m⁶A‐mediated alternative splicing coupled with nonsense‐mediated mRNA decay regulates SAM synthetase homeostasis

Abstract

Alternative splicing of pre‐mRNAs can regulate gene expression levels by coupling with nonsense‐mediated mRNA decay (NMD). In order to elucidate a repertoire of mRNAs regulated by alternative splicing coupled with NMD (AS‐NMD) in an organism, we performed long‐read RNA sequencing of poly(A)⁺ RNAs from an NMD‐deficient mutant strain of Caenorhabditis elegans, and obtained full‐length sequences for mRNA isoforms from 259 high‐confidence AS‐NMD genes. Among them are the S‐adenosyl‐L‐methionine (SAM) synthetase (sams) genes sams‐3 and sams‐4. SAM synthetase activity autoregulates sams gene expression through AS‐NMD in a negative feedback loop. We furthermore find that METT‐10, the orthologue of human U6 snRNA methyltransferase METTL16, is required for the splicing regulation in␣vivo, and specifically methylates the invariant AG dinucleotide at the distal 3′ splice site (3′SS) in␣vitro. Direct RNA sequencing coupled with machine learning confirms m⁶A modification of endogenous sams mRNAs. Overall, these results indicate that homeostasis of SAM synthetase in C. elegans is maintained by alternative splicing regulation through m⁶A modification at the 3′SS of the sams genes.

Full‐length sequences of natural NMD isoforms in C. elegans are elucidated by direct RNA sequencing. SAM synthetase homeostasis is regulated via pre‐mRNA splicing of the sams genes through m⁶A modification at the invariant AG of the distal 3′SS.

Introduction

Alternative splicing of precursor messenger RNAs (pre‐mRNAs) contributes not only to proteome diversity (Nilsen & Graveley, 2010; Ule & Blencowe, 2019) but also to regulation of gene expression levels by generating mRNA isoforms with a premature termination codon (PTC) (Hamid & Makeyev, 2014; Sibley, 2014). Such unproductively spliced mRNAs are unstable and almost undetectable due to an mRNA surveillance system termed nonsense‐mediated mRNA decay (NMD) (Lareau & Brenner, 2015; Kishor et␣al, 2019; Kurosaki et␣al, 2019). Genetic studies in yeast and nematode identified evolutionarily conserved NMD factors including UPF1, UPF2, and UPF3 (Pulak & Anderson, 1993; He & Jacobson, 1995). Searches for mRNAs stabilized under NMD‐deficient conditions in a variety of organisms identified many splicing variants as natural NMD substrates in addition to mRNAs from snoRNA host genes, pseudogenes, long noncoding genes, and viral genes (Chapin et␣al, 2014; Kawashima et␣al, 2014; Celik et␣al, 2017; Colombo et␣al, 2017; Muir et␣al, 2018). Among the genes whose expression is regulated through alternative splicing coupled with NMD (AS‐NMD), many splicing factors and regulators have been shown to negatively autoregulate their own expression at the level of pre‐mRNA splicing (Jumaa & Nielsen, 1997; Lejeune et␣al, 2001; Sureau et␣al, 2001; Lareau et␣al, 2007; Ni et␣al, 2007; McGlincy et␣al, 2010; Turunen et␣al, 2013; Jangi et␣al, 2014; Sun et␣al, 2017).

An advantage of utilizing C. elegans as a model organism for studying AS‐NMD is in that its NMD factors are not essential for the development or fertility, allowing for genetic studies and genome‐wide survey of natural NMD substrates in␣vivo. The C. elegans genome is rich in introns like those of higher organisms (Consortium CeS, 1998), and > 25% of its protein‐coding genes undergo alternative pre‐mRNA splicing (Ramani et␣al, 2009; Ramani et␣al, 2011; Tourasse et␣al, 2017). Early studies with splicing‐sensitive microarrays identified 30 genes regulated by AS‐NMD, including those for splicing regulators (Barberan‐Soler & Zahler, 2008; Barberan‐Soler et␣al, 2009). We previously reported RNA‐seq analysis of mRNAs in an NMD‐deficient mutant smg‐2 and demonstrated that eight of ribosomal protein genes undergo negative autoregulation through AS‐NMD (Takei et␣al, 2016). Recently, AS‐NMD has been demonstrated to play crucial roles in longevity of long‐lived mutants or by dietary restriction and RNA‐seq analyses identified alternative splicing events relevant to longevity (Son et␣al, 2017; Tabrez et␣al, 2017). A caveat of the RNA‐seq analyses, however, is that an entire sequence of each mRNA molecule cannot be accurately reconstructed from the short‐read RNA‐seq data.

Post‐transcriptional base modification of mRNAs has emerged as a new layer of regulatory mechanisms controlling gene expression (Frye et␣al, 2016). Among such modifications, N ⁶‐methyladenosine (m⁶A) is the most prevalent internal modification in mRNAs in higher eukaryotic species (Wei et␣al, 1975). The m⁶A modification is reversible; methyltransferases (“writers”) and demethylases (“erasers”) methylate and demethylate mRNAs, respectively (Duan et␣al, 2019; Shi et␣al, 2019; Zaccara et␣al, 2019; Huang et␣al, 2020a; Huang et␣al, 2020b). Almost all of m⁶A modifications in mRNAs are deposited by a multicomponent m⁶A methyltransferase complex composed of a core component, the methyltransferase‐like 3 (METTL3)/METTL14 heterodimer, and other regulatory factors (Liu et␣al, 2014; Ping et␣al, 2014; Schwartz et␣al, 2014). The m⁶A modifications may alter local RNA structures to switch interaction with RNA‐binding proteins (Liu et␣al, 2015) or may affect splicing, nuclear export, translation initiation, and stability of mRNAs through specific recognition by m⁶A‐binding proteins (“readers”) such as YTHDC1, YTHDF1, and YTHDF2 (Wang et␣al, 2015; Xiao et␣al, 2016; Roundtree et␣al, 2017). The m⁶A modifications in RNAs can be removed by erasers FTO and ALKBH5 (Jia et␣al, 2011; Zheng et␣al, 2013). In C. elegans, however, orthologous genes for the m⁶A writers, erasers, and readers mentioned above are absent from the genome (Cunningham et␣al, 2019; Arribere et␣al, 2020) and recent studies demonstrated that only a limited fraction of mRNAs would have m⁶A modification (van Delft et␣al, 2017; Liberman et␣al, 2020). It is therefore still unclear whether the m⁶A modification also plays regulatory roles in mRNA metabolism in C. elegans.

In this study, we performed high‐throughput long‐read sequencing of mRNAs to reveal their full‐length sequences in an NMD‐deficient mutant strain with a mutation in the smg‐2 gene, encoding the crucial NMD factor UPF1. We summarize features of genes subjected to AS‐NMD. To search for a novel regulatory mechanism of AS‐NMD other than autoregulation by RNA‐binding proteins, we focused on S‐adenosyl‐L‐methionine (SAM) synthetase genes and demonstrate that m⁶A modification at the invariant AG dinucleotide of a 3′ splice site (SS) by a methyltransferase METT‐10 switches the choice of competing 3′SSs, leading to AS‐NMD for homeostasis of the enzyme.

Results

Full‐length sequences of NMD isoforms in C. elegans

In order to reveal full‐length sequences of natural NMD substrates in C. elegans, we performed long‐read sequencing of poly(A)⁺ RNAs from the smg‐2 (yb979) mutant (Kuroyanagi et␣al, 2007) with a direct RNA sequencing protocol on a Nanopore MinION platform (Garalde et␣al, 2018). To minimize over representation of abundant mRNAs from housekeeping genes, we purified newly synthesized RNAs after 6‐h metabolic labeling with 4‐thiouracil (4TU) in synchronized L1 larvae and directly sequenced the nascent poly(A)⁺ RNAs (see Methods). After base‐calling, 1,402,229 reads were mapped to the C. elegans genome and we identified 12,517 mRNA isoforms from 8,028 genes (Dataset EV1) supported by at least 3 reads whose exon‐exon junctions were also supported by short‐read RNA‐seq data from smg‐2 mutants in previous studies (Kuroyanagi et␣al, 2013; Takei et␣al, 2016; Son et␣al, 2017). For comparison, we collected and pooled as many direct RNA sequencing data in a public database as possible for mRNAs from various stages of a wild‐type strain N2 (Roach et␣al, 2020). With the same procedure, 3,511,592 reads were mapped to the genome and 18,376 isoforms from 11,331 genes (Dataset EV1) were supported by short‐read RNA‐seq data from N2 in previous studies (Takei et␣al, 2016; Son et␣al, 2017). Figure 1A summarizes numbers of exon‐exon junctions identified in the smg‐2 mutant and/or N2 as well as those deduced from gene models in WormBase (WS271) in 6,642 genes commonly detected in the two strains. Even though the number of reads and the stage of the worms are limited in our analysis, we detected more than 700 novel junctions that were not detected in N2 or predicted in WormBase (Fig 1A), suggesting that mRNAs with unusual splicing patterns are stabilized in the smg‐2 mutant.

