Refinement does not improve resolution

Dear all,

I am trying to get the structure of a small membrane protein (ca 65 kDa). Despite an initial large dataset (500k particles) and clear 2D classes, I had to reduce the dataset to 40k particles to obtain a useful ab-initio model. Even tho my micrographs look clean and have a CTF fit resolution of 3A, I am stuck at 6A resolution, and my best map actually comes from 22k particles. I can see everything in that map: helices and loops are equally well visible. I tried to play around with NU and local refinement settings, as well to use heterogenous refinement to identify more particles, but I could not yet squeeze out more of this dataset. Does anyone have a suggestion on what I could try?

My settings for NU refinement are the following:
Number of final extra passes: 2-10
Initial lowpass resolution: 9-7

I also tried to pay around with thightening all dynamic masks, minimize over particle-scale, optimize defocus and ctf parameters.

My best settings for local refinement are the following:
Use pose/shift gaussian prior during alignment: ON
Standard deviation (deg) of prior over rotation: 5
Standard deviation (A) of prior over shifts: 3
Re-center rotations each iteration?: ON
Re-center shifts each iteration?: ON
Number of extra final passes: 1
Initial lowpass resolution (A): 7
Dynamic mask far (A): 9

Many thanks,