constant for interconversion in the mutant is on the order of 16 hours.

Intrigued by what they were seeing, Evans et al. made additional mutations to ARNT PAS-B and found that the proportion of protein in each structure can be adjusted within a wide range by mutation. Using a triple-mutant system they were able to drive about 99% of the proteins into the new conformation. Using this mutant the authors were able to assign the resonances in the HSQC spectrum and solve the structure using NOEs. They learned that the chemical shift changes in the mutant are quite widespread, as you can see in the figure I have shamelessly stolen (left) for your benefit. In this figure the chemical shift changes are mapped onto their structure of the new conformation using color, ranging from green (very little or no change) to red (significant changes) — the sites of the three mutations are shown. As you can see, the chemical shift effect is widespread, covering almost the whole β-sheet of the protein and reaching to the helix on the opposite side.

Closer analysis of the structure shows why this is so. The strand of the β-sheet on which Y456 is situated has shifted its register by 3 spots. This has two effects. The first is that all of the hydrogen bonds involving that strand must be broken, at a substantial energetic cost. This is probably the reason the interconversion process is so slow. Moreover, because the register shifts by an odd number of residues, the strand must flip over, exposing to solvent the residues that were buried in the original structure, while burying the residues that were previously solvent-exposed. Although the backbone and side chain orientations in the strand overlay reasonably well between the two structures, the chemical groups available for interactions are completely different. Unsurprisingly, this alternate structure has very low affinity for its natural targets — titration experiments showed that the alternate conformation bound to its partner from hypoxia inducible factor at least 100x worse than the native structure.

Of course, in living cells with a wide variety of surfaces to bind to, the mutant PAS-B might find an alternative partner for which it has high affinity. Studies that attempt to understand protein-protein interfaces from an engineering or evolutionary perspective typically adopt the assumption that point mutations have little effect beyond adding a particular functional group here or there. This study indicates that this attitude underestimates the ability of point mutations to radically remodel interface surfaces. While a Y→T mutation may not seem particularly conservative, a Y→S mutation has similar effects and requires only a single nucleotide base change. It is not inconceivable that this alternate conformation could be reached in vivo, potentially giving rise to completely novel protein-protein interactions.

One might well wonder how common rearrangements of this kind are likely to be. As the authors point out, structural plasticity in the β-sheet is likely to be a common feature of PAS domains, making it difficult to assess whether this kind of mutational effect is widespread in other folds. However, the authors cite several examples of β-strand register shifts in other proteins. In addition, our decades of alanine-scanning mutagenesis have little to tell us about how common these kinds of rearrangements are, for two main reasons. First, the structural effects strongly depend on what a given residue is mutated to (see Table 1); had Evans et al. been content to leave things at alanine mutations they would never have detected this effect. Second, widely-applicable techniques for sensitively detecting small populations of alternate conformations have not been available until recently.

While the conformational transition induced by the Y456T mutation preserves the protein's overall fold and stability, it rearranges the hydrogen bonding contacts of the main β-sheet and shortens a loop. Obviously this is not as dramatic a change as found in lymphotactin. However, this alternate structure has significant consequences for PAS-B function. Unexpectedly, this single point mutation can radically remodel the PAS-B binding surface. Moreover, this result adds to the evidence that new structures (even in the context of known folds) may be accessed with only a few changes in amino acid sequence, without any need to detour through molten-globule intermediates.

M. R. Evans, P. B. Card, K. H. Gardner (2009). ARNT PAS-B has a fragile native state structure with an alternative β-sheet register nearby in sequence space Proceedings of the National Academy of Sciences, 106 (8), 2617-2622 DOI: 10.1073/pnas.0808270106

Read the rest...

Allostery in the CBP KIX domain

Classically, allosteric and cooperative effects have been identified with large complexes of multiple protein subunits, in which the binding of a ligand to one subunit enhances ligand binding in a different subunit. While some features of the models developed to deal with these systems do not translate well to cases of allostery within a single protein or domain, many of their core ideas continue to illuminate these single-subunit systems. In an upcoming paper in the Journal of the American Chemical Society, a team of European researchers examine an allosteric effect based on population shifts in a transcriptional activator, comparing it to a famous model for explaining allostery in hemoglobin (1).

The CREB binding protein (CBP) is a large molecular scaffold that brings pieces of the transcriptional machinery together in order to turn on a gene. Like many scaffold proteins it contains several protein-protein interaction domains linked together by large unfolded regions. One of these domains is KIX, a small bundle of helices that binds other proteins at two distinct sites. In one case, a protein called MLL binds to one site while a protein called c-Myb binds at the other. What is so interesting about this is that KIX is much more likely to bind c-Myb when it is already bound to MLL. Brüschweiler et al. used NMR techniques to try and understand how this happens.

