Thursday 25 April 2013

WATSURF 2013: Day 9 and interesting papers

Today is the second-to-last day of the conference, and was somewhat more simulation-heavy.

Mounir Tarek (U. of Nancy, France) gave a two-part talk.  The first part was titled "Insight into 'bio-systems' hydration water dynamics from Molecular Dynamics simulations".  The main point was to try to make the connection between hydration water in MD with the results of neutron scattering, and in this, he was very convincing.

First, he looked at ribonuclease at "full hydration" (0.35 g water / g protein).  In MD, you see that you can't build even the first hydration layer with this amount of water.  Instead, you have a protein powder with a bit of confined water here and there.  However, A. Oleinikova (JPCB, 2005) has shown that the water *does* form a spanning network, which mediates long-range transmission of correlated motions of the protein.  In any case, if you simulate this system at low T  (100 K), you only see quick beta relaxation in I(Q,t) (single-particle motion), then structural arrest.  As you raise the temperature, around 150-200 K, you start seeing some motion.  In Tarek et al JACS (1999), they show that they can nearly reproduce the protein dynamical transition observed in Doster et al, Nature (?), 1989.  Around 230 K, all sort of groups on the proteins start moving, so it's not just one kind of motion that's starting.  MD of RNase hydration water at 300 K: I(Q,t) at long times is well-fit by stretched exponentials.

So what's happening in the MD simulation?  Can characterize H-bond dynamics in terms of fast motions (survival times tau_HB of a single bond) and slow network rearrangement (correlation time tau_R of the indicator function of a single pair of bonds, that can break and reform many times before definitely breaking after a collective rearrangement).  What they find is that it is tau_R that takes off around and below 200 K, whereas tau_HB is mostly unaffected.  Another computer experiment: put harmonic restraints on oxygen atoms: get a similar effect on tau_R as in cooling.  So there is a connection between slow dynamics and restraining of waters into a particular position.

Glycerol, which is used as a bioprotectant for pharmaceutical industry, slows down motion in proteins (MSD observed in neutron scattering is intermediate between dry powder and hydrated protein powder).  In the MD simulations, what they see is that the protein dynamical transition is happening at the moment when the *solvent* starts moving, whatever the solvent is.  Glycerol just starts moving at a higher temperature than water.

Now for something different: coherent motions of proteins.  If motion is describable by generalized thermodynamics, then S(Q, omega) is a sum of Lorentzians (modes), Rayleigh peak at omega = 0 (heat) and a Brillouin doublet around omega = omega_s (speed of sound).  From S(Q,omega) in simulation, can obtain dispersion curves for phonons in sample (optical and acoustic modes).  In early MD simulations of just water, find different dispersion curves for transverse and longitudinal modes.  Tarek and Tobias have looked at dispersion curves of water next to proteins, find good agreement with experiments.

Of course, getting dispersion relations in this way is an artifact of what you can measure in neutron scattering experiments.  In MD, they can directly measure current correlation functions (<J(x,t) J(0,0)>, where J(x,t) = sum_i v_i(t) delta(x - r_i(t)) to obtain dispersion relations.  They did this for a system of maltose binding protein crystal at 0.42 g / g hydration.  Water dispersion (transverse and longitudinal modes) are similar to bulk.  But if you just focus on waters close to the protein, then you start seeing different things: new modes start appearing at high energies (high frequencies).  You can correlate these new modes in water to collective modes in the protein.  This is a direct demonstration of coupling between protein and water dynamics.

Once you have access to the current correlation functions, you have access to macroscopic elastic properties (e.g., adiabatic bulk modulus, longitudinal modulus, shear modulus).

In the second part of the talk, he discussed "Electroporation of lipid membranes: insight from molecular dynamics simulations".  Electroporation = put a cell in a high electric field, if it's high enough, make pores in cell membrane, either transiently or permanently.  Reversible electroporation can be used to insert drugs into cell, e.g. chemotherapy delivered directly to sick cells, which reduces the total dose => reduction in side effects.  Used medically today.  Irreversible electroporation finds application in food processing: exploding cells from grapes, beets and oranges respectively increases extraction of wine, sugar and orange juice by a factor of 2-3.  There's a big group of people interested in this stuff.

