Selective 40S Footprinting Reveals Cap-Tethered Ribosome Scanning in Human Cells.

Article Selective 40S Footprinting Reveals Cap-Tethered

Ribosome Scanning in Human Cells

Graphical Abstract

d Selective 40S footprinting visualizes regulation of mRNA translation initiation in vivo d Scanning ribosomes are cap-tethered in most human cells d Only one ribosome scans a 50 UTR at a time in most human cells d Ribosomes retain eIFs during early translation, allowing reinitiation after uORFs Bohlen et al., 2020, Molecular Cell 79, 1–14 August 20, 2020 ª 2020 Elsevier Inc. Authors Jonathan Bohlen, Kai Fenzl, G€unter Kramer, Bernd Bukau, Aurelio A. Teleman Correspondence In Brief Bohlen et al. develop 40S selective ribosome footprinting to detect scanning 40S ribosomes in vivo and the position on mRNAs when initiation factors join or disengage from the ribosome. They discover that ribosomes remain attached to initiation factors and the mRNA 50 cap throughout the scanning process and early elongation. ll ll

Selective 40S Footprinting Reveals

Cap-Tethered Ribosome Scanning in Human Cells

Jonathan Bohlen,1,2,3,4,5 Kai Fenzl,6 G€unter Kramer,5,6 Bernd Bukau,5,6 and Aurelio A. Teleman1,2,3,4,5,7,*
1German Cancer Research Center (DKFZ), 69120 Heidelberg, Germany 2CellNetworks - Cluster of Excellence, 69120 Heidelberg University, Germany 3Heidelberg University, 69120 Heidelberg, Germany 4Heidelberg Biosciences International Graduate School (HBIGS), 69120 Heidelberg, Germany 5National Center for Tumor Diseases (NCT) partner site, 69120 Heidelberg, Germany 6Center for Molecular Biology of Heidelberg University (ZMBH) and German Cancer Research Center (DKFZ), DKFZ-ZMBH Alliance, Im Neuenheimer Feld 282, 69120 Heidelberg, Germany 7Lead Contact *Correspondence:

Translation regulation occurs largely during the initiation phase. Here, we develop selective 40S footprinting to visualize initiating 40S ribosomes on endogenous mRNAs in vivo. This reveals the positions on mRNAs where initiation factors join the ribosome to act and where they leave. We discover that in most human cells, most scanning ribosomes remain attached to the 50 cap. Consequently, only one ribosome scans a 50 UTR at a time, and 50 UTR length affects translation efficiency. We discover that eukaryotic initiation factor 3B (eIF3B,) eIF4G1, and eIF4E remain bound to 80S ribosomes as they begin translating, with a decay half-length of 12 codons. Hence, ribosomes retain these initiation factors while translating short upstream open reading frames (uORFs), providing an explanation for how ribosomes can reinitiate translation after uORFs in humans. This method will be of use for studying translation initiation mechanisms in vivo.

Messenger RNA (mRNA) translation efficiency varies substantially between different mRNAs (Ingolia et al., 2009, 2011; Schwanh€ausser et al., 2011), highlighting the regulatory potential of this process. Much of the regulation happens during translation initiation (Hinnebusch, 2011; Jackson et al., 2010; Pelletier and Sonenberg, 2019; Shirokikh and Preiss, 2018). For instance, activation of 4E-BP inhibits translation by blocking ribosome recruitment to the 50 cap (Sonenberg andGingras, 1998). Cellular stresses activate the integrated stress response, thereby inactivating the eukaryotic initiation factor 2 (eIF2)-containing ternary complex (TC) and recruitment of initiator tRNA to ribosomes (Holcik and Sonenberg, 2005). Various 50 UTR sequence elements that affect scanning and initiation have been described, such as upstream open reading frames (uORFs), ribosome shunts, secondary structures such as hairpins and G-quadruplexes, or internal ribosome entry sites (IRESs) (Geballe and Morris, 1994; Leppek et al., 2018; Millevoi et al., 2012; Mitchell and Parker, 2015). One regulatory element in 50 UTRs are uORFs. uORFs can be broadly classified into those with weak Kozak sequences and those with strong ones. The ones with weak Kozak sequences do not strongly impact translation, because ribosomes simply scan past the uORF (leaky scanning). uORFs with strong Kozak sequences, however, pose a challenge, as ribosomes recognize them as bona fide translation start sites and therefore translate the uORF. Upon terminating uORF translation, ribosomes need to reinitiate translation on the main ORF in order to express the encoded protein. This ‘‘translation reinitiation’’ is not well understood but probably involves stabilization of the ribosome on the mRNA, re-recruitment of an initiator tRNA, and resumed scanning of a ribosomal complex that is competent to initiate another round of translation. How this happens mechanistically is not known, yet it happens frequently. Nearly half of all human mRNAs contain at least one uORF (McGillivray et al., 2018), and many of these are translated in vivo (Ingolia et al., 2011). Hence, there is a need to better understand translation reinitiation. Worth noting is that translation reinitiation appears to be different in humans and yeast (Kozak, 2001). In humans, most uORFs are permissive for reinitiation and thus allow translation of the main ORF downstream to occur, whereas in yeast, most uORFs are not (Jackson et al., 2012; Johansen et al., 1984; Kozak, 1984; Liu et al., 1984; Yun et al., 1996). The best-studied example of translation reinitiation in yeast is the GCN4 mRNA, where specific cis-acting elements flanking the uORFs are required to permit reinitiation (Mohammad et al., 2017; Szamecz et al., 2008). After recruitment to the mRNA cap, the small ribosomal subunit starts to scan toward the start codon. Whether the cap Molecular Cell 79, 1–14, August 20, 2020 ª 2020 Elsevier Inc. 1 remains attached to the scanning ribosome is an open question with functional consequences (Jackson et al., 2010; Shirokikh and Preiss, 2018). Cap-severed scanning allows multiple ribosomes to scan a 50 UTR at one time, but cap-tethered scanning does not; hence, increased 50 UTR length reduces mRNA translation efficiency. In yeast, 50 UTR length does not influence translation efficiency (Berthelot et al., 2004), and queuing 43S ribosomes have been observed (Archer et al., 2016), suggesting that scanning is cap-severed. Whether this is also the case in mammalian cells is not known. Some reports suggest 50 UTR length indeed is limiting for translation efficiency in mammals (Chappell et al., 2006; Paek et al., 2015). We develop here selective 40S footprinting to visualize the successive steps of translation initiation in vivo on endogenous mRNAs and pinpoint when translation initiation factors (eIFs) join the ribosome to act, and then later disengage. (Note that for simplicity, we use the term ‘‘40S’’ to denote all variants of the 40S, such as 43S and 48S ribosomes.) We find that in most human cells, 50 UTR scanning is mainly cap tethered, with scanning 40S ribosomes attached to eIF3B, eIF4G1, eIF4E, and the mRNA cap up to the start codon of the main ORF. This implies that only one ribosome can scan a 50 UTR at one time, making 50 UTR length a limiting factor for translation efficiency. We found one mouse cell line (NIH 3T3) where scanning is not cap tethered, suggesting that the mode of ribosome scanning can be modulated. We find that eIFs persist on early 80S translating ribosomes with a half-length of 12 codons. This likely explains how translation reinitiation can occur after a short uORF, but not after main ORFs, which are significantly longer. This approach complements 80S footprinting (Ingolia et al., 2009, 2011) and in vitro systems (Hinnebusch, 2011; Jackson et al., 2010, 2012) for dissecting mechanisms of translation initiation in vivo in the future.

