Principles, mechanisms, and biological implications of translation termination–reinitiation

Abstract

The gene expression pathway from DNA sequence to functional protein is not as straightforward as simple depictions of the central dogma might suggest. Each step is highly regulated, with complex and only partially understood molecular mechanisms at play. Translation is one step where the “one gene–one protein” paradigm breaks down, as often a single mature eukaryotic mRNA leads to more than one protein product. One way this occurs is through translation reinitiation, in which a ribosome starts making protein from one initiation site, translates until it terminates at a stop codon, but then escapes normal recycling steps and subsequently reinitiates at a different downstream site. This process is now recognized as both important and widespread, but we are only beginning to understand the interplay of factors involved in termination, recycling, and initiation that cause reinitiation events. There appear to be several ways to subvert recycling to achieve productive reinitiation, different types of stresses or signals that trigger this process, and the mechanism may depend in part on where the event occurs in the body of an mRNA. This perspective reviews the unique characteristics and mechanisms of reinitiation events, highlights the similarities and differences between three major scenarios of reinitiation, and raises outstanding questions that are promising avenues for future research.

INTRODUCTION

Protein production from a mature eukaryotic messenger RNA (mRNA) appears deceptively simple: each mRNA's protein-coding open reading frame (ORF) is defined by a translation initiation site and a corresponding termination site. This ORF is accurately and precisely recognized by ribosomes that translate the sequence to yield a functional protein. While appealing, this description is inadequate. Although most eukaryotic mRNAs contain a single “main” ORF encoding one polypeptide product, it is common for multiple peptides to be synthesized from a single mRNA. This is often due to “translation reinitiation,” which occurs when a ribosome that has started making protein terminates at a stop codon, does not fully undergo all recycling steps, then starts again (“reinitiates”) on the same mRNA but on a different ORF. Reinitiation can achieve translation of ORFs with start codons downstream from, upstream of, or within the main ORF, but in all cases more than one protein is made from one mRNA by a single ribosome. This form of translation regulation appears to have many important roles, for example to rapidly respond to environmental changes (Lu et al. 2004; Liu and Qian 2014; Young and Wek 2016; Zhou et al. 2018). Reinitiation is therefore an important part of the rich repertoire of mechanisms used to regulate translation in eukaryotes.

Translation is divided into four phases: initiation, elongation, termination, and ribosome recycling. Reinitiation requires that a ribosome initiates and elongates, reaches a stop codon, and terminates, but does not undergo the complete recycling process. Specifically, reinitiation can occur when an event following peptide release either fails or is purposely evaded, resulting in the ribosome remaining associated with the mRNA instead of recycling (Dever and Green 2012). The translation of multiple ORFs within the same mRNA by the same ribosome differentiates reinitiation from other noncanonical translation events (Fig. 1). Broadly, the types of reinitiation can be divided into three categories (Fig. 2). The most well-documented cases involve initiation on relatively short upstream ORFs (uORFs) followed by reinitiation at the main ORF (scenario 1). Reinitiation can also occur downstream from the main ORF (scenario 2) on cellular or viral mRNAs, or following termination within the main ORF (scenario 3) brought about by a nonsense mutation in an mRNA.

FIGURE 1.

Comparison of translation termination–reinitiation to other unique types of initiation and translation events. Cartoon representations of ribosomes on mRNAs undergoing (A) reinitiation, (B) intercistronic internal ribosome entry site (IRES)-driven initiation, (C) frameshifting, (D) stop codon readthrough, (E) 2A-mediated peptide bond skipping, and (F) leaky scanning. All possible protein products produced from each transcript are depicted, and the key features, requirements, and outcomes of the scenarios caused by these translation events are summarized on the right.

FIGURE 2.

Characteristics of mRNAs that support translation reinitiation. Different gene organization contexts lead to three reinitiation scenarios. Short upstream open reading frames (uORFs) are the hallmark of scenario 1, with reinitiation occurring in what would otherwise be considered the 5′ untranslated region. A substantially longer first ORF would lead to scenario 2, where reinitiation accesses a downstream ORF (dORF) in what would otherwise be considered the 3′ untranslated region. Scenario 3 happens when reinitiation occurs within an ORF, typically brought on by a premature termination codon. Green: start codon; red: stop codon; black boxes (scenario 2): sequences/structure motifs for termination/reinitiation coupling.

The state of the reinitiation field was thoroughly reviewed by Gunisova et al. (2018). Rather than presenting a comprehensive review of relevant literature, we briefly compare these three scenarios with an eye toward spotting common themes hinting at underlying principles that link these seemingly disparate scenarios. We put reinitiation into the context of the canonical translation cycle, including recent studies that have clarified the roles of many recycling factors and their contributions to reinitiation (Ahmed et al. 2018; Young and Guydosh 2019, 2022; Bohlen et al. 2020b; Young et al. 2021). We also compare the distinct mechanisms and functional outcomes between reinitiation and other types of noncanonical translation events. Finally, we consider how reinitiation directly affects different biological outcomes and disease states and discuss several lingering questions in the field. Answers to these questions promise not only a deeper understanding of the mechanisms of reinitiation and its importance in translation regulation, but new insight into how reinitiation affects human health.

DISTINGUISHING REINITIATION FROM CANONICAL AND OTHER NONCANONICAL TRANSLATION MECHANISMS

Reinitiation deviates from the canonical translation cycle

A typical round of translation in eukaryotes begins with recruitment of the small (40S) ribosomal subunit and initiation factors to the mRNA, scanning until a start codon is recognized and the large (60S) ribosomal subunit joins, then protein synthesis commences. After elongating through the ORF, a stop codon is recognized, and release factors liberate the protein product. To complete a cycle of canonical translation, ribosomal subunits must dissociate from the mRNA through a process termed recycling to be made available for translation of other mRNAs.

In the simplest sense, translation reinitiation is a consequence of inefficient or defective ribosome recycling at termination sites. Following the failed recycling event, reinitiation entails another round of translation by the same ribosome on the same mRNA but at a different ORF. The functional outcome of reinitiation is two protein products translated from the same mRNA. While reinitiation is an outcome of incomplete ribosome recycling, the process also interfaces with the termination and initiation phases and therefore involves some of the factors associated with those processes; we briefly describe the roles of these factors in canonical translation for comparison later in the review.

Translation initiation is a highly coordinated and complex process involving numerous eukaryotic initiation factors (eIFs) (Jackson et al. 2010; Aitken and Lorsch 2012; Hinnebusch and Lorsch 2012). The 40S subunit of the ribosome along with eIF1, eIF1A, the ternary complex including eIF2, which delivers initiator methionyl-tRNA (Met-tRNA_i), eIF3 (a large complex consisting of many individual proteins), and eIF5 comprise the preinitiation complex. The other major players are the eIF4F complex, which binds the mRNA cap and mediates interactions with the polyA tail, and eIF5B, which mediates joining with the 60S subunit. Different circumstances of reinitiation events lead to large variation in their resemblance to canonical initiation in that some involve many of these factors while others require none of the canonical initiation factors.

