Grad-seq shines light on unrecognized RNA and protein complexes in the model bacterium Escherichia coli

Abstract

Stable protein complexes, including those formed with RNA, are major building blocks of every living cell. Escherichia coli has been the leading bacterial organism with respect to global protein-protein networks. Yet, there has been no global census of RNA/protein complexes in this model species of microbiology. Here, we performed Grad-seq to establish an RNA/protein complexome, reconstructing sedimentation profiles in a glycerol gradient for ∼85% of all E. coli transcripts and ∼49% of the proteins. These include the majority of small noncoding RNAs (sRNAs) detectable in this bacterium as well as the general sRNA-binding proteins, CsrA, Hfq and ProQ. In presenting use cases for utilization of these RNA and protein maps, we show that a stable association of RyeG with 30S ribosomes gives this seemingly noncoding RNA of prophage origin away as an mRNA of a toxic small protein. Similarly, we show that the broadly conserved uncharacterized protein YggL is a 50S subunit factor in assembled 70S ribosomes. Overall, this study crucially extends our knowledge about the cellular interactome of the primary model bacterium E. coli through providing global RNA/protein complexome information and should facilitate functional discovery in this and related species.

INTRODUCTION

RNA-protein interactions leading to the formation of stable cellular complexes underlie many processes that are conserved in all domains of life. In bacteria, these traditionally range from small complexes such as the ∼85 kDa signal recognition particle (SRP) to the ∼450 kDa RNA polymerase (RNAP) and the giant 70S ribosome, which is a ∼2.5 MDa assembly of three ribosomal RNA species and >50 ribosomal proteins (1). CRISPR/Cas systems involved in genome defense are a recent addition to this list and keep surprising with unexpected molecular diversity (2). In contrast with eukaryotic biology where there have been large-scale initiatives to systematically map RNA-protein interactions and complexes in different model species, it is fair to say that we have an incomplete knowledge of the breadth of macromolecular complexes in bacteria, even in primary model species such as Escherichia coli.

Escherichia coli K-12 has been the workhorse not only for bacterial genetics and physiology but also for the identification and functional characterization of complexes of bacterial proteins, with or without RNA. There is a large body of functional in-depth studies in E. coli employing low-throughput biochemical assays, results of which are organized and accessible via curated databases such as EcoCyc (3). In addition, E. coli has been subject to large-scale affinity purification followed by mass spectrometry (AP/MS) in order to determine candidate cellular networks of protein-protein interactions (4–7). Similarly, yeast two-hybrid screens have predicted binary interactions for nearly all possible combinations within the E. coli proteome (8). Beyond these binary methods, the E. coli complexome has been studied with global methods directly, i.e., without epitope-tagging or heterologous expression of proteins (9–12). This includes recent protein-correlation-profiling, whereby complexes of membrane proteins were predicted following the ‘guilt-by-association’ logic after their separation by size exclusion chromatography (13).

An important information that has been missing is which E. coli transcripts and proteins engage in stable cellular complexes, both determined in the same experiment. Such predictions can now be made in a genome-wide manner by combined high-throughput RNA-seq and mass spectrometry (MS) analyses of a bacterial lysate after its fractionation in a glycerol gradient. Referred to as gradient profiling by sequencing (Grad-seq) (14), this method predicts RNA/protein complexes by looking for co-migration of a given RNA and protein in the same gradient fraction (15). In its pioneering application (14), Grad-seq uncovered in Salmonella enterica the ProQ/FinO-domain protein ProQ as a previously overlooked global RNA-binding protein (RBP) that impacts the expression of several hundred different transcripts (16–18).

Here, we use Grad-seq to provide an RNA/protein complexome resource for E. coli. By reporting sedimentation profiles for the vast majority of cellular transcripts and half of the E. coli proteins, we faithfully reproduce hundreds of known protein-protein and RNA–protein interactions including major ribonucleoprotein particles (RNPs). We describe use cases for this resource to be exploited: a stable association with the 30S ribosomal subunit reveals the seemingly noncoding RyeG RNA of prophage origin as an mRNA that encodes a small toxic protein; the broadly conserved uncharacterized protein YggL is revealed to be a 50S subunit factor in assembled 70S ribosomes. The E. coli Grad-seq data are viewable in an online browser (https://helmholtz-hiri.de/en/datasets/gradseqec/) that not only enables easy access to all recorded RNA and protein gradient sedimentation profiles but also permits cross-species comparison with published Salmonella enterica (14,19) and Streptococcus pneumoniae (20) datasets.

MATERIALS AND METHODS

Bacteria and media

For all experiments, E. coli MG1655 was streaked on LB plates and grown overnight at 37°C. Overnight cultures of single colonies were grown in 2 ml LB at 37°C with shaking at 220 rpm. The next day, 1:100 dilutions in fresh LB of the overnight cultures were used to start the cultures for the experiments and grown to an OD₆₀₀ of 2.0 (early stationary phase) at 37°C with shaking at 220 rpm. All strains and plasmids are listed in Supplementary Tables S3 and S4.

Strain construction

Gene inactivation mostly followed a published protocol (21). Briefly, a strain carrying the pKD46 plasmid, which carries the λRED recombinase and is temperature-sensitive, was grown overnight at 28°C. The next day, the overnight culture was diluted 1:300 in 50 ml LB containing 0.2% l-arabinose and grown at 28°C to an OD₆₀₀ of 0.5. Electrocompetent cells were prepared and transformed with 800–1000 ng of a gel-purified PCR product containing a kanamycin resistance cassette. The PCR product was obtained using the pKD4 plasmid and primers containing the flanking regions of the gene of interest. The transformed cells were streaked on LB agar plates and incubated at 37°C overnight. The deletion was verified by PCR. 3xFLAG-tagging of genes followed a published protocol (22), which is similar to the one described for gene inactivation above, except that the PCR product was obtained using the pSUB11 plasmid. Removal of the antibiotics resistance cassettes was performed by transformation of the temperature-sensitive pCP20 plasmid that carries the Flp recombinase (21). All oligo nucleotides are listed in Supplementary Table S5.

Glycerol gradient fractionation

Glycerol gradient fractionation was performed as previously described (20), with the exception of the growth and lysis conditions: 100 ml of E. coli MG1655 wild type were grown to an OD₆₀₀ of 2.0, cooled down in an ice-water bath for 15 min and then harvested by centrifugation for 20 min at 4°C and 4000 rcf. The cells were washed three times in ice-cold 1x TBS, resuspended in 500 μl ice-cold 1× lysis buffer A [20 mM Tris–HCl, pH 7.5, 150 mM KCl, 1 mM MgCl₂, 1 mM DTT, 1 mM PMSF, 0.2% Triton X 100, 20 U/ml DNase I (Thermo Fisher), 200 U/ml RNase inhibitor] and lysed by addition of 750 μl of 0.1 mm glass beads (Carl Roth) and 10 cycles of vortexing for 30 s followed by cooling on ice for 15 s. To remove insoluble debris and the glass beads, the lysate was cleared by centrifugation for 10 min at 4°C and 16 100 rcf.

Of the cleared lysate, 10 μl was mixed with 1 ml TRIzol (Thermo Fisher) for the RNA input control and 20 μl was mixed with 20 μl 5× protein loading buffer for the protein input control. 200 μl of the cleared lysate was then layered on top of a linear 10–40% (w/v) glycerol gradient (in 1× lysis buffer A without DNase I or RNase inhibitor), which was formed in an open-top polyallomer ultracentrifugation tube (Seton Scientific) using the Gradient Station model 153 (Biocomp). The gradient was centrifuged for 17 h at 4°C and 100 000 rcf (23 700 rpm) using an SW 40 Ti rotor (Beckman Coulter), followed by manual fractionation into 20 590 μl fractions and measurement of the A_{260 nm} of each fraction. 90 μl of each fraction and 40 μl of the pellet were mixed with 30 μl of 5× protein loading buffer for protein analysis and stored at −20°C.

