Zebrafish oligodendrocyte lineage cells use the postsynaptic protein Gephyrin to myelinate GABAergic axons and limit myelin sheath growth

Abstract

Axon myelination can tune neuronal circuits through placement and modulation of different patterns of myelin sheaths on distinct types of axons. How myelin formation is coordinated on distinct axon classes remains largely unknown. Recent work indicates neuronal activity and vesicle release promote myelin formation, and myelin-producing oligodendrocytes express canonical postsynaptic factors that potentially facilitate oligodendrocyte-axon interaction for myelin ensheathment. Here, we examined whether the inhibitory postsynaptic scaffold protein Gephyrin (Gphn) mediates selective myelination of specific axon classes in the larval zebrafish. Consistent with this possibility, Gphn was enriched in myelin on GABAergic and glycinergic axons. In gphnb deficient larval oligodendrocytes, myelin sheaths were longer and the frequency of myelin placement on GABAergic axons was reduced. Collectively, our results indicate that oligodendrocyte lineage cells use Gphn to promote myelin formation on GABAergic axons and limit myelin sheath length.

Myelin forms on axons of both inhibitory and excitatory neurons. Here the authors show that a classical neuronal postsynaptic protein, Gphn, functions in oligodendrocytes to promote inhibitory axon myelination in zebrafish.

Introduction

Oligodendrocytes (OLs) are glial cells in the central nervous system (CNS) that produce myelin, a lipid-rich membrane that wraps around axons to provide metabolic and trophic support and increase action potential velocity. A single OL can myelinate dozens of axons simultaneously¹, including different classes of axons defined by distinct neurotransmission profiles. Because different neuron types have different axon lengths, firing rates, and energetic demands, differences in the amount and composition of myelin on axons potentially contribute to the functions of distinct neural circuits². Therefore, OLs and their myelin sheaths may be uniquely positioned to modulate neural circuit output by regulating the timing, strength, or frequency of circuit signals^3–5.

Neural circuits require a balance of excitatory and inhibitory influences to achieve regulated output, such as the coordinated locomotion generated by the spinal cord. Canonically, glutamatergic neurons provide excitatory input and γ-amino butyric (GABAergic) and glycinergic neurons provide inhibitory influence on circuit output^6–8. Critically, glutamatergic, GABAergic, and glycinergic neurons signal through unique molecular machinery where their axon terminals create synapses with the appropriate postsynaptic terminal. Postsynaptic scaffold proteins provide specificity for synapse formation by anchoring receptors and cell adhesion molecules that are enriched at unique synapses. Postsynaptic Density 95 protein (PSD95) is the primary scaffold protein at excitatory glutamatergic synapses, whereas Gephyrin (Gphn) is the postsynaptic scaffold at inhibitory GABAergic and glycinergic synapses^9–13. This specificity of synaptic communication is necessary in complex circuits to coordinate neuronal firing and generate functional behaviors such as locomotion.

Remarkably, OLs produce myelin sheaths with variable lengths and thicknesses on individual axons^14,15, and myelin patterns on distinct classes of axons vary across neuron type and brain region^16–20. What mechanisms might convey specificity in myelin formation on distinct axon classes? One possibility is that OLs engage with axons using mechanisms similar to synapse formation, where a myelin sheath contacts an axon at an axo-sheath interface. Several findings support this possibility. First, neuronal activity promotes myelin formation through vesicle release along the axon^19,21–23. This vesicular release is accompanied by axonal Ca²⁺ events at sites where myelin growth will subsequently occur¹⁹. Second, gene expression profiling studies show that OL lineage cells (OLCs) express many genes that encode postsynaptic proteins such as PSD95 and Gphn^24–27. Third, interfering with postsynaptic protein function in OLs disrupts myelin formation and maintenance^28,29. And fourth, an OL precursor cell (OPC)-specific knockout of GABA_AR γ2 altered myelin profiles on fast-spiking, GABAergic PV interneurons and their subsequent firing rate without impacting the myelin or firing rate of neighboring, glutamatergic spiny stellate cell interneurons³⁰. Thus, we sought to understand whether distinct axon classes use unique mechanisms for myelination. To this end, we hypothesized that OLs and their individual myelin sheaths use postsynaptic signaling machinery to coordinate axon identity-dependent myelination.

In this study, we used larval zebrafish to investigate whether Gphn function mediates myelin sheath formation on specific classes of axons defined by neurotransmitter phenotype. We first used transgenic reporters for glutamatergic, GABAergic, and glycinergic neurons to show that each neuronal class is myelinated in the developing spinal cord. Consistent with our hypothesis, Gphn protein localizes to myelin during development and is enriched in sheaths that wrap GABAergic and glycinergic axons. With the loss of gphnb function, myelin sheaths were abnormally long, indicating that Gphn contributes to a mechanism that limits sheath growth. Additionally, in gphnb mutant larvae, myelin was displaced from GABAergic axons onto glutamatergic axons, and this difference was evident at the earliest stages of myelination. Moreover, oligodendrocyte-specific loss of gphnb also reduced the number of myelinated GABAergic axons. Timelapse imaging revealed no difference in nascent myelin sheath turnover in wild-type and mutant larvae, but there were fewer total nascent sheaths on GABAergic axons, therefore suggesting that Gphn also helps guide GABAergic axon selection for myelination. Together, these observations illustrate that Gphn plays an important role in selecting inhibitory, GABAergic axons for myelination and subsequently modulating myelin sheath length. Prior investigations using electrophysiology established that functional synapses with neurons can be detected in OPCs but not oligodendrocytes^31,32. Our data raise the possibility that as oligodendrocytes differentiate, they repurpose postsynaptic proteins to establish unique, axon class-specific myelin profiles, potentially coordinating neural circuit function.

Results

Glycinergic, GABAergic, and glutamatergic axons are myelinated in the developing spinal cord

As a first step toward investigating axon class-specific myelination, we examined myelination of larval zebrafish spinal cord axons defined by neurotransmitter phenotype. To do this, we used combinations of transgenic reporters to simultaneously visualize class-specific axons and myelin (Fig. 1). With this approach, we found that glutamatergic axons (Fig. 1a), GABAergic axons (Fig. 1b), and glycinergic axons (Fig. 1c) were myelinated. For each neuronal class, myelinated axons occupied both dorsal (Fig. 1a–c dorsal and orthogonal views) and ventral spinal cord (Fig. 1a–c ventral views). Additionally, large myelinated glutamatergic and glycinergic axons occupied positions near the midline of the spinal cord (Fig. 1a, c). Myelinated GABAergic axons were typically smaller in diameter and occupied more lateral positions (Figure b). Thus, OLs myelinate distinct classes of axons in the dorsal and ventral white matter tracts of the zebrafish larval spinal cord.

Fig. 1. Glycinergic, GABAergic, and glutamatergic axons are myelinated in the developing spinal cord.

Representative images of transverse spinal cord sections of 4 dpf larvae expressing the transgene Tg(sox10:mRFP) (myelin, purple) along with a Tg(slc17a6:eGFP) (glutamatergic axons, green); c Tg(gad1b:eGFP) (GABAergic axons, green); or e Tg(slc6a5:eGFP) (glycinergic axons, green). b, d and f Magnification of boxed regions showing individual myelinated axons in the dorsal and ventral myelin tracts, with orthogonal views of dorsal myelinated axons for each neuron class. Dashed outlines indicate myelinated axons in orthogonal views and enlarged panels. a, c and e scale bars = 5 μm. b, d and e scale bars = 2 μm, orthogonal view scale bars = 5 μm.

Gephyrin protein localizes to some, but not all, myelin sheaths

OLCs express many genes encoding postsynaptic proteins, some of which appear critical for myelination²⁸. Recent work showed that the postsynaptic scaffold protein Gphn localizes to OPC processes²⁹ and OLs continue to express Gphn at myelinating stages^24,25. We predicted that if Gphn mediates axon wrapping by myelin membrane, then it would occupy nascent myelin sheaths. To test this prediction, we used immunohistochemistry to detect Gphn in transgenic larvae expressing a membrane-tethered myelin reporter (Supplementary Fig. 1a–f). This revealed Gphn localization within myelin in both dorsal (Supplementary Fig. 1a, d, dorsal views) and ventral tracts (Supplementary Fig. 1a, d, ventral views). Additionally, we determined that the volume of individual Gphn puncta and overall punctal density within myelin increased from 4 dpf to 7 dpf (Supplementary Fig. 1g, h). These data show that Gphn progressively accumulates in myelin sheaths during development, supporting the notion that Gphn contributes to myelin sheath formation.

