High-quality sequencing libraries prepared from HeLa cells expressing different eIF3 subcomplexes

To investigate the impact of different human eIF3 subunits and partial eIF3 sub-complexes, which arise in cells as a result of unbalanced eIF3 expression, on translational efficiency transcriptome-wide, we performed Ribo-Seq together with RNA-Seq, which allowed us to draw conclusions about gene expression at the transcriptional and translational levels and their interdependence. In particular, using the siRNA technology, we examined HeLa cells with reduced levels of eIF3d, eIF3e, or eIF3h subunits and cells treated with non-targeting (NT) siRNA. These eIF3 subunits were selected because the eIF3d depletion leads to a reduction in protein levels of only eIF3d and leaves the rest of the eIF3 complex intact, whereas downregulation of eIF3e or eIF3h generates partial eIF3 subcomplexes as previously described (Wagner et al., 2016). Specifically, targeting eIF3e with siRNA results in reduced protein levels of eIF3e, eIF3d, eIF3k, and eIF3l, whereas downregulation of eIF3h produces a partial eIF3 complex lacking subunits eIF3h, eIF3k, and eIF3l (Figure 1B). In terms of impact on cell growth, both eIF3d^KD and eIF3e^KD strongly reduce proliferation, while the effect of eIF3h^KD is only moderate (Figure 1—figure supplement 1A, Wagner et al., 2016).

We first performed several control experiments with aliquots of cell lysates made to prepare individual Ribo-Seq libraries to document an efficient downregulation of the targeted eIF3 subunit, as well as the expected phenotype (Wagner et al., 2016). Using qPCR, we measured mRNA levels of the siRNA targeted eIF3 subunits (eIF3d, eIF3e, eIF3h), as well as of eIF3b as a negative control and a housekeeping gene ALAS1 for normalization purposes. Routinely, a difference of 4–5 cycles was achieved for each downregulated subunit, corresponding to ~16–32 fold decrease in mRNA levels; the mRNA levels of the control subunit eIF3b remained unchanged (Figure 1—figure supplement 1B).

Next, we determined protein levels of selected eIF3 subunits by Western blotting. Protein levels of all eIF3 subunits targeted individually by siRNA were strongly reduced (Figure 1—figure supplement 1C). Depletion of eIF3e also induced a marked decrease in the eIF3d, eIF3k, and eIF3l protein levels (Figure 1—figure supplement 1C and data not shown), as previously reported (Wagner et al., 2016). Finally, we performed polysome profiling to assess the impact on translation and obtained essentially the same results as before (Wagner et al., 2016). The eIF3d^KD showed the most pronounced defect in translation initiation, marked by a robust increase of the 80 S monosome peak and a concomitant decrease of polysome content, similar to eIF3e^KD, while eIF3h^KD had only a modest effect on polysome formation (Figure 1—figure supplement 1D).

Ribo-Seq libraries were generated from each knock-down and NT control in three biological replicates as described in Ingolia et al., 2012. RNA-Seq libraries were prepared in four biological replicates using the SMARTer smRNA-Seq Kit to normalize for mRNA levels and calculate translational efficiency (TE). Both Ribo-Seq and RNA-seq libraries were highly reproducible as shown by the Spearman correlation coefficient (Figure 2—figure supplement 1A). All Ribo-Seq libraries showed a similar read length distribution with a prominent peak at 30–32 nt and an expected 3 nt periodicity in coding sequences (CDS; Figure 2—figure supplement 1B, C and Figure 2—figure supplement 2A). As expected, Ribo-seq libraries were strongly enriched for CDS reads compared to RNA-Seq libraries (Figure 2—figure supplement 2B, C). The PCA plot and hierarchical clustering (Figure 2A and Figure 2—figure supplement 3) showed clustering of the samples into two main groups: Ribo-Seq and RNA-seq, and also into two subgroups; NT and eIF3h^KD samples clustered on one side and eIF3e^KD and eIF3d^KD samples on the other. These results suggest that the eIF3h depletion has a much milder impact on the translatome than depletion of eIF3e or eIF3d, which agrees with the growth phenotype and polysome profile analyses (Figure 1—figure supplement 1A, D).

eIF3 subcomplexes alter translation efficiency of a large proportion of genes

Using the R DESeq2 package (Love et al., 2014), we identified transcripts with significantly altered TE (P_adjusted <0.05) for each knock-down compared to the NT control (Figure 2B, D, Supplementary file 1), henceforth referred to as Differential Translation Efficiency Genes (DTEGs) (Chothani et al., 2019). In total, we identified 21 DTEGs in eIF3h^KD (19 downregulated and 2 upregulated), 1180 DTEGs in eIF3e^KD (622 down and 558 up), and 1971 DTEGs in eIF3d^KD (1040 down and 931 up) out of a total of 17520 genes that were assigned TE values.

