NMRlipids IVb: Preparing the manuscript about PE & PG results

Thanks to the various contributors, significant progress in the NMRlipids IVb project concerning PE and PG lipids has been made after the previous post.

First of all, as a response to the first ToDo issue in the previous post, two different codes to calculate the order parameters from united atom simulations have been contributed and also thoroughly discussed (see discussion in the previous post and buildH program). This enabled the analysis of previously contributed united atom simulations and the results are now added into the manuscript. Also, significant amount of new data has been contributed and we have now a good collection of force fields for both PE and PG lipids.

The data is still suggesting that the CHARMM36 simulations capture the essential differences between PC, PE, PS, and PG headgroups. A figure of structural ensembles of different headgroups from CHARMM36 simulations is now contributed.

Figure 1: Overlayed snapshots of a glycerol backbone and headgroup conformations of different headgroups from CHARMM36 simulations. Figure is made by Pavel Buslaev.

However, we need also the quantitative analysis of dihedral distributions to finalize the discussion. I think that we should also compare the P-N vector angle with respect to membrane normal between lipids with different headgroups using CHARMM36 simulations. Also, further ideas to characterize the structural ensembles of lipid headgroups in a comprehensible way are welcomed. I have opened an issue in GitHub for further discussions about this topic.

For calcium-binding affinity to PG containing membranes we still need more data. The current data is not fully converged, and I think that we need longer CHARMM36 simulations with the recent NBfix correction for calcium and counterions. I have opened an issue in GitHub with more details for further discussion.

I have opened also other issues in GitHub. All the issues are important, but the most critical for the progress of the manuscript are the above mentioned analysis of structural differences between different lipid headgroups and additional data for PG-calcium interactions. In addition to the issues listed in GitHub, there are also Todo points in the manuscript (tex file) and SI (tex file). For example, we need citations for the force fields in tables I-III, and to finish the method sections for MD simulation and NMR experiments.

The first annual NMRlipids workshop

Twelve contributors participated in the first annual NMRlipids workshop in Berlin from 15th to 17th of May 2019. The flexible schedule included six presentations and a significant fraction of time was devoted to discussions and working towards the goals of the workshop.

The workshop was opened by Samuli Ollila presenting the current status and open issues in the NMRlipids project (slides available here). The presentation was followed by a discussion to determine the main goals of the workshop:

1. Define a quantitative quality measure for the structural quality of a lipid bilayer simulation.
2. Create instructions on how to contribute data into the NMRlipids databank.
3. Create a universal code to visualize lipid structures in simulations (for example, see this figure).

The second day of the workshop was opened by Ivan Gushchin presenting a principal component analysis of lipid structures (slides available from here). This was followed by Patrick Fuchs presenting the status of analysis code to calculate order parameters from united atom simulations, and Tiago Ferreira presenting order parameter experiments using natural abundance ¹³C NMR. At the end of the second day,  Stephanie Dawson presented publishing possibilities offered by www.scienceopen.com. The third day of the workshop was opened by Ivan Gushchin presenting the relation between lipid bilayer densities and scattering form factors (slides available from here).

The main outcomes of the workshop are listed here (topics 1-3 correspond to the goals listed above and 4-7 are additional outcomes):

