#This file contains the R code used to generate gene signatures, tables and figures.
#The TCGA and array express data sets were downloaded, as indicated in the paper, and assembled into lists
#containing the expression data and clinicopathological variables.
#The lists can contain the following variables (depending on data set):
# $data (expression data), $key (key containing microarray probe ID and gene symbol), $devent (OS, deceased=1, alive=0), $dtime (follow-up time for OS)
# $revent (recurrence or NTE=new tumor event, yes=1, no=0), $rtime (follow-up time recurrence), $age (age at diagnosis), $stage (stage at diagnosis),
# $ER (positive/negative=1/0 for estrogen receptor), $node (lymph nodes yes=1, no=0), $tum2 (tumor size >2cm=1)
#The data sets (lists) are not included as supplemental files 
#xclus.RData contains the genes in the signatures identified in Winslow et al, Breast Cancer Res, 2015;10:e0130300

#insert path of datasets
path = ""

#load signatures from Breast Cancer Res paper
load(file=paste(path,"xclus.RData",sep=""))

#name TCGA cancer set files
TBRCA = "TBRCA150918set.RData"
TCOAD = "TCOAD150917set.RData"
THNSC = "THNSC150918set.RData"
TKIRC = "TKIRC150918set.RData"
TLUAD = "TLUAD150917set.RData"
TLUSC = "TLUSC150917set.RData"

#name TCGA normal breast tissue data file
TBRCAnorm = "mRNAnormdata.RData"

#name set files for the microarray sets analyzed
GEOD21653 = "GEOD21653set.RData"
EMTAB365 = "EMTAB365set.RData"

#name library where the GSE6744 data are stored
GSE66744 = "GSE66744\\"


#load TCGA BRCA set
load(file=paste(path,TBRCA,sep=""))

#name TCGA datasets
sets<-c(TBRCA,TCOAD,THNSC,
        TKIRC,TLUAD,TLUSC)
setnames<-c("BRCA","COAD","HNSC","KIRC","LUAD","LUSC")


#get signatures from previous paper
endo<-c(xclus[[4]],xclus[[5]])
ecmg<-c(xclus[[1]],xclus[[2]])

#remove genes with low variance prior to expansion
mRNAtumvarselect<-apply(set$data,1,var)>0.2
mRNAtumvar<-set$data[mRNAtumvarselect,]

#expand potential ECM genes
ng<-NULL
for (i in 1:length(ecmg)){
  for (j in 1:nrow(mRNAtumvar)){
    if (cor(mRNAtumvar[ecmg[i],],mRNAtumvar[j,])>0.84) ng<-c(ng,rownames(mRNAtumvar)[j])
  }
}
ng<-unique(ng)
ecmset<-ng
#ecmset = Table S3

#expand potential endo genes
ng<-NULL
for (i in 1:length(endo)){
  for (j in 1:nrow(mRNAtumvar)){
    if (cor(mRNAtumvar[endo[i],],mRNAtumvar[j,])>0.84) ng<-c(ng,rownames(mRNAtumvar)[j])
  }
}
ng<-unique(ng)
endoset<-ng
#endoset = Table S4

#functions for obtaining compact clusters
cormean<-function(x){
  return(apply(x,1,function (y) (sum(y)-1)/(length(y)-1)))
}
findcors<-function (x) {
  y<-cbind(cormean(cor(t(set$data[x,tumind1]))),cormean(cor(t(set$data[x,tumind2]))),
           cormean(cor(t(set$data[x,tumind3]))),cormean(cor(t(set$data[x,tumind4]))))
  return(y)
}

stp<-0.85
findsetsim<-function (x,stp=0.85){
  lims<-seq(0.5,stp,0.05)
  for(i in 1:length(lims)){
    y<-cormean(cor(t(set$data[x,])))
    yy<-(y>lims[i])
    x<-x[yy]
    y<-cormean(cor(t(set$data[x,])))
    yy<-(y>lims[i])
    x<-x[yy]
  }
  return(x)
}



#obtain ECM clusters
ecmlist<-as.list(NULL)
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  ecmlist[[i]]<-findsetsim(ecmset,0.85)
}
#ecmlist contains ECM clusters for the different cancer forms (Table 1)
#necm contains genes that are present in ECM cluster in all cancer forms
necm<-names(table(unlist(ecmlist)))[table(unlist(ecmlist))==6]

#obtain endo clusters 
endolist<-as.list(NULL)
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  endolist[[i]]<-findsetsim(endoset,0.85)
}
nend<-names(table(unlist(endolist)))[table(unlist(endolist))==6]
#endolist contains endo clusters for the different cancer forms (Table 2)
#nend contains genes that are present in endo cluster in all cancer forms

#make scatterplots ECM-endo (Fig 1)
subnames = c("A","B","C","D","E","F")
tiff(file=paste(path,"fig1AE.tif",sep=""),width=2800,height=2000,res=300,compression="lzw")
par(mfrow=c(2,3),mar=c(5,4,4,1))
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  plot(colMeans(set$data[necm,]),colMeans(set$data[nend,]),main=setnames[i],
       xlab="ECM metagene",ylab="Endothelial metagene",yaxt="n",cex.lab=1.3,cex.axis=1.2)
  axis(2,at=seq(4,12,2),labels=seq(4,12,2),las=1,cex.axis=1.2)
  legend("topleft",legend=signif(cor(colMeans(set$data[necm,]),colMeans(set$data[nend,])),3),bty="n",cex=1.2)
  mtext(subnames[i],side=3,line=1,adj=0,cex=2,outer=F)
}
dev.off()