The number of DTEGs in each knock-down correlated well with the severity of the overall phenotype and with the impact on global translation, as judged from the polysome profiles. eIF3d^KD with the strongest decrease in polysomes had the highest number of DTEGs, whereas eIF3h^KD with only 21 DTEGs had a marginal effect on polysome formation (Figure 1—figure supplement 1D). Consistently, the overall distribution of fold-changes in TE showed most pronounced changes in eIF3d^KD (Figure 2E). It should be noted, however, that in contrast to polysome profiling, differential expression analysis of Ribo-Seq data normalizes out global changes in translational rates, and reveals only specific shifts of ribosome localization on individual mRNAs between the NT and KD datasets. Therefore, the fact that all three knock-downs showed an increase in TE (not just a decrease) for approximately half of the affected genes, clearly points to an impairment of a specific mode of regulation rather than simply to a non-specific global translational shutdown.

Interestingly, there is a considerable overlap between the knock-downs studied, such that only 1045 DTEGs are unique for eIF3d^KD and the rest (926) are shared with other two knock-downs; in particular 923 with eIF3e^KD and 21 with eIF3h^KD – in the latter case, the overlap reaches 100% (Figure 2F, Figure 2—figure supplement 4A–C). Remarkably, eIF3d^KD affects neither the expression of any other eIF3 subunit nor the integrity of the eIF3 complex, yet it has the most pronounced impact on translation (Wagner et al., 2016), correlating well with the highest number of DTEGs identified. In case of eIF3e^KD, only 257 DTEGs are unique and the rest (923) are shared with eIF3d^KD, 18 of which are also shared with eIF3h^KD (Figure 2F, Figure 2—figure supplement 4A–C). The eIF3 subcomplex formed in eIF3e^KD lacks subunits e, k and l and partially also eIF3d (Wagner et al., 2016) – the significantly reduced protein levels of eIF3d are evident in Figure 1—figure supplement 1C. Thus, it is not surprising that the DTEGs identified in eIF3e^KD largely overlap with those identified in eIF3d^KD; in fact, this result supports the robustness and accuracy of this approach. The fact that the magnitude of the knock-down effect is bigger in eIF3d^KD over eIF3e^KD is discussed below.

eIF3d^KD and eIF3e^KD increase TE of mRNAs encoding proteins associated with membrane organelles

To analyze enrichment in KEGG pathways and GO Biological Processes in a meaningful way, we analyzed the DTEGs divided not only into upregulated and downregulated groups, but also into specific subgroups – eIF3d^KD ‘all’, eIF3e^KD ‘all’, eIF3d^KD ‘unique’, eIF3e^KD ‘unique’ and eIF3d^KD/eIF3e^KD ‘common’ (Figure 2F–H). To disentangle the effects of the changes in mRNA abundance vs. translational regulation, we also investigated translationally exclusive DTEGs whose mRNA abundance did not change significantly and therefore the change in TE can be attributed solely to translational regulation; eIF3d^KD ‘translation only’ and eIF3e^KD ‘translation only’, as well as translationally buffered DTEGs whose mRNA abundance changes in opposite direction than translational regulation, that is mRNA increases but TE decreases so the net protein production is not changed or even decreased; eIF3d^KD ‘buffered’ and eIF3e^KD ‘buffered’ (Figure 2—figure supplement 4D–F). For schematic illustrating classification of genes based on fold changes of footprints (FP), mRNA, and TE, see Figure 2—figure supplement 4G, H.

Analysis of DTEGs upregulated in eIF3d^KD and eIF3e^KD showed similar results for almost all groups analyzed (eIF3d^KD ‘all’, ‘unique’, and ‘translation only’, eIF3d^KD/eIF3e^KD ‘common’, as well as eIF3e^KD ‘all’ and ‘translation only’; Figure 3A, B, and Figure 3—figure supplement 1A), with the strongest enrichment in the KEGG ‘Lysosome’ and ‘Protein processing in endoplasmic reticulum’ pathways. The majority of DTEGs in these pathways were solely translationally regulated meaning that their mRNA levels stayed unchanged while the TE increased. For example, for the KEGG pathway ‘Lysosome’, more than 70% (i.e. 37 out of the 51) DTEGs upregulated in the eIF3d^KD ‘all’, fall into the eIF3d^KD ‘translation only’ group. This suggests that when translation initiation is compromised by eIF3d/e^KD, cells may increase protein degradation and turnover by translationally upregulating lysosomal and ER proteins. Interestingly and in accord with our results, Lin et al. detected increased lysosomal load in MCF-10A cells downregulated for eIF3e (Lin et al., 2020). They speculated that this is due to the need to clear defective mitochondria. In support, we also found enrichment of mitochondria-associated DTEGs, especially mitochondrial ribosomal proteins (see below).

