Identifying Small Open Reading Frames in Prokaryotes with Ribosome Profiling

ABSTRACT

Small proteins encoded by open reading frames (ORFs) shorter than 50 codons (small ORFs [sORFs]) are often overlooked by annotation engines and are difficult to characterize using traditional biochemical techniques. Ribosome profiling has tremendous potential to empirically improve the annotations of prokaryotic genomes. Recent improvements in ribosome profiling methods for bacterial model organisms have revealed many new sORFs in well-characterized genomes. Antibiotics that trap ribosomes just after initiation have played a key role in these developments by allowing the unambiguous identification of the start codons (and, hence, the reading frame) for novel ORFs. Here, we describe these new methods and highlight critical controls and considerations for adapting ribosome profiling to different prokaryotic species.

INTRODUCTION

Defining the whole set of protein-coding regions in bacterial and archaeal genomes remains a formidable challenge. Traditionally, coding sequences have been identified in an automated fashion with the help of limiting assumptions: initiation takes place at AUG codons, open reading frames (ORFs) do not overlap, and they must be at least 50 codons in length. When these assumptions are relaxed, the number of potential ORFs quickly becomes overwhelming (there are >100,000 ORFs of between 10 and 50 codons in Escherichia coli) (1). The vast majority of these small ORFs (sORFs), whose number far surpasses the ∼4,300 proteins in the known E. coli proteome, are probably not expressed. On the other hand, given the recent discovery of dozens of proteins in bacteria encoded by sORFs of fewer than 50 codons (reviewed in references 2 and 3), it is increasingly clear that current annotation methods are missing many small proteins that are expressed and may have important biological functions. The challenge in identifying small proteins is further complicated by the fact that they are also difficult to isolate and characterize using traditional biochemical and genetic methods.

The development of new methods has greatly facilitated the genome-wide identification of sORFs and their translation products. In particular, ribosome profiling (Ribo-seq) has tremendous potential to empirically improve the annotations of prokaryotic genomes. Ribo-seq reveals the translational output of the cell through the deep sequencing of ribosome-protected mRNA fragments, so-called “ribosome footprints” (4). Ribo-seq provides direct evidence for protein synthesis from potential ORFs, regardless of their length or whether they start with a canonical AUG codon. Unlike transcriptome sequencing (RNA-seq), Ribo-seq reveals the specific location of translating ribosomes on the mRNA, allowing the accurate mapping of ORFs, including sORFs or alternative ORFs (alt-ORFs) within larger genes. Furthermore, the number of ribosome footprints from a given ORF is highly correlated with the steady-state level of the encoded protein (5). This means that Ribo-seq data obtained from cells grown under various conditions can provide clues about the potential functions of novel sORFs depending on the pattern of their expression. Ribo-seq therefore can help discover, validate, and yield insights into the function of novel bacterial sORFs.

In a typical Ribo-seq experiment (Fig. 1), actively translating ribosomes are first captured on mRNAs by flash-freezing or treatment with translation elongation inhibitors. Following cell lysis, unprotected regions of mRNA are digested with RNases, and ribosome-protected mRNA fragments are purified from monosomes isolated from a sucrose density gradient. cDNA libraries prepared from the ∼30-nucleotide (nt) footprints are then subjected to deep sequencing. Through comparison with an RNA-seq library generated from total RNA from the same culture, ribosome footprint coverage can be used to determine the ORF boundaries and the translational efficiency of mRNAs.

FIG 1.

(Top) Critical steps in the ribosome profiling protocol and suggestions for how to optimize the protocol for new prokaryotic species. (Bottom) Average ribosome density from yeast Ribo-seq data aligned at start codons and stop codons.