Figure 1. Long‐read sequencing reveals putative NMD isoforms enriched in the smg‐2 mutant.

• A
  Venn diagram of exon‐exon junctions in 6,642 common genes identified among the Nanopore reads from the smg‐2 (yb979) mutant and wild‐type N2 as well as those deduced from gene models in WormBase (WS271).
• B
  Volcano plot of 8,701 mRNA isoforms from 2,931 genes with multiple isoforms. X‐axis indicates a difference in percentage of each mRNA isoform within the gene (Δ% _{smg‐2 ‐N2}). Y‐axis indicates ‐log₁₀ of P‐value in Fisher's exact tests. Blue symbols indicate 219 mRNA isoforms significantly depleted from the smg‐2 mutant (P < 0.05, Δ% _{smg‐2 ‐N2}< −10). Red symbols indicate 420 isoforms significantly enriched in the smg‐2 mutant (P < 0.05, Δ% _{smg‐2 ‐N2}> 10).
• C, D
  Length distribution of 3′UTRs (C) and proportions of putative PTC‐containing isoforms (D) in 10,136 mRNA isoforms detected in N2 and/or smg‐2 (All), the 219 isoforms depleted from and the 420 isoforms enriched in the smg‐2 mutant.
• E
  GO terms significantly enriched in 259 genes with smg‐2‐enriched PTC isoforms (top) and in 30 genes with smg‐2‐depleted PTC isoforms (bottom). Modified Fisher's exact P‐values in the DAVID Bioinformatics system are indicated. Note that 14 of the 30 genes with the smg‐2‐depleted mRNA PTC isoforms are common with those having smg‐2‐enriched PTC isoforms. This is why the same GO terms are enriched in these two groups. The mRNAs from such genes likely contain upstream open reading frames (uORFs), which may not be directly involved in or differentially affect NMD.

In order to identify mRNA isoforms whose proportions to the total mRNAs within the genes are significantly different between the smg‐2 mutant and N2, we performed Fisher’s exact tests with read numbers in the Nanopore sequencing data for 2,931 genes with two or more isoforms and with at least 10 total reads in each strain (Dataset EV1). Here, we assumed that the proportions of the NMD isoforms within the genes remain constant throughout development and in nascent RNAs like those of ribosomal protein genes (Takei et␣al, 2016) and used the pooled data for N2. Figure 1B summarizes a difference in percentages of an mRNA isoform within the gene (Δ% _{smg‐2 ‐N2}, percentage in smg‐2 subtracted by percentage in N2) and a P‐value for each of 8,701 isoforms. We identified 219 of smg‐2‐depleted isoforms (Δ% _{smg‐2 ‐N2} < −10, P < 0.05) from 208 genes and 420 of smg‐2‐enriched isoforms (Δ% _{smg‐2 ‐N2} > 10, P < 0.05) from 375 genes (Fig 1B and Dataset EV1). Assuming that an mRNA is translated from the first AUG codon, the smg‐2‐enriched isoforms (median = 613 nt) had significantly longer 3′ untranslated regions (3′UTRs) compared to the smg‐2‐depleted isoforms (median = 194 nt) as well as to total of 10,136 detected isoforms from 4,366 genes with at least 10 total reads in both strains (median = 218 nt) (P < 2e‐16, Wilcoxon rank sum test) (Fig 1C). Assuming that a premature termination codon (PTC) resides > 50 nucleotides (nt) upstream from the downstream‐most exon‐exon junction in an mRNA, 68.8% (289/420) of the smg‐2‐enriched mRNA isoforms had PTCs, whereas only 13.7% (30/219) of the smg‐2‐depleted isoforms and 28.1% (2,851/10,136) of all detected isoforms had PTCs (Fig 1D and Dataset EV1). We performed reverse transcription (RT)‐semi‐quantitative polymerase chain reaction (PCR) analysis of 12 genes with smg‐2‐enriched PTC isoforms and confirmed stabilization of all the PTC isoforms in the smg‐2 mutant except for extra isoforms with retained introns (Appendix␣Fig S1). These results indicate that 259 genes that produces the 289 PTC‐containing and smg‐2‐enriched mRNA isoforms are high‐confidence genes regulated by AS‐NMD in C. elegans.

Gene ontology (GO) analysis of the 259 genes that undergo AS‐NMD revealed enrichment of genes related to RNA translation and processing (Fig 1E). This is consistent with previous findings that many of RNA‐binding proteins (RBPs) including ribosomal proteins can negatively autoregulate their own expression by modulating alternative splicing patterns in both mammals and nematodes (Jangi & Sharp, 2014; Wani & Kuroyanagi, 2017).

Dynamic alternative splicing regulation of SAM synthetase genes upon feeding

Genes related to metabolism are also significantly enriched among those that undergo AS‐NMD (Fig 1E). During the course of analysis of such genes, we found that alternative splicing of the sams‐3 and sams‐4 genes (Fig 2A) drastically changed upon feeding and fasting in the smg‐2 mutant (Fig 2B lanes 1–2 and Appendix␣Fig S2). RT‐quantitative PCR (qPCR) analysis revealed 5.6‐fold and 2.0‐fold increase in the total amounts of sams‐3 and sams‐4 mRNAs, respectively, upon feeding and confirmed divergent changes in the amounts the productive and PTC‐containing mRNA isoforms (Fig EV1A). We confirmed that the PTC‐containing sams‐3 and sams‐4 isoforms are also induced upon feeding and actually degraded by NMD in the wild‐type background by feeding L1 larvae of N2 in the absence and presence of a translation inhibitor emetine that eventually inhibits the translation‐dependent NMD process (Fig 2B lanes 4–6 and Fig EV1B). Alternative splicing of their paralogue sams‐5 was also dynamically regulated upon feeding and fasting (Figs 2 and EV1, Appendix␣Fig S2). We reasoned that expression of these genes is regulated by AS‐NMD under certain environmental conditions and therefore focused on these genes in the following sections.

Figure 2. Alternative splicing of sams‐3, sams‐4, and sams‐5 is dynamically regulated upon feeding.

• A
  Schematic representation of the sams genes in C. elegans. Boxes indicate exons. Coding regions of productive mRNAs are in orange. A truncated coding region with a premature termination codon in a pseudogene sams‐2 is colored cyan. Double arrows illustrate regions deleted in mutant alleles indicated.
• B
  Alternative splicing of sams‐3, sams‐4, and sams‐5 in unfed and fed worms without or with a protein synthesis inhibitor emetine. Synchronized L1 larvae of an NMD‐deficient strain KH1668: smg‐2 (yb979) (lanes 1‐3) and a wild‐type strain N2 (lanes 4‐6) were incubated in S‐complete medium alone (lanes 1, 4), with a standard E. coli strain OP50 (OD₆₀₀ = 10.0) (lanes 2, 5) or OP50 supplemented with 10 mg/ml emetine (lanes 3, 6) for 3 h at 20°C. Total RNAs were extracted from whole animals and subjected to semi‐quantitative RT–PCR, whose products were analyzed by capillary electrophoresis. Representative gel‐like presentation is indicated (n = 3). Schematic structure of each PCR product is indicated on the right. Open reading frames (ORFs) for full‐length and truncated proteins are in orange and cyan, respectively. rpl‐12 was used as an unaffected control.

Figure EV1. RT–qPCR analysis of sams mRNA isoforms in fed and unfed animals.

• A, B
  Relative amounts of sams‐3 (column 1), sams‐4 (column 2), and sams‐5 (column 3) mRNA isoforms as well as sams‐1, sams‐2, fat‐7, and acs‐2 mRNAs (column 4) compared to eef‐1A.1 (n = 3, mean ± standard error of mean) in the smg‐2 (yb979) mutant (A) and in N2 (B). The same RT products were analyzed in Fig 2B. None, S‐complete medium alone; OP50, OP50 in S‐complete (OD₆₀₀ = 10.0); +Eme, OP50 in S‐complete supplemented with 10 mg/ml emetine. Sequences of the primers are available in Appendix␣Table S3.

