Natural loss of function of ephrin-B3 shapes spinal flight circuitry in birds

See allHide authors and affiliations

Science Advances  11 Jun 2021:
Vol. 7, no. 24, eabg5968
DOI: 10.1126/sciadv.abg5968


Flight in birds evolved through patterning of the wings from forelimbs and transition from alternating gait to synchronous flapping. In mammals, the spinal midline guidance molecule ephrin-B3 instructs the wiring that enables limb alternation, and its deletion leads to synchronous hopping gait. Here, we show that the ephrin-B3 protein in birds lacks several motifs present in other vertebrates, diminishing its affinity for the EphA4 receptor. The avian ephrin-B3 gene lacks an enhancer that drives midline expression and is missing in galliforms. The morphology and wiring at brachial levels of the chicken embryonic spinal cord resemble those of ephrin-B3 null mice. Dorsal midline decussation, evident in the mutant mouse, is apparent at the chick brachial level and is prevented by expression of exogenous ephrin-B3 at the roof plate. Our findings support a role for loss of ephrin-B3 function in shaping the avian brachial spinal cord circuitry and facilitating synchronous wing flapping.


A critical aspect of locomotion is left-right coordination. Limb alternation is the predominant gait mode in some amphibians (salamanders), reptiles, and terrestrial mammals, while synchronous movements are manifested in the legs of other amphibians (frogs) and some marsupials and in the wings of bats and birds. On the basis of classical electrophysiological studies in cats and rodents and genetic manipulations of cell fate and axonal trajectories in mice, a modified “half-center oscillator” model has been proposed that explains limb alternations in mammals (1). The core elements are mutually inhibitory neuronal centers on each side of the spinal cord that produce oscillatory excitatory output. This output is simultaneously transmitted to interneurons that innervate motoneurons, termed pre-motoneurons (pre-MNs). These include ipsilateral excitatory pre-MNs that activate motor activity on one side and commissural inhibitory pre-MNs that inhibit motor output on the other side (Fig. 1A) (2).

Fig. 1 Spinal gait circuitry and EFNB3 genomics.

(A and B) Schematic illustration of the morphology and wiring of the spinal cord in WT mouse (A) and in ephrin-B3/EphA4/α2-chimaerin mice collectively referred to as hopB3 mutants (B). (C) Simplified cladogram depicting the relations across representative vertebrate groups examined in this study. Branch lengths are arbitrary and not calibrated for time, red means EFNB3 is absent in galliforms, dark blue indicates branches with solid evidence of EFNB3 sequence divergence from nonavian groups, and light blue indicates branches with limited EFNB3 sequence data. (D) Comparison of EFNB3 synteny across representative organisms where synteny could be examined, with chromosomal location or data source indicated. EFNB3 (red) is absent in galliforms (red square), and DNAH2 (blue) is absent in birds. Nonavian sauropsids: lizards (e.g., green anole) and crocodiles (e.g., Chinese alligator). wgs, whole-genome shotgun contigs. SH2/3, Src homolog 2/3. (E) Schematic structure of the predicted mouse and avian ephrin-B3 proteins. RBD, receptor binding domain; TM, transmembrane domain.

The axon guidance molecule ephrin-B3 and its cognate receptor EphA4 are key regulators of alternating gait. In this family, ligand binding to the receptor typically results in axonal repulsion (forward signaling), but ephrin receptors can also act as ligands for ephrins (reverse signaling), prompting repulsion or adhesion (3). In wild-type (WT) mice, ephrin-B3 serves as a midline repulsive cue, preventing EphA4-expressing axons of excitatory pre-MNs from aberrantly crossing the ventral and dorsal midlines (Fig. 1A) (4). Transition from left-right alternation to synchrony occurs in mice mutated in the genes encoding ephrin-B3, EphA4, or the downstream signaling molecule α2-chimaerin. These mice, collectively referred to as hopB3 mutants here, show ectopic commissural projections of otherwise ipsilaterally projecting excitatory pre-MNs through the ventral and dorsal midlines (Fig. 1B) and walk with synchronous movements of the hindlimbs and forelimbs, producing a hopping gait (58). Furthermore, impairment of the neuroepithelial organization of the roof plate (RP), which leads to partial absence of ephrin-B3 expression, results in rabbit-like hopping (9, 10).

Birds have evolved from dinosaurs that use alternating gait (11). Their adaptation for flight includes substantial structural specializations of the forelimb and of the spinal cord circuits that enable synchronous limb movements. Classical transplantation experiments in chicks (our use of the term chick here refers to embryonic chicken) support a role for neuronal networks at the lumbar and brachial spinal levels in the control of the locomotion pattern of the hindlimbs and forelimbs, respectively. In chicks where brachial segments are surgically replaced by lumbosacral segments, the movement of the wings becomes alternation, whereas coordinated flapping movements are not observed (12). In the complementary experiment, transplantation of brachial segments to the lumbar cord results in synchronous, hopping-like leg movements (13). These findings support the hypothesis that local circuits in the lumbar and brachial spinal cord instruct level-specific alternate or synchronous movements, respectively.

Flight requires spinal circuits that enable the bilateral coordination of forelimb movements. It thus seems reasonable to hypothesize that the emergence of synchronous wing flapping in birds involved a transition from ipsilateral excitatory/contralateral inhibitory inputs to a pattern of bilateral co-excitatory input to wing motoneurons. Such a transition likely required changes in the premotor spinal wiring. Notably, mutations in the midline guidance molecules Netrin1 or ephrin-B3 in mice lead to a reduction in cross inhibition and an increase in cross excitation, respectively (5, 14, 15). Netrins were initially described in chicks and have well-characterized activity and roles in commissural axonal guidance (16, 17). However, little is known about ephrin-B3 in chicks and, more generally, in birds.

To explore a possible role of ephrin-B3 in locomotion patterns in birds, we have conducted a set of genomic, molecular, and neuroanatomical studies, using both chick and zebra finch as experimental organisms. We found an ephrin-B3 gene ortholog (EFNB3) in several bird species and lineages; however, their open reading frames were considerably shorter, and some domains critical for forward or reverse signaling were absent or heavily modified compared to nonavian tetrapods. We also observed a marked loss of dorsal midline ephrin-B3 expression in the brachial spinal cord of zebra finch, likely due to an avian loss of a critical enhancer element. An EFNB3 ortholog could not be found in chicken and related species, suggesting a gene loss in galliforms. Spinal cord interneurons in the chick showed extensive dorsal midline crossing at brachial levels, similar to ephrin-B3 null mice. In contrast, we observed little dorsal crossing at lumbar levels due to the presence of the glycogen body (GB), which serves as a physical barrier to neurite crossing. The crossing at brachial levels was largely prevented by the dorsal midline expression of mouse ephrin-B3. These findings are consistent with a role of ephrin-related mechanisms in spinal wiring patterns that may subserve limb alternation. They also support the notion that diminished barriers to dorsal midline crossing may have favored synchronous limb movements and facilitated the origin and/or maintenance of coordinated flight in birds.


EFNB3 gene in birds

We have conducted an extensive and thorough analysis of the genomic region containing EFNB3 and syntenic genes in birds and nonavian outgroups. Our approach required extensive curation of gene predictions in annotated databases, permissive BLAST searches of genomes, whole-genome shotgun (WGS) databases and preassembled reads (p-reads) (from PacBio assemblies) using avian EFNB3 and syntenic gene queries, and RefSeq alignment searches of unplaced genomic scaffolds and reads (details in Materials and Methods). We found EFNB3 in a broad range of bird species and orders (see Materials and Methods for details on species names), including several songbirds, some eagles, an owl, and a few cormorants, as well as the seriema, a hummingbird, the ruff, a crane, a stork, and some anseriforms (a duck and a goose). WD repeat containing antisense to TP53 (WRAP53) was immediately upstream of EFNB3 in all birds where synteny could be examined and in all nonavian organisms studied, including mammals, nonavian sauropsids, and amphibians (major groups shown in Fig. 1C and synteny details for representative species shown in Fig. 1D), providing strong support for EFNB3 orthology. A gene prediction was present in only a few species, whereas in several others, EFNB3 was partial, possibly due to incomplete sequences in Illumina assemblies.

