Quantitative 3D structural analysis of small colloidal assemblies under native conditions by liquid-cell fast electron tomography

Abstract

Electron tomography has become a commonly used tool to investigate the three-dimensional (3D) structure of nanomaterials, including colloidal nanoparticle assemblies. However, electron microscopy is typically done under high-vacuum conditions, requiring sample preparation for assemblies obtained by wet colloid chemistry methods. This involves solvent evaporation and deposition on a solid support, which consistently alters the nanoparticle organization. Here, we suggest using electron tomography to study nanoparticle assemblies in their original colloidal liquid environment. To address the challenges related to electron tomography in liquid, we devise a method that combines fast data acquisition in a commercial liquid-cell with a dedicated alignment and reconstruction workflow. We present the advantages of this methodology in accurately characterizing two different systems. 3D reconstructions of assemblies comprising polystyrene-capped Au nanoparticles encapsulated in polymeric shells reveal less compact and more distorted configurations for experiments performed in a liquid medium compared to their dried counterparts. A similar expansion can be observed in quantitative analysis of the surface-to-surface distances of self-assembled Au nanorods in water rather than in a vacuum, in agreement with bulk measurements. This study, therefore, emphasizes the importance of developing high-resolution characterization tools that preserve the native environment of colloidal nanostructures.

Fig. 1. Challenges in 3D characterization of colloidal clusters by electron tomography.

a HAADF-STEM image showing an overview of Au@PS colloidal clusters in vacuum, where the polymer shell can be observed as a grey shadow around the bright NPs. b, c HAADF-STEM images acquired at high tilt angle of  ± 70° respectively where the flattening effect of a colloidal cluster is highlighted by light blue dashed lines. d, e 2D HAADF-STEM images of a colloidal cluster before and after dry RT electron tomography tilt series acquisition, where the yellow double-head dashed arrows indicate the volume change. See also Supplementary Movie 1. f 3D reconstructions of colloidal clusters containing 4, 5, and 6 Au NPs via fast electron tomography in vacuum. The stackings of Au NPs within the polymeric shells resemble a tetrahedron, a trigonal bipyramid, and an octahedron, respectively. See also Supplementary Movie 2. Note that a Gaussian blur smooth filter was applied to panel b and c to enhance the signal-to-noise ratio of the raw image.

Fig. 2. Liquid-phase fast electron tomography.

a Schematic illustration of a K-Kit LC used for experimental investigations, highlighting the LC dimensions. b-c Optical micrographs of a K-Kit LC loaded on a single-tilt tomography holder, with, b) 0° and c) 45° tilting view, respectively. d The LP fast electron tomography tilt series acquisition method continuously tilts the sample while recording projection images of the sample. e Comparison of time and electron dose required for acquiring a complete tilt series using fast electron tomography in liquid and vacuum and dry RT electron tomography in vacuum. f-i Fast electron tomography tilt series pre-processing workflow. f Representation of the raw tilt series stack with a sample image from the raw stack displayed. g Illustration of the self-supervised denoising using CAE. E represents the encoder CNN (convolutional neural network) architecture and D represents the decoder CNN. A sample image from the denoised stack is displayed, demonstrating the effectiveness of the autoencoder denoising compared to the original one. h Schematic overview of the iterative process undertaken: refining the tilt series using RPCA, followed by rigid registration using the ICP method (T represents the rigid transformation operator), and then tilt-axis alignment via FBP (θ represents the amount of angular shift required to correct the tilt alignment.). i The final processed stack, which is refined, aligned, and denoised, with a representative image displayed for clarity.

Fig. 3. Quantitative structural comparison between 3D reconstructions of colloidal clusters with different numbers of particles, implemented in liquid and vacuum conditions.

a–c Polyhedra (in blue) computed from the centroid positions of the Au NPs (in dashed yellow pseudo-spheres) obtained through CS-DART reconstructions of three clusters containing a N = 4, b N = 5, and c N = 6 Au NPs in a liquid environment. d–e Quantitative normalized structural comparison between polyhedra formed by the stacking of Au NP obtained from fast electron tomography in vacuum (depicted in orange) and liquid (depicted in blue), including mean interparticle distance (Mean ID), surface area, volume, and regularity index for d N = 4, e N = 5, and f N = 6, respectively. The values are provided in Table 1. Importantly, the Au NPs demonstrated a notable tendency to adopt regular but more condensed configurations when observed in a vacuum environment.

Fig. 4. Quantitative structural comparison of Au NR bilayer assemblies in vacuum and liquid-phase.

2D HAADF-STEM images of self-assembled Au NR superlattices measured in a vacuum and b liquid. c,d 3D reconstructions and e,f orthogonal views of superlattices assembled from two layers of Au NRs in c,e vacuum (orange) and d,f liquid (blue). Insets of panels e and f: zoomed-in views of two adjacent Au NRs, showing the disparity between surface-to-surface distances in vacuum and in liquid environment. Surface-to-surface distances were calculated by subtracting the actual radii of each rod from the distance between the centers of mass (Supplementary Fig. 19 and Supplementary Table 2). Note that the transparency of the 3D renderings was increased for visual clarity.