Progress and Development of Direct Detectors for Electron Cryomicroscopy

A.R. Faruqi1; Richard Henderson; Greg McMullan    MRC Laboratory of Molecular Biology, Francis Crick Ave., Cambridge Biomedical Campus, Cambridge CB2 0QH, UK
1 Corresponding author: email address: arf@mrc-lmb.cam.ac.uk

1 Introduction

The main application of direct electron detectors discussed in this chapter is in single-particle electron microscopy (SPEM) or cryomicroscopy or cryo-EM of frozen hydrated biological specimens, a technique used to obtain structures of biological macromolecules to high (i.e., near-atomic) resolution from large numbers of images of individual molecules without the need for crystals. Electron cryomicroscopy (cryo-EM) in its current form emerged from Jacques Dubochet's development of a method for rapid freezing of single particles in thin films of amorphous ice (Adrian, Dubochet, Lepault, & McDowall, 1984; Dubochet et al., 1988). Single-particle image processing procedures were first applied to negatively stained specimens before being refined for use on images of frozen hydrated specimens (Frank, 2009). Before the advent of direct electron detectors, cryo-EM had been used, mainly with photographic film, to obtain high-resolution structures of relatively large macromolecular assemblies such as viruses. The newly developed direct detectors, together with the associated new methods of data acquisition, produce higher-resolution structures and require fewer single particles. They can also be used to analyze the structures of a range of smaller-molecular-weight molecules that were previously beyond the reach of cryo-EM.

SPEM is a technique used to obtain images from individual molecules frozen in a thin layer of vitreous ice, with their structure being close to native. Images of individual molecules have poor contrast due to the weak scattering by the constituent carbon, nitrogen, oxygen and hydrogen atoms, and poor signal-to-noise ratio due to the limited electron dose permitted before radiation damage builds up. Extracting high-resolution information requires the aligning and averaging of tens of thousands (or more) of individual images. Molecules with high symmetry require fewer images and images of molecules with higher molecular weight have higher-contrast images and are easier to analyze (Grigorieff & Harrison, 2011).

In recent years, a number of outstanding near-atomic-resolution structures of macromolecular complexes have been obtained using single-particle cryo-EM, reviewed briefly in several studies (Faruqi, Henderson, & McMullan, 2013; Kühlbrandt, 2014; Henderson, 2015) and covered more fully in other publications (Agard, Cheng, Glaeser, & Subramaniam, 2014; Liao, Cao, Julius, & Cheng, 2014; Bai, McMullan, & Scheres, 2015). Kühlbrandt (2014) has commented that such unprecedented resolution, though expected on theoretical grounds (Henderson, 1995), still came as a great and pleasant surprise. While better software and increased computing power have contributed to this success, the introduction of direct electron detectors is clearly the most important factor. On the same basis, further improvements are expected once detectors with even better performance become available.

A comparison of the performance of three currently available commercial direct detectors was published recently (McMullan, Faruqi, Clare, & Henderson, 2014). Electronic detectors based on phosphor coupled charge-coupled devices (CCDs) have been used for cryo-EM for the past 20 years, particularly in two-dimensional (2D) electron crystallography (Faruqi & Subramaniam, 2000; Sander, Golas, & Stark, 2005; Bammes, Rochat, Jakana, & Chiu, 2011). It is expected that direct detectors with a higher detective quantum efficiency (DQE) will also lead to improved data for electron crystallography and tomography (Lucic, Forster, & Baumeister, 2005; Szwedziak, Wang, Bharat, Tsim, & Löwe, 2014), as well as for single-particle cryo-EM. Recording electron diffraction data from small three-dimensional (3D) microcrystals has also been made possible by direct detectors based on hybrid detector technology—namely, Medipix2 (Campbell, 2011; Nederlof, van Genderen, Li, & Abrahams, 2013). Since this chapter is focused mainly on biological cryo-EM, many other applications outside this field, such as those in materials science and condensed matter physics are not included, though many of these applications have also benefited from improved detectors. How detectors are evaluated in terms of size, efficiency, resolution, and other factors is described in the section entitled “Detector Basics,” later in this chapter; such measurements are very useful for making an objective direct comparison between different detectors (McMullan, Chen, Henderson, & Faruqi, 2009; Ruskin, Yu, & Grigorieff, 2013; McMullan et al., 2014).

Several publications and reviews have covered in some detail the desirable (and in some cases essential) properties required in a detector for cryo-EM (Faruqi et al., 2005a; Faruqi, 2007; McMullan, Chen, Henderson, & Faruqi, 2009; Faruqi & McMullan, 2011), so this chapter will merely summarize the most important points for this review. As the discussion is dealing with specimens that are damaged easily by radiation, the most important requirement is a high DQE, which incorporates detection efficiency, spatial resolution and noise properties of the detector in its definition. The DQE of a detector is given by the ratio of the square of signal to noise at output compared with input:

QE=S/Nout2/S/Nin2,

where S is the signal and N the noise. It is important to note that, because the detector always adds some noise to the signal, the DQE is always less than 1. When used for imaging, the DQE needs to include the effects of spatial resolution and when given as a function of spatial frequency, ω, it is given by DQE(ω) (Dainty & Shaw, 1974; Meyer & Kirkland, 2000; McMullan, Chen, Henderson, & Faruqi, 2009), where

QEω=DQE0*MTF2ω/NPSω,

with MTF(ω) being the modulation transfer function as a function of spatial frequency and NPS(ω) being the normalized noise power spectrum, also as a function of spatial frequency. For pixelated detectors, the important concept of the Nyquist frequency, given by the inverse of twice the pixel size, is used frequently in the evaluation of individual detector properties. The Nyquist limit is also useful in comparing different detector properties at this spatial frequency, which is independent of the pixel size in different detectors (McMullan, Chen, Henderson, & Faruqi, 2009; Faruqi & McMullan, 2011; Ruskin et al., 2013; McMullan et al., 2014).

A review of high-resolution structures obtained with cryo-EM published several years ago (Grigorieff & Harrison, 2011), before the advent of the direct electron detectors, listed nine different near-atomic-resolution structures of icosahedral viruses. The single particles chosen for that review were all quite large, in the range of 17–150 MDa, some with high symmetry, which makes their orientation determination much easier due to the high-contrast images they produce and consequent ease of averaging data from a number of particles. Apart from one structure, all the research cited used photographic film, which was previously accepted to be the best recording medium with the highest DQE (Sander et al., 2005; McMullan, Chen, Henderson, & Faruqi, 2009). A similar review today, describing the most recent high-resolution structures, would include a majority based on the use of direct electron detectors, usually with back-thinned complementary metal oxide semiconductor (CMOS) technology, as discussed in the “Detector Basics” section [see also Bai et al. (2015)]. There are a number of...