Relaxed initiation pausing of ribosomes drives oncogenic translation – Science Advances

RAS^G12V-mediated oncogenic translation

Cancer-causing RAS mutations impair the GTPase activity, resulting in constitutive activation of the RAS signaling pathway. Since the oncogenic potential of RAS often manifests in a context-dependent manner, the translational status associated with malignant progression is confounded by mixed genetic backgrounds and varying tissue environments. To circumvent this complexity, we took advantage of a human TtH cell line stably expressing a 4-hydroxytamoxifen (4-OHT)–inducible ER:HRAS^G12V fusion protein (21). This system permits direct comparison of translational landscapes with or without oncogenic RAS expression under the same genetic background. Upon treatment with 4-OHT, ER:HRAS^G12V was readily detectable in these cells (Fig. 1A) but not the vector control cells (fig. S1A). As expected, oncogenic RAS expression led to increased phosphorylation of Akt as well as mTOR complex 1 (mTORC1) downstream targets S6K1 and 4EBP1 (Fig. 1A). 4-OHT treatment increased the polysome/monosome (P/M) ratio (Fig. 1B), confirming that hyperactive RAS signaling stimulates protein synthesis (4, 21). In agreement, a firefly luciferase (Fluc) reporter also showed an increased expression in the presence of 4-OHT (Fig. 1C). The enhanced translational capacity was not due to the side effect of 4-OHT because the same treatment had negligible effect in vector control cells (fig. S1, B and C).

It is unclear whether the increased translational output in the presence of RAS^G12V is the sole consequence of augmented mTORC1 signaling. If so, then mTORC1 inhibition is expected to eliminate the translational difference between cells with or without oncogenic RAS expression. We examined the effect of Torin1, a potent active site mTOR inhibitor (22). As expected, Torin1 treatment caused a rapid dephosphorylation of S6K1 and 4EBP1 (Fig. 1A). Although the overall translation was reduced in the presence of Torin1, as evidenced by the disassembly of polysomes, cells with RAS^G12V expression still exhibited a relatively higher P/M ratio than control cells (Fig. 1B). The sustained translational up-regulation caused by RAS^G12V expression was also seen in the Fluc reporter assay (Fig. 1C). As an independent validation, we examined the synthesis of endogenous proteins using puromycin labeling. The total levels of puromycin-labeled nascent chains were increased after 4-OHT treatment (Fig. 1D), confirming the enhanced synthesis of endogenous proteins during oncogenic translation. However, despite the dose-dependent inhibition by Torin1, translation nevertheless remained high in cells expressing RAS^G12V even in the presence of a high dose of Torin1 (100 nM) (Fig. 1D). The similar finding was obtained from [³⁵S] metabolic labeling (fig. S1D). These results suggest that additional mechanisms may contribute to RAS^G12V-mediated oncogenic translation.

Relaxed initiation pausing during oncogenic translation

To illustrate a holistic view of RAS^G12V-stimulated translation, we conducted RNA sequencing (RNA-seq), ribosome profiling (Ribo-seq), and quantitative translation initiation sequencing (QTI-seq) in TtH cells before and after 4-OHT treatment (Fig. 2A). Ribo-seq measures the density of ribosome footprints across the coding region (CDS) (23, 24). By normalizing to the mRNA abundance, the averaged ribosome occupancy often serves as a proxy for translation efficiency. However, RAS^G12V expression altered global protein synthesis with little mRNA specificity (Fig. 2B, middle, and Fig. 2C, right). Unlike Ribo-seq, QTI-seq captures initiating ribosomes in a quantitative manner and permits global mapping of translation initiation sites (TIS) and measurement of the ribosome density at individual start codons (25). To our surprise, most of start codons showed reduced or disappeared ribosome density upon RAS^G12V expression (Fig. 2C, left). The reduced ribosome density at the annotated TIS (aTIS) in the presence of 4-OHT could not be ascribed to inhibited translation initiation because Ribo-seq revealed no evident decrease of footprints in CDS (Fig. 2C, right). Certain transcripts even showed an increased CDS occupancy as exemplified by RPS19 (Fig. 2D) and EEF1G (fig. S2A). While the aTIS density of RPS19 dropped >10-fold after 4-OHT treatment, its CDS occupancy was increased by 2.5-fold. We reasoned that, during oncogenic translation, the initiating ribosome undergoes faster commitment to elongation at the start codon. Hence, QTI-seq is less likely to capture these initiating ribosomes. We are particularly attracted to this possibility, as it would suggest that a faster transition from initiation to elongation increases the translational output under oncogenic RAS signaling.

The ribosome density at the start codon reflects how long an initiating ribosome resides at the main start codon before the elongation commitment. However, it could be influenced by different initiation rates (i.e., the ribosome loading efficiency) and mRNA abundance. To factor out these variations, we divided the total reads at the annotated start codon (aTIS) by the mean background reads in CDS captured by QTI-seq. The resultant aTIS pausing index (PI) represents, on a relative scale, the dwell time of initiating ribosomes at individual start codons. The similar approach was originally designed to compute promoter-proximal pausing of RNA polymerase II (26). As shown in Fig. 2B, nearly half of the transcripts showed decreased aTIS PI (>2-fold) upon RAS^G12V expression. The lowered aTIS density under oncogenic RAS signaling is highly reproducible in biological replicates (fig. S2B). In comparison to Ribo-seq, a substantial amount of transcripts experiencing reduced aTIS PI after 4-OHT treatment exhibited increased ribosome occupancy in CDS (Fig. 2E). In line with the increased translation yield, immunoblotting showed elevated protein levels of RPS19 in cells after 4-OHT treatment (Fig. 2F). Therefore, oncogenic RAS signaling is accompanied with relaxed initiation pausing that results in an increased translational output.