Among the eIF3d^KD ‘unique upregulated’ DTEGs, we identified one interesting KEGG pathway, the ABC transporters, which did not show up in other gene groups (Figure 3—figure supplement 1A, in green). A total of 12 different ABC transporters had elevated TE (9 of them are unique to eIF3d^KD, while 3 were also found in eIF3e^KD), 6 of which (ABCC1-5, ABCC10) belong to the C subfamily, known to confer multidrug resistance with alternative designation as multidrug resistance protein (MRP1-5, MRP7; Sodani et al., 2012). Interestingly, all six of these ABCC transporters were upregulated solely at the translational level (Supplementary file 1).

eIF3e^KD increases production of ribosomal proteins

Interestingly, the KEGG pathway ‘Ribosome’ and GO terms connected with translation and ribosome biogenesis were enriched amongst DTEGs upregulated specifically in eIF3e^KD ‘unique upregulated’ DTEGs (Figure 3C and Figure 3—figure supplement 1B). We then compared the significance of enrichment for ‘Ribosome’ KEGG term between eIF3e^KD and eIF3d^KD ‘all upregulated’ groups. This was the only case where a term or a pathway was more significantly enriched in eIF3e^KD (p=5.199E-10) than in eIF3d^KD (NS, p=0.1183) (Figure 3—figure supplement 1A, in green). Intrigued by this result, we filtered out all ribosomal proteins (RPs) from our data set (according to HGNC database, Tweedie et al., 2021), regardless of the TE changes, which consisted of 89 cytoplasmic and 77 mitochondrial RPs, and analyzed them separately (Supplementary file 1). In total, we identified 32 RPs with significantly changed TE in eIF3e^KD (14 cytoplasmic and 18 mitochondrial) and only 16 in eIF3d^KD (9 cytoplasmic and 7 mitochondrial). Moreover, all the cytoplasmic RPs with increased TE in eIF3e^KD (and most mitochondrial ones) were upregulated solely at the translational level and their mRNA levels were not changed. We selected several RPs with different combinations of changes in TE or FP or mRNA levels in both eIF3e^KD and eIF3d^KD (Figure 3D) and subjected them to western blotting. All ribosomal proteins tested showed visibly increased protein levels in both eIF3e^KD and eIF3d^KD (Figure 3E, F), while their mRNA levels remained unchanged (Figure 3—figure supplement 1C).

As the production of the ribosomes seemed to be increased, we next examined the 60 S/40 S subunit ratio to control for unbalanced ribosome biogenesis. We performed polysome profiling in the presence of EDTA that splits all ribosomes into 40 S and 60 S subunits and showed that the 60 S/40 S ratio remained unaffected (Figure 3G). Furthermore, we performed a similar experiment with exactly 10 million cells per each sample and confirmed that the total amount of ribosomes in cells is indeed increased in eIF3e^KD and eIF3d^KD (Figure 3H).

Overall, our results suggest that eIF3e and, to some extent, its binding partner eIF3d maintain control over production of RPs in the cell. A similar role has recently been attributed to eIF3k, which has been shown to specifically repress the synthesis of the small ribosomal protein RPS15A, and cells depleted of eIF3k showed increased ribosomal content (Duan et al., 2023). As mentioned above, targeting eIF3e with siRNA results in co-downregulation of not only eIF3e and eIF3d, but also of eIF3k and eIF3l (Wagner et al., 2016) which may at least partially explain these findings.

In addition to mitochondrial RPs, we also observed increased TE for several other mitochondrial proteins from the Complex I, Complex IV and Complex V (KEGG term Oxidative phosphorylation hsa00190; rank 17; #of genes 12 out of 134; p=0.0003431) in eIF3e^KD, but many of these transcripts, mainly from the Complex I, are encoded by mitochondrial DNA and translated in mitochondria directly, so these results should be verified by Mitoribosome profiling, as the traditional Ribo-Seq method is not suitable for this (Pearce et al., 2021).

Because RPs and other components of the translational machinery frequently harbor 5’ terminal oligopyrimidine (TOP) motifs that render their translation highly sensitive to mTORC1 (Philippe et al., 2020; Stolovich et al., 2002), we first investigated whether mTORC1 translation is affected in eIF3d^KD or eIF3e^KD but found no significant change in TE for any mTORC1 components. Next, we examined the effect of depletion of eIF3 subunits on translation of all known TOP mRNAs. We observed that transcripts harboring TOP motifs (Philippe et al., 2020) were significantly translationally activated in eIF3d^KD and eIF3e^KD, but not in eIF3h^KD (Figure 4A–C). Interestingly, the effect of eIF3d^KD and eIF3e^KD on translation of non-TOP mRNAs that are mTORC1-sensitive (Gandin et al., 2016) differed from what was observed for TOP mRNAs, as non-TOP transcripts were predominantly translationally offset (buffered), such that translation efficiency was increased to maintain a constant level of polysome association despite decreased total mRNA abundance (Figure 4—figure supplement 1A–C). Together, these findings suggest that eIF3e and eIF3d may play a role in tempering translation of TOP mRNAs under normal cellular conditions, and that the loss of these eIF3 subunits has distinct impact on the translatome as compared to mTORC1.

eIF3d^KD and eIF3e^KD decrease translation efficiency of the key components of the MAPK/ERK and other signaling pathways implicated in cancer

