Pyruvate Kinase M (PKM) binds ribosomes in a poly-ADP ribosylation dependent manner to induce translational stalling

Abstract In light of the numerous studies identifying post-transcriptional regulators on the surface of the endoplasmic reticulum (ER), we asked whether there are factors that regulate compartment specific mRNA translation in human cells. Using a proteomic survey of spatially regulated polysome interacting proteins, we identified the glycolytic enzyme Pyruvate Kinase M (PKM) as a cytosolic (i.e. ER-excluded) polysome interactor and investigated how it influences mRNA translation. We discovered that the PKM-polysome interaction is directly regulated by ADP levels–providing a link between carbohydrate metabolism and mRNA translation. By performing enhanced crosslinking immunoprecipitation-sequencing (eCLIP-seq), we found that PKM crosslinks to mRNA sequences that are immediately downstream of regions that encode lysine- and glutamate-enriched tracts. Using ribosome footprint protection sequencing, we found that PKM binding to ribosomes causes translational stalling near lysine and glutamate encoding sequences. Lastly, we observed that PKM recruitment to polysomes is dependent on poly-ADP ribosylation activity (PARylation)—and may depend on co-translational PARylation of lysine and glutamate residues of nascent polypeptide chains. Overall, our study uncovers a novel role for PKM in post-transcriptional gene regulation, linking cellular metabolism and mRNA translation.


INTRODUCTION
The regulation of mRNA stability, translation and localization is critical for cellular function. Post-transcriptional regulators of mRNAs include RNA-binding proteins (RBPs), ribosome interacting proteins, and nascent-chain associated factors. Various groups have identified post-transcriptional regulators by assaying for mRNA or ribosome interactions. Of particular interest are putati v e post-transcriptional regula tors lacking annota ted RNA-binding domains, known as 'noncanonical', 'unconventional' or 'enigm-' RBPs, and tend to include many metabolic enzymes hinting at a potential connection between metabolism and mRNA regulation (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16). In some cases, these putati v e regulators have been documented to interact with the mRNA, in other cases the ribosomes, and rarely with the nascent chain. Most of these studies have been conducted in model systems, such as yeast, flies, and mammalian cell lines, but recently these hav e been e xtended to more physiological conte xts such as in mouse organs ( 17 ).
Additionally, an o verlook ed aspect of mRN A biolo gy is the spa tial organiza tion of post-transcriptional regula tors within cells. This spatial organization contributes to the local regulation of mRNA translation in response to subcellular demands ( 18 ) and compartment-specific stress such as the unfolded protein response in the lumen of the endoplasmic reticulum (ER) ( 19 ). Although there are known differences between the regulation of mRNA translation on the surface of the ER versus the cytosol (20)(21)(22)(23)(24)(25), most studies of spatially restricted post-transcriptional regulators have focused on ER-bound ribosomes (26)(27)(28). Howe v er, to understand how mRNAs are spatially regulated, we must also understand how these factors are enriched in free cytosolic (i.e. ER-excluded) polysomes.
To address how cytosolic and ER-associated polysomes ar e differ entially r egulated, we combined cellular fractionation and high-speed centrifugation to isolate both cytosolic and ER ribosomes and identified their proteomic composition by mass spectrometry. We focused our efforts on Pyruvate Kinase M (PKM), as our preliminary results suggested that its association to ribosomes was restricted to cytosolic polysomes and sensiti v e to glucose / pyruv ate-starv ation. Canonically, PKM produces p yruvate from phosphoenolp yruva te, while genera ting ATP from substrate-le v el phosphorylation. PKM has been also implicated in the Warburg effect, where it may shunt glycolytic substrates towards anabolic processes rather than oxidati v e phosphorylation (29)(30)(31). Beyond metabolism, PKM has been reported to act as a protein kinase to regulate various processes such as cell proliferation, DNA repair, mitotic progression, and transcription (32)(33)(34)(35)(36). PKM lacks any identifiable RNA-binding domain, making it an ideal model for understanding how these non-canonical RNA-binding proteins are interacting with RNA. Previously, it had been found that PKM bound directly to ribosomes, likely near the A site, and altered mRNA translation ( 10 ), although the details and nature of this regulation were unclear. We found that PKM crosslinked with the open reading frames (ORFs) of mRNAs whose protein products are either cytosolic or nucleoplasmic. These interactions occurred just downstream of regions encoding glutamate or lysine. Furthermore, we demonstrated that PKM promotes ribosome stalling in the vicinity of glutamate-and lysine-encoding regions. Lastly, we found that PKM recruitment to ribosomes is dependent on poly-ADP ribosylation activity and may rely on co-translational PARylation of nascent polypeptides-a completely novel cotransla tional modifica tion. Our da ta suggests tha t this interaction is disrupted by increases in cellular ADP, thus linking the cellular metabolic state to the regulation of mRNA translation.

Cells, growth conditions, and lentiviral mediated depletion of PKM
U2OS, HepG2, HEK293T and HEK293F cells were maintained in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin / streptomycin (P / S) at 37 • C and 5% CO 2 . Short term glucose starvation was carried out via incubation with glucose-and p yruvate-fr ee DMEM (Gibco Cat#11966-025) supplemented with 10% FBS and 1% P / S with either 20 mM 2-deoxyglucose (D8375 Sigma) or vehicle (dH 2 O) for 3 h. For poly-ADP ribosylation inhibition, U2OS cells wer e tr eated with 15 M Olaparib or DMSO vehicle for 25 min. For mitotic arrest experiments, a pproximatel y 8.8 × 10 6 U2OS cells were synchronized toward G1 / S by growing in medium supplemented with 2 mM thymidine for 16 h, followed by 24 M deoxycytidine for 8 h, followed by 2 mM thymidine for 16 h, and lastly by 24 M deoxycytidine for 2 h. 100 ng / ml of nocodazole was added for 18 h to mitotically arrest synchronized cells. Mitotic cells were collected by vigorously washing tissue culture dish with PBS.

