We created ribosome profiling libraries from eIF3 mutant and corresponding isogenic WT strains for both the eIF3a/b Degron (YAJ34: MAT a trp1∆ leu2-3,112 ura3-52 gcn2::hisG P _GAL1 -myc-UBR1::TRP1::ubr1 P _CUP1 -UBI-R-HA-tif32 ^td ::URA3::tif32 P _CUP1 -UBI-R-DHFR ^ts -HA-prt1 ^td ::URA3::prt1 and YAJ3: MATa trp1∆leu2-3,112 ura3-52 gcn2::hisG P _GAL1 -myc-UBR1::TRP1::ubr1 pRS316 [URA3]) (Jivotovskaya et al., 2006) and eIF3i DDKK (H450: MATa leu2-3,-112 ura3-52::GCN2 trp1Δ tif34Δ hc TIF34 URA3 transformed with YCp-i/TIF34-D207K-D224K-HA or YCp-i/TIF34-HA, respectively) (Herrmannová et al., 2012) as described previously (Ingolia, 2010; Sen et al., 2015; Sen et al., 2016). We grew two biological replicates of each strain and its matching isogenic WT strain under permissive conditions before harvesting and adding to pre-warmed restrictive media for a duration resulting in an ∼90% decrease in bulk translation (as judged by polysome:monosome ratios, Supplementary Figure S1) and a final cell density at mid-log phase (OD₆₀₀ = ∼0.6). We grew eIF3a/b Degron cells at 25°C in SC_Raff + Cu²⁺ before shifting them to pre-warmed SC_Raff/Gal + BCS at 36°C for 90 min. We grew eIF3i DDKK at 30°C in SC media before shifting them to pre-warmed SC media at 37°C for 30 min. We added cycloheximide to a final concentration of 100 μg/ml 2 min prior to harvesting by filtration through a Kontes filtration apparatus and flash freezing in liquid nitrogen with 2 ml of ribosome footprinting buffer (20 mM Tris pH 8.0, 140 mM KCl, 1.5 mM MgCl2, 1% Triton, 100 μg/ml cycloheximide).

We generated sequencing libraries of ribosome footprints and total mRNA as previously described (Ingolia, 2010; Sen et al., 2015; Sen et al., 2016). Briefly, we lysed cells using a freezer mill and then prepared lysates by centrifuging 5 min at 3,000 × g, collecting the supernatant and then centrifuging 12 min at >20,000 × g. We then collected the supernatant and flash-froze in 30 OD₂₆₀ aliquots. We purified ribosome footprints by adding 5 µL RNase1 to one aliquot of purified lysate and incubating 60 min at 26°C with mixing at 700 rpm. We then added 5 µL SuperAsin (Thermo Fisher) and loaded on a 10–50% sucrose gradient and centrifuged at 40,000 rpm for 3 h and then collected the monosome peak using a gradient fractionator. We then purified RNA from purified monosomes via hot phenol extraction. We purified total mRNA from one 30 OD₂₆₀ aliquot of purified lysate using the miRNeasy mini kit (Qiagen) per the vendor’s instructions and then randomly fragmented at 70°C for 8 min using Fragmentation Reagent (Invitrogen). We then performed subsequent steps (size selection, linker ligation, reverse transcription, circularization, rRNA subtraction, and PCR amplification) as previously described and had libraries sequenced using an Illumina HiSeq system.

We processed and analyzed sequencing libraries of ribosome footprints and total mRNA as described previously (Ingolia, 2010; Sen et al., 2015; Sen et al., 2016). We then employed DESeq2 (Love et al., 2014) for statistical analysis of differences in ribosome footprint and RNA-seq read counts, and TE_rel values between WT and mutant samples, as previously reported (Martin-Marcos et al., 2017; Kulkarni et al., 2019). We excluded genes with fewer than 128 total mRNA reads in the four samples combined (two replicates of both WT and mutant strains) from the calculation of TE_rel values. We then performed subsequent analysis of mRNA features and characteristics within R, using custom scripts, together with previously-reported datasets reporting 5′-UTR lengths, PARS values, closed-loop-forming propensity of individual mRNAs, previously identified uORFs, or ∆TE_rel values observed in the presence of mutations targeting eIF4A, eIF4B, Ded1, eIF1, or eIF1A. We performed Gene Ontology analysis using the Gene Ontology Resource PANTHER classification system. Statistical tests were performed as described in the main text and figures.

