Co-option of the piRNA pathway to regulate neural crest specification.

Across Metazoa, Piwi proteins play a critical role in protecting the germline genome through piRNA-mediated repression of transposable elements. In vertebrates, activity of Piwi proteins and the piRNA pathway was thought to be gonad specific. Our results reveal the expression of Piwil1 in a vertebrate somatic cell type, the neural crest. Piwil1 is expressed at low levels throughout the chicken neural tube, peaking in neural crest cells just before the specification event that enables epithelial-to-mesenchymal transition (EMT) and migration into the periphery. Loss of Piwil1 impedes neural crest specification and emigration. Small RNA sequencing reveals somatic piRNAs with sequence signatures of an active ping-pong loop. RNA-seq and functional experiments identify the transposon-derived gene ERNI as Piwil1's target in the neural crest. ERNI, in turn, suppresses Sox2 to precisely control the timing of neural crest specification and EMT. Our data provide mechanistic insight into a novel function of the piRNA pathway as a regulator of somatic development in a vertebrate species.

INTRODUCTION

Small noncoding RNAs and their protein partners, Argonaute proteins, play central regulatory roles in transcriptional and posttranscriptional gene expression in all domains of life (). The Piwi clade of Argonaute proteins is unique to Metazoa, where they are required for maintenance of stemness, self-renewal, and safeguarding of the genome by repressing transposable elements (TEs) in the germ cell lineage (–). TEs are mobile genetic elements that can replicate and reinsert themselves in the genome, threatening genomic integrity. By keeping these “selfish genes” in check, the Piwi-interacting RNA (piRNA) pathway helps preserve genomic stability and thus plays a critical role in the arms race between TEs and their host genomes (). piRNAs recognize TE transcripts via sequence complementarity, which, in the cytoplasm, leads to TE target cleavage by the Piwi protein. In several organisms, Piwi proteins have gained the ability to enter the nucleus and instigate chromatin modifications at target loci. piRNA biogenesis differs from that of their better-known relatives, microRNAs (miRNAs) and small interfering RNAs, and relies in part on cleavage by Piwi proteins themselves in an amplification cycle termed the ping-pong amplification loop (, ), as well as on other cytoplasmic factors that are unique to this pathway (, ). These differences lead to characteristics of piRNAs that are distinct from other small RNAs: piRNAs are slightly longer than miRNAs at around 23 to 30 nucleotides (nt), predominantly map to TE sequences, have a 1-U bias at their 5′ end, and, in the case of ping-pong–generated piRNAs, a 10-A bias and a characteristic 10-nt overlap between complementary sequences (, , ). Due to 3′ end processing, they also carry a 2′-O-methyl residue, which enables differential cloning ().

TEs are extremely prevalent and found in all metazoan genomes. New lineage-specific TEs emerge often during the course of evolution, and the repertoire of active transposons can vary widely among species (, ). Despite their negative impact on genomic stability when left unchecked, TEs provide a steady source of germline mutations and are considered an important driver of evolution. There are many examples of host genome co-option of retroviral genes, as well as domestication of long terminal repeats (LTRs) that bind host transcription factors to rewire gene regulatory networks, many of which act in somatic tissues (, , ).

While Piwi proteins and the piRNA pathway perform critical functions in the germ line to repress TEs, their potential role in somatic cells is not as well understood. It has long been queried whether the piRNA pathway might be active outside the germ line or used to regulate genes other than TEs. In invertebrate models such as Hydra and Planaria with high regenerative capacity, somatic stem cells have been shown to use the canonical piRNA pathway to actively repress TEs (, ). In addition, Planaria Piwi protein SMEDWI-3 can regulate non-transposon mRNAs (). In the sea slug Aplysia, a piRNA-mediated mechanism has been shown to mediate epigenetic regulation of synaptic plasticity (). piRNAs and piRNA pathway genes are also expressed in somatic tissues of numerous arthropods (–). These findings suggest that somatic piRNA-mediated regulation may be widespread. In contrast to these invertebrate models, a functional role for the piRNA pathway in somatic cells of vertebrates has been debated (, ). Piwi protein expression has been observed in cancer cells () and some adult tissues including hematopoietic stem cells (), brain, and heart (, ). Loss of mouse Piwi protein Mili has been correlated with behavioral deficits (), and Piwil1 has been implicated in neuronal migration in rats (), although mechanistic insights remain elusive and whether there is piRNA involvement in these processes is unclear.

Here, we report somatic Piwi and piRNA expression in a vertebrate embryonic cell type, the neural crest. The neural crest is a rapidly evolving, migratory population of stem cells that is unique to vertebrates and essential for their development and evolution (). While neural crest cells undergo specification within the forming central nervous system during neurulation, they subsequently leave the neural tube to migrate extensively throughout the embryo and contribute to a diverse array of tissues (, ). Our functional analysis shows that chick Piwil1 (Chiwi) is required for neural crest cell emigration from the neural tube. Chiwi regulates expression of TE-derived gene ERNI, a regulator of Sox2 expression during early nervous system formation, through an active somatic piRNA pathway. Our results indicate that the gene regulatory network controlling neural crest development has co-opted a transposon-derived sequence and its piRNA-mediated regulation to precisely time neural crest specification and initiation of the epithelial-to-mesenchymal transition (EMT) from the neural tube to begin neural crest migration.