The remaining 500 μl of each fraction were used for RNA isolation by addition of 50 μl of 10% SDS (25 μl for the pellet) and 600 μl of acidic phenol/chloroform/isoamylalcohol (P/C/I; 300 μl for the pellet). The fractions were then vortexed for 30 s and let rest at room temperature for 5 min before separating the phases by centrifugation for 15 min at 4°C and 16 100 rcf. The aqueous phases were collected, 1 μl GlycoBlue (Thermo Fisher) and 1.4 ml of ice-cold ethanol/3 M sodium acetate, pH 6.5 (30:1) were added and precipitated for at least 1 h at −20°C. The RNA was collected by centrifugation for 30 min at 4°C and 16 100 rcf and washed with 350 μl ice-cold 70% ethanol, followed by centrifugation for 15 min at 4°C and 16 100 rcf. The lysate RNA sample stored in TRIzol was purified according to the manufacturer's protocol, except that the precipitation was performed using the mentioned ethanol mix. After drying of the RNA pellet, it was dissolved in 40 μl DEPC-treated H₂O and DNase-digested by addition of 5 μl DNase I buffer with MgCl₂ (Thermo Fisher), 0.5 μl RNase inhibitor, 4 μl DNase I and 0.5 μl DEPC-treated H₂O, followed by incubation for 45 min at 37°C. The DNase-treated RNA was purified by addition of 150 μl DEPC-treated H₂O and 200 μl P/C/I as described above. The purified, DNase-treated RNA was dissolved in 35 μl DEPC-treated H₂O and stored at −80°C.

Sucrose polysome gradient fractionation

50 ml of E. coli MG1655 was grown to an OD₆₀₀ of 2.0, followed by rapid filtration and immediate freezing in liquid nitrogen. The cells were then resuspended in 1 ml of ice-cold 1× lysis buffer B [20 mM Tris–HCl, pH 7.5, 100 mM NH₄Cl, 10 mM MgCl₂, 1 mM DTT, 1 mM PMSF, 0.4% Triton X 100, 20 U/ml DNase I, 200 U/ml RNase-inhibitor] and lysed using a FastPrep-24 instrument (MP Biomedicals) and a 2 ml lysing matrix E tube (MP Biomedicals) for 15 s at 4 m/s. To remove insoluble debris and the beads, the lysate was cleared by centrifugation for 10 min at 4°C and 16 100 rcf. Of the cleared lysate, 10 μl was mixed with 1 ml TRIzol for the RNA input control. Fifteen A_{260 nm} per ml of the cleared lysate were then layered on top of a linear 10–55% (w/v) sucrose gradient (in 1× lysis buffer B without DNase I or RNase inhibitor and with addition of 5 mM CaCl₂), which was formed in an open-top polyclear ultracentrifugation tube (Seton Scientific) using the Gradient Station model 153. The gradient was centrifuged for 2.5 h at 4°C and 237 000 rcf (35 000 rpm) using an SW 40 Ti rotor, followed by automated fractionation into 20 fractions using an FC 203B fractionator (Gilson). RNA extraction was performed as for the glycerol gradient, except that the vortexing step was performed for 15 s and that DNase treatment of the purified RNA was skipped.

RNA gel electrophoresis and northern blotting

Equal volumes of the gradient RNA samples (glycerol or sucrose) were separated by 6% denaturing PAGE in 1× TBE and 7 M urea and stained with ethidium bromide. For northern blotting, unstained gels were transferred to Hybond+ membranes (GE Healthcare Life Sciences) and probed with RNA-specific radioactively labeled DNA oligonucleotides.

Protein gel electrophoresis and western blotting

Equal volumes of the gradient protein samples were separated by 12% SDS-PAGE and stained with Coomassie. For western blotting, unstained gels were transferred to PVDF membranes (GE Healthcare Life Sciences) and probed with protein-specific antisera against 3xFLAG (Sigma-Aldrich, cat# F1804), 6xHis (Sigma-Aldrich, cat# H1029), GroEL (Sigma-Aldrich, cat# G6532), RpoB (BioLegend, cat# 663905) or RpoD (BioLegend, cat# 663202). Visualization of the primary antibodies was performed using anti-mouse (Thermo Fisher, cat# 31430) or anti-rabbit (Thermo Fisher, cat# 31460) secondary antibodies.

RNA-seq

RNA-seq was performed as described before (20). Briefly, 5 μl of the gradient samples were diluted in 45 μl DEPC-treated H₂O. 10 μl of the resulting 1:10 dilution were mixed with 10 μl of a 1:100 dilution of the ERCC spike-in mix 2 (Thermo Fisher) and subjected to library preparation for next-generation sequencing (vertis Biotechnologie). Briefly, the RNA samples were fragmented using ultrasound (4 pulses of 30 s at 4°C) followed by 3′ adapter ligation. Using the 3′ adapter as primer, first strand cDNA synthesis was performed using M-MLV reverse transcriptase. After purification, the 5′ Illumina TruSeq sequencing adapter was ligated to the 3′ end of the antisense cDNA. The resulting cDNA was PCR-amplified to about 10–20 ng/μl using a high-fidelity DNA polymerase followed by purification using the Agencourt AMPure XP kit (Beckman Coulter). The cDNA samples were pooled with ratios according to the RNA concentrations of the input samples and a size range of 200–550 bp was eluted from a preparative agarose gel. This size-selected cDNA pool was finally subjected to sequencing on an Illumina NextSeq 500 system using 75 nt single-end read length.

RNA-seq data analysis

Pre-processing steps like read trimming and clipping were done with cutadapt (23). Read filtering, read mapping, nucleotide‐wise coverage calculation and genome feature‐wise read quantification was done using READemption (24) (v0.4.3; https://doi.org/10.5281/zenodo.250598) and the short read mapper segemehl (25) (v0.2.0‐418) with the Escherichia coli MG1655 genome (accession number: NC_000913.3) as reference. The annotation provided was extended by ncRNAs predicted by ANNOgesic (26). The analysis was performed with the tool GRADitude (Di Giorgio, S., Hör, J., Vogel, J., Förstner, K.U., unpublished; v0.1.0; https://foerstner-lab.github.io/GRADitude/). Only transcripts with a sum of ≥100 reads in all fractions within the gradient were considered for the downstream analyses. Read counts for each fraction were normalized by calculating size factors following the DESeq2 approach (27) generated from the ERCC spike‐in read counts added to each sample (see above). To remove left‐over disturbances in the data, the size factors were then manually adjusted by multiplication based on quantified northern blots: 1.5 (fraction 5), 4.5 (fraction 7) and 28 (fraction 8). In order to make all the transcript counts comparable, they were scaled to the maximum value.

After normalization, analyses based on the detected transcripts were performed. t‐SNE dimension reduction (28) was performed using the Python package scikit‐learn (29). All default parameters provided by the sklearn.manifold.TSNE class were used. A file collection representing the analysis workflow, including Unix Shell calls, Python scripts, documentation as well as resulting files have been deposited at Zenodo (https://doi.org/10.5281/zenodo.3876866). Files related to the Grad-seq browser have been deposited at Zenodo as well (https://doi.org/10.5281/zenodo.3955585).