40S Footprinting Visualizes Scanning Ribosomes In Vivo Weaimed to visualize scanning 40S ribosomes onmRNAs in vivo and identify at which point during this process eIFs join the ribosome to act and then leave. We first adapted for human cells translation complex profile sequencing (TCP-seq), a version of ribosome footprinting that includes paraformaldehyde (PFA) crosslinking and enables the detection of 40S ribosome footprints (Archer et al., 2016; Shirokikh et al., 2017). For crosslinking in HeLa cells, we tested different concentrations and combinations of PFA with dithiobis(succinimidyl propionate) (DSP), which stabilizes protein-protein interactions. Although we found that 40S footprints can be observed without any crosslinking (see below), insufficient crosslinking fails to stabilize eIF-40S interactions important for selective 40S footprinting. Since excessive crosslinking causes ribosome aggregation, we found an optimal concentration of 0.025% PFA + 0.5 mM DSP that does not produce ribosome aggregates (as observed on a polysome gradient; Figure S1A) yet stabilizes eIFs on ribosomes (Figure S1B–S1F) without cross-linking nonspecific interactions (tubulin; Figure S1B). After in vivo crosslinking, cell lysis, and RNase digestion (Figure S1G), we separated scanning 40S ribosomes from elongating 80S ribosomes on a sucrose gradient 2 Molecular Cell 79, 1–14, August 20, 2020 and sequenced their footprints (Figure 1A). As expected, 40S footprints localize predominantly to 50 UTRs (Figure 1B). This allows for an mRNA-resolved transcriptome-wide view of scanning ribosomes in human cells. For instance, on the eIF5A mRNA, scanning 40S ribosomes are detected in the 50UTR, which convert to 80S ribosomes at the ORF start (Figure 1C). After translation termination on main ORFs, 40S ribosomes could either fall off or scan off the 30 end of the mRNA (Bertram et al., 2001; Dever and Green, 2012). Since we see almost no 40S ribosomes in 30 UTRs (Figures 1B and 1C), this means they are falling off the mRNA at the stop codon. As observed in yeast (Archer et al., 2016), 40S footprints have a size distribution distinct from 80S footprints (Figures 1D and 1E). A two-dimensional (2D) ‘‘start-codon metagene plot’’ resolving footprint position (x axis) versus footprint length (y axis) of 40S footprints relative to all main-ORF start codons shows that scanning ribosomes (positions 100 to 60 relative to the start codon) have heterogeneous footprint sizes ranging from 20 to 80 nt (Figure 1D). As expected, 40S footprints show no triplet periodicity (Figure 1D), whereas 80S footprints do (Figure 1E). When scanning ribosomes reach the start codon, they pause, as observed by a strong enrichment of 40S footprints overlapping the start codon, indicating initiation is significantly slower than scanning in vivo in human cells. Two main populations of footprints can be observed on start codons, with footprint lengths of 20 and 30 nt. Each of these has a ‘tail’ that extends diagonally up and to the left, representing a series of footprints that have the same 30 end but varying 50 ends (bottom of Figure 1D). This suggests that the 50 ends of initiating 40S ribosomes protect mRNA less robustly than the 30 ends. As described in yeast (Archer et al., 2016), we also observe 40S footprints on the stop codon of ORFs (Figures S1H and S1I), with footprint lengths of 20 and 30 nt, which likely represent post-termination intermediates. Biological replicates confirmed that the results are reproducible (Figures S2A–S2D). We prepared most 40S libraries with the Takara SMARTer smRNA-Seq kit. When compared to the ligation-based library preparation protocol by Ingolia et al. (Galmozzi et al., 2019; McGlincy and Ingolia, 2017), only small differences were observed (Figures S2E, S2F, S2F0, S2H, and S2I). The Ingolia et al. library preparation allows introduction of unique molecular identifiers (UMIs) in the sequenced footprints to remove PCR duplicates. When we removed duplicates, the distribution of 40S footprints was largely unchanged (Figure S2F0 and inset). Without crosslinking, 40S footprints are obtained with a similar distribution as with crosslinking, both across and within transcripts (Figures S2H and S2J), but their length distribution collapses, yielding footprints of 30 nt (Figure S2E and S2G), similar to those of 80S ribosomes. These data are consistent with results presented below indicating that the large footprints of 40S ribosomes are due to areas upstream of the ribosome protected by eIFs. In sum, 40S footprinting enables the investigation of ribosome scanning and initiation in human cells. Selective 40S Footprinting Reveals Steps of Translation Initiation In Vivo We next introduced an immunoprecipitation step to isolate 40S complexes that contain proteins of interest, similar to 80S (A) Schematic diagram illustrating selective and total 40S and 80S ribosome footprinting in human cells. (B) 40S ribosome footprints are located mainly in 50 UTRs. Metagene plot of all reads mapped to all human protein coding transcripts with 50 UTR length >33 nt (n = 35,921). (C) Footprint distribution on eIF5A mRNA (ENST00000336458). Curves smoothened with 10-nt sliding window. Black box, ORF. (D and E) 40S ribosome footprinting reveals stalling and processing of 40S ribosomes on translation start sites. Metagene plots of length resolved 40S (D) or 80S (E) footprints aligned to main ORF start codons on >41,000 human transcripts. Color scale is linear. Bottom: Schematic representation of footprint species of different lengths on the start codon. See also Figure S1. selective ribosome profiling (Oh et al., 2011; Schibich et al., 2016). We immunoprecipitated 40S ribosomes that contain endogenous eIF2S1, eIF3B, or eIF4G1 (Figures S3A–S3C). eIF2S1/eIF2a is a component of the TC that is responsible for recruiting initiator tRNA to ribosomes. eIF3B is a core subunit of the eIF3 complex that attaches scanning ribosomes to mRNA and serves as a docking platform for many eIFs. eIF4G1 is the core scaffold of the eIF4F complex that links eIF4E and hence the 50cap to the ribosome (Jackson et al., 2010). This successfully identified different subpopulations of 40S ribosomes; a metagene plot of main ORF start codons showed that ribosomes overlapping the start codon are partially depleted of eIF2 (Figure S3D, which is normalized to library size, and Figure 2A, which normalizes the selective footprints down to the total footprints in the scanning region). This is expected, because the eIF2a-containing TC is evicted prior to 60S subunit joining (Kapp and Lorsch, 2004; Pisarev et al., 2006; Unbehaun et al., 2004); hence, 40S ribosomes spend part of their time on the start codon with eIF2 and part of their time without eIF2. Selective footprinting allows annotation of the 40S populations of the 2D metagene plot (Figure 2B). Scanning ribosomes (population 1) contain eIF2a (Figures 2C and 2C0), whereas 40S ribosomes overlapping the start codon (2 and 3) are partially depleted of eIF2a. Scanning ribosomes containing eIF3B (Figure 2D) have larger footprints than the average scanning 40S ribosome (the upper region is green and the lower region Molecular Cell 79, 1–14, August 20, 2020 3 (A) eIF2S1 dissociates from the ribosome during start codon recognition. Metagene plot of total 40S and eIF3B-, eIF4G1-, or eIF2S1-selective 40S ribosome footprints aligned to the start codon of all human protein coding transcripts (n = 41,244). Reads are mapped to their 50 end. Curves are scaled to total read counts in the region 98 to 69, representing scanning ribosomes and show the average of two to five biological replicates (dots). Graphs normalized only to library size are in Figure S3D. (B) Distinct populations are visible: (1) scanning 40 ribosomes, (2 and 3) 40S ribosomes on start codons, and (4) background signal from 80S ribosome disassembly during sample preparation. (C–E0) Length resolved start codon metagene plots of selective 40S footprints of eIF2S1 (C and C0 ), eIF3B (D andD0 ), or eIF4G1 (E and E0) containing 40S ribosomes. Intensity scales are adjusted to the normalization in (A). (C0), (D0), and (E0) show ratiometric images of selective footprints (green) versus total 40S footprints (red). See also Figure S2. is red in the ratiometric image in Figure 2D0), indicating that part of the RNA protection of scanning ribosomes is caused by the eIF3 complex (Erzberger et al., 2014). eIF3B is enriched in the ‘‘tail’’ of population 3 on the start codon (Figure 2D0), representing footprints with extended RNA protection on the 50 end but unchanged 30 ends. Hence, eIF3 protects loosely a region of 40 nt on the 50 end of the 40S ribosome. Since no structure is available for eIF4F on the ribosome, it is unclear on which side of the ribosome this complex sits (Shirokikh and Preiss, 2018). Like eIF3, eIF4G1 protectsmRNA on the 50 end of the ribosome (Figures 2E and 2E0). This suggest eIF4F contacts the mRNA most strongly where it exits the ribosome, raising the possibility that eIF4A may function by pulling mRNA through the ribosome rather than pushing it in from the front. The 20-nt footprints in population 3 may represent an initiation intermediate, as observed in yeast (Archer et al., 2016), or a conformation where the A-site is not occupied and hence accessible to RNase (Wu et al., 2019). In sum, selective 40S footprinting enables us to localize scanning 40S ribosomes on individual mRNAs and to identify when eIFs join or disengage from the scanning 40S. 4 Molecular Cell 79, 1–14, August 20, 2020 50 UTR Scanning Occurs Mainly in a Cap-Tethered Fashion The 50 cap of mRNAs binds 40S ribosomes via eIF4G1 and eIF4E (Figure 3A). As ribosomes scan toward the main ORF start codon, they could either release the 50cap by severing an eIF interaction or remain attached to it (Figure 3A). In the latter case, the 50 UTRwould have to loop.Which of these two options happens in vivo is not known, but this issue has conceptual and functional consequences for translation regulation (Jackson et al., 2010; Shirokikh and Preiss, 2018). We reasoned that eIF4E-selective 40S footprinting could provide insight into this open issue. If ribosomes release eIF4E during scanning, then they should become depleted of eIF4E as they scan in the 30 direction, causing a drop in eIF4E-selective footprint density (Figure 3A). If instead they remain bound to eIF4E, then eIF4E-selective footprint densities should remain uniform throughout the 50 UTR. To distinguish these two possibilities, we performed eIF4E-selective 40S footprinting (Figures S4A and S4B). Surprisingly, a metagene plot of 50 UTRs, where each 50 UTR length is scaled to 100%, shows that ribosomes retain eIF3B, eIF4G1, and eIF4E in constant proportion throughout the entire scanning process from the cap to the main ORF start codon (Figure 3B). On the start codon of main ORFs, 40S ribosomes retain both eIF4G1 and eIF4E (Figures 3C, S4C, and S4D). Hence, in most cases, ribosomes remain associated to eIF4E during scanning up to the start codon in human cells in vivo. A small percentage of ribosomes may let go of eIF4E at the start codon itself (14% drop in area under the curve from 45 to 5 for the eIF4E-selective profile compared to total 40S; Figure 3C). However, this does not fit with the quantitative (legend on next page) Molecular Cell 79, 1–14, August 20, 2020 5 retention of eIF4E throughout 50 UTRs (Figure 3B), even though 50% of 50 UTRs contain uORFs. If eIF4E were lost at start codons, this would happen on translated uORFs thereby causing a reduction in eIF4E binding in the 50 UTR. eIF4E might release the mRNA cap during scanning (Figure 3A), despite the fact that binding of eIF4E to eIF4G1 increases eIF4E affinity for the cap (von Der Haar et al., 2000). To test this, we treated cells with harringtonine and performed 40S footprinting. Harringtonine arrests initiating 80S ribosomes on start codons (Fresno et al., 1977; Ingolia et al., 2012). If ribosomes release the cap during scanning, then the cap should be free to recruit multiple new rounds of 40S ribosomes, which should accumulate on the 50 UTR, likely queuing in front of the stalled 80 ribosome. Hence, 40S footprints should increase with harringtonine treatment over time, and we would expect a 40S peak forming in front of the start codon. If instead the ribosome on the start codon remains cap tethered, then this will block recruitment of another ribosome to the mRNA. Hence over time, 50 UTRs should become progressively depleted of 40S ribosomes, and the number of footprints in the 50 