Upregulation of 5′-terminal oligopyrimidine mRNA translation upon loss of the ARF tumor suppressor

Tumor cells require nominal increases in protein synthesis in order to maintain high proliferation rates. As such, tumor cells must acquire enhanced ribosome production. How the numerous mutations in tumor cells ultimately achieve this aberrant production is largely unknown. The gene encoding ARF is the most commonly deleted gene in human cancer. ARF plays a significant role in regulating ribosomal RNA synthesis and processing, ribosome export into the cytoplasm, and global protein synthesis. Utilizing ribosome profiling, we show that ARF is a major suppressor of 5′-terminal oligopyrimidine mRNA translation. Genes with increased translational efficiency following loss of ARF include many ribosomal proteins and translation factors. Knockout of p53 largely phenocopies ARF loss, with increased protein synthesis and expression of 5′-TOP encoded proteins. The 5′-TOP regulators eIF4G1 and LARP1 are upregulated in Arf- and p53-null cells.

Increased translation efficiency of 5′-TOP mRNAs upon ARF depletion. Previously ARF has been identified to not only regulate translation globally but to specifically regulate the translation of a subset of mRNAs 10,17 . To map global translational regulation by ARF we employed ribosome profiling to measure changes in translation following ARF depletion. We performed ribosome profiling and RNAseq for WT MEFs transduced with either shSCR or shARF in addition to Arf −/− MEFs transduced with shSCR, Fig. 2a. The shRNA used to knockdown ARF has been described previously and has no effect on INK4A expression 16 . Replicateto-replicate comparisons show strong correlations between RNAseq or ribosome profiling replicates (Supplemental Fig. 2b). We used DESeq2 to identify differential translation efficiency (TE) as described previously 32 . Most mRNAs showed no significant difference in TE upon loss of ARF, Fig. 2b,c. However, a subset of mRNAs showed a slight but significant increase in translation following ARF depletion by either knockdown with shARF or knockout of Arf. There is a significant correlation between fold change of TE following ARF knockdown and knockout, Fig. 2d. Gene ontology (GO) analysis of the genes with increased TE following knockout of Arf identified multiple GO terms associated with translation, Fig. 2e. A closer look at the genes with the biggest increase in TE revealed many ribosomal proteins and translation factors. It has been well established that most ribosomal proteins and many translation factors contain a 5′-TOP motif 21,22 . Analysis of our ribosome profiling results revealed that many of the mRNAs with increased TE are known to contain a 5′-TOP motif, Fig. 2f. 21 . Furthermore, mRNAs known to contain a 5′-TOP motif show a general increase in TE following knockout of Arf, Fig. 2g.
To validate our ribosome profiling findings, we assessed the mRNA and protein abundance of several 5′-TOP genes in MEFs following depletion of ARF. By immunoblot we observed increased protein abundance for the 5′-TOP genes PABPC1, TPT1, EEF2, RPL23A and RPL22, Fig. 3a,b and Supplemental Figs. 3, 4. The increased protein abundance measured by immunoblot was generally consistent with the increased TE measured by ribosome profiling, Fig. 3d. The antibody used for PABPC1 was raised against a peptide from PABPC1, but may cross-react with other PABPCs, as such we have labeled this blot and others 'PABP' . Because the expression of PABPC1 is much higher than that of the other PABPCs (Supplemental Dataset) it likely that most of the immunoblot signal is coming from PABPC1. The expression of PABPC1, TPT1, EEF2, RPL23A and RPL22 www.nature.com/scientificreports/ was unchanged at the mRNA level, Fig. 3c. Combined with our ribosome profiling results these data support increased translation of 5′-TOP mRNAs in ARF-null MEFs. To further validate these findings, we utilized a previously described 5′-TOP reporter 26 . The reporter contains the promoter and 5′-UTR of the 