Elsevier

Ultramicroscopy

Volume 230, November 2021, 113311
Ultramicroscopy

Relative roles of multiple scattering and Fresnel diffraction in the imaging of small molecules using electrons, Part II: Differential Holographic Tomography

https://doi.org/10.1016/j.ultramic.2021.113311Get rights and content

Highlights

  • The proposed 3D reconstruction method, termed Differential Holographic Tomography, is developed theoretically.

  • The method is also demonstrated computationally on several numerical models of biological molecules.

  • The method is shown to be capable of accurate reconstruction of the locations of atoms in a molecule from TEM data.

  • Input TEM images can be collected at random orientations of the molecule, with one or more defocus images per orientation.

Abstract

It has been argued that in atomic-resolution transmission electron microscopy (TEM) of sparse weakly scattering structures, such as small biological molecules, multiple electron scattering usually has only a small effect, while the in-molecule Fresnel diffraction can be significant due to the intrinsically shallow depth of focus. These facts suggest that the three-dimensional reconstruction of such structures from defocus image series collected at multiple rotational orientations of a molecule can be effectively performed for each atom separately, using the incoherent first Born approximation. The corresponding reconstruction method, termed here Differential Holographic Tomography, is developed theoretically and demonstrated computationally on several numerical models of biological molecules. It is shown that the method is capable of accurate reconstruction of the locations of atoms in a molecule from TEM data collected at a small number of random orientations of the molecule, with one or more defocus images per orientation. Possible applications to cryogenic electron microscopy and other areas are briefly discussed.

Introduction

Cryogenic electron microscopy (cryo-EM) has recently reached the true atomic spatial resolution milestone [1], [2]. As the spatial resolution in high-energy transmission EM (TEM) approaches one ångström, the corresponding depth of focus (also known as the depth of field) tends to become shallower than the longitudinal (along the direction of propagation of the illuminating electron wave) extent of a typical imaged sample, such as a protein molecule. For example, at a spatial resolution of Δ = 1 Å and a wavelength of λ = 0.025 Å (for electrons at 200 keV energy), the depth of focus is equal to z=Δ2/(2λ)=20Å [3]. It has been suggested previously that, as a result of the shallow focus, the in-molecule Fresnel diffraction (free-space propagation) becomes a significant factor that cannot be ignored in the reconstruction of the sample from its defocused images [3], [4], [5], [6]. In other words, the propagation of the incident electron wave through the sample can no longer be satisfactorily approximated by projection integrals of the electrostatic potential along straight lines. Instead, the Fresnel diffraction of the electron beam inside the molecule must be explicitly taken into account in order to achieve an accurate 3D reconstruction. The introduction of specially designed aberrations into the EM illumination system can increase the depth of focus [7], [8], but this is still unlikely to fully eliminate the in-molecule Fresnel diffraction effects at atomic resolution. Moreover, as we argue below (see also [9]), attempts to increase the depth of focus in EM tomography may in some cases be counter-productive, since a shallow focus improves the spatial resolution by enhancing the longitudinal localization of individual atoms in the reconstruction of the molecule.

It is known that, in the transmission electron microscopy of sufficiently thin non-crystalline specimens, "diffraction tomography" (DT) [10], [11], [12] can be used to properly account for the Fresnel diffraction inside the sample and correctly reconstruct the three-dimensional (3D) distribution of the complex refractive index, or, equivalently, the spatial distribution of the electrostatic potential in electron imaging. The DT approach can be based on the first Rytov or first Born approximation, instead of the projection approximation utilized in conventional computed tomography (CT). The use of DT for 3D reconstruction of small molecules was discussed in detail in the first part of this work [9]. Here, we develop and test a different method, termed "Differential Holographic Tomography" (DHT), aimed at reconstruction of the 3D structure of small molecules from TEM defocus series collected at a small number of different orientations of the molecule. This method is developed in detail in the next section, and here we only briefly discuss the main physical considerations that underpin the method.