Cell fractionation and oligo-dT affinity chromatography
To isolate crude polysomal fractions, 75 × 10 6 U2OS cells wer e tr eated with growth medium supplemented with 10 g / ml cy clohe ximide for 30 min. Cells were collected by trypsiniza tion, then sedimented a t 800 g for 2 min, washed twice with ice cold PBS containing 10 g / ml cy clohe ximide, washed once with ice cold Phy Buffer (150 mM Potassium Aceta te, 5 mM Magnesium Aceta te, 20 mM HEPES-KOH pH 7.4, 5 mM DTT, protease inhibitor cocktail [Roche], 10 g / ml cy clohe ximide). Cell pellets wer e then r esuspended in 1 ml Phy Buffer. Cells were extracted by adding an equal volume of cold Phy Buffer + 0.08% digitonin and gently inverting the tube to allow the detergent to mix. The solution was then centrifuged at 800 g for 2 min to produce a suspension (cytosolic fraction, C1) and pellet (P1). The pellet was resuspended in additional 1 ml of cold Phy Buffer and extracted with an equal volume of cold Phy Buffer and 0.08% digitonin. The solution was then centrifuged at 800 g for 2 min to produce a suspension (C2) and pellet (ER + nuclear fraction, P2). The pellet was then resuspended in 1 ml cold Phy Buffer and extracted by adding an equal volume (1 ml) of Phy Buffer + 0.05% TritonX-100. This sample was then centrifuged at 800 g for 2 min to produce a suspension (ER fraction) and pellet (nuclear fraction). Cytosolic (C1) and ER fractions were then centrifuged at 10 000 g for 10 min to remove contaminating organelles such as mitochondria and nuclei. These fractions were analyzed for protein or further fractionated.
For crude ribosome fractionation, 0.5ml of the C1 or P2 fractions were layered over a 0.5 ml Phy-sucrose buffer (Phy buffer containing 1 M sucrose), and centrifuged at 90 000 RPM for 40 min in a TLA120.2 rotor, to produce a suspension (non-ribosomes) and a pellet (ribosomes). The polysome fraction was then resuspended in 100 l of Phy Buffer supplemented with 10 l RNases and incubated for 10 min at room temperatur e. Tr eated fractions were centrifuged 60 000 RPM for 1 h to produce a suspension (mRNA-dependent interactors) and pellet (Ribosome-proteins) (see Figure 1 B).
For the oligo-dT affinity chromato gra phy 0.5 ml of C1 or P2 fractions were incubated with an equal volume of a 50% slurry of oligo(dT) beads (NEB #S1408S) in Phy buf fer, or unconjuga ted Protein A beads (Thermo Fisher Cat#101041) overnight at 4 • C. Beads were washed 5 times with 1 ml cold Phy buffer. Proteins were eluted off the beads by incubating them with 2 × Laemmli buffer at 65 • C for 5 min.

Polysome analysis by sucrose centrifugation
18 × 10 6 U2OS were treated with cy clohe ximide (100 g / ml) or Homoharringtonine for 10 min. Cells were collected via trypsinization and sedimented at 800 g and washed with ice cold PBS (supplemented with 100 g / ml cy clohe ximide) three times. Cells were resuspended in 1 ml pol ysome l ysis buffer (20 mM HEPES-KOH pH 7.4, 5 mM MgCl 2 , 100 mM KCl, 1% Triton X-100, 100 g / ml cycloheximide) supplemented with 20 U / ml Superase RNAse Inhibitor (Thermo Fisher Cat# AM2694), and protease inhibitor cocktail (Roche). Lysates wer e clear ed via centrifuga tion a t 16 000 g for 10 min. 500 l of lysate saved for RNA input or total. The remaining 500 l was layered on a 20-60% sucrose gradient buffer (20 mM HEPES-KOH pH 7.4, 5 mM MgCl 2 , 100 mM KCl, 100 g / ml cy clohe ximide, and either 20% or 60% sucrose weight / volume) generated using the Biocomp Gradient Master. Lysates centrifuged at 36 000 g for 2 h in a SW-41 rotor. Samples were collected and OD260 was continuously measured using the Biocomp Piston Gradient Fractionator. For RNA-seq analysis, polysome fractions (disomes and heavier) were pooled together and RNA was extracted using Trizol-LS (Thermofisher Cat#10296010) protocol. For total fraction, RNA was extracted from saved input. For immunoblotting, proteins from individual fractions were salted out via a TCA precipitation, washed in 100% acetone and re-suspended in 5 × Laemmli buffer.

Ribosome co-sedimentation
18 × 10 6 U2OS cells wer e pr etr eated with either cy clohe ximide (100 g / ml) for 10 min or puromycin (200 M) for 30 min and then collected by trypsinization. The cells were washed twice in ice cold PBS and lysed in ribosome lysis buffer (125 mM KCl, 5 mM MgCl 2 , 20 mM HEPES-KOH pH 7.4, 250 mM sucrose, 0.08% digitonin, 100 g / ml cycloheximide). Unlysed cells were removed by centrifugation at 800 g for 10 min, and the resultant supernatant was centrifuged for 16 000 g to remove cellular debris. For ADP, ATP, PEP and F-1,6-BP ribosome sedimentations, the indicated amount of metabolite was added to cleared lysate. The concentration of KCl in the cleared supernatant was adjusted to 500 mM for high salt conditions or remained at 125 mM for physiological conditions. 0.5 ml of cleared supernatant was layered on 0.5 ml of either a high salt (500 mM KCl) or low salt (125 mM KCl) sucrose cushion (1 M sucrose, 5 mM MgCl 2, 20 mM HEPES-KOH pH 7.4, 100 g / ml cy clohe ximide) in a 1 ml polycarbonate tube and then centrifuged at 90 000 RPM for 1 hour in a TLA-120.2 rotor. The pellet was washed twice in ice cold dH 2 O prior to solubilization in suspension buffer (125 mM KCl, 5 mM MgCl 2 , 20 mM HEPES-KOH pH 7.4).

Salt-washed ribosome isolation and In vitro co-sedimentation
To generate salt washed ribosomes, 500 ml of HEK293F cells were collected via centrifugation at 800 g , and washed 5 times with ice cold PBS. Cells were lysed in modified ribosome lysis buffer (125 mM KCl, 5 mM MgCl 2 , 50 mM Tris-HCl pH 7.4, 250 mM sucrose, 1% NP-40). Lysates were cleared at 16 000 g for 10 min. The concentration of KCl in the cleared supernatant was adjusted to 500 mM KCl. 0.5 ml of cleared supernatant was layered on 0.5 ml of high salt (500 mM KCl) sucrose cushion (1 M sucrose, 5 mM MgCl 2, 50 mM Tris-HCl pH 7.4) in a 1 ml polycarbonate tube and then centrifuged at 90 000 RPM for 1 h in a TLA-120.2 rotor. The pellets wer e r e-suspended in modified suspension buffer (500 mM KCl, 5 mM MgCl 2, 50 mM Tris-HCl pH 7.4). Re-suspended pellets were subjected to an additional round of centrifugation --layering 0.5 ml of the re-suspended pellet solution over 0.5 ml of high salt cushion buffer in a 1 ml polycarbonate tube and then centrifuged at 90 000 RPM for 1 h in a TLA-120.2 rotor. Pellets were then re-suspended in suspension buffer (25 mM KCl, 5 mM MgCl 2, 50 mM Tris-HCl pH 7.4). Bicinchoninic acid (BCA) assay was used to measure relati v e ribosome concentrations. Equal amounts of salt-washed ribosomes, wer e mix ed with equal molar amounts of either recombinant GST or GST-PKM1 in a 0.1 ml suspension buffer and incubated on ice for 30 min. This binding solution was layered on a 0.5 ml sucrose cushion (1 M sucrose, 5 mM MgCl 2, 50 mM Tris-HCl pH 7.4) in a 1 ml polycarbonate tube and then centrifuged at 90 000 RPM for 1 hour in a TLA-120.2 rotor. The supernatant was discarded and the pellet washed twice in ice-cold water. The pellet was re-suspended in 1 × Laemmli buffer and denatured at 95 • C for 5 min.

