Effector gene expression underlying neuron subtype-specific traits in the Motor Ganglion of Ciona

Abstract

The central nervous system of the Ciona larva contains only 177 neurons. The precise regulation of neuron subtype-specific morphogenesis and differentiation observed during the formation of this minimal connectome offers a unique opportunity to dissect gene regulatory networks underlying chordate neurodevelopment. Here we compare the transcriptomes of two very distinct neuron types in the hindbrain/spinal cord homolog of Ciona, the Motor Ganglion (MG): the Descending decussating neuron (ddN, proposed homolog of Mauthner Cells in vertebrates) and the MG Interneuron 2 (MGIN2). Both types are invariantly represented by a single bilaterally symmetric left/right pair of cells in every larva. Supernumerary ddNs and MGIN2s were generated in synchronized embryos and isolated by fluorescence-activated cell sorting for transcriptome profiling. Differential gene expression analysis revealed ddN- and MGIN2-specific enrichment of a wide range of genes, including many encoding potential “effectors” of subtype-specific morphological and functional traits. More specifically, we identified the upregulation of centrosome-associated, microtubule-stabilizing/bundling proteins and extracellular guidance cues part of a single intrinsic regulatory program that might underlie the unique polarization of the ddNs, the only descending MG neurons that cross the midline. Consistent with our predictions, CRISPR/Cas9-mediated, tissue-specific elimination of two such candidate effectors, Efcab6-related and Netrin1, impaired ddN polarized axon outgrowth across the midline.

1. Introduction

Genetic information is a major determinant of the morphological and physiological properties of individual neurons, as well as the connectivity and function of a nervous system (Baker et al., 2001; Bargmann, 1993; Manoli et al., 2006). Although these can be influenced by external cues and activity-dependent mechanisms (Thompson et al., 2017; Zhang and Poo, 2001), the clearest evidence for genetic determination of neurodevelopment comes from the stereotyped neural circuits that underlie innate behaviors (Kim and Emmons, 2017; Yamamoto and Koganezawa, 2013), or behavioral phenotypes caused by genetic mutations that result in changes to neuronal cell biology or connectivity (Branicky et al., 2014; White et al., 1992). Although a major focus of modern neuroscience is to dissect behavior at the level of individual genes, neurons, and specific synaptic connections (Luo et al., 2008), we have yet to decipher even the simplest nervous systems. Part of this difficulty stems from the fact that few organisms studied so far have proven tractable enough for the simultaneous investigation of gene function, neuronal activity, circuit connectivity and behavior.

The first synaptic connectivity network, or “connectome” (Sporns et al., 2005) to be fully mapped was that of Caenorhabditis elegans, a nematode that has only 302 neurons (White et al., 1986). Also a genetic and developmental model organism, C. elegans has delivered many key neurobiology breakthroughs, many of which were prompted by specific insights gleaned from the connectome (Chalfie et al., 1985; Jang et al., 2012). A second connectome, that of the larva of the tunicate Ciona intestinalis, was recently completed (Ryan et al., 2016, 2017, 2018). Vertebrates are the sister group to the tunicates within the chordate phylum (Delsuc et al., 2006), and this close genetic relationship has prompted the study of conserved, chordate-specific mechanisms of neurodevelopment in Ciona (Nishino, 2018). The central nervous system (CNS) of the Ciona intestinalis larva has only 177 neurons (Ryan and Meinertzhagen, 2019), making it the smallest described in any animal. This minimal, but comprehensive, connectome has neatly dovetailed with cell lineage and gene regulatory network studies performed on the closely related Ciona robusta (Cole and Meinertzhagen, 2004; Horie et al., 2018b; Ikuta and Saiga, 2007; Imai et al., 2009; Nicol and Meinertzhagen, 1988a, b; Sharma et al., 2019). The previously cryptic C. robusta was referred to as “C. intestinalis” in past studies, but recent evidence suggests splitting the two into distinct species (Pennati et al., 2015). However, the species’ embryonic cell lineages appear identical, and all the neurons identified in C. robusta have been identified in the C. intestinalis connectome (Ryan et al., 2016), suggesting a high degree of conservation of the CNS in the Ciona species complex (referred to as simply “Ciona” from now on).

Within this CNS, the development of Motor Ganglion (MG) has been studied in greatest detail. Situated at the base of the tail, just dorsal to the notochord, the neurons of the MG form a simple central pattern that drives the swimming behaviors of the larva (Nishino et al., 2010)(Fig. 1A). Of these, a core of only 7 bilaterally symmetric left/right pairs of neurons from the majority of the synaptic connectivity of the MG (Ryan et al., 2016)(Fig. 1B), and can all be traced to the A7.8 pair of blastomeres of the 64-cell stage embryo (Cole and Meinertzhagen, 2004; Navarrete and Levine, 2016)(Fig. 1C). We refer to these as the “core” MG, as additional cells traditionally assigned to the MG have either been shown to be quite removed from the motor network, e.g. AMG neurons, which serve as peripheral nervous system relay neurons (Ryan et al., 2018), or have yet to be visualized by light microscopy, e.g. Motor Neurons 3 through 5 and MG Interneuron 3 (Ryan et al., 2016).

Fig. 1.

Neuronal subtypes in the Motor Ganglion of Ciona. A) Diagram of a Ciona larva, with Motor Ganglion (MG) highlighted and magnified in inset. MG neuron morphologies are drawn according to Ryan and Meinertzhagen (2019). Larva illustration by Lindsey Leigh. B) Diagram of core MG connectome, adapted from Ryan et al. (2016). Colored arrows indicate major chemical synapses, dashed lines indicate major gap junctions (electrical synapses). Thickness of lines is proportional to cumulative depth of synaptic contact, except for neuromuscular synapses. Only synaptic connections shown are chemical synapses of cumulative depth >1 μm, and gap junctions of cumulative depth >6 μm. ACIN left/right symmetry is portrayed even though original connectome was missing a second ACIN on right side. ddN: descending decussating neuron, MGIN1: MG interneuron 1, MGIN2: MG interneuron 2, MN1: motor neuron 1, MN2: motor neuron 2, ACIN: ascending contralateral inhibitory neuron, Mus.: muscles. C) Cell lineage diagram of core MG neurons inferred from Cole and Meinertzhagen (2004) and Stolfi and Levine (2011), updated by results from Navarrete and Levine (2016) showing A10.64 identity of MN2, and from Nishitsuji et al. (2012) showing descent of ACINs from A9.29 (Cole and Meinertzhagen, 2004; Navarrete and Levine, 2016; Nishitsuji et al., 2012). Unresolved cell divisions in ACIN lineage indicated by dashed lines. D) ddN pair labeled with Dmbx reporter construct, arrow indicating axons crossing the midline. E) MGIN2 pair labeled with Pitx reporter construct. F) Supernumerary ddNs generated by early Notch/late FGF inhibition. This condition was used to isolate ddNs by FACS. G) Supernumerary MGIN2s generated by early FGF inhibition, condition used to isolate MGIN2s by FACS. Panels D, E, and G adapted from Stolfi and Levine (2011). Panel F adapted from Stolfi et al. (2011).