Notably, we could not identify EFNB3 in any galliforms (Fig. 1D, red square), even in species with high-quality PacBio assemblies and no local gaps. In chicken, the genes immediately syntenic to EFNB3 in other birds were either on chromosome 31 (KDM6B) or on an unplaced scaffold (WRAP53 and CHD3), with no trace of EFNB3 (Fig. 1D, chicken). This indicates that the syntenic genes are not adjacent in this species, but we cannot distinguish whether this resulted from an inter- or intrachromosomal rearrangement compared to other birds. EFNB3 was also absent in the chicken WGS and PacBio p-reads, as well as in numerous RNA sequencing (RNA-seq) databases. In a quail, a WRAP53 prediction and an unannotated KDM6B (found by BLAST) were found on the same incomplete scaffold. Several PacBio p-reads from the current assembly (18) bridged the local gaps but revealed no trace of EFNB3 (Fig. 1D, Japanese quail). EFNB3 was also not detected in the assemblies or WGS of other galliforms (turkey, pheasant, and guinea fowl), even though the syntenic genes were sometimes present in unplaced scaffolds. A scaffold in the unannotated genome of a grouse contained traces of WRAP53 and KDM6B but no evidence of EFNB3 sequences (Fig. 1D, Gunnison sage grouse). We also could not find EFNB3 in ratites, but the data were too sparse to conclusively prove or disprove an EFNB3 loss in this basal group.

In most birds where EFNB3 synteny could be verified, KDM6B was immediately downstream, whereas in nonavian sauropsids and mammals, DNAH2 was interposed between EFNB3 and KDM6B (Fig. 1D, blue). Given that there were no gaps in several avian assemblies between EFNB3 and KDM6B, this observation supports an avian-specific loss of DNAH2, as previously indicated (19). As we show later, this finding has important implications for EFNB3 expression regulation. The exception among birds was the kakapo, where the receptor gene EphB4 was downstream of EFNB3, the only such case we are aware of. In sum, the EFNB3 gene is clearly present in a broad range of Neoaves and in Anseriformes but seems absent in Galliformes as a group (Fig. 1, C and D). We suggest that this is possibly due to a loss that occurred after their split from Anseriformes, estimated to have occurred between 55 and 72.5 million years ago (2022), but before their later diversification.

EFNB3 gene and protein structure in songbirds

Songbirds were the avian group with the most complete information on EFNB3. In zebra finches, EFNB3 is largely conserved compared to nonavian tetrapods, consisting of five exons and a large 3′ untranslated region (23) (fig. S1A). However, the percent identity of the predicted ephrin-B3 proteins in songbirds compared to human is low (48.9 to 50.6%; fig. S2), and their sizes are shorter compared to the amphibian, reptilian, and mammalian proteins (247 to 267 versus 332 to 340 amino acid residues; fig. S2). Functional domains in the extracellular and intracellular parts of the ephrin-B3 protein are missing in songbirds. In the intracellular domain, while the PDZ motif is intact, the entire Src homology 2/3 binding motif, which is required for reverse signaling (24, 25), and the tyrosine residues that are phosphorylated upon activation of the reverse signaling in other organisms are missing (Fig. 1E and fig. S2), suggesting that reverse signaling is impaired. Birds also lack an extracellular heparin-binding motif within the region juxtaposed to the transmembrane domain (26, 27) and several residues in the tetramerization interface within the receptor binding domain, required for normal signaling through the formation of receptor-ligand (Eph-ephrin) complexes (28) (Fig. 1E and fig. S2). The region in the dimerization interface where an L to P (L111P) substitution leads to a hopping mouse phenotype (29) is conserved in birds and shows other substitutions (DLDL_108-111_QRDV) compared to mammals (fig. S2). Unlike ephrin-B3, the predicted avian ephrin-B1 and ephrin-B2 proteins are highly similar to the mammalian proteins in terms of size, sequence, and subdomain partitioning. The mouse ephrin-B1 is 77.4 and 78% identical to the chicken and zebra finch proteins, respectively, and the mouse ephrin-B2 is 87.4 and 88.3% identical to the chicken and zebra finch proteins, respectively.

To test whether these structural differences affect cell trafficking of the avian ephrin-B3, we expressed an amino-myc–tagged isoform of the mouse and zebra finch genes in COS-7 cells. Surface staining revealed that both proteins are presented on the cell membrane (Fig. 2A). Next, we used an assay where a chimeric protein composed of the extracellular domain of chicken EphA4 fused to placental alkaline phosphatase is tested for binding to different EFNB3 orthologs expressed in transiently transfected COS-7 cells, noting that chicken and the zebra finch EphA4 proteins share 98.4% identity (fig. S3). We found that chicken EphA4 binds mouse ephrin-B3 but not zebra finch ephrin-B3 (Fig. 2, B and C). Hence, the alterations in the sequence and structure of the extracellular domain of ephrin-B3 in songbirds have impaired its binding to the EphA4 receptor.

Fig. 2 Zebra finch ephrin-B3 protein and mRNA expression.

(A) The mouse and zebra finch proteins are presented on the cell membrane. An anti-myc antibody (9E10) was used for surface staining of COS-7 cells transiently transfected with N-terminal myc-tagged ephrin-B3 isoforms. The sequence of the full-length zebra finch ephrin-B3 transcript was reconstructed from RNA-seq reads using the Trinity platform (60) (fig. S1). DAPI, 4′,6-diamidino-2-phenylindole. (B) Binding assay shows EphA4-AP binding to mouse ephrin-B3 protein but not to GFP or to zebra finch ephrin-B3. (C) Representative EphA4-AP/ephrin-B3 binding curves assessed by AP activity in a colorimetric assay. Apparent KD value for mouse ephrin-B3 was 40.15 nM (n = 3). (D) In situ hybridization for ephrin-B3 mRNA in P0 zebra finch spinal cord. High expression is evident in the FP but declines along the RP. (D’) Normalized optical density (OD) profile of zebra finch ephrin-B3 midline expression; shown is the mean from 58 cross sections. (E) In situ hybridization for ephrin-B3 mRNA in E15.5 mouse spinal cord. High expression is evident in both the FP and RP. (E’) Normalized optical density profile of mouse ephrin-B3 midline expression; shown is the mean from 160 cross sections. (F and G) Analysis of the activity of the ephrin-B3 enhancer element, at E9 (F) and E15 (G). The cherry reporter gene was cloned downstream to the putative enhancer and electroporated into the chick spinal cord. GFP, under the control of the ubiquitous CAGG enhancer-promoter, was coelectroporated. The reporter is detected in both the FP and RP (F and G), similar to the pattern of ephrin-B3 expression in mice, while GFP is expressed ubiquitously.

Expression of ephrin-B3 in zebra finch

The presence of the EFNB3 gene in songbirds prompted us to examine its expression pattern in the zebra finch spinal cord. In situ hybridization with antisense RNA probes revealed ephrin-B3 mRNA expression in both the floor plate (FP) and RP of the brachial spinal cord in post hatching day 0 (P0) zebra finches, with apparent lower expression in the RP (Fig. 2D). Quantification confirmed that ephrin-B3 expression is higher in the FP and declines along the RP in the ventral-to-dorsal direction (Fig. 2D’). This contrasted with mouse at a comparable developmental stage, where expression is similarly high in both the FP and RP (Fig. 2E), with no dorsal decline (Fig. 2E’).

We reasoned that the modified ephrin-B3 expression seen in zebra finch compared to mouse might relate to changes in regulatory elements. Using the UCSC Genome Browser, we searched for enhancer hallmarks in mouse EFNB3 and its flanking genes. We also examined the existing mouse Hi-C datasets using an online tool (; details in Materials and Methods). We found that mouse DNAH2, located immediately downstream of EFNB3 (Fig. 1D, blue), contains several features consistent with the presence of an enhancer element in its first intron: binding of modified histones that are associated with enhancer elements, deoxyribonuclease I hypersensitivity, genomic conservation, and the fact that the putative enhancer and the EFNB3 gene reside within the same topologically associating domain (30) (fig. S4). Because the DNAH2 gene is not present in birds [Fig. 1D, blue; (19)], birds may lack an important EFNB3 enhancer element. To test whether this DNAH2-embedded region in mice would be sufficient to confer midline RP expression in the spinal cord of a bird species, a reporter gene cloned downstream of the putative enhancer element was electroporated into the chick spinal cord at embryonic day 3 (E3). Expression of the reporter in both the FP and RP was evident at E9 and E15 (Fig. 2, F and G), similar to mouse where endogenous ephrin-B3 midline expression occurs during embryogenesis and early postgestation (5, 31) (E15 chick is equivalent to mouse P2) and contrasting with the more ubiquitous expression of the coelectroporated green fluorescent protein (GFP) under a ubiquitous promoter (Fig. 2, F and G, insets). Notably, DNAH2 is missing in all bird species examined to date, but it is present in reptiles [Fig. 1D; (19)], suggesting that the loss of the EFNB3 regulatory element is specific to the avian lineage.