UTR should diminish. Indeed, the latter was the case. Treatment with harringtonine, which led to an arrest of initiating 80S ribosomes and a runoff of elongating 80S ribosomes (Figures S4E– S4G), did not cause an accumulation of 40S ribosomes upstream of the main start codon (Figure 3D). In fact, it caused a reduction. Furthermore, there was a progressive depletion of 40S footprint reads on 50 UTRs. This could be seen both as a reduction in the total number of footprints per 40S ribosome (assayed by not depleting rRNA in the library preparation and calculating the percentage of mRNA footprints in the library; inset, Figure 3E) and as a drop in 40S footprints in 50 UTRs even after normalizing to footprints in the library (Figure 3E). Thus, we conclude that most 40S ribosomes do not let go of the mRNA cap during scanning, i.e., scanning is mainly cap tethered in human cells. A consequence of cap-tethered scanning is that 50 UTR length should affect mRNA translation efficiency. The time it takes to scan an entire 50 UTR puts a minimum limit on how quickly a new ribosome can be recruited to the mRNA and hence how quickly a new round of protein synthesis can be initiated. Thus, all things equal, the longer the 50 UTR of an mRNA, the lower its translation efficiency should be. To test this, we synthesized Figure 3. In HeLa Cells, Only One Cap-Tethered 40S Ribosome Scans (A) Schematic diagram of cap-severed versus cap-tethered scanning. (B) eIF2S1, eIF3B, eIF4G1, and eIF4E are retained on scanning 40S ribosomes th 250 nt (n = 13,439). Position along 50 UTR is scaled from 0% (50 cap) to 100% (s ribosomes (shown in C). Curves normalized to scaling factors of Figure 2A and R mRNA abundance along the 50 UTR due to alternative transcription start sites. S (C) eIF3B, eIF4G1, and eIF4E are retained on 40S ribosomes onmain ORF start co on all human protein coding transcripts (n = 41.244). Reads are mapped to their (D and E) Harringtonine block of initiating 80S ribosomes does not cause 40S queu in 50 UTRs (E). Start codon (D) and 50 UTR (E) metagene plots of total 40S ribosome (2 mg/mL, 37 C). Counts are normalized to the total number of reads mapped to co (B). Inset: Percentage of reads in the library (not rRNA depleted) that map to mR (F and G) 50 UTR length limits mRNA translational output in HeLa cells. Renilla lu containing 50 UTRs of varying lengths illustrated in (F). 50 UTRswere created by ins ORF. Renilla luciferase mRNA levels were normalized to actin B mRNA levels. Va See also Figures S3 and S4. 6 Molecular Cell 79, 1–14, August 20, 2020 a series of Renilla luciferase (RLuc) reporters with 50 UTRs of homogeneous quality but increasing length by multimerizing a 26-mer sequence that lacks secondary structure or uORFs (Figure 3F). We then transfected HeLa cells and quantified RLuc activity normalized to reporter mRNA levels. This revealed that indeed the longer the 50 UTR, the lower the RLuc expression from these mRNAs (Figure 3G). The data fit a simple model whereby the time it takes to initiate translation on an mRNA (t) is equal to the length of the 50 UTR (l) divided by a scanning velocity (v), plus a fixed time (c) for subunit joining on the ATG (t = l/ v + c), and hence, the number of initiations per unit time is 1/t (r > 0.88; Figure 3G). As expected, this length dependence was not observed when the 26-mer repeats were inserted in the 30 UTR of the luciferase reporter (Figure S5A), and it was still observed when the 26-mer was mutated to remove two weak nearcognate initiation sites (Figure S5B). Altogether, these data validate the finding that in HeLa cells ribosomes scan mainly in a cap-tethered manner, making 50 UTR length one important parameter in determining translation efficiency of an mRNA. To study if this is a general phenomenon, we tested the reporters with varying 50 UTR lengths in multiple mammalian cell lines. The majority of human and mouse cell lines displayed a cap-tethered phenotype of decreasing translation efficiency with increasing 50 UTR length (Figure 4A). Interestingly, we found one exception. In mouse NIH 3T3 cells, reporter translation was not dependent on 50 UTR length (purple line, Figure 4A), suggesting that scanning in these cells is not entirely cap tethered. Consistent with this, harringtonine treatment of NIH 3T3 cells (Figures S5C and S5D) did not reduce 40S ribosome density in 50 UTRs (Figure 4B and inset) and induced a peak of 40S ribosomes queuing in front of the start codon (Figure 4C). Together, these results indicate scanning is cap tethered in most mammalian cells but mainly cap severed in NIH 3T3 cells. We aimed to understand the underlying molecular difference in NIH 3T3 cells. KRPC cells are also mouse, yet they do not show this behavior (Figure 4A), indicating it is not a species-specific difference between mice and humans. One option is that 40S ribosomes release eIF4E while scanning in NIH 3T3 cells. This could happen due to 4E-BP, which binds eIF4E and blocks its interactionwith eIF4G. However, this is not the case. Just as in HeLa cells, eIF4E-selective 40S footprinting in NIH 3T3 cells revealed that eIF4E is retained on scanning 40S ribosomes an mRNA 50 UTR at One Time roughout the entire 50 UTR. Metagene plot of footprints on 50 UTRs longer than tart codon 60 nt). The last 60 nt are excluded because they contain initiating NA sequencing (RNA-seq) read counts at each position to account for varying hown is the average of two to five biological replicates (dots). dons. Start codonmetagene plot of total and selective 40S ribosome footprints 50 end. Shown is the average of two to five biological replicates (dots). ing in front of the start codon (D) but instead causes depletion of 40S ribosomes footprints as in (B) and (C) at different time points after harringtonine treatment ding mRNAs in each library and in (E) to RNA-seq counts at each position as in NAs. ciferase luminescence normalized to mRNA levels (G) for translation reporters ertingmultiple copies of an unstructured 26-mer in front of the Renilla luciferase lues represent the average ± SD of three biological replicates. (Figures S5E and S5F). Furthermore, the mTORC1 inhibitor torin, which activates 4E-BP, did not cause scanning to become cap severed in HeLa cells (Figure S5G). Alternatively, binding of eIF4E to the mRNA cap may become severed during scanning. Phosphorylation of eIF4E by MNK1/2 decreases eIF4E affinity to the cap (Scheper et al., 2002). To test a role of MNK1/2, we either activated MNK1/2 with epidermal growth factor (EGF) (in EGF-responsive HEK293T cells) or inhibitedMNK1/2 pharmacologically with eFT508, but we did not observe any effect on the reporters (Figure S5H). Thus, further research will be required to understand the mechanism causing scanning to become cap severed. eIF3B, eIF4G1, and eIF4E Persist on Elongating 80S Ribosomes with a Decay Half-Length of 12 Codons Since eIF3B, eIF4G1, and eIF4E are retained on 40S ribosomes up to the start codon, we asked if they can remain bound to 80S ribosomes after subunit joining. We performed selective 80S footprinting, immunoprecipitating each of these eIFs from 80S fractions of a sucrose gradient (Figures S3A–S3C and S4A). Footprint sequencing revealed an enrichment of eIF3B-, eIF4G1-, and eIF4E-containing 80S ribosomes on main ORF start codons compared to total 80S ribosomes (Figure S6A). In contrast, eIF2S1 is depleted (Figure S6A), consistent with it falling off at the start codon (Figure 2A). To analyze the dissociation of these eIFs from elongating 80S ribosomes, we normalized down the height of the selective 80S peaks to equal the total 80S on the start codon (Figure 5A). This revealed that eIFs do not dissociate immediately from 80S ribosomes as they start elongating (Figure 5A). By calculating the ratio of selective versus total 80S footprints at each position of the ORF, we found an exponential dissociation of eIFs from the 80S ribosome, with a half-length of 36 nt (Figures 5B and S6B for mouse eIF4E) corresponding to 12 elongation cycles. An exponential decay (r2 > 0.8) suggests the dissociation process is stochastic. In sum, eIF3 and eIF4 complexes remain associated to initiating and early translating 80S ribosomes, with functional consequences, which we analyze below. eIFs Persist Past uORFs on Translating Ribosomes Roughly half of human mRNAs contain uORFs, and ribosome footprinting has shown that many uORFs are translated (Calvo et al., 2009; Ingolia et al., 2011; Johnstone et al., 2016). In such cases, ribosomes need to terminate translation on the uORF, resume scanning, and reinitiate downstream on the main ORF (Guni sová et al., 2018; Jackson et al., 2012). Themolecular steps Molecular Cell 79, 1–14, August 20, 2020 7 (legend on next page) 8 Molecular Cell 79, 1–14, August 20, 2020 are not fully understood. For instance, it is unclear how ribosomes re-recruit eIFs after uORF translation. Our eIF selective footprinting data provide a possible explanation; if the uORFs are short enough, then eIFs such as eIF3 and eIF4G might be retained on the 80S ribosome up to the uORF stop codon. This would make the ribosome competent to reinitiate by stabilizing mRNA association and providing a platform to re-recruit the eIFs that were lost at 60S joining (eIF1, eIF1A, eIF2, and eIF5). To study this, we analyzed uORFs with detectable 80S footprints on their start codon, indicating they are translated. From these, we selected only uORFs with an intercistronic spacing (ICS) R 80 nt to the main ORF start codon to resolve uORF translation from main ORF translation (6,663 uORFs in 3,916 transcripts). Metagene plots of such uORFs confirmed that 40S and 80S ribosomes accumulate on their start codons (Figures 5C and 5D), indicating they are recognized as translation start sites. Indeed, 80S ribosomes translating these uORFs display triplet periodicity when the curves are not smoothened (Figures S6C and S6D). Just like 40S ribosomes on main-ORF start codons, 40S ribosomes on translated uORF start codons have eIF3B, eIF4G1, and eIF4E but are depleted of eIF2S1 (blue trace is below the others; Figure 5C). All 40S and 80S graphs in Figures 5C–5D00 are normalized using the same values as in Figures 2A and 5A, respectively, to make them directly comparable. Note that the coding sequences of the uORFs in Figure 5C have varying lengths (6–1,284 nt, with amean of 36 nt); hence the start codons are aligned, but the stop codons are not and occur at various downstream positions. As a result, 40S ribosomes and eIF2S1 are depleted directly after the start codon but progressively return to baseline further downstream as the uORFs asynchronously finish (Figure 5C). If we align translated uORF stop codons, several observations can be made (Figure 5C0 and 5D0). First, 80S ribosomes disappear and scanning 40S ribosomes re-appear after the stop codon. This suggests that reinitiation is performed by scanning 40S, not 80S, ribosomes, in agreement with initial propositions, but not recent suggestions (Zhou et al., 2018). In this metagene plot, uORF start codons occur at different upstream positions, since the uORFs are of varying lengths but aligned to the stop codon. Hence, the high 40S signal from start codon peaks is visible throughout the region 60 to 20 in Figure 5C0. Second, we see that eIF2S1 is quickly re-recruited to 40S ribosomes after termination on the uORF stop codon (the blue trace rejoins the other traces after the stop codon; Figure 5C0). Third, after the uORF stop codon, eIF3B, eIF4G, and eIF4E are not depleted from 40S ribosomes. Relative to total 40S ribosomes, these eIFs are present at similar levels af- ter the uORF stop codon as they were before the start codon (Figures 5C and C0). Indeed, as these 40S ribosomes move downstream and reach the main ORF start codon, they have eIF3B, eIF4G1, and eIF4E, and they become depleted of eIF2S1, just as on transcripts lacking uORFs (Figures 5C00 and 5D00). Similar results were obtained by analyzing a set of uORFs found to be translated in a meta-analysis of 35 ribo-seq experiments (Figures S6E-F00) (Scholz et al., 2019). These data are consistent with a model of translation reinitiation whereby most eIFs are retained on 80S ribosomes as they translate short uORFs, thereby enabling the post-termination 40S ribosomes to re-recruit eIF2 and to recommence scanning (Figure 5E). In contrast, on main ORFs that are much longer, the eIFs are depleted by the time the 80S reaches the stop codon, causing ribosomes to release the mRNA after terminating and not scan the 30 UTR (Figure 1B).

