RNA tales – how embryos read and discard messages from mom

Early development of all metazoans is characterized by massive reprogramming to enable a smooth cellular transition from the fertilized egg to transient totipotent and pluripotent embryonic states, and towards re-establishment of differentiation (Giraldez, 2010). The newly created zygote inherits parental genomes that are in a transcriptionally quiescent state. However, the loading of maternal RNA and protein from the mature oocyte into the embryo compensates for the absence of RNA supply. Therefore, maternal stores power the initial cell cycles and support embryogenesis until several thousand cells are formed, depending on the species. The transition to the zygotic developmental program requires regulated degradation of the old maternal transcriptome (maternal mRNA clearance) and its replacement by factors produced de novo through transcriptional activation of the zygotic genome (zygotic genome activation, ZGA) (Tadros and Lipshitz, 2009). This aspect of early development comprises the maternal-to-zygotic transition (MZT; Fig. 1A). The mechanisms that underlie ZGA have been reviewed elsewhere (Langley et al., 2014; Lee et al., 2014; Pálfy et al., 2017). Recently, the establishment and adaptation of novel high-throughput methods for use in embryos have facilitated progress in our understanding of post-transcriptional regulation of maternal mRNAs at the molecular level. Here, we will review new insights into the control of maternal mRNA stabilities during MZT. To that end, we will focus on how microRNAs (miRNAs), RNA-binding proteins (RBPs), codon optimality and RNA modifications contribute to this process.

Given that most mRNAs have half-lives of the order of hours (Schwanhäusser et al., 2011), one might assume that maternal mRNAs simply undergo decay without any special regulation by the embryo and are thereby gradually replaced by zygotic mRNAs. However, in fact, the stability of maternal mRNAs is tightly controlled, such that particular pools of messages are cleared through different mechanisms at different time points during embryogenesis. Therefore, maternal mRNA clearance represents a major aspect of post-transcriptional gene regulation and is a hallmark of MZT (Tadros and Lipshitz, 2009). The extent of maternal mRNA clearance varies across species: 30% and 60% of maternally provided mRNAs are degraded during the course of MZT in the nematode Caenorhabditis elegans and the fruit fly Drosophila melanogaster, respectively (Baugh et al., 2003; Thomsen et al., 2010).

Why is the regulation of maternal mRNA clearance important? The ‘permissive model’ posits that the elimination of ubiquitously expressed maternal transcripts enables spatially and temporally restricted expression of their zygotic counterparts (Tadros and Lipshitz, 2009). This is supported by observations that zygotically expressed mRNAs in Drosophila have more highly patterned expression in comparison to more ubiquitously expressed maternal counterparts that are subjected to decay during MZT (De Renzis et al., 2007). Alternatively, the ‘instructive model’ holds that the selective elimination of maternal mRNAs restricts their functions; for example, prolonged stabilities of maternal mRNAs throughout embryogenesis could impair cell cycle regulation during MZT or potentially be deleterious for later phases of development (Tadros et al., 2007; Takacs and Giraldez, 2016). These models are not mutually exclusive. It is currently unclear which one of these two models dominates or whether they are assisted by additional processes of maternal mRNA clearance. Instead, the focus of many recent studies has been the timing and mechanisms of maternal mRNA clearance during MZT. Below, we will summarize the state of current results and discuss future perspectives of the field.