To investigate the transcriptome-wide roles of eIF3, we performed ribosome profiling in two S. cerevisiae strains in which the eIF3 complex is partially or entirely compromised. The eIF3a/b Degron strain expresses temperature-sensitive degron variants of the eIF3a and eIF3b subunits. Depletion of these subunits provokes the loss of the entire eIF3 complex (Jivotovskaya et al., 2006). The eIF3i DDKK strain expresses a variant of the eIF3i subunit that is unable to bind stably to eIF3b (Herrmannová et al., 2012). Because both eIF3i and eIF3g depend on this interaction to associate with the remainder of the eIF3 complex, this results in the loss of both subunits, which normally contribute to the mRNA-entry-channel arm of eIF3 (Figure 1A). This arm has been observed in two distinct locations of the PIC, depending on its functional state (Llácer et al., 2015; Llácer et al., 2018).

Before constructing ribosome profiling libraries, we first investigated the effects of the eIF3a/b Degron and eIF3i DDKK mutations on global translation levels, as assayed by polysome profiling. For each strain, we grew cells first under permissive conditions and then shifted them to restrictive conditions for 30, 60, 90, and 120 min. Both strains exhibit no growth defect under permissive conditions but manifest severe growth defects at restrictive conditions (Jivotovskaya et al., 2006; Herrmannová et al., 2012). Consistent with this, polysome profiles collected under permissive growth conditions for both strains were similar to those collected for isogenic wild-type strains (Supplementary Figure S1A). In contrast, we observed strong decreases in polysome to monosome ratios (P/M) upon shifting to restrictive conditions for both strains, with eIF3a/b Degron and eIF3i DDKK cells exhibiting an approximately 90% decrease in P/M (as compared to isogenic WT cells grown under the same conditions) after 30 min and 90 min at restrictive conditions, respectively (Supplementary Figure S1B,C).

Given the marked decrease in global translation levels we observed in both strains, we next asked how these global affects translate to individual mRNAs across the transcriptome. To that end, we constructed ribosome profiling and RNA-seq libraries from both the eIF3i DDKK and eIF3a/b Degron strains (and their corresponding isogenic WT strains) grown under restrictive conditions and calculated relative translational efficiency (TE_rel) values for coding sequences (CDS), ignoring reads obtained from the initial 15 codons and from the final 5 codons to avoid cycloheximide-induced artifacts (Gerashchenko and Gladyshev, 2014). Owing to the absence of an internal read-count standard, read counts from both ribosome profiling and RNA-seq libraries are normalized to the total library size for each condition. The TE_rel values we calculate from these normalized read counts thus do not enable direct comparison of absolute TE between samples. Instead, TE_rel values provide a measure of the translational status of individual mRNAs as compared to the overall population of mRNAs within each sample. We also calculated TE_rel values for a set of previously identified uORFs (Martin-Marcos et al., 2017; Kulkarni et al., 2019) but did not attempt to identify novel translated uORFS owing to the inclusion of cycloheximide in our library preparation. Both ribosome footprint and RNA-seq libraries were highly reproducible across replicates for each condition (Supplementary Figure S2).