To detect Gphn in living animals, we modified a genetically encoded Gphn intrabody³³ to analyze the sub-cellular localization in OLs at 4 dpf (Fig. 2a, c) and 7 dpf (Fig. 2b, d). Importantly, OLs of larvae lacking Gphn function (see below) had significantly fewer Gphn.FingR puncta than OLs of wild-type larvae, validating the specificity of this labeling method (Supplementary Fig. 2). Like our findings with Gphn immunohistochemistry, Gphn.FingR puncta per OL increased from 4 dpf to 7 dpf (Fig. 2e), an effect that was not influenced by OL sheath number (Fig. 2f). At both 4 dpf and 7 dpf, not all sheaths contained Gphn.FingR signal, therefore, to examine sub-cellular localization patterns, we quantified the number of Gphn.FingR puncta per sheath. This showed that individual myelin sheaths had different amounts of Gphn.FingR labeling (Fig. 2a–d), with some sheaths containing a high density of puncta (yellow boxes) and others with fewer puncta (magenta boxes). The frequency distribution shifted dramatically between 4 and 7 dpf, reflecting an overall increase in Gphn.FingR expression by 7 dpf (Fig. 2g). Notably, some sheaths lacked the Gphn reporter at both 4 dpf and 7 dpf (Fig. 2g). These results show that Gphn differentially accumulates in nascent myelin sheaths, raising the possibility that it mediates myelination of distinct classes of axons.

Fig. 2. Variable localization of Gphn in myelin sheaths during development.

a, b Fluorescent images of the individual OLs expressing mbpa:eGFP-CAAX (myelin sheaths, purple) and a genetically encoded Gphn intrabody (zfmbpa:Gphn.FingR-mScarlet-IL2RGTC-KRAB(A), orange) at 4 dpf and 7 dpf. c, d Three dimensional surface models of the individual OLs in (a) and (b), respectively, illustrating Gphn intrabody signal within mbpa:eGFP-CAAX+ membranes (purple) at 4 dpf and 7 dpf. Yellow boxes indicate an individual myelin sheath with abundant Gphn intrabody signal, while magenta boxed regions indicate an individual sheath with little Gphn intrabody signal. Quantification of e Gphn puncta counts per OL (4 dpf min = 165.0, 1st quartile = 265.5, median = 536.0, mean = 499.7, 3rd quartile = 715.5, max = 841.0; 7 dpf min = 445.0, 1st quartile = 877.8, median = 1277.0, mean = 1370.7, 3rd quartile = 1821.5, max = 2674.0), p-value = 0.0011, 95% CI: −1301 - −291. f Sheath number per OL (4 dpf min = 8.0, 1st quartile = 15.5, median = 17.0, mean = 17.6, 3rd quartile = 22.0, max = 23.0; 7 dpf min = 7.0, 1st quartile = 10.3, median = 13.0, mean = 14.1, 3rd quartile = 19.0, max = 22.0), p-value = 0.14, 95% CI: −1.0–9.0. g Frequency distribution of Gphn puncta per sheath at 4 dpf (orange) and 7 dpf (purple), with a dashed line representing the mean (4 dpf mean±SD = 28.5±20.7, 7 dpf mean±SD = 97.2±83.5). Puncta per OL significance determined by two-sided Wilcoxon rank sum test, and sheaths per OL by two-sided Student’s T-Test, with p < 0.05 is considered significant. 4 dpf n = 11 OLs and larvae; 7 dpf n = 10 OLs and larvae. Scale bars = 5 μm. dpf days post fertilization, Gphn Gephyrin, OL oligodendrocyte, SD standard deviation. Source data are provided as a Source Data file.

Previously, we showed that PSD95, a canonical postsynaptic scaffolding protein at glutamatergic synapses, localizes to nascent myelin sheaths. Because Gphn localizes to GABAergic and glycinergic synapses, the presence of these unique scaffolds in specific OLC processes and myelin sheaths^28,29 could provide a mechanism for selective, axon class-specific myelination. Therefore, to determine whether Gphn and PSD95 occupy the same or different myelin sheaths we examined dorsal OLs that simultaneously expressed Gphn and PSD95 intrabodies (Supplementary Fig. 3a–d). At 7 dpf, the density of each type of scaffold varied among myelin sheaths. In particular, some sheaths contained both PSD95 and Gphn puncta, some had more PSD95 puncta than Gphn puncta (Supplementary Fig. 3a–d, blue boxes), and others had more Gphn puncta than PSD95 puncta (Supplementary Fig. 3a–d, orange boxes). Generally, OLs that had higher levels of Gphn also had higher levels of PSD95 (Supplementary Fig. 3e). The amount of Gphn and PSD95 per sheath also correlated, but expression levels were highly variable within individual sheaths and between individual OLs (Supplementary Fig. 3f). Altogether, these data indicate that Gphn and PSD95 are not uniformly distributed among newly formed myelin sheaths. Instead, myelin sheaths contain different amounts of these scaffold proteins, supporting the possibility that they are equipped to mediate myelin sheath interactions with distinct axon subtypes.

Myelin on GABAergic and glycinergic axons has more Gphn than myelin on glutamatergic axons

Because Gphn functions at inhibitory neuronal synapses, we predicted that it localizes to myelin sheaths on GABAergic and glycinergic axons. To test this prediction, we investigated Gphn localization within myelin on glutamatergic, GABAergic, and glycinergic neurons marked by transgenic reporter gene expression (Fig. 3a–c). We used immunohistochemistry to identify Gphn puncta in myelin on axons corresponding to each neuronal class (Fig. 3a–c, yellow arrows). Remarkably, myelin on GABAergic and glycinergic axons contained significantly more Gphn than myelin on glutamatergic axons (Fig. 3d). This further supports our model that postsynaptic proteins mediate OL interactions with specific classes of axons.

Fig. 3. Enrichment of Gphn protein in myelin on GABAergic and glycinergic axons.

Representative transverse images of a Tg(slc17a6:eGFP); Tg(mbp:mCherry-CAAX) larva, b Tg(gad1b:eGFP); Tg(mbp:mCherry-CAAX) larva, and c Tg(slc6a5:eGFP); Tg(mbp:mCherry-CAAX) larva processed to detect Gphn at 7 dpf. Magnified panels of boxed regions indicate Gphn puncta (orange) in myelin (purple) on the respective axon reporter (green). In all magnifications, yellow arrows point to Gphn puncta in myelin on the labeled axon class. d Quantification of the percent of Gphn puncta in myelin wrapping each axon class, using the Kruskal-Wallis test for global significance followed by Bonferroni-corrected Wilcoxon multiple comparisons (glutamatergic min = 17.8, 1st quartile = 21.2, median = 23.2, mean = 24.5, 3rd quartile = 28.3, max = 30.9; GABAergic min = 36.0, 1st quartile = 38.6, median = 39.9, mean = 40.6, 3rd quartile = 42.7, max = 47.9; glycinergic min = 34.0, 1st quartile = 37.0, median = 39.5, mean = 39.9, 3rd quartile = 42.0, max = 47.3), global p-value = 2.03e−05; glutamatergic vs GABAergic p-value = 8.5e−06, glutamatergic vs glycinergic p-value = 8.5e−06, GABAergic vs glycinergic p-value = 1. p < 0.05 is considered significant. All groups n = 11 larvae. Scale bars = 5 μm, zoom panel scale bars = 1 μm. Gphn = Gephyrin. Source data are provided as a Source Data file.

Gphn regulates myelin sheath length

To investigate Gphn function in myelination, we created loss-of-function gphn mutants using CRISPR/Cas9 genome editing. Zebrafish have 2 gphn paralogs, gphna and gphnb, likely from an ancestral genome duplication event^34,35. By simultaneously targeting gphna and gphnb we generated 2 lines with mutations in both genes: gphna^co91; gphnb^co94 and gphna^co92; gphnb^co95 (Supplementary Fig. 4). Homozygous double mutant larvae do not survive past 4 dpf, likely due to the role Gphn plays in molybdenum cofactor (MoCO) biosynthesis throughout the body³⁶. We therefore segregated gphna and gphnb alleles by outcrossing and then used immunohistochemistry to detect Gphn in larvae with homozygous mutations of each paralog. Whereas gphna mutant larvae expressed Gphn in the spinal cord, gphnb mutant larvae expressed very little (Supplementary Fig. 4d–f). This is consistent with RNA in situ hybridization data showing that gphnb expression is specific to the central nervous system in zebrafish, whereas gphna is globally expressed^37,38. These data indicate that gphnb mutation mostly eliminates Gphn from the zebrafish nervous system, and we therefore used gphnb mutant larvae for our experiments.

To test whether Gephyrin contributes to myelination, we used mbpa:eGFP-CAAX expression to label individual OLs in gphnb mutant and wild-type larvae and analyzed them at different developmental stages (Fig. 4a–f). We focused on dorsal OLs because we could assay all sheath characteristics from individual cells. Neither sheath length nor sheath number differed between homozygous gphnb mutant and wild-type larvae at 3 dpf (Fig. 4g, h) or 4 dpf (Fig. 4g, h). However, by 7 dpf, myelin sheaths were significantly longer in gphnb mutant larvae than in wild-type larvae (Fig. 4g, h). To confirm that this myelin phenotype is due to loss of Gphnb function and not an off-target mutagenic event, we performed a complementation test using two gphnb alleles derived from independent founders (Supplementary Fig. 4g). At 7 dpf, there was no difference in sheath length or number between trans-heterozygous gphnb^co94/95 mutant larvae and homozygous gphnb^co95 mutant larvae (Supplementary Fig. 4h, i). Together, these data indicate that Gphn limits sheath growth.

Fig. 4. Loss of gphnb results in long myelin sheaths across development.