The sams genes are members of a gene family encoding SAM synthetase that produces SAM from L‐methionine (L‐Met) and adenosine tri‐phosphate (ATP). The C. elegans genome contains five members of the SAM synthetase genes (Fig 2A) (Tamiya et␣al, 2013). sams‐3 and sams‐4 are highly homologous to each other (91.9% nucleotide sequence identity in the coding region), share exon organization (Fig 2A), and are divergent neighbors in the C. elegans genome. sams‐5 also shares the gene structure and the alternative splicing patterns (Fig 2). We noticed that nucleotide sequences of intron 2, the site of the alternative splicing regulation, in these genes are extraordinarily conserved for introns (87.5% identity between sams‐3 and sams‐4) (Fig EV2). In contrast, sams‐1 lacks an intron at this position (Figs 2A and EV2) and is therefore constitutively expressed. sams‐2 is the most homologous to sams‐3 (94.2% overall nucleotide sequence identity), yet is a pseudogene because of a frame‐shifting insertion in exon 3 (Fig EV2) and is constitutively spliced due to a substitution at the proximal 3′ splice site (SS) for intron 2 (Fig EV2).

Figure EV2. Nucleotide sequence alignment of genomic regions spanning from exon 2 through exon 3 of the sams genes in C. elegans .

The sequences were aligned with Clustal V algorithm followed by manual adjustment. Conserved nucleotides are highlighted in black. Exons are boxed in magenta. Canonical and cryptic 5′SSs as well as proximal/unproductive and distal/productive 3′SSs are boxed in green. A yellow box indicates a sams‐3/sams‐4 RNA fragment used in in␣vitro methylation analysis with recombinant METT‐10 (Fig 5). Cyan boxes indicate sams‐3/sams‐4 and sams‐5 RNA fragments used in Nanopore direct sequencing after in␣vitro methylation (Fig 6). A red arrowhead indicates the m⁶A modification site. A black arrowhead indicates a frame‐shifting insertion in sams‐2. Note that sams‐1 lacks intron 2.

To ask whether the feeding‐induced change in the alternative splicing patterns of the sams genes requires newly synthesized proteins, we fed the smg‐2 worms in the presence of emetine. The ratios and the amounts of the mRNA isoforms changed to some extent even in the presence of emetine (Figs 2B and EV1). This is in good contrast to the amounts of fat‐7 and acs‐2 mRNAs whose induction and reduction, respectively, were completely suppressed by the emetine treatment (Fig EV1A). These results indicated that induction and dynamic splicing regulation of the sams genes do not depend on synthesis of new proteins upon feeding.

Negative feedback regulation of alternative splicing of the SAM synthetase genes

To address biological roles for the dynamic AS‐NMD of the sams genes in␣vivo, we tested whether or not SAM synthetase activity autoregulates SAMS expression. Because sams‐1 is the only sams gene that is constitutively expressed and because sams‐3 and sams‐4 are neighboring genes, we tested effects of deleting sams‐1 and sams‐5 on splicing patterns of sams‐3 and sams‐4 in the smg‐2 mutant background (Fig 3A). In unfed worms, the splicing patterns of sams‐3 and sams‐4 were not so much different between the smg‐2 (lane 1) and the smg‐2; sams‐5; sams‐1 (lane 5) mutants. Upon feeding, however, proportions of the enzyme‐coding isoforms were apparently higher in the smg‐2; sams‐5; sams‐1 mutant (lane 6) compared to the smg‐2 mutant (lane 2). RT–qPCR analysis revealed comparable induction of total sams‐3 (5.6‐fold in smg‐2 vs 6.4‐fold in smg‐2; sams‐5; sams‐1) and sams‐4 (1.7‐fold vs 1.6‐fold) mRNAs (Fig EV3, EV4, EV5), indicating that disruption of sams‐1 and sams‐5 did not affect transcriptional regulation of sams‐3 or sams‐4. Isoform‐specific qPCR analysis confirmed divergence of the feeding‐induced splicing change between the two strains (Fig EV3A and B). We also fed these worms in the presence of a substrate or a competitive inhibitor of SAM synthetase. The substrate L‐Met had little effect (lanes 3 and 7) while the inhibitor cycloleucine (Lombardini & Talalay, 1970) slightly but significantly increased the proportions (lanes 4 and 8) and the amounts (Fig EV3A) of the productive isoform of sams‐3 in either strains. These results indicated that SAM synthase activity regulates AS‐NMD of the sams‐3 and sams‐4 genes.

Figure 3. Expression of sams‐3, sams‐4, and sams‐5 is negatively regulated by SAM synthetase activity through alternative splicing.

• A
  Alternative splicing of sams‐3, sams‐4, and sams‐5 in smg‐2 (yb979) and smg‐2 (yb979); sams‐5 (gk147); sams‐1 (ok2946) mutants. Synchronized L1 larvae of each strain were incubated in S‐complete medium alone (lanes 1 and 5), with OP50 (OD₆₀₀ = 10.0) (lanes 2 and 6) and with OP50 supplemented with 25 mM L‐Met (lanes 3 and 7) or 25 mM cycloleucine (cLeu) (lanes 4 and 8) for 3 h at 20°C. The splicing patterns were analyzed and presented as in Fig 2B (n = 3).
• B, C
  Western blot analysis of SAMS‐1, SAMS‐3, and SAMS‐4 during larval development in the smg‐2 (yb979) (B) and wild‐type (C) backgrounds. Genotypes of the worms are smg‐2 (w) and smg‐2; sams‐5; sams‐1 (s) in (B) and wild‐type (w) and sams‐1 (s) in (C). Synchronized L1 larvae of each strain were incubated with OP50 at 20°C and subjected to Western blot analysis at indicated time points. Anti‐β‐tubulin (B) or Coomassie Brilliant Blue (CBB) staining (C) was used as a loading control. Specificity of the antibodies is confirmed in Appendix␣Fig S12. Note that upregulation of SAMS‐1 protein in the wild type during larval development in (C) is consistent with feeding‐induced upregulation of sams‐1 mRNA in Fig EV3D.

Figure EV3. RT–qPCR analysis of sams mRNA isoforms in smg‐2 and smg‐2; sams‐5; sams‐1 mutants fed in the presence of L‐Met or cycloleucine.

• A–C
  Relative amounts of sams‐3 (A), sams‐4 (B), and sams‐5 (C) mRNA isoforms compared to eef‐1A.1 (n = 3, mean ± standard error of mean) in smg‐2 (yb979) and smg‐2 (yb979); sams‐5 (gk147); sams‐1 (ok2946) mutants. The same RT products were analyzed in Fig 3A. None, S‐complete medium alone; OP50, OP50 in S‐complete (OD₆₀₀=10.0); +Met, OP50 in S‐complete supplemented with 25 mM L‐Met; +cLeu, OP50 in S‐complete supplemented with 25 mM cycloleucine. *P < 0.05; **P < 0.01 (Student's t‐test). Sequences of the primers are available in Appendix␣Table S3.
• D
  Relative amounts of sams‐1 mRNA and sams‐3 and sams‐4 productive mRNA isoforms compared to eef‐1A.1 (n = 3, mean ± standard error of mean) in L1 larvae of N2 and the sams‐1 (ok2946) mutant incubated without (None) or with OP50 in S‐complete (OD₆₀₀ = 10.0) for 3 h at 20°C.

Figure EV4. Expression of SAMS‐3 and SAMS‐4 is upregulated in a mett‐10 single mutant.

• A
  Relative amounts of sams‐1, sams‐3, and sams‐4 productive mRNAs compared to eef‐1A.1 (n = 3, mean ± standard error of mean) in N2 and the mett‐10 (ok2204) mutant. Synchronized L1 larvae of each strain were incubated in S‐complete medium alone (None) or with OP50 (OD₆₀₀ = 10.0) for 3 h at 20°C.
• B
  Western blot analysis of SAMS‐1, SAMS‐3, and SAMS‐4 during larval development of N2 (w) and the mett‐10 (ok2204) mutant (m). Synchronized L1 larvae of each strain were incubated with OP50 at 20°C and subjected to Western blot analysis at indicated time points. Coomassie Brilliant Blue (CBB) staining was used as a loading control. Representative images of three replicated experiments are shown. Note that upregulation of SAMS‐1 in both strains during larval development is consistent with feeding‐induced upregulation of sams‐1 mRNA shown in (A).