The 7-methylguanosine (m⁷G) cap and the poly(A) tail (a stretch of variable numbers of non-templated adenosine residues) represent two major features of mRNA that promote translation and transcript stabilities. Thus, these modifications on the ends of mRNA serve as ‘bodyguards’ that protect the message from 5′-3′ and 3′-5′ exonucleases (Wilusz et al., 2001). The m⁷G cap and poly(A) tail are added in the nucleus to 5′- and 3′-ends of transcripts, respectively, by capping enzymes and canonical poly(A) polymerases (Shatkin and Manley, 2000). Conventional mRNA decay occurs in the cytoplasm and usually starts with the shortening of poly(A) tails (deadenylation), the rate-limiting step in the clearance of transcripts that precedes removal of the 5′-cap structure from mRNA (decapping), and degradation by, for example, XRN1 and the exosome complex (Decker and Parker, 1993; Couttet et al., 1997; Li and Kiledjian, 2010; Lykke-Andersen et al., 2011). In some instances, mRNAs can be targeted for degradation through decapping that occurs independently of deadenylation (Badis et al., 2004). Because shortening of the poly(A) tail often serves as a trigger for transcript turnover, deadenylation is highly controlled – either positively or negatively – by various regulatory factors, for example, RBPs and the sequences they bind (Matoulkova et al., 2012) (Fig. 1B).

In embryos, the poly(A) tail length undergoes dynamic changes to regulate both mRNA translation and stability (Weil, 2015). Poly(A) tail length is largely determined by the balance between deadenylation and cytoplasmic polyadenylation, the post-transcriptional elongation of poly(A) tails by non-canonical poly(A) polymerases (Zhang et al., 2010). Although it is evident that both processes occur in embryos of various species, it is presumed that regulated cytoplasmic polyadenylation represents the predominant force with a wide impact on poly(A) tail lengths in Drosophila and vertebrate embryos, particularly after fertilization (Subtelny et al., 2014; Eichhorn et al., 2016). Interestingly, shortening of the poly(A) tail in embryos has two different outcomes for maternal mRNA fate, depending on when during early embryogenesis deadenylation occurs. Although cytoplasmic polyadenylation predominates after fertilization, active deadenylation does not result in the destabilization of mRNA. Instead, transcripts with shortened tails generally exhibit high stabilities (Audic et al., 1997; Voeltz and Steitz, 1998; Graindorge et al., 2006). This is particularly evident in Xenopus embryos, where fertilization-triggered deadenylation of transcripts that contain destabilizing AU-rich elements (AREs) in 3′ untranslated regions (UTRs) results in a delayed mRNA clearance hours after the tails have been shortened. Furthermore, in the zebrafish Danio rerio, both endogenous and exogenously introduced miRNAs cause deadenylation, with predominant effects on translation and on mRNA decay several hours later (Bazzini et al., 2012; Subtelny et al., 2014). Because transcriptionally inactive embryos lack the capacity to replace maternal products before ZGA, this dynamic regulation of the poly(A) tail length is thought to fine-tune the translational output of an mRNA without compromising stability. At ZGA, embryos add new pools of mRNA to the maternal mRNA pool, and poly(A) tail shortening becomes coupled to transcript destabilization. It has been shown that the CCR4-NOT complex represents a major deadenylase that triggers degradation of most maternal mRNAs during MZT (Mishima and Tomari, 2016). Without CCR4-NOT, maternal mRNAs become stabilized, confirming that maternal mRNA decay strongly depends on deadenylation. What underlies uncoupling between deadenylation and degradation at earlier steps of development and how stability is conferred to transcripts with short poly(A) tails remains to be determined. Moreover, how mRNA deadenylation and decay become coupled at later steps of development also remains elusive.