Representative images of mosaically labeled OLs of wild-type larvae at a 3dpf, c 4 dpf, and e 7dpf. Representative images of individual OLs in gphnb mutant larvae at b 3 dpf, d 4 dpf, and c, 7 dpf. Statistical comparisons of sheath characteristics at 3 dpf, 4 dpf, and 7 dpf in wild-type (teal) and gphnb mutant (purple) larvae for g individual sheath length (μm) (3 dpf, wildtype min = 2.2, 1st quartile = 18.8, median = 26.9, mean = 28.8, 3rd quartile = 37.1, max = 65.2; gphnb min = 1.2, 1st quartile = 18.8, median = 27.8, mean = 29.0, 3rd quartile = 38.6, max = 77.4; p-value = 0.879) (4 dpf, wildtype min = 2.0, 1st quartile = 24.2, median = 33.3, mean = 35.3, 3rd quartile = 44.5, max = 81.4; gphnb min = 4.6, 1st quartile = 25.6, median = 35.7, mean = 37.7, 3rd quartile = 48.0, max = 94.8; p-value = 0.075) (7 dpf, wildtype min = 6.0, 1st quartile = 27.1, median = 38.2, mean = 39.1, 3rd quartile = 49.9, max = 94.9; gphnb min = 8.0, 1st quartile = 30.4, median = 40.8, mean = 42.6, 3rd quartile = 51.9, max = 111.7; p-value = 0.009) and h total sheaths per cell (3 dpf, wildtype min = 8.0, 1st quartile = 11.0, median = 12.0, mean = 13.1, 3rd quartile = 14.0, max = 23.0; gphnb min = 7.0, 1st quartile = 11.0, median = 13.5, mean = 13.4, 3rd quartile = 14.8, max = 23.0; p-value = 0.829) (4 dpf, wildtype min = 10.0, 1st quartile = 13.5, median = 16.0, mean = 16.1, 3rd quartile = 18.5, max = 22.0; gphnb min = 6.0, 1st quartile = 11.0, median = 13.0, mean = 14.0, 3rd quartile = 16.8, max = 25.0; p-value = 0.192) (7 dpf, wildtype min = 7.0, 1st quartile = 9.0, median = 12.5, mean = 13.1, 3rd quartile = 15.5, max = 25.0; gphnb min = 8.0, 1st quartile = 11.0, median = 16.0, mean = 15.4, 3rd quartile = 17.0, max = 29.0; p-value = 0.126). Representative images of individual OLs expressing Cas9 and i no gRNA control and j gphnb gRNAs at 7 dpf. Quantification of k individual sheath length (control min = 3.9, 1st quartile = 22.0, median = 30.3, mean = 31.1, 3rd quartile = 38.8, max = 76.4; gphnb min = 6.1, 1st quartile = 27.6, median = 37.7, mean = 38.4, 3rd quartile = 47.4, max = 71.6; p-value = 0.006) and l sheaths per cell (control min = 8.0, 1st quartile = 13.5, median = 18.0, mean = 18.3, 3rd quartile = 22.0, max = 32.0; gphnb min = 7.0, 1st quartile = 10.3, median = 13.0, mean = 12.5, 3rd quartile = 14.8, max = 17.0; p-value = 0.024) for control (teal) and gphnb gRNA (purple) conditions. a–h 3 dpf: wildtype n = 13 OLs and larvae, gphnb n = 30; 4 dpf: wildtype n = 15, gphnb n = 18; 7 dpf: wildtype n = 16, gphnb n = 21; i–l control n = 14 OLs and larvae, gphnb gRNA n = 14 OLs and larvae. g, h significance determined by a Type-III sum of squares test followed by Bonferroni-corrected multiple comparisons; k significance determined by a non-linear mixed effects model, and l significance determined by a two-sided Wilcoxon Rank Sum Test. p < 0.05 is considered significant. dpf days post fertilization. Scale bars 10 μm. Source data are provided as a Source Data file. min. =, 1st quartile =, median =, mean =, 3rd quartile =, max =.

Because neurons express Gphn, we performed two complementary experiments to test whether myelin sheath length is limited by Gphn function in OLs. First, we expressed human GPHN in OLs of gphnb mutant larvae using mbpa:GPHN-2A-mApple-CAAX (Supplementary Fig. 5a, b). This rescued individual sheath length (Supplementary Fig. 5), with no change in sheath number (Supplementary Fig. 5d). Second, we injected embryos with a plasmid engineered to express two gphnb sgRNAs that target exon 1 and exon 2 under control of U6 promoters, and Cas9 coupled with membrane-tethered mScarlet fluorescent protein driven by mbpa regulatory DNA²⁹. In this experiment, a subset of OLs, revealed by mScarlet, express Cas9 to promote sgRNA-guided mutagenesis of gphnb. As a control, we injected a plasmid that is identical to the experimental plasmid except that it lacks the sgRNA-encoding sequences. OLs expressing the gphnb sgRNA plasmid had longer myelin sheaths than OLs expressing the control plasmid, providing important validation that Gphn function in OLs limits sheath length (Fig. 4i–k). In contrast to OLs of gphnb mutant larvae, OLs expressing the gphnb sgRNA plasmid formed fewer myelin sheaths than control OLs (Fig. 4l). Together, these data provide strong evidence that Gphn functions in OLs to mediate myelin sheath formation.

The excess myelin sheath length of gphnb mutant larvae is not driven by hyperactivity

Gphn anchors GABA and glycine receptors at inhibitory neuronal synapses. Because inhibition is critical to curtail excitatory output, we predicted that gphnb mutant larvae would be hyperactive as a result from reduced inhibition of locomotive circuits. Therefore, we tracked swimming behavior, and at 7 dpf (Fig. 5a), gphnb mutant larvae spent more time in motion (Fig. 5b) and swam at increased velocity compared to wild-type control larvae (Fig. 5c). Because neuronal activity promotes myelination^{19,22,23,39–49} we tested the possibility that neuronal activity drives the formation of long myelin sheaths in gphnb mutant larvae. To do this, we blocked neuronal activity using tetrodotoxin (TTX), which inhibits voltage-gated sodium channels. We injected TTX or a control solution into the yolk sac and selected paralyzed fish in the TTX group for imaging at 7 dpf (Fig. 5d). In wild-type larvae, TTX-induced silencing reduced sheath number and increased individual sheath length (Fig. 5e, f, i, j). By contrast, TTX-induced silencing did not change sheath number or length in gphnb mutant larvae relative to controls (Fig. g--j). We conclude that the excessively long myelin sheaths of gphnb mutant larvae do not result from elevated neuronal activity. Additionally, these data raise the possibility that loss of Gphn function impairs neuronal activity-dependent modulation of myelin sheath characteristics.

Fig. 5. gphnb mutant hyperactivity does not account for long myelin sheaths.

a Tracking data from a behavioral trial with 6 wild-type (left, boxed in teal) and 6 gphnb mutant (right, boxed in purple) larvae at 7 dpf. Red lines are traces of the swimming path of an individual larva during the 10 minute video recording. Quantification of b time in motion (s) (wildtype min = 0.00, median = 3.80, mean = 7.62, max = 33.28; gphnb min = 0.00, median = 16.76, mean = 26.38, max = 91.44; p-value = 0.0001) and c swimming velocity (mm/s) (wildtype min = 0.0007, median = 0.072, mean = 0.107, max = 0.381; gphnb min = 0.0052, median = 0.145, mean = 0.222, max = 0.792; p-value = 0.0003). d Experimental timeline for tetrodotoxin (TTX) experiments: (1) mosaic expression of mbpa:eGFP-CAAX at the single-cell stage; (2) eGFP sorting and TTX or control injections at 6 dpf; (3) confocal imaging at 7 dpf. Representative images of OLs in e wild-type control, f wild-type +TTX, g gphnb mutant control, and h gphnb + TTX conditions. i Quantification of sheaths per OL (wildtype control min = 9.0, 1st quartile = 13.0, median = 15.0, mean = 16.2, 3rd quartile = 18.0, max = 28.0; wildtype TTX min = 8.0, 1st quartile = 11.0, median = 12.0, mean = 12.2, 3rd quartile = 14.0, max = 18.0; p-value = 0.0115) (gphnb control min = 9.0, 1st quartile = 11.5, median = 16.0, mean = 14.9, 3rd quartile = 17.0, max = 21.0; gphnb TTX min = 9.0, 1st quartile = 12.0, median = 15.0, mean = 14.7, 3rd quartile = 17.0, max = 30.0; p-value = 0.5246). j Quantification of individual sheath length (wildtype control min = 5.2, 1st quartile = 26.6, median = 38.8, mean = 39.3, 3rd quartile = 50.7, max = 90.8; wildtype TTX min = 7.1, 1st quartile = 33.8, median = 48.0, mean = 49.2, 3rd quartile = 62.6, max = 114.2; p-value = 0.020) (gphnb control min = 3.0, 1st quartile = 33.3, median = 44.7, mean = 46.2, 3rd quartile = 55.5, max = 113.8; gphnb TTX min = 8.1, 1st quartile = 36.4, median = 45.5, mean = 48.2, 3rd quartile = 57.6, max = 110.8; p-value = 0.562). b and c comparisons performed with two-tailed Mann-Whitney U tests; b wild-type n = 49, gphnb mutant n = 81; c wild-type n = 52 larvae; gphnb mutant n = 76. i comparisons performed with two-sided Wilcoxon rank sum tests, j comparisons performed with non-linear mixed effects model; wildtype control n = 17; wildtype +TTX n = 13; gphnb control n = 14; gphnb + TTX n = 13. p < 0.05 is considered significant. Scale bars = 10 μm. dpf days post fertilization, TTX tetrodotoxin. Source data are provided as a Source Data file.