Within the core MG, each neuron is uniquely delineated by its invariant lineage, molecular profile, morphology, and synaptic connectivity (Cole and Meinertzhagen, 2004; Ryan et al., 2016, 2017; Stolfi and Levine, 2011). Here we focus on the comparison between two very different MG interneuron types: the descending decussating neuron (ddN) and MG Interneuron 2 (MGIN2). As their name implies, ddNs are the only neurons whose axons cross the midline before descending towards the tail (Fig. 1D)(Stolfi and Levine, 2011; Takamura et al., 2010). They receive synaptic inputs from peripheral nervous system (PNS) relay neurons and in turn synapse onto other MG neurons, each in particular forming synapses with their respective contralateral Motor Neuron 2 (MN2)(Ryan et al., 2017). The development and synaptic connectivity of the ddNs support homology with the Mauthner cells (M-cells) of the hindbrain of various fish and amphibian species. M-cells initiate the startle reflex during swimming, suggesting the ddNs could be mediating similar unilateral tail muscle flexions (“flicks”) in tunicates, in response to mechanosensory stimuli (Ryan et al., 2017). On the other hand, MGIN2s (Fig. 1E) are ipsilaterally-projecting descending interneurons whose most salient morphological trait is their extensive dendritic arborization (Stolfi and Levine, 2011). Like the ddNs, they also form conspicuous synapses with MN2s (Ryan et al., 2016), but receive synaptic inputs mainly from photoreceptor relay neurons and other interneurons of the brain, where the larval light- and gravity-sensing organs are located. Thus, these two MG neurons subtypes might modulate asymmetric swimming behaviors in response to sensory cues processed by distinct thigmotactic (ddNs) and phototactic/geotactic (MGIN2s) pathways (Kourakis et al., 2019; Rudolf et al., 2019; Salas et al., 2018).

Recent advances in single-cell RNAseq have allowed for identification of rare cell types in the Ciona nervous system (Cao et al., 2019; Horie et al., 2018b; Sharma et al., 2019), but the transcriptional profiles of ddNs and MGIN2s during their differentiation remains elusive. Here we use previously validated genetic tools to convert a majority of the Motor Ganglion to ectopic ddNs or MGIN2s, and isolate and profile these neurons at a time when their differentiation and morphogenesis is occurring (Stolfi and Levine, 2011; Stolfi et al., 2011). By analyzing and comparing the transcriptomes of isolated ddNs and MGIN2s, we identified differentially-expressed transcripts enriched in either neuron type and validated the ddN- or MGIN2-specific expression of several genes by mRNA in situ hybridization. We hypothesize that many of the genes thus identified are rate-limiting effectors of their unique morphological and physiological characteristics. More specifically, we identify centrosomal microtubule-stabilizing proteins and extracellular guidance cues as potential effectors, all expressed by the ddNs themselves, that together might direct the unique mode of polarization and axon outgrowth that we document in the ddNs.

2. Materials and methods

2.1. Ciona robusta collection and handling

Adult Ciona robusta (intestinalis Type A) were collected from Pillar Point Marina (Half Moon Bay, CA) or San Diego, CA (M-REP). Dechorionated embryos were obtained and electroporated as previously described (Christiaen et al., 2009a, b). Sequences of plasmids and in situ hybridization probe templates not previously published can be found in the Supplemental Sequences file. Fluorescent, whole-mount in situ hybridization and immunostaining were carried out as previously described (Beh et al., 2007; Ikuta and Saiga, 2007; Stolfi et al., 2011). Images were captured using Nikon, Leica, or Zeiss epifluorescence compound microscopes.

2.2. Whole embryo dissociation for FACS

Embryos were electroporated with the following combinations of plasmids, previously published and described (Stolfi et al., 2011). A mix of 60 μg Fgf8/17/18 > Su(H)^DBM + 60 μg Engrailed > dnFGFR + 61 μg Dmbx > Unc-76∷GFP + 58 μg Twist-reLated.b > RFP was to generate and isolate ectopic ddNs (“ddN” condition). Briefly, Fgf8/17/18 > Su(H)^DBM suppresses Notch signaling in the A9.30 lineage, resulting in duplication of the A11.120 cell, while Engrailed > dnFGFR suppresses FGF signaling in the anterior A9.30 lineage (A11.119 and A11.120), thus converting their progeny to ddNs (Stolfi et al., 2011). A mix of 60 μg Fgf8/17/18 > dnFGFR + 60 μg En > lacZ + 61 μg Vsx > Unc-76∷GFP + 58 μg Twist-related.b > RFP was used to generate and isolate ectopic MGIN2s (“IN2” condition) Briefly, Fgf8/17/18 > dnFGFR suppresses FGF signaling early in the A9.30 progenitor, converting the whole lineage into 4 MGIN2 cells on either side of the embryo (Stolfi and Levine, 2011; Stolfi et al., 2011). En > lacZ is a neutral construct used to normalize the total amount of plasmid DNA electroporated when compared to the ddN condition. Unc-76 tags (Dynes and Ngai, 1998; Stolfi and Levine, 2011) are used to label the entire cytoplasm, as opposed to untagged GFP, which tend to accumulate in the nucleus and is not efficiently transported into the axons. Additional unelectroporated embryos were raised in parallel, for sorted, unlabeled cells from whole embryos (“Whole” condition) Embryos were grown to 15.5 h post-fertilization (hpf) at 16 °C (Hotta Stage 23)(Hotta et al., 2007a) and dissociated in trypsin and Ca++/Mg + -free artificial sea water as previously described (Wang et al., 2018). We chose to dissociate right before ddN polarity inversion occurs (see Fig. 5), to maximize chance of observing upregulation of polarization effector genes. All plasmid amounts are given in μg per 700 μl of total electroporation solution.

Fig. 5.

Intracellular polarity dynamics in ddNs during contralateral axon outgrowth. A) MG cells labeled with Ebf > Unc-76∷mCherry and ddN labeled with Dmbx > Unc-76∷GFP, on one side of mosaic transgenic embryo. Arrow indicated ddN axon growth cone extending posteriorly. B) A9.30 lineage cell Golgi apparatuses labeled by Fgf8/17/18 > Galnact∷YFP (Stolfi et al., 2015a) and nuclei labeled by Fgf8/17/18 > H2B∷mCherry (Gline et al., 2009) showing intracellular polarity inversion in ddN between 15.75 and 16.5 h post-fertilization at 16 °C. C) Plot showing inversion of Golgi apparatus position in the ddNs, showing a shift from a more medial, apical position to a more lateral, basal position relative to cell nuclei. Each time point was analyzed in three independent replicates. In each replicate, 31 < n < 100. Full data set contained in Supplemental Table 4. D) Three-color labeling showing Golgi apparatuses, nuclei, and cell body before (15.5 h, left panels) and after (17.5 h, right panels) inversion of polarity and axon extension across the midline (dashed lines).