Using the R DESeq2 package (Love et al., 2014), we then identified transcripts whose TE_rel was significantly changed (P_adj < 0.05) in each strain, as compared to an isogenic WT strain (Figure 1B). Owing to the marked decrease in global translational levels we observed in each mutant strain, as well as the normalization to total ribosomal footprint reads performed when calculating TE_rel, we interpreted mRNAs exhibiting significant TE_rel decreases as having a greater than average dependence on the either the eIF3i and eIF3g subunits lost in eIF3i DDKK cells or on the entire eIF3 complex disrupted in eIF3a/b Degron cells. We interpreted those mRNAs exhibiting significant TE_rel increases as instead having a weaker than average dependance on the regions of the eIF3 complex targeted by each mutation. The significant effects on global translational levels that we observed in both eIF3i DDKK and eIF3a/b Degron cells in fact suggest that most mRNAs likely experience decreases in their absolute TE. Nonetheless, comparison of these significant changes in TE_rel (∆TE_rel) enables identification of those mRNAs whose translation is most or least sensitive to disruption of eIF3 or its mRNA-entry-channel arm in a background where global translational levels are repressed.

In eIF3a/b Degron cells—in which the entire eIF3 complex is disrupted—we identified 1,455 transcripts whose TE_rel decreased and 1,340 transcripts whose TE_rel increased (5,466 total with significant read counts), as compared to TE_rel values in an isogenic WT strain (Figure 1B). Because eIF3 has been implicated in mediating the translation of specific mRNAs in a number of cell types (Sha et al., 2009; Lee et al., 2015; Rode et al., 2018; Lin et al., 2020; Lamper et al., 2020), we investigated the gene ontology (GO) terms associated with these affected mRNAs. The set of transcripts whose TE_rel decreased in eIF3a/b Degron cells was enriched for mRNAs with GO terms involved in mitochondrial translation and gene expression, as well as a variety of metabolic processes (Supplementary Figure S3). These most sensitive mRNAs were under-enriched in mRNAs involved in RNA processing and ribosome biogenesis. Consistent with this, mRNAs whose TE_rel increased in eIF3a/b Degron cells were enriched in GO terms associated with RNA processing and under-enriched in terms associated with mitochondrial translation and gene expression (Supplementary Figure S3).

In eIF3i DDKK cells—in which the eIF3i and eIF3g subunits of the eIF3 mRNA-entry-channel arm are lost from the complex—we identified 139 transcripts whose TE_rel decreased and 133 transcripts whose TE_rel increased, as compared to TE_rel values in an isogenic WT strain (Figure 1B). We did not observe any significant over- or under-enrichment of specific GO terms in affected transcripts in these cells, perhaps because the strong global effects on translation in eIF3i DDKK cells are more uniformly distributed amongst all mRNAs, resulting in widespread but uniform decreases in absolute TE levels across the transcriptome with more limited effects on the relative TE of individual mRNAs.

In both eIF3i DDKK and eIF3a/b Degron cells, we observed strong increases in uORF translation (Supplementary Figure S4A). Of 4,830 uORFs identified previously, 529 displayed significant (P_adj < 0.05) increases in TE_rel in eIF3a/b Degron cells, whereas 21 displayed significant decreases in TE_rel. In eIF3i DDKK cells, 391 uORFs displayed significant (P_adj < 0.05) increases in TE_rel, whereas 5 displayed significant decreases in TE_rel. This global increase in uORF translation in both strains is likely a result of the previously described effects of cycloheximide on read counts near the start codon (Gerashchenko and Gladyshev, 2014), which cannot be discarded as in the calculation of ORF TE_rel values owing to the short length of these regions. Thus we focused subsequent analysis of uORF TE_rel values on any observed differential behavior between uORFs within the same strain.

Because the eIF3 complex is either partly or entirely disrupted in both cell lines, we investigated the degree to which the TE_rel effects we observed were similar in eIF3i DDKK and eIF3a/b Degron cells. We observed a significant correlation (R = 0.63, P < 2.2 × 10⁻¹⁶) between the ∆TE_rel values observed in both strains (Figure 1C). Additionally, we found a strong overlap between those transcripts exhibiting TE_rel decreases in eIF3i DDKK and eIF3a/b Degron cells (Figure 1D), with the changes observed in eIF3i DDKK cells appearing to represent a subset of the changes observed in eIF3a/b Degron cells. Moreover, the overall portfolio of ∆TE_rel values we observe is similar for both strains, as evidenced by heat map comparison of the magnitude and direction of observed changes (Figure 1E). Together, these results are consistent with strong translational effects upon disruption of the eIF3 mRNA-entry-channel arm (as in eIF3i DDKK cells) or depletion of the entire eIF3 complex (as in eIF3a/b Degron cells) and a role for eIF3 in promoting the translation of mRNAs with mitochondrial roles.

