Developmental regulation of canonical and small ORF translation from mRNAs

Abstract

Background: Ribosomal profiling has revealed the translation of thousands of sequences outside annotated protein-coding genes, including small open reading frames of less than 100 codons, and the translational regulation of many genes. Here we present an improved version of Poly-Ribo-Seq and apply it to Drosophila melanogaster embryos to extend the catalog of in vivo translated small ORFs, and to reveal the translational regulation of both small and canonical ORFs from mRNAs across embryogenesis.

Results: We obtain highly correlated samples across five embryonic stages, with nearly 500 million putative ribosomal footprints mapped to mRNAs, and compare them to existing Ribo-Seq and proteomic data. Our analysis reveals, for the first time in Drosophila, footprints mapping to codons in a phased pattern, the hallmark of productive translation. We propose a simple binomial probability metric to ascertain translation probability. Our results also reveal reproducible ribosomal binding apparently not resulting in productive translation. This non-productive ribosomal binding seems to be especially prevalent amongst upstream short ORFs located in the 5' mRNA leaders, and amongst canonical ORFs during the activation of the zygotic translatome at the maternal-to zygotic transition.

Conclusions: We suggest that this non-productive ribosomal binding might be due to cis-regulatory ribosomal binding and to defective ribosomal scanning of ORFs outside periods of productive translation. Our results are compatible with the main function of upstream short ORFs being to buffer the translation of canonical canonical ORFs; and show that, in general, small ORFs in mRNAs display markers compatible with an evolutionary transitory state towards full coding function.

Keywords: Maternal to zygotic transition; Non-canonical translation; Poly-Ribo-Seq; Regulation of translation; Ribosomal binding; Ribosomal profiling; sORFs; smORFs; uORFs.

Fig. 1

Experimental design and quality-control of Poly-Ribo-Seq data. aDrosophila melanogaster embryos were collected in three contiguous 8-h time-windows spanning the whole of embryogenesis. Two biological replicates (T and B) were collected for each window, for both Poly-Ribo-Seq and RNA-Seq. pro: procephalon; gb: germ band; df: dorsal fold; mg: presumptive midgut (modified from ref. [15]). b Percentage of 5′UTR, 3′UTR, and CDS-aligned reads for Poly-Ribo-Seq, across the annotated transcriptome of Drosophila melanogaster. c Reproducibility of RPKM values for canonical ORFs by Spearman correlation (r) across replicates T (x-axis) and B (y-axis), per stage, for both Poly-Ribo-Seq (top panels) and RNA-Seq (bottom panels). d Density of genome-aligned reads for all experiments, both Poly-Ribo-Seq (FPs) and RNA-Seq across the polycistronic tarsal-less (tal) transcript. As expected, FPs map to ORFs 1A, 2A, 3A, and AA (previously shown to be translated), see ref. [30] and Table S3A for details

Fig. 2

Detection of translation in annotated and non-annotated ORFs. a Pipeline used in this study. An ORF was considered translated, in each stage, if it showed transcription (RPKM > 1 in one of two RNA-Seq replicates), reproducible ribosomal binding (RPKM > 1 in both Poly-Ribo-Seq replicates) and a significant framing of FPs (p value < 0.01 in the binomial test for framed-reads). b Metagene plot for codon framing of 32-nt ribosome footprints across all ORFs analyzed in this study, showing three-nucleotide periodicity (framing) in the third position of each codon (frame 2, blue). Numbers on top denote the distances between the 5′-end of ribosomal footprints and START/STOP codons. c The observed framing for 32-nt ribosome footprints is consistent across embryonic stages (combined replicates per stage). d Logic of per-ORF analysis of framing probabilities for each ORF at a given stage using the binomial test. e Density of 32-nt genome-aligned Poly-Ribo-Seq reads for all experiments, per frame, and corresponding translation probability for the scl-A ORF, as measured by the binomial test (see Table S3B for details). f Numbers of developmentally transcribed, ribosome-bound, and translated canonical ORFs (annotated ORFs with length > 100 codons), short CDSs (annotated ORFs ≤ 100 codons), and uORFs