#survival function, right-censored
library(survival)
survrc<-function(time,event,rc){
  event[time>rc]<-0
  time[time>rc]<-rc
  output<-Surv(time,event)
}

#uni- and multivariate Cox of ECM and endo signatures, data stored in variable y (Table 3A-B)
options(scipen=10)
y<-NULL
i = 1
load(paste(path,sets[i],sep=""))
ecuv<-summary(coxph(survrc(set$rtime,set$revent,1825)~scale(colMeans(set$data[necm,]))+
                      strata(set$age)+strata(set$stage)))$coefficients[,c(2,5)]
enuv<-summary(coxph(survrc(set$rtime,set$revent,1825)~scale(colMeans(set$data[nend,]))+
                      strata(set$age)+strata(set$stage)))$coefficients[,c(2,5)]
ecmv<-summary(coxph(survrc(set$rtime,set$revent,1825)~scale(colMeans(set$data[necm,]))+scale(colMeans(set$data[nend,]))+
                      strata(set$age)+strata(set$stage)))$coefficients[1,c(2,5)]
enmv<-summary(coxph(survrc(set$rtime,set$revent,1825)~scale(colMeans(set$data[necm,]))+scale(colMeans(set$data[nend,]))+
                      strata(set$age)+strata(set$stage)))$coefficients[2,c(2,5)]
y<-rbind(y,c(ecuv,enuv),c(ecmv,enmv))
for (i in 2:length(sets)){
  load(paste(path,sets[i],sep=""))
  ecuv<-summary(coxph(survrc(set$dtime,set$devent,1825)~scale(colMeans(set$data[necm,]))+
                        strata(set$age)+strata(set$stage)))$coefficients[,c(2,5)]
  enuv<-summary(coxph(survrc(set$dtime,set$devent,1825)~scale(colMeans(set$data[nend,]))+
                        strata(set$age)+strata(set$stage)))$coefficients[,c(2,5)]
  ecmv<-summary(coxph(survrc(set$dtime,set$devent,1825)~scale(colMeans(set$data[necm,]))+scale(colMeans(set$data[nend,]))+
                        strata(set$age)+strata(set$stage)))$coefficients[1,c(2,5)]
  enmv<-summary(coxph(survrc(set$dtime,set$devent,1825)~scale(colMeans(set$data[necm,]))+scale(colMeans(set$data[nend,]))+
                        strata(set$age)+strata(set$stage)))$coefficients[2,c(2,5)]
  y<-rbind(y,c(ecuv,enuv),c(ecmv,enmv))
}
rownames(y)<-rep("",nrow(y))
for(i in 1:length(setnames)){
  rownames(y)[2*i-1]<-paste(setnames[i],"univariate")
  rownames(y)[2*i]<-paste(setnames[i],"multivariate")
}
colnames(y)<-c("ECM HR","ECM p-val","Endo HR","Endo p-val")

#generate three decimals from data in y
yy = NULL
for (i in 1:length(setnames)){
  yy = rbind(yy,y[grep(setnames[i],rownames(y)),][1,1:2])
}
yy = round(yy,3)
yy = NULL
for (i in 1:length(setnames)){
  yy = rbind(yy,y[grep(setnames[i],rownames(y)),][2,1:2])
}
yy = round(yy,3)
yy = NULL
for (i in 1:length(setnames)){
  yy = rbind(yy,y[grep(setnames[i],rownames(y)),][1,3:4])
}
yy = round(yy,3)
yy = NULL
for (i in 1:length(setnames)){
  yy = rbind(yy,y[grep(setnames[i],rownames(y)),][2,3:4])
}
yy = round(yy,3)

#multivariate Cox, ECM, endo, ER, node, size (TCGA BRCA) Table 3C
load(file=paste(path,TBRCA,sep=""))
n = set$stage<4
n[is.na(n)] = F
coxph(survrc(set$rtime,set$revent,1825)~scale(colMeans(set$data[necm,]))+scale(colMeans(set$data[nend,]))+set$ER+set$tum2+set$node,subset=n)


#boxplot ECM sign i +/-node, tum size, +/- ER (Figure 2)
load(file=paste(path,TBRCA,sep=""))
tiff(file=paste(path,"fig2.tif",sep=""),width=2200,height=2200,res=300,compression="lzw")
par(fig=c(0,0.33,0.5,1))
par(mar=c(4,4,2,1))
par(cex=0.8)
y<-colMeans(set$data[necm,])
boxplot(colMeans(set$data[necm,])~set$ER,col=c(2,4),las=1,ylab="ECM metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("ER neg","ER pos"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[necm,])~set$ER)$p.value,3),sep=""),adj=0)
mtext("A",side=2,line=1,at=22.5,cex=2,las=1,outer=F)
par(fig=c(0.33,0.67,0.5,1),new=T)
boxplot(colMeans(set$data[necm,])~set$node,col=c(4,2),las=1,ylab="ECM metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("node neg","node pos"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[necm,])~set$node)$p.value,3),sep=""),adj=0)
mtext("B",side=2,line=1,at=22.5,cex=2,las=1,outer=F)
par(fig=c(0.67,1,0.5,1),new=T)
boxplot(colMeans(set$data[necm,])~set$tum2,col=c(4,2),las=1,ylab="ECM metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("size<2cm","size>2cm"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[necm,])~set$tum2)$p.value,3),sep=""),adj=0)
mtext("C",side=2,line=1,at=22.5,cex=2,las=1,outer=F)