In early Ribo-seq studies in eukaryotes, the observation of ribosome density on what were thought to be long noncoding RNAs raised questions about how to confidently establish that the signal in Ribo-seq experiments truly reflects translating ribosomes (6, 7). The consensus is that there are distinct hallmarks of true translation that can be seen in plots of Ribo-seq density averaged across thousands of genes aligned at their start or stop codons (Fig. 1) (8). First, there are typically strong peaks at start codons due to slow steps following the assembly of the initiation complex. These peaks of initiating ribosomes can be enhanced by treating cells with certain antibiotics, as described below. Second, there are strong peaks at stop codons due to the low rates of peptidyl hydrolysis by release factors or the disassembly of posttermination complexes by recycling factors. Third, since ribosomes move 3 nt at a time during elongation, ribosome footprints typically show 3-nt periodicity throughout ORFs, with higher ribosome density at the first nucleotide than at the second and third nucleotides of the codon. These three lines of evidence could validate the translation of a given ORF.

Taking advantage of these hallmarks of translation, Ribo-seq studies have revealed extensive, previously unrecognized coding potential in genomes of diverse organisms, including yeast, mammals, plants, and viruses (2). Although progress in annotating prokaryotic genomes has been slower, recent Ribo-seq studies have started to identify cryptic ORFs, including sORFs, in diverse species (Table 1) such as E. coli (reviewed in reference 3), Salmonella (9, 10), mycobacteria (11), Bacteroides thetaiotaomicron (12), and even the archaeon Haloferax volcanii (13). As described below, recent refinements of the Ribo-seq method use antibiotics to trap ribosomes at start codons, revealing a genome-wide map of translation initiation sites (1, 14, 15). Here, we describe several considerations for carrying out ribosome profiling in prokaryotic organisms and highlight new methods for identifying initiation and termination sites with the goal of improving genome annotations and identifying sORFs.

TABLE 1.

Examples of ribosome profiling studies in prokaryotes^a

Approach and organism(s)	Type of Ribo-seq	Reference(s)
General ribosome profiling approaches
Escherichia coli	First application to bacteria, selective ribosome profiling	17
E. coli-B. subtilis	Ribosome pausing	56
E. coli	Selective ribosome profiling	16
E. coli	Quantification of protein synthesis rates	5
E. coli	New harvesting method	19
E. coli	Ribo-seq-based analyses of translation termination by apidaecin	50
E. coli	sRNA target identification	57
E. coli	sRNA target identification	58
E. coli-Salmonella	Examination of the regulon of the translational regulator CsrA	59, 60
Phage T4-infected E. coli	Studying mechanisms of translation	61
Bacteriophage lambda	Bacteriophage gene expression analysis	62
Staphylococcus aureus	Ribo-seq-based analysis of ribosome hibernation factor effects	63
Bacillus subtilis	Protein synthesis rate quantification	64
Streptomyces coelicolor	Protein synthesis rate quantification	65
Listeria monocytogenes	Ribo-seq-based analyses of antibiotic action	66
Pseudomonas aeruginosa	Examination of gene regulation	67
Pseudomonas fluorescens	Examination of posttranscriptional regulation by Hfq	68
Halobacterium salinarum	Examination of regulation and efficiency of translation	69
Genome annotation/sORF identification based on Ribo-seq
Bacteroides thetaiotaomicron		12
Caulobacter crescentus		70
E. coli	TetRP to map TISs	15
E. coli	Ribo-RET	14
E. coli	TIS profiling using retapamulin/Onc112	1
EHEC		71
Haloferax volcanii		13
Mycobacterium abscessus		72
Mycobacterium smegmatis		11
Salmonella enterica serovar Typhimurium		9, 10
Microbiome	metaRibo-seq	73

^aThis table provides an overview of several example studies that employed general ribosome profiling approaches as well as studies that employed Ribo-seq for genome annotation and sORF annotation. Due to space restrictions, only selected examples are listed, and we apologize to all authors whose studies were not mentioned. sRNA, small RNA; TetRP, tetracycline-inhibited ribosome profiling; TISs, translation initiation sites; Ribo-RET, retapamulin-enhanced Ribo-seq; EHEC, enterohemorrhagic E. coli.