Figure EV5. Direct RNA sequencing with Nanopore technologies coupled with machine learning‐based classification reveals m⁶A modification of endogenous sams mRNAs.

• A
  Summary of machine learning‐based classification of sams mRNA reads in Nanopore direct RNA sequencing data with thirteen algorithms. Machine learning was conducted, and the results are shown as in Fig 6E. Numbers of reads used in the tests are indicated at the top; percentages of reads classified as “m⁶A‐modified” are shown with color code.
• B–E
  Relative importance of the mean (top), standard deviation (middle), and duration time (bottom) of normalized Nanopore currents at nucleotide positions −50 through +50 relevant to the m⁶A site for LightGBM (B), XGBoost (C), Random Forest (D), and Decision Tree (E).

To test whether increase in the amounts of the productive mRNA isoforms leads to increase in the amounts of SAMS proteins, we performed Western blot analysis with antibodies specific to each of the four SAMS proteins. The amounts of SAMS‐3 and SAMS‐4 remain constant throughout larval development in the smg‐2 mutant, whereas they gradually increased in the smg‐2; sams‐5; sams‐1 mutant (Fig 3B). Consistent upregulation of SAMS‐3 and SAMS‐4 mRNAs and proteins in the sams‐1 mutant was observed in the wild‐type background (Figs 3C and EV3D). These results indicated that homeostasis of SAM synthetase activity in C. elegans is maintained by negative feedback through AS‐NMD of the sams genes.

mett‐10 regulates alternative splicing of the sams genes in␣vivo

Because SAM is the major donor of a methyl group for methyltransferases that methylate proteins, DNAs, RNAs, or lipids (Fontecave et␣al, 2004), we speculated that altered SAM synthetase activity affects alternative splicing of the sams genes through altered methylation levels of specific target molecule(s) for certain methyltransferase(s). We first analyzed histone methylation levels because some of histone marks are known to affect alternative splicing by recruiting splicing factors or by affecting the transcription elongation rate in mammals (Allo et␣al, 2009; Schor et␣al, 2009; Luco et␣al, 2010; Saint‐Andre et␣al, 2011). Chromatin immunoprecipitation coupled with high‐throughput sequencing (ChIP‐seq) analyses of histone H3 with dimethylation on lysine 4 (H3K4me2), trimethylation on lysine 27 (H3K27me3) or trimethylation on lysine 36 (H3K36me3), however, did not reveal a significant difference at the alternatively spliced exon in the sams‐3 or sams‐4 locus between the smg‐2 and smg‐2; sams‐5; sams‐1 mutants (Appendix␣Fig S3), precluding the scenario.

We next speculated methylation of the sams pre‐mRNAs themselves. Although the most abundant internal chemical modification in higher eukaryotic mRNAs is m⁶A (Wei et␣al, 1975), the C. elegans genome lacks orthologous genes for METTL3, METTL14, and other components of mammalian m⁶A methyltransferase complex that are responsible for the vast majority of the m⁶A modifications (Cunningham et␣al, 2019; Arribere et␣al, 2020). Recently, it has been shown that another human methyltransferase METTL16 specifically catalyzes m⁶A modification in a consensus sequence UACAGARAA within a loop of six hairpin structures on 3′UTR of human MAT2A mRNA that is translated into SAM synthetase (Pendleton et␣al, 2017; Shima et␣al, 2017; Doxtader et␣al, 2018). We noticed that each of sams‐3, sams‐4, and sams‐5 pre‐mRNAs potentially forms a hairpin structure with a loop sequence similar to the METTL16 consensus and that the potential m⁶A site is exactly at the invariant AG dinucleotide of the distal 3′SS that is specifically used for the productive mRNA isoform (Figs 4A and B and EV2). We therefore tested whether the METTL16 orthologue of C. elegans, METT‐10 (Appendix␣Fig S4), is involved in the splicing regulation of sams‐3, sams‐4, and sams‐5 in␣vivo (Fig 4C). The proportions of the NMD isoforms upon feeding drastically decreased in a smg‐2; mett‐10 double mutant (lane 4) compared to those in the smg‐2 single mutant (lane 2), indicating that mett‐10 is required for shifting the splice site choice from the distal/productive to the proximal/unproductive 3′SSs upon feeding. We also found that the productive isoforms of sams‐3 and sams‐4 mRNAs are upregulated in a mett‐10 single mutant (Fig EV4A) and SAMS‐3 and SAMS‐4 protein levels are higher in the mett‐10 mutant (Fig EV4B) compared to the wild‐type strain, confirming that mett‐10 is required for negative regulation of sams‐3 and sams‐4 in the wild‐type background.

Figure 4. mett‐10 is required for effective production of NMD isoforms from the sams genes in␣vivo .

• A
  Nucleotide sequence alignment of putative stem‐loop structures in the C. elegans sams‐3, sams‐4, and sams‐5 genes together with six hairpin structures in 3′UTR of the human MAT2A gene. Conserved residues are shaded in black. Loop, transition, and stem regions of the MAT2A hairpins (Doxtader et␣al, 2018) are boxed in magenta, green, and gray, respectively. An arrowhead indicates adenine bases in the MAT2A hairpins that are specifically modified into m⁶A by METTL16 in␣vitro (Shima et␣al, 2017).
• B
  Schematic representation of the predicted hairpin structures in sams‐3/sams‐4 (left) and sams‐5 (right) pre‐mRNAs. The boundaries between intron 2 and exon 3 are indicated. Arrowheads indicate the m⁶A modification sites.
• C
  Alternative splicing of sams‐3, sams‐4, and sams‐5 in smg‐2 (yb979) and smg‐2 (yb979); mett‐10 (ok2204) mutants. Synchronized L1 larvae of each strain were incubated in S‐complete medium alone (lanes 1 and 3) or with OP50 (lanes 2 and 4) for 3 h at 20°C, and the splicing patterns were analyzed and presented as in Fig 2B.

METT‐10 specifically catalyzes the m⁶A modification at the distal 3′SS in␣vitro

We next asked whether C. elegans METT‐10 can directly methylate the sams pre‐mRNAs in␣vitro. We prepared a 127‐nucleotide (nt) RNA spanning from intron 2 to exon 3 of the sams‐3/sams‐4 genes (Fig EV2) and performed in␣vitro methylation reactions with recombinant full‐length METT‐10 protein as well as methyltransferase domain of human METTL16. As expected, liquid chromatography (LC) coupled with tandem mass spectrometry (MS/MS) revealed that these recombinant proteins specifically and efficiently methylated the adenine base within the AG dinucleotide at the distal 3′SS in a SAM‐dependent manner (Fig 5A and B, Appendix␣Figs S5 and S6).

Figure 5. METT‐10 catalyzes m⁶A modification at the invariant AG dinucleotide of the distal 3′SS in␣vitro .

• A
  Mass chromatograms of RNA fragments to detect m⁶A (top) or unmodified A (bottom) after in␣vitro incubation of 127‐nt sams‐3/sams‐4 pre‐mRNA (sequence available in Fig EV2) with recombinant full‐length METT‐10 protein in the presence (left) or absence (right) of 1 mM SAM. The sequence, m/z value, and charge state for each fragment are indicated on the right. Asterisks indicate non‐specific signals.
• B
  The negatively charged ions of RNase T1 fragment is decomposed in the instrument by collision‐induced dissociation (CID) using helium gas. The product ions produced by CID are assigned on the sequence illustrated on the top right inset panel. Nomenclature of the product ions are described in the literature (McLuckey et␣al, 1992). Product ions of c and y series derive from the 5′ and 3′ termini of the fragment, respectively.

Endogenous sams mRNAs have the m⁶A modification at the distal 3′SS

We finally investigated m⁶A modification of endogenous sams mRNAs. The potential m⁶A modification site is at the end of intron 2 and is excised from the productive mRNA, whereas it is retained in all the NMD isoforms. RNA immunoprecipitation with an antibody specific to m⁶A (m⁶A‐IP) followed by semi‐quantitative RT–PCR revealed depletion of the productive mRNA and enrichment of all the NMD isoforms for sams‐3, sams‐4, and sams‐5 (Fig 6A). RT–qPCR analysis confirmed specific enrichment of the NMD isoforms (Fig 6B), consistent with the idea that the NMD isoforms have the m⁶A modification in the NMD‐isoform‐specific regions.