2.3. Fluorescence-activated cell sorting (FACS) and microarray profiling

FACS was performed on a Coulter EPICS Elite ESP sorter (Coulter Inc.), as previously described (Christiaen et al., 2008), with GFP + cells selected and RFP + cells counterselected. “Whole embryo” cells corresponding to cells dissociated from whole unelectroporated embryos were isolated in parallel. RNA extraction was performed using RNAqueous-micro kit (ThermoFisher) as per manufacturer protocol and analyzed by BioAnalyzer (Agilent). Total RNA amounts calculated to correspond to 1364 GFP + cells was used to prepare each cDNA sample, normalized based on the GFP + sample that yielded the lowest RNA concentration. Due to even lower RNA concentrations for GFP- “whole embryo” sorted cells, 833 cells were used for replicates 1 and 2, and 2083 for replicate 3, maxing out the volume of RNA solution allowed for cDNA synthesis. See Supplemental Table 1 for detailed information about each sample used for cDNA synthesis. We used the Ovation Pico WTA System (NuGen) and the Encore Biotin Module (NuGen) to prepare cDNAs and target probes for microrarray, following the manufacturer’s instructions and as previously described (Razy-Krajka et al., 2014). Microarray hybridization, washing, staining, and scanning were performed on a custom Affymetrix GeneChip (ArrayExpress accession A-AFFY-106) according to NuGen protocols and as previously described (Christiaen et al., 2008; Razy-Krajka et al., 2014).

Raw expression values (e.g. .CEL files) for each probe over three biological replicates for each condition (“IN2”, “ddN”, and “Whole”) were used to normalize and compute probe set expression estimates using the robust multi-chips analysis (RMA) algorithm (RMAexpress software) (Bolstad et al., 2003; Smyth, 2004). RMA estimates are available at https://osf.io/n7vr2/and Supplemental Table 2. RMA estimates were averaged across replicates and then converted to Log2 and pair-wise fold-change comparisons (LogFC) calculated (e.g. Log2[condition X RMA average] – Log2[condition Y RMA average) and P-values given by 1-tailed type 1 T-test. Probesets were matched to KyotoHoya (KH) gene models (Satou et al., 2008), or prior gene models when KH gene models did not appear to match a probeset sequence. See Supplemental Table 3 for our new annotation of correspondences between probesets and gene models.

2.4. CRISPR/Cas9-mediated mutagenesis

CRISPR/Cas9 was performed as established, using Foxa.a>Cas9 to drive expression in vegetal hemisphere lineages (A/B)(Di Gregorio et al., 2001), or Fgf8/17/18 > Cas9 to drive expression in the A9.30 lineage, though it is also active in mesenchyme, tail nerve cord, and tail tip neurons (Imai et al., 2009). Single-chain guide RNAs (sgRNAs) were designed and selected using CRISPOR (Haeussler et al., 2016), and vectors constructed as described (Gandhi et al., 2018). The sgRNA vectors were validated using the peakshift method (Gandhi et al., 2017), so-called due to overlapping “peaks” seen in Sanger sequencing of amplicons from the targeted locus amplified from pooled embryos; the overlapping “peaks” result from a mix of molecules containing indels of different sizes (Hsiau et al., 2019). We also validated sgRNAs by a novel “plasmid cutting assay”, using GFP fusions (Efcab6-r∷GFP, Netrin1∷GFP) driven by an Ebf driver that drives expression in Ebf + neuronal progenitors (Stolfi and Levine, 2011). Efficient sgRNAs will cleave all the GFP fusion-encoding plasmids, resulting in loss of GFP fluorescence. All sgRNA sequences and peakshift primers are in the Supplemental Sequences file. Efcab6r.157 sgRNA was validated as cutting Efcab6-related (Supplemental Fig. 1), and Netrin1.364 was validated as cutting Netrin1 (Supplemental Fig. 2). For Efcab6-related axon outgrowth assay, we electroporated 30 μg of Foxa.a>Cas9 90 μg + Dmbx > Unc-76∷YFP + 50 μg of U6>Efcab6r.157. The “control” for this experiment was identical but using 50 μg U6>Cesa4.1 (gift from Lionel Christiaen), designed to target Cellulose synthase, a gene expressed only in epidermis (Nakashima et al., 2004) and therefore not of predicted functional importance in the vegetal lineages where Foxa.a>Cas9 was to be expressed. For Netrin1 axon trajectory assay, we electroporated 70 μg Fgf8/17/18 > Cas9 + 90 μg Dmbx > Unc-76∷GFP + 70 μg of U6>Netrin1.364. For Netrin1 polarity assay, we electoporated 70 μg Fgf8/17/18 > Cas9 + 25 μg Fgf8/17/18 > Galnact∷YFP + 25 μg Fgf8/17/18 > H2B∷mCherry + 70 μg U6>Netrin1.364. The controls for these experiments were identical but using instead 70 μg of U6>ControlF + E plasmid (control “target” sequence that does not exist in Ciona genome: gctttgctacgatctacatt)(Stolfi et al., 2014). All plasmid amounts are given in μg per 700 μl of total electroporation solution.

3. Results and discussion

We used fluorescence-activated cell sorting (FACS) to isolate specific MG neuron types from synchronized Ciona robusta (intestinalis Type A) embryos, allowing us to profile their transcriptomes. We took advantage of different genetic manipulations to convert the majority of MG neurons into either supernumerary ddNs or supernumerary MGIN2s, which normally arise from a common progenitor at the neurula stage, the A9.30 pair of blastomeres of the neural plate (Fig. 1A)(Stolfi and Levine, 2011). We previously established that irreversibly inhibiting early Notch and late FGF signaling in the anterior cells of the A9.30 lineage converts their progeny into ectopic ddNs, all expressing the ddN marker Dmbx (Fig. 1F). In contrast, irreversibly inhibiting early FGF signaling converts the entire A9.30 lineage into ectopic MGIN2s, all expressing the MGIN2 marker Vsx (Fig. 1G). To generate ectopic ddNs or MGIN2s for fluorescence-activated cell sorting (FACS)-mediated cell type isolation, we recapitulated these perturbation conditions by co-electroporating synchronized embryos with specific combinations of plasmids (see Materials and methods for details). Ectopic ddNs were then isolated based on Dmbx > GFP expression, while ectopic MGIN2s were isolated by Vsx > GFP expression. In both conditions, co-electroporation with Twist-related.b(KH.C5.554)>RFP (Abitua et al., 2012) was used to counterselect mesenchyme cells potentially contaminating of our otherwise pure populations of MG neuron types. Embryos were dissociated at 15.5 h post-fertilization (hpf) at 16 °C. Control “whole embryo” cells were dissociated from un-electroporated embryos and subjected to FACS without selection. Total RNA was extracted from sorted cells, followed by cDNA synthesis and transcriptome profiling by microarray, in three independent biological replicates for each condition (see Materials and methods for details).

cDNA libraries were hybridized to Custom-designed Affymetrix GeneChip microarrays (ArrayExpress accession A-AFFY-106)(Christiaen et al., 2008) to quantify transcripts in sorted ddN, MGIN2, and mixed whole-embryo cells, and calculate the enrichment or depletion of ~21,000 individual transcript models in each cell population (Supplemental Table 2). Although the custom-made Affymetrix GeneChip microarrays we used were designed prior to the release of the most recent C. robusta genome assembly and associated transcript models (KyotoHoya, or KH)(Satou et al., 2008), we re-linked probesets to KH gene models where possible (see Materials and methods for details). Pairwise comparison of ddN and MGIN2 probeset expression values revealed a list of candidate genes that are differentially up- and/or down-regulated in either MG neuron subtype (Supplemental Table 2).

