Identifying Conserved Cell-Type-Specific Transcriptional Programs in Hydra and Clytia
This document covers our comparative analysis of the Hydra and Clytia single-cell atlases. This analysis entailed using reciprocal principal component analysis to align the two datasets, quantifying overall transcriptional similarities between cell types using mutual nearest neighbor analysis, and identifying orthologs with conserved expression patterns.
Identifying Conserved Cell-Type-Specific Transcriptional Programs in Hydra and ClytiaAligning the Clytia and Hydra Single-Cell AtlasesQuantifying Transcriptional Similarities Between Clytia and Hydra Cell TypesIdentifying Orthologous Genes with Conserved Expression PatternsFiles Associated with This Document
Aligning the Clytia and Hydra Single-Cell Atlases
The process of aligning single cell datasets from different species uses the same approach as correcting for inter-sample batch effects. Essentially, we simply treat inter-species differences as (dramatic) batch effects. For this to be possible, the two datasets need to use the same gene IDs. In order to unambiguously link orthologous genes between Hydra and Clytia, we limited our alignment analysis to only genes that had a one-to-one orthology between the two species, as determined through our Orthofinder analysis. We then simply converted the gene IDs for the Clytia data to their Hydra equivalents.
In the following code chunk, we identify one-to-one ortholog pairs, import and perform initial filtering on the Clytia and Hydra gene expression matrices, convert the Clytia gene IDs to their Hydra equivalents, and concatenate all of the separate samples (15 Hydra libraries and one Clytia library) into a list of Seurat objects.
(snippet from 01_alignment/aepCl.R)
x
library(Seurat)library(tidyverse)library(rstudioapi)library(glmGamPoi)library(plotly)library(RColorBrewer)library(patchwork)library(plyr)library(SeuratDisk)library(cowplot)library(fields)library(BiocNeighbors)library(networkD3)library(reticulate)
#utility to convert transcript ID to gene IDt2g <- function(x){ vapply(x, function(y) gsub('HVAEP1_T(\\d+)[.]\\d','HVAEP1-G\\1',y),"")}
setwd(dirname(getActiveDocumentContext()$path))
#import list of non-doublet Hydra cell IDsnonDub <- read.delim('../dropSeqMapping/finalNonDub.txt')[,1]
#import orthofinder results to identify one-to-one orthologs between clytia and hydraorthos <- read.delim('../../Orthofinder/Results_Sep15_1/Phylogenetic_Hierarchical_Orthogroups/N14.tsv')
orthos <- orthos[,c('C_hemisphaerica','H_vulgarisAEP')]
#drop genes without orthologs in the other speciesorthos <- orthos[orthos$C_hemisphaerica != '' & orthos$H_vulgarisAEP != '',]
#drop genes that don't have one-to-one orthologyorthos <- orthos[!(grepl(',',orthos$C_hemisphaerica) | grepl(',',orthos$H_vulgarisAEP)),]
orthos$H_vulgarisAEP <- gsub('HVAEP1_T(\\d+)[.].*','HVAEP1_G\\1',orthos$H_vulgarisAEP)
#import data and convert#get list of hydra count matrix files (include path in name)readMats <- list.files(path = '../dropSeqMapping', pattern = 'dge.txt.gz', recursive = T, full.names = T)
#read in each individual read matrix, generate a separate seurat object for each, and do some initial filteringinAep <- lapply(readMats, function(x) { y <- read.delim(x,stringsAsFactors = F,header = T, row.names = 1) #get batch name from parent folder of read matrix libName <- gsub('[.]+\\/dropSeqMapping\\/(D\\d+[^\\/]+)\\/.*','\\1',x) #apply batch names to cell barcodes to prevent redundant barcodes colnames(y) <- paste0(colnames(y),'-',libName) #drop doublet cells y <- y[,colnames(y) %in% nonDub] #only keep genes that have one-to-one orthology with clytia genes y <- y[rownames(y) %in% t2g(orthos$H_vulgarisAEP),] #initialize seurat object tmpDS <- CreateSeuratObject(counts = y, project = libName, min.cells = 3, min.features = 200) #perform preliminary filtering subset(tmpDS, subset = nFeature_RNA > 300 & nFeature_RNA < 7500 & nCount_RNA > 500 & nCount_RNA < 75000)})
#import raw count matrix for remapped clytia datacl <- Read10X_h5('../cl/remap/raw_feature_bc_matrix.h5')
#drop genes without one-to-one orthology with Hydra genescl <- cl[rownames(cl) %in% orthos$C_hemisphaerica,]
#convert clytia gene names to their hydra equivalentsrownames(cl) <- mapvalues(rownames(cl),from = orthos$C_hemisphaerica, to = orthos$H_vulgarisAEP, warn_missing = F)
#import the original annotated clytia datasetclPP <- readRDS('../cl/remap/annotatedCl.rds')
#remove any clytia cells that weren't in the original publicationcl <- cl[,colnames(cl) %in% rownames(clPP@meta.data)]
rm(clPP)
cl <- CreateSeuratObject(counts = cl, project = 'clDS', min.cells = 3, min.features = 200)
cl <- subset(cl, nFeature_RNA < 4000 & nCount_RNA > 500 & nCount_RNA < 100000)
#combine clytia object with list of hydra objectsaepCl <- inAep
aepCl <- append(aepCl,cl)
rm(inAep)We then performed standard dataset integration using reciprocal PCA analysis as implemented through Seurat. Following integration we performed PCA on the aligned dataset.
