The geometry of limb motor innervation is controlled by the dorsal-ventral compartment boundary in the chick limbless mutant

Abstract

Precise control of limb muscles, and ultimately of limb movement, requires accurate motor innervation. Motor innervation of the vertebrate limb is established by sequential selection of trajectories at successive decision points. Motor axons of the lateral motor column (LMC) segregate at the base of the limb into two groups that execute a choice between dorsal and ventral tissue: medial LMC axons innervate the ventral limb, whereas lateral LMC axons innervate the dorsal limb. We investigated how LMC axons are targeted to the limb using the chick mutant limbless (ll), which has a dorsal transformation of the ventral limb mesenchyme. In ll the spatial pattern of motor projections to the limb is abnormal while their targeting is normal. While extensive, the dorsal transformation of the ll ventral limb mesenchyme is incomplete whereas the generation, specification and targeting of spinal motor neurons are apparently unaffected. Thus, the dorsal-ventral motor axon segregation is an active choice that is independent of the ratio between dorsal and ventral tissue but dependent on the presence of both tissues. Therefore, the fidelity of the motor projections to the limb depends on the presence of both dorsal and ventral compartments, while the geometry of motor projections is controlled by the position of limb dorsal-ventral compartment boundary.

INTRODUCTION

Neural circuits that enable precise motor control depend on accurate connections between motor neurons and their corresponding muscles. The connectivity pattern is represented in a topographic map with motor neurons, whose somas are organized hierarchically in nested spatial domains in the spinal cord, projecting axons to a corresponding hierarchy of muscles (Jessell, 2000; Landmesser, 2001). This map is established during embryonic development as somatic motor axons grow along shared trajectories that are subdivided at intermediate positions into progressively finer branches that reflect subdivisions of both the motor neuron population and their targets (Jessell, 2000). Understanding how the neuromuscular map is constructed thus requires understanding how intermediate trajectory decisions are controlled during embryonic development.

The motor neurons that innervate the limb musculature are located at limb axial levels in the lateral motor column (LMC) of the spinal cord. The LMC is separated into medial and lateral divisions that are anatomically and molecularly distinct (Hollyday and Jacobson, 1990; Lance-Jones and Landmesser, 1980; Landmesser, 1978). Axons of LMC neurons project ventrolaterally from the spinal cord into a plexus near the base of the limb, and sort into two nerve branches (Tang et al., 1992; Tosney and Landmesser, 1985b; Wang and Scott, 2000; Whitelaw and Hollyday, 1983). Lateral LMC axons form a nerve branch that chooses a trajectory into dorsal limb mesenchyme, while medial LMC axons form a nerve branch choosing a trajectory into ventral limb mesenchyme (Bennett et al., 1983; Lance-Jones and Landmesser, 1981; Tsuchida et al., 1994). The fidelity of the LMC motor axon choice between dorsal limb tissue (where extensor muscles are) and ventral limb tissue (where flexor muscles are) is essential to the control of limb movement in the adult, which relies on the alternate activation of extensor and flexor muscles (Akay et al., 2006; Kao et al., 2015; Monaco et al., 2010).

An important but incompletely understood aspect of motor axon projection patterns pertains to their geometry, and particularly to how the angle between the two nerve branches at limb entry is controlled. Strikingly, in mutants in which axon guidance molecules or limb mesenchyme patterning are affected, the fidelity of the dorsal/ventral choice made by LMC axons is perturbed, while the geometry of motor nerve branches, defined as the relative spatial arrangement of branches, is unchanged (Eberhart et al., 2002; Huber et al., 2005; Kania et al., 2000; Kao et al., 2009; Kramer et al., 2006; Luria et al., 2008; Luria and Laufer, 2007; Wang and Scott, 2000). This suggests that nerve branch positioning is established independently of the factors controlling branch choice, with distinct effectors influencing branch geometry. One candidate effector is the cartilage condensation in the center of the limb, which precludes axon growth and thus might constrain the available projection paths, creating a forced decision point between limited trajectories (Oakley and Tosney, 1991). Another possibility is that guidance cues common to dorsal and ventral territories, such as Slit proteins (Vargesson et al., 2001), define permissive axonal “highways” (Tosney and Landmesser, 1985a). Alternatively, factors located at the boundary between dorsal and ventral limb compartments (Arques et al., 2007) might influence nerve geometry. Identifying conditions with altered branch geometry but normal LMC axon targeting might help distinguish between these options.