CONSIDERATIONS FOR SETTING UP Ribo-seq IN NEW PROKARYOTIC SPECIES

Established protocols for Ribo-seq in E. coli (16, 17) cannot always be directly applied to other bacterial and archaeal species. Each step of the workflow has to be carefully adapted and optimized for new species. Critical steps that may require methodological adaptation include harvesting of cultures, cell lysis, and the generation and isolation of ribosome footprints (Fig. 1). While species-specific optimization is needed to appropriately detect translation on all mRNAs, they are particularly important for the identification of translated sORFs.

Harvesting cells and arresting translation.

The first critical step in Ribo-seq experiments is harvesting cells in a way that arrests translating ribosomes without distorting the natural translational landscape. Although mammalian cells can be lysed with detergents, stopping translation immediately, bacterial and yeast cells have to be harvested from culture and lysed in a later step. To arrest ribosomes prior to harvesting, antibiotics that inhibit elongation were often added in early studies. However, treating bacteria with chloramphenicol (Cm) or Saccharomyces cerevisiae with cycloheximide prior to harvesting cells introduces artifacts by trapping ribosomes near the 5′ ends of ORFs as elongation is blocked while initiation continues (18, 19). These artifacts skew ribosome coverage on genes depending on their length. Moreover, Cm alters the footprint distribution along ORFs because its activity is context dependent: it is more effective at stopping ribosomes at some codons than at others (19, 20). We recommend rapidly filtering cultures without antibiotic pretreatment followed by flash-freezing cell pellets in liquid nitrogen (18). Filtration and flash-freezing arrest the ribosomes fast enough for Ribo-seq analyses in most species, including sORF detection, without the need for antibiotics.

Different methods may be required to harvest cells and arrest translation in establishing Ribo-seq in new species. An alternative harvesting method is to spray ∼50 mL of culture directly into liquid nitrogen (19). This method was developed in E. coli to prevent artifactual ribosome pauses at Ser and Gly codons induced by filtration. Even though these pauses do not interfere with genome annotation, direct flash-freezing of cultures is useful for harvesting of cultures of species that cannot be filtered efficiently, such as Haloferax volcanii (13). If neither filtration nor flash-freezing is appropriate (e.g., for pathogens where the spraying method could generate aerosols), a third approach using fast chilling can be applied. This approach is based on the rapid chilling of cultures in ice-water baths with shaking, collecting the cells by centrifugation in a prechilled rotor, and freezing the cell pellets in liquid nitrogen. The latter approach has been successfully applied to several bacterial and archaeal species that did not yield robust polysomes by filtration or flash-freezing (C. M. Sharma, unpublished data).

Cell lysis.

When setting up Ribo-seq in a new organism, different cell lysis methods might need to be tested in order to identify a protocol that prevents polysome disassembly while efficiently lysing cells. Different lysis methods include sonication, cryomill-based grinding, high-pressure homogenization, bead beating, or lysozyme treatment coupled with freeze-thaw cycles. Since cell lysis releases RNA into the lysate, the efficiency of cell lysis can be monitored by quantifying the RNA concentration in the lysates based on A₂₆₀ values (1 A₂₆₀ unit corresponds to roughly 40 μg/mL RNA): higher values correspond to better lysis. Alternatively, lysates can be plated; if large numbers of viable bacteria are recovered, lysis was incomplete. It is important to balance the strength of the lysis procedure by optimizing either the time of lysis (e.g., short sonication cycles) or the intensity of lysis (e.g., the frequency of grinding when using a cryomill) because lysis conditions that are too harsh can lead to mRNA degradation or split ribosomes into subunits. Another key consideration is the composition of the lysis buffer: while the concentrations of Mg²⁺ and monovalent salt ions are critical for ribosome stability, they should also be compatible with the nucleases used to generate ribosome footprints in the subsequent step. Therefore, prior to performing the actual Ribo-seq experiment, we recommend optimizing the cell harvesting and lysis methods for the efficient recovery of polysomes, whose integrity can be monitored by sucrose gradient fractionation. It should also be kept in mind that growth conditions may impact the yield and quality of polysomes.

