In this study, we have examined the genome-wide consequences for translational efficiency of simultaneously eliminating eIF4G2 and depleting eIF4G1 from yeast cells. The conditional depletion of eIF4G1 achieved using a degron-tagged version of this protein was highly effective and reduced the polysome content and rate of translation to only 20-30% of WT levels, indicating a substantial reduction in the rate of translation initiation. We used genome expression microarrays to measure the abundance of each mRNA in heavy polysomes (with 4 or more translating ribosomes) relative to its level in total mRNA to calculate translational efficiencies of 5868 different genes. The results indicated that the overwhelming majority of mRNAs experienced only a moderate change in translational efficiency on eIF4G depletion. Less than 2% of the genes showed a statistically significant decrease in TE in the mutant by a factor of 1.4 of more, and the genes in this group that were affected the most displayed reductions of a factor of ~2.5 or less. While the actual percentage of genes affected to this extent is probably higher, only ~10% of genes exhibited decreases in TE of this magnitude for each biological replicate, which likely represents the upper size limit for this category. Thus, we did not detect even a small group of mRNAs that are dramatically dependent on eIF4G for translation in vivo.
We made the unexpected observation, however, that depletion of eIF4G narrows the range of translational efficiencies for a large fraction of mRNAs, decreasing the number with efficiencies that are substantially higher or lower than the genome-average TE. This trend is well illustrated in the log-log plots of mean TE values in WT versus mutant cells (Figure ), and also by the fact that depleting eIF4G reduced (by several fold) the numbers of mRNAs with TE values either 1.5-fold higher, or 1.5-fold lower, than unity (Table ). Furthermore, the bulk of mRNAs with TE values ≥ 1.5 in WT cells are, at least to some extent, dependent on eIF4G for their higher than average TE values (Table ). This dependence is consistent with a significant role for eIF4G in stimulating one or more steps of initiation for the most efficiently translated mRNAs in the cell, presumably the activation of mRNA for recruitment of the 43S PIC, scanning the 5'UTR, or start codon recognition. Unexpectedly, we found that many mRNAs with lower than average TE values in WT cells exhibit an increased translational efficiency on eIF4G depletion. It is conceivable that eIF4G directly impairs the translation of these latter mRNAs. However, we favor an indirect mechanism involving competition among all mRNAs for limiting initiation factors or PICs, coupled with the role of eIF4G in stimulating efficiently translated mRNAs at the expense of those with lower than average efficiencies. In the absence of eIF4G, this competitive edge would be eliminated for the first group and thereby enable the second group of mRNAs to compete better for limiting factors/PICs.
The small group of ~100 genes we identified that are most dependent on eIF4G for their higher than average TEs in WT cells contain a mean 5'UTR length that is slightly below the genome-average for all mRNAs (Figure ), a feature that should facilitate efficient scanning and AUG recognition. This was surprising because we expected to find that the mRNAs most dependent on eIF4G would have long or highly structured 5'UTRs, requiring the eIF4G·eIF4A complex for unwinding secondary structure to promote 43S attachment or scanning. In fact, the ~100 genes we identified whose translation is stimulated the most by eliminating eIF4G contain a mean 5'UTR length substantially larger than the genome average (Figure ). The fact that these latter mRNAs display a lower than average TE in WT cells and benefit from the absence of eIF4G seems to indicate that they function inefficiently at steps of initiation not significantly enhanced by eIF4G. Given their long 5' UTR lengths, it seems likely that scanning to the start codon is relatively inefficient for these mRNAs. If so, then the fact that depleting eIF4G does not exacerbate this deficiency suggests that factors besides eIF4G are more critically required for efficient scanning through long 5'UTRs in yeast.
This last suggestion is consistent with our finding that none of the 17 mRNAs predicted by the Randfold program to contain the most stable secondary structures among yeast 5' UTRs [16
] displayed a significant reduction in TE on eIF4G depletion--in fact, four such mRNAs appear to be translated more efficiently on eIF4G depletion. Thus, other initiation factors besides eIF4G might also be more critically involved in removing secondary structures in advance of the scanning PIC. This view is supported by the fact that in a mammalian reconstituted system, eIF4G, eIF4A and eIF4B are sufficient for 43S attachment and scanning on β-globin mRNA, which harbors a relatively unstructured 5'UTR, whereas the DExH-box protein DHX29 is required for initiation complex assembly on mRNAs containing more structured 5'UTRs [7
]. Similarly, there is evidence that yeast DEAD-box protein Ded1 contributes more than eIF4A does to the processivity of scanning in vivo [6
]. These findings are in agreement with the possibility that the eIF4E/eIF4G/eIF4A complex (eIF4F) is more critical for 43S PIC attachment near the 5' end of the mRNA than for subsequent scanning to the start codon.