We next investigated the structural features (Figure 2A) of mRNAs most or least sensitive to the eIF3i DDKK and eIF3a/b Degron mutations, as identified by DESeq2 analysis of TE_rel changes in each strain. eIF3 is required for the overall process of mRNA recruitment both in vivo (Jivotovskaya et al., 2006) and in vitro (Mitchell et al., 2010; Aitken et al., 2016), and mutations to several eIF3 subunits elicit defects in mRNA recruitment or its component events of initial docking, scanning, and start-codon recognition (Valášek et al., 2002; Nielsen et al., 2004; Chiu et al., 2010; Cuchalova et al., 2010; Elantak et al., 2010; Herrmannová et al., 2012).

To shed light on the contribution of the eIF3 complex and the eIF3i and eIF3g subunits of the mRNA-entry-channel arm to scanning processivity, we first asked if the ∆TE_rel values we observed in each strain correlated with 5′-UTR length. Upon restricting our analysis to mRNAs previously shown to have a dominant 5′-UTR isoform (defined as mRNAs for which one isoform accounts for at least 40% of all transcripts and is present at an abundance at least twice that of the next most abundant isoform) (Pelechano et al., 2013; Zinshteyn et al., 2017), we observed significant (P < 10⁻⁵⁹) negative correlations between 5′-UTR length and ∆TE_rel in both the eIF3i DDKK and eIF3a/b Degron strains (Figure 2B). Consistent with this effect, we also observed a negative correlation between ∆TE_rel and distance from the 5′ end for a set uORFs identified in previous studies (Martin-Marcos et al., 2017; Kulkarni et al., 2019), though this correlation was significant only in eIF3i DDKK cells (Supplementary Figure S4B).

To interrogate the roles of eIF3 and the eIF3i and eIF3g subunits in resolving structural impediments during initial mRNA docking or scanning, we next determined the relationship between ∆TE_rel and the propensity of specific regions of an mRNA to form secondary structures, as measured by their differential sensitivity in vitro to nucleases specific for single- or double-stranded RNA (PARS, Figure 2A) (Kertesz et al., 2010). In eIF3a/b Degron and eIF3i DDKK cells, we observed a significant (P < 10⁻⁸ and P < 10⁻⁷, respectively) negative correlation between ∆TE_rel and mean 5′-UTR PARS values, as well as with mean PARS scores determined within specific 30 nucleotide (nt) windows located at the first 30 nt of the 5′-UTR (First30; P < 10⁻³ for both), 30 nt centered around the AUG start codon (Start30; P < 10⁻⁴ for both), and the first 30 nt downstream of the Start30 window (Plus30; P < 10⁻³ and P < 10⁻², respectively) (Figure 2B) (Sen et al., 2015; Sen et al., 2016). Of the correlations we observed between ∆TE_rel and PARS measures, the strongest and most significant was with the maximum PARS score observed within any 30 nt window within the 5′-UTR (Max30; P < 10⁻³² and P < 10⁻³⁸, respectively) (Figure 2B).

