Translational buffering by ribosome stalling in upstream open reading frames

Abstract

Upstream open reading frames (uORFs) are present in over half of all human mRNAs. uORFs can potently regulate the translation of downstream open reading frames through several mechanisms: siphoning away scanning ribosomes, regulating re-initiation, and allowing interactions between scanning and elongating ribosomes. However, the consequences of these different mechanisms for the regulation of protein expression remain incompletely understood. Here, we performed systematic measurements on the uORF-containing 5' UTR of the cytomegaloviral UL4 mRNA to test alternative models of uORF-mediated regulation in human cells. We find that a terminal diproline-dependent elongating ribosome stall in the UL4 uORF prevents decreases in main ORF protein expression when ribosome loading onto the mRNA is reduced. This uORF-mediated buffering is insensitive to the location of the ribosome stall along the uORF. Computational kinetic modeling based on our measurements suggests that scanning ribosomes dissociate rather than queue when they collide with stalled elongating ribosomes within the UL4 uORF. We identify several human uORFs that repress main ORF protein expression via a similar terminal diproline motif. We propose that ribosome stalls in uORFs provide a general mechanism for buffering against reductions in main ORF translation during stress and developmental transitions.

Fig 1. Models of uORF regulation considered in this study.

(A) Constitutive repression. The uORF constitutively siphons away a proportion of scanning ribosomes from the main ORF. (B) 80S-hit dissociation. Elongating ribosomes that collide with 3′ scanning ribosomes cause the leading scanning ribosome to dissociate from the mRNA. (C) Queuing-mediated enhanced repression. Scanning or elongating ribosomes form a queue behind a 3′ stalled elongating ribosome. If the queue correctly positions a scanning ribosome at the uORF start codon, then the proportion of scanning ribosomes that initiate translation at the uORF start codon increases. (D) Collision-mediated 40S dissociation. Scanning ribosomes that collide with a 3′ stalled ribosome dissociate from the mRNA. (E) Regulated re-initiation. Ribosomes initiate translation at the first uORF start codon, and scanning continues after termination at the stop codon of the first uORF. Ribosomes re-initiate at the main ORF start codon or the second downstream uORF start codon when phosphorylated eIF2α levels are high or low, respectively. The schematic is depicted in a low phosphorylated eIF2α state.

Fig 2. An experimental and computational platform for assessing uORF-mediated regulation of main ORF translation.

(A) The 236 nt 5′ UTR of UL4 mRNA from human cytomegalovirus contains 3 uORFs. The terminal proline and stop codons of uORF2 at which the P- and A-sites of the stalled ribosome are positioned are highlighted as uORF2 stall. (B) A dual-luciferase reporter system for measuring 5′ UTR repressiveness in HEK293T cells. FLuc signal serves as an internal control for transfection efficiency. (C) The reporter system recapitulates the known elongating ribosome stall-dependent repression of protein expression by the UL4 uORF2 [21]. The indicated mutations improve the uORF2 Kozak context (ACCATGG instead of GTGATGC), remove the start codon (ACC instead of ATG), or remove the elongating ribosome stall by mutating the terminal proline codon to an alanine codon (GCT instead of CCT). Error bars show standard error of mean NLuc / FLuc ratios over 3 biological replicates. Data are normalized to the no-uORF start codon control. (D) Computationally predicted uORF regulation in the 80S-hit dissociation, queuing-mediated enhanced repression, and collision-mediated 40S dissociation models. Data are normalized to the no-uORF start codon control. The parameter combination that best recapitulated the control behavior in Fig 2C is displayed in Table 1. Error bars for simulated data are smaller than data markers.

Fig 3. Modeling workflow.

