RNA-binding protein Elavl1/HuR is required for maintenance of cranial neural crest specification

While neural crest development is known to be transcriptionally controlled via sequential activation of gene regulatory networks (GRNs), recent evidence increasingly implicates a role for post-transcriptional regulation in modulating the output of these regulatory circuits. Using available single-cell RNA-sequencing datasets from avian embryos to identify potential post-transcriptional regulators, we found that Elavl1, which encodes for an RNA-binding protein with roles in transcript stability, was enriched in the premigratory cranial neural crest. Perturbation of Elavl1 resulted in premature neural crest delamination from the neural tube as well as significant reduction in transcripts associated with the neural crest specification GRN, phenotypes that are also observed with downregulation of the canonical Wnt inhibitor Draxin. That Draxin is the primary target for stabilization by Elavl1 during cranial neural crest specification was shown by RNA-sequencing, RNA immunoprecipitation, RNA decay measurement, and proximity ligation assays, further supporting the idea that the downregulation of neural crest specifier expression upon Elavl1 knockdown was largely due to loss of Draxin. Importantly, exogenous Draxin rescued cranial neural crest specification defects observed with Elavl1 knockdown. Thus, Elavl1 plays a critical a role in the maintenance of cranial neural crest specification via Draxin mRNA stabilization. Together, these data highlight an important intersection of post-transcriptional regulation with modulation of the neural crest specification GRN.


Introduction
Neural crest cells are an essential, multipotent cell population in the vertebrate embryo. During development, these cells undergo coordinated induction, specification, and epithelial-mesenchymal transition (EMT) events to migrate and ultimately form a myriad of tissues, including craniofacial structures, components of the peripheral nervous system, as well as many other derivatives (Gandhi and Bronner, 2018). The transcriptional control of these events has been dissected and mapped into modules of a feed-forward gene regulatory network (GRN), which helps explain the detailed sequence of events involved in neural crest development (Martik and Bronner, 2017;Simões-Costa and Bronner, 2015a;Williams et al., 2019). Recently, in addition to transcriptional events, there has been growing appreciation for the role that post-transcriptional regulation plays in the establishment, maintenance, and regulation of neural crest formation (Bhattacharya et al., 2018;Cibi et al., 2019;Copeland and Simoes-Costa, 2020;Forman et al., 2021;Sánchez-Vásquez et al., 2019;Ward et al., 2018;Weiner, 2018).
Given that RNA-binding proteins play an essential role in post-transcriptional regulatory processes (Dassi, 2017), we sought to broadly identify those with early roles in neural crest development. To this end, we analyzed existing single-cell RNA-sequencing (scRNA-seq) data (Williams et al., 2019) from specification-stage avian embryos to identify enriched RNA-binding protein candidates. Using this approach, we identified Elavl1 as an enriched transcript in newly formed neural crest cells. Elavl1 is a nucleocytoplasmic shuttling protein from the ELAV (embryonic lethal abnormal vision) family of RNA-binding proteins, which have conserved roles in neural development (Ma et al., 1996;Yao et al., 1993). It is a well-established stabilizer of mRNA, a function often mediated via association with the 3′-untranslated region (3′-UTR) of its mRNA targets (Abdelmohsen and Gorospe, 2010;Rothamel et al., 2021).
Elavl1 is essential for mammalian development and embryonic survival; Elavl1 null mouse embryos exhibit lethality due to abnormal placental morphogenesis, and conditional epiblast-null embryos display a broad array of phenotypes, ranging from defects in ossification and craniofacial development to asplenia. Interestingly, despite the myriad of tissue systems affected by Elavl1 knockout and relatively broad expression in wild-type embryos, mechanistic insights suggest Elavl1 acts on specific gene networks in a spatiotemporally controlled manner (Katsanou et al., 2009). Thus, due to its complexity and specificity of function, much remains to be discovered with eLife digest As an embryo develops, different genetic programs become activated to give cell populations a specific biological identity that will shape their fate. For instance, when certain sets of genes get switched on, cells from the outermost layer of the embryo start to migrate to their final destination within the body. There, these 'neural crest cells' will contribute to bones and cartilage in the face, pigmented skin spots, and muscles or nerves in the gut.
When genes responsible for the neural crest identity are active, their instructions are copied into an 'RNA molecule' which will then relay this information to protein-building structures. How well the RNA can pass on the message depends on how long it persists within the cell. Certain RNA-binding proteins can control this process, but it is unclear whether and how this regulation takes place in neural crest cells. In their work, Hutchins et al. therefore focused on identifying RNA-binding proteins involved in neural crest identity.
Exploratory searches of genetic data from chick embryos revealed that, even before they started to migrate, neural crest cells which have recently acquired their identity produced large amounts of the RNA-binding protein Elavl1. In addition, these cells did not behave normally when embryos were deprived of the protein: they left the outer layer too soon and then switched off genes important for their identity. Genetic studies of neural crest cells lacking Elavl1 revealed that this effect was due to having lost the RNA molecule produced from the Draxin gene.
