FIB-SEM datasets of the Anaeramoeba flamelloides BUSSELTON2 symbiosome

FIB-SEM Cells were grown on Sapphire discs and high-pressure frozen in 100 µm deep golden carriers in an HPM100 (Leica microsystems, Wetzlar, Germany) with 20% BSA in ASW as cryo-protectant. The cells Freeze substitution was performed as in Guo et al¹⁰, but with shorter washing steps for the sake of time. After freeze substitution, the samples were infiltrated with TAAB embedding resin hard grade (TAAB Laboratories, Aldermaston, England). The sample was mounted on an SEM-stub with epoxy and silver glue. The sample was further coated with 5nm Pt to reduce charging. Volumes were acquired using a Scios dualbeam (FEI, Eindhoven, The Netherlands) with the electron beam operating at 2 kV/0.2 nA, detected with the T1 In-lens detector. To automate the acquisition, we use the Auto Slice and View 4 software provided with the microscope. A 700 nm protective layer of platinum was deposited on the selected area before milling. The volume was further registered and processed by the ImageJ plugins Linear alignment by SIFT and Multistackreg. After registration, the volumes were converted to mrc-files, and the header was modified to recover the pixel size that got lost during conversion. The resolution of the two volumes and the number of slices were: 9235_1 (8.43 nm x 8.43 nm x 8.0 nm), 1,725 slices, and 9235_2 (6.744 nm x 6.744 nm x 7.0 nm), 1,300 slices. Segmentation The A. flamelloides BUSS2 cells were segmented by a combination of manual segmentation and Deep Learning Segmentation in Microscopy Image Browser v2.91 beta 45¹¹,¹². Training sets were assembled using the Segment Anything Model 2 (SAM 2). The training dataset for endoplasmatic reticulum and endocytic compartments was assembled by using the whole model and gradually using the different individual models to cut out the model. The final cut out introduced to the mask and used for mask-restricted BW thresholding to yield the respective training slices. Briefly, Desulfobacter sp. symbionts, whole cells, symbiosome, endoplasmatic reticulum and endocytic compartments and hydrogenosomes were manually annotated throughout the volume every 50-slices where the FIB-SEM volume was abundant in those structures. Image segments and the central model (patch size 256x256) were extracted 5 slices deep as and used to train using the 2.5D semantic segmentation approach using 5 slices deep in Z2C+DLv3 architecture with the ResNet50 model. The predicted structures were manually refined extensively. Nucleus – segmented using the SAM2 model interactive 3D model. Nuclear pores – The nucleus model was dilated by 15 pixels and BW thresholding was used to select the pixels that corresponded to nuclear pores and manually refined. Plasma membrane and symbiosome membrane– the whole cell and symbiosome model respectively were used to create an eroded mask to create a 5 px cutout of the respective membranes. Sharply shifting membrane sections were manually refined. Microtubule Organizing Centre and microtubules – the MTOC was manually segmented using SAM2. Microtubules were traced manually using a 3 px brush. Other structures (digestive vacuoles/inclusions/secondary symbionts) – manually segmented using the SAM2 model. Symbiosome pores – symbiosome openings were identified manually in all three orientations any labeled using 3D balls. Three model files (exported from MIB¹¹,¹²) were compiled for each volume, one for the whole cell, one for the full symbiosome and one with the remaining segmentations (9235_1, 10 segments: nucleus, nuclear pores, endoplasmatic reticulum and endocytic compartments, hydrogenosomes, symbiosome membrane, other structures (digestive vacuoles/inclusions/secondary symbionts), Desulfobacter sp. symbionts, plasma membrane, microtubule organizing centre and microtubules, symbiosome pores and 9235_2, 7 segments: nucleus, nuclear pores, hydrogenosomes, symbiosome membrane, Desulfobacter sp. symbionts, plasma membrane, symbiosome pores.) The segmentations were rendered in Dragonfly v.2022.2.0 Build 1399.