We showed in [9] that, in agreement with the statements found in [3], [4] and elsewhere, in high-resolution electron microscopy (EM) of "sparsely localized" structures, such as small biological molecules, the total contrast distribution in a defocused image can typically be well approximated by an incoherent sum of contrast functions corresponding to individual atoms. This fact can be used as a basis for reconstructing the 3D distribution of the electrostatic potential in a molecule from transmission images collected at different defocus distances and different orientations of the molecule, effectively performing the reconstruction separately for each atom before adding together the individual reconstructed atomic potentials. This general approach is similar to the "independent atoms model" which has been used for crystallographic phase retrieval, sometimes in conjunction with the atomicity constraint, for a long time (see e.g. [13], [14]). However, we note that the independent atoms approximation, which ignores molecular effects, is often applied in a far-field diffraction regime, where the interference of the waves scattered by different atoms is essential. In contrast, the present method is concerned with the Fresnel diffraction region and ignores the interference effects, thus using an incoherent independent atoms approximation. The validity of such an approximation was studied previously in the context of the first Born approximation [9] and is also considered in the next section of this paper.

In order to be able to perform independent localized reconstruction of the potential in the vicinity of each atom, one can collect, for a given orientation of the molecule, one or more defocused images along the propagation direction of the transmitted electron wave. These images can be used to perform computational phase retrieval, e.g. with the help of the Iterative Wave Function Reconstruction (IWFR) method from [15], the contrast transfer function (CTF) based L2-difference minimization method from [16] or the single-image phase retrieval method from [17]. The complex wave amplitude thus reconstructed at a given rotational orientation of the molecule can then be numerically back-propagated to the vicinity of each atom and the contrast function can be evaluated in that vicinity. As the local distribution of the electrostatic potential in the neighborhood of an atomic nucleus is essentially spherically symmetric, its reconstruction from the localized contrast function becomes relatively trivial. In particular, we will show that such a reconstruction can be obtained by summing up slightly defocused local contrast functions obtained at different orientations (that is, averaging them over the illumination directions) and then taking the inverse 3D Laplacian of the sum. The latter operation effectively compensates for the local Transport of Intensity equation (TIE) type differential phase contrast and recovers the potential from its Laplacian [18]. The differential nature of this method for reconstruction of the electrostatic potential is essential in the case of phase objects, such as atoms illuminated by high-energy electron waves, since they do not produce intensity contrast in the exact in-focus position.

By construction, the electrostatic potential reconstructed by the proposed method is localized to narrow regions around the atomic positions along the direction of propagation of the electron wave through the molecule at each orientation. This localization property of the reconstruction algorithm allows one to significantly relax the sampling conditions with respect to the number of required rotational orientations. In the case of conventional CT [19], the usual Nyquist conditions relate the required number of angular steps, na, over 180 degrees of rotation to the number of pixels in the detector rows, nx, via the well-known sampling relationship na=(π/2)nx. In contrast, in our DHT algorithm, the number of required angular positions becomes almost independent of the number of detector pixels, due to the longitudinal localization of the back-propagated contrast function. Indeed, the relevant number of rotational positions here is related to the size of the vicinity of an individual atom inside which its electrostatic potential is significant, instead of being related to the size of the whole molecule. The quality of reconstruction using the DHT method is still likely to depend on the size of the molecule. However, the primary mechanism for the deterioration of the DHT reconstruction quality with the increase in the size of the molecule is in this case related to the increase of the contribution of multiple scattering effects to the contrast function. These effects include the "shading" of atoms by one another along the lines of tomographic projections. The negative influence of these multiple scattering effects on the reconstruction quality can be alleviated by increasing the number of angular orientations in the scan. Several papers have been published recently on the incorporation of multiple scattering into methods for 3D EM reconstruction [20], [21], [22]. Potentially, some of these techniques could be used in conjunction with DHT in the future to take the effects of multiple scattering in the input data into account explicitly and make the resultant algorithm even more robust in this respect.

Apart from its connections with DT [10], [11], [12] stated above, the proposed DHT method has some similarities with the Big Bang Tomography technique developed by D. Van Dyck and co-authors [23], [24], [25], [26]. In that technique, the "depth" position (i.e. the position along the direction of the optical axis) of different atoms in the imaged sample is determined using the CTF, which is similar in principle to the approach used in DHT. However, the reconstruction technique used in [23], [24], [25], [26] is different from that of DHT. Another method, known as holotomography [27], [28], utilizes multiple defocused images to retrieve the complex wave amplitude in the Fresnel region before the subsequent numerical back-propagation and tomographic reconstruction. Unlike the DHT method developed below, however, holotomography employs conventional CT for reconstruction of the 3D distribution of the real and imaginary parts of the refractive index in the imaged sample from the recovered defocused complex amplitudes. The DHT method is also closely related to the problem of the Ewald sphere curvature correction which has attracted significant attention in recent years in the context of high-resolution TEM [4], [5], [6].

