eIF1 and eIF5 dynamically control translation start site fidelity

Abstract

Human translation initiation requires accurate recognition of translation start sites. While AUG codons are canonical start sites, non-AUG codons also are used, typically with lower efficiency. The initiator tRNA and initiation factors eIF1 and eIF5 control recognition. How they distinguish different start sites, yet allow flexible recognition remains unclear. Here, we used real-time single-molecule assays and an in vitro reconstituted human system to reveal how eIF1 and eIF5 direct start site selection. eIF1 binds initiation complexes in two modes: stable binding during scanning, followed by transient, concentration-dependent rebinding after start site recognition. Termination of eIF1 rebinding requires transient and concentration-dependent binding by eIF5, which allows formation of translation competent ribosomes. Non-AUG start sites differentially stabilize eIF1 and destabilize eIF5 binding, blocking initiation at multiple points. We confirmed these opposing effects in human cells. Collectively, our findings uncover that eIF1 and eIF5 directly compete to bind initiation complexes and illuminate how their dynamic interplay tunes the fidelity of start site recognition, which has broad connections to health and disease.

INTRODUCTION

Human translation initiation canonically requires an AUG codon at the start of the coding region of a messenger RNA (mRNA). The initiation machinery recognizes the start site with single-nucleotide precision, which maintains the proper reading frame and avoids synthesis of truncated and potentially toxic peptides. Yet, initiation also can proceed via start sites that differ by a single nucleotide (e.g., CUG, UUG, GUG)^1,2. These non-AUG start sites typically yield less synthesized protein and have regulatory roles^3–5. How the translation initiation machinery distinguishes different potential start sites and the mechanistic basis for the decreased efficiency on alternative start sites remain unclear.

Translation initiation is directed by eukaryotic initiation factors (eIFs) and the initiator methionyl-tRNA_i^Met (Met-tRNA_i^Met). The process starts when a 43S initiation complex – the small (40S) ribosomal subunit bound by eIF1, eIF1A, eIF3, and the eIF2–GTP–Met-tRNA_i^Met ternary complex – loads onto the mRNA, which is facilitated by eIF4 proteins bound near the 5’-m⁷G cap^6–8. While directionally moving 5’ to 3’ along the 5’UTR, the complex scans for translation start sites^9–11, with selection enhanced by favorable Kozak sequence context¹². A start site correctly placed at the ribosomal P site and base paired with the anticodon of Met-tRNA_i^Met remodels the complex from a scanning-permissive conformation (P_scan) to a scanning-arrested one (P_in)^9,10,13–15. While the timing remains unknown, the remodeling permits eIF5-stimulated hydrolysis of GTP and release of eIF1 and inorganic phosphate^16–19, which triggers eIF2 departure. To complete initiation, eIF5B binds and collaborates with eIF1A to reposition Met-tRNA_i^Met and allow the large (60S) ribosomal subunit to join and form the 80S ribosome^20–24.

eIF1 and eIF5 play a central role during the unimolecular recognition of the start site by the anticodon of Met-tRNA_i^Met on the initiation complex. In yeast, genetic and biochemical studies indicate that eIF1 discriminates ideal from suboptimal translation start sites; increased eIF1 levels suppress usage of non-AUG start sites, whereas decreased eIF1 levels or activity enhance their usage^25–33. For eIF5, the opposite likely occurs; increased eIF5 activity enhances initiation on non-AUG sites^27,34, perhaps by increasing the rate of eIF1 release from initiation complexes¹⁷. Similar effects may occur in mammalian systems^35–38. These apparent concentration-dependent effects of eIF1 and eIF5 may be due to improper formation of 43S initiation complexes prior to their loading onto the mRNA^{29,30,39–42}. However, the presence and role of eIF5 during formation of human 43S initiation complexes remains ambiguous. Moreover, human initiation complexes that lack eIF1 improperly stall upstream of the start site, which can be rescued by subsequent addition of eIF1^43,44. The concentration dependence of eIF1 and eIF5 effects therefore also could be explained by an alternate model where eIF1 and eIF5 dynamically bind and release initiation complexes to guide start site selection using distinct mechanisms.

The binding sites for eIF1 and eIF5 within initiation complexes overlap. Prior to recognition of the start site, eIF1 binds initiation complexes proximal to the ribosomal P site and the anti-codon stem and loop of Met-tRNA_i^{Met 45–54}. The presence of eIF1 restricts Met-tRNA_i^Met to a scanning competent (P_scan) conformation. Density for eIF1 disappears upon recognition of the translation start site, which correlates with movement of the 40S head region and Met-tRNA_i^Met to the scanning-arrested (P_in) conformation, partially occluding the eIF1 binding site^{46,47,55–57}. By contrast, eIF5 is not observed in structures of initiation complexes prior to recognition of the start site^45–54. Its absence could be due to structural flexibility or absence of the protein within the complex. Once an initiation complex stalls at the start site in the scanning arrested (P_in) conformation, the N-terminal domain of eIF5 occupies the eIF1 binding site and directly contacts the anti-codon stem and loop of Met-tRNA_i^{Met 56,57}. However, the timing of eIF1 and eIF5 binding and release from initiation complexes, the nature of their exchange, and how their potential interplay might underpin recognition of the start site remain unknown.

To define how translation start sites are selected, we examined eIF1 and eIF5 directly during human translation initiation. Using real-time single-molecule Förster resonance energy transfer (FRET) assays and our in vitro reconstituted human initiation system, we monitored eIF1 and eIF5 during initiation on mRNAs with AUG and non-AUG start sites. We reveal unexpected binding dynamics that demonstrate eIF1 and eIF5 compete to bind initiation complexes with opposite outcomes, which we confirmed is conserved in human cells. Our findings provide a biophysical framework that the underlies the fidelity of start site selection and its tunable regulation.

RESULTS

Initiation complexes contain two modes of eIF1 binding

To examine human eIF1 function, we first established a single-molecule FRET signal to monitor eIF1 binding. We labeled the flexible U46 nucleotide of synthetic Met-tRNA_i^Met with Cy3 fluorescent dye (tRNA_i-Cy3, FRET donor) and Cys94 of eIF1 with Cy5 dye (eIF1-Cy5, FRET acceptor) (Extended Data Fig. 1a,b). These two positions are within ~50 Å in structures of initiation complexes^45–54 and yielded the expected FRET efficiency using our reconstituted human system²⁴ (Fig. 1a and Extended Data Fig. 1c–g).

Figure 1. Initiation complexes contain two modes of eIF1 binding.

(a) Structure of a human translation initiation complex prior to recognition of the translation start site (PDB ID: 6ZMW)⁴⁸. The positions of U46 on Met-tRNA_i^Met (green) and C94 of eIF1 (red) are indicated, which are separated by about 50 Å. (b) Schematic of the real-time single-molecule human translation initiation assay. (c) Example single-molecule fluorescence data where tRNA_i-Cy3 (green), eIF1-Cy5 (red), and eIF5B-Cy3.5 (orange) were monitored during translation initiation on β-globin mRNA. (d) Heat map of tRNAi-to-eIF1 FRET signal on all analyzed complexes synchronized to its initial appearance. (e, f, h, i) Cumulative probability plots of the indicated parameters (defined as in panel c). Lines represent fit to exponential functions, which we used to derive the reported rates and mean times. In all experiments here, eIF1-Cy5 was present at a final concentration of 40 nM (by Cy5) and unlabeled eIF5 was present at 290 nM. See Supplementary Table 1 for all rates, rate constants, and the number of complexes and binding events analyzed in each experiment. (g) Plot of observed eIF1 reassociation rate constants at the indicated concentrations. Error bars represent the 95% confidence interval (CI) of the observed association rate constants. Line represents fit via linear regression analysis to derive the indicated bimolecular rate.

We next monitored eIF1 continuously during human translation initiation using a real-time single-molecule FRET assay. We assembled 43S initiation complexes that contained the eIF2-GTP-tRNA_i-Cy3 ternary complex (green) and eIF1-Cy5 (red). Separately, full-length human β-globin mRNA with a 5’-m⁷G cap was tethered to an imaging surface within zero-mode waveguides (ZMWs), which facilitate four-color fluorescence detection (Fig. 1b)⁵⁸. Saturating concentrations of eIF4 proteins (4A, 4B, 4E, 4G), ATP, and GTP were added to the tethered mRNA. Upon start of data acquisition, a mixture containing the 43S initiation complex and 60S subunits was added to the surface. We also included eIF5B-Cy3.5 (gold) to demarcate initiation complexes that successfully progressed to late initiation steps²⁴. In these experiments, the presence of tRNA_i-Cy3 (and thus the 43S initiation complex) and eIF5B-Cy3.5 was indicated by direct excitation and emission of the conjugated dyes, whereas eIF1-Cy5 binding was indicated by tRNA_i-to-eIF1 FRET (Extended Data Fig. 2a). Dye labeling of synthetic tRNA_i, which lacks native modifications, and eIF1 did not affect translation initiation, as tRNA_i-Cy3 and eIF1-Cy5 function identical to their unlabeled versions; the mean elapsed time (10 ± 0.9 s) from loading of the 43S initiation complex (initial Cy3 appearance) until eIF5B bound matched the published time obtained using unlabeled Met-tRNA_i^Met and eIF1 (Extended Data Fig. 2b)²⁴. To facilitate comparisons, we report mean binding lifetimes (i.e., mean duration of a binding event) and rebinding times (i.e., mean time elapsed between binding events). The rates and rate constants used to derive the mean times are aggregated in Supplementary Table 1.

eIF1 first stably bound, dissociated from, and then transiently rebound the same initiation complexes. As expected, eIF1 loaded with the 43S initiation complex onto the mRNA, which was signaled by appearance of tRNA_i-Cy3 to eIF1-Cy5 FRET (Fig. 1c,d). eIF1 remained bound for 2 ± 0.1 s, which we defined as the initial eIF1 binding lifetime (Fig. 1e). Following its departure, eIF1 rebound most initiation complexes (~60%) at least once for 10-fold briefer durations (0.2 ± 0.02 s) (Fig. 1e and Extended Data Fig. 2c). The time elapsed between individual eIF1 rebinding events decreased as the eIF1 concentration increased (9.5 ± 0.04 μM⁻¹ s⁻¹) (Fig. 1f,g and Extended Data Fig. 2d). Furthermore, we observed complete loss of eIF1 signal between binding events when the protein was directly monitored using a 640 nm laser, which directly excites the Cy5 dye conjugated to eIF1 (Extended Data Fig. 2e,f). eIF1 therefore rebinds initiation complexes in a concentration-dependent manner and completely dissociates between all binding events. Final departure of eIF1 occurred 9 ± 0.7 s after initial loading of the 43S initiation complex onto the mRNA, which we defined as the eIF1 binding window (Fig. 1h). Following a delay of 3 ± 0.2 s, eIF5B (at 40 nM) bound the initiation complex to catalyze 60S ribosomal subunit joining (Fig. 1i).

