Translation efficiency calculation

Data analysis was performed with custom scripts written for R 2.15.2 and Python 2.6.6.

To calculate mRNA abundance, the number of mRNA-seq reads mapped to the open reading frame (ORF) of a gene, following a Winsorization applied to remove the top and bottom 5% of reads, was normalized by the length of the ORF and the total number of mapped reads to yield the number of mRNA-seq reads corresponding to the gene per kilobase of message per million mapped reads (RPKM).

The protein synthesis rate of individual ORFs was measured by average ribosome footprint density of the ORF calculated as described in (

Li et al., 2014). First, genes with less than 128 ribosome profiling reads mapped to the ORFs and genes with unconventional translation events were excluded from the analysis, which include (1) genes encoding selenoproteins (e.g., fdhF, fdoG, fdnG); (2) protein pairs with nearly identical coding sequences (e.g., gadA and gadB, ynaE and ydfK, ldrA and ldrC, ybfD and yhhI, tfaR and tfaQ, rzoD and rzoR, pinR and pinQ). Second, ribosome profiling reads mapped to the first and last five codons of the gene were excluded when calculating the average ribosome footprint density to remove effects of slower ribosome kinetics during translation initiation and termination. Third, correction for the variations in translation elongation rate was done in the following three steps as described in (Li et al., 2014): (1) Counts of ribosome footprints for each gene was first corrected for the elevated density observed for the first 50-100 codons (Oh et al., 2011). A metagene analysis for the relative ribosome density as a function of the distance to start codons was used as a calibration. (2) The resulting counts were corrected for the elevated density from internal Shine-Dalgarno-like sequences (Li et al., 2012). The affinity of the upstream hexameric sequence to the anti-Shine-Dalgarno (aSD) sequence of each position on the gene was calculated and the calibration curve was obtained empirically by fitting the observed average ribosome density of the hexameric sequences as a function of the hybridization energy to the aSD sequence. (3) The resulting counts were further corrected for other possible ribosome pausing events using 90% Winsorization, by removing the top and bottom 5% of the ribosome profiling signal for each gene. Finally, the average ribosome footprint density of a gene was calculated by dividing the corrected total number of mapped ribosome footprint reads by the corrected length of the gene body (excluding first and last five codons). Translation efficiency of a gene was calculated by normalizing the average ribosome footprint density by the mRNA abundance of the gene (defined above). The average ribosome footprint density (i.e. protein synthesis rate), mRNA abundance, and translation efficiency of genes from different samples are listed in Supplementary file 1-4.