The DEAD-box RNA helicase Ddx39ab is essential for myocyte and lens development in zebrafish

The DEAD-box RNA helicase family is a large group of proteins characterized by the presence of an Asp-Glu-Ala-Asp (DEAD) motif that is highly conserved from bacteria to humans (Bleichert and Baserga, 2007; Rocak and Linder, 2004). Using the energy derived from ATP hydrolysis, these proteins modulate RNA topology and association/dissociation of RNA-protein complexes. DEAD-box RNA helicases play important roles in all aspects of RNA metabolism, including transcription, pre-mRNA splicing, rRNA biogenesis, RNA transport and translation (Calo et al., 2015; Jarmoskaite and Russell, 2011, 2014; Linder and Jankowsky, 2011). Recently, a fuller appreciation of the functions of these RNA helicases in variant physiological or developmental scenarios has started to emerge. Numerous reports have described dysregulation of expression or function of DEAD-box RNA helicases in cancer development or progression, indicating that DEAD-box proteins may be involved in key processes in cellular proliferation (Fuller-Pace, 2013; Sarkar and Ghosh, 2016). Recent reports have further shown that DEAD-box RNA helicases play diverse roles in developmental events ranging from body axis establishment (Meignin and Davis, 2008) to germ cell, blood, digestive organ and brain development (Hirabayashi et al., 2013; Hozumi et al., 2012; Payne et al., 2011; Zhang et al., 2012). These studies indicated that DEAD-box RNA helicase family members can regulate specific developmental processes by affecting pre-mRNA splicing, ribosomal biogenesis or RNA transport.

As a member of the DEAD-box family of ATP-dependent RNA helicases, Ddx39a was originally identified in a screen for proteins that interact with the essential splicing factor U2AF65 (Fleckner et al., 1997). Studies in different organisms indicated that, besides playing an important role in pre-mRNA splicing (Fleckner et al., 1997; Shen et al., 2008, 2007), Ddx39a also acts in mRNA nuclear export (Luo et al., 2001), cytoplasmic mRNA localization (Meignin and Davis, 2008) and maintenance of genome integrity (Yoo and Chung, 2011). However, how Ddx39a functions during embryogenesis, and which RNAs are processed by Ddx39a in different developmental scenarios, remain open to investigation.

In the current study, we examined the function of ddx39a in vertebrate development using a newly identified zebrafish ddx39ab gene-trap line. Transcriptome profiling and phenotypic analysis of the mutant showed that ddx39ab is indispensable for the development of heart, trunk muscle and eyes. Further experiments revealed that Ddx39ab can bind to mRNAs encoding a set of epigenetic regulatory factors, including members of the KMT2 family. Loss of ddx39ab results in aberrant pre-mRNA splicing of these epigenetic regulator transcripts, leading to failure in establishment of the proper epigenetic status of multiple structural genes, and eventually hampers the terminal differentiation of cardiomyocyte, myoblast and lens fiber cell lineages.

In a Tol2 transposon-mediated gene-trapping screen to identify novel genes involved in cardiovascular system development (Hou et al., 2017), we identified a zebrafish line, RT-011, in which embryos demonstrated a dynamic GFP expression pattern, with strong signal evident in somites and eyes from the 8-somite stage [13 hours post-fertilization (hpf)]. As development proceeded, GFP signal also emerged in the heart (Fig. 1A). Multiple incrosses of RT-011 heterozygotes yielded wild-type, heterozygous, and homozygous mutant embryos at the expected Mendelian ratios, with all homozygous mutant embryos dying by 4 days post-fertilization (dpf). This indicated that the RT-011 trap line carried a recessive lethal allele.

Homozygous mutant embryos from incrosses of heterozygous RT-011 fish showed no obvious morphological defects until 24 hpf (Fig. 1B), at which point contraction of the definitive heart tube was extremely weak and irregular in mutants (Movie 1). Homozygous mutant embryos were completely paralyzed at 24 hpf, and did not exhibit spontaneous tail movements or any response to tactile stimulation (Movie 2). At later stages of development, further defects became apparent, including a failure to establish blood circulation, a curved body axis, prominent cardiac edema, disorganized myotome and extensive cell death (Fig. 1B, Fig. S1).