Oligodendroglial Gphn aids in selecting GABAergic axons for myelination

In neurons, Gphn functions specifically at synapses that engage in GABAergic and glycinergic signaling. Could Gphn function similarly in nascent myelin sheaths to mediate specific interactions with GABAergic and glycinergic axons? To test this, we used transgenic reporters to determine whether the sheath length phenotype of gphnb mutant larvae is specific to axon class (Fig. 6a–d). This analysis revealed no difference in sheath lengths between wild-type and gphnb mutant larvae on glutamatergic axons (Fig. 6e) or on GABAergic axons, although we found a trend toward longer sheaths on GABAergic axons in gphnb mutant larvae (p = 0.076) (Fig. 6f).

Fig. 6. Myelin sheath formation on glutamatergic and GABAergic axons in gphnb mutant larvae.

a, b Representative images of individual OLs mosaically labeled with mbpa:mApple-CAAX (purple) in (a) Tg(slc17a6:eGFP) glutamatergic and (b) Tg(gad1b:eGFP) GABAergic (green) wildtype larvae. Representative images of single OLs in (c) glutamatergic and d GABAergic gphnb mutant larvae, at 7 dpf. Yellow arrows point to the edges of an individual myelin sheath that wraps the labeled axon class, and dashed outlines indicate the perimeter of the myelin on axon class. Comparison of sheath length on axon class for e glutamatergic axons (wildtype glutamatergic min = 5.0, 1st quartile = 28.5, median = 35.2, mean = 38.7, 3rd quartile = 47.4, max = 97.2; gphnb glutamatergic min = 13.9, 1st quartile = 28.9, median = 35.5, mean = 37.6, 3rd quartile = 43.8, max = 81.1; p-value = 0.431) and f GABAergic axons (wildtype GABAergic min = 10.1, 1st quartile = 22.5, median = 33.0, mean = 36.5, 3rd quartile = 48.1, max = 70.1; gphnb GABAergic min = 8.4, 1st quartile = 35.2, median = 43.77, mean = 43.9, 3rd quartile = 53.7, max = 70.1; p-value = 0.076). Statistical comparisons were performed with a non-linear mixed effects model; p < 0.05 is considered significant; wildtype glutamatergic n = 10 OLs and larvae; gphnb glutamatergic n = 10; wildtype GABAergic n = 12; gphnb GABAergic n = 9. Scale bars = 5 μm. Source data are provided as a Source Data file.

We then examined whether Gphn influences myelination of unique axon classes. Quantification of myelinated glutamatergic (Fig. 7a, b) and GABAergic axons (Fig. 7c, d) in transverse spinal cord sections showed that gphnb mutant larvae had more myelinated glutamatergic axons and fewer myelinated GABAergic axons compared to wild-type control larvae (Fig. 7e), in both dorsal (Fig. 7f) and ventral spinal cord (Fig. 7g). Notably, there was no difference in myelin volume between gphnb mutant and wild-type larvae (Supplementary Fig. 6). This indicates that the global loss of gphnb function shifts myelin placement onto glutamatergic axons from GABAergic axons without substantially changing the amount of myelin. Interestingly, we found more PSD95 puncta in myelin of gphnb mutant larvae compared to wildtype (Supplementary Fig. 7), possibly reflecting the displacement of myelin from GABAergic to glutamatergic axons.

Fig. 7. Loss of gphnb biases myelin formation onto glutamatergic axons.

Representative images of glutamatergic neurons (TgBAC(slc17a6:eGFP), green) and myelin (Tg(sox10:mRFP), purple) transgenic reporters in (a) wildtype and (b) gphnb larvae, and GABAergic neurons (TgBAC(gad1b:eGFP), green) and myelin transgenic reporters in (c) wildtype and d gphnb larvae at 7 dpf. Magnified regions boxed in yellow are enlarged to show dorsal and ventral regions in detail. Dashed outlines in magnified panels indicate individual myelinated axons. Quantification of myelinated axon counts for a single side of the spinal cord for e total myelinated axons (wildtype GABAergic min = 63.00, 1st quartile = 66.17, median = 67.83, mean = 68.54, 3rd quartile = 69.75, max = 77.33; gphnb GABAergic min = 56.00, 1st quartile = 57.67, median = 58.67, mean = 58.54, 3rd quartile = 59.33, max = 61.00; p-value = 0.00092) (wildtype glutamatergic min = 81.00, 1st quartile = 89.08, median = 94.50, mean = 92.54, 3rd quartile = 95.92, max = 100.33; gphnb glutamatergic min = 93.33, 1st quartile = 97.58, median = 100.83, mean = 101.92, 3rd quartile = 104.58, max = 115.67; p-value = 0.01041), f dorsal myelinated axons (wildtype GABAergic min = 28.33, 1st quartile = 29.33, median = 29.67, mean = 31.46, 3rd quartile = 33.50, max = 37.33; gphnb GABAergic min = 24.33, 1st quartile = 24.67, median = 25.50, mean = 25.58, 3rd quartile = 26.33, max = 27.33; p-value = 0.00089) (wildtype glutamatergic min = 19.00, 1st quartile = 21.08, median = 23.83, mean = 22.79, 3rd quartile = 24.08, max = 26.33; gphnb glutamatergic min = 24.00, 1st quartile = 25.17, median = 26.33, mean = 27.12, 3rd quartile = 27.17, max = 35.67; p-value = 0.00723), and g ventral myelinated axons (wildtype GABAergic min = 32.00, 1st quartile = 34.17, median = 38.67, mean = 37.08, 3rd quartile = 39.50, max = 40.00; gphnb GABAergic min = 29.67, 1st quartile = 32.25, median = 33.17, mean = 32.96, 3rd quartile = 34.17, max = 34.67; p-value = 0.035) (wildtype glutamatergic min = 62.00, 1st quartile = 66.25, median = 70.83, mean = 69.75, 3rd quartile = 73.42, max = 76.33; gphnb glutamatergic min = 68.00, 1st quartile = 72.58, median = 75.50, mean = 74.79, 3rd quartile = 77.08, max = 80.00; p-value = 0.052). Significance was determined with two-sided Wilcoxon rank sum tests with p < 0.05 considered significant. All groups n = 8 larvae. Scale bars = 5 μm. Magnified panel scale bars = 2 μm. Source data are provided as a Source Data file.

Though our cell-specific loss of gphnb approach phenocopied the long individual myelin sheaths formed in larvae lacking gphnb function, we identified a difference in the number of sheaths that these cells made compared to our analysis of homozygous gphnb mutants. Therefore, we used the OL-specific gphnb CRISPR/Cas9 plasmid to reduce Gphn function in OLs and counted the number of sheaths that individual cells formed on GABAergic and glutamatergic axons (Fig. 8a–d). OL-specific loss of gphnb function reduced the number of GABAergic axons that were myelinated (Fig. 8c–e), supporting our conclusion that Gphn expressed by OLs promotes GABAergic axon myelination. By contrast, there was no difference in the number of myelinated glutamatergic axons (Fig. 8a, b, e). We interpret these data to mean that Gphn function in OLs enhances GABAergic but not glutamatergic axon selection for myelination. Moreover, the elevated neuronal activity in mutant larvae likely increases the total number of myelin sheaths formed by individual OLs (Fig. 5), in agreement with prior work showing that neuronal activity and vesicle release regulate OL sheath number²³, thereby placing additional sheaths onto glutamatergic axons.

Fig. 8. Oligodendrocyte-specific loss of gphnb reduces GABAergic axon myelination.

Representative images of glutamatergic axons and individual dorsal oligodendrocytes expressing Cas9 and a no gRNAs control and b gphnb gRNAs. Representative images of GABAergic axons with individual oligodendrocytes expressing Cas9 and c no gRNAs control and d gphnb gRNAs at 7dpf. Orthogonal y-z views are to the right of each representative image. x-y views are a single 0.19 mm z-slice. Dashed yellow outlines indicate single myelin sheaths that wrap the reporter axon. e Quantification of the number of myelin sheaths on the reporter axon per oligodendrocyte (control GABAergic min = 5.0, 1st quartile = 7.8, median = 9.0, mean = 8.7, 3rd quartile = 10.0, max = 12.0; gphnb GABAergic min = 2.0, 1st quartile = 3.6, median = 4.0, mean = 4.6, 3rd quartile = 5.3, max = 8.0; p-value = 1.339e−09) (control glutamatergic min = 6.0, 1st quartile = 8.0, median = 9.0, mean = 10.0, 3rd quartile = 12.0, max = 16.0; gphnb glutamatergic min = 7.0, 1st quartile = 9.0, median = 10.5, mean = 11.2, 3rd quartile = 13.0, max = 18.0; p-value = 0.2142). Significance was determined with a two-sided Student’s t-test. p < 0.05 is considered significant. GABAergic control n = 20, gphnb n = 20; glutamatergic control n = 17, gphnb n = 22. Scale bars = 5 μm. Source data are provided as a Source Data file.

