An inter-tissue recognition process and the transition from asymmetric to symmetric gene expression patterns at the Arabidopsis graft junction involves differential auxin and sugar responses
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
R
README_files
data
inst/extdata
man
.Rbuildignore
.gitignore
DESCRIPTION
GraftingScripts.Rproj
LICENSE
NAMESPACE
README.Rmd
README.md

README.md

Reproducible script for the publication

Alexander Gabel 12 9 2017

Charles W. Melnyk, Alexander Gabel, Thomas J. Hardcastle, Sarah Robinson, Shunsuke Miyashima, Ivo Grosse and Elliot M. Meyerowitz. Transcriptome dynamics at Arabidopsis graft junctions reveal an intertissue recognition mechanism that activates vascular regeneration. PNAS. (2018). https://doi.org/10.1073/pnas.1718263115.

Contents

Getting started

Installation of necessary packages

Install and load packages containing the functions for the analyses

install.packages("devtools")
install.packages("readxl")
devtools::install_github("GraftingScripts")
library(GraftingScripts)
library(readxl)

Load expression data

Loading length normed expression data generated by BaySeq

data(exp.data)

Calculate TPMs and summarize replicates

Next, we calculate the corresponding TPM values based on the length normed data and also the TPM expression matrix based on the summarized replicates.

exp.data.tpm <- apply(exp.data,2,function(i)i/sum(i)*10^6)
exp.data.tpm.sum <- sapply(seq(1,by=2,length.out=ncol(exp.data)/2),function(i){
                           # calculating the geometric mean of biological replicates
                           exp_col <- exp(rowMeans(log(exp.data[,c(i,i+1)]))) 
                           exp_col/sum(exp_col) * 1e6
                           })
colnames(exp.data.tpm.sum) <- gsub(pattern = "A|B", replacement = "h", 
                                   x = colnames(exp.data)[seq(1, by=2, length.out = ncol(exp.data)/2)])

Visualization

Generating dendrogram based on summarized replicates

To visualize the similarity of the different samples (conditions) we calculated the pearson correlation coefficient between each sample and used the measure 1-correlation as a similarity score to perform a hierarchical clustering. The following code snippet applies the plot_colored_dendrogram function to perform these calculations and plot the resulting dendrogram. The data groupS is necessary for the function for later color coding the different samples in the tree.

data(groupS)
plot_colored_dendrogram(exp_data = exp.data.tpm.sum, groups = groupS)

Generating PCA plot

Another way of visualizing the data and how the different samples cluster is the principle component analysis, which can be applied by the following function call. Alternatively to the PCA this function can also generate a MDS plot if the parameter do.MDS = TRUE.

GraftingScripts::plotPCA(exp_data = exp.data.tpm.sum, plot_time_points = T, groups = groupS, do.MDS = F, cex=5, do.legend = T, log=T)

Define differentially expressed genes

Calculate Fold changes against the intact samples

In the next step we calculated for each time point of the grafted and separated samples the foldchanges to the corresponding intact samples, beginning at 6 hours after separation/grafting.

fc.exp.data <- exp.data.tpm.sum[,-which(colnames(exp.data.tpm.sum) %in% "0h Intact")]
fc.list <- list()
sample.Indices <- 9:16
for(i in 1:(ncol(fc.exp.data)/8-1)){
  fc.list[[i]] <- log2(fc.exp.data[,sample.Indices]+1) - log2(fc.exp.data[,1:8]+1)
  sample.Indices <- sample.Indices + 8 
}
names(fc.list) <- unique(unlist(lapply(strsplit(x=colnames(fc.exp.data),split=" "),function(i)paste(i[-1],collapse = " "))))[-1]

For the definition of differentially expression we used additionally the marginal likelihoods from our baySeq analysis. The data is stored in two lists ml.list.up containing the marginal likelihoods of grafted and separated samples in the case that they are upregulated compared to the intact samples. The list ml.list.down contains the marginal likelihoodsin the case that the grafted or separated samples are downregulated compared to the intact samples. ### Integrate the marginal likelihoods from BaySeq

data(ml.list.up)
data(ml.list.down)