5′ RACE was used to identify the gene trapped in the RT-011 line. Sequencing indicated that the gene-trapping element was integrated within the second intron of the ddx39ab locus (Fig. 1C). Zebrafish Ddx39ab protein contains DEXDc (for ATP binding and hydrolysis) and HELICc (helicase superfamily C-terminal) domains, both of which are highly conserved (over 90% amino acid identity) between zebrafish and human (Fig. S2). The gene-trap insertion resulted in a transcript that encodes a fusion protein containing the first N-terminal 69 amino acids of Ddx39ab (Fig. 1D). Since this fusion protein lacked both the DEXDc and HELICc domains, the allele that we identified from the RT-011 line should act as a true null allele. RT-PCR analysis clearly showed the presence of the ddx39ab-GFP fusion transcript in heterozygous RT-011 embryos and the absence of wild-type ddx39ab transcript in homozygous embryos (Fig. 1E).

To confirm that mutation of ddx39ab represents the causal event in the homozygous RT-011 phenotype, the expression level of ddx39ab, ddx39aa and ddx39b was examined, and only expression of ddx39ab was found to be absent in homozygous gene-trap embryos (Fig. 1F). Furthermore, when mRNA encoding wild-type Ddx39ab was injected into embryos from incrosses of heterozygous RT-011 fish, the morphological defects of the homozygous RT-011 embryos were efficiently rescued (Fig. 1G,H), with injected mutant embryos surviving up to 9 dpf. Taken together, these findings indicated that the developmental abnormalities of RT-011 homozygous embryos result from mutation of ddx39ab.

To further confirm the endogenous embryonic expression of ddx39ab, whole-mount RNA in situ hybridization was carried out on wild-type embryos. ddx39ab was strongly expressed during early embryogenesis, with abundant transcript localized to the myotome, heart and eyes. As development proceeded, ddx39ab expression became restricted to several tissues, including the pharyngeal arches and liver (Fig. S3). The dynamic expression pattern of ddx39ab indicated possible roles in multiple developmental stages and tissues.

Based on the expression pattern during embryogenesis and defects displayed in ddx39ab homozygous mutants, we examined development of the heart, skeletal muscle and lens in ddx39ab mutant embryos.

At 36 hpf, the heart tube in wild-type embryos had undergone looping and chamber ballooning, becoming a functional two-chamber pump. By contrast, hearts of ddx39ab mutants were morphologically abnormal, with little looping or chamber emergence evident (Fig. 2A, left). Immunostaining revealed that the cardiac sarcomere was severely disorganized in ddx39ab mutants as compared with control embryos (Fig. 2A, middle and right).

Gene expression analysis of a number of cardiac markers at 26 hpf demonstrated that loss of ddx39ab did not cause a readily apparent decrease in the expression of many cardiogenic regulatory genes, including nkx2.5, gata5, tbx5a and bmp4 (Fig. 2B; data not shown). Specification of both heart chambers occurred properly in ddx39ab mutants, as shown by expression of myh6 and myh7 (Fig. 2C). By contrast, ddx39ab mutant embryos demonstrated significantly reduced expression of nppa, a gene associated with maturation of the heart tube (Auman et al., 2007). Expression of a number of cardiac sarcomere structural genes was also distinctly downregulated in ddx39ab mutant embryos, including myh6, cmlc1, acta1b and ttn.2 (Fig. 2C). Decreased expression of these cardiac sarcomere components might represent the causal factor for the weak contractility observed in ddx39ab mutant hearts.

Next, we examined whether the locomotion defect in ddx39ab mutant embryos reflected altered skeletal muscle organization. Immunohistochemical staining for myosin heavy chain (MF20 antibody) and F-actin revealed that myofibrillar protein assembly was severely disrupted in ddx39ab mutant embryos when compared with wild type (Fig. 3A). To determine which step of muscle development was affected, ddx39ab mutant embryos were analyzed for myoblast differentiation by RNA in situ hybridization for genes encoding myogenic regulatory factors (Bentzinger et al., 2012). The expression of early myogenic specification markers (myod1 and myf5) and late differentiation markers (myogenin and myf6) appeared normal until 32 hpf (Fig. 3B; data not shown). Similar to the myocardium, the expression of a battery of sarcomeric components, including tnnt2d, myhz2 and smyhc1, was significantly reduced in the ddx39ab mutant embryos at 32 hpf (Fig. 3C,D). This indicated that maturation of both fast muscle and slow muscle was compromised in ddx39ab mutants. We also found reduction in transcript levels of casq1a, which encodes a calcium-binding protein of the skeletal muscle sarcoplasmic reticulum (Yazaki et al., 1990), and slc25a4, which encodes a muscle cell-specific mitochondrial ATP-ADP carrier (Gutiérrez-Aguilar and Baines, 2013) (Fig. 3C). These results suggested that ddx39ab plays an important role in skeletal muscle maturation and function.

