Wednesday, October 2, 2013

Welcome! If you are new here, reading the posts Our aim and Why not submit to a journal? might be helpful to find out what this is all about. Also the post On credits might be of interest. Below is a short version of the actual science done so far.

First. Samuli Ollila wrote a preliminary manuscript demonstrating that the widely used Berger force field appeared to have some trouble correctly reproducing the lipid head group conformations measured in NMR experiments. Also responses to changes in hydration levels, ion concentrations and cholesterol content differed between experiments and simulations with the Berger model. Other force fields were not tested.

Then. During the last two weeks data for Stockholm-, Kukol-, OPLS-AA-, and the unpublished Maciejewski–Rog force fields have been made available by commenters (thanks guys!) of this blog.

Now. The plots below summarize the current status of our knowledge:
Figure 1. NMR results, original Berger data, contributions obtained through the blog so far, and the GAFF and CHARMM results taken from the publications for pure fully hydrated bilayers.
(Figure last updated on October 26th 2013.)


Figure 2. Data of systems containing NaCl for the Kukol and OPLS-AA force fields. (For these we currently have no data of pure fully hydrated bilayers and thus decided not include them in Figure 1.)


It seems that in addition to the GAFF and CHARMM, which were known to be good, also the Maciejewski–Rog force field gives significantly better order parameters than the Berger model.

Next. As always, all contributions are warmly welcome. However, here are maybe the most critical things we should obtain next:
  • Response of the good all-atom force fields (GAFF, CHARMM, Maciejewski–Rog) to changing conditions: hydration levels, ion concentrations, cholesterol content. If you have done simulations with these force fields and would be willing to either (1) share the input files for Gromacs or (2) report the order parameters here, it would be a great benefit for the project. (All contributors will become authors of the final paper, see the post On credits.)
  • Order parameters from Chiu et. al. united-atom force field. They have put effort in parametrizing headgroup torsions, thus this could be the united atom force field that gives a significant improvement over Berger. Here the input files seem to be available, but if ready trajectories already exist, posting order parameter data would be very useful. Update on Oct 8th: This FF has now also been tested, see comment by Samuli below. Figure 1 has been updated accordingly.
  • Comments, questions, criticism. There are errors in all manuscripts, and we would be very grateful if you would report should you have found one—no matter how small or big. For example, based on the feedback we have got outside this blog, there are people who have known for a long time that the headgroup is not well described in the Berger model. We would be very grateful if you could share your insights and, e.g., point out missing relevant literature in the manuscript. Another cool example of a very relevant contribution could be if an impartial NMR expert could provide us with their insight and write a post on the reliability of the experimental data used for the comparisons. But these are just examples. Please let us know if there is anything that you think we should have taken into account.

markus & Samuli

Tuesday, September 10, 2013

The manuscript is now available. What next?

This blog has been started to make improvements on the manuscript titled "Response of the hydrophilic part of lipid membranes to changing conditions — a critical comparison of simulations to experiments", written by O. H. Samuli Ollila, and openly available at http://arxiv.org/abs/1309.2131.

Here is the abstract of the manuscript: "We compare the order parameters predicted for the hydrocarbon segments in lipid bilayer headgroup region by the Berger molecular dynamics simulation model to those measured by Nuclear Magnetic Resonance (NMR) experiments. We first show results for a fully hydrated POPC bilayer, and then focus on changes of the order parameters as a function of hydration level, NaCl and CaCl2 concentrations, and cholesterol content. The experimental headgroup order parameters are never reproduced. This indicates that under all of these conditions the used model is unable to correctly reproduce the headgroup structure. Consequently, many of the conclusions drawn over the years from this model might be erroneous."

Although I think that the manuscript is a high quality scientific document, it has not been submitted to any journal, instead its contents are discussed here. The reasons for this decision you can find in the post Why not submit to a journal?

The ultimate goal of this blog is to find an atomistic (preferably united-atom) force field that reproduces the experimental properties discussed in the manuscript. Naturally the optimal situation would be that some of the already available force fields would fulfill this goal. If this, however, turns out not to be the case, the goal will be to find the appropriate modifications.