The eIF3d^KD or eIF3e^KD ‘all downregulated’ and eIF3d^KD/eIF3e^KD ‘common downregulated’ groups were enriched for a similar set of KEGG pathways (Figure 5A and Figure 5—figure supplement 1A). The effect is generally stronger for eIF3d^KD, which simply means that this knock-down has a higher number of DTEGs in the same pathway than eIF3e^KD (for example ‘MAPK signaling pathway’ has 42 DTEGs in eIF3d^KD vs. 30 DTEGs in eIF3e^KD). Therefore, we focused primarily on DTEGs identified in eIF3d^KD ‘all’ (Figure 5A and Figure 5—figure supplement 1A). Most of the top 10 KEGG results are related to and/or encompass the ‘MAPK signaling pathway’ term (Figure 5—figure supplement 1B), like ‘Chronic myeloid leukemia’ and the ‘Neurotrophin signaling pathway’. Depletion of eIF3d affected multiple components of the MAPK signaling pathways, which plays a critical role in the regulation of cell proliferation (Zhang and Liu, 2002) and cancer development and progression (reviewed in Dhillon et al., 2007). These include the docking protein Grb2 (encoded by GRB2) and guanine nucleotide exchange factor Sos (SOS1), through kinases c-Raf (RAF1), MEK1 (MAP2K1), Erk2 (MAPK1), MNK1 (MKNK1), Rsk2 (RPS6KA3), to the transcription factor CREB (ATF4). In addition, Rac1 and PAK1, that activate c-Raf by increasing phosphorylation of Ser338 (King et al., 1998) were also significantly affected (Figure 1A, Figure 5—figure supplement 1C and Supplementary file 1). Consistent with the observed translational suppression of MNK1, which phosphorylates eIF4E (Waskiewicz et al., 1997), mRNAs that critically depend on phospho-eIF4E levels (Karampelias et al., 2022) were translationally offset in eIF3d^KD and eIF3e^KD cells, such that translation efficiency was decreased to maintain a constant level of polysome association despite increased total mRNA abundance (Figure 5—figure supplement 2A, B).

Consistently, Gene Ontology (GO) analysis revealed strong association of downregulated DTEGs with biological processes like ‘protein phosphorylation’ and ‘regulation of transcription’ and the molecular function term ‘protein serine/threonine kinase activity’ (Figure 5—figure supplement 3A, B). Key integrin-mediated signaling proteins that were implicated in cancer cell plasticity, adhesion, migration, and survival (reviewed in Cooper and Giancotti, 2019) also had significantly decreased TE, specifically, proto-oncogenes Crk (CRK) and Crk-L (CRKL), PRKCA – a catalytic subunit of PI3-kinase, Rac1 GTPase (RAC1), JNK2 (MAPK9), as well as Abl1 kinase (ABL1) and BCR activator of Rho-GEF (BCR). The TGF-beta signaling, important for a wide range of cellular processes as well as for tumor cell invasion, metastasis, and resistance to chemotherapy (reviewed in Colak and Ten Dijke, 2017), was affected at the level of the intracellular transducer Smad4 (SMAD4) and receptor activated SMADs, namely Smad3 (SMAD3) and Smad5 (SMAD5) and TGF beta (TGFB3) itself. The NFkB pathway, a regulator of immune and inflammatory responses (Liu et al., 2017), also implicated in cancer cell proliferation, survival, angiogenesis and plasticity (reviewed in Xia et al., 2014), was downregulated at the level of NFkB (NFKB1) itself and IKK kinase (CHUK). Notably, a direct link between activation of NFkB and eIF3 has very recently been described (She et al., 2023). Finally, the canonical Wnt signaling pathway, which shapes the stem cells-like properties of cancer cells and boosts proliferation (reviewed in Zhang and Wang, 2020), was downregulated at the level of a protein Dishevelled (DVL3) and GSK3 beta kinase (GSK3B). Physiological relevance of all these hits remains to be seen; for the complete list of all DTEGs, please see Supplementary file 1.

eIF3d^KD and eIF3e^KD reduce protein levels of key components of the activated MAPK/ERK pathway

Interestingly, more than half (26 of the 42) downregulated DTEGs belonging to the ‘MAPK signaling pathway’ can be classified as translationally buffered; that is, their mRNA levels increase while their TE decreases (Kusnadi et al., 2022). However, this is not a typical case of translational buffering because mRNA upregulation is not only nullified by translational repression, but completely overruled, so that the total number of FPs is also reduced. To verify that the reduction in TE and FP counts is indeed manifested by a reduction in cellular protein levels, we selected a set of genes with different combinations of changes in TE or FP or mRNA levels, including translationally exclusive and translationally buffered genes, as well as genes with unchanged TE, and subjected them to western blotting (Figure 5B, D).

We found, that AKT1, GRB2, MAP2K1 (MEK1), PAK1, RPS6KA1 (RSK1), RPS6KA3 (RSK2), RAF1 and SOS1 had visibly decreased protein levels in eIF3d^KD and most of them also in eIF3e^KD. An interesting exception is the transcription factor c-Jun, which showed increased mRNA and FP levels, unchanged TE, but robust increases in protein levels in all three knock-downs (Figure 5C, D). The mRNA levels of all selected genes were verified by qPCR, showing a good correlation with our sequencing data; whereas PAK1 showed significant decrease, JUN mRNA was significantly increased in both eIF3d^KD and eIF3e^KD (Figure 5—figure supplement 3C). To assess the functional consequences of the downregulation of MAPK/ERK pathway components, we examined the phosphorylation status of effector kinases ERK1/2, which directly phosphorylate several transcription factors including c-Jun (Leppä et al., 1998). As with other MAPK proteins, ERK1/2 expression was reduced in eIF3d^KD and eIF3e^KD, but its phosphorylation at activation sites Thr202/Tyr204 was unexpectedly increased in all three knock-downs (Figure 5E). Next, we examined the phosphorylation status of the upregulated c-Jun protein. As phosphorylated c-Jun translocates to nucleus to activate transcription of specific genes, we performed western blot of both cytoplasmatic and nuclear fractions. In accord with increased ERK1/2 phosphorylation, c-Jun was found to be activated by phosphorylation at Ser63 and largely accumulated in the nucleus (Figure 5F). Thus, despite reduced MAPK/ERK pathway protein levels, MAPK/ERK signaling appears to be activated when eIF3 function in translation is compromised.