RESULTS

Chiwi exhibits a distinct expression pattern in the developing neural tube

As a first step in examining whether the piRNA pathway plays a role in regulating vertebrate developmental events, we assessed the expression of Piwi genes at early embryonic stages. Vertebrates have two conserved Piwi proteins, Piwil1 and Piwil2, which play central roles in the germline piRNA pathway (–). To test whether Piwi transcripts are expressed in somatic cells of the early vertebrate embryo, we examined Piwil1 (Chiwi) and Piwil2 (Chili) expression during chick embryogenesis by quantitative reverse transcription polymerase chain reaction (qRT-PCR) using cDNA derived from whole embryos from stages ranging from Hamburger and Hamilton (HH) stages () HH4 to HH23 (fig. S1). We observed extremely low Chili transcript levels at all stages. In contrast, Chiwi transcript levels were markedly higher, especially at early stages, peaking at HH6 to HH8, corresponding to neurulation stages, and dropping after HH10 to HH12, corresponding to the time of active cranial neural crest migration. Primordial germ cells are extraembryonic until stage HH13 (~2 days of development), when they enter the embryonic bloodstream through which they migrate to the gonadal anlagen by around HH18 (3 days) (). Thus, high Chiwi mRNA levels before this embryonic stage indicate that Chiwi is expressed in somatic cells.

To confirm the RT-qPCR results and assess Chiwi expression with spatiotemporal resolution, we performed hybridization chain reaction (HCR) on HH6 to HH12 embryos in whole mount (Fig. 1A and fig. S2A). Transverse sections revealed low but ubiquitous Chiwi expression throughout the cranial region at HH6, which then becomes primarily constrained to the neural tube at HH9 and onward (Fig. 1B and fig. S2B). This shift from ubiquitous to tissue-specific expression in the cranial region mirrors the drop in Chiwi expression that we observed between HH6 to HH8 and HH10 to HH12 cDNA via qPCR. Chiwi expression appears to be differentially regulated in the dorsal tube at HH9, where there are distinguishable subdomains of neural crest precursors that differ in their expression of neural crest specifier genes (Fig. 1, A and B) (, ). In the dorsal midline, there are cells that express both Snai2 and Pax7, corresponding to premigratory neural crest cells in the process of undergoing EMT (, ). In this domain, Chiwi expression is reduced compared to the rest of the neural tube. In contrast, in the immediately lateral domain that feeds into the specified neural crest pool (), cells are marked by high Pax7 but no Snai2 and relatively high levels of Chiwi expression.

Fig. 1.

Piwi protein and piRNA expression in the cranial region.

(A) HCR reveals the expression of Chiwi and the neural crest markers, Pax7 and Snai2; scale bars, 50 μm. (B) Quantification of the intensity of HCR signal of Chiwi, Pax7, and Snai2 from the ventral midline to the dorsal neural folds of the neural tube (n = 4 to 5 sections each from two HH9 embryos). (C) Schematic diagram of the small RNA cloning strategy from the midbrain region of HH9− embryos (n = 2 biological replicates). (D) Annotation of small RNAs mapping to the genome. Orientation is relative to the annotated feature. “Other” category includes reads that could not be assigned to a feature, as well as reads mapping to simple repeats, satellite repeats, small cytoplasmic RNA, and small nuclear RNA, which together account for <1% of mapped reads in all samples. (E) Length distribution of all reads mapping to the genome in total and oxidized libraries. (F) Length distribution of reads from oxidized libraries mapping to TEs in sense and antisense orientation. (G) Analysis of 5′ to 5′ distance of complementary small RNA reads mapping to TEs in total and oxidized libraries. (H) Sequence logos of oxidized, collapsed sequences mapping to TEs in antisense (left) and sense (right) orientation.

We next confirmed the presence of Chiwi in the neural crest region by RNA sequencing (RNA-seq). To this end, we used three RNA-seq datasets, one from dissected cranial neural folds (two replicates) and two previously published fluorescence-activated cell sorted specified neural crest datasets (), which include early migrating cranial neural crest cells from HH9 stage embryos (three replicates) and trunk neural crest cells from HH18 embryos (three replicates). All three datasets reveal notable expression levels of Chiwi mRNA, with respective transcripts per million (TPM) of 6.1 and 6.5 in the specified cranial and trunk datasets and a TPM of 13.9 in the cranial neural fold dataset. Chili was detected in neural folds with a TPM of 4.5 but was not detectable in sorted migrating neural crest cells (fig. S2C). Together, these results support the somatic expression of Chiwi in both neural crest precursors and early migrating neural crest cells of the chicken embryo, as well as low-level expression of Chili in dorsal neural folds.

Somatic piRNAs target transposons

To address whether Chiwi could exert a regulatory function in the neural crest that is directed by associated piRNAs, we tested for the presence of piRNAs in the cranial region. Due to a lack of appropriate antibodies, it was not feasible to immunoprecipitate Chiwi and directly analyze associated small RNAs. As an alternative, we cloned and sequenced small RNAs from the midbrain region of HH9 heads, where Chiwi expression is very pronounced and confined to the neural tube. In parallel, to specifically enrich for piRNAs, we performed small RNA cloning that included an oxidation step to select for RNA species that are 2′-O-methylated at their 3′ end, a characteristic of piRNAs (Fig. 1C).