Does this selection bias arise because of differential sheath stability on distinct axon classes or is it established during initial axon ensheathment? We reasoned that if Gphn is important for stabilizing nascent sheaths then myelin should be normally distributed among axon classes in mutant larvae at 3 dpf, soon after ensheathment initiation. However, we found that gphnb mutant larvae had fewer myelinated GABAergic axons and more myelinated glutamatergic axons by 3 dpf (Supplementary Fig. 8), similar to our observations made at 7 dpf. We next investigated sheath stability directly by performing timelapse imaging beginning at 55 hours post fertilization, a period of myelin sheath initiation and growth (Fig. 9). To quantify our observations, we counted the number of myelin sheaths that were removed after initiating growth along axons, which we characterized as ensheathment failures. This analysis revealed no difference in the total number of ensheathment failures between wild-type and gphnb mutant larvae (Fig. 9i, supplementary movies 1 and 2, orange traces). Furthermore, we found no difference in the number of ensheathment failures on GABAergic axons (Fig. 9b, c, e, f, j). However, gphnb mutant larvae had fewer stabilized sheaths on GABAergic axons compared to wild-type larvae (Fig. 9k, Supplementary Movies 3 and 4, blue traces). We therefore conclude that the lack of Gphn does not reduce nascent myelin sheath stability or stability on GABAergic axons. Rather, these data indicate that Gphn facilitates GABAergic axon recognition by OLC processes leading to ensheathment. Notably, a recent study found an association between Gphn puncta accumulation within OPC processes and sites of myelin sheath formation²⁹. Collectively, we interpret these data to mean that Gphn localized within processes biases the selection of GABAergic axons for myelination, indicating a neurotransmitter identity-dependent function for Gphn in myelination.

Fig. 9. Loss of gphnb reduces the number of nascent myelin sheaths on GABAergic axons.

Representative z-stack images of a wild-type and d gphnb mutant Tg(sox10:mRFP) larvae immediately prior to myelination onset. Boxed regions indicate where representative sheath initiation in magnified panels (b, c, e, f) subsequently fail to stabilize on GABAergic axons. b, c Representative magnified single 0.5 μm slice of a nascent wild-type sheath on a GABAergic axon that fails to stabilize by the next imaging frame c. e Nascent gphnb sheath on a GABAergic axon that fails to stabilize by the next imaging frame, f Yellow dashed outline indicates the nascent sheath, and orange dashed outlines indicate where the nascent sheath was in the preceding imaging frame (sox10:mRFP+ processes still visible). T values indicate sequential 30 min frame increments from the beginning of the imaging session at around 55 hpf. Quantification of the number of i ensheathment failures (wildtype min = 9.0, 1st quartile = 17.8, median = 26.5, mean = 27.4, 3rd quartile = 35.0, max = 50.0; gphnb min = 7.0, 1st quartile = 14.8, median = 19.0, mean = 26.8, 3rd quartile = 31.3, max = 74.0; p-value = 0.9393), j ensheathment failures that occurred on GABAergic axons (wildtype min = 0.0, 1st quartile = 4.3, median = 6.5, mean = 6.8, 3rd quartile = 9.0, max = 16.0; gphnb min = 0.0, 1st quartile = 1.3, median = 4.0, mean = 6.7, 3rd quartile = 7.0, max = 24.0; p-value = 0.3618), and k the number of stabilized sheaths on GABAergic axons (wildtype min = 1.0, 1st quartile = 3.3, median = 5.0, mean = 4.4, 3rd quartile = 6.0, max = 6.0; gphnb min = 1.0, 1st quartile = 1.0, median = 1.0, mean = 1.6, 3rd quartile = 2.0, max = 4.0; p-value = 0.0029). Significance was determined by two-sided Student’s t-test for i, and two-sided Wilcoxon Rank Sum Test for j and k; p < 0.05 considered significant; wildtype n = 10 larvae, gphnb n = 10 larvae. a and d scale bar = 10 μm; b, c, e, f, and g, h scale bars = 5 μm. Source data are provided as a Source Data file.

Discussion

OLs can myelinate fixed axons and synthetic substrates in vitro^48–52, indicating that OLs can myelinate axons without the need for specific molecular or cellular cues that distinguish them. However, not all axons are myelinated in vivo, and different types of axons are covered by distinct patterns of myelin^14–18. Furthermore, OLs preferentially place myelin on axons that are more electrically active^{19,22,23,45,46,48,49}. These observations suggest that OLs can discriminate between the many different types of axons they encounter in a developing nervous system, but the mechanisms by which they do so remain unknown. In this study, we sought to understand whether OLs use unique molecular machinery to selectively myelinate axons of distinct neurotransmitter classes. Building on prior evidence that synaptic-like mechanisms promote myelin sheath formation, we focused our investigation on Gphn, a scaffolding protein that functions at neuronal postsynaptic terminals that receive inhibitory signals. Altogether, our data support a model where myelinating OLCs repurpose classical postsynaptic machinery to facilitate myelination of unique axon classes with specificity characteristic of neuronal synapses (Supplementary Fig. 9).

Electrophysiological measurements revealed that neurons make glutamatergic and GABAergic synaptic connections with OPCs, which can mediate excitable responses within OPCs^53–58. Strikingly, calcium activity in OPC processes correlated with Gphn localization, preceding and predicting which processes eventually formed myelin sheaths²⁹. Thus, neuronal activity might convey pro-myelinating signals to OPCs via Gphn-mediated synaptic interactions. Whether synaptic mechanisms also mediate axon-OL interactions during myelin sheath formation has been unclear. Although electrical activity was not detected in OLs³², activity-dependent calcium transients were evident in nascent myelin sheaths^59,60. OLs express genes that encode postsynaptic proteins following their differentiation from OPCs^24,25 and our prior work²⁸, together with data we present here, show that the postsynaptic proteins Cadm1b, Caska, PSD95, and Gphn occupy nascent myelin sheaths in zebrafish. Thus, newly differentiating OLs might have the capacity to interact with presynaptic molecules displayed on axons during myelin sheath formation.

Consistent with prior observations using mice^15–18, we determined that OLs myelinate inhibitory GABAergic and glycinergic axons as well as excitatory glutamatergic axons in the zebrafish spinal cord. Neurons use different molecular complexes to assemble different types of synapses, particularly excitatory and inhibitory synapses. These complexes, anchored by the scaffold proteins Gphn at inhibitory synapses and PSD95 at excitatory synapses, are necessary for the specificity of neurotransmission. Neurons receive both inhibitory and excitatory inputs by segregating these molecular complexes to different postsynaptic terminals. In this study, we found that myelin on GABAergic and glycinergic axons has more Gphn than myelin on glutamatergic axons. This raises the intriguing possibility that, similar to neurons, distinct postsynaptic complexes within individual myelin sheaths facilitate interaction with specific types of axons. How then do OLs use postsynaptic factors for myelin sheath formation? Here we learned that the global absence of Gphn function causes myelin sheaths to wrap GABAergic axons less frequently and glutamatergic axons more frequently, without changing the total amount of myelin. OL-specific gphnb mutagenesis similarly reduced the number of myelin sheaths placed on GABAergic axons, indicating that GABAergic axon selection for myelination is driven by Gphn function in OLs. With our observation that Gphn is enriched in myelin on GABAergic axons relative to myelin on glutamatergic axons, these data potentially provide a mechanistic explanation for a prior observation that some OLs preferentially myelinate inhibitory axons in the mouse cortex¹⁸. Furthermore, this reduction in myelination of GABAergic axons was evident as early as 3 dpf, indicating that Gphn functions early in myelin sheath formation to select axonal targets. Our time-lapse data suggest this selection mechanism is not based on nascent sheath stabilization, as we found no change in initial turnover with the loss of gphnb, and specifically no change in sheath loss on GABAergic axons. However, we did find that gphnb mutants had fewer stabilized sheaths on GABAergic axons during the initial stages of ensheathment. This raises the possibility that Gphn biases initial ensheathment, such that fewer GABAergic axons are wrapped in the absence of gphnb. Consistent with this, Gphn localized within OPC processes was correlated with sites of myelin sheath formation²⁹. Together, these results point to a model where postsynaptic proteins function to aid myelin targeting and formation on distinct classes of axons.