1. A quantitative measure for the structural quality of lipid bilayer simulations. The aim was to define a simple machine and human readable measure for the structural quality of lipid bilayer in simulations using scattering form factors and order parameters. The main applications would be the NMRlipids III project (see post and post and issue) and the lipid databank. For order parameters, the proposal was to measure the distance from experimental values using the size of the error bar as a unit (see the first page in the document). For the form factor, the task turned out to be more complicated and universal quality measure was not yet defined (see the pages 1-2 in the document).
2. Instructions to contribute data into the NMRlipids databank. The aim was to generate straightforward instructions on how to contribute data into the NMRlipids project in the way to enable automatic analysis and indexing. The suggestion was to update the current indexing system and read the metadata directly from hashtags given in Zenodo descriptions. Preliminary script to pull data from Zenodo was created, but it is not publicly available yet. I have opened an issue for further discussion.
3. Visualize structural differences between force fields and lipids. The aim was to create a code which automatically visualizes lipid structure by generating figures like this, and analyze the differences between PC, PS, PG and PE headgroup structures as suggested in the NMRlipids IVb project. The code for this was generated, but it is not yet publicly available.
4. The analysis code to calculate order parameters from united atom models by Patrick Fuchs is progressing, but it is not yet publicly available. The merging with the current order parameters calculation code was superficially discussed. See also the discussion in the blog after the workshop.
5. In the discussion following the presentation by Tiago Ferreira, we concluded that the sn-1 chain order parameters from natural abundance ¹³C NMR may be less accurate than 0.02 (the previously used quantitative accuracy of order parameters) due to the spectral overlap. The order parameters from specifically deuterated samples are probably more accurate for this region. This is relevant for the quantitative quality measure and for the NMRlipids III project. Further discussion is in the issue.  
6. In the discussion following the presentation by Ivan Gushchin about lipid bilayer density profiles and scattering form factors, we concluded that some of the complications in the NMRlipids III project could be potentially resolved by separately analyzing the effect of POPC and cholesterol densities to the form factors from POPC/cholesterol mixture. 
7. The possibilities of using Markov state modeling presented by Ivan Gushchin (slide 5) to understand lipid structures and transition rates were discussed. 

The workshop was overall successful and the second NMRlipids workshop will be most likely organized at least partially with the same concept.

Explanation how to measure order parameters with ¹³C NMR

NMRlipids III: Quantifying intermolecular interactions in binary lipid mixtures

After trying to rationalize the data in the NMRlipids III manuscript, I think that we should emphasize more the evaluation of lipid-cholesterol interactions in binary lipid bilayer against experiments. As far as I know, the quantitative comparison of intermolecular interactions between simulations and experiments has not been done much. Therefore, the procedures in this manuscript could be useful also for other than POPC/cholesterol mixtures.

To facilitate this, I plotted the absolute values of POPC sn-1 acyl chain order parameters as a function of cholesterol concentration from experiments (Fig. 1) and different simulations (slopes*pdf, see also the SI).

Fig 1: Absolute values of sn-1 acyl chain order parameters from experiments (points). The lines are fitted to the values with 0-50 mol % of cholesterol.

The increase is approximately linear below equimolar concentration in both simulations and experiments. To compare the ordering effect of cholesterol between different simulations and experiments, I made a linear fit to the data as a function of cholesterol and plotted the slopes for each carbon segment (Fig. 2).

Fig 2: Slopes of order parameters as a function of cholesterol. Determined by fitting equation S_CH(C_chol) =k_OPC_chol + S_CH(0) to the data in Fig 1 (experiment) and slopes*pdf (currently S2-S7 in the SI).

Such a plot enables the comparison of cholesterol ordering effect between different simulations and experiments, even though the data would not be measured exactly at the same concentrations. We could also add some literature data with fewer cholesterol concentrations or other lipid media in this figure for comparison.

In addition, the quantitative quality measure for the force field quality, recently contributed by Hanne Antila, could help to evaluate the intermolecular interactions. Some preliminary results are already in the manuscript, but some details in the analysis may need refinement, see issues 65 and 64 in the MATCH repository.

The most important ToDos to progress the manuscript are listed here:

1. We need to define a appropriate quantitative quality measure to be used in this manuscript, for discussion see issues 65 and 64 in the MATCH repository.
2. The changes in cholesterol order parameters upon dilution with POPC could be also analyzed from experiments and different simulations.
3. Results from Lipid14 simulations with cholesterol model would be useful, see issue 7.
4. Additional data from the literature could be added in the figure showing acyl chain order parameters slopes (Fig. 2).
5. Figure showing form factors should be updated after all the form factors are calculated properly and the scaling method for experimental form factors is decided, see issue 18.
6. Section S1 in the supplementary information about reproducibility of CHARMM36 simulations should be written based on the discussion in issue 4. 
7. Section S3 in the supplementary information about the effect of undulations on the order parameters should be written based on the discussion in issue 16.
8. The used accuracy for the experimental acyl chain order should be decided, see issue 15.