Expression of ephrin-B family ligands and EphA4 in chick spinal cord

We next used in situ hybridization to examine the expression of ephrin-B1 and ephrin-B2 at the chick brachial cord, as both can mediate axonal repulsion through the EphA4 receptor and could thus potentially compensate for an EFNB3 loss. Both ephrin-B1 and ephrin-B2 were expressed in the FP and RP at E10 (Fig. 3, A and C). At E16, when the spinal circuitry elicits simultaneous wing flapping (32), ephrin-B1 expression persisted in the FP, but dorsally, it became restricted to the ventral aspect of the RP (Fig. 3B), whereas ephrin-B2 expression was detected only at the ventral tip of the RP (Fig. 3D). Expression of ephrin-B genes in the chick spinal cord is thus largely restricted to the FP, low in the ventral RP, and absent in the dorsal RP.

Fig. 3 Expression of ephrin-B1, ephrin-B2, and EphA4 mRNAs in the chick spinal cord.

In situ hybridization for ephrin-B1 (A and B), ephrin-B2 (C and D), and EphA4 (E and F) mRNAs at the brachial level in E10 (A, C, and E) and E16 (B, D, and F) cords. The FP is indicated by black arrows and the RP by magenta arrows. (A and B) Ephrin-B1 at E10 is high in the FP and low in the RP; at E16, it is high in the FP and low in ventral RP. (C and D) Ephrin-B2 at E10 is low in the FP and RP; at E16, it is low in ventral RP. (E and F) EphA4 expression at E10 is widespread in the gray matter, with higher levels in scattered interneurons, the lateral subdivision of the lateral motor columns (LMCl) and the medial motor columns (MMC). At E16, expression is detected in scattered interneurons and motoneurons. Interneurons invading the dorsal midline are apparent [white arrow in inset in (F)]. Insets in (E) and (F) are enlargements of the areas boxed in dashed yellow. (E’ and F’) Cell density distribution, obtained from 10 cross sections from the regions flanking the dorsal midline at E10 and E16, respectively [indicated by white dashed boxes in (E) and (F)]. (G) Summation of cell in situ hybridization relative intensity in the area indicated in (E’) (E10, blue) and (F’) (E16; red values are significantly different at the dorsal midline) (t test, P < 0.05, n = 10 per group).

Next, we studied the expression of the receptor EphA4. In mice, interneurons that express EphA4 are excluded from the dorsal midline, but many of these cells invade the dorsal midline in EphA4 and ephrin-B3 null mutants (33). In the chick, expression is detected throughout the spinal cord gray matter at E10, with high levels in scattered interneurons, in the lateral subdivision of the lateral motor column, and in the medial motor column, but absent in the dorsal midline (Fig. 3E). At E16, EphA4 shows high expression in discrete cells distributed all over the gray matter, including adjacent to and within the dorsal midline (Fig. 3F, white arrow in inset).

When we assigned x-y coordinates to EphA4-expressing cells, we observed their midline invasion at E16 compared to E10 (Fig. 3, E’, F’, and G). Thus, the presence of EphA4-expressing neurons in the dorsal midline correlates with the temporal reduction of ephrin-B1 and ephrin-B2 levels. Together with the lack of ephrin-B3, we conclude that a molecular barrier that prevents midline crossing is missing in the chick dorsal spinal cord. This contrasts markedly with mice, where midline expression of ephrin-B3 persists at high levels both ventrally and dorsally at postgestation stages (5).

Extended RP of the chick brachial cord

To further investigate how the natural loss of EFNB3 might have affected the organization of the avian spinal cord, we examined several parameters that differ substantially between WT and hopB3 mutants, namely the size and axonal decussation at the RP and the somata distribution and neurotransmitter phenotype of commissural pre-MNs (cpre-MNs) (Figs. 4 to 6). In WT mice, the length of the FP was relatively constant along the longitudinal axis, but the RP was substantially shorter at limb levels, resulting in low RP:FP ratios at those levels (fig. S5). Thus, the RP:FP ratio can serve as a reasonable proxy for the size of the RP. The length of the RP at limb levels was markedly enlarged in EphA4 knockout mice (Fig. 4, B and D) in comparison to control mice (Fig. 4, A and C), with a significantly higher RP:FP ratio in the homozygous mutant (1.68 ± 0.22, mean of RP:FP ratios ± SD) than in control mice (0.34 ± 0.13). In birds, an elongated RP was evident at brachial levels in E15 chick (Fig. 4E) and P8 zebra finch (Fig. 4G), with RP:FP ratios of 2.18 ± 0.2 and 1.53 ± 0.08, respectively, thus comparable to EphA4 knockout mice (Fig. 4H). Hence, the morphology of the avian brachial RP is similar to that of hopB3 mutants. At the lumbar sciatic level, the dorsal midline in birds is occupied by an ovoid gelatinous mass, the GB (Fig. 4F), an avian-specific organ. Anatomical and histological studies demonstrate the presence of the GB in various bird species belonging to different clades (34, 35). These features differ from those in alternating limbed reptiles (lizard and turtle) where the RP is short, as in WT mice (36, 37), and no GB has been described.

Fig. 4 Spinal cord shows enlarged brachial RPs in hopB3 mutants, chick, and zebra finch.

(A to D) Dark-field microphotographs of transverse sections of adult mouse spinal cord at lumbar and brachial levels. Compared are control heterozygous (EphA4lx/lx−) (A and C) and homozygous null mice (EphA4lx/lx−/PGK-Cre+) (B and D) (33). Right insets (enlargements of the boxed areas) depict the central canal and the RP and FP. (E and F) Transverse sections at the brachial (E) and lumbar (F) levels of E15 chick spinal cord. The white matter is labeled with the anti-neurofilament antibody 3A10, the gray matter with nuclear DAPI staining. Right inset [in (E)] is an enlargement of the boxed area. (G) Transverse section at the brachial level of P8 zebra finch. The white matter is labeled with the anti-neurofilament antibody 3A10. Right inset is an enlargement of the boxed area. (H) Box plot representation of the RP-to-FP length ratios at the brachial level in adult EphA4+/− mouse (N = 3, 36 cross sections), adult EphA4−/− mouse (N = 3, 36 sections), E15 chick (N = 2, 22 sections), and P0 zebra finch (N = 1, 25 sections). Comparing the RP/FP ratios across groups using Dunnett’s method (which takes into account multiple comparisons) shows significant differences between the heterozygous mouse and the null mouse, chick, and zebra finch. The circle charts shows the significance. No overlapping between the heterozygous mouse (red circle) and other groups (black circles) is indicative of P < 0.05 (see Supplementary Statistics).

Fig. 5 Midline crossing at the brachial and lumbar levels of chick spinal cord.

(A) Dorsal midline crossing at brachial level in E15 chick demonstrated by neurofilament (3A10) immunostaining in transverse sections. Detailed views of three examples in insets; left inset is an enlargement of the boxed area in (A). Axons cross at the FP (yellow arrows) and the dorsal part of the RP (white arrows). (B) Dorsal midline crossing at brachial level in E17 chick; the boxed area is enlarged in (B’) and (B”). Anti-3A10 labels all axons, and anti-TAG1/axonin labels sensory axons. (C) Dorsal midline crossing is evident following unilateral expression of a reporter gene. GFP under the control of the CAGG enhancer/promoter construct was electroporated at E3 into the left side of the spinal cord. At E15, midline-crossing axons are evident in the brachial FP (yellow arrows) and RP (white arrows). Detailed views of three examples in insets; left inset is an enlargement of the boxed area in (C). (D and E) Transverse sections at the sciatic plexus level at E15 (D) and E17 (E). Axons cross only at the FP (white arrows). The GB lacks axonal labeling. (F) Transverse section at the crural plexus level at E17. Right inset is an enlargement of the boxed area in (F). Axons cross at the FP (yellow arrow) and RP (white arrow). (G) Box plot of the ratio between axons crossing the RP and the FP at different axial levels in E15 chick. Numbers of midline-crossing axons were scored at cervical (15 sections), brachial (31 sections), thoracic (27 sections), lumbar crural (23 sections), and lumbar sciatic (14 sections) levels of one embryo. At the sciatic level, axons did not cross the GB (see Supplementary Statistics).

