Translation of upstream open reading frames in a model of neuronal differentiation

Abstract

Background: Upstream open reading frames (uORFs) initiate translation within mRNA 5' leaders, and have the potential to alter main coding sequence (CDS) translation on transcripts in which they reside. Ribosome profiling (RP) studies suggest that translating ribosomes are pervasive within 5' leaders across model systems. However, the significance of this observation remains unclear. To explore a role for uORF usage in a model of neuronal differentiation, we performed RP on undifferentiated and differentiated human neuroblastoma cells.

Results: Using a spectral coherence algorithm (SPECtre), we identify 4954 consistently translated uORFs across 31% of all neuroblastoma transcripts. These uORFs predominantly utilize non-AUG initiation codons and exhibit translational efficiencies (TE) comparable to annotated coding regions. On a population basis, the global impact of both AUG and non-AUG initiated uORFs on basal CDS translation were small, even when analysis is limited to conserved and consistently translated uORFs. However, uORFs did alter the translation of a subset of genes, including the Diamond-Blackfan Anemia associated ribosomal gene RPS24. With retinoic acid induced differentiation, we observed an overall positive correlation in translational shifts between uORF/CDS pairs. However, CDSs downstream of uORFs show smaller shifts in TE with differentiation relative to CDSs without a predicted uORF, suggesting that uORF translation buffers cell state dependent fluctuations in CDS translation.

Conclusion: This work provides insights into the dynamic relationships and potential regulatory functions of uORF/CDS pairs in a model of neuronal differentiation.

Keywords: 5′ untranslated region; Near-cognate start codon; Neuronal differentiation; Ribosome profiling; Translation; Upstream open reading frame.

Fig. 1

Retinoic acid treatment induces differential translation in SH-SY5Y human neuroblastoma cells. a Schematic of experimental design and data acquisition work-flow. b Immunocytochemistry performed on Non-Diff and RA-Diff SH-SY5Y cells with antibodies against neurofilament (red) and β-actin (green). Nuclei were DAPI stained (blue). c β-actin expression was decreased in RA-Diff cells. Individual cell fluorescence was quantified and represented as a corrected total cell fluorescence (CTCF) for β-actin; n = 119 for Non-Diff and n = 118 for RA-Diff. d Primary neurite length measured by neurofilament staining; n = 109 for Non-Diff and n = 100 for RA-Diff. e FMRP expression by immunoblot before and after RA treatment, quantified in F); n = 4 for both conditions. For panels C), D), and f Student’s t test, ****p ≤ 0.0001. Graphs represent mean ± S.E.M. g Differential mRNA abundance based on Non-Diff versus RA-Diff TPM. Transcripts were defined as significantly up-regulated (cyan) or down-regulated (gold) in the RA-Diff condition based on rank-change in abundance compared to the Non-Diff condition. h Volcano plot of transcripts with differential translation by translational efficiency (TE) by Riborex analysis. Significantly up-regulated genes (cyan) and down-regulated genes (gold) in RA-Diff cells are defined by an absolute log2-normalized fold-change cutoff of ±1 (vertical lines), and a multiple testing corrected p-value cutoff of 0.1 (horizontal line). i Gene sets (biological process) with significantly downregulated TE in RA-Diff cells. Genes upregulated in RA-Diff cells is shown in (j). The top five biological processes with significant change using a multiple testing corrected p-value cutoff of 0.05 (vertical line) are shown on the graph. k Plot shows normalized mRNA reads (grey) and RPF (cyan/gold) over the 5’leader (thin line, left), and CDS (thick line, middle). The axon guidance gene, PLXNA2, is representative of a transcript with higher translational efficiency and RPF in the RA-Diff condition.

Fig. 2

Computational prediction and filtering of upstream-initiated open-reading frames. ORFs were predicted from AUG and non-AUG, near-cognate translation initiation sites in the 5′ leader of annotated protein-coding genes, and computationally extended to the first termination site encountered in the 5′ leader (upstream-terminated ORFs) or CDS (overlapping ORFs). Predicted ORFs were then screened through a series of heuristic filters including: 1) minimum RPF coverage in the 5′ leader, 2) minimum mRNA-seq coverage in CDS, 3) in-frame N-terminal extensions, 4) redundant isoforms, 5) minimum length with optimal RPF coverage, 6) sufficient SPECtre signal, and 7) removal of ambiguously annotated protein-coding transcripts.

Fig. 3

Characterization and validation of predicted uORFs. a The number of genes with at least one predicted ORF (bar plot) in the 5′ leader of evaluated protein-coding genes. The number of genes with a predicted ORF terminated upstream in the 5′ leader only (orange), terminated in the CDS only (blue), or with both a predicted upstream- and CDS-terminated ORF (overlap). b Distribution of predicted ORF translation initiation position relative to the annotated protein-coding CDS start site. c Distribution of predicted ORF lengths. d Distribution of uORF translation start sites (TIS). AUG represents all AUGs predicted by SPECtre, or upstream/downstream 30-nt from the SPECtre predicted start site if no intervening stop codon is present. Near-cognate start codons are utilized in the majority of uORFs, while AUG is the single most common start site. e Plots show mRNA reads (grey) and RPF counts (cyan/gold) for ARF4. The annotated uORF is characterized by the presence of consistent RPF coverage in the 5′ leader. f Schematic of the uORF nanoluciferase (nLuc) reporters used in this study. GGG-nLuc serves as a negative control, as its AUG initiation start codon is mutated to a GGG codon. This reporter supports very little translation. A table of the predicted start sites for each uORF reporter. g nLuc assays performed in SH-SY5Y cells confirmed expression of these uORFs (teal). 5′ leaders not included in the uORF dataset (black) are below the GGG-nLuc reporter activity and considered not translated. All values are normalized to the GGG-nLuc control performed in parallel, data for individual reporters was collected in triplicate in multiple experiments. Student’s t test, all teal uORFs in panel F) have a p value ≤0.0001. Graph represents mean ± S.E.M. h Frameshifting the uORF relative to nLuc decreases translation of the reporter. The reporter was cloned so that the nLuc tag was frameshifted (f.s.) out of frame with the predicted uORF and the CDS start site. n = 3, Student’s t test, ****p ≤ 0.0001. Graph represents mean ± S.E.M.