We detected 982 probesets that were significantly enriched (LogFC> 0.6, p < 0.05) in the ddNs vs. the MGIN2s, and 1245 probesets significantly enriched in MGIN2s vs. ddNs (LogFC < −0.6, p < 0.05). This is a slight overestimation of enriched genes, since many genes appear to be represented by more than one probeset, due to imprecise gene annotation. By perusing previously published expression patterns of some of the top differentially-expressed genes, we deduced that our FACS-isolated MGIN2 population contained contaminating cells that we failed to counterselect. MGIN2s appeared to be contaminated with epidermis midline cells, based on the presence of epidermal midline markers Dlx.c and KLf1/2/4 (Imai et al., 2004) in our top 25 MGIN2-enriched genes. This was likely due to the weak expression of Vsx > GFP in the dorsal epidermis midline (Supplemental Fig. 3). This contamination might explain the slightly higher number of genes enriched in the MGIN2 samples. Although Vsx > GFP is also expressed in MGIN1 (A13.474 cell)(Stolfi and Levine, 2011), such contamination was not a concern, due to the fact that we previously established that our molecular perturbation (Fgf8/17/18 > dnFGFR) converts the whole lineage to MGIN2 neurons, abolishing MGIN1 neurons (Stolfi et al., 2011).

We next selected a subset of the top differentially expressed genes in either MGIN2s (Table 1) or ddNs (Table 2) to validate by whole-mount fluorescent mRNA in situ hybridization (ISH). Some genes were selected on the basis of their statistically significant enrichment in either population, or based on their potential interest to us as candidates for follow-up functional studies.

Table 1.

MGIN2-enriched genes selected for validation by in situ hybridization.

Gene name	Gene model ID	LogFC* MGIN2 vs. ddN	p-value	Confirmed by ISH?
Chrnb	KYOTOGRAIL2005.771.2.1	3.7	0.009	YES
Kcna. a	KH.C1.232	3.1	0.002	YES
Slc24a4	KH.L132.15	2.9	0.016	YES**
Protocadherin. e	KH.C9.518	2.7	0.044	YES
Gabr.d	KH.C1.1254	2.6	0.014	YES
Grin	KH.S1211.3	2.2	0.066	YES
Ncs	KH.C1.1067	2.0	0.015	YES**

^*for genes represented by more than one probeset, we indicate the highest statistically-significant (p < 0.05) LogFC value. LogFC values are inverted relative to Table 1 and Supplemental Table 2, to denote enrichment in MGIN2 vs. ddN.

^**Slc24a4 and Ncs were enriched in MGIN2 and MN2 relative to ddN.

Red font highlights either p > 0.05, or “NO” in situ hybridization validation.

Table 2.

ddN-enriched genes selected for validation by in situ hybridization.

Gene name	Gene model ID	LogFC* ddN vs. MGIN2	p-value	Confirmed by ISH?
Lhx1/5	KH.L107.7	3.2	0.004	YES
Fam167	KH.C2.629	3.1	0.044	YES
Efcab6-related	KH.C1.1218	2.9	0.033	YES
Mnx	KH.L128.12	2.6	0.031	YES**
Saxo	KH.C10.475	2.6	0.008	YES
Calmodulin1-related	KH.C8.573	2.3	0.064	YES
Nckap5	KH.C9.229	2.2	0.009	YES
Ephrin a.d	KH.C3.716	2.1	0.013	YES**
Fibrillin	KH.C1.184	1.9	0.042	NO
Fibronectin-related	KH.C2.667	1.7	0.036	YES
Pou4	KH.C2.42	1.7	0.005	YES
Myosin10	KYOTOGRAIL2005.152.13.2	1.6	0.010	NO
Scna.a	KH.C9.462	1.6	0.029	YES**
Mpc	KH.C1.85	1.3	0.016	YES
Netrin1	KH.C12.72	1.3	0.052	YES

^*for genes represented by more than one probeset, we indicate the highest statistically-significant (p < 0.05) LogFC value. Complete dataset available in Supplemental Table 2.

^**Mnx was enriched in ddN, MN1, and MN2 relative to MGIN2. Ephrin a.d and Scna.a were enriched in ddN and MN1 relative to MGIN2.

Red font highlights either p > 0.05, or “NO” in situ hybridization validation.

3.1. MGIN2-enriched transcripts

While several ddN-specific markers have been previously described, there are relatively few known markers of MGIN2 other than Vsx. Although Vsx was indeed the gene that was most enriched in MGIN2s in our dataset (Table 1, Supplemental Table 2), this enrichment could be attributed to expression of the VSx > GFP plasmid used to sort these cells, which contains portions of the Vsx coding sequence.

The second-most enriched transcript we identified in MGIN2 was Chrnb (LogFC = 3.7), encoding a beta (non-alpha) subunit of the neuronal nicotinic acetylcholine receptor. Chrnb was not found in the current KyotoHoya genome assembly, but is represented by previous gene models. We designed an ISH probe based on the KYOTOG-RAIL.2005.771.2.1 gene model, which revealed highly specific expression in differentiating MGIN2s (Fig. 2A). According to the C. intestinalis connectome, MGIN2 receives synaptic inputs primarily from Photoreceptor Relay Neurons (prRNs), which in turn receive inputs primarily from Group I photoreceptors (Kourakis et al., 2019; Ryan et al., 2016). Recent experimental evidence suggests that the negative phototactic behavior of Ciona larvae is mediated by directional light detected Group I photoreceptors, while Group II photoreceptors mediate a light dimming “shadow response”. In the most recent model of the negative phototactic neural circuit, acetylcholine neurotransmission from prRNs onto the MG might provide the synaptic link between visual processing and motor control (Kourakis et al., 2019). Thus, acetylcholine receptors formed in part by Chrnb subunits might be mediating this crucial step in the visuomotor pathway of Ciona.

Fig. 2.