Sample preparation for mass spectrometry

Sample preparation for mass spectrometry was performed as described before (20). Briefly, the gradient protein samples (diluted in 1.25× protein loading buffer) were homogenized using ultrasound [5 cycles of 30 s on followed by 30 s off, high power at 4°C (Bioruptor Plus, Diagenode)]. Insoluble material was then removed by centrifugation for 15 min at 4°C and 16 100 rcf. 20 μl of the cleared protein sample were mixed with 10 μl of UPS2 spike-in (Sigma-Aldrich) diluted in 250 μl 1.25× protein loading buffer. The samples were then reduced in 50 mM DTT for 10 min at 70°C and alkylated with 120 mM iodoacetamide for 20 min at room temperature in the dark. The proteins were precipitated in four volumes of acetone overnight at −20°C. Pellets were washed four times with acetone at −20°C and dissolved in 50 μl 8 M urea, 100 mM ammonium bicarbonate. Digestion of the proteins was performed by addition of 0.25 μg Lys-C (Wako) for 2 h at 30°C, followed by dilution to 2 M urea by addition of 150 μl 100 mM ammonium bicarbonate and overnight digestion with 0.25 μg trypsin at 37°C. Peptides were desalted using C-18 Stage Tips (30). Each Stage Tip was prepared with three disks of C-18 Empore SPE Disks (3M) in a 200 μl pipet tip. Peptides were eluted with 60% acetonitrile/0.3% formic acid, dried in a laboratory freeze-dryer (Christ) and stored at −20°C. Prior to nanoLC-MS/MS, the peptides were dissolved in 2% acetonitrile/0.1% formic acid.

NanoLC–MS/MS analysis

NanoLC–MS/MS analysis was performed as described before (20) using an Orbitrap Fusion (Thermo Scientific) equipped with a PicoView Ion Source (New Objective) and coupled to an EASY-nLC 1000 (Thermo Scientific). Peptides were loaded on capillary columns (PicoFrit, 30 cm × 150 μm ID, New Objective) self-packed with ReproSil-Pur 120 C18-AQ, 1.9 μm (Dr Maisch) and separated with a 140 min linear gradient from 3% to 40% acetonitrile and 0.1% formic acid at a flow rate of 500 nl/min. Both MS and MS/MS scans were acquired in the Orbitrap analyzer with a resolution of 60 000 for MS scans and 15 000 for MS/MS scans. HCD fragmentation with 35% normalized collision energy was applied. A Top Speed data-dependent MS/MS method with a fixed cycle time of 3 s was used. Dynamic exclusion was applied with a repeat count of 1 and an exclusion duration of 60 s; singly charged precursors were excluded from selection. Minimum signal threshold for precursor selection was set to 50 000. Predictive AGC was used with a target value of 2 × 10⁵ for MS scans and 5 × 10⁴ for MS/MS scans. EASY-IC was used for internal calibration.

MS data analysis

MS data analysis was performed as described before (20), with a few exceptions. Raw MS data files were analyzed with MaxQuant version 1.5.7.4 (31). The search was performed against the UniProt database for E. coli MG1655 (organism identifier: ECOLI), a database containing the UPS2 spike-in and a database containing common contaminants. The search was performed with tryptic cleavage specificity with 3 allowed miscleavages. Protein identification was under control of a false-discovery rate of 1% on both protein and peptide level. In addition to the MaxQuant default settings, the search was performed against the following variable modifications: Protein N-terminal acetylation, Gln to pyro-Glu formation (N-terminal Q) and oxidation on Met. For protein quantitation, the LFQ intensities were used (32). Proteins with <2 identified razor/unique peptides were dismissed.

Normalization of the proteins across the fractions was performed using the UPS2 spike-in. For this, only spike-in proteins with detectable intensities in all fractions were used. The spike-in proteins showing the highest variance (median average deviation of log₁₀ intensities >1.5× lQR) were eliminated. Following this, for each spike-in protein, the median log₁₀ intensity was subtracted from the log₁₀ intensities of each fraction. The fraction-wise median of the resulting values was then subtracted from the log₁₀ intensities for each bacterial protein in the corresponding fractions. Finally, all log₁₀ intensities smaller than the 5% quantile of all intensities in the data set were replaced by the value of the 5% quantile of all intensities in the data set.

Estimation of in vivo RNA copy numbers

Estimation of in vivo copy numbers of RyeG was performed as described previously (33). Briefly, RNA was extracted at the given time points by collecting 4 OD₆₀₀ of cells. The RNA was diluted in 40 μl water and 10 μl from each time point (≈10⁹ cells) was probed on a northern blot. For reference, in vitro-transcribed RyeG was loaded (0.05, 0.1, 0.5, 1 and 2.5 ng). RNA levels per cell were based on determination of viable cell counts per OD₆₀₀ as described in (34).

30S subunit toeprinting analysis

30S subunit toeprinting was performed as previously published (35,36) with few changes. Briefly, 0.2 pmol unlabeled, in vitro-transcribed RyeG and 0.5 pmol of a 5′-labeled DNA oligonucleotide (JVO-16833) were denatured for 1 min at 95°C in the presence of 0.8 μl SB 5× -Mg (50 mM Tris-acetate, pH 7.6, 500 mM potassium acetate, 5 mM DTT) in a total volume of 3 μl. After incubation on ice for 5 min, 1 μl dNTPs (5 mM each) and 1 μl SB 1× Mg60 (10 mM Tris-acetate, pH 7.6, 100 mM potassium acetate, 1 mM DTT, 60 mM magnesium acetate) were added and the samples were incubated for 5 min at 37°C. Next, 4 pmol purified 30S subunits (pre-activated for 20 min at 37°C) was added to the samples [SB 1× Mg10 (10 mM Tris-acetate, pH 7.6, 100 mM potassium acetate, 1 mM DTT, 10 mM magnesium acetate) was added to the control]. After incubation for 5 min at 37°C, 10 pmol uncharged fMet-tRNA^Met_i was added to the corresponding sample. Reactions were continued at 37°C for 15 min, followed by addition of 100 U SuperScript II reverse transcriptase (Thermo Fisher) and incubation for 20 min at 37°C.

Reactions were stopped by addition of 100 μl toeprint stop buffer (50 mM Tris–HCl, pH 7.5, 0.1% (w/v) SDS, 10 mM EDTA, pH 8). DNA was extracted by addition of 110 μl P/C/I. Next, 5 μl 3 M KOH was added and the RNA digested at 90°C for 5 min. 10 μl 3 M acetic acid, 1 μl GlycoBlue and 300 μl ethanol/3 M sodium acetate, pH 6.5 (30:1) were added and the DNA precipitated at −20°C overnight. Extraction was finished and the pellet washed once with 100 μl of 70% ethanol. The purified pellet was dissolved in 10 μl 1× RNA loading buffer, denatured for 3 min at 90°C and subjected to separation using a denaturing 8% sequencing gel in presence of a RyeG-specific sequencing ladder prepared using the DNA Cycle Sequencing kit (Jena Bioscience) according to the manufacturer's instructions. Gels were run for 1.5 h at 40 W, dried and exposed on a phosphor screen.

Purification of ribosomes

Crude purification of ribosomes mostly followed a previously published protocol (37). Briefly, 800 ml of an E. coli ΔyggL culture was grown to an OD₆₀₀ of ∼0.5–0.7 and washed once with 25 ml of ice-cold 1× TBS. The cell pellets were snap-frozen in liquid nitrogen and stored at −80°C. The pellets were then resuspended in 6 ml ice-cold lysis buffer C (20 mM Tris×HCl, pH 7.5, 100 mM NH₄Cl, 10.5 mM MgCl₂, 0.5 mM EDTA, 3 mM DTT) on ice. Lysis was performed by two lysis steps using a french press at 10 000 psi. 75 μl of 100 mM PMSF was added and the lysates cleared by centrifugation for 30 min at 4°C and 30 000 rcf using an SW 40 Ti rotor. 12.5 ml of the supernatant was subsequently layered on top of a 12.5 ml 1.1 M sucrose cushion made up in lysis buffer C. Next, the sample was centrifuged for 16 h at 4°C and 100 000 rcf using a type 70 Ti rotor (Beckman Coulter). The pellet was gently washed with 500 μl of storage buffer (lysis buffer C + 10% (v/v) glycerol) and finally dissolved in 1 ml of storage buffer by gentle shaking for 2.5 h at 4°C. After centrifugation for 5 min at 16 100 rcf and 4°C, the concentration was measured, the purified ribosomes aliquoted, snap-frozen in liquid nitrogen and stored at −80°C.