Introducing an additional source of Draxin into mutant embryos missing Elavl1 was enough to restore normal neural crest behaviour. Further biochemical experiments then showed that the RNA for Draxin decayed quickly in the absence of Elavl1. This suggests that the protein normally allows Draxin's RNA to persist long enough to pass on its message.
These results reveal a new mechanism controlling the identity and behaviour of the neural crest. Since many cancers in adulthood arise from the descendants of neural crest cells, Hutchins et al. hope that this knowledge could lead to improved therapies in the future. respect to Elavl1's essential roles and targets during embryonic development across tissue-specific contexts.
Here, we sought to determine the role of Elavl1 during cranial neural crest specification by taking advantage of the chick embryo model, an amniote system in which it is possible to perturb Elavl1 function with precise spatiotemporal control via unilateral knockdown. The results show that perturbation of Elavl1 led to premature neural crest delamination, as well as significant reduction in the expression of genes within the neural crest specification GRN. We find that these effects were mediated by loss of Draxin, a direct mRNA target for stabilization by Elavl1. Our data demonstrate a critical role for Elavl1, and RNA-binding protein-mediated post-transcriptional control, in the regulation of a critical neural crest specification module.

Results
The RNA-binding protein Elavl1/HuR is expressed in cranial neural crest Cranial neural crest cells are indispensable for proper craniofacial development (van Limborgh et al., 1983;Vega-Lopez et al., 2018). Whereas transcription factors have been well-established critical regulators of neural crest development and craniofacial morphogenesis reviewed in Gou et al., 2015, growing evidence indicates an essential role for post-transcriptional regulation in these processes (Cibi et al., 2019;Copeland and Simoes-Costa, 2020;Dennison et al., 2021;Forman et al., 2021). To identify RNA-binding proteins with potential roles in cranial neural crest specification, we analyzed scRNA-seq data for cranial neural crest isolated from avian embryos at the 5-6 somite stage (Williams et al., 2019) we identified three distinct clusters (neural, premigratory [pNC], and delaminating/migratory [mNC]) among which genes associated with the gene ontology (GO) term 'binds to 3′-UTR' were differentially expressed ( Figure 1A-E). To isolate potential positive regulators of neural crest specification, we then performed GO term analysis for 'stabilizes RNA' and 'regulates translation' among the identified RNA-binding proteins; only the Elavl1 gene was associated with all three GO terms and abundantly expressed in the isolated cranial neural crest cells ( Figure 1F, G).
Given that Elavl1 knockout mice often display defects in craniofacial structures (Katsanou et al., 2009), and the transcript appeared enriched in premigratory cranial neural crest cells ( Figure 1E), we hypothesized a potential role for Elavl1 during cranial neural crest specification. To test this possibility, we first examined the expression pattern of Elavl1 in the developing chick embryo. Early in neurulation, when the neural plate border (NPB) is established within the rising neural folds, Elavl1 expression was detected in the anterior open neural tube and closing neural folds surrounding the anterior neuropore but absent from Pax7-expressing NPB cells ( Figure 2A). As the neural tube closed, when neural crest specification is complete, Elavl1 expression became enriched throughout the neural tube and overlapped with Pax7 expression in premigratory cranial neural crest cells ( Figure 2B-D). Following cranial neural crest EMT, Elavl1 remained expressed in the migratory neural crest cells, as well as throughout the brain and neural tube ( Figure 2E-G). Thus, Elavl1 is expressed in specified, premigratory cranial neural crest cells following establishment of the NPB and is retained during the onset of EMT and in early migrating cranial neural crest cells.

Elavl1 downregulation alters cranial neural crest specification and delamination
To determine what, if any, role Elavl1 has in cranial neural crest specification, we perturbed Elavl1 function in the early embryo using a translation-blocking antisense morpholino oligo (MO). We electroporated control or Elavl1 MOs bilaterally into gastrula stage chick embryos and analyzed neural crest specification using quantitative fluorescent hybridization chain reaction (HCR) to measure expression of markers of specified neural crest at HH9 ( Figure 3A, B). Given Elavl1's association with Wnt signaling (Kim et al., 2015) and the essential roles Wnt signaling plays during early neural crest development (Milet and Monsoro-Burq, 2012;Rabadán et al., 2016;Simões-Costa and Bronner, 2015a;Steventon and Mayor, 2012;Wu et al., 2003;Yanfeng et al., 2003), we focused on the Wnt effector Axud1, its target and neural crest specifier FoxD3, and the Wnt antagonist Draxin (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019;Simões-Costa et al., 2015b). Following Elavl1 knockdown ( Figure 3-figure supplement 1; 61.4 ± 0.9% of the control side, p < 0.001, paired t-test, n = 5 embryos, 15 sections), we observed significant reduction in the levels of Axud1 ( Figure 3C; 73.7 ± 3.1% of the control side, p = 0.03, Wilcoxon signed-rank test, n = 6 embryos), FoxD3 ( Figure 3D; 61.3 ± 5.5% of the control side, p = 0.002, Wilcoxon signed-rank test, n = 10 embryos), and Draxin ( Figure 3E; 66.6 ± 1.8% of the control side, p = 0.03, Wilcoxon signed-rank test, n = 6 embryos) transcripts compared to contralateral control sides. To parse whether this was a specific effect or a broad defect in neural crest development, we also examined additional neural crest genes and found no significant difference in expression between control or Elavl1 knockdown sides for Pax7 ( Figure 3F; 87.4 ± 3.1% of the control side, p = 0.1, Wilcoxon signed-rank test, n = 5 embryos) or Tfap2b ( Figure 3G; 98.2 ± 7.0% of the control side, p = 0.6, Wilcoxon signed-rank test, n = 4 embryos), indicating a specific defect in a subset of genes required for cranial neural crest specification.