Fig. 6 Patterns of pre-MNs in the brachial and lumbar chick spinal cord.

(A to C) Cell distribution maps in the brachial and lumbar segments of E15 spinal cord sections of chicks in which the hindlimb (A), distal wing (B), or pectoral (C) musculature was injected with PRV-cherry at E14. Data are from 1038 ipre-MNs and 88 cpre-MNs in (A) (three embryos), 620 ipre-MNs and 210 cpre-MNs in (B) (four embryos), and 734 ipre-MNs and 392 cpre-MNs in (C) (two embryos). Ellipses in dashed magenta represent the areas occupied by motoneurons. (A’ to C’) Images of PRV-positive neurons supplemented with immunostaining for motoneurons (ChAT); right insets are enlargements of the midline boxed area showing axonal decussation of pre-MNs at the FP (yellow arrows) and RP (white arrows); white ellipses represent the central canal. (D) Identification of a subpopulation of pre-MNs that innervate pectoral muscle motoneurons bilaterally. PRV-GFP and PRV-cherry were injected bilaterally at E14 into the pectoral muscles. Cell distribution map of bilaterally innervating pectoral pre-MNs seen at E15 is shown in gray. (D’) Examples of bilaterally innervating pre-MNs. Left: Merged images. Middle: PRV-cherry. Right: PRV-GFP. Out of 500 and 730 cpre-MNs on each side (18 sections), 11.9% ± 0.05% and 10.9% ± 0.04% were double-labeled.

Patterns of axonal midline crossing in the chick spinal cord

Axon fibers that cross the RP in rodents originate mostly from sensory neurons and inhibitory interneurons (38, 39). To further study the extent of dorsal midline crossing in the chick embryonic spinal cord, we examined immunoreactivity for neurofilament (3A10) and for sensory axons (TAG1) at E15 and E17. We observed neurofilament-positive axons crossing the brachial FP in a tight bundle (Fig. 5A, yellow arrows in insets), whereas those crossing the dorsal midline were less fasciculated and sparsely distributed along the dorsal RP (Fig. 5A, white arrows in insets). Quantification showed that 44.8 ± 23% of midline-crossing axons at the brachial level were in the RP (Fig. 5G, brachial). TAG1-expressing sensory axons did not appear to cross the midline at this level (Fig. 5B). To further interrogate midline axonal crossing, a membrane-tethered GFP was electroporated at E3 into one side of the spinal cord. At E15, GFP-labeled axons crossing the ventral and dorsal midline were evident (Fig. 5C, yellow and white arrows in insets). A few labeled cells were apparent on the contralateral side dorsally, suggesting that some cells may migrate through the RP at the brachial level (Fig. 5C, white arrowheads in insets).

At the lumbar sciatic level, axonal decussation was apparent only through the FP (Fig. 5, D and E), with no axons crossing the GB. At the lumbar crural level, which is devoid of GB, dorsal midline crossing was apparent, with 31 ± 12% of crossing fibers occurring through the RP (Fig. 5, F and G, crural). Notably, the sciatic nerve innervates the muscles of the shank and foot, while nerves originating from the crural plexus innervate the thigh muscles (40, 41). In the swing phase of birds, the ankle flexion leads to the elevation of the feet, while the knee is relatively stable. Hence, the sciatic nerve mediates the flexion and extension of the ankle and foot joints during stepping. Significant dorsal midline decussation was also apparent at the cervical, brachial, and thoracic levels (Fig. 5G), as well as sacral levels (not shown), suggesting that the GB serves as a physical barrier for axonal decussation at the sciatic plexus level, thus allowing hindlimb alternation. In contrast, dorsal midline crossing at brachial levels correlates with synchronous wing (forelimb) movements, similar to that seen in hopB3 mice.

Connectivity of pre-MNs in the chick spinal cord

The synchronous gait of hopB3 mice results from ectopic midline crossing of excitatory pre-MN through the FP and RP (Fig. 1B) (5, 42). To analyze the patterns of connectivity of pre-MNs at the lumbar and brachial levels of the chick, we injected the transsynaptic pseudorabies virus (PRV) (43, 44) into the leg, the distal wing musculature, and the main flight muscle, the pectoralis major. We observed substantial differences in the densities and laminar distributions of cpre-MNs at lumbar versus brachial levels (Fig. 6, A to C, right). cpre-MNs accounted for 7.81% of labeled cells at the sciatic level and for 25.3 and 34.8% of brachial level cells labeled from the distal wing and the pectoral muscle, respectively.

The distribution of ipsilateral pre-MNs (ipre-MNs) was similar at lumbar and brachial levels, occupying the medial spinal cord [Fig. 6, A to C (left), and fig. S6]. In contrast, the distribution of cpre-MNs was markedly different across levels. At lumbar levels (both sciatic and crural), we observed cpre-MNs ventromedially, with most cells ventral to the central canal (Fig. 6A, right, and fig. S6). At the brachial level, cpre-MNs of the distal wing musculature and the pectoral muscle were distributed ventrally and dorsally to the central canal (Fig. 6, B and C, right). Likewise, decussation of pre-MNs differed markedly between lumbar and brachial levels, with axons crossing only the FP at the lumbar level (Fig. 6A’) but both the RP and the FP at the brachial level (Fig. 6, B’ and C’).

Synchronous gait in mice hopB3 mutants is also associated with pre-MNs that innervate motoneurons bilaterally (42). To directly test whether the chick brachial cord contains these cells, we injected PRV-GFP and PRV-cherry into the left and right pectoral muscles, respectively (Fig. 6D, left). A total of 10.9% of pectoral cpre-MNs were double-labeled, most located above the central canal and adjacent to the dorsal midline (Fig. 6D), with neurite processes oriented toward the dorsal midline (Fig. 6D’). Hence, like in hopB3 mice, a subpopulation of the brachial pre-MNs innervate motoneurons bilaterally (42).

To examine the neurotransmitter phenotype of spinal pre-MNs, we processed sections from hindlimb and wing PRV-injected embryos for in situ hybridization with a marker for excitatory neurons (vGlut2). At the lumbar level, 46.5% of the ipre-MNs and 38.6% of the cpre-MNs were excitatory. At the brachial level, 35% of the ipre-MNs and 38% of the ventral cpre-MNs were excitatory. Of the brachial-specific cpre-MNs that reside dorsal the central canal, 47% were excitatory (table S1 and fig. S7). The chick brachial-specific pre-MNs are thus comparable to the ectopic cpre-MNs in hopB3 mutants in location, wiring, and excitatory phenotype (10, 42).

Exogenous ephrin-B3 expression in the chick RP prevents midline axonal crossing

Our observations, so far, suggest that dorsal midline crossing and the wiring pattern of brachial pre-MNs in the chick spinal cord might be linked to a lack of ephrin-B3 expression due to the genomic loss. To test this hypothesis, we examined whether expression of mouse ephrin-B3 in the chick would affect RP neurite crossing. Mouse ephrin-B3 or GFP, cloned into a Cre-dependent expression vector under the RP-specific Wnt1 enhancer element (45), was coelectroporated into the E3 chick spinal cord (Fig. 7A), and the extent of RP neurite crossing was analyzed at E15. To exclude possible effects of exogenous ephrin-B3 in reverse signaling, we also used an ephrin-B3 isoform that lacks the intracellular domain (ephrin-B3-ΔC). Ephrin-B3 expression was confined to the RP (Fig. 7B), and the number of midline-crossing neurites was significantly lower for ephrin-B3 (7.1 ± 3.1) or ephrin-B3-ΔC (7.1 ± 2.6) than for GFP (16.5 ± 4.6) (Fig. 7, C to E). Hence, as in mammals, midline ephrin-B3 expression through forward signaling (5, 8) is sufficient to prevent midline crossing.

Fig. 7 Expression of exogenous ephrin-B3 in the chick RP prevents midline crossing.