(snippet from 01_alignment/aepCl.R)
x
#normalize data using SCTransformaepCl <- lapply(aepCl, FUN = SCTransform, method = "glmGamPoi")
#integrate data (both across batches and across species) using reciprocal pcafeatures <- SelectIntegrationFeatures(aepCl,assay=rep('SCT',length(aepCl)))
aepCl <- PrepSCTIntegration(object.list = aepCl, anchor.features = features,assay=rep('SCT',length(aepCl)))
aepCl <- lapply(X = aepCl, FUN = RunPCA, features = features, verbose = F, assay = 'SCT', npcs = 50)
dsAnchors <- FindIntegrationAnchors(object.list = aepCl, normalization.method = "SCT", anchor.features = features, dims = 1:50, reduction = "rpca", k.anchor = 20)
rm(aepCl)gc()
aepCl.int <- IntegrateData(anchorset = dsAnchors, normalization.method = "SCT", dims = 1:50)
rm(dsAnchors)gc()
#determining the dimensionality of the dataaepCl.int <- RunPCA(aepCl.int, npcs = 80,verbose=F)
ElbowPlot(aepCl.int,ndims = 80)
We then performed Louvain clustering and UMAP dimensional reduction using 30 PCs. We then visualized the clustering results using the UMAP plot.
(snippet from 01_alignment/aepCl.R)
x
#generate UMAP projection and perform louvain clusteringndUse=1:30aepCl.int <- RunUMAP(aepCl.int, reduction = "pca", dims = ndUse, min.dist = 0.18,spread = 0.2, seed.use = 4321)aepCl.int <- FindNeighbors(aepCl.int, reduction = "pca", dims = ndUse)aepCl.int <- FindClusters(aepCl.int, resolution = 0.4, graph.name = 'integrated_snn')
DimPlot(aepCl.int) + NoAxes() + NoLegend()
We also split the plot by species to see the contribution of each species to different cell populations
(snippet from 01_alignment/aepCl.R)
x
aepCl.int@meta.data$species <- grepl('^D',aepCl.int@meta.data$orig.ident)
aepCl.int@meta.data$species[aepCl.int@meta.data$species == T] <- 'AEP'
aepCl.int@meta.data$species[aepCl.int@meta.data$species == 'FALSE'] <- 'Cl'
DimPlot(aepCl.int,group.by = 'species') + NoAxes()
(snippet from 01_alignment/aepCl.R)
x
DimPlot(aepCl.int,split.by = 'species') + NoAxes() + NoLegend()
To determine if our integrated dataset was properly aligning homologous cell types in the two species, we imported the cell annotations from the unintegrated atlases for each species. Specifically, we used the cell type annotations from our analysis of the AEP-mapped Hydra atlas (described in 05_hydraAtlasReMap.md) and the annotations from the original Clytia atlas publication.
(snippet from 01_alignment/aepCl.R)
x
####Bring in curated cell identity annotations####clPP <- readRDS('../cl/remap/annotatedCl.rds')
ds <- readRDS('../dropSeqMapping/nonDubLabeledSeurat.rds')
annotTab <- rbind(data.frame(ID=colnames(clPP), curatedIdent = as.character(clPP@meta.data$annos),curatedIdentSub=as.character(clPP@meta.data$annosSub)), data.frame(ID=colnames(ds), curatedIdent = as.character(ds@meta.data$curatedIdent),curatedIdentSub = as.character(ds@meta.data$curatedIdent)))
#only keep cells that made it into the integrated objectannotTab <- annotTab[annotTab$ID %in% colnames(aepCl.int),]
#reorder cells so they are the same order as the integrated metadata tablerownames(annotTab) <- annotTab$IDannotTab <- annotTab[colnames(aepCl.int),]
aepCl.int@meta.data$curatedIdent <- annotTab$curatedIdent
aepCl.int@meta.data$curatedIdentSub <- annotTab$curatedIdentSub
#colors for cluster plotclustPal <- c("#45adec","#cbee2f","#d565f5","#85e538","#f452cd", "#57e958","#997cfb","#eadc25","#528efb","#abd533", "#d07be8","#65c737","#a585f0","#54e170","#e27cd1", "#4eb13a","#f2699f","#3ee698","#f76748","#3be6ea", "#f57417","#3295e9","#d0cc36","#6e8de9","#def469", "#7d9af7","#95ca42","#47a2f7","#e8b027","#2499d7", "#e8802b","#9d94e5","#a1f078","#db93d2","#83ce5a", "#f2686c","#51dbb4","#e77f4a","#56c88d","#e87c86", "#6bc565","#e8886e","#82e596","#c98b23","#a6f4b4", "#c78b3c","#5ab267","#eaaa43","#99cd86","#df9f64", "#ade486","#e8aa59","#91a824","#d6e58f","#b39420", "#cce679","#b0953a","#efda63","#8eac57","#d8b644", "#94af45","#dcbd61","#aaa126","#ceca7b","#c9ca4d", "#a4a154","#e8de7a","#afa746")
set.seed(5321)clustPal <- clustPal[sample(length(clustPal))]
#plot using broad cluster labels from clytia paperDimPlot(aepCl.int,group.by = 'curatedIdent',split.by = 'species', label = T, repel = T, cols = clustPal,pt.size=0.5) + NoAxes() + NoLegend()
(snippet from 01_alignment/aepCl.R)
x
#plot using more granular cluster labels from clytia paper#(a bit crowded)DimPlot(aepCl.int,group.by = 'curatedIdentSub',split.by = 'species', label = T, repel = F, cols = clustPal,pt.size=0.5) + NoAxes() + NoLegend()