Figure 6. Endogenous unproductive sams mRNAs have m⁶A modification at the AG dinucleotide of the distal 3'SS.

• A, B
  m⁶A‐IP specifically enriched sams mRNA isoforms that retain the AG dinucleotide at the distal/productive 3′SS. (A) Representative gel‐like images (n = 3) of semi‐quantitative RT–PCR analysis of sams‐3, sams‐4, and sams‐5 in input and immunoprecipitated (IP) RNAs from the smg‐2 (yb979) mutant. The splicing patterns were analyzed and presented as in Fig 2B. (B) Relative enrichment of total and each of mRNA isoforms compared to eef‐1A.1 mRNA by RT–qPCR. Error bars indicate standard error of mean (n = 3).
• C
  Examples of normalized Nanopore currents for unmodified and m⁶A‐modified in‐vitro transcribed sams‐3/sams‐4 RNAs at nucleotide positions −4 through +4 relevant to the m⁶A site. One hundred reads are plotted for each. Color codes are indicated.
• D
  Distribution of mean (top) and standard deviation (bottom) of normalized Nanopore currents at nucleotide positions −4 through +4 relevant to the m⁶A site for 5,000 of the unmodified and 2,680 of the m⁶A‐modified sams‐3/sams‐4 RNAs and 85 of an endogenous sams‐3 mRNA isoform E2/E3L. Color codes are indicated. Central bands represent the median, boxes represent the 25th and 75th percentiles, and whiskers represent the lowest and highest values. A red line at position 0 in the top panel indicates a cutoff line (mean current = 1.7557) that discriminates between the unmodified and m⁶A‐modified sams‐3/sams‐4 RNAs with accuracies of 64.58% for both.
• E
  Machine learning‐based classification of sams RNA reads in Nanopore direct RNA sequencing data. The statistics in (D) as well as duration time of the Nanopore current at nucleotide positions −50 through +50 of up to 80% of mapped reads from the unmodified and methylated in␣vitro transcribed sams‐3/sams‐4 and sams‐5 RNAs were pooled and used for training with a machine learning algorithm Gradient Boosting and accuracy of the classifier was tested with 20% of the mapped reads. The classifier was also applied to reads identified as unproductive sams‐3, sams‐4, or sams‐5 mRNA isoforms in the analysis of endogenous mRNAs shown in Fig 1. Numbers of reads used in the tests are indicated at the top; percentages of reads classified as “m⁶A‐modified” are shown with color code.
• F
  Relative importance of the statistics at each position for Gradient Boosting.

We further investigated m⁶A modification of the endogenous sams mRNAs by utilizing our direct RNA sequencing data. For comparison, we prepared barcoded, unmodified and m⁶A‐modified 161‐nt sams‐3/sams‐4 and 163‐nt sams‐5 pre‐mRNAs in␣vitro (see Methods for details; sequences are available in Fig EV2). Specific and efficient (> 97.5%) m⁶A modification at the AG dinucleotide of the distal 3′SS of the sams‐3/sams‐4 and sams‐5 RNAs methylated with the METTL16 methyltransferase domain was confirmed by LC‐MS/MS (Appendix␣Fig S7). These four RNAs were pooled and subjected to Nanopore direct RNA sequencing, which records ion current signals as an RNA strand passes through a nanopore (Garalde et␣al, 2018). The sams‐3/sams‐4 and sams‐5 RNAs were successfully distinguished by standard base‐calling and mapping. There was not a crucial difference, however, between unmodified and m⁶A‐modified RNAs in mean and standard deviation of the fluctuating current and duration time at each nucleotide position (Fig 6C and D, Appendix␣Fig S8); the unmodified and m⁶A‐modified sams‐3/sams‐4 RNAs were classified with only 64.58% of accuracy based only on the mean of the current at the m⁶A site (Fig 6D). We therefore conducted machine learning with a variety of algorithms to classify modification status of each of the sams RNA reads. Most of the thirteen algorithms tested successfully classified methylated test reads as “m⁶A‐modified” with 86–90% sensitivity and unmethylated test reads as “unmodified” with 11–19% false positives (Figs 6E and EV5A). As expected, the means of the current and the standard deviations at and near the m⁶A modification site contributed to the successful classification as demonstrated by some of the algorithms (Figs 6F and EV5B–E). With these classifiers, 73–100% of the endogenous sams‐3 or sams‐4 mRNA isoforms carrying the putative m⁶A modification site were classified as “m⁶A‐modified” (Figs 6E and EV5A). These results indicated that most, if not all, of the NMD isoforms of the endogenous sams mRNAs have the m⁶A modification at the distal/productive 3′SS.

Discussion

In the present study, we demonstrated that homeostasis of SAM synthetase in C. elegans is regulated at the pre‐mRNA splicing level of the sams‐3, sams‐4, and sams‐5 genes through m⁶A modification at the invariant AG dinucleotide of the distal 3′SS of intron 2 (Fig 7A). At low levels of SAM, the distal 3′SS remains unmodified and is preferentially used as a splice acceptor for the productive mRNA, which is translated into active SAM synthetase proteins that eventually increase the SAM level. At excess concentrations of SAM, the methyltransferase METT‐10 specifically catalyzes the m⁶A modification at the distal 3′SS, which leads to selection of the proximal 3′SS and production of the NMD isoforms.

Figure 7. A model for negative feedback regulation for homeostasis of SAM synthetase genes in C. elegans .

• A
  A model for AS‐NMD regulation of the sams genes in C. elegans. See text for details.
• B
  Modeled structure of C. elegans UAF‐2 binding to 5′‐UAGGU‐3′. The structure was modeled based on sequence homology to U2AF23 from S. pombe (Appendix␣Fig S9) and crystal structure of S. pombe U2AF23/U2AF59 complex bound to the RNA (Yoshida et␣al, 2020). N‐terminal Zn finger 1 (ZnF1), U2AF homology motif (UHM), and C‐terminal Zn finger 2 (ZnF2) domains are colored in blue, yellow and green, respectively. Red spheres indicate zinc ions. The position of the amino group methylated upon m⁶A modification (N ⁶) is indicated with an arrowhead.

This is the first demonstration that the m⁶A modification at the 3′SS interferes with splicing in any organisms. The invariant AG dinucleotide at the 3′SS is recognized by a small subunit of an evolutionarily conserved heterodimeric protein complex U2 auxiliary factor (U2AF) (Merendino et␣al, 1999; Wu et␣al, 1999). In C. elegans, the consensus 3′SS sequence UUUUCAG is recognized by U2AF composed of the large subunit UAF‐1 and the small subunit UAF‐2 (Zorio & Blumenthal, 1999; Hollins et␣al, 2005). Crystal structure of the small subunit of U2AF binding to a 3′SS sequence has been solved only for the orthologue in fission yeast Schizosaccharomyces pombe, U2AF23, in complex with a part the large subunit U2AF59 (Yoshida et␣al, 2015; Yoshida et␣al, 2020). It has also been demonstrated that m⁶A modification of a 3′SS sequence 5′‐UUAGGU‐3′ at the position −2 (UUm⁶AGGU) dramatically decreased affinity to the U2AF23 complex in␣vitro (Yoshida et␣al, 2020). To ask whether m⁶A modification at the AG dinucleotide would also affect 3′SS recognition by C. elegans UAF‐2, we modeled three‐dimensional structure of UAF‐2 binding to 5′‐UAGGU‐3′ after its homology to S. pombe U2AF23 (Appendix␣Fig S9). The amino group of the adenine base at position −2 that is methylated upon m⁶A modification is embedded in a pocket on the surface of Zn finger domain 1 (Fig 7B) and is intimately surrounded by the identical residues as S. pombe U2AF23 forming the pocket (Yoshida et␣al, 2020), consistent with our finding that the m⁶A modification at the invariant AG dinucleotide interferes with its use as the 3′SS (Fig 7A) even in the absence of a reader protein in C. elegans. As for the function of the m⁶A modification in alternative splicing regulation of mammalian pre‐mRNAs, conflicting results have been reported (Xiao et␣al, 2016; Ke et␣al, 2017b; Louloupi et␣al, 2018; Zhou et␣al, 2019), yet there is not a reported case where the invariant AG dinucleotide at the 3′SS has the m⁶A modification.