Candidate effector genes preferentially expressed in MGIN2 vs. ddN. In situ hybridization of neurotransmitter receptor subunit-encoding transcripts A) Chrnb (neuronal nicotinic acetylcholine receptor, beta/non-alpha subunit), B) Gabrd (GABA receptor subunit delta), C) Grin (NMDA-type ionotropic glutamate receptor). D) Larva electroporated with Slc17a6/7/8(Vglut)>RFP (magenta) labeling glutamatergic neurons and Vsx > GFP (green) labeling MGIN1 and MGIN2. Axons from unidentified glutamatergic neurons (possibly Apical Trunk Epidermal Neurons, or ATENs) extend and contact MGIN2, suggesting an unknown glutamatergic sensory relay input into the MG via MGIN2. E) In situ hybridization of transcripts encoding the Shaker-type voltage-gated potassium channel (Kcna.a). F) In situ hybridization of Protocadherin.e. OSP: oral siphon primordium. Arrows in all panels indicate MGIN2.

In addition to cholinergic transmission, GABAergic transmission has been proposed to play a minor role in the negative phototactic pathway, and a major role in the shadow response pathway (Brown et al., 2005; Kourakis et al., 2019). We detected enrichment of transcripts from the GABA receptor subunit delta-encoding gene Gabrd (KH.C1.1254, LogFC = 2.6) in MGIN2, which we confirmed by ISH (Fig. 2B). In the future it will be interesting to ascertain whether this localized expression confirms the requirement of GABAergic neurotransmission in negative phototaxis or whether it implicates a cryptic role for MGIN2 in the shadow response.

In contrast to acetylcholine and GABA, no role for direct glutamate neurotransmission onto the MG has been proposed. However, we identified MGIN2-specific enrichment of transcripts for the NMDA-type ionotropic glutamate receptor-encoding gene Grin, which was validated by ISH (KH.S2302.1, LogFC = 2.2, Fig. 2C). While most prRNs that provide synaptic input onto MGIN2 are predicted to be cholinergic and/or GABAergic, we detected the presence of putative glutamatergic neurons projecting to and making contact onto MGIN2 in larvae co-electroporated with Slc17a6/7/8(Vglut)>tagRFP (Horie et al., 2008; Stolfi et al., 2015a) and VSx > GFP reporter plasmids (Fig. 2D). The identity of these MGIN2-contacting brain neurons remains elusive, but could be Apical Trunk Epidermal Neurons (ATENs)(Imai and Meinertzhagen, 2007). According to the connectome (Ryan et al., 2016), MGIN2 also receives substantial input from other classes of brain neurons, including antenna relay neurons, which presumably relay positional information from the otolith-attached antenna cells, and coronet relay neurons, which presumably relay information of unknown nature from dopaminergic coronet cells (Moret et al., 2005). Larval swimming and attachment are modulated by gravity, which is lost in larvae lacking an otolith (Jiang et al., 2005; Tsuda et al., 2003). Furthermore, the shadow response can be altered by pharmacological treatments predicted to interfere with dopamine that is presumably released by the coronet cells (Razy-Krajka et al., 2012). Therefore, it is possible that Grin expression in MGIN2 is necessary for the modulation of swimming behavior by as of yet unidentified glutamatergic neurons.

Other potential effectors of MGIN2 function that we detected as highly enriched in MGIN2 and validated by ISH included Kcna.a (KH.C1.232, LogFC = 3.1, Fig. 2E), encoding a Shaker-related, voltage-gated potassium channel, closely related to TuKv1 (Ono et al., 1999) from Halocynthia roretzi, a distantly related tunicate species. Another candidate was Protocadherin.e (Pcdh.e, KH.C9.518, LogFC = 2.7, Fig. 2F). Protocadherins play numerous roles in morphogenesis, including being extensively implicated in dendrite morphogenesis and dendritic arborization (Keeler et al., 2015). Thus, its specific expression in MGIN2 could be related to its relatively elaborate dendrites, a morphological hallmark of MGIN2 (Stolfi and Levine, 2011)(Fig. 1B).

All the above were confirmed to be expressed in the MGIN2 by two-color double in situ hybridization with Vsx as the second probe (Supplemental Fig. 4) or immunostaining+in situ hybridization in the case of Kcna.a (Supplemental Fig. 5). We also found 2 genes that by ISH seemed to be enriched in both MGIN2 and MN2 (Supplemental Fig. 6), but not in other MG neurons: Slc24a4 (KH.L132.15, LogFC = 2.9) and Neuronal calcium sensor (Ncs, also known as Frequenin, KH.C1.1067, LogFC = 2.0), further suggesting that many effectors might not be strictly neuron subtype-specific, but that specific combinations of unique and shared effectors may precisely delineate MG neuron functions.

3.2. ddN-enriched regulators

Our basic strategy was validated by the presence of previously characterized ddN markers among top differentially expressed genes. Although Dmbx was indeed the most enriched transcript in the ddNs (LogFC = 6.1), this may have been a result of the probeset detecting portions of the Dmbx > GFP reporter that was used to select these cells by FACS. In addition to Dmbx, known ddN markers Lhx1/5 (LogFC = 3.2, Fig. 3A) Pou4 (LogFC = 1.7, Fig. 3B), and Hox1 (LogFC = 1.6) were among the statistically significant (p < 0.05), top 35 genes most enriched in ddNs relative to MGIN2s. These are all transcription factor-encoding genes and have been previously validated by ISH (Imai et al., 2009; Stolfi et al., 2011). We also found that another transcription factor-encoding gene, Mnx (Fig. 3C), and an Ephrin signaling molecule-encoding gene, Ephrin a.d (Efna.d, KH gene model identifier KH.C3.716, Fig. 3D) were also highly enriched in the ddNs relative to MGIN2s (LogFC = 2.6 and 2.1, respectively). Expression of Mnx and Efna.d had not been previously reported in ddNs, but our ISH validation confirmed their expression in the ddN (Fig. 3C,D). Mnx transcripts were also detected in MN1 and MN2, confirming previous ISH (Imai et al., 2009), while Efna.d was also expressed in the sister cell of the ddN (A12.240, which does not give rise to a differentiated neuron in the larval stage), and MN1.

Fig. 3.

Genes preferentially expressed in ddN vs. MGIN2. Immunostaining of Beta-galactosidase or H2B∷mCherry (magenta) in A9.30 lineage progeny nuclei (Fgf8/17/18 reporters, which also label mesoderm) coupled to in situ hybridization (green) of transcripts encoding transcription factors A) Lhx1/5, B) Pou4, C) Mnx, D) signaling molecule Ephrin a.d, and E) voltage-gated sodium channel subunit Scna.a. Arrows indicate ddNs. nnsc: non-neuronal sister cell of ddN (A12.240).

