The study
The question
Chemical fixation is the standard way to preserve tissue for electron microscopy, and it is known to distort the extracellular space. Rapid high-pressure freezing should distort it less. This dataset measures how much the two differ, tissue by tissue and region by region.
The unit
A crop is a small annotated cube of volume EM in which every cell and the extracellular space between them has been segmented. There are 55, across cortex, heart, kidney and liver, roughly half from each preparation. Each is measured independently; no crop appears in both arms.
The two preparations
These two colours mean the same thing on every chart on this site. Orange is always chemical fixation, blue is always rapid high-pressure freezing.
The three runs
Every metric family is computed three ways. Which one you should look at depends on what you are asking.
- Native resolution
- Each crop measured at the resolution it was acquired at — 2, 4 or 8 nm. Voxel size is not balanced across preparations, so a difference here can be a resolution effect rather than a biological one. Check it against the matched run before believing it.
- Resolution-matched (8 nm)
- Every crop downsampled to a common 8 nm voxel before measuring, so Chemical and HPF are compared on equal footing. The honest comparison for anything that depends on fine structure. Downsampling loses the narrowest channels in every crop equally.
- Degradation series
- One preparation measured repeatedly at 2, 4, 8 and 16 nm to show how each metric drifts as resolution is thrown away. This is the control that tells you how much of any difference could be resolution alone. Read the slope, not the value.
How to read the charts
In the analysis page, every dot is one crop and every row is a group of crops sharing a tissue, region or anatomy. Chemical sits above the line, HPF below. Groups share one horizontal scale so they can be compared against each other, not just within themselves. A vertical rule marks the median, and is omitted where a group has only one crop, because the median of one crop is that crop.
Cliff's δ, which the summary cards use
For a metric and a slice of crops, δ is the chance that a chemical crop reads higher than a rapid-HPF one, minus the chance it reads lower. It runs from −1 to +1; zero means the two are indistinguishable. It is non-parametric, which matters here because most groups have a handful of crops and no reason to be normal.
The words on the cards follow the usual thresholds: |δ| under 0.147 is negligible, under 0.33 small, under 0.474 medium, and above that large. They describe the size of the separation, not whether it would survive a significance test — with two crops against three, nothing would.
The two surfaces in the viewer
The crop page can show either side of the same boundary. They are built differently, and the difference matters when you compare them.
Membrane
One cell's ECS-facing skin, in the crop's own frame. The cell with the most ECS-facing surface is meshed at 16 nm; the patch is the part of it that faces extracellular space.
Coloured by signed curvature, protrusion / indentation against a 60 nm smoothed reference, or the gap to the nearest neighbouring cell.
Faces at the crop wall are trimmed: marching cubes caps the volume there, and the cap is not membrane.
ECS
The space itself, meshed at 8 nm from an 800 nm cube of every crop — equal volume, so two crops can be compared on a shared camera. The cube is not the middle of the crop: it is the position whose ECS fraction is closest to the whole crop's, because the middle of a liver crop can sit inside one hepatocyte and hold no ECS at all.
Coloured by curvature, protrusion / indentation, thickness or width. Thickness is the chord — march the normal through the space to the far wall. Width is the largest ball that fits, which the nearest wall in any direction bounds, so it stays smaller where sheets meet.
The cut faces are kept, in a darker grey, so the space reads as the solid it is. Curvature is averaged over 24 nm before it is drawn. The methods in full has the rest.
Both surfaces use one sign convention: positive curvature and positive deviation mean the membrane bulges into the extracellular space. Where a reading cannot be trusted — a kernel that reached the crop wall, a ray that left the box — the vertex is grey rather than wrong, and at 8 nm the narrowest resolvable channel is 16 nm, so the low end of the thickness scale is the sampling and not the tissue.
The datasets
The 55 crops come from nine imaged volumes. A crop is a hand-annotated cube inside one of these; the volume itself is the raw electron microscopy it was cut from.
| Dataset | Tissue | Preparation | Voxelnm | Crops | EM array | Public S3 | Open |
|---|
Metric dictionary
94 metrics in 7 families. The grey name under each entry is its column name in the CSVs. Use the filter to find one by name, unit or wording.