Having observed these correlations between ∆TE_rel and 5′-UTR length or structural complexity in both the eIF3i DDKK and eIF3a/b Degron strains, we next investigated if we could detect significant differences in the median values of these measures when comparing mRNAs whose TE_rel was either significantly decreased or increased in either strain (Figure 2C). As before, we restricted our analysis to mRNAs with one dominant 5′ isoform. In eIF3a/b Degron cells, mRNAs displaying significant negative ∆TE_rel values possess longer 5′-UTRs and those displaying significant positive ∆TE_rel values possess shorter 5′-UTRs, as compared to mRNAs whose TE_rel was not significantly affected (P_adj < 10⁻²⁸ for all pairwise comparisons, Wilcoxon test). Similarly, in eIF3i DDKK cells, mRNAs displaying significant negative ∆TE_rel values possess longer 5′-UTRs than both unaffected mRNAs (P_adj < 10⁻¹⁸) and mRNAs displaying significant positive ∆TE_rel values (P_adj < 10⁻³). However, we did not observe a significant difference in 5′-UTR lengths when comparing mRNAs displaying significant positive ∆TE_rel values and unaffected mRNAs in eIF3i DDKK cells, perhaps because the set of mRNAs expressed as a dominant transcript isoform and exhibiting significant positive ∆TE_rel values in these cells is relatively small (n = 57). In both cell lines, we observed similar results when comparing 5′-UTR values reported in a separate study (Kertesz et al., 2010) (Supplementary Figure S5A).

When comparing measures of structural complexity (as measured by PARS values across the 5′-UTR and in distinct windows), we again observed differences between mRNAs whose TE_rel either increased or decreased in eIF3a/b Degron cells. mRNAs displaying significant TE_rel decreases in these cells have higher mean and max30 5′-UTR PARS values, as compared to unaffected mRNAs (P_adj < 10⁻³ and P_adj < 10⁻⁶, respectively; ANOVA and post-hoc Tukey test) and mRNAs displaying significant TE_rel increases (P_adj < 10⁻⁹ and P_adj < 10⁻⁶, respectively). Similarly, mRNAs whose TE_rel decreased in eIF3a/b Degron cells have higher PARS values at the 5′ end of their 5′-UTRs (first30) and around their start codons (start30) as compared to mRNAs whose TE_rel either increased (P_adj < 10⁻⁷ and P_adj < 10⁻², respectively) or was unaffected (P_adj < 10⁻² for both) in these cells. However, these differences appear more modest than those observed for 5′-UTR length, mean and max30 PARS values. In contrast, there is no significant difference in the PARS values downstream of the start-codon window (plus30) between mRNAs whose TE_rel either increased, decreased, or was unaffected in eIF3a/b Degron cells.

In eIF3i DDKK cells, we did not observe significant differences in most measures of 5′-UTR structural complexity when comparing mRNAs whose TE_rel either decreased, increased, or was unaffected. The one exception is the max30 PARS values of mRNAs whose TE_rel decreased in these cells, which is higher than for mRNAs whose TE_rel was not significantly affected (P_adj < 10⁻³) and modestly different than for transcripts whose TE_rel increased in these cells (P_adj < 0.02), perhaps because the number of transcripts displaying significant TE_rel increases in these cells and with available PARS scores is limited (n = 17). Nonetheless, these observations are consistent with a role for eIF3 and its mRNA-entry-channel arm in processive scanning through longer 5′-UTRs. eIF3 also appears to contribute to resolving structural complexity within the 5′-UTR during initial docking or scanning, though the mRNA-entry-channel arm may play a more peripheral role in these events.

Because various subunits of eIF3 appear to play roles in start-codon recognition (Nielsen et al., 2004; Valasek et al., 2004; Chiu et al., 2010; Cuchalova et al., 2010; Herrmannová et al., 2012), we next asked whether there was a correlation between the observed ∆TE_rel values in the eIF3i DDKK and eIF3a/b Degron strains and the strength of the Kozak consensus sequence surrounding the AUG start codon for each mRNA. However, we did not observe a significant correlation between context scores (calculated for nucleotides −6 to +4) and ∆TE_rel values in either eIF3i DDKK or eIF3a/b Degron cells (Supplementary Figure S6).

Consistent with this, we observed no significant difference in the median context scores for mRNAs whose TE_rel either increased, decreased, or was unaffected in eIF3a/b Degron cells (Figure 3A). Nonetheless, we did observe modest but significant differences between the median context scores of mRNAs displaying TE_rel increases in eIF3i DDKK cells, which are weaker than either those of mRNAs displaying TE_rel decreases or those whose TE_rel was unaffected (P_adj < 10⁻⁴ and P_adj < 10⁻⁵, respectively).