Initiation complexes that load onto an mRNA thus contain two distinct modes of eIF1 binding: an initial stable eIF1 binding event followed by a switch to a transient and concentration-dependent rebinding mode that terminates prior to eIF5B binding. The dramatic change in initial versus subsequent eIF1 binding events indicate a substantial change occurred within initiation complexes between the first and second eIF1 binding events.

Start site recognition triggers switch to transient binding

In the canonical pathway, the 43S initiation complex loads near the 5’-m⁷G cap and scans the 5’UTR to search for the translation start site. We hypothesized that eIF1 remains stably bound throughout scanning, with its initial departure triggered by preliminary recognition of the start site. To test this model, we generated 5’ capped model mRNAs with either 50 nt or 200 nt long 5’UTRs comprised of CAA repeats. This design minimized potential RNA structures and enabled precise control of the start site identity and location without potential near-cognate start codons (Fig. 2a). On both model RNAs, the translation start site was present in ideal Kozak context (ACCAUGGA).

Figure 2. Termination of transient eIF1 binding requires a translation start site and concentration-dependent binding by eIF5.

(a) Cartoon schematic of the model mRNAs with either 50 nt (black), 200 nt (orange) long 5’UTRs or no AUG (purple). (b-e) Cumulative probability plots of the indicated eIF1 binding parameters observed on eIF5B-bound initiation complexes; the exceptions are the No AUG and TC-GDPNP data, which represent complexes that rarely or never progressed to eIF5B binding. Lines represent fits to exponential functions, and the derived mean times are reported on the plots. The 5’UTR length effects on eIF1 kinetics are contrasted in panel b. Panel c shows the effects of removing the AUG start site. Panel d contrasts use of TC-GTP against TC-GDPNP on β-globin mRNA. Panel e contrasts molar excess 290 nM eIF5 against a limiting concentration of 10 nM eIF5. For all conditions, 40 nM of Cy5-eIF1 and 290 nM eIF5 (unless as stated in panel e) were used. Panels b and c plot the same AUG (50 nt) data, and panels d and e plot the same β-globin data with 290 nM eIF5 and TC-GTP present. See Supplementary Table 1 for all rates, rate constants, and the number of complexes and binding events analyzed in each experiment.

eIF1 binding depended on the length of the 5’UTR and the presence of a start site. The initial eIF1 binding lifetime extended by 2-fold on the longer 5’UTR (from 2.7 ± 0.1 s to 5.3 ± 0.1 s) (Fig. 2b). Following the switch to transient eIF1 rebinding, the different length 5’UTRs contained nearly identical eIF1 rebinding times and lifetimes. Consistently, the eIF1 binding window lengthened from 8.1 ± 0.5 s to 11 ± 0.2 s on the longer 5’UTR, which matched the difference in initial eIF1 lifetimes. To examine the role of ATP hydrolysis by eIF4A (the only ATP-dependent enzyme in our reaction), we also examined eIF1 binding in the presence of a non-hydrolyzable ATP analog (ADPNP) on the longer 5’UTR. Inclusion of 100-fold molar excess ADPNP further lengthened the duration of both the initial and subsequent eIF1 binding events by approximately 2- and 4- fold (to 11 ± 1 s and 2.2 ± 0.2 s respectively) (Extended Data Fig. 3a). Inclusion of ADPNP also inhibited and delayed loading of the 43S initiation complex (Extended Data Fig. 3b,c), as observed in yeast¹¹. These findings suggest that ATP hydrolysis by eIF4A plays a role in multiple initiation steps, some of which impact eIF1 binding dynamics. Similarly, reduction to a limiting eIF1A concentration also lengthened all eIF1 binding events by ~2–3 fold and delayed eIF5B binding by 10-fold (Extended Data Fig. 3d & Table 1), consistent with its central role throughout initiation. In dramatic contrast, elimination of the start site in the shorter 5’UTR lengthened the initial eIF1 binding lifetime by at least 10-fold (to >23 s) (Fig. 2c). This measurement was limited by photostability of the Cy5 dye and should be considered an underestimate. All subsequent eIF1 binding events also lengthened by 20-fold (to 4.7 ± 0.5 s) on the RNA without a start site.

Our findings thus suggest that eIF1 remains stably bound during scanning and its initial departure is triggered by initial recognition of the start site. This interpretation agrees with prior biochemical data in yeast²⁵ and the presence of a steric clash between eIF1 and Met-tRNA_i^Met that would occur upon transition of the initiation complex from the P_scan to P_in state after recognition of the start site^{46,47,55–57}. Our findings further highlight the dependence of 43S initiation complex loading on multiple rounds of ATP hydrolysis by eIF4A and suggest that the protein may regulate subsequent initiation steps^57,59.

Termination of transient eIF1 rebinding

A critical commitment step of initiation occurs when eIF2 hydrolyzes GTP, which is at least partially coupled to recognition of the start site. To examine how this step affects eIF1, we replaced eIF2-GTP with eIF2-GDPNP (a non-hydrolyzable GTP analog). On β-globin mRNA, the initial eIF1 binding lifetime using eIF2-GDPNP essentially matched the initial lifetime using eIF2-GTP (2.3 ± 0.3 s) (Fig. 2d). Following its initial departure, eIF1 transiently rebound the initiation complex, and the kinetics of individual rebinding events also matched those with eIF2-GTP. However, eIF1 continuously rebound for an at least 10-fold longer period with eIF2-GDPNP present, which was limited by photostability of the tRNA_i-Cy3 signal (Fig. 2d and Extended Data Fig. 3e). These complexes never progressed to eIF5B binding, consistent with prior work^20,24. Thus, the switch from stable to transient eIF1 rebinding occurred independent of GTP hydrolysis by eIF2, but termination of transient eIF1 rebinding required it.

Since eIF5 serves as the GTPase activating protein for eIF2, eIF5 activity and subsequent hydrolysis of GTP by eIF2 may terminate transient rebinding by eIF1. We therefore monitored eIF1 in the presence of eIF2-GTP and different concentrations of eIF5 on β-globin mRNA. At each eIF5 concentration, the duration of the initial eIF1 binding event remained about 2.5 s (Fig. 2e and Extended Data Fig. 3f). By contrast, the duration of the eIF1 binding window inversely correlated with the concentration of eIF5. Individual eIF1 rebinding events had similar kinetics, but the eIF1 binding window extended from 9.1 ± 0.8 s at the highest eIF5 concentration to 33 ± 8 s at the lowest due to the presence of 6-fold more eIF1 rebinding events (Extended Data Fig. 3f,g & Table 1). Termination of eIF1 binding and progression to ribosomal subunit joining thus requires concentration-dependent binding by eIF5.

eIF5 transiently binds initiation complexes

We next developed a real-time, single-molecule FRET assay to monitor eIF5 binding to initiation complexes directly. We fluorescently labeled the N-terminal domain of eIF5 as it binds proximal to our labeling position on tRNA_i-Cy3 (U46) upon recognition of the translation start site and yielded the expected high FRET (Fig. 3a and Extended Data Fig. 4)^56,57. Given that yeast eIF5 can co-purify with eIF3⁶⁰, we confirmed that purified human eIF3 lacked eIF5 (and eIF1) using mass spectrometry (Supplementary Table 2). To monitor eIF5, we added to the eIF4-bound β-globin mRNA a mixture that contained the 43S initiation complex (with tRNA_i-Cy3, green), eIF5-Cy5.5 (purple), eIF5B-Cy3.5 (gold), and 60S subunits labeled with Cy5 (red) (Fig. 3b). During imaging, we simultaneously excited the surface with 532 nm and 640 nm lasers. eIF5-Cy5.5 occupancy thus was indicated by appearance of Cy3-to-Cy5.5 FRET (via the 532 nm laser) and by direct excitation and emission of the Cy5.5 dye (via the 640 nm laser). This dual-excitation strategy ensured detection of eIF5 regardless of any potential structural flexibility of the protein within the complex incompatible with FRET. Importantly, eIF5B and 60S subunit binding kinetics (relative to 43S complex loading) with tRNA_i-Cy3 and eIF5-Cy5.5 present in the reaction were identical to when the unlabeled versions were present (Extended Data Fig. 5)²⁴. Fluorescently labeled Met-tRNA_i^Met and eIF5 therefore function identically to their unlabeled versions.

Figure 3. eIF5 transiently binds initiation complexes.

(a) Structure of a human translation initiation complex present at the translation start site (PDB ID: 8OZ0)⁵⁷. The labeling positions of U46 on Met-tRNA_i^Met (green) and the N-terminus of eIF5 (purple) are indicated, which are separated by about 35 Å. (b) Schematic of the real-time single-molecule human translation initiation assay. (c) Example single-molecule fluorescence data where tRNAi-Cy3 (green), eIF5-Cy5.5 (purple), eIF5B-Cy3.5 (orange), and 60S-Cy5 (red) were monitored during translation initiation on β-globin mRNA. The dotted box indicates the analogous region depicted in panel i. (d) Heat map of the tRNA_i-Cy3 (top) and eIF5-Cy5.5 (bottom) fluorescent signals on all analyzed complexes (the 40 nM eIF5 experiment) synchronized to initial appearance of the tRNA_i-Cy3 signal. (e) Heat map of the eIF5-Cy5.5 (top), eIF5B-Cy3.5 (middle), and 60S-Cy5 (bottom) fluorescent signals on all analyzed complexes (the 40 nM eIF5 experiment) synchronized to appearance of the eIF5-Cy5.5 signal. (f) Cumulative probability plot of the observed eIF5 lifetime in the 40 nM experiment. The line represents the fit to a single-exponential function, which accounted for 90% of observed binding events. (g) Plot of the observed eIF5 association rate constants at the indicated concentrations, with error bars representing the 95% CI. Lines represent fits via linear regression analysis to derive the indicated rates. ‘First’ and ‘Final’ indicates the association rate constant of the first and final eIF5 binding event measured relative to loading of the 43S initiation complex. ‘Rebinding’ represents the rate constant for eIF5 to rebind the initiation complex measured relative to departure of the previous eIF5 binding event. (h) Cumulative probability plot of the observed eIF5B binding time in the 40 nM experiment. The line represents a fit to an exponential function. (i) Example single-molecule data where the eIF5-Cy5.5 signal extended into the eIF5B-Cy3.5 binding event. In all experiments, the final concentration of unlabeled eIF1 was 290 nM. See Supplementary Table 3 for all rates, rate constants, and the number of complexes and binding events analyzed in each experiment.