All the work will be progressed openly in this blog, and anyone is free to join in. In the end of the project the manuscript, updated with the improvements made via this blog, will be submitted to an appropriate scientific journal. Authorship will be shared among the blog contributors, as described in the post On credits.

My current ideas about things to be done in order to achieve the goal are described below. Some of these issues are straighforward to do, especially for the people who already have done something similar. For example, if you have done CHARMM simulations on a lipid bilayer with ions it is relatively easy to calculate headgroup order parameters and share the results through this blog. Some of the issues are not that straighforward. Below I have listed some tasks which would improve the manuscripts significantly. In addition of these issues, any other comments, questions and contributions are very welcome.

Things to do:

- The CHARMM [Klauda et al. JPCB 114, 7830 (2010)] and the GAFF [Dickson et al. Soft Matter 8, 9617 (2012)] force fields give better order parameters for the headgroup under full hydration. Interesting question now is, would these force fields also reproduce the correct changes in the order parameters as a function of dehydration, ion cholesterol concentration? At least for Sodium [Valley et al. J. Membrane Biol. 244, 35 (2011)] and cholesterol [Lim et al. JPCB 116, 203 (2012)] simulations have already been done with the recent CHARMM model. The calculations of headgroup and glycerol order parameters from these simulations would be very informative.

- For the all-atom Stockholm lipids force field [Jämbeck et al. JPCB 116, 3164 (2012)] I have not seen order parameters calculated for headgroup or glycerol. These would be interesting to see.

- I have not seen the headgroup and glycerol order parameters for the united atom lipid force fields based on GROMOS by Kukol [JCTC 5, 615 (2009)] and by Chiu et al. [JPCB 113, 2748–2763 (2009)]. Poger et al. [J Comput Chem 31: 1117–1125, (2010)] comment the order parameters briefly, but more information would be interesting. Fore more details and discussion see the post on Some points concerning the POPC full-hydration results.

- If it turns out that none of the existing united-atom force fields can reproduce the experimental order parameters under the studied conditions, we should find the way to fix the models. This is probably not a straightforward task. I have presented some of my ideas and opinions in the post on Some points concerning the POPC full-hydration results.

- The ion parameters play naturally a large role in the interaction between ions and lipids, making this a particularly complicated issue. I have written some ideas in the post on Some points concerning the ion-membrane interaction results.

All the files which are used to run the simulations and analysis for the manuscript can be downloaded from: http://dx.doi.org/10.6084/m9.figshare.790722

Everyone is invited to discuss, criticize and contribute to the manuscript through this blog. Communication and contributions can be done by writing the comments to the blog posts. If you pick one of the already determined tasks, it might be good to inform readers to avoid overlapping contributions.

Samuli Ollila
Aalto University

Some points concerning the POPC full-hydration results

Here is a summary of my current understanding on how the order parameters are reproduced by other force fields than the Berger:

In all-atom force fields CHARMM [Klauda et al. JPCB 114, 7830 (2010)] and GAFF [Dickson et al. Soft Matter 8, 9617 (2012)], the headgroup and glycerol group order parameters are relatively well reproduced. However, in the case of CHARMM it is not clearly reported if order parameter splitting is observed for the alpha and beta carbons.

For the Stockholm lipids [Jämbeck et al. JPCB 116, 3164 (2012)], I have not seen order parameters calculated for headgroup or glycerol. Similarly I have not seen comparison for the united atom models by Kukol [JCTC 5, 615 (2009)] or Chiu et al. [JPCB 113, 2748–2763 (2009)].

Poger et al. [J Comput Chem 31: 1117–1125, (2010)] write about their united atom model: "At 323K they found values of ν of approximately 5, 6, and 28 kHz for the methylenes α, β, and δ, respectively, yeilding the corresponding |SCD| values of 0.05, 0.04, and 0.22. The deuterium order parameters calculated from the simulations are consistent with the experiments with |SCD| values of 0.09 ± 0.01, 0.01 ± 0.01, and 0.16 ± 0.01 for the methylenes at positions α, β, and δ, respectively." They do not mention about splittings. To me it seems that these results are rather similar to those in the manuscript, but my interpretation was that the simulations and experiments are not consistent.

