Spatial Analyses

In this tutorial we will show you how you can analyze the spatial configuration of proteins on the cell surface of single cells, using PNA data.

Pixelator will generate a number of metrics from the graph structure inherent to the PNA data. In this tutorial we will show you how you can approach spatial analysis using the PNA proximity scores to get a deeper understanding of how proteins are organized on the cell surface.

After completing this tutorial, you will be able to:

• Interact with PNA proximity scores to analyze protein clustering and colocalization on the cell surfaces.

• Quantify changes in protein clustering: Utilize PNA proximity scores and differential tests to measure and compare protein clustering between conditions, revealing spatial rearrangements upon cellular stimulation.

• Identify changes in protein colocalization: Conduct differential proximity analysis to pinpoint protein pairs with significant changes in co-occurance across different samples, uncovering potential protein interactions under specific conditions.

• Visualize and interpret data: Effectively communicate your findings through volcano plots and heatmaps for differential analyses.

Setup

library(pixelatorR)
library(dplyr)
library(tibble)
library(stringr)
library(ggrepel)
library(Seurat)
library(here)
library(pheatmap)
library(ggplot2)
library(tidygraph)
library(ggraph)
library(tidyr)
library(patchwork)

Load data

We begin by loading our resting PBMC and PHA-stimulated PBMC dataset from previous tutorials.

DATA_DIR <- "path/to/local/folder"
Sys.setenv("DATA_DIR" = DATA_DIR)

# Load the object that you saved in the previous tutorial. This is not needed if it is still in your workspace.
pg_data_combined <- readRDS(file.path(DATA_DIR, "combined_data_annotated.rds"))
pg_data_combined

An object of class Seurat 
474 features across 1970 samples within 3 assays 
Active assay: PNA (158 features, 158 variable features)
 3 layers present: counts, data, scale.data
 2 other assays present: clr_assay, log_assay
 4 dimensional reductions calculated: pca, umap, harmony, harmony_umap

Obtaining proximity scores for target cells

Since T cells are the primary target of PHA, let us limit our analysis of spatial metrics to CD8 T cells. We use the cell types from the cell annotation tutorial.

pg_data_cd8 <- pg_data_combined %>% 
  subset(cell_type == "CD8 T cells")
ProximityScores(pg_data_cd8) %>% 
  head() %>% 
  knitr::kable()

marker_1	marker_2	join_count	join_count_expected_mean	join_count_expected_sd	join_count_z	join_count_p	log2_ratio	component
CD229	CD229	0	0.23	0.4682872	-0.2300000	0.4090459	0.0000000	resting_c76253fa939d5450
CD229	CD36	0	0.18	0.4353102	-0.1800000	0.4285763	0.0000000	resting_c76253fa939d5450
CD229	CD37	0	0.06	0.2386833	-0.0600000	0.4760778	0.0000000	resting_c76253fa939d5450
CD229	CD81	1	0.15	0.3859921	0.8500000	0.1976625	0.0000000	resting_c76253fa939d5450
CD229	CD277	0	0.38	0.5822284	-0.3800000	0.3519727	0.0000000	resting_c76253fa939d5450
CD229	CD73	3	2.57	1.5907244	0.2703171	0.3934582	0.2231941	resting_c76253fa939d5450

The measures in each row of the proximity scores’ table correspond to a protein pair (marker_1 and marker_2) in a given component. When marker_1 and marker_2 are the same, the measure indicates the clustering degree of that protein in the component; when they are different, the measure indicates the colocalization degree of those two proteins.

Proximity measures

When analyzing spatial proximity, we primarily examine two key values. First, “join_count_z” is a z-score indicating the statistical significance of the observed number of links between marker_1 and marker_2 antibodies compared to random expectation. A higher absolute z-score suggests a more significant deviation from chance, but it reflects significance, not the degree of proximity itself. Higher total counts can lead to higher z-scores even for similar proximity degrees. Second, “log2_ratio” represents the log2 of the ratio between observed and expected (by chance) edges between marker_1 and marker_2. This metric provides a more direct measure of the degree of proximity. Therefore, to answer questions like “Are these two proteins colocalized?” or “Is this protein significantly clustered?”, focus on the “join_count_z” value. For questions such as “Has this marker’s clustering increased after stimulation?”, the log2_ratio metric is more informative.

