Code release accompanying Kraft, Lee et al., Mesoscale molecular architecture of the striatum across cell types and adult aging (Cell, 2026). Nomenclature note: "MSN" in script and variable names corresponds to "SPN" (spiny projection neuron) in the paper.
Code for the spatially constrained ligand–receptor analysis of human striatal slide-tags data (18 donors), comparing MSN/glial signaling between striatal Zone 1 and Zone 4, including the full radius/neighbor sensitivity battery.
Each candidate interaction (CellChat v2 database, CellChatDB.human; Jin et al.
2025) is scored per sender cell as ligand expression × mean receptor expression
among neighbor cells within a spatial radius. Multi-subunit receptors use the
geometric mean of subunit expression (full-complex scoring). Per donor, the
Zone 1 − Zone 4 difference (Δ) of this score is computed on a 10% trimmed mean.
Statistics, in two stages:
- Discovery — for each interaction, a donor-paired t-test on the per-donor Δ values, Benjamini–Hochberg FDR across interactions, q < 0.05.
- Robustness — for each discovered interaction, a within-donor permutation
null (zone labels shuffled within donor, 200 shuffles) yields each donor's
95th-percentile null Δ.
Robust_95= fraction of donors whose observed |Δ| exceeds their own 95th-percentile null. Reported interactions require Robust_95 ≥ 0.5 (≥ half of donors beat their own null).
Sensitivity design: the primary radius is 200 µm; the sweep covers 50–1,000 µm, a fixed-neighbor-count (kNN) arm holds neighbor count constant across cell types, and a density-matched arm assigns each sender its own radius from its kNN saturation distance.
| script | role |
|---|---|
March_LR_rerun_pipeline_v3.R |
v3.1 pipeline: scoring, radius sweep {50,100,200,400,1000} µm, kNN arm, paired-t + BH discovery, permutation-null robustness. All downstream arms source their functions from here (or from the radius-only variant). |
build_filtered_donor_objs.R |
one-time loader/filter/cache: per-donor Seurat objects → filtered .qs inputs (cell types, zone labels, spatial coordinates). |
run_LR_v3.R |
one-shot driver (build/load donor objects, source pipeline, run). |
run_LR_v3_stage2.R |
stage 2: adds the fixed-neighbor-count (kNN) arm (k = 5,10,20,40) additively; recomputes the null on the combined grid. |
test_pipeline_correctness.R |
unit tests of the scoring/stat/null machinery on synthetic data. |
| script | role |
|---|---|
low_k.R + run_lowk.R, run_lowk2.R |
extends the kNN arm downward (k = 1, 3, 5). |
matched_radii.R + run_matched.R |
density-matched radius arm: each sender scored at the physical distance of its own kNN-saturation k*, rounded to 100 µm. |
radius600.R + run_radius600.R |
single fixed 600 µm run (completes the per-sender radius grid). (Header comment is inherited from matched_radii.R; the script runs radius = 600 µm.) |
extract_coords.R |
exports per-cell spatial coordinates used by the neighbor-distance summaries. |
| script | role |
|---|---|
LR_pipeline_radius_only.R |
the v3.1 pipeline stripped to the radius arm only (identical scoring/stat/null functions); sourced by the two launchers below. |
run_explode_compare.R |
full-complex (geometric mean) vs exploded (per-subunit) receptor scoring at 200 µm, stock + curated database. |
run_tgfb2_variants.R |
targeted scoring of every TGFB2 receptor definition (including the curated TGFB2_TGFBR3 betaglycan single-receptor and TGFB2_TGFBR1_TGFBR3) through the same permutation null. |
R scripts for the published panels: the 4-panel sensitivity figure
(fig_4panel_combo.R), the composite G/H + sensitivity row
(Fig_composite_GH_sensitivity.R), pathway-family and ephrin/semaphorin panels,
custom Hedgehog/TGFβ and MSN-ligand panels, and Zone 1/Zone 4 network/chord
maps. These consume CSV outputs of the pipeline (paths need adjusting; see
below).
Run on a Linux compute cluster via a conda environment (seurat_env):
- R 4.3.3
- Seurat 5.3.0, SeuratObject
- qs 0.27.3, data.table 1.17.8, dbscan 1.1.11
- dplyr, tidyr, readr, tibble, Matrix, stringr
- Figure scripts additionally: ggplot2, patchwork, scales, ragg, extrafont
Interaction database: CellChatDB.human from CellChat v2 (Jin et al. 2025),
subset to the "Secreted Signaling" and "Cell-Cell Contact" categories.
01_core_pipeline/run_LR_v3.R(thenrun_LR_v3_stage2.Rfor the kNN arm)- Sensitivity arms in
02_sensitivity_arms/(eachrun_*.Ris standalone; all source the core pipeline for its functions) 03_complex_handling_custom_db/run_explode_compare.R(once withLR_EXPLODE=false, once withLR_EXPLODE=true) andrun_tgfb2_variants.R- Figure scripts in
04_figures/
The pipeline runs on per-donor Seurat objects saved as .qs, one file per
donor, named s<N>_with_master.qs, in the directory set by RAW_DIR at the
top of build_filtered_donor_objs.R. Each object's @meta.data must carry:
| column | content |
|---|---|
x_um2, y_um2 |
spatial coordinates in µm (cells with NA are dropped) |
Spatial_Domain |
domain/zone assignment; integer-coercible 1–6 (cells outside 1–6, incl. white matter, are dropped). The two compared domains are set in CONFIG (as run: 1 vs 4). |
Cell_types |
cell type per cell; used directly as the analysis type. The sender/receiver set is defined in CONFIG$cell_types (as run: MSN, Astrocyte, OPC, Oligo, Microglia, Vascular). |
For exact reproduction of the paper runs, the internal as-run column names are
also accepted and take effect when the generalized ones are absent:
Final_Zone_Assignments_No_Smooth (≙ Spatial_Domain) and Class_name +
AK_celltype (collapsed by the pipeline into the six analysis types).
The interaction database is not bundled: CellChatDB.human.rda is
downloaded automatically from the CellChat GitHub repository on first run and
cached under 00_db/ (an internet connection is needed once; subsequent runs
use the cache).
- Scripts contain absolute paths to the original cluster filesystem
(
/broad/macosko/...); adjust the path constants at the top of each script (FIG6,LRDIR,OUT_DIR, andRAW_DIR/OUT_BASEinbuild_filtered_donor_objs.R) to your local layout. Paths were deliberately left as-run for provenance. - To run on your own data, provide donor objects with
x_um2/y_um2,Spatial_Domain, andCell_typesmetadata — no other renaming needed. - Scripts are as-run except for a small input adapter (added for this release)
that accepts the generalized
Spatial_Domain/Cell_typescolumns; the scoring, statistics, and null machinery are untouched. - The permutation null shuffles zone labels within donor, so donor identity and per-donor density structure are preserved under the null.