Given these modest effects, we investigated the sequence logos in the vicinity of the start codons of these distinct mRNAs (Figure 3B). Here again, we observed no difference when comparing mRNAs whose TE_rel either increased, decreased, or was unaffected in eIF3a/b Degron cells. However, consistent with the differences we observed in context scores in eIF3i DDKK cells, mRNAs displaying TE_rel increases in this background show a weaker preference for an adenine at the −3 nt than do mRNAs whose TE_rel is either unaffected or decreases; optimal Kozak consensus sequences contain a purine base (most often adenine) at position −3 and a guanosine at position +4 (Hinnebusch, 2014). Together, these results suggest a peripheral role for the eIF3i and eIF3g subunits of the mRNA-entry-channel arm in discriminating against start codons in poor Kozak context.

In light of previous work implicating eIF3 in both mRNA recruitment and start-codon recognition and our results here, we next compared the effects we observed in eIF3a/b Degron and eIF3i DDKK cells to those previously reported for mutations targeting other initiation factors. The effects of mutations targeting eIFA, eIF4B, and Ded1—which appear to contribute either to initial PIC docking or scanning—have previously been investigated using ribosome profiling (Sen et al., 2015, 2016). Specifically, these studies investigated the effect of temperature-sensitive alleles of eIF4A and Ded1, and a deletion of eIF4B. Importantly, the sequencing libraries from which these datasets were obtained were prepared by addition of cycloheximide to media in which cells exhibited strong translational defects, as were our sequencing libraries.

We found significant overlaps (Figure 4A, left panel) in the specific mRNAs experiencing significant TE_rel decreases in eIF3i DDKK or eIF3a/b Degron cells and those mRNAs whose TE_rel was significantly decreased in the presence of mutations targeting eIF4A (P < 10⁻⁶ and P < 10⁻¹⁹ for eIF3i DDKK and eIF3a/b Degron, respectively; HyperGeometric test), eIF4B (P < 10⁻²⁰ for both), and Ded1 (P < 10⁻²⁰ and P < 10⁻¹¹). We similarly found significant overlaps when comparing the sets of mRNAs whose TE_rel increased in the presence of these mutations (eIF4A, P < 10⁻²⁰ and P < 10⁻¹⁷; eIF4B, P < 10⁻¹³ and P < 10⁻²⁰; Ded1, P < 10⁻¹⁰ and P < 10⁻¹¹). Consistent with these overlapping effects, we also observed significant positive correlations between the ∆TE_rel values we observed in both eIF3i DDKK and eIF3a/b Degron cells and ∆TE_rel values observed in the presence of mutations targeting eIF4A, eIF4B, and Ded1 (Figure 4B and Supplementary Figure S7). Global comparison of the magnitude and direction of observed ∆TE_rel values from these distinct experiments further revealed a similar transcriptome-level portfolio of effects (Figure 4C, right panel).

Having observed these similarities, we next compared the effects observed in eIF3i DDKK and eIF3a/b Degron cells to those obtained from cells expressing an eIF1 variant (L96P) that increases recognition of both AUG start codons in poor context and near-cognate uORF start codons (Zhou et al., 2020). The sequencing libraries giving rise to this dataset were also prepared under conditions similar to those we employed in our experiments. In contrast to the significant overlaps we observed when comparing to datasets from eIF4A, eIF4B, or Ded1 mutant cells, we only found a significant overlap between mRNAs displaying TE_rel decreases in eIF3i DDKK and L96P eIF1 cells (P < 10⁻⁵) and not in eIF3a/b Degron cells or for mRNAs displaying TE_rel increases in either eIF3 background (Figure 4A). Consistent with this, we observed a significant but negative correlation between ∆TE_rel values from eIF3a/b Degron and L96P eIF1 cells and no significant correlation between eIF3i DDKK and L96P eIF1 cells (Figure 4B and Supplementary Figure S7). Moreover, global comparison of the observed ∆TE_rel values from these datasets further revealed a distinct pattern of transcriptome-wide effects in L96P eIF1 cells when compared to either eIF3i DDKK or eIF3a/b Degron cells (Figure 4C, left panel). We also observed weak but modestly significant correlations in ∆TE_rel values when comparing our eIF3 datasets to a ribosome profiling dataset obtained from cells expressing R13P eIF1A, in which discrimination against near-cognate codons or AUG codons in poor context was increased (Supplementary Figure S7). However, the sequencing libraries for this dataset were prepared from cells harvested in the absence of cycloheximide, which complicates their comparison to our sequencing results.