(A) Strategy for ectopic expression of the mouse ephrin-B3 at the RP. (B) Cross section from brachial spinal cord of E17 embryo expressing ephrin-B3 (magenta) at the dorsal midline, immunostained for neurofilament (cyan). (C and D) Magnifications of three examples of dorsal midline region of spinal cords electroporated with GFP (C, C’, and C”) or ephrin-B3 (D, D’, and D”) (both in magenta), immunostained for neurofilament (cyan). The black and white images are the neurofilament (3A10) staining. Some neurites cross the GFP-expressing RP (C) but not the ephrin-B3–expressing RP. (E) Quantification of dorsal midline crossing in spinal cords expressing GFP (N = 43 sections, three embryos), mouse ephrin-B3 (N = 71 sections, three embryos), or ephrin-B3-ΔC (N = 59 sections, three embryos). Comparing the number of neurites in the ephrin-B3 and ephrin-b3-ΔC RP to the control (GFP-expressing) RP using Dunnett’s method shows a significant difference between the control group and the experimental groups. The circle charts show the significance of these results. No overlapping between the control group (red circle) and the experimental groups (black circles) is indicative of P < 0.05 (see Supplementary Statistics). Thus, the numbers of neurites crossing ephrin-B3– or ephrin-B3-ΔC–expressing RPs were significantly lower than in GFP-expressing controls (see Supplementary Statistics).

Next, after targeting ephrin-B3 expression to the RP at E3, PRV was injected into the wing musculature at E14 (Fig. 8A), and the numbers and distribution of cpre-MNs were examined at E15.5. In controls, both ipre-MNs and cpre-MNs were seen, the latter both dorsal and ventral to the central canal (Fig. 8B), and their neurites crossed at both the FP and RP (Fig. 8B’). In ephrin-B3–expressing embryos, fewer cpre-MNs were seen dorsal to the central canal (Fig. 8, C, C’, C”, D, and E), and most neurites were observed to cross the FP (Fig. 8, C, C’, and C”, right insets). Short cell extensions, possibly dendrites originating from ipre-MNs, did cross the RP (Fig. 8, C and C’, right insets). We note that in ovo electroporation results in mosaics where only a portion of the cells expresses the transgene. Our efficiencies, assessed as sections with midline expression of ephrin-B3, varied from 5 to 64% (Fig. 8D); thus, axons circumvent ephrin-B3 and cross the midline close to non-EFNB3–expressing RP cells (movie S1). The ectopic expression of ephrin-B3 did not affect the dorsoventral extent of the RP (Figs. 7, C and D, and 8, B and C). The mosaic noncontinuous expression of the exogenous ephrin-B3, attained via in ovo electroporation, may explain the lack of an effect on the size of the RP, in the manipulated chicks.

Fig. 8 Expression of exogenous ephrin-B3 in the chick RP prevents pre-MN crossing.

(A) Schematic of the experimental setup. GFP or ephrin-B3 was expressed in the midline at E3. At E14, PRV-cherry was unilaterally injected into the wing musculature, and embryos were fixed and processed 38 hours later. ChAT-immunostained (cyan) MNs and PRV-labeled (magenta) pre-MNs are indicated; dashed line represents cpre-MN axons that cross the RP. (B and C) Transverse sections of control (B) and Wnt1::ephrin-B3–expressing E15 embryos [three examples in (C), (C’), and (C”)]. The black and white images are in (B’), and the insets in (C), (C’), and (C”) are PRV staining. In the control (B), cpre-MN somata are located ventral and dorsal to the central canal, and neurites of cpre-MNs are apparent at the FP [yellow arrow in (B’)] and RP [white arrow in (B’)]. In Wnt1::mEFNB3 embryos (C, C’, and C”), the somata of pre-MNs are mainly located ventral to the central canal. Axons crossing the FP (yellow arrows) and few short neurites (white arrowheads) crossing the dorsal midline are shown in the right insets. (D) Density maps of brachial cpre-MNs in control (GFP, N = 627 cells from three embryos) and five ephrin-B3–expressing embryos (N = 341, 547, 348, 48, and 72 cells). The fraction of sections containing ephrin-B3 expression along the entire dorsal midline is indicated at the top of each spinal cord. A ventral shift of the distribution is apparent in the high-efficiency electroporated spinal cords of Wnt1-B3-4 and Wnt-1-B3-5. (E) Plot demonstrating the correlation between the percentage of dorsal cpre-MNs from all pre-MNs and the electroporation efficiency [as described in (D)]. The average of three GFP-expressing embryos is indicated as a green dot.

Next, we analyzed the spatial distribution of cpre-MNs in five embryos with variable efficiencies of wnt1::ephrin-B3 expression (Fig. 8D). A dorsal to ventral shift in the location of cpre-MNs was apparent when comparing the embryos with lower versus higher exogenous ephrin-B3 expression in the RP (1-3 versus 4-5 in Fig. 8D), suggesting a reduction in the size of the brachial-specific dorsal population of cpre-MNs. To further quantify this observation, we subdivided cpre-MNs into ventral and dorsal groups, based on location relative to the central canal, and examined their numbers compared to ipre-MNs. The distribution of pre-MNs in three GFP-expressing embryos served as a control. The percentage of dorsal cpre-MNs from the total number of pre-MNs was calculated and correlated with the efficiency of the midline expression of ephrin-B3, using linear regression (Fig. 8E and table S2); the average number of pre-MNs in the three GFP-expressing embryos was included as having null midline ephrin-B3 expression efficiency. The Pearson correlation coefficient between efficiencies of ephrin-B3 expression and the percentage of dorsal cpre-MN was −0.95 (P = 0.0035). Thus, the dorsal cpre-MN population is reduced in proportion to higher ephrin-B3 expression at the RP.

We also calculated the ratios between pre-MNs populating the different zones (ventral cpre-MNs, dorsal cpre-MNs, and ipre-MNs) and correlated these ratios with the efficiencies of midline ephrin-B3 expression (fig. S8). The ratio of ventral cpre-MNs to ipre-MNs was similar in controls and all experimental embryos (fig. S8A); thus, ephrin-B3 RP expression did not affect this population. The ratio of dorsal cpre-MNs to ipre-MNs was reduced in proportion to higher ephrin-B3 expression (fig. S8B), as was their ratio to ventral cpre-MNs (fig. S8C). The ratio of all cpre-MNs to the ipre-MNs was also negatively correlated with increasing ephrin-B3 expression (fig. S8D). We conclude that robust ephrin-B3 RP expression results in fewer dorsal cpre-MNs that cross through the RP, while ventral cpre-MNs that cross through the FP are not affected.


During vertebrate evolution, most terrestrial species adopted alternating limb motion as an effective gait strategy, while adaptation to powered flight in birds emerged via morphological changes that patterned wings from forelimbs and required transition from alternate gait to synchronous flapping. Our genomic analysis indicates that the gene encoding the midline axonal repellent ephrin-B3 is missing in chicken and other galliforms. In songbirds, the EFNB3 gene encodes an ephrin-B3 protein that does not bind to the EphA4 receptor and whose expression in the spinal cord is reduced along the RP, probably due to the loss of a downstream enhancer. Morphological and circuit analyses revealed that the chick spinal cord shares similarities with WT mice at the lumbar level (Fig. 9, A and C) and with hopB3 mutant mice at the brachial level (Fig. 9, B and D). Dorsal midline expression of exogenous (rodent) ephrin-B3 in the chick brachial spinal cord prevents crossing of cpre-MN axons (Fig. 9E). Together, our findings support the notion that a reduction in ephrin-B3 expression and/or function plays an important role in shaping the wing level circuitry in the avian spinal cord.

Fig. 9 Spinal circuits that enable limb alternation, hopping, or flying in mouse and chicken.

(A and B) Schematic illustration of the morphology and circuit wiring of the spinal cord in WT mouse (A) and in hopB3 mutants. (C and D) Schematic illustration of the morphology and circuit wiring of the avian spinal cord at the lumbar (C) and brachial (D) levels. We propose that dorsal midline crossing is prevented at the lumbar level by the GB but enabled at the brachial level due to the loss of ephrin-B3 activity. The brachial avian spinal cord shares significant morphological and wiring similarities with mouse hopB3 mutants. (E) Schematic illustration of the circuitry at the brachial level of the chick following expression of the mouse ephrin-B3 at the RP. The circuitry resembles that of the WT mouse (A) and the chick lumbar level (C).