The 43S initiation complex loaded onto β-globin mRNA without eIF5. Loading of the 43S complex was indicated by appearance of tRNA_i-Cy3 fluorescence and occurred independent of eIF5-Cy5.5 signal at all tested eIF5 concentrations (10, 20, 40, 80, and 250 nM) (Fig. 3c,d and Extended Data Fig. 6a & Table 3). After a delay, initiation complexes typically contained one or two bursts of eIF5-Cy5.5 signal lasting 1 ± 0.1 s in duration that universally corresponded to a decrease in tRNA_i-Cy3 intensity (Fig. 3c,e,f,i and Extended Data Fig. 6b–d). This observation indicates that eIF5 bound in a conformation that yielded FRET with tRNA_i-Cy3, rather than binding distally and rearranging later. The duration of the delay prior to eIF5 binding depended on the concentration of eIF5 (0.8 ± 0.1 μM⁻¹ s⁻¹), which corresponded to an observed delay of 9.2 ± 0.7 s at 80 nM eIF5 (Fig. 3g and Extended Data Fig. 6b). On complexes with multiple eIF5 binding events, eIF5 rebound 6-fold faster than its initial binding (6 ± 0.25 μM⁻¹ s⁻¹) (Fig. 3g and Extended Data Fig. 6c,d). We thus conclude that eIF5 transiently binds initiation complexes in a concentration-dependent manner after the scanning and start site recognition steps.

eIF5 binding progressed initiation complexes to subsequent eIF5B binding and 60S ribosomal subunit joining. On about half of the initiation complexes (38–49%), eIF5B binding occurred prior to or simultaneous with eIF5 departure (at 100 ms temporal resolution) (Fig. 3h,i and Extended Data Fig. 6e). For the remainder of complexes, eIF5B rapidly bound after eIF5 departure (within 2.4 ± 0.4 s at 40 nM). Given the incomplete labeling of eIF5 with dye (estimated at ~50% labeled) and the rapid eIF5 rebinding rate, most of these slower eIF5B binding events very likely occurred on complexes that contained an unlabeled eIF5 protein. Our finding agrees with a recent structural analysis that visualized eIF2, eIF5, and eIF5B simultaneously bound to initiation complexes⁶¹. Once eIF5B bound, the 60S subunit joined and eIF5B departed the 80S initiation complex identical to previous work (Extended Data Fig. 6b)²⁴. By contrast, eIF5 continuously rebound initiation complexes and failed to progress to eIF5B binding when eIF2-GDPNP replaced eIF2-GTP (Extended Data Fig. 6d). eIF5 binding thus allows and may mediate subsequent eIF5B binding, with any direct interactions between the proteins occurring transiently (< 0.5 s)⁶².

Given that eIF1 binding depended on the concentration of eIF5, we further examined the interplay between the proteins using two strategies. First, we examined eIF5 binding at a substochiometric concentration of eIF1 (10 nM). At this concentration, about 4-fold fewer initiation complexes progressed to eIF5 and eIF5B binding (Extended Data Fig. 7a). This finding suggests that a large population of complexes completely lacked eIF1, consistent with its low nanomolar equilibrium dissociation constant and critical role during initiation^42,43. Indeed, we observed a small increase in the number of 43S initiation complexes loading onto mRNA with eIF5 bound (from 0 to 12%), which never progressed to eIF5B binding. On complexes that ultimately progressed to eIF5 binding, eIF5 bound 4-fold more rapidly than when eIF1 was in relative excess (290 nM) (Extended Data Fig. 7b). Second, we simultaneously monitored both eIF1-Cy5 and eIF5-Cy5.5 binding on initiation complexes. On complexes that progressed to eIF5B binding (i.e. successful), eIF1 and eIF5 binding events were mutually exclusive (Extended Data Fig. 7c–f). Furthermore, an eIF5 binding event was the final event observed prior to eIF5B binding on ~95% of initiation complexes. The very rare complexes (~5 %) where eIF1 and eIF5 binding overlapped never progressed to eIF5B binding and likely represent stalled complexes or otherwise aberrant events (Extended Data Fig. 7g). Thus, eIF5 directly competes with eIF1 to bind initiation complexes paused at translation start sites.

Non-AUG start sites differentially affect eIF1 and eIF5

Given their key role in the discrimination of potential translation start sites, we examined eIF1 and eIF5 binding on model RNAs with non-AUG start sites. To design our experiments, we first analyzed eIF1 and eIF5 binding in the presence of an AUG start site either in ideal or poor Kozak context. In our assays, we observed nearly identical eIF1 binding kinetics regardless of Kozak context (Extended Data Fig. 8a). While the delay prior to eIF5 binding also was similar, the duration of eIF5 binding events decreased by 2-fold to 0.4 ± 0.03 s in poor context (Extended Data Fig. 9a).

We next examined non-AUG start sites using ideal Kozak context to enhance their usage. We replaced the single AUG with AUC, CUG, UUG, or GUG and compared eIF1 and eIF5 binding kinetics (Fig. 4a). We monitored binding of eIF1-Cy5 (at 40 nM) in the presence of excess unlabeled eIF5 (290 nM), or eIF5-Cy5.5 (at 40 nM) with excess unlabeled eIF1 (290 nM). Given the heterogenous nature of the observed effects (see below), we analyzed the data in multiple ways. First, to capture disparate outcomes, we examined eIF1 binding on all initiation complexes that loaded onto an mRNA, regardless of whether eIF5B bound (successful) or not (unsuccessful). Second, we quantified the percentage of initiation complexes that contained an eIF5 binding event. Third, we hypothesized that eIF1 and eIF5 kinetics would vary for successful initiation complexes on different start sites. We therefore also examined their binding kinetics on the subset of complexes that progressed to eIF5B binding, which was present at identical concentrations in both experimental schemes.

Figure 4. non-AUG start sites differentially affect eIF1 and eIF5 binding kinetics.

(a) Schematic of the model RNAs with the indicated start sites. (b) Example single-molecule fluorescence data that monitored eIF1-Cy5 (at 40 nM) and depicts the three broad categories quantified in panel c. Note the different lengths of times depicted in each panel. (c) Plot of the percent of initiation complexes that classified into the three broad categories of eIF1-Cy5 binding outlined in panel b. ‘Indeterminant’ indicates complexes that were too ambiguous to classify. The No AUG model lacked any potential start site. (d) Plot of the percentage of initiation complexes that were bound by eIF5-Cy5.5 on the indicated model RNAs. The error bars represent the 95% CI derived from binomial bootstrapping of the observed events. (e) Plot of the relative fold change of the indicated eIF1 binding parameters on eIF5B-bound initiation complexes on the indicated model RNAs. (f) Plot of the mean eIF5B binding time measured relative to loading of the 43S initiation complex onto the indicated mRNAs in the two predominate experimental regimes used to monitor eIF1-Cy5 (top) or eIF5-Cy5.5 (bottom). Error bars represent the derived 95% CI of the mean time. (g) Plot of the relative fold change of the indicated eIF5 binding parameters on eIF5B-bound initiation complexes on the indicated model RNAs. See Supplementary Tables 1 & 3 for all rates, rate constants, and the number of complexes and binding events analyzed in each experiment.

The AUC start site yielded eIF1 and eIF5 binding kinetics analogous to the RNA without any start site. Relative to successful complexes on β-globin and AUG model RNAs, most initiation complexes stalled with eIF1 stably bound on the model RNA with an AUC site (Fig. 4b,c). Indeed, initial departure of eIF1 occurred about 9-fold slower (19 ± 2 s) and subsequent rebinding events lengthened by 12-fold (6.2 ± 0.7 s (Extended Data Fig. 8b,c & Table 3). These stabilizing effects collectively extended the eIF1 binding window to more than 40 s, which represents an underestimate due to the limited photostability of the dyes. Consistently, eIF5 binding was reduced to a level nearly identical to the RNA without any start site, which precluded further analyses (Fig. 4d). The combined eIF1 and eIF5 effects agree with the critical role of the G in start codon triplets for translation initiation.

Initiation on CUG, UUG, and GUG start sites led to heterogeneous effects. Relative to β-globin and the AUG version, these three non-AUG sites increased by ~2 fold the population of complexes where eIF1 remained stably bound (to ~30 % of complexes) (Fig. 4b,c). In parallel, the non-AUG sites increased by 2–4 fold the subset of complexes that contained continuous eIF1 rebinding events, which peaked with GUG (~18 % of complexes). Thus, even in the presence of molar excess eIF5 relative to eIF1-Cy5, the fraction of successful initiation complexes decreased on the non-AUG sites. These distinct effects on eIF1 binding suggest that initiation complexes stalled in at least two different conformations, both of which were incompatible with eIF5B binding. In agreement, the fraction of initiation complexes that contained eIF5 binding events progressively decreased on the RNAs with CUG, UUG, and GUG sites (Fig. 4d).

We next compared eIF1 binding kinetics on initiation complexes that successfully progressed to eIF5B binding on the CUG, UUG, and GUG start sites. Relative to β-globin and AUG, the initial and subsequent eIF1 binding lifetimes lengthened by about 1.5–2 fold (to 5–7 s) and 1–1.7 fold (to 0.4–0.9 s), respectively (Fig. 4e and Extended Data Fig. 8d). The time elapsed between eIF1 rebinding events remained similar (1.6–2.3 s). The lengthened duration of all eIF1 binding events extended the eIF1 binding window by 1.3–2 fold (to 11–20 s). These collective effects lengthened the dwell time from loading of the 43S initiation complex until eIF5B bound by more than 2-fold (to 22–26 s) (Fig. 4f). Further consistent with a competition model, the eIF1 binding window extended to more than 30 s when the eIF5 concentration was reduced to a limiting concentration (Extended Data Fig. 8e,f). These findings indicate that non-AUG sites prolong initial eIF1 binding and subsequent rebinding to initiation complexes. These stabilizing effects delay progression to downstream steps at two distinct points – initial eIF1 departure and eIF1 rebinding – which ultimately delay and inhibit eIF5B binding.

We also estimated eIF5 binding kinetics on CUG, UUG, and GUG start sites, which were more rare since eIF1 was in molar excess. The delay until eIF5 productively bound initiation complexes lengthened by 2–3 fold on the mRNAs with CUG (19 ± 3 s), UUG (26 ± 5 s), and GUG (29 ± 6 s) sites relative to AUG (7.7 ± 0.6 s) (Fig. 4g, Extended Data Fig. 9b & Table 3). Further consistent with a competition model, the delay until eIF5 first bound shortened by 4-fold on the GUG site when the concentration of eIF1 was limiting (Extended Data Fig. 9c). Moreover, the duration of eIF5 binding events shortened by 4-fold on UUG (0.2 ± 0.04 s) and GUG (0.2 ± 0.1 s) start sites relative to the CUG and AUG sites (both ~0.8 ± 0.05 s) (Fig. 4g and Extended Data Fig. 9b). Thus, eIF5 generally bound more slowly and departed more rapidly from initiation complexes on RNAs with non-AUG start sites, except when eIF1 was limiting. We estimate that the apparent affinity of eIF5 is ~2-fold lower for CUG start sites and at least 14-fold lower for UUG and GUG sites, consistent with the relative efficiencies we observed in cellular translation assays (see below). In each case, destabilization of eIF5 also delayed eIF5B binding by at least 1.5-fold (relative to 43S complex loading), further delaying initiation (Fig. 4f and Extended Data Fig. 9d–f). Once initiation complexes progressed to eIF5B binding, though, downstream initiation events on mRNAs with alternative start sites proceeded at rates similar to those on the mRNA with an AUG start site (Extended Data Fig. 9b).