Together with the more pronounced overlap we observed when comparing our datasets with those obtained from cells expressing mutant versions of eIF4A, eIF4B, and Ded1, these results suggest that eIF3 and its mRNA-entry-channel arm contribute to initial docking and scanning of the mRNA, whereas their contributions to start-codon recognition may be less critical or peripheral to those of eIF1 and eIF1A.

Because of the similarities we observed in the effects of eIF3i DDKK and eIF3a/b Degron mutations and mutations targeting eIF4A, eIF4B, and Ded1, we investigated whether these similarities extended to the observation that long transcripts and transcripts with lower closed-loop-forming potential are particularly sensitive to deletion of eIF4B or to mutations of eIF4A or Ded1 (Sen et al., 2016). In fact, we observed significant negative correlations between ∆TE_rel and both overall transcript length and coding sequence (CDS) length in both eIF3i DDKK and eIF3a/b Degron cells (Figure 5A). Comparing the overall length of mRNAs whose TE_rel decreased in each mutant eIF3 background reveals them to be significantly longer than both unaffected mRNAs (P_adj < 10⁻¹³ and P_adj < 10⁻¹⁶ for eIF3i DDKK and eIF3a/b Degron, respectively; Wilcoxon test) and mRNAs whose TE_rel increased (P_adj < 10⁻¹⁴ and P_adj < 10⁻¹⁶) (Figure 5B). Similarly, the CDS lengths of mRNAs whose TE_rel decreased in the presence of each eIF3 mutation is longer than both unaffected mRNAs (P_adj < 10⁻¹¹ and P_adj < 10⁻¹⁶) and mRNAs whose TE_rel increased (P_adj < 10⁻⁵ and P_adj < 10⁻¹⁶).

We further observed that, as in cells expressing mutations of eIF4A or Ded1 or in which the gene coding for eIF4B was deleted, mRNAs identified in a previous study (Costello et al., 2015) as having strong closed-loop-forming potential owing to their enrichment in eIF4G, eIF4E, and PABP (group 3) are less sensitive than other mRNAs to both the eIF3i DDKK (P_adj < 10⁻⁷; P_adj < 10⁻¹⁴; and P_adj < 10⁻¹⁴ for comparison to groups 1, 2, and 4 respectively) or eIF3a/b Degron (P_adj < 10⁻¹²; P_adj < 10⁻⁴; and P_adj < 10⁻¹⁵) mutations (Figure 5C). Moreover, we observed positive median ∆TE_rel values for strong closed-loop mRNAs (group 3) and negative median ∆TE_rel values for weak closed-loop mRNAs (groups 1 and 2), suggesting that mRNAs with strong closed-loop-forming potential compete more effectively for the initiation machinery when eIF3 or its mRNA-entry-channel arm are disrupted, whereas mRNAs less likely to form closed-loop structures are disadvantaged under these conditions. Taken together, these results suggest that, like eIF4A, eIF4B, and Ded1, eIF3 and its mRNA-entry-channel arm may contribute to driving initiation on long mRNAs less likely to form eIF4G-, eIF4E-, and PABP-dependent closed loop structures in vivo. While the translation of most mRNAs likely depends on contributions from these factors, translation of these long and closed-loop-dependent mRNAs is particularly sensitive to their disruption.