Intact eIF3 contributes to translational regulation of ATF4 and MDM2

Knocking down eIF3h generates an eIF3 subcomplex lacking eIF3h and two nonessential subunits eIF3k and eIF3l (Figure 1B), having only negligible impact on overall translation. Accordingly, we identified 19 transcripts whose TE was decreased and only 2 transcripts whose TE was increased compared to the NT control (Figure 2D). Given the low number of DTEGs, we observed no significant enrichment of any KEGG pathway within the upregulated DTEGs and only a few significant enrichments for the eIF3h^KD ‘all downregulated’ DTEGs in pathways like ‘Prostate cancer’, ‘Viral carcinogenesis’ or ‘PI3-Akt signaling pathway’, all of which included only two common hits – MDM2 and ATF4 (Figure 6A, Supplementary file 2A). Additionally, we also noticed enrichment of GO terms associated with ubiquitination (Figure 6B, Supplementary file 2B), which was mainly due to the MDM2, WAC, RNF10, RMND5A, USP38, and ZRANB1 genes, all of which showed decreased TE. MDM2, WAC, RNF10 and RMND5A are all E3 ubiquitin ligases while USP38 and ZRANB1 are de-ubiquitinases. Note that all of the above downregulated DTEGs are common for all three knock-downs.

Perhaps the most interesting of these genes is the proto-oncogene MDM2, an E3 ubiquitin ligase that acts as a negative regulator of the p53 protein (Marine and Lozano, 2010), whose overexpression has been reported in many different cancer types (Rayburn et al., 2005). We did confirm that the MDM2 protein levels were reduced in all three knock-downs (Figure 6C).

An equally interesting DTEG is ATF4, which is among the top 10 most significant DTEGs in all three knock-downs (Figure 2B, D). In eIF3h^KD, it can be classified as a translationally buffered gene because it has strongly upregulated mRNA levels, but this increase is not accompanied by a concomitant increase in the FP count, thereby resulting in decreased TE. In eIF3d^KD and eIF3e^KD, it is even significantly translationally repressed (FPs are decreased). Using qPCR, we confirmed ATF4 mRNA upregulation in all three knock-downs (Figure 6—figure supplement 1A). Because ATF4 protein levels are on the verge of the western blot detection limit under non-stress conditions, we examined whether all three knock-downs tested had any effect on ATF4 induction upon 3 hour-long ER stress provoked by 1 µM Thapsigargin (Harding et al., 2000). Indeed, stress-mediated ATF4 induction was almost abolished in eIF3d^KD and eIF3e^KD, and strongly decreased in eIF3h^KD (Figure 6D). This differential effect correlates well with our sequencing data, where the decrease in TE was also stronger in eIF3d^KD and eIF3e^KD. The ATF4 protein levels in DMSO control are shown as a no stress reference. Overall, these findings further confirm that an intact eIF3 complex is required for proper translational control of ATF4, as we and others have previously suggested (Guan et al., 2017; Hronová et al., 2017; Mukhopadhyay et al., 2023; Shu et al., 2022; Wagner et al., 2020), and as discussed below.

As both ATF4 and MDM2 are known to contain uORFs in their 5’ UTR that regulate their translation under stress vs. non-stress conditions (Akulich et al., 2019; Dey et al., 2010; Lu et al., 2004; Smirnova et al., 2024; Vattem and Wek, 2004), we examined the footprint coverage in the 5’ UTR of these genes in more detail. In the case of ATF4, a markedly increased accumulation of footprints was observed at the most 5’ Start-stop (St-st) element, specifically in eIF3d^KD and eIF3e^KD, while coverage of uORF1 decreased, and so did also the basal level of uORF2 (modestly) and ATF4 (Figure 6E). This agrees well with our observations that ribosomes are stalled at this St-st element (Rendleman et al., 2023; Wagner et al., 2020) and further suggests that both eIF3e and eIF3d are required to clear these St-st-trapped ribosomes. Regarding MDM2, footprint coverage was increased specifically for uORF1 in all three knock-downs (Figure 6F), consistent with its several-fold stronger inhibitory role on MDM2 expression compared to uORF2 (Jin et al., 2003), which seems to be magnified in the absence of fully functional eIF3. Collectively, these results are in line with the well-established role of eIF3 in translation reinitiation (reviewed in Valášek et al., 2017). In addition, we found similarly accumulated footprints also at uORF1 and uORF2 of the RAF1 kinase, one of the downregulated components of the MAPK pathway (Figure 6—figure supplement 1B).