Load the gene sets from previous publications

In this step we load the gene sets from previously published datasets which we investigated in our study. In this code snippet we extract all ids from the excel sheets and put them into our gene_list. If the ids seem to be probe ids we use the biomartR package to match the corresponding gene id to the probe id. After all gene sets are in the list we add two additional random lists, which later should not show any statistical significance. The gene_list object is stored in the GraftingScripts package and can be loaded by data(gene_lists).

xl_file <- system.file("extdata", "all_gene_subsets_from_manuscript_original.xlsx", package="GraftingScripts") 
sheet_names <- excel_sheets(xl_file)
gene_lists <- list()

for(i in seq_along(sheet_names)){

  gset <- as.matrix(read_excel(xl_file,sheet = sheet_names[i], col_names = F))[,1]
  gset <- unlist(lapply(strsplit(x = gset,split = ";| /// "),function(i)i[1:length(i)]))
  gset <- gsub(pattern = "\\*| ",replacement = "",x = gset)

  # separate gene_ids (beginning with AT) from possible probe_ids (beginning with numbers)
  gset_gene_ids <- gset[grepl(pattern = "^AT", x = gset, ignore.case = T)]
  gset_probe_ids <- gset[grepl(pattern = "^[0-9]", x = gset, ignore.case = T)]

  if(length(gset_probe_ids) > 0){
    # convert probe ids into gene ids with biomartr
    gset_gene_ids_from_probes <- biomartr::biomart(genes = gset_probe_ids, mart = "plants_mart", 
                                                   dataset = "athaliana_eg_gene", filters = "affy_ath1_121501", 
                                                   attributes = "ensembl_gene_id")[,2]
    gset <- unique(toupper(gset_gene_ids_from_probes), toupper(gset_gene_ids))
  }else{
    gset <- unique(toupper(gset_gene_ids))
  }
  gene_lists[[i]] <- gset
}
names(gene_lists) <- sheet_names

set.seed(10)
gene_lists[[length(gene_lists)+1]] <- sample(250, x=rownames(exp.data))
gene_lists[[length(gene_lists)+1]] <- sample(500, x=rownames(exp.data))
 
names(gene_lists)[(length(gene_lists)-1):(length(gene_lists))] <- c("250_random","500_random")

Calculate transcriptional overlap plots and check significance

In this step we calculate the transcriptional overlap of up- and down- regulated genes from previously published datasets compared to our transcriptome dataset. First, we defined a foldchange threshold and a marginal likelihood threshold to define if a gene is differentially expressed. Afterwards we go through the lists of published gene sets and take only those genes which are differentially expressed in our dataset. Second, we tested if the ratio of up- and down- regulated differentially expressed genes in a published is significantly different to the ratio of up- and down- regulated genes in the complete transcriptome dataset. This is done by the function do_fisher_up_down. After all test are performed and all p-values are calculated we correct fo multiple testing with the Benjamini-Yekutieli method. This is done by applying the function adjust_and_split which summarizes all p-values, correct them and split the p-values to their original gene list. When all tests are done we call barplot_up_down to visualize the transcriptional overlap which means we plot the relative number of published genes which are actually differentially expressed in our dataset.

fc <- 1

plot_dir <- paste0("Histogram_abs_logfc_gt_",fc,"_and_ML_gt_0.9")

if(!file.exists(plot_dir)){
  dir.create(plot_dir)
}

fisher_res_list <- p_val_list <- list()

for(i in seq_along(gene_lists)){
  fisher_res_list[[i]] <- do_fisher_up_down(fc.list, fg.ids=gene_lists[[i]], ml.list.up, ml.list.down, 
  fc.threshold = fc, ml.threshold=0.9, alternative = "two.sided")
  p_val_list[[i]] <- fisher_res_list[[i]]$p.val.mat
}

names(fisher_res_list) <- names(p_val_list) <- names(gene_lists)
# adjusting p-values with Benjamini & Yekutieli
adjusted_p_mat_list <- adjust_and_split(p_val_list, "BY")
names(adjusted_p_mat_list) <- names(p_val_list)

for(i in seq_along(gene_lists)){

  glist_name <- names(gene_lists)[i]

