RNA analysis

RNA was prepared using RNAeasy Mini kit (Qiagen 74106) and 1 μg was converted to cDNA using qScript cDNA SuperMix (Quanta 95048-100). Relative transcript levels were determined by qPCR, performed on a 7500 Fast Real Time PCR System (Applied Biosystems) using SYBR Green PCR Master Mix (Applied Biosystems). No RT or template control and inspection of dissociation curves verified amplifications. Arbitrary units were determined by normalizing to Gapdh levels. Primer sequences are shown in Supplementary Methods.

RNA was prepared similarly for RNA Sequencing and the quality was checked on an Agilent Technologies 2100 Bioanalyzer. Libraries were prepared using the TruSeq RNA sample Preparation Kit (Illumina Rs-122–2001) and sequenced as 2 × 51 bp reads at the Weill Cornell Genomics Core Facility with the Illumina HiSeq2000 sequencer using paired-end module. After quality control using the Illumina pipeline, reads were mapped using Tophat with default parameters and mouse genome build mmp9 (ref. ⁴⁰). Cufflinks with upper-quartile normalization and sequence-specific bias correction was used to generate Fragments per kilobase of transcript per million fragments (FPKM) values⁴¹.

Hierarchical clustering and principal component analysis were performed in R using log2 transformed FPKM values with distances calculated by subtracting the Pearson's correlation value from 1. Those values were also used to generate the bar graphs indicating distances to an average cultured lung EC sample. Clustering was unsupervised except when only genes associated with the GO term ‘angiogenesis' were used. Pathway analysis was performed using the set of differentially expressed genes for an indicated comparison. Genes were considered differentially expressed if their log2 fold change was greater than or less than 1 for the comparison of their averaged FPKMs and if the P-value was<0.05 according to a two-sided t-test, and not assuming equal variance. Terms among the top 10 GOTERM_BP_FAT category, as determined by DAVID, were used⁴². Heatmaps were generated using the pheatmap function in R after normalizing FPKM values by the maximum value for a given transcript.