We also found that in the absence of Gphn function, myelin sheaths were abnormally long. This effect on sheath length was only evident at 7 dpf, after myelin sheath extension is normally complete in the spinal cord of larval zebrafish⁶¹. Prior investigations also revealed that disruptions of postsynaptic protein functions have consequences for myelin sheath formation. For example, we showed that OL-specific expression of dominant negative forms of the synaptogenic adhesion proteins Nlgn2b, Cadm1b, and Lrrtm1 caused either longer (Nlgn2b) or shorter (Cadm1b, Lrrtm1) myelin sheaths²⁸. Using a transient, cell-type-specific CRISPR/Cas9 mutagenesis approach, knockdown of dlg4a/b, which encodes PSD95, gphnb, and nlgn3b in OPCs in zebrafish, reduced the total length of myelin sheaths at 5 dpf^29,62. Finally, an OLC-specific γ2-GABA_A receptor subunit knock-out resulted in long myelin sheaths in the mouse barrel cortex³⁰. This effect was evident on inhibitory, GABAergic parvalbumin (PV)+ interneurons, while neighboring excitatory, glutamatergic spiny stellate cells (SSC) retained normal myelin. Notably, none of these functional manipulations prevented myelination, but they consistently altered myelin sheath length. Neuronal activity also can modulate sheath length^22,45,46, and sheath length can influence neuronal firing patterns^3–5,63,64. Thus, postsynaptic proteins expressed by OLs might be key mediators of myelin plasticity, like their roles in synaptic plasticity. In support of this possibility, we found that in gphnb mutant larvae, myelin sheaths were unresponsive to TTX-mediated inhibition of neuronal activity, as they are in wild-type larvae.

One difference between our global gphnb mutant and cell-specific loss of gphnb results is the number of myelinated glutamatergic axons. In gphnb mutant larvae, glutamatergic axons were more frequently myelinated compared to controls (Fig. 7 and Supplementary Fig. 8), whereas there was no difference with the cell-specific gphnb mutations (Fig. 8). Prior work showed that pharmacological elevation of neuronal activity increased OL sheath number, dependent on synaptic vesicle release²³. Because gphnb mutant larvae are hyperactive and lack machinery for selecting GABAergic axons, we speculate that the hyperactivity of mutant larvae (Fig. 5a–c) drove additional myelin formation that was deposited onto glutamatergic axons. Supporting this interpretation, there was no change in the total amount of myelin in gphnb mutant larvae (Supplementary Fig. 6) and myelin contained more PSD95 (Supplementary Fig. 7), likely due to the increase in overall activity and glutamatergic myelination. Altogether, these data are consistent with the idea that neuronal activity and oligodendroglial postsynaptic proteins together modulate myelin sheath number and axon selection.

The OL-specific CRISPR/Cas9 approach we used for this study has important limitations. First, it is difficult to validate the creation of loss-of-function mutations in OLs using a transient expression approach. Nevertheless, this approach phenocopied the sheath length and GABAergic axon selection phenotypes of gphn mutant larvae, and OL-specific expression of human GPHN restored myelin sheath length to normal in gphn mutant larvae, providing confidence that this approach effectively reduces Gphn function in OLs. In the future, this cell type-specific transient expression approach might benefit from using the RNA-targeting Cas13 system, enabling RNA quantification techniques to assess knockdown efficacy⁶⁵. Second, we have been unable to establish stable transgenic lines to inactivate Gphn function in all OLs, precluding our ability to investigate myelination of neural circuits and to test behavior. Because we routinely create other transgenic expression lines, we speculate that Cas9 transgenes are highly susceptible to genomic silencing. Placing Cas9 transgenes in safe harbor sites using the pIGLET site-specific recombination system⁶⁶ might help overcome this limitation.

Another limitation to our study is that we did not exhaustively characterize the potential changes in myelin patterns on all major axon classes in the spinal cord. Specifically, we did not test glycinergic axons. In neurons, Gphn differentially interacts with glycine receptors (GlyR) and GABA_A receptors^67,68. For example, Gphn localizes to all glycinergic synapses and is required for GlyR clustering^36,67 but Gphn does not localize to all GABAergic synapses, and it is only required for a subset of GABA_AR subunit clustering⁶⁹. This may account for the lack of a clear sheath length phenotype on GABAergic and glutamatergic axons (Fig. 6) that was prominent when axon identity was ambiguous (Fig. 4). Potentially, the longest sheaths reside on glycinergic axons in the absence of Gphn function. In future work, we plan to compare myelin formation on GABAergic and glycinergic axons.

Collectively, our data reveal a role for canonical postsynaptic proteins in myelination of specific neurotransmitter axon classes. We provide evidence that Gphn biases the selection of inhibitory axons for myelin targeting and influences myelin sheath growth. Because myelin can alter axonal conduction velocity^3–5, myelin targeting and plasticity can provide a powerful influence on circuit development by fine-tuning the activity of individual axons within developing neural circuits^70–72. This is supported by evidence where the specific loss of the γ2-GABA_A subunit in OLCs disrupted PV+ firing rate without altering neighboring excitatory SSC frequency³⁰. Consequently, the loss of myelin on GABAergic axons with oligodendrocyte loss of gphnb, and redistribution of myelin patterns on specific axon classes we find in gphnb mutant larvae could have important consequences for neural circuit function by disrupting the balance of excitatory and inhibitory signaling (E/I balance)^70–75. This raises the possibility that genetic variants of synaptic genes that are significantly linked to neuropsychiatric disorders, such as autism and schizophrenia, alter myelination, thereby contributing to the E/I imbalance characteristic of those diseases.

Methods

Zebrafish Lines and Husbandry

All fish and larvae were handled in accordance with the University of Colorado Institutional Animal Care and Use Committee (IACUC), protocol #0370. Previously generated transgenic lines used in this manuscript included: TgBAC(slc17a6b:eGFP)^76,77; Tg(slc6a5:eGFP)⁷⁸; TgBAC(gad1b:eGFP)⁷⁹; Tg(sox10:mRFP)^{vu234 80}; Tg(mbpa:eGFP-CAAX)^{co58 81,82}; and Tg2(mbpa:mCherry-CAAX)^{co13 76} (see Supplementary Table 1 for Zfin IDs). After egg collection or injections, embryos were stored at 28.5 °C in petri dishes at a density of roughly 60 embryos per plate or less in egg water (6 g Instant Ocean in 20 L miliQ water). Larvae hatched from chorions were stored at a similar density in screen media (1X E3, 60μg/mL NaHCO₃, 145μg/mL CaCl₂) up to 7 days post fertilization (dpf). For embryo injections, eggs were collected within 30 minutes of breeding and injected at the single-cell stage. For Tol2 transgenesis injections, 6–25 ng/mL of expression plasmids were injected in 2–3 nL drops. Larvae were then screened prior to imaging for transgenesis reporters or fluorescent protein expression in the spinal cord. When animals for fixed experiments needed genotyping, larvae were anesthetized and quickly tail clipped, and the body was fixed in 4% paraformaldehyde (PFA) while the tail was lysed and used for genotyping. Otherwise, larvae for fixed experiments were euthanized with 4% Tricaine mesylate prior to fixation.

Mutagenesis

We used the CRISPR/Cas9 system to generate our gphn loss-of-function mutations. We used CRISPOR⁸³ to design guide RNAs that target the coding sequence of exon 1 for both gphna and gphnb (Fig. 5, see Supplementary Table 1 for gRNA sequences). We followed the “Alt-R CRISPR-Cas9 System: In vitro cleavage of target DNA with ribonucleoprotein complex” protocol to synthesize each gRNA (http://sfvideo.blob.core.windows.net/sitefinity/docs/default-source/protocol/alt-r-crispr-cas9-protocol-in-vitro-cleavage-of-target-dna-with-rnp-complex.pdf?sfvrsn=88c43107_2, 2017). The RNP complex was created immediately prior to injections with 0.5 μg final concentration of Cas9 (IDT) diluted in Cas9 working buffer (20 mM HEPES, 150 mM KCl, pH 7.5), 100 ng each of gphna and gphnb gRNAs, and incubated at 37 °C for 10 min. This solution was then kept at room temperature until injected, with gphna and gphnb guides co-injected into wild-type embryos with 2-3 nL at the single-cell stage to generate double mutants. Several F0 injected larvae were screened for evidence of insertions and deletions created by DNA cutting and repair with PCR primers that span the guide target site (see Supplementary Table 1 for primers and sequences). F0 larvae were grown to adulthood and individually outcrossed to wild-type fish. Individual F1 larvae were anesthetized, and 50 mM NaOH was used for genomic DNA extraction, followed by 1 M Tris, pH 9 neutralization. Mutant allele identification was performed on several individual F1 animals from at least two unique F0 founders each via TA cloning and sequencing, followed by alignment with the published genome. F1 adults carrying selected gphna^co91, gphna^co92, gphnb^co94, and gphnb^co95 mutant alleles were then outcrossed, and F2 generation larvae were grown for experiments. All wild-type and mutant larvae used for mutant experiments were genotyped to confirm alleles.

Behavior

Individual larvae were placed into single wells in a 24-well plate at 7 dpf and given an equal volume of screen media per well. Larvae in 24-well plates were then transferred to the CU Anschutz Neurotechnology Animal Behavior Bore in a covered box and allowed to acclimate to the room for 10 min. For the behavior recording, the plate was placed in the DanioVision (Noldus) recording box, and larvae were allowed to habituate to the box for 10 min prior to recording start. Larvae were then recorded for 10 min. Following behavior, larvae were anesthetized and lysed for genotyping.