Generating and isolating ribosome footprints.

The digestion of unprotected mRNA by nucleases creates ribosome footprints. Optimizing this reaction is important: if there is too little nuclease activity, the footprints will not be trimmed to the ribosome boundaries; if there is too much, cleavage of rRNA will lead to fragmentation and decomposition of 70S ribosomes. Both the selection of the RNase and optimal digestion conditions are important. Several RNases have been used for Ribo-seq; their efficiencies in mRNA digestion and their effects on ribosome stability differ between organisms (21, 22). For example, RNase I, which is the nuclease of choice in yeast, leads to the rapid degradation of ribosomes in Drosophila melanogaster. RNase I is inactive in E. coli lysates, likely because it is derived from E. coli and is inhibited by binding to the 30S ribosomal subunit (23). RNase I, however, is active in the lysates of some other bacteria, including Helicobacter pylori (L. Hadjeras and C. M. Sharma, unpublished data) and Bacillus subtilis (A. R. Buskirk, unpublished data). Building on early E. coli studies, most Ribo-seq studies in bacteria have used micrococcal S7 nuclease (MNase). While MNase can be used in many different organisms, the conditions have to be optimized by varying the concentration of the enzyme, the duration and temperature of the reaction, and the Ca²⁺ concentration. For some organisms, mRNA can be extremely unstable in lysates, in particular during the digestion step, even in the absence of added nucleases (e.g., MNase). The stability of polysomes can be assessed by mock digestions lacking added nucleases. We recommend varying the nuclease treatment conditions and looking for evidence of successful digestion based on the nuclease-specific collapse of polysomes to a single monosome peak on sucrose gradients. To prevent overdigestion, the integrity of 16S and 23S rRNAs can be checked by gel electrophoresis before and after RNase treatment.

Following RNA extraction, ribosome footprints are selected based on their length from a denaturing polyacrylamide gel. In bacteria, ribosome footprints vary in length (15 to 40 nt) regardless of the nuclease used. This broad distribution seems to be due in part to the ribosome itself. One contributing factor is that the 16S rRNA 3′-terminal segment can base pair with the mRNA during nuclease treatment (24). This base pairing can affect the footprint size: because purine-rich motifs at the 5′ ends of mRNAs base pair with the pyrimidine-rich anti-Shine-Dalgarno (anti-SD) sequence in 16S rRNA, the resulting footprints are protected from digestion and tend to be longer than other footprints. In addition, the state of the ribosome, A-site occupancy, and salt concentrations in the lysis buffer can also contribute to the heterogeneity of the footprint length (1, 25). Until we have a better understanding of the source of heterogeneity in ribosome footprint lengths, we recommend isolating and sequencing a broad range of footprints arising from monosomes (15 to 40 nt); unwanted reads can always be filtered out computationally during the analysis of the data.

One limitation of bacterial Ribo-seq studies that arises from this broad distribution of footprint lengths is that the data typically do not show the 3-nucleotide periodicity characteristic of the ribosome’s reading frame as is common for eukaryotic data sets. This lack of information about reading frames makes it difficult to tease apart the translation of overlapping ORFs given that the density of ribosomal footprints in overlapping regions cannot be unambiguously assigned to either ORF, as can be done in yeast or higher eukaryotes (8). A word of caution: E. coli Ribo-seq data show a modest 3-nucleotide periodicity that could easily be misinterpreted as evidence of elongating ribosomes. Unlike in yeast, where >90% of reads align to the first nucleotide of codons, the periodicity in E. coli is weak: 40% of reads align to the first nucleotide, 30% align to the second, and 30% align to the third. The periodicity in the E. coli data arises from artifacts of the nuclease digestion and not from the reading frame of the ribosome. Total RNA-seq samples prepared by MNase digestion show the same weak periodicity exclusively in ORFs even in the absence of intact ribosomes (26). MNase cleaves more efficiently before A and T. Because codons are used at different frequencies within ORFs, A and T occur more often than expected at the second nucleotide of codons in the genome. Together, this bias in the nucleotide composition of the genome and the sequence specificity of MNase yield the weak periodicity seen in Ribo-seq data from E. coli (26).