5′-TOP gene EEF2, Fig. 3e. The 5′-TOP reporter and a control that has a mutated 5′-TOP motif were transfected in parallel into WT or Arf −/− MEFs along with a firefly luciferase control. We observed increased luciferase activity for our 5′-TOP reporter in Arf −/− MEFs, Fig. 3f. This contrasts with the 5′-TOP-mutant reporter which had consistent activity between WT and Arf −/− MEFs. Analysis of the mRNA abundance of the 5′-TOP reporter revealed no change in mRNA expression in Arf −/− MEFs relative to WT, Fig. 3g. This indicates that the increased luciferase activity observed for the 5′-TOP reporter is caused by increased translation efficiency-not transcription or RNA stability. To provide further evidence we generated a second reporter system; the promoter and 5′UTR from Rpl23a, a 5′-TOP containing gene that showed increased TE upon ARF depletion in our ribosome profiling data, was cloned upstream of firefly luciferase, Supplemental Fig. 4b. The RPL23A reporter showed increased luciferase activity in ARF knockdown and knockout MEFs relative to WT, Supplemental Fig. 4c,d. Together these data confirm increased translation efficiency of 5′-TOP containing mRNAs following loss of ARF.  www.nature.com/scientificreports/ Increased translation efficiency of 5′-TOP mRNAs upon ARF loss is dependent on p53. ARF is a well-known activator of p53 4,5 . To test whether p53 is required for ARF-dependent regulation of 5′-TOP mRNA translation we performed some of the experiments described above in p53 −/− MEFs. Knockdown of ARF in p53 −/− MEFs had no effect on protein abundance of the evaluated 5′-TOP genes, Fig. 4a,b and Supplemental Fig. 5b. Furthermore, knockdown of ARF had no effect on the translation of the EEF2 5′-TOP reporter or the RPL23A reporter, Fig. 4c and Supplemental Fig. 5c. There was no difference in polysome abundance between shSCR and shARF transduced p53 −/− MEFs indicating no changes in total protein synthesis, Fig. 4d and Supplemental Fig. 5a. However, there were some changes in the abundance 40S and 60S ribosomal subunits based on variations in peak heights. This observation mirrored that seen following knockout of Arf in WT MEFs. These findings indicate that p53 is required for ARF-dependent translational regulation of 5′-TOP mRNAs.
Loss of ARF modestly affects the expression or activity of 5′-TOP regulators. The mTOR pathway is known to regulate 5′-TOP mRNAs through the phosphorylation of LAPR1, which changes LARP1 from a repressor of 5′-TOP mRNA translation to an enhancer 23,24,27,28 . LARP1 is phosphorylated by mTORC1 and the mTORC1 substrate S6-kinase 24 . Immunoblot analysis of mTORC1 pathway activation in WT and ARF-null MEFs revealed a slight, though inconsistent and non-significant increase in mTORC1 activity, Fig. 5a,b and Supplemental Fig. 7a. S6-kinase showed the largest increase, though statistically insignificant, in phosphorylation following ARF depletion, indicating LARP1 may be phosphorylated in those cells. However, inhibition of . Luciferase activity was normalized to firefly luciferase transfection control and set relative to WT. Mean ± SD, n = 3. (g) qPCR shows no increase in mRNA expression of the 5′-TOP reporter. Normalized to firefly luciferase transfection control. Mean ± SD, n = 3. *, p-value < 0.05, ** p-value < 0.01; two-tailed t-test with Bonferoni correction. www.nature.com/scientificreports/ S6-kinase had no effect on the expression of 5′-TOP encoded proteins, Supplementary Fig. 9. This result is consistent with previous studies that have shown S6-kinase to be dispensable for regulation of 5′-TOP mRNA translation 33 . Interestingly, we observed increased phosphorylation of S6-kinase, but no increased phosphorylation of other mTORC1 markers following ARF-knockdown in p53 −/− MEFs, Supplemental Fig. 6a,b. Treatment with the mTOR inhibitor rapamycin was employed to assess the importance of mTORC1 pathway activity in Arf −/− MEFs. Consistent with previous reports, cells treated with rapamycin show reduced mTOR autophosphorylation and reduced expression