Purification of YggL

The purification of recombinant YggL was performed by the Recombinant Protein Expression core unit at the Rudolf Virchow Center, University of Würzburg, Germany. Briefly, to purify recombinantly expressed Cbf1, the CDS was cloned into pETM-14, which adds an N‐terminal His‐tag and a 3C protease cleavage site to the sequence. After transformation into E. coli BL21, cells were grown to an OD₆₀₀ of 0.65 and induced with 0.5 mM IPTG for 3.5 h at 30°C. After centrifugation, the cell pellet was dissolved in 20 mM Tris–HCl, pH 7, 300 mM NaCl, 0.4 mM PMSF and 1.5 U/ml DNase I and subjected to affinity purification using Protino Ni‐NTA Agarose (Macherey Nagel). Following elution, the eluate was further purified by size exclusion chromatography using a Superdex 16/600 column (GE Healthcare Life Sciences). Purified YggL was stored in 50 μl aliquots of 20 mM Tris–HCl, pH 7, 300 mM NaCl and 10% glycerol at −80°C (∼0.9 mg/ml).

Analysis of in vitro-reconstituted complexes

To test binding of YggL to purified ribosomes, given amounts of recombinant YggL were mixed with purified ribosomes extracted from the ΔyggL strain. Then, the volume was increased to 200 μl with lysis buffer C and the samples were incubated for 10 min at 30°C with shaking at 330 rpm to allow complex formation. The samples were subsequently loaded on 10–40% (w/v) sucrose gradients (in 20 mM Tris–HCl, pH 7.5, 100 mM NH₄Cl, 10 mM MgCl₂, 3 mM DTT), which were formed in open-top polyclear tubes. Gradients were centrifuged for 14 h at 4°C and 71 000 rcf (20 000 rpm) using an SW 40 Ti rotor. Fractionation was performed as described for the sucrose gradients above.

Co-immunoprecipitation of YggL followed by MS

Cells representing 50 OD₆₀₀ of the yggL-3xFLAG or wild-type strains were collected and washed once with 1 ml of lysis buffer A. After resuspension in 800 μl lysis buffer A, the cells were transferred to a 2 ml FastPrep tube with lysing matrix E and lysed using a FastPrep-24 instrument for 20 s at 4 m/s. The lysate was cleared for 10 min at 16 100 rcf and 4°C. 40 μl of magnetic protein A/G beads (Thermo Fisher) were washed with 1 ml of lysis buffer A, resuspended in 400 μl lysis buffer A and 3 μl anti-FLAG antibody was added. After rotating for 45 min at 4°C, the beads were washed twice with 400 μl lysis buffer A. 600 μl of the lysate was added to the beads with the coupled antibody and rotated for 1.5 h at 4°C. The beads were washed five times with 400 μl lysis buffer A and briefly spun down. The lysis buffer was removed, the beads were resuspended in 35 μl 1× LDS sample buffer (Thermo Fisher) with 50 mM DTT and the proteins eluted by incubation at 95°C for 5 min.

The samples were alkylated in presence of 120 mM iodoacetamide for 20 min in the dark and run on a precast 4–12% Bolt Bis–Tris plus gel (Thermo Fisher) using 1x MES buffer (Thermo Fisher). The gel was stained with SimplyBlue Coomassie (Thermo Fisher) and each lane of the gel was cut into 15 pieces. To prepare the gel pieces for LC/MS-MS, they were destained with 30% acetonitrile in 100 mM ammonium bicarbonate, pH 8. Next, the pieces were shrunk using 100% acetonitrile and dried. Digestion was performed by addition of 0.1 μg trypsin per gel piece and incubation overnight at 37°C in 100 mM ammonium bicarbonate, pH 8. The supernatant was removed and the peptides were extracted from the gel pieces with 5% formic acid. Finally, the supernatant was pooled with the extracted peptides and subjected to MS.

RESULTS

Grad-seq reveals sedimentation of the soluble RNA and protein content of E. coli

We performed Grad-seq on E. coli grown to early stationary phase (OD₆₀₀ of 2.0) in rich medium, to be consistent with our previous Salmonella Grad-seq study (14) and other RBP-related data sets for E. coli and Salmonella from this laboratory (17,18,38,39). Inferring from Salmonella, close to 70% of genes are robustly expressed in this growth phase (40). As before (14,20), soluble particles from a lysate were separated on a linear 10–40% glycerol gradient, followed by RNA-seq and MS analyses of all 20 gradient fractions and the pellet (Figure 1A). The A_{260 nm} UV profile of the gradient showed the typical three peaks, one bulk peak around low molecular weight (LMW) fraction 2 and the two peaks representing the small (30S) and large (50S) ribosomal subunits (Figure 1B) (14,20). Fully assembled 70S ribosomes as well as inactive 100S ribosomes (41,42) sedimented in the pellet (P).

Figure 1.

Grad-seq reveals the E. coli RNA/protein complexome. (A) Overview of the Grad-seq workflow. (B) A_{260 nm} profile of the gradient. Low‐molecular‐weight complexes (bulk peak) and ribosomal subunits (30S, 50S) are highlighted. Particles larger than the 50S subunit were pelleted. (C) Ethidium bromide‐stained RNA gel. Bands corresponding to abundant housekeeping RNAs are indicated. (D) Coomassie‐stained SDS-PAGE. Bands corresponding to abundant housekeeping proteins are indicated. (E) Western blot. The β-subunit of RNAP (RpoB) and the major σ factor σ⁷⁰ (RpoD) co-migrate. (F) Heat map of digital in‐gradient distributions of known RNA-protein complexes derived from RNA‐seq and LC–MS/MS data. For each molecule, the spike-in-normalized sedimentation profiles are normalized to the range from 0 to 1 by dividing the values of each fraction by the maximum value of the corresponding molecule. M, size marker. L, lysate (input control). P, pellet fraction. (G) Sucrose polysome gradient of a wild-type lysate followed by northern blotting. 6S RNA, ChiX and CsrB are only present in the bulk peak, whereas GcvB and Spot 42 show additional abundances around the polysomes (compare to (F)). The lpp mRNA is only present in ribosomal fractions.

RNA gel analysis showed tRNAs, 16S rRNA and the 5S/23S rRNAs to individually overlap with those three major UV peaks, as expected (Figure 1C). Other abundant house-keeping RNAs forming stable RNPs such as 6S RNA, tmRNA or RnpB (the RNA part of RNase P (43)), were also readily detected in their expected fractions (Supplementary Figure S1A and (14)), confirming RNA integrity. Likewise, SDS-PAGE of the extracted proteins confirmed intactness of several of those major RNPs, exemplified by co-sedimentation of RNA polymerase (RNAP) proteins with 6S RNA (44) and ribosomal proteins with rRNAs (Figure 1D). The position of RNAP was further refined by western blot detection of its β-subunit (RpoB) and its major σ factor, σ⁷⁰ (RpoD) (Figure 1E).