Characterizing ribosome initiation pausing

The unexpected “lingering” of an 80S ribosome at the start codon before productive elongation suggests a previously unappreciated mechanism by which initiation pausing regulates translation efficiency, at least for some transcripts. It is conceivable that, under the physiological condition, initiation pausing limits the translational output by acting as a barrier for productive elongation (fig. S2C). Supporting this notion, a negative correlation (R = −0.22) was evident between the aTIS PI and the CDS ribosome occupancy (Fig. 3A). This salient feature is highly reproducible in biological replicates, in different cell types, and in mouse liver samples (fig. S2D). Therefore, initiation pausing represents an inherent feature for individual transcripts. We searched for common biological themes among transcripts with relatively high aTIS PI using gene ontology (GO) analysis. Genes involved in cell growth and proliferation, such as translation and cell cycle, were overrepresented among mRNAs exhibiting strong initiation pausing (fig. S2E). Pathway analysis also revealed the same functional category for transcripts with high aTIS PI values (fig. S2F).

An immediate outstanding question is what, if any, sequence features could contribute to varied ribosome pausing during the transition from initiation to elongation. We first examined the structural features of mRNAs with high or low aTIS PI. However, mRNA regions surrounding the start codon (−100 to +300 nt) exhibited comparable thermal dynamic stability between transcripts with or without initiation pausing (fig. S3A). This was further corroborated by the similar secondary structures determined experimentally via either parallel analysis of RNA structures (PARS) score measurement or dimethyl sulfate sequencing (DMS-seq) (27, 28) (fig. S3, B and C). We found that transcripts with shorter 5′UTR length tend to have higher aTIS PI values (Fig. 3B). Since longer 5′UTR can accommodate multiple scanning ribosomes, it is possible that a direct “push” by subsequent ribosomes helps release initiation pausing. However, the static mRNA sequence feature unlikely explains the altered aTIS PI in response to oncogenic RAS signaling. We hypothesize that additional regulatory factors must be at play to influence the behavior of scanning ribosomes at the start codon.

5′UTR methylation regulates initiation pausing

We recently reported that the scanning process of translation initiation is subjected to regulation by 5′UTR methylation in the form of m⁶A (29). To illustrate the m⁶A topology for mRNAs undergoing active translation, we collected ribosome-associated mRNAs followed by m⁶A sequencing (m⁶A-seq). On a transcriptome-wide scale, we observed an expected landscape of methylation with m⁶A sites highly enriched around the stop codon (fig. S3D). To our surprise, transcripts with high aTIS PI exhibited a marked enrichment of m⁶A in the vicinity of start codon with the peak density of >4-fold higher above the average (Fig. 3C). For example, in TtH cells, EEF1G bears methylation only near the start codon region and consequently displayed a prominent initiation pausing (Fig. 3D). By contrast, PARK7 is primarily methylated around the stop codon and shows minimal aTIS peak. Notably, in human embryonic kidney (HEK) 293 cells, PARK7 shows two m⁶A peaks in the vicinity of both start and stop codons. As a result, the initiation pausing becomes evident (fig. S3E). Therefore, despite the cell type–specific variations of m⁶A, there is a strong correlation between initiation pausing and start codon–associated m⁶A. Intriguingly, nearly half of the transcripts showed no identifiable TIS sites, although some of these messages were abundant and highly translated (Fig. 2C). We found that this group of mRNAs bears the lowest m⁶A levels in the vicinity of start codons (Fig. 3C). The distinct m⁶A enrichment near the start codon of transcripts with differential aTIS PI also holds true in other cell types including mouse liver tissues (fig. S3D). The antibody-based m⁶A-seq might have limitations (30). We queried recent m⁶A datasets using antibody-free mapping approach, such as deamination adjacent to RNA modification targets (DART-seq) (31), m⁶A-SEAL (32), and m⁶A-label-seq (33). Despite considerable variation, m⁶A around the start codon exhibits higher levels for mRNAs with aTIS peaks (fig. S3F). These results collectively suggest a functional correlation between start codon–associated m⁶A modification and initiation pausing (Fig. 3E).

To independently test this hypothesis, we created reporters by fusing the first 200 nt of EEFIG containing the full-length 5′UTR and the first 54 nt of CDS to the Fluc and introduced A→G mutation into the m⁶A site (A143). In transfected cells, the A143G mutant exhibited higher Fluc levels than the wild-type reporter (Fig. 3F). To examine the behavior of endogenous EEF1G, we used single-base elongation- and ligation-based qPCR amplification method (SELECT) to quantify the site-specific m⁶A levels. SELECT is a sensitive method based on the ability of m⁶A to hinder the single-base elongation activity of DNA polymerases and the nick ligation efficiency of ligases (34). In TtH cells, 4-OHT treatment resulted in reduced m⁶A levels at A143 (Fig. 3G). As a negative control, the neighboring A156 showed little difference. We then examined the protein levels of endogenous EEF1G and found an evident increase of EEF1G upon oncogenic RAS signaling (Fig. 3H). These results collectively support the notion that m⁶A modification in the start codon vicinity restricts the translational output.