Thus, our results are consistent with the model that 43S attachment is a rate-limiting step for a large proportion of mRNAs with higher than average TEs, and that this step is stimulated by eIF4G, particularly for the ~100 genes we identified with the greatest dependence on eIF4G that contain relatively short 5'UTRs. By contrast, scanning or AUG recognition would be rate-limiting for mRNAs with longer than average 5'UTRs whose translation is enhanced by depletion of eIF4G, because these steps are not critically dependent on eIF4G. The fact that eliminating eIF4G mitigates the lower than average translational efficiencies of this second group of mRNAs can be explained by proposing that the negative effect of depleting eIF4G on 43S attachment is outweighed by their enhanced ability to compete with other mRNAs for limiting factors that promote scanning or AUG recognition.
Fulfilling this last stipulation of our model would be facilitated if the inefficient mRNAs with long 5'UTRs are relatively ineffective at exploiting eIF4G function in 43S attachment. That is, if eIF4G contributes relatively less to 43S attachment by these inefficient mRNAs in WT cells, then depleting eIF4G would produce relatively smaller reductions in their translation rate. One reason for thinking that this condition holds is our finding that this group of mRNAs also displays unusually long coding sequences, whereas the mRNAs we identified with the greatest dependence on eIF4G exhibit smaller than average ORF lengths. Recent findings by Jacobson et al [20
] indicate that shorter yeast mRNAs produce more stable eIF4F-cap interactions than do longer mRNAs, which is fully dependent on an extended poly(A) tail and PABP. Presumably, shorter mRNAs more efficiently assemble a closed-loop mRNP via PABP-eIF4G interaction, which stabilizes eIF4F binding to mRNA [3
]. In fact, the possibility of less efficient 5'-3' interaction for larger mRNAs was advanced previously as one explanation for the inverse correlation between ribosome density and ORF length [14
], which we confirmed here using TE values (Figure S3A in Additional file 2
). Hence, we suggest that longer mRNAs are affected less than shorter mRNAs by the elimination of eIF4G because the eIF4F-cap interaction is inherently less stable for longer transcripts and, hence, less efficacious in promoting 43S recruitment when eIF4G is present. The fact that depleting eIF4G diminishes, but does not eliminate the correlation between TE and ORF length (Figure S3B in Additional file 2
) indicates that reduced eIF4G-PABP interaction is not the only factor limiting the translation of mRNAs with longer ORFs, and limited processivity of elongating ribosomes or less efficient termination have been suggested as other possibilities [14
We showed previously that depletion of eIF4G did not lower the amounts of native 48S complexes containing the RPL41A
], both very short transcripts, which is ostensibly at odds with the idea that eIF4G has an important function in 43S attachment to mRNA. Examining the results we obtained for these mRNAs in the LP dataset (from small polysomes) reveals that they both exhibit mean TE4G
values ~90% of their TEWT
values (Additional file 3
). Thus, even if we assume that these two mRNAs require eIF4G only at the step of 43S attachment to achieve their maximum translation rates, it would have been very difficult to detect a 10% decrease in the levels of their free 48S complexes with the techniques employed in the previous study [8
]. It remains to be determined what features in mRNA, besides a short 5'UTR and short ORF length, are responsible for the more pronounced requirement for eIF4G displayed by the small fraction of yeast mRNAs identified here.
Considering that eIF4G is essential in yeast, and also noting its role as a protein bridge linking the eIF4E-mRNA-PABP mRNP to components of the 43S complex (eIF5 and eIF1) [22
], it is surprising that a significant amount of translation still proceeds in the absence of this factor. Based on our microarray data, it appears that eIF4G is dispensable for the translation of most, if not all mRNAs in vivo, indicating that it is rate-enhancing rather than essential in budding yeast. This stands in contrast to the critical requirement for the eIF3 complex, which is required for nearly all translation in yeast, and is crucial for attachment of native 43S complexes to mRNAs (RPL41A
that can assemble 48S PICs in cells depleted of eIF4G [8
]. Of course, we cannot exclude the possibility that a compensatory initiation pathway comes into play during the 8 h of incubation in the non-permissive conditions used to thoroughly deplete eIF4G. It is also impossible to eliminate the possibility that a very small fraction of the WT amount of eIF4G, below the detection limit of our Western analysis, is sufficient to catalyze the residual protein synthesis that occurs in the depleted cells. This seems unlikely, however, because the eIF4G level in WT cells is already lower than those of nearly all other initiation factors [23
On the other hand, the 3 to 4-fold reduction in the rate of translation, and the narrowed range of translational efficiencies evoked by depletion of eIF4G, could have serious consequences for a subset of dosage-sensitive proteins with essential functions in the cell. Moreover, cell division could be blocked under these conditions by regulatory mechanisms that respond to a drop in the rate of synthesis of a key cell cycle controlling factor, eg., the G1 cyclin Cln3 [24
]. Considering that cell division is not blocked by a decrease in the overall translation rate of ~70% occurring in response to hyperosmotic stress [27
], eIF4G depletion might evoke a comparatively greater reduction in translation of a key protein(s) required for cell division than occurs during osmotic stress.
