Streamlined structure determination by cryo-electron tomography and subtomogram averaging using TomoBEAR

Determining the structures of macromolecules in their native cellular environment provides crucial insights into their function. Cryo-electron tomography (cryo-ET), coupled with subtomogram averaging (StA), offers a powerful approach to achieve this, yielding angstrom-scale resolutions for certain macromolecules. This technique has already revealed structural details of diverse biological entities, including viral and bacterial proteins, eukaryotic protein coats, and even ribosomes within intact cells. Recent advancements in both hardware and software have further improved the resolution attainable through cryo-ET/StA. However, several significant obstacles limit the widespread adoption of this method. These include the multifaceted workflow involving multiple specialized software packages requiring intricate interfacing, the frequent need for manual intervention in steps like tilt-series alignment and particle picking, and the challenges associated with managing and processing the vast amounts of data generated. The limited copy numbers of many macromolecules within cells necessitate the processing of numerous tomograms to reach sufficient resolution, adding another layer of complexity. TomoBEAR aims to address these limitations by providing a streamlined, automated workflow for high-throughput cryo-ET data processing.

Existing methods for cryo-ET and subtomogram averaging (StA) often rely on a complex combination of software packages, demanding significant expertise and manual intervention. This makes the process time-consuming and prone to user-introduced errors. While several software packages like IMOD, Dynamo, and RELION offer powerful functionalities, integrating and managing them effectively for large-scale data processing poses a substantial challenge. The literature highlights a need for automated workflows capable of handling the massive datasets generated by modern cryo-ET, especially considering the increasing throughput of parallel data acquisition techniques. Previous efforts have focused on improving specific aspects of the workflow, such as automated tilt-series alignment or particle picking, but a comprehensive, integrated solution for high-throughput processing remained lacking. This gap in the current literature underscores the critical need for user-friendly, automated workflows, such as the one proposed by TomoBEAR, that can address the inherent complexities of cryo-ET/StA.

TomoBEAR is designed as a modular pipeline runner, executing modules sequentially or in parallel for tilt-stacks, tomograms, or particle sets. The workflow starts with the raw microscope movie frames. It uses MotionCor2 for motion correction, followed by tilt-series assembly. Tilt-series alignment is performed using either IMOD's BatchRunTomo or Dynamo, or AreTomo for fiducial-less alignment (suitable for FIB-milled lamellae). Defocus determination is automated using GCTF or CTFFIND4, with 2D CTF correction done using IMOD's Ctfphaseflip. Tomographic reconstruction leverages IMOD, with optional post-processing steps such as denoising and CTF deconvolution using IsoNet. Particle picking can be automated through template matching (accelerated using GPU computation within Dynamo) or deep learning-based methods like crYOLO. Template matching utilizes user-provided or automatically generated templates. Alternatively, the Dynamo Catalogue system can be used for semi-automated particle picking. Subtomogram averaging uses Dynamo, with automated generation and execution of multi-reference alignment and classification projects. Parameters are specified in a JSON configuration file, enabling flexible customization. Intermediate results are stored in a transparent manner allowing for easy monitoring and re-running of modules. The workflow also includes modules for data management, cleanup, and the ability to generate tomographic reconstructions 'live' during data acquisition (live mode), which accelerates the feedback process and quality assessment. The workflow allows export to RELION and SUSAN for downstream refinement steps where more manual fine-tuning may be beneficial.

TomoBEAR's performance was benchmarked on four datasets: purified 80S ribosomes (EMPIAR-10064), purified human apoferritin (EMPIAR-11543), ion channel RyR1 (EMPIAR-10452), and ribosomes from FIB-milled HeLa cells (EMPIAR-11306). For the 80S ribosomes, TomoBEAR achieved a resolution of 11.0 Å, comparable to previously published results. For purified human apoferritin, a resolution of 2.8 Å was obtained, nearing the Nyquist limit. The RyR1 dataset yielded a resolution of 8.9 Å, slightly better than previously reported. Finally, for the FIB-milled lamellae dataset, a resolution of 6.2 Å was achieved using template matching and RELION refinement. The processing time was significantly reduced compared to manual processing, demonstrating TomoBEAR's efficiency. The 'live' data processing mode significantly sped up the initial tomogram generation process by a factor of three, providing immediate feedback during data collection. Overall, TomoBEAR demonstrates the capability of achieving high-resolution structures with minimal user intervention, enabling high-throughput cryo-ET and StA.

TomoBEAR successfully addresses the need for a streamlined and automated workflow for cryo-ET and StA. The results demonstrate that high-resolution structures can be obtained with significantly reduced manual intervention and processing time. This is crucial for tackling the challenges associated with processing large datasets necessary for achieving meaningful resolution with low-copy-number macromolecules. The modular design and flexible configuration options of TomoBEAR make it adaptable to various experimental setups and targets. While near-complete automation is achieved for many steps, the system also incorporates functionalities for manual inspection and adjustment where necessary, providing a balance between automation and user control. The comparison with previously published results validates the accuracy and efficiency of the TomoBEAR workflow, showing comparable or improved resolutions with significantly reduced processing time. The integration of multiple existing software packages into a user-friendly interface lowers the barrier to entry for researchers seeking to utilize this powerful technique.

TomoBEAR provides a significant advancement in cryo-ET data processing for subtomogram averaging. Its modular design, automated functionalities, and flexible configuration options effectively address the limitations of existing workflows, enabling high-throughput processing with minimal user intervention. The demonstrated ability to achieve high-resolution structures across diverse datasets, coupled with significant time savings, establishes TomoBEAR as a valuable tool for accelerating in situ structural biology research. Future directions may include tighter integration with existing pipelines, development of a user-friendly web interface, and further incorporation of advanced AI-powered tools for tasks such as particle picking and classification.

While TomoBEAR significantly automates the cryo-ET/StA workflow, some steps, particularly in the final stages of refinement, might still require user expertise and manual adjustments. The performance of automated particle picking methods, such as template matching and crYOLO, can be influenced by sample characteristics (e.g., density, shape, orientation) and the quality of the tomograms. In cases with particularly challenging samples or low signal-to-noise ratios, more manual intervention might be necessary. The reliance on external software packages means that TomoBEAR's functionality is dependent on the availability and proper installation of these packages.