(snippet from 01_alignment/aepCl.R)
x
#a very course level of clustering for the integrated data just to group together broad cell typesaepCl.int <- FindClusters(aepCl.int, resolution = 0.05, graph.name = 'integrated_snn')DimPlot(aepCl.int,group.by = 'integrated_snn_res.0.05',split.by = 'species', repel = F,pt.size=0.5) + NoAxes() + NoLegend()
#save object for latersaveRDS(aepCl.int,'aepClInt.rds')
Based on these plots, our integration was successful, as homologous cell types colocalize in the UMAP and are clustered together by the Louvain algorithm.
Quantifying Transcriptional Similarities Between Clytia and Hydra Cell Types
Although UMAPs can give an indication of similarity between different cells, it is not a quantitative way of assessing similarity. To assess similarity quantitatively, we adopted the alignment score metric proposed by Tarashansky et al. (2021).
To generate this score, the 30 nearest inter-species neighbors for each cell in the aligned principal component space are identified. Then, for a given cell type in one species, the number of nearest cross-species neighbors that were from each cell type in the other species is tabulated. The final alignment score is defined as the portion of total cross-species pairs for a cell type in one species that belonged to a particular cell type in another species. Thus, a higher alignment score indicates that the two cell types being compared shared a higher number of neighbors in the aligned PC space.
Because we were interested in identifying possible similarities between neuronal subtypes in Clytia and Hydra, and because the whole-animal version of the Clytia atlas did not resolve individual neuronal subtypes, we incorporated cell type labels from the Clytia neuronal subclustering analysis before performing the alignment quantification.
To do this we first had to download the neuronal sub clustering object (available here) and convert it to a Seurat object.
(snippet from 01_alignment/aepCl.R)
x
#convert neuronal subclustering object from clytia paper into seurat object#only needs to be run onceuse_condaenv("rScanpy", required = T)
sc <- import("scanpy")
adata <- sc$read_h5ad('../cl/remap/neuron_subpops_fs.h5ad')
exprs <- t(adata$X)colnames(exprs) <- adata$obs_names$to_list()rownames(exprs) <- adata$var_names$to_list()# Create the Seurat objectseurat <- CreateSeuratObject(exprs)# Set the expression assayseurat <- SetAssayData(seurat, "data", exprs)# Add observation metadataseurat <- AddMetaData(seurat, adata$obs)# Add embeddingembedding <- adata$obsm["X_umap"]rownames(embedding) <- adata$obs_names$to_list()colnames(embedding) <- c("umap_1", "umap_2")seurat[["umap"]] <- CreateDimReducObject(embedding, key = "umap_")
#save converted object to save time latersaveRDS(seurat,'annotatedNeuroCl.rds')
clN <- seuratrm(seurat)
clN <- readRDS('annotatedNeuroCl.rds')
DimPlot(clN, group.by = 'louvain_neur', label = T) + NoAxes() + NoLegend()
We then updated the cell labels for Clytia neurons using the higher resolution cluster labels from the neuron sub-clustering
(snippet from 01_alignment/aepCl.R)
x
#for neuronal cells, replace the cell labels from the#from the whole-animal analysis with the labels#from the neuronal subcluster analysisannotTab <- annotTab[!(annotTab$ID %in% colnames(clN)),]
annotTab <- rbind(data.frame(ID=colnames(clN), curatedIdent = 'Neuron',curatedIdentSub=paste0('neuro_',as.character(clN$louvain_neur))),annotTab)
annotTab <- annotTab[annotTab$ID %in% colnames(aepCl.int),]
rownames(annotTab) <- annotTab$ID
annotTab <- annotTab[colnames(aepCl.int),]
aepCl.int@meta.data$curatedIdent <- annotTab$curatedIdent
aepCl.int@meta.data$curatedIdentSub <- annotTab$curatedIdentSubWe then extracted the principal component cell scores from the integrated seurat object and separated them by species. We then performed a mutual nearest neighbor analysis to identify the Clytia cells that was most similar to each Hydra cell. We then converted these cell IDs to cell type labels and tabulated the number of cells from each Clytia cell type associated with each Hydra cell type. Finally, we used these values to generate a Sankey diagram showing all alignment scores greater than 0.05 (indicating that greater than 5% of all nearest neighbors for a Hydra cell type were made up of a given Clytia cell type).