Microscale thermophoresis
Purified PKM1 was fluorescently labeled with fluorescein-5-ex succinimidyl ester (Invitrogen) on lysine residues according to the manufacturer's protocol by mixing PKM1 at a final concentration of 3.5 M protein with an 8-fold molar excess of dye at room temperature for 30 min in the dark. Free dye was eliminated through e xtensi v e dialysis and an 80 nM stock solution of labeled PKM in MST buffer (PBS + 5mM MgCl 2 ) was pr epar ed.
Prior to mixing, PKM1 and salt-washed ribosome stock solutions were spun at 21 000 g for 5 min to remove any aggregated species. We then performed a 1:1 serial dilution of purified ribosomes in MST buffer. All dilutions were performed in 200 l PCR strips (Starstedt). Purified ribosome dilutions were mixed 1:1 with labeled PKM1, yielding a final PKM1 concentration of 40 nM and a ribosome  Supplementary Table S1 for the complete list). (G-H) Analysis of the ER / cytosolic distribution of various proteins in the crude ribosome fractions ( G ) and oligo-dT affinity purified fractions ( H ) by immunoblot. ( I ) For all proteins expressed in U2OS cells, a comparison of the total number of peptides identified by mass spectrometry in the oligo-dT chromato gra phy experiments (both ER and cytosol; y-axis ) was plotted against the estimated le v el of proteins ( x-axis ; data from ( 15 )). Gl ycol ytic enzymes are labeled in red, and ribosomal proteins in magenta. Proteins that did not appear in the oligo-dT chromato gra phy experiments were set to 0.1 peptides, while those proteins that did not appear in the analysis of protein le v els were set to 100 molecules / cell. (J-L) For each protein present in either the crude polysome or oligo-dT purifications, the total number of peptides ( y-axis ) identified by mass spectrometry was plotted against the percentage of peptides found in the ER [100% × peptides in ER / (peptides in ER + peptides in cytoplasm); '% on ER'x-axis ]. This analysis was performed on the RNA-dependent ( J ) and ribosome-bound ( K ) fractions of the crude polysome preparations, and on the oligo-dT associated proteins ( L ). Note that the x-axis is the average of 5 biological replicates for (J, K) and two biological replicates for (L), while the y-axis is the total sum of peptides in all experiments (J-L). Classes of proteins that are enriched in either the ER or the cytoplasm are highlighted --carbohydrate metabolic proteins (red), eIFs (yellow) and cytoskeletal-associated proteins (green). concentr ation r ange from 0 to ∼1.5 M. Binding reactions were incubated for 30 min before loading into standard glass capillaries (NanoTemper Technologies). MST measur ements wer e performed on a Monolith NT.115 microscale thermophoresis instrument (NanoTemper Technologies) at room temperature using a BLUE LED power of 60% and a medium MST-IR laser power. The resulting dose response curves from three biological replicates were exported from the NanoTemper Technologies Analysis Soft-ware (Version 2.3) into GraphPad Prism 6 where the K d was calculated by globally fitting the data from the three curves to a sigmodal dose response assuming a 1:1 binding model.

Mass spectrometry
Protein samples were separated by electrophoresis on a 10% SDS-PAGE gels and stained with Coomassie brilliant blue staining (BioBasic). Each lane on the gel was cut into 16 samples and destained using 50% acetonitrile with 0.2% ammonium bicarbonate. The gel pieces were shrunk with acetonitrile and air dried. The gel pieces were incubated in digestion mixture of trypsin (10 ng / ul) in 50mM ammonium bicarbonate and incubated at 37 • C overnight. The f ollowing da y the gel pieces were centrifuged and the digestion mixture removed. The gel pieces were soaked in 1% formic acid in 50% acetonitrile: 49% water for 15 min, centrifuged and the supernatant combined with the digestion mixture. The samples were lyophilized and resuspended in 50mM ammonium bicarbonate and analyzed by liquid chromato gra py-tandem MS using either an LTQ-XL linear ion-trap mass spectrometer (Thermo Fisher Scientific) or Proxeon Easy-nLC 1200 pump in-line with a hybrid LTQ-Orbitrap velos mass spectromter (Thermo Fisher Scientific). Raw files from LTQ-XL mass spectrometer were uploaded to Prohits database ( 39 ) and converted to MGF. The data was analyzed and searched using Mascot (2.3.02; Matrix Science). Raw files from LTQ-Orbitrap mass spectrometer was uploaded to Prohits ( 39 ) and converted to mzXML. The data was analyzed and searched using X!Tandem ( 40 ). Raw files available on MassIVE (MSV000090941) and pro-teomeXchange (PXD038978).
We analyzed ER and cytosolic fractions from fiv e crude polysomes preparations (both RNA-dependent and ribosome-bound fractions) and two oligo-dT affinity purifications by mass spectrometry analysis. Note that due to the lower overall levels of protein in the Mock Bead pulldowns (Figure 1 D), the peptide counts in this fraction are likely to be over-sampled in comparison to the oligo-dT purified samples. We scored proteins as present in the ribosome and RNA-dependent fractions if r epr esented by peptides in at least two of three separate biological replicates. For the oligo-dT experiments, proteins had to contain peptides that appeared in two of three biological replicates and be at least two-fold enriched over the mock bead pulldown in both experiments. Manual curation removed 30 proteins that were likely contaminants (e.g. Keratin, RNase, Albumin, mitochondrial proteins). Percent ER was calculated by determining the fractional r epr esentation of peptides from a protein in a gi v en pool (pools included: ER RNAdependent, Cyto RNA-dependent, ER Ribosome-bound, Cyto Ribosome-bound, ER mRNP(oligo-dT)-bound, Cyto mRNP(oligo-dT)-bound) and averaging these between experiments.

Computational analysis of eCLIP-seq data
Trimming and mapping. Sequencing r eads wer e processed essentially as described ( 42 ). Reads were adapter trimmed and mapped to human-specific repetiti v e elements from RepBase (version 18.04) by STAR ( 43 ). Repeat-mapping r eads wer e r emoved and r emaining r eads mapped to the human genome assembly hg19 with STAR. PCR duplicate reads were removed using the unique molecular identifier (UMI) sequences in the 5 adapter and remaining r eads r etained as 'usable reads'. Peaks were called on the usable reads by CLIPper ( 44 ) and assigned to gene regions annotated in Gencode (v19). Each peak was normalized to the sizematched input (SMInput) by calculating the fraction of the number of usable reads from immunoprecipita tion to tha t of the usable reads from the SMInput. Peaks were deemed significant at > 8-fold enrichment and P -value < 10 −5 (Chi-square test). All sequencing and processing statistics are in Supplementary Table S3.
Irr epr oducibility discovery r ate. To test the quality of the datasets, Irreproducib le Discov ery Rate (IDR) analysis as described in ( 45 ) was performed, yielding 1907 common peaks between the two replicates (consistency ratio: 5.37; rescue ratio: 2.40).