In order to pull this off they performed relaxation-dispersion experiments on the amide nitrogen, α-carbon, and some methyl carbon atoms of the KIX domain bound to a peptide derived from MLL. Many of the amino acids in the protein showed a significant contribution to R₂ from exchange, suggesting a global conformational switch between two states. In order to cover their bases, the authors performed experiments to prove that this behavior was not related to the unfolding of the protein. Satisfied that the protein was stable, they used standard methods to calculate the rate of the conformational change, the population of the two states, and the chemical shift difference between them. They found that the minor state of the KIX-MLL complex is 7% of the total population of protein molecules. They also noticed that the chemical shift difference between the two states correlates very well with the chemical shift difference between the KIX-MLL complex and the KIX-MLL-c-Myb complex. Assuming that the conformation of KIX is the primary determinant of chemical shift in the bound state, this suggests that the dynamics are sensing a switch between a state that doesn't bind c-Myb and a state that does.

In order to determine whether MLL binding gave rise to this conformational switching behavior, the authors measured relaxation dispersion in KIX at several different concentrations of MLL. Excluding residues highly sensitive (by chemical shift) to MLL binding, they found that the exchange contribution to relaxation increases as MLL is added. Although Brüschweiler et al. were unable to fit this small number of residues quantitatively, these results strongly suggest that the addition of MLL increases the population of the c-Myb binding state. Moreover, under conditions where KIX forms a saturated complex with MLL and a peptide from another protein (pKID), the chemical exchange contribution to relaxation vanishes, suggesting that the protein has been pushed completely to the binding-competent state.

In order to identify the pathway by which the MLL binding site communicates to the c-Myb binding site, the authors examined the residues in KIX that had the largest chemical shift change associated with the chemical exchange behavior. As it happens, residues satisfying these criteria cluster in a region stretching from the MLL site to the c-Myb site, as you can see to the right (explore this structure at the PDB). Here, KIX is blue, the MLL peptide is red, and the c-Myb peptide is green. The side chains of the residues Brüschweiler et al. identify are shown as sticks inside the pink atomic surface. As you can see, these residues constitute a contiguous body stretching from one site to the other. Presumably, this set of residues provides a pathway for communication between the two sites. A trio of isoleucines at the core of this region (I 611, 660, and 657) are present in KIX domains from many different species (supporting information), suggesting that this communication pathway is evolutionarily conserved. Mutational studies centered on this trio of residues may teach us more about the mechanism of information flow in this domain.

Although this allosteric pathway is known to work in reverse (binding of c-Myb enhances the binding of MLL), the authors were unable to detect any exchange contribution to R₂ when only c-Myb or pKID was bound. While this may suggest that communication in the opposite direction uses a completely different mechanism, such that KIX has two unidirectional allosteric pathways, that's not a necessary conclusion from this result. Alteration of R₂ due to conformational exchange is dependent on the populations of the two states, the difference in chemical shift between them, and the rate of the switch. Actually detecting a dispersion curve requires that all these parameters lie within a 'sweet spot' that allows observation. This doesn't always happen, even when a dynamic process is occurring with a μs-ms rate. Because the chemical shift changes that result from MLL binding appear to be quite large (2) the exchange process may be slow on the NMR timescale.

One minor concern I have with the paper is that the experiments were carried out at a pH of 5.8, which is lower than the pH of cytosol (7.2). The only groups likely to change their charge over that range are histidines, but one of the key residues for this paper is H651 in the KIX domain. The experiments that established the allosteric effect of MLL on c-Myb binding (2) were performed at pH 7.0 so it is formally possible that the dynamics and allostery are a coincidence (although the chemical shift data argue against this). It would probably be worthwhile to perform HMQC experiments to clarify the protonation state of the histidine, or to repeat the binding experiments at a lower pH. The latter might be preferable; I assume that mildly acidic conditions are used for the NMR experiments because KIX has undesirable spectral characteristics nearer neutral pH. Additionally, it might be interesting to perform experiments that explore the effects MLL has on the kinetics of binding, seeing as this appears to be a dynamic process.

Brüschweiler et al. identify their results with the Monod-Wyman-Changeux model of allostery. Although this model was formally developed for systems with multiple subunits, what the authors really wish to emphasize is the idea from the MWC model that proteins in solution exist in an equilibrium of high-affinity and low-affinity forms. The evidence from the relaxation-dispersion experiments indicates that a very small proportion of free KIX exists in a (unfavorable) conformation that's ready to bind c-Myb. The binding of MLL enhances KIX affinity for c-Myb by stabilizing this structure — the allosteric effect arises because MLL binding defrays the energetic cost of adopting this fold. This manifests as a shift in the population of KIX proteins towards the binding-competent state. This kind of binding cooperativity may play a significant role in CBP's transcriptional activation function.