#boxplot endo sign i +/-node, tum size, +/- ER (Figure)
par(fig=c(0,0.33,0,0.5),new=T)
par(mar=c(4,4,1,1))
par(cex=0.8)
y<-colMeans(set$data[nend,])
boxplot(colMeans(set$data[nend,])~set$ER,col=c(2,4),las=1,ylab="Endothelial metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("ER neg","ER pos"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[nend,])~set$ER)$p.value,3),sep=""),adj=0)
mtext("D",side=2,line=1,at=13.2,cex=2,las=1,outer=F)
par(fig=c(0.33,0.67,0,0.5),new=T)
boxplot(colMeans(set$data[nend,])~set$node,col=c(4,2),las=1,ylab="Endothelial metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("node neg","node pos"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[nend,])~set$node)$p.value,3),sep=""),adj=0)
mtext("E",side=2,line=1,at=13.2,cex=2,las=1,outer=F)
par(fig=c(0.67,1,0,0.5),new=T)
boxplot(colMeans(set$data[nend,])~set$tum2,col=c(4,2),las=1,ylab="Endothelial metagene",xaxt="n",ylim=c(min(y)*0.95,max(y)*1.1))
axis(1,at=1:2,labels=c("size<2cm","size>2cm"),tick=F)
text(0.5,1.09*max(y),labels=paste("p=",signif(t.test(colMeans(set$data[nend,])~set$tum2)$p.value,3),sep=""),adj=0)
mtext("F",side=2,line=1,at=13.2,cex=2,las=1,outer=F)
dev.off()


#read scores of TCGA BRCA tumors (stroma histology type)
load(file=paste(path,"stromtype.RData",sep=""))
load(file=paste(path,TBRCA,sep=""))

#generate figure of ECM signature vs stroma scores (Figure)
tiff(file=paste(path,"fig3.tif",sep=""),width=2200,height=1500,res=300,compression="lzw")
par(fig=c(0,0.22,0,1))
par(mar=c(2.5,4,1,1))
boxplot(colMeans(set$data[necm,TCGAscore$ID])~stromtype,las=1,xaxt="n",col=c(4,2),
        ylab="ECM metagene")
axis(1,at=c(1:2),tick=F,labels=c("S","M"),line=-1)
title(xlab="stroma type",line=1)
mtext("A",side=2,line=2.5,at=20.7,cex=2,las=1,outer=F)
par(fig=c(0.22,0.5,0,1),new=T)
boxplot(colMeans(set$data[necm,TCGAscore$ID])~TCGAscore$stroma,las=1,xaxt="n",col=c("gray80","gray70","gray60"),
        ylab="ECM metagene")
axis(1,at=c(1:3),tick=F,labels=1:3,line=-1)
title(xlab="stroma score",line=1)
mtext("B",side=2,line=2.5,at=20.7,cex=2,las=1,outer=F)
par(fig=c(0.5,1,0,1),new=T)
boxplot(colMeans(set$data[necm,TCGAscore$ID])~TCGAscore$stroma+stromtype,las=1,xaxt="n",
        col=c("blue2","blue3","blue4","red2","red3","red4"),
        ylab="ECM metagene")
axis(1,at=c(1:6),tick=F,labels=c("S1","S2","S3","M1","M2","M3"),line=-1)
title(xlab="stroma type+score",line=1)
mtext("C",side=2,line=2.5,at=20.7,cex=2,las=1,outer=F)
dev.off()

#identify genes that are prognostic in multivariate model (stratify for node and ER)
#top pairs in topres
load(file=paste(path,TBRCA,sep=""))
ng<-ecmset
gi<-NULL
gj<-NULL
res<-NULL
n = set$stage<4
n[is.na(n)] = F
for (i in 1:(length(ng)-1)){
  for (j in (i+1):length(ng)){
    cx<-summary(coxph(survrc(set$rtime,set$revent,1825)~set$data[ng[i],]+set$data[ng[j],]+strata(set$node)+strata(set$ER)+strata(set$tum2),subset=n))
    gi<-c(gi,ng[i])
    gj<-c(gj,ng[j])
    res<-rbind(res,c(cx$coefficients[1,c(2,5)],cx$coefficients[2,c(2,5)],cx$logtest[3]))
  }
}
options(scipen=10)
res[,c(1,3)]<-formatC(round(res[,c(1,3)],2),format="f",digits=2)
res[,c(2,4)]<-formatC(round(as.numeric(res[,c(2,4)]),6),format="f",digits=6)
res[,5]<-formatC(round(as.numeric(res[,5]),5),format="f",digits=5)
resfin<-data.frame(cbind(as.character(gi),as.character(gj),res),stringsAsFactors=F)
resfin<-resfin[order(resfin[,7]),]
resfin[resfin[,3]>1,]<-resfin[resfin[,3]>1,c(2,1,5,6,3,4,7)]
colnames(resfin)<-c("gene 1","gene 2","HR gene 1","p-val gene 1","HR gene 2","p-val gene 2","p-val likelihood")
topres<-resfin[1:100,]

#genes that cause a lower HR in ecmn
gcox<-c(topres[topres[,3]<1,1],topres[topres[,5]<1,2])
table(gcox)
ecmn<-names(table(gcox))[table(gcox)>5]

