Translation exhibits bursts with low average ribosome density.

(A) Schematic of the SunTag system used for single-imaging of translation. (B) SunTag reporters differing in their coding sequence inserts: no insert, PPG (proline-rich), AAG (alanine-rich) and Renilla. (C) Representative live-cell image of a HeLa cell expressing the PPG reporter, showing JF646 (mRNA), GFP (nascent peptide), and merged signals. The image is part of a larger field of view, acquired with a 60X objective in spinning-disk confocal set-up. Scale bar, 10 μm. (D) Representative long traces for each reporter, showing stable JF646 (mRNA, red) and fluctuating GFP (SunTag, green) intensities over time. (E) Average fraction of translated traces for each reporter. Colored dots represent averages from individual experiments; black dots with bars show the mean ± SEM across multiple experiments (n traces, n experiments): no-insert (669, 4), PPG (440, 2), AAG (1235, 6), Renilla (804, 5). (F) Example of the SunTag signal of a translated mRNA (no-insert), highlighting translated periods (shaded yellow). (G) Average duration of translated periods (left) and untranslated periods (right) for traces longer than 20 min. Colored dots represent averages from individual experiments; black dots with bars show the mean ± SEM across multiple experiments (n traces, n experiments): no-insert (41, 4), PPG (45, 2), AAG (90, 5), Renilla (62, 5). (H) Estimated average ribosome density (percentage of transcript covered by ribosomes) for traces in (G). Significance tests performed with Mann-Whitney U test.

Figure 1—figure supplement 1. SunTag traces in control conditions.

Figure 1—figure supplement 2. Experimental measurement of mature protein intensities to estimate number of translating ribosomes.

Figure 1—video 1. Time-lapse imaging of HeLa cells expressing the AAG reporter.

TASEP-based inference of translation dynamics from single run-off traces.

(A) Schematic of the Hidden Markov Model (HMM) used to analyze run-off experiments. α = initiation rate, λ = average elongation rate, L = reporter length, ⟨N⟩ t=0 = average number of ribosomes at t = 0 (60 s after HT addition). (B) Representative slow (top) and fast (bottom) decaying traces from one run-off experiment with the PPG reporter; intensity in green and predicted number of ribosomes in red. C) Comparison between the analytical correction factor γ (black line, Methods Equation 19) and the results obtained by numerical simulations of run-off traces (grey dots). The latter represent (I(N, t) − b0)/NiMP averaged over 2000 simulated run-off traces, with I(N, t) simulated intensity, b0 offset, N number of ribosomes and iMP mature protein intensity. Kinetic parameters used in simulations: initiation α = 1/60 s−1, elongation λ = 3.0 aa/s. D) Comparison between the analytical correction factor ν (black line, Methods Equation 21) and the results obtained by numerical simulations of run-off traceseLife (coloured dots). The latter represent the normalized intensity variance for different values of N, same simulated traces as in C). E) Relative error on model parameters (α, λ, iMP) vs average ribosome density ρ, for different values of the noise parameter s0 (Methods Equation 5), for simulated run-off traces. Each data point is obtained with a different combination of λ ∈ {0.5, 1, 3, 5} aa/s and α ∈ {1/120, 1/60, 1/30} s−1. Blue dots represent the relative error on each parameter, the black curve is a moving average, and the dashed line indicates the density ρ at which the relative error goes beyond 0.2. Simulation parameters in C) – E): iMP = 10 a.u., reporter length L = 1066 aa, ribosome size ℓ = 10 aa.

Figure 2—figure supplement 1. SunTag run-off traces in control conditions after HT addition.

Figure 2—figure supplement 2. TASEP-based inference accurately estimates ribosome density on simulated data despite correlations between α and iMP.

Figure 2—video 1. Time-lapse imaging of HeLa cells expressing the PPG reporter after harringtonine treatment.

Figure 2—video 2. Time-lapse imaging of HeLa cells expressing the AAG reporter after harringtonine treatment.

Low ribosome density arises from coordinated translation initiation and elongation.

In all panels the error bars indicate 95% confidence intervals. (A) Run-off time distribution (top) and percentage of incomplete run-offs (bottom) for each reporter. For incomplete runoffs we include the total duration of the trace, as it represents a lower-bound to the total run-off time (red dots). (B) Inferred elongation rates (λ). (C) Inferred initiation rates (α). (D) Inferred average ribosome density. (E) Correlation between elongation and initiation rates for each experiment. The shaded background indicates approximate ribosome density regimes, ranging from low density (light yellow) to high density (dark yellow), as shown by the adjacent color bar. ρ indicates the average density ± standard deviation, r is the Pearson correlation coefficient.