We also performed immunostaining for Pax7 in cross-section to assess if Elavl1 knockdown altered neural crest cell number or dorsal neural tube morphology. Interestingly, the total number of Pax7 + cells was unaffected with Elavl1 knockdown (101.3 ± 4.5% of the control side, p = 0.6, one-sample Wilcoxon signed-rank test, n = 4 embryos, 11 sections); however, we found a significant increase in the number of Pax7 + cells that delaminated from the neural tube (139.4 ± 8.4% of the control side, p = 0.002, one-sample Wilcoxon signed-rank test), and concomitant decrease in the number of Pax7 + cells retained within the dorsal neural tube (66.0 ± 5.3% of the control side, p = 0.001, one-sample Wilcoxon signed-rank test: Figure 3H, I). To determine what impact premature delamination might have on cranial neural crest migration, we also performed whole mount immunostaining for Pax7 at HH9 + and found significant decrease in cranial neural crest emigration away from the midline (Figure 3-figure supplement 2; 63.6 ± 4.1% of the control side, p < 0.001, Wilcoxon matched-pairs signed-rank test, n = 5 embryos, 5 measurements per embryo averaged). Taken together, these data suggest that Elavl1 is required during early cranial neural crest development to regulate specification and prevent premature delamination.
Draxin is the primary target of Elavl1 during cranial neural crest specification To parse the mechanism of Elavl1 function during cranial neural crest specification, we sought to more broadly identify the RNA targets of Elavl1 using bulk RNA-sequencing (RNA-seq) and differential gene expression analysis ( Figure 4A). Given that Elavl1 is known to bind to and stabilize its RNA targets via 3′-UTR interaction (Chen et al., 2002;Dormoy-Raclet et al., 2007;Katsanou et al., 2009;Rothamel et al., 2021;Shi et al., 2020), we expected expression of potential targets to be reduced with Elavl1 knockdown. Among the differentially expressed genes we identified (Figure 4-figure supplement 1), 12 were significantly downregulated, with four having established roles in neural crest development and which we validated using HCR-Axud1, Draxin, BMP4, and Msx1 ( Figure 4B, Figure 4-figure supplement 1). We also identified several canonical neural crest genes that were unaffected with Elavl1 knockdown (e.g., Pax7, Tfap2b, Snai2, Sox9, and Zeb2), consistent with HCR data (Figure 3, Figure 4-figure supplement 1). Together these data suggest Elavl1 does not broadly bind and stabilize the transcripts of neural crest genes, rather it targets specific RNAs to drive cranial neural crest specification.
Notably, FoxD3 failed to meet our stringency cutoff during RNA-seq analysis due to low expression at the examined stages, and is likely downregulated with Elavl1 knockdown ( Figure 3D) due to indirect effects from loss of Axud1 (Simões-Costa et al., 2015b) and therefore unlikely to be a bona fide target of Elavl1. To determine whether Elavl1 directly or indirectly interacts with Axud1, Draxin, BMP4, and Msx1 mRNAs, we first measured the rate of mRNA decay with actinomycin D treatment to assess the stability of these RNAs, and Pax7 as a nontarget, with or without Elavl1 knockdown identified three distinct clusters neural, premigratory (pNC), and delaminating/migratory (mNC). Expression of marker genes (Sox3 and Cdh2 for neural, Pax7 and Snai2 for premigratory cranial neural crest, and Sox10 and Ets1 for early migratory cranial neural crest) was used to label the three subclusters. (D) The majority of the FoxD3-NC1 + cells were also positive for the expression of transcription factors Pax7 and Snai2, which label the dorsal neural tube and premigratory/delaminating neural crest cells. These triple positive NC1 + /Snai2 + /Pax7 + were further processed for gene ontology analysis to identify post-transcriptional regulators. (E) A strip-plot showing expression and abundance of a subset of genes that are associated with the gene ontology term 'binds to 3′-UTR'. (F) A three-way Venn diagram shows overlap between genes associated with the gene ontology terms 'binds RNA', 'regulates translation', and 'stabilizes RNA'. Only three genes, Elavl1, Dazl, and Igf2bp1, were associated with all three. (G) Feature plots showing the expression distribution of the three genes identified in (F). Only Elavl1 is abundant among all NC1 + /Pax7 + /Snai2 + cells.  Figure 4C); if Elavl1 is required for transcript stability, target RNAs should decay at a faster rate with loss of Elavl1 compared to control. Interestingly, among all the transcripts tested, only Draxin had a significant reduction in transcript stability (p = 0.001, Mann-Whitney test; Figure 4D), suggesting that Draxin is the primary target bound and stabilized by Elavl1 during cranial neural crest specification.