(1) Sven Brüschweiler, Paul Schanda, Karin Kloiber, Bernhard Brutscher, Georg Kontaxis, Robert Konrat, Martin Tollinger (2009). Direct Observation of the Dynamic Process Underlying Allosteric Signal Transmission Journal of the American Chemical Society DOI: 10.1021/ja809947w

(2) N. K. Goto, T. Zor, M. Martinez-Yamout, H. J. Dyson, P. E. Wright (2002). Cooperativity in Transcription Factor Binding to the Coactivator CREB-binding Protein (CBP). Journal of Biological Chemistry, 277 (45), 43168-43174 DOI: 10.1074/jbc.M207660200

Read the rest...

Blog of the Round Table: Hero's Blade

For the February Blogs of the Round Table, Corvus asks us to take a design provided by another contributor for the January Round Table and build on it. The idea is to develop a similar theme or mood in this second game, without necessarily remaining true (or relevant) to the literary work that inspired the initial pre-make post. In that vein, I wanted to riff off Corvus' "A Lego Orange" with an eye primarily towards its endgame. At that point he's trying to get across a sense of disenchantment with a once-loved lifestyle, which is what I'm going for in this idea for an unconventional fantasy RPG called Hero's Blade.

Hero's Blade is set in a fairly large valley with a road running through it. There's a village at the center and several outlying farms, a forest area, and a few fields and streams. The player controls an older hero in the "grizzled ranger" mode, armed with an excellent sword and bow, good armor, and so on. The game begins as the Hero wakes up in the village inn, learning that he got a knock on the head during a landslide that has blocked the road at one end of the valley. Don't worry, he doesn't have amnesia. The road at the other end has recently been taken over by bandits, so he is trapped in the valley until the road is cleared or he removes the bandits. The game world should at this point be presented in desaturated colors, as if the player is looking at the world through a filter that gives everything a gray tinge. Other than this, the graphics should be standard realistic 3D stuff.

The inhabitants of the valley are all farmers and, with one exception, are neither willing nor able to fight the thieves on their own. The exception is a single character (who I'll call the Kid) who is eager to go out and get rid of the bandits. He (or she) should have similar armaments to the Hero, but of much lower quality. The Kid is a very unskilled and totally untrained fighter, but possesses the same set of combat skills as the Hero. Combat in the game is turn-based and perfunctory — as little removed from a classic turn-based model as possible. There are no magical items or magic users, no elemental strength/weakness system, no need for any kind of strategy. The Hero's abilities are so far above those of the bandits that one of them can only stand against him for a turn or two. He earns no experience from battle at all. By contrast, the Kid starts off as a pretty bad fighter and earns experience points and skill boosts as is typical for an RPG. The Kid is AI-controlled, aggressive, and foolhardy; the Hero's only means of controlling him is a command that can knock the Kid down and take him out of the battle for a few turns. If either the Kid or the Hero dies in combat the game ends.

Each turn of combat and each bandit killed desaturates the color palette by a certain fixed percentage. If the saturation drops below a certain level, the game whites out, then fades to a steel-gray color and ends. This cutoff applies without respect to bonuses based on relationships or locations (see below). Thus, a player can get a game over by walking into the cemetery.

While the valley has several places where the bandits patrol, the rest of it is fairly open and features no real threats. The Hero can converse with the inhabitants of the village and the farm, and can establish relationships with each (about 50 total). Early stages of building a relationship involve just talking with the inhabitants, using a conversation system that allows the player to convey an attitude towards the NPC. For the most part these NPCs will do their own daily tasks without seeking out the Hero. The exception is the Kid, who can never be found if you're looking for him, but appears immediately if you call him (using a button solely dedicated to this task in the open world). The Kid always shows up out of nowhere to join in or watch if the Hero enters a fight or does a job that might be related to fighting or adventuring. As the Hero's relationship with an NPC improves, the inhabitant will offer quests to the player, some of them minor and some of them more involved, but never requiring combat. Several characters will also offer jobs.

These side-quests and jobs should take be the focus of the development. Each job should have some depth of gameplay, so that it carefully rewards strategy and player skill. Also, there should be some degree of challenge to each of them. Similarly, each quest should have extended moments of interaction between the character and the world. For instance, if an NPC has a missing cow, the player's job will be to find the cow and then guide it back carefully by coaxing it along and constantly nudging it in one direction or another. The goal is to require continual input from the player during each of these quests while varying significantly from quest to quest so that the player is always engaged during them (as opposed to the combat).

The performance of each job or sidequest will be discretized in some way so that as the Hero does more of each the colors of the world increase in saturation. Similarly, each time the Hero "levels up" his relationship with one of the NPCs, he the world saturates. In addition, NPCs with whom the Hero has a very good relationship will boost the color saturation in nearby areas. Spending time near his friends will literally make the Hero's day brighter. Only the Kid does not have this effect.