A way to improve the 3-nucleotide periodicity to determine the authentic reading frames in bacterial Ribo-seq studies could be the use of alternative nucleases. The endonuclease RelE, for example, cleaves mRNA within the ribosomal A site precisely after the second nucleotide of the A-site codon (27). The addition of both MNase (to digest naked mRNA back to the ribosome 5′ boundary) and RelE (to generate clean 3′ ends) to E. coli lysates generates footprints that show robust 3-nucleotide periodicity, enough to detect even programmed frameshifting events (26). This approach is generally applicable because RelE is known to bind ribosomes and cleave mRNA in a variety of species, including eukaryotes (28).

Depletion of rRNA fragments.

Approximately 30 to 90% of the 15- to 40-nt RNA fragments extracted from the sizing gel in a typical Ribo-seq library correspond to rRNA fragments. The first bacterial Ribo-seq protocols suggested hybridizing several biotinylated oligonucleotides to the most problematic rRNA contaminants, depleting them using streptavidin beads. While kits based on hybridization and subtraction were reasonably effective at removing rRNA (to just ∼20% of the library), their continuous commercial availability is uncertain, and their efficiency in removing rRNA in the diverse bacterial and archaeal species remains to be tested. The common alternative rRNA depletion kits rely on annealing oligonucleotide probes complementary to rRNA followed by nuclease digestion. This is problematic for Ribo-seq because the nuclease degrades the ends of the ribosome footprints, reducing the resolution and the 3-nucleotide periodicity in the data (29). Therefore, we recommend avoiding rRNA removal kits involving nucleases in Ribo-seq experiments. Hybridization and subtraction with several oligonucleotides (optimized for the species of interest) remain a practical solution, although skipping rRNA depletion altogether and simply increasing the sequencing depth may be a more universal and cost-effective solution.

The required sequencing depth of the resulting ribosome footprint libraries depends on the downstream analyses. To get a reasonable idea of library quality and changes in gene expression, 1 million reads that uniquely map to the genome (4.6 Mb in E. coli) are usually sufficient. Ten million uniquely mapped reads are enough for most applications addressing gene expression, including sORF discovery. Detailed analyses of ribosome pausing and other translational events leading to the redistribution of ribosomes along mRNAs would benefit from an even higher number of uniquely mapped reads. In addition, RNA-seq data based on sequencing libraries of total RNA generated from fragments of lengths similar to those of the ribosome footprints (∼30 nt) or using commercial RNA-seq kits (e.g., TruSeq) facilitate mapping transcript boundaries, assessing the enrichment of ribosome footprints in ORFs, and carrying out differential expression analysis.

Analysis of Ribo-seq data still represents one of the major obstacles for the technique to be widely adopted. Although such analyses are typically done primarily with custom R or Python scripts, software packages that handle Ribo-seq data are now available. (A comprehensive review of these resources and their strengths and weaknesses can be found in reference 30.) A software package that is particularly relevant for bacterial genome annotation is HRIBO (high-throughput annotation by Ribo-seq) (31); it provides a reproducible and high-throughput workflow for analysis of bacterial Ribo-seq data. The HRIBO software performs all required preprocessing and quality control steps, provides annotation-independent ORF predictions based on two complementary bacterium-focused prediction tools (REPARATION [32] and DeepRibo [33]), and integrates them with additional feature information and expression values.

REFINED Ribo-seq PROTOCOLS: MAPPING TRANSLATION INITIATION AND TERMINATION SITES

Identifying sORFs from conventional Ribo-seq data can be difficult. The low ribosome density associated with sORFs is one limitation. A more significant problem is assigning the observed ribosome footprints to a specific sORF because multiple such ORFs in different reading frames are often compatible with the ribosome density observed in conventional Ribo-seq data sets (Fig. 2A). One approach to improve the detection of authentically translated sequences is the use of antibiotics that, by arresting the ribosome at specific sites within an sORF, define its boundaries as well as verify its “translatability.”

