PMC

a, CryoFM of MX04-labelled amyloid pathology targeted for cryoFIB-SEM lift-out showing in top left, top right, and lower left, reflection, MXO4 fluorescence and merged overview, respectively. Scale bar, 200 μm. Yellow rectangle in top right, closeup in bottom right. Scale bar, 20 μm. b-c, Alignment of cryoFM, FIB and SEM images to target lift-out in ZEN Connect. Images on left were aligned with middle image using fiducials indicated by open arrowheads. b, Alignment between cryoSEM normal overview and cryoSEM normal high-magnification image of MX04-labelled lift-out target. Scale bar, 200 μm. c, Alignment between cryoSEM normal high-magnification and cryoFIB normal image. Scale bar, 200 μm. Left and right, scale bar, 200 μm. Middle, scale bar, 100 μm. d, Alignment between cryoFIB normal image and cryoFIB high-magnification after cryoFIB milling trench. Left and right, scale bar, 100 μm. Middle, scale bar, 20 μm. e, Alignment between cryoFIB images before and after surface cleaning, sputter coating and cold deposition of platinum precursor. Scale bar, 20 μm. f, Alignment of cryoFIB images before and after trenches were milled in front, behind and to the right side to prepare tissue chunk for targeted lift-out. Scale bar, 20 μm. g, Images showing the result of alignments (from a to h) to target MX04-labelled tau for cryoFIB-SEM lift-out. Left, cryoFIB image of tissue chunk prepared for lift-out. Middle, confocal cryoFM of MX04-labelled amyloid plaque and tau tangles. Brown and yellow rectangles, left and right regions of serial chunk lift-out, respectively. Cyan line, location of lamella. Scale bar, 20 μm. h-l, CryoFIB-SEM lift-out of MX04-labelled post-mortem AD brain. Lines and rectangles tracing the tissue chunk during cryoFIB cuts and thinning were used for cryoCLEM. Red semi-transparent line, length of tissue chunk before lift-out. Green semi-transparent line, length of tissue chunk during lift-out. Brown rectangle, left tissue chunk. Yellow rectangle, right tissue chunk (lost during sample transfer from cryoFIB-SEM to cryoEM). Cyan line, region targeted for cryoFIB milling lamellae within tissue chunk. h, CryoFIB image of tissue chunk after final left side cut to detach tissue chunk. Magenta closed arrowhead, tissue chunk. Blue closed arrowhead, lift-out tool (copper block linking micromanipulator needle) attached to right side of tissue chunk. Scale bar, 10 μm. i, CryoSEM images showing attachment of the left tissue chunk to EM grid and cryoFIB cut between the left and right tissue chunks. Purple closed arrowhead, EM grid. Brown closed arrowhead, left tissue chunk. Yellow closed arrowhead, right tissue chunk. Blue closed arrowhead, lift-out tool. Scale bar, 10 μm. j, CryoFIB normal view image (56° stage tilt) of the tissue chunk after cryoFIB thinning to produce two 130–200 nm thick, ~8 μm wide, ~15 μm deep lamellae windows. Cyan closed arrowhead, lamella. Brown closed arrowhead, tissue chunk. Scale bar, 10 μm. k, CryoSEM showing serial attachment of the left and right tissue chunks. Yellow and brown closed arrowheads, left and right tissue chunks, respectively. Scale bar, 50 μm. Inset, overview image showing clipped half-moon Omniprobe EM grid. Scale bar, 0.5 mm. l, CryoEM overview showing left and right cryoFIB-milled tissue chunk attachment positions, left was lost and right remained during transfer cryoFIB to Krios TEM, respectively. Arrowheads, same as k. Scale bar, 50 μm.

For each modality, the chunk size of the benchmark dataset was chosen as a compromise between the size of individual chunks and the total number of chunks in the Zarr dataset. The plots above show typical power of 2 chunk sizes: between 32 and 1024 for the 2D data and between 16 and 128 for the 3D data. We chose a 2D chunk size of 256×256 for the CyIF-like dataset and a 3D chunk size of 32x32x32 for the LSM-like dataset. Note that due to the planar limitation of TIFF, the LSM dataset was stored as 2D TIFF tiles of size 32×32 but the benchmark looped over 32 tiles to measure the access time of the same chunk size.

Chunk patterns and chunk rates for the 10-digit digit string 4259522560. The chunk pattern is 2222, with chunk rates of 1.8 Hz (inside δ) and 2.6 Hz (outside). Each chunk was synthesized as a two-digit unit, using the AT&T Text-to-Speech System accentuation (see text). Note that a particular two-digit chunk has the same acoustics, regardless of whether it occurs in the 1.8- or 2.6-Hz 2222 chunk condition (red box). The 1.8-Hz stimulus is generated by increasing the gap between the chunks (with identical chunk acoustics).

Labelling of chunks: Beginning of the chunk; Inside the chunk; Outside the chunk. The unitary chunk formed of the token shows (highlighted in magenta in the top tree) is labelled O. Possible chunks to attach it to are tried in order of their NPMI. This results in the unitary chunk being appended to the chunk consisting of My and schedule. The resulting chunk forms My schedule shows.