Global measurements of mRNA levels in different species have revealed different cohorts of maternal mRNAs with respect to the onset of their elimination during MZT (Baugh et al., 2003; Tadros et al., 2003, 2007; Hamatani et al., 2004; Alizadeh et al., 2005; Mathavan et al., 2005; De Renzis et al., 2007; Ferg et al., 2007; Thomsen et al., 2010; Aanes et al., 2011). In Drosophila, a handful of studies have suggested that maternal mRNA clearance begins shortly after fertilization, and is almost completed by 3 h after fertilization (Bashirullah et al., 1999; Tadros et al., 2007; Thomsen et al., 2010), whereas a recent study has reported the first evidence of maternal mRNA decay only at 2–3 h post-fertilization (Eichhorn et al., 2016). Similarly, reports of the onset of maternal mRNA clearance in zebrafish embryos differ: whereas some studies suggest that maternal mRNA elimination occurs in different waves following fertilization (Mathavan et al., 2005; Aanes et al., 2011), others propose that the majority of maternal mRNAs remain stable for several hours (Vesterlund et al., 2011). The latter findings agree with previous reports of delayed degradation of mRNAs with AREs in Xenopus, and miRNA targets in zebrafish embryos (Voeltz and Steitz, 1998; Bazzini et al., 2012; Subtelny et al., 2014). The existence of an early wave of maternal mRNA degradation has been validated for a number of individual endogenous genes (Bashirullah et al., 1999), yet it remains unclear when the clearance of bulk maternal mRNA begins. The current consensus is that distinct pools of maternal mRNA decay with different kinetics (Fig. 2A). However, it will be important to determine the precise timing of maternal mRNA destabilization to decipher the underlying mechanisms that mediate mRNA decay during MZT. In the future, this could be accomplished by performing total RNA-seq experiments on samples from tightly staged embryos. Note that some discrepancies that have been found might be due to the differences in RNA-seq and analysis methodologies that have been used to quantify changes in mRNA levels during MZT. When studying mRNA decay rates at embryonic stages after ZGA, experimental care must be taken to clearly separate the decay of maternally provided transcripts from that of their zygotically expressed counterparts with identical transcript architecture. Several methods have successfully been employed to differentiate between maternal and zygotic transcripts, such as tracking accumulation of RNA that contains paternal single-nucleotide polymorphisms (SNPs) (Harvey et al., 2013) or intronic sequences from unprocessed zygotic transcripts (Lee et al., 2013), or through labeling of newly synthesized transcripts with 4-thioUTP (4-sUTP) (Heyn et al., 2014).

Two mechanisms of maternal mRNA clearance during MZT can be identified based on the requirement of the transcriptional activity of the embryo for mRNA degradation: ZGA-independent mRNA decay occurs independently of zygotic transcription and relies strictly on maternally provided instructions for efficient mRNA destabilization, whereas ZGA-dependent mRNA decay requires factors of zygotic origin and thereby strongly depends on transcriptional activity of the embryo (Fig. 2B). In Drosophila, activation of unfertilized eggs that do not show signs of transcriptional activity and fertilized embryos that carry specific chromosomal arm deletions from which zygotic products are lacking have been useful systems for investigating the independent contributions of these two different modes of mRNA clearance (Bashirullah et al., 1999; Tadros et al., 2003; De Renzis et al., 2007). In zebrafish, total RNA-seq has been used to profile the response of destabilized mRNAs to transcription inhibition mediated by α-amanitin injection into one-cell-stage embryos (Bazzini et al., 2016; Mishima and Tomari, 2016). Transcripts that continue to decay when transcription is blocked are said to undergo ZGA-independent mRNA decay, whereas those whose clearance is prevented by α-amanitin follow ZGA-dependent decay. However, this separation of transcript destabilization into two decay modes is likely a simplification, because many maternal transcripts may be undergoing degradation through the combined action of ZGA-dependent and ZGA-independent decay pathways.