Beyond genes encoding transcription factors and signaling molecules, no other potentially ddN-specific transcripts have been previously assayed in detail by ISH. We therefore focused on these, hoping to identify candidate effectors of ddN development and function. For instance, we detected an enrichment for transcripts from the Sodium voltage-gated channel subunit alpha a gene (Scna.a, KH.C9.462, LogFC = 1.6), encoding a voltage-gated sodium channel orthologous to vertebrate NaV1 channels (Katsuyama et al., 2005; Nishino and Okamura, 2018; Okamura et al., 2005). By ISH, we found that Scna.a is upregulated specifically in the ddN and in MN1, but not in MGIN2 nor in any other MG neurons at around stage 23 (~15.5 hpf at 16 °C), when most are differentiating (Fig. 3E). Ciona NaV1 (encoded by Scna.a) possesses a short, chordate-specific “anchor motif” that in vertebrates is required for dense clustering it the axon initial segment (AIS) through ankyrin-mediated interactions with an actin-spectrin network (Garrido et al., 2003; Hill et al., 2008; Lemaillet et al., 2003). This clustering is crucial for rapid action potential initiation in the proximal axon (Kole et al., 2008). Vertebrate neurons that show dense AIS-specific clustering of NaV1 channels include M-cells and spinal motor neurons (Hill et al., 2008), proposed homologs of ddN and MN1/2, respectively (Ryan et al., 2017). The expression of Scna.a in Ciona ddN and MN1 suggests that the excitability of these neurons (and ultimately their function) might be similar to those of their vertebrate counterparts, regulated by a conserved, chordate-specific mechanism of subcellular compartmentalization of voltage-gated sodium channels.

3.3. Centrosome-enriched proteins are upregulated in ddNs

Two genes encoding homologs of centrosome-enriched, microtubule-stabilizing proteins were identified among the top ddN-expressed transcripts: Stabilizer of axonemal microtubules (Saxo, KH.C10.475, LogFC = 2.6) and Nck-associated protein 5 (Nckap5, KH.C9.229). We confirmed the upregulation of both genes in Ciona ddNs by ISH (Fig. 4A,B). Saxo is the sole C. robusta ortholog of human SAXO1 and SAXO2, previously known as FAM154A and FAM154B respectively. In humans, SAXO1 was found to bind to centrioles and stabilize microtubules (Dacheux et al., 2015). Similarly, Nckap5 is an ortholog of the closely related human paralogs NCKAP5 and NCKAP5L. In mammals, NCKAP5L encodes Cep169, a centrosome-enriched protein that also stabilizes microtubules (Mori et al., 2015a, 2015b). In our profiling, a single probeset detected enrichment of Saxo, but fragmented annotation of earlier versions of the C. robusta genome resulted in at least 5 independent probesets that we manually annotated as covering the updated Nckap5 gene model (KH.C9.229). These 5 probesets were all significantly enriched in ddNs relative to MGIN2s (LogFC 1.6-2.2, average 1.8).

Fig. 4.

Genes encoding centrosome-localized microtubule-binding proteins are enriched in ddNs. In situ hybridization of A) Nckap5, B) Saxo, and C) Efcab6-related. Arrows indicate ddNs. BTNs: Bipolar Tail Neurons. D) Efcab6-related∷GFP (driven by Ebf promoter) labeling the centrosome in a differentiating Bipolar Tail Neuron. F) Co-electroporation of Ebf > Efcab6-related∷GFP and Ebf > Galnact∷mCherry reveals association of centrosome with Golgi apparatus in the BTNs, as seen in vertebrate cells. Efcab6-related∷GFP labels a small punctum in the axon growth cone of a ddN as it extends across the midline. G) Efcab6-related∷GFP punctum also in the growth cone of the ddN axon turning and extending posteriorly down the tail. H) Diagram of Efcab6-related and Dmbx loci and shared cis-regulatory sequences. I-L) Expression of various Efcab6-related/Dmbx reporters in the ddN, color coded according to the diagram in panel H.

In addition to these orthologs of genes encoding previously characterized centrosomal proteins, we also noticed enrichment of KH.C1.1218 (LogFC = 2.9, Fig. 4C), encoding an EF-hand calcium-binding domain-containing protein (See Supplemental Sequences). The predicted protein is weakly similar to human EFCAB6 (also known as DJ-1 Binding Protein, or DJBP) but much more similar to many vertebrate genes annotated as “Efcab6-like”. However, KH.L125.4 (and not KH.C1.1218) is the predicted Ciona ortholog of human EFCAB6 according to the inParanoid ortholog prediction program (O’Brien et al., 2005). Alignment and phylogenetic analysis in MAFFT (Katoh et al., 2017) of several related genes suggests KH.C1.1218 closely resembles platypus Efcab6-like (XP_028932998.1), which lacks a clear 1-to-1 ortholog in human but still clusters with KH.L125.4 and human EFCAB6 (Supplemental Sequences). This suggests that the ortholog of KH.C1.1218 may have been lost in placental mammals. For these reasons, we refer to KH.C1.1218 as Efcab6-related, to indicate a close but unresolved phylogenetic relationship to human EFCAB6, as per the tunicate gene nomenclature guidelines (Stolfi et al., 2015b).

In human cells, EFCAB6 can inhibit the transcriptional activity of androgen receptor (Niki et al., 2003) through its association with DJ-1, a regulator of oxidative stress response and mitochondrial function (Canet-Avilés et al., 2004; Wang et al., 2012). A mouse knockout line from the Knock Out Mouse Phenotyping Program (KOMP², Jackson Laboratory) for Efcab6 shows inserted reporter gene staining in the developing hindbrain (https://www.mousephenotype.org/data/genes/MGI:1924877), suggesting a potentially conserved role in M-cell/ddN development. Given its seven EF-hand domains, we hypothesized that Efcab6-related might be localized to centrosomes, much like other EF hand-containing, calcium-binding proteins like Centrin or Calmodulin (Ito and Bettencourt-Dias, 2018). Indeed, an Efcab6-related∷GFP fusion was specifically localized to the centrosome of the bipolar tail neurons (Stolfi et al., 2015a)(Fig. 4D), co-localizing with the Golgi apparatus (Fig. 4E). In ddNs, we detected a small punctum of Efcab6-related∷GFP in the axon growth cone during extension over the midline (Fig. 4F) and later down the tail (Fig. 4G), but we were unable to visualize its localization earlier. These localization patterns hint at previously unrecognized roles for Efcab6-like proteins in regulating centrosome function, microtubule stabilization, and/or axon extension. Localization to both the centrosome and later to the growth cone might be linking centrosome position to axon outgrowth, with Efcab6-related in the growth cone required later to promote further axon growth or guidance.

Of further note, Ciona Efcab6-related and Dmbx are neighboring genes, arrayed in a “head to head” manner and transcribed in opposite directions (Fig. 4H). A reporter construct spanning this putative shared cis-regulatory module and the translation start site of Efcab6-related was sufficient to drive expression in the ddNs (Fig. 4I). This fragment is overlapping with the Dmbx upstream cis-regulatory region that was originally used to drive reporter expression in the ddNs (Stolfi and Levine, 2011)(Fig. 4J). It also overlaps the smaller fragment used in this study to FACS-isolated the ddNs initially (Stolfi and Levine, 2011)(Fig. 4K), as well as the minimal cis-regulatory element that contains the Pax3/7 binding site required for Dmbx activation (Stolfi and Levine, 2011; Stolfi et al., 2011)(Fig. 4L). Since the minimal Pax3/7-binding module is roughly equidistant and 5’ to both Dmbx and Efcab6-related (2.1 kb and 1.9 kb respectively, Fig. 4H), these two genes likely share a common regulatory element for ddN-specific expression. Given that Dmbx itself is a transcription factor that appears to inhibit proliferation and promote mitotic exit (Stolfi et al., 2011; Wong et al., 2015), this shared cis-regulatory element might be essential for coordination of genetically linked, but mechanistically distinct specification and morphogenetic processes in the ddNs.