Cellular Stress Induces Mainly Low tRNA-iMet Binding to the TC Compared to Low TC Binding to 40S
Inactivation of eIF2 is one of the principal modes of translation regulation. Many different cellular stresses lead to phosphorylation of eIF2S1/eIF2a and consequently inactivation of the eIF2 complex (McConkey, 2017; Pakos-Zebrucka et al., 2016). This causes a global drop in translation while activating translation of stress response proteins such as ATF4 via a reinitiation mechanism. In brief, ribosomes translate a uORF in the ATF4 50 UTR and then recharge with the eIF2$GTP$inititator-tRNA TC. If eIF2 activity is high, then they recharge quickly and translate a second, decoy uORF. If they recharge slowly, then they skip past the decoy uORF and translate the ATF4main ORF (Harding et al., 2000; Lu et al., 2004; Vattem and Wek, 2004). Hence, the speed of recruitment of the initiator tRNA is key. In vitro, two modes of initiator tRNA recruitment are possible (Sokabe et al., 2012). Initiator tRNA can first bind eIF2$GTP to form the TC, which then binds the ribosome. Alternatively, eIF2 can first bind the ribosome and then recruit initiator tRNA. Which happens predominantly in vivo is not known. Note that for regulation of ATF4 by stress, either a delay in eIF2 recruitment to the 40S or a delay in initiator tRNA recruitment to eIF2 would lead to a lack of initiator tRNA on the decoy uORF and hence would be compatible with current models of ATF4 reinitiation. To test this, we performed eIF2S1-selective 40S footprinting on cells treated with tunicamycin to induce endoplasmic reticulum (ER) stress (Figure S3B). We then analyzed eIF2 recruitment after translated uORFs. As expected, both in the presence and absence of stress, eIF2a is depleted from 40S ribosomes on ting 80S Ribosomes Leads to Initiation Factor Retention after Trans- (legend on next page) 10 Molecular Cell 79, 1–14, August 20, 2020 uORF start codons (blue and pink curves, Figure 6A). Interestingly, eIF2a is re-recruited after translated uORF stop codons both in the unstressed and stressed conditions (Figure 6A0). Instead, the amount of initiator-tRNA detected in our eIF2S1-selective 40S pull-downs was reduced in the presence of stress (Figure 6B). These data indicate that in vivo, upon stress, many ribosomes bind eIF2; however, these eIF2 complexes contain less initiator tRNA, which is then recruited to the eIF2$40S complex in a delayed manner. In comparison, there may be a mild delay in eIF2 recruitment to the 40S upon stress that we cannot exclude.