Following elongation, a stop codon (UAA, UAG, UGA) in the A-site of the ribosome is recognized by eukaryotic release factor 1 (eRF1) which, in conjunction with eRF3 performing a GTP hydrolysis event, mediates release of the polypeptide chain from the tRNA in the P-site (Dever and Green 2012; Hellen 2018). The first step of recycling is removal of the 60S subunit, which involves ABCE1 (Rli1 in yeast) and eIF3j (Hcr1 in yeast) (Pisarev et al. 2010; Young et al. 2015; Heuer et al. 2017; Young and Guydosh 2019). The termination and recycling processes are closely linked, in that ABCE1 requires that eRF1 remain bound in the A-site of the ribosome after eRF3 departs for successful 60S splitting and ABCE1 speeds up peptide release during termination (Khoshnevis et al. 2010; Pisarev et al. 2010; Shoemaker and Green 2011). To complete recycling, the deacylated P-site tRNA and mRNA must be removed from the 40S subunit, which can then go on to perform subsequent rounds of translation after 60S rejoining. These later steps of recycling are primarily performed by eIF2D (Tma64 in yeast; also referred to in many previous studies as Ligatin) or a heterodimer consisting of multiple copies in T-cell lymphoma-1 (MCT-1; Tma22 in yeast) and density regulated protein (DENR; Tma20 in yeast) (Skabkin et al. 2010; Hellen 2018; Young et al. 2018, 2021; Bohlen et al. 2020b; Gaikwad et al. 2021; Young and Guydosh 2022). The DENR/MCT-1 heterodimer resembles eIF2D, with MCT-1 and DENR sharing significant homology with the amino- and carboxy-terminal portions of eIF2D, respectively (Lomakin et al. 2017; Weisser et al. 2017). Recycling steps can also be performed by canonical initiation factors eIF3, eIF1, eIF1A, and eIF3j in vitro (Pisarev et al. 2007). The role of many “initiation” factors in recycling and direct interactions between recycling and initiation factors (Chen et al. 2006; Pisarev et al. 2007; Heuer et al. 2017; Young et al. 2018; Young and Guydosh 2019), highlight the intersection of recycling and initiation by preparing the 40S for the next initiation event as it is removed post-termination. Reinitiation events lie at the intersection of recycling and initiation and many recycling and initiation factors—especially eIF2D and DENR/MCT-1—have been strongly implicated in reinitiation (Schleich et al. 2014, 2017; Ahmed et al. 2018; Castelo-Szekely et al. 2019; Bohlen et al. 2020b), as explored in more detail in subsequent sections.

Comparing reinitiation with other noncanonical translation events

Other alternative translation mechanisms can cause some of the same functional outcomes as a reinitiation event (i.e., translation of a downstream coding region), but none of these depend on a preceding translation termination event. Another characteristic of translation termination–reinitiation that distinguishes this mechanism is the production of two polypeptide chains from a single RNA transcript (excluding start-stop uORFs), by the same ribosome. Below we briefly discuss some of these other types of translation initiation and elongation events, including similarities with reinitiation (Fig. 1).

Some or nearly all of the requirements for canonical translation initiation can be bypassed when an internal ribosome entry site (IRES) (Fig. 1B) is present upstream of the coding region (Filbin and Kieft 2009; Mailliot and Martin 2018; Martinez-Salas et al. 2018). Certain types of IRESs require few or no initiation factors to recruit ribosomes and need not be in the 5′ untranslated region (UTR) to initiate translation (Jan and Sarnow 2002; Pestova and Hellen 2003; Jaafar et al. 2016). Such IRESs can initiate translation of a downstream coding region from the intergenic region (Wilson et al. 2000). Functionally, this appears like reinitiation. The key distinction in the case of the IRES is that translation of the downstream IRES-driven coding region does not need to be performed by the same ribosome as the upstream coding region and termination at the stop codon of the upstream ORF is not a requirement for translation of the downstream ORF.

Programmed ribosomal frameshifting (Fig. 1C) accomplishes translation of downstream coding regions that would not normally be accessed during canonical translation elongation (Harger et al. 2002; Ketteler 2012; Caliskan et al. 2015; Riegger and Caliskan 2022). When the ribosome slips into a different frame, an alternative polypeptide sequence is synthesized in accordance with the new reading frame and termination happens at the next in-frame stop codon. Programmed ribosomal frameshifting only involves one initiation and one termination event per ribosome passthrough of that transcript. For an mRNA containing a frameshifting signal, the portion of the protein from the amino terminus up to the slippery sequence is constant while the carboxy-terminal portion of the protein differs by accessing two different reading frames, depending on whether frameshifting occurs or not. In contrast to reinitiation, the two different protein products that can be produced from a transcript containing a frameshifting signal are mutually exclusive—a ribosome that translates the mRNA and maintains the 0 frame will create one protein product whereas another ribosome that undergoes frameshifting will produce the other protein product.

Programmed stop codon readthrough (Fig. 1D) also accomplishes translation of an alternate downstream coding region (Wills et al. 1991; Li and Rice 1993; Steneberg and Samakovlis 2001; Firth et al. 2011; Loughran et al. 2018). This mechanism evades the termination process at the first stop codon and instead elongation continues until the next in-frame stop codon is reached. As with frameshifting, only one “round” of translation occurs per ribosome pass, as the peptide is not cleaved and released when readthrough occurs at the first stop codon, leading to a carboxy-terminal extension of the resulting protein.

The 2A self-cleaving peptide (Fig. 1E) is another example of a translation event that enables the production of two separate polypeptides from a single mRNA. These viral-derived elements depend on a specific amino acid sequence encoded within the mRNA that promotes peptide bond skipping at a specific position (de Felipe et al. 2003; Luke et al. 2008; Brown and Ryan 2010), leading to the liberation of the amino-terminal portion of the peptide before the entire round of translation elongation has occurred. The functional outcome of two separate polypeptides resulting from one “round” of translation by each ribosome per transcript is the same for 2A peptide-containing mRNAs as reinitiation. Mechanistically, 2A peptide-induced peptide bond skipping is likely distinct from reinitiation. While termination factors have been shown to play a role in 2A-mediated peptide bond skipping (Doronina et al. 2008), the mechanism as well as other potential factor requirements are still largely uncharacterized, making it difficult to fully draw a comparison with reinitiation at this time.

Leaky scanning is the process by which the 48S preinitiation complex scans past a putative start codon and instead initiates at an alternative downstream start codon (Fig. 1F). The mechanism and characteristics for leaky scanning are distinct from translation termination–reinitiation, primarily in that there is only one “round” of translation per transcript. Different functional outcomes can be achieved by leaky scanning depending on factors such as the proximity, arrangement, and relative reading frames accessed by each putative start codon. Some functional consequences of leaky scanning are regulating the amount of protein produced from the main ORF or the production of a protein with an amino-terminal truncation (as depicted in Fig. 1F). Additionally, it can be extremely difficult to experimentally distinguish and/or exclude leaky scanning as the mechanism for certain apparent 5′ proximal reinitiation events. We discuss the situations where leaky scanning can induce analogous outcomes and address these experimental challenges in later sections.

Many of these noncanonical translation mechanisms occur at a specifically dictated frequency. Some ribosomes that translate the mRNA will undergo a completely canonical round of translation from initiation to recycling while other ribosomes on that same transcript will be affected by a certain signal in a way that causes the noncanonical translation event. Many known cases of reinitiation, frameshifting, stop codon readthrough, and leaky scanning occur in the minority of translation events on transcripts containing those signals. Another commonality between these mechanisms is that they are all used by viruses to create different viral protein products and/or regulate relative expression of different viral proteins. Furthermore, these noncanonical translation events can be influenced by, or in some cases are wholly dependent on, RNA sequences and structures within the viral mRNAs.

THE THREE REINITIATION SCENARIOS

Scenario 1: Reinitiation after translation of an upstream open reading frame

It is estimated that 30%–60% of plant and metazoan mRNAs possess one or more short ORFs of 10–30 codons upstream of the main ORF (Fig. 2; Calvo et al. 2009; Von Arnim et al. 2014; Causier et al. 2022; Chothani et al. 2022). Because these uORFs are usually out of frame with the main ORF, ribosome access to the main ORF depends on either bypass of the uORFs via leaky scanning or a reinitiation event after termination at a uORF stop codon. This scenario thus creates a way to regulate expression of the main ORF. uORF-associated reinitiation can be affected by features including the peptide product encoded by the uORF (Gaba et al. 2001; Bhushan et al. 2010), the availability of the tRNA^Met_i-containing ternary complex (Grant et al. 1994), the nucleotide sequence within and around the uORFs (Grant et al. 1995; Munzarová et al. 2011; Gunišová and Valášek 2014; Gunišová et al. 2016), and the distance between the uORF and the start codon of other ORFs (Grant et al. 1994). A shorter distance between the uORF and main ORF may favor reinitiation if specific initiation factors are still bound to the ribosome while longer distances may allow time for these factors to be rerecruited, as discussed in a later section on initiation factor requirements.