(snippet from 01_alignment/aepCl.R)
x
#extract cell score for principal components from seurat objectpcSpace <- aepCl.int@reductions$pca@cell.embeddings[,1:80]
#split cell PC scores by speciespcSpace.a <- pcSpace[aepCl.int$species == 'AEP',]
pcSpace.c <- pcSpace[aepCl.int$species == 'Cl',]
#specify the number of nearest neighbors to use for MNNkUse <- 30
#find nearest neighbors for all cells in aligned PC spacemnnRes <- findMutualNN(pcSpace.a,pcSpace.c, k1 = kUse)
#get the cross-species neighbor pair cell IDsmnnRes <- data.frame(AEP=mnnRes[[1]],Cl=mnnRes[[2]])
mnnRes$AEP <- rownames(pcSpace.a)[mnnRes$AEP]mnnRes$Cl <- rownames(pcSpace.c)[mnnRes$Cl]
mnnRes.orig <- mnnRes
#convert neighbor pair cell IDs to cell type labelsmnnRes$Aepclust <- mapvalues(mnnRes$AEP,from = rownames(aepCl.int@meta.data), to = aepCl.int$curatedIdentSub, warn_missing = F)mnnRes$ClClust <- mapvalues(mnnRes$Cl,from = rownames(aepCl.int@meta.data), to = aepCl.int$curatedIdentSub, warn_missing = F)
write.csv(mnnRes,'mnnRes.csv')
#split pair assignments by AEP cellmnnRes.a <- split(mnnRes, mnnRes$Aepclust)
#for a given hydra cell type, calculate the portion of#it's neighbors made up by each clytia cell typemnnRes.a <- lapply(mnnRes.a, function(x) { res <- as.data.frame(table(x$ClClust)) res$score <- (res$Freq/length(unique(x$AEP)))/kUse return(res)})
#set a minimum score cutoff of 0.05mnnRes.a <- lapply(mnnRes.a,function(x){ x[x$score >= 0.05,]})
#drop cell types without any meaningful alignmentmnnRes.a <- mnnRes.a[sapply(mnnRes.a,nrow) > 0]
#add a column with the hydra cell ID#(to keep track of stuff when the list is collapsed)mnnRes.a <- lapply(1:length(mnnRes.a), function(x){ df <- mnnRes.a[[x]] df$aepClust <- names(mnnRes.a)[x] return(df)})
mnnRes.a <- do.call(rbind,mnnRes.a)
#these objects are formatted in a particular way to fit the input requirements#of the sankey plotting funciton#this object simply lists all of the cell types in the analysis (both clytia and hydra)nodeDf <- data.frame(name=unique(c(as.character(mnnRes.a$Var1),mnnRes.a$aepClust)))
#need to manually set the order so that the resulting plot has a logical ordernodeDf <- nodeDf[c(1,3,4,5,2,12,13,14,11,16,15,17,22,10,6,20,8,7,9,21,18,19,23, 24,27,26,28,30,29,46,25,34,36,35,37,39,38,47,41,33,43,48,44,31,32,45,49,40,42,50),,drop=F]
#this object specifies the cell type pairs and the alignment score between themlinkDf <- data.frame(source = mnnRes.a$Var1, target=mnnRes.a$aepClust, value=mnnRes.a$score)
#these encode the index for the cell type in the nodeDF object (zero indexed)linkDf$IDsource <- match(linkDf$source, nodeDf$name)-1 linkDf$IDtarget <- match(linkDf$target, nodeDf$name)-1
#plot using plotly sankey plotfig <- plot_ly( type = "sankey", orientation = "h", node = list( label = nodeDf$name, color = clustPal[1:48], pad = 20, thickness = 20, line = list( color = "black", width = 0.5 ) ), link = list( source = linkDf$IDsource, target = linkDf$IDtarget, value = linkDf$value ))
fig
As an alternative approach to visualize and evaluate the degree of transcriptional similarity between Hydra and Clytia cells, we also calculated a distance metric that captured how different the overall transcriptional profiles of cells from one species were when compared to their most similar transcriptional neighbors from the other species. To generate this metric, we calculated the average distance to the 30 nearest cross-species nearest neighbors in aligned principal component space for each cell. We then plotted these values on the cross-species UMAP to visually link these distance values to cell type annotations.