As ddx39ab demonstrated strong expression in developing eyes, we next examined the development of retina and lens in ddx39ab mutants. Sectioning of retinas at 32 hpf revealed that pigmented epithelium formed and retinal neuroepithelium displayed normal histological features in ddx3ab mutants. Further analysis revealed that retinal neuron production (marked by atoh7 expression) and ganglion cell differentiation (marked by lhx3 and Alcama expression) were initiated normally in ddx39ab mutant retina (Fig. S4). These data indicated that development of the retina was largely normal. At 28 hpf, the lens of ddx39ab mutant embryos displayed no obvious defects at the level of gross morphology, with cell number being comparable in ddx39ab mutant and wild-type embryos (Fig. 4A).

During eye morphogenesis, cells in the center of the lens mass retain spheroidal morphology, differentiating as primary fibers (which originate from the central lens placode) that elongate and wrap around the ovoid-shaped cells in the center, resulting in crescent-shaped layers of fibers (Greiling and Clark, 2009). In contrast to wild type, in ddx39ab mutants a majority of lens fiber cells did not form crescent-shaped layers, instead showing a relatively irregular and convex shape with significantly higher convexity (Fig. 4A, bottom; Fig. S5). We next analyzed the expression of lens-specific genes in order to explore the nature of the disorganization of primary fiber cells in ddx39ab mutant embryos. We first examined the expression of a cascade of upstream transcription factors that drive fiber cell differentiation, including prox1a, foxe3 and pitx3 (Cvekl and Duncan, 2007; Greiling and Clark, 2012; Pillai-Kastoori et al., 2015). In ddx39ab mutants, the transcription of foxe3 was mildly upregulated, whereas no overt changes in pitx3 and prox1a expression were observed (Fig. 4B). During lens fiber cell elongation, soluble proteins known as crystallins are abundantly expressed in lens fibers to increase the refractive index and contribute to transparency (Clark, 2004). We found that the expression of crystallin genes, including cryaa, crygm2d10 and crygm2d1, was dramatically downregulated in ddx39ab mutant embryos (Fig. 4C). Differentiated lens fiber cells express a set of cell-cell adhesion molecules required for refractive index matching of lens membranes and cytoplasm (Bassnett et al., 2011). In situ hybridization and qPCR demonstrated that in ddx39ab mutant embryos the expression of lim2.4, a lens-specific receptor for Calmodulin involved in cell junction organization, was notably downregulated (Fig. 4C). Bfsp1 (also known as Filensin) and Bfsp2 (also known as Phakinin) are assembly partners of the beaded-chain filament, a type of lens-specific cytoskeletal element. Significant downregulation of these two genes in ddx39ab mutant lens was also observed (Fig. 4C). These data indicated that ddx39ab is also indispensable for lens fiber cell terminal differentiation.

The above results indicated that ddx39ab deletion causes dysregulated expression of structural genes during terminal differentiation of cardiomyocyte, myocyte and lens fiber cells. This hampered cell differentiation in turn leads to multiple defects in the muscular organs and eyes that are likely to contribute collectively to the lethal phenotype of ddx39ab mutant zebrafish embryos.

DEAD-box RNA helicases regulate multiple facets of RNA metabolism (Linder and Jankowsky, 2011). In order to determine how mutation of ddx39ab affects the zebrafish embryo transcriptome, we performed RNA-seq on wild-type and ddx39ab mutant embryos at 24 hpf. A total of 878 genes, consisting of 548 with decreased and 330 with increased expression, were significantly altered in the ddx39ab mutant embryos [fold change (FC)>2, false discovery rate (FDR)<0.05] (Fig. 5A; Table S3). Following gene ontology (GO) analysis, we found that genes downregulated in ddx39ab mutants were enriched for GO terms linked to development of muscular tissue and lens (Fig. 5B). Interestingly, consistent with previous observations revealed by in situ hybridization, in all three cell types (cardiomyocyte, myocyte and lens fiber cells), the expression of upstream regulatory factors showed no evidence of change, whereas the expression of many structural constituents was greatly diminished (highlighted in Fig. 5B). These results further suggested that ddx39ab plays a common and important role in the terminal differentiation of cardiomyocyte, myocyte and lens fiber cells.

