Ribo-seq analysis

We applied Bowtie 1.0 (REF) to map reads to rRNA, genomic and transcriptomic sequences from the Ensembl database (version 75). rRNA reads and reads mapping to the mitochondrial genome were discarded. All alignments were mapped to genomic coordinates. Fractional counts were used for ambiguous alignments (with regard to genomic coordinates). We then used the probabilistic model implemented in Price (version 1.0.3b) [26] to map reads to their P site codons using default parameters. All read counts corresponding to translation start site profiling (lactimidomycin-treated samples) were discarded. Next, we removed (i) PAS that were located inside of Ensembl 75 exons (n = 110), (ii) PAS without an annotated or Price-identified open reading frame (ORF) in the upstream exon (n = 64), (iii) PAS without an in-frame stop codon in the intron in between the exon boundary and the PAS (n = 9), (iv) PAS where the first in-frame stop codon was in the first five intronic codon triplets (n = 100), and (v) PAS downstream of very weakly translated ORFs (<0.5 reads per exonic codon in all Ribo-seq samples pooled; n = 104). For the remaining n = 132 PAS, we counted codon mapped reads for the partial open reading frame in the upstream exon (reads mapped to codons in the same frame as the ORF, and in the other two frames), for its extension into the intron up to the first in-frame stop codon (reads mapped to codons in the same frame as the ORF, and in the other two frames), and reads mapped in between the stop codon and the PAS (Fig 6C). Read counts were normalized to the total number of Ribo-seq reads mapped to the human genome.