Collectively, our single-molecule findings indicate that both eIF1 and eIF5 kinetics respond differentially to the identity of the translation start site. The combined effects at multiple initiation steps delay binding by eIF5B and 60S ribosomal subunit joining (Fig. 4f). These results are consistent with the direct proximity of both proteins to the initiator tRNA anticodon-mRNA base pairing interactions within initiation complexes. Our kinetic findings suggest that eIF1-eIF5 competition to bind initiation complexes paused at putative start site determines whether the site is selected to begin protein synthesis.

eIF1 and eIF5 have opposing roles in cells

Based on our kinetic model and prior studies examining overexpression of eIF1 and eIF5^37,38, we hypothesized that changes in the relative abundance of eIF1 and eIF5 would impact translation in human cells. We therefore examined how both knockdown and overexpression of eIF1 and eIF5 impacted translation of mRNAs with ideal and alternative translation start sites. To enable precise control of the start site, we fused our panel of unstructured model 5’UTRs comprised of CAA repeats to the firefly luciferase coding region (Fig. 5a). We also examined a reporter that encoded the native eIF5 5’UTR, which contains two inhibitory uORFs with AUGs in poor context. On this reporter, we normalized its translation relative to a version that lacked the uORF start codons.

Figure 5. eIF1 and eIF5 have opposing roles in cells.

(a) Plot of the percent relative translation activity of the indicated reporter mRNAs. The top 6 mRNAs contained unstructured 5’UTRs 50 nt in length, which were comprised of CAA repeats. For all 6, translation activity was quantified relative to model mRNA with an AUG in ideal Kozak context. The bottom mRNA was derived from the native 5’UTR of eIF5, which contains two upstream start sites in poor context. For this mRNA, translation activity was quantified relative to the same mRNA except the potential upstream start sites were eliminated. Error bars represent standard deviation (SD) from n = 6 experiments. (b) Plot of the fold change of the translation activity of the indicated mRNAs upon shRNA-mediated knockdown of eIF1 or eIF5. Data are plotted relative to a control scrambled shRNA. Error bars represent SD from n = 6 experiments. (c) Plot of the fold change of the translation activity of the indicated mRNAs upon overexpression of eIF1 or eIF5. Data are plotted relative to an empty vector control. Error bars represent SD from n = 6 experiments. In panels b and c, * indicates p < 0.01 for Student’s t-test (2 tailed) tests that compared the effects of eIF1 versus eIF5 on the given mRNAs. The precise p-values are (from top to bottom): (panel b) 0.00018, 0.0029, 0.0088, 0.0000031, 0.0072, 0.0072, 0.00040, and 0.000082; (panel c) 0.000025, 0.0000042, 0.000015, 0.00000063, 0.0000076, 0.0000053, and 0.00011. (d) Proposed kinetic model and potentially associated conformational rearrangements derived from our single-molecule assays. The model highlights that concentration-dependent binding by eIF1 and eIF5 underlies recognition of the translation start site, which can be tuned by changes in the abundance or activity of either protein. * indicates that the eIF1 and eIF5 association rates correspond to association rates determined in the presence of molar excess eIF5 and eIF1 (both at 290 nM).

Alterations of eIF1 and eIF5 abundances differentially impacted translation of mRNAs that contain alternative start sites. As predicted, the presence of an alternative translation start site reduced translation of the reporter mRNA (Fig. 5a). However, relative to control cells, partial knockdown of eIF1 modestly enhanced translation of mRNAs with alternative start sites by about 2-fold, whereas eIF5 partial knockdown modestly inhibited translation of those same mRNAs (Fig. 5b and Extended Data Fig. 10). Strikingly, overexpression of either protein had the inverse effect. Overexpression of eIF1 inhibited translation of the mRNAs with alternative start sites by 2–4 fold, and eIF5 overexpression enhanced translation by 4–16 fold (Fig. 5c and Extended Data Fig. 10). In each case, eIF1 versus eIF5 effects were significantly different (t-test, p < 0.01). Translation of the eIF5 reporter responded inversely relative to the mRNAs with alternative start sites upon changing eIF1 and eIF5 abundance. This observation agrees with the known auto- and cross-regulation of eIF5 mRNA translation by eIF5 and eIF1, respectively, which is mediated by the inhibitory uORFs that have AUGs in poor context³⁸.

DISCUSSION

We defined a biophysical framework for recognition of the translation start site during initiation in humans. Multiple kinetic steps – initial eIF1 release, subsequent competitive binding by eIF1 and eIF5, final eIF5 release, and ultimately eIF5B binding – underlie the process. We suggest that these kinetic steps couple to dynamic conformational rearrangements within the initiation complex (rev. in^9,10,13,15) (Fig. 5d).

As it loads onto an mRNA, the human 43S initiation complex contains the 40S ribosomal subunit, eIF1, eIF1A, and Met-tRNA_i^Met. Each factor has been visualized directly as the complex loads in real time (this study and²⁴). Extensive prior biochemical and structural data indicate that the complex also includes eIF2 and eIF3 (rev. in^7–10,13,15). By contrast, the human 43S initiation complex lacked eIF5 as it loaded onto mRNA. Our finding agrees with the absence of eIF5 in structures of mammalian initiation complexes prior to recognition of the start site^45–54. This distinction from yeast, where eIF5 stably incorporates into 43S initiation complexes^39,40, may reflect the divergence of S. cerevisiae and human eIF3 (5 subunits versus 12). Regardless, such a difference in kinetics has precedence in translation initiation. Universally-conserved eIF5B binds initiation complexes with inverted kinetics in yeast and humans, while still guiding joining of the 60S subunit in both organisms^21,24.

eIF1 remained stably bound to the initiation complex until initial recognition of the start site. The timing of its initial departure depended on the length of the 5’UTR and required a translation start site. After accounting for the eIF1 dissociation rate (3.1 ± 0.3 s⁻¹) and assuming a 35 nt footprint of the initiation complex^63,64, we estimate a scanning rate of about 30 nt per s on the 200 nt model 5’UTR. To recognize the start site during scanning, the initiation complex must engage (P_in) and disengage (P_scan) the Met-tRNA_i^Met anticodon with the mRNA on a tens of milliseconds or briefer timescale. The presence of eIF1 and absence of base-pairing interactions rapidly shifts the complex back to the disengaged state (P_scan). By contrast, proper recognition of the start site – and its stabilization from three base-pair interactions with the anticodon – lengthens the residence time of the complex in the engaged state (P_in). This stabilized conformational shift ejects eIF1. Consistently, the anticodon and stem of Met-tRNA_i^Met shift by ~12 Å into the eIF1 binding site when paired properly with a start site^{46,47,55–57}. The initial eIF1 lifetimes we quantified thus provide an upper bound for the timing of this movement on the mRNAs we examined. The identical initial eIF1 binding lifetimes on complexes with eIF2-GTP and eIF2-GDPNP also indicate that this movement occurs independent of and likely prior to GTP hydrolysis by eIF2. The increased population of initiation complexes with very long-lived initial eIF1 lifetimes on mRNAs with non-AUG start sites indicates that many complexes bypass such sites (i.e. leaky scanning).

After initial start site recognition, eIF1 and eIF5 directly compete to bind initiation complexes with opposite outcomes. The competition also may involve dynamic conformational shifts of the initiation complex that rapidly engage and disengage Met-tRNA_i^Met with the mRNA while presumably paused at (or very near) the start site. Such fluctuations would permit stable rebinding of eIF1 to the disengaged state (P_scan) and potential resumption of scanning at mis-identified start sites⁶⁵. They also may underpin rare reading frame shifts observed in ribosome profiling studies of 80S initiation complexes⁶⁶. Consistently, we observed broad FRET distributions at equilibrium for both eIF1 and eIF5 bound initiation complexes stalled at start sites, which suggests the complexes are structurally flexible at a rapid (ms or faster) timescale. During this period, eIF5 very likely competes with eIF1 to capture the engaged conformation (P_in) and commit the complex to ribosomal subunit joining. The inverse effects by non-AUG start sites on eIF1 and eIF5 initiation complex interactions (stabilized and destabilized, respectively) suggests that the energy barrier between the various conformational states is low. Population of the states within the conformational landscape is modulated by a single base pair, independent of GTP hydrolysis by eIF2. Furthermore, both eIF1 and eIF5 kinetics were differentially affected by different non-AUG start sites, which at least partially explains the differential efficiency of these sites in human cells. The disruption of multiple initiation steps uncovered in our single-molecule assays will compound in cells, where individual mRNAs likely undergo many rounds of initiation.

Collectively, our biophysical model rationalizes more than a decade of genetic, biochemical, and structural data. The presence of multiple, competing bimolecular reactions on initiation complexes paused at potential start sites not only builds fidelity, but also allows flexible tuning of start site selection via changes in the relative abundance of eIF1 or eIF5. Indeed, our findings illuminate how dynamic sequestration and release of eIF1 from the nucleus during the cell cycle rapidly reprograms the cellular proteome through differential usage of non-AUG start sites⁶⁷. They also suggest that multiple offramps from the productive initiation pathway likely exist. Thus, our results underscore the dynamic plasticity of translation and its control in health and disease.

METHODS

Labeled eIF1.