This study demonstrated that homeostasis of SAM in C. elegans is maintained by an indirect negative feedback loop relying on METT‐10 that utilizes SAM to catalyze m⁶A modification for AS‐NMD of SAM synthetase pre‐mRNAs (Fig 7A). In human cells, the mRNA level of the SAM synthetase gene MAT2A is also indirectly regulated by the SAM level and dual roles have been reported for METTL16 that methylates MAT2A mRNA on six hairpin structures in the 3′UTR; prolonged occupancy of METTL16 on the upstream‐most hairpin due to lack of SAM promotes splicing of a retained intron to maturate the MAT2A mRNA (Pendleton et␣al, 2017), whereas m⁶A modification of the six hairpins by METTL16 leads to nuclear degradation of the mRNA via a nuclear reader protein YTHDC1 (Bresson et␣al, 2015; Shima et␣al, 2017). SAM is the major methyl‐donor reagent in all living organisms and is also used as a source of methylene groups, amino groups, ribosyl groups, and aminopropyl groups (Fontecave et␣al, 2004). Altered SAM levels affect a variety of biological processes including longevity (Hansen et␣al, 2005), lipid homeostasis (Li et␣al, 2011; Walker et␣al, 2011), and immune response (Ding et␣al, 2015) in C. elegans as well as longevity in Drosophila (Obata & Miura, 2015), maintenance and differentiation of human‐induced pluripotent stem cells (iPSCs) (Shiraki et␣al, 2014), and osteoclast differentiation in the mouse (Nishikawa et␣al, 2015). Autoregulation of SAM synthetase gene expression by alternative pre‐mRNA splicing in C. elegans demonstrated in this study emphasizes the importance of maintaining the SAM level in eukaryotes.

We demonstrated that METT‐10 as well as human METTL16 methyltransferase domain specifically methylated the adenine base at the AG dinucleotide of the distal 3′SS of sams‐3/sams‐4 and sams‐5 pre‐mRNAs in␣vitro. A recent study on crystal structures of the METTL16 catalytic domain bound to the hairpins of the MAT2A mRNA revealed structural basis for recognition of the target RNAs (Doxtader et␣al, 2018). Most of the amino acid residues directly involved in recognition of the RNAs are conserved in METT‐10 (Appendix␣Fig S4A). Each nucleotide in a UACAG stretch in the loop is directly recognized by METTL16 (Doxtader et␣al, 2018) and is perfectly conserved in the sams genes (Fig 4A). There is not a direct contact between METTL16 and the stem structures (Doxtader et␣al, 2018), consistent with less conserved sequences in the stem region (Fig 4A). The transition region between the loop and the stem is highly conserved among the human MAT2A hairpins (Fig 4A) and forms three unusual base pairs between GU and AGAA motifs (Doxtader et␣al, 2018). However, the sequence of the transition region is not well conserved in the sams genes (Fig 4A) and a key residue Arg200 that recognizes the transition region is not conserved in METT‐10 (Appendix␣Fig S4A). Nevertheless, METTL16 as well as METT‐10 can specifically target the AG dinucleotide of the sams pre‐mRNAs (Fig 5 and Appendix␣Figs S5–S7). This is consistent with an idea that the transition region of the stem‐loop is a key region to tune the methylation efficiency and not the target recognition based on previous findings that a G‐to‐A substitution in the latter motif or an R200Q substitution rather increased the in␣vitro methylation activity of METTL16 (Doxtader et␣al, 2018). Considering that regulation of sams gene expression is based on balanced competition between splicing and m⁶A modification (Fig 7A), it is reasonable to suggest that highly conserved nucleotide sequences flanking the distal as well as proximal 3′SSs of intron 2 (Appendix␣Fig S10) play critical roles in the genus Caenorhabditis. MAT2A and sams mRNAs are so far the only known mRNA targets for human METTL16 and C. elegans METT‐10, respectively. Although the m⁶A modification is very rare (0.0008%) in C. elegans mRNAs (Liberman et␣al, 2020), this study revealed the critical and specific role for the m⁶A modification in the mRNA metabolism. Quite recently, Mendel et␣al (2021) have reached essentially the same conclusion with distinct approaches.

We performed long‐read direct RNA sequencing of poly(A)⁺ RNAs from L1 larvae of the smg‐2 mutant of C. elegans and revealed full‐length sequences of 3,056 splice variants that are not registered as gene models in WormBase including 1,532 variants with novel junctions and 1,963 putative NMD isoforms with PTCs. This result indicates that the long‐read sequencing is a powerful tool to elucidate the full‐length sequences of mRNAs and hence the repertoire of the isoforms.

We demonstrated that a metabolite SAM can negatively regulate expression of the SAM‐synthesizing enzyme genes through AS‐NMD for homeostasis. It is common in bacteria that metabolites directly and specifically bind to mRNAs to regulate their transcription and/or translation for feedback regulation (Pavlova et␣al, 2019), including SAM (McDaniel et␣al, 2003; Winkler et␣al, 2003; Montange & Batey, 2006). This study identified many non‐RBP genes whose expression may be regulated through AS‐NMD in C. elegans (Fig 1E). mett‐10 is among the genes whose novel isoforms were discovered in this study and have a putative PTC. We found that the PTC‐containing isoform of mett‐10 is stabilized in the smg‐2 mutant (Appendix␣Fig S4B and C), confirming that mett‐10 is also regulated by AS‐NMD. Further studies on splicing regulation of such genes will elucidate further examples of gene expression control by environmental conditions at the pre‐mRNA splicing level.

We conducted machine learning to classify m⁶A modification status at a single specific position in individual reads for the sams mRNA isoforms from the direct Nanopore RNA sequencing data (Fig 6D) because the properties of the electric current were not so much different between unmodified and m⁶A‐modified RNAs prepared in␣vitro (Fig 6C). Sequencing with the Nanopore technology deconvolutes changes in the electric current as a single stranded RNA transverses a pore protein (Garalde et␣al, 2018). Since the measured current fluctuated even while the RNA strand stays in a given position, we considered not only the mean but also the standard deviation of the recorded electric current as well as the duration time at each position in the machine learning. Unbiased parameter tuning for higher accuracy revealed major contribution of the mean and the standard deviation of the electric current at the relevant positions (Figs 6E and EV5B–E), indicating that these features are useful for classifying the base modifications. The direct Nanopore RNA sequencing technology has recently been applied by other groups to detecting the m⁶A modifications. Liu et␣al (2019) reported that systematic errors and decreased base‐calling qualities can predict m⁶A RNA modifications. Parker et␣al (2020) also reported higher base‐calling error rates for an m⁶A‐modified RNA than an unmodified RNA and identified > 17,000 possible m⁶A modification sites with a motif DRAYH in mRNAs from a plant Arabidopsis thaliana by comparing the error rates for an m⁶A‐writer mutant line to those for a complemented line. The differential error rate and base quality, however, were not evident in our direct RNA sequencing data for in␣vitro‐modified sams‐3/sams‐4 or sams‐5 RNAs (Appendix␣Fig S11), suggesting that the differential error rate analysis approach may be context‐dependent. Lorenz et␣al developed a software MINES to detect m⁶A modifications in mammalian mRNAs de novo (Lorenz et␣al, 2020), yet it relies on rich data of verified m⁶A modification sites from cross‐linking and immunoprecipitation with high‐throughput sequencing (CLIP‐seq) and a well‐established motif DRACH for the METTL3/METTL14 complex (Linder et␣al, 2015; Ke et␣al, 2017b). Thus, classification of the base modification status at a given or new site in individual direct sequencing reads is still challenging and requires a high‐quality training data set or a large number of reads with a writer mutant control.

Materials and Methods

Worm strains

Genotypes of the strains used are summarized in Appendix␣Table S1. Some of the strains were obtained from Caenorhabditis Genetics Center (CGC) and National BioResource Project (NBRP), Japan. All worms were cultured at 20°C unless otherwise mentioned. KH2418: smg‐2 (yb979); sams‐1 (ok2946), KH2421: smg‐2 (yb979); sams‐5 (gk147); sams‐1 (ok2946); and VC2428: sams‐1 (ok2946) X were maintained in the presence of 20 mM choline chloride (Nacalai Tesque) (Walker et␣al, 2011).

Synchronized worm preparation

Gravid worms raised in S‐complete medium (Lewis & Fleming, 1995) supplemented with an E. coli strain OP50 were bleached with standard bleaching solution (Lewis & Fleming, 1995). Embryos were harvested and washed three times with M9 buffer. The embryos were incubated in M9 buffer for 18 h (except for the smg‐2; mett‐10 mutant that was incubated for 24 h) with gentle agitation for hatching and the L1 larvae were harvested and washed with S‐complete.