Reduced mRNA methylation in cancers with RAS mutations

To investigate whether the relaxed initiation pausing during oncogenic translation is a result of reduced mRNA methylation, we purified poly(A)⁺ RNAs from cells with or without RAS^G12V expression followed by m⁶A dot blotting. In cells treated with 4-OHT, we found approximately 40% reduction of total mRNA m⁶A levels (Fig. 4A). This was not due to the side effect of 4-OHT because the same treatment did not alter m⁶A levels in vector control cells (fig. S4A). The lowered methylation upon oncogenic RAS signaling was further confirmed by a commercial m⁶A quantification kit (fig. S4B). We next measured m⁶A methyltransferase and demethylase activities in these cells. While 4-OHT treatment had little effect on the demethylase activity, we observed a 43% reduction of the methyltransferase activity in cells expressing ER:RAS^G12V (Fig. 4B). To probe the mechanism by which oncogenic RAS signaling reduces mRNA methylation, we examined the integrity of the m⁶A machinery via immunoprecipitation (IP). Despite the similar steady-state levels, METTL3 exhibited decreased association with Wilms’ tumor 1-associating protein (WTAP) in cells with oncogenic RAS signaling (Fig. 4C). Transcriptome-wide m⁶A-seq revealed a similar topology of m⁶A distribution before and after 4-OHT treatment, suggesting that oncogenic RAS signaling is associated with a global reduction of mRNA methylation (fig. S4C).

The role of METTL3 in tumorigenesis is complex as different phenotypes have been observed from different types of cancers. From The Cancer Genome Atlas database, recurrent missense mutations of METTL3 and METTL14 have been identified in several types of human cancers, in particular pancreatic and endometrial cancers (Fig. 4D). The most commonly mutated sites are within the conserved methyltransferase domain MT-A70. The mutant form of METTL14 shows reduced methyltransferase activity (35). This finding strongly suggests a tumor suppressor role for m⁶A methyltransferase, at least in certain types of cancers. Furthermore, Kaplan-Meier plotter revealed that high expression of METTL3 correlates with increased overall survival of patients with pancreatic cancer but not liver cancers (Fig. 4E). Since more than 90% of pancreatic cancers bear RAS mutations (19), the cancer type–specific clinical correlation between METTL3 and patient survival supports the notion that reduced mRNA methylation is positively linked to RAS-mediated oncogenic translation and subsequent cancer progression.

Reduced mRNA methylation promotes oncogenic translation

To validate the tumor suppressor role for METTL3 in cancers driven by oncogenic RAS signaling, we knocked down METTL3 from TtH cells bearing inducible RAS^G12V. With >80% depletion of METTL3, we observed approximately 40% reduction of mRNA m⁶A levels (fig. S4D). This manipulation resembles the physiological status of mRNA methylation under oncogenic RAS expression. Translation of endogenous proteins was enhanced in cells with METTL3 knockdown as revealed by puromycin labeling (Fig. 4F). This result suggests that METTL3 depletion mimics the increased translation by oncogenic RAS. Consistently, real-time luciferase assay recorded elevated Fluc levels (1.8-fold) in the absence of METTL3. Upon 4-OHT treatment, despite the evident increase of overall Fluc levels, METTL3 knockdown did not further boost Fluc translation (Fig. 4G). Therefore, RAS-triggered oncogenic translation encompasses the effect of METTL3 reduction.

To test whether lowered m⁶A levels on mRNAs directly leads to relaxation of initiation pausing, we conducted QTI-seq and Ribo-seq in TtH cells before and after METTL3 knockdown. A substantial number of transcripts (67.2%) showed reduced aTIS PI in cells lacking METTL3 (fig. S4E). Many of these transcripts had similar changes of aTIS PI in response to either METTL3 knockdown or RAS^G12V expression (fig. S4F), indicating that relaxed initiation pausing is a common feature shared by METTL3 depletion and oncogenic translation. In line with this notion, in cells lacking METTL3, the positive correlation of aTIS PI before and after 4-OHT treatment indicates that RAS^G12V expression had limited effects on initiation pausing (fig. S4G). We conclude that RAS-mediated oncogenic translation involves two distinct mechanisms: (i) activated eIF4F that enhances ribosome loading and (ii) relaxed initiation pausing that facilitates elongation commitment. The coordination between these two mechanisms maximizes translational flux during oncogenic translation (fig. S4H).

Increased mRNA methylation suppresses oncogenic translation

We next asked whether knocking down m⁶A demethylases would counteract oncogenic translation by restoring initiation pausing via increased mRNA methylation. In TtH cells without oncogenic RAS expression, knocking FTO had little effects on global protein synthesis measured by puromycin labeling (fig. S5A). This is likely due to the fact that transcripts with strong initiation pausing have limited room to increase methylation further. Cells lacking FTO showed no response to activated oncogenic RAS signaling after 4-OHT treatment as evidenced by comparable endogenous protein synthesis (Fig. 5A). Notably, these cells were still responsive to Torin1 treatment, indicating that the effect of FTO knockdown is independent of eIF4F. Consistently, real-time reporter assay revealed that the increased Fluc translation imposed by oncogenic RAS signaling was eliminated in the absence of FTO (Fig. 5B). To substantiate this finding further, we knocked down ALKBH5 and observed the similar translational consequence (fig. S5B). Therefore, elevated mRNA methylation effectively counteracts oncogenic translation induced by RAS signaling, presumably by canceling the relaxation of initiation pausing.

