5. Microglia-attributable proteins modulated by APP-NLGF

Overview

This notebook identifies the subset of microglia-attributable proteins (i.e. proteins lost when microglia are depleted in the FIRE-KO background) that are further modulated by the APP-NLGF mutation. Two complementary views:

  1. Interaction-based — proteins where the NLGF effect differs depending on microglia presence (interaction contrast, single statistical test).
  2. Set-based — overlap of significant proteins across the NLGF (hAppNLGF vs hApp) and microglia-depletion (hApp_FIRE vs hApp) contrasts, visualised as an UpSet plot.

The intersection of microglia-attributable hits and interaction-significant hits is the set of “microglia-amplified APP-NLGF responses” — the closest proteomic proxy for a DAM-like reactive signature in this design.

Libraries

Directories

Code 
base_dir <- "/nemo/lab/destrooperb/home/shared/zanettc/giulia_proteomics/bulk_proteomics"
run_num  <- "run4"

results_dir <- file.path(base_dir, "results", run_num)
objects_dir <- file.path(base_dir, "data", "processed", run_num)

interaction_dir <- file.path(results_dir, "Interaction_Microglia_APP")
dir.create(interaction_dir, recursive = TRUE, showWarnings = FALSE)

# Significance thresholds (kept consistent with notebook 03)
FDR_CUT   <- 0.05
LOGFC_CUT <- 0.5

Load DE results, processed abundance, and protein dictionary

Code 
res_df <- fread(file.path(results_dir, "Full_GroupComparison_Results.csv"))
processed_data     <- qs_read(file.path(objects_dir, "processed_msstats_data.qs2"))
protein_dictionary <- fread(file.path(objects_dir, "protein_dictionary.csv"))

# Map gene names onto results
res_ann <- res_df %>%
  filter(!is.na(adj.pvalue), is.finite(log2FC)) %>%
  left_join(protein_dictionary, by = "Protein") %>%
  mutate(Gene = ifelse(is.na(Gene) | Gene == "", Protein, Gene))

cat("Contrasts present:\n")
Contrasts present:
Code 
print(unique(res_ann$Label))
[1] "hAppNLGF_vs_hApp"            "hApp_FIRE_vs_hApp"          
[3] "hAppNLGF_FIRE_vs_hAppNLGF"   "hAppNLGF_FIRE_vs_hApp_FIRE" 
[5] "Interaction_APP_x_Microglia"

Define protein sets

Code 
# Helper: significant proteins in a given contrast (returns Gene-level set)
sig_genes <- function(df, contrast, direction = c("any", "up", "down")) {
  direction <- match.arg(direction)
  d <- df %>% filter(Label == contrast, adj.pvalue < FDR_CUT)
  d <- switch(direction,
              any  = d %>% filter(abs(log2FC) > LOGFC_CUT),
              up   = d %>% filter(log2FC >  LOGFC_CUT),
              down = d %>% filter(log2FC < -LOGFC_CUT))
  unique(d$Gene)
}

set_NLGF_in_micro      <- sig_genes(res_ann, "hAppNLGF_vs_hApp")            # NLGF effect, microglia present
set_FIRE_hApp          <- sig_genes(res_ann, "hApp_FIRE_vs_hApp")           # microglia depletion in hApp
set_FIRE_hAppNLGF      <- sig_genes(res_ann, "hAppNLGF_FIRE_vs_hAppNLGF")   # microglia depletion in hAppNLGF
set_NLGF_in_FIRE       <- sig_genes(res_ann, "hAppNLGF_FIRE_vs_hApp_FIRE")  # NLGF effect, no microglia
set_interaction        <- sig_genes(res_ann, "Interaction_APP_x_Microglia")

# Microglia-attributable: significantly *down* when microglia depleted in either APP background
microglia_attributable <- union(
  sig_genes(res_ann, "hApp_FIRE_vs_hApp",         direction = "down"),
  sig_genes(res_ann, "hAppNLGF_FIRE_vs_hAppNLGF", direction = "down")
)

cat("Set sizes:\n")
Set sizes:
Code 
print(c(
  NLGF_in_micro          = length(set_NLGF_in_micro),
  FIRE_hApp              = length(set_FIRE_hApp),
  FIRE_hAppNLGF          = length(set_FIRE_hAppNLGF),
  NLGF_in_FIRE           = length(set_NLGF_in_FIRE),
  Interaction            = length(set_interaction),
  microglia_attributable = length(microglia_attributable)
))
         NLGF_in_micro              FIRE_hApp          FIRE_hAppNLGF 
                    17                     74                     58 
          NLGF_in_FIRE            Interaction microglia_attributable 
                     5                      5                     72

UpSet plot — overlap of significant proteins across key contrasts

The user-of-interest set: union of NLGF (hAppNLGF vs hApp) and microglia-depletion (hApp_FIRE vs hApp). Adding the interaction contrast and the NLGF-without-microglia contrast lets us see which hits are microglia-dependent vs microglia-independent.