Volume fraction
How much of the crop is extracellular space · 10 metrics
Voxel-count ratios: what fraction of the volume is extracellular space and what fraction is cell. The most resolution-robust family here, because it counts voxels rather than measuring distances or surfaces.
10 metrics
ecs_bm_voxelscell_fractioncell_volume_um3cell_voxelstotal_volume_um3ecs_fractionecs_volume_um3ecs_primary_voxelsecs_voxelstotal_voxelsECS width
How wide the extracellular gaps are · 15 metrics
For every extracellular voxel, the Euclidean distance to the nearest cell voxel, summarised as percentiles across the crop.
15 metrics
n_ecs_voxelsnarrow_mean_nmn_narrow_voxelsnarrow_fractionfull_percentiles_nm_p10narrow_percentiles_nm_p10full_percentiles_nm_p25narrow_percentiles_nm_p25full_percentiles_nm_p75narrow_percentiles_nm_p75full_percentiles_nm_p90narrow_percentiles_nm_p90full_percentiles_nm_p50narrow_percentiles_nm_p50narrow_std_nmCell-to-cell gap
How far apart neighbouring cells sit · 14 metrics
Every extracellular voxel is assigned to its nearest cell; where two differently-assigned voxels touch, the gap between those cells is estimated from both distances plus the voxel spacing.
14 metrics
percentiles_nm_p10percentiles_nm_p25percentiles_nm_p75percentiles_nm_p90percentiles_nm_p50contact_fractions_p160contact_fractions_p20contact_fractions_p320contact_fractions_p40contact_fractions_p80gap_std_nmgap_mean_nmgap_min_nmn_boundary_facesSurface area to volume
How much ECS-facing membrane per unit of cell · 8 metrics
ECS-facing membrane area divided by cell volume, pooled across every cell in the crop that passes a physical size filter.
8 metrics
cell_density_per_um3total_cell_volume_nm3n_cells_totaln_cells_passingtotal_cell_cell_sa_nm2total_ecs_facing_sa_nm2total_outer_sa_nm2sa_v_ecs_per_nmMembrane shape
How folded, rough and bumpy the membrane is · 21 metrics
A mesh is fitted to each cell surface, and curvature, roughness at three spatial scales, and protrusion/indentation counts are measured on the ECS-facing part of it. All vertices from all cells in a crop are pooled into one sample, weighted by surface area.
21 metrics
curvature_abs_p25_per_nmcurvature_abs_p75_per_nmcurvature_abs_median_per_nmn_cells_consideredn_cells_includedfraction_concavefraction_convexcurvature_iqr_per_nmcurvature_std_per_nmfraction_flatindentation_density_per_um2indentation_counttotal_ecs_surface_nm2protrusion_density_per_um2protrusion_countroughness_rms_nm_p120roughness_rms_nm_p30roughness_rms_nm_p60curvature_signed_p10_per_nmcurvature_signed_p90_per_nmcurvature_signed_median_per_nmMembrane shape (mesh-based)
The same shape questions, measured on one representative cell · 23 metrics
A second, independent implementation working from a single ECS-facing membrane patch per crop, with its own boundary handling. This is what the 3D viewers show.
23 metrics
abs_curvature_p90_nm-1abs_curvature_p50_nm-1abs_deviation_p90_nmabs_deviation_p50_nmfrac_concavefrac_convexn_curvature_keptn_deviation_keptdeviation_p10_nmdeviation_p90_nmdeviation_p50_nmecs_fracn_gap_keptgap_bounded_fracfrac_indentpatch_facesgap_p10_nmgap_p90_nmgap_p50_nmfrac_protrusioncurvature_p10_nm-1curvature_p90_nm-1curvature_p50_nm-1Basement-membrane sensitivity
How much the kidney result depends on one annotation choice · 3 metrics
Kidney ECS fraction recomputed with basement membrane included and excluded, to bound how much of the kidney pattern is annotation rather than biology.
3 metrics
bm_voxelsecs_fraction_bm_excludedecs_fraction_with_bm