Roughly 64 and 60% of the reads in the total and oxidized samples mapped to the genome with no mismatches. Among these, both the total and oxidized samples showed comparable numbers of reads mapping to ribosomal RNA (rRNA) at around 6% (Fig. 1D). Total small RNA reads mostly mapped to miRNAs, with some mapping to tRNAs and to genes in sense orientation (Fig. 1D). Sense mapping reads are unable to target the corresponding mRNAs and likely represent degradation products, while antisense mapping reads have the potential to silence transcripts via the piRNA pathway. Less than 1% of the mapped reads in the total small RNA samples corresponded to transposon sequences, mostly in antisense orientation, confirming that piRNA expression in this tissue is extremely low relative to other small RNA species. The representation of different TE families in total TE-mapping reads, however, was similar to the oxidized samples, indicating that the total small RNA library contains piRNAs, even if they represent only a small fraction of all reads (fig. S3A). Consistent with predominance of miRNAs and degradation products in the non-oxidized libraries, the size distribution of the total small RNA samples peaked at 22 nt and showed a broader distribution across all sizes (Fig. 1E). In contrast, 44% of the reads in the oxidized samples mapped to TEs, with 96% of these in antisense orientation (Fig. 1D), and the size distribution of these libraries peaked at 28 nt (Fig. 1E), consistent with the size profile of previously published chick embryonic piRNA libraries (). Interestingly, we noted an enrichment of tRNA mapping reads in addition to TE-mapping reads in the oxidized libraries, consistent with tRNAs being heavily 2′-O-methylated (Figs 1D) (). Upon plotting the size distribution of tRNA mapping reads, we saw a strong enrichment of 23- and 25-nt sequences in the oxidized samples, possibly suggesting the presence of tRNA-derived small RNAs with 2′-O-methylated 3′ ends (fig. S3B).

To further characterize the presumptive piRNA population, we analyzed the size profiles of sense and antisense TE-mapping reads in the oxidized samples. Interestingly, we found that sense reads peaked between 25 and 27 nt, while the more abundant antisense reads had a strong peak at 28 nt (Fig. 1F and fig. S3C). While the RT-qPCR and RNA-seq data indicated very low Chili expression (Fig. 1A and fig. S2C), these size profiles suggest that more than one Piwi protein might be present, albeit at different expression levels, and contributing to piRNA biogenesis via the ping-pong amplification mechanism. Consistent with this idea, we observed a strong enrichment for 10–base pair (bp) overlap between complementary reads in the oxidized libraries when we analyzed 5′ to 5′ overlap of TE-mapping reads (Fig. 1G), as well as a notable 1-U bias of antisense and 10-A bias of sense sequences upon collapsing the libraries (Fig. 1H). These findings provide strong support for an active somatic piRNA pathway in the embryo and imply that somatic piRNAs are, at least in part, generated by a ping-pong mechanism, possibly through the interplay of two Piwi proteins.

To determine whether somatic piRNAs might be generated by a similar mechanism to germline piRNAs, we looked for expression of genes involved in germline piRNA biogenesis in RNA-seq data from dorsal neural folds (fig. S4). We found that several key factors are expressed at low levels, comparable to those of Chiwi and Chili. These include Hen1 (TPM 4.0), which imparts 2′-O-methylation to piRNAs (), Tudor Domain Containing 9 (TDRD9) (TPM 1.4), which is involved in ping-pong piRNA biogenesis (), PNLDC1 [poly(A)-specific ribonuclease-like domain containing 1] (TPM 7.6), which trims the 3′ ends of piRNAs (), and, possibly, the Zucchini homolog phospholipase D6 (PLD6; TPM 9.0) (), although RNA-seq tracks of PLD6 are noisier than those of the other genes. Interestingly, the transcription factor A-MYB, which mediates expression of both piRNA clusters and piRNA pathway genes, is also present (TPM 3.0), as are the two Tudor domain–containing genes TDRD1 (TPM 7.7) and TDRD3 (TPM 53.5), which A-MYB has been shown to regulate along with Chiwi in rooster testes (). Together, expression of piRNA pathway genes in the dorsal neural tube raises the possibility that a similar mechanism of piRNA biogenesis to that seen in the germ line might be at play.

Loss of Chiwi disrupts neural crest development

On the basis of Chiwi’s distinct expression pattern, we hypothesized that it plays a role in neural crest development. To probe Chiwi’s function in the neural crest, we first performed loss-of-function experiments using a translation-blocking Chiwi morpholino oligomer (MO). After electroporating Chiwi MO into the prospective neural crest region on one side of the embryo and control MO on the other side at HH4, we allowed the embryos to grow until HH9, when cranial neural crest begins to delaminate and migrate. Chiwi appears to be relatively strongly expressed in the dorsal neural tube at this stage (Fig. 2A and fig. S5). By immunostaining for the neural crest marker Pax7, we observed a reduction in neural crest migration distance from the midline on the Chiwi-depleted side. Upon sectioning, we saw significant reduction in the numbers of Pax7-positive cells on the Chiwi-depleted side (Fig. 2B).

Fig. 2.

Perturbation of Chiwi disrupts neural crest development.

(A) Schematic diagram of the chick embryo electroporation strategy. (B) Loss of Chiwi impedes neural crest migration and reduces neural crest cell count, as measured by Pax7-expressing cells. Left: Examples of whole mount and cross sections upon MO and CRISPR knockout of Chiwi. Right: Quantification of image analysis. Each data point represents measurements from a single embryo, with the right (experimental) divided by the left (control) side. (C) Overexpression (OE) of Chiwi increases the number of Pax7-positive cells, though migration distance is not significantly altered, while overexpression of the YK mutant, which is unable to bind piRNAs, impedes neural crest migration and reduces cell number. The blue stars and brackets denote the control side, while the red stars and brackets denote the experimental side. Scale bars, 50 μm. Box plots indicate the interquartile range, while whiskers extend to minimum and maximum values. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and ****P ≤ 0.0001. Asterisks (*) represent the difference between control and experimental measurements for each treatment.