In my opinion, it would be useful to have a united atom model which would reproduce the correct structure for the headgroup and glycerol. It might be that either the Kukol or the Chiu force field already achieves this. This should, obviously, be tested first.

If these two do not work, there are many possible approaches that could be tried to fix the issue:

- Both CHARMM and GAFF developers relate the issue into the dihedral potentials. One way to approach the problem would be to take an all-atom dihedral potential that reproduces the experimental order parameters and transform it into the united-atom form. I do not have a clear idea, however, how this should be done in practise.

- It would be interesting to try to construct possible dihedral angle distributions from the experimental order parameters by some numerical fitting methods,  for example, with the methods presented by Thaning et al. [JPCB 111, 13638 (2007)].

- There are some suggestions in the literature how the headgroup should behave in order to generate the measured order parameters. Quoting from Ferreira et al. [PCCP, 15, 1976 (2013)]: " The first detailed model for the structure of the choline headgroup and glycerol backbone was built based on 2H and 31P NMR results from DPPC bilayers and crystallography studies [Seelig et al. BBA 467, 109 (1977), Pearson et al. Nature 281, 499 (1979)]. This model assumes rapid transitions between two enantiomeric choline states, a free rotation around the Cg1–Cg2–Cg3–O dihedral, and the assumption that the Cg2–Cg3 bond is on average perpendicular to the plane of the bilayer [Seelig et al. BBA 467, 109 (1977)]. Such description captures almost all NMR parameters; however, the last two assumptions are not compatible with the two distinct order parameters for Cg3, and thus other models were proposed e.g. in which the angle of the Cg1–Cg2–Cg3–O dihedral was completely fixed [Hong et al. Biochemistry 35, 8335 (1996)]." One option for us would be to directly include these ideas into the MD model.

- Finally, it might also be useful to remember that based on very recent NMR results, it seems that the headgroup and glycerol dynamics is too slow in the Berger force field [Ferreira PhD thesis].

Samuli Ollila
Aalto University

Some points concerning the ion-membrane interaction results

The topic of ion interactions with membranes is complicated since in the model it depends on both ion and lipid parameters. There are probably inaccuracies in both of these.

I think that the ion issue discussed in the manuscript could be divided into two different problems:
1) the partitioning of sodium ions is probably too high and
2) the membrane response to the ion partition is wrong.
It could be that the first issue is solely related to the ion force field and the second issue to the lipid force field, however also cross-relation is possible. For example, it could be that the Sodium partitioning is too high because membrane response has too low energy cost. Several other scenarios are also possible.

There is at least one thing which could be assessed to clarify the issue: The electronic polarizability. Recently, it has been suggested that this could be partially taken into account by scaling the charges [Leontyev et al. PCCP 13, 2613 (2011)]. It would be interesting to try out if this approach would improve the partitioning issue (1). I do not believe, however, that the membrane response issue (2) could be fixed simply by this approach; there is, namely, a problem in the membrane energies in all the other cases as well.

Samuli Ollila
Aalto University

Why not submit to a journal?

We believe that the manuscript Samuli has written is a high quality scientific document that could already in its current form be submitted to a respectful scientific journal. One might then rightly wonder, why did Samuli decide to make his manuscript freely available (http://arxiv.org/abs/1309.2131) and open for discussion in this blog, instead of following the traditional route of submission, peer review, revisions, publication? In the following we try to clarify the reasons for the approach taken.

Our answer to the question is threefold.

Speed-up of delivering the information. Samuli's manuscript shows that the widely used Berger model does not describe the lipid headgroup behaviour properly, and that this probably has led to several wrong conclusions being made in the field of lipid bilayer research. This is extremely important information for the field. If the manuscript would be submitted to a journal, the information would be available for the audience only several months later. We believe that this time would be better spent trying to use the information than waiting for the wheels of the publishing machinery to turn. This brings us to our second point.