Genetic and surgical experiments have revealed relatively simple rules governed by neuronal and target tissue effectors that control how LMC axons choose trajectories. Surgical experiments showed LMC axons are actively guided by local mesenchymal cues at the base of the limb (Ferguson, 1983; Ferns and Hollyday, 1993; Lance-Jones and Landmesser, 1981; Tosney and Landmesser, 1984; Whitelaw and Hollyday, 1983). Genetic experiments showed the identity of medial and lateral LMC neurons and of their target dorsal or ventral limb mesenchyme is specified by LIM homeodomain transcription factors (Kania et al., 2000; Tsuchida et al., 1994). These control the differential expression in peripheral targets and motor neurons of axon guidance ligands and their receptors. Members of the Eph and ephrin superfamily are the predominant guidance receptors and ligands (Dudanova et al., 2012; Gatto et al., 2014; Kania and Jessell, 2003; Kania and Klein, 2016; Luria et al., 2008), while semaphorins and cadherins are also implicated (Chai et al., 2014; Huber et al., 2005). Most pairs of surface-bound Eph receptors and ephrin ligands function as axon guidance cues that are generally repellant (Kania and Jessell, 2003; Luria et al., 2008) and occasionally attractive (Dudanova et al., 2012; Kao and Kania, 2011; Marquardt et al., 2005). These work in conjunction with the ligand GDNF and its cognate tyrosine kinase receptor (Kramer et al., 2006) and act synergistically with other molecules such as semaphorins (Morales and Kania, 2017; Poliak et al., 2015) to guide the axonal projections into the appropriate limb nerve branch.

Both surgical interventions and targeted mutations can alter the dorsoventral character of limb mesenchyme and thereby the spatial organization and structure of its derivative tissues, as well as the myoblast-derived musculature. These include chick embryos with surgically-generated bidorsal limbs, in which host ventral limb mesenchyme is replaced with donor dorsal mesenchymal tissue (Lance-Jones, 1986), the mouse Lmx1b^−/− mutant that develops with biventral limbs (Kania et al., 2000), and the conditional mouse Brn4^cre; BmprIa^flox/- mutant that develops with bidorsal limbs (Luria and Laufer, 2007). In each of these cases the motor nerves do not significantly change their geometry, even though in the two mouse mutants the dorsal-ventral projection decision executed by the LMC motor axons becomes randomized.

The chick limbless mutant might provide further insight into how the geometry of motor axon branches is controlled in relation to targeting accuracy. The ll mutation is autosomal recessive, fully penetrant, lethal perinatally and of unknown molecular nature (Lanser and Fallon, 1984; Prahlad et al., 1979). While it has been suggested that ll is encoded by the Sp8 gene, which encodes a transcription factor (Haro et al., 2014), the specific genetic lesion remains undefined. In ll mutants limb outgrowth is initiated but limbs quickly atrophy (Fallon et al., 1983; Lanser et al., 1986) because the apical ectodermal ridge (AER), which is required for limb outgrowth, fails to form (Carrington and Fallon, 1988). Motor neurons in ll mutant embryos are generated on schedule, extend axons to the periphery and subsequently die in the absence of target tissue (Lanser et al., 1986; Lanser and Fallon, 1984). The AER promotes proximal-distal limb outgrowth via the production of fibroblast growth factors (FGF) (Niswander and Martin, 1992; Savage et al., 1993) and can be functionally replaced by recombinant FGF protein (Fallon et al., 1994; Laufer et al., 1994; Niswander et al., 1993), which rescues proximal-distal outgrowth. In ll mutant embryos, limb outgrowth can be rescued with recombinant FGF protein application, and resultant limbs develop distally well beyond the point of motor axon entry (Grieshammer et al., 1996; Ros et al., 1996). Surprisingly, the ventral limb mesenchyme of these FGF-rescued limbs has an extensive dorsal morphological transformation (Grieshammer et al., 1996; Ros et al., 1996). Molecular analysis using the dorsal limb mesenchyme marker Lmx1b revealed this transformation begins prior to limb rescue and extends rostrocaudally into inter-limb flank tissue (Grieshammer et al., 1996; Ros et al., 1996). However, the precise extent of the dorsal transformation in proximal, ventral mesenchyme is unclear. Furthermore, neither the LMC projection patterns, the fidelity of dorsal-ventral trajectory selection choice of LMC axons, nor the geometry of the limb nerves were analyzed in FGF-rescued ll limbs.

Here we investigated the relationships between motor innervation targeting and geometry and dorsal-ventral limb identity in ll, by determining the fidelity and spatial patterns of motor projections in FGF-rescued ll embryonic limbs. ll limbs have an extensive but proximally incomplete dorsal transformation. Limb nerves reach the forelimb and split into a dorsal branch that projects into the dorsal limb territory and a ventral branch that projects into the small ventral proximal limb remnant. Ventral branch ll medial LMC axons track the displaced border between dorsally transformed mesenchyme and the small ventral limb remnant, but never enter the dorsally transformed tissue in the ventral limb. Thus, the geometry of the limb nerves is disrupted, and the angle between dorsal and ventral branches is much larger in ll than in wild type embryos. Nonetheless the LMC axons make a normal dorsal-ventral projection choice, with medial LMC axons entering the ventral nerve branch, and lateral LMC axons entering the dorsal branch. At thoracic levels, thoracic motor axons can innervate FGF-induced supernumerary limbs, but the geometry of these ectopic limb nerves is similarly disrupted. Our results show that the geometry of the limb motor nerves is controlled by the position of the border between dorsal and ventral limb compartments. If this border moves, so do the nerves, changing their relative positions and the overall geometry of limb nerves. Thus, we demonstrate that the fidelity of the motor projections to the limb depends on the presence of both dorsal and ventral compartments, while the geometry of motor projections is controlled by the position of limb dorsal-ventral compartment boundary.

