Preparation of example histological annotations in the Loupe Browser

Next, we will first demonstrate how to create an annotation .csv file from the Loupe Browser. The user will then read in a .csv file containing histological annotations and convert it to the R data.frame format. Histological annotations in our manuscript were performed at ‘per-spot’ level by clinical histopathologists who are expert trained to assess morphology of prostate samples. These detailed annotations permit clinical translation in a spatio-biological context. These annotations are primarily used in our analyses to enable groups of spots to be selected according to specific morphological classification (e.g. tumour versus benign; or epithelial versus stroma). We have included specific details of our histological annotation work-flow in the manuscript (Update with publication link upon publication).

It is important to emphasize, that for our manuscript, histological annotation work was not done in R, but done using the 10X Genomics LoupeBrowser. For this userguide, we used version 5.1.0.

For the sample in the above code chunk, 10X Genomics provides a Breast Cancer .cloupe Loupe Browser file file.

After opening the LoupeBrowser file, click the 3 white dots at upper right, and select “New Category”.

Type in “Histology” and click the green checkbox

Select the Polygonal Selection Tool at the center top.

Next, left click and drag around the 12 spots at the lower right of the section.

Select “Histology” from the dropdown menu, and type in an annotation, in this example “Userguide_12”, and then click “Create new cluster”Userguide_12", and then click the save button to save the annotation.

If this all was done properly, you should see the following:

Finally, export the annotations. Start by selecting the 3 white dots at the upper right, and select “Export Histology” from the dropdown menu.

Select “Exclude Unlabeled”.

Type in “10xBreast_UserguideHistologyAnnotations.csv”, and select “Export”.

Importing Histological Annotation Data

Next, the user will import the histological annotation .csv file containing histological annotations and convert it to the R data.frame format. This was created in the previous step, but we also provide this .csv file in the directory if you wish to skip to this step.

The motivations for annotations is described above and in our manuscript. These annotations are typically used in analyses to subselect or filter for certain histological populations.

# Reading the histological annotation file created above, and appending the string "Breast10X" before each barcode with ImportHistologicalAnnotations()
Userguide_10xBreast_Histology <- ImportHistologicalAnnotations("Breast10X", "./UserGuideFiles/10xBreast_UserguideHistologyAnnotations.csv")

# Showing the first part of the Userguide_10xBreast_Histology dataframe 
head(Userguide_10xBreast_Histology)

Merging histological annotation data with count data

This next step demonstrates functionality of MergingCountAndAnnotationData(). This function requires 3 parameters: a section annotation string (this needs to be identical from the annotation and count files), the annotation data.frame, and the count data.frame)

This step will also apply a QC filter, to remove any barcodes that fall below the recommended QC threshold of a minimum of 500 counts.

# Merging annotations, and counts, and applying QC with MergingCountAndAnnotationData()
Userguide_10xBreast_Joined_Counts <- MergingCountAndAnnotationData("Breast10X",Userguide_10xBreast_Histology, Breast_ENSBMLID_Counts)

# Setting the "Genes" column as the rownames
Userguide_10xBreast_Joined_Counts <- Userguide_10xBreast_Joined_Counts %>% column_to_rownames(var = "Genes")

# Removing the pre-merged count file as it is no longer needed
rm(Breast_ENSBMLID_Counts)

# Showing the first part of the Userguide_10xBreast_Histology dataframe 
head(Userguide_10xBreast_Joined_Counts)

Selecting finalized annotations for export for use in infercnv::run

During pre-processing, you may filter out certain barcodes (either due to selection of certain histological populations, or due to not passing QC). The function FinalAnnotations() compares the processed count file (which has all of the count data to be analyzed) with the original parent histological annotation file, removes any extra annotations not included in the count file, and returns a finalized annotation dataframe for export.

In this example, we will simulate this, by dropping the first barcode from the count file processed above.