FIG 2.

(A) E. coli ribosome density upstream of corA is difficult to reconcile with putative ORFs starting with AUG and GUG start codons. Putative start codons (AUG and GUG) and stop codons (UAA, UAG, and UGA) in each of the three frames are indicated by short and long bars, respectively. The precise start peak in Ribo-seq from retapamulin-treated cells reveals that the density can be attributed to a novel sORF (ysgDI) in frame 1 (dotted arrow) that starts with UUG. The complete sequence of the ysgD gene is shown in orange at the bottom, with the putative SD sequence boxed. The 3′-mapped ribosome footprint data were reported previously (1, 14). (B) Ribosome density on the ndk gene in a negative-control sample and samples treated with the antibiotics tetracycline (Tet), retapamulin (Ret), and apidaecin (Api) (with and without additional puromycin [Pmn] treatment). The 3′-mapped ribosome footprint data were reported previously (14, 15, 50).

Antibiotics that trap ribosomes at start codons.

Translation inhibitors that arrest the ribosome at start codons have been particularly useful for mapping cryptic genes in eukaryotes. Harringtonine and lactimidomycin trap ribosomes at start codons but allow elongating ribosomes to run off, enriching the ribosome density at initiation sites (6, 34). While harringtonine works in Haloferax volcanii (13), neither of these compounds is active in bacteria, and antibiotics with a similar mechanism of action initially were unavailable. Although a number of antibiotics are classified as translation initiation inhibitors in bacteria, most of them either prevent the association of mRNA or initiator tRNA with the small ribosomal subunit or interfere with the formation of the 70S initiation complex (35) and are therefore not practical for the generation of Ribo-seq footprints, which are specific for 70S ribosomes bound to mRNA. However, in recent years, several antibiotics have been identified that can be applied for genome-wide mapping of start codons. Each of these drugs has advantages and drawbacks, as discussed below.

The first antibiotic successfully used for mapping E. coli start codons by Ribo-seq was tetracycline (15). Tetracycline binds to the small ribosomal subunit, hindering the binding of aminoacyl-tRNA in the ribosomal A site (35–37). Although traditionally, this drug was considered to be an elongation inhibitor (38), pioneering experiments of Nakahigashi and coworkers showed that in E. coli cells briefly treated with tetracycline, a significant fraction of ribosomes are arrested at start codons (Fig. 2B) (15). The broad spectrum of tetracycline makes it suitable for use with a wide range of Gram-positive and Gram-negative bacteria, assuming that its mode of action does not change dramatically from species to species. However, the arrest of ribosomes at start codons in E. coli by tetracycline is imperfect, and a relatively high background of ribosome footprints within the coding sequence makes interpretation of the data difficult, especially when applied to sORFs.

The peptide antibiotic oncocin (or, rather, its semisynthetic derivative Onc112) offers a “cleaner” option than tetracycline. Oncocin, which belongs to the class of proline-rich antimicrobial peptides (PrAMPs), binds in the nascent peptide exit tunnel of the large ribosomal subunit and inhibits protein synthesis (39–42). Onc112 traps ribosomes at start codons by invading the A site of the peptidyl-transferase center (PTC), blocking the binding of aminoacyl-tRNA, and preventing the formation of the very first peptide bond. The binding of Onc112 requires the exit tunnel to be vacant. Onc112 cannot associate with elongating ribosomes (39–42), whose exit tunnel is occupied by nascent peptides, and thus functions as a specific inhibitor of the initiating ribosome. Onc112 has been successfully used for mapping start codons in E. coli, identifying many new short ORFs in the E. coli genome (1). The major drawback of Onc112 is that it is imported into the bacterial cell by a special transporter (SbmA in E. coli). While a number of Gram-negative species express SbmA or similar transporters and are susceptible to Onc112 inhibition (43), the majority of Gram-positive bacteria and many Gram-negative species are intrinsically resistant. In a promising study, the sequence of PrAMPs was altered to identify peptides that no longer rely on SbmA for transport, broadening their spectrum of action (44). Currently, these PrAMPs and Onc112 are not available commercially, but they can be custom ordered from a number of synthetic peptide providers.