Well-studied examples of scenario 1 are found in the mRNAs encoding Activating transcription factor 4 in mammals (ATF4) and its yeast equivalent, General control nonderepressible protein 4 (GCN4) (Mueller and Hinnebusch 1986; Lu et al. 2004; Vattem and Wek 2004; Hinnebusch 2005). In these mRNAs, stress conditions trigger cellular changes that favor reinitiation on the main ORF following termination at the uORF (Liu and Qian 2014). While the mechanisms of GCN4 and ATF4 regulation are the best understood examples for this scenario, numerous other uORF-containing mRNAs have also been described in detail and reviewed previously (Morris and Geballe 2000; Barbosa et al. 2013; Somers et al. 2013).

Scenario 2: Reinitiation after long open reading frames

In contrast to the short uORF scenario, reinitiation can also occur at the end of a long, “main” ORF. In these cases, what would normally be considered the 3′-UTR is a second protein-coding region whose translation depends on reinitiation, resulting in translation of a second protein product following translation of the main ORF (Fig. 2). Reinitiation this far downstream from the start codon usually relies on additional elements such as specific sequences or RNA secondary structural elements in close proximity to the overlapping ORFs.

The best-documented cases of this type of reinitiation are found in members of the Caliciviridae family and influenza B virus. These viral RNAs contain overlapping ORFs near their 3′ end, with the upstream ORF stop codon and downstream ORF start codon in very close proximity and out of frame, sometimes even sharing nucleotides (e.g., UAAUG, AUGA) (Horvath et al. 1990; Meyers 2003; Luttermann and Meyers 2007; Pöyry et al. 2007; Powell et al. 2008). In these viral examples, a structured RNA element upstream of the start-stop overlap is required to achieve reinitiation (Powell et al. 2011; Luttermann and Meyers 2014; Wennesz et al. 2019). These “termination upstream ribosome binding site” (TURBS) RNA elements require a conserved sequence complementary to a portion of ribosomal RNA, specifically in the loop of expansion segment 9 in the 18S rRNA (Meyers 2003; Luttermann and Meyers 2007; Pöyry et al. 2007). The current model involves tethering of the post-termination 40S subunit to the mRNA through base-pairing with the TURBS structure and subsequent delivery of Met-tRNA^Met_i, likely by either eIF2/1/1A or eIF2D (Zinoviev et al. 2015), inducing reinitiation at a nearby start codon (Meyers 2003).

TURBS-mediated reinitiation is not perfectly efficient, in that not every ribosome that translates the upstream ORF will reinitiate and translate the downstream ORF, leading to different expression levels for each of the two encoded viral proteins. In caliciviruses, reinitiation occurs at a defined frequency, ranging from 3%–20% depending on the virus (Meyers 2003; Luttermann and Meyers 2014), which creates a precise ratio of major capsid protein (upstream ORF) to minor capsid protein (downstream ORF). Reports of reinitiation in viruses outside of the Caliciviridae family that infect diverse eukaryotic hosts, described in more detail below, hint that reinitiation could be a widespread method for viruses to express specific proteins at different levels while maintaining a very compact genome. Although many of these viral reinitiation events depend on specific portions of the viral mRNA, not all appear to rely on interactions with rRNA and each might use unique strategies and cellular and/or viral factors to impede recycling and promote reinitiation.

Reinitiation events in viral mRNAs from metazoan-infecting respiratory syncytial virus (RSV) and other pneumoviruses (Ahmadian et al. 2000; Gould and Easton 2005, 2007), fungal-infecting Cryphonectria hypovirus 1 (CHV1) (Guo et al. 2009), and Helminthosporium victoriae virus 190S (HvV190S) (Li et al. 2015) all depend on a portion of coding region sequence upstream of the stop codon to achieve reinitiation; HvV190S has been proposed to use an RNA pseudoknot structure in this upstream region to achieve reinitiation whereas it is currently unclear exactly which RNA sequences, secondary structure, or encoded amino acids in the upstream region of the viral mRNA are necessary for RSV and CHV1. Reinitiation at a downstream ORF in the polycistronic viral mRNA of cauliflower mosaic virus is dependent on the viral-encoded “transactivator/viroplasmin” protein, which has been found to interact with the 60S subunit, eIF3, and a “reinitiation supporting protein” in the plant host (Park et al. 2001; Thiébeauld et al. 2009; Mancera-Martínez et al. 2021).

Reinitiation also appears to be exploited by some retrotransposons such as SART1, in which RNA structure downstream from the ORF1 stop codon was found to be necessary (Kojima et al. 2005). Bioinformatic analyses of human mRNAs revealed thousands with potential downstream ORFs >50 amino acids in length that overlap the main ORF and extend into the 3′-UTR (Gould et al. 2014). Peptides corresponding to these downstream ORFs were detected for most candidates and relied on specific sequences in the upstream coding region, but in this case reinitiation appears to require specific amino acid sequences rather than RNA structure (Gould et al. 2014). Finally, ribosome profiling in diverse types of eukaryotic cells suggests translation in 3′-UTRs is more common than previously appreciated (Bazzini et al. 2014; Guydosh and Green 2014; Ji et al. 2015; Duffy et al. 2022), possibly due to reinitiation. Interestingly, inhibition of recycling in vivo leads to higher levels of reinitiation in 3′-UTRs (Young et al. 2015; Meydan and Guydosh 2020), but it is unclear if these 3′-UTR translation events or resulting peptides are products of a perturbed cellular state or play specific biological roles.

Scenario 3: Reinitiation within a main open reading frame

Reinitiation within a main ORF becomes relevant when a mutation introduces a premature termination codon (PTC) into an mRNA's protein coding region. PTCs can lead to nonfunctional or detrimental truncated protein products and usually activate nonsense-mediated RNA decay (NMD), which degrades the aberrant mRNAs (Miller and Pearce 2014; Forster et al. 2015; Kurosaki and Maquat 2016; Celik et al. 2017; Nasif et al. 2018; Kishor et al. 2019; Dyle et al. 2020). However, sometimes PTCs in essential genes are far less detrimental than expected if post-termination ribosomes at some PTCs fail to recycle and instead reinitiate at a downstream start codon (Zhang and Maquat 1997; Pereira et al. 2015; Moey et al. 2016). Depending on the location of the PTC and the reinitiation frame, the predominant protein product can retain some residual activity, leading to less-severe disease phenotypes (Fig. 2; Paulsen et al. 2006; Moey et al. 2016). This is not the case for all aberrant mRNAs, as most reinitiation products do not seem to significantly affect the cell's fate (Zoppi et al. 1993). Reinitiation can also increase the half-life of some PTC-containing mRNAs, allowing them to accumulate to near wild-type levels (Zhang and Maquat 1997). When a ribosome reinitiates translation downstream from a PTC, it displaces RBPs that are critical for triggering NMD, such as exon junction complexes and other RBPs. Without the RBPs serving as triggers for NMD, these PTC-containing transcripts escape NMD completely (Hogg and Goff 2010; Annibaldis et al. 2020; Zhu et al. 2020; D'Orazio et al. 2021). Thus, reinitiation and NMD are mutually exclusive fates for PTC-containing mRNAs. Examples of PTCs that evade NMD are found in genes such as ATRX and ATP7A, for which a PTC causes only a mild version of the disease that manifests upon complete loss of the protein product's function (Howard et al. 2004; Paulsen et al. 2006).

