Code
source("R/setup.R")
library(compareGroups)
data <- load_slides_and_distances("full")
slide1 <- data$slide1; slide2 <- data$slide2
cell_distances_1 <- data$distances_1; cell_distances_2 <- data$distances_2Co-registration — Akhil (single annotator)
ThioS quantification and HALO analysis
HALO analysis
All annotations performed by Akhil Palleria (single annotator; no inter-observer variability).
Metrics per sample
Thioflavin-S (ThioS)-positive amyloid plaques were quantified across six brain samples (Brain1–6; IgG control, n=3; Aducanumab-treated, n=3) using HALO image analysis software, with all annotations performed by a single rater (A.P.) to eliminate inter-observer variability.
Two plaque phenotypes were identified: cerebrovascular (CAA) and parenchymal plaques. Across all brains, parenchymal plaques were substantially more numerous than CAA plaques, with per-brain counts ranging from 2,681 to 3,961 (parenchymal) versus 284 to 446 (CAA). Consistent with this, parenchymal plaques accounted for the majority of total annotated area in every sample.
Code
t1 <- bind_rows(thios_1, thios_2) |>
group_by(brain, sample, plaque_type) |>
summarise(
n_plaques = n(),
total_area = sum(Area),
.groups = "drop"
)
t1 |> kbl() |> kable_styling("striped")| brain | sample | plaque_type | n_plaques | total_area |
|---|---|---|---|---|
| Brain1 | KK4_465 | CAA | 345 | 37005.25 |
| Brain1 | KK4_465 | Parenchymal | 3961 | 417446.50 |
| Brain2 | KK4_492 | CAA | 446 | 58097.50 |
| Brain2 | KK4_492 | Parenchymal | 2681 | 512135.00 |
| Brain3 | KK4_504 | CAA | 366 | 46653.50 |
| Brain3 | KK4_504 | Parenchymal | 2760 | 338232.25 |
| brain4 | KK4_496 | CAA | 389 | 35432.25 |
| brain4 | KK4_496 | Parenchymal | 3559 | 367226.00 |
| brain5 | KK4_502 | CAA | 370 | 44558.75 |
| brain5 | KK4_502 | Parenchymal | 3187 | 441707.00 |
| brain6 | KK4_464 | CAA | 284 | 22620.00 |
| brain6 | KK4_464 | Parenchymal | 2743 | 276220.25 |
Group-level comparison revealed no statistically significant difference between Aducanumab-treated and IgG-control animals in either total CAA area (41,759 ± 17,904 vs. 39,697 ± 6,076 µm²; p=0.865), total parenchymal plaque area (410,021 ± 121,107 vs. 374,302 ± 40,078 µm²; p=0.668), or the proportion of total plaque burden attributable to CAA (8.97 ± 1.32% vs. 9.69 ± 2.13%; p=0.653).
Code
t1_wide <- t1 |>
pivot_wider(id_cols = c(brain, sample), names_from = plaque_type, values_from = total_area) |>
mutate(CAA_percent = CAA / (Parenchymal + CAA) * 100) |>
inner_join(sample_dict, by = c("brain", "sample"))
caa1 <- compareGroups(treatment ~ CAA + Parenchymal + CAA_percent, data = t1_wide)
export2md(createTable(caa1, show.p.overall = T), caption = "")| Adu | IgG | p.overall | |
|---|---|---|---|
| N=3 | N=3 | ||
| CAA | 41759 (17904) | 39697 (6076) | 0.865 |
| Parenchymal | 410021 (121107) | 374302 (40078) | 0.668 |
| CAA_percent | 8.97 (1.32) | 9.69 (2.13) | 0.653 |
Plaque centroids
To enable spatial co-registration with single-cell Xenium transcriptomic data, the centroid of each plaque was computed as the midpoint of the HALO bounding box and converted from pixel coordinates to micrometres using a calibration factor of 0.2125 µm/pixel. This coordinate transformation provides a common spatial reference frame for subsequent nearest-neighbour analyses.
Centroid and distance functions are defined in R/setup.R and computed by load_slides_and_distances().
Cell-to-plaque distances (k-d tree)
Minimum Euclidean distances from each Xenium-profiled cell to its nearest CAA plaque (dist_caa) and nearest parenchymal plaque (dist_paren) were computed using a k-d tree algorithm. The resulting distance distributions were markedly different between plaque types. For CAA plaques, the distribution had a modal distance of approximately 200–300 µm and a long tail extending to ~1,500 µm, reflecting the relatively sparse and focal distribution of vascular amyloid deposits. By contrast, the majority of cells were within ~50 µm of a parenchymal plaque, consistent with the dense and spatially pervasive distribution of diffuse parenchymal amyloid observed histologically.