Speed-up of solving the actual problem. Samuli's manuscript spotlights old experimental data and shows that one of the currently widely used models does not reproduce these data. This information, while crucial, is only an intermediate step towards the real scientific progress, which in this case would be to find an atomistic force field that can correctly describe the headgroup behaviour. Once this goal has been achieved, the current manuscript will have only historical value. We believe that it is better for the scientific community that the manuscript is openly updated as the project progresses, instead of writing-submitting-reviewing-revising-publishing several manuscripts in order to correct possible errors and improve the previous work. Since the project is open for everybody and the manuscript can be openly addressed with critisism and suggestions, there is no need for peer-review process. Also all the files and information are shared such that the experts in the field can reproduce the results.

Experiment with the way we do science. Finally, along its main goal, this blog is also an experiment on new ways to progress scientific understanding. These new ways have only been made possible in the recent years by the modern networking tools which allows us to communicate, share ideas and data very rapidly indendently on the physical location. Indeed, we feel that the Gutenbergian concept of an article printed in a journal, although having served the scientific community well in the past centuries, is not how optimal scientific communications would look like if they were to be redesigned today. Rather, the communications would be closer to adding small contributions in a collaborative manner to an open base of information. In lack of a better metaphor, what we see could be shortly described as something like a Wikipedia for science, a tree of knowledge in which new contributions (text, raw data, source code, new open questions, connections between fields, ...) are being constantly added. This particular blog is of course still far from this futuristic view, however, we feel that it could be a small step to that direction. In fact, there are already promising examples of cases where this kind of open collaborative approach has been successful in solving complicated problems, most famous probably being the Polymath Project; for readers interested in these topics we can recommend the book Reinventing Discovery: The New Era of Networked Science by Michael Nielsen.

As was done with the Polymath Project, also our plan is to eventually submit the final manuscript to a regular scientific journal. This is because the current citation (and thus funding) system in science still appreciates only articles published in peer-reviewed journals; other types of contributions, no matter how significant, are practically invisible to it. We hope, however, that the experiences from this blog will in their own part help bringing the scientific rewarding system from the good old times of the printing press to the internet age.


Markus Miettinen,
FU Berlin

Samuli Ollila,
Aalto University

Thursday, July 11, 2013

On credits


Samuli Ollila is currently writing a manuscript comparing the order parameters measured from NMR experiments to those calculated from the Berger POPC lipids, and he will share it via this blog in August 2013.

During the project, the manuscript will be improved based on  contributions to this blog. In particular, we hope to find an improved force field, but we stress that contributions can be anything that improves the scientific content. In the end of the project, the manuscript will be submitted to an appropriate peer reviewed journal.

Everyone who has contributed to the project by communicating via this blog will be offered coauthorship. The final decision on authorship will be to the invited individuals themselves. They should base their decision on self-assessment of their scientific contribution to the project.

We hope that all communications would be written such that they are assignable to a person and affiliation, as this will remove unnecessary ambiguity upon publishing.

Based on discussions in this blog, Samuli will handle the correspondence with the journal. The order of all names in the final manuscript will be alphabetical.

Finally, it is the responsibility of the people communicating via this blog to make sure that in case they write something in which other people have had a scientific contribution (such as supervisors writing comments based on discussions with their students), the identities of all contributors should be made known. These secondary contributors will also be offered coauthorship in the manuscript.



Markus Miettinen,
FU Berlin

Samuli Ollila,
Aalto University

Wednesday, July 3, 2013

Our aim

As Samuli Ollila pointed out in his talk (at the Biological membranes: challenges in simulations and experiments -meeting in Paris on June 28th 2013), there are plenty of high quality NMR data available on lipid order parameters. Our aim is to develop an all-atom force field that will match these data. Everyone is free to join us. This blog is our prime communication platform.