The best-known triggers of the ZGA-dependent maternal mRNA clearance program are zygotically expressed miRNAs, which have been identified in fly (miR-309), frog (miR-427) and zebrafish (miR-430) embryos (Giraldez et al., 2006; Bushati et al., 2008; Lund et al., 2009) (Fig. 2B,C). Zebrafish miR-430 is expressed very early in ZGA and causes deadenylation-dependent translational repression of a few hundred maternal transcripts before triggering delayed mRNA degradation 2 h later (Bazzini et al., 2012; Heyn et al., 2014). Similarly, miR-309 is responsible for degradation of a comparable number of maternal messages in Drosophila embryos (Bushati et al., 2008). In Xenopus, mRNA reporters have been used to assess the role of the miR-430 ortholog in maternal mRNA clearance (Lund et al., 2009). Based on these studies, miR-427 inevitably acts as a zygotic factor for maternal mRNA clearance; however, a transcriptome-wide analysis of miR-427-mediated decay is still lacking. Although miR-309 deletion causes stabilization of hundreds of maternal transcripts, embryos that lack this cluster of miRNAs are viable and reach adulthood (Bushati et al., 2008). By contrast, miR-430 mutants exhibit more drastic phenotypes in the adulthood manifested through severe changes in brain morphology (Giraldez et al., 2005). Although it is evident that these two miRNAs contribute to maternal mRNA destabilization during MZT, the significance of maternal mRNA clearance through this pathway for development progression is unclear, as the effects of their absence on development are either not seen or displayed much later in development, far beyond MZT. Interestingly, the production of zygotic miRNAs in mouse and C. elegans embryos does not have a major impact on maternal mRNA destabilization. Instead, endogenous small interfering RNAs (endo-siRNAs) appear to be required for maternal mRNA clearance (Svoboda and Flemr, 2010; Stoeckius et al., 2014).

RBPs are components of post-transcriptional regulatory networks that bind to the cis-acting elements in coding and 3′UTR regions of mRNA in a sequence-specific manner to regulate mRNA stability and translation (Glisovic et al., 2008) (Fig. 2C). In many embryos, maternally provided RBPs appear to be potent regulators of maternal mRNA fates. For example, Drosophila Smaug (Smg), Brain Tumor (Brat) and Pumilio (Pum) proteins associate with various numbers of translationally repressed and destabilized maternal mRNAs that are cleared by CCR4-NOT during MZT (Gerber et al., 2006; Tadros et al., 2007; Benoit et al., 2009; Chen et al., 2014; Laver et al., 2015). In addition, Brat and Pum proteins bind to non-overlapping sets of mRNAs; whereas Brat tends to associate with maternal mRNAs that are cleared in both a ZGA-independent and ZGA-dependent manner, Pum targets are mostly enriched in mRNAs that require zygotic transcription for their decay. This implies that there is a role for Pum in the ZGA-dependent mode of maternal mRNA clearance (Laver et al., 2015). Moreover, the maternally provided Piwi protein Aubergine also promotes transcript clearance in Drosophila embryos in concert with Piwi-interacting RNA (piRNAs) (Rouget et al., 2010; Barckmann et al., 2015). In Xenopus, the embryonic deadenylation element-binding protein (EDEN-BP; also known as CUGBP1) recognizes specific sequences in maternal mRNAs and mediates their degradation through the PARN deadenylase (Audic et al., 1998; Paillard et al., 1998; Osborne et al., 2005; Cosson et al., 2006). In mouse embryos, absence of the ARE-binding protein ZFP36L2 leads to an early developmental arrest at the two-cell stage, possibly owing to the control of maternal mRNA clearance by ZFP36L2 (Ramos et al., 2004). Roles for AREs in maternal mRNA decay during Xenopus MZT have also been demonstrated (Voeltz and Steitz, 1998). Recently, BTG4 has been shown to mediate degradation of thousands of maternal mRNAs in a CCR4-NOT-dependent manner during MZT in mice (Liu et al., 2016). It is noteworthy that BTG4 also associates with maternal mRNAs during zebrafish MZT (Despic et al., 2017) and therefore represents a potentially conserved factor in the ZGA-independent program of mRNA clearance in vertebrates. Importantly, although in vivo mRNA targets of RBPs often overlap with groups of transcripts that are destabilized during MZT, RBP mutants tend to only show mild defects in transcript stability, indicating that there is redundancy (Gerber et al., 2006). This strongly suggests the existence of a complex network of RNA-protein interactions within the post-transcriptional regulatory code of maternal mRNA stability and decay.

