To demonstrate the utility of the protocol for refinement in
high-resolution maps, two maps from a 2.5 A ˚ resolution X-ray
structure of E.coli Adenylate Kinase (PDB ID: 1AKE) were
simulated at resolutions of 2.5 and 3.5 A ˚ (Fig. 8a). A comparative
model based on 46% sequence identity to a mutant adenylate kinase
from S. cerevisiae at a different conformation (distinct from the
experimentally determined structure) was used as the input. Following refinement into the maps, the Cα- and all-atom RMSD
between the model and X-ray structures were 0.41 A ˚ and 0.96 A ˚ ,
respectively, for the 2.5 A ˚ resolution map, and 0.64 A ˚ and 1.19 A ˚ ,
respectively, for the 3.5 A ˚ resolution map [35].
Additionally, this protocol was successfully used to model the
structure of D. discoidium initiation factor 6 (eIF6) into a 3.3 A ˚
experimental cryo-EM map of its complex with 60S ribosomal
subunit (EMD-3145). An initial comparative model was used
based on the structure of a M. jannaschii eIF6. At each stage of
refinement, the SMOCf score, which was used to assess the local fit
of the model to the density, was improved (Fig. 8). The final model
was assessed using QMEAN score [24] and seen to satisfy typical
spatial constraints for proteins of similar size.
The hierarchical methodology has also been shown to be applicable to the fitting of atomic models into intermediate resolution
EM maps, for example, in the refinement of nicotinic acetylcholine
receptor structures [54]. Recently, it was applied to model a
Kinesin-8/inhibitor complex bound to microtubules, using an
initial model derived from X-ray crystallography in an MT-free
state (PDB ID: 3LRE) [55]. SSEs of the model were refined into
the cryo-EM map, followed by an all-atom refinement step.
3.6.1 Loop Optimization
A major challenge for automated model refinement is the accurate
modeling and placement of loops into the density. Scores such as
SMOCf may identify flexible regions in the structure as represented
by a low value. Multiple loop models can then be generated for the
identified region, assessed locally, and be further refined. The
assessment can also be done using a Z-score. In the example of
E. coli adenylate kinase described above (see Subheading 3.5), following the all-atom refinement step, an SMOCf profile was generated using Z-scores, which were able to better capture two regions
of poor fit quality (Fig. 8b). These regions corresponded to loops of
low model quality as identified by the QMEAN analysis. For both
regions, 200 loop models were generated with MODELLER and
an all-atom refinement was conducted with Flex-EM on the model
with the best SMOCf score. A significant improvement in the
SMOCf profile was seen at these regions, indicating the comparative model errors were the limiting factor in generating a good fit
for the model.
204
Tristan Cragnolini et al.
high-resolution maps, two maps from a 2.5 A ˚ resolution X-ray
structure of E.coli Adenylate Kinase (PDB ID: 1AKE) were
simulated at resolutions of 2.5 and 3.5 A ˚ (Fig. 8a). A comparative
model based on 46% sequence identity to a mutant adenylate kinase
from S. cerevisiae at a different conformation (distinct from the
experimentally determined structure) was used as the input. Following refinement into the maps, the Cα- and all-atom RMSD
between the model and X-ray structures were 0.41 A ˚ and 0.96 A ˚ ,
respectively, for the 2.5 A ˚ resolution map, and 0.64 A ˚ and 1.19 A ˚ ,
respectively, for the 3.5 A ˚ resolution map [35].
Additionally, this protocol was successfully used to model the
structure of D. discoidium initiation factor 6 (eIF6) into a 3.3 A ˚
experimental cryo-EM map of its complex with 60S ribosomal
subunit (EMD-3145). An initial comparative model was used
based on the structure of a M. jannaschii eIF6. At each stage of
refinement, the SMOCf score, which was used to assess the local fit
of the model to the density, was improved (Fig. 8). The final model
was assessed using QMEAN score [24] and seen to satisfy typical
spatial constraints for proteins of similar size.
The hierarchical methodology has also been shown to be applicable to the fitting of atomic models into intermediate resolution
EM maps, for example, in the refinement of nicotinic acetylcholine
receptor structures [54]. Recently, it was applied to model a
Kinesin-8/inhibitor complex bound to microtubules, using an
initial model derived from X-ray crystallography in an MT-free
state (PDB ID: 3LRE) [55]. SSEs of the model were refined into
the cryo-EM map, followed by an all-atom refinement step.
3.6.1 Loop Optimization
A major challenge for automated model refinement is the accurate
modeling and placement of loops into the density. Scores such as
SMOCf may identify flexible regions in the structure as represented
by a low value. Multiple loop models can then be generated for the
identified region, assessed locally, and be further refined. The
assessment can also be done using a Z-score. In the example of
E. coli adenylate kinase described above (see Subheading 3.5), following the all-atom refinement step, an SMOCf profile was generated using Z-scores, which were able to better capture two regions
of poor fit quality (Fig. 8b). These regions corresponded to loops of
low model quality as identified by the QMEAN analysis. For both
regions, 200 loop models were generated with MODELLER and
an all-atom refinement was conducted with Flex-EM on the model
with the best SMOCf score. A significant improvement in the
SMOCf profile was seen at these regions, indicating the comparative model errors were the limiting factor in generating a good fit
for the model.
204
Tristan Cragnolini et al.