#Diplaying the dimensions of the original annotation file: 12 rows, 2 columns
dim(Userguide_10xBreast_Histology)

#Dropping the first barcode count data
Userguide_10xBreast_Joined_Counts <- Userguide_10xBreast_Joined_Counts %>% select(-Breast10X_AATAACGTCGCGCCCA.1)

#Running FinalAnnotations() to return only the barcodes in the count file
FinalAnnotationsForExport <- FinalAnnotations(Userguide_10xBreast_Histology, Userguide_10xBreast_Joined_Counts)

#Diplaying the dimensions of the final annotation: 11 rows, 2 columns
dim(FinalAnnotationsForExport)

Manual selection of output groups from a infercnv run plot for visualization and further analysis

In order to run the next function, the user needs to run infercnv. InferCNV requires 3 inputs: a count file, an annotation file, and a gene loci file. The code will output the first annotation and count file created above. Documentation to create a gene loci file can be found here at the infercnv wiki documentation. To save you time, we provide a file download in the chunk. Author Note: given that the repo is private, this is currently commented out: manually download the siCNV_GeneOrderFile.tsv file from the repo.

# Download the siCNV_GeneOrderFile to the local directory

# Author Note: given that the repo is private, this is currently commented out: manually download the file from the repo
#download.file("https://raw.githubusercontent.com/aerickso/SpatialInferCNV/main/FigureScripts/siCNV_GeneOrderFile.tsv?token=GHSAT0AAAAAABRUPXQSKNJYB6FRBX2GLUGMYR77RCQ", "./siCNV_GeneOrderFile.tsv", mode = "wb")

# Write the Userguide_10xBreast_Joined_Counts.tsv to the local directory
write.table(Userguide_10xBreast_Joined_Counts, "Userguide_10xBreast_Joined_Counts.tsv", sep = "\t")

# Write the Userguide_10xBreastFinalAnnotationsForExport.tsv to the local directory
write.table(FinalAnnotationsForExport, "Userguide_10xBreastFinalAnnotationsForExport.tsv", 
            sep = "\t",
            quote = FALSE, 
            col.names = FALSE, 
            row.names = FALSE)

# Create the inferCNV object from the 3 files, without the reference group and mitochondria 
Userguide_10xBreastCancer_infCNV <- infercnv::CreateInfercnvObject(raw_counts_matrix="./Userguide_10xBreast_Joined_Counts.tsv", 
                                               gene_order_file="./UserGuideFiles/siCNV_GeneOrderFile.tsv",
                                               annotations_file="./Userguide_10xBreastFinalAnnotationsForExport.tsv",
                                               delim="\t",
                                               ref_group_names=NULL, 
                                               chr_exclude = c("chrM")) 

# Run inferCNV
Userguide_10xBreastCancer_infCNV = infercnv::run(Userguide_10xBreastCancer_infCNV, 
                                              cutoff=0.1, #(see infercnv::run documentation)
                                              out_dir="./InferCNVrun_outputs", 
                                              cluster_by_groups=FALSE, #unsupervised analysis
                                              HMM = FALSE, 
                                              denoise=TRUE) #denoising applies noise reduction for the plot

Further documentation on running infercnv can be found here.

Within the created InferCNVrun_outputs subfolder, this is what the resultant infercnv.21_denoised.png looks like:

This is where SpatialInferCNV’s SelectingSubTreeData() function can help: it allows you to identify barcodes associated with certain clusters by manual selection from the dendrogram at left. For this The dendrogram, visualized at left in the above denoised output plot, is also generated in infercnv::run, and can be found at “./InferCNVrun_outputs/infercnv.21_denoised.observations_dendrogram.txt”.

Please note: this next step requires manual interpretation of the images (often done in an image viewer or editor outside of R).