Code
bind_rows(cell_distances_1, cell_distances_2) |>
pivot_longer(c(dist_caa, dist_paren), names_to = "plaque_type", values_to = "min_dist") |>
ggplot(aes(x = min_dist)) +
geom_histogram(binwidth = 50, fill = "lightgray", col = "black") +
labs(x = "min. dist (µm)") +
facet_wrap(~plaque_type, scales = "free") +
theme_minimal() +
theme(strip.text = element_text(size = 14))
Proximity classification
Each cell was assigned to either the CAA or parenchymal plaque neighbourhood based on which plaque type was spatially closer.
Across the six samples (n = 51,959–69,664 cells per brain), the large majority of cells were classified as parenchymal-proximal (87.5–92.9% per sample), with 7.4–12.5% classified as CAA-proximal, broadly reflecting the disparity in plaque abundance between types. Between 4.1–6.6% of cells per sample fell within the 50 µm CAA proximity threshold, while 38.5–52.2% of cells were within 50 µm of a parenchymal plaque.
Each cell is assigned to whichever plaque type is nearest: CAA if dist_caa < dist_paren, Parenchymal otherwise.
Code
classify_proximity <- function(dist_data) {
dist_data |>
column_to_rownames("cell") |>
mutate(
caa = case_when(
dist_caa < 50 ~ "<50um",
dist_caa >= 50 ~ "≥50um"),
parenchymal = case_when(
dist_paren < 50 ~ "<50um",
dist_paren >= 50 & dist_paren < 150 ~ "50-150um",
dist_paren >= 150 ~ "≥150um"),
classification = if_else(dist_caa < dist_paren, "CAA", "Parenchymal")
) |>
select(dist_caa:classification)
}
slide1 <- AddMetaData(slide1, classify_proximity(cell_distances_1))
slide2 <- AddMetaData(slide2, classify_proximity(cell_distances_2))Code
md <- bind_rows(slide1@meta.data, slide2@meta.data)
c1 <- compareGroups(sample_ID ~ caa + parenchymal + classification, data = md, max.ylev = 6)
export2md(createTable(c1, show.p.overall = FALSE), caption = "Per sample")| KK4_464 | KK4_465 | KK4_492 | KK4_496 | KK4_502 | KK4_504 | |
|---|---|---|---|---|---|---|
| N=60064 | N=57104 | N=71401 | N=60182 | N=53359 | N=56289 | |
| caa: | ||||||
| <50um | 2540 (4.23%) | 2658 (4.65%) | 3865 (5.41%) | 3452 (5.74%) | 2665 (4.99%) | 3590 (6.38%) |
| ≥50um | 57524 (95.8%) | 54446 (95.3%) | 67536 (94.6%) | 56730 (94.3%) | 50694 (95.0%) | 52699 (93.6%) |
| parenchymal: | ||||||
| <50um | 24409 (40.6%) | 29621 (51.9%) | 26340 (36.9%) | 32095 (53.3%) | 25575 (47.9%) | 24012 (42.7%) |
| ≥150um | 8559 (14.2%) | 4908 (8.59%) | 11155 (15.6%) | 3287 (5.46%) | 3824 (7.17%) | 6522 (11.6%) |
| 50-150um | 27096 (45.1%) | 22575 (39.5%) | 33906 (47.5%) | 24800 (41.2%) | 23960 (44.9%) | 25755 (45.8%) |
| classification: | ||||||
| CAA | 5975 (9.95%) | 3962 (6.94%) | 8227 (11.5%) | 5028 (8.35%) | 4249 (7.96%) | 5781 (10.3%) |
| Parenchymal | 54089 (90.1%) | 53142 (93.1%) | 63174 (88.5%) | 55154 (91.6%) | 49110 (92.0%) | 50508 (89.7%) |
Code
c2 <- compareGroups(Treatment.Group ~ caa + parenchymal + classification, data = md)
export2md(createTable(c2), caption = "Per treatment group")| Adu | IgG | p.overall | |
|---|---|---|---|
| N=184824 | N=173575 | ||
| caa: | <0.001 | ||
| <50um | 9070 (4.91%) | 9700 (5.59%) | |
| ≥50um | 175754 (95.1%) | 163875 (94.4%) | |
| parenchymal: | 0.000 | ||
| <50um | 76324 (41.3%) | 85728 (49.4%) | |
| ≥150um | 23538 (12.7%) | 14717 (8.48%) | |
| 50-150um | 84962 (46.0%) | 73130 (42.1%) | |
| classification: | <0.001 | ||
| CAA | 18451 (9.98%) | 14771 (8.51%) | |
| Parenchymal | 166373 (90.0%) | 158804 (91.5%) |
Pseudoreplication warning: these cell-level statistics treat each cell as independent, but the true unit of replication is the brain (n = 3 per group). P-values are therefore inflated and should not be used for formal inference. See pseudobulk DE in notebooks 7 and 8 for properly replicated treatment comparisons.