To confirm the specificity of our results, we used a second approach to knock out Chiwi via a plasmid-based CRISPR-Cas9 system optimized for use in chicken (). To this end, we designed two CRISPR guide RNAs against sequences corresponding to the first exon junction and the 3′ piRNA binding site of Chiwi. We electroporated a Cas9 construct along with control guide on one side and Chiwi guides on the other side of HH4 stage embryos. HCR for Chiwi after CRISPR-Cas9 knockout confirmed a reduction in Chiwi mRNA levels compared to the control side at stage HH9, just before the onset of neural crest migration (fig. S6). CRISPR-Cas9–mediated loss of Chiwi gave the same reduction in neural crest migration distance and cell count at HH9 as did the Chiwi morpholino (Fig. 2B).

Chiwi’s piRNA binding activity is necessary for its function in the neural crest

Due to the presence of piRNAs in the cranial midbrain region, we hypothesized that Chiwi identifies its targets via the associated piRNAs. To test whether Chiwi’s piRNA binding activity is necessary for its function in the neural crest, we created wild-type and mutant Chiwi overexpression constructs. The mCMV-YK-Chiwi construct contains two amino acid substitutions in positions 574 and 578, resulting in impaired piRNA binding (, ). To express these at levels similar to those of endogenous Chiwi, we cloned cDNA amplicons into a construct containing a minimal cytomegalovirus (mCMV) promoter, which has weak global expression in chick. The vector also contains an Histone2B-RFP (H2B-RFP) sequence following an internal ribosome entry site downstream of the Chiwi sequence. As a control, we used the empty vector, which expresses only the H2B-RFP under the mCMV promoter. While overexpression of wild-type Chiwi did not significantly alter neural crest migration, it did result in an increase in Pax7-positive cells. In contrast, overexpression of the piRNA binding mutant resulted in reduced neural crest migration distance and cell count (Fig. 2C) compared to the control, recapitulating the Chiwi loss of function phenotype and indicating that Chiwi’s piRNA binding ability is required for its function in the neural crest.

Chiwi regulates a single, transposon-derived gene, ERNI, in the dorsal neural tube

To investigate Chiwi targets in the neural crest, we performed RNA-seq experiments after CRISPR-Cas9–mediated knockout. Differential mRNA expression analysis of control and knockout RNA-seq libraries generated from total RNA of neural folds at stage HH9 revealed that the most significantly up-regulated gene upon Chiwi knockout is the retrotransposon embryonic normal stem cell (ENS-1), also called Soprano (Fig. 3A). ENS-1 appears to have expanded recently within the Galliforme lineage and has numerous, highly homologous copies in the genome, most of which only contain the LTR (, ). Several copies, however, contain only the 5′ end of the internal domain, which codes for ERNI (, ), a chicken-specific transposon-derived gene and a known regulator of neural induction (, ).

Fig. 3.

Chiwi regulates a single, transposon-derived gene, ERNI, in the neural crest.

(A) Differential expression analysis of RNA-seq from HH9− Chiwi CRISPR knockout versus control cranial neural folds. Dots represent genes (blue) and transposon families (orange). FC, fold change. (B) HCR depicting ERNI, Chiwi, and Snai2 expression in wild-type HH9 cranial midbrain region; scale bar, 50 μm. (C) HCR reveals changes to ERNI expression upon Chiwi loss (MO and CRISPR), as well as Chiwi and YK mutant overexpression. Each data point represents the average fluorescent intensity of the right dorsal fold region (experimental) divided by the left (control) side from three nonadjacent sections from the cranial region of a single embryo. Scale bar, 50 μm. Box plots indicate the interquartile range, while whiskers extend to minimum and maximum values. *P ≤ 0.05 and **P ≤ 0.01. Asterisks (*) represent the difference between control and experimental measurements for each treatment. (D) Normalized small RNA read counts mapping to the ERNI mRNA sequence in total versus oxidized (Ox) small RNA libraries. Error bars indicate SDs from two biological replicates. (E) Analysis of 5′ to 5′ distance of complementary small RNA sequences mapping to the ERNI mRNA sequence in the oxidized small RNA libraries. (F) Small RNA-seq (top) and RNA-seq (bottom) tracks depicting sequences mapping to ENS-1 loci and the ERNI mRNA sequence (left). Oxidized small RNAs mapping in sense (teal) and antisense (red) orientation are depicted separately. Cranial neural fold total RNA-seq [control libraries from (A)] is depicted in blue. Replicate tracks are overlaid.

To validate our RNA-seq results, we performed HCR against the ERNI mRNA sequence (Fig. 3B). In wild-type embryos, this revealed ERNI expression throughout the neural tube and ectoderm, albeit at varying levels. In particular, we noted higher expression in the dorsalmost region of the neural tube, corresponding to the Snai2+ domain in which Chiwi transcripts are down-regulated. Consistent with this, ERNI was reduced compared to the control side after Chiwi overexpression, whereas Chiwi depletion and overexpression of the piRNA binding mutant resulted in an increase in ERNI (Fig. 3C).