Synthetic DNA that encoded human eIF1 with C69 substituted for alanine was purchased from IDT and inserted into the 6His-MBP expression vector (v1C) from the UC Berkeley QB3 Macrolab. Using maleimide chemistry, this construct was used to conjugate Cy5 dye on the sole remaining cysteine94 residue in eIF1. The plasmid was transformed into Rosetta2 cells purchased from the UC Berkeley QB3 MacroLab and grown overnight at 37 °C on LB agar plates supplemented with 50 μg/mL kanamycin. Liquid cultures of single colonies were grown to OD₆₀₀ ≈ 0.5 at 37 °C in LB supplemented with kanamycin. After addition of 0.5 mM IPTG, the cultures were grown for 4 hours at 30 °C. Cells were harvested by centrifugation at 5,000 × g for 15 min at 4 °C in a Fiberlite F9 rotor (ThermoFisher, cat. # 13456093). Cells were lysed by sonication in lysis buffer (20 mM Tris-HCl pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 20 mM imidazole, and 5 mM β-mercaptoethanol), and lysates were cleared by centrifugation at 38,000 × g for 30 min at 4 °C in a Fiberlite F21 rotor followed by filtration through a 0.22 μm syringe filter. Clarified lysate was loaded onto a Ni-NTA gravity flow column equilibrated in lysis buffer, washed with 20 column volumes (CV) of lysis buffer, 20 CV of wash buffer (20 mM Tris-HCl pH 8.0, 1 M NaCl, 10% (v/v) glycerol, 40 mM imidazole, and 5 mM β-mercaptoethanol), and 10 CV of lysis buffer. Recombinant proteins were eluted with five sequential CV of elution buffer (20 mM Tris-HCl pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 300 mM imidazole, and 5 mM β-mercaptoethanol). Fractions with recombinant protein were identified by SDS-PAGE analysis. The relevant fractions were dialyzed overnight at 4 °C into TEV cleavage buffer (50 mM HEPES-KOH pH 7.5, 200 mM NaCl, 10% (v/v) glycerol, and 1 mM DTT) in the presence of excess TEV protease. TEV protease and the cleaved 6His-MBP tag were removed via a subtractive Ni-NTA gravity column equilibrated in TEV buffer, with the flow-through collected. The protein sample was diluted to 50 mM NaCl, applied to 1 mL SP HP ion exchange column, and eluted using a 50 mM to 500 mM NaCl gradient in the absence of reducing agents. Fractions that contained eIF1 at > 95% purity were concentrated to ~ 22 μM, frozen in liquid N₂, and stored at −80 °C. To fluorescently label the purified protein, we added 700 μL of purified protein (~15 nmoles) to 1 mg of malemide-Cy5 in the presence of 2.5 mM TCEP. The reaction was incubated at room temperature for 20 min in the dark followed by an overnight incubation on ice. The unreacted Cy5 dye was removed using a 10DG-desalting column (Bio-Rad, cat.# 7322010) equilibrated in size-exclusion buffer (20 mM HEPES-KOH pH 7.5, 150 mM KOAc, 10% (v/v) glycerol, and 1 mM DTT). The labeled protein then was purified using a 23 mL SD75 size-exclusion chromatography column. Fractions that contained eIF1 at > 95% purity were concentrated to ~ 10 μM (by Cy5; ~50% labeled), frozen in liquid N₂, and stored at −80 °C.

Labeled eIF5.

Synthetic DNA that encoded human eIF5 with an N-terminal ybbR tag was purchased from IDT and inserted into the pET28 backbone, which contained a TEV protease cleavage site between the 6His-tag and the ybbR-tag. Rosetta2 cells were grown to OD 0.5 and induced with 0.5 mM IPTG for 4 hours at 30 °C. ybbR-eIF5 was purified essentially as above for eIF1, except 10 μM zinc sulfate was added to all buffers. Prior to ion exchange purification, ybbR-eIF5 was labeled with either Cy5 or Cy5.5 fluorescent dye using Sfp synthase as described²⁴. Briefly, ~10 μM ybbR-eIF5 was supplemented with 10 mM MgCl₂ and incubated at 37 °C for 90 min in the presence of 2–4 μM Sfp synthase enzyme and 20 μM of Cy5-CoA or Cy5.5-CoA substrate. Free dye was removed via purification over 10DG-desalting columns (Bio-Rad, cat.# 7322010) equilibrated in TEV cleavage buffer supplemented with 20 mM imidazole. TEV protease, Sfp synthase, and any remaining cleaved 6His tag were removed via a second subtractive Ni-NTA gravity column equilibrated in TEV buffer, with the flow-through collected. The labeled ybbR-eIF5 proteins were diluted to 50 mM NaCl and purified using a 1 mL Q HP column. eIF1 was eluted using a 50 mM to 500 mM gradient of NaCl. Fractions that contained ybbR-eIF5 at > 95% purity were concentrated to ~ 10 μM (by dye; ~50% labeled), frozen in liquid N₂, and stored at −80 °C.

Labeled eIF5B.

N-terminally truncated (587–1220) eIF5B with an N-terminal ybbR-tag was purified and labeled with Cy3.5 dye using Sfp synthase as described²⁴.

Labeled initiator tRNA.

HPLC purified, synthetic human Met-tRNA_i with U46 replaced with Uridine-C6 Amino linker conjugated to Cyanine 3 SE was purchased from TriLink. Prior to aminoacylation, the fluorescently-labeled tRNA_i-Cy3 was purified using phenol:chloroform extraction and ethanol precipitation. Aminoacylation with methionine was performed as described⁴². The charging efficiency was > 90% based on acid urea PAGE analyses.

Unlabeled eIFs.

Recombinant eIF1⁶⁸, eIF1A⁶⁸, eIF3j⁶⁸, eIF4AI^68,69, eIF4B²⁴, eIF4G_165–1599²⁴, eIF4E^69,70, eIF5⁴² proteins were purified as described. Native human eIF2 and eIF3 were purified as described^68,71. Importantly, purified human eIF3 lacked any detectable presence of eIF1 or eIF5 by mass spectrometry analysis (Supplementary Table 2). Unlabeled human Met-tRNA_i^Met was in vitro transcribed and aminoacylated as described²⁴.

Ribosomal subunits.

Human 40S and 60S ribosomal subunits were purified from either wild-type or the appropriately edited HEK293T cell lines (RPS15-ybbR for 40S-Cy3) and (RPL5-ybbR for 60S Cy5/Cy5.5) and labeled with fluorescent dyes as described^72,73. Cell line identities were confirmed by PCR assays, Sanger sequencing, western blotting, and fluorescence gels (after labeling). Cells were not tested for mycoplasma contamination.

mRNAs.

Full-length human β-globin mRNA (NM_000518.5, 628 nts) was in vitro transcribed and prepared as described²⁴ to yield a 5’ m⁷G capped and 3’-biotinylated mRNA with a 30 nt poly(A) tail. To generate the short, unstructured model RNAs, DNA that encoded the reverse complement of the RNA and T7 promoter sequences were purchased from IDT as single-stranded oligos. After hybridizing a second oligo that encoded the sense T7 promoter sequence, the RNAs were in vitro transcribed using T7 RNA polymerase for 3 hrs at 37 °C and treated with Turbo DNase. The transcribed RNAs were prepared identically to β-globin mRNA, except the RNAs lacked the poly(A) tail and all purifications were done using standard phenol:chloroform extractions and ethanol precipitations. To generate the mRNA with a 200 nt unstructured 5’UTR, we replaced the 5’UTR on β -globin mRNA with 198 nt of CAA repeats, leaving only the Kozak context around the AUG (CACCAUGGA). The mRNA was in-vitro transcribed and prepared identical to β -globin, producing a 5’-m7G capped, 3’-biotinylated mRNA with a 202 nt 5’UTR and a 30 nt poly(A) tail.

Real-time single-molecule analyses.

All real-time imaging was conducted using a modified Pacific Biosciences RSII DNA sequencer and the Maggie software (v. 2.3.0.3.154799), which were described previously⁵⁸. All experiments were performed at 30 °C (unless otherwise noted) using a 532 nm excitation laser at 0.32 μW/μm², which directly excited Cy3 and Cy3.5 dyes. Cy5 and Cy5.5 dyes were excited either directly via a 642 nm laser (0.1 μW/μm²) or via FRET as indicated. Four-color fluorescence emission (Cy3, Cy3.5, Cy5, and Cy5.5) was detected at 10 frames per second for 600 s. ZMW chips were purchased from Pacific Biosciences, which were passivated by reaction with polyvinylphosphonic acid to form a covalent Al-polyphosphonate coating 60. Prior to imaging, all ZMW chips were washed with 0.2% Tween-20 and TP50 buffer (50 mM Tris-OAc pH 7.5, 100 mM KCl). Washed chips were coated with neutravidin by a 5 min incubation with 1 μM neutravidin diluted in TP50 buffer supplemented with 0.7 mg mL⁻¹ UltraPure BSA and 1.3 μM of pre-annealed DNA blocking oligos (CGTTTACACGTGGGGTCCCAAGCACGCGGCTACTAGATCACGGCTCAGCT) and (AGCTGAGCCGTGATCTAGTAGCCGCGTGCTTGGGACCCCACGTGTAAACG). The imaging surface then was washed with TP50 buffer at least four times. All experiments, which are outlined below, were performed essentially as described²⁴.

The ‘initiation reaction buffer’ was: 20 mM HEPES-KOH, pH 7.5, 70 mM KOAc, 2.5 mM Mg(OAc)₂, 0.25 mM spermidine, 0.2 mg mL⁻¹ creatine phosphokinase, 1 mM ATP•Mg(OAc)₂, and 1 mM GTP•Mg(OAc)₂. The ‘imaging buffer’ was the initiation reaction buffer supplemented with casein (62.5 μg mL-1), 5 mM TSY, and an oxygen scavenging system⁷⁴: 2 mM protocatechuic acid (PCA) and 0.06 U/μL protocatechuate-3,4-dioxygenase (PCD).

To prepare the eIF2–GTP–Met-tRNA_i^Met ternary complex (TC), 3.3 μM eIF2 was incubated in initiation reaction buffer (excluding ATP•Mg(OAc)₂) for 10 min at 37 °C to saturate eIF2 with GTP. The eIF2–GTP complex then was incubated with either 2.3 μM of unlabeled Met-tRNA_i^Met (as a control) or Met-tRNA_i^Met-Cy3 (tRNA_i-Cy3) for 5 min at 37 °C to form TC. In a few instances, GTP was replaced with a non-hydrolyzable analog (GDPNP) during formation of TC, which prevents GTP hydrolysis and subsequent departure of eIF2 from the initiation complex.

To prepare the 43S pre-initiation complex (43S PIC), we typically incubated 1 μM eIF1, 1 μM eIF1A, 500 nM TC (by eIF2), 1 μM eIF5, 400 nM eIF3, 1.2 μM eIF3j, and 240 nM 40S subunits for 5 min at 37 °C in initiation reaction buffer. In most experiments with eIF1-Cy5 or eIF5-Cy5/5.5, the unlabeled protein was replaced with the labeled version at the same concentration (1 μM, determined via dye fluorescence). For eIF1-Cy5 and eIF5-Cy5.5 titration experiments (Fig. 1 and Fig. 3), eIF1-Cy5 was included at 250 nM, and eIF5-Cy5.5 was included at 62 nM, in their respective experiments. Additional eIF1-Cy5 or eIF5-Cy5.5 was supplemented during imaging (see below) to yield the reported final concentrations. Finally, in a few experiments (Fig. 2, Extended Data Fig. 3, and Extended Data Fig. 9), the concentrations of eIF1, eIF5, and eIF1A were reduced to 250 nM (eIF1 and eIF5) and 125 nM (eIF1A) to yield the final concentrations of 10 nM and 5 nM respectively.