Because nonsense mutations are random, the outcome of introducing a PTC may vary from one transcript to another depending on location and context. PTCs near the 5′-most AUG in the main ORF reinitiate and bypass NMD more readily (Zhang and Maquat 1997; Neu-Yilik et al. 2011; Jagannathan and Bradley 2016; Lindeboom et al. 2016). This suggests that brief elongation increases the likelihood of ribosome-bound initiation factors remaining bound to facilitate reinitiation (Bohlen et al. 2020a; Lin et al. 2020; Wagner et al. 2020). In addition, the number of nucleotides between the PTC and the downstream AUG also affects reinitiation and the optimal intercistronic distance varies between mRNAs. Finally, mRNAs predicted to have less stable secondary structure (such as α-globin), which contain nonsense mutations within 32 codons of the AUG—which closely resembles a transcript that falls into the category of scenario 1—have both higher translation rates and reinitiation levels than ORFs with more stable structures (Pereira et al. 2015). Overall, the factors that influence the dissociation rate of ribosome-bound initiation factors during elongation, and how they affect PTC-associated reinitiation remain poorly understood but have direct relevance to many human diseases.

MECHANISTIC PRINCIPLES OF REINITIATION

The three scenarios described above (Fig. 2) illustrate the diverse ways in which reinitiation occurs in biology in terms of the regulatory pathways exploited, spatial organization, complexes involved, and variety of signals produced. Despite these differences, there are also common characteristics between these examples that can provide mechanistic insight and the basis for new hypotheses, as discussed below (Fig. 3).

FIGURE 3.

Internal and external factors and features that influence reinitiation mechanisms. (A) Depending on the step at which recycling fails, the 60S subunit may or may not dissociate (mediated by ABCE1 and eIF3j during ribosome recycling) before a reinitiation event occurs. Failure to remove the P-site tRNA and/or the mRNA from the 40S subunit (mediated by eIF2D or DENR/MCT-1 during ribosome recycling) can also lead to reinitiation events. (B) The identity of the amino acids of the nascent peptide from the upstream ORF that are still in the exit channel at the time of termination/failure to recycle may influence reinitiation frequency. (C) Post-transcriptional modifications of the mRNA, such as N⁶-Methyladenosine (m⁶A) or pseudouridine (Ψ), can act as a regulatory mechanism to regulate the frequency of reinitiation. (D) Depending on the scenario, the start codon can be upstream, overlapping, or downstream from the first stop codon. The requirements vary greatly between the scenarios of reinitiation and a canonical AUG start codon is not always necessary. (E) The proximity of the stop codon of the upstream ORF with relation to the start codon for the downstream ORF is a major determinant for whether reinitiation occurs. (F) Intramolecular secondary and tertiary structures in the mRNA can influence reinitiation frequency or start codon selection, and in some cases reinitiation is entirely dependent on these structural elements. (G) The set of initiation factors necessary for reinitiation have yet to be determined definitively and likely vary greatly between different scenarios or circumstances of reinitiation. If required, initiation factors either remain bound through the first round of translation or might need to be rerecruited. (H) Met-tRNA_i delivery, typically performed by eIF2, could occasionally be carried out by other factors such as eIF2D or DENR/MCT-1 under different conditions and scenarios of reinitiation, although sometimes reinitiation occurs independent of this component.

Reinitiation requires successful termination and failed recycling

The occurrence of two initiation and two termination events per mRNA, performed by the same ribosome (or at least the same 40S ribosomal subunit) is a key feature that is consistent between all three scenarios of reinitiation while distinguishing it from all other noncanonical translation events. The full recycling process can fail for a variety of reasons and at various steps in the recycling process, but the failure to undergo the full course of recycling is a prerequisite for all types of reinitiation.

The location of this failed recycling event within the mRNA and the protein complexes associated with the ribosome at that point influence whether reinitiation occurs. This accounts for some of the resulting mechanistic differences between the three scenarios. The step at which recycling fails likely also contributes to some of the variance observed between different examples within the same reinitiation scenario. The step at which recycling fails to proceed depends on which factors are depleted or inhibited (e.g., ABCE1, eIF3j, eIF2D, DENR/MCT-1), which then determines whether the post-termination 80S complex remains intact, the 60S subunit dissociates but the deacylated P-site tRNA remains, or only the 40S subunit remains associated with the mRNA (Fig. 3A). The state of the post-termination, unrecycled ribosome has significant repercussions in terms of the possible downstream ORFs accessed through reinitiation. If recycling proceeds to the point at which only the 40S remains bound to the mRNA, then many of the stringencies imposed by canonical initiation factors (scanning, start site selection) are reflected in the reinitiation event. In contrast, if the first step of recycling is impeded, then translation by the intact post-termination 80S ribosome can resume in a manner that seemingly bypasses the initiation process and proceeds straight to elongation under certain circumstances.

In general, the point at which recycling fails or the specific factors that are perturbed, depleted, or otherwise manipulated leading to reinitiation are not well understood. Some possible mechanisms that would affect specific steps in the recycling process include regulation of ABCE1 function by oxidative stress (Zhu et al. 2020; Makeeva et al. 2023) overexpression (Gao et al. 2020) or decreased expression during differentiation (Mills et al. 2016), regulation of DENR function through phosphorylation (Clemm von Hohenberg et al. 2022), congenital mutations of recycling factors, such as those to DENR that disrupt reinitiation and cause neurological disorders (Haas et al. 2016), or preventing 40S dissociation from the mRNA through direct tethering by the mRNA, as in TURBS-mediated reinitiation (Luttermann and Meyers 2007; Powell et al. 2008).

Overall, the connections and shared factors between termination, recycling, and initiation reveal how reinitiation can occur when many of the individual processes in translation are manipulated or perturbed. More global changes to translation also affect reinitiation, as demonstrated by the finding that the rate of reinitiation (as well as stop codon readthrough) is elevated under stress conditions, which also coincides with a number of factors (such as ABCE1, eIF2D, MCT-1, DENR) localizing to stress granules (Makeeva et al. 2023).

Relative positioning of termination and reinitiation start site

The three reinitiation scenarios have strict requirements for the placement of the stop codon (Fig. 3E) relative to the reinitiation start codon, that is, the intercistronic distance. However, the requirements differ substantially between scenarios in terms of proximity, overlap, and relative reading frame.

Regarding proximity, in both scenarios 1 and 3, the reinitiation start codon is always downstream from the stop codon, while in scenario 2 the reinitiation start codon can be very close to or even upstream of the stop codon (Fig. 2). For example, all viral genomes that use a TURBS RNA-dependent reinitiation mechanism (scenario 2) have the two codons within 20 nt of each other and reinitiation efficiency decreases if they are moved more than ∼30 nt apart (Luttermann and Meyers 2007; Pöyry et al. 2007). The distance requirements between the stop and start codons for scenario 2 in eukaryotic mRNAs are currently unknown; reinitiation on these mRNAs has mostly been investigated in the context of ORFs with some overlap (Gould et al. 2014). For scenario 1, the reinitiation start site can be hundreds of nucleotides downstream from the first termination codon. In the case of GCN4, reinitiation after translation of uORF1 can occur either at uORF2–4 (up to 199 nt downstream from the uORF1 stop codon) or at the main ORF (350 nt downstream from the uORF1 stop codon) depending on eIF2 availability (Mueller and Hinnebusch 1986; Hinnebusch 2005). The preferred distance between the stop and start codon can vary from transcript to transcript for mRNAs that undergo reinitiation after encountering a PTC (scenario 3). For example, the preproinsulin gene favors distances between 79–140 nt (Kozak 1987), while the p53 gene preferentially reinitiates after 39–42 nt (Cohen et al. 2019).

Overlapping ORFs are common in translational control mediated by uORFs, but this does not always reflect a reinitiation mechanism. For example, when a short uORF is out of frame and substantially overlaps the main ORF, the uORF typically acts as a negative regulator of translation. The downstream main ORF is accessed only when the overlapping uORF is skipped through a regulated leaky scanning mechanism. Conversely, if the uORF stop codon is upstream of the main ORF start codon, translation of the main ORF occurs by a reinitiation mechanism. Therefore, while overlapping ORFs are common with uORFs, translation at each of the two overlapping ORFs is mutually exclusive and reinitiation is only involved when the downstream ORF is spatially separated from the ORF where the initial round of translation occurred. For scenario 2, the two translated ORFs often slightly overlap, positioning the ORF2 start codon upstream of the ORF1 stop codon, implying a coordinated mechanism that allows limited retrograde ribosome movement. In scenario 3, reinitiation start sites are always downstream from the PTC and therefore overlapping protein products have not been detected in these cases.