Filtering the low-abundance markers

Not all proteins will be expressed by every cell, i.e. some proteins might have very low abundance in some cells. For such proteins, the proximity scores should be excluded in the corresponding cells. Typically, filtration should be performed before spatial proximity analysis to reduce the potential noise from low abundance markers.

One strategy is to define the filter threshold based on the isotype control fraction. The isotype control fraction is the fraction of counts that are attributed to the isotype control antibody, which should not bind to any specific target. Any value lower than this threshold is considered noise and should be filtered out.

We can use the median isotype control fraction plus two median absolute deviations (MAD) as a cutoff. If the value is smaller than 0.001, we fall back to a threshold of 0.001 to make sure that the threshold is conservative.

isotype_control_fraction_cutoff <- median(pg_data_combined$isotype_fraction) + 2*mad(pg_data_combined$isotype_fraction)
isotype_control_fraction_cutoff <- max(isotype_control_fraction_cutoff, 0.001)

pg_data_combined[[]] %>% 
  select(condition, isotype_fraction) %>% 
  ggplot(aes(condition, isotype_fraction, fill = condition)) +
  geom_violin(draw_quantiles = 0.5) + 
  geom_jitter(size = 0.4, alpha = 0.5) +
  theme_bw() +
  labs(y = "Fraction of isotype control counts") +
  scale_y_continuous(labels = scales::percent) +
  geom_hline(yintercept = isotype_control_fraction_cutoff, linetype = "dashed")

pixelatorR provides a method FilterProximityScores() to filter out protein pairs based on a predefined isotype control fraction threshold. To use this method, we need to set add_marker_proportions = TRUE when fetching the proximity scores. This will create two new columns p1 and p2 that represent the proportion of counts for marker_1 and marker_2 in the component. When we apply FilterProximityScores() with a predefined isotype control fraction threshold, it will filter out any protein pair where the minimum proportion is lower than the threshold.

As an example, if our isotype control fraction threshold is 0.001 and the component (cell) has 20,000 nodes, any protein pair where the minimum count is less than 0.001 * 20,000 = 20 will be filtered out.

filtered_cd8_proximity_scores <- ProximityScores(
  pg_data_cd8, 
  meta_data_columns = "condition", 
  add_marker_counts = TRUE,
  add_marker_proportions = TRUE
) %>%
  FilterProximityScores(background_threshold_pct = isotype_control_fraction_cutoff)

Proximity score profiling

We can now analyze the proximity profiles of the CD8 T cells. The SummarizeProximityScores method can be used to compute various summary statistics from a proximity score table. By default, it calculates the mean log2 ratio while setting missing observations to 0. It also reports the fraction of cells where the protein pair is detected, which is useful for filtering and visualization.

Below we summarize the proximity scores for the CD8 T cells in each condition and visualize the results for the stimulated cells in a heatmap. Note that area of the dots are scaled by the fraction of cells where the pair is detected, which is useful for understanding if the pair is detected in the whole population or a fraction of it.

summarized_cd8_proximity_scores <- filtered_cd8_proximity_scores %>% 
  SummarizeProximityScores(group_vars = "condition")

selected_markers <- summarized_cd8_proximity_scores %>% 
  filter(pct_detected > 0.5) %>% 
  arrange(desc(abs(mean_log2_ratio))) %>%
  slice_head(n = 60) %>% 
  select(marker_1, marker_2) %>% 
  unlist() %>% 
  unique()
  
summarized_cd8_proximity_scores %>% 
  filter(condition == "stimulated") %>% 
  filter(marker_1 %in% selected_markers & marker_2 %in% selected_markers) %>% 
  ColocalizationHeatmap(
    marker1_col = "marker_1", 
    marker2_col = "marker_2",
    value_col = "mean_log2_ratio",
    size_col = "pct_detected",
    type = "dots",
    clustering_method = "ward.D2",
    legend_range = c(-1, 1)
  ) +
  ggtitle("Colocalization in CD8 T cells")