Loss of ephrin-B3 function in birds

Changes in gene copy number, enhancers, and protein sequence—arguably some of the foremost molecular events in evolution—are thought to have influenced locomotion patterns in vertebrates. For example, a point mutation that leads to a premature stop codon in DMRT3 enabled Icelandic horses to develop their characteristic Tölt ambling gait (46). A modification and subsequent loss of a regulatory element of GDF6 during the transition to bipedalism in humans is thought to have played a role in foot digit shortening (47). Limb loss in snakes has been shown to be a result of mutations in a limb-specific enhancer of Sonic hedgehog (48). Loss of function for ephrin-B3 leads to midline crossing in the spinal cord of rodents, resulting in a remarkable hopping gait (5, 6). Our findings now indicate that the ephrin-B3 gene has undergone significant modifications in birds compared to nonavian groups, likely affecting flight control circuits in the avian spinal cord.

Absence of DNAH2 is seen in all birds where this genomic region could be assessed, suggesting that the loss of a putative ephrin-B3 enhancer was an early event in avian evolution or possibly even in dinosaurs. It seems reasonable to hypothesize that the enhancer loss preceded the loss/degradation of functional domains, but at present, it is hard to conclusively answer that question given the lack of data on EFNB3 gene structure and synteny in the most basal avian clade, Palaeognathae. The enhancer that we characterized is probably also not the only EFNB3 expression regulator, since zebra finches show endogenous EFNB3 expression in the FP. Targeted deletion of the DNAH2-embedded enhancer from the mouse genome may further reveal its contribution to the pattern and level of EFNB3 expression. The apparent loss of EFNB3 in chicken and other galliforms was likely a more recent event specific to that basal clade, since EFNB3 is present in Anseriformes and in several Neoaves, including the more recently evolved Passeriformes. It is hard to estimate the timing of mutations in the coding region, since high-quality sequence is limited to a few species. Nonetheless, similar protein changes compared to mammals and nonavian sauropsids are present in eagles and songbirds (fig. S2), avian groups whose split from their sister taxa is estimated to have occurred ~63 to 64 and ~55 to 61 million years ago, respectively (2022). High-quality sequences of this region in other avian clades will be critical for better understanding the timing and extent of avian EFNB3 changes.

In zebra finch, besides a predicted lack of reverse signaling due to a loss of specific domains, we provide clear evidence that forward signaling through the EphA4 receptor is impaired. The lack of binding to avian EphA4, for example, is likely related to substitutions in residues 108 to 111 (in the mouse protein). This region is within the G β strand, which forms a loop with the H β strand, and is part of the ligand-receptor dimerization interphase (49). In amphibians and nonavian sauropsids, this region is also quite divergent compared to mammals (fig. S2), even though other domains are more conserved than in birds. Future experiments using chimeric protein constructs across these groups may reveal further insights into the evolution of ephrin-B3 and its possible role in controlling spinal cord wiring and pattern of locomotion across vertebrates. We also note that we currently do not know whether the lack of functional domain in the intracellular portion of avian ephrin-B3 causes impairments to other functions that depend on reverse signaling in ephrin-B3–expressing cells.

Phenotype of avian and hopB3 mouse spinal cords

Our data show that the brachial spinal cord in chicken, which lacks an EFNB3 gene, shares similarities with mouse hopB3 mutants in its morphology and in the wiring and excitatory/inhibitory balance of cpre-MNs (Fig. 9, B and D). The significance of the dorsal midline expansion, anatomically the most profound alteration in hopB3 mutants, is unclear. While synchronous motoneuron activity is seen in the spinal cord of hopB3 mutants using an in vitro fictive locomotion assay, severing of the dorsal midline does not restore left-right alternation (4, 50), arguing that bilateral coupling is facilitated by ventral midline crossing. However, in vivo disruption of the RP integrity is sufficient to induce synchronous gait in mice (9, 10). It is conceivable that in hopB3 mutants, proprioceptive signals from weight-load receptors, which are used only in vivo, inhibit the bilateral coupling mediated by the dorsally crossing axons. In support of this hypothesis, removal of EphA4 from a subpopulation of dorsal interneurons results in ectopic dorsal midline crossing. Furthermore, the overground gait of these mice is normal, while during air-walking and swimming, their limbs move in synchrony (42). In birds, dorsal midline crossing might be sufficient to instruct the bilateral coupling of the wing motoneurons, since weight load without ground support during flight does not affect the wings.

The importance of excitatory neurons for synchronous gait in mouse hopB3 mutants is shown by conditional targeting of EphA4 in vGlut2 neurons, which results in hopping gait (4). Furthermore, a rather small shift in the excitatory/inhibitory balance in cpre-MNs, from 21% excitatory and 46% inhibitory in WT to 26% and 35% in the EphA4 null mouse, is sufficient to induce synchronous gait (4, 50). We found that the brachial dorsal population of cpre-MNs in chick is moderately but significantly more enriched (−47%) with excitatory neurons than in WT mice. Furthermore, the difference between this population and the brachial ventral and lumbar cpre-MNs, both corresponding to ~38% of excitatory neurons (Fig. 6 and table S1), is similar to the difference between WT and hopB3 mutant mouse. These observations support the notion that the brachial-specific population of cpre-MNs is sufficient to support synchronous activation of motoneurons involved in wing flapping.

Intriguingly, the GB (Fig. 9C) is a prominent but poorly understood dorsal lumbar gelatinous structure that arises from RP cells, is present across birds, and may have been present in dinosaurs (34, 35, 51). Our study has uncovered evidence that the GB serves as a major physical barrier to dorsal midline crossing at sciatic lumbar levels in the chick. We thus suggest that the GB is likely a major contributor to the alternating gait in birds, analogous to the molecular barriers to midline crossing in mice. Further comparative studies are needed to further elucidate the exact role of the GB across birds with predominantly hopping versus alternating gaits.

Alternative mechanisms and further comparative notes

Our findings do not exclude the possibility that other genetic mechanisms might play important roles in shaping avian spinal circuits underlying synchronous wing movement. For example, some molecular pathways involved in midline crossing at the FP have been described as missing or altered in birds. DCC, a receptor for the axon guidance modulator Netrin-1, is missing in chicken and zebra finch but is preserved in other avian groups such as ducks (5254). Robo3, a receptor for Slit ligands that are involved in axonal guidance and required for midline crossing, shows distinct features in birds compared to mammals. While avian Robo3 binds Slits, mammalian Robo3 specifically interacts with DCC (55). Presently, it is unclear whether these modifications contribute to species-specific spinal functions or whether they preceded or followed the ephrin-B3 function loss described here. Other mechanisms that could contribute to the modulation of flight-related circuitry also remain to be investigated, including changes in synaptic input, axonal trajectory, or neurotransmitter identity of the interneuron populations involved in the alternation-to-synchronization transition.

Some vertebrate lineages other than birds have evolved synchronous limb movement, including flight in bats and hopping in postmorphogenesis anurans (frogs and toads) and macropods (kangaroos and wallaby). Protein alignment revealed high levels of conservation of bat and wallaby ephrin-B3 compared to mouse (fig. S2). Thus, mechanisms unrelated to ephrin-B3 changes are more likely to instruct synchrony-mediating spinal circuitry in these mammals. Among amphibians, salamander walks via alternation and shows higher ephrin-B3 similarity to mice, whereas anurans move by synchronous leg movements and show lower ephrin-B3 similarity to mice (fig. S2). Studies of ephrin-B3 expression and activity in amphibians are needed to test whether modulation of this signaling pathway has shaped the spinal circuitry for synchronous leg movements in anurans. Last, we note that, because of technical limitations, it has not been possible to directly assess the role of this pathway in avian flight behavior. Approaches other than electroporation, such as germline transgenesis, will likely provide a more suitable experimental paradigm to test the physiological role of ephrin-B3 gain of function in birds such as chicken or quail.

In summary, we provide evidence linking loss of ephrin-B3/EphA4 signaling to the emergence of synchrony-mediating circuitry in the avian brachial spinal cord. While our findings do not exclude the importance of other molecular pathways, they suggest that modulation of ephrin-related mechanisms may have been an important contributor to the origin of flight in birds.



The EFNB3 genomic region in birds has presented several difficulties for sequencing and assembling, likely due to very high GC content and density of poorly characterized repeat elements. The gene and its syntenic region are not present, or present only in partial form, in most Sanger- and Illumina-based avian assemblies, but are better represented in the more recent PacBio assemblies (23, 56). Furthermore, this region is likely part of a short avian microchromosome, which, until the recent Swainson’s thrush (Catharus ustulatus) genome, had not been assembled and/or annotated in any species. Our current study of EFNB3 in birds has therefore required considerable search, annotation, and curation efforts.