To assess whether increased mRNA methylation upon FTO knockdown is accompanied with restored initiation pausing, we took advantage of QTI-seq datasets obtained from HEK293 cells with either FTO knockdown or FTO overexpression (fig. S5, C and D). The aTIS ratio averaged across the entire transcriptome showed an approximate 1.5-fold increase in the absence of FTO, but nearly 40% decrease under FTO overexpression (Fig. 5C). When the methylated transcripts are considered, the fold change of aTIS PI at individual start codons is significantly increased in the absence of FTO compared to FTO overexpression (Wilcoxon signed-rank test, P < 2.2 × 10⁻¹⁶; Fig. 5D). For messages without m⁶A modification near the start codon, changing FTO levels minimally affected initiation pausing (fig. S5D). The Ribo-seq variation could be due to the pleiotropic effect of FTO in mRNA stability and translation. Together with the effect of METTL3 knockdown (Fig. 4), these results indicate that mRNA methylation controls the translational output by modulating initiation pausing.

Restored initiation pausing prevents tumorigenesis

With only a subset of transcripts exhibiting initiation pausing, it is interesting to find the global translational effect in response to oncogenic RAS signaling. This is likely due to the fact that initiation pausing primarily restrains the translational flux for genes involving cell growth, especially the translation machinery (fig. S2, D to F). It is conceivable that increased translational capacity leads to increased global protein synthesis, which is essential for active proliferation of cancer cells. We next asked whether the lack of this translational “brake” alone contributes to cellular transformation and tumorigenesis. As expected, TtH cells with activated RAS^G12V expression readily formed colonies in soft agar (Fig. 6A). Since FTO knockdown restores initiation pausing, we explored the possibility of reversing oncogenic translation by depleting FTO from cells expressing RAS^G12V. Notably, FTO depletion mediated marked, sustained inhibition of colony formation (Fig. 6A).

To extend this finding from TtH cells to cancer cell lines derived from human patients, we examined two pancreatic cancer cell lines harboring KRAS^G12V mutation (CFPAC-1) and KRAS^G12D mutation (AsPC-1), respectively. Similar to TtH cells bearing inducible RAS^G12V, FTO knockdown significantly inhibited the colony formation in soft agar for both CFPAC-1 and AsPC-1 (Fig. 6B). To examine the tumorigenesis in vivo, we conducted xenograft experiments using immunocompromised severe combined immunodeficient (SCID)–Beige mice. Consistent with the colony formation assay, FTO knockdown significantly suppressed tumor growth following subcutaneous injection (Fig. 6C). We also selected a pancreatic cancer cell line BxPC-3 that harbors the wild-type KRAS. Unlike CFPAC-1 and AsPC-1, BxPC-3 failed to form colonies in soft agar. Intriguingly, FTO depletion exhibited little effects on the tumor growth of BxPC-3 in vivo (Fig. 6C, right). To substantiate this finding further, we conducted orthotopic implantation of pancreatic cancer cell lines with or without FTO knockdown into SCID mice. The orthotopic xenograft models have similar tumor microenvironment as the original tumor and more closely resemble the natural tumorigenesis in human. In agreement with the heterotopic tumor models, silencing FTO significantly inhibited the growth of tumor cells bearing RAS mutation but not the wild type (Fig. 6D). These results not only confirmed the cancer type specificity in terms of oncogenic translation but also excluded the possibility that the efficacy of FTO knockdown on mutant RAS cancers was a result of nonspecific effects.

Given the relaxed initiation pausing in the absence of METTL3, we wondered whether silencing METTL3 would promote tumorigenesis. In both orthotopic and heterotopic tumor models, depleting METTL3 further increased the tumor burden of CFPAC-1 and AsPC-1 despite their progressive nature (Fig. 6D and fig. S6). Once again, little difference was observed for the tumorigenesis of BxPC-3 cells. This result is in line with the tumor suppressor role for METTL3 in mutant RAS cancers.

Therapeutic potential of FTO inhibitors

Given the antitumor effect of FTO depletion, at least in RAS mutant cancers, we were encouraged to explore therapeutic potential of small-molecule inhibitors of FTO. Meclofenamic acid (MA), an anti-inflammatory drug, was previously shown to act as a selective inhibitor of FTO (36). However, the inhibition efficiency of MA is not satisfied. More recently, a more specific FTO inhibitor FB23-2 was developed and showed therapeutic potential for AML (37). Following the dose regime for AML, we assessed the therapeutic effects of FB23-2 in vivo with xenografted pancreatic cancers in SCID-Beige mice. A week after xenotransplantation, FB23-2 (10 mg/kg) or vehicle control was injected intraperitoneally into the individual mice every other day. FB23-2 administration significantly delayed the tumor growth of CFPAC-1 and AsPC-1 (Fig. 7, A and B). The little effect on the tumorigenesis of BxPC-3 not only confirms the therapeutic specificity of FB23-2 (Fig. 7C) but also indicates that the in vivo side effect of FTO inhibitors is minimal.