To gain further insight into the impact on transcript levels caused by loss of ddx39ab, we examined alternative splicing using the OLego program and the Quantas pipeline. 12,236 significant alternative splicing events (|dI|>0.1; Table S4) were scored in ddx39ab mutant embryos compared with controls. These events include skipping/inclusion of single or tandem cassette exons, intron retention and the use of alternative 5′ or 3′ splice sites (Table S4; number of events shown in Fig. 5C). These data showed that loss of Ddx39ab leads to extensive intron retention events and cassette skipping, implicating Ddx39ab functions in intron definition. The mixed splicing activities (skipping and inclusion) observed in ddx39ab mutants clearly suggest that context-dependent splicing defects result from the loss of Ddx39ab.

We proposed that defining the RNA interactome of Ddx39ab would reveal insights into the molecular mechanisms underlying the phenotypes of ddx39ab mutant embryos. To systematically identify Ddx39ab-associated RNAs, RNA immunoprecipitation sequencing (RIP-seq) was performed in ddx39ab mutant embryos injected with mRNA encoding Flag-tagged Ddx39ab (Flag-Ddx39ab). Sequencing results showed that Ddx39ab interacts with a diverse set of RNAs (Table S6), of which mRNAs were highly represented (73.1%), with snoRNAs and rRNA contributing 16.8% and 4.5%, respectively, of bound transcripts (Fig. 5D). Comparison of data from RIP-seq and RNA-seq revealed that 84% of Ddx39ab-associated mRNAs showed alternative splicing in ddx39ab mutants (which we refer to as high-confidence Ddx39ab target transcripts), underscoring the function of Ddx39ab in pre-mRNA splicing. GO term and pathway analysis linked these Ddx39ab-associating mRNAs to histone modification. Among these potential Ddx39ab target mRNAs, members of the KMT2 gene family (kmt2a, kmt2ba, kmt2bb, kmt2ca and kmt2cb) were prominent. KMT2 family members methylate lysine 4 on the histone H3 tail, a crucial regulatory step in myogenesis associated with the modulation of chromatin structure and DNA accessibility (Lee et al., 2013; Rao and Dou, 2015).

To confirm interactions between Ddx39ab and KMT2 family member mRNAs, RNA immunoprecipitation and quantitative reverse-transcription PCR (RIP-qPCR) was applied. This further verified that Ddx39ab could bind to these mRNAs (Fig. 5E). We examined splicing events of KMT2 family genes in control and ddx39ab mutant embryos by RT-PCR analysis (Ríos et al., 2011; Rösel et al., 2011). This showed that unspliced mRNAs were retained at higher levels in ddx39ab mutants versus controls at 24 hpf (Fig. 6A), suggesting that the pre-mRNA splicing of these genes was defective. As a control, we confirmed that in ddx39ab mutants the splicing of the housekeeping gene actb1 was normal. These results suggest that in ddx39ab mutants the effect on pre-mRNA splicing might be specific to a certain set of genes.

KMT2s are the major histone methyltransferases responsible for mono-methylation at lysine 4 of histone H3 (H3K4me1) at distal enhancers and regions flanking the transcription start site (TSS) (Rao and Dou, 2015). To investigate whether loss of ddx39ab affects the presence of H3K4me1 on actively transcribed genes, ChIP-qPCR was performed for a selected set of genes that showed altered transcript levels in ddx39ab mutant versus control embryos at 24 hpf. Whereas global H3K4me1 levels in ddx39ab mutants showed only minor decreases compared with wild type (Fig. S6), the H3K4me1 occupancy at regions flanking the TSS of myocyte- and cardiomyocyte-specific genes (acta1b, myhz2, nppa, myom1a, tnnt2d, mylpfb and smhyc1) was significantly reduced in the mutants (Fig. 6B). Interestingly, the H3K4me1 occupancy level in the TSS region of myogenic regulatory factors (such as myod1) was comparable in control and ddx39ab mutant embryos, consistent with the analysis of transcript and protein levels by RNA-seq, in situ hybridization and western blot (Fig. 6B, Fig. S6).

