What is the Preferred Orientation Problem in Cryo-EM And How to Deal with It?

1.       What is the preferred orientation problem?

Preferred orientation problem occurs when most of the particles presented the same view in the EM image. It has been a tricky problem for high-resolution structure determination with cryo-EM, especially for samples without symmetry. It is caused by the adherence of the sample to the air-water interface or to the substrate of the grid (carbon film or graphene oxide). Let’s take 20S Proteasome as an example.

Figure 1. The side view and top view (A) of the 20S proteasome (EMD-8726). The simulated image with the preferred side view and top view (B).

When most of the proteasome particles prefer to “lie on the grid”, we will see a lot of side views like showing in Fig.1B (left panel). And when most of them “stand on the grid”, only top views are observed (Fig1B, right panel). 

2.       How does the preferred orientation affect the resolution?

To answer this question, we will need to first understand the “central slice theorem”: The Fourier transform of the 2D projection is equal to the central slice of the Fourier transform of the 3D object. The EM images that we collect with TEM, are 2D projections of the real particle. The cryo-EM image processing is a process to use 2D projections of the real particle to recover the Fourier transform of the real particle, and with invert Fourier transform, we can get the structure of the real particle from its counterpart in Fourier space.

 

Figure 2. The central slice theorem.

For cylindrical samples like 20S proteasome, if a dataset has preferred side views, it is usually not a big deal. However, if a dataset has preferred top views, it would degrade the resolution. And here is why.

For cylindrical particles with preferred side views, it has the freedom to rotate along the y-axis as shown in Fig.3A. Thus, we will have a set of 2D projections of different rotation angles along the y-axis. After Fourier transforms, the Fourier transform of the 2D projections with different rotation angles along y-axis can form a 3D object in Fourier space. In this way, all the structural information is presented in the 3D object in Fourier space and can be easily inverted back to real space.

On the contrary, if a dataset has preferred top views, the z dimensional information is lost and couldn’t be recovered during the image processing. In this scenario, every particle has the freedom to rotate along z-axis rather than y-axis (If a particle rotates along the y-axis, it will be lying on the grid which will produce side view. To simplify the case, we assume that only top views can be observed in a top view preferred dataset. So a particle cannot rotation along the y-axis.). After Fourier transforms, a set of 2D projections with different rotation angles along z-axis are obtained. However, they are on the same plane (XY plane). If we combine images on the same plane, we will still get a single plane which means we fail to generate the 3D Fourier transform of the real particle. In other words, we couldn’t resolve the structure of our real particle.

Figure 3. The reconstruction process of side view preferred dataset (A) and top view preferred dataset (B).

 

3.       How to deal with preferred orientation problem?

Figure 4. The sample orientation on grid non-tilted (left) and tilted (right).

The most common practice to tackle the preferred orientation problem is to tilt the grid during data collection. By tilting the grid which is equivalent to tilt the particle (Fig.4), we aim to record more z dimensional information in the projection images.