A Comprehensive View of Translation Initiation on a
Single Endogenous Transcript In Vivo Many of the observations mentioned above derive from metagene analyses combining data frommany transcripts. We asked whether selective 40S footprinting can yield insights into translation initiation of individual transcripts. We analyzed the eIF4G2 mRNA, which interestingly has a uORF and a GUG start codon on the main ORF (Figure 6C). On a 2D 40S plot, the uORF start codon has a characteristic peak with a tail (feature 1; Figure 6E), indicating increased 40S dwell time and start codon recognition. This is accompanied by depletion of eIF2a, visible on the 1D 40S plot (feature 2; Figure 6D), in agreement with some initiation on this uORF. Indeed, 80S footprints are visible on the uORF (feature 3; Figure 6F). Scanning 40S footprints are also visible within the uORF (feature 4), indicating there is also leaky scanning past the uORF start codon. On the uORF stop codon 80S footprints decrease and eIF2a is re-recruited to 40S ribosomes (blue trace, feature 5). All the while, the proportion of 40S ribosomes containing eIF3B, eIF4G1, and eIF4E remains roughly constant, and 40S ribosomes scanning between the uORF and the main ORF are still bound to these eIFs (feature 6). On the near-cognate GUG start codon of the main ORF, something non-canonical is happening. Scanning 40S ribosomes recognize the start codon (high 40S peak at this location, feature 7, and diagonal lines on the 2D plot, feature 8); however, eIF2a is not depleted (feature 7). This is analyzed more below. Leaky scanning can be seen by 40S footprints downstream of the GUG start codon (feature 9), confirming it is a weak initiation signal. These ribosomes become ‘‘trapped’’ by an internal out-of-frame ORF (iORF; Figure 6C), which to our knowledge is not yet annotated. On this iORF, initiation occurs, seen as a 40S peak that is depleted of eIF2a (feature 10) and the appearance of 80S footprints (feature 11). Figure 6. Initiation Factor 2 Complex Is Slowly Recharged with iMet-tR Near-Cognate Start Codons (A–A00) eIF2S1 is re-recruited to scanning 40S ribosomes after translation of a u footprints relative to the start (A) or stop codon (A0) of translated uORFs (n = 6,663) (UT) or treated with tunicamycin (TM; replicate 1: 250 ng/mL; replicate 2, 1 mg/mL number of scanning 40S footprints (positions 98 to 69) as in Figure 2A. (B) Binding of methionine initiator-tRNA to eIF2S1 is stress dependent. Count of 40S footprint libraries from untreated and tunicamycin-treated cells. (C) Diagram of ORF positions in the eIF4G2 mRNA (ENST00000339995). (D–F) Reinitiation, near-cognate GTG initiation, and leaky scanning occur on the eI resolved counts of total and selective 40S ribosome footprints. Reads normalized (G) Start codons on eIF4G2 mRNA revealed by 80S accumulation upon harringto depth and smoothened with a 10-nt sliding window. We considered two explanations for the lack of a drop in eIF2a-selective 40S footprints on the GUG start codon. The first is that translation is not initiated with eIF2a. The second is that initiation is eIF2a dependent. However, start codon recognition is slow due to poor pairing between GUG and initiator-tRNAMet. Hence, eIF2a remains bound to the 40S for a long time before leaving, and a larger proportion of 40S footprints contain eIF2a. We aimed to distinguish these two options. In the first scenario, there are two populations of 40S ribosomes: those with eIF2a that pause but scan past the GUG start codon, and those with another unknown eIF that initiate translation on the GUG and convert to 80S ribosomes. In this case, downstream of the start codon eIF2a-containing 40S ribosomes should become comparatively enriched compared to total 40S ribosomes, because the 40S population containing the unknown eIF becomes depleted. Roughly the same proportion of ribosomes, however, contain eIF2a on the GUG start codon as they do further downstream (at 540 nt, feature 9). Hence, if there is a pool of 40S ribosomes containing an unknown eIF, then it must be a small fraction of the total 40S. In this case, only a small fraction of 40S ribosomes should initiate on the GUG while most 40S ribosomes should scan past and initiate downstream. Harringtonine treatment, however, revealed that most ribosomes initiate on the GUG start codon and not on the downstream iORF (Figure 6G). Hence the majority of ribosomes, which are the ones containing eIF2a, initiate on the GUG. From this, we conclude that translation initiation on this GUG start codon is mediated by eIF2a, consistent with recent findings that TC inhibition blocks initiation on near-cognate start codons (Kearse et al., 2019). In sum, selective 40S and 80S footprinting allows a detailed view of the sequential steps of translation initiation occurring in vivo and on a single endogenous transcript.