Given that depletion of eIF4G reduces the translation rate by 3 to 4-fold, it is surprising that the average TE calculated for all 5868 genes decreased only a small amount, from 1.100 ± 0.006 in WT cells to 1.05 ± 0.004 in the mutant. Of course, many genes translated with higher than average efficiencies in WT exhibit much larger reductions in TE values on depletion of eIF4G, but this effect was counterbalanced by increased translation of many genes with lower than average TEWT
values. As noted above, the fact that microarray results are normalized to give each array the same average signal intensity will dampen the reduction in polysomal mRNA abundance in the eIF4G mutant, and the amounts of total mRNA might also decline on eIF4G depletion, which would offset the effect of decreased polysomal mRNA on the calculated TE values. It is also conceivable that eIF4G depletion triggers a signal transduction response that decreases the rate of elongation, counteracting the effect of reduced initiation on polysome size. For example, oxidative stress reduces the rates of both initiation and elongation in yeast [29
Because we examined cells lacking eIF4G2 and depleted of eIF4G1, it could be argued that the changes in translational efficiencies we observed result primarily from the absence of only eIF4G1 or eIF4G2 rather than the elimination of both eIF4G isoforms. This is unlikely in view of recent findings by Clarkson et al on mutant strains expressing only eIF4G1 or eIF4G2 and engineered to express each isoform at a level equivalent to the combination of both isoforms in WT. These strains displayed almost no changes in translational efficiency genome-wide [30
], providing strong evidence against the possibility that eIF4G1 or eIF4G2 is specifically required to support the translation of particular mRNAs. In this same study, two groups of protein-coding genes (of ~150 each) displayed a significant change in translational efficiency on deletion of only TIF4631
, encoding the major isoform (eIF4G1), which reduced the growth rate and polysome content relative to the isogenic WT strain. Only 10% of the genes with significantly repressed translational efficiencies in tif4631Δ
cells thus identified by Clarkson et al belong to the group of ~100 genes we identified here with mean TE4G
ratios of ≤ 0.71. However, the group of translationally repressed genes in the Clarkson et al study displayed an average TE4G
ratio in our experiments (0.89 ± 0.01) that is significantly below the genome-average TE4G
ratio (1.05 ± 0.004) and also the average TE4G
ratio determined in our experiments for the group of translationally enhanced genes identified by Clarkson et al (1.17 ± 0.02). Thus, the translational efficiencies of at least a subset of genes are affected similarly by the absence of eIF4G1 alone and the elimination of both eIF4G1 and eIF4G2 simultaneously. This is consistent with the conclusion that eIF4G1 and eIF4G2 perform essentially identical functions [30
A recent analysis of the consequences of depleting eIF4GI and eIF4GII with siRNAs in cultured mammalian cells [31
] reached certain conclusions congruent, and others that seem to differ, from our findings. It was found that depleting both eIF4GI and eIF4GII reduced overall translation by only ~20%, whereas depleting two eIF3 subunits provoked a stronger (~50%) reduction, consistent with the greater requirement for eIF3 versus eIF4G we observed in yeast [8
]. eIF4GI depletion reduced the translational efficiencies of a subset of mammalian mRNAs, including a group whose products function in mitochondrial regulation, bioenergetics, and cell proliferation. In accordance with our observations, there was no significant correlation between the presence of long or structured 5'UTRs and the degree of eIF4GI-dependence. This is consistent with the aforementioned suggestion that eIF4GI is more important for 43S attachment than for subsequent scanning through the 5'UTR. At odds with our results, however, the eIF4GI-dependent class of mRNAs appeared to be somewhat enriched in those containing uORFs, and the presence of an uORF was shown to increase the eIF4GI dependence on translation. One possibility is that the majority of uORF-containing mRNAs in yeast do not support appreciable reinitiation in WT cells, as this process has strict requirements for uORF length and cis-acting sequences surrounding the stop codon [32
]. In this event, eliminating the potential role of eIF4G in stimulating reinitiation would be difficult to detect on a genome-wide basis in yeast.