Non-uniform refinement: Adaptive regularization improves single particle cryo-EM reconstruction

Ali Punjani; Haowei Zhang; David J. Fleet

Zebrafish STRA6-CaM [4] is a 180kDa C2 symmetric membrane protein complex comprising the 74kDa STRA6 protein bound to calmodulin (CaM). STRA6 mediates the uptake of retinol in various tissues. We present results of uniform and non-uniform refinement on a dataset of 28,848 particle images (pixel size 1.07Å) of STRA6-CaM in a lipid nanodisc, courtesy of Oliver Clarke and Filippo Mancia [6] . Note that this dataset differs from the older set used in [4] which consisted of 56,615 particle images of STRA6-CaM in amphipol. Figure 3A shows FSC curves from uniform and non-uniform refinement, computed with the same mask. There is a substantial improvement in nominal resolution, from 4.0Å to 3.6Å. This indicates an improvement due to non-uniform refinement in the average global signal-to-noise over the entire structure. It is well known that different regions of a protein will have different resolution characteristics [2] and this is apparent in Figure 3C . The filtered and sharpened 3D maps from uniform and non-uniform refinements are shown from two viewing directions, colored by local resolution estimates. Both maps are filtered using their respective FSC curves and sharpened using a B-factor of −140Å 2 . No local filtering or sharpening is used. The maps are thresholded to contain the same total volume in both cases. Non-uniform refinement resolves significantly improved structural detail in several regions of the molecule, while peripheral and flexible regions remain at low resolutions. Figure 3D shows detailed views of individual helices from within the structure with a docked atomic model, also courtesy of Oliver Clarke [6] . The visible difference in structure quality between uniform and non-uniform refinement is especially relevant during atomic model building, where in many cases, protein backbone and side-chains can only be traced with confidence in the non-uniform refinement map.