Julian, you are quite right that a disordered region of the protein chain will occupy a large region of conformational space, but there are some additional things to consider.

The terms "disordered" and "random" when it comes to protein structure are basically misnomers. The structures are never truly random and local sequence effects can be extremely important. "Disordered" proteins often populate ensembles of partially ordered/partially folded structure that can be defined, albeit with great difficulty experimentally, hence why MD or MC simulations are very valuable as predictive tools. The interconversion of these local secondary structure elements is particularly interesting, as such dynamics have implications for the physics of the remainder of the polypeptide chain, aggregation and disease, etc. Local effects can propagate global effects, especially when a flexible polypeptide links protein domains.

For "disordered" proteins, there is not an "equilibrium structure," per se, but there is definitely an equilibrium ensemble that is sampled. How extensive that ensemble is depends on the nature of the protein. Techniques like REMD (temperature and Hamiltonian exchange) have been used very effectively to sample the conformational ensembles of amyloidogenic peptides like A-beta and alpha-synuclein, and PrPSC, among others. The energetics of conversion between different conformational ensembles is very interesting and can give us mechanistic information about how structural transitions of "disordered" species propagate (dys)functional dynamics.

Difficulty

GROMACS and NAMD softwares can do molecular dynamics analysis. The problem is that this type of study can be a time simulation (greater than microseconds) and errors may start to arise if not treated properly.

Flexible-meccano: a tool for the generation of explicit ensemble descriptions of intrinsically disordered proteins and their associated experimental observables.

http://www.ibs.fr/science-213/scientific-output/software/flexible-meccano?lang=en

Techniques like locally enhanced sampling or replica exchange would probably be useful here. REMD would be pretty expensive with a system of that size, though. If structures are known or predicted for the terminal domains, you could restrain their configurations and allow the flexible region to move freely.

Well, it is a very challenging task. It also depends a lot on the size of your intrinsically disordered domain. You should first isolate just the region of your interest since it sounds like you have a multi-domain protein here. Then, in principle all-atom classical MD is not the best strategy since the conformational landscape of an IDP is quite complicate and you will end up having sampled just a small region of the FES of your protein. Also classical all-atom force fields encounter the risk of overcompaction for an IDP, even if some recent development in the fields are making available forcefields that might be good for IDPS (see for example A preformed binding interface in the unbound ensemble of an intrinsically disordered protein: evidence from molecular simulations.

Knott M, Best RB. PLoS Comput Biol. 2012;8(7):e1002605.) If you don't want to renounce to an all-atom description and the explicit solvent and your domain is very small, perhaps REMD approches can say you something useful (for example in terms of propensity to form secondary structures), but it has to be short domain otherwise I think (although I am not an expert on REMD), you will not sample that good the landscape also in this case. The best strategy is however to combine experimental data, as the ones from NMR or other biophysical spectroscopies to the simulations and the literature on this is quite broad. Also you can consider to apply some enhanced sampling techniques (for example metadynamics, or others... but again it is better to have a small target system).

I believe that in any cases simulations alone will not help that much and you will need some experimental validation.

i was wondering would elastic fluctuation models be a good approach. However, modeling of the ID region and energy minimization (EM) seems to be a problem as several residues do not fit in the favourable region of the ramachandran plot even after EM

Is challenging. All depends on your systems. Some methods have been tested with this purpose, with good results, at least were able to reproduce somehow the experimental data with they were calibrated. So, you have couple of options. You can use MD based methods, like REMD, or unbiased MD , or normal MD. Also some CG methods, like Martini model, has been used. Besides, you have the option of make Monte Carlo simulations, that I think that due to the rought enertgy landscape of these proteins are very suitable because avoid you to fall in a local minimu and remain there for the whole simulation or crash the simulation. Among these I recommend you a couple of codes used to this purpose. PROFASI, that incorporates also the Parallel tempering algorithm, like a Replica exchange, and CAMPARI. Both have been used for simulate IDPs.

But, you have take into account that until now all of them presents advantages and drawbacks over the others, so mostly depends on your systems and the experimental data that you have, although I recommend you MC techniques.

400 aa is a bit large. You can run molecular dynamics, but this may take several weeks depending on the your system.