The ddN-specific expression of known centrosome-enriched microtubule stabilizing proteins Nckap5 and Saxo, and the previously unrecognized centrosome marker Efcab6-related identified in this study, is interesting given the cellular processes that appear to underlie the unique contralateral axon projection of the ddNs. The ddN axon begins as an initial outgrowth that is oriented towards the neural tube lumen, extending across the midline (Fig. 5A). This is immediately preceded by a precisely timed, 180° re-orientation of the intracellular polarity of the cell, as visualized by the position of the Golgi apparatus, starting from an apical position apposing the neural tube lumen to a basal position near the neural tube basal lamina (Fig. 5B–D, Supplemental Table 4). In all other MG neurons, the Golgi apparatus remain on the apical side (lumen), and the direction of axon outgrowth is instead oriented away from the midline, resulting in an ipsilateral axon trajectory. A similar positioning of the Golgi apparatus on the opposite side of the nucleus relative to the site of axon extension was previously documented in migrating Ciona Bipolar Tail Neurons (BTNs), in which a precisely timed, 180° re-orientation of Golgi apparatus position also correlates with the direction of axon extension from an initially anterior orientation to a posterior one (Stolfi et al., 2015a). In these cases, the position of the Golgi apparatus is marker for centrosome position, which are tightly linked in vertebrate cells (Sütterlin and Colanzi, 2010) and in Ciona (Fig. 4E).

The relationship between centrosome/Golgi apparatus position and site of axonogenesis has been subject to long-running debates. Studies on cells in vitro suggested that the centrosome is positioned proximal to the site of axonogenesis (de Anda et al., 2005). However, more recent evidence suggests that in vivo, and depending on neuron type, centrosome position does not determine axon outgrowth and can even be distal to the site of axonogenesis (i.e. on the opposite side of the nucleus)(de Anda et al., 2010; Distel et al., 2010; Stolfi et al., 2015a; Zolessi et al., 2006). Since centrosome repositioning has been shown to depend on microtubule stabilization (Pitaval et al., 2017) the repositioning of ddN centrosomes that we observe might be effected in part by microtubule stabilization, driven by ddN-specific upregulation of Saxo and/or Nckap5 (see follow up experiments below). Saxo transcripts were also detected in migrating BTNs by ISH (Fig. 4A), and by single-cell RNAseq analysis (Horie et al., 2018a), hinting at the possible involvement of Saxo in centrosome repositioning in BTNs too.

3.4. ddNs upregulate the axon guidance cue Netrin1

We found that the major axon guidance molecule-coding gene Netrin1 (KH.C12.72, LogFC = 1.3)(Boyer and Gupton, 2018) is enriched in the ddN. We confirmed, by ISH, Netrin1 expression specifically in the ddNs, among MG neurons (Fig. 6A,B). Netrin1 is also highly expressed in the notochord (Hotta et al., 2000, 2007b), as is another axon guidance molecule, Sema3a (Kugler et al., 2008), supporting also a potential role of the notochord in guiding MG axons into the tail. Although Netrin was long thought to be exclusively a long-range cue (Kennedy et al., 1994; Serafini et al., 1994), tissue-specific targeting of Netrin1 in the vertebrate hindbrain and spinal cord recently revealed that the role of netrin1 protein in guiding midline crossing is consistent with its function as a short-range cue. More specifically, netrin1 from the floorplate of the developing hindbrain is dispensable for midline crossing (Dominici et al., 2017; Yamauchi et al., 2017), which is mostly regulated by netrin1 distributed along the axon path and derived from ventricular zone progenitor progenitors instead. In the spinal cord, netrin from both sources act synergistically as short- and long-range cues to guide midline crossing (Dominici et al., 2017; Moreno-Bravo et al., 2019; Varadarajan et al., 2017; Wu et al., 2019). Furthermore, UNC-6/netrin in C. elegans is instructive for neuronal polarization and defines the site of axonogenesis (Adler et al., 2006). These short-range functions might explain the potential of ddN-deposited Netrin1 to specify the nascent axon medially and to serve as a short-range cue to drive axon extension towards the neural tube lumen and across the midline.

Fig. 6.

Netrin1 expressed by ddNs and model for autocrine mechanism of ddN polarization. A) In situ hybridization of axon guidance cue-encoding Netrin1, showing expression in ddN (arrow) and notochord. C) Two-color in situ hybridization showing co-expression of Netrin1 (magenta) and known ddN marker Dmbx (green). D) Cartoon diagram describing proposed model for an intrinsic program for ddN polarization, based on autocrine deposition of Netrin1. Briefly: 1) MG neural precursors on one side of the embryo depicted with their basal side pointing laterally, attached to the basal lamina of the neural tube, and their apical side (marked by centrosome and Golgi apparatus) pointing medially, exposed to the lumen of the neural tube. 2) the ddN expresses and apically secretes Netrin1, inverting its own polarity relative to other MG cells. 3) The ddN axon subsequently grows medially, away from basal lamina and towards the midline, eventually crossing the midline. All other neurons extend their axons laterally along the basal lamina of the neural tube.

We also detected enrichment of transcripts from a gene encoding a relatively short, cysteine-rich predicted extracellular protein whose closest BLAST hits were the N-terminal heparin-binding and collagen-binding domains of Fibronectin-like proteins from various organisms (see Supplemental Sequences). We termed this gene Fibronectin-related (Fn-related, KH.C2.667, LogFC = 1.7). However, expression of Fn-related was detected by ISH in the sister cell of the ddN, A12.240, and its progeny, but not in the ddN itself (Supplemental Fig. 7). Fn-related thus may have been a false-positive in our profiling, due to contamination by A12.240 or A12.240-like cells.

Among other poorly studied genes or genes without any obvious or specific function in establishing ddN-specific traits that were confirmed by ISH were Fam167a (KH.C2.629, LogFC = 3.1), Calmodulin1-related (KH.C8.573, LogFC = 2.3) and Mitochondrial pyruvate carrier (KH.C1.85, LogFC = 1.3)(Supplemental Fig. 8A–C). Additionally, we could not detect with any certainty the expression of two candidate genes in the ddNs by ISH, Myosin10 and Fibrillin (Supplemental Fig. 8D and E). These negative results may have been due to poor probe design, which were prepared from short synthetic sequences (~500 bp). However, they also represent potentially false positives in the differential expression dataset, suggesting caution in interpreting such analysis devoid of any confirmatory ISH data.