Together with Wagner et al. (2020) (in this issue of Molecular Cell), we present here a method to analyze translation initiation on endogenous mRNAs in vivo. It allows translation initiation to be investigated in a cellular context that includes regulatory mechanisms such as eIF2 inactivation. It can be applied to study other aspects of translation initiation, e.g., how initiation is altered if m6A methylation is reduced (Coots et al., 2017; Ozkurede et al., 2019; Patil et al., 2018). Likewise, the function of less well-studied eIFs such as eIF2A could be revealed.

NA on 40S Ribosomes during Stress and Can Initiate Translation on
ORF independent of cellular stress. Metagene plots of total or selective 40S or the start codon of themainORF downstream (A00). Cells were either untreated ) for 16 h. Reads were mapped to their 50 end. Graphs were normalized to the iMet tRNA reads per 1,000 sequenced reads in total 40S and eIF2S1-selective F4G2mRNA. (D) Counts, smoothenedwith a 10-nt sliding window, or (E) length- using scaling factors from Figure 2A. (F) Length-resolved total 80S footprints. nine treatment (2 mg/mL, 37 C). Graphs were normalized to library sequencing Molecular Cell 79, 1–14, August 20, 2020 11 Using this method, we discover that (1) scanning is mainly cap tethered in human cells; and (2) eIF3B, eIF4G1, and eIF4E persist on translating 80S ribosomes with a half-length of 12 codons, in agreement with Lin et al., 2020 (in this issue of Molecular Cell). This has several implications. First, usually only one ribosome will scan a 50 UTR at a time. This agrees with our harringtonine data, where we do not see 40S ribosomes accumulating in front of a stalled initiating 80S ribosome in human cells (Figures 3D and 3E). In contrast, 40S queuing has been observed in rabbit reticulocyte lysates (Kozak, 1991), suggesting a difference in vitro versus in vivo. Second, this implies 50 UTR length influences translation efficiency, as we see in luciferase assays where we only modulate 50 UTR length (Figures 3F and 3G). Third, it provides a possible explanation for how translation reinitiation works in human cells; the eIFs necessary for mRNA binding, TC recruitment, and scanning directionality (Pöyry et al., 2004, 2007; Skabkin et al., 2013) are still present on ribosomes after uORFs, which tend to be short. Finally, 80S ribosomes that stall near the start codon of an ORF (as with harringtonine) will prevent a new 40S ribosome from scanning the 50 UTR, whereas 80S ribosomes that stall >12 codons downstream of a start codon will release the cap, allowing a new 40S to scan and then queue, as has been observed (Ivanov et al., 2018; Kearse et al., 2019). Although we find 40S scanning is mainly cap tethered in most mammalian cells, this is not the case in NIH 3T3s. This raises the possibility that the mode of 40S scanning could be modulated. Whether this happens physiologically or only due to mutations remains to be investigated. The scanning process has some differences in humans and yeast. In yeast, ribosomes are generally not competent to reinitiate after uORFs (Yun et al., 1996), unless specific cis-acting elements are present, as on the GCN4 mRNA (Mohammad et al., 2017; Szamecz et al., 2008). In contrast, mammalian cells do not seem to require cis-acting elements on mRNAs; they are generally reinitiation competent (Johansen et al., 1984; Kozak, 1984; Liu et al., 1984). The difference between the two systems may be quantitative (i.e., how strongly ribosomes remain eIF associated during scanning and elongation). If ribosomes shed eIFs more quickly in yeast than in humans, for instance after translating 1 or 2 codons, then this may render them incapable of uORF reinitiation. In contrast to what we observe in human cells (Figures 3F and 3G), in yeast, 50 UTR length does not affect translation efficiency (Berthelot et al., 2004), and 40S queuing has been observed in vivo (Archer et al., 2016). All this could be explained if yeast ribosomes detach from the cap more quickly than in humans, enabling parallel scanning of a 50 UTR by multiple ribosomes. This would enable higher translation rates, which would be beneficial for yeast cells with a rapid cell cycle. Binding of eIF3 and eIF4F tomammalian 80S ribosomes is sterically possible, since eIF4F directly binds eIF3 (Shirokikh and Preiss, 2018) and eIF3 binds the 40S ribosomemainly on the solvent-exposed side unaffected by 60S subunit joining (Aylett et al., 2015; des Georges et al., 2015). In yeast, eIF3 consists of fewer proteins than human eIF3 (Hinnebusch, 2006; Phan et al., 1998), and eIF4F binds the ribosome via eIF5 and eIF1 (Asano et al., 2001; He et al., 2003; Shirokikh and Preiss, 12 Molecular Cell 79, 1–14, August 20, 2020 2018), which leave the ribosome upon 60S subunit joining. Thus, it seems likely that eIF4F cannot persist on 80S ribosomes in yeast. Some of these molecular differences may account for differences in the scanning process. It is unclear if mRNAs are stably circularized in vivo, with the 50 cap contacting the poly(A) tail (Adivarahan et al., 2018; An et al., 2018; Gallie, 1991; Khong and Parker, 2018; Wells et al., 1998). If so, this might help re-recruit terminated ribosomes to the cap for a new round of translation. Our data indicate that 40S ribosomes do not scan down the 30 UTR to the poly(A) tail after terminating on the main ORF but rather fall off at the stop codon. Hence, to help re-recruit ribosomes to the cap, circularization would need to bring the main ORF stop codon close to the cap. In sum, we provide here insights into scanning and translation initiation in human cells. Selective 40S footprinting may develop into a powerful approach to study translation initiation mechanisms in vivo on endogenous mRNAs in the future. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d KEY RESOURCES TABLE d RESOURCE AVAILABILITY B Lead Contact B Materials Availability B Data and Code Availability d EXPERIMENTAL MODEL AND SUBJECT DETAILS B Cell lines and Culture Conditions d METHOD DETAILS B Cloning B Immunoblotting B Quantitative RT-PCR B Translation Reporter Dual-Luciferase Assay B 40S and 80S Ribosome footprinting B Deep-sequencing library preparation d QUANTIFICATION AND STATISTICAL ANALYSIS SUPPLEMENTAL INFORMATION Supplemental Information can be found online at molcel.2020.06.005.

We thank Georg Stoecklin, Johanna Schott, and Kathrin Leppek for sharing with us their ribosome footprinting protocol, for scientific discussion, and for giving us the eIF3B antibody; Oliver Hoppe for designing the graphical abstract; and Wilhelm Palm for providing the KRPC cells. This work was funded in part by a Deutsche Forschungsgemeinschaft (DFG [German Research Foundation]) Collaborative Research Center grant (project ID 201348542, SFB 1036) to B.B. and A.A.T., a DKFZ NCT3.0 Integrative Project in Cancer Research (NCT3.0_2015.54 DysregPT) grant to B.B. and A.A.T., an ERC Advanced grant (Transfold) to B.B., and a Cell Networks – Cluster of Excellence (EXC81) grant to J.B. High-throughput sequencing was carried out at the DKFZ Genomics and Proteomics Core Facility or using an Illumina NextSeq 550 system funded by the Klaus Tschira Foundation (Heidelberg, Germany).

Almost all experiments were performed by J.B., except the Ingolia/Bukau sequencing libraries, which were prepared by K.F. All authors designed the work, analyzed data, interpreted data, and wrote the manuscript.

The authors declare no competing interests. Received: November 6, 2019 Revised: April 10, 2020 Accepted: May 18, 2020 Published: June 25, 2020

REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Tubulin T9026 Sigma RRID AB_477593 RPS15 Atlas HPA054510 RRID AB_2682510 RPLP0 Atlas HPA003512 RRID AB_1079844 eiF2S1 Cell Signaling #5324 RRID AB_10692650 eIF3B Santa Cruz 16377 (Discontinued) RRID AB_671941 eIF3A Cell Signaling #3411 RRID AB_2096523 eIF4E MBL RN001P RRID AB_1570634 eIF4G1 MBL RN002P RRID AB_1570635 eIF1 OriGene TA502844 RRID AB_11141763 eIF3D Atlas HPA063330 RRID AB_2684988 ATF4 Cell Signaling 11815 RRID AB_2616025 anti phospho 4E-BP1 (Ser65) Cell Signaling 9451 RRID AB_330947 4E-BP1 Cell signaling 9452 RRID AB_331692 Bacterial and Virus Strains One Shot TOP10 Chemically Competent E. coli ThermoFisher Cat# C404006 Chemicals, Peptides, and Recombinant Proteins Torin1 Cayman Chemical Cat# Cay10997-50 cOmplete Protease Inhibitor Cocktail Roche Diagnostics Cat# 11836145001 cOmplete Protease Inhibitor Cocktail (EDTA-free) Roche Diagnostics Cat# 11836170001 PhosSTOP Roche Diagnostics Cat# 4906845001 Dithiobis(succinimidyl propionate) (DSP) ThermoFisher Cat# 22585 Lipofectamine 2000 ThermoFisher Cat# 11668 Dynabeads Protein A ThermoFisher Cat# 10002D Dynabeads Protein A ThermoFisher Cat# 10003D SMARTer smRNA-Seq Kit for Illumina Takara/Clonentech Cat# 635031 Tunicamycin Pan Reac AppliChem Cat# A2242,0005 16% PFA ThermoFisher Cat# 28906 Deposited Data Sequencing Data from this Study NCBI Geo GEO: GSE139391 Uncropped Immunoblots on Mendeley Data Mendeley Data Experimental Models: Cell Lines Human: MCF7 Dr. Boutros (DKFZ) N/A Human: HEK293T Dr. Boutros (DKFZ) N/A Human: HeLa ATCC RRID:CVCL_0030 Mouse: KRPC Dr. Wilhelm Palm (DKFZ) N/A Mouse NIH 3T3 DKFZ N/A Oligonucleotides OJB0440 Sigma-Aldrich N/A OJB0441 Sigma-Aldrich N/A OJB0494 Sigma-Aldrich N/A OJB0495 Sigma-Aldrich N/A OJB0496 Sigma-Aldrich N/A OJB0497 Sigma-Aldrich N/A (Continued on next page) Molecular Cell 79, 1–14.e1–e5, August 20, 2020 e1