And for sure if you want to simulate this huge structure, use some CG or classical MD, or make it by parts, The C-terminal, the main core and so on..., or also trying to divide the main core into smaller parts, try to take some experimental approach some NMR, PRE or SAXS, in that probably they break the protein in some fragments... use to be the normal strategy with IDPs

You may consider NMR. You may focus on a specific region rather than the full length protein. To study the dynamics of the specific region, you can use the synthetic peptide or produce it in E.coli.

Absolutely, as the colleagues already stated, Molecular Dynamics is a highly efficient method nowadays. I myself use it to study several conformational changes in small peptides to large protein ensembles, in different environment conditions. I suggest you to look for GROMACS or NAMD (both open-source - Free), or CHARMM, AMBER programs. It seems you're going to need to simulate lipid bilayers also, so I sugegst you to use GROMACS with Charmm36 force field, or the CHARMM program itself.

NMR spectroscopy

NMR is not likely to work on 400 residue membrane bound protein. If you remove the transmembrane N-terminal segment, you may be able to get a coarse grained description of the protein through small angle x-ray scattering or diffusion NMR if the remainder of the protein is small enough.

http://www.ncbi.nlm.nih.gov/pubmed/22821520

Thank you for the suggestions and references

I'm speaking from almost complete ignorance of the ways disordered (regions) of proteins have been studied so far. From the previous answers I gather that people have done simulations on disordered regions using MD simulations, or Monte Carlo simulations, and the like. I clearly realize that there will be sampling problems and potentential-energy-accuracy problems, since by definition a disordered region occupies a large portion of conformational space, too large to sample completely.

Now I wonder, as a chemical physicist, a lot of disorder is usually easy to deal with! The statistical mechanics of gases, for example, was the first to be solved. In classical mechanics, the two limits of complete order (absolute temperature going to 0) and complete disorder (T going to infinite) are easy: at T=0 nothing moves, at T=Infinite, the distribution of conformations is uniform.

There're a few ideas that come to mind, sorry if they're so trivial people have already considered them, I reckon this is a rather informal forum in which one can be somewhat speculative and hand waving...

1) If a region of a protein is disordered, why does it matter to simulate it? Couldn't one just assume that that region behaves as a typical random coil, they way most linear polymers behave in liquid phase, a problem that has been studied for decades, and of which the statistical physics is essentially solved? I mean, why do simulations for a problem that can be dealt with perhaps analytically?

2) If (1) doesn't work, i.e. if a disordered protein or region of a protein does not behave as a random coil: Why? To what extent? Can we still talk about disorder in such a case?

3) Manan Desai mentioned, in a previous comment, the possible use of Elastic Network Models. The very idea of an elastic network model is that there's an equilibrium conformation around which the network (protein) fluctuates. ENM's can and have been used to estimate the equilibrium distribution of the protein in conformational space. You can do that analytically, no simulation needed. There are regions of a protein (the network) that fluctuate with more amplitude than others (e.g. a loop, or an unfolded terminal region). However, to build an ENM you need an equilibrium structure. And I reckon by definition a disordered region of the protein does not have an equilibrium structure, so I don't think an ENM would work here...

4) If only part of the protein is disordered and part is ordered, i.e. if I have a protein that has, say, a well-defined ordered "structural core", so to speak, plus other parts of the sequence that move a lot (which is seen in pdb files as lots of sites that are just not there: no coordinates. I would, a priori, consider building a model that is an ENM for the well-defined regions, plus "random links" for the disordered regions....

Anyway, I just wanted to pose some issues that might lead to an interesting discussion...

Julian, you are quite right that a disordered region of the protein chain will occupy a large region of conformational space, but there are some additional things to consider.

The terms "disordered" and "random" when it comes to protein structure are basically misnomers. The structures are never truly random and local sequence effects can be extremely important. "Disordered" proteins often populate ensembles of partially ordered/partially folded structure that can be defined, albeit with great difficulty experimentally, hence why MD or MC simulations are very valuable as predictive tools. The interconversion of these local secondary structure elements is particularly interesting, as such dynamics have implications for the physics of the remainder of the polypeptide chain, aggregation and disease, etc. Local effects can propagate global effects, especially when a flexible polypeptide links protein domains.