of the 5′-TOP encoded proteins RPL22, PABPC1 and TPT1, Fig. 5c,d 34,35 . These results show that mTORC1 activity is required for maximal expression of 5′-TOP mRNAs in ARF-null MEFs. mTORC1 has been shown to enhance 5′-TOP translation through phosphorylation of 4EBP1 25 . There was a slight but insignificant increase in p-4EBP1 levels upon ARF knockdown, but no change upon knockout of ARF, Fig. 5a,b and Supplemental Fig. 7a. However, there was also an increase in total 4EBP1 levels upon ARF knockdown, Fig. 5a,b and Supplemental Fig. 7b. As unphosphorylated 4EBP1 is a translational repressor, the increased expression of 4EBP1 upon ARF knockdown is unlikely to contribute to increased 5′-TOP translation. The expression of eIF4E was unchanged upon knockout or knockdown of ARF, Fig. 5a Supplementary Fig. 6c,d. We attempted to overexpress eIF4G1 in WT MEFs via lentiviral transduction, but due to the size of the coding sequence the transduction efficiency was extremely low (data not shown). As such, we took the reverse approach and knocked-down eIF4G1 in Arf −/− MEFs. Knockdown of eIF4G1 caused a slight reduction of protein expression for several 5′-TOP genes  www.nature.com/scientificreports/ at 4 days post transduction, Fig. 5g,h, and Supplemental Fig. 8b. These data suggest that increased eIF4G1 expression is unlikely to be driving increased translation of all 5′TOP mRNAs, but may have a role in regulating specific mRNAs like TPT1. LARP1 is a known regulator of 5′-TOP mRNA translation 24,27,28 . Immunoblot analysis of WT and ARF-null MEFs showed increase protein abundance for LARP1, Fig. 5i,j and Supplemental Fig. 8c. There was no increase in LARP1 mRNA expression in ARF-null MEFs relative to WT MEFs, Fig. 5j. Knockdown of ARF in p53 −/− MEFs had no effect on LARP1 expression, Supplementary Fig. 5e,f. Given the dual role of LARP1 as both an enhancer and repressor of 5′-TOP mRNA translation 24 it is possible that elevated expression of LARP1 in this context could drive 5′-TOP mRNA translation. However, knockdown of LARP1 had little to no effect on the expression of several 5′-TOP genes in Arf −/− MEFs, Fig. 5k,l. Only one of the two shRNAs targeting LARP1 had a significant effect on expression but for only one of the 5′-TOP genes assessed -TPT1. These findings indicate that elevated LARP1 expression is not likely to be the driving force behind increased 5′-TOP mRNA translation following loss of ARF, though it may have a role in the regulation of specific 5′-TOP mRNAs.

Loss of p53 causes upregulation of 5′-TOP mRNAs. Having observed a role for ARF in regulation
5′-TOP mRNA translation, we next sought to determine if p53, which is tightly coupled to ARF expression, has a similar role. Polysome profiling revealed increased translation in p53 −/− MEFs, as indicated by higher polysome peaks, Fig. 6a and Supplemental Fig. 10a-b. We used immunoblot analysis and qPCR to determine the protein and mRNA expression of some of the 5′-TOP genes studied above. We observed an increase in protein expression for three of the 5′-TOP genes (PABPC1, EEF2 and TPT1) in p53 −/− MEFs, there was no increase in RNA expression, Fig. 6b,e,f, and Supplemental Figs. 10 and 11. This is consistent with our findings from ARFnull MEFs. Analysis of the luciferase reporters described above showed increased activity for the EEF2 5′-TOP reporter and the RPL23A reporter in p53 −/− MEFs, Fig. 6c

Discussion
Together these data fit well with what is known about the functions of ARF in ribosome biogenesis. ARF is known to repress ribosome biogenesis by affecting rRNA transcription, processing and nuclear export [7][8][9][10][11][12][13][14][15][16] . Changes in 5′-TOP mRNA translation upon loss of Arf would help to balance concomitant changes in rRNA production by providing increases in the necessary ribosome subunit proteins. Increased translation of 5′-TOP mRNAs following loss of the ARF tumor suppressor provides a missing mechanism for the increased rates of translation and ribosome biogenesis seen in many human cancers 36 .