These results support a context-dependent requirement of Ddx39ab for proper pre-mRNA splicing of KMT2 family members, with loss of ddx39ab leading to failure in establishment of the epigenetic status required for terminal differentiation.

Members of the DEAD-box RNA helicase family have been shown to be involved in nearly all aspects of RNA metabolism, from transcription to mRNA decay. Recent studies have started to delineate different functions of DEAD-box RNA helicases in a broader context, including animal development. For example, ddx46 is expressed in the digestive organs and brain and has been shown to be required for the development of these organs (Hozumi et al., 2012). Researchers have also reported that ddx18 is essential for hematopoiesis (Payne et al., 2011). The generation of the ddx39ab gene-trapping allele provided the opportunity to study the function of this gene in vertebrate development. Here we demonstrated that ddx39ab is required for normal gene expression and differentiation of cardiomyocyte, myocyte and lens fiber cells. In previous studies, the development and function of the heart was shown to be sensitive to defects in RNA metabolism (Ding et al., 2004; Xu et al., 2005). Our observations corroborate these results and suggest that myocyte and lens fiber differentiation similarly bear this cell type-specific susceptibility. One possible explanation for this observation is that these three cell types exploit a similar, KMT2 family-dependent mechanism to establish epigenetic status during differentiation. An alternative hypothesis is that, compared with other cell types (such as neurons), a larger fraction of splicing undergoes Ddx39ab-related regulation in these three types of cell. FACS followed by RIP-seq might identify cell type-specific Ddx39ab-associated transcripts, and this information might help to clarify the underlying mechanisms of cell type-specific susceptibility.

RNA-binding proteins (RBPs) including DEAD-box RNA helicases modulate splicing primarily by positively or negatively regulating splice site recognition by the spliceosome. Recognition by RBPs relies on distinct regulatory sequences in pre-mRNAs that function as splicing enhancers or silencers. Our data showed that Ddx39ab can bind to a battery of mRNAs, although more sophisticated molecular biology and bioinformatics efforts need to be exploited to unfold the detailed mechanism of Ddx39ab-mediated splicing events, such as how the specificity is defined.

During development, ddx39ab shows a complex and dynamic expression pattern (Fig. 1, Fig. S3). This prompts future analysis of the role of ddx39ab in later developmental events (e.g. during formation of the pharyngeal arches and the development of digestive organs). However, the severe defects observed in ddx39ab mutants at early stages of development complicates the analysis of later developmental events. CRIPSR/Cas9-based generation of tissue-specific ddx39ab mutants, or generation of a conditional ddx39ab allele, will be required to study ddx39ab function in other organs/tissues.

Previous studies showed that Ddx39ab acts as a growth-associated factor in cancer cells that is required for genome integrity and telomere protection (Sugiura et al., 2007; Yoo and Chung, 2011). We did not detect abnormal mitoses in ddx39ab mutant embryos (Fig. S7). However, analysis of later stage ddx39ab mutants might be required to observe telomeric defects, after a sufficient number of cell divisions have occurred. It might also be the case that the telomere protection function of Ddx39ab is not conserved between fish and mammals.

It was interesting to note that among a relatively small number of genes significantly downregulated in ddx39ab mutant embryos, a large proportion of them encoded structural constituents, including sarcomeric components in muscle cells and crystallin genes in lens fiber cells. In addition to regulating mRNA splicing, it is possible that Ddx39ab is involved in transactivation of structural components via an uncharacterized mechanism that is exploited in both muscular and lens fiber cells. Several studies have shown that DEAD-box proteins play important roles as regulators of transcription, particularly as co-activators or co-suppressors of transcription (Fuller-Pace and Nicol, 2012; Huang et al., 2015). Further investigation to determine which proteins Ddx39ab directly binds to in various cell types, and the functional consequences of these interactions, should provide important insight into how the specificity of Ddx39ab function is regulated.

Zebrafish were maintained and handled in accordance with approved guidelines of the Institutional Animal Care and Use Committee of Nanjing University and as per Canadian Council on Animal Care and Hospital for Sick Children Laboratory Animal Services guidelines.