Digital Grad-seq recovers the majority of E. coli transcripts and proteins

RNA-seq of the 20 fractions and the pellet reported the sedimentation profiles of 4095 transcripts, comprising 3699 mRNAs, 287 ncRNAs and all tRNAs and rRNAs (Supplementary Table S1). Similar to Salmonella Grad-seq (14), mRNAs strongly co-sedimented with the 70S ribosome and were additionally present throughout the gradient, suggesting decay products or possibly the presence of a non-translated mRNA population that might form non-ribosomal RNPs upon entering stationary phase (Supplementary Figure S1B). By contrast, ncRNAs showed disparate behaviors (Supplementary Figure S1C). The well-known class of Hfq-binding sRNAs (45) mostly sedimented in medium-sized complexes around fraction 5, with an additional peak in the pellet (Supplementary Figure S1D). ProQ-binding sRNAs peaked earlier, around fraction 4 (Supplementary Figure S1E). In contrast, the CsrA antagonists CsrB and CsrC were confined to the gradient region around fraction 5 (Supplementary Figure S1F). Northern blotting confirmed the sedimentation profiles of the RNA-seq (average Spearman's ∼0.88), thereby verifying our dataset (Supplementary Figure S2).

Parallel MS analysis detected a total of 2145 proteins with high confidence, representing ∼49% of the proteome as annotated on UniProt (46) (Supplementary Table S2). The majority of proteins sedimented in LMW fractions, suggesting involvement in no or small complexes (Supplementary Figure S1G). Of all detected proteins, ∼71% have known or predicted cytoplasmic and periplasmic localization, indicating enrichment of soluble proteins during our sample preparation (Supplementary Figure S3A and B).

Grad-seq visualizes RNPs

The possibility of analyzing complex formation of RNAs is one of the main benefits of Grad-seq compared to other methods (15). RNAP consisting of the α-, β-, β’- and ω-subunits (RpoA, RpoB, RpoC and RpoZ, respectively) co-migrated with the noncoding 6S RNA (Figure 1F), which controls transcription by competing for promoter binding of RNAP-σ⁷⁰ (47). We note that in the MS analysis σ⁷⁰ (RpoD) showed an intriguing second peak around fraction 10, outside RNAP, which was not detected in western blot analysis (Figure 1E). In addition, while σ⁷⁰, σ²⁴ (RpoE) and σ²⁸ (FliA) generally occurred in the same fractions as did RNAP, this was only true for a fraction of the measured σ⁵⁴ (RpoN) and σ^S (RpoS) intensities (Figure 1F). The ribosomal subunits, the SRP (consisting of 4.5S RNA and Ffh (48)) and the SmpB-tmRNA RNP (49) are other major RNPs for which the RNA and protein components showed excellent correlation. In contrast, RnpA being the protein factor of RNase P (43) was barely detected in the first 3 fractions, away from RnpB. This resembles previous results with Salmonella Grad-seq (14) and argues that RNase P has a tendency to disintegrate under the present Grad-seq conditions.

The three major regulatory RBPs of E. coli—CsrA, Hfq and ProQ—are known to bind specific subsets of ncRNAs (16,18,50–51). CsrA exhibited a broader peak, perhaps caused by its associations with diverse target mRNAs (52) in addition to its complexes with the major CsrB and CsrC RNAs (Figure 1F). The absence of CsrA from the pellet fraction (containing 70S ribosomes) underscores its main function as an RBP that inhibits mRNA translation.

Hfq is responsible for most sRNA-based regulation in E. coli (45,53,54). In our Grad-seq data, Hfq showed peaks in fraction 4 and the pellet (Figure 1F), echoing early biochemical results from even before the sRNA-related major function of Hfq was discovered (55). In general, this pattern was also seen with the Hfq-binding sRNAs (Supplementary Figure S1D). Individual sRNAs, however, exhibited disparate sedimentation profiles (Figure 1F). For example, ChiX, which is both abundant and perhaps the strongest Hfq binder (56), peaked in fraction 4 and was found in the pellet to some degree. In contrast, GcvB was almost exclusively present in the pellet fraction. This suggested that GcvB was preferentially associated with ribosomes and/or ribosome-associated Hfq. Testing this prediction, we probed for GcvB on a northern blot of a sucrose instead of a glycerol gradient, and indeed found this sRNA to be abundant in both the 70S monosome and the polysome fractions (Figure 1G).

ProQ is the least understood of the three sRNA-associated RBPs. In our E. coli Grad-seq data, ProQ-binding sRNAs showed a higher average abundance toward the top of the gradient around fraction 4 (Supplementary Figure S1E), which is also where ProQ was found to peak (Figure 1F). Antitoxins of type I toxin-antitoxin (TA) systems are noncoding antisense RNAs that form a well-characterized class of ProQ ligands (14,16,18) and function by repressing the translation of their corresponding toxins (57–59). Of these antitoxins, SibA, SibB and SibC coincided with ProQ, whereas RyeA, which was proposed to serve as an antitoxin to SdsR (60), sedimented away from ProQ (Figure 1F). We note that, similar to Hfq, ProQ was abundant in the pellet, supporting an earlier report of ProQ association with polysomes (61).

For a bird's eye view of possible in vivo complexes of other RBPs, we filtered the MS data for proteins with predicted RNA-binding properties based on UniProt (46) and Gene Ontology (62,63) information (Supplementary Figure S4). Interestingly, these known and predicted RBPs populated the whole gradient, revealing that some are likely to act without a stable partner, whereas others are involved in complexes of different sizes. As a general observation predictive of function, we note that ribosomal proteins were generally most abundant in the pellet (where 70S ribosomes sediment), whereas proteins involved in ribosome maturation were not found in the pellet and rather co-sedimented with the 30S or 50S subunit fractions, or found elsewhere.

RNA sedimentation profiles give functional insight

To learn more about the in-gradient behavior of RNA molecules, we performed t-stochastic neighbor embedding (t-SNE; (28)) to globally cluster all detected transcripts (Figure 2A). As expected, tRNAs as well as the rRNAs of the 30S or 50S subunits each accumulated, showing that t-SNE correctly identified their respective sedimentation profiles to be almost congruent. sRNAs mostly clustered in proximity to tRNAs, whereas mRNAs populated the whole map, as previously observed in Salmonella (14).

Figure 2.

Global analysis of transcript sedimentation reveals transcripts with unexpected properties. (A) t-SNE plot of all transcripts detected in the gradient. Proximities in the plot are not proportional to the distances in the original space. (B) Zoomed-in region of the plot shown in (A). RyeG is in close proximity to the 16S rRNAs, indicating similar sedimentation behavior.

Interestingly, in between the 16S rRNAs and the tRNAs, we found all 16 of the toxin mRNAs of type I TA systems we detected in the gradient (Figure 2A, green). Unlike the typical mRNA (Supplementary Figure S1B), these toxin mRNAs seemed to be generally excluded from 70S ribosomes, i.e. they were not found in the pellet (Supplementary Figure S5A). Additionally, several of the antitoxin RNAs co-migrated with their respective toxin mRNAs, which would suggest they present translationally inactive RNA-RNA complexes (Supplementary Figure S5A). However, such RNA–RNA complexes of type I TA systems are substrates of RNase III (57,58) and thus unlikely to be stable. Therefore, we interpret this sedimentation to represent association with ProQ (Figure 1F) (14,16,18). In contrast to type I TA systems, both the toxins and the antitoxins of type II TA systems are proteins, meaning that the antitoxins have to be translated in order to combat the harmful effects of the toxins (59). Consequentially, both partners of type II TA systems were found to have their peak abundance in the pellet fraction (Supplementary Figure S5B).

RyeG is a prophage-encoded RNA that co-sediments with the 30S subunit

Our t-SNE map (Figure 2A) placed several transcripts close to the 16S rRNAs, suggesting co-migration with 30S ribosomes (Figure 2B). Among these, we noticed many mRNAs coding for small proteins such as mgrB (64), mgtS (65) or sra (a.k.a. rpsV; ribosomal protein S22). Indeed, compared to all CDSs, the median length of the CDSs of these mRNAs was significantly shorter (282 aa vs. 107 aa; Supplementary Figure S5C). Furthermore, our t-SNE map revealed several ncRNAs to co-sediment with the 30S subunit. Of these, RyeG was the only sRNA that had previously been used in studies analyzing libraries of sRNA-overexpressing strains (66–69), whereas the others were low-confidence predictions. Northern blot detection of RyeG in the gradient fractions recapitulated the 30S association as well as a weaker signal in the pellet (Figure 3A, B), indicating that this noncoding RNA undergoes translation.