REAGENT or RESOURCE SOURCE IDENTIFIER Recombinant DNA Plasmids: 50UTR Length Reporters 1-32 repeats This paper N/A Plasmids: 30UTR Length Reporters 1-32 repeats This paper N/A Plasmids: 50UTR Length Reporters Near-cognate mutated 1-28 repeats This paper N/A Software and Algorithms Adobe Photoshop photoshop.html RRID:SCR_014199 GraphPad Prism 7 RRID:SCR_002798 Image Lab Software RRID:SCR_014210 Microsoft Excel v16.16.19 RRID:SCR_016137 R & R studio RRID:SCR_001905 Cutadapt RRID:SCR_011841 Bowtie2 index.shtml RRID:SCR_005476 Bbmap RRID:SCR_016965 ImageJ RRID:SCR_003070 Custom C programs This Paper N/A


Lead Contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Aurelio Teleman (

Materials Availability
Unique reagents generated in this study can be obtained by emailing the Lead Contact.

Data and Code Availability
All custom software used in this study is available on GitHub at All sequencing data are available at NCBI Geo (GSE139391). Uncropped immunoblots are available at Mendeley Data at the following URL http://dx.doi. org/10.17632/h8kdp823dg.1. A table summarizing read counts per transcript for different experiments is provided as Table S7.


Cell lines and Culture Conditions
HeLa, NIH 3T3,MCF7, KRPC andHEK293T cells were cultured in DMEM+10% fetal bovine serum+100U/ml Penicillin/Streptomycin (GIBCO 15140122). Cells were sub-cultured using Trypsin-EDTA for dissociation. Cellular stress conditions were induced by treatment with Tunicamycin at 1 mg /ml or 250 ng /ml. Cells weremaintained at 30%–90%confluence. Cells were treatedwith 25 nMTorin to inhibit mTORC1/2 and 1 mM eFT508 to inhibit MNK1/2. HeLa cells were authenticated using SNP typing and tested negative for mycoplasma.


Sequences of oligos used for cloning are provided in Table S3 at the end of the Materials & Methods. Translation reporters with increasing 50UTR length were generated as follows. Oligo pairs were annealed and oligo cloned into a pcDNA3 vector in between a CMV promoter and the Renilla luciferase ORF using HinDIII and Bsp119l sites. The resulting plasmid has a 50UTR of 60 nt. To generate reporters with increasing 30UTR lengths, first, EcoR1 and Xho1 sites were introduced between the stop codon and the SV40 polyadenylation signal, then primer pairs were ligated into this junction as explained above. These plasmids were then opened with EcoR1 and Sal1/Xho1, separately the 26-mer was excised using EcoR1 and Xho1, and then the 26-mer was inserted into the opened plasmid to double the number of 26-mers in the 50UTR or 30UTR. This procedure was repeated multiple times to yield the panels of reporters shown in Figures 3G, 4A, S5A, and S5B. e2 Molecular Cell 79, 1–14.e1–e5, August 20, 2020

Protein solutions from sucrose gradients, immunoprecipitations and lysates were run on SDS-PAGE gels and transferred to nitrocellulose membrane with 0.2 mm pore size. After Ponceau staining membranes were incubated in 5% skim milk PBST for 1 hour, briefly rinsed with PBST and then incubated in primary antibody solution (5% BSA PBST or 5% skim milk PBST) overnight at 4 C. Membranes were then washed three times, 15 minutes each in PBST, incubated in secondary antibody solution (1:10000 in 5% skim milk PBST) for 1 hour at room temperature, then washed again three times for 15 minutes. Finally, chemiluminescence was detected using ECL reagents and the Biorad chemidoc. No membranes were stripped. Antibodies used for immunoblotting are listed in Table S1.

Quantitative RT-PCR
Total RNA was isolated from cells using QIAGEN RNAeasy spin columns, including on column DNase digestion to remove plasmid DNA. To synthesize cDNA, 1 mg of total RNAwas used for oligo dT primed reverse-transcription usingMaximaHMinus Reverse Transcriptase. Quantitative RT-PCR was run with Maxima SYBR Green/ROX mix on a StepOnePlus Real-Time PCR System. Actin B (human) and RPL13a (mouse) were used as a normalization controls and all samples were run in technical triplicates. Non-reverse transcribed RNA, H2O and cDNA from non-transfected cells were assayed to ensure that Renilla luciferase qPCR signal originates from Renilla Luciferase mRNA, not plasmid DNA or unspecific amplification. Sequences of oligos used for Q-RT-PCR are provided in Table S2.

Translation Reporter Dual-Luciferase Assay
For translation reporter assays after plasmid transfection, HeLa cells were seeded at 8.000-10.000 cells per 96-well. 16-20 hours after seeding, these cells were transfected with three plasmids using lipofectamine 2000. Per well, 60 ng of either GFP expression plasmid, 70 ng of renilla luciferase reporter plasmid and 70 ng of firefly reporter plasmid were used. 3-5 hours after transfection, the medium was exchanged. Renilla luciferase plasmids always contained the 50UTR of interest. 0.4 ml Lipofectamine reagent was used per 96-well. Cells were always transfected in six replicates. 16-20 hours after transfection, luciferase activity was assayed using the Dual-Luciferase Reporter Assay System by Promega according to the manufacturer’s instructions. To calculate Renilla Luciferase signal per mRNA, only the Rluc signal (not normalized to Fluc) was used and normalized to Rluc mRNA levels from qPCR experiments on cells transfected with the same transfection mixture.

40S and 80S Ribosome footprinting
Two days before cell harvest, HeLa cells were seeded at 1.5 million cells per 15 cm dish in 20 mL growth medium. For stress conditions, cells were treated with 1 mg/ml or 250 ng/ml tunicamycin 16 hours before cell harvest. For cell harvest, growthmediumwas poured off and cells were quickly washed with ice-cold washing solution (1x PBS 10 mMMgCl2 800 mM Cycloheximide). For harringtonine block experiments, cells were treated with 2 mg/ml harringtonine at 37 C for 1-4 minutes before washing. Washing solution was immediately poured off and freshly prepared crosslinking solution (1x PBS, 10 mM MgCl2, 400 mM Cycloheximide, 0.025%PFA, 0.5mMDSP) was added to the cells. When cells were not crosslinked, lysis was carried out directly after this washing step. Cells were incubated with crosslinking solution for 15 minutes at room temperature while slowly rocking. Crosslinking solution was then poured off and remaining crosslinker was inactivated for 5 minutes with ice-cold quenching solution (1x PBS, 10 mM MgCl2, 200 mM Cycloheximide, 300 mM Glycine). Quenching solution was poured off and 150 ml of lysis buffer (0,25 M HEPES pH 7.5, 50mMMgCl2, 1MKCl, 5%NP40, 1000 mMCHX) was added to each 15 cm dish, resulting in 750mL of lysate. Lysis was carried out at 4 C. Cells were scraped off the dish and the lysate was collected. After brief vortexing, lysates were clarified by centrifugation at 20.000xg for 10 minutes at 4 C. Supernatant was collected and approximate RNA concentration was determined using a Nanodrop photo-spectrometer. 100 U of Ambion RNase 1 was added per 120 mg of measured RNA. To obtain polysome profiles, no RNase was added. Lysates were incubated for 5 minutes at 4 C and then loaded onto 17.5%–50% sucrose gradients and centrifuged for 5 hours at 35.000 rpm in a Beckman Ultracentrifuge in the SW40 rotor. Gradients were fractionated using a Biocomp Gradient Profiler system. 40S and 80S fractions were collected for immunoprecipitation and footprint isolation. 40S and 80S fractions corresponding to roughly one or two 15 cm dishes were used for direct extraction of RNA for total footprint samples. 40S and 80S fractions corresponding to roughly ten 15 cm dishes were used for immunoprecipitation of initiation factor bound ribosomes, NP40 was added to these fractions to 1% final concentration. For immunoprecipitation, antibodies were bound to protein A or protein G magnetic dynabeads (Thermo) according to the manufacturers instructions. Antibodies & amounts used for ribosome immunoprecipitation are listed in Table S5. Beads were washed three times and then added to the 40S or 80S fractions. Fractions with beads were incubated for 2 hours, rotating at 4 C. Then beads were washed three times with bead wash buffer (20 mM Tris pH 7.4, 10 mM MgCl2, 140 mM KCl, 1% NP40), including a change of reaction-vessels during the last wash. Bead volume was increased to 500ml with bead wash buffer. Total footprint fractions and IPed fractions were then subjected to crosslink removal and RNA extraction: 55 ml (1/9th of volume) of crosslink-removal solution (10% SDS, 100 mM EDTA, 50 mM DTT) was added, 600 ml Acid-Phenol Chloroform (Ambion) was added and mixture was incubated at 65 C, 1300 rpm shaking for 45 minutes. Tubes were then placed on ice for 5 minutes, spun for 5 min at 20.000 g and supernatant was washed once with acid-phenol chloroform and twice with chloroform, then RNA was precipitated with isopropanol and subjected to library preparation (see below). The organic phase was used to isolate the precipitated or total proteins. 300 ml Ethanol were added, then 1,5 mL isopropanol were Molecular Cell 79, 1–14.e1–e5, August 20, 2020 e3 added and solutions were incubated at 20 C for 1 hour. Proteins were sedimented by centrifugation at 20.000 g for 20 minutes, washed twice with 95% Ethanol 0,3 M Guanidine HCl, dried and resuspended in 1x Laemmli buffer.