Ribosome footprint profiling
HEK293T cell line was obtained from ATCC and maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Cells were washed twice with 10 ml of ice-cold PBS and placed on ice. Lysis was carried out on plates using 400 l volume (Tris-HCl 20 mM pH 7.4, 150 mM NaCl, 5 mM MgCl 2 , 1 mM DTT, 100 g / ml cy clohe ximide, 1% Triton-X). 5 l of RNaseI (Invitrogen AM2249) was added to the lysates, followed by a 1 h digestion a t 4 • C . Immedia tely following digestion, ribonucleoside vanadyl complex (NEB S1402S) (20 mM) was added to inhibit further digestion. Digested lysates (400 l each) were loaded on a sucrose cushion (20 mM Tris-Cl titrated to pH 7.4, 150 mM NaCl, 5 mM MgCl 2 , 1% TritonX-100, 34% sucrose, 1 mM DTT, 100 ug / ml cy clohe ximide) and centrifuged in a SW41Ti rotor (Beckman Coulter) at 38000 RPM at 4 • C for 2.5 h to pellet ribosomes. 700 l QIAzol Lysis Rea gent (Qia gen) was used to resuspend the ribosome pellet, which was was transferred to a fresh 1.5 ml tube and frozen until further processing. To extract RNA, samples were thawed at room temperature and processed with the Qiagen miRNeasy Kit per manufacturer's instructions. RNA was eluted in 20 l nuclease-free water and size-selected by electrophoresis of 3 g inputs in a Nov e x denaturing 15% polyacrylamide TBE-urea gel. The 26-34 nt RNA fragments were excised, crushed with a pestle, and extracted in 310 l gel extraction buffer (300 mM NaOAc pH 5.2, 1 mM EDTA, 0.25% w / v SDS) by incubating on dry ice for 30 min then overnight at room temperature. The gel-liquid mixture was passed through a Spin-X filter (Corning 8160) and the elution was precipitated with 1.5 l Gly cob lue (5 mM MgCl 2 , 75% ethanol). RNA was pelleted by centrifugation at 23 300 g at 4 • C for 1 h and resuspended in 10 l nuclease-free water. 25 ng inputs of sizeselected ribosome footprints were immediately processed for library preparation using the D-Plex Small RNA-Seq Kit (Diagenode) per manufacturer's instructions, with some modifica tions. Approxima tely half of the resulting cDNA (14 l) was amplified for 9 cycles. Following quantitation of the target libraries with Agilent HS DNA Kit and Bioanalyzer, the resulting libraries were pooled to equimolar concentrations and cleaned with AMPure XP beads at 1.8x concentration (Beckman Coulter) and eluted in 30 l TE. To enrich for ribosome footprints in the libraries, the entire elution was size selected in a 3% agarose precast gel (Sage Science) in the BluePippin system using a 172-206 nt range with tight settings. The resulting libraries were sequenced with NovaSeq SP SE 100 (Illumina). All sequence data was deposited to GEO (GSE202881).

RNA sequencing
To extract RNA, samples w ere thaw ed at room temperature and processed with the Qiagen miRNeasy Kit per manufacturer's instructions. RNA was eluted in 25 l nuclease-free water and processed with the CATS RNA-seq Kit v2 per manufacturer's instructions (Diagenode C05010045). Approximately half of the resulting cDNA (14 l) was amplified for 11 cycles. PCR products were cleaned as outlined in the Diagenode protocol and the resulting libraries were sequenced with NovaSeq SP SE 100 (Illumina). All sequence data was deposited to GEO (GSE202881).

Computational analyses of ribosome profiling data
Ribosome profiling data were processed using RiboFlow ( 47 ). We extracted the first 12 nucleotides from the 5 end of the reads using UMI-tools ( 48 ) with the following parameters: 'umi tools extract -p ' ∧ (?P < umi 1 > . { 12 } )(?P < discard 1 > . { 4 } ).+ $ ' -extract-method = regex'. The four nucleotides downstream of the UMIs are discarded as they are incorporated during the re v erse transcription step. Next, we used cutadapt ( 49 ) to clip the 3 adapter AAAAAAAAAACAAAAAAAAAA. After UMI extraction and adapter trimming, ribosomal and transfer RNAs wer e filter ed by alignment using Bowtie2. The r emaining r eads wer e mapped to human transcriptome and alignments with mapping quality greater than two were retained. UMIs were used for deduplication and .ribo files ar e cr eated using RiboPy ( 47 ). Identification of differential ribosome occupancy and RNA expression was carried out as previously described ( 50 , 51 ).

Pause site detection
P-site adjustment was carried out using the metagene plots to determine offsets as a function of read length. We selected genes for which normalized CDS coverage is equal to or larger than one in all experimental replicates. We then fitted negati v e binomial distributions for each genes' ribosome occupancy coverage vector at nucleotide resolution after removing 5% top and bottom outliers. Using these estimated parameters, we calculated P -values for each nucleotide resolution that captures the probability of the observed read counts are deri v ed from this distribution. We the combined P -values from experimental replicates with Fisher's method. Finally, positions with a combined Pvalue < 1 × 10 −7 were defined as putati v e pause sites.

Detection of differential pause sites
To find differential pause sites, we compared coverage of the pause sites between shPKM and shCtrl set with edgeR ( 52 ) and filtered pause sites with an adjusted P -value threshold of 0.05. Some of these pause sites can potentially be attributed to the overall ribosome occupancy coverage difference between shPKM and shCtrl treatments. To filter out such cases, we determined pause sites that were detected in genes with differential ribosome occupancy. We then removed pause sites where the absolute value of the difference between a pause site's log2 fold-change and the log2 foldchange of the gene's ribosome occupancy was < 0.8.

Ribosome foot printing northern blots
16 million U2OS cells were pretreated with cy clohe ximide (100 g / ml) for 10 min and then collected by trypsinization. Cells were washed twice in ice cold PBS followed by the addition of 800 l lysis buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM MgCl 2 , 1 mM DTT, 100 g / ml cycloheximide, 1% TritonX). Cells were lysed on ice by vigorous pipetting for 10 min. Lysed cells were centrifuged for 10 min at 800 g and the collected supernatant centrifuged for 10 min at 16 000 g . The resulting supernatants were treated with 14 l of RNaseI (Invitrogen AM2294) for 1 h at 4 • C to degrade unprotected mRN A fragments. RN A digestion was inhibited by addition of 20 mM ribonucleoside vanadyl complex (NEB S1402S). 500 l of RNAse digested samples was layered on 500 l sucrose cushion (1 M sucrose, 5 mM MgCl 2 , 20 mM Tris-HCl pH 7.4, 100 g / ml cy clohe ximide) in a 1 ml polycarbonate tube and then centrifuged at 90 000 RPM for 1 h in a TLA-120.2 rotor. Resulting pellet was washed twice in ice cold dH 2 O prior to extraction of RNA from ribosome pellet by TRIzol (ThermoFisher 15596026).