The relative reading frame of the two ORFs also plays a critical role in different ways for each scenario of reinitiation. In scenarios 1 and 2, the start codons for the downstream ORF are usually out of frame compared to the first ORF (Grant et al. 1994; Meyers 2003; Gould and Easton 2007; Luttermann and Meyers 2007; Powell 2010; Gould et al. 2014; Luttermann and Meyers 2014). This apparent preference for different relative reading frames is intriguing as it should not matter in these scenarios, except perhaps to exclude expression by another mechanism such as stop codon readthrough. In contrast, reinitiation events within an ORF after a PTC (scenario 3) show a strong preference for in-frame start codons (Paulsen et al. 2006; Jagannathan and Bradley 2016; Moey et al. 2016). It is unclear how the post-termination ribosome would distinguish in-frame AUGs from those out of frame. The recent structural evidence for a role for eEF2 as a pawl to maintain reading frame (Djumagulov et al. 2021) suggests that a ribosome that has failed to recycle would not be capable of registering the reading frame. While reinitiation could occur at all possible downstream start codons regardless of frame, the apparent preference for in-frame downstream start codons following a PTC is likely due to NMD being triggered by additional PTCs that are encountered during out-of-frame translation events, whereas reinitiating in frame within a main ORF avoids additional PTCs.

Reinitiation start site selection and specificity

As mentioned above, reinitiation start sites can be upstream or downstream and in-frame or out-of-frame compared to the first ORF, depending on the scenario (Fig. 3D). The requirement for an authentic AUG start codon is likely heavily dependent on the composition of the ribosome and/or initiation factors that remain on the mRNA or are rerecruited following termination rather than on the scenario of reinitiation. In general, post-termination 40S complexes have a stricter requirement for an AUG start codon to reinitiate, reminiscent of canonical initiation. It has been proposed (Young et al. 2015; Gunisova et al. 2018) that the site of 80S-mediated reinitiation might depend on the nucleotides in the A-site and how they match with an available tRNA–eEF1A complex. Other studies indicate that ribosome collisions may be involved, especially under conditions of impaired recycling, wherein collisions near the stop codon push unrecycled ribosomes into the 3′-UTR and the 80S reinitiates translation in any frame (Annibaldis et al. 2020; Meydan and Guydosh 2020).

In the many cases studied to date, reinitiation is most efficient with a canonical AUG start codon in a favorable context (Meyers 2003; Luttermann and Meyers 2007; Powell et al. 2008; Neu-Yilik et al. 2011; Moey et al. 2016; Schleich et al. 2017; Bohlen et al. 2020b). Characteristics that determine a favorable context are discussed in great detail in subsequent sections. For TURBS-dependent reinitiation, viral RNAs with mutations to introduce near-cognate start codons (e.g., CUG) undergo reinitiation at a frequency of ∼45% or less compared to wild-type (Luttermann and Meyers 2007). Reinitiation at a main ORF following a round of translation at a uORF is less efficient at near-cognate start codons, which subsequently affects the levels of gene expression (Glass 2017). In the context of reinitiation after a PTC, mutating the downstream AUG can increase sensitivity to NMD (Neu-Yilik et al. 2011), indicating fewer productive downstream translation events. Without the original downstream AUG, reinitiation might occur at out-of-frame AUGs, encountering other PTCs that trigger NMD, or there may not be another AUG that satisfies the distance requirements necessary for efficient reinitiation. Taken together, the overall preference for AUG strongly suggests that reinitiation in these cases is happening in the context of a 40S complex and many of the same underlying mechanisms as canonical initiation are used. In contrast, cellular states with impaired ribosome recycling, such as ABCE1 depletion, that result in post-termination 80S complexes on the mRNA seem to cause reinitiation events with little preference for a start codon or reading frame (Young et al. 2015).

Delivery of Met-tRNA_i for reinitiation

The preference for AUG as the reinitiation start codon for many cases of reinitiation is consistent with an observed general requirement for Met-tRNA_i to deliver the first amino acid during reinitiation (Gunisova et al. 2018). For example, limiting Met-tRNA_i can reduce reinitiation after translation of uORFs, long ORFs and within the main ORF (Dever et al. 1995; Skabkin et al. 2013). This dependence on Met-tRNA_i is maintained even when the reinitiation start codon is not an AUG. As mentioned above, TURBS-dependent reinitiation can occur at near-cognate start codons, but the observation of continued incorporation of amino-terminal methionine residues in the context of these mutations suggests that Met-tRNA_i is used in initiation (Luttermann and Meyers 2007).

Interestingly, although Met-tRNA_i appears essential for certain types of reinitiation, evidence suggests that its delivery mechanism to reinitiating ribosomes can vary (Fig. 3H). As reinitiation follows a translation termination event, the post-terminating complex must acquire this tRNA de novo. While the canonical initiation factor eIF2 can deliver Met-tRNA_i in the context of reinitiation, other methods of Met-tRNA_i delivery can also be used. For example, TURBS-dependent reinitiation was observed in vitro when Met-tRNA_i was supplied in the absence of eIF2 (Zinoviev et al. 2015).

Factors with roles in recycling, such as DENR/MCT-1 or eIF2D may have the ability to deliver Met-tRNA_i under certain conditions (Skabkin et al. 2010; Zinoviev et al. 2015; Ahmed et al. 2018) for both scenarios 1 and 2. The binding site of eIF2D or DENR/MCT-1 on the ribosome overlaps with that of eIF1, eIF2β, eIF2γ, eIF5B, and parts of eIF3 (Lomakin et al. 2017; Weisser et al. 2017). This raises the possibility that eIF2D or DENR/MCT-1 delivery of Met-tRNA_i for reinitiation could bypass the requirement for those factors although, given the dynamic nature of these processes and complexes, canonical initiation factors could have temporally distinct interactions to achieve this type of reinitiation. How different mechanisms of Met-tRNA_i delivery might operate in the context of translational repression in response to viral infection remains to be established. Interestingly, some viral IRES RNAs have been shown to use a factor-free mechanism or alternative factors to recruit Met-tRNA_i during certain cellular conditions, suggesting parallels between these mechanisms (Jaafar et al. 2016; Jaafar and Kieft 2019).

Initiation factor requirements for reinitiation

Dependence on canonical initiation factors varies substantially between different examples of reinitiation (Fig. 3G). Several initiation factors such as eIF1, eIF1A, eIF3, and eIF4F have been shown to be involved in certain types of reinitiation (Skabkin et al. 2013). eIF4G was recently shown to be retained on ribosomes elongating and terminating on short uORFs and to enhance reinitiation efficiency in yeast (Mohammad et al. 2021). The requirement for eIF4F only in certain scenarios of reinitiation may be due in part to its role in enforcing 5′ to 3′ directionality of ribosome motion (referred to as either scanning or traversing) following termination (Skabkin et al. 2013), which would only be necessary when the reinitiation start site is downstream from the ORF1 stop site.

eIF3, which is the largest of the initiation factors comprising a multisubunit assembly, appears to have a key role in many forms of reinitiation (Cate 2017). Interestingly, eIF3, or possibly only some of its components, can remain bound to the ribosome after initiation in yeast, plants and mammals, and promote reinitiation events (Kim et al. 2007; Roy et al. 2010; Mohammad et al. 2017; Valášek et al. 2017; Bohlen et al. 2020a; Lin et al. 2020; Poyry et al. 2020; Wagner et al. 2020). For example, one study found eIF3 remains associated with 80S ribosomes during elongation through the first 60 codons (Lin et al. 2020) while another found the half-length of eIF3 association with 80S ribosomes to be 12 codons (Bohlen et al. 2020a). Furthermore, eIF3 retention on the 80S can be regulated, such as through O-linked N-acetylglucosamine modification to eIF3a during nutrient stress leading to altered reinitiation after uORF translation (Shu et al. 2022). A key determinant as to which initiation factors remain bound to the ribosome during reinitiation is the length of the first ORF. Reinitiation efficiency decreases as the length of the uORF increases (Luukkonen et al. 1995; Kozak 2001), supporting the notion that longer elongation increases the likelihood that initiation factors have departed, which then disfavors subsequent reinitiation events.