In the real-time single-molecule initiation assays, the indicated 5’ m⁷G capped and 3’-biotinylated mRNA was tethered to a prepared, neutravidin-coated ZMW imaging surface. Immediately after tethering and washing the surface with imaging buffer, a mixture of 2 μM eIF4A, 440 nM eIF4B, 260 nM eIF4G, and 320 nM eIF4E in 20 μL of imaging buffer was added to the surface. The surface and reaction mixture were incubated at room temperature for 5–10 minutes as the instrument initialized. At the start of data acquisition, a 20 μL mixture of 20 nM 43S PIC (by the 40S subunit; a 12.5-fold dilution of the complex assembled above), 80 nM eIF5B-Cy3.5, and 300 nM 60S subunits in imaging buffer supplemented with 500 nM of unlabeled eIF1, eIF1A, and/or eIF5 was added to the surface. Importantly, addition of this second reaction mixture to the surface doubled the total reaction volume (to 40 μL; since 20 μL with eIF4 proteins was already present); thus, unless indicated otherwise, the final concentrations of individual components in the experiments were: 1 μM eIF4A, 220 nM eIF4B, 230 nM eIF4G, 320 nM eIF4E, 10 nM 40S subunits, 20 nM TC, 16 nM eIF3, 48 nM eIF3j, 290 nM eIF1A, 290 nM eIF1 (unlabeled), and 290 nM eIF5 (unlabeled). In experiments with labeled eIF1-Cy5 and eIF5-Cy5.5, the unlabeled versions of the proteins were excluded from all steps and solutions.

Equilibrium TIRF microscopy.

A home-built, prism-based TIRFM instrument has been described previously^74–76. Emission data were collected in both the Cy3 (donor) and Cy5 (acceptor) channels following excitation of the Cy3 donor dye with the 532 nm laser. All complexes were prepared as above, and TC-GDPNP-tRNA_i-Cy3 was used. The pre-assembled 43S PIC (75 nM) was incubated with the indicated biotinylated model mRNA (AUG or No AUG at 50 nM) for 15 min at 37 °C in the initiation reaction buffer. The mRNA-43S complex was tethered to the neutravidin coated imaging surface for 5 min at room temperature at 5 nM final concentration (by 40S subunits). Following a 50 μL wash of the surface using initiation reaction buffer, the sample was incubated with the imaging buffer supplemented with 300 nM eIF1A and either 20 nM eIF1-Cy5 or 20 nM eIF5-Cy5, as indicated. Emission data were collected in both the Cy3 (donor) and Cy5 (acceptor) channels following excitation of the Cy3 donor dye with the 532 nm laser. Fluorescence intensities from single complexes were analyzed as outlined below.

Single-molecule data analyses.

Experimental movies that captured fluorescence intensities over time were processed using MATLAB R2018-b as described previously. In all analyses, spots or ZMWs with the desired fluorescence signals were identified by filtering for the desired signals.

To determine association and dissociation kinetics, binding events of individual components were assigned manually based on the appearance and disappearance of the respective fluorescence signals. The observed times for an event to occur typically from 100–200 individual molecules (i.e., ZMWs) were used to calculate cumulative probability functions of the observed data (cdfcalc, MATLAB), which were fit to single- or double-exponential functions in MATLAB (cftool, non-linear least squares method) as described. All derived association rates, lifetimes, and the number of molecules examined are reported in Supplementary Tables 1 & 3. The base exponential function was defined as:

$$C\left(t\right)=a\left(1-e^{-b\left(t+d\right)}\right)+(1-a)\left(1-e^{-c\left(t+d\right)}\right)$$

where t is time (in s), b and c rates, and d an adjustment factor. If a parameter yielded a phase that represented less than 10% of the population, only a single phase was used to derive the respective rate and reported (i.e., single-exponential function). For eIF5, we focused kinetic analyses on the fastest 90% of observed molecules to enable fits to single-exponential functions and facilitate comparisons.

To facilitate visualization of events in main text figures, raw fluorescent intensities were manually corrected to account for spectral bleedthrough across the various emission channels. Reported errors for derived rates represent 95% C.I. yielded from fits to linear, single-exponential, or double-exponential functions, as indicated. The reported mean times were defined as the reciprocal of the derived rate constant. Estimates of the 95% C.I. of the number of observed complexes with a given binding event (e.g., eIF5 in Figure 4d) were derived using binomial bootstrapping.

To produce FRET efficiency and fluorescence intensity heat maps, the data is normalized by computing the maximum and minimum for each data set (top and bottom 10–20^th percentile respectively). The raw data is then subtracted by the minimum and divided by the range (maximum minus minimum), normalizing the data set to be within the range of 0 to 1.

eIF3 sample preparation for mass spectrometry analysis.

The eIF3 protein solutions were brought up to 50 μL with 40 μL 50 mM ammonium bicarbonate. The solutions were reduced with dithiothreitol (10 mM final in 50 mM ammonium bicarbonate) at 56 °C for 45 min. The solutions were alkylated with 2-chloroacetamide (20 mM final in 50 mM ammonium bicarbonate) and incubated in the dark at ambient temperature for 30 min. Trypsin (Promega, Madison, WI) 250 ng, was added and the solutions were incubated overnight at 37 °C with mixing. Samples were allowed to equilibrate to ambient temperature and were taken to dryness on a speedvac. The dried samples were brought up in 30 μL 85% Acetonitrile/15 mM ammonium formate pH 2.8. The peptide solution was desalted over 1cm TopTip packing material (Poly-LC-, Columbia, MD) packed in a 10 μL pipette tip. Material was washed with 50 μL 85% Acetonitrile/15 mM ammonium formate pH 2.8, 3 times and eluted with 50 μL 15 mM ammonium formate pH 2.8. The desalted material was taken to dryness in a speed vac.

Orbitrap Fusion LC/MS/MS analysis.

Desalted samples were brought up in 2% acetonitrile in 0.1% formic acid (20 μL) and were analyzed (2 μL) by LC/ESI MS/MS with a Thermo Scientific Easy-nLC 1200 (Thermo Scientific, Waltham, MA) nano HPLC system coupled to a tribrid Orbitrap Fusion (Thermo Scientific, Waltham, MA) mass spectrometer. In-line de-salting was accomplished using a reversed-phase trap column (100 μm × 20 mm) packed with Magic C18AQ (5-μm 200Å resin; Michrom Bioresources, Bruker, Billerica, MA) followed by peptide separations on a reversed-phase column (75 μm × 270 mm) packed with Reprosil C18AQ (3-μm 100 Å resin; Dr. Maisch, Germany) directly mounted on the electrospray ion source and warmed to 37 °C with a column heater. A 45-minute gradient from 9% to 44% B at a flow rate of 300 nL/minute was used for chromatographic separations, with A as 0.1% formic acid and B 80% acetonitrile in 0.1% formic acid. The heated capillary temperature was set to 300 °C and a static spray voltage of 2200 V was applied to the electrospray tip. The Orbitrap Fusion was operated in the data-dependent mode, switching automatically between MS survey scans in the Orbitrap (AGC target value 500,000, resolution 120,000, and maximum injection time 50 milliseconds) with MS/MS spectra acquisition in the linear ion trap using quadrupole isolation. A 3 second cycle time was selected between master full scans in the Fourier-transform (FT) and the ions selected for fragmentation in the HCD cell by higher-energy collisional dissociation with a normalized collision energy of 27%. Selected ions were dynamically excluded for 45 seconds and exclusion mass by mass width +/− 10 ppm.

LC/MS/MS Data Analysis.

Data analysis was performed using Proteome Discoverer 3.1 (Thermo Scientific, San Jose, CA). The data were searched against Uniprot Human (Uniprot UP000005640 Nov 11, 2023) (http://www.thegpm.org/crap/) fasta files. Trypsin was set as the enzyme with maximum missed cleavages set to 2. The precursor ion tolerance was set to 10 ppm and the fragment ion tolerance was set to 0.6 Da. Variable modifications included oxidation on methionine (+15.995 Da). Dynamic modifications on the protein N-terminus included acetylation (+42.011 Da), methylation (+14.016 Da) and methionine loss plus acetylation (−89.030 Da). Static modifications included carbamidomethyl on cysteine (+57.021 Da). Data were searched using Sequest HT. All search results were run through Percolator for scoring and identified peptides were filtered for 1% peptide-level false discovery rate using q value of 0.01.

Cell Culture and Transfections.

U2OS cells were kindly provided by Nancy Kedersha (Harvard University). U2OS cells were maintained in DMEM (Corning) supplemented with 1 mM L-glutamine, 1 mM sodium pyruvate, 10% FBS (Gibco) and penicillin/streptomycin (Quality Biological) at 37°C in 5% CO₂. For luciferase assays, U2OS cells were grown overnight in 10 cm plates to ~70% confluence, washed, treated with trypsin, and then transfected using Lipofectamine 2000 reagent (Invitrogen) in a one-day protocol in which suspension cells were added directly to the DNA mixtures in 96-well half-area white plates (Costar). For eIF1/eIF5 overexpression experiments, 0.2 μl Lipofectamine, 10 ng Renilla control reporter, and 10 ng firefly test reporter plasmid were mixed with 25 ng eIF1 or eIF5 overexpressing plasmid or empty control plasmid in 25 μl Opti-MEM (Gibco) and then dispensed to each well along with 10⁴ cells suspended in 25 μl DMEM. The transfection was terminated 24 h later by removing the media and lysing the cells using 25 μl 1x Passive Lysis Buffer (Promega). For eIF1/eIF5 knockdown experiments, 0.2 μl Lipofectamine, 5 ng Renilla control reporter, and 5 ng firefly test reporter plasmid were mixed with 25 ng mixture of eIF1 (four different plasmids, two each targeting EIF1 and EIF1B respectively, with every shRNA plasmid contributing 6.25 ng DNA to the total), eIF5 (two different plasmids, with every shRNA plasmid contributing 12.5 ng DNA to the total) or control (shc002) shRNA expressing plasmids in 25 μl Opti-MEM (Gibco) and then dispensed to each well along with 10⁴ cells suspended in 25 μl DMEM. The transfection was terminated 48 h later by removing the media and lysing the cells using 25 μl 1x Passive Lysis Buffer.

For Western analyses, U2OS cells were cultured in the same DMEM medium described above in 6-well tissue-culture plates. After 24 h, cells were transfected by adding 250 μL mixture of Opti-MEM, 5 μL Lipofectamine 2000 and 800 ng shRNA or eIF1/eIF5 overexpression plasmid. After 48 hr, cells were lysed in 100 μl RIPA lysis buffer (Thermo Scientific 89900) supplemented with cOmplete EDTA-free protease inhibitor tablet (Roche 04693159001). Repeated pipetting was performed to ensure complete lysis. The cell lysates were transferred to 1.5 mL Eppendorf tubes containing ~40 mg glass beads, vortexed briefly, and spun at 10,000 × g for 1 min.

Plasmids.

Leader sequences matching those of the single-molecule in vitro experiments were synthesized and cloned between the HindIII and BamHI restriction sites of the vector p2luc⁷⁷ by LifeSct LLC (Rockville, MD). The sequence of the WT leader insert is: 5’-aagcttGGACAACAACAACAAACCCTCGAACAACAACAACAACAACAACAACACCATGGACAACAcCAACAcCAcCggatcc-3’ (HindIII and BamHI sites in italics, nucleotides in bold were mutated as shown in Fig. 5a to generate alternative start sites). The Renilla expressing plasmid used in these experiments, pSV40-Renilla, as well as the phRL-based eIF1 and eIF5 overexpressing plasmids and empty vector control plasmid were described previously³⁸. The shRNAs for eIF1 and eIF5 knockdown experiments were purchased from Sigma. Each mRNA was targeted by two shRNAs - TRCN0000310709 and TRCN0000303471 targeting eIF1, TRCN0000149959 and TRCN0000149658 targeting eIF1B, and TRCN0000338465 and TRCN0000338404 targeting eIF5. For control, shc002 shRNA was used. The firefly reporters with the eIF5 wild-type and deregulated control leaders were described previously³⁸.

