from X-ray crystallography, PDB 3WHE) shows a much poorer
global resolution of ~21.0 Å (see Fig. 1b, green). This discrepancy
is problematic, and, assuming that the model is properly docked
and fitted into the map, there could be several possible explanations: either the model is severely underfit into the density (too
much weight has been placed onto idealizing model parameters,
rather than the experimental density) or the map contains errors. In
this particular case, due to the large amount of anisotropy, coupled
to reinforcement of misassigned orientations during refinement,
the main reason for the discrepancy is the worsening of the map.
The fact that the half-map resolution belies the actual resolution of
the reconstructed density also highlights some of the limitations of
a simple global self-consistency metric. On the other hand, the
reconstruction from 40
-tilted images shows a slightly decreased
half-map (see Fig. 1b, red) but much improved map-model (see
Fig. 1b, orange) global resolution when compared with the reconstruction from the 0
images. Now, there is less discrepancy
between the half-map and map-model FSC curves, and the resulting values are consistent with the structural features (see Fig. 9).
The origin underlying these issues is partially described in [6] and is
also subject to ongoing investigation.
3.3.2 Interpreting
the Local Resolution
of the Map
We selected three experimental cases with which to demonstrate the
result of local resolution analysis using the FSC-based method. In
the case of the icosahedrally symmetric reconstruction of AAV2
capsid (EMDB: EMD-9012) [24], the resolution varies only
slightly about the global resolution (see Fig. 2a). Most areas of the
capsid proteins are resolved approximately evenly, although some
loops on the periphery show slightly lower resolutions. This is not
surprising because many of these loops are known to be dynamic
and functionally relevant, as they interact with cell surface receptors
and are also most prone to variation between viral subtypes
[25]. The second experimental data analyzed is the ternary complex
of Cas13d with CRISPR RNA and target RNA (EMDB:
EMD-9014) [26]. In this structure, Cas13d associates with RNAs
to form a stable ribonucleoprotein assembly with a global resolution value of 3.3 Å and a local resolution that ranges from 2.9 to
3.7 Å (see Fig. 2b). Here, the variation in local resolution is more
significant. The regions that form extensive electrostatic interactions with RNA have the highest resolution of 3.1 Å, while most of
the remaining areas range from 3.3 to 3.5 Å. The worst resolution
of 3.7 Å is only found on a solvent exposed RNA loop, which is
expected to be dynamic. The third experimental case is a helical
reconstruction of the SgrAI oligomer with associated doublestranded DNA (dsDNA) (EMDB: EMD-20015) [27]. The global
resolution within an asymmetric unit is 3.5 Å. The best resolved
regions are again located within a tightly packed core, as is typical
174
Sriram Aiyer et al.
global resolution of ~21.0 Å (see Fig. 1b, green). This discrepancy
is problematic, and, assuming that the model is properly docked
and fitted into the map, there could be several possible explanations: either the model is severely underfit into the density (too
much weight has been placed onto idealizing model parameters,
rather than the experimental density) or the map contains errors. In
this particular case, due to the large amount of anisotropy, coupled
to reinforcement of misassigned orientations during refinement,
the main reason for the discrepancy is the worsening of the map.
The fact that the half-map resolution belies the actual resolution of
the reconstructed density also highlights some of the limitations of
a simple global self-consistency metric. On the other hand, the
reconstruction from 40
-tilted images shows a slightly decreased
half-map (see Fig. 1b, red) but much improved map-model (see
Fig. 1b, orange) global resolution when compared with the reconstruction from the 0
images. Now, there is less discrepancy
between the half-map and map-model FSC curves, and the resulting values are consistent with the structural features (see Fig. 9).
The origin underlying these issues is partially described in [6] and is
also subject to ongoing investigation.
3.3.2 Interpreting
the Local Resolution
of the Map
We selected three experimental cases with which to demonstrate the
result of local resolution analysis using the FSC-based method. In
the case of the icosahedrally symmetric reconstruction of AAV2
capsid (EMDB: EMD-9012) [24], the resolution varies only
slightly about the global resolution (see Fig. 2a). Most areas of the
capsid proteins are resolved approximately evenly, although some
loops on the periphery show slightly lower resolutions. This is not
surprising because many of these loops are known to be dynamic
and functionally relevant, as they interact with cell surface receptors
and are also most prone to variation between viral subtypes
[25]. The second experimental data analyzed is the ternary complex
of Cas13d with CRISPR RNA and target RNA (EMDB:
EMD-9014) [26]. In this structure, Cas13d associates with RNAs
to form a stable ribonucleoprotein assembly with a global resolution value of 3.3 Å and a local resolution that ranges from 2.9 to
3.7 Å (see Fig. 2b). Here, the variation in local resolution is more
significant. The regions that form extensive electrostatic interactions with RNA have the highest resolution of 3.1 Å, while most of
the remaining areas range from 3.3 to 3.5 Å. The worst resolution
of 3.7 Å is only found on a solvent exposed RNA loop, which is
expected to be dynamic. The third experimental case is a helical
reconstruction of the SgrAI oligomer with associated doublestranded DNA (dsDNA) (EMDB: EMD-20015) [27]. The global
resolution within an asymmetric unit is 3.5 Å. The best resolved
regions are again located within a tightly packed core, as is typical
174
Sriram Aiyer et al.