Figure 3—figure supplement 1. Global measurement noise estimation from CHX traces.

Figure 3—figure supplement 2. HMM approach predicts lower elongation speed than simple linear regression and suggests the presence of bursting and stalling.

eIF5A perturbations minimally affects ribosome density and translational bursting.

(A) Top: western blot of h-eIF5A levels in control cells and cells treated with 1 μM and 10 μM GC7 for 24 hours, with 1 mM AG. Bottom: quantification of h-eIF5A signal relative to total eIF5A. Error bars represent standard deviation between replicates. (B) Total eIF5A and h-eIF5A expression in CRISPR/Cas9 knock-out (KO) clones compared to wild-type (WT) HeLa cells. (C) Representative live-cell images of a HeLa cell expressing the PPG reporter after 24-hour treatment with 1 μM GC7 and 1 mM AG, showing JF646 (mRNA, red), GFP (SunTag, green), and merged signals. The image is part of a larger field of view, acquired with a 60X objective in a spinning-disk confocal set-up. Scale bar, 10 μm. (D) Same as (C) for eIF5A KO cells. (E) Representative traces of the PPG reporter under control conditions, GC7 treatment, and eIF5A KO, showing GFP (SunTag, green) and J646 (mRNA, red) intensities over time. (F) Fraction of translated mRNAs (n traces, n experiments): no-insert CTRL (669, 4), GC7 (511, 3); PPG CTRL (440, 2), GC7 (392, 2); AAG CTRL (1235, 6), GC7 (900, 5); Renilla CTRL (804, 5), GC7 (1476, 6); PPG WT (396, 2), KO (343, 2). (G) Average duration of translated periods and (H) untranslated periods for traces longer than 20 min. (I) Average ribosome density during the translated periods analyzed in (G). (G – I) (n traces > 20 min, n experiments): no-insert CTRL (41, 4), GC7 (20, 3); PPG CTRL (45, 2), GC7 (28, 2); AAG CTRL (90, 5), GC7 (26, 4); Renilla CTRL (62, 5), GC7 (112, 5); PPG WT (33, 2), KO (26, 2). Coloured dots represent averages from individual experiments, black dots with bars show the mean ± SEM across multiple experiments. Mann-Whitney U significance test.

Figure 4—figure supplement 1. SunTag traces in perturbed conditions.

Figure 4—video 1. Time-lapse imaging of HeLa cells expressing the PPG reporter after 24h GC7 treatment.

Figure 4—video 2. Time-lapse imaging of EIF5A KO HeLa cells expressing the PPG reporter.

eIF5A-driven elongation changes induce initiation regulation to preserve ribosome density.

(A) Fraction of translated mRNAs over time in control (solid lines) and perturbed conditions (GC7 treatment and eIF5A KO, dashed lines) for each reporter. Thick lines represent the mean across experiments, shaded areas represent standard deviation between experiments. Full circles and squares represent the fraction of mRNAs still translated after 15 min from HT addition (control and perturbed conditions, respectively). (n is the number of experiments) (B) Fraction of translated traces showing slow run-off (still translated after 15 min of HT treatment). (C) Run-off time distribution (top) and percentage of incomplete run-offs (bottom) for each reporter in control conditions (circles) and perturbed conditions (GC7 treatment and eIF5A KO, squares). For incomplete runoffs we include the total duration of the trace, as it represents a lower-bound to the total run-off time (red dots). Number of translated traces (control, perturbed): no-insert (58, 35), (22, 24); PPG (45, 67); AAG (93, 80), (20, 32); Renilla (24, 24), (24, 40); PPG WT/KO (39, 28), (16, 25). (D) Inferred elongation rates. (E) Inferred initiation rates. (F) Inferred average ribosome density. (G) Representation of changes in translation kinetic parameters upon eIF5A perturbation in the initiation-elongation (α – λ) plane. The shaded background indicates approximate ribosome density regimes, ranging from low density (light yellow) to high density (dark yellow), as shown by the adjacent color bar. Significance tests performed with Mann-Whitney U test.

Figure 5—figure supplement 1. SunTag run-off traces in perturbed conditions after HT addition.

Figure 5—figure supplement 2. Inferred mature protein intensities are in good agreement with the experimental measurement.

Figure 5—video 1. Time-lapse imaging of HeLa cells expressing the PPG reporter after 24h GC7 treatment and harringtonine addition.