Dual Luciferase Assay.

Luciferase activities were determined by the Dual Luciferase Stop and Glo Reporter assay system (Promega), injecting 25 μl of each buffer per well. Relative light units were measured on a CentroXS3 LB960 microplate luminometer fitted with two injectors (Berthold Technologies). Light emission was measured after injection of 25 μL firefly luciferase substrate followed by the same volume of Renilla luciferase substrate. For the firefly (test) luciferase reporters, values were first normalized to a co-transfected Renilla luciferase control plasmid. Subsequently, the normalized value for each reporter was expressed as percentage relative to the firefly reporter initiated by AUG in optimal (Kozak) context. For the eIF5 reporter, the normalized luciferase value was expressed as a percentage relative to a reporter in which the uORF AUGs were mutated to AAA. The fold change in reporter expression was calculated by dividing the corresponding values in co-transfected eIF1 or eIF5 overexpressing (or eIF1 or eIF5 shRNA) plasmids to cells co-transfected with empty vector (or scrambled control shRNA) plasmids. Significance was calculated for the luciferase values in cells overexpressing eIF1 vs eIF5 (or expressing the eIF1 vs eIF5 shRNAs). All reported p-values are derived from Student’s t-tests (2 tailed, type III).

Western analyses.

Immunoblot analyses of eIF1 and eIF5 levels were conducted as previously described with minor modifications⁷⁸. Samples were electrophoresed on precast 12% SDS-polyacrylamide gels (Bio-Rad) at 100 V for 1.5 h and then transferred onto 0.45 μm nitrocellulose (Bio-Rad 1620094). Membranes were cut to the appropriate size, blocked in 5% milk in 1x TBST (1x TBS plus 0.1% Tween 20) for 60 min at room temperature (RT), washed with 1x TBST, and then incubated for 1 h at RT with rabbit anti-eIF1 (Cell Signaling 12496) or rabbit anti-eIF5 (Cell Signaling 2480) antibodies at 1:1000 dilution in 1x TBST supplemented with 5% milk for 1 h at RT. Following washes with 1x TBST, the membrane was incubated for 1 h at RT with horseradish peroxidase conjugated anti-rabbit IgG (GE Healthcare NA9340V) at 1:10,000 in TBST supplemented with 5% milk, washed in TBST, and then developed with SuperSignal^™ West Femto maximum sensitivity substrate (Thermo Scientific 34096). Chemiluminesce was detected with Azure BioSystems 280 machine. To probe for β-Actin, the membrane was stripped by incubation for 20 min at RT in stripping buffer (1.5% glycine, 0.1% SDS, 1% Tween 20). After washing, the membrane was blocked with 5% milk for 1 h at RT then re-probed with mouse β-actin monoclonal antibody (Proteintech 66009–1-Ig) at 1:750 dilution for 1 h at RT. Following washes with 1x TBST, the membrane was incubated for 1 h at RT with horseradish peroxidase conjugated anti-mouse IgG (GE Healthcare NA9331V) at 1:10,00 dilution in TBST supplemented with 5% milk. The bands were quantitated using ImageJ. The eIF1 and eIF5 signal in each lane was normalized to the signal for β-actin. The values for the three biological replicates were used to calculate the mean and then the significance.

Extended Data

Extended Data Figure 1. An eIF1 FRET signal.

(a) Image of a gel from SDS-PAGE analysis of purified eIF1 labeled on C94 with Cy5 fluorescent dye. The left and right images are total protein and Cy5 fluorescent scans of the same gel. (b) Image of an acid TBE-Urea gel that examined Met-tRNA_i-Cy3 before and after in vitro aminoacylation with methionine. The gel was scanned for Cy3 fluorescence. No molecular weight ladder was used. (c) Schematic of the equilibrium total internal reflection fluorescence microscopy (TIRFm) experiments used to validate the eIF1 FRET signal. Samples were excited with a 532 nM laser. (d) Example fields of view of TIRFm experiments. (e, f) Example fluorescence data of initiation complexes equilibrated on the model mRNA without a start site (No AUG model, panel e) or with a start site (AUG model, panel f). The complexes contain tRNA_i-Cy3 (green) to eIF1-Cy5 (red) FRET events. The top plots represent the fluorescence intensities (arbitrary units) of both signals, and the bottom plots represent the calculated FRET efficiency (E_FRET) during eIF1-Cy5 binding events. The right panels plot the E_FRET distribution observed on the indicated number of eIF1 binding events. The mean E_FRET ± standard deviations are indicated. (g) Cumulative probability plot of the eIF1 FRET lifetime observed on the two model mRNAs. The lines represent fits to double-exponential functions, which were used to derive the indicated rates.

Extended Data Figure 2. eIF1 stably binds and dynamically samples individual complexes.

(a) Schematic of predicted single molecule data for eIF1-Cy5 experiments. Initial appearance of tRNA_i-Cy3 (green) to eIF1-Cy5 (red) FRET signifies that a 43S initiation complex loaded onto the tethered mRNA. This initial FRET event is defined as the ‘initial eIF1 binding event’ and the duration of this event defined as the ‘initial eIF1 binding lifetime’. The complex then contains multiple subsequent tRNA_i-Cy3 to eIF1-Cy5 FRET events, which are defined as subsequent eIF1 binding events. The dwell time between loss of the previous eIF1 signal and appearance of the next eIF1 signal is defined as the ‘eIF1 rebinding time’. The duration of all subsequent eIF1 binding events is defined as the ‘subsequent eIF1 binding lifetime’. Appearance of eIF5B-Cy3.5 (orange) signal signifies successful entry to the final initiation steps that culminate with joining of the 60S ribosomal subunit. The dwell time from loss of the final eIF1-Cy5 signal to appearance of eIF5B-Cy3.5 signal is defined as the ‘eIF5B binding time’. (b) Cumulative probability plot of the eIF5B binding time measured relative to initial loading of the 43S initiation complex, which was signified by either appearance of tRNA_i-Cy3 to eIF1-Cy5 FRET (this study) or initial appearance of 40S-Cy3 signal. The overlapped plots indicate that labeled tRNAi-Cy3 and eIF1-Cy5 function analogously as their unlabeled versions. (c) Plot of the percentage of loaded 43S initiation complexes that contain multiple eIF1 binding events on β-globin mRNA in the indicated conditions. ‘Productive’ indicates complexes that progressed to eIF5B binding, whereas ‘any’ indicates that all complexes were analyzed regardless of whether eIF5B bound. (d) Cumulative probability plots of the indicated parameters at differing Cy5-eIF1 concentrations, observed on any loaded initiation complexes (both successful and unsuccessful). Lines represent fits to exponential functions. (e) Schematic of the real-time single-molecule assay with direct excitation of all fluorophores present using dual 532 nm and 640 nm lasers. This excitation scheme examined whether eIF1-Cy5 fully departed initiation complexes. (f) Example single-molecule data of the direct excitation assay showing termination of eIF1 binding is correlated to loss of FRET. The total number of initiation complexes analyzed is indicated (n = 115) and all 744 eIF1-Cy5 binding events ended with complete loss of eIF1-Cy5 signal. See Supplementary Table 1 for the number of complexes and binding events analyzed in each experiment.

Extended Data Figure 3. Termination of transient eIF1 binding requires a translation start site and concentration-dependent binding by eIF5.

(a) Cumulative probability plots of the indicated parameters for eIF1-Cy5 binding at the indicated molar ratios of ATP:ADPNP on the 200 nt 5’UTR model RNA. Lines represent fits to exponential functions. (b) Plot of the 43S initiation complex loading efficiency on the mRNA at the indicated ATP:ADPNP ratios (1 mM total concentration) relative to 1 mM ATP. (c) Cumulative probability plot of the 43S initiation complex loading time (from start of data acquisition to appearance of tRNA_i-Cy3 fluorescence signal) at the indicated molar ratios of ATP:ADPNP on the 200 nt 5’UTR model RNA. Lines represent fits to exponential functions. (d) Cumulative probability plots of the indicated kinetic parameters in experiments that monitored eIF1-Cy5 on the 50 nt 5’UTR with an AUG start site in the presence of a limiting concentration (5 nM) of eIF1A. Lines represent fits to exponential functions. The rightmost plot quantifies the percentage of loaded 43S initiation complexes that ultimately progressed to eIF5B binding, which was corrected for the eIF5B labeling efficiency. (e, g) Representative single-molecule fluorescence data that monitored eIF1-Cy5 either in the presence of TC-GDPNP (panel e) or in the presence of a limiting concentration of eIF5 (10 nM) (panel g). Both experiments used the β-globin mRNA. (f) Cumulative probability plots of the indicated kinetic parameters in experiments that monitored eIF1-Cy5 on the on β-globin mRNA in the presence of different concentrations of eIF5. Lines represent fits to exponential functions. See Supplementary Table 1 for the number of complexes and binding events analyzed in each experiment.

Extended Data Figure 4. An eIF5 FRET signal.

(a) Image of a gel from SDS-PAGE analysis of purified eIF5 labeled on its N-terminal ybbR tag with Cy5 (or Cy5.5) fluorescent dye. The left and right images are total protein and Cy5 fluorescent scans of the same gel. Identical results are obtained with either Cy5 or Cy5.5. (b) Schematic of the equilibrium total internal reflection fluorescence microscopy (TIRFm) experiments. Samples were excited with a 532 nM laser. (c) Example field of view of TIRFm experiments. (d) Example fluorescence data of initiation complexes equilibrated on the model mRNA with an AUG start site. The complexes contain tRNA_i-Cy3 (green) to eIF5-Cy5 (purple) FRET events. The top plot represents the fluorescence intensities (arbitrary units) of both signals, and the bottom plot represents the calculated FRET efficiency (E_FRET). The right panel plots the E_FRET distribution observed during the indicated number of eIF5 binding events. The mean E_FRET ± standard deviations are indicated. (e) Cumulative probability plot of the eIF5 FRET lifetime observed on the AUG model mRNA. The line represents a fit to a double-exponential function, which was used to derive the indicated rates.

Extended Data Figure 5. eIF5-Cy5.5 functions analogously to its unlabeled version.

