A systematically-revised ribosome profiling method for bacteria reveals pauses at single-codon resolution

Abstract

In eukaryotes, ribosome profiling provides insight into the mechanism of protein synthesis at the codon level. In bacteria, however, the method has been more problematic and no consensus has emerged for how to best prepare profiling samples. Here, we identify the sources of these problems and describe new solutions for arresting translation and harvesting cells in order to overcome them. These improvements remove confounding artifacts and improve the resolution to allow analyses of ribosome behavior at the codon level. With a clearer view of the translational landscape in vivo, we observe that filtering cultures leads to translational pauses at serine and glycine codons through the reduction of tRNA aminoacylation levels. This observation illustrates how bacterial ribosome profiling studies can yield insight into the mechanism of protein synthesis at the codon level and how these mechanisms are regulated in response to changes in the physiology of the cell.

Keywords: E. coli; bacteria; biochemistry; chemical biology; chromosomes; gene expression; pausing; ribosome profiling; serine.

Figure 1.. Comparison of ribosome profiling data from yeast and E.coli.

(A) Average ribosome density on genes aligned at the start codon using the 5’-end of reads in yeast (library SRR1042864), or the center or 3’-end of reads from E. coli (library SRR1734438). (B) Length distribution of yeast and E. coli ribosome-protected fragments mapping uniquely to coding sequences. (C) The fraction of reads at the first, second, or third nt within codons in yeast profiling data (blue), E. coli profiling data (grey), RNA-seq from total RNA digested with MNase (yellow), and profiling data in which nucleases RelE and MNase were used to generate ribosome-protected footprints (red). See also Figure 1—figure supplement 1.

Figure 1—figure supplement 1.. Preferential isolation of long RPFs increases ribosome density at SD-like motifs within open reading frames.

Cross-correlation plots were generated by first, calculating a genome-wide map of SD affinity using an eight nt sliding window, and then taking the correlation of the SD-affinity map with ribosome density at different offset values (Li et al., 2012). The strong peak at −22 in L18 (black) indicates a positive correlation between SD affinity and ribosome density as would be expected if SD motifs caused ribosomes to pause in ORFs. The −22 offset is consistent with the known spacing between the SD motif and the 3’-boundary of the ribosome during initiation (top). In contrast to the strong positive correlation at −22 seen in L18, a negative correlation is observed in L17 (blue). This difference arises from how footprints were isolated: L18 contains exclusively long RPFs (>28 nt) whereas L17 contains exclusively short RPFs (20–30 nt). Given that footprints that interact with rRNA tend to be longer (30–35 nt), isolating only long RPFs leads to artificial enrichment of ribosome density at SD-like motifs. No SD pauses are observed in our libraries (e.g. L26, red) that capture the whole distribution of footprint sizes, nor were they observed in samples prepared using our new methods with high MgCl₂ lysis buffers with cells harvested by either filtration (L29) or direct freezing (L33) (purple and green respectively). In addition to the peaks at −22 initially attributed to pauses on SD motifs, two other peaks are also observed. The peak at −15 arises from pauses at Gly codons (Figure 3 and Mohammad et al., 2016) because Gly codons are G-rich, giving a spurious but strong SD affinity. In a similar fashion, the peak at 0 arises from cloning bias because the nucleotide G is enriched at the 3’-ends of reads.

Figure 2.. Heat map of the distribution of read lengths in published E.coli ribosome profiling libraries from several labs.

See Figure 2, Figure 2—source data 1 for details.

Figure 3.. Chloramphenicol (Cm) alters ribosome density at the gene and codon level in published E.coli ribosome profiling libraries.

(A) Cultures are harvested by centrifugation or filtration. L1-L10 were treated with Cm in the media prior to harvesting; all samples were prepared with Cm in the lysis buffer. (B) Distribution of asymmetry scores, the log2 value of the number of reads in the second half of a gene divided by the number of reads in the first half. Genes with more ribosomes at the 5’-end than the 3’-end have negative values. (C) Heat map of pause scores for the codon in the ribosomal E site (corresponding to the penultimate amino acid in the nascent peptide). See also Figure 3—figure supplement 1.

Figure 3—figure supplement 1.. Treating cultures with Cm prior to harvesting skews estimates of protein synthesis levels in different ways depending on the gene length.

(A) We ranked genes by length and divided them into six subsets with between 460–470 genes in each set. We took the ratio of footprints per gene (RPKM) for a sample treated with Cm in the media (L1) compared with a similar sample without Cm treatment (L26). Both samples were filtered and prepared using the standard protocol. (B) We performed the same analysis using two samples without Cm in the media as a control.

