Cryo-electron tomography (cryo-ET) enables visualization of macromolecular complexes in their native cellular environment. However, deriving high-resolution structural information requires state-of-art subtomogram averaging (StA) pipelines. A persistent challenge is particle picking, which is typically done by manual picking or using template matching where the template may not reflect the true conformation of the in situ protein or protein complex. Manual picking is time-consuming and subjective, while homologous templates risk introducing bias, especially for membrane proteins for which structures differ when extracted from native membranes. We developed a segmentation-driven workflow that uses Dragonfly, where a neural network is trained on minimal ground truth (∼0.12% of a tomogram; 19 particles) with F-type ATPase protein complex as a model target. A full data set of 82 tomograms generated more than 400,000 particle coordinates out of which ∼390,000 are on target. After particle cleanup and refinement in Relion, generated native template were used for high throughput template matching with GAPSTOP software. From a single tomogram, the native template results in 81.9% of correct picks as opposed to the publicly available template with 11.52% of correct picks from the same tomogram. Next, the native template was used to generate a final particle set in GAPSTOP that was again cleaned and refined in Relion, and after refinement in M, we obtained the final density at subnanometer resolution. The approach was also generalized and successfully implemented on crowded cardiomyocyte tomograms acquired from the generated lamellae, further demonstrating robustness and applicability in situ. In summary, segmentation-derived references reduce operator bias, outperform homologous templates, and remove manual picking as a bottleneck in StA pipelines. This accessible workflow integrates seamlessly with existing cryoET and StA practices, runs on modest hardware, and provides a practical route to democratize StA for diverse macromolecular targets in native environments.
Mounteer et al. (Sun,) studied this question.