For the purposes of the userguide, lets assume that we have identified signal in certain groups of Visium spots (here termed “clones”). Lets proceed with about extracting the identities of the barcodes for clone 1, 2 and 3 highlighted in boxes the following image (edited in Microsoft Paint).

library(ape)
library(phylogram)

#Use read.dendrogram() to import the dendogram file
BreastCancer10x_for_clustering <- read.dendrogram(file = "./UserGuideFiles/infercnv.21_denoised.observations_dendrogram.txt")

#Convert to a phylo object using as.phylo()
BreastCancer10x_for_clustering_phylo <- as.phylo(BreastCancer10x_for_clustering)

#Use subtrees() to enable further interaction with the dendrogram
my.subtrees = subtrees(BreastCancer10x_for_clustering_phylo)  # subtrees() to subset

#Output an image to visualize all of the dengdrogram nodes 
png("BreastCancer10x_forclustering_phylo.png",width=10000,height=2500, res = 300)
plot(BreastCancer10x_for_clustering_phylo,show.tip.label = FALSE)
nodelabels(text=1:BreastCancer10x_for_clustering_phylo$Nnode,node=1:BreastCancer10x_for_clustering_phylo$Nnode+Ntip(BreastCancer10x_for_clustering_phylo))
dev.off()

If we manually inspect the image above, we can identify the dendrogram nodes associated with the clone groups of interest.

We then select these nodes data using SelectingSubTreeData(), with the my.subtrees object created above, and the numerical node number in the output image. The output of SelectingSubTreeData() is a data.frame, with barcodes in the first column, and the node number (identified above), in the second column.

#Selecting Node 9
Node9 <- SelectingSubTreeData(my.subtrees, 9)

#Selecting Node 5
Node5 <- SelectingSubTreeData(my.subtrees, 5)

#Selecting Node 3
Node3 <- SelectingSubTreeData(my.subtrees, 3)

#Merging All Nodes Together
MergedNodes <- rbind(Node5, Node9)
MergedNodes <- rbind(MergedNodes, Node3)

#Note, since the "additional spot" for Clone 3 did not have a node, identify it by joining the original annotations to the MergedNodes dataframe
MergedNodes <- full_join(MergedNodes, FinalAnnotationsForExport)

#The additionally joined barcode does not have a node identity, so manually change it to "Clone 3" (in line with the image above)
MergedNodes$Node <- ifelse(is.na(MergedNodes$Node), "Clone 3", MergedNodes$Node)

#Then drop the Histology column
MergedNodes <- MergedNodes %>% select(-Histology)

#This then renames "Node" to "Clone"
MergedNodes$Node <- ifelse(MergedNodes$Node == "Node_9", "Clone1",
                     ifelse(MergedNodes$Node == "Node_5" , "Clone2",
                     ifelse(MergedNodes$Node == "Node_3" , "Clone2", MergedNodes$Node)))

#Show the first part of the finalized dataframe for this step
head(MergedNodes)

The MergedNodes data.frame object can be used, for example, to output a new annotation file, to be imported into inferCNV, and then run inferCNV with cluster_by_groups=TRUE, to output a plot, clustered by clones.

The identified clones now can also be visualized in the LoupeBrowser for analysis or generation of figure images. To do so, they need to be exported as a .csv file, and the “section name” that we appended to the barcodes needs to be removed.

#Copy the MergedNodes dataframe to a new dataframe called "ForLoupeBrowser"
ForLoupeBrowser <- MergedNodes

#Use the function substr() to remove the first 9 characters in the barcode column (removes "Breast10X_")
ForLoupeBrowser$Barcode <- trimws(substr(ForLoupeBrowser$Barcode, 11, 100))

#Replace the "." in the current barcode dataframe to "-" (required for LoupeBrowser)
ForLoupeBrowser$Barcode <- gsub("\\.", "\\-", ForLoupeBrowser$Barcode)

#Write the dataframe to csv
write.csv(ForLoupeBrowser, "Userguide_BreastCancer_Clones_ForLoupeBrowser.csv", row.names = FALSE)

Open the .cloupe file that was downloaded earlier. Select the 3 white dots at the right, and select “Import Categories…”

Select the Userguide_BreastCancer_Clones_ForLoupeBrowser.csv created above, and click “Import Categories…”

The spots should be now visualized with the clone identity labeled.