Recently, we have demonstrated the transcription-dependent relocalization of RBPs during zebrafish MZT, which may globally impact on maternal mRNA stabilities (Despic et al., 2017). Before ZGA, heterogeneous nuclear ribonucleoprotein A1 (Hnrnpa1) associates with the 3′UTRs of maternal mRNA in a sequence-specific manner and promotes translation through the regulation of poly(A) tail length (Despic et al., 2017). At ZGA, Hnrnpa1 dissociates from maternal transcripts and translocates to the nucleus, where it regulates processing of zygotically expressed pri-miR-430 into active miR-430 (Fig. 2D). Thus, Hnrnpa1 subserves several gene regulatory mechanisms before, during and after ZGA. The transcription-dependent translocation of Hnrnpa1 and other conserved RBPs during MZT is a shared phenomenon across many species (Dreyer et al., 1982; Dequin et al., 1984; Sanford and Bruzik, 2001; Vautier et al., 2001; Despic et al., 2017). Because these RBPs also have roles as pre-mRNA splicing factors, one rationale is that they are repurposed at ZGA to assist the processing of nascent zygotic transcripts in the nucleus. Although the mechanism that underlies nuclear translocation of RBPs is unknown, it may represent an evolutionarily conserved feature that is involved in maternal mRNA stability control.

Previously, comparative genome analysis of different organisms has revealed that certain codons are present in much higher frequencies than their synonymous counterparts (Grantham et al., 1980). The unequal frequency of synonymous codons was further found for genes within genomes of individual organisms (Ikemura, 1985). This phenomenon is called codon usage bias. Moreover, the frequency of certain codons correlates with cellular concentrations of distinct tRNAs, or the copy number of the tRNA genes that decode them (Ikemura, 1985; dos Reis et al., 2004). Therefore, codons have different optimality with respect to the rate of their decoding by the ribosome. Optimality of a codon depends on the balance between the concentration of tRNAs (tRNA supply) and the frequency of the codon being decoded by that tRNA (tRNA demand) (Pechmann and Frydman, 2013) (Fig. 3A). Emerging evidence that non-optimal codons slow translation across the mRNA sequence support a model whereby codon optimality determines translation elongation rates and translation efficiency of individual mRNAs (Nedialkova and Leidel, 2015; Yu et al., 2015) (Fig. 3B). These changes in translation elongation rates that are driven by codon optimality can in turn influence total protein output, folding of the nascent polypeptide chain, and the activity of the final protein product (Zhou et al., 2013; Sander et al., 2014).

The idea that codon optimality might influence mRNA stability originated three decades ago from seminal studies in budding yeast (Hoekema et al., 1987). Recent genome-wide studies of mRNA stabilities in bacteria and yeast have shown that codon optimality is highly correlated with the half-life of mRNA, such that mRNAs with a high percentage of optimal codons are more stable (Presnyak et al., 2015; Boel et al., 2016) (Fig. 3B). A mechanistic link between codon optimality, ribosome-decoding speed and mRNA decay has recently been elucidated (Radhakrishnan et al., 2016). A conserved DEAD-box RNA helicase, yeast Dhh1p, associates with mRNAs that have low codon optimality and are translated by slowly elongating ribosomes. Moreover, these transcripts are degraded in a Dhh1p-dependent fashion. Therefore, it has been postulated that Dhh1p senses the low elongation rates of the ribosome across non-optimal codons and triggers the decay of the respective mRNAs through conventional mRNA decay pathway, which involves mRNA deadenylation and decapping (Radhakrishnan et al., 2016) (Fig. 3B).

What about codon optimality in early embryos? Two independent studies have recently shown that translation of zebrafish maternal mRNAs with a high percentage of non-optimal codons leads to their degradation during MZT in a CCR4-NOT-dependent manner (Bazzini et al., 2016; Mishima and Tomari, 2016). These studies used different metrics to define codon optimality, yet both conclude that codon identity influences the half-lives of maternal mRNAs during MZT. Bazzini et al. further showed that the translation-dependent decay through codon identity operates during the MZT in other species, such as Drosophila, Xenopus and mouse (Bazzini et al., 2016), thereby suggesting that this decay program is evolutionarily conserved (Fig. 3C). Furthermore, codon-mediated decay is part of the ZGA-independent mode of maternal mRNA clearance, because it mostly affects transcripts whose degradation occurs in embryos injected with α-amanitin (Mishima and Tomari, 2016). Interestingly, the decay of mRNAs with non-optimal codons can be antagonized by 3′UTR length, such that mRNAs with non-optimal codons and longer 3′UTRs display higher stabilities (Mishima and Tomari, 2016). Thus, codon optimality is part of a multi-component regulatory system that determines mRNA half-life in embryos.