3.5. Insights into ddN effector gene functions

To test the potential functions of certain ddN-specific effector genes, we used tissue-specific CRISPR/Cas9 (Stolfi et al., 2014) to knock them out in F0 embryos. First, we tested the function of Efcab6-related, since this gene family has been largely unstudied. We validated a single-chain guide RNA (sgRNA) that appeared to cut Efcab6-related most efficiently (Efcab6r.167, see Supplemental Sequences), as evidenced by short indels in vivo and elimination of Ebf > Efcab6-related∷GFP expression in a lineage-specific manner (Supplemental Fig. 1). Using Foxa.a>Cas9 to perform CRISPR/Cas9-mediated knockouts specifically in the vegetal lineages, from where the ddNs arise (Cole and Meinertzhagen, 2004), we assayed the effect of knocking out Efcab6-related on ddN development. Using Dmbx > GFP to visualize ddN axon outgrowth at the late tailbud stage (Stage 25, 12hpf/22 °C), we found that targeting Efcab6-related resulted in 45 of 100 (45%) ddNs with no clearly visible axon outgrowth (Fig. 7A,B). In control embryos (co-electroporated with Cesa4.1sgRNA vector instead, see Materials and methods), only 22 of 100 (22%) ddNs did not have a visible axon extending at this stage (Fig. 7A,B). We repeated this experiment, and found that Efcab6-related CRISPR resulted in 45 of 83 embryos with no visible axon (54%), compared to only 24 of 75 embryos in the control condition (32%) (Fig. 7B). We also quantified the lengths of randomly selected ddN axons from Efcab6-related and control animals, showing that embryos in the Efcab6-related CRISPR condition had a statistically significant (p = 0.039) reduction in axon length (Fig. 7C). Taken together, these results suggest that Efcab6-related is an important effector for axon outgrowth in the ddNs.

Fig. 7.

CRISPR/Cas9-mediated knockout of ddN effectors Efcab6-related and Netrin1. A) A-line-specific knockout of Efcab6-related using Foxa.a>Cas9 co-electroporated U6>Efcab6-r.157 and Dmbx > Unc-76∷YFP to visualize ddN axons. Top panel: “Control” condition performed side-by-side with Efcab6-related knockout, using instead 50 μg of U6>Cesa4.1. Axon growth cone indicated by arrow. B) Scoring percentage of embryos with visible axon growth in two biological replicates of the conditions depicted in (A). C) Plot of axon lengths measured in 26 embryos randomly selected from “Control” and Efcab6-r CRISPR conditions each. Statistical significance (p = 0.039) determined by one-tailed type 3 T-test. D) Knocking out Netrin1 in A9.30 lineage using Fgf8/17/18 > Cas9 co-electroporated with U6>Netrin1.364 and Dmbx > Unc-76∷GFP. Top panel: the control condition using U6>Control instead. Midlines indicated by dashed blue line. Nascent ddN axons indicated by arrows. E) Scoring percentage of embryos with visible axon decussation in two biological replicates of “Control” and Netrin1 CRISPR conditions. F) Scoring percentage of embryos with visible ddN polarity inversion (visualized by H2B:mCherry to label nuclei and Galnact∷YFP to label Golgi apparatus) in “Control” and Netrin1 CRISPR conditions. G) Representative “Control” embryo showing inverted polarity of ddN (arrow) at 12 h/20 °C. Bottom panels: two instances of ddN nucleus translocation across the midline seen in Netrin1 CRISPR embryos. Dashed lines indicate midline.

To test the potential role of Netrin1 as an intrinsic but extracellular positional cue that is required for ddN polarized axon outgrowth in an autocrine manner, we also used CRISPR/Cas9-mediated mutagenesis, but used instead Fgf8/17/18 > Cas9 in order to carry out the Netrin1 knockout specifically in the A9.30 lineage and not in other lineages expressing this gene (Fig. 7D). In one replicate, this resulted in only 42 of 97 (43%) ddNs with axons clearly projecting across the midline at 12 hpf/21 °C, compared to 27 of 43 (63%) in the control (co-electroporated with “Control” sgRNA vector instead). We repeated this, and in our second replicate we obtained only 14 of 65 (21%) ddNs visibly projecting across the midline in Netrin1 CRISPR, compared to only 45 of 71 (63%) in the control. Both replicates were scored blindly (Fig. 7D,E).

To test our model that Netrin1 may be an intrinsic cue deposited by the ddN to polarize itself, we assayed Golgi apparatus position in Netrin1 CRISPR mutants (Fig. 7F). In the control condition, we observed 51 of 77 (66%) embryos had at least one ddN with inverted Golgi position at 12hpf/20 °C, but in Netrin1 CRISPR, only 26 of 73 (36%) embryos did. These numbers are inversely correlated with the Golgi positions scored in wild-type embryos at the equivalent stage (16.5 hpf/16 °C, Fig. 5C). In at least two CRISPR mutant embryos, a ddN was observed to translocate its nucleus towards the midline, instead of away from the midline, a highly unusual phenotype that we have never observed before (Fig. 7G). This may reflect a non-cell-autonomous attractive effect of Netrin1 secreted from the ddN in the non-electroporated half of these embryos (thus also unlabeled by the fluorescent reporters), though a more careful investigation will be needed to parse cell-autonomous versus non-cell-autonomous effects of lineage-specific Netrin1 knockouts. Nonetheless, these data suggest that Netrin1, expressed by the ddN itself, is key to its characteristic polarization and subsequent axon outgrowth towards and across the midline.

4. Conclusions

Here we have used molecular perturbations, embryo dissociation, and FACS to isolate specific neuronal progenitors in the developing Ciona MG, and compared their transcriptomes by microarray. Specifically, we have compared ddN and MGIN2 neurons, which are predicted by the connectome to serve as major conduits for various sensory modalities to modulate the larva’s swimming and escape response behaviors. Our transcriptome profiling points to possible effectors of ddN/MGIN2-specific electrophysiological properties (e.g. Scna.a, Kcna.a), morphology (e.g. Saxo, Nckap5, Efcab6-related, Pcdh.e), and functional connectivity (e.g. Chrnb, Gabrd, Grin). These now comprise an attractive set of targets for tissue-specific CRISPR/Cas9 somatic knockouts (Gandhi et al., 2017), in future functional studies of the gene regulatory networks regulating neurodevelopmental processes in the Ciona larva. We show that CRISPR/Cas9-mediated knockout of Efcab6-related and Netrin1 produce ddN axon outgrowth defects consistent with the predicted involvement of these genes in ddN polarization and extension across the midline. Given the polarity defects observed in Netrin1 CRISPR embryos, we predict that Netrin1 acts upstream of axon specification to polarize the ddNs 180° relative to neighboring, non-Netrin1-expressing cells. Given the different phenotype observed in Efcab6-related CRISPR embryos, in which ddNs more often failed to grow an axon or had shorter axons, we predict that Efcab6-related instead effects axon growth, possibly through its physical interaction with the centrosome and/or growth cone.

Our results suggest that both intracellular and extracellular effectors (e.g. Efcab6-related and Netrin1, respectively) are part of the same intrinsic program that is deployed largely through ddN-specific transcriptional activation, blurring the line between fully extrinsic or intrinsic modes of neuronal polarization or axon guidance.