Section snippets

Differential holographic tomography algorithm

Consider an imaging setup with a monochromatic plane wave Iin1/2exp(i2πkz) illuminating a weakly scattering object, where k=1/λ is the wave number, Iin=constant is the intensity of the incident wave, r(x,y,z) is a Cartesian coordinate system in 3D space and z is the optical axis. The complex amplitude U(r) of the wave inside the object satisfies the time-independent Schrödinger equation: 2U(r)+4π2n2(r)k2U(r)=0, where n(r) is the refractive index. In the case of electron microscopy, one has n(r

Numerical simulations

In this section we describe numerical simulations performed to test the potential of practical reconstruction using the DHT method of sparse atomic structures from TEM defocus image series collected at multiple rotational orientations of the structure. The code used for the calculation of defocused images in these simulations was based on the freely available TEMSIM C++ source code developed by E.J. Kirkland [34]. The TEMSIM programs allow one to simulate defocused TEM images using the

Conclusions

In the first part of this work [9], we argued that in many examples of TEM imaging of small biological molecules or other “sparsely localized” weakly scattering structures, multiple scattering tends to have only a moderate effect and therefore can be safely ignored in the reconstruction procedures, without introducing large errors into the results. Importantly, the in-molecule free-space propagation (Fresnel diffraction) cannot be ignored because of the shallow depth of focus under the imaging

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

The authors wish to thank Dr. Andrew Martin for sharing his software code that was used in the course of this research.

References (50)

  • T. Nakane et al.

    Single-particle cryo-EM at atomic resolution

    Nature

    (2020)
  • K.M. Yip et al.

    Atomic-resolution protein structure determination by cryo-EM

    Nature

    (2020)
  • R.M. Glaeser

    How good can single-particle cryo-EM become? What remains before it approaches its physical limits?

    Ann. Rev. Biophys.

    (2019)
  • R. Glaeser

    How good can cryo-EM become?

    Nat. Methods

    (2016)
  • M. Lentzen

    Contrast transfer and resolution limits for sub-angstrom high-resolution transmission electron microscopy

    Microsc. Microanal.

    (2008)
  • R. Erni

    Aberration-Corrected Imaging in Transmission Electron Microscopy: An Introduction

    (2015)
  • G. Gbur et al.

    Diffraction tomography without phase information

    Opt. Lett.

    (2002)
  • V. Elser

    Phase retrieval by iterated projections

    J. Opt. Soc. Am. A

    (2003)
  • H.M. Quiney et al.

    Biomolecular imaging and electronic damage using X-ray free-electron lasers

    Nature Phys.

    (2011)
  • D. Paganin et al.

    Quantitative phase-amplitude microscopy. III. The effects of noise

    J. Micros.

    (2004)
  • D.A. Thompson et al.

    Fast three-dimensional phase retrieval in propagation-based X-ray tomography

    J. Synch. Rad.

    (2019)
  • F. Natterer

    The Mathematics of Computerized Tomography

    (2001)
  • H.G. Brown et al.

    Structure retrieval at atomic resolution in the presence of multiple scattering of the electron probe

    Phys. Rev. Lett.

    (2018)
  • D. Ren et al.

    A multiple scattering algorithm for three dimensional phase contrast atomic electron tomography

    Ultramicroscopy

    (2019)
  • J.J. Donatelli et al.

    Inversion of many-beam bragg intensities for phasing by iterated projections: removal of multiple scattering artifacts from diffraction data

    Phys. Rev. Lett.

    (2020)
  • Cited by (4)

    • Perspective: Emerging strategies for determining atomic-resolution structures of macromolecular complexes within cells

      2022, Journal of Structural Biology
      Citation Excerpt :

      We therefore consider it worthwhile to investigate whether focal series reconstruction can approach the Shannon sampling limit, in which images are recorded from 2D/d physical sections of an object. Nevertheless, a hybrid imaging scheme involving both tilts and focal series may be more beneficial than using just a focal series of images, as was illustrated recently for aberration-corrected STEM (Yalisove et al., 2021) and explored theoretically in TEM (Gureyev et al., 2021; Ren et al., 2020). Such a hybrid approach could still benefit from reducing the number of tilts, and, by adding a known amount of parallax, it could better exploit the range of z positions and orientations of protein copies in a cryo-tomography sample.

    • Method for virtual optical sectioning and tomography utilizing shallow depth of field

      2022, Journal of the Optical Society of America A: Optics and Image Science, and Vision
    View full text