x
#get distance between mnn pairs
#make distance matrixpcSpace.a <- pcSpace[aepCl.int$species == 'AEP',1:80]
pcSpace.c <- pcSpace[aepCl.int$species == 'Cl',1:80]
cDist <- rdist(pcSpace.a,pcSpace.c)
rownames(cDist) <- rownames(pcSpace.a)
colnames(cDist) <- rownames(pcSpace.c)
#for every hydra cell, get the average distance for the 30 closest clytia cellsaepDist <- vapply(rownames(pcSpace.a), function(x){ dists <- cDist[x,] dists <- dists[order(dists)] dists <- mean(dists[1:30]) return(dists)}, numeric(1))
#for every clytia cell, get the average distance for the 30 closest hydra cellsclDist <- vapply(rownames(pcSpace.c), function(x){ dists <- cDist[,x] dists <- dists[order(dists)] dists <- mean(dists[1:30]) return(dists)}, numeric(1))
dist.df <- data.frame(cellID = c(rownames(pcSpace.a),rownames(pcSpace.c)), dist = c(aepDist,clDist))
dist.df <- dist.df[match(colnames(aepCl.int),dist.df$cellID),]
dist.df$curatedIdent <- aepCl.int$curatedIdentSub
write.csv(dist.df, file = 'crossSpecDist.csv',row.names = F)
aepCl.int$dist <- dist.df$dist
FeaturePlot(aepCl.int, 'dist',order = T)
xxxxxxxxxxFeaturePlot(aepCl.int, 'dist', order = T, split.by = 'species', pt.size = 0.4)
We also summarized the distance metric results using a boxplot that grouped distance scores by cell type in each species. The first boxplot presents distances for Hydra cell types.
xxxxxxxxxxplotDf <- aepCl.int@meta.data
plotDf$curatedIdentSub <- factor(plotDf$curatedIdentSub)
plotFacts <- levels(plotDf$curatedIdentSub)
factOrder <- c("I_ISC","I_FemGC","I_MaleGC", "I_GlProgen","I_ZymoGl","I_GranGl", "I_SpumMucGl","I_Neuro","I_Ec1/5N", "I_Ec1N","I_Ec2N","I_Ec3N", "I_Ec4N","I_En1N","I_En2N", "I_En3N","I_EarlyNem","I_DesmoNB", "I_StenoNB","I_IsoNB","I_DesmoNC", "I_StenoNC","I_IsoNC","Ec_BasalDisk", "Ec_Peduncle","Ec_BodyCol/SC","Ec_Head", "Ec_Tentacle","En_Foot","En_BodyCol/SC", "En_Head","En_Tentacle", "i-Cells","Very Early Oocytes","Small Oocytes", "Medium Oocytes","Gland Cells-A","Gland Cells-B", "Gland Cells-C","Gland Cells-D","Gland Cells-E", "neuro_0","neuro_1","neuro_2", "neuro_3","neuro_4","neuro_5", "neuro_6","neuro_7","neuro_8", "neuro_9","neuro_10","neuro_11", "neuro_12","neuro_13","neuro_14", "Nematocyte Precursors","Early Nematoblasts","Mid Nematoblasts", "Late Nematoblasts","Differentiating Nematocytes","Terminal Differentiating Nematocytes", "Mature Nematocytes","Exumbrella Epidermis","Manubrium Epidermis", "Gonad Epidermis","Tentacle Epidermis","Tentacle GFP Cells", "Radial Smooth Muscles","Striated Muscle of Subumbrella","Striated Muscle of Velum", "Endodermal Plate","GastroDigestive-A","GastroDigestive-B", "GastroDigestive-C","GastroDigestive-D","GastroDigestive-E", "GastroDigestive-F","Tentacle Bulb Distal Gastroderm")
plotDf$curatedIdentSub <- factor(plotDf$curatedIdentSub, levels = factOrder)
ggplot(plotDf[plotDf$species == 'AEP',],aes(x=curatedIdentSub,y=dist,fill=curatedIdentSub)) + geom_boxplot() + #geom_jitter(height = 0, width = 0.1, size = 0.1,alpha=0.4) + theme_bw() + theme(legend.position="none") + coord_flip()
The second boxplot presents distances for Clytia cell types.
xxxxxxxxxxggplot(plotDf[plotDf$species == 'Cl',],aes(x=curatedIdentSub,y=dist,fill=curatedIdentSub)) + geom_boxplot() + #geom_jitter(height = 0, width = 0.1, size = 0.1,alpha=0.4) + theme_bw() + theme(legend.position="none") + coord_flip()
Identifying Orthologous Genes with Conserved Expression Patterns
The alignment score we calculated above could be used to holistically examine transcriptional similarities, but it did not give us access to the genes whose expression patterns were conserved in Hydra and Clytia. To identify genes with conserved expression patterns, we needed a way to correlate the expression patterns in Hydra and Clytia cells.