Last but not least, eIF3d was shown to mediate mechanism of alternative cap-dependent translation initiation in conjunction with DAP5 (p97/NAT1) (Alard et al., 2023; de la Parra et al., 2018a; Volta et al., 2021) and, importantly, DAP5-dependent mRNAs also frequently contain uORFs (David et al., 2022). Consistent with this, the effects of depletion of DAP5 on the translatome were comparable to those observed in eIF3d-depleted cells (Figure 6—figure supplement 2A–C); that is, no changes in mRNA levels and the FP coverage moving in the same direction, either up or down. Since similar effects were also observed with eIF3e^KD and eIF3h^KD, it seems that the DAP5-dependent mechanism may rely upon not only eIF3d but also on the entire eIF3 complex.

Presence of uORFs, UTR length and GC content correlate with the change in translation efficiency

eIF3 is critically required for mRNA recruitment to the 40 S ribosomal subunit (reviewed in Valášek et al., 2017) and several eIF3 subunits have been proposed to regulate translation of specific subsets of mRNAs (Lee et al., 2015; Lee et al., 2016). We therefore asked whether the changes in TE that were observed in the eIF3d, eIF3e or eIF3h knock-downs correlate with specific mRNA properties. Indeed, in all three knock-downs, the change in TE was negatively correlated with the length of 5’ and 3’ UTR, and with the presence of uORFs in 5’ UTR (Spearman correlation, p<1 × 10^–45) (Figure 7A). Moreover, we found a significant positive Spearman correlation between TE changes and the GC content in the coding region (CDS) and 3’ UTRs.

To explore these correlations in more detail, we focused only on those mRNAs that are exclusively regulated at the translational level in each knock-down and therefore are more likely to contain distinct features that facilitate their translational regulation. We divided them into those with significantly changed TE (decreased and increased) and those with unchanged TE and searched for significant differences in the median values of the tested traits (Figure 7B, C). In accordance with the Spearman correlation, in eIF3d^KD and eIF3e^KD, mRNAs of downregulated DTEGs had longer 5’ UTRs, while upregulated DTEGs possessed shorter 5’ UTRs, when compared to mRNAs with unchanged TE. The same trend is seen for the 3’ UTR length in eIF3e^KD, but with lower significance. In case of eIF3d^KD, only the downregulated DTEGs showed a significant change in 3’ UTR length, but not DTEGs that were upregulated.

In terms of GC content, while we observed little significant difference in GC content in the 5' UTR of eIF3d^KD and eIF3e^KD, there was a highly significant difference in the CDS and 3' UTR of these knock-downs in both downregulated and upregulated DTEGs, indicating that mRNAs with increased TE tend to have more GC-rich 3' UTRs and, surprisingly, CDS and vice versa. For eIF3h^KD, we only observed differences of low significance due to the low number of DTEGs that were only translationally regulated (only 8 downregulated and 1 upregulated) (data not shown).

In summary, mRNAs with decreased TE in cells wherein eIF3d and eIF3e were depleted exhibited on average longer 5’ and 3’ UTRs and lower GC content in the CDS and 3’ UTRs. In turn, mRNAs with increased TE have on average shorter 5’ UTRs and a higher GC content in the CDS and 3’ UTRs.

We next asked whether 5’ UTRs of upregulated or downregulated DTEGs contain any specific sequence motifs or short upstream open reading frames (uORFs). We did not identify any specific sequence motif, but expectedly, the 5’ UTRs of ‘downregulated translation only DTEGs have a strong tendency to contain uORFs starting with AUG (Figure 7D, E). These findings suggest that the intact eIF3 complex is primarily required for translational control of subset of mRNAs containing short uORFs in their 5’ UTR even under normal cellular conditions, like for example MDM2, ATF4 or RAF1, which is in line with its role in translation reinitiation (Valášek et al., 2017), and further discussed below.

Cell lines, culture conditions, and transfection

HeLa cells (ATCC Cat# CCL-2, RRID:CVCL_0030) were grown at 37 °C and 5% CO₂ in Ø 15 cm dishes in DMEM (Sigma, cat #D6429) supplemented with 10% FBS (Gibco, cat # 10270–106). Twenty-four hours after seeding, cells were transfected with the ON-TARGETplus siRNA cocktail system from Dharmacon at a final concentration of 5 nM. Catalog numbers for all siRNAs used in this study are listed in Supplementary file 3A. INTERFERin (Polyplus, cat # 101000016) was used as a transfection reagent, 100 µl per dish. Cells were harvested 72 hr after transfection at approximately 80% confluency and cytoplasmatic lysates were prepared as described previously (Herrmannová et al., 2020). For monitoring levels of phosphorylated c-Jun protein, cytoplasmatic and nuclear lysates were prepared using NE-PER Nuclear and Cytoplasmic Extraction Reagents from Thermo Scientific.