Deep-sequencing library preparation
During development of the 40S and 80S selective ribosome footprinting method, we optimized and hence changed several parameters. Here we outline first the final, optimized protocol, which we recommend people use for future experiments. Afterward, we briefly explain the variations of the method which apply to some of the datasets in this manuscript. In Table S6 we indicate which deep-sequencing libraries were prepared with which protocol and the numbers of reads with adapters, rRNA, tRNA and mapped reads. Optimized protocol: After RNA extraction from total and IP-purified fractions, RNA quality and integrity were determined on an Agilent Bioanalyzer using the total RNA Nano 6000 Chip. For size selection, RNA was run on 15% Urea-Polyacrylamide gels (Invitrogen) and fragments of size 20-60 nt (80S libraries) and 20-80 nt (40S libraries) were excised using the Agilent small RNA ladder as a reference. RNA was extracted from the gel pieces by smashing the gels into small pieces with gel smasher tubes and extracting the RNA in 0.5 mL of 10 mM Tris pH 7 at 70 C for 10 minutes. Gel pieces were removed and RNA was precipitated using isopropanol (with 150 mM Sodium Acetate pH 5.2, 150 mM NaCl). Footprints were then dephosphorylated using T4 PNK (NEB) for 2 hours at 37 C in PNK buffer without ATP. Footprints were then again precipitated and purified using isopropanol. For 40S footprints, contaminating 18S rRNA fragments were depleted as follows. Prevalent 18S rRNA fragments from the first round of 40S footprinting were used to design complementary Biotin-TEG-DNA oligonucleotides (sequences listed in Table S4, ordered from Sigma-Aldrich). 100ng of RNA footprints were then hybridized to a mixture (proportional to occurrence of the fragment, listed in Table S4) of these DNA oligos (in 40x molar excess) in (0.5M NaCl, 20mM Tris pH7.0, 1mM EDTA, 0.05% Tween20) by denaturing for 90 s at 95 C and then annealing by reducing the temperature by 0.1 C/sec down to 37 C, then incubating 15min at 37 C. Hybridized species were pulled out using Streptavidin magnetic beads (NEB) by incubating at room temperature for 15 minutes, and the remaining RNA was purified by isopropanol precipitation. For harringtonine experiments, no rRNA depletion was carried out, as to not bias the rRNA to footprint ratio. Footprints were then assayed using an Agilent Bioanalyzer small RNA chip and Qubit smRNA kit. 25 ng or less of footprint RNA was used as input for library preparation with SMARTer smRNA-SeqKit for Illumina from Takara / Clontech Laboratories according to the manufacturer’s instructions. Deep-sequencing libraries were sequenced on the Illumina Next-Seq 550 system. For RNA-seq libraries, total cell RNA was extracted using TRIzol and library preparation was performed using the Illumina TruSeq Stranded library preparation kit. These RNA-seq libraries were also sequenced on the Illumina Next-Seq 550 system. Protocol variants used for some of the datasets: Two 80S footprint libraries and one RNA-seq library (as listed in Table S6) were prepared using a the NEXTflex Small RNA-Seq Kit v3 library preparation kit instead of the Takara/Clonentech kit. For this, size selected, rRNA depleted RNA was phosphorylated using T4 PNK. Footprints were then assayed using an Agilent Bioanalyzer small RNA chip and a Qubit smRNA kit. Deep sequencing libraries were prepared from these RNA fragments using the Bio-Scientific NEXTflex Small RNA-Seq Kit v3. Deep-sequencing libraries were sequenced on the Illumina Next-Seq 550 system. In some cases (Table S6), the Ribo Zero Gold rRNA depletion kit (Illumina) was used instead of our custom rRNA depletion protocol mentioned above. However, this kit was discontinued half-way though the project and is no longer available. Two footprinting samples - one 40S total and one 40S eIF4E selective - were each split into two halves to prepare libraries using both the protocol detailed above and the Ingolia et al. protocol as implemented by the Bukau lab (Galmozzi et al., 2019; McGlincy and Ingolia, 2017).

Adaptor sequences and randomized nucleotides (Nextflex) or polyA stretches (Clonentech/Takara) were trimmed from raw reads using cutadapt ( Nuclear and mitochondrial Ribosomal RNA and tRNA reads were removed by alignment to human tRNA and rRNA sequences using bowtie2 (22388286). Then, the remaining reads were separately aligned to the human transcriptome (Ensemble transcript assembly 94) and human genome (hg38) using BBmap ( Multiple mappings were allowed. Secondary mappings were counted when analyzing single transcripts, but not counted in metagene plots to avoid biasing genes with many transcript isoforms or reads with low sequence complexity. Metagene plots, single transcript traces and grouped analyses were carried out or created with custom software written in C, supplied in Data S1. Read counts for metagene plots of whole transcripts that encompass 50UTRs, ORFs and 30UTRs (Figure 1B) were normalized for the length of each of these features to make them comparable. 70 of the 41.314 transcripts were excluded from the analyses (listed in Table S8) because PCR artifacts mapped to these transcripts. 2-D metagene plots were visualized using Fiji (Schindelin et al., 2012). The same, linear color gradient was always used to indicate read counts at each position in these 2-D metagene plots. Importantly, the maximum intensity of the color gradient was adjusted for the library size of the sample e4 Molecular Cell 79, 1–14.e1–e5, August 20, 2020 displayed so that 2-Dmetagene plots can be compared across panels. For 80S plots, only reads with footprint lengths between 26 to 37 nt were counted. For 50UTR plots, only 40S footprints larger than 29 nt were counted because of the occurrence of artifacts and misaligned reads in the shorter population. Translated uORFs were defined by the presence of any 80S ribosome footprints in a 10-nucleotide window around the uORF start codon. Only ATG initiated uORFs were considered. uORFs annotated by uORF-Tools (Scholz et al., 2019) were downloaded from Table S2 and filtered for uORFs with an intercistronic space > 80 nt. Molecular Cell 79, 1–14.e1–e5, August 20, 2020 e5

Jonathan Bohlen, Kai Fenzl, Günter Kramer, Bernd Bukau, Aurelio A Teleman
Molecular cell
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6