To test this hypothesis, we searched the 3′-UTR sequences of Draxin, Axud1, BMP4, Msx1, Pax7, and Tfap2b mRNAs for putative Elavl1-binding sites. The Draxin 3′-UTR contained four high probabilitybinding sites, whereas the other 3′-UTRs contained only one (BMP4, Msx1) or none (Axud1, Pax7, Tfap2b; Figure 4-figure supplement 2). Given that Elavl1 contains three RNA recognition motifs (RRMs) that cooperate for RNA recognition (Pabis et al., 2019), it is possible that multiple contacts are required for Elavl1 to bind and stabilize RNAs in vivo, and by extension is unlikely for Elavl1 to bind RNAs with only a single putative binding site. Thus, we hypothesize that Elavl1 specifically targets and stabilizes Draxin in cranial neural crest through multiple contact sites within its 3′-UTR.
To confirm that Draxin mRNA is bound by Elavl1 in vivo, we first performed an RNA immunoprecipitation (RIP) followed by quantitative reverse transcription-PCR (qRT-PCR) to pull down endogenous Elavl1 ribonucleoprotein (RNP) complexes. To this end, we incubated lysate generated from wild-type HH9 embryonic heads with magnetic beads coated with either Elavl1 antibody or a rabbit IgG nonspecific control antibody, then eluted bound RNA and performed qRT-PCR, comparing immunoprecipitated RNAs ('IP') with RNAs extracted from a fraction of the input lysate ('Input'). We expected that RNAs bound by Elavl1 would be enriched in the IP compared to the Input, whereas nonspecifically associated RNAs, that is nontargets, would not ( Figure 4E). We found that Pax7 and FoxD3 mRNAs (nontargets) were neither enriched in the IP, nor significantly different from each other (p = 0.89, oneway analysis of variance (ANOVA) with Tukey's post hoc test); however, Draxin mRNA was significantly enriched with Elavl1 IP compared to Pax7 and FoxD3 (p < 0.001, one-way ANOVA with Tukey's post hoc test; Figure 4F), suggesting that Elavl1 specifically associated with endogenous Draxin mRNA. However, this assay could not distinguish whether Draxin was bound directly or indirectly, or where within the transcript it might be associating with Elavl1.
To test whether Draxin was directly bound by Elavl1 within the 3′-UTR, we performed a proximity ligation assay (PLA), wherein in situ fluorescent signal can be detected as puncta only when two proteins are in close proximity (<40 nm), indicating a direct interaction in vivo. Taking advantage of the MS2-MCP reporter system (Tutucci et al., 2018), we electroporated a construct encoding a GFPtagged MS2 bacteriophage coat protein (MCP-GFP) alone ('Control') or in combination with ('Experimental') a construct containing a Luciferase coding region, MS2 stem loops (bound by MCP when transcribed), and the endogenous Draxin 3′-untranslated region (MS2-Draxin 3′-UTR) and performed PLA with antibodies against Elavl1 and GFP ( Figure 4G). We observed significantly more PLA puncta with expression of MS2-Draxin 3′-UTR ( Figure 4H-I), indicating a specific and direct interaction between Elavl1 and the Draxin 3′-UTR.

Discussion
Understanding of neural crest development has been greatly enhanced by the identification of key transcriptional circuits that control its developmental progression. Recent studies suggest a critical role for post-transcriptional regulation in the refinement of the expression outputs of these GRNs. Here, we identified and characterized Elavl1 as an RNA-binding protein essential for the maintenance of cranial neural crest specification via its stabilization of the Wnt antagonist Draxin. We found that loss of Elavl1, and by extension its target, Draxin, interferes with output from multiple nodes of the GRNs required for neural crest specification, including those driven by Wnt and BMP (Hovland et al., 2020;Simões-Costa et al., 2015b;Tribulo et al., 2003). Together, our data implicate Elavl1 as a point of integration to coordinate signaling from parallel but independent GRNs.