TTX injections

For neuronal activity experiments, the voltage-gated sodium channel blocker tetrodotoxin (TTX) was injected into the yolk sac at 6 dpf. All animals were anesthetized with Tricaine and sorted for fluorescent reporter expression using a stereomicroscope equipped with epifluorescence optics. While anesthetized, larvae were then placed into molds made of 2% agarose in egg water with individual larval wells. Larvae were positioned in the wells such that their yolk sacs faced up, and 3–4 nL buffered TTX (0.6 mM TTX, 0.4 M KCL, 0.05% phenol red, and water) or control solution (0.4 M KCL, 0.05% phenol red, and water) was injected into the yolk sac below the swim bladder. Animals were then removed from the molds and placed into fresh egg water and allowed to recover fully for 24 hrs prior to imaging.

Cloning

Several plasmids, listed in Supplementary Table 1, were generated using the Gateway system and the Tol2 kit⁸⁴. To create the Gephyrin intrabody middle entry vector, we used pCAG_GPHN.FingR-mKate2-IL2RGTC (Addgene plasmid #46297; RRID: Addgene_46297) and replaced the mKate2 fluorophore with mScarlet in silico. Using this plasmid as a template, we custom-ordered pEXPR-zfmbpa:GPHN.FingR-mScarlet-IL2RGTC-KRAB(A)-pA-CC2-Tol2 and a modified version of the PSD95 intrabody we previously published²⁸ for co-expression experiments (pEXPR-zfmbpa:PSD95.FingR-EGFP-CCR5TC-KRAB(A)-pA-CC2-Tol2) from VectorBuilder (see Supplementary Table 1 for Vector IDs). To create an entry plasmid encoding human GPHN protein, we used plasmid pECE-M2-GPHN (Addgene plasmid #31665; RRID: Addgene_31665) and designed primers that introduced a Kozak sequence and attB sites that flank the coding sequence. We amplified human GPHN using Phusion polymerase, and then performed a gateway BP reaction with pDONR-221 to create a middle entry pME-GPHN. We also generated middle- and 3’-entry vectors containing mApple-CAAX sequence. We used Gateway cloning to create a pME-mApple-CAAX vector and InFusion cloning (Takara Bio, 638944) for a p3E-mApple-CAAX-pA vector (see Supplementary Table 1 for cloning primers and entry plasmids). The Gateway system was subsequently used to clone these middle-entry vectors and others into expression plasmids with the 5’ entry p5E-mbpa plasmid to provide cis-regulatory elements and 3’ reporter fluorophores following a T2A sequence (see expression vectors in Supplementary Table 1).

To create the cell-specific CRISPR plasmid, we modified the 10xUASE1b:myrmScarlet-P2A-Cas9polyA;U6-ctrl-sgRNA1;U6-ctrl-sgRNA2-pA plasmid from Li et al. ²⁹ (gift from Dr. Kelly Monk)²⁹. We annealed single-stranded oligos with 4 base-pair overhangs compatible to the plasmid backbone to generate each double-stranded gRNA (see Supplementary Table 1 for single-stranded oligo sequences; overhang sequence is underlined)⁸⁵. Oligos were annealed in IDTE buffer (10 mM TRIS (pH 8.0), 0.1 mM EDTA) with the following thermalcycler conditions: 95° 5 min, ramp down to 50° at 0.1°/s, hold at 50° for 10 min. The plasmid backbone was then digested with BsaI and the annealed gphnb gRNA1 oligo was ligated with NEB Quick Ligase according to the manufacturer’s instructions, and the resulting plasmid was transformed. This process was subsequently repeated for gphnb gRNA2 using BsmBI. We used XmaI and XhoI to replace the 10xUASE1b sequence with mbpa regulatory DNA using restriction cloning. Briefly, the mbpa regulatory DNA was PCR amplified using primers to introduce XmaI and XhoI sites (see Supplementary Table 1 for primer sequences). We then digested this PCR product and the 10xUASE1b:myrmScarlet-P2A-Cas9polyA;U6-ctrl-sgRNA1;U6-ctrl-sgRNA2-pA plasmid and used the NEB Quick Ligation Kit to ligate the vector and insert followed by transformation. Plasmid construction was validated using whole plasmid sequencing.

Immunohistochemistry

Larvae were fixed in 4% PFA at each time point and washed with 1x PBS after 24 hours. Larvae were then embedded in 1.2% agarose in sucrose, cut into blocks, and soaked in 30% sucrose overnight. Blocks were then dried and frozen on dry ice and stored at -80° until sectioning. Blocks were sectioned on a Leica 1950 cryostat in 20 μm thick sections and thaw-mounted to polarized slides. Slides were mounted in a Sequenza rack, followed by immunostaining as previously described^28,77 (Thermo Scientific, Waltham, MA). Briefly, slides were washed with 1x PBSTx (0.1% Triton-X 100) 3 times for 5 min. Then, a block was added (2% goat serum and 2% bovine serum albumin in 1xPBSTx) for 1 hour. Primary antibodies (see Supplementary Table 1 for antibodies and concentrations) were diluted in block and added to slides, and then stored overnight at 4°. The next day, slides were washed in 1x PBSTx every 10 min for 1.5 hr, then secondary antibodies were diluted in block and added and incubated at room temperature for 2 hr (see Supplementary Table 1 for antibodies and concentrations). Secondaries were washed off in 1x PBSTx every 10 min for 1.5 hr, followed by DAPI for 5 min (optional, 1:2000 diluted in 1x PBSTx). For DAPI labeling, slides were washed 3 times for 5 min with 1x PBSTx, followed by mounting. Two drops of Vectashield (Vector Laboratories, H-1000-10) were added, followed by application of a No.1 coverslip. Slides were sealed with clear nail polish and allowed to dry at room temperature for 15 minutes, and stored at 4° until imaging. Axon class myelination experiments and total myelin in wildtype and mutant experiments were performed with fixed, transverse sections that were washed with PBSTx and mounted without immunostaining.

Imaging

Live animal fluorescent imaging

Anesthetized larvae sorted for transgenesis reporters or identified as positive for spinal cord fluorescent reporter expression were imaged at 3 dpf, 4 dpf, and 7 dpf. Larvae were mounted in 0.9% agarose with 0.6% tricaine on their sides pressed against a coverslip and imaged on a spinning disk confocal microscope (Carl Zeiss, Oberkochen, Germany) or a laser scanning confocal microscope (Carl Zeiss), with a 40x water objective. Individual dorsal OLs were imaged above the yolk sac extension. Standard confocal imaging was used to capture Z-stacks with 0.5 μm step intervals for sheath analysis. Super resolution Airyscan confocal imaging was used on the Zeiss 880 to acquire Z-stacks of individual OLs in gphnb and wild-type larvae carrying transgenic neuron reporters to analyze mutant and wild-type sheath length on axon classes with 0.225 μm step intervals. Airyscan imaging was also used to capture all intrabody images with 0.5 μm step intervals. For the cell-specific loss of gphnb myelinated axon count experiment, larvae were imaged on an Andor Dragonfly 200 Spinning Disk Confocal (Oxford Instruments) DMi8 inverted microscope (Leica) with a 25x water objective with 0.19 μm step intervals. If larvae required genotyping, individual larvae were extracted from the agarose following imaging and lysed with NaOH and Tris and genotyped as described above. Otherwise, larvae were euthanized after imaging.

Timelapse imaging

Anesthetized larvae sorted for fluorescent reporters were mounted sagittally for live imaging as described above in 50 ×7 mm WillCo-dish glass bottom dishes (cat. #GWST-3512), with a layer of 0.6% tricaine in egg water and covered with the dish lid to prevent dehydration during the imaging session. Larvae were acclimated to the microscope for 2 hrs prior to imaging onset to allow solidification of agarose and equilibration of temperature. After acclimation, imaging positions were selected to capture half of the spinal cord, purposefully over-sampling the z-stack to account for any drift. Imaging started at ~55 hours post fertilization, capturing z-stack images with 0.5 μm step intervals every 30 min for 19 hrs. Genotyping was performed after imaging as stated above.

Fixed tissue imaging

Transverse imaging was carried out on fixed, 20 μm thick cryosectioned tissue with a Zeiss LSM 880 Airyscan confocal microscope with a 40x oil objective and at least 1.8x zoom or an Andor Dragonfly 200 Spinning Disk Confocal with a 63x oil objective. For immunostaining, all images were captured within 2 weeks of staining. For each larva, 3–5 images of the spinal cord were captured serially above the yolk extension, and data were either summed or averaged by animal. All imaging parameters, including Z-step interval, frame size, bit depth, laser power, and gain, were kept consistent within experiments.

Data analysis

Image blinding

All images were blinded prior to analysis. We used the FIJI (Fiji Is Just ImageJ)⁸⁶ image blinding plugin Lab-utility-plugins/blind-files developed by Nick George and Wendy Macklin (https://github.com/Macklin-Lab/imagej-microscopy-scripts), or the File Name Encrypter from the Blind Analysis Tools FIJI plugin, and only unblinded once analysis was complete.