Cell lines and reagents

Primary TtH cells were stably infected to generate polyclonal populations with retrovirus derived from pBabe Puro by subcloning into complementary DNA (cDNA) encoding fused ER:HRas^G12V. The primary TtH cells and ER:HRas^G12V TtH cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum (FBS). ER:HRas^G12V TtH cells were treated with 1 μM 4-OHT (Sigma-Aldrich) for 48 hours for oncogenic HRas^G12V expression; BxPC-3 (KRas^WT), and AsPC-1 (KRas^G12D) cells were grown in RPMI 1640 medium with 10% FBS; CFPAC-1 (KRas^G12V) cells were grown in Iscove’s modified Dulbecco’s medium with 10% FBS. Antibodies used in this study are listed as follows: anti–pan-Ras [Milipore MABS195, 1:1000 Western blotting (WB)], anti-Akt (Cell Signaling Technology, 9272; 1:1000 WB), anti–phospho (Ser⁴⁷³)–Akt (Cell Signaling Technology, 9271; 1:1000 WB), anti-p70 S6 kinase (Cell Signaling Technology, 9202; 1:1000 WB), anti–phospho-S6 (Thr³⁸⁹) kinase (Cell Signaling Technology, 9205; 1:1000 WB), anti–4E-BP1 (Cell Signaling Technology, 9452; 1:1000 WB), anti–phospho (Thr^37/46) 4E-BP1 (Cell Signaling Technology, 2855; 1:1000 WB), anti-METTL3 (Abnova, H00056339-B01P; 1:1000 WB), anti-METTL14 (Sigma-Aldrich, HPA038002; 1:1000 WB), anti-WTAP (Santa Cruz Biotechnology, sc-374280; 1:1000 WB), anti-FTO (PhosphoSolutions, 597-FTO; 1:1000 WB), anti-ALKBH5 (Proteintech, 16837-1-AP; 1:1000 WB), anti-YTHDF1 (Proteintech, 17479-1-AP; 1:1000 WB), anti-YTHDF2 (Proteintech, 24744-1-AP; 1:1000 WB), anti-m⁶A (Millipore, ABE572; 1:1000 WB), anti-puromycin (Developmental Studies Hybridoma Bank, PMY-2A4; 1:100 WB) and anti–β-actin (Sigma-Aldrich, A5441; 1:2000 WB).

Lentiviral short hairpin RNAs

Short hairpin RNA (shRNA) targeting sequences based on The RNAi Consortium at Broad Institute (https://portals.broadinstitute.org/gpp/public/) are listed below: METTL3 (human): 5′-ATTCTGTGACTATGGAACCA-3′; FTO (human): 5′-GCCAGTGAAAGGGTCTAATAT-3′; and scramble control sequence: 5′-AACAGTCGCGTTTGCGACTGG-3′. shRNA targeting sequences were cloned into DECIPHER pRSI9-U6-(sh)-UbiC-TagRFP-2A-Puro (Cellecta, CA). Lentiviral particles were packaged using Lenti-X 293 T cells (Clontech). Virus-containing supernatants were collected at 48 hours after transfection and filtered to eliminate cells. Cells were infected by the lentivirus for 48 hours before selection by puromycin (1 to 2 μg ml⁻¹).

Immunoblotting

Immunoblotting was conducted using a method described previously (44). In brief, cells were lysed on ice in tris-buffered saline (TBS) buffer [50 mM tris (pH 7.5), 150 mM NaCl, and 1 mM EDTA] containing a protease inhibitor cocktail tablet, 1% Triton X-100, and deoxyribonuclease (2 U ml⁻¹). After incubating on ice for 30 min, the lysates were heated for 5 min, 95°C in SDS–polyacrylamide gel electrophoresis (SDS-PAGE) sample buffer [50 mM tris (pH 6.8), 100 mM dithiothreitol (DTT), 2% SDS, 0.1% bromophenol blue, and 10% glycerol]. Proteins were separated on an SDS–polyacrylamide gel and transferred to Immobilon-P membranes (Millipore). Membranes were blocked for 1 hour in TBS containing 5% nonfat milk and 0.1% Tween 20, followed by incubation with primary antibodies overnight at 4°C. After incubation with horseradish peroxidase (HRP)–coupled secondary antibodies at room temperature for 1 hour, immunoblots were visualized using enhanced chemiluminescence (ECL^Plus, GE Healthcare).

Soft agar colony formation assay

Bottom agar layer was made with 0.8% noble agar (Sigma-Aldrich, A5431) in growth medium with 10% FBS in a six-well plate (1.5 ml per well). Cells were detached with trypsin-EDTA, washed twice with Dulbecco’s Phosphate Buffered Saline (DPBS) and counted, and 0.4% noble agar in growth medium mixed with single suspended cells (10⁴ per well) was placed on top of a solidified bottom agar. Cells were fed twice a week with 500-μl growth medium with 10% FBS. After 3 to 4 weeks of growing, the colonies were counted and imaged using a dissecting microscope.

Xenograft injections

Cells at 70 to 80% confluence were changed with fresh medium to remove the dead cells 3 to 4 hours before harvesting, then detached with trypsin-EDTA, and washed twice with DPBS. Cells were counted after collection, and 5 × 10⁶ cells were suspended in 50-μl DPBS in combination with 50-μl Matrigel (BD Biosciences, 356234). A total of 100 μl was injected subcutaneously into each side of the lower flank of the SCID-Beige mice (6 to 8 weeks of age) using insulin syringes. For testing the therapeutic potential of FTO inhibitor, 100-μl FB23-2 [dissolved in dimethyl sulfoxide (DMSO), 10 mg/kg mouse weight] or same volume of DMSO was injected intraperitoneally into mice every other day. Mice were examined every other day for evidence of tumor growth. Tumor diameters were measured with calipers, and the tumor volume in cubic millimeter was calculated by the formula: Volume = (width)² × length/2. Tumor tissues were excised after tumors reached a diameter of 10 mm (3 to 4 weeks). Mice were housed in a temperature- and humidity-controlled facility with a 12-hour light-dark cycle. The experimental protocols were approved by Cornell Institutional Animal Care and Use Committee (no. 2008-0167).