Analysis of small RNA reads mapping to the ERNI mRNA sequence, which includes portions of both the ENS-1 internal domain and LTR, showed enrichment in the oxidized piRNA sample compared to total small RNAs, accounting for 0.01% of mapped total small RNA reads and 0.30% of mapped oxidized reads, and displayed the 5′ to 5′ complementarity signature of the ping-pong pathway (Fig. 3, D and E), indicating that Chiwi is regulating ERNI via a piRNA-mediated mechanism. Both the RNA-seq and the small RNA samples showed enrichment specifically over the LTR and ERNI sequence and did not map to other internal parts of the ENS-1 transposon (Fig. 3F). Together, these results indicate that ERNI expression in the dorsal neural tube is spatially regulated by Chiwi in a piRNA-dependent fashion.

Perturbation of ERNI recapitulates Chiwi phenotypes

To functionally test whether dysregulation of ERNI could account for the observed neural crest defects upon Chiwi perturbation, we directly disrupted ERNI expression in the dorsal neural tube and analyzed its effect on neural crest cell count and migration. To this end, we generated an overexpression vector, enh195-FLAG-ERNI, which encodes an N-terminally FLAG-tagged ERNI coding sequence under control of the Pax7 enhancer enh-195 (Fig. 4A) (). We chose to perturb ERNI only within the PAX7+ domain to avoid disturbing the earlier process of neural induction, where ERNI is known to play a pivotal role. We electroporated this construct into one side of the embryo at HH4, with a control version driving citrine on the other side. Pax7 staining of HH9 embryos revealed a reduction in neural crest cell count and migration distance, similar to that seen after Chiwi knockout (Fig. 4B); this is consistent with Chiwi regulating ERNI expression. The ERNI protein harbors an N-terminal coiled-coil domain, which is responsible for recruitment of ERNI to the Sox2 N2 enhancer, and a C-terminal Heterochromatin Protein 1 (HP1) box, which recruits HP1-γ, thereby inducing transcriptional repression (). To further test ERNI’s function, we created a dominant negative ERNI construct, enh195-N150-FLAG-ERNI-NLS (N150), which only contains the first 150 amino acids of ERNI. This includes the Sox2 localizing coiled-coil domain but lacks the HP1-γ binding domain. We also added a nuclear localization signal on the C terminus to ensure nuclear localization, which we confirmed by immunostaining (fig. S7). Overexpression of this dominant negative construct imparted a reciprocal phenotype to wild-type ERNI overexpression, with an increased number of Pax7+ neural crest cells at HH9 (Fig. 4B), recapitulating the Chiwi overexpression phenotype. Together, these data suggest that a primary role of Chiwi during neural crest development is to regulate ERNI expression in a spatiotemporal manner.

Fig. 4.

Perturbation of ERNI recapitulates Chiwi phenotypes.

(A) Schematic diagram of ERNI overexpression and dominant negative (N150) expression construct products. (B) Overexpression of ERNI recapitulates loss of Chiwi phenotype, with a reduction in neural crest migration distance and cell number, while overexpression of the N150 truncated ERNI sequence increases cell count and migration distance, as measured by Pax7-expressing cells. Left: Examples of whole mount and cross sections upon ERNI and N150 overexpression. Right: Quantification of image analysis. Each data point represents measurements from a single embryo, with the right (experimental) divided by the left (control) side. (C) HCR reveals that ERNI overexpression leads to a reduction in Sox2 expression in the dorsal neural tube, while N150 dominant negative overexpression increases Sox2 expression. Each data point represents the average fluorescent intensity of Sox2 signal in the right dorsal fold region (experimental) divided by the left (control) side from three nonadjacent sections from the cranial region of a single embryo. The teal stars and circle denote the control side, while the orange stars and circle denote the experimental side. Scale bars, 50 μm. Box plots indicate the interquartile range, while whiskers extend to minimum and maximum values. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and ****P ≤ 0.0001. Asterisks (*) represent the difference between control and experimental measurements for each treatment. (D) Schematic diagram of the neural crest piRNA pathway, depicting a neural tube with the PAX7+/Snai2− stem cell niche in teal, which feeds into the Snai2+-specified neural crest region in dark blue. Chiwi represses ERNI in the PAX7+/Snai2− stem cell niche via a piRNA-mediated mechanism to maintain Sox2 expression and proliferation. In specified neural crest, Chiwi expression is reduced, permitting ERNI expression, which, in turn, represses Sox2 to allow for neural crest specification and EMT.

ERNI regulates Sox2 during neural crest specification

Since ERNI has been shown to regulate Sox2 expression at early gastrula stages (, ), we next tested whether it also plays a role in Sox2 regulation in the dorsal neural tube during neural crest development. To address this, we performed HCR for Sox2 on embryos into which we electroporated the ERNI overexpression and ERNI dominant negative constructs. We noted a decrease in Sox2 expression in the Pax7+ domain of embryos, where ERNI expression is increased, and an increase in Sox2 expression in this domain of embryos that expressed the N150 dominant negative construct (Fig. 4C); this is consistent with ERNI inducing Sox2 repression. Together, these results suggest that ERNI functions to repress Sox2 in the dorsal neural tube, thereby enabling activation of the gene program to specify bona fide neural crest cells, which is inhibited by high levels of Sox2 (). Chiwi, in turn, is responsible for repressing ERNI to maintain Sox2 expression in the Pax7+/Snai2− dorsal neural tube cells, prior to the onset of specification and emigration from the neural tube (Fig. 4D).

