Simultaneous Determination of Protein Structure and Dynamics Using Cryo-Electron Microscopy

Abstract

Cryo-electron microscopy is rapidly emerging as a powerful technique to determine the structures of complex macromolecular systems elusive to other techniques. Because many of these systems are highly dynamical, characterizing their movements is also a crucial step to unravel their biological functions. To achieve this goal, we report an integrative modeling approach to simultaneously determine structure and dynamics of macromolecular systems from cryo-electron microscopy density maps. By quantifying the level of noise in the data and dealing with their ensemble-averaged nature, this approach enables the integration of multiple sources of information to model ensembles of structures and infer their populations. We illustrate the method by characterizing structure and dynamics of the integral membrane receptor STRA6, thus providing insights into the mechanisms by which it interacts with retinol binding protein and translocates retinol across the membrane.

Figure 1

Validation of the metainference approach on a conformationally heterogeneous ensemble of the chaperonin GroEL. The crystal structure of Apo-GroEL (63) (A, blue) and a comparative model built from the structure of GroEL in complex with ADP (64) (A, red) were used to create synthetic cryo-EM maps at near-atomistic resolution (B). An average map was then computed by mixing contributions from the two models in ratio 1:1 (C). The metainference approach was capable of disentangling the contribution of the two states (D), determining their relative populations in the mixture, and inferring the local level of noise in the map (E). To see this figure in color, go online.

Figure 2

Structure, dynamics, and noise characterization of the STRA6 membrane complex. Compared to the single-structure model (PDB: 5SY1, A), the metainference ensemble displays a higher degree of flexibility (C). We calculated the predicted cryo-EM maps from the single-structure model (B) and metainference ensemble (D) and evaluated the global and local CC with the experimental map. The metainference map provides a better CC with the experimental map (global CC = 0.91) compared to the single-structure map (global CC = 0.86), especially in the more dynamical regions of STRA6. Cross-correlation maps are visualized at a threshold of 3.5. The pie charts report the distributions of local CC in the regions of the single-structure and metainference maps with density between 3.4 and 3.6. The level of relative error in the experimental map inferred by metainference is rather uniform, with the exception of the regions occupied by cholesterol and amphipols (E). To see this figure in color, go online.

Figure 3

Structural insights into the mechanism of RBP binding. To understand the mechanism of RBP binding (A), we projected all conformations of the metainference ensemble along the two collective variables d₁ and d₂, which were defined as the distances between residues N441 in the TM8-TM9 loop and L323 in LP, in each of the two identical monomers. The resulting free energy landscape indicates an equilibrium among different conformations (B). The close state observed in the single-structure model (LP1, C), in which the two LPs are close together, has a relatively low population. A more stable state is an open conformation in which the two LPs fold back to interact with the TM8-TM9 loop (LP2, D). States in which only one of the two LPs folds back are also visible (LP3 and LP4, B). To see this figure in color, go online.

Figure 4

Structural insights into the mechanism of retinol release. To investigate the role of JM in retinol binding and release (A and B), two collective variables were defined as the distance between the geometric centers of residues P248-D252 in JM and V535′-F538′ in JML (d₁) and the distance between the geometric centers of residues P248-D252 in JM and L366-R376 in TM7 (d₂). The associated free energy landscape indicated an equilibrium among different conformations (C). JM, which in the PDB model resides far apart from JML and TM7 (JM1, D), can transiently interact with both JML (JM2, E) and TM7 (JM3, F), suggesting a possible role of JM in facilitating retinol release by weakening the JML-IM interaction and the stability of the binding site situated between the IM helices (B). To see this figure in color, go online.