Sheath analysis

Sheath length was measured using FIJI with the Neuroanatomy plugin SNT (simple neurite tracer)⁸⁷ as described previously^22,28,81,88. A background subtraction with a rolling ball radius of 50 μm was performed prior to analysis. Briefly, for individual OLs, the length of each sheath was traced, and the length in μm was measured. Each length measurement entry was then counted as an individual sheath, and sheaths per cell were summed. Sheath length was plotted for individual sheaths across all OLs analyzed, whereas sheath number was plotted by cell. For sheath length on axon class analysis in wild-type and gphnb mutant larvae, images were first Airyscan processed using Zen Black (Carl, Zeiss, version 2.3 SP1). Imaris x64 (Oxford Instruments, 9.9.1-10.2.0) was used to trace sheaths and classify sheaths on reported axons. Sheaths were traced using the Filaments tool, using the manual tracing feature. Individual sheaths were manually traced, and detailed statistics were captured for each filament length. Each sheath was then assessed to determine whether it was wrapping reporter axons. Individual sheath lengths on reporter axons were then plotted by genotype and neuron reporter conditions.

Immunohistochemistry analysis

We used Imaris to quantify all immunostaining. With Gphn antibody staining in wild-type and mutant larvae, images were first Airyscan processed for deconvolution in Zen Black. We then performed background subtraction with a 15 μm radius, then cropped the stack to 20 μm. We then created a surface object to quantify puncta with a background subtraction of 0.5 μm, and threshold 5% the maximum value, followed by an intensity-based object splitting and a final volume of ≥0.00125 μm filter. A spinal cord surface was created using the manual creation feature by drawing around the spinal cord on the last slice, duplicating this surface onto the first slice, and creating the surface object from the drawn ROI. The Gphn surface object was then filtered to signal within the spinal cord. We then gathered the specific, detailed volume statistics for total puncta and volume per puncta for each genotype. Three images per larvae were analyzed, with the Gphn puncta measure corrected for spinal cord volume for each image. The mean per fish was then calculated and plotted as puncta per 100 μm³.

For Gphn and PSD95 puncta in myelin experiments, images were Airyscan deconvoluted in Zen Black, followed by Imaris background subtraction with a 15 μm radius for all channels. Images were then cropped to the inner 20 μm slices from the Z-stack. A surface is created of the myelin channel, similar to that described above for single OLs, with a background subtraction with a diameter of the largest sphere of 1 μm, followed by a manually tuned threshold and a split objects filter of 2 μm to best match the fluorescent signal. Extraneous background surfaces were manually deleted. Then, surfaces were created of the Gphn or PSD95 channel as described above in wild-type and mutant conditions. This Gphn or PSD95 surface object was then filtered by distance to the myelin surface with a distance of ≤0 μm and duplicated to a new surface. This distance to the myelin filter captured the puncta within the myelin surface, and any puncta obviously only touching the outer surface of the myelin were manually deleted. The specific, detailed volume statistics were then generated to provide total puncta and volume per puncta for Gphn of PSD95 in myelin.

To quantify the Gphn puncta in myelin on axon classes, deconvolution was performed in Imaris using 10 iterations, followed by a background subtraction with a 15 μm radius for the Gphn channel. We then created myelin and Gphn surface objects as described above for the Gphn in myelin analysis. Once puncta in myelin were duplicated to a new surface object, the slice view option was used with the neuron and myelin fluorescent channels toggled on, and each punctum was manually assessed for whether it is in myelin on the reporter axon. Puncta determined to be in myelin on the reporter axon were selected and duplicated to a new surface object, and the specific, detailed volume statistics were generated. The proportion of puncta in myelin on axon class was then calculated by totaling the puncta in myelin on the reporter axons and dividing by the total puncta in myelin (Fig. 3d).

Intrabody analysis

Intrabody images were Airyscan deconvolution processed and analyzed in Imaris. Background subtraction with a 15 μm radius was performed, then images were cropped to isolate single, dorsal OLs. A surface was created of the OL with a background subtraction with a diameter of the largest sphere of 1 μm, followed by a threshold that was manually tuned to best match the fluorescent signal. The split objects filter of 2 μm was also used to capture the OL surface more precisely. Intrabody surfaces were created using a background subtraction with a diameter of the largest sphere of 0.35 μm, and the threshold was manually adjusted to best match the fluorescent signal. The split objects filter of 0.5 μm was used, followed by a voxel filter to further eliminate the creation of any background surfaces. The intrabody signal was then filtered to within the OL surface, and any puncta where the majority of the volume resided outside of the OL surface with minimal volume inside the surface, and puncta just touching the outside of the OL surface were manually deleted. Puncta in individual sheaths were then manually duplicated to new surface objects, and the detailed, specific statistics were used for puncta per sheath quantification. Total puncta per OL were calculated with the summation of the puncta in individual sheaths. This eliminated the bright, self-regulating expression in the cell body. To quantify the Gphn and PSD95 intrabodies per OL, we generated surfaces for puncta as described above, then manually selected all puncta within a single sheath and duplicated this to a new object for both intrabodies. We then collected the detailed statistics and calculated the total Gphn and PSD95 puncta per sheath.

Myelinated axon counts analysis

Transverse images were processed and analyzed in Fiji. First, a background subtraction with a 15 μm rolling ball radius was performed. Then, we created a new sub-stack of the inner 20 μm slices and used the multi-point tool to identify every myelinated axon on one side of the spinal cord. These counts were totaled and separated by region (dorsal white matter tract, medial white matter, and ventral white matter tract). We analyzed 3 images per larva that were then averaged. Single 0.5 μm sections were used for representative images, and the same region was isolated for magnified views between wild-type and gphnb mutant larvae. For myelinated axon counts with cell-specific loss of gphnb, a single oligodendrocyte was analyzed per fish. We performed a Gaussian filter of 0.75 µm, merged the myelin and neuron channels, and used the Cell Counter plugin to label sheaths that were on the reporter axon. Sheaths were counted if a myelin signal was observed on both sides of the axon signal in the z-plane, in addition to surrounding the axon signal in the x- and y- planes.

Timelapse analysis

We used Imaris software (v 10.2) to analyze time-lapse data. Images were cropped to the white matter region across the 19 hr imaging window. All images were then processed with a background subtraction using a 50 μm radius, followed by a Gaussian filter of 0.1. We then used the filaments tool to manually trace all sheaths that initiated for a single OL at each 30-minute time point. Traces that failed or disappeared were selected and duplicated to a new object. Detailed length statistics were saved, and the number of failures was summed for the individual OL. We then used the slice view to manually determine whether failed traces resided on GABAergic axons; these were duplicated to a new object and saved detailed statistics. Finally, we used the slice view to manually assess whether stabilized sheaths resided on GABAergic axons, duplicated these traces to a new object, and save the detailed statistics.

Behavior analysis

The DanioVision software (Noldus) recorded 10-min videos and used the EthoVision TX to automatically track the movement of individual larvae. Total time in motion and velocity were generated and used for analysis between wild-type and gphnb mutant larvae.

Statistical analysis

All data analysis, statistics, and plotting were performed using R in R Studio (v 2024.04.2 + 764), except for behavior analysis, which was performed in GraphPad Prism (version 10). Data and statistical analysis were performed using the ggpubr, emmeans, and nlme packages. Plots were generated with the dplyr and ggplot2 packages. Data from each experiment were tested for normality with the Shapiro-Wilk test. If data were normally distributed (p > 0.05), parametric tests were used to compare groups. If data were not normally distributed, non-parametric tests were used. For tests with two groups, the Student’s t-test, Wilcoxon rank sum test, or Mann-Whitney U test was used. For datasets with multiple comparisons, global significance was tested with a one-way ANOVA, Kruskal-Wallis test, or a type III two-way ANOVA for unbalanced design, followed by Bonferroni-corrected pair-wise T-Tests or Wilcoxon rank sum tests. For comparisons of individual sheath lengths, a non-linear mixed effects model was used to account for the dependence of individual sheaths coming from a single OL. For double intrabody correlation within individual sheath analysis, data did not fit linear regression assumptions, so a Spearman’s Correlation was used. For behavioral experiments, outliers were detected with the ROUT method in Prism, with a maximum false discovery rate of Q = 1%. All experiments were performed by 7 dpf when sex is not yet determined in the zebrafish, and all n are provided in figure legends and refer to the number of larvae per group for each experiment. All experiments were performed across at least three biological replicates.

Author contributions

N.J.C. and B.A. conceived the project. C.A.D. performed behavior experiments and analysis. C.A.D. aided time-lapse experiments and analysis. B.A. made the pME-mApple-CAAX and p3E-T2A-mApple-CAAX plasmids. B.A. and N.J.C. made the mbpa:myrmScarlet-P2A-Cas9polyA;U6-ctrl-sgRNA1;U6-ctrl-sgRNA2-pA and mbpa:myrmScarlet-P2A-Cas9polyA;U6-gphnb-sgRNA1;U6-gphnb-sgRNA2-pA plasmids. N.J.C. designed and performed all other experiments and analyzed all other data. N.J.C. wrote, and N.J.C., C.A.D., and B.A. edited the manuscript.

Peer review

Peer review information

Nature Communications thanks Tim Czopka and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by National Institutes of Health (NIH) grants 1F31NS125915 to N.J.C. and 1R35NS122191 to B.A. and a gift from the Gates Frontiers Fund to B.A.

Data availability

All data are available in the manuscript, and source data are provided with this paper. Raw images, plasmid constructs, and zebrafish lines are available upon request. Source data are provided with this paper.

Code availability

R code for statistical analysis in plotting is available upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74902-3.