 

NMRlipids IV: Toward submission of the manuscript about PS lipids (2)

The manuscript describing results about PS lipids from the NMRlipids IV project has progressed substantially since the previous blog post, and I believe that it is almost ready for submission now. However, we cannot proceed with the submission before we have simulation details from the data delivered by Tom Piggot, Jesper Madsen and Josef Melcr, see ToDo points in the supplementary information.

I have now changed the title of the manuscript to "Headgroup structure and cation binding in phosphatidylserine lipid bilayers". Let me know if you think that we should add molecular dynamics and/or NMR experiments in the title, or if you have other suggestions.

If you have any comments regarding the manuscript, please let me know. I would like to proceed with the submission as soon as possible. I think that a suitable journal for the manuscript would be, e.g., The Journal of Physical Chemistry B or Physical Chemistry Chemical Physics. Let me know if there are opinions about this.

NMRlipids IVb: Assembling the PE & PG results

As discussed in the previous post, the NMRlipids IV project concerning PE, PG and PS lipids was divided into two parts. The first part about the results from PS lipids is approaching to the submission, while the second part about PE and PG lipids is in a more preliminary stage.

I have now started to assemble the contributed data for PE and PG lipids into a manuscript in a new GitHub repository. From now on, I will call this part of the project as NMRlipids IVb. Current status of the manuscript and most important things to todo are summarized in this post.

Headgroup and glycerol backbone order parameters from lipids with different headgroups in experiments

Experimental order parameters delivered by Tiago Ferreira with the information about the sign revealed that the 𝛽-carbon order parameter of PG lipids is positive, while other lipids have negative order parameter for this carbon (Fig. 1). Therefore, the structure of PG lipid seems to be distinct from PC and PE, in contrast to the conclusions from the order parameter data without sign information in the previous post.

Figure 1: Experimental order parameters of the headgroup and glycerol backbone region C-H bonds from experiments. For literature references, see the manuscript draft.

 

Headgroup and glycerol backbone order parameters of PE and PG lipids in simulations

The experimental information about order parameter signs enables a similar comparison between different simulation models and experiments for PE and PG lipids as done previously for PC and PS lipids in NMRlipids I and NMRlipids IV projects. This comparison (Fig. 2) and the results in NMRlipids I and NMRlipids IV projects suggest that the differences between PC, PE, PG, and PS headgroups are well captured in CHARMM36 simulations, although order parameters of all C-H bonds are not within the experimental error for any of the lipids. Therefore, the structural differences between lipid headgroups could be analyzed from the CHARMM36 simulations, perhaps in a similar way as done for the PS headgroup in the NMRlipids IV project. The contributed data for PE and PG lipids from united atom simulations is not yet analyzed, partly because of the lack of suitable programs for the automatic analysis of the data.

Figure 2: Headgroup and glycerol backbone order parameters from PE (left) and PG (right) lipid bilayers from experiments and different simulation models. For literature references, see the manuscript draft.

 

Cation binding to lipid bilayers containing PE and PG lipids

Sodium binding to PE lipid bilayers seems to be slightly weaker than in corresponding simulations for PC lipids, but experimental data is not available. For calcium binding to PE lipids, we do not have experimental nor simulation data yet.

As discussed in the NMRlipids IV project, the counterion binding to negatively charged lipid bilayer is complicated to assess against experimental data. However, the results from PC:PG mixtures from CHARMM36 and Slipids simulations indicate that sodium ions could be slightly overbinding to PG lipids, similarly to PS lipids in the NMRlipids IV project. For calcium binding to PG containing lipid bilayers, we have data only from CHARMM36 simulations and the conclusions are not yet fully clear.

 

ToDo:

1. Analyze united atom simulations of PE lipids (GROMOS, CHARMM36ua, OPLS-UA, Berger, and GROMOS-CKP) listed in Table I in the manuscript with or without the new code for united atom data.
2. Analyze structural differences between POPC, POPE, POPG and POPS headgroups from CHARMM36 simulations. The most reasonable first step would be probably to follow a similar analysis that was done for PS lipids in the NMRlipids IV project.
3. Simulation data for PG lipids with different force fields would be useful. Currently, we have only CHARMM36 and Slipids.
4. Simulation data from POPC:POPG (1:1) mixtures with sodium counterions and different CaCl₂ concentrations (0-1M) from different force fields would be useful. Currently, we have results only from CHARMM36.