TL;DR
This paper introduces novel methods within the RELION software to detect and correct for previously unaddressed optical aberrations (symmetrical and antisymmetrical) and anisotropic magnification in cryo-EM data. By incorporating these corrections, researchers can achieve significantly higher resolution reconstructions from existing datasets, even those collected with older equipment.
What Problem Does This Paper Address?
Electron cryo-microscopy (cryo-EM) is a powerful technique for determining the 3D structure of biological molecules. However, achieving high resolution is crucial for understanding molecular function, and this is often limited by imperfections in the electron microscope's optics. While some basic aberrations like defocus and astigmatism have been accounted for, higher-order aberrations and anisotropic magnification (where the image is stretched or compressed differently in perpendicular directions) have remained significant challenges. These issues can distort the data, leading to lower resolution reconstructions and obscuring fine structural details. Existing methods for addressing these problems often require specific test samples or additional experimental setups, making post-hoc correction of already collected data difficult.
Methods
The authors developed three distinct, yet complementary, methods implemented in RELION-3.1 to estimate and correct for:
- Antisymmetrical Aberrations: These affect the phase of the electron beam and break Friedel symmetry in the data. The method extends previous work on beam-tilt correction by modeling these aberrations using Zernike polynomials. It involves a two-step optimization process: first, a computationally intensive step to derive per-pixel optimal phase shifts and weights, followed by a faster step to estimate the Zernike coefficients.
- Symmetrical Aberrations: These alter the shape of the contrast transfer function (CTF), causing deviations from the standard elliptical rings. The approach models these aberrations using symmetrical Zernike polynomials, again employing a two-step optimization. This method involves a novel substitution to transform a non-linear problem into a linear one, making it computationally tractable.
- Anisotropic Magnification: This refers to differential magnification in perpendicular directions, leading to image distortion. The method estimates this by comparing predicted particle images to observed ones. It models the distortion using a 2x2 matrix and employs a first-order Taylor expansion to linearize the problem, allowing for efficient estimation of per-pixel displacements and the global magnification matrix.
Crucially, all these methods are designed to be applied a posteriori to existing cryo-EM datasets, requiring only a preliminary 3D reconstruction.
Datasets
The authors evaluated their methods using five publicly available cryo-EM datasets:
- Rabbit muscle aldolase (EMPIAR-10181)
- T. acidophilum 20S proteasome (EMPIAR-10185)
- T. acidophilum 20S proteasome (EMPIAR-10078) - used for phase-plate experiment
- Mouse heavy-chain apoferritin (EMPIAR-10216) - high-resolution dataset
- Synthetic data derived from a human apoferritin dataset (EMPIAR-10200) - used to simulate anisotropic magnification.
These datasets were collected on various microscopes (Thermo Fischer Talos Arctica, Thermo Fischer Titan Krios) and with different detectors (Gatan K2 Summit, Falcon 3), demonstrating the broad applicability of the methods.
Key Insights and Ideas
- Post-hoc Correction is Powerful: The ability to correct for these complex aberrations after data collection, using only a preliminary reconstruction, is a major advancement. This democratizes access to higher resolution, as it doesn't require re-imaging or specialized equipment.
- Aberrations are Common: The study found evidence of significant antisymmetrical and symmetrical aberrations in multiple datasets, suggesting these are not rare artifacts but rather common issues that can limit resolution.
- Zernike Polynomials as a Flexible Basis: The use of Zernike polynomials provides a robust and flexible framework for modeling these optical distortions, allowing for the capture of complex, non-uniform aberrations.
- Two-Step Optimization is Key: The clever separation of the optimization problem into two stages (per-pixel intermediate calculations followed by global parameter fitting) makes the computationally intensive task of correcting these aberrations tractable for large datasets.
- Anisotropic Magnification Affects High Resolution: Even small amounts of anisotropic magnification can become significant at higher resolutions, impacting the accuracy of structural models.
Key Results
- Resolution Improvement: Correction of aberrations led to significant resolution improvements. For example, the aldolase dataset resolution increased from 2.5 Å to 2.1 Å, and the proteasome dataset from 3.1 Å to 2.3 Å. For the high-resolution apoferritin dataset, considering both aberrations and anisotropic magnification yielded a resolution of 1.57 Å, an improvement of three shells over the RELION-3.0 reconstruction.
- Measurable Aberrations: The methods successfully detected and quantified antisymmetrical (trefoil, axial coma) and symmetrical (four-fold astigmatism) aberrations in various datasets.
- Accurate Anisotropic Magnification Estimation: The anisotropic magnification estimation was shown to be accurate, even with relatively small numbers of particles and moderate anisotropy (e.g., 1.35% anisotropy measured).
Limitations and Future Work
- Model Limitations for Extreme Cases: The authors note that their model for symmetrical aberrations may not fully capture extreme effects, such as those observed with phase plates, which can create complex "afterimages."
- Computational Cost: While optimized, the methods still require significant computational resources, especially for large datasets.
- Optics Groups: The implementation in RELION-3.1 uses "optics groups" to handle variations in optical properties across different datasets or experimental conditions. This is a powerful feature for merging data but requires careful management.
- Future Work: The authors suggest that further investigation is needed to understand the prevalence and origins of these aberrations. They also highlight the potential for their methods to be crucial for processing data from lower-energy microscopes, where aberrations are more pronounced. The ability to combine datasets with unknown relative magnifications is also a promising avenue for future applications.
Summary updated