Fig. 3

Comparisons across genome-wide datasets. a Per-stage Spearman’s correlations across Poly-Ribo-Seq (“FP”), RNA-Seq (“mRNA”), and mass-spectrometry datasets (“E”—early, “M”—mid, “L”—late stages correspondence in Casas-Vila et al. [26]—see “Materials and methods”). “Maternal” datasets correspond to concatenated mature oocyte and activated egg datasets from Kronja et al. [18]. S2-cells correspond to the Aspden et al. 2014 dataset plus new sequencing. Numbers denote Spearman’s rho. b Numbers and translation fates of maternal ribosome-bound canonical ORFs in the maternal-to-zygotic transition. “Constant translation” denotes ORFs detected as translated by our pipeline in both maternal and early embryo datasets (see Fig. 2a). “Poised for translation” denotes ORFs showing maternal ribosome-binding which only appear translated in the early embryo. “Translation down” denotes ORFs with ribosome-binding and translation in the Maternal dataset, which do not appear translated in the early embryo dataset. c Overall correlation between average embryo Poly-Ribo-Seq RPKM and average embryo Mass-Spec imputed LFQ intensities (both datasets are log₂-transformed). All detected ORFs are included in this analysis. Red points and line show the data binned in intervals of 0.1 RPKM. Note that the linear correlation is lost below RPKM = 1 (log₂ = 2) and above RPKM = 1024 (log₂ = 10). d Differential detectability of canonical ORFs between Poly-Ribo-Seq and mass spectrometry techniques. e Differential detectability of short CDSs between Poly-Ribo-Seq and mass spectrometry. f S2-tagging experiments of nine uncharacterized short CDSs showing translation in either cortical or reticular-punctate patterns (FLAG antibody: green, F-actin stained with phalloidin: red. Scale bar denotes 5 μm)

Fig. 4

Regulated translation during embryogenesis. a Number of transcribed, ribosome-bound, and translated ORFs per class across Drosophila melanogaster embryonic development, showing stage-specificity in ON/OFF translation patterns. b Changes in average translation efficiency, per ORF class, across analyzed developmental time-points. cZ-score ratio analysis of translation efficiencies (TE) during successive developmental stages pinpointing ORFs subjected to significant developmental modulations of translation across ORF classes. d Number of developmentally regulated canonical ORFs subjected to downregulation (bottom, Z-ratio ≤ − 1.5) or upregulation (top, Z-ratio ≥ 1.5), per developmental transition. e Metagene plot for codon framing of 32-nt ribosome footprints across reproducibly “ribosome-bound-only” uORFs, showing a lack of three-nucleotide periodicity (framing) similar to that of mRNA-Seq (compared with translated uORFs in Additional file 1: Fig. S3C). f Distribution of log₂(RPKM^FP) values of ribosome-bound-only uORFs (red) and translated (framed) uORFs (green). Median nucleotide sizes of each pool are indicated above

Fig. 5

uORFs as translational regulators. a Variation in canonical TE levels as a function of the number of cistronic uORFs showing ribosomal binding. b Comparison of translational changes between uORFs and their canonical mORFs across developmental stages. Numbers denote the number of Z-ratio significant uORF-mORF pairs in each quadrant. Vertical arrows: uORF translational regulation. Horizontal arrows: canonical mORF translational regulation. Green arrows: upregulated; red arrows: downregulated. c Correlation between the number of uORFs per mRNA and 5′UTR length (Spearman’s r = 0.7). d Comparison of Kozak-context scores between distinct ORF classes and uORF subclasses. Graphs display 10–90 percentile range. Mann-Whitney p < 0.001. e Comparison of phyloP conservation scores (27-way alignments) between distinct ORF classes and uORF subclasses. Graphs display 10–90 percentile range. f Correlations in amino acid usage across distinct uORF subclasses, short CDSs, and canonical ORFs; values denote pairwise Spearman’s r. g Model for uORF function and evolution. 5′ leaders in blue. t: evolutionary time. Red lines denote ribosomal binding signal; white boxes correspond to novel uORFs; salmon boxes correspond to ribosomal-bound uORFs; red boxes correspond to translated uORFs. Purple scribble denotes produced peptide