DISCUSSION

Host co-option of TE components is a relatively common occurrence and an important driver of evolution. Until now, the piRNA pathway had not been implicated in this process. Here, we show that the piRNA pathway, canonically functioning in the germ line to repress deleterious expression of TEs, has been co-opted in the chicken embryo to spatiotemporally regulate the expression of TE-derived gene ERNI and precisely time neural crest specification and EMT. To our knowledge, this is the first demonstration of co-option of the piRNA pathway to regulate a developmentally relevant gene, expanding our understanding of the piRNA pathway from solely being an antagonistic force against TEs to playing a critical role in the domestication of a TE-derived gene.

ERNI is known to regulate Sox2 at the onset of neural induction in the chicken embryo () and is derived from an endogenous retrovirus sequence that appears to be unique to the Galliforme lineage (). ERNI expression has also been observed in chicken embryonic stem cells and in the embryonic gonads, where it appears to be a marker of pluripotency (, ). The fact that ERNI has embedded itself in the highly conserved processes of neural crest specification and neural induction raises the possibility that similar regulation of development and differentiation by piRNA-mediated regulation of TE sequences might be happening in other species. Intriguingly, another TE-derived gene, Crestin, is expressed in specified neural crest cells of zebrafish, although its function is unclear (). We hypothesize that the piRNA pathway plays a conserved role in vertebrate somatic development, although the transposon sequences through which it exerts its function likely change more frequently, in concert with the ever-evolving TE landscape of vertebrates. By constantly updating the pool of piRNAs to match the evolving TE profiles, the piRNA pathway naturally provides the needed plasticity to repress whichever TE is co-opted by the gene regulatory network in a given species or tissue. Consistent with this idea, we observed piRNAs from several TE families in the chick midbrain, suggesting that despite the fact that only ERNI is being regulated by the piRNA pathway in this context, the machinery necessary to regulate other TEs in the same spatiotemporal manner is already present.

In the dorsal region of the chick neural tube, we posit that down-regulation of Sox2 by ERNI permits Pax7-positive cells to complete the neural crest specification program and undergo EMT, becoming migratory neural crest cells. However, when we overexpress ERNI, we see less Sox2 in the dorsal neural tube but fewer neural crest cells, which challenges our understanding of Sox2 as a promoter of neural fate and inhibitor of neural crest in this tissue (, ). One possible explanation for this observation is that ERNI might also regulate other genes in addition to Sox2 in the dorsal neural tube. We postulate, however, that very precise Sox2 levels are required to maintain the proliferative ability of Pax7+/Snai2− cells, such that reducing Sox2 in this region of the neural tube–which expresses pluripotency markers and is thought to act as a stem cell niche, contributing to both neural and neural crest fates (, )–leads to a loss of its ability to self-renew. This model of Sox2’s role in neural crest development mirrors its function in other cell types. Sox2 is a well-known stem cell factor, required for maintenance of several developmental stem cell niches, and precise maintenance of its expression levels can have a notable effect on cell fate. For example, a recent study found that Sox2 levels in the chick tail bud modulate a stem cell population driving secondary neurulation, with very low levels of Sox2 required to maintain proliferation in the stem cell niche and overexpression of Sox2 instigating differentiation into neural epithelium and also reducing the self-proliferative properties of these cells (). Thus, we hypothesize that the piRNA pathway may function as a guardian of the stem cell niche that feeds into the specified neural crest region, regulating proliferation by repressing ERNI to maintain Sox2 expression and only permitting ERNI to switch off Sox2 when it is time for a cell to undergo EMT and become bona fide migratory neural crest.

The piRNA pathway has long been thought to be confined to the germ line in most organisms because expression of its components is typically not observed in other tissues, particularly in vertebrates. Unlike the germline piRNA pathway, it is worth noting that the neural crest piRNA pathway requires relatively low levels of Piwil1 and its associated piRNAs to function. The fact that Piwil1 can have a substantial effect at these levels suggests that there may be other somatic piRNA pathways previously missed due to low expression profiles. This, in turn, raises the intriguing possibility that repurposing of the piRNA pathway for somatic gene regulation may be a widespread occurrence, perhaps shedding light on the myriad reports of Piwi protein expression in diverse cancers (). Somatic regulation by Piwi proteins not only leads to interesting new functions of the piRNA pathway outside of the gonads, but also implies that activation of the piRNA pathway, and thus its adaptation as an epigenetic tool in research and therapy, may be an attainable goal.

MATERIALS AND METHODS

Cloning of expression vectors

The mCMV-H2B-RFP construct was generated by replacing the rabbit β-actin promoter of PCI-H2B-RFP with an mCMV promoter sequence using the restriction sites Spe I and Xba I. This promoter also contains a 5xTetO site, which we left uninduced to achieve weak expression. The Chiwi coding sequence (Ensembl transcript ENSGALT00000004171.6) was PCR-amplified from HH10 to HH12 whole-embryo cDNA, prepared using oligo(dT) primers and the SuperScript III reverse transcriptase kit, and subsequently cloned into the mCMV-H2B-RFP vector using the restriction sites Asc I and Xho I to create mCMV-Chiwi. The mutant construct (mCMV-YK-Chiwi) was generated using the QuikChange Lightning Multi Site-Directed Mutagenesis Kit to generate the Y574I and K578E amino acid substitutions, which have been previously described to inhibit piRNA binding (, ). As a marker for electroporation efficiency, pCI-H2B-RFP was coelectroporated alongside the mCMV constructs, as mCMV-H2B-RFP expression is too low to pick up after methanol dehydration of embryos.