Working with 1k array spatial array ST

SpatialInferCNV functions also work with 1K array ST data from the original version of spatial transcriptomics. This is functionally similar to the above ImportCountData(), but instead ImportOriginalSTCountData imports count data in .tsv format, appends a section identifier to the barcodes, and returns an R data.frame.

We will download data from this publicly available BreastCancer 1k array spatial transcriptomics dataset.

#Downloading the files from the published Mendeley
download.file("https://data.mendeley.com/public-files/datasets/29ntw7sh4r/files/cbf4607b-1b8f-4b23-9a7e-27f94bdfd0fc/file_downloaded", "./BC23209_C1_stdata.tsv.gz", mode = "wb")

#Importing counts from the downloaded .gz file
BC23209_C1_counts <- ImportOriginalSTCountData("BC23209_C1", "./BC23209_C1_stdata.tsv.gz")

#Displaying first part of the dataframe
head(BC23209_C1_counts)

Import histological selections

The STPipeline outputs a spot_data-selection.tsv file, which is used to identify which spots actually had tissue covering them. This step is automatically performed in the SpaceRanger pipeline, so is not needed for Visium data.

The Mendeley data repository also provides the spot selection file.

#Downloading the files from the published Mendeley
download.file("https://data.mendeley.com/public-files/datasets/29ntw7sh4r/files/01568b4e-950d-4bcd-9a01-cd509e557625/file_downloaded", "./spots_BC23209_C1.csv.gz", mode = "wb")

#Importing the downloaded spot selection .csv file
CSV <- read.csv("./spots_BC23209_C1.csv.gz")

#The file is slightly different than the STPipeline spot_data-selection_aligned.tsv file structure, which has explicit "x" and "y"
CSV <- CSV %>% separate(X.1, into = c('x', 'y'), sep = "x")

#Writing the Dataframe to .tsv (to mirror the actual STPipeline output)
write.table(CSV, "spots_BC23209_C1.tsv", 
                              sep = "\t",
                              quote = FALSE,
                              row.names = FALSE)
rm(CSV)

#Read in the selected barcodes and appending the section name
BC23209_C1_SelectedBarcodes <- ImportHistologicalOriginalSTSelections("BC23209_C1", "./spots_BC23209_C1.tsv")
#rewrite for X and Y

#Displaying first part of the dataframe
head(BC23209_C1_SelectedBarcodes)

Merging 1k array histological annotation data with count data

In this chunk, the user will subselect 10 spots at random from the selected barcodes, append an annotation for these 10 spots, and then run OriginalST_MergingCountAndAnnotationData() to select count data only from these 10 spots. This function also passes a QC check. This function is similar to MergingCountAndAnnotationData() ran above on Visium Data

#Select the first 10 barcodes and rename the first column back to barcode
BC23209_C1_SelectedBarcodes <- as.data.frame(BC23209_C1_SelectedBarcodes[1:10,])
names(BC23209_C1_SelectedBarcodes)[1] <- "Barcode"

#Add a second column to the selected barcodes dataframe called "Histology", which will title BC23209_Random10
BC23209_C1_SelectedBarcodes <- BC23209_C1_SelectedBarcodes %>% 
                  select(Barcode) %>%
                  mutate(Histology = "BC23209_Random10")

#Finally, then subselecting the count data for only the BC23209_Random10 spots
BC23209_C1_JoinedCounts <- OriginalST_MergingCountAndAnnotationData(BC23209_C1_SelectedBarcodes, BC23209_C1_counts)

# Setting the "Genes" column as the rownames
BC23209_C1_JoinedCounts <- BC23209_C1_JoinedCounts %>% column_to_rownames(var = "Genes")

#Displaying first part of this dataframe
head(BC23209_C1_JoinedCounts)

Spatial Visualization of Original ST Data - Using publicly available Breast Cancer Data

SpatialInferCNV provides functions to visualize the number of genes with an inferred copy number alteration (as presented in Figure 1 of our manuscript). The user needs to generate the 1k array, spot level inferCNV HMM results.