#genes that cause a higher HR in ecmp
gcox<-c(topres[topres[,3]>1,1],topres[topres[,5]>1,2])
table(gcox)
ecmp<-names(table(gcox))[table(gcox)>5]


#generate HR plot of different cancer datasets
sets<-c(TBRCA,TCOAD,THNSC,
        TKIRC,TLUAD,TLUSC,GEOD21653,EMTAB365)

setnames<-c("BRCA","COAD","HNSC","KIRC","LUAD","LUSC","GEOD21653","EMTAB365")

#generate HR plot of different cancer datasets (Fig 4)

y<-NULL
yuni<-NULL
for(i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  x12<-unlist(lapply(ecmn,function(x) which(set$key[,2]==x)))
  x13<-unlist(lapply(ecmp,function(x) which(set$key[,2]==x)))
  x12<-scale(colMeans(set$data[set$key[x12,1],]))
  x13<-scale(colMeans(set$data[set$key[x13,1],]))
  es<-sum(as.integer(set$ER),na.rm=T)
  ns<-sum(as.integer(set$node),na.rm=T)
  if(i==1){
    n = set$stage<4
    n[is.na(n)] = F
    cx<-summary(coxph(Surv(set$rtime,set$revent)~x12+x13+set$node+set$ER,subset=n))
    cx1<-summary(coxph(Surv(set$rtime,set$revent)~x12+set$node+set$ER,subset=n))
    cx2<-summary(coxph(Surv(set$rtime,set$revent)~x13+set$node+set$ER,subset=n))
  }
  if(i==7|i==8){
    cx<-summary(coxph(Surv(set$time,set$event)~x12+x13+set$node+set$ER))
    cx1<-summary(coxph(Surv(set$time,set$event)~x12+set$node+set$ER))
    cx2<-summary(coxph(Surv(set$time,set$event)~x13+set$node+set$ER))
  }
  if(i>1&i<7){
    n = rep(T,length(set$stage))
    cx<-summary(coxph(Surv(set$dtime,set$devent)~x12+x13+strata(set$stage)+strata(set$age),subset=n))
    cx1<-summary(coxph(Surv(set$dtime,set$devent)~x12+strata(set$stage)+strata(set$age),subset=n))
    cx2<-summary(coxph(Surv(set$dtime,set$devent)~x13+strata(set$stage)+strata(set$age),subset=n))
  }
  y<-rbind(y,c(cx$conf.int[1,c(1,3,4)],cx$conf.int[2,c(1,3,4)]))
  yuni<-rbind(yuni,c(cx1$conf.int[1,c(1,3,4)],cx2$conf.int[1,c(1,3,4)]))
}
y<-y[nrow(y):1,]
yuni<-yuni[nrow(yuni):1,]

tiff(file=paste(path,"fig4.tif",sep=""),width=2000,height=1600,res=300,compression="lzw")
plot(-1,-1,ylim=c(0,nrow(y)+1),log="x",xlim=c(0.1,10),yaxt="n",xlab="Hazard ratio",ylab="",cex=0.7)
abline(v=1)
for(i in 1:nrow(y)){
  lines(y[i,2:3],c(i-5/6,i-5/6),lty=1,col=4,lwd=2)
  lines(y[i,5:6],c(i-4/6,i-4/6),lty=1,col=2,lwd=2)
  lines(yuni[i,2:3],c(i-2/6,i-2/6),lty=1,col="blue4",lwd=2)
  lines(yuni[i,5:6],c(i-1/6,i-1/6),lty=1,col="red4",lwd=2)
  abline(h=i,lty=2)
  text(0.1,9-i-1/3,labels=setnames[i],cex=1,adj=0)
}
lines(c(0.1,0.15),rep(nrow(y)+0.8,2),lty=1,lwd=2,col="blue4")
lines(c(0.15,0.225),rep(nrow(y)+0.8,2),lty=1,lwd=2,col="red4")
lines(c(0.1,0.15),rep(nrow(y)+0.4,2),lty=1,lwd=2,col=4)
lines(c(0.15,0.225),rep(nrow(y)+0.4,2),lty=1,lwd=2,col=2)
text(0.24,nrow(y)+0.85,labels="univariate",pos=4)
text(0.24,nrow(y)+0.45,labels="multivariate",pos=4)
dev.off()







#plot HR of ecmn and ecmp genes in breast cancer microarray sets (Fig 5)
sets<-c(GEOD21653,EMTAB365)
x1 = NULL
x2 = NULL
x3 = NULL
x4 = NULL
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  probesp = NULL
  for (j in 1:length(ecmp)){
    x = set$key[which(set$key[,2]==ecmp[j]),1]
    if(length(x)>1){
      rm = rowMeans(set$data[x,])
      probesp = c(probesp,x[which.max(rm)])
    } else {
      probesp = c(probesp,x)
    }
  }
  probesn = NULL
  for (j in 1:length(ecmn)){
    x = set$key[which(set$key[,2]==ecmn[j]),1]
    if(length(x)>1){
      rm = rowMeans(set$data[x,])
      probesn = c(probesn,x[which.max(rm)])
    } else {
      probesn = c(probesn,x)
    }
  }
  for(k in 1:length(probesp)){
    cx = summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[probesp[k],])+
                         scale(colMeans(set$data[probesn,]))+strata(set$ER)+strata(set$node)))
    x1=rbind(x1,cx$conf.int[1,c(1,3,4)])
    cx = summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[probesp[k],])+
                         +strata(set$ER)+strata(set$node)))
    x2 = rbind(x2,cx$conf.int[1,c(1,3,4)])
  }
  for(k in 1:length(probesn)){
    cx = summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[probesn[k],])+
                         scale(colMeans(set$data[probesp,]))+strata(set$ER)+strata(set$node)))
    x3=rbind(x3,cx$conf.int[1,c(1,3,4)])
    cx = summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[probesn[k],])+
                         +strata(set$ER)+strata(set$node)))
    x4 = rbind(x4,cx$conf.int[1,c(1,3,4)])
  }
}