The ERNI coding sequence [National Center for Biotechnology Information (NCBI) RefSeq: NM_001080874.1] was PCR-amplified from cDNA obtained from reverse transcribing HH9 head RNA using oligo(dT) and cloned with an N-terminal FLAG tag into the enh195-citrine vector () using fusion PCR, Eco RI, and Bsr GI (enh195-FLAG-ERNI). The coding sequence for the first 150 amino acids (N150) was similarly cloned but with an N-terminal FLAG tag and a C-terminal nuclear localization signal (enh195-N150-FLAG-ERNI-NLS), which was confirmed to localize to the nucleus by FLAG staining (fig. S7).

Electroporation

Fertilized chicken eggs were acquired from various providers, most recently from Sun State Ranch (Sylmar, CA), and grown at 37°C for 18 to 20 hours to reach HH4 and HH5. Ex ovo electroporations were performed on stage HH4 and HH5 embryos as previously described. Embryos were dissected onto rings of filter paper in Ringer’s, and a solution of DNA expression construct or MO was injected into the space between the vitelline membrane and ectoderm (Fig. 3A) and electroporated into the ectoderm with five pulses of 5.2 V for 50 ms, with 100 ms between each pulse. Embryos were then cultured at 37°C in thin albumen with penicillin-streptomycin until HH9. All embryos were bilaterally electroporated with the control on the left side and experimental on the right side, allowing for direct comparison, and electroporated with either a fluorescent reporter or tagged protein to allow for confirmation of electroporation coverage.

Expression constructs were injected at a concentration of 1 μg/μl, while MOs were used at a concentration of 0.25 mM with pCIG-GFP (green fluorescent protein) (1.0 μg/μl) as carrier DNA. Fluorescein isothiocyanate–labeled MOs used include standard control MO (5′-CCTCTTACCTCAGTTACAATTTATA-3′) and Chiwi translation blocking MO (5′-TCTGGCTCTAGCTCTTCCTGTCATG-3′) from Gene Tools.

CRISPR-mediated knockout was performed as described previously (). A Cas9-expressing construct (pCAG-nls-hCas9-nls-eGFP) was coelectroporated alongside two guide RNA constructs targeting Chiwi, one in the first exon (5′-GGGAGGTCTCCCTCTCGCTC-3′) and the other in the piRNA binding region (GAATGTGACGGTAGGACCTG), or alongside a nonbinding control guide (5′-GCACTGCTACGATCTACACC-3′).

Reverse transcription quantitative polymerase chain reaction

RNA was extracted from batches of wild-type embryos dissected within the area pellucida using TRIzol and reverse transcribed using SuperScript III reverse transcriptase with random hexamers according to the manufacturer’s suggestions. qPCR was performed on an Eppendorf RealPlex using MyTaq mix, SYBR Green, and the respective primers averaging the Ct values for technical triplicates. Chiwi and Chili Ct values at different developmental stages were normalized to 18S rRNA (delta Ct), and fold difference was calculated to Chiwi levels at HH4 to analyze relative expression.

Small RNA-seq

The midbrain region of wild-type HH9− chicken embryo heads was dissected, and RNA was purified using TRIzol. Two replicates of several pooled embryos each were collected. Two micrograms of total RNA was then run on a 15% denaturing polyacrylamide gel, and small RNAs within a 19- to ~38-nt range were isolated and gel extracted as previously described (). Half of each sample was then oxidized in borate buffer [5× solution (pH 8.6); 150 mM borax and 150 mM boric acid] and 25 mM sodium periodate for 25 min at 25°C, while the other half of the sample was incubated in buffer only. Samples were then ethanol precipitated, and libraries were cloned using the NEBNext Small RNA Library Prep Set for Illumina (E7330S) according to the protocol with NEBNext Multiplex Oligos for Illumina (set 1 E7335S). To optimize for the extremely low concentration of RNA and avoid overamplification of adapter dimers, adapters were diluted 1:10 for cloning of the oxidized samples. Libraries were sequenced on an Illumina HiSeq X Ten (150-bp reads, single end) at a sequencing depth of ~20 million reads. Reads were trimmed and mapped to the chicken genome (galGal6) using Bowtie (), with no mismatches end to end and multimapping reads included (-V 0 -k 1 --best). Reads were analyzed for sequence length, gene type that they mapped to, and whether they mapped sense or antisense to a feature. Reads mapping to TEs were extracted and analyzed for 5′ to 5′ complementarity using a previously published script that we altered to include reads ranging from 19 to 34 nt (), and the ping-pong z score was calculated as described previously () by taking the difference of the value at position 10 and the mean of the background values (values of all positions but 10), divided by the SD of the background values. TE-mapping reads were then analyzed for mapping orientation by read length, and after deduplication of libraries, sequence logos were generated for the first 18 nt of each sequence using WebLogo (). Due to the low number of sense mapping reads and overrepresentation of some sequences, collapsing of libraries was necessary to resolve the 10-A bias. Reads were normalized to reads per million mapped reads (to the genome) unless otherwise stated.