Briefly, we initially examined the annotated databases [National Center for Biotechnology Information (NCBI) RefSeq and Ensembl] available for >50 avian species, searching for gene predictions annotated as EFNB3 or ephrin-B3–like and using reciprocal BLAT alignments and verification of conserved synteny of the predictions to confirm their orthology with humans and nonavian vertebrates [e.g., green anole (Anolis carolinensis) and Chinese alligator (Alligator sinensis)]. We were able to identify EFNB3 and verify its orthology in several songbird species [zebra finch (Taeniopygia guttata), Bengalese finch (Lonchura striata domestica), common starling (Sturnus vulgaris), Tibetan ground-tit (Pseudopodoces humilis), small tree finch (Camarhynchus parvulus), and Swainson’s thrush (C. ustulatus)], in the golden eagle (Aquila chrysaetos) and bald eagle (Haliaeetus leucocephalus), and in a parrot, the kakapo (Strigops habroptilus). Notably, in the Swainson’s thrush, EFNB3 and syntenic genes have been assigned to chr37, the first case any genes have been assigned to that avian microchromosome. We also note that two apparently identical zebra finch EFNB3 models (discontinued Gene IDs: 115492780 and 115492629) were present in an unplaced scaffold (NW_022045341.1) of an early PacBio assembly [bTaeGut1_v1.p; (56)] but have been recently discontinued as this entire region is not present in the latest zebra finch PacBio assemblies in NCBI (as of March 2021). We used the discontinued XM_030261704.1 transcript prediction in our study, noting that it is well supported by expressed sequence tags (ESTs) and transcriptome data (fig. S1, A and B). Ensembl has EFNB3 predictions for the zebra finch, kakapo, and the Eurasian sparrowhawk (Accipiter nisus), the latter in a short segment and right upstream of KDM6B. We also BLAST-searched avian and nonavian RefSeq databases, using as queries the genomic regions just downstream of avian gene predictions annotated as WRAP53 (or WD repeat containing antisense to TP53, a.k.a. telomerase Cajal body protein 1) or just upstream of avian gene predictions for KDM6B (or lysine-specific demethylase 6B). This effort identified EFNB3 in the barn owl (Tyto alba) and in another songbird species, even though no EFNB3 predictions were present. We also BLAST-searched all existing avian genomes (May 2020) and the respective WGS databases, using as queries the existing avian EFNB3 predictions, and manually examined all hits with scores of >50, excluding cases that did not show top reciprocal alignments (typically other EFNB family members). This effort identified EFNB3 in the double-crested cormorant (Phalacrocorax auritus) and in a few other passerines and anseriforms with limited synteny information, as well as in several species with incomplete data for synteny verification [mallard (Anas platyrhynchos), swan goose (Anser cygnoides), several other passerines, other cormorants, the red-legged seriema (Cariama cristata), the ruff (Calidris pugnax), the gray crowned crane (Balearica regulorum), and the Oriental stork (Ciconia boyciana)]. As EFNB3 was not present in the current Anna’s hummingbird (Calypyte anna) PacBio assembly (the only such case for a non-galliform avian PacBio assembly), we also BLAST-searched the p-reads for this assembly and obtained evidence for an ortholog with similar synteny as in other bird groups. We also conducted BLAST searches for EFNB3 in the current assemblies of chicken (Gallus gallus; PacBio-based GRCg6a) and Japanese quail (Coturnix japonica; hybrid C. japonica 2.1), as well as in the corresponding PacBio p-reads and previous assemblies. Similar searches were also conducted in the Illumina-based assemblies and WGS databases of other galliforms, namely turkey (Meleagris gallopavo), golden pheasant (Chrysolophus pictus), helmeted guineafowl (Numida meleagris), and Gunnison sage grouse (Centrocercus minimus). These searches were extensive and permissive (>30 hit scores), followed by manual verification of detected hits. We also searched the BBSRC ChickEST Database and several brain transcriptome databases at E10, E17, P0, and 10-week-old chick available in NCBI (ERR2576391, ERR2576392, ERR2576465, ERR2576464, ERR2576488, ERR2576584, ERR2576585, and ERS2480079) but found no evidence of EFNB3.

The predicted ephrin-B3 proteins that were used for the protein alignment (fig. S2) were derived from the following: human (Homo sapiens), accession: NM_001406.4; mouse (Mus musculus), NM_007911.5; Egyptian fruit bat (Rousettus aegyptiacus), XM_016154770; wallaby (Macropus eugenii), from two EST clones FY22448 and FY578350; Chinese alligator, XM_02519434; Chinese softshell turtle (Pelodiscus sinensis), XM-025188382; Iberian ribbed newt (Pleurodeles waltl), Trinity_DN303912 (57); African clawed frog (Xenopus laevis), XM_018239717; Tibetan frog (Nanorana parkeri), XM_018574039; zebra finch (T. guttata) (fig. S1); small tree finch (C. parvulus), XM_030970743; Swainson’s thrush (C. ustulatus), XM_033084252; starling (S. vulgaris), XM_014893375; kakapo (S. habroptilus), XM_030474552.


Fertilized White Leghorn chicken eggs (Gil-Guy Farm, Israel) were incubated under standard conditions at 38°C. Zebra finch hatchlings were obtained from A. Barnea (Department of Natural and Life Sciences, The Open University of Israel, Ra’anana 43710, Israel; School of Zoology, Tel-Aviv University). All experiments involved with animals were conducted in accordance with the designated Experiments in Animals Ethic Committee policies and under its approval. Experiments were performed on adult male and female mice. EphA4lox mice (58) were crossed to PGK-Cre mice (59) to generate full knockout of the EphA4 gene. Animals were kept and used in accordance with regulations from the government of Upper Bavaria.


For the binding assay in COS cells, the mouse ephrin-B3 was cloned into pSecTag-B plasmid (Thermo Fisher Scientific, Waltham, USA). Six repeats of the myc epitope were cloned at the carboxyl end of ephrin-B3. For cell surface detection of the mouse ephrin-B3, the κ chain signal sequence and three repeats of the myc epitope were cloned upstream to amino acid 30 of the mouse gene. The sequence of the full-length zebra finch ephrin-B3 transcript was reconstructed from RNA-seq reads (fig. S1, B and C) using the Trinity platform (60). The sequence matched the exonic sequence of the EFNB3 transcript prediction (XM_030261704.1). For cloning the zebra finch ephrin-B3, an optimized sequence (available upon request) of residues 21 to 171 (fig. S2), predicted to encode the extracellular domain, was synthesized (IDT, Coralville, USA) and cloned downstream to an immunoglobulin (Ig) κ chain signal sequence in the pSecTag-B plasmid. The transmembrane and intracellular portions of the zebra finch transcript were obtained by fully sequencing the EST DV945855 complementary DNA (cDNA) clone. The synthetic 5′ portion of the gene and the 3′ DV945855–derived sequence were fused using a polymerase chain reaction (PCR)–generated Eco RV site at the synthetic 5′ part and an internal Sma I site in the 3′ DV945855–derived sequence. For detection of the in vitro transcribed protein, four repeats of the myc epitope were inserted between the Ig κ chain signal sequence and predicted residue 21 of zebra finch ephrin-B3. The EphA4-AP fused gene was generated by cloning amino acids 1 to 543 of the chick EphA4 (61) into the pSecTag-B plasmid. The SEAP gene was cloned downstream to EphA4.

Cell culture and transfection

COS-7 cells (American Type Culture Collection) were cultured in Dulbecco’s modified Eagle’s media (Sigma-Aldrich, Saint Louis, USA) supplemented with 10% fetal bovine serum, penicillin (100 U/ml), streptomycin (100 mg/ml), and 2 mM glutamine (Biological Industries, Beit HaEmek, Israel). COS-7 cells were transiently transfected with appropriate cDNA using polyethylenimine (linear, MW 25000; Polysciences Inc.).

AP fusion protein-binding experiments

To produce secreted EphA4-AP (alkaline phosphatase; placental) fusion protein, COS-7 cells were transiently transfected with appropriate cDNA. Posttransfection cells were supplemented with Opti-MEM (Gibco), and media was collected after 48 hours. The media was concentrated through a 30K Ultracel filter (Merck Millipore, Tullagreen, Ireland). The amount of the AP fusion protein was quantified by SDS–polyacrylamide gel electrophoresis analysis and Coomassie staining. The amount of EphA4-AP was calibrated by comparison to known amounts of EphA4-Fc fusion protein (R&D, Indianapolis, USA). AP activity was measured using para-nitrophenylphosphate substrate (pNPP; Sigma-Aldrich).