tiff(file=paste(path,"fig5A.tif",sep=""),width=2000,height=1600,res=300,compression="lzw")
plot(-1,-1,ylim=c(0,6),log="x",xlim=c(0.3,3.3),yaxt="n",xlab="Hazard ratio",ylab="",cex=0.7)
abline(v=1)
for(i in 1:5){
  lines(x3[i,2:3],c(i-5/6,i-5/6),lty=1,col="blue",lwd=2)
  lines(x4[i,2:3],c(i-4/6,i-4/6),lty=1,col="blue4",lwd=2)
  lines(x3[i+5,2:3],c(i-2/6,i-2/6),lty=1,col="red",lwd=2)
  lines(x4[i+5,2:3],c(i-1/6,i-1/6),lty=1,col="red4",lwd=2)
  abline(h=i,lty=2)
  text(0.3,6-i-1/3,labels=ecmn[i],cex=1,adj=0)
}
lines(c(0.3,0.36),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red4")
lines(c(0.36,0.432),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="blue4")
lines(c(0.3,0.36),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="red")
lines(c(0.36,0.432),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue")
text(0.44,nrow(x3)/2+0.85,labels="univariate",pos=4)
text(0.44,nrow(x3)/2+0.45,labels="multivariate",pos=4)
lines(c(1.05,1.25),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red4")
lines(c(1.25,1.5),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red")
lines(c(1.05,1.25),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue4")
lines(c(1.25,1.5),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue")
text(1.6,nrow(x3)/2+0.85,labels="GEOD21653",pos=4)
text(1.6,nrow(x3)/2+0.45,labels="EMTAB365",pos=4)
mtext("A",side=3,line=0,cex=2,adj=0,outer=F)
dev.off()

tiff(file=paste(path,"fig5B.tif",sep=""),width=2000,height=1600,res=300,compression="lzw")
plot(-1,-1,ylim=c(0,6),log="x",xlim=c(0.3,3.3),yaxt="n",xlab="Hazard ratio",ylab="",cex=0.7)
abline(v=1)
for(i in 1:5){
  lines(x1[i,2:3],c(i-5/6,i-5/6),lty=1,col="blue",lwd=2)
  lines(x2[i,2:3],c(i-4/6,i-4/6),lty=1,col="blue4",lwd=2)
  lines(x1[i+5,2:3],c(i-2/6,i-2/6),lty=1,col="red",lwd=2)
  lines(x2[i+5,2:3],c(i-1/6,i-1/6),lty=1,col="red4",lwd=2)
  abline(h=i,lty=2)
  text(0.3,6-i-1/3,labels=ecmp[i],cex=1,adj=0)
}
lines(c(0.3,0.36),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red4")
lines(c(0.36,0.432),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="blue4")
lines(c(0.3,0.36),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="red")
lines(c(0.36,0.432),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue")
text(0.44,nrow(x3)/2+0.85,labels="univariate",pos=4)
text(0.44,nrow(x3)/2+0.45,labels="multivariate",pos=4)
lines(c(1.05,1.25),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red4")
lines(c(1.25,1.5),rep(nrow(x3)/2+0.8,2),lty=1,lwd=2,col="red")
lines(c(1.05,1.25),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue4")
lines(c(1.25,1.5),rep(nrow(x3)/2+0.4,2),lty=1,lwd=2,col="blue")
text(1.6,nrow(x3)/2+0.85,labels="GEOD21653",pos=4)
text(1.6,nrow(x3)/2+0.45,labels="EMTAB365",pos=4)
mtext("B",side=3,line=0,cex=2,adj=0,outer=F)
dev.off()


#plot expression tumors in normal tissue and in tumor tissue (Fig 6)
load(paste(path,TBRCAnorm, sep=""))
mod<-as.matrix(log2(mRNAnormdata+1))
rownames(mod)<-mRNAID[,1]
load(file=paste(path,TBRCA,sep=""))
x<-match(colnames(mod),colnames(set$data))
np<-mod[,!is.na(x)]
x<-x[!is.na(x)]
tp<-set$data[,x]