Previously, it has been shown that loss of ARF caused enhanced translation of two mRNAs: DROSHA and VEGFA 10,17 . We did not observe any significantly increased translation efficiency for either DROSHA or VEGFA following knockdown or knockout of ARF. One possible explanation for the discrepancy regarding VEGFA could be the cell line used. The VEGFA experiments were performed in established NIH3T3 and Arf/p53 DKO MEFs with overexpression of ARF, not knockdown/knockout 17 .
While many of the functions of ARF in regulating ribosome biogenesis are p53-independent, we show here that the regulation of 5′-TOP mRNAs by ARF requires p53 expression. Furthermore, we show that loss of p53 partially phenocopies the effects of ARF depletion on 5′-TOP mRNA translation. These data suggest that canonical ARF-MDM2-p53 axis plays an important role in 5′-TOP mRNA translation. Interestingly, free ribosomal proteins, such as RPL11, are known to activate p53 through binding MDM2 18 . Elevated expression of 5′-TOP containing ribosomal proteins following loss of p53 would complete a feedback loop. Recently it has been shown that the cyclin-dependent kinase CDK1 can phosphorylate LARP1 and 4EBP1, and has a general role in enhancing translation 37 . As p53 represses CDK1 activity through transcriptional upregulation of p21 (Cdkn1a), the ability p-mTOR = p-S2484, p-S6K = p-T389 (p70), p-S6 = p-S240/S244, p-4EBP1 = p-T37/T46. Total protein was used for normalization. Mean ± SD, n = 4-5 (c) Immunoblot analysis showing expression of 5′-TOP mRNA encoded proteins following treatment with rapamycin (Selleck Chemicals) and quantified in (d). Arf −/− MEFs were treated with 5 nM rapamycin for 4 days. Total protein was used for normalization. Mean ± SD, n = 3. (e) Immunoblot analysis showing increased expression of eIF4G1 in ARF-null MEFs. f Quantitation of blot in (e) and qPCR for eIF4G1, total protein was used for normalization of immunoblot, Mean ± SD, n = 4. (g) Immunoblot analysis showing the effect of eIF4G1 knockdown on 5′-TOP gene expression. h Quantitation of blot in (g), total protein was used for normalization of immunoblot, Mean ± SD, n = 3. (i) Immunoblot analysis showing increased expression of LARP1 in ARF-null MEFs. (j) Quantitation of blot in i and qPCR for LARP1, total protein was used for normalization of immunoblot, Mean ± SD, n = 4. (k) Immunoblot of Arf −/− MEFs following knockdown of LARP1 with two different shRNAs. (l) Quantitiation of blot in (k), total protein was used for normalization of immunoblot, Mean ± SD, n = 3. *, **, *** p-value < 0.05, < 0.01, < 0.001; two-tailed t-test with Bonferoni correction. www.nature.com/scientificreports/ of CDK1 to phosphorylate LARP1 and 4EBP1 could contribute to increased 5′-TOP translation following loss of p53 or ARF-though we did not see evidence of increased 4EBP1 phosphorylation (relative to total 4EBP1) following loss of p53 or ARF. In support of this hypothesis, we observed decreased RNA expression of p21 upon knockout of Arf (Supplementary Dataset). The connection between ARF, p53, p21, CDK1 and regulation of 5′-TOP translation deserves further investigation.