In order to run the next function, the user needs to run infercnv. InferCNV requires 3 inputs: a count file, an annotation file, and a gene loci file. We will output the first annotation and count file created above. Documentation to create a gene loci file can be found here at the infercnv wiki documentation. To save you time, we provide a file download in the chunk. Author Note: given that the repo is private, this is currently commented out: manually download the siCNV_GeneOrderFile.tsv file from the repo.

# Download the siCNV_GeneOrderFile to the local directory

# Author Note: given that the repo is private, this is currently commented out: manually download the file from the repo
#download.file("https://raw.githubusercontent.com/aerickso/SpatialInferCNV/main/FigureScripts/siCNV_GeneOrderFile.tsv?token=GHSAT0AAAAAABRUPXQSKNJYB6FRBX2GLUGMYR77RCQ", "./siCNV_GeneOrderFile.tsv", mode = "wb")

# Write the Userguide_10xBreast_Joined_Counts.tsv to the local directory
write.table(BC23209_C1_JoinedCounts, "BC23209_C1_JoinedCounts.tsv", sep = "\t")

#One spot was filtered from the count array by the QC step, so extract the final barcodes for
FinalAnnotationsForExport1karray <- FinalAnnotations(BC23209_C1_SelectedBarcodes, BC23209_C1_JoinedCounts)

# Write the Userguide_10xBreastFinalAnnotationsForExport.tsv to the local directory
write.table(FinalAnnotationsForExport1karray, "BC23209_C1_SelectedBarcodes.tsv", 
            sep = "\t",
            quote = FALSE, 
            col.names = FALSE, 
            row.names = FALSE)

# Create the inferCNV object from the 3 files, without a reference group nor running mitochondrial genes
Userguide_1kArrayBreastCancer_infCNV <- infercnv::CreateInfercnvObject(raw_counts_matrix="./BC23209_C1_JoinedCounts.tsv", 
                                               gene_order_file="./UserGuideFiles/siCNV_GeneOrderFile.tsv", 
                                               annotations_file="./BC23209_C1_SelectedBarcodes.tsv", 
                                               delim="\t", 
                                               ref_group_names=NULL, 
                                               chr_exclude = c("chrM"))

# Run inferCNV
Userguide_1kArrayBreastCancer_infCNV = infercnv::run(Userguide_1kArrayBreastCancer_infCNV,
                                              cutoff=0.1, 
                                              out_dir="./InferCNVrun_outputs_1karray", 
                                              cluster_by_groups=TRUE, #We want to run the analysis by spot, so need this parameter = TRUE
                                              denoise=TRUE, 
                                              HMM=TRUE, #We need the HMM data for the visualization
                                              analysis_mode = "cells", #We want to run the analysis by spot (inferCNV was designed for scRNAseq)
                                              HMM_report_by = "cell") #we need to set the HMM report to output by cell

Having then run inferCNV and obtained HMM outputs, the user then needs to further format the data prior to using the functions for 1k array visualization

#initializing additional libraries
library(reshape2)
library(grid)
library(svglite)

#We want a dataframe of all possible barcodes, not just the 10 random sections with HMM data
AllBarcodes <- ImportHistologicalOriginalSTSelections("BC23209_C1", "./spots_BC23209_C1.tsv")
names(AllBarcodes)[1] <- "Barcode"
AllBarcodes$Histology <- "AllBarcodes"