They use MD simulations to understand the basic physics of electroporation.  To put E-fields from MD into context, he mentioned that you can now experimentally apply huge E-field pulses (O(1 MV/cm)) for about a nanosecond.

Inside membrane, there is a huge potential difference (about 1V, exp 0.3-1.5 V) between outside and inside of hydrophobic core.  Experimentally, this rationalizes different solubilities of fat-soluble ions.  In simulations, measured directly by Wang et al, PNAS, 2006.  So around the membrane edge, have an E-field of about 1V/nm.  If you apply an electric field across the membrane (no ions), you rearrange the water dipoles on both sides of the membrane, result in a potential difference between both sides of the membrane.

Ideally, you would look in simulations at vescicles 10-20nm in size: not possible, do it in planar membranes.  Observe pores forming in membranes in simulations (after a few ns) upon applied electric field (about 2-2.5 V across membrane).  The physics: dipoles of waters in both sides reorient, create a large sustained electric field inside hydrophobic core, to which waters outside are strongly attracted.  Once water finds its way there, head groups on membrane rearrange to "seal" edges of pore.

They've simulated transport of siRNAs across membrane via a few ns E-field pulse: it works and it's been experimentally realized.

If you replace amphiphilic lipids by just negatively charged head group, then head groups *do not* stabilize the pores that form, so pores exist only during the length of the E-field pulse.



The following talk was by Benjamin Bouvier (IBCP, Lyon, France) talked about "Water in biomolecular recognition processes: the forgotten partner?"  He began with a summary of what makes water interesting for the purposes of biology, which I thought was quite nice.  Here were the main properties he suggested: a) High heat capacity => oceans act as heat sinks & sources, stabilising climate; b) Density maximum at 4 C => lakes and rivers freeze from the top, life can continue below. c) Water vapor absorbs IR, "greenhouse effect", keeps Earth warm (a bit too warm these days!);  d) High surface tension = important for dynamics in biology e) Universal solvent.

What about water in the living cell?  50-400g/L of biomolecules, occupy 20-40% of cellular volume => high crowding.  Lots of contacts form randomly.  Can water help distinguish random contacts from productive one?  Upon structural rearrangements, there are large numbers of waters either released or taken up, so they make a large contributions to association/dissociation energetics, despite low contribution of individual molecules.

From here on, most of the talk was a pedagogical introduction to MD from the point of view of biological simulations (the conference is, after all, supposed to be a school).  But there were some results at the end:

His work focuses on the role of water in protein-protein and protein-DNA interactions.  To motivate, he showed an example of one antifreeze protein, which recognizes one of the faces of ice Ih and attaches to it, stopping growth of crystals (Madura, J. Mol. Recognition, 2000).

DNA and water: in the minor groove, you have a fairly ordered chain of water ("the hydration spine"), H-bonded to bases or bridging phosphate groups on either strand.  The next two layers are somewhat structured as a consequence, and this transmits sequence information to outside the DNA.  Binding lifetimes vary depending on sequence.  There are specific water cluster structures at certain sequences, e.g. at the boundary between a GC-rich and AC-rich region.

He first mentioned looking at ubiquitin.  Binds other ubiquitin via lots of lysine residues on surface, building poly-ubiquitin chains.  Different polyubiquitins signal different thing, e.g. a protein bound to di-ubiquitin (linked by the Lys 48 residue) is marked for degradation by the proteasome.  The same hydrophobic interface in ubiquitin is used to bind other ubiquitins and the proteasome.  They calculated a PMF for ubiquitin dimer as a function of minimum distance between any atom pair (much smaller bias than, say, distance between monomer CoMs).  The result is unpublished (so can't talk about it here!) but it's submitted to JCTC.

Now for protein-protein recognition.  The current idea is based on "hotspots": <5% of residues contribute non-negligibly to energy of binding (Bogan, Thorn, J. Mol. Biol., 1998).  Can hotspots be predicted? (would be useful for drug targets).  How about monitoring the behavior of water around proteins?  Can areas of low solvent flow correspond to free energy hotspots (maximize the hydrophobic effect)?  Such residues would be called "dehydrons" (Scheraga et al, PNAS, 2003).  One possible statistic would be to calculate average inverse lifetimes of waters in contact with each residue.  Hard to sample and requires expensive simulations.  What about evolutionarily conserved residues?  Very little statistical power.  However, there is some correlation between dry residues and conserved residues.