Global effects of codon optimality on maternal mRNA stabilities in zebrafish were measured during the late MZT (around 6 h post-fertilization) when significant drops in maternal mRNA abundance can be observed (Bazzini et al., 2016; Mishima and Tomari, 2016). It remains unknown whether codon optimality dictates mRNA decay at earlier time points in development. This possibility is unlikely, given that unbiased measurements of maternal mRNA stabilities suggest that maternal mRNA clearance begins after ZGA (Vesterlund et al., 2011). Thus, it remains unclear why the effects of codon optimality on maternal mRNA stabilities are delayed when codons are ubiquitous features of maternal transcripts present from the time of fertilization? There are several possible explanations. First, as previously discussed, deadenylation is evidently uncoupled from mRNA decay before ZGA in some species (Audic et al., 1997; Voeltz and Steitz, 1998). Therefore, the codon-mediated shortening of poly(A) tails may be used to fine-tune translational output from mRNAs without compromising their stabilities. Secondly, non-optimal codons may not be sensed earlier. As mentioned above, optimality of a codon depends on both tRNA supply and demand (Fig. 3A). One possibility is therefore that tRNA supply may vary over time during development, such that supply exceeds demand at fertilization; tRNA levels may become limiting later, when codon identity starts to trigger decay (Fig. 3C). This could be accomplished either through selective degradation of tRNAs, or through selective transcription of tRNA genes at ZGA. Previous studies have shown that transcription of tRNA genes by RNA polymerase III may be occurring at ZGA (Heyn et al., 2014). Therefore, the interplay between tRNA supply and demand may evolve as early embryogenesis proceeds (Fig. 3C). Thirdly, zebrafish embryonic rRNA sequences differ substantially from somatic rRNAs (Locati et al., 2017), as previously observed in other species (Komiya et al., 1986). Given the recent insights into differences in ribosomal composition and function during development (Shi and Barna, 2015; Fujii et al., 2017), it is worth considering that maternal mRNAs may interact with molecularly distinct ribosomes during their lifetimes. Interestingly, the shift to expression of somatic rRNAs in zebrafish occurs much later [∼9 h post fertilization (hpf)] than ZGA (∼3 hpf) (Heyn et al., 2017; Locati et al., 2017). This raises the question of whether the transition to somatic ribosomes could participate in late maternal mRNA decay. In order to decipher the mechanism by which codon identity influences transcript stability during MZT, it will be important to determine the precise onset of mRNA destabilization through ZGA-independent mRNA decay, as well as quantitative features, such as the kinetics of poly(A) tail shortening, dynamic tRNA concentrations and the elongation rates of embryonic and somatic ribosomes.