It is unclear what factors drive increased 5′-TOP mRNA translation following loss of ARF. We observed increased expression and activity for several regulators of 5′-TOP mRNA translation. Depletion of ARF had little effect on mTORC1 activity, though there was a slight and non-significant increase in phosphorylation of some pathway components. By inhibiting mTOR with rapamycin we showed that maximal expression of 5′-TOP encoded proteins requires mTOR activity. mTORC1 enhances 5′-TOP mRNA translation through phosphorylation of LARP1. LARP1 acts to balance 5′-TOP mRNA translation through inhibition of translation when dephosphorylated and enhancing translation when phosphorylated 24 . Dephosphorylated LARP1 binds the 5′-TOP motif and blocks binding of eIF4E, thus blocking assembly of the eIF4F complex 27 . A component of the eIF4F complex, eIF4G1 has also been shown to be required for 5′-TOP translation 26 . We observed increased expression of both eIF4G1 and LARP1 following depletion of ARF. Because of technical limitations described above, it is difficult to assess the direct role of eIF4G1 in regulating endogenous 5′-TOP mRNAs via overexpression in this system. Knockdown of eIF4G1 caused a slight and somewhat inconsistent reduction in 5′-TOP gene expression at the protein level. Knockdown of LARP1 in Arf −/− MEFs had only modest effects on 5′-TOP mRNA expression, with one shRNA causing a significant reduction in the expression of the 5′-TOP mRNA encoded protein TPT1. This suggests that at least for TPT1, LARP1 may modestly enhance translation. Because phosphorylated LARP1 is known to enhance 5′-TOP mRNA translation, this result is consistent with active mTORC1 in Arf −/− MEFs. Increased 5′-TOP mRNA translation could be driven by the combinatorial effect of small increases in eIF4G1 and LARP1 expression, as well as increased mTORC1 activity. A more intriguing possibility is that an unknown factor-downstream of ARF-controls 5′-TOP mRNA translation. Further work is needed to elucidate this pathway. Because many 5′-TOP encoded proteins are essential components of the translation machinery, it is important to understand how loss of ARF can drive 5′-TOP mRNA translation and global protein synthesis. A complete understanding of this pathway may lead to future therapeutic interventions aimed at reducing the protein synthesis capacity of ARF-null tumors.
Regulation of 5′-TOP mRNA translation independent of mTORC1 has been recently observed in mouse liver. Knockdown of the translation elongation factor eEF2 increased the translation efficiency of 5′-TOP containing mRNAs, without any change in mTORC1 activity 38 . The findings described here and those from knockdown of eEF2 in mouse liver suggest that there is another pathway regulating 5′-TOP mRNA translation.
One potential factor driving increased 5′-TOP translation in p53 −/− MEFs could be the cap-binding protein eIF4E. We observed a slight increase in eIF4E expression upon p53 knockout, however the same was not true following ARF depletion. The affinity of eIF4E for the mRNA cap can be influenced by cap-proximal sequences. In fact, 5′-TOP motif containing sequences are sensitive to reductions in eIF4E levels and eIF4E has a lower affinity for RNAs containing 5′-TOP motifs 39 . Future studies are needed to explore the connection between p53, eIF4E and 5′-TOP mRNA translation.

Materials and methods
For measurement of translation rates by puromycin incorporation, MEFs were treated with 10 µM puromycin at room temperature for the times indicated. As a control, cells were also treated with 100 µg/mL cycloheximide for 5 min at 37 °C prior to addition of puromycin. Immediately following treatment, the cells were washed with 1 × PBS containing 100 µg/mL cycloheximide to inhibit further puromycin labeling. The cells were harvested and lysed in RIPA buffer with 1 × HALT (Pierce) and 100 µg/mL cycloheximide. Immunoblot was performed as described below with anti-puromycin antibody (Developmental Studies Hybridoma Bank, University of Iowa).
Plasmid construction. The 5′-TOP reporter for mouse Rpl23a was made by PCR amplification of a region encompassing 500 bp upstream of the annotated transcription start site to the start codon using primers RPL23A Promoter Forward: 5′-GTA CCT CGA GGA GCT ATA AAG GGA AAC CCT GTCTC -3′ and RPL23A 5′UTR Reverse: 5′-GTA CCC ATG GTG CTT GGC TGA AAA GGA TGG CCC-3′. The PCR product was digested with XhoI and NcoI and ligated into pGL3-control (Promega). The resulting plasmid, pGL3-RPL23A-FF, was confirmed by Sanger sequencing.