(A) Molecules in the kinetic model. Molecules have components each of which has state values or binding sites for other molecules (called bond in BioNetGen). Components are abbreviated in parentheses by how they are referenced in the model specification. For example, the mRNA (M) initiation footprint (c₁ to c_n where n is equal to the ribosome footprint size in nt) can either be clear of ribosomes, and therefore free for a PIC_43S loading reaction to occur, or blocked by a ribosome, preventing this reaction. (B) Visual representations of the reactions in the kinetic model. Re-initiation necessitates several additional reactions. PIC_43S formation (R_40S binding TC) can occur if the R_40S is bound to the mRNA; this TC re-binding is required for start codon selection competence. R_40S molecules can scan forward or backward. Some reactions in the kinetic model, such as different types of collision and dissociation reactions, are not depicted here.

Fig 4. Kinetic modeling predicts translational buffering by uORFs.

Buffering refers to a smaller than expected decrease (small positive slope), or even increase (negative slope), in main ORF protein output with reduced ribosome loading. (A) The constitutive repression model, without an elongating ribosome stall, has no buffering behavior. uORFs simply siphon away scanning ribosomes from the main ORF. (B) Buffering in the 80S-hit dissociation model depends on uORF initiation and re-initiation frequencies [38]. For buffering to occur in this model, uORFs must initiate well enough to have elongating ribosomes hit 3′ scanning ribosomes (yellow-green line). uORFs must also not continue scanning at high frequencies following termination (left panel); frequent continuation of scanning coupled with high uORF initiation allows many scanning ribosomes to make it to the main ORF. Buffering occurs better for longer uORFs that have more time for elongating ribosomes to hit 3′ scanning ribosomes (S2A Fig, yellow-green line). Here, the uORF is 100 codons long. The dissociation rate is 200s⁻¹, so 99% of scanning ribosomes hit by 5′ elongating ribosomes dissociate rather than continue scanning. The scanning rate is 2 nucleotides/s, and the elongation rate is 2 codons/s. There is no elongating ribosome stall in this model. (C) Buffering in the queuing-mediated enhanced repression model depends on d_stall: the distance between the uORF start codon and elongating ribosome stall. In this model, uORF initiation can increase above baseline with increased ribosome loading when d_stall is an integer multiple of the ribosome footprint (30 nt, left panel). When this condition is met, buffering occurs. For d_stall values of 60 and 63 nt, the uORF length is 21 and 22 codons, respectively. (D) Buffering in the collision-mediated 40S dissociation model depends on the dissociation rate. Here, d_stall is 63 nt; with a low dissociation rate, this model reduces to the queuing-mediated enhanced repression model. (E) Buffering in the regulated re-initiation model depends on uORF initiation and continued scanning frequencies. For buffering to occur, several conditions must be met. At least 2 uORFs are required, both of which must be well-translated (yellow-green line). Continued scanning following termination at the first uORF must be frequent, and continued scanning following termination at the second downstream uORF must be rare (right panel). The second downstream uORF is 3 codons long. There is no elongating ribosome stall in this model. uORFs are located 25 nt from the 5′ cap. 99% of scanning ribosomes that make it to the main ORF will initiate translation; 1% will leaky scan. Unless otherwise stated, parameters (Table 1) obtained from calibrating models to reporter measurements on wild-type or mutant uORF2 (Fig 2C) are used here. Ribosome loading is the k_{cap bind} rate for non-regulated re-initiation models. We model changes in ribosome loading via changes in k_{cap bind} as that rate is easier to match to in vivo estimates of ribosome loading. However, buffering in the regulated re-initiation model is dependent on an eIF2α phosphorylation mechanism; we instead vary the number of ternary complexes in this model. Error bars of simulated data are smaller than data markers.

Fig 5. The human cytomegaloviral uORF2 buffers against reductions in main ORF protein output.