Figure 4.. High salt buffers arrest translation after cell lysis better than Cm.

(A) We added [14C]Lys-tRNA^Lys to frozen lysate that was then thawed for 15 min. [14C]Lys that was incorporated into nascent peptides can be selectively precipitated with TCA after tRNA hydrolysis under alkaline conditions. (B) Lysates were made with buffers containing 1 mM Cm, 50 mM MgCl₂, or 1M NaCl. The boiled sample was denatured prior to addition of Lys-tRNA. Error bars reflect the standard deviation of four technical replicates. The boiled and Cm samples were compared using a one-tailed, paired t-test.

Figure 5.. Pausing is crystal clear in samples prepared with high salt buffers instead of Cm.

(A) Heatmap of pause scores for codons for all 20 amino acids in either the E or A site of the ribosome from samples prepared with lysis buffers containing Cm, 150 mM MgCl₂, or 1 M NaCl (libraries L26, L27, and L28 respectively). (B) Average ribosome occupancy aligned at Ile codons for samples treated with mupirocin, an inhibitor of Ile-tRNA synthetase, and an untreated control (L29), using lysis buffers with either high MgCl₂ or Cm (L30 and L32 respectively). (C) Average ribosome occupancy aligned at Ser codons in untreated cells using lysis buffers containing Cm, high MgCl₂, or high NaCl (libraries L26, L27, and L28 respectively).

Figure 5—figure supplement 1.. Incorporating high salt lysis buffers into ribosome profiling.

(A) High salt buffers interfere with MNase activity; to exchange buffers, we pellet ribosomes and resuspend them in the standard buffer (without Cm). (B) Polysome profiles of samples pelleted over a sucrose cushion. The ratio of polysomes to monosomes and subunits is also given. (C) Comparison of rpkm values for a sample prepared with 1 M NaCl (L28) versus the standard protocol with Cm in the buffer (L31, also pelleted). (D) Comparison of rpkm values for a sample prepared with 150 mM MgCl₂ (L29) versus the standard protocol with Cm in the buffer (L31, also pelleted). (E and F) Top: average ribosome occupancy aligned at start codons (E) or Ser codons (F) in library L27 prepared with 150 mM MgCl₂ and pelleted prior to digestion. Bottom: heat-map of RPF lengths at each position. In pelleted samples, nuclease cleavage of mRNA in the ribosome yields shorter RPFs with shifted 3’-ends compared to RPFs that span the whole ribosome.

Figure 6.. Filtering cells leads to ribosome pausing at Ser codons due to reduced levels of aminoacylated tRNA^Ser.

(A) Heatmap of ribosome density downstream of Ser codons in samples harvested by filtration using lysis buffers containing Cm, 150 mM MgCl₂, or 1 M NaCl (L26, L27, and L28, respectively). (B) Model of how pausing affects ribosome density. Downstream of a pause site (shown in red), ribosomes continue elongation and are released at stop codons, such that downstream density drops until a new steady state is reached. (C) Heatmap of ribosome density downstream of Ile codons in untreated cells (L29) and after 10 min of mupirocin treatment (L30). (D) Schematic of the method used for panels 6E and 6F: from a single culture, samples were harvested by rapid filtration or by directly freezing the culture. Ribosomes were then pelleted through a sucrose cushion. (E) Plots and heatmaps of average ribosome density aligned at Ser codons in untreated cells that were filtered (L29) or frozen (L33). (F) Northern blot of Ser, Gly, and Ile tRNAs after periodate oxidation and β-elimination, a treatment that removes the final nt of tRNAs that are not charged. As a negative control, an aliquot of tRNA from filtered or frozen samples were pretreated in alkaline conditions to deacylate tRNA.

Figure 7.. Samples harvested by direct freezing and lysed in high MgCl₂ buffer reveal subtle ribosome pauses that reflect known biology, pauses at polyproline motifs and at rare codons.

(A) Heatmap of pause scores in two biological replicates harvested by direct freezing (L33 and L35). (B) Ranking of all tripeptide motifs by their pause scores with the motif occupying the A, P and E sites of the ribosome in library L35. (C) Sequence logo of the top 50 tripeptide motifs from 7B. (D) Spearman correlation between ribosome density at each codon and the inverse value of its codon-adaptation index (CAI), a measure of codon usage and optimality. The correlation was calculated for codons within the ribosome (E, P, and A-site codons) and two codons on either side. The E. coli data are from libraries L29 (filtered, MgCl₂), L31 (filtered, Cm), and L33 (frozen, MgCl₂) and the S. cerevisiae data are from SRR1049521 (Subtelny et al., 2014).