Nascent polypeptide analysis
Nascent polypeptides were isolated by adapting the protocol from Aviner et al. ( 53 ). 50 million U2OS cells were collected by trypsinization. Cells washed twice with icecold PBS and re-suspended in 1 ml PUNCH-P polysome buffer (50 mM HEPES pH 7.5, 10 mM MgCl 2 , 25 mM KCl, supplemented with protease inhibitor cocktail [Roche] and RNase inhibitor [AM2694]). 120 ul of lysis buffer (11% sodium deoxycholate, 11% Triton-X) added to resuspended cells, and incubated on ice for 20 min with occasional pipetting. Cell lysate clarified at 16 000 g for 15 min. 500 ul Clarified lysate layered on 500 ul sucrose cushion (PUNCH-P polysome buffer containing 1 M sucrose). Translating ribosomes pelleted by centrifugation at 37 000 RPM for 160 min. Ribosome pellet washed with 500 ul dH 2 O and resuspended in 90 ul polysome buffer. Resuspended ribosomes were supplemented with either 100 pmol of puromycin or biotin-puromycin analogue (Jena Bioscience, Biotin-dCpuromycin) per 1 OD254 of detected ribosomes and incuba ted a t 37 • C for 15 min to release nascent chains. 5 ul of each reactions were kept to assess biotinylation efficiency. Streptavidin beads (ThermoFisher, 20347) were added to remaining reaction volumes ( ∼40 uL) at 5 ul of slurry per 1 OD254 of detected ribosomes to capture nascent chains. Streptavidin-biotin capture was performed overnight at 4 • C. After captur e, str eptavidin beads were washed with 6 times in PUNCH-P polysome buffer. To elute biotinylated proteins, streptavidin beads were incubated with a solution of 95% formamide, 10 mM EDTA, and 25 mM free biotin at 95 • C for 10 min. Eluted nascent chains mixed with 5X Laemmli buffer and analyzed by SDS-PAGE.

Proteomic analysis of cytosolic and ER polysome-associated factors
The cytosol and ER r epr esent the two major subcompartments where cellular protein synthesis occurs, each containing distinct pools of mRNAs and unique translational regulatory systems (20)(21)(22)(23)(24)(25)(26)(27)(28). To determine the spatial distribution of post-transcriptional regulatory factors we isolated polysomes from ER and cytosolic fractions (Figure 1 A) from human osteosarcoma (U2OS) cells (Figure 1 B). These isolated polysomes were treated with RNase to liberate putati v e RNA-binding proteins ('RNA-dependent fraction'), and resedimented to pellet ribosomes and associated proteins ('Ribosome-bound fraction'). We analyzed the composition of the RNA-dependent and ribosome-bound fractions (Figure 1 C) by mass spectrometry, as previously described ( 26 ). In parallel we also isolated messenger ribonuclear protein (mRNP) complexes from the ER and cytosol using oligo-dT af finity chroma to gra phy ('mRNP-bound' ;  Figure 1 B, D ), or a mock bead control, and analyzed these fractions by mass spectrometry. Our purification conditions, done in the absence of crosslinking, enabled recovery of proteins that are directly and indirectly bound to mR-NAs or ribosomes, although we cannot fully rule out the co-purification of other large molecular complexes in our "Ribosome-bound' fraction.
To ensure that we only classify true interactions, we considered proteins present in (a) the mRNP-bound and (b) either the RNA-dependent or ribosome-bound fractions, as high confidence interactors that associate with either mRNA, or ribosomes. After statistical processing and manual curation (see methods), we identified 496 polysome interactors (Figure 1 E, Supplementary Table S1). Nearly all of these proteins (98%; 484 / 496) had been identified in previous global analyses of proteins tha t associa te with RNA or ribosomes ( 1-16 ) (Supplementary Table S2), indicating that our list of polysome interactors is in agreement with the literature.
Our high confidence polysome interactors fell into several broad categories (Figure 1 F). These include a wide array of metabolic enzymes, especially those involved in carbohydrate metabolism, similar to what had been uncovered by se v eral genome-wide RNA and ribosome binding protein surveys from a variety of species ( 1-16 , 54 ). The most abundant interactors by spectral counts were gl ycol ytic enzymes, including both spliced isoforms of pyruvate kinase M (PKM1 and PKM2). The detected gl ycol ytic enzymes were not nascent polypeptides emerging from the ribosome, a concern for abundantly expressed proteins, as we found no bias for N-terminal peptides (Supplementary Figure S1), which would be the case for partially synthesized proteins. The presence of full-length gl ycol ytic enzymes (as opposed to partially synthesized polypeptides) in these fractions was further confirmed by immunoblot (Figure 1 G, H ). Additional gl ycol ytic enzymes that were identified in the mass spectrometry but did not meet our cut-off (e.g. PGAM1, PFKL, PFKB) were also found in the RNA-dependent and ribosome-bound fractions (Figure 1 G, H ). Lastl y, w hen the number of peptide spectral counts in the oligo-dT pulldown were compared to estimates of overall protein le v els in U2OS cells ( 55 ), gl ycol ytic enzymes were on par with ribosomal proteins (Figure 1 I, gl ycol ytic enzymes are in red, ribosomal proteins are in magenta), suggesting that they were not trace contaminants.
We calculated both the ER and cytosolic distribution of post-transcriptional regulators from our RNA-, ribosome-, and mRNP-bound fractions (Supplementary Table S1) and found that they differed significantly from what would be expected if peptides were randomly assorted ( Figure 1 J-L , Supplementary Figure S2). As expected, ER-resident proteins were biased towards the ER, whereas ribosomal proteins were not biased suggesting equal sampling of ribosomes from both compartments (Figure 1 G, H , Supplementary Figure S3). The ER / cytosolic distribution of individual proteins was highly correlated between the different fractionation protocols (Supplementary Figure S4). Interestingly, eukaryotic translation initiation factors (eIFs) were slightly enriched in the ER for all fractions (Figure 1 J-L , yellow data points), and this was verified by immunoblot (Figure 1 G). We speculate that this may be due to the pres-ence of non-translating ribosomes that are predominantly in the cytosol. In contrast, cytoskeletal components including actin, tubulin, motor and cytoskeletal-binding proteins were enriched in the cytosol for all fractions (Figure 1 J-L , green data points; also see ␣-tubulin in Figure 1 G, H and ␤-actin in Figure 1 H), which may indica te tha t ribosomes and mRNPs that are transported along cytoskeletal filaments must be disengaged from the ER. Lastly, carbohydrate metabolic enzymes, including gl ycol ytic enzymes, were also enriched in the cytosol (Figure 1 J-L , red data points). Indeed, most of these co-sedimented with cytosolic ribosomes in an RNA-dependent manner, suggesting that they bound directly or indirectly to RNA, and this was verified by immunoblot (Figure 1 G, H ).