Clustering analysis

Let’s look at the significantly clustered proteins in each sample. We can get the protein clustering scores by filtering our summary table on rows where marker_1 = marker_2. We can also filter the table to include proteins with high average log2 ratio that are detected in a large fraction of cells. This will help us identify the most clustered proteins.

summarized_cd8_proximity_scores %>% 
  filter(marker_1 == marker_2) %>% 
  arrange(desc(mean_log2_ratio)) %>% 
  filter(mean_log2_ratio > 0.3) %>% 
  select(condition, marker_1, marker_2, pct_detected, mean_log2_ratio) %>%
  pivot_wider(
    names_from = condition, 
    values_from = c(pct_detected, mean_log2_ratio)
  ) %>% 
  filter(pmin(pct_detected_stimulated, pct_detected_resting) > 0.5) %>% 
  head(n = 20) %>% 
  knitr::kable()

marker_1	marker_2	pct_detected_stimulated	pct_detected_resting	mean_log2_ratio_stimulated	mean_log2_ratio_resting
CD81	CD81	0.9364548	0.9716312	2.1724251	1.3652747
CD82	CD82	0.9933110	0.9007092	1.9469293	1.3364326
CD26	CD26	0.6956522	0.7163121	0.9097436	0.7058541
CD58	CD58	0.9698997	0.6382979	0.5903486	0.9033571
CD52	CD52	1.0000000	1.0000000	0.8713160	0.5629012
CD48	CD48	1.0000000	1.0000000	0.5802723	0.3626769
CD162	CD162	0.9364548	1.0000000	0.5559009	0.3941196
CD352	CD352	0.9966555	0.9929078	0.4627413	0.3184433
CD45RA	CD45RA	0.8996656	0.8581560	0.3970703	0.3549928
CD59	CD59	1.0000000	1.0000000	0.3426679	0.3856913
CD156c	CD156c	0.9464883	0.9929078	0.3385367	0.3798949
CD50	CD50	1.0000000	1.0000000	0.3744366	0.3448415

Here, we can for example see that CD81 and CD82 clustering goes up after PHA stimulation.

Differential clustering analysis

Another question is whether we can observe an increase in clustering of any proteins upon stimulation. We already expect CD82 to be one of those but we recall that join_count_z is an indication of significance, so this time, we compare degrees of clustering using the log2_ratio.

test_results <- DifferentialProximityAnalysis(
  filtered_cd8_proximity_scores,
  metric_type = "self",
  target = "stimulated",
  reference = "resting",
  contrast_column = "condition", 
  proximity_metric = "log2_ratio",
  min_n_obs = 25
)

# Visualize this in a volcano plot
test_results %>%
  left_join(
    summarized_cd8_proximity_scores,
    by = c("marker_1", "marker_2", "target" = "condition")
  ) %>% 
  ggplot(aes(diff_median, -log10(p_adj), label = marker_1, color = mean_log2_ratio)) +
  geom_point() +
  geom_text_repel(max.overlaps = 5) +
  theme_minimal() +
  guides(color = guide_colorbar(title.position = "top")) +
  labs(x = "Difference in medians",
      y = expression(-log[10] ~ (adj. ~ p - value)), 
      color = "Mean clustering\nafter stimulation") +
  geom_vline(xintercept = 0, linetype = "dashed")

We see that CD82 and CD52 have indeed increased in clustering after PHA-stimulation as well as other markers including CD50, CD44, CD45, HLA-ABC, and B2M. We also see that CD5 and CD27 have decreased in clustering. Let us now look at the clustering distributions for these markers.