Direct RNA sequencing

Synchronized L1 larvae of KH1668: smg‐2 (yb979) were fed with OP50 in S‐complete medium supplemented with 4 mM 4‐thiouracil (4TU) (Tokyo Chemical Industry) for 6 h with gentle agitation. The worms were harvested and washed with M9, and total RNAs were extracted from the whole animals by using RNeasy Plus Mini (QIAGEN) as described previously (Kuroyanagi et␣al, 2010; Kuroyanagi et␣al, 2013). Thio‐labeled RNAs were purified essentially as described previously (Duffy et␣al, 2015) with minor modification; 4TU‐labeled RNAs were biotinylated with MTSEA biotin‐XX (Biotium) in 90% dimethylformamide at 10°C for 30 min and purified by using Streptavidin Mag Sepharose (GE Healthcare). Ribosomal RNAs were then removed with Ribo‐Zero Gold rRNA Removal Kit (Epidemiology) (Illumina), and the remaining RNAs were subjected to direct RNA sequencing. Nanopore libraries were prepared from 500 ng RNAs and sequenced using SQK‐RNA002 kit and FLO_MIN107 flow cells on MinION of Oxford Nanopore Technologies (ONT) according to manual instructions. Nanopore reads of Poly(A)⁺ RNA from the smg‐2 (yb979) mutant and in␣vitro transcribed sams RNAs were base‐called using Guppy version 2.3.4 and 3.0.3 (ONT), respectively, with default RNA parameters. Demultiplexing reads of the in␣vitro methylated RNAs and unmodified RNAs was performed by using fast5_subset (ONT) and in‐house R scripts by checking 5′‐GUCAUCCC‐3′ (unmodified) and 5′‐GUCAUGGG‐3′ (m⁶A) sequences within 40 nt region from the 3′‐end of the reads with 1 nt mismatch/indel allowance.

Transcript annotations from Nanopore data

Nanopore direct RNA sequencing reads for the smg‐2 (yb979) mutant and those for N2 (Roach et␣al, 2020) were mapped to ce11 reference genome in WormBase (https://parasite.wormbase.org) by using minimap2 version 2.17 (Li, 2018) with “‐x splice”. For the N2 data, reads of all developmental stages were pooled before mapping. Full‐length transcript annotations were generated from minimap2 bam outputs using spliced_bam2gff and cluster_gff in Pinfish version 0.1.0 (ONT) with settings “‐c 3 ‐d 2 ‐e 30”. The transcripts were collapsed using collapse_partials in Pinfish with “‐d 2 ‐e 30 ‐f 1000”. The generated annotations were labeled with gene names in WormBase (WS271) using bedtools intersectBed version 2.27.1 (Quinlan & Hall, 2010). To filter truncated transcripts, annotations with 5′‐ends not within +/−100 nt of annotated 5′‐ends in WormBase (WS271) were discarded. Furthermore, artificial transcripts with splice junctions caused by base‐calling error were filtered using supporting junction reads of Illumina RNA‐seq data and annotated junctions in WormBase (WS271). The supporting junctions were generated from published Illumina RNA‐seq data of N2 and the smg‐2 mutants with DDBJ accession numbers DRX002713 (Kuroyanagi et␣al, 2013), DRX021821 and DRX021822 (Takei et␣al, 2016), SRX2516749, SRX2516750, SRX2516753, and SRX2516754 (Son et␣al, 2017). The Illumina reads were mapped to ce11 reference genome using STAR version 2.5.3a (Dobin et␣al, 2013) with options “‐‐alignIntronMin 20 ‐‐alignIntronMax 10000 ‐‐alignMatesGapMax 10000”. The STAR mapping outputs include a␣list of splice junctions. Transcripts containing not‐supported, not‐annotated junctions were discarded by in‐house R and Python␣scripts.

Analysis of putative NMD isoforms using Nanopore data

Differential isoform analysis was performed on isoforms generated from pooled Nanopore data for N2 (Roach et␣al, 2020) and the smg‐2 (yb979) mutant by using Pinfish in the same way as described above. The read number of each isoform was extracted by using Pinfish and compared between N2 and the smg‐2 mutant by Fisher’s exact tests with BH adjustment (Benjamini & Hochberg, 1995) to calculate FDR. Open reading frames (ORFs) in the isoforms were predicted by using predORF in systemPipeR version 3.10 (Backman & Girke, 2016) and the upstream‐most ORFs were used. ORFs in isoforms deposited in WormBase (WS271) were re‐defined in the same way. A premature termination codon (PTC) was annotated when the stop codon locates > 50 nt upstream from the last exon‐exon junction. Gene ontology (GO) analysis was performed on smg‐2‐enriched PTC isoforms (Δ% _{smg‐2 ‐N2} > 10, P < 0.05), and smg‐2‐depleted PTC isoforms (Δ% _{smg‐2 ‐N2} < −10, P < 0.05) by using DAVID version 6.8 (Huang et␣al, 2009) with the genome background “Caenorhabditis elegans”.

Total RNA extraction and RT–PCR

Synchronized L1 larvae were incubated in 400 µl of S‐complete medium (Lewis & Fleming, 1995) in 1.5‐ml non‐stick tubes (Ambion) with continual agitation at 1,400 rpm on Thermomixer (Eppendorf). Worms were washed three times with M9 buffer, and total RNAs were extracted as described previously (Kuroyanagi et␣al, 2010; Kuroyanagi et␣al, 2013). RT–PCR was performed essentially as described previously (Kuroyanagi et␣al, 2010; Kuroyanagi et␣al, 2013). The PCR products were analyzed by using 2100 BioAnalyzer (Agilent Technologies). qPCR was performed with LightCycler 480 (Roche) and SYBR Premix DimerEraser (TaKaRa). Sequences of the primers used in the semi‐quantitative RT–PCR and RT–qPCR experiments are available in Appendix␣Tables S2 and S3, respectively. Sequences of the PCR products were confirmed by direct sequencing or by cloning and sequencing.

Western blotting

Synchronized L1 larvae were incubated in 400 µl of S‐complete medium (Lewis & Fleming, 1995) in 1.5‐ml non‐stick tubes (Ambion) with continual agitation at 1,400 rpm on Thermomixer (Eppendorf). Worm lysates were extracted from the synchronized worms, separated by neutral polyacrylamide gel electrophoresis (NuPAGE, Invitrogen), and transferred to nitrocellulose membrane (Protran NC, Amersham). Western blotting was performed with 1∶1,000‐diluted rabbit anti‐SAMS antisera and 1∶1,000‐diluted HRP‐conjugated anti‐rabbit IgG antibody (Pierce). Chemiluminescence signals (West Dura, Thermo or Western BLoT Ultra Sensitive HRP Substrate, Takara) were detected by using LAS4000 (GE Healthcare). The rabbit anti‐SAMS‐1, anti‐SAMS‐3, and anti‐SAMS‐4 antisera were raised with synthetic peptides KALKISPALLEKAKGNPI, DVELLKKIGGKTISNGN, and SADMLAKSQGPAQPDV, respectively, by Sigma‐Aldrich Japan.

ChIP‐seq analysis of histone methylations

Synchronized L1 larvae of the smg‐2 and smg‐2; sams‐5; sams‐1 mutants were fed with OP50 in S‐complete medium for 30 min. The worms were then washed with M9, frozen in liquid nitrogen, and ground to powder with mortar and pestle. The worm powder was treated with 1% formaldehyde at 20°C for 10 min, and cell nuclei were collected by centrifugation at 1,000 g. The chromatin was solubilized with Covaris M220 (Peak Incident Power = 75 W; Duty Factor = 20%; Cycles Per Burst = 200; time = 1,800 s). Nucleosomes with modified histones were immunoprecipitated with mouse monoclonal antibodies (MABI0303, MABI0323, or MABI0333, MBL, Nagoya, Japan) and DNAs were purified with NucleoSpin Gel and PCR Clean‐up (MACHEREY‐NAGEL) after proteinase K digestion of proteins at 55°C and reverse crosslinking at 65°C overnight. Sequencing libraries were prepared with TruSeq ChIP Library Prep Kit (Illumina), and sequencing (Single End 50 bp) was performed with Hiseq3000 (Illumina). Low‐quality reads were trimmed by using Trimmomatic version 0.36 (Bolger et␣al, 2014) with options “‐phred33 LEADING:30 TRAILING:30 MINLEN:50”. Mapping to the ce11 reference genome was done by using Bowtie2 version 2.4.1 (Langmead & Salzberg, 2012) with default settings. Differentially bound regions were detected by using csaw (Lun & Smyth, 2016). In the csaw analysis, the ce11 genome was binned into 200‐bp windows with a step size of 100 bp. Normalized read counts of each window were compared between the smg‐2 and smg‐2; sams‐5; sams‐1 mutants. Significantly changed regions were selected as those windows with P < 0.01. Genome browser snapshots were generated by using Integrative Genomics Viewer (Robinson et␣al, 2011).