Code 
upset_input <- list(
  "NLGF effect (+microglia)" = set_NLGF_in_micro,
  "NLGF effect (−microglia)" = set_NLGF_in_FIRE,
  "Microglia effect (WT)"    = set_FIRE_hApp,
  "Microglia effect (NLGF)"  = set_FIRE_hAppNLGF,
  "APP × Microglia"          = set_interaction
)

# UpSetR uses fromList()
upset_obj <- upset(
  fromList(upset_input),
  order.by = "freq",
  nsets    = length(upset_input),
  nintersects = 30,
  text.scale = c(1.4, 1.2, 1.2, 1.0, 1.3, 1.1),
  point.size = 3,
  line.size  = 0.8,
  mainbar.y.label = "Shared significant proteins",
  sets.x.label    = "Significant per contrast"
)
print(upset_obj)

Code 
pdf(file.path(interaction_dir, "UpSet_significant_proteins.pdf"), width = 10, height = 6)
print(upset_obj)
dev.off()
png 
  2

Microglia-attributable + APP-NLGF-modulated proteins

Code 
# Two operational definitions of "modulated by APP-NLGF":
modulated_by_NLGF_pairwise   <- microglia_attributable %>% intersect(set_NLGF_in_micro)
modulated_by_NLGF_interaction <- microglia_attributable %>% intersect(set_interaction)

cat("Microglia-attributable AND significant in (hAppNLGF vs hApp):  ",
    length(modulated_by_NLGF_pairwise),  "\n")
Microglia-attributable AND significant in (hAppNLGF vs hApp):   1
Code 
cat("Microglia-attributable AND significant in interaction contrast:",
    length(modulated_by_NLGF_interaction), "\n")
Microglia-attributable AND significant in interaction contrast: 1
Code 
target_genes <- union(modulated_by_NLGF_pairwise, modulated_by_NLGF_interaction)
cat("Union (target heatmap set):", length(target_genes), "\n")
Union (target heatmap set): 2
Code 
# Export the table
target_table <- res_ann %>%
  filter(Gene %in% target_genes) %>%
  select(Gene, Protein, Description, Label, log2FC, adj.pvalue) %>%
  pivot_wider(
    id_cols     = c(Gene, Protein, Description),
    names_from  = Label,
    values_from = c(log2FC, adj.pvalue)
  ) %>%
  arrange(Gene)

fwrite(target_table,
       file.path(interaction_dir, "Microglia_attributable_NLGF_modulated.csv"))

Heatmap of microglia-attributable, APP-NLGF-modulated proteins

Uses MSstats’s TMP-summarised, median-normalised log2 abundance per sample (ProteinLevelData). Z-scored per protein.

Code 
abund <- as.data.table(processed_data$ProteinLevelData)

# MSstats stores abundance under either 'LogIntensities' or 'ABUNDANCE' depending
# on the package version — pick whichever exists.
abund_col <- intersect(c("LogIntensities", "ABUNDANCE"), names(abund))[1]
stopifnot(!is.na(abund_col))

abund_wide <- abund %>%
  as.data.frame() %>%
  select(Protein, RUN = originalRUN, GROUP, all_of(abund_col)) %>%
  rename(value = !!abund_col) %>%
  pivot_wider(id_cols = Protein, names_from = RUN, values_from = value) %>%
  left_join(protein_dictionary, by = "Protein") %>%
  mutate(Gene = ifelse(is.na(Gene) | Gene == "", Protein, Gene))

# Subset to target gene set; resolve duplicate Genes by taking the most-detected row
mat_df <- abund_wide %>%
  filter(Gene %in% target_genes) %>%
  group_by(Gene) %>%
  slice_max(order_by = rowSums(!is.na(across(starts_with("GA_")))),
            n = 1, with_ties = FALSE) %>%
  ungroup()

mat <- mat_df %>%
  select(starts_with("GA_")) %>%
  as.matrix()
rownames(mat) <- mat_df$Gene

# Z-score per protein
mat_z <- t(scale(t(mat)))
mat_z[!is.finite(mat_z)] <- NA

# Sample annotation — parse GA_N robustly from whatever survived in mat_z
sample_meta <- data.frame(
  Sample    = colnames(mat_z),
  SampleNum = as.integer(str_extract(colnames(mat_z), "(?<=GA_)\\d+"))
) %>%
  mutate(Condition = factor(case_when(
    SampleNum %in% 1:4   ~ "hApp",
    SampleNum %in% 5:8   ~ "hAppNLGF",
    SampleNum %in% 9:12  ~ "hAppNLGF_FIRE",
    SampleNum %in% 13:16 ~ "hApp_FIRE"
  ), levels = c("hApp", "hAppNLGF", "hApp_FIRE", "hAppNLGF_FIRE")))

stopifnot(
  ncol(mat_z) > 0,
  nrow(sample_meta) == ncol(mat_z),
  !anyNA(sample_meta$Condition)
)

# Reorder mat_z and sample_meta together so they stay aligned
samp_order  <- order(sample_meta$Condition)
mat_z       <- mat_z[, samp_order, drop = FALSE]
sample_meta <- sample_meta[samp_order, , drop = FALSE]

col_anno <- HeatmapAnnotation(
  Condition = sample_meta$Condition,
  col = list(Condition = c(
    "hApp"          = "#1b9e77",
    "hAppNLGF"      = "#d95f02",
    "hApp_FIRE"     = "#e7298a",
    "hAppNLGF_FIRE" = "#7570b3"
  ))
)

ht <- Heatmap(
  mat_z,
  name = "z-score",
  col  = colorRamp2(c(-2, 0, 2), c("navy", "white", "firebrick3")),
  top_annotation       = col_anno,
  cluster_columns      = FALSE,
  cluster_rows         = TRUE,
  show_row_names       = nrow(mat_z) <= 80,
  show_column_names    = TRUE,
  row_names_gp         = gpar(fontsize = 7),
  column_names_gp      = gpar(fontsize = 9),
  column_split         = sample_meta$Condition,
  column_title_gp      = gpar(fontsize = 10, fontface = "bold"),
  heatmap_legend_param = list(title = "z-score (log2 abundance)")
)

draw(ht)

Code 
pdf(file.path(interaction_dir, "Heatmap_microglia_attributable_NLGF_modulated.pdf"),
    width = 8, height = max(6, min(20, nrow(mat_z) * 0.18)))
draw(ht)
dev.off()
png 
  2
Code 
cat("Heatmap proteins:", nrow(mat_z), "\n")
Heatmap proteins: 2