Figure 3.

RyeG is a prophage-encoded transcript that binds the 30S subunit. (A) Sedimentation profile of RyeG. RyeG sediments around the 30S subunit. Northern blot (NB) data are quantified from (B) and a replicate. n = 2. (B) Verification of the sedimentation profile of RyeG by northern blotting. (C) Genetic locus of ryeG. Genes within the prophage CPS-53 are shown in blue. Genes outside of CPS-53 are shown in gray. ryeG is shown in orange. Asterisks denote pseudogenes. (D) Sequence of the ryeG locus. The predicted extended –10 box is highlighted in gray. The transcriptional start site (TSS) is indicated by an arrow. Lower case letters indicate the sequence upstream of the TSS, whereas capital letters indicate the sequence of RyeG. (E) RyeG expression during growth. Northern blotting of RyeG shows that its expression is constant during growth and is downregulated at late stationary phase. RNA from three independent biological replicates was loaded. 5S rRNA was used as loading control. (F) Estimation of in vivo copy numbers of RyeG. Northern blotting of RyeG compared to in vitro-synthesized RyeG reveals low levels of ∼1–5 copies/cell. 5S rRNA was used as loading control. (G) Predicted secondary structure of RyeG. Secondary structure prediction by RNAfold (108) reveals several stem loops and a ρ-independent terminator. Nucleotides highlighted in orange represent the coding sequence of ORF2 and its SD sequence is indicated (compare to Figure 4C). Visualization was performed using VARNA (109).

RyeG was first reported as IS118 in an early bioinformatics sRNA search in E. coli (70) and later shown to decrease biofilm formation (68) and motility (69) when overexpressed. The ryeG gene lies within the cryptic prophage CPS-53, on the antisense strand between yfdI and tfaS (Figure 3C). The CPS-53 prophage is only present in E. coli K-12 strains and shows signs of gene erosion (71). CPS-53 has been reported to increase H₂O₂ and acid resistance (72) and to possess genes that inhibit initiation of chromosomal replication when overexpressed (73). The ryeG gene carries an extended -10 box (74) indicative of transcription by the E. coli housekeeping RNAP-σ⁷⁰ (Figure 3D). Northern blot analysis showed that RyeG accumulates to 3–5 copies per E. coli cell during growth in rich medium, dropping to ∼1 copy/cell in late stationary phase (Figure 3E, F). This 199 nt long sRNA is predicted to be highly structured (Figure 3G), which might also explain its recently described interaction with ProQ (16).

RyeG encodes a toxic small protein

To assess whether ryeG was a functional gene, we overexpressed it from a high copy plasmid and determined effects on bacterial growth (Figure 4A). We observed a much longer lag-phase, with RyeG overexpressing cells reaching mid-log phase ∼2 h later than wild-type E. coli grown in parallel. Most importantly, this growth retardation was also observed when RyeG was overexpressed in Salmonella (Figure 4B), which lacks CPS-53. Thus, the bacteriostatic effect of RyeG is independent of any other prophage-encoded genes.

Figure 4.

RyeG encodes for a toxic small protein. (A, B) Growth curves comparing E. coli ΔryeG and wild-type Salmonella against their corresponding RyeG overexpression strains. Overexpression of RyeG leads to strongly prolonged lag times. The used plasmid is a pZE12 derivative (containing a high copy colE1 origin) that expresses full-length RyeG under control of a P_LlacO-1 promoter. Growth was monitored using a microplate reader. Data was obtained from three biological replicates each. (C) Sequence of RyeG with ORF2 (recently designated yodE (75)) and ORF3 highlighted in bold (related to Supplementary Figure S6). The corresponding predicted SD sequences are highlighted in gray. The A nucleotide at which a stop was detected using 30S subunit toeprinting (D) is highlighted in blue. (D) 30S subunit toeprinting of RyeG. A specific toeprint in presence of initiator tRNA and 30S subunits can be detected by a strong stop at an A nucleotide at position +37 (relative to the transcriptional start) of RyeG, indicating formation of an initiation complex. The overlapping start codons of ORF2 and ORF3 are indicated. (E) Mutants of RyeG. To test which ORF is responsible for the toxic effect of RyeG, three mutants were created: SD-mut eliminates the SD sequences of ORF2 and ORF3 without changing the predicted secondary structure (compare to Figure 3G). ORF2-stop and ORF3-stop introduce stop codons in codon 8 and 9 of ORF2 and ORF3, respectively, by single nucleotide exchanges. (F) Growth curves comparing E. coli ΔryeG with a control plasmid against different inducible RyeG versions (E). Expression of SD-mut and ORF2-stop show similar growth as the control, whereas ORF3-stop shows a similar phenotype as expression of wild-type RyeG, indicating that ORF2 is responsible for the phenotype of RyeG expression. Induction of the tetracycline-inducible plasmids was performed by addition of 200 ng/ml doxycycline to the medium at the start of the experiment. Growth was monitored using flasks. Data was obtained from three biological replicates. Error bars show SD from the mean.

Next, we followed up on the observed strong 30S association of RyeG, asking whether the RNA itself or an unrecognized open reading frame (ORF) was responsible for the observed toxicity. The ORFfinder algorithm (https://www.ncbi.nlm.nih.gov/orffinder/) returned five different possible small ORFs (Supplementary Figure S6A-E) in RyeG, of which ORF2 (48 aa) and ORF3 (19 aa) were preceded by a potential Shine-Dalgarno (SD) sequence (Figure 4C). To experimentally test translation initiation, we performed toeprinting assays (35) using in vitro-synthesized RyeG and purified 30S subunits (Figure 4D). This revealed a strong toeprint at position +37 (relative to the transcriptional start) in presence of 30S and charged tRNA^fMet but not without the initiator tRNA, indicating assembly of an initiation complex. The adenosine at position +37 is located 14 nt or 16 nt upstream of ORF2 or ORF3, respectively (Figure 4C), suggesting that both ORFs can be translated, in principle.

To pinpoint the ORF that is being translated and whether it causes the observed toxicity, we constructed three different mutant versions of RyeG (Figure 4E). These had to be cloned in a tetracycline-inducible plasmid since some of them were impossible to maintain in a constitutive overexpression plasmid. As before, wild-type RyeG strongly delayed growth when expressed from the tetracycline-inducible plasmid (Figure 4F). This growth phenotype was largely abrogated by the SD mutant (SD-mut), and fully so by a premature stop codon in ORF2 (ORF2-stop). In contrast, a premature stop codon in ORF3 (ORF3-stop) did not alleviate toxicity of RyeG; if at all, we observed a longer lag time than with wild-type RyeG. Therefore, RyeG is a previously unrecognized mRNA encoding a 48 aa growth-inhibitory protein. Of note, a recent global survey of candidate small proteins also showed ORF2 of RyeG to be translated, designating it yodE (75). This ORF2 must be exceptionally toxic, for all our attempts to clone it on its own (i.e., using the strong ribosome binding site of the plasmid), even under control of a tight tetracycline-dependent promoter, have failed thus far.