Taking a relationship with an NPC to a certain level will cause the Hero to open up to the NPC, with the Hero revealing some of his past. These flashbacks should take the form of slideshows of images illustrated in striking style and vibrant color, voiced over by the Hero narrating the event from his life. The life events he narrates will take a variety of tones, appropriate to the history and nature of the NPC. However, each of them will have a sad epilogue. His old friends have all died in battle, the women he danced with in taverns across the land have settled down and forgotten him, he outlived his favorite horse, etc. etc. The image shown in each epilogue should use the washed-out palette from the start of the game.

Taking a relationship to a high level will also help keep the Hero from dying by whiteout. In cases where the game would normally end due to the color desaturation, a good friend can find the Hero and boost the saturation level back to safety.

All of this is meant to make combat an unattractive thing to do, but after every couple of days that the Hero does not find and fight a bandit in the valley, the raiders will become emboldened and attack the town. If the Hero does not decimate them, one of the villagers will be killed. On the day after the attack, any bandits or villagers killed will be buried in the town cemetery in a scripted event; depending on who is getting buried the Hero's attendance or non-attendance will affect his relationship with all the NPCs. In the vicinity of the cemetery itself (as well as the bandit hideout) the color desaturates by a certain fixed amount.

In addition to these spots, the color will desaturate at any spot outside of the village where the Hero killed a bandit previously. These will be easily identified because the bodies will persist, slowly decaying with time. In town, bandits killed indoors will leave permanent bloodstains. No matter how faint the rest of the world gets these will always be clearly visible; they will fade to black rather than white.

The population of bandits is finite, however, and once they get reduced to about a fifth of their initial number they will hole up in their little fort and start sending out individuals to burn farms or kill villagers while the Hero is too far away to help. At this point the Hero will be urged to attack their stronghold. If he refuses to go, the Kid will go by himself and die there; the game ends. If the Hero accepts, the villagers will contrive a way to make the bandits open their fort and the Hero and Kid will take them on. The "boss" bandit leader is only marginally more skilled than the bandits themselves.

After this fight, victory celebration, yada yada yada, the Hero is awakened one morning by the Kid, who says that he is going to leave the valley and adventure across the world. The player can choose either "I'll go with you" or "You should take my sword." For the first option, the Hero's better friends drop by and say farewell. The Hero will start to walk out of the valley with the Kid at his heels, and the player's ability to redirect his motions will gradually diminish. As they pass by the bandit hideout, the player loses control entirely. A few straggling bandits attack shortly thereafter, with both the Hero and the Kid being completely scripted. Each turn desaturates the view significantly, and when the bandits are defeated in the 5th turn or so the screen whites out. Game Over.

If the player chooses the second option, however, the camera leaves the Hero behind and starts to follow the Kid, who is not yet in the control of the player. The colors return to the same washed-out look they had at the beginning of the game. As the Kid gets closer to the bandit hideout, however, the palette becomes more vibrant. As he continues on past, the saturation increases. When the bandits attack, the player is finally given control of him; at this point the Kid is better than them by a fair margin. Each turn of combat increases color saturation, as does each kill. After he triumphs, the Kid looks away from the valley into the very bright and vibrant lands beyond. Heroic music; game over.

The thing I'm really going for in this idea is the weariness of the Hero. He's good at fighting, much better by far than anyone in the valley. Once upon a time he found combat thrilling, but now it's lost whatever luster it had. He's tired of killing, but he's trapped in a particular life by the sword. The desaturation of colors with each moment of combat and each kill represents his growing weariness and depression. The combat is intentionally made un-fun and un-rewarding to encourage the player to feel the same way about it that the character does. The life the Hero finds in the valley refreshes and revitalizes him, hence the increased color, and so the conversations, jobs, and sidequests should be made rewarding and engaging to encourage a similar feeling in the player. The Hero finds one last use for his fighting skills and then passes his sword along to the Kid, who resembles the Hero's younger self in his outlook. Or, the Hero allows his spirit to fade and consigns himself to an empty life he has grown to hate. The choice at the end isn't really about player input, but about asking the player to reflect on his experience in the game and correctly identify what the character values.

Obviously, I'm also tweaking a couple of things about RPGs here. A typical RPG would be about the fresh-faced Kid who has made a decision to go out into the world and have adventures, and this bit back home in the valley with the weak enemies would just be the first chapter. The pointless monotony of combat grind is also something I'm trying to get at with the desaturation and the first ending to the game. Of course, this could end up just being a collection of minigames with a story attached, but that might be refreshing in its own right.

Please visit this month's other entries:

Read the rest...