Another inhibitor that efficiently arrests ribosomes at start codons is the small-molecule antibiotic retapamulin, which belongs to the class of pleuromutilins. These antibiotics bind in the PTC active site spanning both P and A sites (45, 46). Pleuromutilins clash with peptidyl-tRNA and therefore, like Onc112, cannot bind to elongating ribosomes, which are always associated with tRNAs carrying the growing protein chain. However, retapamulin can bind to ribosomes carrying the initiator fMet-tRNA, blocking an aminoacyl-tRNA from entering the A site and arresting ribosomes at start codons (Fig. 2B). Retapamulin-assisted Ribo-seq, dubbed “Ribo-RET,” helped to identify new sORFs in commensal and pathogenic E. coli strains (1, 14, 47). This approach was also useful in mapping internal start codons in some known protein-coding genes and identifying sORFs present within the alternative reading frames enclosed in the body of several E. coli genes (1, 14). One advantage of retapamulin in comparison with Onc112 is its availability from commercial sources at a relatively low cost. However, even though retapamulin acts upon a broad range of Gram-positive organisms, it has limited activity against Gram-negative bacteria due to poor uptake and active efflux by the multidrug transporters. In the case of E. coli, this called for the use of strains with an inactivated tolC transporter gene (14). The newer pleuromutilin drug lefamulin, which was recently approved for clinical use and which, like retapamulin, is commercially available, has been reported to have increased activity against Gram-negative species and may potentially surpass retapamulin for mapping translation start sites in a broader array of bacterial species.

While arresting ribosomes at start codons by antibiotics is useful for mapping 5′ ends of open reading frames in bacterial genomes, none of the described inhibitors yields an absolutely clear picture of ribosome footprints. As is common with Ribo-seq, spurious peaks of ribosome density can be observed at seemingly random sites. On many occasions, high footprint density can be found at noncanonical start codons. It is unclear whether peaks at these sites reflect true translation initiation or whether higher levels of nontranslating ribosomes in the drug-treated cells lead to their spurious (illegitimate) binding to mRNA. This question can be addressed in part by comparing Ribo-seq data obtained from drug-treated cells with those from untreated controls: the footprints of elongating ribosomes downstream from the putative noncanonical translation initiation site validate the translation of the ORF.

Antibiotics that trap ribosomes at stop codons.

Our confidence in identifying protein-coding sequences would be significantly improved if start codon mapping approaches could be supplemented with a comparable technique for mapping termination sites. Unfortunately, until recently, no specific inhibitors of translation termination were known (35). A new lead came from the studies of another PrAMP, apidaecin (Api), an 18-amino-acid-long unmodified antibacterial peptide produced by honeybees. A modified version of apidaecin, Api137, exhibits activity superior to that of the native peptide (48). Studies of the mechanism of Api137 action showed that it binds in the nascent peptide exit tunnel of ribosomes after the release of the completed protein but prior to the departure of the release factors (49). Api137 interacts with release factors, sequestering them on the ribosome, so that ribosomes arrest at stop codons due to the absence of release factors.

Identifying Api as a translation termination inhibitor raised hopes that Ribo-seq analysis of cells treated with Api137 could allow mapping the ends of the translated sequences. However, the results of the pilot Api137-assisted Ribo-seq profiling experiments turned out to be fairly complicated because ribosomes trapped at stop codons caused queuing of the trailing elongating ribosomes, often generating ramps of ribosome density at the 3′ ends of ORFs (50). Ribosome footprints were also observed downstream of stop codons in Api137-treated cells because the sequestration of release factors by Api137 causes stop codon readthrough (Fig. 2B). Potentially, both of these problems, ribosome queuing and stop codon readthrough, could be eliminated by treatment with puromycin, which is expected to release ribosomes carrying peptidyl-tRNA without affecting ribosomes directly trapped by Api137 at stop codons. Indeed, puromycin treatment significantly enhanced the relative height and definition of the peak of ribosomal footprints at stop codons (Fig. 2B). However, the optimal conditions for the combined Api137 and puromycin treatment are not yet worked out.