PKM associates with polysomes in a glucose / p yruv atedependent manner
Since we observed that gl ycol ytic enzymes were present in cytosolic polysomes, we asked if their association was sensiti v e to changes in gl ycol ysis, as the canonical functions of gl ycol ytic enzymes ar e often r egulated by metabolites. We found that PKM, which catalyzes the last step of glycolysis, pr efer entially co-fractionates with polysomes isolated from cells fed with glucose and pyruvate in comparison to cells starved of these two metabolites for 3 h (Figure 2 A-C ). PKM-polysome co-fractionation was diminished in U2OS cells pr e-tr eated with homoharringtonine (HHT), which eliminates pol ysomes, w hen compared to lysates from cy clohe ximide-treated cells (Supplementary Figure S5A, B) indica ting tha t PKM migration in the gradient is polysome-dependent. PKM also cosedimented with isolated ribosomes by sucrose cushion centrifugation in a glucose / pyruvate-dependent manner (Figure 2 D). Glucose / pyruvate starvation did not significantly change total le v els of PKM, ATP or lactic acid (Figure 2 E-G ), suggesting that this treatment did not activate major stress pathways in our cells. In contrast, treatment with 2deoxyglucose (2-DG), an inhibitor of gl ycol ysis, impaired mTOR signaling, as monitored by the phosporylation of S6K, and caused a decrease in the le v els of ATP and lactic acid (Figure 2 E-G).
The PKM gene produces two proteins due to alternati v e splicing. Since the M2 isoform (PKM2) is expressed at a higher le v el than the M1 (PKM1) ( 56 ), we tested whether PKM sedimentation is affected by the exogenous addition of the M2 specific allosteric regulator fructose-1,6-bisphosphate but found that it had no effect on PKM sedimentation (Figure 2 H). As we detected both PKM1 and PKM2 peptides in our cytosolic polysome screen, and gi v en the insensitivity of PKM ribosome interaction to fructose-1,6-bisphosphate, we belie v e that the PKM ribosome interaction is isoform-independent. Glucose starvation is known to ra pidl y increase the le v els of ADP ( 57 ), w hich directl y binds to and is a substrate of PKM. We found that the exogenous addition of ADP to lysate from unstarved cells disrupted the co-sedimentation of PKM with ribosomes (Figure 2 I). In contrast, exogenous addition of ATP, also a PKM substrate, significantly increased PKM co-sedimentation with ribosomes although this was highly variab le between e xperiments (Supplementary Figure S5C). Polysomes were sedimented through a sucrose cushion and probed for pan-PKM and ribosomes (RPL4 and RPS6). Total proteins were also monitored by Ponceau stain. ( E ) U2OS cells were incubated in DMEM that contained or lacked glucose / pyruvate, with or without 2-deoxyglucose ('2-DG'; 20 mM) for 3 h. Lysates wer e monitor ed by immunoblot for pan-PKM, phosphorylated S6 Kinase (p-S6K, a downstream product of mTOR kinase) and Actin. (F, G) U2OS cells were incubated in DMEM tha t contained or lacked glucose / pyruva te, with or without 2DG for 3 h. Cell lysates were assessed for ATP ( F ) and lactate ( G ) le v els. Each bar r epr esents the mean of thr ee biological r eplica tes plotted alongside standard error, as whiskers. P -value calcula ted from paired Student's t -test. (H-J) U2OS lysate were spiked with 1 mM fructose-1,6-bisphosphate ( H ) or a combination of either 10 mM PEP, ADP or both ( I , J ). Polysomes were sedimented and probed for pan-PKM and ribosomes (RPL4 and / or RPS6). Total proteins were also monitored by Ponceau stain. ( K ) Densitometry analysis of (J). Each bar r epr esents the mean of at least three biological replicates plotted alongside standard error, as whiskers, and individual replicates, as dots. P -value calculated from paired Student's t -test.
The other substra tes, pyruva te and phosphoenolpyruva te (PEP), had no effect on PKM pelleting but when PEP was added with ADP ther e appear ed to be a loss in ADP dependent regulation, although this was highly variable between experiments (Figure 2 J, K ). Overall, these suggests that changes in gl ycol ytic flux, as seen in transient glucose and p yruvate r estriction, can modulate PKM-ribosome interactions through changes in ADP, and possibly ATP, le v els.

PKM associates with ribosomes that translate mRNAs coding for glutamate-or lysine-rich tracts
Gi v en that we and others ( 1 , 6 , 10-15 , 54 ) have identified PKM as a putati v e RN A-binding protein, we imm unoprecipitated the major isoform, PKM2 (Figure 3 A), under very stringent conditions from lysates of UV-crosslinked HepG2 cells, followed by limiting RNase trea tment. W hen these immunoprecipitates were labeled with [ ␥ 32 P]-ATP using polynucleotide kinase, one major RNase-sensiti v e band, corresponding to PKM2 was observed (Figure 3 B), indicating that this protein was in close proximity to endogenous RNAs. Next, we isolated the cross-linked RNAs and analyzed these by eCLIP-Seq ( 41 ). We identified a pproximatel y 4000 enriched crosslinking peaks in comparison to the sizematched inputs (8-fold enrichment, P -value < 10 −5 ), distributed over transcripts from 961 genes from two independent replicates (Supplementary Table S3). Both replicates displayed similar read distributions, gene targets and reproducible peaks (Figure 3 C, D , Supplementary Figure S6). Strikingly, we found that PKM2 was crosslinked primarily to the coding sequence (CDS) of target transcripts encoding primarily cytosolically localized proteins (Figure 3 C-E ). This suggested that PKM2 is in close proximity to translating mRNAs, as expected for our results and a previous study which identified PKM as a ribosome-associated protein ( 10 ). Additionally, we found that purified GST-PKM1 co-sedimented with salt-washed ribosomes in vitro , again suggesting that PKM1 and PKM2 likely behave similarly in their capacity to interact with polysomes (Figure 3 F). We were able to further verify the binding of GST-PKM1 to ribosomes using microscale thermophoresis (MST) ( 58 , 59 ), which has been used previously to characterize interactions between ribosomes and their binding proteins ( 60 , 61 ). When the thermophoretic mobility of fluorescein labeled GST-PKM1 was measured in increasing concentrations of ribosomes (Figure 3 G), we were able to obtain a dose response curve (Figure 3 H) that was used to estimate a dissociation constant ( K d ) of ≈ 0.9 ± 0.2 M. Although we wer e able measur e interactions at r elati v ely high concentrations of ribosomes (1.5 M at the highest MST concentration), w e w ere unable to fully sa tura te PKM binding. In cell extracts, endogenous PKM co-sedimented with puromycin tr eated ribosomes, which ar e not bound to mRNA and nascent chains; howe v er, this association was disrupted by high salt (Figure 3 I). PKM also co-sedimented with ribosomes from cell extracts treated with cycloheximide regardless of salt concentration (Figure 3 I). Since cy clohe ximide maintains the association between ribosomes, mRNA and nascent chains, our observations suggest that PKM may form additional contacts with either the mRNA or nascent polypeptide chain. In contrast, poly(A) binding protein (PABP) co-sedimented with ribosomes under all conditions (Figure 3 I), consistent with previous findings ( 62 ). Since the binding of PKM to polysomes is partially RNase-sensiti v e (Figure 1 G), it is possible that PKM reco gnizes rRN A, as suggested by previous work ( 10 ). While we observe PKM2 eCLIP peaks on sections of mature rRNA (Supplementary Figure S7A-C) they do not align with previously-identified PKM2 CLIP reads, and was not significantly enriched over the input (Supplementary Figure S7D). Ne v ertheless, it remains possible that the detection of PKM-rRNA association is highly dependent on the crosslinking procedure used.
Although our metagenome plot would suggest that PKM2 has no underlying sequence bias (Figure 3 D) we found that it had well defined crosslinking sites to individual target mRNAs. For example, in the case of the G3BP2 mRNA, we found PKM2 crosslinking to regions encoding peptides they are rich in glutamate and lysine (Figure 3 J). We analyzed all PKM2 mRNA targets and found that binding to sequences encoding charged residues was a general feature of PKM2 crosslinking sites (Figure 3 K, L ). A close examination of the distribution of encoded amino acids revealed that the peak enrichment was slightly upstream of the eCLIP crosslink (Figure 3 L, left of dotted line) suggesting that the recruitment of PKM to polysomes occurs after the synthesis of charged polypeptide chains. We plotted the enrichment of all amino acids proximal to the PKM2 eCLIP crosslinking site and found that only glutamate and lysine were significantly enriched over background (Figure 3 M, Supplementary Figure S8) suggesting that these two amino acids predominantly contributed to the bias in charged residues.