Zebrafish embryos were maintained and staged using standard techniques (Westerfield, 1993). The RP-T gene-trap vector was modified from RP2 (Clark et al., 2011) by switching the monomeric RFP to monomeric GFP. RP-T plasmid (25 ng/μl) and Tol2 transposase mRNA (50 ng/μl) were injected (1 nl each) into 1-cell stage embryos as described (Hou et al., 2017). To identify the affected gene in the gene-trap lines, inverse PCR and 5′ RACE were performed as described (Clark et al., 2011). The gene-trapping insertion position in the ddx39ab locus was determined by sequencing. For PCR genotyping of single embryos, primers were designed to differentially amplify the wild-type allele or gene-trap allele. After immunostaining or in situ hybridization, genotyping PCR was applied as previously described (Kawakami et al., 2016). Primer pairs and detailed PCR conditions for genotyping are listed in Table S1.

Total RNA was prepared using TRIzol (Invitrogen, Life Technologies), with DNase-treated RNA reverse transcribed using random 16-mer priming and SuperScript II reverse transcriptase (Life Technologies). PCR was performed with primers specifically amplifying cDNA from ddx39ab mRNA or fusion transcript. Primer pairs and detailed PCR conditions are listed in Table S1.

Embryos from three different clutches were collected as biological replicates and total RNA was prepared using TRIzol. RNA was reverse transcribed with a mix of oligo(dT) and random 16-mer priming and SuperScript II reverse transcriptase. Quantitative PCR (qPCR) assays were performed in triplicate with SYBR Green Master Mix (Takara) according to the manufacturer's instructions. Melting curves were examined to ensure primer specificity. Results were analyzed using the standard ΔΔCT method (Schmittgen and Livak, 2008). 18S rRNA served as the reference gene in all analyses, and changes in mRNA levels relative to 18S rRNA were confirmed using actb2 as an alternate reference gene in independent experiments. Primers used for qPCR analysis are listed in Table S1.

Two-sided, paired Student's t-tests were applied for RT-qPCR and ChIP-qPCR results; two-sided, unpaired Student's t-tests were applied for other quantification experiments. Significance of differences was calculated with GraphPad Prism6. P<0.05 was considered significant.

pCS2+ vectors carrying a cDNA fragment encoding membrane RFP (mRFP), Ddx39ab, Flag-Ddx39ab and Tol2 transposase were used. Capped mRNA was synthesized using the SP6 mMESSAGE mMACHINE Kit (Ambion, Life Technologies). For phenotype rescue experiments, ddx39ab mRNA (100 pg) was injected at the 1-cell stage.

Embryos were fixed in 4% paraformaldehyde at 4°C overnight. Embryos were blocked with blocking solution (1× PBS, 1% BSA, 1% Triton X-100, 0.1% DMSO) for 2 h then incubated with primary antibodies diluted in blocking solution overnight at 4°C. Embryos were washed three times in 1× PBS with 1% Triton X-100 for 15 min each. Embryos were incubated with secondary antibodies diluted in blocking solution for 2 h. Primary antibodies specific to Myh1a, Myh1e, Cardiac troponin T (cTnT; Tnnt2), and acetylated Tubulin were used. Fluorescent immunocytochemistry was performed using anti-mouse antibody conjugated with actin filaments, as visualized with Rhodamine-conjugated phalloidin. Unless otherwise stated, manipulations were performed at room temperature. Detailed information on antibodies and dilutions is provided in Table S2.

Embryo whole-mount imaging was performed using a Leica DFC320 camera on a Leica M205FA stereomicroscope. Confocal images were taken using a Zeiss LSM880 confocal microscope.

RNA in situ hybridization using DIG-labeled antisense RNA probes was carried out as previously described (Thisse and Thisse, 2008).

Western blotting analysis of myocyte or cardiomyocyte transcriptional regulators and H3K4Me1 in control and ddx39ab mutant embryos at different stages was performed as described in the supplementary Materials and Methods using the primary antibodies listed in Table S2.