For "disordered" proteins, there is not an "equilibrium structure," per se, but there is definitely an equilibrium ensemble that is sampled. How extensive that ensemble is depends on the nature of the protein. Techniques like REMD (temperature and Hamiltonian exchange) have been used very effectively to sample the conformational ensembles of amyloidogenic peptides like A-beta and alpha-synuclein, and PrPSC, among others. The energetics of conversion between different conformational ensembles is very interesting and can give us mechanistic information about how structural transitions of "disordered" species propagate (dys)functional dynamics.

Those are some really interesting points that you have discussed here Julian. I can provide a justification to only the first point so far... as the question is posed why to simulate a disordered region, well a part of this disordered region appears to cover/occupy space immediately in front of a binding pocket of the C terminal region. Previous structural studies have been performed..both NMR and XRD of this c terminal region alone and have shown the binding region essential for effective protein-protein interaction. Now my thinking is that if this disordered region or a part of it (to be precise) is covering this previously reported binding site.. then either the disordered are is involved in PPI or the binding pocket need not really be the binding pocket in reality. By removing the disordered region only a domain has been considered for interaction studies so far which according to me appears rather inaccurate. Therefore a simulation to observe the movement of the disordered are would help understand if not clarify entirely the true picture of PPI ..

As for ENM well again as correctly said, requires an equilibriated structure, minimizing the energy of the entire structure as much as possible and try to fix the side chain rotation to satisfy the Ramachandran plot..could be then used to observe fluctuations.. since there is no known way to equiliibriate disorderedness but at least this EM treated structure would be near equilibriated... however, this is my guessing based on what i have understood about protein dynamics...

Justin's answer is very instructive, thanks! As I said, I've no experience on disordered regions of proteins. To summarize, they're disordered in the sense that their equilibrium conformational distribution is multi-peak, and the interconversion interesting and possibly functional. I have the bias that I tend to think in terms of energy landscapes and equilibrium distributions, hence I'd never have come up with the term "disordered" for a multi-peak distribution... Hence my confusion. So, disordered regions need to be simulated because they're not so random as I thought...

Now, regarding Manan's issue, I'm not sure I understand how a "disordered" region would cover the binding site... If it is disordered, even if it's not completely random, it will spend only so much time in the minimae corresponding to conformations in which it covers the binding site, and the rest of the time elsewhere... I know nothing about your particular protein, but I reckon having a disordered region near a binding pocket might have functional importance: for regulation, for example, or for specificity. For regulation, I imagine a mechanism by which the substrate binds, such interaction stabilizes some particular conformation of the disordered region, and this switches the whole protein structure perhaps activating or inhibiting some active site far away... For specificity, the disordered region may "cover" the binding site just enough for the right substrate to bind... So, I wouldn't rule away a proposed binding site just because there's a disordered region nearby. On the contrary, I would look for possibly more interesting functional roles of the disordered region in substrate binding and, even more potentially interesting, switching the whole protein to an alternative conformation to regulate its activity.

thank you Justin and julian for the explanation and suggestions.

Another potential route to consider would be to generate a monoclonal antibody that recognises the protein in one of its disordered states but does not bind in its changed conformation. These are difficult to obtain but perhaps by employing peptides within any recognisable Molecular Recognition Features (MoRFs) within the disordered region your chances can be improved. My colleagues used this approach to identify a useful monoclonal to follow the involvement of the DELLA motif in protein-protein interactions of plant GRAS proteins (Sheerin et al. 2011 Biochem J. 2011 May 1;435(3):629-39. doi: 10.1042/BJ20101941.)

We recently developed a tool called DynaMine (doi:10.1038/ncomms3741) that predicts backbone dynamics from sequence alone. In our experience it gives a good indication of what the protein is 'capable of' in terms of dynamics, sort of a statistical potential - this might be helpful for you to at least get an idea of which regions are completely flexible and which ones might have a tendency towards (some) local order. We're working on a web server, should be online soon; I'm happy to post an update here if there is interest.