Figure 5—video 2. Time-lapse imaging of HeLa cells expressing the AAG reporter after after 24h GC7 treatment and harringtonine addition.

Figure 5—video 3. Time-lapse imaging of EIF5A KO HeLa cells expressing the PPG reporter after harringtonine addition.

Technical specification of the live-cell imaging set-up.

SunTag traces in control conditions.

(A) Representative SunTag traces for different reporters in control conditions. (B) Duration of translated traces in control conditions (in black, average between experiments ± SEM).

Experimental measurement of measuring mature protein intensities to estimate number of translating ribosomes.

A) Diffraction-limited spots in puromycin-treated cell expressing SunTag-Renilla-RH1-MS2. Black, yellow and red arrows indicates spots of different sizes and intensities; the image is part of a larger field-of-view, acquired at 100% laser power, with a 60X oil-objective. Scale bar: 10 μm. B) Histogram showing the average intensity distribution of 31 spots of lower intensity, after puromycin treatment, at 100% laser power. C) Intensity at increasing laser power of CHX-treated mRNAs. (D) Number of ribosomes per translated Operiod in traces longer than 20 min (in black, average between experiments ± SEM).

SunTag run-off traces in control conditions after HT addition.

(A) Representative SunTag traces for different reporters in control conditions after HT addition.

TASEP-based inference accurately estimates ribosome density on simulated data despite correlations between α and iMP.

A) Agreement of γ(t) (top) and ν(t) (bottom) with simulations for different values of λ, with α = 1/30s−1. While λ decreases (left to right) the average ribosome density ρ increases and the agreement between ν(t) and simulations significantly deteriorates. B) Correlation between the relative error on α and the relative error on iMP, for L = 1066aa and L = 714aa and s0 = 0.4. Black and red dots represents data with ρ > 0.10 and ρ ≤ 0.10, respectively. The Pearson correlation coefficient r and the p-value P is indicated for all data (n = 12 simulations) and the black line shows the linear fit. The initiation rate α and on the mature protein intensity iMP are negatively correlated, but at low densities (ρ < 0.10) the mature protein intensity is inferred more accurately and the correlation is less important. Simulation parameters: λ ∈ {0.5, 1, 3, 5} aa/s, α ∈ {1/120, 1/60, 1/30} s−1, iMP = 10 a.u. C) Inferred average density ρ = αℓ/λ at the beginning of run-off vs true average density as measured in simulations (s0 = 0.4). The black line shows the y = x line. In all panels error bars represent 95%-confidence intervals.

Global measurement noise estimation from CHX traces.

A) Examples of GFP intensity traces after CHX treatment (black) and corresponding mRNA signal (grey). B) Autocorrelation K(τ) normalized by the mean intensity square ⟨I⟩ 2 for each CHX-treated trace (black lines) and median over all traces (red line), for different time delays τ. K(0)/ ⟨I⟩ 2 averaged over 236 traces yields an estimate of the multiplicative noise factor.

HMM approach predicts lower elongation speed than simple linear regression and suggests the presence of bursting and stalling..

(A) Harringtonine run-off traces of PPG and Renilla (2 acquisitions). The black dashed line represents a linear regression of the average intensity decay (red). Vertical lines indicates the total elongation time as predicted by the linear fit (black) or by the HMM (blue). (B) Elongation rate predicted by the HMM vs elongation rate predicted by the linear-regression approach. (C) Q-Q plot of the waiting times between termination events, compare to an exponential distribution with mean the inverse initiation rate (as inferred by the HMM). Each plot represents one acquisition, and each dot the time between two termination events. The dashed line represents the total elongation time as inferred by the HMM model, and the full line the bisector. Waiting times lying above the dashed line are likely to be strong stalling events.

SunTag traces in perturbed conditions.

(A) Representative SunTag traces for different reporters in perturbed (GC7/EIF5A KO) conditions. (B) Duration of translated traces in perturbed conditions (average between experiments ± SEM).

SunTag run-off traces in perturbed conditions after HT addition.

(A) Representative SunTag traces for different reporters in perturbed conditions (GC7/EIF5A KO) after HT addition.

Inferred mature protein intensities are in good agreement with the experimental measurement.

A) Representative PPG traces in control and GC7-treated conditions, together with the inferred number of ribosomes. B) Mature protein intensity iMP for each experiment, subdivided by reporter. The red dashed line indicates the mature protein intensity as measured by an independent calibration experiment and the filled area the uncertainty on the measurement. C) Initiation rate vs mature protein intensity, for both control and perturbed conditions (n = 10 experiments): α initiation rate and iMP mature protein intensity.