Because the eIF3i DDKK and eIF3a/b Degron mutations mimic the loss of either the eIF3i and eIF3g subunits (eIF3i DDKK) or the entire eIF3 complex (eIF3a/b Degron), we reasoned that comparing those mRNAs whose TE_rel was affected uniquely in eIF3a/b Degron cells to those whose TE_rel was affected in both eIF3a/b Degron and eIF3i DDKK cells might disentangle the roles of these distinct regions of the eIF3 complex. mRNAs whose TE_rel was affected in both cell lines might depend more heavily on the contributions of the eIF3i and eIF3g subunits of the eIF3 mRNA-entry-channel arm, whereas mRNAs whose TE_rel was affected solely in eIF3a/b Degron cells might depend more heavily on the contributions of other eIF3 subunits (or the collaboration of subunits within the intact complex) for their translation.

To this end, we compared the features of mRNAs whose TE_rel significantly increased or decreased either in both strains or exclusively in eIF3a/b Degron cells (Figure 6A). mRNAs whose TE_rel decreased in both eIF3i DDKK and eIF3a/b Degron cells possess longer 5′-UTRs than those whose TE_rel decreased only in eIF3a/b Degron cells, when restricting our analysis to mRNAs previously identified as having a dominant 5′ transcript isoform (P < 10⁻⁹). In contrast, there is no significant difference in the 5′-UTR lengths of mRNAs whose TE_rel increased. We observed similar results when comparing 5′-UTR lengths reported in a separate study (Kertesz et al., 2010) (Supplementary Figure S5B). We also observed no difference in the degree of structural complexity, as measured by various PARS metrics (Figure 2A), of mRNAs whose TE_rel was affected either in both eIF3 mutant backgrounds or solely in eIF3a/b Degron cells. These results suggest that the eIF3i and eIF3g subunits of the mRNA-entry-channel arm may be specifically required for the contributions of eIF3 to processive scanning through long 5′-UTRs, whereas the other subunits of the eIF3 complex, either independently or in collaboration with eIF3i and eIF3g, participate in its contribution to the resolution of structural impediments during initial mRNA docking and scanning.

We next compared the differential sensitivity of affected mRNAs within these groups to mutations targeting eIF4A, eIF4B, Ded1, or eIF1 (Sen et al., 2015; Sen et al., 2016; Zhou et al., 2020). mRNAs whose TE_rel increased or decreased in both eIF3i DDKK and eIF3a/b Degron cells were significantly more sensitive to mutations targeting eIF4A, eIF4B, or Ded1 than those uniquely affected in eIF3a/b Degron cells: mRNAs whose TE_rel decreased in both mutant eIF3 cell lines displayed greater TE_rel decreases in response to mutations targeting these factors (P < 10⁻⁹, P < 10⁻⁸, and P < 10⁻¹² for eIF4A, eIF4B, and Ded1, respectively; ANOVA) and mRNAs whose TE_rel increased displayed stronger TE_rel increases in these datasets (P < 10⁻¹⁶ for all comparisons, Figure 6B). In contrast, we observed no significant differences in the relative sensitivity of affected mRNAs to a mutation targeting eIF1. These differential sensitivities are consistent with a role for the mRNA-entry-channel arm in collaborating with eIF4A, eIF4B, and Ded1 during mRNA recruitment.

Finally, mRNAs whose TE_rel increased in both eIF3i DDKK and eIF3a/b Degron cells possess significantly weaker context scores than transcripts whose TE_rel increased uniquely in eIF3a/b Degron cells (P < 10⁻⁸, Figure 6C). Consistent with this, sequence logos reveal that mRNAs whose TE_rel increased in both mutant eIF3 cell lines display a weaker preference for adenine at the -3 position (Figure 6D). These observations suggest that eIF3i and eIF3g, and by extension the eIF3 mRNA-entry-channel arm, may play a role in discriminating against AUG codons in poor context.