@Manan

"Previous structural studies have been performed..both NMR and XRD of this c terminal region alone and have shown the binding region essential for effective protein-protein interaction.Now my thinking is that if this disordered region or a part of it (to be precise) is covering this previously reported binding site.. then either the disordered are is involved in PPI or the binding pocket need not really be the binding pocket in reality. "

If you have an approximate prediction of which region of the disordered region is covering the binding site, you can check your hypothesis by a paramagnetic relaxation enhancement (PRE) nmr experiments by putting a spin probe near your predicted region If this region approaches the binding site, the signal in the NMR spectrum will decrease as the time averaged distance between the movable flap and the binding site (r^6 dependence, usually useful between 15-25 A distances).

Assumes the membrane binding region can be removed without consequence and you can get adequate resolution to identify the binding site in the spectrum

http://pubs.acs.org/doi/abs/10.1021/ja101645g

@Wim Vranken, it would be great if you could provide me with the URL for the web server.

Dear Manan,

We have recently developed a new tool that predicts backbone dynamics from sequence. Using this tool you get a picture of the stable and flexible regions that may occur in a protein.

"DynaMine is able to accurately distinguish regions of different structural organization within proteins, such as folded domains and disordered linkers of different sizes. Additionally, it can identify disordered regions within proteins with an accuracy comparable to the most sophisticated existing disorder predictors. Remarkably, DynaMine achieves this high performance without depending on prior disorder knowledge or three-dimensional structural information, which makes it a unique approach on the field as well as providing independent proof of the relationship between dynamics and structural disorder in protein regions."

I think this may be very useful for you. This is the reference;

Cilia, E., Pancsa, R., Tompa, P., Lenaerts, T., & Vranken, W. F. (2013). ncomms3741. Nature Communications, 4, 1–10. doi:10.1038/ncomms3741

We are also in the process of setting up a web server which you can find here.

http://dynamine.ibsquare.be/

Kind regards

Tom

Thank you Tom for the URL

@Lanaerts, @Vranken had already pointed out the possible use of your tool DynaMine. Quite frankly, I think this is not a forum for researchers to promote their own work so shamelessly. We're most of us involved in a candid and interesting discussion on the ways disorder can be studied, either experimentally, theoretically or both, what the functional implications are, and so on, and you guys keep advertising your own "tool".

You should have a look at forum etiquette rules on Advertising, for example in http://www.n-etiquette.com/forum.htm

I quote:

"The purpose of a Forum is a free and open exchange of ideas. It is understood that there are times when the idea itself requires that a product be mentioned, critiqued or evaluated."

"There are also times when the recommendation of a product is the answer to a posted problem."

"Blatant advertising on a Forum, however, is rude. Such postings should be removed."

I think you're guilty of Blatant advertising here, however rather than being removed, I'd like to rise this issue for discussion.

Anyway, since you keep insisting, let me tell you my opinion about the use of your tool: even though it may be helpful to "predict" disorder from sequence alone without reference to structure and/or knowledge about disorder, as you claim you can do, disorder is a feature of the ensemble of conformations of the protein, and a serious study of disorder must involve experimental and theoretical work at structure and dynamics levels, not just predictions from sequence, except may be just as a first step.

Regarding your "tool", however reliable it may be, it is a fit to empirical data, and, as such, it does very little to actually understand the underlying mechanisms of disorder, their possible effects and so on. Such an understanding will come from rigorous biochemical and biophysical experimental and theoretical/simulation studies based on mechanistic models, not from empirical models such as yours. Again, your model (or "tool" as you call it) may be useful for a first assessment, especially when you have no data other than the sequence, but it is of little help for further understanding of structural, dynamical, and functional issues.

@Julian my apologies for posting the link and reference. The only reason for doing his was to provide help to Manan with information to answer his question. I must admit I did not see the post of my colleague before I posted mine. I will keep the rules in mind next time I post something. Feel free to remove my post if you wish.

I take your point concerning the explanatory power of the predictive model. A predictor is off course not enough to unravel the mysteries of protein disorder. Yet i want to underline that the values predicted by the model have a specific meaning; they are the predicted S2 order parameters values, which normally are derived from NMR chemical shift data for the backbone. As far as i know this is one of the experimental ways to gain insight into the dynamics of proteins. As such the predictive model provides for everyone a first image of what one could expect. So, different from traditional predictors of disorder and order, we do not give a binary classification: we provide a value expressing the tendency of a residue within the context of the other residues. We are currently exploring whether these predicted S2 values also provide hints to some functional features of proteins.

Anyway, to anyone disturbed by the message I posted, I apologize.