At a mechanistic level, our study is consistent with previous work examining Elavl1 function in other cellular contexts, with respect to its role as a stabilizing RNA-binding protein via 3′-UTR interaction (Chen et al., 2002;Dormoy-Raclet et al., 2007;Katsanou et al., 2009;Rothamel et al., 2021;Shi et al., 2020). In the context of embryonic development, our study aligns well with data from knockout mouse consistent with an important role for Elavl1 in intersecting signaling cascades; interestingly, this prior work similarly observed indirect downregulation of BMP4 with loss of Elavl1, though an upstream mediator remained unidentified (Katsanou et al., 2009). In neural crest specification, we identified a single biologically relevant target of Elavl1, whereas prior high-throughput studies found many RNAs directly bound by Elavl1 (Mukherjee et al., 2011;Rothamel et al., 2021). Whether this is a feature specific to neural crest is unclear, though the knockout mouse work suggests that during embryonic development Elavl1 function and RNA targets are driven by spatiotemporal determinants (Katsanou et al., 2009), which may be due RNP complex heterogeneity as a result of tissue-specific expression of other interacting RNA-binding proteins. Taken in the context of these prior studies, we speculate that despite broad tissue expression of Elavl1 across embryonic development, specificity in neural crest is achieved through spatiotemporally regulated combinatorial expression of post-transcriptional regulators and RNA regulons (Keene, 2007;Keene and Tenenbaum, 2002).
It is important to note that, while Elavl1 expression persists in cranial neural crest during the initiation of EMT and migration, Draxin must be rapidly downregulated for these processes to proceed Quantification of the ratio of Pax7 + cells on Elavl1 MO (right) versus control MO (left) sides of cross-sections. Data are from individual sections; sections from same embryo are displayed in same color (n = 4 embryos, 11 sections). *p ≤ 0.002, one-sample Wilcoxon signed-rank test.
The online version of this article includes the following figure supplement(s) for figure 3:      (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019;Hutchins et al., 2021). Thus, we hypothesize that Elavl1 becomes endogenously displaced from Draxin at the onset of EMT, though it is yet unclear how this is achieved. RNA-binding proteins are known to alter association with targets due to post-translational modifications such as phosphorylation or alternative RNA-binding protein competition (Dassi, 2017;García-Mauriño et al., 2017;Liu et al., 2015). Indeed, inhibition of serine-threonine kinases has been shown in neural crest to increase cell-cell adhesions and negatively impact cell migration (Monier-Gavelle and Duband, 1995), suggesting kinase-driven signaling pathway activation coincident with neural crest EMT; this is interesting given established roles for serine-threonine phosphorylation in modulating Elavl1-RNA-binding activity and target selection (Grammatikakis et al., 2017). Thus, we speculate that Elavl1 serine-threonine phosphorylation, either in combination with or as an alternative to competitive RNA-binding protein displacement, likely facilitates Draxin release and turnover at the onset of EMT.
Because its primary target during specification is downregulated while Elavl1 expression persists, this suggests there are likely additional targets and roles for Elavl1 beyond Draxin stabilization. Elavl1 has been shown in other contexts to stabilize Snail1 (Dong et al., 2007) and matrix metalloprotease-9 (MMP-9) (Yuan et al., 2011), factors with well-established roles in neural crest EMT (Cano et al., 2000;Kalev-Altman et al., 2020;Monsonego-Ornan et al., 2012;Strobl-Mazzulla and Bronner, 2012;Taneyhill et al., 2007). Given that the neural crest specification GRNs is proceeded by activation of an EMT GRN also coinciding with Elavl1 expression, we speculate that Elavl1 likely intersects with additional GRNs following specification. It is possible one or more of the downregulated genes we identified through RNA-seq that were not important for specification may yet be targets of Elavl1, with functions later in neural crest development. To fully understand how Elavl1 exerts control over key developmental processes, these possibilities will need to be explored. expressed following Elavl1 knockdown (12 upregulated, 12 downregulated), four genes (Draxin, Axud1, Msx1, BMP4) have established roles in neural crest development and were significantly downregulated. (C) Schematic diagram illustrating experimental design for RNA stability assay. Gastrula stage chick embryos were electroporated bilaterally with control or translation-blocking morpholino (MO) targeting Elavl1. Dorsal neural folds were dissected from stage HH8+/9− embryos; left neural folds were used as the 0 hr time point, whereas right neural folds were treated with actinomycin D for 30 min, 2 hr, or 4 hr prior to total RNA extraction and quantitative reverse transcription-PCR (qRT-PCR) to measure RNA decay. (D) Transcript stability plots show Draxin mRNA stability is significantly reduced (*p = 0.001, Mann-Whitney test) with Elavl1 knockdown (blue) compared to control (gray), whereas other neural crest mRNAs (Axud1, Msx1, BMP4, Pax7) are not (ns,nonsignificant,p > 0.37,. (E) Schematic illustrating experimental design of RNA-binding protein/RNA co-immunoprecipitation (RIP) to test RNA association with Elavl1 in vivo for neural crest targets. Lysates generated from HH9 heads were incubated with antibody-coated beads for Elavl1 or a nonspecific IgG to co-immunoprecipitate protein with bound RNAs. In qRT-PCR, specifically bound RNAs would be more abundant and reach threshold before RNAs that were nonspecific, and therefore would have smaller C T values. C T , threshold cycle. (F) Fold enrichment of RNAs eluted from RIP (n = 16 embryos), quantified by qRT-PCR, performed in triplicate. ns, nonsignificant, p = 0.89, one-way analysis of variance (ANOVA) with Tukey's post hoc test. *p < 0.001, one-way ANOVA with Tukey's post hoc test. Error bars, standard error of the mean (SEM). (G) Schematic diagram illustrating experimental design for proximity ligation assay (PLA). Gastrula stage chick embryos were electroporated bilaterally with a construct expressing a nuclear localized, GFP-tagged MS2 bacteriophage coat protein (MCP-GFP) alone (left) or in combination with a construct containing a Luciferase coding region, MS2 stem loops (which are bound by MCP when transcribed), and the endogenous Draxin 3′-untranslated region (MS2-Draxin 3′UTR) (right). Following fixation and cross-sectioning at HH9, tissues were incubated with primary antibodies made in goat and rabbit that recognized GFP and Elavl1, respectively. Secondary antibodies against goat and rabbit IgG were labeled with complementary oligonucleotides that generate a fluorescent signal due to rolling circle amplification only when in close proximity (<40 nm). Thus, fluorescence signal (magenta) would indicate in vivo interaction between MCP-GFP and endogenous Elavl1. (H) Representative confocal maximum intensity projection micrograph of dorsal neural folds from cross-sectioned HH9 embryo bilaterally electroporated with MCP-GFP (green) alone ('control', left)    Scale bar, 50 μm. (E) Quantitation of HCR processed embryos for control versus Elavl1 knockdown, calculated as ratio of Elavl1 MO versus control MO integrated density. *p < 0.02, Wilcoxon matched-pairs signed rank or paired t-test as indicated in text. Representative confocal maximum intensity projection micrographs of HCR processed embryos for FoxD3 (n = 6) (F), Axud1 (n = 6) (G), Msx1 (n = 5) (H), and BMP4 (n = 5) (I) transcripts in whole mount embryos bilaterally electroporated with control morpholino (left) or Elavl1 morpholino and a Draxin overexpression construct (right). Dotted white line indicates midline. MO, morpholino. Scale bar, 50 μm. (J) Quantitation of HCR processed embryos for control versus Elavl1 knockdown with Draxin rescue, calculated as ratio of Elavl1 MO + Draxin versus control MO integrated density. ns, nonsignificant, p > 0.09, Wilcoxon signed-rank test. Grayed values indicating Elavl1 MO data were originally presented in Figure 3 and are reproduced here to illustrate rescue. *p < 0.05, one-tailed paired t-test.

Model organism and embryo collection
Fertile chicken eggs (Gallus gallus) were purchased locally (Sun State Ranch, Monrovia, CA), and incubated in a humidified 37°C incubator to the specified Hamburger-Hamilton (HH) stage (Hamburger and Hamilton, 1951). Live embryos were removed from eggs with Whatman filter paper as described (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019) and stored in Ringer's solution until further processing.

Immunohistochemistry and HCR
For whole mount immunohistochemistry, embryos were fixed at room temperature for 20 min with 4% paraformaldehyde in sodium phosphate buffer. For cross-sections, embryos were fixed at room temperature for 1 hr, then washed, embedded, and cryosectioned as described (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019) prior to immunohistochemistry. Washes, blocking (10% donkey serum), and antibody incubations were performed in TBSTx (0.5 M Tris-HCl/1.5 M NaCl/10 mM CaCl 2 /0.5% Triton X-100/0.001% Thimerosal) as described (Chacon and Rogers, 2019;Manohar et al., 2020). Primary antibodies are listed in the Key Resources Table. Species-specific secondary antibodies were labeled with Alexa Fluor 568 and 647 (Invitrogen) and used at 1:1000 or 1:500, respectively. For nuclear staining on cross-sections, DAPI (4′,6-diamidino-2-phenylindole) was added to the secondary antibody solution at [14.3 μM] final concentration. Coverslips were mounted using Fluoromount-G (SouthernBiotech). HCR was performed as described (Gandhi et al., 2020). Embryos were fixed at room temperature for 1 hr with 4% paraformaldehyde in phosphate-buffered saline (PBS) prior to HCR processing. Custom HCR probes were designed and ordered through Molecular Technologies.