tiff(file=paste(path,"fig6.tif",sep=""),width=2700,height=1600,res=300,compression="lzw")
options(scipen=3)
gns = c(ecmn,ecmp)
par(mfrow=c(3,6))
par(mar=c(1,3,2,1))
par(oma=c(0,2,0,0))
par(font.main=4)
for (i in 1:length(gns)){
  ymin = min(c(tp[gns[i],],np[gns[i],]))
  ymax = max(c(tp[gns[i],],np[gns[i],]))
  y1 = ymin - 0.1*(ymax-ymin)
  y2 = ymax + 0.1*(ymax-ymin)
  boxplot(tp[gns[i],],np[gns[i],],main=gns[i],ylim=c(y1,y2),xaxt="n",col=c(2,4),las=1)
  tt = t.test(tp[gns[i],],np[gns[i],])
  text(1.3,ymax,labels=signif(tt$p.value,2),pos=3)
  if(i==1){mtext("A",side=2,line=0.8,cex=1.6,at=20,las=1,outer=F)}
  if(i==7){mtext("B",side=2,line=0.8,cex=1.6,at=17.5,las=1,outer=F)}
  if(i==11){
    #frame()
    plot(0,0,type="n",xlim=c(0,10),ylim=c(0,10),xaxt="n",yaxt="n",xlab="",ylab="")    # type="n" hides the points  
    box("plot", col="white")  
    legend(-1.5,8,bty="n",pch=15,col=c("red","blue"),legend=c("cancer","normal"),cex=1.6)
  }
}
gns = c(ecmp)
for (i in 1:length(gns)){
  ymin = min(c(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,])))
  ymax = max(c(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,])))
  y1 = ymin - 0.1*(ymax-ymin)
  y2 = ymax + 0.1*(ymax-ymin)
  boxplot(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,]),main=gns[i],
          ylim=c(y1,y2),xaxt="n",col=c(2,4),las=1)
  tt = t.test(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,]))
  text(1.3,y2,labels=signif(tt$p.value,2))
  #text(1,1.1*ymax,labels=signif(tt$p.value,2))
  if(i==1){mtext("C",side=2,line=0.8,cex=1.6,at=(-0.8),las=1,outer=F)}
}
mtext("log2 expression",side=2,line=0,cex=1,outer=T)
mtext("log2 expression",side=2,line=0,cex=1,outer=T,adj=0.93)
mtext("log2 ratio",side=2,line=0,cex=1,outer=T,adj=0.12)
dev.off()

#find and plot expression levels in mouse and human cells (stroma vs tumor) - xenografts
pathx = paste(path,GSE66744,sep="")
x = dir(pathx)
samples = x[grep("Mammary",x)]
samples = c(samples,x[grep("Mouse",x)])
samples = samples[3:18]
fact = c(1,2,1,2,1,2,1,2,3,4,3,4,3,4,3,4)

ecmn_mouse = sapply(ecmn,function(x) paste(substr(x,1,1),tolower(substr(x,2,nchar(x))),sep=""))
ecmp_mouse = sapply(ecmp,function(x) paste(substr(x,1,1),tolower(substr(x,2,nchar(x))),sep=""))
contr = c("ESR1","YWHAZ","SDHA","UBC")
contr_mouse = sapply(contr,function(x) paste(substr(x,1,1),tolower(substr(x,2,nchar(x))),sep=""))

gdata = NULL
for (i in 1:length(samples)){
  x = read.table(paste(pathx,samples[i],sep=""),
                 stringsAsFactors=F,header=T,sep="\t")
  xl = log2(10e+5*x[,3]/sum(x[,3])+1)
  if (length(grep("human",samples[i]))>0){
    gdata = rbind(gdata,xl[match(c(ecmn,ecmp,contr),x[,1])])
  }
  if (length(grep("mouse",samples[i]))>0){
    gdata = rbind(gdata,xl[match(c(ecmn_mouse,ecmp_mouse,contr_mouse),x[,1])])
  }
}
tiff(file=paste(path,"fig7.tif",sep=""),width=2500,height=1600,res=300,compression="lzw")
par(mfrow=c(3,6))
par(mar=c(1,2,2,0.5))
par(oma=c(0,2,0,0))
for (i in 1:ncol(gdata)){
  ymax = max(gdata[,i])
  plot(jitter(fact),gdata[,i],col=c(2,4,2,4),xaxt="n",las=1,ylab="log2 RPKM",pch=19,xlab="",main=c(ecmn,ecmp,contr)[i],
       font.main=4,ylim=c(0,ymax*1.3),xlim=c(0.5,4.5))
  lines(c(0.7,2.3),c(ymax*1.02,ymax*1.02))
  lines(c(2.7,4.3),c(ymax*1.02,ymax*1.02))
  lines(c(2.5,2.5),c(0,ymax*1.05),lty=2)
  text(1.5,ymax*1.01,pos=3,labels="MCF7")
  text(3.5,ymax*1.01,pos=3,labels="  MDA-\nMB-231")
  if (i==11){
    frame()
  }
  if (i==15){
    plot(0,0,type="n",xlim=c(0,10),ylim=c(0,10),xaxt="n",yaxt="n",xlab="",ylab="")    # type="n" hides the points  
    box("plot", col="white")  
    legend(-1,8,bty="n",pch=16,col=c("red","blue"),legend=c("cancer \ncells","stroma"),cex=1.6)
    frame()
  }
}
mtext("log2 expression",side=2,line=0,cex=1,adj=0.48,outer=T)
mtext("log2 expression",side=2,line=0,cex=1,outer=T,adj=0.91)
mtext("log2 expression",side=2,line=0,cex=1,outer=T,adj=0.06)
mtext("A",side=2,line=0,cex=1.6,outer=T,at=0.98,las=1)
mtext("B",side=2,line=0,cex=1.6,outer=T,at=0.64,las=1)
mtext("C",side=2,line=0,cex=1.6,outer=T,at=0.30,las=1)
dev.off()