In vitro methylation

Recombinant His‐tagged full‐length METT‐10 and human METTL16 catalytic domain were expressed in E. coli and purified with HisTrap and Capto SP cation exchange columns. A template for the 127‐nt sams‐3/sams‐4 RNA was amplified from a synthetic oligonucleotide 5′‐AACAATATGTTTCTTTACCTGTTACAACGGTTACTTTGATTACAGAAACGGTGACTAAAACGGGTATGAT‐3′ with 5′‐GCTAATACGACTCACTATAGGGAAACGGGGCTGAAGCTTATCGCAACAATATGTTTCTTTACCTGTTAC‐3′, and 5′‐CAGCTTTCGAGGTGATCTCACCGCACAACATGATCATACCCGTTTTAGTCACC‐3′. The template DNA was transcribed in␣vitro with recombinant T7 RNA polymerase. In vitro methylation of the sams‐3/sams‐4 RNA was performed with 0.2 µM recombinant protein and 2 µM RNA in methylation buffer (20 mM HEPES‐KOH pH 7.5, 150 mM NaCl, and 1 m DTT) with or without 1 mM SAM at 37°C for 3 h in a 50 μl reaction mixture.

Mass spectrometry

Identification of modified bases was performed essentially as described previously (Shima et␣al, 2017). Collision‐induced dissociation (CID) spectrum analysis was performed as described previously (Shima et␣al, 2017).

Preparation of in‐vitro methylated RNAs for direct RNA sequencing

sams‐3/sams‐4 and sams‐5 genomic DNA fragments were amplified from N2 genomic DNA by using PrimeStar GXL (Takara). The sams‐3/sams‐4 and sams‐5 DNA fragments were further amplified with a forward primer 5′‐TAATACGACTCACT ATAGGGAAACGGGGCTGA AGC‐3′ (sams‐3/sams‐4) or 5′‐TAATACGACTCACTATAGGGTGACAAGA GGAT GAAAC‐3′ (sams‐5) and a barcoded reverse primer 5′‐TTTTTTTTTTTTTTTTTTTTGGGGGGGGGATGACGTTTCGGACGAG AAC‐3′ (C9A20) or 5′‐TTTTTTTTTTTTTTTTT TTTCCCCCCCCCATGACGTTTCGGACGAGAAC‐3′ (G9A20) to attach a T7 promoter and a C9A20 or G9A20 tail. The four DNA fragments were utilized as templates for in␣vitro transcription with T7 RNA polymerase, and the RNAs were purified with PAGE and gel extraction. The G9A20‐tagged sams‐3/sams‐4 and sams‐5 RNAs were subjected to methylation reaction with 30 µM of recombinant human METTL16 catalytic domain protein in the methylation buffer described above with 1 mM SAM at 37°C for 3 h in a 50 μl reaction mixture and were purified again with PAGE and gel extraction. The four RNA samples were then pooled and subjected to direct RNA sequencing.

m⁶A RNA immunoprecipitation

Total RNAs were extracted from synchronized L1 larvae of the smg‐2 mutant after feeding with OP50 in S‐complete medium for 3 h. Immunoprecipitation of m⁶A RNA was performed by using Rabbit anti‐m⁶A antibody (ab151230, Abcam) and Dynabeads Protein A (Invitrogen). Antibody‐bound magnetic beads were prepared by following manual instructions and washed with Binding buffer (10 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.5% NP‐40) three times. Total RNAs in 30 μl of TE buffer were denatured for 5 min at 65°C, and then, the RNAs were added to the antibody‐bound beads in 270 μl of Binding buffer and incubated on a mixer for 2 h at 4°C (40 μg of total RNAs and 3 μg of anti‐m⁶A antibody per sample). After incubation, the beads were washed with Binding buffer twice and Wash buffer (10 mM Tris–HCl pH 7.5, 1 M NaCl, 1% NP‐40) three times. Washed beads were suspended in 50 μl of RLT buffer (QIAGEN) and incubated on a mixer for 5 min at r.t. to elute the beads‐bound RNAs. Finally, the eluted RNAs were purified by using RNA Clean & Concentrator‐5 kit (Zymoresearch) and were subjected to semi‐quantitative RT–PCR and RT–qPCR analysis.

Classification of m⁶A‐modified sams RNAs

Raw Nanopore current data for in␣vitro prepared RNAs and the endogenous mRNAs from the smg‐2 mutant were mapped to sams transcript annotations generated by Pinfish using Tombo version 1.5 (ONT). Mean, standard deviation (STD), and duration time of the normalized current at each genomic position were extracted by using Tombo Python API. For machine learning, the data of the normalized current at positions from −50 through +50 relevant to the m⁶A site were used. Among 15,706 reads for the unmodified RNAs and 6,063 reads for the m⁶A‐modified RNAs, up to 80% were used to train classifiers, and 20% were used for tests. The reads for the unmodified RNAs in the training data set were downsampled to the same number as those of the m⁶A‐modified RNAs in order to prevent biased learning. To select the best performing models, we compared classifiers xgboost version 1.0.0 (https://github.com/dmlc/xgboost) (Chen & Guestrin, 2016) and LightGBM version 2.3.2 (https://github.com/microsoft/LightGBM) (Ke et␣al, 2017a) and those in scikit‐learn version 0.22 (https://github.com/scikit‐learn/scikit‐learn) (Pedregosa et␣al, 2011). Scikit‐learn includes a variety of algorithms including Decision Tree, Random Forest, Logistic Regression, K‐Nearest Neighbor, SVM, Linear Discriminant Analysis, Quadratic Discriminant Analysis, Multilayer Perceptron, Gaussian Naive Bayes, and Adaptive Boosting. Each classifier was tuned by grid‐search of hyperparameters by using hyperopt version 0.2.2 (https://github.com/hyperopt/hyperopt) (Bergstra et␣al, 2013).

Protein sequence analysis

Multiple sequence alignment with the Clustal V or W algorithm was performed by using a module MegAlign in Lasergene 15 (DNASTAR). Modeling of the three‐dimensional structure of UAF‐2 was performed by using Modeller 9.24 (Webb & Sali, 2016) after the structure of S. pombe U2AF23 in complex with U2AF59 and 5′‐UAGGU‐3′ (PDB Acc. No. 7C06). The 3D structures were analyzed and visualized with a module Protean 3D in Lasergene 17 (DNASTAR).

Author contributions

EW contributed to sequencing library preparation, bioinformatics analysis, RNA‐IP, and Western blotting. MT‐O and HK contributed to RT–PCR and RT–qPCR analysis. YI and TS contributed to in␣vitro methylation, and LC‐MS/MS. SW, MK‐A, SH, and ST contributed to preliminary data acquisition. KH and AF contributed to antibody preparation and validation. YS contributed to RNA and ChIP sequencing. HK organized the study and prepared worm strains. EW and HK wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Dataset EV1

Review Process File

The EMBO Journal (2021) 40: e106434.

Data availability

• RNA‐Seq data: DDBJ Sequence Read Archive (DRA) DRX184564 (https://ddbj.nig.ac.jp/DRASearch/experiment?acc=DRX184564).

• RNA‐Seq data: DDBJ Sequence Read Archive (DRA) DRX184565 (https://ddbj.nig.ac.jp/DRASearch/experiment?acc=DRX184565).

• Chip‐Seq data: DDBJ Sequence Read Archive (DRA) DRA009050 (https://ddbj.nig.ac.jp/DRASearch/submission?acc=DRA009050).

• Codes used in the sequence analysis: GitHub KUROYANAGI‐Lab/WATABE‐SAMS (https://github.com/KUROYANAGI‐Lab/WATABE‐SAMS).

• Nucleotide sequence: DDBJ/ENA/GenBank LC603057 (http://getentry.ddbj.nig.ac.jp/getentry/na/LC603057?filetype=html).

• The materials generated and/or analyzed during the current study are available from KUROYANAGI Hidehito (hidehito@med.u-ryukyu.ac.jp).