Gene expression constructs and perturbation
Translation-blocking antisense MO for Elavl1 (Gene Tools; Key Resources Table) was designed to span the Elavl1 (GenBank: NM_204833.1) start codon from nucleotide −20 to +5, and electroporated at [2 mM]. Draxin MO was described previously (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019), and electroporated at [1 mM]. The standard control MO (Gene Tools) was used for contralateral control electroporation. MOs were co-electroporated with pCIG (Megason and McMahon, 2002) or pCI-H2B-RFP (Betancur et al., 2010) to increase electroporation efficiency and to visualize successfully electroporated cells. The Draxin-FLAG overexpression construct (Draxin OE; Hutchins and Bronner, 2018;Hutchins and Bronner, 2019) and the MCP-GFP construct Hutchins et al., 2020 have been previously described. The MS2-Draxin 3′-UTR construct used in this study was generated by exchanging the mTurq2 coding region for a Luciferase coding region in a previously described MS2-Draxin 3′-UTR construct . Electroporations were performed on HH4 gastrula stage chicken embryos as described previously (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019).

RNA-seq and data analysis
For single-cell RNA-seq analysis, raw data from Williams et al., 2019 were downloaded from Gene Expression Omnibus (GEO; GSE130500) and processed for quality assessment using FastQC (Andrews, 2014). The reads were trimmed and filtered using Cutadapt (Martin, 2011), following which they were aligned to the chicken galgal6 (GRCg6a) genome assembly using Bowtie2 (Langmead and Salzberg, 2012). The alignment files generated were used to count features using HTSeq-Count (Anders et al., 2015). All further analyses were performed in R using Seurat (Butler et al., 2018) as previously described (Gandhi et al., 2020). GO terms associated with all genes in the chicken genome were identified using the biomaRt package (Durinck et al., 2009). This dataset was queried for the ontology terms expected to be associated with RNA-binding proteins, such as 'binds mRNA', 'binds 3′-UTR', 'regulates translation', and 'stabilizes RNA'. All feature plots and strip plots were made using Seurat.
For bulk RNA-seq, dorsal neural folds from morpholino-electroporated embryos grown to stage HH8+/9− were dissected in DEPC-treated PBS and pooled (n = 3) before total RNA was extracted using an RNAqueous-Micro kit (Thermo). cDNA libraries (three biological replicates per treatment) were prepared from 10 ng total RNA per pooled sample using the Takara Bio SMART-Seq v4 Ultra Low Input cDNA kit, according to the manufacturer's instructions. Single-end RNA-sequencing was performed at the Caltech Millard and Muriel Jacobs Genetics and Genomics Laboratory at a depth of 20 million reads. Reads were processed and differential expression analysis was performed as described Piacentino et al., 2021). We subsequently filtered the gene lists to exclude lowly expressed genes (average normalized count values <250), resulting in a filtered list of 24 differentially expressed genes (12 downregulated, 12 upregulated) with greater than 1.5-fold change and p < 0.05 cutoff.

RNA decay assay
Embryos were electroporated bilaterally with either control MO or Elavl1 MO and grown to HH8+/9−. Dorsal neural folds were dissected in DEPC-treated PBS, pooled (n = 3), and suspended in explant culture media (Dulbecco's modified Eagle medium supplemented with 10% fetal bovine serum, 10% chick embryo extract, and 1% penicillin/streptomycin). Actinomycin D [10 μg/ml] or DMSO (dimethyl sulfoxide) alone (for 0 hr time point) was added to neural fold samples, which were then incubated at 37°C/5% CO 2 for 0 hr, 30 min, 2 hr, or 4 hr. Following the specified incubation time, neural fold samples were centrifuged at 300 × g for 3 min to remove media then washed three times with DEPCtreated PBS.
Total RNA was then extracted using the RNAqueous-Micro kit (Thermo) and 50 ng of RNA was reverse transcribed using SuperScript III and oligo dT priming. Following reverse transcription, we performed qPCR using FastStart Universal SYBR Green Master (Rox) with cDNA (diluted 1:10) and gene-specific primers (Key Resources Table) on a QuantStudio 3 Real-Time PCR System (Applied Biosystems) in triplicate. Average C T values were calculated for 0 hr time points. Individual C T values for 30 min, 2 hr, and 4 hr actinomycin D time points were subtracted from the 0 hr average C T , and fold change was calculated for each technical replicate from the start of transcription inhibition (t = 0 hr) to each actinomycin D time point for each target. RNA decay rates were calculated and plotted using one phase decay in Prism9 (n = three biological replicates).