filtered_cd8_proximity_scores %>% 
  filter(marker_1 == marker_2) %>% 
  filter(marker_1 %in% c("CD50", "CD5", "CD27", "CD52", "CD82", "CD102", "HLA-ABC", "B2M")) %>% 
  ggplot(aes(marker_1, log2_ratio, fill = condition)) +
  geom_violin(draw_quantiles = 0.5) +
  facet_wrap(~marker_1, scales = "free", ncol = 4) +
  theme_bw() +
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank()) +
  labs(title = "Clustering distributions") +
  geom_hline(yintercept = 0, linetype = "dashed", color = "black")

Colocalization analysis

Another aspect of spatial organization is whether pairs of proteins colocalize. So now, we focus on protein-pairs and their proximity scores. Here, we consider a protein pair to be significantly colocalized if it’s mean log2_ratio is above 0.3 and is detected in at least 80% of cells in both conditions.

summarized_cd8_proximity_scores %>% 
  filter(marker_1 != marker_2) %>% 
  arrange(desc(mean_log2_ratio)) %>% 
  filter(mean_log2_ratio > 0.3) %>% 
  select(condition, marker_1, marker_2, pct_detected, mean_log2_ratio) %>%
  pivot_wider(
    names_from = condition, 
    values_from = c(pct_detected, mean_log2_ratio)
  ) %>% 
  filter(pmin(pct_detected_stimulated, pct_detected_resting) > 0.8) %>% 
  knitr::kable()

marker_1	marker_2	pct_detected_stimulated	pct_detected_resting	mean_log2_ratio_stimulated	mean_log2_ratio_resting
CD81	CD82	0.9331104	0.8794326	1.4808266	0.8959510
CD49D	CD81	0.8528428	0.8936170	0.5615125	0.4217757
CD156c	CD81	0.8963211	0.9645390	0.5208844	0.4941323
CD156c	CD82	0.9431438	0.8936170	0.4693419	0.3361408
CD52	CD59	1.0000000	1.0000000	0.3965301	0.4127493

Here, we can for example see that CD52/CD59 have roughly the same colocalization strength in both unstimulated and PHA-stimulated CD8 T cells whereas CD81/D82 has a much higher colocalization score in PHA-stimulated CD8 T cells.

Given that there can be several colocalizing protein-pairs, a protein-pair graph illustration is one alternative for getting an overview of colocalization patterns in a sample. Below we use the same filtering procedure as above to filter for significant colocalization pairs.

markers_to_plot <- c("CD52", "CD59", "CD81", "CD82", "CD43", "CD44", "CD45", "CD45RA", "CD45RB", "CD8")

df_net <- summarized_cd8_proximity_scores %>%  
  filter(marker_1 %in% markers_to_plot & marker_2 %in% markers_to_plot) %>% 
  filter(abs(mean_log2_ratio) > 0.1 & pct_detected > 0.5) %>% 
  mutate(from = marker_1, to = marker_2) %>% 
  select(from, to, mean_log2_ratio, condition)

tidygraph::as_tbl_graph(df_net)  %>%
  ggraph(layout = "kk") +
  geom_edge_link(aes(color = mean_log2_ratio, width = abs(mean_log2_ratio))) +        
  geom_node_label(aes(label = name), size = 2.5) +
  scale_edge_color_gradientn(
      colours = c("#1F395F", "#647DA3", "#BBC8DA", 
                  "#FFFFFF", "#F0BCCD", "#BE5F7D", 
                  "#781534"),
      limits = c(-abs(max(df_net$mean_log2_ratio)), abs(max(df_net$mean_log2_ratio)))
  ) +
  scale_x_continuous(expand = c(0.2, 0.2)) +
  scale_y_continuous(expand = c(0.2, 0.2)) +
  facet_edges(~condition, nrow = 1) +
  theme(aspect.ratio = 1,
      panel.background = element_rect(fill = "white", color = "black"),
      panel.spacing = unit(1, "lines")) +
  labs(edge_color = "Mean\nlog2 ratio") +
  guides(edge_width = "none")

note

Keep in mind that the decision to remove protein pairs present in fewer cells should be guided by your experimental goals. If you aim to identify colocalization changes in rare cell subpopulations, retaining these pairs for downstream analysis might be crucial.