Orthotopic mouse model of pancreatic tumor growth

All animals were maintained in facilities approved by the Association for Assessment and Accreditation of Laboratory Animal Care International in accordance with the current regulations and standards of the U.S. Department of Agriculture and Department of Health and Human Services. The sample sizes of the animals were justified by statistical considerations and statistical power analyses. The animals were randomized to different experimental groups. For tumorigenic assay using an orthotopic mouse model, AsPC-1, CFPAC-1, or BxPC-3 cells (1 × 10⁶ cells per mouse, 5 mice per group) in 0.1 ml of Hanks’ balanced salt solution were injected into the pancreases of 7- to 8-week-old female athymic BALB/c nude mice (National Cancer Institute, Fredrick, MD), respectively. The mice were euthanized 4 to 5 weeks after tumor cell inoculation, and their tumors were harvested and weighed. The investigators were blinded to allocation during experiments and outcome assessment.

In vitro m⁶A methyltransferase activity assay

ER:HRas^G12V cells with or without 48-hour 4-OHT treatment (1 μM) were lysed in denaturing lysis buffer [20 mM tris-HCl (pH 7.5), 137 mM NaCl, 1% NP-40, and 2 mM EDTA] after three times wash with cold DPBS, then incubated at 4°C for 30 min with rotation, followed by centrifuging at 12,000 rpm for 15 min. The supernatant were collected and the concentration of total protein was measured using 660-nm protein assay reagent (Pierce). The in vitro methylation assay was performed in a 50-μl reaction mixture containing 400 nM RNA probe (commercially synthesized in vitro, Thermo Fisher Scientific), cell lysate with same amount of total protein, 20 mM tris (pH 7.5), 50 μM ZnCl₂, 1 mM DTT, RNaseOUT (0.2 U/μl), 1% glycerol, and 0.5 μCi [methyl-3H]AdoMet (PerkinElmer). The reaction was incubated at 30°C for 1 hour and then stopped by adding TRIzol reagent (Invitrogen). RNA after reaction was precipitated and purified using sodium acetate at −20°C for at least 2 hours. The precipitated RNA was subjected to radioactivity measurement using scintillation counting (Beckman). Levels of ³H-methyl–incorporated RNA are shown as disintegrations per minute.

Real-time quantitative polymerase chain reaction

Real-time quantitative polymerase chain reaction (qPCR) was conducted using a method described previously (44). In brief, total RNA was isolated with TRIzol reagent (Invitrogen), and reverse transcription was performed using the High Capacity cDNA Reverse Transcription Kit (Invitrogen). Real-time PCR analysis was conducted using Power SYBR Green PCR Master Mix (Applied Biosystems) and carried on a LightCycler 480 Real-Time PCR System (Roche Applied Science). All primers used in this study are listed in table S1.

Real-time luciferase assay

Real-time luciferase assay was conducted using a method described previously (44). In brief, cells grown in 35-mm dishes were transfected with 1-μg plasmids including luciferase gene (PGL3-luciferase or pcDNA3-CMV-luciferase) at 70 to 80% confluence. Luciferase substrate d-luciferin (1 mM, Regis Technologies) was added into the culture medium immediately after transfection. Luciferase activity was monitored continuously and recorded using Kronos Dio Luminometer (Atto).

Puromycin labeling

As described previously (55), cells at 70 to 80% confluence were treated with puromycin (10 μg ml⁻¹) for 10 min. After washing twice with ice-cold DPBS, cells were lysed with SDS-PAGE sample buffer, and proteins were separated on SDS-PAGE and transferred to Immobilon-P membranes. Membranes were blocked for 1 hour in TBS containing 5% nonfat milk and 0.1% Tween 20, followed by incubation with puromycin antibodies (1:100 dilution) overnight at 4°C. After incubation with HRP-conjugated anti-mouse immunoglobulin G (IgG) (1:5000 dilution) for 1 hour at room temperature, the membrane was visualized using enhanced chemiluminescence.

[³⁵S] methionine radiolabeling

ER:HRas^G12V cells with or without 4-OHT treatment (1 μM for 48 hours) and Torin1 treatment (10 nM or 100 nM for 2 hours) were washed with DPBS before incubation in methionine-free DMEM for 15 min. Cells were resuspended in labeling media {methionine-free DMEM supplemented with 10% FBS and methionine (30 μCi ml⁻¹ [³⁵S])} for 15 min. Labeling was stopped by ice-cold DMEM containing cycloheximide (100 μg ml⁻¹). Cells were washed with DPBS containing cycloheximide (100 μg ml⁻¹), lysed with SDS sample loading buffer. Cell lysates were heated and then resolved on a 10% tris-glycine SDS-PAGE gel, and radiography captured was by Typhoon 9400.

m⁶A dot blotting

Total cellular RNAs were isolated with TRIzol reagent, and mRNAs were purified using Dynabeads Oligo (dT)₂₅ (Thermo Fisher Scientific). Equal amounts of mRNA were spotted to a Hybond-N⁺ membrane (GE Healthcare), followed by ultraviolet (UV) cross-linking at UV 254 nm (0.12 J/cm²). After blocking in DPBS containing 5% nonfat milk and 0.1% Tween 20 for 1 hour, the membrane was incubated with anti-m⁶A antibody (1:1000 dilution) overnight at 4°C. The membrane was incubated with HRP-conjugated anti-rabbit IgG (1:5000 dilution) at room temperature for 1 hour and visualized using enhanced chemiluminescence.