EXPERIMENTAL PROCEDURES

AER removal and rescue with FGF-soaked beads in chick embryos

Wild type fertile SPF White Leghorn chick eggs were used (Charles River Laboratories, MA). Fertile ll eggs were obtained from intercrosses between heterozygous ll/+ parents (from a flock maintained by Dr. Jackie Pisenti, University of California, Irvine).

Surgery on wild type and ll embryos

Surgery was performed as described (Grieshammer et al., 1996; Ros et al., 1996). FGF-soaked beads were implanted at stage HH17–18 (Hamburger and Hamilton, 1951) at forelimb or thoracic level in the lateral plate in the posterior forelimb region or in the mid-thoracic region, as described (Lance-Jones and Landmesser, 1981; Ros et al., 1996; Turney et al., 2003). Exogenous FGF sustains significant ll limb outgrowth (Grieshammer et al., 1996; Ros et al., 1996). Embryos were harvested at various times through HH34 for further analyses. One to three FGF-soaked beads were implanted where the AER is expected to form in wild type embryos. ll mutant embryos were identified by the absence of limbs contralaterally, where beads were not implanted (Grieshammer et al., 1996; Ros et al., 1996). Results from wild type and heterozygous ll/+ embryos were pooled. FGF bead implants generated 75 mutant embryos for analysis (Table of surgeries).

Table of ll surgeries in the limb fields.

	Embryos harvested	Embryos harvested at HH24	Embryos harvested at HH27–29	Total
Surviving embryos / operated embryos	Total#	94 / 132 (71 %)	243 / 459 (53 %)	337 / 591 (57 %)
	WT##	60 / 83 (72 %)	178 / 329 (54 %)	238 / 412 (58 %)
	ll ###	34 / 49 (69 %)	65 / 130 (50 %)	99 / 179 (55 %)
	Total*	25 / 34 (74 %)	50 / 65 (77 %)	75 / 99 (76 %)
ll with limb field outgrowths	FL	21	46	67
	Thoracic	4	4	8

Surgeries indicate the number of embryos operated.

Percentages indicate:

^#,embryos surviving relative to the total number of operated embryos;

^##,wild type embryos surviving relative to the total number of operated wild type embryos;

^###,ll embryos surviving relative to the total number of operated ll embryos;

^*,ll embryos with outgrowths relative to the total number of operated ll embryos.

The ll mutant frequency (36 %) was slightly higher than expected (25 %) because in some cases it was possible to identify the nascent AER and discard excess control embryos.

Bone/cartilage identification

Since the proximal limb is necessary for a correct DV axon choice, we confirmed the presence of proximal skeletal elements in mutants. Alcian green was used to stain skeletal elements at E8-E10 as described (Firulli et al., 2005).

Retrograde nerve labeling of limb and flank nerves in mouse and chick embryos

Retrograde nerve labeling with horseradish peroxidase (HRP) injected into dorsal or ventral limb of cultured explanted HH28–30 embryos was performed as described (Kania et al., 2000; Lance-Jones and Landmesser, 1980). After 4–6 hr the tissue was fixed and cryosectioned transversely. Control sections were examined by HRP and neurofilament (NF) co-immunostaining and evaluated at the dorsoventral branch point for targeting accuracy. Only embryos with a single labeled nerve branch were analyzed further.

Immunofluorescence

Antibody stains on 12–25 μm transverse cryosections were performed as described (Tsuchida et al., 1994). Cell nuclei were counterstained with Hoechst 33258 (2 μg/ml).