Grad-seq resolves a wide range of protein complexes

Our focus on interactions of RNAs notwithstanding, Grad-seq also enables the analysis of multi-protein complexes. For many such complexes, we observed well-correlated profiles of their corresponding subunits (Figure 5A). For example, the succinyl-CoA synthetase consisting of SucC and SucD partitioned as a small complex around fraction 3, whereas the >900 kDa FtsH/HflKC metalloprotease complex sedimented as a particle almost the size of 30S subunits, matching a previous observation (76). This illustrates the wide range of complexes resolved by Grad-seq. For a more global assessment of the quality of such predictions, all heterocomplexes, for which all subunits could be detected in the MS data, were tested for co-sedimentation. Of those 107 heterocomplexes, 79 (∼74%) showed high correlation (Spearman's ≥ 0.7), indicating intact complexes (Figure 5B). Thus, following the ‘guilt-by-association’ logic, Grad-seq profiles might be able to predict whether a given E. coli protein is part of a cellular complex. For orientation, a <20 kDa protein with a slightly elongated shape will sediment around <3S (77), i.e., at the top of the gradient (Supplementary Figure S1A). Conversely, if a protein <20 kDa occurs in higher fractions, it is likely involved in a complex.

Figure 5.

Grad-seq resolves a wide range of protein-protein complexes. (A) Heat map showing the sedimentation profiles of exemplary intact complexes spanning ∼140–1,600 kDa. For each protein, the spike-in-normalized sedimentation profiles are normalized to the range from 0 to 1 by dividing the values of each fraction by the maximum value of the corresponding protein. (B) Violin plot showing the distribution of the mean Spearman's correlation of all heterocomplexes, for which all subunits were detected in the gradient (n = 106). 79 of these show a correlation ≥0.7, indicating intact complexes. The solid line indicates the median, whereas the dashed lines indicate the upper and lower quartiles.

To predict new complexes, we filtered our MS data (proteins <20 kDa with a peak in fraction ≥4) to obtain 97 proteins with unexpected in-gradient occurrences (Figure 6). Unsurprisingly, 42 of these were ribosomal proteins, and an additional four known to be ribosome-associated: Hsp15 (HslR), Rmf, RsfS and the L31 paralog YkgM. Hsp15 and RsfS co-sedimented with the 50S subunit, as reported earlier (78–81). YkgM could only be detected in fraction 15, overlapping with the height of the 50S subunit peak. Given its probable function as an alternative L31 protein (82), this may reflect a tight 50S association of YkgM. In contrast, Rmf primarily occurred in the pellet fraction, which agrees with its function in the formation of inactive 70S dimers, so-called 100S ribosomes (41,42). Other expected proteins included the RNAP-interacting proteins RpoZ, GreB (a transcription elongation factor) (83) and CedA (a regulator of cell division) (84), all of which co-sedimented with RNAP (Figure 1F). The membrane-associated proteins Lpp, OmpX and SecG were almost exclusively detected in the pellet fraction, indicating the formation of insoluble aggregates (Figure 6).

Figure 6.

Small proteins are involved in large complexes. Heat map showing the sedimentation profiles 97 proteins <20 kDa whose peak abundance is detected in fraction 4 or higher. The profile of YggL (highlighted in orange) is congruent with the proteins of the large ribosomal subunit (Rpl* and Rpm* proteins). For each protein, the spike-in-normalized sedimentation profiles are normalized to the range from 0 to 1 by dividing the values of each fraction by the maximum value of the corresponding protein.

A ribosome-associated function of protein YggL

Searching for proteins with unrecognized association with larger complexes, we homed in on YggL. This small ∼13 kDa protein constitutes its own family of DUF469 proteins (85), is extremely conserved in the class of γ-proteobacteria and is further present in the orders of Burkholderiales and Neisseriales within the class of β-proteobacteria (Figure 7A and Supplementary Figure S7A). According to our global MS data, YggL was most abundant in the pellet but showed an additional broad peak similar to 50S subunit components (Figure 6). To test this global MS-based prediction, we chromosomally tagged the yggL gene with a 3xFLAG epitope and performed western blot analysis on two different sorts of gradient samples of this yggL-3xFLAG strain. These analyses verified both the 50S (Figure 7B) and 70S association of YggL (Figure 7C).

Figure 7.

YggL is a 70S ribosome-interacting protein. (A) Phylogenetic analysis of YggL based on 150 protein sequences deposited in eggNOG 4.5.1 (COG3171; (110)). YggL homologs were only found in γ-proteobacteria (orange) and β-proteobacteria (blue). (B) Glycerol gradient analysis of a yggL-3xFLAG strain followed by western blotting. YggL-3xFLAG co-migrates with the 50S subunit and is found in the pellet. L, lysate (input control). P, pellet. (C) Sucrose polysome gradient analysis of a yggL-3xFLAG strain followed by western blotting. YggL-3xFLAG co-migrates exclusively with the 70S ribosome. L, lysate (input control). (D) SDS-PAGE of YggL-3xFLAG co-immunoprecipitation (PD) and the wild type (wt). Specific bands for the PD between 10 and 25 kDa can be detected. (E) MS analysis of the co-immunoprecipitation shown in (E). Apart from the expected enrichment of YggL (blue), several proteins of the large ribosomal subunit (red) as well as two of the small ribosomal subunit (red) were enriched compared to the wild type. Other enriched proteins are shown in yellow and proteins not considered enriched are shown in gray. (F) Sucrose polysome gradient analysis of in vitro-reconstituted YggL-ribosome complexes followed by western blotting. 100 pmol (black) or 400 pmol (orange) recombinant YggL was allowed to bind to 400 pmol of purified ribosomes obtained from a ΔyggL strain. Subsequent sucrose polysome gradient analysis shows that YggL specifically binds to 70S ribosomes. Probing for purified YggL was performed using an antibody against its 6xHis-tag, which was used for the purification of YggL and not cleaved off. (G) yggL expression during growth. Northern blotting of yggL shows that its expression is constant during growth and is shut off at late stationary phase (OD 2.0 + 4 h). RNA from three independent biological replicates was loaded. 5S rRNA was used as loading control. Note that the same membrane as in Figure 3E was used, explaining the recurrence of the control lanes. (H) YggL-3xFLAG expression in different phases of bacterial growth. Western blot analysis shows that expression of YggL-3xFLAG protein is constant during growth, including late stationary phase (OD 2.0 + 4 h). Protein content from three independent biological replicates was loaded (0.1 OD of cells per well). A wild-type culture was sampled in the same way and loaded as control. GroEL was used as loading control. (I) Sucrose polysome gradient analysis of wild-type and ΔyggL strains. A_{260 nm} profiles show that the knockout of yggL increases the amount of free 50S subunits.

The ribosome association of YggL found support in additional experiments. First, immunoprecipitation of YggL-3xFLAG from an E. coli lysate (Figure 7D) strongly enriched 12 proteins of the 50S subunit (Figure 7E). Interestingly, with the exception of L2 (RplB), all of them were from the ribosome's cytosolic side (86), suggesting this is the side where YggL binds as well. Second, we confirmed the YggL-70S association by in vitro reconstitution of purified YggL with purified ribosomes obtained from a ΔyggL strain run in a sucrose gradient (Figure 7F).

Next, investigating yggL mRNA levels in different growth phases, we observed yggL to be expressed only until the early stationary phase of growth (Figure 7G), which matches published gene expression data for Salmonella (40). This analysis revealed three transcripts containing yggL, which could be attributed to two primary transcripts (both of which contain a predicted σ⁷⁰ binding site) and one processed transcript (Supplementary Figure S7B–D). Interestingly, YggL protein levels were not reduced during late stationary phase (Figure 7H), which was also reported by a previous MS study (87), possibly due to increased translational efficiency or protein over stabilization.

We then followed up earlier predictions by others (88) that YggL might be involved in late 50S subunit assembly or final maturation of the 70S ribosome. To do so, we constructed an E. coli ΔyggL strain to test a loss-of-function effect on ribosomes. Comparing profiles of polysome gradients (Figure 7I), ΔyggL bacteria exhibited a strong increase in free 50S subunits, as compared to the wild-type strain. This change in ribosome profile is unlikely to be a growth effect, since the yggL knockout grew indistinguishably from wild-type E. coli (Supplementary Figure S8A).

