Sharp fsc drop in NU refinement

3D Refinement
1

Hi,

I didn’t understand a sharp drop in fsc curve in in 3D refinement (NU refine)
This resolution drop was shown in three different (but all membrane proteins) samples repeatably on Titan Krios K3, tiff movies, EF 20 eV

titan_data
titan_data874×1069 127 KB

However,
Before data collection on Titan, I always screened the same samples (but different grid) on Glacios, Falcon IV, MRC. It showed common, good feature of fsc and signals in each three samples.

Here, the same samples from Glacios corresponding to the above data set

Glacios_data
Glacios_data836×1069 140 KB

All data preprocessed in cryosparc (patch-motion, patch-ctf estimation) and NU refine

What factors affect signal loss between the same sample grids…?
I’m so frustrated that I can’t get over the invisible wall in fsc…

thank you

2

Is there really that different considering the x-axis scales are different - they look quite comparable…

3

Hi @Jadyn

Have you checked for beam tilt issues? When we see this kind of vertical drop in the FSC (in masked & unmasked) it often indicates that beam tilt is a limiting factor. Try refining global CTF parameters and see if that helps?

Cheers
Oli

1 Like
4

thank you for your comment @schow

yes, their x-axis scales are different due to different apix (1.1A, 0.85A) so, this makes the sharp drop seem a little exaggerated.
But, the absolute value of no mask and corrected curve are sharp signal loss still…

Here are another one of the three kinds of my data I collected (almost same apix, same sample batch, same grid preparations, same processing)

another
another826×1091 192 KB

5

Thank you @olibclarke

That’s a great point. In fact, I’ve tried some things like beam-tilt refinement (also, tried other CTF refine), further classification with cryoSPARC and RELION, local Motion Correction, polishing, and new picking… but, It didn’t show improvement on a vertical drop in the FSC and resolution limit compared to the Glacios data.

It seems like the signal after a certain frequency (~3.5 A) is dead…
This Titan Krios data ctf-fit distribution is like this

ctf_fit

It may simply be the limit of the particle qualities, but similar results were obtained for all three different sample collections (also all Glacios/Falcon IV data showed normal features).

The other posts you wrote on the cryoSPARC Discuss were very helpful! I got a lot of help. Thank you.

6

What resolution of apoF or T20S can you get on this Titan? Are diffraction spots from gold etc. observed out to near the information limit?

7

yes, I’m also curious about it so, I asked about the test samples (apoF…) but, I couldn’t get details except for just sub 2 A apoF at the time.

I will ask the Insitute for more details and beam condition later

thank you!

8

The box and mask look really tight - have you tried a bigger box? Especially on the Krios dataset, which has the smaller pixel size.

9

Thank you @DanielAsarnow

In fact, the box size of the Glacios data set is 300 pix (1.1 apix) and the Titan data set is 400 pix (0.85 apix). The dimension of my protein (GPCR-G protein complex) is about 100 ~120 A.
I think that these box sizes are about 3x larger than particles so, enough to apply CTF signals. You mean the new box size for Krios dataset, for example, ~512 box size?

Also, the masks of two data are automatically generated by NU refinement in cryoSPARC. I’ll simpely test the NU refinement with more soft mask and new box

10

Oh that sounds good, I was just assuming you were using the same circle mask as in the 2D classes you posted.

1 Like