Primary antibodies, dilutions and sources: rabbit anti-EphA4 (sc-921 s-20, Santa Cruz Biotechnology, Santa Cruz, CA) 1:1000; mouse anti-Pax6 1:10, mouse anti-Pax7 1:5, mouse antirat Nkx6.1 1:4000, guinea pig anti mouse Nkx6.1 1:8000, rabbit anti-Nkx2.2 1:4000, mouse antiNkx2.2 (75–5A5) 1:50, guinea pig anti-chick Olig2 1:5000, rabbit anti-Hb9 1:10000, rabbit anti-Irx3 1:3000, mouse anti-chick MNR2/HB9 (81–5C10) 1:50, rabbit anti-Lim1/2 1:4000, rabbit anti-Lhx2/9 (LH2A/B) 1:8000, guinea pig anti-Lhx2/9 1:8000, rabbit anti-Lbx1 1:1000, rabbit antiLim3 1:4000, guinea pig anti-Lim3 1:4000, mouse anti-Lim3 (61–8C10) 1:50, rabbit anti-Chx10 1:4000, rabbit anti-Isl1/2 (K4) 1:2500, guinea pig anti-Isl1/2 1:16000, mouse anti-Isl1/2 (4D5) 1:100, rat anti-Math3 1:2000, rabbit anti-mouse Dbx1 1:16000, guinea pig anti-En1 1:8000, guinea pig anti-Evx1/2 1:8000, mouse anti-NF related protein (3A10) 1:100, rabbit anti-chick Lmx1 (E1) 1:2000 (Briscoe et al., 2000; Dasen et al., 2005; Kania et al., 2000; Muller et al., 2002; Tsuchida et al., 1994); goat anti HRP 1:1000 (Jackson ImmunoResearch Labs, West Grove, PA); mouse anti-human Plzf (2A9) 1:1,000 (Barna et al., 2000; Barna et al., 2005); rabbit anti-ephrin-A5 (1.6 μg/ml) (Eberhart et al., 2002); mouse anti-NCAM (5e) 1:1000 and mouse anti-PSA-NCAM IgM (5A5) (Tang et al., 1992).

Secondary antibodies (raised in donkey) coupled to fluorophores Alexa-488 (Molecular Probes, OR) and Cy3 were used at 1:1000, those coupled to Cy5 used at 1:500 (Cy3, Cy5 - coupled antibodies were from Jackson ImmunoResearch Labs, West Grove, PA).

RNA in situ hybridization

Non-radioactive RNA in situ hybridization on cryosectioned mouse tissue was performed as described (Laufer et al., 1997). Probes used: Plzf (Cook et al., 1995); Tbx4 and Tbx5 (Logan et al., 1998); Lmx1 (Vogel et al., 1995); Meis1 and Meis2 (Mercader et al., 1999); Sim1 (Pourquie et al., 1996); ephrin-A2, ephrin-A5 and EphA4 (Araujo et al., 1998; Iwamasa et al., 1999).

Imaging and morphometric analysis

Expression patterns were compared using multichannel immunofluorescence imaging or overlaying of pseudo-colored RNA in situ hybridization images from adjacent cryosections (Kania and Jessell, 2003). Measurements of limb areas, nerve and limb angles, fluorescence area intensity were conducted on pictures using the ImageJ program as described (Hanson and Landmesser, 2004; Kania et al., 2000).

Quantification of motor neuron populations and retrogradely labeled neurons

Motor neurons were counted in every fifth section, spanning 4–5 spinal cord segments from the beginning of the LMC (marked by the appearance of Lim1+ lateral LMC motor neurons). Neurons retrogradely labeled by HRP and co-labeled with LMC markers were counted (Lin et al., 1998). Four or more sections from three or more embryos encompassing at least 260 HRP+ cells were counted for each condition.

Statistics

Statistical comparisons of neuronal counts were performed using a 2-tailed heteroscedastic Student’s t-test, with a significance limit of P < 0.05. Statistical comparisons of limb areas and nerve and limb angles were carried out by performing an analysis of variance followed by 2-tailed t-tests. Values are reported as mean ± SEM (standard error of the mean).

RESULTS

To address how limb motor innervation is disrupted when the dorsal-ventral (DV) polarity of the limb is perturbed in the chick, we analyzed limbless embryos with FGF-rescued limbs, which are known to have a dorsal transformation of ventral limb mesenchyme (Grieshammer et al., 1996; Ros et al., 1996). We examined the motor innervation pattern, limb morphology and DV polarity of FGF-rescued limbless forelimbs at HH27, after motor neuron cell bodies complete their migration and motor axons enter the limb (Sockanathan and Jessell, 1998), but before the onset of extensive secondary branching and target-dependent motor neuron cell death (Chu-Wang and Oppenheim, 1978; Ferns and Hollyday, 1995; Lanser et al., 1986; Wang and Scott, 2000). FGF-rescued mutant limbs are similar in size to wild type limbs and have at least the proximal limb segment, since a humerus is present on the rescued side if allowed to develop to HH 35 (Supplementary Fig. 1 A–B). This proportion of limb tissue is compatible with sustaining motor axon innervation and eliciting a DV motor axon choice (Ferguson, 1983; Ferns and Hollyday, 1993; Lance-Jones, 1986; Tosney and Landmesser, 1984; Whitelaw and Hollyday, 1983).