To further investigate potential growth defects, we monitored bacterial growth of the ΔyggL strain in minimal medium at 37°C and in rich medium at 25°C (Supplementary Figure S8B and C). While there was no obvious effect in minimal medium, at 25°C the yggL knockout grew considerably slower than did wild-type bacteria, which is reminiscent of the previously reported cold-sensitivity of E. coli knockout strains of ribosome-associated proteins (89–91). Taken together, based on its Grad-seq profile, YggL emerges as a 70S ribosome-associated protein with a potential function in particle assembly.

Data visualization and accessibility

Grad-seq provides a global overview of RNA and protein interactions obtained from a single experiment. To facilitate data accessibility und usability, we set up an online browser (https://helmholtz-hiri.de/en/datasets/gradseqec/) for interactive exploration of these datasets (Figure 8). Visualization can be performed based on a user-selected group of genes and displayed as bar chart, line plot or heat map. Importantly, the browser also allows to view Grad-seq data for Salmonella (14,19) and S. pneumoniae (20), which permits a cross-species comparison of the sedimentation profiles of selected entries. Comparison of the closely related enterobacteria E. coli and Salmonella whose proteins tend to be generally similar will be useful for a fine-grained analysis of in-gradient distributions. Comparing potential complex formation of distantly related homologs of Gram-negative and Gram-positive bacteria may give clues for broadly conserved functions of a protein or RNA molecule.

Figure 8.

Overview of the Grad-seq online browser, accessible at https://helmholtz-hiri.de/en/datasets/gradseqec/.

DISCUSSION

Our Grad-seq analysis of E. coli provides the first comprehensive landscape of stable RNA and protein complexes in this important model bacterium. As a valuable resource, our Grad-seq dataset adds the previously missing knowledge about potential RNA interactions to the ever-growing pool of information about E. coli. Exploring this data, we have identified a phage-encoded, K-12-specific toxic protein as well as a conserved ribosome-binding protein. As more Grad-seq data for other species become available, we will develop a better understanding of the scope of interactions and complexes of RNA and proteins in bacteria.

Although there have been several global studies of the protein interactome of E. coli (4–13), global information about RNA complex formation has been lacking. Our Grad-seq data readily reproduce the major cellular RNPs such as the SRP or RNAP, while also providing information about regulatory RBPs such as CsrA, Hfq and ProQ, which together bind the majority of sRNAs within the cell (Figure 1F) (92).

While mRNAs were expected to accumulate in the pellet together with actively translating ribosomes (Supplementary Figure S1B), sRNAs were not (Figure 1F and G and Supplementary Figure S1C-F). Straight-forward explanations for the surprisingly abundant ribosome associations of sRNAs include their activities as activators of mRNA translation (such as DsrA (93,94)) and translational repressors within a polycistronic mRNA (such as Spot 42 (95)). Moreover, ribosome association of sRNAs can hint at a dual function of an RNA, as first describe for Staphylococcus aureus RNAIII, which is both a regulatory RNA and the mRNA of δ-hemolysin (96). Here, we observed SgrS—the best-characterized dual-function RNA of E. coli (97)—almost exclusively in the pellet fraction (Figure 1F).

The present work highlights strong association with the 30S subunit as a predictor of coding potential. We have discovered that the seemingly noncoding prophage-specific RyeG sRNA encodes a toxic 48 aa protein (Figures 3 and 4), agreeing with recent analysis of ribosome footprints by others (75). At this point, however, we have no indication for a dual function of RyeG, i.e. that the RyeG RNA itself serves as regulator, judging by the fact that a single nucleotide change creating a premature stop codon rendered RyeG non-toxic (Figure 4F). The exceptional 30S subunit association of RyeG could indicate that initiation complex formation takes place but formation of actively translating ribosomes is somehow inhibited. In our dataset, other mRNAs with potential 30S subunit association (Supplementary Figure S9A) were enriched in pseudogenes, toxins and phage-encoded genes (Supplementary Figure S9B), suggesting there might indeed be a mechanism preventing the translation of non-functional or detrimental RNAs. Alternatively, 30S subunit association could be an intrinsic feature of mRNAs with especially small CDSs that is readily detected by Grad-seq (Supplementary Figure S5C).

The toxicity of RyeG is visible by a much prolonged lag time before bacterial growth takes off after fresh inoculation (Figure 4A and B). Increased lag times can be detrimental to a bacterium because it might be outcompeted by others in the same environment that are quicker at utilizing the available nutrients (98). However, lag phase can also be beneficial by conferring stress tolerance to, e.g. antibiotics (99) or by possibly increasing immune evasion (100). The latter might be one benefit of the continuous presence of the defective prophage CPS-53 in the chromosome of E. coli K-12, since CPS-53 was shown to increase H₂O₂ and acid resistance (72). The low in vivo copy numbers of RyeG/YodE (Figure 3F and (75)) suggest that its function might depend on its abundance within the cell. Yet, the exact function of RyeG and how the cell overcomes its toxic effect upon overexpression remain obscure.

Although our previous Grad-seq analysis of related Salmonella bacteria already included the global proteomics component (14), this part remained underexplored. Complementing data obtained via binary co-purification, Grad-seq provides an overview of the major complexes of E. coli (Figure 5), and so lends itself to cross-comparison with existing global data obtained from AP/MS and two-hybrid screens (4–8). As a first example, we identified the well-conserved YggL as a 50S ribosome-binding protein (Figures 6 and 7), agreeing with binary interactome studies that reported interactions between YggL and the ribosomal proteins L2, L28 and L32 (5,7). Intriguingly, another previous study used a pulse labeling approach in combination with sucrose gradient centrifugation and MS to identify proteins implicated in ribosome assembly, which suggested YggL to be involved in late 50S subunit assembly or final maturation of the 70S ribosome (88). Taking note of this, we found the knockout of yggL to increase the amount of free 50S subunits within the cell (Figure 7I). This implies an increase in biogenesis of 50S subunits, possibly to compensate for a defect in 70S assembly. Similarly, others observed abnormal ribosome profiles upon knockout of other ribosome-associated proteins (101–103).

Our discovery of the ribosome association of YggL further emphasizes the diversity of ribosomes across different species. For example, RbgA is an essential GTPase for 50S assembly in B. subtilis (103) but absent from E. coli, showing that different species use different proteins for ribosome assembly. Reciprocally, YggL is strongly conserved within the γ-proteobacteria but absent from other bacterial classes. Therefore, might YggL be involved in the formation of subpopulations of specialized ribosomes (104) or exert a function not needed in other groups of bacteria? In this regard, the emerging protein catalogs from Grad-seq in different species promise to yield new types of protein functions in building and shaping the full diversity of bacterial ribosomes.

DATA AVAILABILITY

The sequencing data have been deposited in NCBI's Gene Expression Omnibus (105) and are accessible through GEO Series accession number GSE152974 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE152974). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (106) via the PRIDE (107) partner repository with the dataset identifier PXD019900 (https://www.ebi.ac.uk/pride/archive/projects/PXD019900). The used software and the resulting files have been deposited at Zenodo (https://doi.org/10.5281/zenodo.3876866 and https://doi.org/10.5281/zenodo.3955585).

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

Deutsche Forschungsgemeinschaft (DFG) (to J.V.); Leibnitz Award [DFG875-18]; SPP 2002 project [DFG Vo875-20-1]. The open access publication charge for this paper has been waived by Oxford University Press – NAR Editorial Board members are entitled to one free paper per year in recognition of their work on behalf of the journal.

Conflict of interest statement. None declared.