To do this, we used a very high resolution Louvain clustering analysis to generate ad-hoc 'pseudo-cells'. These pseudo-cells grouped together cells (regardless of species) that were close together in the aligned principal component space. We calculated average gene expression values in each species for each pseudo-cell and then, by matching pseudo-cell labels from each species, identified genes with similar expression patterns.
We started by first importing the Seurat object containing the aligned Clytia and Hydra data and performing a high resolution Louvain clustering analysis. This generated a total of 132 'pseudo-cell' clusters.
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxxlibrary(Seurat)library(tidyverse)library(rstudioapi)library(glmGamPoi)library(plotly)library(RColorBrewer)library(patchwork)library(plyr)library(SeuratDisk)library(cowplot)library(gplots)library(viridis)
#utility to convert transcript ID to gene IDt2g <- function(x){ vapply(x, function(y) gsub('HVAEP1_T(\\d+)[.]\\d','HVAEP1-G\\1',y),"")}
setwd(dirname(getActiveDocumentContext()$path))
#import integrated hydra and clytia seurat objectaepCl.int <- readRDS('aepClInt.rds')
#perform a high resolution louvain clustering analysis#this generates ad-hoc 'pseudo-cells' that group #together small groups of cells with similar expression#(total of 132 clusters at this resolution)aepCl.int <- FindClusters(aepCl.int, resolution = 10, graph.name = 'integrated_snn')
#pseudo-cell UMAPDimPlot(aepCl.int) + NoAxes() + NoLegend()
We then extracted the normalized read matrix from the Seurat object, split cells both by species and pseudo-cell ID, and calculated average read counts. We then calculated correlation scores for pseudo-cell expression when comparing the two species. We used the resulting scores as a readout of how conserved the cell-type-specificity was for a pair of Clytia and Hydra orthologs.
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#extract normalized expression dataexpDat <- t(as.matrix(aepCl.int@assays$SCT@data))
#get the average expression of each gene, aggregated by both pseudo-cell ID#and speciesexpDat.cl <- aggregate(expDat,list(aepCl.int$seurat_clusters,aepCl.int$species),mean)
#only keep the clusters that had cells from both speciesexpDat.cl <- expDat.cl[expDat.cl$Group.1 %in% expDat.cl[expDat.cl$Group.2 == 'Cl','Group.1'],]expDat.cl <- expDat.cl[expDat.cl$Group.1 %in% expDat.cl[expDat.cl$Group.2 == 'AEP','Group.1'],]
write.csv(expDat.cl,file='pcExpression.csv',row.names = F)
#split data by species (make an AEP df and a Clytia DF)expDat.cl.list <- split(expDat.cl,expDat.cl$Group.2)
#droup group.1 and group.2 columnsexpDat.cl.list <- lapply(expDat.cl.list,function(x){ x <- x[,c(-1,-2)] return(x)})
#calculate the correlation in pseudocell expression patterns across#the two speciescorRes <- lapply(1:ncol(expDat.cl.list[[1]]),function(x){ cor(expDat.cl.list[[1]][,x],expDat.cl.list[[2]][,x],method = 'pearson')})
#generate DF of correlation scores for all genescorRes <- do.call(c,corRes)
corRes <- data.frame(id = colnames(expDat.cl.list[[1]]),cor=corRes)
#set NAs to 0corRes[is.na(corRes$cor),'cor'] <- 0
#order by correlation scorecorRes <- corRes[order(-corRes$cor),]
write.csv(corRes,file='geneCoreRes.csv',row.names = F)For identifying conserved expression patterns, we used high resolution clusters, which allowed us to perform a relatively more refined comparison of expression in the two species. To visualize these conserved expression patterns, we opted to use a gene-by-cell-type heatmap. In order to avoid having a high number of columns (cell types) in the heatmap, thus improving overall readability, we used a lower resolution clustering to summarize the data. We generated descriptive labels for these lower resolution clusters based on the identities of both the Hydra and Clytia cells making up each cluster, although we showed some preference for labels based on Hydra annotations as Hydra cell types are generally better characterized.