Using the different classes of IRESs, which require distinct sets of eIFs to initiate translation (Kieft 2008; Mailliot and Martin 2018; Martinez-Salas et al. 2018), it has been demonstrated that downstream reinitiation at the main ORF is significantly less efficient when eIFs 1, 1A, 4G, and 4A are not involved in the first initiation event at a uORF (Poyry et al. 2004; Russell et al. 2023). Reinitiation levels increased with a longer intercistronic distance for the uORFs initiated with IRESs that do not use these initiation factors (Russell et al. 2023), suggesting that more time for these factors to be rerecruited following termination increases the likelihood for reinitiation. The role of eIF3 in these studies is unclear, as certain IRESs requiring that initiation factor reposition it (Hashem et al. 2013) in a way that could be unproductive for a subsequent reinitiation event. While eIF3 was not found to be necessary for TURBS-dependent reinitiation (scenario 2) in vitro (Zinoviev et al. 2015), it might be part of the reinitiation complex in vivo, as it was demonstrated to crosslink with the viral TURBS RNA in lysate and the addition of excess eIF3 led to higher reinitiation levels (Pöyry et al. 2007). Overall, analysis of which factors are bound during various types of reinitiation scenarios and if they must be rerecruited following the first initiation event is crucial to uncover how and why reinitiation occurs. Studies that examine the occupancy of partial or whole eIF complexes on ribosomes as a function of their location on mRNAs or different forms of reinitiation are necessary to better understand each of their individual roles.

Effects of the amino acid sequence and nascent peptide on reinitiation

During elongation, interactions between the growing polypeptide chain and the ribosome exit tunnel are associated with events such as frameshifting (Yordanova et al. 2015) and peptide bond skipping (de Felipe et al. 2003; Yan et al. 2010), thus it seems that similar mechanisms may affect reinitiation (Fig. 3B). The encoded amino acid sequence or even specific codons upstream of the stop codon can affect reinitiation, likely by modulating the last few rounds of elongation or the termination process.

In the case of scenario 1, uORF encoded peptides can affect translation to alter expression levels of the main ORF. For example, the CPA1 (S. cerevisiae) and arg-2 (N. crassa) mRNAs contain a single uORF encoding an arginine attenuator peptide (Gaba et al. 2001), which stalls ribosomes and blocks access to the main ORF under high arginine conditions. For GCN4, the sequence of the termination region around uORF1, including that of the penultimate codon, affects reinitiation efficiency in that specific serine and arginine codons are permissive for reinitiation while others are not, and a penultimate proline is inhibitory (Grant and Hinnebusch 1994; Grant et al. 1994; Gunišová et al. 2016). These results are particularly intriguing in the context of reports of different penultimate codons showing differential dependence on recycling factors (Bohlen et al. 2020b; Young et al. 2021), suggesting a plausible link between the removal of specific tRNAs from the P-site during termination and the propensity for a downstream reinitiation event.

In scenario 2, reinitiation at downstream ORFs in some human mRNAs may depend on the sequence of the nascent peptide. Specifically, stretches of aspartate residues near the carboxyl terminus of the upstream encoded peptide of multiple genes including CASQ1 and CASQ2 were responsible for coupled translation. The mechanism for the influence of these nascent peptides is unclear but was demonstrated via mutational analysis that the encoded amino acid, but not particular codons, causes changes to reinitiation efficiency (Gould et al. 2014). For some cases of downstream reinitiation on viral mRNAs, it has been conclusively shown that the RNA sequence and not the encoded amino acid sequence is responsible for promoting reinitiation (Luttermann and Meyers 2007, 2014; Firth and Brierley 2012), but other cases have yet to be tested in enough detail to rule out a role for the encoded peptide sequence.

The role of the nascent peptide in scenario 3 is yet to be determined. The random locations of nonsense mutations suggest any effect would be highly variable and unpredictable but could explain discrepancies as to which PTC-containing mRNAs successfully evade NMD through reinitiation.

Overall, it appears that the recently decoded nascent peptide sequence can, but does not always, affect reinitiation. For some of the above cases where a particular amino acid has an effect, it has not been shown whether the process of decoding those amino acids or the physical interactions between the amino acids and the ribosome exit tunnel cause the observed effects on reinitiation. The interplay of codon-specific effects on elongation kinetics and the efficiency of subsequent reinitiation in each scenario requires further investigation. In the absence of a comprehensive list of sequences that promote reinitiation, there is currently no way to predict the effect of a given sequence or engineer a sequence to increase or decrease reinitiation, especially given the added layer of complexity where the mRNA sequence also affects reinitiation, as discussed in the next section. A combination of detailed biochemical, genetic, and structural methods will be necessary to investigate the relative contributions of all the possible effects of nascent peptides, encoded amino acid sequences, and specific codon usage in each reinitiation scenario.

Effects of mRNA sequence and structure on reinitiation

mRNAs are not passive players during reinitiation, as the RNA sequence and structures surrounding the stop codon and reinitiation start codon can influence reinitiation in all three scenarios (Fig. 3F). However, the mechanisms by which these features elicit their effects remain unclear. One example of a primary sequence-based effect is that of the Kozak consensus sequence. This sequence promotes start codon recognition in canonical translation initiation and also does so at reinitiation start sites in scenarios 1 and 3 (Kozak 1986). Mutations that introduce poor Kozak consensus sequence surrounding the start codon for the downstream ORF in human norovirus did not alter reinitiation efficiency, indicating that TURBS RNA-mediated reinitiation does not rely on Kozak context (Luttermann and Meyers 2014). It is currently unclear if reinitiation in the 3′-UTR of cellular mRNAs benefits from favorable Kozak context.

Secondary and tertiary structures within the mRNA can also regulate reinitiation. As discussed in the section on scenario 2, reinitiation events in some viral mRNAs and the SART1 retrotransposon are highly dependent on specific RNA structures (Kojima et al. 2005; Luttermann and Meyers 2007; Li et al. 2015). In scenario 1, reinitiation promoting elements (RPEs) are RNA structural elements that determine if traversing occurs following termination on an uORF (Grant et al. 1995; Munzarová et al. 2011; Gunišová et al. 2016). In the case of GCN4, mRNA sequence and likely structural elements play a significant role in reinitiation efficiency (Grant and Hinnebusch 1994) and certain regions are proposed to interact with eIF3a to promote traversing after termination at uORF1 followed by reinitiation downstream (Szamecz et al. 2008). Given that strong hairpin structures are known to affect scanning and start codon selection, RNA structures might similarly affect traversing ribosomes looking to reinitiate (Babendure et al. 2006; Wang et al. 2022). Nonetheless, little is known about potential contributions of surrounding mRNA sequence or local secondary structure to reinitiation associated with a PTC.