PKM promotes ribosome pausing and decay of substrate mRNAs
Previously, it was found that PKM depletion resulted in a decrease in ribosomal occupancy for a subset of mRNAs ( 10 ). We depleted PKM in HEK293T cells by RNAi using lenti viral-deli v ered shRNAs and performed ribosome profiling sequencing (Figure 4 A-C , Supplementary Figure S9). As expected, the peak fraction of reads fell in the 28-32 nucleotide range. Metagene plots indicated high coverage across the ORF alongside conventional trinucleotide periodicity and strong correlations were found between replicates (Supplementary Figure S9A-D). Overall, PKM depletion significantly affected the ribosome occupancy of mRNAs from almost 3000 genes (1682 decreased occupancy versus 1300 increased occupancy, P -value < 0.01; Figure 4 A, Supplementary Table S4). PKM2 eCLIP targets were disproportionally affected and tended to have decreased ribosome occupancy in PKM-depleted cells (Figure 4 A, yellow dots). Additionally, PKM depletion significantly affected the steady-state le v el of mRNA from 4186 genes (2147 upregulated versus 2039 downregulated, Pvalue < 0.01) (Figure 4 B). Again, we found that PKM2 eCLIP targets were disproportionally affected and tended to be stabilized by PKM depletion (Figure 4 B, yellow dots).
Since ribosome profiling only provides a snapshot of translation, w e w er e unable to differ entiate whether high ribosome footprint counts on a gi v en mRNA was due to  an increase in transla tion initia tion or elongational stalling ( 63 ). The former possibility would lead to an increase in protein production, while the latter would result in a decr ease. In particular, ther e is growing evidence that suboptimal elongation is coupled to mRNA decay ( 64 ) which may be the case for PKM eCLIP targets. This would explain the stabilization of these mRNAs upon PKM depletion (Figure 4 B). To distinguish between these two possibilities, we measured the synthesis of proteins (G3BP2, HSP90AA1 and eIF3A) that are encoded by mRNAs that are both PKM eCLIP targets and whose ribosome occupancy is affected by PKM depletion. Newly synthesized proteins were labelled by a 35 S-methionine / cysteine pulse, and isolated by immunoprecipitation. PKM depletion increased the synthesis of G3BP2, HSP90AA1 and eIF3A (Figure 4 D, E ). We also observed a general increase in the steady le v el of these proteins, although these were not statistically significant (Figure 4 F, G ). Overall, these observations are consistent with a model that PKM pr efer entially interferes with translational elongation on a subset of mRNAs.
Using our ribosome profiling dataset, we mapped putati v e pause sites in mRNAs present in all control and PKMdepleted replicates with high expression (RPKM > 1), with mRNAs from 605 genes meeting this cut-off. We then identified pause sites as regions with an unexpectedly higher preponderance of ribosome footprints than expected. In total, we mapped 5244 pause sites in mRNAs from 541 genes. Of these pause sites, 1298 were PKM-dependent (present in 440 mRNAs), with almost all (98.3%) having lowered ribosome read counts upon PKM depletion. Of the PKM-dependent sites, 500 were found across 131 PKM eCLIP target mR-NAs and all had lower read counts after PKM depletion. We found that PKM-sensiti v e pauses in eCLIP-targets were pr esent in r egions coding for polypeptides enriched in glutamate and lysine (magenta bar, Figure 4 H), similar to PKM crosslinking sites (Figure 3 L). To validate this finding, we performed a northern blot and probed for a predicted ribosome pause site in the HSP90AA1 mRNA (Figure 4 I) in nuclease treated l ysates-w hich should be pr otected fr om degradation by paused ribosomes. We found a decrease in signal for the monosome and disome fragments in PKM depleted cells (Figure 4 J, K ). In contrast, mRNA signal from probes against non-pause sites in the HSP90AA1 mRNA was not affected by PKM depletion (Figure 4 J, K ).

Nascent polypeptide chains are PARylated, which is r equir ed for PKM-polysome association
Lysine-rich nascent chains encoded by poly(A)-stretches are known to induce translational stalling ( 65 , 66 ), howe v er, the impact of glutamate-rich nascent chains on translation is unclear. Despite this, four studies have identified glutamate to be enriched near endogenous stall sites (67)(68)(69)(70). Gi v en the disparate chemical characteristics of both lysine and glutamate, we suspected that a transient modification on the nascent polypeptide rather than the underlying amino acid sequence recruited PKM to the ribosome. Both positi v e and negati v ely charged amino acids are known to be the target of a cellular modification known as poly-ADP ribosyla tion (PARyla tion) ( 71 ). Notab ly, a pre vious study found that PKM can directly bind to PARylated polypeptides ( 35 ). Furthermore, PKM was identified in a mass spectrometry screen for PAR-binding proteins ( 72 ). Additionally, many PKM eCLIP targets and mRNAs whose pauses are PKM sensiti v e are significantly enriched for PARylated proteins (Figure 5 A, respecti v e P -values for enrichment: 4.00 × 10 −43 , 3.83 × 10 −18 ). Incidentally, we found the major PARylation polymerase (PARP-1) in our polysome mass spectrometry screen, and further validated this interaction by immunoblot (Supplementary Table S1, Figure 5 B). Probing for PARylated proteins in our polysome pellets re v eals a slow migrating smear near the top of the gel (Figure 5 C). Additionally, we also identified banding pattern akin to ribosomal proteins (Figure 5 C)-consistent with a study indicating that ribosomal subunits are PARylated ( 73 ). Remar kab ly, pr e-tr eating cells with the PARP inhibitor, olaparib, for 15 min eliminates the slow migrating species suggesting that these PARylated substrates have a short half-life, in contrast to the putati v e PARylated ribosomal bands (Figure 5 C). We hypothesized that the slow migrating species consisted of PARylated nascent polypeptides w hose a pparent molecular weight is drastically altered by PARylation. To test this hypothesis, we tagged nascent polypeptides using a biotinylated-puromycin analogue and isolated them via a streptavidin pulldown. We were able to recover a slow migrating PAR-enriched species in our pulldown ( Figure 5 D, PAR blot), which was of higher apparent molecular weight than total nascent polypeptides as detected by probing for biotin ( Figure 5 D). In contrast, we did not detect the numerous low molecular weight PARylated bands in the streptavidin pulldown, concomitant with a loss in ribosomal proteins as detected by ponceau stain (Figure 5 D). Finally, we found that olaparib treatment disrupted the co-sedimentation of PKM with polysomes, suggesting tha t PARyla tion may be r equir ed for PKM binding (Figure 5 E).