Eukaryotic RNAs contain over a hundred chemical modifications (Machnicka et al., 2013). Until recently, their biological significance remained largely unknown owing to the lack of methods for determining their abundance, the specific location within RNA and their dynamics within cells. N⁶-methyladenosine (m⁶A) has been intensively studied in recent years owing to the development of two antibody-based methods for transcriptome-wide identification of m⁶A positions: methylated RNA immunoprecipitation sequencing (MeRIP) and m⁶A individual-nucleotide-resolution crosslinking and immunoprecipitation (miCLIP) (Dominissini et al., 2012; Meyer et al., 2012; Linder et al., 2015). These studies revealed that m⁶A is abundant in mRNA; it decorates thousands of transcripts in coding regions and 3′UTRs and shows particular enrichment in 5′ ends of terminal exons and near stop codons (Dominissini et al., 2012; Meyer et al., 2012; Schwartz et al., 2013; Luo et al., 2014; Ke et al., 2015). Consequently, m⁶A has emerged as a potential regulator of numerous RNA processing events; these include pre-mRNA splicing, pri-miRNA processing, 3′ end processing, mRNA decay and translation and X-chromosome inactivation (Wang et al., 2014; Alarcón et al., 2015a,b; Ke et al., 2015, 2017; Liu et al., 2015; Meyer et al., 2015; Wang et al., 2015; Haussmann et al., 2016; Lence et al., 2016; Molinie et al., 2016; Patil et al., 2016; Xiao et al., 2016; Li et al., 2017) (Fig. 4A).

RNA modifications have ‘writers’, ‘erasers’ and ‘readers’ to regulate the dynamics of modifications and interpret their activities (Wu et al., 2017). The deposition of methyl groups occurs on the N⁶ position of adenosines within conserved RAC or DRACH sequence motifs, where D is not C, R is a purine, and H is not G (Ke et al., 2017). N⁶-adenosine methyltransferase activity is contained in the nuclear m⁶A writer proteins, methyltransferase-like 3 (METTL3) and methyltransferase-like 14 (METTL14), which modify RNA co-transcriptionally. Together, METTL3 and METTL14 constitute a minimal, core methyltransferase complex (Liu et al., 2014). METTL3 harbors the catalytic activity, whereas METTL14 serves as an important structural component that promotes substrate recognition and METTL3 activity (Sledz and Jinek, 2016; Wang et al., 2016). Additional proteins associate with the core and are important for both methyltransferase substrate-specificity and activity (Ping et al., 2014; Schwartz et al., 2014; Patil et al., 2016) (Fig. 4A). The discovery of an m⁶A demethylase or an eraser, ALKBH5, demonstrated that the addition of m⁶A is reversible and thereby has regulatory potential (Zheng et al., 2013).

Finally, m⁶A readers are RBPs that bind to RNAs that contain m⁶A through a YTH domain. This domain is a conserved RNA-binding module that exhibits 20- to 50-fold higher binding affinity for m⁶A than for non-methylated RNA (Luo and Tong, 2014; Theler et al., 2014; Xu et al., 2014, 2015). YTH proteins can be separated into three different classes based on their cellular localization: YTHDC1 is a nuclear m⁶A reader, whereas YTHDC2 and the YTHDF proteins (YTHDF1, YTHDF2 and YTHDF3) are predominantly cytoplasmic (Patil et al., 2017). These ‘readers’ are thought to mediate the effects of m⁶A on RNA processing in accordance with their subcellular localization. Indeed, YTHDC1 has been implicated in the regulation of pre-mRNA splicing and nuclear mRNA export (Xiao et al., 2016; Roundtree et al., 2017), whereas the mechanistic effects of cytoplasmic YTHDF proteins have largely been ascribed to their diverse roles in translation and mRNA turnover (Wang et al., 2014, 2015; Du et al., 2016; Kennedy et al., 2016). The extent to which YTHDC1 affects pre-mRNA splicing in an m⁶A-dependent manner awaits further evaluation, as recent studies have shown that the loss of m⁶A in cells only mildly affects pre-mRNA splicing (Geula et al., 2015; Ke et al., 2017). The current data suggests different roles for cytoplasmic m⁶A readers and often implicates a particular YTHDF protein in the regulation of divergent processes. However, YTHDF proteins are paralogs that are highly similar and share most of the mRNA-binding sites in mammalian cells (Patil et al., 2016), which suggests overlapping functions. Owing to these discrepancies in the results obtained by different groups, further work is required to resolve the functions of individual YTHDF proteins.