One idea to identify "dry residues" almost statically is to quickly solvate the protein, equilibrate for a tiny amount of time, then perform a Voronoi tesselation of space according to heavy atom positions (some correction for different atom sizes).  Using this, you can map the "depth" of a residue inside a protein by measuring distances from residue to bulk water region in units of Voronoi cells (the details went by too quickly for me to take notes, but it's described in Bouvier, Proteins, 2009).  Apparently, there's a good correlation between this "depth" and dryness of the residues.

Now about protein-DNA recognition.  Two mechanisms: direct recognition (bonds formed with interfaces) and indirect recognition (structural or mechanical differences w.r.t. B-DNA due to sequence).  Looked at one particular DNA-protein interaction (Bouvier & Lavery, JACS 2009).  The protein barely deforms upon binding/unbinding, but the DNA bends a lot.  Also, they find that the hydration "spine" in the minor groove that the protein has to displace moves collectively.

They also simulation the pathway of a drug (daunmycin) sliding along target DNA sequence, PMF minimum correlates with experimental binding position, and barriers can be correlated to firmly attached waters on the DNA that are hard to displace by the drug.



Now for some interesting papers that came out today:

* G. Grazziano, "On the signature of the hydrophobic effect at a single molecule level", Phys. Chem. Chem. Phys., 15, 7389-7395 (2013): A putative alternative explanation for the experiments of Li and Walker on the free energy changes upon hydrophobic collapse of a homopolymer, measured by AFM.  He claims that their rationalization by LCW is illusory; as far as I could tell from a cursory reading, he claims that the results can be understood entirely in terms of changes in the total solvent-excluded volume, and surface tension plays basically no role.  The parabolic temperature dependence of per-residue solvation free energy changes (in the LCW-language, due to a shift in the crossover distance from above to below the monomer size) is then explained in terms of temperature dependences on the effective size of each monomer and some link to scaled-particle theory, as well as a rotational entropic loss of the monomers when going from extended to collapsed states.  To my ears, this all sounds extremely fishy, but perhaps it's worth a closer look some day during a slow afternoon.

* Burns et al, "Self-Assembled DNA Nanopores that Span Lipid Bilayers", Nano Lett., Just Accepted Manuscript (2013): A 3D DNA-origami of an open tube, where the DNA in the middle is chemically modified to be hydrophobic.  The barrel inserts into lipid bilayers and acts as a pore.

* Abascal et al, "Homogeneous bubble nucleation in water at negative pressure: A Voronoi polyhedra analysis",  J. Chem. Phys. 138, 084508 (2013): Not from today, but Chantal Valeriani sent me this paper of hers and her co-workers after reading our work on evaporation.  They look at the nucleation of vapor bubbles in TIP4P/2005 water at 280 K under negative pressure.  They have a nice trick for identifying which molecules are at the cavity interface based on a Voronoi tessellation of space, so can monitor the sizes of small, transient cavities.  Just above the spinodal line, using a neat mean-first passage time (MFPT) analysis due to Wedekind et al, they measure the nucleation rate and critical bubble size in a way that's only mildly sensitive to their exact definition of cavity volumes.  They find that CNT describes the critical bubble size well, but underestimates the rate by 8 orders of magnitude (not too surprising, given that CNT assumes that the cavity free energy is of the form gamma * area + Delta mu * volume, which isn't true for small cavities in water).  Almost by definition, they find that just below the spinodal line, bubble growth proceeds through spinodal decomposition, but they show how this growth shows up in an MFPT vs. bubble volume plot.  I wonder if they would find similar results using the Willard-Chandler interface to estimate cavity sizes.

* Feng et al, "DNA Patchy Particles", Adv. Mater. (2013): An experimental realisation by the NYU group of a colloid coated with one type of DNA on most of the surface, but a different type in a single (small) patch region.  It's conceptually easy (Fig 1 of their paper says more than I could by writing), but it's good to see it done in experiments.

No comments:

Post a Comment