Like oncocin, apidaecin sensitivity seems to be limited to Gram-negative species that express SbmA-like transporters (43, 49, 51). However, due to the “biological nature” of these PrAMPs, they can be directly expressed in the bacterial cell after introducing their expression from an appropriate plasmid vector. Induction of PrAMP expression followed by Ribo-seq would help to bypass the uptake problem. Although the utility of such an approach for mapping the start or stop codons has not been experimentally tested, the expression of functionally active Onc112 and apidaecin in bacteria has been amply demonstrated (52–54). The endogenous expression of the PrAMP gene under the control of an inducible promoter may expand the PrAMP-based mapping of start and stop codons to species that are naturally resistant to these inhibitors.

In addition to inhibitors of translation initiation or termination, inhibitors of translation elongation can also be employed to increase the accuracy of assigning ribosome footprints to a specific sORF and defining the coding frame. In cells treated with inhibitors of specific aminoacyl-tRNA synthetases, ribosomes stall at “hungry” codons corresponding to the missing aminoacyl-tRNA (19, 55). The resulting codon-specific pauses could help to define which of the putative sORFs in alternative frames at the regions of high Ribo-seq density corresponds to the translated sequence. This approach would work only for sORFs that contain the codons affected by the inhibitors.

In summary, the use of antibiotics with specific mechanisms of action in conjunction with the Ribo-seq approach may significantly increase the confidence and precision of the identification of sORFs in bacterial genomes.

CONCLUDING REMARKS

The advent of Ribo-seq technology has revealed the existence of previously hidden genes in prokaryotic genomes that have been overlooked because of their small size, the use of noncanonical start codons, or their unusual location (e.g., encased within other coding sequences). Here, we have described some of the general challenges in applying Ribo-seq for mapping the cryptic proteome in bacteria (more specifically sORFs) and have presented approaches for solving such problems. At present, most of these approaches have been tested or optimized using only a few conventional laboratory species. Moving into new organisms will likely require optimizing the critical steps in the Ribo-seq protocols as well as expanding experimental and bioinformatic strategies for mapping the boundaries of the protein-coding sequences. Advancing Ribo-seq for mapping the translatome in archaea has just begun (13) and will certainly bring about new challenges requiring innovative solutions.

Even with the refinements in the Ribo-seq method described here, it is still important to ask critical questions about whether a signal in a Ribo-seq data set really represents the translation of a given sequence in the cell. Ideally, having several lines of evidence from different Ribo-seq methods (i.e., with and without antibiotics) increases the likelihood of correct annotation. The presence of ribosome binding sites upstream of the start codon and the conservation of the sORF in closely related species also increase the likelihood that an sORF is expressed. But since all of these characteristics of sORFs have their shortcomings, ultimately, the expression of candidate sORFs should be validated by the detection of the small protein product by proteomic methods or by inserting an epitope tag or a translational reporter fusion to green fluorescent protein (GFP) or lacZ at the genomic locus and detecting the tagged protein with antibodies on a Western blot.

Undoubtedly, the rate of accumulation of new bacterial and archaeal Ribo-seq data sets will increase. The treasure trove of candidate sORFs in these data will offer unprecedented opportunities for unraveling principles of differential translation under various growth conditions and stresses and the functions of small proteins. Therefore, there is an urgent need for powerful and user-friendly tools for analysis of these data and integrating them with genome annotations. Overall, the combination of the new Ribo-seq and bioinformatics approaches will illuminate the role of the previously cryptic proteome in bacterial physiology.