  up_fc <- fisher_res_list[[i]]$fg.up_fc
  down_fc <- -fisher_res_list[[i]]$fg.down_fc

  rownames(up_fc) <- names(ml.list.up)
  rownames(down_fc) <- names(ml.list.down)

  rel_up_mat <- t(up_fc/length(fisher_res_list[[i]]$fg.ids))
  rel_down_mat <- t(down_fc/length(fisher_res_list[[i]]$fg.ids))

  names_arg <- gsub(pattern = " ",replacement = "\n",rownames(up_fc))
  rect_names <- unlist(lapply(strsplit(colnames(fc.list[[1]])," "),function(l)l[1]))

  mainText <- paste0(glist_name," (",length(fisher_res_list[[i]]$fg.ids)," genes)")

  filename_prefix <- paste0(plot_dir,"/Histogram_", glist_name)

  pdf(paste0(filename_prefix,"_abs_logFC_gt_",fc,"_ml_gt_09.pdf"),width = 9,height = 6)
    par(mar=c(3.1,4.1,3.1,.1), cex=1.5)
    barplot_up_down(up_mat = rel_up_mat, down_mat = rel_down_mat, p_val_mat = adjusted_p_mat_list[[i]], names.arg = rep(rect_names,4), 
    main = mainText, labels = names_arg)
  dev.off()
}

Defining symmetrically and asymmetrically expressed genes (Table 1)

In this step we are interested in the number of genes from previous publications that are overlapping with symmetrically and asymmetrically expressed genes in our dataset defined by a baySeq clustering. The results from this baySeq analysis can be found in the object sym_asym_bS. In the next lines we use each published geneset and calculate the number of intersecting genes with the baySeq clustering. The object overlap_stats contains the results from the Fisher's exact test, which was used to test if the number of overlapping genes is higher than expected. The resulting p-values from all tests were aagin summarized and corrected against multiple testing with the the Benjamini-Yekutieli method. In the last loop of this code snippet the results were summarized in a table and each table is stored in the list sym_asym_table_list which can also be loaded from the GraftingScripts package.

# load genes that are symmetrically or asymmetrically expressed based on BaySeq filtering
data(sym_asym_bS)
number_genes <- nrow(exp.data)

p.val.mat <- exp_num_overlap_genes <- obs_num_overlap_genes <- matrix(nrow=length(gene_lists), ncol=length(sym_asym_bS))
num_gene_sym_asym_bS <- unlist(lapply(sym_asym_bS, length))

for(i in seq_along(gene_lists)){

  glist_name <- names(gene_lists)[i]
  gset <- gene_lists[i]
  
  print(glist_name)
  for(j in seq_along(sym_asym_bS)){
  
    overlap_stats <- test_sym_asym(gene_set_1 = gene_lists[[i]], gene_set_2 = sym_asym_bS[[j]], number_genes_complete = number_genes, alternative = "greater")
    p.val.mat[i,j] <- overlap_stats$p.value
  }
}
colnames(p.val.mat) <- names(sym_asym_bS)
rownames(p.val.mat) <- names(gene_lists)

# Correction of p-values
p.adjusted.mat <- matrix(nrow = nrow(p.val.mat), ncol = ncol(p.val.mat), p.adjust(p.val.mat, method = "BY")) 

rownames(p.adjusted.mat) <- rownames(p.val.mat)
colnames(p.adjusted.mat) <- colnames(p.val.mat)

# Summarizing the results, equivalent to Table 1
baySeq_conditions <- c( "Symm", "Asym_tGb", "Asym_bGt")
names(baySeq_conditions) <- c("Graft Bottom = Top", "Graft Top > Bottom", "Graft Bottom > Top")

sym_asym_table_list <- list()

for(i in seq_along(gene_lists)){
  
  glist_name <- names(gene_lists)[i]
  