#boxplot of ECM and endothelial clusters in TCGA tumors (Figure S3)
sets<-c(TBRCA,TCOAD,THNSC,
        TKIRC,TLUAD,TLUSC)
setnames<-c("BRCA","COAD","HNSC","KIRC","LUAD","LUSC")

x<-NULL
y<-NULL
ind<-NULL
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  x<-c(x,colMeans(set$data[necm,]))
  y<-c(y,colMeans(set$data[nend,]))
  ind<-c(ind,rep(i,ncol(set$data)))
}
tiff(file=paste(path,"Fig S3.tif",sep=""),width=1600,height=3200,res=300,compression="lzw")
par(mfrow=c(2,1))
boxplot(x~ind,xaxt="n",main="ECM metagene",ylab="mean log2 expression")
axis(1,at=1:6,tick=F,labels=setnames)
boxplot(y~ind,xaxt="n",main="Endothelial metagene",ylab="mean log2 expression")
axis(1,at=1:6,tick=F,labels=setnames)
dev.off()

#find and plot expression levels of P4HAs in mouse and human cells (stroma vs tumor) - Fig 8A
pathx = paste(path,GSE66744,sep="")
x = dir(pathx)
samples = x[grep("Mammary",x)]
samples = c(samples,x[grep("Mouse",x)])
samples = samples[3:18]
fact = c(1,2,1,2,1,2,1,2,3,4,3,4,3,4,3,4)
genes = c("P4HA1","P4HA2","P4HA3")
genes_mouse = sapply(genes,function(x) paste(substr(x,1,1),tolower(substr(x,2,nchar(x))),sep=""))
gdata = NULL
for (i in 1:length(samples)){
  x = read.table(paste(pathx,samples[i],sep=""),
                 stringsAsFactors=F,header=T,sep="\t")
  xl = log2(10e+5*x[,3]/sum(x[,3])+1)
  if (length(grep("human",samples[i]))>0){
    gdata = rbind(gdata,xl[match(c(genes),x[,1])])
  }
  if (length(grep("mouse",samples[i]))>0){
    gdata = rbind(gdata,xl[match(c(genes_mouse),x[,1])])
  }
}
tiff(file=paste(path,"fig8A.tif",sep=""),width=1600,height=800,res=300,compression="lzw")
par(oma=c(0,2,0,0))
par(mfrow=c(1,4))
par(mar=c(1,2,2,0.5))
par(font.main=4)
for (i in 1:ncol(gdata)){
  ymax = max(gdata[,i])
  plot(jitter(fact),gdata[,i],col=c(2,4,2,4),xaxt="n",las=1,ylab="log2 RPKM",pch=19,xlab="",main=c(genes)[i],
       ylim=c(0,ymax*1.3),xlim=c(0.5,4.5))
  lines(c(0.7,2.3),c(ymax*1.02,ymax*1.02))
  lines(c(2.7,4.3),c(ymax*1.02,ymax*1.02))
  lines(c(2.5,2.5),c(0,ymax*1.05),lty=2)
  text(1.5,ymax*1.01,pos=3,labels="MCF7")
  text(3.5,ymax*1.01,pos=3,labels="  MDA-\nMB-231")
  #legend(0.2,ymax*1.4,pch=19,col=c(2,4),legend=c("Cancer cells","Stroma"),bty="n")
}
mtext("log2 expression",side=2,line=0,cex=1,adj=0.48,outer=T)
mtext("A",side=2,line=0,cex=1.6,outer=T,at=0.95,las=1)
plot(0,0,type="n",xlim=c(0,10),ylim=c(0,10),xaxt="n",yaxt="n",xlab="",ylab="")    # type="n" hides the points  
box("plot", col="white")  
legend(-1,8,bty="n",pch=16,col=c("red","blue"),legend=c("cancer \ncells","stroma"),cex=1.2)
dev.off()


#plot expression in tumors and in normal (normalized to ECM metagene) tissue of P4HAs - Fig 8B
load(paste(path,TBRCAnorm,sep=""))
mod<-as.matrix(log2(mRNAnormdata+1))
rownames(mod)<-mRNAID[,1]
load(file=paste(path,TBRCA,sep=""))
x<-match(colnames(mod),colnames(set$data))
np<-mod[,!is.na(x)]
x<-x[!is.na(x)]
tp<-set$data[,x]
options(scipen=3)
gns = c("P4HA1","P4HA2","P4HA3")
tiff(file=paste(path,"fig8B.tif",sep=""),width=1600,height=800,res=300,compression="lzw")
par(oma=c(0,2,0,0))
par(mfrow=c(1,4))
par(mar=c(1,2,2,0.5))
par(font.main=4)
for (i in 1:length(gns)){
  ymin = min(c(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,])))
  ymax = max(c(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,])))
  y1 = ymin - 0.1*(ymax-ymin)
  y2 = ymax + 0.1*(ymax-ymin)
  boxplot(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,]),main=gns[i],
          ylim=c(y1,y2),xaxt="n",col=c(2,4),las=1)
  tt = t.test(tp[gns[i],]-colMeans(tp[necm,]),np[gns[i],]-colMeans(np[necm,]))
  text(1.3,y2,labels=signif(tt$p.value,2))
  #text(1,1.1*ymax,labels=signif(tt$p.value,2))
}
mtext("log2 ratio",side=2,line=0,cex=1,adj=0.48,outer=T)
mtext("B",side=2,line=0,cex=1.6,outer=T,at=0.95,las=1)
plot(0,0,type="n",xlim=c(0,10),ylim=c(0,10),xaxt="n",yaxt="n",xlab="",ylab="")    # type="n" hides the points  
box("plot", col="white")  
legend(-1,8,bty="n",pch=15,col=c("red","blue"),legend=c("cancer","normal"),cex=1.2)
dev.off()