Nerve branching patterns are abnormal in FGF-rescued limbless forelimbs

We first analyzed the patterns of nerve growth in FGF-rescued ll mutant limbs. We compared the morphology of limb nerves with the expression of the dorsal limb determinant Lmx1 (Chen et al., 1998; Riddle et al., 1995; Vogel et al., 1995). In wild type embryos, the spinal nerve splits at the base of the limb into a dorsal branch, which enters Lmx1+ territory, and a ventral branch, which enters Lmx1– territory (Fig. 1A) (Hollyday, 1995; Kania et al., 2000). In mutants, the spinal nerve also splits, such that the dorsal branch invades a ventrally expanded Lmx1+ limb territory, and the ventral branch straddles the border between Lmx1+ and Lmx1– tissue (Fig. 1A, n=33/35 embryos examined). In 2 of 35 ll embryos, the spinal nerve defasciculated and axons invaded the limb in a disorganized, fan-like, structure (Supplementary Fig. 2A). In most embryos (33/35), angles between the nerve branches differ between mutants and controls (Fig. 1B, D). While the exit angle from the spinal cord is similar in controls and mutants, the angle between the dorsal nerve branch and the spinal nerve is smaller in ll (102° ± 2°, mean ± standard error of the mean (SEM)) than in controls (125° ± 3°, P<0.001, 2-tailed heteroscedastic t-test). Moreover, the angle between the dorsal and the ventral branches is significantly wider in ll (98° ± 4°) than in controls (63° ± 2°, P<0.001).

Figure 1. The limb nerve bifurcates at the base of ll limbs.

(A) The limb nerve bifurcates at the base of the limb into a dorsal and a ventral nerve branch. In transverse sections of HH27 embryos, axons visualized with anti-neurofilament (NF) antibody have invaded the limb in both wild type and mutant. (i-ii) In wild type the dorsal half of the limb expresses the dorsal marker Lmx1 that stains cell nuclei (red). The dorsal nerve branch invades the Lmx1+ territory and the ventral branch invades the Lmx1– territory. (iii-iv) In ll, most limb territory is Lmx1+. The dorsal nerve branch invades the Lmx1+ territory while the ventral branch tracks the border between Lmx1+ and Lmx1– territory.

Panels are at increasing magnification: 4x (i, iii); 10x (ii, iv).

(B) Diagrams of nerve angles. Angles were measured at HH27 in 7 wild type (i) and 16 ll embryos (ii). The angles measured are: A, between the ventral root of the limb nerve and the spinal cord (wild type 135 ± 2°, ll 138 ± 2°, mean ± SEM; P=0.47, 2-tailed t-test); B, between the limb nerve and the dorsal branch (wild type 126 ± 2°, ll 102 ± 2°, P<<0.001); G, between the dorsal and ventral nerve branches (wild type 63 ± 2°, ll 98 ± 3°, P<<0.001). All figures are mean values ± SEM (standard error of the mean). The B and G angle values in red in ll are significantly different from the corresponding ones in wild type (P<0.001).

(C) The angle D between the sagittal body axis (running from back to belly) and the main limb axis (the line running through the most distal extremity of the limb and through the DV motor axon choice point) is smaller in FGF-rescued ll limbs than in wild type (wild type, 121 ± 3°, n=11; ll 96 ± 4°, n=14, P<0.001).

(D) Angle measurements shown as box and whisker plots. Boxes indicate the 25%−75% interquartile range, the median is drawn as a horizontal line, and the whiskers indicate the 5^th and 95^th percentile. Each dot represents the average angle from an individual embryo.

* Indicates the cross-reactivity of the E1 anti-Lmx1 antibody with muscle fibers.

D: dorsal nerve branch. V: ventral nerve branch.

Dorsal to the top, medial to the left.

The spinal cord is dorsal and medial (A).

We noted that FGF-rescued ll limbs protrude from the body at a different orientation than in wild type. This prompted us to measure the angle between the sagittal body axis and the main limb axis. Wild type limbs protrude at an obtuse angle from the body (121° ± 3°), while FGFrescued ll limbs protrude almost perpendicularly (96° ± 4°, P<0.001) (Fig. 1C, D). Nerve branch angles and limb orientation angle are thus changed in opposite directions in ll. This results in altered relative positions of nerves and limb, such that the ventral nerve branch runs almost parallel to the base of the mutant limb, and the angle between nerve branches is wider in mutants than in controls.

The DV motor axon projection choice is normal in FGF-rescued ll forelimbs

To address whether the abnormal spatial arrangement of nerves in ll is accompanied by errors in DV motor axon targeting, we investigated whether the DV choice is correctly executed. We determined the origin of axons in the spinal cord in each limb nerve branch using horseradish peroxidase (HRP) retrograde labeling of limb nerves (Landmesser, 1978) at HH 28–29. Following nerve branch labeling, HRP-labeled neurons are identified by immunostaining for LMC subcolumnar markers (Tsuchida et al., 1994). Lateral LMC neurons express Lim1 and not Isl1 (Lim1+ Isl1–); conversely, medial LMC neurons express Isl1 and not Lim1 (Lim1– Isl1+). Injecting HRP into the dorsal limb (Fig. 2A, C) labels about 94% lateral LMC neurons in both wild type and ll embryos (Fig. 2B, 2G; Table 1; P = 0.55). Injecting the ventral limb (Fig. 2D, F) labels about 92% medial LMC neurons in both wild type and ll embryos (Figs. 2E, H; Table 1; P =0.88). These data indicate that in both wild type and ll embryos the dorsal nerve branch is composed of lateral LMC axons while the ventral nerve branch is composed of medial LMC axons. Thus, in ll mutants the dorsal-ventral segregation of motor axons is normal despite the presence of an abnormal limb morphology and nerve branch pattern.