Transfection of MEFs and luciferase assay. The EEF2 (pIS1-Eef25UTR-renilla) and EEF2-mutant (pIS1-Eef25UTR-TOPmut-renilla) reporters were purchased from Addgene and have been described previously 26 . The day before transfection 1 × 10 5 cells were plated per well in a six well dish. For transfection of the EEF2 reporters, the cells were transfected with Fugene6 (Promega) and 1 µg each of either pIS1-Eef25UTRrenilla or pIS1-Eef25UTR-TOPmut-renilla and pGL3-Control. After 24 h the cells were washed briefly with 1 × PBS prior to measurement of luciferase activity by Dual Luciferase Reporter Assay (Promega). For transfection of the RPL23A or SV40 reporters, the cells were transfected with Fugene6 and 1 µg each of either pGL3control (firefly luciferase with SV40 promoter) or pGL3-RPL23A and pGL3-Renilla-deltaFF. After 24 h the cells were washed briefly with 1 × PBS prior to measurement of luciferase activity by Dual Luciferase Reporter Assay.
Ribosome profiling and RNAseq. For RNAseq, total RNA was isolated from the appropriate cells using the Direct-zol RNA MiniPrep (Zymo) with Trizol (Invitrogen) for initial cell lysis. Contaminating genomic DNA was removed by treatment with Turbo DNA-free Kit per the manufacturers protocol (Invitrogen). The RNA concentration was determined by Qubit RNA BR Assay (Invitrogen) and ribosomal RNA was removed using the RiboZero Gold rRNA Removal Kit (Illumina) per the manufacturers protocol.
The ribo-depleted RNA (10 µL) was then fragmented by adding one volume of 2 × Fragmentation Buffer (0.5 M EDTA, 0.1 M Na 2 CO 3 , 0.1 M NaHCO 3 ) and incubating at 95 °C for 20 min. Fragmentation was inhibited by the addition of 280 µL of Stop Solution (3 M NaOAc pH 5.5, 15 mg/mL GlycoBlue (Invitrogen)). The RNA was the precipitated by the addition of ethanol. The RNA was resuspended in 10 mM Tris pH 8.0 and resolved on a 15% acrylamide Urea-TBE gel (Bio-Rad). A region containing fragments between 17-34 nt (the same size as those isolated for ribosome profiling 46 ) was excised and extracted overnight in RNA Extraction buffer (0.3 M NaOAc pH 5.5, 1 mM EDTA and 0.25% Sodium dodecylsulfate). The extracted RNA was precipitated by ethanol precipitation and subsequently used for sequencing library preparation as described below. www.nature.com/scientificreports/ Ribosome footprinting was performed as described previously 46 , with the use of RNase T1 (Thermo Scientific) instead of RNase I for the footprinting step, see Supplemental Fig. 2a. The lysates were treated with 700 units of RNase T1 for 1 h at room temperature with gentle mixing. For both ribosome profiling and RNAseq, a small amount of yeast lysate was spiked-in to the lysis buffer for later normalization. We chose not to use the yeast spike-in for normalization as the replicate-to-replicate comparisons were weak (data not shown).
Following purification of the RNA footprints, sequencing library production was carried out for both the fragmented total RNA and ribosome footprints using the previously described protocol 46 . The RNAseq and ribosome profiling libraries were multiplexed and sequenced on a HiSeq 3000 (Illumina) by the Washington University Genome Technology Access Center.
Data and code availability. Raw sequencing reads are available at https ://www.ncbi.nlm.nih.gov/geo (GSE156749). Scripts used for analysis of differential expression and TE are available at GitHub (https ://githu b.com/cottr ellka /ARF_5TOP). The DESeq2 results (see below) for Fold Change of TE and RNA are available in the Supplemental Dataset.