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#after identifying genes with similar expression patterns#convey the data using a heatmap based on a lower resolution clustering#otherwise it's just an insane number of columns
#import more descriptive names for lower resolution clusterscorPlotLabels <- read.csv('corPlotClustsLables.csv',header=F)
#make umap with descriptive names for clusters used for heatmap plotsaepCl.int$renameClust <- mapvalues(aepCl.int$integrated_snn_res.0.4,from = corPlotLabels$V1, corPlotLabels$V2, warn_missing = F)
DimPlot(aepCl.int,group.by = 'renameClust',label = T, repel = T) + NoLegend() + NoAxes()
Using these broader clusters we again averaged gene expression by clusters (split by species) to generate the values that would populate the heatmap plot. We restricted these heatmaps to only those ortholog pairs with high correlation values (correlation score > 0.65)
We first plotted the Hydra data:
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#generate averaged expression data for lower resolution clusteringheatObj <- aggregate(expDat,list(aepCl.int$integrated_snn_res.0.4,aepCl.int$species),mean)
heatObj <- heatObj[heatObj$Group.1 %in% heatObj[heatObj$Group.2 == 'Cl','Group.1'],]heatObj <- heatObj[heatObj$Group.1 %in% heatObj[heatObj$Group.2 == 'AEP','Group.1'],]
#bring in the descriptive cluster namesheatObj$Group.1 <- mapvalues(heatObj$Group.1, from=corPlotLabels$V1, to=corPlotLabels[,2], warn_missing = F)
#split by speciesheatObj.list <- split(heatObj,heatObj$Group.2)
#drop group.1 and group.2 columnsheatObj.list <- lapply(heatObj.list,function(x){ rownames(x) <- x[,1] x <- x[,c(-1,-2)] return(x)})
#get data for AEP heatmapheatObj.1 <- heatObj.list[[1]]
#only plot genes with high correlation scoresheatObj.1 <- heatObj.1[,colnames(heatObj.1) %in% corRes[corRes$cor > 0.65,'id']]
#import table that specifies a reasonably logical way of ordering#the heatmap columns (cell clusters)corPloOrder <- read.csv('corPlotClustsOrder.csv',header=F)
heatObj.1 <- heatObj.1[corPloOrder$V2,]
pdf('heat1.pdf',width = 8, height = 40)#plot heatmap, save the output, which specifies how rows are ordered#which is needed to keep the clytia heatmap consistentdgm <- heatmap.2(t(heatObj.1), scale = 'row', dendrogram = 'none', Colv = F, col = viridis(30), trace='none', key = F, keysize = 0.1, margins = c(10,6), distfun = function(x) as.dist(1-cor(t(x))), hclustfun = function(x) hclust(x, method="average"))dev.off()
write.csv(t(heatObj.1),file='crossSpecMatHydra.csv',row.names = F)
We then plotted the Clytia data
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#extract clytia heatmap dataheatObj.2 <- heatObj.list[[2]]
#keep only genes with good correlation scoresheatObj.2 <- as.data.frame(t(heatObj.2[,colnames(heatObj.2) %in% corRes[corRes$cor > 0.65,'id']]))
#reorder data to have the same order as the AEP matrixheatObj.2 <- heatObj.2[rev(dgm$rowInd),corPloOrder$V2]
#change gene names back to original clytia ortho nameorthos <- read.delim('../../Orthofinder/Results_Sep15_1/Phylogenetic_Hierarchical_Orthogroups/N14.tsv')
orthos <- orthos[,c('C_hemisphaerica','H_vulgarisAEP')]
orthos <- orthos[orthos$C_hemisphaerica != '' & orthos$H_vulgarisAEP != '',]
orthos <- orthos[!(grepl(',',orthos$C_hemisphaerica) | grepl(',',orthos$H_vulgarisAEP)),]
orthos$H_vulgarisAEP <- gsub('HVAEP1_T(\\d+)[.].*','HVAEP1-G\\1',orthos$H_vulgarisAEP)
rownames(heatObj.2) <- mapvalues(rownames(heatObj.2), from = orthos$H_vulgarisAEP, to = orthos$C_hemisphaerica,warn_missing = F)
#plot clytia heatmappdf('heat2.pdf',width = 8, height = 40)heatmap.2(as.matrix(heatObj.2), Rowv = F, Colv = F, scale = 'row', dendrogram = 'none', col = magma(30), trace='none', key = F, keysize = 0.1, margins = c(10,6), distfun = function(x) as.dist(1-cor(t(x))), hclustfun = function(x) hclust(x, method="average"))dev.off()
write.csv(t(heatObj.2),file='crossSpecMatClytia.csv',row.names = F)
We also generated a table including correlation scores for all high scoring (correlation score > 0.65) ortholog pairs as well as functional annotations (putative vertebrate orthologs, protein domains, etc.)