@Julian you are completely right, my apologies, Tom and I got carried away in our enthusiasm. The more general point I was trying to make is that before you embark on (expensive) experimental and/or computational studies, it is worthwhile to get a 'predicted picture' of your protein, based on its sequence, that is as accurate as possible with regard to its probable behaviour. This can be done for membrane regions, secondary structure, possible structure homology, order/disorder, dynamics, functional sites, ... . It helps in making an informed decision on the regions to focus on.

Your problem sounds a lot like juxta-membrane segments in protein kinases. Usually such segments have functional roles, but may not be ordered in crystal structures. However, if they perform a specific function, residues in the segment may co-evolve with another part of the protein. If you know it has a specific function, then you can use co-evolution (for instance http://bioinf.gen.tcd.ie/caps/) to arrive at a decent model for inter-residue interactions between the disordered region and the rest of the protein. Using such distance restraints might give you a biologically meaningful structural model.

@Tom and @Wim, Predicting values of s2 does not amount to measuring them. I agree that it may be a first cheap thing to do, but the insight gained will be very limited, since your model is an empirical model, not a mechanistic one. By the way, s2 order parameters correlate with B factors and there's a large body of work on their prediction using sequence information, especially if alignments are used as input. My view is that such prediction methods are black boxes that provide very little understanding. I anticipate you will even have some success in "predicting" active sites, since they're usually among the most rigid (low S2) regions, so in so far as your model predicts S2...

@ Julian, It is clear that there is no substitute for experimental data. However, prediction methods based on empirical data and statistics can be extremely useful and quite accurate, think, for example, dihedral angle prediction based on NMR chemical shift values. This, of course, still uses experimental data, and I also agree that if you move to the sequence-only level the reliability of any method will be limited. That doesn't take away from their importance in assessing likely sequence characteristics before going experimental, and there is no real alternative when you get to the genome scale; a 'statistical' view can have its advantages.

I'd also like to point out that almost all experimental studies at the moment are oversimplified in that they use a reductionist approach to the protein environment (X-ray has its crystals, NMR uses artificially high concentrations of a particular protein without all the other proteins in there), so there are problems there as well, especially for proteins with disorder. It is also very hard to use mechanistic computational models in these cases, as conformational space is gigantic and hard to sample well; (limited) NMR data can clearly help here and is probably the way to go in Manan's case.

That said, it all depends on what the hypothesis is that you are trying to prove using experimental or computational approaches: I recommend to Manan to define the question he's trying answer more clearly before committing to particular methodologies (and here sequence-based approaches are useful because they give you an idea of what is probably going on).

@Wim Vraken, the question i seek to answer is about how this protein, which is more than 50% disordered, functions as there are very limited experimental data available on this protein. The purpose of playing with disordered regions is because of their importance in protein-protein interactions. As a result, i am trying to find a way to understand the role of these random coils via observing backbone fluctuations or anisotropic fluctuations.

@Wim, I completely agree with your last post. Especially when "experimental" data are not model-free. In X-ray, for example, data are diffraction patterns, all the rest, including "structures" and "B-factors" are model-fits, that depend on the method used. And there's also the issues you rise about the protein's environment not being their natural environment, and the dependence on experimental conditions (Temperature, pH, etc). "NMR" structures of the pdb are ensembles of model-dependent conformations constrained by NMR data (e.g. distance constraints) but otherwise generated using some model, so they're model-dependent too.

Hi Manan,

A recent paper of ours explores this very issue.

https://www.researchgate.net/publication/241690743_Differences_in_-strand_Populations_of_Monomeric_A40_and_A42

We found that using REMD to explore the conformational space of an IDP, and then using NMR data to refine the ensemble of structures sampled works well for characterizing a very heterogenous IDP. We also have a paper under review that explicitly compares different methods for developing the IDP ensemble using experimental data vs. MD. You may be interested in looking out for this work.

Article Differences in β-strand Populations of Monomeric Aβ40 and Aβ42

Thank you

Your best bet for a flexible structure is computation combined with small angle X-ray scattering - see Nature. 2013 Apr 25;496(7446):477-81. doi: 10.1038/nature12070 and Biopolymers. 2011 Aug;95(8):559-71. doi: 10.1002/bip.21638.