To quantify lens equatorial width, bright-field images from live embryos were taken with a Nikon ECLIPSE Ni microscope and measured with the Nikon Elements BR measurement tool. At least 20 embryos of each genotype were measured under identical conditions. To quantify the number and convexity of lens fiber cells, mRFP mRNA was injected into embryos at the 1-cell stage and at 28 hpf, and confocal images for equatorial sections from control and ddx39ab mutant embryos were taken. Cell numbers on optical sections were counted manually. Cell convexity was calculated with the 3D Convex Hull plug-in of ImageJ (Sheets et al., 2011). Convexity was defined as the ratio between convex surface area and total surface area.

Total RNA from 100 control and ddx39ab mutant 24 hpf embryos was isolated using TRIzol reagent. Two biological replicates for each group (control and ddx39ab mutant) were processed and sequenced. Sequencing libraries were prepared using the Nextera sample preparation kit (Illumina) and subjected to HiSeq paired-end 100 bp plus sequencing. Resulting reads were aligned to the zebrafish reference genome (GRCz10) and gene expression quantified using TopHat V2.2.1 and Bowtie2 v2.2.3 (Kim et al., 2013; Langmead and Salzberg, 2012). Differential gene expression was analyzed using HTSeq v0.6.1p1 (Anders et al., 2015). Genes showing altered expression with adjusted P<0.05 were considered differentially expressed. For the set of differentially expressed genes a functional analysis was performed using Ingenuity Pathway Analysis software and DAVID (Huang et al., 2009), and some of the enriched processes were selected according to relevant criteria related to the biological process studied. Using the R visualization package GOPlot (Walter et al., 2015), a chord plot was generated to better visualize the relationships between genes and the selected enriched processes. OLego (Wu et al., 2013) and Quantas pipelines were used for alternative splicing analysis. Transcript structure was inferred between paired-end reads. Alternative splicing was quantified by separating genomic and junction reads and scoring the output from transcript inference. Finally, statistical tests were run to filter the significant alternative splicing events (Fisher's exact test and Benjamini FDR).

RNA immunoprecipitation (RIP) was performed as previously described (Jain et al., 2011) with the specific modifications detailed below. Flag-Ddx39ab mRNA was injected into embryos from ddx39ab heterozygous in-crosses, with mutant embryos sorted based on GFP brightness. Deyolked embryos were homogenized in RIP buffer and briefly sonicated using the probe tip of a Branson sonicator to solubilize chromatin. Each sample was normalized for total protein amount and then Flag-Ddx39ab and associated RNA was isolated by incubation with anti-Flag agarose beads (Sigma) for 6 h at 4°C with gentle rotation. Samples were washed sequentially in high-stringency buffer, high-salt buffer and RIP buffer. Ddx39ab-associated RNA was extracted with Trizol and processed for sequencing. Sequencing libraries were prepared using the Nextera sample preparation kit and subjected to HiSeq paired-end 100 bp plus sequencing. Data analysis was performed as above.

Total RNA was prepared from control and ddx39ab mutant larvae at 36 hpf and RT-PCR performed to monitor the splicing of pre-mRNAs. Primer pairs and the PCR conditions used to amplify each of the genes are listed in Table S1.

Chromatin immunoprecipitation (ChIP) assays were performed as described (Lindeman et al., 2009). In brief, at 24 hpf, ddx39ab mutant and control embryos were collected, deyolked and cross-linked with 1% formaldehyde for 10 min at room temperature and subsequently quenched with glycine to a final concentration of 0.125 M for another 10 min. Chromatin was sonicated with a Bioruptor (Diagenode), cleared by centrifugation (10 min, 4°C, 8000 g), and incubated overnight at 4°C with 5 mg anti-H3K4me1 antibody (Abcam). Immunocomplexes were immobilized with 100 μl protein-G Dynal magnetic beads (Abcam) for 4 h at 4°C, followed by stringent washes and elution. Eluates were subject to reversal of cross-links overnight at 65°C and deproteinated. DNA was extracted with phenol chloroform, followed by ethanol precipitation. H3K4me1-occupied regions at 24 hpf were retrieved from a previously reported data set (Bogdanovic et al., 2012). ChIP-qPCR analyses were performed using a Light Cycler 480II (Roche). ChIP-qPCR signals were calculated as percentage of input. Primers used in qPCR analyses are shown in Table S1.

We thank Qi Xiao for drawing schematic pictures, Peipei Yin for zebrafish husbandry and members of the I.C.S. lab at the University of Toronto and the Di Chen lab at Nanjing University for feedback and help during this project.