Figure 2. The dorsal-ventral choice of LMC motor axon is normal in ll.

Retrograde labeling of limb nerves at HH28 marks the spinal cord (SC) cells that innervate the limb.

(A) Diagram of the experiment in (B, C). Boxes correspond to images in (B, C).

(B, C) Dorsal nerves originate from lateral LMC in both wild type (n=4 embryos, 266 neurons counted) and ll (n=5 embryos, 289 neurons counted). (B) In both wild type (i-iii) and mutant (Aiv-vi) SC, most HRP+ cells are (Lim1+ Isl1–) lateral LMC cells. (C) In both wild type (i) and mutant (ii) limb, only the dorsal (D) nerve branch is labeled by the HRP injection.

(D) Diagram of the experiment in (D, E). Boxes correspond to images in (D, E).

(E, F) Ventral nerves originate from medial LMC in both wild type (n=5 embryos, 389 neurons counted) and ll (n=4 embryos, 393 neurons counted). (E) In both wild type (i-iii) and mutant (Aiv-vi) SC, most HRP+ cells are (Lim1– Isl1+) medial LMC cells. (F) In both wild type (i) and mutant (ii) limb, only the ventral V nerve branch is labeled by the HRP injection.

(G, H) Quantification of the data in (B, C, E, F). The percentage of labeled cells that are of lateral or medial LMC identity marked by dorsal (G) or ventral (H) nerve labeling does not differ significantly between wild type and mutant (Table 1).

The pink circle (A, D) represents the FGF-bead implanted in ll limbs.

Table 1.

Retrograde nerve labeling in ll.

	Dorsal labeling		Ventral labeling
	% LMCl HRP+	% LMCm HRP+	% LMCl HRP+	% LMCm HRP+
WT	93.5 ± 1	6.5 ± 1	7.5 ± 0.9	92.5 ±0.9
	n=4		n=5
ll	94.4 ± 1.5	5.6 ± 1.5	7.7 ± 2.2	92.1 ± 2.2
	n=5		n=4
P	0.55		0.88

Lateral LMC (LMCl) neurons were identified as Lim1+ Isl1– cells; medial LMC (LMCm) were identified as Lim1– Isl1+ cells.

Percentages indicate the number of HRP-labeled neurons belonging to the lateral LMC or the medial LMC, relative to the total number of motor neurons labeled. The values are reported as mean ± SEM (standard error of the mean). n = number of embryos labeled. P values are calculated using a 2-tailed t-test.

We next examined the distribution of PSA-NCAM, previously described as a dorsal branch marker (Hanson and Landmesser, 2004; Tang et al., 1992; Tang et al., 1994). In wild type embryos, we found most dorsal axons and some ventral axons express PSA-NCAM, while a discrete s subset of ventral axons lack PSA-NCAM staining (Supplementary Fig. 2B). These axons separate from the ventral nerve branch after reaching the DV choice point and innervate flank muscles (Haase et al., 2002; Luria and Laufer, 2007). Thus, we find PSA-NCAM marks nerve branches that innervate dorsal and ventral limb but not the flank, rather than just the dorsal nerve branch. In mutants, PSANCAM is present on both dorsal and ventral limb nerve branches and no flank nerve branch is present (Supplementary Figure Fig. 1C), raising the possibility that ventral and flank branches are fused.

To determine if FGF treatment alone altered LMC axon projection decisions, the AER of wild type limb buds was replaced with an FGF-soaked bead. In these embryos, both PSA-NCAM on axons and the nerve branching pattern were indistinguishable from wild type (Supplementary Fig. 2B). We also performed retrograde labeling of the dorsal or the ventral nerve branch and observed no targeting difference between FGF-rescued and unoperated control embryos (Supplementary Fig. 3). Therefore, FGF rescue of limb outgrowth has no detectable influence on either DV projection choice or spatial nerve patterns.

The ll dorsal limb transformation, while extensive, is incomplete

The observation that the DV projection decision is normal in ll limbs differs from that seen in the two mouse mutants with DV-altered limb mesenchyme (Kania et al., 2000; Luria and Laufer, 2007), which display a randomized DV choice. One possibility is that in ll limbs, the dorsal transformation of ventral limb mesenchyme is incomplete and that a ventral limb remnant influences the DV projection decision. To determine the degree of dorsal limb transformation in FGF-rescued ll forelimbs, we compared the expression of dorsal, ventral and general limb markers at the time of axon arrival and DV choice (HH23–25).