Colocalization analysis

Let’s run a differential colocalization test to find protein-pairs that differ significantly in proximity scores between conditions.

coloc_test_results <- DifferentialProximityAnalysis(
  filtered_cd8_proximity_scores,
  metric_type = "co",
  target = "stimulated",
  reference = "resting",
  contrast_column = "condition", 
  proximity_metric = "log2_ratio",
  min_n_obs = 50
)

# Visualize this in a volcano plot
coloc_test_results %>%
  mutate(pair = paste0(marker_1, "/", marker_2)) %>% 
  arrange(-diff_median) %>% 
  mutate(label = case_when(
    row_number() %in% 1:5 ~ pair,
    row_number() %in% (n() - 4):n() ~ pair,
    TRUE ~ NA_character_
  )) %>% 
  ggplot(aes(diff_median, -log10(p_adj), label = label)) +
  geom_point() +
  geom_text_repel(max.overlaps = 10) +
  theme_minimal() +
  guides(color = guide_colorbar(title.position = "top")) +
  labs(
      x = "Difference in medians",
      y = expression(-log[10] ~ (adj. ~ p - value))
  )

We observe several differentially colocalized protein pairs. Let us look at a clustered heatmap to get a better grasp of these protein pair colocalization patterns.

ColocalizationHeatmap(
  coloc_test_results %>% 
    mutate(diff_median = case_when(
      p_adj < 0.001 ~ diff_median,
      TRUE ~ 0
    )),
  marker1_col = "marker_1", 
  marker2_col = "marker_2", 
  value_col = "diff_median", 
  type = "dots", 
  size_col_transform = \(x) {-log10(x)}
) &
  labs(size = "-log10p(p_adj)") &
  scale_size(range = c(1, 5), limits = c(0, max(-log10(coloc_test_results$p_adj)))) &
  labs(fill = "Difference\nin medians")

Now let’s take a holistic view of colocalization scores and make a UMAP. We will only include protein pairs that display a large difference in colocalization score between our two conditions.

# Make sure that we are using the correct assay
DefaultAssay(pg_data_cd8) <- "PNA"
# Convert colocalization table to a new assay
pg_data_cd8 <- ProximityScoresToAssay(pg_data_cd8, values_from = "log2_ratio", missing_obs = 0)
DefaultAssay(pg_data_cd8) <- "proximity"

# Filter test results to define pairs of interest
proteins_of_interest <- coloc_test_results %>%
  filter(p_adj < 0.01) %>%
  arrange(-diff_median) %>%
  select(marker_1, marker_2) %>%
  unlist() %>%
  unique()

# Find pairs of interest
pairs_of_interest <- coloc_test_results %>%
  select(marker_1, marker_2) %>%
  filter(
      marker_1 %in% proteins_of_interest &
      marker_2 %in% proteins_of_interest
  ) %>%
  tidyr::unite(marker_1, marker_2, col = "pairs", sep = "/") %>%
  pull(pairs)

# Set variable features
VariableFeatures(pg_data_cd8) <- pairs_of_interest

# Create UMAP
pg_data_cd8 <- pg_data_cd8 %>%
  ScaleData() %>%
  RunPCA(reduction.name = "pca_coloc") %>%
  RunUMAP(dims = 1:10, reduction = "pca_coloc", reduction.name = "umap_coloc")

# Plot
DimPlot(pg_data_cd8, group.by = "condition", reduction = "umap_coloc") +
    theme_classic() +
  labs(title = "UMAP of CD8 T cells based on colocalization scores")

Building upon the insights gained from Proximity Network Assay data analysis, we now turn our attention to the spatial organization of proteins on the cell surface and how their clustering and interactions vary across conditions. The next tutorial in this series introduces cell graph visualizations, an exciting method that allows us to visualize and intuitively understand the dynamic clustering of proteins, providing deeper insights into their roles in cellular function.