  cond_list <- list()
  for(cond in seq_along(baySeq_conditions)){
  
    indices <- grepl(x = colnames(p.adjusted.mat), pattern = baySeq_conditions[cond])
    cond_list[[cond]] <- data.frame(HAG = unlist(lapply(strsplit(x = grep(x = colnames(p.adjusted.mat), 
                                                                          pattern = baySeq_conditions[cond], value = T), 
                                                                 split = "_"), function(c_name)c_name[1])),
                                    Treatment = rep(length(gene_lists[[i]]),8),
                                    baySeq_Condition = num_gene_sym_asym_bS[indices],
                                    Overlap = obs_num_overlap_genes[i, indices], 
                                    check.names = F)
    cond_list[[cond]] <- data.frame(cond_list[[cond]], '%' = round((cond_list[[cond]]$Overlap/cond_list[[cond]]$baySeq_Condition)*100), check.names=F )
    sig_overlap <- p.adjusted.mat[i,indices] < 0.05
    cond_list[[cond]]$`%`[sig_overlap] <- paste0(cond_list[[cond]]$`%`[sig_overlap], "*")  
    colnames(cond_list[[cond]])[2:3] <- c(names(gene_lists)[i],names(baySeq_conditions)[cond])
  }
   sym_asym_table_list[[i]] <- data.frame(cond_list, check.names=F)
}
names(sym_asym_table_list) <- names(gene_lists)
HAG S2_S3h_CSt3h_down Graft Bottom = Top Overlap % HAG S2_S3h_CSt3h_down Graft Top > Bottom Overlap % HAG S2_S3h_CSt3h_down Graft Bottom > Top Overlap %
6 1998 4988 107 2 6 1998 6679 53 1 6 1998 4971 1563 31*
12 1998 3473 68 2 12 1998 7111 112 2 12 1998 5657 1526 27*
24 1998 4135 88 2 24 1998 6873 72 1 24 1998 4942 1525 31*
48 1998 3689 113 3 48 1998 6601 93 1 48 1998 4915 1427 29*
72 1998 10421 530 5 72 1998 2459 111 5 72 1998 2019 538 27*
120 1998 15012 979 7 120 1998 1510 84 6 120 1998 941 210 22*
168 1998 20620 1615 8* 168 1998 464 25 5 168 1998 339 26 8
240 1998 22586 1736 8* 240 1998 321 32 10 240 1998 230 26 11

Expression of DEGs during graft formation

In this step we were interested in genes that are specifically expressed in grafted top, grafted bottom and grafted bottom and top. These three gene sets were also determined by baySeq and can be found in the object baySeq_grafting_lists. The aim was now to use these gene sets and find an overlap to previously published datasets of graft formation studies. The procedure is almost the same as in the previous step. We took each of the gene sets from graft formation studies and calculated for each time point the overlap to baySeq clusters of graft specific differentially expressed genes. To test if the overlap is significant we used the same Fisher's exact test as in the previous step and performed after all tests the Benjamini-Yekutieli correction method. After the siginicances are calculated we apply the function barplot_graft_formation to visualize the number of overlapping genes from each publication with the grafting clusters defined by baySeq at each given time point.

data(baySeq_grafting_lists)
data(graft_gene_lists)
geneUniverse <- rownames(exp.data)
intersect_lists <- list()
for(i in seq_along(graft_gene_lists)){
  
  pub_gene_set <- graft_gene_lists[[i]]
  
  count_mat <- matrix(nrow=length(baySeq_grafting_lists), ncol=8) # 8 time points
  pval_mat <- matrix(nrow=length(baySeq_grafting_lists), ncol=8)
  col_names <- c()
  intersect_lists[[i]] <- list()
  
  for(treat in seq_along(baySeq_grafting_lists)){
  
    sub_list <- baySeq_grafting_lists[[treat]]
    cur_col <- 1
    
    merge_gene_ids <- c()
    
    for(j in seq_along(sub_list)){
      
      check_list <- list(sub_list[[j]], pub_gene_set)
      names(check_list) <- c(names(sub_list)[j], names(graft_gene_lists)[i])
      
      inter_stats <- intersect.analysis(gene.list = check_list, geneUniverse = geneUniverse, alternative = "greater")
      
      count_mat[treat, j] <- inter_stats$stat[1]
      pval_mat[treat, j] <- inter_stats$stat[2]
      
      if(treat == 1){
        col_names <- c(col_names, paste0(names(sub_list)[j],"h_",names(graft_gene_lists)[i]))
      }
    }
  }
  