We previously defined a set of four molecular markers for evaluating the DV identity of the proximal limb mesenchyme in mouse at the time when the motor axons choose between dorsal and ventral compartments (Luria and Laufer, 2007). The transcription factor Plzf (Avantaggiato et al., 1995; Barna et al., 2000; Barna et al., 2005) marks the entire limb mesenchyme (Luria and Laufer, 2007). The transcription factor Lmx1b marks dorsal limb mesenchyme and controls dorsal limb identity and DV motor axon choice (Kania et al., 2000; Luria et al., 2008) while the receptor tyrosine kinase EphA4 is expressed in the proximal, dorsal limb and is positively controlled by Lmx1b (Dottori et al., 1998; Helmbacher et al., 2000; Iwamasa et al., 1999; Kania and Jessell). The Eph ligand ephrin-A5 is expressed in the ventral limb (Eberhart et al., 2004; Eberhart et al., 2000; Swartz et al., 2001) and is repressed dorsally by Lmx1b (Kania and Jessell, 2003). We showed that dorsal limb mesenchyme is defined as Plzf+ Lmx1+ EphA4+ ephrin-A5– and ventral limb mesenchyme is defined as Plzf+ Lmx1– EphA4– ephrin-A5+ (Luria and Laufer, 2007). Here we extended these findings in chicken at the time motor axons begin entering the limb mesenchyme (HH 23–25 in hindlimb, HH 22–24 in forelimb) (Tosney and Landmesser, 1984). First, we found that Plzf is expressed in the entire limb mesenchyme (Fig. 3A) and excluded transcription factor Tbx5 as a possible general limb marker since its expression extends beyond the limb in the ventral flank (Fig. 3B). Second, we found EphA4 marks the proximal dorsal limb (Fig. 3C) while ephrin-A5 marks the ventral limb (Fig. 3D). Thus, the same DV molecular code applies to chicken and mouse limbs, despite 318 million years of evolutionary divergence (Kumar et al., 2017).

Figure 3. FGF-rescued ll limbs have an extensive but incomplete dorsal transformation at the time of DV motor choice.

A) FGF-rescued ll forelimbs have altered DV polarity. In wild type HH24 limbs, Plzf expression is present in all limb territory (i, iv) and aligns proximally with that of Lmx1 (ii, iv), indicating Plzf is a general limb marker. Extended dorsal (Plzf+ Lmx1+) domain and reduced ventral (Plzf+ Lmx1–) domains are found in mutants (viii), indicating mutant forelimbs have an extensive but incomplete dorsal transformation.

(B) Tbx5 is expressed in both limb and flank tissues. At HH24, Tbx5 is expressed in the limb and its dorsal proximal border aligns with those of Lmx1 and Plzf. It also is expressed outside the limb in the flank.

(C) In wild type embryos (i-iii), EphA4 is a proximal dorsal limb marker and its proximal border coincides with that of Lmx1. In FGF-rescued ll embryos (iv-iv), the proximal border of EphA4 coincides with that of Lmx1 only dorsally, indicating that the proximal ventral tissue does not have dorsal character.

(D) In wild type embryos (i-iii) ephrin-A5 is a ventral limb marker and does not overlap with Lmx1. In ll embryos (iv-vi) most limb territory is dorsal (Plzf+ Lmx1+ ephrin-A5–) but a small proximal ventral (Plzf+ Lmx1– ephrin-A5+) limb territory persists.

Dorsal is to the top, medial to the left. The limb outline and the border between dorsal and ventral territories are highlighted with dotted white lines. The proximal limb border is indicated in A (iv, viii), D (iii, vi).

(E) Quantification of DV areas at the time of DV motor choice in wild type (white), FGF-rescued ll (gray). The proportion of dorsal limb tissue is larger in ll but there still is some ventral proximal limb tissue left (see also Table 2).

(F) A summary of the expression data. Plzf is a general limb marker. Half of the wild type limb and most of the ll limb are dorsal (Plzf+ Lmx1+). EphA4 and ephrinA5 are dorsal and ventral limb markers.

* Denotes P<0.001 in 2-tailed t-tests comparing wild type and ll.

We next determined the relative dorsal and ventral limb areas at HH25 in wild type and ll embryos using these markers (Fig. 3E, Table 2). In wild type embryos, about half (54%) of the limb is dorsal. In FGF-rescued ll limbs, the majority of the limb is Plzf+ Lmx1+ ephrin-A5–, indicating that the limb exhibits an extensive dorsal transformation (Figs. 3A, 3D). However, we also found that a small ventral proximal domain Plzf+ Lmx1– ephrin-A5+ remains in ll (Figs. 3A, 3D), while EphA4 expression in ll is confined to Lmx1+ territory and does not extend into the small ventral Plzf+ Lmx1– area (Fig. 3C). In FGF-rescued ll 80% of the limb territory is dorsal (Fig. 3E, Table 2; P << 0.01, wild type vs. ll). Interestingly, the dorsalization is even more extensive (89%) on the contralateral side not treated with FGF. Thus, the dorsal transformation of FGF-rescued limbs is extensive albeit incomplete (Fig. 3F).

