Fig. 2: Synaptic connectivity and single-bouton properties.

coCATS of hippocampal mossy fiber/CA3 PN synapses in adult mouse CA3 stratum lucidum with in vivo microinjection. a, Automated synapse detection guided by synaptic immunostaining. High-intensity 3D features in coCATS are segmented and classified as pSCRs if co-localized with pre-synaptic and post-synaptic markers and associated with manual volume segmentations of MFBs. Schematic (top) and single xy planes of volumetric data (bottom) including coCATS (gray, z-STED, STAR RED-NHS), BASSOON (magenta, confocal, AF488) and SHANK2 (turquoise, z-STED, AF594) (N2V, raw data: Supplementary Fig. 4). Imaging data are representative of in vivo microinjection into the LV in n = 10 animals. b, 3D renderings of 22 MFBs segmented from coCATS data at near-isotropic resolution (z-STED). MFB surface areas occupied by pSCRs (white) were automatically segmented and manually proofread. 3D scale bars refer to bouton center. N_MFB = 30 MFBs were reconstructed (10 from each of three imaging volumes recorded across two brain sections (one animal); additional renderings: Supplementary Fig. 7). c–e, MFB volume (V_MFB) (c), surface area (A_MFB) (d), absolute area (A_pSCR/MFB) and relative area occupied by pSCRs on individual MFBs (A_pSCR/MFB/A_MFB) (e) (mean ± s.d., n_MFB = 30). Data points: individual MFBs. f,g, A_pSCR/MFB as function of bouton volume (f) and surface area (g) with linear regression (n_MFB = 30). h, One of the imaging volumes used for MFB characterization (N2V, raw data: Supplementary Fig. 4) with coCATS (gray, z-STED), BASSOON (magenta, confocal) and SHANK2 (turquoise, z-STED), including manually segmented MFBs and automatically detected pSCRs. i, Deep learning pSCR identification with training on paired structural (coCATS) and molecular (BASSOON immunostaining) super-resolved data. Prediction of synaptic marker ___location in unseen datasets is based on structural data alone. pSCRs are segmented similarly as in a but using predicted BASSOON instead of immunostainings. j, Immunostained (orange, z-STED) and predicted BASSOON distribution (blue) in a dataset not included in the training. Corresponding pSCRs (yellow) segmented from coCATS data (gray, z-STED, N2V), guided by immunostained (pSCRs_immuno) or predicted BASSOON (pSCRs_prediction). k, Similarity between pSCRs_immuno and pSCRs_prediction quantified by F1 score (range: 0–1, combining precision and recall; Methods) as a function of IOU threshold. No manual proofreading was applied in j and k. Training was performed on n = 13 imaging volumes recorded across four brain sections from n = 3 animals and testing on n = 1 dataset.