#WGSize <-  3049315783

#We also want a dataframe of all possible barcodes in a 1k array assay. We use the original barcode file provided by SpatialTranscriptomicsResearch
L2_Barcodes <- read.table("https://github.com/SpatialTranscriptomicsResearch/st_pipeline/raw/master/ids/1000L2_barcodes.txt", sep = "\t")
names(L2_Barcodes)[1] <- "Barcode"
names(L2_Barcodes)[2] <- "X"
names(L2_Barcodes)[3] <- "Y"
L2_Barcodes$ModfiedBarcode <- paste0(L2_Barcodes$X, "x", L2_Barcodes$Y)
L2_Barcodes <- L2_Barcodes %>% 
      select(ModfiedBarcode)
names(L2_Barcodes)[1] <- "Barcode"


#Then use the first threshold to optimize signal-to-noise for spatial visualization across all genes with an inferred CNV, all genes genes with an inferred CNV (for this dataset, 35%).

#Uncomment, and use this line if you want to use the file you generated above
#CNV_Genes_Userguide <- read.delim("./InferCNVrun_outputs_1karray/17_HMM_predHMMi6.hmm_mode-cells.pred_cnv_genes.dat")

#Use this comment if you want to use the file that we have provided 
CNV_Genes_Userguide <- read.delim("./UserGuideFiles/17_HMM_predHMMi6.hmm_mode-cells.pred_cnv_genes.dat")

#Adding section name as a column
CNV_Genes_Userguide <- CNV_Genes_Userguide %>% mutate(section = substr(cell_group_name, 1, 10))

#Displaying the first part of this dataframe 
head(CNV_Genes_Userguide)

Extracting sectionwise, spot level HMM data

ExtractSectionWise() is designed to extract specific section, spot level HMM data: while not relevant for the userguide as we are using data from only 1 section, this is useful in the modified output HMM data.frames containing multiple sections data.

Having now obtained the necessary inputs, you can run ExtractSectionWise().

BC23209_C1_Sectionwise_CNVsGenes_Counted <- ExtractSectionWise("BC23209_C1", #The section name
                                                               CNV_Genes_Userguide, #The HMM dataframe file that we modified to add "section" above
                                                               AllBarcodes, #All barcodes that had tissue (imported above)
                                                               0.45) #A threshold, to filter only for HMMs that are detected in 45 percent of spots with HMM data

Creating a visualization matrix for the HMM 1k array data

Using the output from ExtractSectionWise(), you can then create a matrix for spatially plotting the 1k array data with Output_PGA_Visualization_MatrixGreyNA().

PGA_Matrix <- Output_PGA_Visualization_MatrixGreyNA("BC23209_C1", #The section name
                                                    BC23209_C1_Sectionwise_CNVsGenes_Counted, #The output from ExtractSectionWise() above
                                                    L2_Barcodes) #All barcodes in the 1k array, used to generate the matrix for the plot

Spatially plot the number of genes with an inferredCNV, per spot from 1k array data

Using the output from Output_PGA_Visualization_MatrixGreyNA(), we can finally spatially plot the 1k array data with Plot_PGA_Visualization_Matrix().

#Defining Max Value, this is more relevant when working with multiple sections
BC23209_C1Max <- as.numeric(max(BC23209_C1_Sectionwise_CNVsGenes_Counted$PercentageGenomeAltered))
MaxVal<- max(BC23209_C1Max)

#Plotting the visualization
Plot_PGA_Visualization_Matrix("BC23209_C1", PGA_Matrix, MaxVal)

#Saving the plot with ggsave
ggsave(paste0("BC23209_C1", "_", "PGA_SpatialVisualization_", Sys.Date(), ".png"),
         plot = last_plot() + labs(x=NULL, y=NULL),
         device = NULL,
         path = NULL,
         scale = 1,
         width = NA,
         height = NA,
         dpi = 300,
         limitsize = TRUE)

The spots should be now visualized with the number of genes with an inferredCNV. Remember that we are using data from <10 spots, so the majority of the image should be grey.

This ends the user guide, complementary information is embedded in the provided scripts.

Spatial InferCNV Userguide

Andrew Erickson, Ludvig Larsson

Userguide

Initialize SpatialInferCNV

Downloading and importing publicly available Visium Datasets (10x Genomics)