Ribosome profiling (Ribo-Seq) and RNA-seq library preparation

For footprint (FP) libraries cycloheximide was added to a final concentration of 100 µg/ml 1 min prior to harvesting. Cells were washed with ice cold 1 x PBS and lysed in buffer A (10 mM HEPES [pH 7.5], 62.5 mM KCl, 2.5 mM MgCl₂, 1 mM DTT, 1 mM PMSF, 1 µg/ml Aprotinin, 1 µg/ml Leupeptin, 1 µg/ml Pepstatin, Complete Mini EDTA-free [Roche, cat # 11836170001] – 1 tablet/5 ml, 1% Triton X-100, 100 µg/ml Cycloheximide) on the dish. The resulting lysate was collected and cleared by centrifugation. Aliquots of twelve absorbance units (AU) OD₂₆₀ of WCE were flash frozen in liquid nitrogen and used for control polysome profile and footprint isolation. An aliquot of 1 mg total protein was taken for control western blot, and another 200 µl aliquot was mixed with 1 ml of RNA Blue (TopBio, cat # R013) and taken for RNA isolation and control qPCR. When the lysate passed all three control experiments (polysome profile, western blot, qPCR), one aliquot of 12 AU OD₂₆₀ was thawed on ice, digested with 5 µl RNAseI (Ambion, cat # AM2295) for 30 min at 24 °C with mild shaking (300 rpm) and inactivated with SuperaseIN (Ambion, cat #AM2696). 80 S ribosomes were separated by high-velocity sedimentation through a 5% to 45% sucrose gradient at 39,000 rpm for 2.5 hr using the SW41Ti rotor. The gradients were scanned at 254 nm to visualize the ribosomal species and fractions containing 80 S ribosomes were collected and mixed with RNA Blue. RNA was isolated according to vendor’s instructions, FP RNA was size selected on 15% TBE-Urea PAGE (Biorad, cat # 3540092) in range 20–40 nt and footprint library was prepared according to Ingolia et al., 2012. FP libraries were sequenced on NovaSeq6000 (Illumina) with read length 50 nt.

For mRNA libraries the growth media was aspirated and cells were harvested into 5 ml of RNA Blue per Ø 15 cm dish. RNA was isolated according to vendor’s instructions. mRNA was isolated from the total RNA using Poly(A)purist MAG kit (Ambion, cat # AM1922), fragmented using RNA Fragmentation Reagents (Ambion, cat # AM8740) and size selected on 15% TBE-Urea PAGE in range 50–100 nt. Size selected mRNA was then dephosphorylated and library was prepared using SMARTer smRNA-Seq Kit (TAKARA, cat # 635029) according to vendor’s instructions. mRNA libraries were sequenced on NovaSeq6000 (Illumina) with read length 150 nt.

RNA isolation, reverse transcription, and qPCR

Total RNA was isolated using RNA Blue reagent (TopBio, cat # R013) 72 hr post-transfection according to the manufacturer’s instructions. After DNase I digestion (NEB, cat # M0303L), cDNA was synthesized from 1 µg RNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, # 4368813). qPCR was performed in triplicates using 5×HOT FIREPol EvaGreen qPCR Mix Plus (Solis BioDyne # 08-25-00020). The obtained data were normalized to ALAS1 mRNA levels and non-transfected control. All qPCR primers are listed in Supplementary file 3B.

Western blotting

All samples were resolved using SDS-PAGE followed by western blotting. All primary antibodies used in this study are listed in Supplementary file 3C. The signal was developed using SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, cat # 34096) and detected in a G-Box imager (Syngene) using a series of varying exposure times. Signals were processed with Quantity One (Bio-Rad). Only signals from the same strips and with the same exposure times were compared.

Polysome profiles

One aliquot of 12 AU OD₂₆₀ was thawed on ice and separated by high-velocity sedimentation through a 5% to 45% sucrose gradient at 39,000 rpm for 2.5 hr using the SW41Ti rotor. The gradients were scanned at 254 nm to visualize the ribosomal species.

60S/40S ratio calculation

Cell lysates were prepared as above by on dish lysis in buffer A. To 6 AU OD₂₆₀ EDTA was added to final concentration of 50 mM and the lysate was separated by high-velocity sedimentation through a 5% to 50% sucrose gradient at 39,000 rpm for 3 h using the SW41Ti rotor. The gradients were scanned at 260 nm to visualize the ribosomal species. Area under the curve was quantified to calculate the 60 S:40 S ratio to check for ribosome biogenesis defect.

Ribosome content profiles

Cells were counted by Corning Automated Cell Counter and 10 million cells were pelleted in microcentrifuge tube and lysed in buffer A (as above). EDTA was added to the lysate to final concentration of 50 mM and the lysate was separated by high-velocity sedimentation through a 5% to 50% sucrose gradient at 39,000 rpm for 3.5 hr using the SW41Ti rotor. The gradients were scanned at 260 nm to visualize the ribosomal species. Area under the curve was quantified to calculate the total amount of ribosomes between samples.