Preparation of cell lysates for Ribo-seq and QTI-seq

As described previously (25), at least four 10-cm dishes of cells were harvested in 400-μl ice-cold polysome buffer [10 mM Hepes (pH 7.4), 100 mM KCl, and 5 mM MgCl₂] containing cycloheximide (100 μg ml⁻¹, for Ribo-seq) or lactimidomycin (5 μM, for QTI-seq). As described previously (25), cells were then disrupted by vortexing six times for 20 s using lysing matrix D (Fisher), followed by a 40-s interval each time on ice and then centrifuged at 12,000g, 4°C for 10 min. For QTI-seq, to dissociate noninitiating ribosomes, cell lysates were incubated in a solution containing 16 mM Hepes buffer (pH 7.4), 10 mM creatine phosphate, 0.1 mM spermidine, creatine phosphokinase (40 μg ml⁻¹), 0.8 mM adenosine triphosphate (ATP), and 25 μM puromycin at 35°C for 15 min. Samples were then subjected to sucrose gradient sedimentation.

Polysome profiling

As described previously (25), sucrose solutions were prepared in polysome buffer [10 mM Hepes (pH 7.4), 100 mM KCl, 5 mM MgCl₂, and 2% Triton X-100]. Fifteen to 45% (w/v) sucrose density gradients were freshly prepared in SW41 ultracentrifuge tubes (Backman) using a Gradient Master (BioComp Instruments). Five hundred microliters of supernatant from cell lysates prepared as described above was loaded onto sucrose gradients followed by centrifugation for 2.5 hours at 32,000 rpm, 4°C in a SW41 rotor. Separated samples were fractionated at 1.5 ml min⁻¹ through an automated fractionation system (Isco) that continually monitors values of optical density at 254 nm.

RNA-seq and m⁶A-seq

Both RNA-seq and m⁶A-seq procedures have been described previously (44). For RNA-seq, total RNA was first isolated using TRIzol reagent followed by fragmentation using freshly prepared RNA fragmentation buffer [10 mM tris-HCl (pH 7.0) and 10 mM ZnCl₂]. Five-microgram fragmented RNA was saved as input control. For m⁶A IP, 1-mg fragmented RNA was incubated with 15-μg anti-m⁶A antibody (Millipore ABE572) in 1× IP buffer [10 mM tris-HCl (pH 7.4), 150 mM NaCl, and 0.1% Igepal CA-630] for 2 hours at 4°C. The m⁶A-IP mixture was then incubated with protein A beads for additional 2 hours at 4°C on a rotating wheel. After washing three times with IP buffer, bound RNA was eluted using 100-μl elution buffer (6.7 mM N⁶-methyladenosine 5′-monophosphate sodium salt in 1× IP buffer), followed by ethanol precipitation. Precipitated RNA was used for cDNA library construction and high-throughput sequencing described below.

SELECT

All primers used in SELECT are listed in table S2. Five-microliter total RNA was mixed with 40 nM Up Primer, 40 nM Down Primer, and 5 μM deoxynucleotide triphosphate (dNTP) in 17-μl 1× CutSmart buffer [50 mM KAc, 20 mM tris-HAc, 10 mM MgAc₂, and bovine serum albumin (100 μg/ml; pH 7.9) at 25°C]. The RNA and primers were annealed by incubating at 90°C for 1 min, 80°C for 1 min, 70°C for 1 min, 60°C for 1 min, 50°C for 1 min, and then 40°C for 6 min. Subsequently, 3 μl of mixture containing 0.01 U of Bst 2.0 DNA polymerase, 0.5 U of SplintR ligase, and 10 nmol of ATP was added to the final volume of 20 μl. The final reaction mixture was incubated at 40°C for 20 min and then denatured at 80°C for 20 min. Ten-microliter qPCR reaction was set up with 2× Power SYBR Green PCR Master Mix (Applied Biosystems), 200 nM qPCR-F primer, 200 nM qPCR-R primer, and 2.5 μl of the final reaction mixture. qPCR was run at the following conditions: 95°C, 1 min; (95°C, 10 s; 60°C, 45 s) × 45 cycles.

cDNA library construction

As described previously (25), Escherichia coli ribonuclease I (Ambion, 750 U per 100 units of absorbance at 260 nm) was added into the pooled fractions from ribosome profiling and incubated at 4°C for 1 hour, and then total RNAs were extracted using TRIzol reagent. RNA extracts (Ribo-seq and QTI-seq), fragmented RNAs (RNA-seq), and m⁶A-IP elutes (m6A-seq) were dephosphorylated for 2 hours at 37°C in a 15-μl reaction (1× T4 polynucleotide kinase buffer, 10 U of SUPERase_In, and 20 U of T4 polynucleotide kinase). The products were separated on a 15% polyacrylamide tris-borate EDTA (TBE)–urea gel (Invitrogen) and visualized using SYBR Gold (Invitrogen). Selected regions in the gel corresponding to 40 to 60 nt (for RNA-seq and m⁶A-seq) or 25 to 35 nt (for Ribo-seq and QTI-seq) were excised. The gel slices were disrupted by using centrifugation through the holes at the bottom of the tube. RNA fragments were dissolved by soaking overnight in 400-μl RNA elution buffer [300 mM NaOAc (pH 5.2), 1 mM EDTA, and SUPERase_In (0.1 U ml⁻¹)]. The gel debris was removed using a Spin-X column (Corning), followed by ethanol precipitation. Purified RNA fragments were resuspended in nuclease-free water. Poly(A) tailing reaction was carried out for 45 min at 37°C [1× poly(A) polymerase buffer, 1 mM ATP, SUPERase_In (0.75 U μl⁻¹), and 3 U of E. coli poly(A) polymerase].