Effects of post-transcriptional modifications on reinitiation

The identification of diverse post-transcriptional modifications throughout mRNAs has exploded in recent years, begging the question of their role in regulating reinitiation frequency (Fig. 3C). In fact, Zhou and colleagues used the ATF4 system to show that N⁶-Methyladenosine (m⁶A) modification at certain positions fine tunes uORF start codon usage, perhaps by slowing the ribosome (Zhou et al. 2018). This work emphasizes how post-transcriptional RNA modifications can affect reinitiation, but the mechanism is unclear. It is possible m⁶A modifications slow ribosomes through direct recognition by eIF3a or other “readers” (Shi et al. 2019), or by altering local mRNA structure. By extension, other ubiquitous modifications like 5-methylcytosine and pseudouridine may also alter reinitiation; this remains to be tested. Likewise, it will be important to understand the magnitude and specificity of this effect in cells: how many copies of a given mRNA harbor the modification, and at what positions it affects reinitiation. Global mapping of modifications, sequencing of individual RNAs including detection of modifications by nanopore sequencing, and studies of translation on single molecules with known modifications will ultimately be necessary to address these questions.

BIOLOGICAL IMPLICATIONS AND IMPACT OF REINITIATION

Reinitiation increases a genome's coding capacity and creates additional modes of translation regulation. What is less clear is how reinitiation might more broadly affect translation and which peptide products generated by a reinitiation event are functional. Reinitiation might affect the overall translational state of the mRNA and alter relative expression levels, especially during stress as was observed for IRES-driven downstream translation (Koch et al. 2020). In the following section, we describe how the three scenarios of reinitiation lead to different types of peptide products with different implications for function and to different questions to frame future exploration, as discussed below (Fig. 4).

FIGURE 4.

Functional consequences of translation reinitiation. The three reinitiation scenarios have different outcomes in terms of the functions of encoded proteins. Short uORFs in scenario 1 are considered primarily regulatory, with the act of their translation and not the peptide sequence itself dictating expression of the main ORF, which has critical cellular functions. Scenario 2 produces two functional proteins at a precise stoichiometric ratio, which so far has mainly been observed in viruses. It is unclear what, if any, roles the peptide sequences encoded by downstream ORFs play in eukaryotic genes. Scenario 3 can have different consequences—neutral, beneficial, or detrimental—depending on the position of the PTC and the nature of the two protein products that would be translated as a single protein in the absence of a nonsense mutation. Green: start codon; red: stop codon; black open boxes (scenario 2): sequences/structure motifs for termination/reinitiation coupling.

Functional consequences of multiple protein products from reinitiation events

Multiple peptide products originating from translation of a single transcript is a hallmark of reinitiation. In this section, we discuss the proteins produced from mRNAs that undergo reinitiation, whether the products from the ORF downstream or upstream of the “main” ORF have known functional roles, and how to study these alternative peptide products.

In Scenario 1, translation of a short uORF regulates translation of the downstream main ORF that encodes a functional and critical protein (Fig. 4; Mueller and Hinnebusch 1986; Gaba et al. 2001; Hinnebusch 2005). In this case, ribosomes can access the “canonical” ORF either by reinitiation or leaky scanning. The peptides produced from these uORFs can act during translation (Causier et al. 2022), such as to help stall termination (Gaba et al. 2001; Jousse et al. 2001). They may also have functions outside translation regulation (Samandi et al. 2017; Chen et al. 2020). For example, uORF-encoded peptides have been characterized that regulate transcription (Koh et al. 2021), metabolism (Laing et al. 2015; van der Horst et al. 2019; Huang et al. 2021), signaling (Jayaram et al. 2021), and DNA repair (Slavoff et al. 2014).

In the case of Scenario 2, the well-characterized reinitiation events at downstream ORFs in viral mRNAs produce two functional and essential proteins. In this case, the peptides are both clearly functional, and reinitiation serves to control the precise stoichiometry of the two protein products (Fig. 4). For some cellular mRNAs, peptide products resulting from reinitiation of overlapping downstream ORFs have been detected experimentally (Gould et al. 2014), although their biological role, if any, are yet to be determined. As mentioned earlier, translation in the “3′-UTR” of cellular mRNAs is more common than previously appreciated, especially under certain cellular conditions (Bazzini et al. 2014; Guydosh and Green 2014; Ji et al. 2015). In some cases, downstream ORFs have been shown to encode a functional peptide (Schwab et al. 2003), some of which up-regulate translation of the main ORF (Wu et al. 2020), although reinitiation has not been shown to be the mechanism through which these downstream ORFs are translated. Reinitiation into the 3′-UTR caused by faulty or absent recycling factors would yield peptides randomly from three possible reading frames (Young et al. 2015). It is tempting to speculate that such downstream-encoded peptides could have regulatory functions, such as rescuing recycling defects by setting up feedback loops to globally affect other initiation, termination, or recycling events. As in viruses, eukaryotic cells could also use programmed reinitiation downstream from a main ORF to couple the expression of two proteins at a defined stoichiometry, but this remains speculative.

In Scenario 3, reinitiation due to a mutation-generated PTC within an ORF creates additional considerations because reinitiation is on an aberrant mRNA (Fig. 4). Since 5′-proximal PTCs appear to be preferred substrates for reinitiation, the resulting protein may only lack a small portion of the amino terminus (Zhang and Maquat 1997; Neu-Yilik et al. 2011; Jagannathan and Bradley 2016; Lindeboom et al. 2016). These truncated proteins could retain function, perhaps partially overcoming an otherwise extremely harmful mutation (Paulsen et al. 2006), and PTCs in this region could be less prone to negative selection (Howard et al. 2004; Stump et al. 2013). However, when the amino-terminal region provides critical subcellular localization signals or other functions, the nearly full-length protein could be nonfunctional or toxic. If reinitiation occurs further along in the ORF, two truncated polypeptides may result in place of one full-length functional protein. It is theoretically possible that the two halves of a protein could stably interact to function like the full-length protein in whole or in part (He et al. 2013). While a compelling idea, there is no evidence for this effect and most PTC locations are likely not conducive to accomplishing a split yet functional bimolecular protein.

A key challenge in fully exploring the functional consequence of the protein products derived from mRNAs that undergo reinitiation events is the lack of suitable quantitative methods. Larger protein products, such as those often resulting from scenario 2, can be isolated and characterized easily. However, the upstream protein products from scenario 1 and 3 can be much smaller. Previous studies have incorporated epitope tags to aid in the isolation and detection of such proteins (Cohen et al. 2019), but changes to the length of these relatively small ORFs as well as the RNA sequence and encoded peptide products could significantly perturb downstream reinitiation. Detecting extremely short peptides that could be rare and/or short-lived is a challenge that is particularly relevant when differentiating between true reinitiation events and leaky scanning that bypasses upstream start sites (Howard et al. 2004; Poyry et al. 2004; Neu-Yilik et al. 2011; Cohen et al. 2019). However, the underlying mechanisms of reinitiation and leaky scanning are distinct (Fig. 1), and could influence how, for example, a clinically relevant nonsense mutation is explored in terms of therapeutic modulation.

Leaky scanning can alleviate deleterious effects of a nonsense mutation when the initiation happens at a downstream, in-frame start codon that is past the location of the PTC. This creates the same amino terminally truncated protein product (as depicted in Fig. 1F) as reinitiation at that same downstream start codon. In such cases, methods involving a thorough and systematic mutational analysis and careful quantitation can help determine the relative contributions of leaky scanning and reinitiation. Other methods such as cutting-edge mass spectrometry and/or single molecule imaging approaches to detect short peptides are necessary to truly establish whether reinitiation or leaky scanning results in the translation of the downstream ORF. Forward and reverse genetics coupled with phenotyping and eventually more directed functional assays may be needed to tease out any biological roles for these peptides. The likelihood that some peptides may have subtle or unexpected effects, may be transient in nature, or may operate both within and external to reinitiation, render such experiments potentially challenging.

Evolution of reinitiation

Given the challenges in experimentally determining the function of reinitiation products, their evolutionary history may offer important clues for their importance as a protein product or as a regulatory mechanism (Dever et al. 2020; Takahashi et al. 2020b; Zhang et al. 2021).