Embryonic heads were dissected in Ringer's solution, washed in RNase-free PBS, and dissociated in Accumax (Innovative Cell Technologies) for 15 min at room temperature. Following dissociation, cells were pelleted at 2000 × g for 4 min at 4°C, washed in RNase-free PBS, and resuspended in polysome lysis buffer (0.1 M KCl, 5 mM MgCl 2 , 10 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid), 0.5% NP-40, 200 U RNaseOUT, 1× cOmplete, Mini EDTA-free Protease Inhibitor). Cells were frozen at −80 °C overnight to complete lysis and reduce adventitious binding. Lysate was then thawed on ice, vortexed, and centrifuged at 14,000 × g for 10 min at 4°C to remove cellular debris, then cleared lysate was added to antibody-coated beads in NET-2. Immediately following addition of cleared lysate, 10% was removed to serve as an Input control. IP reaction was tumbled at room temperature for 1 hr, beads were then washed in NT-2, and RNA was eluted in proteinase K buffer (NT-2, 1% sodium dodecyl sulfate, 1.2 mg/ml proteinase K) for 30 min at 55°C and phenol/chloroform extracted.
RNA from Input and IP samples was reverse transcribed using SuperScript III and oligo dT priming. Following reverse transcription, we performed qPCR using FastStart Universal SYBR Green Master (Rox) with cDNA (diluted 1:5) and gene-specific primers (Key Resources Table) on a QuantStudio 3 Real-Time PCR System (Applied Biosystems) in triplicate. We determined ΔC T (ΔC T = Input C T − IP C T ) for Draxin, FoxD3, and Pax7 for Elavl1 and Control IgG RIP samples, then calculated fold enrichment values (=2 (Average Control IgG ΔCT− Elavl1 ΔCT) ) for each target and replicate.

Proximity ligation assay
Embryos were electroporated bilaterally with MCP-GFP alone ('control'), or MCP-GFP and MS2-Draxin 3′-UTR ('experimental'), grown to stage HH9, then fixed at room temperature for 1 hr, washed, embedded, and cryosectioned as described (Hutchins and Bronner, 2018;Hutchins and Bronner, 2019). Sections were processed using a DuoLink PLA kit (Millipore/Sigma) with Anti-Rabbit MINUS PLA probe, anti-Goat PLUS PLA probe, and Far Red PLA detection reagent, and primary antibodies for Elavl1 and GFP, according to the manufacturer's DuoLink PLA Fluorescence protocol.

Image acquisition and analysis
Confocal images were acquired using an upright Zeiss LSM 880 at the Caltech Biological Imaging Facility, and epifluorescence images were acquired using a Zeiss Imager.M2 with an ApoTome.2 module. Images were minimally processed for brightness/contrast and pseudocolored using Fiji (ImageJ, NIH) and Adobe Photoshop.
Relative fluorescence intensity was determined in Fiji. For each whole mount image, the line tool was used to draw an ROI surrounding the area of neural crest indicated by positive HCR fluorescence for the genes examined. For cross-sections, ROIs were drawn surrounding the neural crest and neural tube based on tissue morphology from nuclear staining. Following background subtraction (50-pixel rolling ball radius), integrated density was quantified for the ROIs on the control electroporated (left) and experimental electroporated (right) sides from the same embryo. Relative fluorescence intensity was then calculated by dividing the integrated density measurements for the experimental versus the control side of the same embryo.
Pax7 cell counts were performed as described (Hutchins and Bronner, 2018). The limit of the dorsal neural tube and characterization of cells as 'delaminated' or 'neural tube' was determined based on tissue morphology from nuclear staining. For relative migration distance determined from Pax7-stained embryos, distance of migration was measured in Fiji as described (Hutchins and Bronner, 2018).
For PLA analysis, confocal micrographs were processed in Fiji. Following background subtraction (50-pixel rolling ball radius), ROIs were drawn surrounding the neural crest and neural tube based on tissue morphology from nuclear staining, then puncta (PLA signal) were counted using 'analyze particles' function following global manual thresholding to reduce the appearance of nonspecific puncta outside of electroporated regions.

Statistical analyses
Statistical analyses were performed using Prism (8,9;GraphPad Software). p values are defined in the text, and significance was established with p < 0.05. p values were calculated using Wilcoxon signedrank, Mann-Whitney, one-sample t-tests, one-way ANOVA with post hoc Tukey, unpaired or paired t-tests as indicated; tests were two-tailed unless otherwise specified in the text/legend. Data were confirmed to be normally distributed using Kolmogorov-Smirnov or Kruskal-Wallis tests for parametric tests. Data measuring fluorescence intensities or cell counts for Experimental/Control sides are presented as box plots with individual data points shown. Bar graphs representing qPCR fold enrichment are presented as mean values, with error bars indicating standard error of the mean. The number of embryos and replicates is indicated in figure legends and/or text. Post hoc power analyses (Faul et al., 2007) confirmed sufficient statistical power was reached (≥0.8) for reported p values and sample sizes.

Supplementary files • Transparent reporting form
Data availability RNA-sequencing datasets have been deposited on NCBI under the accession number PRJNA861325. The 3' untranslated region (UTR) sequence for Axud1 has been deposited to GenBank under accession number ON920861.
The following datasets were generated: