I am working with a membrane protein, which until put on the grid, looks homogeneous and stable but once vitrified, gets severely aggregated, see attached picture of a representative micrograph.
What can I change to get nicer particle distribution and minimize aggregations? I also lose many particles on the carbon.
I am using holy carbon grids here. I glow discharge my grids at 20 mA for 60 sec.
I also need to add a compound in DMSO and the final concentration of DMSO in my to-be-vitrified sample is 6%. Is there a trick to having less DMSO on the grid, given the compound is very hydrophobic but has good affinity to the protein? P.S. aggregation happens regardless of the DMSO but gets worse with it.
in my opinion, this actually doesnt look too bad. I have seen worse grids that ended up yielding less than 2.5 A. I would give it a try and collect a test dataset (1-2 hours depending on your system) to see if you can get some high-resolution 2D classes.
For DMSO 6% DMSO is a bit high final concentration. Going from 100% DMSO 100x stock → 1% DMSO 1X in buffer can be too large a change leading to buffer shock and crashing out, while an intermediate 10% DMSO will work. For the intermediate for some hydrophobic compounds having 10X 10% DMSO in buffer won’t work because of the salt while 10X 10% DMSO in water will work. With water intermediate this ends up reducing the final buffer conditions when doing the 10% DMSO 10X → 1% DMSO 1X as your 500 mM salt becomes 450 mM salt in the 10x dilution. Working with less salt can certainly help, say going down to 150mM or less. Can try other solvents DMF,EtOH etc with the usual buffer non-drug control for comparison. Finally the way you mix it can matter, adding 1 ul 10X to 9 ul of buffer versus 9 ul to the 1 ul can have an effect for precipitation. One last note for membrane proteins if under detergent conditions, the detergent can have a large effect on the hydrophobic drugs, seen a drug with very little density in one detergent and clear density in another.
I agree with @mruetter; I’ve seen grids with worse conditions reach high resolutions, so depending on availability of microscope time it might be worth trying a short run anyway.
Comment regarding DMSO additive from @T_Bird is good too; like ammonium sulphate precipitation, local concentration can cause big headaches when adding DMSO or similar.
Losing protein on carbon film is pretty usual; if distribution is a problem, have you tried a carbon support film?
thank you for your reply. I have already collected on a similar grid and did not get many good particles, that is why I am trying to solve the problem. Hopefully using an intermediate DMSO stock would help. About the carbon, I think it’s a minor issue compared to my aggregation problem, so I will try to solve the latter first and see if I can get enough particles.
Can you show us some of the 2Ds from the previous grid? That could help us understand what is going on in the processing side of things (i.e. if box size, particle size, etc. are correct for your particle).
What detergent are you using for the membrane protein? Playing around with different detergents (as well as lowering the DMSO concentration as others have mentioned) can help. You can take a look at some of the “Troubleshooting Specimen Preparation” tips here: Cryo-EM Posters | Chemical Biology TEM at Dana-Farber Cancer Institute (aggregation on the grid is usually a side effect of AWI denaturation). We also have a list of common detergents of different charges to try on the pricing page of our website, although concentrations are aimed at non-membrane proteins: Pricing | Chemical Biology TEM at Dana-Farber Cancer Institute .
However, I would want to see the 2Ds from the last data set you collected first before you start messing around with detergents. Generally especially for membrane proteins, if they are happy then that is a good detergent to keep them in. You could try comparing +/- compound with negative stain, and also do a no-compound cryo condition to understand at what step of the process the “aggregation” is occurring.
Hey, Thank you for responding. Here are some 2D classes: These are from 7000 micrographs. I use LMNG/CHS, in which my protein is very happy before vitrification. For my particle size, it’s around 60 kDa inside the micelle and ~130 kDa overall. I measured the average particle diameter to be 210 Å and box size to be 510 px (twice particle size). pixel size is 0.828Å
Maybe some membrane folks can chime in here, but those look pretty promising for a first pass to me. Have you tried template or Topaz picking the micrographs after selecting only the good 2D classes and using those as templates to re-pick?
Can you specify what you mean by “better 2Ds”? Looks to me like you have ~50k single particles that are in OK 2D classes for a first round of 2Ds for a membrane protein. You clearly have a lot of classes that are just averaging to nothing at all, which is probably a sign that your picking parameters need to be optimized, but you can also try increasing your classification uncertainty factor to separate out the good particles further (or just do fewer classes in the beginning and multiple rounds of 2D, which you’ll almost always have to do anyways). And potentially not having a circular mask diameter on, which I think you do, but that may be fine for the single particles in this case.
All this to say, it doesn’t look like a terrible first 2D classification to me, but membrane protein focused people will probably be able to help a lot more! There are also some membrane protein-specific tutorials floating around that searching the CS documentation will find.
Yes I agree, for the first pass this looks fine. Pick all of the classes that look like your membrane protein and do an ab initio to get an initial map. Reproject those classes and use denoised micrographs to redo the template picking. Don’t be afraid to really crank the NCC score to a high value and adjust the power scores so you don’t get so many bad picks.
Then the next step is to do several rounds of heterogeneous refinement with your best map and several junk maps as decoys to get a good stack of particles that you can use for refining further.
Membrane protein aggregation on CryoEM grids could be more prominent depending on the type of metal used. From experience, copper would make my protein aggregate in the same way, whereas gold+carbon or UltrAuFoil would have minimal or no aggregation. I’ve heard colleagues reporting an inverse (glod worse, copper better) relationship for their proteins. So screening different type of grids is a good idea.
In addition your protein seems to be very small, micelle dominates, it will be difficult to align it even if you get better looking particles.
I would suggest you to check these resources to help with small membrane protein data processing:
Is your protein in detergent or reconstituted? I had a serious issue before where the compounds dissolved in DMSO caused severe aggregation and smearing with a membrane protein solubilized in detergent. I ended up “solving” this problem by reconstituting the protein in MSPs, and the aggregation caused by the compounds ceased to be an issue.
I mean that with template picker doesn’t find more particles. I think it’s because of the severe aggregation. I agree that these initial 2Ds are not terrible but in my case it isn’t enough. That is why I am trying to solve my aggregation problem.
Those classes look quite well for initial ones after generous blob picking… select the classes which have actual signal and run them again in a 2D classification with 100 classes, 3A max. resolution, uncertanty 1, 60 O-em iterations, 4 full iterations and a batch size of 400. As soon as you have a good stack, I recommend training crYOLO which gave much better results for membrane proteins in my hand (distinguishing membrane proteins from micelles).