The roles of m⁶A modifications in cellular transitions have recently been demonstrated. In mouse embryonic stem cells, m⁶A regulates the transition from naïve pluripotency to differentiation; it does so by enforcing the degradation of mRNAs that encode factors that promote pluripotency, presumably by means of cytoplasmic m⁶A reader proteins (Geula et al., 2015). Interestingly, transcriptome-wide m⁶A mapping, together with CRISPR/Cas9-targeted mutagenesis and RNA-seq, has shown that maternally provided mRNAs are m⁶A-rich and degraded in an YTHDF2-dependent manner during zebrafish MZT (Zhao et al., 2017). Both maternally and zygotically provided YTHDF2 proteins contribute to the clearance of hundreds of maternal mRNAs that contain m⁶A, which makes YTHDF2 an important mRNA destabilization factor (Zhao et al., 2017) (Fig. 4B, red dotted line). Moreover, many transcripts that are destabilized in an YTHDF2-dependent manner are also known miR-430 targets (Zhao et al., 2017). This indicates that both maternal and zygotic factors act in a combined fashion to promote maternal mRNA clearance (Fig. 4C). However, only 25% of maternal, m⁶A-bearing transcripts were affected by the loss of YTHDF2 in that study (Kontur and Giraldez, 2017). In mammalian cells, YTHDF2 binds almost all m⁶A sites in mRNA (Patil et al., 2016). Thus, it remains unclear how a broad m⁶A binder like YTHDF2 fails to regulate the expression of a large number of m⁶A-containing mRNAs (Zhao et al., 2017) (Fig. 4B, green dotted line). It is conceivable that other RBPs act in combination with m⁶A readers to counteract the effects of YTHDF2 – either by promoting continuous mRNA polyadenylation, or by specifically blocking deadenylase activities (Fig. 4B,C). Finally, the largest proportion of maternal mRNAs that are affected by the loss of YTHDF2 do not contain m⁶A modifications (Kontur and Giraldez, 2017) (Fig. 4B, red line). This suggests either indirect effects of YTHDF2 on this population of transcripts, or that YTHDF2 may exert mRNA decay functions independently of m⁶A. Our recent mRNA interactome capture study that was carried out during zebrafish MZT identified all three YTHDF proteins as maternal mRNA-binding partners (Despic et al., 2017) (Fig. 4C). However, a precise contribution of YTHDF1 and YTHDF3 in maternal mRNA regulation remains to be determined. As seen in zebrafish, maternal mRNAs in Drosophila embryos are m⁶A-rich (Lence et al., 2016; Kan et al., 2017). Although a significant drop in m⁶A levels occurs at ZGA and coincides with a major wave of maternal mRNA decay (Lence et al., 2016), further analysis is necessary to confirm that the m⁶A-mediated maternal mRNA clearance exists in this species.

In this Review, we have discussed recent evidence showing that an elaborate interplay between codon optimality and translation regulation, RBP and miRNA binding, and downstream effects of RNA modifications like m⁶A determine the levels of discrete pools of maternally provided mRNAs (Fig. 5). However, many questions remain to be answered. Although the extensive identification of embryonic RBPs has begun through mRNA interactome capture, we lack comprehensive data of RBP activities in embryos. To understand the regulatory network that is specific for embryogenesis and repurposes RBPs for new functions in embryos, it will be necessary to obtain the individual binding patterns and activities of RBPs. In addition, we need to know how RBPs translocate to the nucleus upon transcription activation and whether the need for maternal mRNA clearance supports the establishment of zygotic mRNA localization patterns. Superimposed on these points is the overall question of kinetics in this rapidly evolving system, where mRNA and tRNA levels, poly(A) tail lengths and ribosome identity may be in a continuous state of flux. Our current knowledge indicates that the intricate and highly orchestrated post-transcriptional regulation of embryonic mRNA is critically important for development, making future studies seem all the more urgent.

We thank our colleagues in the Neugebauer laboratory for helpful discussions and Dahyana Arias Escayola for comments on the manuscript. We also thank Olivia Howard for her help with the artwork.