In higher eukaryotes, as many as 60% of all mRNAs have identified uORFs (Lu et al. 2004; Gunisova et al. 2018; Takahashi et al. 2020a), and more have putative downstream open reading frames accessed through programmed reinitiation. Likewise, ribosome occupancy in regions of mRNAs annotated as noncoding (e.g., 5′- and 3′-UTRs) are more common than previously appreciated (Guydosh and Green 2014; Ji et al. 2015). New uORFs could readily emerge from a point mutation that creates a new start site within a 5′ UTR (Fig. 5A). Most new start codons within 5′ UTRs are likely prone to negative selection as they inhibit the use of the downstream main ORF (Zysow et al. 1995; Labrouche-Colomer et al. 2020; Coursimault et al. 2022). However, a new uORF could also benefit by providing a new layer of regulation. In addition to random mutations, transposable elements can introduce new uORFs that are then subjected to natural selection (Zhang et al. 2019); these are proposed to be the source of 10% of human uORFs (Kitano et al. 2018). Whether introduced through mutation or transposon, new uORFs can become fixed and essential for cellular homeostasis, as evidenced by the existence of diseases that originate from deleterious mutations in the uORF of oncogenes and proto-oncogenes (McGillivray et al. 2018; Schulz et al. 2018).

FIGURE 5.

Evolution and therapeutic applications of translation reinitiation. (A) Sequences or structures can evolve in the viral genome to promote expression of different or new open reading frames (scenario 2). uORFs can be created via mutations or transposon insertion (scenario 1). The acquisition of a downstream AUG can allow an otherwise detrimental PTC to be tolerated (scenario 3). (B) Small molecule or RNA-based therapeutics targeting RNA sequences, RNA structures, or protein factors that impact reinitiation could be developed to inhibit viral proliferation or alleviation of a genetic disease due to acquisition of an upstream AUG or a PTC. Green: start codon; red: stop codon; black boxes: sequences/structure motifs for termination/reinitiation coupling.

Reinitiation signals at the end of ORFs, such as the TURBS found in caliciviruses also display conservation crafted by evolution (Powell 2010). It is particularly intriguing that the TURBS RNA element found in Caliciviridae (positive-send RNA viruses) also appears in the distantly related influenza B virus (segmented negative-sense RNA virus) but not influenza A virus. While both influenza A and B viruses encode two proteins in the segment 7 mRNA, influenza A virus expresses the downstream ORF through an alternative splicing event (Tsai et al. 2013) and influenza B virus uses reinitiation. It is therefore tempting to speculate that the TURBS RNA motif in influenza B virus arose independently of those in caliciviruses and fortuitously contains the same ribosome-binding nucleotides flanked by base pairing upstream of the stop codon.

Interestingly, the mechanism of TURBS RNA binding to the ribosome shares features with that of the Hepatitis C virus IRES, in that both bind the same stem–loop in the 18S rRNA, suggesting a link between RNA structure-mediated reinitiation and internal initiation. Whether one evolved from the other, or this shared mechanism of ribosome binding arose separately, is not clear. However, this suggests that direct interactions with the translational machinery are convenient ways for new modes of translational control to evolve in viruses, and perhaps in cellular mRNAs. Reinitiation occurs in viruses that infect plant, animal, and fungal hosts, supporting the notion that reinitiation is quite widespread throughout biology (Park et al. 2001; Guo et al. 2009; Li et al. 2015). More viral reinitiation signals likely remain to be discovered, especially considering small downstream ORFs might easily go unannotated. Reinitiation provides a mechanism in viruses as well as eukaryotes for new coding sequences to be expressed, creating new peptide products that are subjected to evolutionary pressures to determine whether these new proteins persist in the population (Fig. 5A).

Finally, nonsense mutations can theoretically arise in any mRNA and, in the right context, promote reinitiation. Random mutations that introduce a new AUG downstream from a PTC could provide benefits such as avoiding NMD and would be evolutionarily favorable in those contexts (Fig. 5A). The presence of such downstream ORFs might allow the PTC to be tolerated and maintained in the population despite its otherwise deleterious effects. Evidence for this phenomenon is found in the observation that PTCs are more common in the first 10% of the coding region, especially when there is an in-frame downstream start codon (Jagannathan and Bradley 2016). This would suggest that reinitiation downstream from PTCs is pervasive and influences the mutational landscape of a gene. However, we do not yet know what percentage of mRNAs in a transcriptome support reinitiation in some capacity and at what frequency.

In contrast to eukaryotic systems, polycistronic transcripts are more of the rule than the exception in bacteria, which raises the intriguing question as to the origin of reinitiation mechanisms and whether they have simply persisted through evolution or reemerged as a translation strategy. It is unclear whether this mechanism is more frequently exploited by viruses, or if viral examples are more easily detected. There may be many other viral and cellular RNA structures involved in reinitiation that we have not yet identified because their sequence and structure are sufficiently divergent as to evade detection by computational homology searches. Finding them may only be accomplished through functional assays. In summary, the total fraction of translation activity at any moment due to reinitiation is probably highly variable and small, yet critical for proper cellular function. The global and specific fluctuations in reinitiation frequency, the variables that affect them, and differences in how reinitiation is accomplished and regulated between species are key areas to be explored by further studies.

Clinical implications of understanding reinitiation

The presence of specific reinitiation signals in human-infecting viruses has obvious clinical relevance, as the use of reinitiation in these viruses is critical to create the proper stoichiometry of certain viral proteins (Luttermann and Meyers 2007, 2014). Similar to programmed ribosomal frameshifting, reinitiation by TURBS elements depends on a specific folded RNA element. Frameshifting elements have been the focus of recent studies to develop small-molecule inhibitors that target these RNA structures for therapeutic application, with some promising success (Haniff et al. 2020). Likewise, TURBS or other viral reinitiating signals could perhaps be the target of therapies (Fig. 5B). If elements analogous to TURBS exist in cellular mRNAs, their proper function could be key for regulating gene expression, with implications for the health of the cell.

Reinitiation downstream from a PTC can sometimes alleviate disease by resulting in a partially functional protein (Howard et al. 2004; Paulsen et al. 2006), but can also cause disease by creating a dominant negative protein product (Stump et al. 2012, 2013). When reinitiation generates a partially functional protein, inhibiting NMD via specific small molecule inhibitors or by inducing PTC readthrough (Hamid et al. 2010) could enhance the production of this beneficial product (Fig. 5B). When a PTC-containing mRNA produces a dominant negative gene product via reinitiation, destruction of the mRNA could provide therapeutic benefit. Better understanding of factors that determine the frequency and efficiency with which reinitiation occurs downstream from a PTC is necessary to accurately interpret the functional outcome of a nonsense mutation.

Being able to induce, enhance, or suppress reinitiation of all types could have significant diagnostic and therapeutic implications for genetic and viral diseases if it can be achieved (Fig. 5B). Much remains to be learned about how global translational reprogramming in different cellular states (oxidative stress, viral infection, and specific disease contexts) and mRNA localization influence reinitiation rates, especially considering reinitiation can vary in a tissue-dependent manner (Van De Velde et al. 2021). Such efforts to comprehensively understand reinitiation events will require a deeper understanding of the basic underlying molecular mechanisms than we currently possess.

CONCLUDING REMARKS

Translation reinitiation is an important part of the repertoire of strategies used to regulate translation in eukaryotic cells. Reinitiation occurs in different contexts and for different purposes and is used both to regulate cellular gene expression and by viruses as part of their infection strategy. Although different mechanisms govern each scenario, some common themes underlie all forms of reinitiation, and this can help focus future studies addressing key unknowns. Critical areas of inquiry include cataloging the full global extent of reinitiation used in healthy and diseased cells, new modes or locations of reinitiation in mRNAs, and the details of how different variables affect the amount or efficiency of specific reinitiation events. Likewise, understanding how reinitiation affects the cell's biology or allows it to evade deleterious mutations will be key to the development of clinical approaches that target both mRNA and viral RNA-based reinitiation. Over the next decade, we expect to see the tools of transcriptome-wide analysis, ribosome profiling, high-throughput and high-sensitivity mass spectrometry, forward and reverse genetics, and others contribute to solving these important mysteries.