For reverse transcription, the oligos containing barcodes were listed in table S3. In brief, the tailed-RNA sample was mixed with 0.5 mM dNTP and 2.5 mM synthesized primer and incubated at 65°C for 5 min, followed by incubation on ice for 5 min. The reaction mix was then added with 20 mM tris (pH 8.4), 50 mM KCl, 5 mM MgCl₂, 10 mM DTT, 40 U of RNaseOUT, and 200 U of SuperScript III. Reverse transcription reaction was performed according to the manufacturer’s instruction. Reverse transcription products were separated on a 10% polyacrylamide TBE-urea gel as described earlier. The extended first-strand product band was expected to be approximately 100 nt, and the corresponding region was excised. The cDNA was recovered by using DNA gel elution buffer (300 mM NaCl and 1 mM EDTA). First-strand cDNA was circularized in 20 μl of reaction containing 1× CircLigase buffer, 2.5 mM MnCl₂, 1 M betaine, and 100 U of CircLigase II (Epicentre). Circularization was performed at 60°C for 1 hour, and the reaction was heat-inactivated at 80°C for 10 min and then was precipitated by ethanol.

Deep sequencing

As described previously (25), circular template was amplified by PCR by using the Phusion high-fidelity (HF) enzyme (New England Biolabs) according to the manufacturer’s instructions. The oligonucleotide primers listed in table S3 were used to create DNA suitable for sequencing. The PCR contains 1× HF buffer, 0.2 mM dNTP, 0.5 μM oligonucleotide primers, and 0.5 U of Phusion polymerase. PCR was carried out with an initial 30-s denaturation at 98°C, followed by 12 cycles of 10-s denaturation at 98°C, 20-s annealing at 60°C, and 10-s extension at 72°C. PCR products were separated on a nondenaturing 8% polyacrylamide TBE gel as described earlier. Expected DNA at 120 base pairs was excised and recovered as described earlier. After quantification by Agilent BioAnalyzer DNA 1000 assay, equal amounts of barcoded samples were pooled into one sample. Approximately 5 pM mixed DNA samples were used for cluster generation followed by sequencing by using sequencing primer 5′-CGACAGGTTCAGAGTTCTACAGTCCGACGATC-3′ (Illumina HiSeq).

Alignment of sequencing reads

The 3′ adapters and low-quality bases were trimmed by Cutadapt as described previously (48). The trimmed reads with length <15 nt were excluded. For Ribo-seq and QTI-seq, reads without adapters were discarded. The remaining reads were mapped to the human transcriptome using Bowtie with parameters: -a –best -m1 –strata. The annotation file downloaded from ENSEMBL database (GRCh38) was used to construct the transcriptome index file. For each gene, the transcript with longest CDS was selected. In the case of equal CDS length, the longest transcript was used. For read alignment, two mismatches were permitted. To avoid ambiguity, the reads mapped to multiple positions were disregarded for further analyses. The 13th, 14th, and 15th positions (12-nt offset from the 5′ end) of the resulting uniquely mapped read was defined as the ribosome P-site position. For Ribo-seq, the reads mapped to CDS were used to calculate the RPKM (reads per kilobase of transcript per million mapped reads) values for translation levels. For RNA-seq, the reads mapped to entire transcript were used to calculate RPKM. Transcripts with RPKM <1 were excluded. Translation efficiency was defined as the ratio of RPKM of Ribo-seq over RPKM of RNA-seq.

Calculation of PI at the start codon

For each transcript, QTI-seq reads with the P-site in a window centering the annotated start codon (−5, +5 nt) were used to represent the abundance of translation initiation signal at the start codon, which we termed aTIS density. PI was defined as the ratio of mean aTIS density in the window over mean read captured by QTI-seq in the remaining CDS region. To reduce the noises from insufficiently translating mRNAs, which usually result in a large variation on PI estimation, we only considered the mRNAs that meet the following criteria: (i) RPKM estimated by Ribo-seq was higher than 1. (ii) aTIS density in the window either in wild-type or 4-OHT treatment sample was higher than 3. For the transcripts with 0 value of aTIS density, a pseudo-number 1 was used. The mean value by averaging PIs of two replicates was used in analyses. The transcripts with top (bottom) 20% of aTIS PI were defined as high (low) aTIS PI transcripts.

m⁶A sites identification and aggregation plot

m⁶A sites were identified using the method described previously (48). m⁶A coverages at individual sites were normalized by the mean coverage of the transcript. To avoid possible biases caused by insufficient coverage, analyses were limited to the transcripts with at least 30 reads per 1000 nt. Then, all transcripts were divided into three segments: 5′UTR, CDS, and 3′UTR. Each segment was then divided into 100 bins with equal length. All transcripts were aggregated by averaging the normalized m⁶A coverages across all available transcripts embracing the same bins. Besides aggregation plot in normalized mRNA regions, we also aggregated m⁶A coverages around the start codon (−300, +400), without normalizing the length of mRNA regions.

RNA secondary structure analysis

A sliding window of 30 nt with a step of 3 nt was used to calculate RNA minimum fold free energy (MFE) along transcript. For each window, MFE was calculated by ViennaRNA (56), using default parameters. For aggregation plot, a mean MFE in each window was calculated by averaging MFE values of the windows at the same position. DMS-seq data were used to investigate mRNA folding in vivo. DMS signal and Gini score were calculated using the methods in the work of Rouskin et al. (57). PARS data were used to analyze mRNA secondary structure in vitro (58).

Gene GO analysis

DAVID was used for GO enrichment analysis. The genes with fold change higher than 2 or lower than 0.5 were used for GO analysis. The expressed genes (FPKM >1) were used as the background gene set. Pathway analysis was performed by ClueGO (59) based on the annotation of Reactome (60).

Code availability

All statistical analyses were performed by R. The heatmap plots were made by heatmap.2 in R. All procedures but sequencing mapping were completed using custom Perl scripts. These scripts are available upon request.