Analysis of sequencing data

We processed the ribosome footprint libraries as follows. Adapter sequence ‘CTGTAGGCACCATCAAT’ and all sequences 3’ of the adapter were trimmed from the 3’ end of each read by Cutadapt v. 3.5 (Martin, 2011). Flanking Ns were trimmed from both 5’ and 3’ ends of each read. Reads without adapter sequence and reads shorter than 15 nt after trimming were discarded. Next, rRNA and tRNA reads were discarded after alignment with Bowtie2 v. 2.4.4 (Langmead and Salzberg, 2012). For rRNA, we used RNA5S1, RNA5-8SN5, RNA18SN5 and RNA28SN5 from NCBI (Sayers et al., 2022), and ENSG00000211459 and ENSG00000210082 from Ensembl (Howe et al., 2021). For tRNA we used all human tRNA sequences from tRNAdb (Jühling et al., 2009) and all high confidence hg38 tRNA sequences from GtRNAdb (Chan and Lowe, 2016). Remaining reads were aligned to genome and transcriptome from Ensembl (GRCh38.p13, annotation release 104) by the STAR aligner v. 2.7.9 a_2021-06-25 (Dobin et al., 2013) using local alignment with 10% mismatches, and with indels and soft clipping allowed for transcriptomic alignment. Gene counts were counted from genomic alignment using Samtools index v. 1.13 (Danecek et al., 2021) and Htseq-count v. 0.13.5 (Anders et al., 2015) within CDS regions. Transcriptomic alignments were further processed using MANE Select set from MANE project v. 0.95 (Morales et al., 2022) and custom programs ‘mane2ensembl_gtf’, ‘filter_reverse_reads’, ‘filter_ambiguous_genes’ and ‘select_transcripts’. See Supplementary file 4 for detailed overview of raw read processing.

The RNA-seq libraries were processed the same as above, except that raw read processing by Cutadapt had to be adapted to different library construction method (SMARTer smRNA-Seq Kit, TAKARA). In particular, first 3 nucleotides on the 5’ end were removed, and poly(A) sequence (10xA) was trimmed from the 3’end.

Quality checks were done by FastQC v. 0.11.9 (Andrews, 2010) and Ribo-seQC v. 1.1 (Calviello et al., 2019). Triplet periodicity was checked on transcriptomic alignments using custom modified riboWaltz (Lauria et al., 2018) and a custom script ‘Triplet_periodicity.R’. Gene counts were used to check correlation between replicates using a custom script ‘Correlation-samples.R’. Read length counts were used to check read length distribution using a custom script ‘Lengths.R’.

PCA plot, sample distances clustering and the differential expression analysis were performed on gene counts by DESeq2 (Love et al., 2014) with biomaRt library (Durinck et al., 2009) used to download gene annotations. Differential translation efficiency (TE) values between NT and the knock-downs were calculated as described in Chothani et al., 2019. The p_adj threshold for differentially translated transcripts was set to 0.05. For details see custom script ‘Differential analysis.R’. Results of differential expression analysis were further used to generate heatmap and scatterplots of differentially expressed genes and genes with defined p-adjusted value were used to analyze dependency between mRNA characteristics and changes in TE using custom scripts ‘Significant_TE_heatmap.R’, ‘Significant_TE_intersection.R’ and ‘Correlation-features.R’. To get mRNA characteristics we used MANE Select set to select a single representative transcript for each protein-coding gene. uORF database uORFdb (Manske et al., 2023) was used to get number of ATG uORFs per mRNA.

Metagene analysis was done as described in Lin et al., 2020, but 5’ ends of reads were used instead of P-sites. Samtools and custom programs ‘transcripts_startstop_positions’, ‘read_counts’ ‘nd ‘region_readcounts’ were used to get read counts relative to the start codon. These were normalized to a number of reads in a region starting 75 bases after the start codon and ending 15 bases before the stop codon and averaged over the samples. Only transcripts that met the following parameters were considered: CDS longer than 450 nt and RPKM higher than 10. Weighted counts of biological replicates were further used to compute average read counts for each sample. Moving average method was used with a 21 bases window size.

For footprint coverage plots of individual mRNAs, MANE transcripts of examined genes (ENST00000674920-ATF4; ENST00000258149-MDM2; ENST00000251849-RAF1) were selected. Reads were counted for each position using Samtools and a custom script read_counts and normalized to number of read counts over all positions of transcript. Weighted counts of biological replicates were further used to compute average read counts for each sample. Graphs were smoothened with a sliding window of 30nt. All used custom scripts are available on GitHub, copy archived at Jelinek, 2024.

KEGG/GO pathway enrichment analysis

We analyzed enrichment of the differentially translated transcripts in KEGG pathways (Kanehisa and Sato, 2020) and Gene Ontology (Ashburner et al., 2000; Carbon et al., 2021), using Enrichr (Kuleshov et al., 2016), ShinyGO v0.741 (Ge et al., 2020) and KEGG Mapper Search (Kanehisa et al., 2022). Venn diagrams were created using InteractiVenn (Heberle et al., 2015).

Analysis of gene signatures

To assess changes in the regulation of known gene signatures in the eIF3 KDs, empirical cumulative distribution functions (ECDFs) of log₂ fold changes in FP and total mRNA were plotted independently. Fold change ECDFs for genes belonging to a particular signature were compared to those for all other genes, and differences in distributions were calculated at the quantiles. Significant directional shifts between the signatures and the background were identified using the Wilcoxon rank-sum test.