Analysis of sequencing data. The sequences for the linkers and reverse transcriptase primer used in this study are available in the Supplementary Information. The ribosome profiling and RNAseq reads were processed to remove short reads, reads lacking the sequencing adapter and adapter only reads using fastx_clipper 47 (fastx_clipper -a AGA TCG GAA GAG CAC ACG TCT GAA -c -n -v -l 13). Reads were split by barcodes introduced by the linkers using fastx_barcode_spliter (example: -bcfile Barcodes.txt -prefix/Results/-suffix_ribo. fastq-eol < riboseq_trimmed_umi_trimmed.fastq). Barcodes and other sequences introduced during library production were removed by umi-tools 48 (to remove UMIs: umi_tools extract -bc-pattern = NNNNNXXXXX -3prime) and cutadapt (to remove barcodes: cutadapt -cut -5) 49 . Sequencing reads mapping to rRNA or ncR-NAs were removed using Bowtie2 50 (example: bowtie2 rrna_seqs results.fastq -un results_norrna.fastq -S nor-rna_results). The processed reads were then aligned to the mouse genome (UCSC, mm10) using TopHat 51 (example: tophat -no-novel-juncs -output-dir dir/ -GTF genes.gtf mm10/Sequence/Bowtie2Index/genome results_norrna_notrna_noncrna.fastq). Read counts of those aligning to the coding sequence of each gene were then determined by HTSeq 52 (example: htseq-count -t CDS accepted_hits.sam genes.gtf > htseq_results. txt). Differential TE was determined using DESeq2 in which the design included an interaction term for assay and condition (~ assay + condition + assay:condition), this allows for determining differential TE via a ratio of ratios (treatment_RiboSeq/control_Riboseq)/(treatment_RNAseq/control_RNAseq) 32,53 (see GitHub repository above for scripts used).
For classification of 5′-TOP genes, we used a previously described list of known 5′-TOP genes which is available in Supplementary Information 21 .
Gene ontology (GO) analysis was performed using the PANTHER Over Representation test via web browser http://geneo ntolo gy.org/ [54][55][56] . For this analysis, the annotation file used was GO version: 2020-03-23. Significance was determined by Fisher Exact test and multiple-testing correction was performed by FDR. Genes with fold change of TE > 0 and FDR adjusted p-value of < 0.05 where used to identify enriched GO terms.
Quantitative PCR. Total RNA was isolated using the Nucleospin RNA (Macherey-Nagel) with on column DNase treatment. Reverse transcription to make cDNA was performed with iScript Supermix (Bio-Rad). For qPCR the primers listed in the Supplementary Information were used with iTaq SYBR Green (Bio-Rad). Fold change in RNA expression was determined by the ΔΔCt method with normalization to PSMA5, TOMM20 and ATP5B. The normalization genes were chosen based on steady expression across all samples as measured by RNA-seq. For qPCR of the luciferase reporters: RNA was treated with the restriction enzyme AluI for 30 min at 37 °C in 1 × Turbo DNase Buffer (Thermo Fisher). After treatment 1 µL of Turbo DNase (Thermo Fisher) was added and remaining DNA was digested for 30 min at 37 °C. Turbo DNase was removed using the Turbo DNA-Free Kit (Thermo Fisher) and the RNA was used for cDNA synthesis as above, with the addition of a 20 min incubation at 80 °C following first strand synthesis to inhibit the restriction enzyme. This approach allowed for complete removal of plasmid DNA from the RNA sample as determine by qPCR of a no-reverse transcriptase control reaction (data not shown). Quantitative PCR was performed as above using Renilla and firefly luciferase qPCR primers (Supplementary Information). Fold change in firefly (RPL23A reporters) or Renilla (EEF2 reporters) luciferase RNA was determined by the ΔΔCt method with normalization to Renilla (RPL23A reporters) or firefly (EEF2 reporters) luciferase, respectively.