approach depends on classifying the 2D images based on the eigen images/
eigenvectors [74, 108–111] without any starting model. First, classify using MSA
to obtain orientation classes, and then, the major variation among the picked particles in each orientation classes can be identified in the low-order eigen images by
MSA, and using these, information particles can be classified into homogenous
classes, leading to preliminary 3D reconstruction from a class containing majority
of homogenous particles as shown in Fig. 3. The preliminary 3D reconstructions
can be projected as references for competitive alignment. Further, the quality of 3D
reconstruction can be iteratively improved until the eigen images show no major
variations within the class and the particles stabilize from jumping to another class
during competitive projection matching. In this manner, three class reconstructions
were obtained as shown in Fig. 4. The second method depends on detection in 2D
variations using starting model [41]. The third method also needs initial starting
model and uses a statistical approach to obtain 3D classification. In this case, large
number of 3D maps are calculated from randomly selected subset of particles (with
previously assigned orientation based on initial 3D map). Determination of the 3D
variance can be used to assess the heterogeneity, and estimation of covariance
enables one to carry out the 3D classification according to variable regions.
Alternatively, the molecular states can be separated using maximum likelihood
classification [104, 112] or by the latest multi-body refinement method [113].
5 Single-Particle Cryo-EM Applications in SBDD
There are five 3D reconstructions in the EMDB with bound inhibitors or ligands at
resolution better than 2.5 Å as shown in Table 1. We have focused at this resolution
since this is at the center of medium (3.0 Å) and high resolution (1.8 Å), which is
desired resolution for SBDD studies. Although we have highlighted reconstructions
better than 2.5 Å, ligands have been visualized in the 3D reconstructions better than
4.0 Å. While there are only 10s of EM reconstructions with bound inhibitors at
2.5 Å or better, there are several 100s of structures in the EMDB at resolution
between 2.5 and 4 Å with bound inhibitors or ligands. Here are couple of examples
of 3D reconstructions with resolution better than 2.5 Å: In the Sect. 2.3 we have
already come across the example of 3D reconstruction (by the Subramaniam group
[75]) of the inhibitor PETG bound to beta-galactosidase at 1.9 Å resolution. His
group used a similar approach to solve the cryo-EM structure of human
p97ATPase, an important target for cancer, in complex with its allosteric inhibitor
UPCDC30245 [10] as shown in Fig. 7c. Although they could not see a part of the
inhibitor in the EM electron potential map, they could see at 2.3 Å resolution the
other part where the inhibitor snugly fits into the protein pocket and proposed how
the allosteric inhibitor UPCDC30245 inhibits the conformational changes necessary
for the function of p97. They further could see three coexisting functional states of
p97 in the presence and absence of ATPcS. Here are couple of examples of 3D
reconstructions of proteins bound to inhibitors, with 3D reconstruction resolution
392
R. Natesh
eigenvectors [74, 108–111] without any starting model. First, classify using MSA
to obtain orientation classes, and then, the major variation among the picked particles in each orientation classes can be identified in the low-order eigen images by
MSA, and using these, information particles can be classified into homogenous
classes, leading to preliminary 3D reconstruction from a class containing majority
of homogenous particles as shown in Fig. 3. The preliminary 3D reconstructions
can be projected as references for competitive alignment. Further, the quality of 3D
reconstruction can be iteratively improved until the eigen images show no major
variations within the class and the particles stabilize from jumping to another class
during competitive projection matching. In this manner, three class reconstructions
were obtained as shown in Fig. 4. The second method depends on detection in 2D
variations using starting model [41]. The third method also needs initial starting
model and uses a statistical approach to obtain 3D classification. In this case, large
number of 3D maps are calculated from randomly selected subset of particles (with
previously assigned orientation based on initial 3D map). Determination of the 3D
variance can be used to assess the heterogeneity, and estimation of covariance
enables one to carry out the 3D classification according to variable regions.
Alternatively, the molecular states can be separated using maximum likelihood
classification [104, 112] or by the latest multi-body refinement method [113].
5 Single-Particle Cryo-EM Applications in SBDD
There are five 3D reconstructions in the EMDB with bound inhibitors or ligands at
resolution better than 2.5 Å as shown in Table 1. We have focused at this resolution
since this is at the center of medium (3.0 Å) and high resolution (1.8 Å), which is
desired resolution for SBDD studies. Although we have highlighted reconstructions
better than 2.5 Å, ligands have been visualized in the 3D reconstructions better than
4.0 Å. While there are only 10s of EM reconstructions with bound inhibitors at
2.5 Å or better, there are several 100s of structures in the EMDB at resolution
between 2.5 and 4 Å with bound inhibitors or ligands. Here are couple of examples
of 3D reconstructions with resolution better than 2.5 Å: In the Sect. 2.3 we have
already come across the example of 3D reconstruction (by the Subramaniam group
[75]) of the inhibitor PETG bound to beta-galactosidase at 1.9 Å resolution. His
group used a similar approach to solve the cryo-EM structure of human
p97ATPase, an important target for cancer, in complex with its allosteric inhibitor
UPCDC30245 [10] as shown in Fig. 7c. Although they could not see a part of the
inhibitor in the EM electron potential map, they could see at 2.3 Å resolution the
other part where the inhibitor snugly fits into the protein pocket and proposed how
the allosteric inhibitor UPCDC30245 inhibits the conformational changes necessary
for the function of p97. They further could see three coexisting functional states of
p97 in the presence and absence of ATPcS. Here are couple of examples of 3D
reconstructions of proteins bound to inhibitors, with 3D reconstruction resolution
392
R. Natesh