To generate a heatmap of small RNA reads mapping to TEs, reads mapping to different TE families were counted with FeatureCounts () using a RepeatMasker GTF. Reads were then normalized by reads per million mapped reads to the genome, and hierarchical gene clustering was performed using Cluster 3.0 (). Normalized read counts were then adjusted by log2, and a heatmap was generated using Morpheus software ().

To analyze small RNAs mapping to the ENS-1 and ERNI loci, reads were mapped to chicken TE consensus sequences from Repbase () using Bowtie with three mismatches allowed and reporting all valid alignments (-V 3 -a --best --strata). Reads were separately mapped to the ERNI mRNA sequence (RefSeq: NM_001080874.1) using the same parameters. Reads were normalized to reads per million reads that map to the genome.

mRNA-seq

Cranial neural folds of two replicate batches of HH9− embryos electroporated at HH4 with CRISPR control constructs on the left side and Chiwi CRISPR constructs on the right side were dissected, and RNA was isolated using the RNAqueous-Micro Total RNA Isolation Kit (Ambion). Libraries were prepared and sequenced on an Illumina HiSeq 2500 by the Millard and Muriel Jacobs Genetics and Genomics Laboratory at Caltech at a depth of ~60 million reads (50-bp reads, single end). Reads were trimmed for adapter sequences and mapped to the galGal6 genome using Bowtie2 (), and reads mapping to genes and TEs were counted with FeatureCounts. Differential expression analysis was performed using DESeq2 (). To analyze Chiwi and Chili levels in sorted cranial and trunk neural crest cells, previously published raw data (NCBI BioProject no. PRJNA497902) () were obtained and mapped to galGal6 using Bowtie2.

TPM counts were generated with TPMCalculator () using the Ensembl galGal6 GTF (GRCg6a, International Nucleotide Sequence Database Collaboration (INSDC) Assembly GCA_000002315.5) (), with the coordinates for the predicted RefSeq Chili locus (XM_025142807.1) added, as Chili is not currently annotated in Ensembl. Read coverage plots were generated using the UCSC Genome Browser () and normalized to reads per million mapped reads. Ensembl tracks were used for gene models in the figures for all transcripts except Chili, which is not annotated in Ensembl, and PLD6, which is annotated slightly differently in RefSeq than Ensembl, with the RefSeq version more closely matching transcript sequences from other vertebrates and our RNA-seq data.

Immunofluorescence

All embryos were fixed for 20 min at room temperature in 4% paraformaldehyde, and subsequently blocked in 10% goat or donkey serum in PBST (phosphate-buffered saline and 0.5% Tween 20) for 2 hours at room temperature. Both primary and secondary antibody incubations occurred at 4°C for two nights in 10% goat or donkey serum, with four 1-hour washes in PBST at room temperature after primary and two 30-min washes in PBST after secondary antibody incubation. After imaging, whole-mount embryos were postfixed in 4% paraformaldehyde overnight at 4° before sectioning. The following primary antibodies were used: mouse immunoglobulin G1 anti-Pax7 (1:10; Developmental Studies Hybridoma Bank), rabbit anti-FLAG (1:1000; Sigma-Aldrich, F7425), and goat anti-GFP (1:500; Rockland, 600-101-215). The following secondary antibodies were used: Molecular Probes donkey or goat secondary antibody conjugated to Alexa Fluor 488, Alexa Fluor 568, or Alexa Fluor 647 (1:1000).

In situ HCR

All HCR was performed with probes designed by Molecular Technologies and following the published V3 protocol (). Twenty probe sets were used for all genes except Sox2, for which a 12-probe set was used. Before sectioning, embryos were postfixed in 4% paraformaldehyde overnight at 4°C. Since Chiwi expression was very low and had not been previously reported in the neural tube, we confirmed that the signal was not due to autofluorescence or background amplification by performing a negative control with only the odd HCR probe set, which imparted no signal when imaged under the same conditions as the even+odd probe experiments (fig. S8).

Sectioning

Cryosectioning was performed at a thickness of 18 μm on a Microm HM550 cryostat. Embryo preparation included fixation in 4% paraformaldehyde overnight at 4°C (either from live embryos or postfix processed embryos), followed 15% sucrose overnight at 4°C and 7.5% gelatin overnight at 37°C before mounting in silicone molds and snap freezing in liquid nitrogen.

Imaging and statistical analysis

All images were taken using a Zeiss Axio Imager M2 with an ApoTome.2. Quantification of HCR intensity in wild-type neural tube images was performed using the Plot Profile feature in Fiji (), which was then normalized to the length of the neural tube segment from ventral to dorsal midline and binned at 200. Fluorescent intensity was normalized to the highest value for each channel. Four to five cranial sections each from two HH9 embryos were analyzed. Whole-mount images of cranial neural crest stained for PAX7 were analyzed in Fiji by measuring the area of the migratory crest on the experimental (right) side of the embryo and dividing it by the area of the migratory neural crest on the left (control) side. Cell counts of PAX7-positive cells in sections of the cranial region were taken with the Analyze Particles feature in Fiji as previously described (). Fluorescence intensity for experimental quantification was measured from maximum intensity projections of Z-stack images by manually drawing regions of interest to measure average intensity and subtracting average intensity of background regions. Experimental values were then divided by control values from the same image. For fluorescent intensity and cell count quantification, three nonadjacent cranial sections were measured and averaged to create a representative value for each embryo. For all statistical analysis on images, a paired two-tailed Student’s t test was performed to compare two values (experimental and control) within single embryos.