DISCUSSION
Her e, we pr esent evidence tha t PKM associa tes with cytosolic polysomes to promote translational pausing in the presence of glucose and pyruvate. We propose a model (Figur e 5 F) wher e under normal conditions, PKM binds to PARylated nascent chains via its ADP-binding site. At this moment, we are unable to distinguish whether PKM stalls translation in cis (i.e. its bound ribosome) and / or in trans (i.e. neighbouring ribosomes). Gi v en the flexibility and length of nascent chains and PAR-chains, both mechanisms may be happening simultaneously. The recruitment of PKM to nascent chains allows it to stall elongation, likely by associating with the A-site of nearby ribosomes as indicated by Simsek et al ( 10 ). This model is reminiscent of how the signal recognition particle inhibits elongation by occluding the A site and thus pre v enting the entry of tRNAs ( 74 ). In agreement with this model, it has been recently found that the E. coli homologue of PKM interacts with the ribosome A site, suggesting that this may be an evolutionary conserved mechanism ( 75 ). Then upon glucose / pyruv ate starv ation, increased le v els of ADP compete with PAR for PKM-association and relie v e this elongation inhibition. This model of ADP regulation is in or Cy clohe ximide (100 g / ml) for 10 min and then lysed and incubated in isotonic (125 mM KCl) or a high salt (500 mM KCl) buffer. Samples were either directly analyzed ('Input') or polysomes were isolated by centrifugation through a sucrose cushion and the pellets were analyzed by immunoblot for PARP-1 and RPS6 le v els. Ponceau was used to monitor total protein. ( C ) U2OS cells were treated with PARP-1 inhibitor Olaparib for 25 min. Polysomes were isolated by sucrose cushion sedimentation and pellets were analyzed by immunoblot for poly-ADP ribosylated (PAR) proteins. ( D ) Polysome pellets were incubated with either puromycin or biotin-tagged puromycin. Streptavidin pulldown was performed to isolate biotin-tagged puromycinylated nascent pol ypeptides. Pulldown anal yzed by imm unoblot and probed for poly-ADP ribosylated nascent chains (PAR), biotin and total proteins monitored by Ponceau. ( E ) U2OS cells treated with PARP-1 inhibitor Olaparib for 25 min. Samples were either directly analyzed ('Input') or polysomes were isolated by sucrose cushion sedimentation and pellets were analyzed by immunoblot for PKM, RPL4 and total proteins monitored by Ponceau. PKM signal was quantified relati v e to RPL4 le v els in ribosome pellets. Each bar r epr esents the mean of at least thr ee biological r eplicates plotted alongside standard error, as whiskers, and individual replicates, as dots. P -value calculated from paired Student's t -test. ( F ) Model depicting the recruitment of PKM to polysomes and their subsequent fate. agreement with previous observa tions tha t ADP and PAR bind to the same pocket on PKM ( 35 ), and this is likely due to fact that PAR chains are in part composed of ADPmoieties.
Our da ta indica tes tha t this PKM-dependent regula tion is restricted to cytosolic pol ysomes. Notabl y, ER-associated polysomes have their nascent chains translocated into the ER lumen and thus would not be accessible to either PARP-1 or PKM. This is in agreement with our observations that PKM does not associate with ER-associated polysomes (Figure 1 G). While our model suggests that nascent chains are co-transla tionally PARyla ted, we are unable to rule out an indirect effect gi v en that an increase in PAR polymers can affect cellular metabolism via a decrease in NAD+ lev-els ( 76 ). While our PKM2 eCLIP targets do not completely overlap with the set of PARylated proteins from Martello et al., we suspect that this may reflect the difficulties of detecting PARyla tion ra ther than an absence of PARylation, especially those occurring on nascent polypeptides. It must be noted that a previous group had reported ADP-ribosylation activity associated with cellular ribosomes that could be partially reduced by inhibiting protein synthesis ( 77 ).
We suspect that elongation rates may affect protein folding and that both of these processes r equir e acti v e feedback from gl ycol ysis. Under normal conditions, PARylation may allow poorly folded domains to remain soluble as they exit the ribosome and may induce translational stalling through PKM to further promote proper folding of the nascent polypeptides. This is analogous to how protein glycosylation may help to fold proteins in the lumen of the ER ( 78 ). Poorly folded nascent polypeptides would e v entually promote PKM-dependent ribosome collisions and thus activ ate cleav age of the translated mRNAs and the decay of partially synthesized unfolded nascent chains. Notabl y, PAR accum ulates in stress granules ( 79 ), large cytoplasmic biomolecular condensates of mRNA and RNAbinding proteins that form in response to cellular stress ( 80 ).
Howe v er, under starvation conditions when ADP le v els rise, we specula te tha t this process is inactivated to allow the cell to conserve energy by pre v enting the decay of energeticall y costl y mRN As and nascent pol ypeptides. Consistent with this model is the observation that nearly a quarter of newly synthesized cytosolic proteins are rapidly degraded ( 81 ). Furthermore, it has been shown that PKM depletion enhances sensitivity to proteasome inhibition in mammalian cells by mediating the formation of the CHIP-HSP70-BAG3 complex with ubiquitinated proteins ( 82 ). Since BAG3 has been shown to associate with polypeptides as they exit the ribosome, this finding is in agreement with our model that PKM is r equir ed to eliminate poorly folded nascent proteins.
Overall, PKM depletion induces a variety of cellular phenotypes that include d ysregula ted carbohydra te metabolism, mitotic defects, DNA damage defects and mRNA translation ( 10 , 30 , 32 , 34 , 35 , 83-87 ), suggesting that it may have many roles in se v eral distinct cellular processes. It is possible that our newly characterized role of PKM may explain how these processes are linked. Among the transcripts whose ribosome occupancy are PKM-sensiti v e and / or interact with PKM by eCLIP, we find an enrichment for proteins involved in DN A repair, DN A replication, and mitotic regulation (Supplementary Table S3 and  S4). Furthermore, the PKM-ribosome interaction is lost during mitosis (Supplementary Figure S5D), at the exact same time when ribosome-associa ted PARyla tion drops ( 77 )-suggesting that elongation of mitotic-encoding transcripts may be enhanced during cell division. Additionally, it must be noted that PARylation has a close relationship with DNA damage as it is an important transient modification which acts to recruit DNA repair factors to damaged DNA regions. It has been shown that DNA-damage can reduce protein synthesis ( 73 ) and trigger stress granule assembly, a hallmark of reduced mRNA translation ( 88 , 89 ). Interestingl y, DN A damage has been shown to recruit PKM to the nucleus in a PAR-dependent manner ( 35 ). Whether PKM and / or PAR mediate how DNA damage triggers the r epr ession of mRNA translation r emains to be elucidated.
The ribosome profile sequencing data can be access through the GEO database (GSE202881).