Fig. 4

uORFs shift translationally with Retinoic acid induced differentiation. a K-means clustering analysis of log₂(uORF TE) in Non-Diff and RA-Diff cells, reveals differentiation-associated shifts. Three clusters of uORF translation emerge: those that are up-regulated in RA-Diff cells (cyan), up-regulated in Non-Diff cells (gold), and uORFs with no change in translational potential (gray). b Clustering in (A) does not correlate with directional CDS changes. Kernel density estimation analysis of changes in TPM over annotated protein-coding CDS as a function of changes in TPM over predicted upstream-initiated ORFs. Cluster identity of predicted ORF changed in translational potential as scored by SPECtre predicted ORFs enriched for translation in RA-Diff cells (cyan), predicted ORFs with enriched translation in Non-Diff cells (gold), and those with static translation across the two conditions (black) are annotated to protein-coding CDS with higher RPF abundance in Non-Diff cells (above horizontal line), and those with higher RPF abundance in RA-Diff cells (below horizontal line). c Analysis of transcripts with a cORF reveals a positive correlation of cORF TE and CDS TE. Pearson correlation, R² = 0.13. d The same is true for oORFs with a Pearson correlation, R² = 0.59.

Fig. 5

Impact of highly translated uORFS on coding sequence translation. a SPECtre identified uORFs were filtered to include only uORFs that have significant coverage in all four Non-Diff and RA-Diff libraries; these are considered highly translated. b Average TE values for cORFs and oORFs in the Non-Diff (left) and RA-Diff (right) conditions. c Conservation analysis of annotated 5′ leaders in all three reading frames (orange), annotated CDS regions over all three frames (grey), predicted AUG-initiated uORFs (dark blue), and predicted non-AUG uORFs (light blue). d Average GC nucleotide content is shown for 5′ leader regions (orange), CDSs (grey), AUG uORFs (dark blue), and non-AUG uORFs (light blue). For oORFs, only the 5′ leader region of the oORF is included. 5′ leaders are significantly more GC rich than both AUG uORFs and non-AUG uORFs, p = 5.72e-12 and 1.54e-07, respectively. Non-AUG uORFs are significantly more GC rich than CDSs and AUG uORFs, p = 7.92e-18 and 2.16e-06. e Average TE for AUG uORFs and non-AUG uORFs reveals no difference between the two subtypes. f Empirical cumulative distribution of TE in all CDSs (black) versus CDSs from transcripts with two subsets of uORFs: those with an AUG initiation site (red) and those with a non-AUG initiation site (Blue). g Empirical cumulative distribution of TE in all CDSs (black) versus CDSs from transcripts with two subsets of non-AUG uORFs: those in the highest quartile of conservation (Conserved, red) and those in the lowest quartile of conservation (Non-Conserved, Blue). Distributions are significantly different with p-values annotated on graph. h GC content of non-AUG uORFs grouped by conservation. All versus conserved: p = 5.89e-08, all versus non-conserved: p = 2.54e-06, conserved versus non-conserved: p = 1.62e-18.

Fig. 6

Ribosomal transcripts are enriched in uORF datasets. a Top: Chart shows the percent of total actively translated ribosomal protein transcripts (n = 70) with a predicted uORF (n = 59). Bottom: Chart shows the percent of ribosomal transcripts with uORFs that are in the highly translated dataset (n = 23). Ribosomal transcripts are enriched in both sets by Fisher’s exact test, ****p > 0.0001. b nLuc assay in SH-SY5Y cells transfected with uORF reporters for ribosomal transcripts: RPS8 and RPS18. c Immunoblot of RPS8 and RPS18 reporters show an increase in molecular weight relative to AUG-nLuc control, confirming translation initiation within the 5′ leader of these ribosomal transcripts. d Removal of the 5′ leader portion of the RPS24 uORF from nLuc reporters for the RPS24 CDS increased nLuc signal relative to reporters with the WT RPS24 5′ leader.

Fig. 7

uORFs buffer against differentiation-induced shifts in CDS TE. a Analysis of the relationship between cORF and CDS translation (log₁₀TE Non-Diff/RA-Diff) for the highly translated dataset reveals that the translational efficiency of these two regions positively correlate in response to RA-Differentiation, R² = 0.08. Regression coefficients were calculated from untransformed TE ratios. b This was also seen for oORFs, R² = 0.18. C-E) 40% (118/295, see Additional file 5: Figure S5) of uORF/CDS pairs in the highly translated dataset exhibit TE shifts in the opposite direction for the uORF and the CDS with differentiation (“inverse”), while 60% of uORF/CDS pairs exhibit TE shifts in the same direction with differentiation (“positive”). c Relative proportion of “positive” or “inverse” oORFs and cORFs. Chi-square, p = 0.038. d Distribution of uORFs by length. e Start site codon distribution for “positive” or “inverse” uORFs. f uORFs with an “inverse” relationship to their associated CDS have a higher average TE than those with a “positive” relationship, p = 0.00923. g Histograms of log₂(CDS TE, RA-Diff/TE Non-Diff) for transcripts with a uORF in the highly translated set (uORF) or no uORF (none). ANOVA, p = 0.000389.