  rownames(count_mat) <- rownames(pval_mat) <- names(baySeq_grafting_lists)
  colnames(count_mat) <- colnames(pval_mat) <- col_names
  
  intersect_lists[[i]][[1]] <- count_mat
  intersect_lists[[i]][[2]] <- pval_mat
  names(intersect_lists[[i]]) <- c("Counts","Pvalues")
}
names(intersect_lists) <- names(graft_gene_lists)

adjusted_p_mat_list <- GraftingScripts::adjust_and_split(lapply(intersect_lists, function(mat_list) mat_list[[2]]), "BY")
names(adjusted_p_mat_list) <- names(graft_gene_lists)

for(i in seq_along(intersect_lists)){
  #par(mar=c(3.1,3.1,3.1,.1), xpd=F, mgp=c(2,1,0), cex=1.5)
  GraftingScripts::barplot_graft_formation(count_mat = intersect_lists[[i]]$`Counts`, 
                                           pval_mat = adjusted_p_mat_list[[i]], main = names(intersect_lists)[i])
  intersect_lists[[i]][[3]] <- adjusted_p_mat_list[[i]]
  names(intersect_lists[[i]])[3] <- "adjusted.Pvalues"
}

GO enrichment

The GO enrichment analysis is the last step of this script. Here we used each by baySeq predefined cluster of grafted top, grafted bottom, and grafted top and bottom gene set at each time point and determined the corresponding GO terms. To test if a certain GO term is significantly enriched in a defined gene set we used the hypergeometric test from the R package GOstats to calculate the corresponding p-value. After all GO terms were tested against a gene set, we used the Bonferroni method to correct against multiple testing. To call a GO term significant enriched the adjusted p-value had to be smaller than 0.05. The results of the GO enrichment analysis can be loaded with data(go_result_lists).

go_result_lists <- list()
go_ontologies <- c("BP","MF","CC")
geneUniverse <- rownames(exp.data)

sapply(seq_along(baySeq_grafting_lists), function(i){
  go_result_lists[[i]] <<- list() 
  sapply(seq_along(baySeq_grafting_lists[[i]]), function(j){
  
    go_result_lists[[i]][[j]] <<- list() 
    if(length(baySeq_grafting_lists[[i]][[j]]) > 0){
      sapply(seq_along(go_ontologies), function(o){
        
          print(paste0(names(baySeq_grafting_lists)[i]," - ",names(baySeq_grafting_lists[[i]])[j],"hrs - ",go_ontologies[o]))
          go_result_lists[[i]][[j]][[o]] <<- GraftingScripts::Compute_GO_Enrichment(geneUniverse = geneUniverse, selectedGeneIds = baySeq_grafting_lists[[i]][[j]], 
                                                                                    pvalueCutoff = 0.05, ontology = go_ontologies[o])
            
      })
      names(go_result_lists[[i]][[j]]) <- go_ontologies
    }
  })
  names(go_result_lists[[i]]) <- names(baySeq_grafting_lists[[i]])
})
names(go_result_lists) <- names(baySeq_grafting_lists)

48 hours after grafting; genes expressed only in grafted bottom samples. Top 20 BP GO terms:

GOBPID Pvalue Pvalue.adjusted OddsRatio ExpCount Count Size Term
GO:0010200 4.77e-70 4.10e-67 3.57e+01 3.65e+00 69 421 response to chitin
GO:0010243 6.10e-69 5.25e-66 3.42e+01 3.78e+00 69 436 response to organonitrogen compound
GO:1901698 7.30e-51 6.28e-48 1.71e+01 6.83e+00 69 789 response to nitrogen compound
GO:0009719 4.14e-41 3.56e-38 9.27e+00 1.67e+01 86 1929 response to endogenous stimulus
GO:0002679 2.72e-37 2.34e-34 4.74e+01 1.05e+00 31 121 respiratory burst involved in defense response
GO:0045730 2.72e-37 2.34e-34 4.74e+01 1.05e+00 31 121 respiratory burst
GO:1901700 3.47e-36 2.98e-33 7.70e+00 2.04e+01 88 2355 response to oxygen-containing compound
GO:0010033 6.52e-34 5.61e-31 6.88e+00 2.47e+01 93 2851 response to organic substance
GO:0006952 6.45e-31 5.54e-28 7.71e+00 1.39e+01 69 1603 defense response
GO:0002376 1.15e-30 9.90e-28 9.53e+00 8.54e+00 56 986 immune system process
GO:0002252 3.44e-29 2.96e-26 1.93e+01 2.42e+00 34 280 immune effector process
GO:0042221 6.25e-26 5.38e-23 5.14e+00 3.26e+01 95 3760 response to chemical
GO:0035556 3.84e-25 3.30e-22 1.28e+01 3.73e+00 36 431 intracellular signal transduction
GO:0006950 1.40e-21 1.21e-18 4.43e+00 3.35e+01 90 3872 response to stress
GO:0009611 9.09e-21 7.81e-18 1.27e+01 2.90e+00 29 335 response to wounding
GO:0007165 1.34e-20 1.15e-17 5.66e+00 1.36e+01 56 1576 signal transduction
GO:0070887 6.89e-20 5.93e-17 5.90e+00 1.16e+01 51 1344 cellular response to chemical stimulus
GO:0050896 1.71e-19 1.47e-16 3.92e+00 5.53e+01 114 6386 response to stimulus
GO:0044700 1.89e-19 1.63e-16 5.31e+00 1.44e+01 56 1668 single organism signaling
GO:0023052 1.94e-19 1.67e-16 5.31e+00 1.45e+01 56 1669 signaling

120 hours after grafting; genes expressed only in grafted top + grafted bottom samples. Top 20 BP GO terms:

GOBPID Pvalue Pvalue.adjusted OddsRatio ExpCount Count Size Term
GO:0044036 8.54e-11 9.86e-08 6.05e+00 4.33e+00 23 309 cell wall macromolecule metabolic process
GO:0048507 5.78e-10 6.67e-07 4.50e+00 7.02e+00 28 501 meristem development
GO:0010067 5.84e-10 6.75e-07 1.44e+02 1.26e-01 6 9 procambium histogenesis
GO:0010065 1.44e-09 1.67e-06 1.08e+02 1.40e-01 6 10 primary meristem tissue development
GO:0010087 3.45e-09 3.99e-06 1.01e+01 1.50e+00 13 107 phloem or xylem histogenesis
GO:0010410 3.87e-09 4.47e-06 6.84e+00 2.82e+00 17 201 hemicellulose metabolic process
GO:0010413 6.43e-09 7.42e-06 7.11e+00 2.55e+00 16 182 glucuronoxylan metabolic process
GO:0045492 6.95e-09 8.03e-06 7.07e+00 2.57e+00 16 183 xylan biosynthetic process
GO:0045491 8.79e-09 1.01e-05 6.95e+00 2.61e+00 16 186 xylan metabolic process
GO:0071554 9.67e-09 1.12e-05 3.27e+00 1.24e+01 36 883 cell wall organization or biogenesis
GO:0044038 1.72e-08 1.99e-05 6.59e+00 2.73e+00 16 195 cell wall macromolecule biosynthetic process
GO:0070592 1.72e-08 1.99e-05 6.59e+00 2.73e+00 16 195 cell wall polysaccharide biosynthetic process
GO:0070589 1.72e-08 1.99e-05 6.59e+00 2.73e+00 16 195 cellular component macromolecule biosynthetic process
GO:0010014 1.98e-08 2.29e-05 7.07e+00 2.40e+00 15 171 meristem initiation
GO:0010383 4.80e-08 5.54e-05 5.68e+00 3.34e+00 17 238 cell wall polysaccharide metabolic process
GO:0042546 1.11e-07 1.28e-04 4.54e+00 4.86e+00 20 347 cell wall biogenesis
GO:0010075 1.41e-07 1.62e-04 6.54e+00 2.40e+00 14 171 regulation of meristem growth
GO:0010051 1.59e-07 1.84e-04 8.99e+00 1.40e+00 11 100 xylem and phloem pattern formation
GO:0007389 1.60e-07 1.85e-04 4.43e+00 4.98e+00 20 355 pattern specification process
GO:0009933 1.72e-07 1.99e-04 5.16e+00 3.64e+00 17 260 meristem structural organization