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#generate table of correlation scores for export (only high scoring genes)
#bring in functional annotation information to make browsing results easier to interpretannots <- read.csv('../../Orthofinder/HVAEP1_annotation.csv')
annots$H_vulgarisAEP <- t2g(annots$H_vulgarisAEP)
corRes.annot <- merge(corRes,annots,by.x = 'id', by.y = 'H_vulgarisAEP', all.x = T)
corRes.annot <- corRes.annot[order(-corRes.annot$cor),]
corRes.annot <- corRes.annot[corRes.annot$cor > 0.65,]
corRes.annot$clOrtho <- mapvalues(corRes.annot$id, from = orthos$H_vulgarisAEP, to = orthos$C_hemisphaerica, warn_missing = F)write.csv(corRes.annot,'crossSpecExpCor.csv',row.names = F)Finally, we generated another set of heat maps that included only putative transcription factors (predictions were based on interproscan results as described in 03_aepGenomeAnnotation.md). First for AEP:
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxx#subset correlation heatmap to just look at trancription factorstfList <- read.delim('../../Genome_annotation/functionalAnnotation/tfIDs.txt',header=F)[,1,drop=T]
tfList <- gsub('_','-',tfList)
heatObj.1 <- heatObj.list[[1]]
heatObj.1 <- heatObj.1[,colnames(heatObj.1) %in% corRes.annot[corRes.annot$cor > 0.65,'id']]
heatObj.1 <- heatObj.1[,colnames(heatObj.1) %in% tfList]
heatObj.1 <- heatObj.1[corPloOrder$V2,]
pdf('tfHeat1.pdf',width = 8, height = 10)dgm <- heatmap.2(t(heatObj.1), scale = 'row', Colv = F, dendrogram = 'none', col = viridis(30), trace='none', key = F, keysize = 0.1, margins = c(5,10), distfun = function(x) as.dist(1-cor(t(x))), hclustfun = function(x) hclust(x, method="average"))dev.off()
write.csv(t(heatObj.1),file='crossSpecTfMatHydra.csv',row.names = F)
Then for Clytia:
(03_expressionConservation/aepCl_Cor.R)
xxxxxxxxxxheatObj.2 <- heatObj.list[[2]]
heatObj.2 <- as.data.frame(t(heatObj.2[,colnames(heatObj.2) %in% corRes.annot[corRes.annot$cor > 0.65,'id']]))
heatObj.2 <- heatObj.2[rownames(heatObj.2) %in% tfList,]
heatObj.2 <- heatObj.2[rev(dgm$rowInd),corPloOrder$V2]
pdf('tfHeat2.pdf',width = 8, height = 10)heatmap.2(as.matrix(heatObj.2), Rowv = F, Colv = F, scale = 'row', dendrogram = 'none', col = magma(30), trace='none', key = F, keysize = 0.1, margins = c(5,10))dev.off()
write.csv(t(heatObj.2),file='crossSpecTfMatClytia.csv',row.names = F)
Files Associated with This Document
x12_crossSpeciesAtlasAlignment/├── 01_alignment│ ├── aepClInt.rdsR binary file containing the aligned Hydra and Clytia single-cell atlasesformatted as a Seurat object.│ └── aepCl.RR script that aligns the Clytia and Hydra single-cell atlases and quantifiesand visualizes transcriptional similarities across homologous cell types usinga mutual nearest neighbor-based alignment score.├── 02_similarityQuant│ ├── aepClIntNeuroLab.rdsR binary file containing a modified version of aepClInt.rds that includesneuronal subtype annotations from the original Clytia atlas publication.│ ├── crossSpecDist.csvTable that provides the average distance to the 30 nearest cross-speciesnearest neighbors for every cell in the cross-species atlas. The tablealso includes the barcode and the curated cell type annotation for eachcell.│ ├── mnnRes.csvTable that lists the mutual nearest cross-species neighbors for all cellsin the aligned Clytia and Hydra atlas.│ ├── mnnSankeyLink.csvTable used for generating a sankey plot visualizing the alignment scores generatedby systematically comparing Clytia and Hydra cell types in the aligned cross-speciessingle-cell atlas. The table includes all cross-species cell type pairs that had analignment score ≥ 0.05. In addition to specifying the two cell types, it also includesthe alignment score under the 'value' column.│ └── mnnSankeyNode.csvTable used for generating a sankey plot visualizing the alignment scores generatedby systematically comparing Clytia and Hydra cell types in the aligned cross-speciessingle-cell atlas. This table simply lists all cell types included in the alignmentscore sankey plot (required by the function that made the plot).└── 03_expressionConservation├── aepCl_Cor.RR script that identifies genes with conserved expression patterns in the cross-species atlas and visualizes the results using heatmaps.├── corPlotClustsLables.csvTable containing descriptive names for the Louvain clusters in the cross-speciesatlas used for generating the heatmaps visualizing conserved expression patterns.├── corPlotClustsOrder.csvTable that specifies the order that the clusters listed in corPlotClustsLables.csvwill be displayed in the heatmaps visualizing conserved expression patterns.├── crossSpecExpCor.csvTable listing all genes with a cross-species expression correlation score above 0.65.The table also includes functional annotation data for each gene.├── crossSpecMatClytia.csvTable containing the values used to generate the heatmap depicting the expressionpatterns of all Clytia genes with a cross-species expression correlation scoreabove 0.65.├── crossSpecMatHydra.csvTable containing the values used to generate the heatmap depicting the expressionpatterns of all Hydra genes with a cross-species expression correlation scoreabove 0.65.├── crossSpecTfMatClytia.csvTable containing the values used to generate the heatmap depicting the expressionpatterns of all Clytia transcription factors with a cross-species expressioncorrelation score above 0.65.├── crossSpecTfMatHydra.csvTable containing the values used to generate the heatmap depicting the expressionpatterns of all Hydra transcription factors with a cross-species expressioncorrelation score above 0.65.├── geneCoreRes.csvTable containing the cross-species expression correlation scores for all genes inthe cross-species atlas.└── pcCellIDs.csvTable containing the pseudo-cell assignments for all cells in the cross-speciesatlas.