#find cor coefficients for P4HAs and ECM signature in TCGA data Fig 8D
sets<-c(TBRCA,TCOAD,THNSC,
        TKIRC,TLUAD,TLUSC)
setnames<-c("BRCA","COAD","HNSC","KIRC","LUAD","LUSC")
x<-NULL
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  x<-cbind(x,c(cor(set$data["P4HA1",],colMeans(set$data[necm,])),
               cor(set$data["P4HA2",],colMeans(set$data[necm,])),
               cor(set$data["P4HA3",],colMeans(set$data[necm,]))))
}
colnames(x)<-setnames
rownames(x)<-c("P4HA1","P4HA2","P4HA3")
x<-x[nrow(x):1,]
cols<-c("red4","orange4","green4","cyan4","blue4","purple4")
cols<-c("red1","orange1","green1","cyan1","blue1","purple1")
tiff(file=paste(path,"fig8D.tif",sep=""),width=1600,height=1400,res=300,compression="lzw")
par(mar=c(5,6,4,2))
plot(1000,1000,ylim=c(0,3),xlim=c(0,1),yaxt="n",xlab="correlation coefficient with ECM metagene",ylab="")
for(i in nrow(x):1){
  points(x[i,],rep(i*0.5,length(sets)),col=cols,pch=19)
}
axis(2,at=c(0.5*(1:3)),labels=rownames(x),tick=F,las=1,font=3)
legend(0.2,3,pch=19,col=cols[1:3],legend=setnames[1:3],bty="n")
legend(0.5,3,pch=19,col=cols[4:6],legend=setnames[4:6],bty="n")
mtext("D",side=2,line=2,cex=1.6,at=3,las=1)
dev.off()

#plot HRs of P4HAs in breast cancer microarray sets and TCGA BRCA - Fig 8E
gns=c("P4HA1","P4HA2","P4HA3")
sets<-c(GEOD21653,EMTAB365)
yy<-NULL
for (i in 1:length(sets)){
  load(paste(path,sets[i],sep=""))
  y<-NULL
  for (j in 1:length(gns)){
    x12<-which(set$key[,2]==gns[j])
    rm<-rowMeans(set$data[set$key[x12,1],])
    x12<-names(rm)[which.max(rm)]
    x13<-set$key[unlist(sapply(necm,function(x) which(set$key[,2]==x))),1]
    y<-c(y,summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[x12,])+strata(set$ER)+strata(set$node)))$conf.int[1,c(1,3:4)])
    y<-c(y,summary(coxph(survrc(set$time,set$event,1825)~scale(set$data[x12,])+scale(colMeans(set$data[x13,]))+strata(set$ER)+strata(set$node)))$conf.int[1,c(1,3:4)])
  }
  yy<-cbind(yy,y)
}
rn<-c(rep(gns[1],6),rep(gns[2],6),rep(gns[3],6))
rn<-paste(rn,rep(c("uni HR","l","h","multi HR","l","h"),length(gns)))
rownames(yy)<-rn
y<-NULL
load(file=paste(path,TBRCA,sep=""))
for(j in 1:length(gns)){
  y<-c(y,summary(coxph(survrc(set$rtime,set$revent,1825)~scale(set$data[gns[j],])+strata(set$ER)+strata(set$node)))$conf.int[1,c(1,3:4)])
  y<-c(y,summary(coxph(survrc(set$rtime,set$revent,1825)~scale(set$data[gns[j],])+scale(colMeans(set$data[necm,]))+strata(set$ER)+strata(set$node)))$conf.int[1,c(1,3:4)])
}
yy<-cbind(yy,y)
rn<-c(rep(gns[1],6),rep(gns[2],6),rep(gns[3],6))
rn<-paste(rn,rep(c("uni HR","l","h","multi HR","l","h"),length(gns)))
rownames(yy)<-rn
library(RColorBrewer)
setnames<-c("GEOD21653","EMTAB365","TCGA BRCA")
cols = c("red4","red","blue4","blue","black","gray50")
ltypes=c(2,1)
tiff(file=paste(path,"fig8E.tif",sep=""),width=1800,height=1400,res=300,compression="lzw")
plot(1000,1000,ylim=c(-28,0),xlim=c(0.6,6),log="x",xaxt="n",yaxt="n",xlab="HR",ylab="")
axis(1,at=c(0.7,1,1.5,2.5,4))
abline(v=1)
ind<-c(1,2,3)
for (i in 1:3){
  for(k in 1:3){
    j<-ind[k]
    points(yy[i*6-5,j],-((i-1)*11+2*j-1),col=cols[2*j-1])
    points(yy[i*6-2,j],-((i-1)*11+2*j),col=cols[2*j])
    lines(c(yy[i*6-4,j],yy[i*6-3,j]),c(-((i-1)*11+2*j-1),-((i-1)*11+2*j-1)),col=cols[2*j-1],lty=1,lwd=2)
    lines(c(yy[i*6-1,j],yy[i*6,j]),c(-((i-1)*11+2*j),-((i-1)*11+2*j)),col=cols[2*j],lty=1,lwd=2)
  }
  text(0.63,-(i*11)+8,labels=gns[i],bty="n",font=3)
}
lgds<-paste(c(sapply(setnames,function(x) rep(x,2))),rep(c("uni","multi"),2))
legend(2.2,-1.1,legend=lgds,bty="n",lty=1,lwd=2,col=cols)
mtext("E",side=2,line=0.4,cex=1.6,at=0,las=1)
dev.off()