Binding assays were carried in COS-7 cells, transiently transfected with the mouse or zebra finch ephrin-B3. Following overnight (O/N) incubation, cells were incubated with varying amount of EphA4-AP in binding buffer [Hanks’ buffer saline supplemented with 20 mM Hepes (pH 7.3), 0.05% bovine serum albumin, 5 mM CaCl2, 1 mM MgCl2, and heparin sodium salt (10 μg/ml; Sigma-Aldrich)] at room temperature (RT) for 70 min. For visualization of AP staining, cells were plated on a glass coverslip coated with poly-lysine (0.005%; Sigma-Aldrich), and following AP-ligand binding as described above, cells were washed with binding buffer, fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) for 15 min at RT, and then washed with hepes buffered saline (HBS) solution [20 mM Hepes (pH 7.3) and 150 mM NaCl]. Endogenous AP activity was inactivated by incubation at 65°C for 100 min. For color reaction, cells were incubated with bromochloroindolyl phosphate–nitro blue tetrazolium substrates (Roche, Basel, Switzerland) in AP buffer [100 mM tris (pH 9.5), 100 mM NaCl, and 5 mM MgCl2] O/N. After washing in HBS and fixation for 15 min in 4% PFA and an additional washing in HBS, coverslips were mounted under mounting medium (Thermo Fisher Scientific, USA).

For colorimetric assays, following AP-ligand binding and washing, cells were incubated O/N at 4°C in lysis buffer [10 mM tris (pH 8.0) with 1% Triton X-100]. The supernatant was subjected to heat inactivation for 15 min at 65°C. Lysates were transferred to 96-well high-protein absorbance plate (Thermo Fisher Scientific, Nunc International, USA), and the level of AP activity was determined following the addition of equal volume of pNPP substrate (2 mg/ml) in reaction buffer [200 mM tris (pH 9.5), 10 mM MgCl2, and 200 mM NaCl]. Plates were incubated at RT for 30 min to 1 hour, and optical density was determined at 405 nM using an enzyme-linked immunosorbent assay plate reader.

Ephrin-B3 surface staining

COS-7 cells were transiently transfected with the mouse or zebra finch myc-EphrinB3 expression vectors. Twenty-four hours after transfection, cells were incubated with mouse anti-myc (9E10) antibody in PBS, washed with PBS, subsequently incubated with Rhodamine Red X–labeled anti-mouse antibody, and washed again. Cells were fixed in 4% PFA/0.1 M phosphate buffer and mounted before imaging.

In ovo electroporation

A DNA solution of 5 mg/ml was injected into the lumen of the neural tube at hamburger hamilton stages 17 to 18 (E2.75 to E3). Electroporation was performed using 3 × 50 ms pulses at 25 to 30 V, applied across the embryo using a 0.5-mm tungsten wire and a BTX electroporator (ECM 830). Following electroporation, 150 to 300 μl of antibiotic solution containing penicillin (100 U/ml) in Hanks’ balanced salt solution (Biological Industries, Beit HaEmek) were added on top of the embryos. Embryos were incubated for 3 to 19 days before further treatment or analysis. RP-specific expression was achieved using the Wnt1 or DNAH2 enhancers, coupled with a dorsal electrode positioning.

Immunohistochemistry and in situ hybridization

Embryos were fixed O/N at 4°C in 4% PFA/0.1 M phosphate buffer, washed twice with PBS, incubated in 30% sucrose/PBS for 24 hours, and embedded in optimal cutting temperature compound (Scigen, Gardena, USA). Cryostat sections (20 μm) were collected on Superfrost Plus slides and kept at −20°C. For thicker sections, spinal cords were isolated from the fixed embryos and subsequently embedded in warm 5% agar (in PBS), and 100-μm sections (E12 to E17) were cut with a vibratome. Sections were collected in wells (free-floating technique) and processed for immunolabeling. The following primary antibodies were used: rabbit polyclonal anti-GFP (1:1000; Molecular Probes, Eugene, OR, USA), mouse anti-GFP (1:100), goat anti-GFP (1:300; Abcam, Cambridge, UK), rabbit anti–red fluorescent protein (1:1000; Acris, Hiddenhausen, Germany), goat anti-ChAT (choline acetyltransferase) (1:300; Chemicon, Temecula, CA, USA), mouse anti-TAG1 4-5 (DSHB, University of Iowa), mouse anti-neurofilament (1:100; 3A10), mouse anti-myc (1:10; 9E10), and rabbit anti-myc (1:100; Abcam). The following secondary antibodies were used: Alexa Fluor 488/647 AffiniPure donkey anti-mouse, anti-rabbit, and anti-goat (Jackson) and Rhodamine Red X donkey anti-mouse and anti-rabbit (Jackson). Images were taken under a microscope (Eclipse Ni; Nikon) with a digital camera (Zyla sCMOS; Andor) or a confocal microscope (FV1000; Olympus).

In situ hybridization was performed as described (62). The following probes were used: chicken ephrin-B1, ephrin-B2, EphA4, and Vglut2; zebra finch ephrin-B3; and mouse ephrin-B3. Probes were PCR-amplified from brain cDNAs of E6 chicken embryo, P4 zebra finch, and E15 mouse, respectively, using the following primers in Table 1. PCR products were verified by sequencing before use.

Table 1 List of primers that were used for generating the RNA probes.
View this table:

PRV injection

We used two isogenic recombinants of an attenuated PRV strain (PRV Bartha) that express enhanced GFP (PRV152) and monomeric red fluorescent protein (PRV614). The viruses were harvested from Vero cell cultures at titers 4 × 108, 7 × 108, and 1 × 109 plaque-forming units/ml, respectively. Viral stocks were stored at −80°C. For pre-MN infection, injections of 3 μl of PRV152 or PRV614 were made into the thigh or wing musculature of E13 or E14 chicken embryos, using Hamilton syringes (Hamilton; Reno, NV, USA) equipped with a 33-gauge needle. The embryos were incubated for 36 to 40 hours and euthanized for analysis. For bilateral pre-MN infection, PRV152 and PRV614 were each injected into the pectoralis musculature on opposite sides of the embryo midline.

Cell distribution and density plot construction

The codes for the cell distribution and density plot imaging and analysis were written in MATLAB. The distribution plots were constructed by calculating the relative midline expression, after background subtraction and a subsequent length normalization between all analyzed sections. The density plots were generated on the basis of cross-sectional images transformed to a standard form. The background was subtracted, and the cells were filtered automatically on the basis of their soma area or using a manual approach. Subsequently, the neurons were depicted by semitransparent circles, or their distribution was visualized by two-dimensional kernel density estimation, using the MATLAB function “kde2d.” Unless indicated otherwise, a contour plot was drawn for density values between 20 and 80% of the estimated density range, in six contour lines. An additional script was used for pre-MN classification based on their neurotransmitter identity and laminar location.

A similar density visualization procedure was done for the EphA4 density plots (Fig. 3, E’ and F’) but was preceded by signal binarization, and then an averaging filter (5 cycles) was used to achieve a smoother density signal. The intensity of EphA4 signal as a function of distance from the midline was obtained by summation of the binary signal over the horizontal axis, followed by a rescaling step for each image. The difference at the midline between the intensity profiles of E10 and E16 data was evaluated by t test.


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We dedicate this paper to the memory of H. Falk, our dear collaborator and friend, who passed away on 22 November 2019 while this research was in progress. We thank T. Lederman for technical assistance; J. Gros for providing the PacBio p-reads of the quail genome; W. Warren for providing the PacBio p-reads of the chicken genome; G. Friedlander, M. Yassour, and Y. Drier for assistance in the analysis of the expression and regulation data; and A. Barnea for providing zebra finch hatchlings. Funding: This work was supported by a grant to A.K. and C.V.M. from the U.S.-Israel Binational Science Foundation (grant no. 2017/172), an NSF graduate fellowship to S.F., and grants to A.K. from the Israel Science Foundation (grant no. 1400/16) and the Avraham and Ida Baruch endowment fund. Author contributions: A.K., C.V.M., and B.H.: Research design and data analysis. B.H.: Circuit analysis and in situ hybridization. O.M.: Circuit analysis. R.S.-M.: Enhancer analysis. G.E.: Protein studies. S.F. and P.V.L.: Genomics and gene structure analysis. S.P. and R.K.: Generated conditional EphA4 knockout mouse. A.K. and C.V.M.: Wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article