(a) Example single-molecule fluorescence data from an experiment that monitored loading of the 43S initiation complex as signaled by 40S-Cy3 (green) in the presence of 40 nM eIF5B-Cy3.5 (orange), and 60S-Cy5 (red). In this experiment, unlabeled eIF1 and eIF5 were present at 290 nM each. (b-d) Cumulative probability plot of the eIF5B binding time, 60S subunit joining time, and the lifetime of eIF5B on the 80S initiation complex. The eIF5B binding time was measured relative to initial loading of the 43S initiation complex, which was signified by either appearance of tRNA_i-Cy3 signal (this study) or initial appearance of 40S-Cy3 signal²². The overlapped plots indicate that labeled tRNA_i-Cy3 and eIF5-Cy5.5 function analogously as their unlabeled versions.

Extended Data Figure 6. eIF5 transiently binds initiation complexes.

(a) Schematic of predicted single-molecule data that monitor eIF5-Cy5.5. Initial appearance of tRNA_i-Cy3 (green) fluorescence indicates loading of the 43S initiation complex onto the mRNA. Either direct appearance of eIF5-Cy5.5 (purple) fluorescence or appearance of tRNA_i-to-eIF5 FRET indicates the presence of eIF5 on initiation complexes. Appearance of eIF5B-Cy3.5 (orange) indicates binding of eIF5B. Appearance of 60S-Cy5 (red) fluorescence indicates the 60S subunit joined to form the 80S ribosome; the relative proximity of tRNA_i-Cy3 and 60S-Cy5 labeling sites in the 80S initiation complex yields low FRET. The kinetic parameters were defined as follows. ‘eIF5 first binding time’ as the time elapsed from initial appearance of the tRNA_i-Cy3 signal to the first appearance of eIF5-Cy5.5 signal. ‘eIF5 first lifetime’ as the duration of the first eIF5-Cy5.5 binding event. ‘eIF5 rebinding time’ as the time elapsed from departure of the previous eIF5 protein until the start of the next binding event. ‘eIF5 subsequent lifetimes’ as the duration of the subsequent eIF5 binding events. ‘eIF5 final binding time’ as the time elapsed from appearance of the tRNA_i-Cy3 signal until appearance of the final eIF5-Cy5.5 signal prior to eIF5B binding. ‘eIF5 final lifetime’ was the duration of the final eIF5 binding event. Since most complexes contained a single eIF5 binding event, the ‘first’ and ‘final’ eIF5 binding events were identical on most complexes. ‘eIF5B binding time’ was defined as the time elapsed from disappearance of the final eIF5 signal until appearance of the eIF5B-Cy3.5 signal. ‘60S joining time’ was defined as the time elapsed from appearance of eIF5B-Cy3.5 signal until appearance of 60S-Cy5 signal. ‘eIF5B-80S lifetime’ was defined as the duration of the eIF5B-Cy3.5 signal on the 80S initiation complex. (b) Cumulative probability plots of the indicated kinetic parameters on β-globin mRNA at the indicated concentrations of eIF5-Cy5.5. (c) Example single-molecule data of an initiation complex that was bound multiple times by eIF5-Cy5.5. (d) Plot of the percent of initiation complexes that contained either a single or multiple eIF5-Cy5.5 binding events at the indicated eIF5-Cy5.5 concentrations. (e) Plot of the percent of initiation complexes where eIF5B bound either during (overlapped), within 100 ms of (< 0.1 s), or after 100 ms (> 0.1 s) the final eIF5 binding event at the indicated concentrations of eIF5-Cy5.5. In all experiments, unlabeled eIF1 was present at a final concentration of 290 nM and eIF5-Cy5.5 was present at the indicated final concentrations (by Cy5.5 dye). See Supplementary Table 3 for the number of complexes and binding events analyzed in each experiment.

Extended Data Figure 7. eIF5 competes with eIF1 to bind initiation complexes.

(a) Plot of the percentage of initiation complexes bound by eIF5 in the presence of 290 nM or 10 nM eIF1 on the AUG model RNA. (b) Cumulative probability plot of the eIF5 first binding time on initiation complexes in the presence of 290 nM or 10 nM eIF1 on the AUG model RNA. (c) Schematic of the single-molecule translation initiation assay to monitor eIF1-Cy5 and eIF5-Cy5.5 binding simultaneously using the AUG model RNA. In this experiment, only the 532 nm excitation laser was used. This experimental scheme directly detected loading of the 43S initiation complex (via tRNA_i-Cy3, green) and eIF5B-Cy3.5 (gold), as the conjugated dyes are directly excited by the 532 nm laser. By contrast, eIF1-Cy5 (red) and eIF5-Cy5.5 (purple) are not excited directly by the 532 nm laser and thus their binding is detected solely via FRET with the tRNA_i-Cy3 donor. (d) Representative single-molecule data of initiation complexes that progressed to eIF5B binding using the experimental scheme outlined in panel c. In total, 77 initiation complexes were analyzed, and all eIF1 and eIF5 binding events were mutually exclusive. Furthermore, eIF5 was the final binding event prior to eIF5B binding on 95% of the complexes. We suspect that the remaining 5% where eIF1 was last were likely followed by binding of an unlabeled eIF5 protein immediately prior to eIF5B binding. (e) Schematic of an alternative single-molecule translation initiation assay to monitor eIF1-Cy5 and eIF5-Cy5.5 binding simultaneously using the AUG model RNA. In this experiment, 532 nm and 640 nm excitation lasers were used simultaneously. This experimental scheme directly detected binding of all fluorescently-labeled components in the assays, independent of FRET. (f) Representative single-molecule data of initiation complexes that progressed to eIF5B binding using the experimental scheme outlined in panel e. As in panels c and d, eIF1 and eIF5 binding was mutually exclusive on initiation complexes that progressed to eIF5B binding. (g) On about 5% of complexes, we observed eIF1 and eIF5 binding events that overlapped; however, these complexes never progressed to eIF5B binding (i.e., were unsuccessful) and very likely represent aberrant complexes or non-specific binding events.

Extended Data Figure 8. eIF1 binding kinetics depend on the identity of the translation start site.

(a) Cumulative probability plots of the indicated eIF1 kinetic parameters on eIF5B-bound initiation complexes (successful) on the model AUG mRNAs in ideal (ACCAUGGA) or poor (CCCAUGCA) Kozak context. Lines represent fits to exponential functions. (b) Cumulative probability plots of the indicated eIF1 parameters observed on any loaded initiation complexes (both unsuccessful and successful) on the indicated model RNAs. Lines represent fits to exponential functions. (c) Plot of the population-weighted mean elapsed times for the indicated eIF1 parameters on the indicated mRNAs. All values plotted here were derived from events that occurred on any loaded initiation complex (both unsuccessful and successful). (d, e) Cumulative probability plots of the indicated kinetic parameters on eIF5B-bound initiation complexes on the indicated model mRNAs in the presence of 290 nM (panel d) or 5 nM (panel e) unlabeled eIF5. Lines represent fits to exponential functions. (f) Plot of the population-weighted mean eIF1 binding time on the indicated model RNAs in the presence of molar excess eIF5 (290 nM) or a limiting eIF5 concentration of 5 nM. (g) Plot of the percent of initiation complexes that classified into the three broad categories of eIF1-Cy5 binding outlined in Fig. 4b on β-globin mRNA. ‘Indeterminant’ captures complexes that were too ambiguous to classify. In all experiments, eIF1-Cy5 was present at a final concentration of 40 nM (by Cy5) and unlabeled eIF5 was present at either 290 or, when indicated, at 5 nM. See Supplementary Table 1 for the number of complexes and binding events analyzed in each experiment.

Extended Data Figure 9. eIF5 binding kinetics depend on the identity of the translation start site.

(a) Cumulative probability plots of the indicated eIF5 kinetic parameters on eIF5B-bound initiation complexes on the model AUG RNAs in ideal (ACCAUGGA) or poor (CCCAUGCA) Kozak context. Here, the eIF5B binding time was measured relative to loss of signal for the final eIF5-Cy5.5 binding event. The negative times represent complexes where eIF5B bound before eIF5 departed. Lines represent fits to exponential functions. (b) Cumulative probability plots of the indicated eIF5 kinetic parameters on eIF5B-bound initiation complexes on the model RNAs with various non-AUG start sites in ideal Kozak context. Here, the eIF5B binding time was measured relative to loss of signal for the final eIF5-Cy5.5 binding event. The negative times represent complexes where eIF5B bound before eIF5 departed. Lines represent fits to exponential functions. Lines represent fits to exponential functions. (c) Cumulative probability plots of the indicated eIF5 kinetic parameters on eIF5B-bound initiation complexes on the AUG or GUG model RNAs in the presence of a limiting concentration of eIF1 (10 nM). Here, the eIF5B binding time was quantified relative to loading of the 43S initiation complex to demonstrate initiation proceeded at similar rates on AUG and GUG start sites when the concentration of eIF1 was limiting. (d, e) Cumulative probability plots of the eIF5B binding time measured relative to loading of the 43S initiation complex when eIF1-Cy5 was monitored in the presence of excess eIF5 (panel d) or eIF5-Cy5.5 was monitored in the presence of excess eIF1 (panel e). (f) A replot of cumulative probability data depicted in panels d and e to facilitate cross-comparisons of the data. See Supplementary Table 3 for the number of complexes and binding events analyzed in each experiment.

Extended Data Figure 10. Perturbing eIF1 and eIF5 expression in human cells.

Immunoblots measuring eIF1 and eIF5 levels in U2OS cells transfected with plasmids overexpressing eIF1, eIF5, or a control empty vector (panels a-c); or eIF1 shRNA, eIF5 shRNA, or a scrambled control shRNA (panels d-f). The blots depict three replicates for each sample, each run on a single lane of the gel. Bands for eIF1, eIF5, and β-actin were quantified using ImageJ and then averaged (mean). eIF1/β-actin and eIF5/β-actin ratios were calculated based on the quantifications and plotted as normalized values to the control samples. All error bars denote SD. * denotes p < 0.05 from student’s two-tailed t test for n = 3 replicates relative to the control sample. Precise p-values are (from left to right): panel b, 0.0047; panel c, 0.025; panel f, 0.039 and 0.042.

Supplementary Material

Supplementary Tables 1-3

Supplementary Table 1. Contains all rates, rate constants, and the number of complexes and binding events analyzed in each real-time single molecule experiment using eIF1-Cy5.

Supplementary Table 2. Lists all proteins identified in the final eIF3 protein sample used in the single-molecule assays. eIF1 and eIF5 were not detected.

Supplementary Table 3. Contains all rates, rate constants, and the number of complexes and binding events analyzed in each real-time single molecule experiment using eIF5-Cy5.5.

DATA AVAILABILITY:

Processed single-molecule data and source data needed to recapitulate single-molecule figure plots throughout the manuscript are available for download from GitHub: https://github.com/LapointeLab/eIF1-eIF5-2024-paper. Specific requests can be requested using the Issue feature or by email request to C.P.L.. Structure images were generated using published models (PDB IDs: 6ZMW, 8OZ0) and ChimeraX-1.71.1 software. Source data are provided with the manuscript.

CODE AVAILABILITY:

All code needed to analyze the single-molecule data are available on GitHub: https://github.com/LapointeLab/eIF1-eIF5-2024-paper.