The human cytomegaloviral UL4 uORF2 is used in the dual-luciferase assay (Fig 2B) in conjunction with three experimental strategies to reduce ribosome loading. (A) Ribosome loading is reduced using stem-loops [76] with the indicated GC percentages. All stem-loops are positioned 8 nt from the 5′ cap and have the same predicted stability of -30 kcal/mol. The no-stem-loop construct has a CAA repeat instead of a stem-loop. The 5′ UTR is 287 nt long. Data are normalized to a no-uORF start codon control without a stem-loop. (B) Ribosome loading is reduced using the drug thapsigargin (1X = 1 μM) [77], which induces the integrated stress response (ISR) by triggering ER stress. NLuc has a C-terminal PEST tag to turnover [78] of protein produced prior to the 6-hour drug treatment. The 5′ UTR is 236 nt long. Data are normalized to a no-uORF start codon control without a PEST tag. Error bars show standard error of mean NLuc / FLuc ratios over 4 biological replicates. (C) Ribosome loading onto the uORF2-NLuc portion of the transcript is reduced using a 5′ synthetic uORF: ATG GGG TAG. The synthetic uORF Kozak is varied to alter ribosome loading. The variants are vertically ordered by the no-stall means. The 5′ UTR is 262 nt long. Data are normalized to a no-uORF start codon control without a synthetic uORF. Right panels in A,B, C show wild-type (WT) mean values normalized by the corresponding no-stall values. The no-stall uORF2 mutants lack their terminal diproline motifs (P22A mutation). Unless stated otherwise, error bars show standard error of mean NLuc / FLuc ratios over 3 biological replicates.

Fig 6. Changes to the distance between the human cytomegaloviral uORF2 start codon and the elongating ribosome stall do not change repressiveness or buffering, consistent with the collision-mediated 40S dissociation model.

(A) Computational modeling predicts greater changes in uORF repressiveness with changes in d_stall in the queuing-mediated enhanced repression model. Fast backward scanning abolishes this periodicity. d_stall refers to the distance between the start codon and the elongating ribosome stall. As backward scanning increases in rate (moving right along panels), the collision-mediated enhanced repression model loses periodicity (middle panel, purple line) before the queuing-mediated enhanced repression model (right panel, yellow-green line). Parameters that best recapitulated reporter measurements on wild-type or mutant uORF2 (Fig 2C and Table 1) are used here. The forward scanning rate is 5 nucleotides/s. Data are normalized to a no-uORF start codon control. Error bars of simulated data are smaller than data markers. (B) Experimentally varying the distance between the human cytomegaloviral uORF2 start codon and the elongating ribosome stall does not systematically affect its repression of main ORF protein output. The human cytomegaloviral UL4 uORF2 is used in the dual-luciferase assay (Fig 2B) in conjunction with various length inserts from the N-terminus of the EYFP main ORF. The EYFP main ORF sequence is inserted directly 3′ to the uORF2 start codon. The added sequence increases the distance between the uORF2 start codon and the elongating ribosome stall. The bottom three controls improve the uORF2 Kozak context, remove the start codon, and remove the elongating ribosome stall. Error bars show standard error of mean NLuc / FLuc ratios over 3 biological replicates. Data are normalized to the no-uORF start codon control. (C) Experimentally varying the human cytomegaloviral uORF2 d_stall does not strongly regulate the capacity of buffering against reductions in main ORF protein output. Ribosome loading is reduced with a 5′ synthetic uORF: ATG GGG TAG. The no-stall uORF2 mutants lack their terminal diproline motifs (P22A mutation). No-synthetic uORF mutants (ATG to AAG) are depicted by transparent, gray bars with red Xs and have a higher relative ribosome loading rate onto the uORF2-NLuc portion of the transcript. The distance between the uORF2 start codon and the elongating ribosome stall is varied as indicated by adding 6 nt, GTC AGC, from the N-terminus of the EYFP main ORF. Data are normalized to a no-uORF start codon control without a synthetic uORF.

Fig 7. Several human uORFs have repressive terminal diproline motifs.

Terminal diproline motif-containing human uORFs are used in the dual-luciferase assay (Fig 2B). The terminal proline codon in each uORF is mutated to an alanine codon in the P to A mutant. Start codons are mutated to ACC for the no-AUG mutants. P values comparing the indicated mutants to the wild-type are from a two sample t-test: * (0.01 < P < 0.05), ** (0.001 < P < 0.01), *** (P < 0.001). Error bars show standard error of mean NLuc / FLuc ratios over 5 biological replicates. Data are normalized to a no-UL4-uORF2 start codon control.