Table 2.

Dorsal and ventral areas in ll limbs.

	% Dorsal area	% Ventral area
WT	54 ± 1	46 ±1
ll	89 ± 2	11 ± 2
Rescued ll	80 ± 2	20 ± 2

The dorsal limb area is defined as Plzf+ Lmx1+, ventral limb area as Plzf+ Lmx1–. Percentages indicate the dorsal or ventral limb areas relative to the total measured limb area and are reported as mean ± SEM. Three sections were measured per embryo. Areas were measured in 11 wild type, 4 ll and 14 FGF-rescued ll embryos.

These results raise the possibility that the ll ventral limb remnant influences the DV targeting of motor axons. Previous experiments showed that limbs truncated on the proximal axis (but where the amounts of proximal dorsal and ventral tissue were still substantially equal) still supported a correct DV choice (Tosney and Landmesser, 1984). Our ll data extend these observations and further show that even very unequal proportions of dorsal and ventral tissue (80%:20%) still support a correct DV motor axon choice.

The extent of ll dorsal limb transformation varies with time

The differences in the proportion of ventral tissue between FGF-treated and untreated ll limbs led us to ask if the dorsal ll character varies with developmental time. We analyzed DV identity from early limb outgrowth (HH 19) until after nerves invade the limb mesenchyme (HH27). At HH19 in non-rescued ll limbs, there is a small proximal ventral Lmx1– domain (Fig. 4A), accounting for 20% of the limb (Fig. 4E, Table 3). The ventral proximal tissue is expanded at HH 24 in FGF-rescued ll limbs (Figs. 4B, 3A, 3E; 4E, 27% of the limb) and also is proportionally larger at HH 25+, after motor axons start invading the limbs (Fig. 4C). At HH 27, the fraction of ll ventral tissue is 18% and thus smaller (Figs. 4D, 4E) than at HH 24. Thus, the ll dorsal limb transformation is continuously incomplete and the proportion of ventral tissue is highest around the time of DV choice.

Figure 4. The extent of the ll dorsal limb transformation varies with time.

(A) At HH19, before motor axons exit the SC, half of the wild type limb (i-iv) is dorsal. Most mutant limb territory (v-viii) is dorsal (Plzf+ Lmx1+) and very little tissue is ventral (Plzf+ Lmx1–).

(B) At DV choice time (HH24), while half of the wild type limb territory (i-iii) is dorsal Plzf+ Lmx1+, most FGF-rescued ll limb territory (iv-vi) is dorsal.

(C) As axons start to invade the limbs (HH25+), half of the wild type limb territory (i-iv) is dorsal Plzf+ Lmx1+ and most FGF-rescued ll limb territory (v-viii) is dorsal. The ventral nerve branch in ll projects towards the ventral (Plzf+ Lmx1–) territory.

(D) At HH27, nerves have invaded the limb. Half of the wild type limb (i-iv) is dorsal (Plzf+ Lmx1+) and most FGF-rescued ll limb territory (v-viii) is dorsal. Note that Plzf expression is weaker in the mutant than in wild type at this stage.

The arrows mark the ventral nerve branch.

(E) Quantification of DV areas at three developmental times: H19, shortly after the limb bud forms; HH 24, the time of DV motor choice; HH27, after the axons have invaded the limb. At all times, the proportion of dorsal limb tissue is significantly larger in ll (gray) than in wild type (white) (HH19, P = 0.024; HH24, P=0.0007; HH27, P=0.0003). Nevertheless, there still is some ventral proximal limb tissue left in ll.

Table 3.

Dorsal and ventral areas in ll limbs across time.

HH 19	% Dorsal area	% Ventral area
WT	56 ± 1.4	44 ±1.8
Rescued ll	80 ± 3	20 ± 1.3

HH 24	% Dorsal area,	% Ventral area
WT	56 ± 1.5	44 ±1.6
Rescued ll	72.5 ± 2.1	27.5 ± 5.4

HH 27	% Dorsal area	% Ventral area
WT	57 ± 1.6	43 ±2.1
Rescued ll	82 ± 2	18 ± 4

The dorsal and ventral areas (as defined in Table 2) were measured at HH 19 (before axon entry), HH 24 (when exons decide between dorsal and ventral limb) and HH 27 (after axons have invaded the limb). n=3 embryos examined in all cases except n=4 HH24 ll embryos.

Limb nerves branch abnormally in ll thoracic supernumerary limbs, which have an extensive but incomplete dorsal transformation

At thoracic levels, the flank tissue (that, like limb mesenchyme, is derived from the lateral plate mesoderm) is innervated by the lateral division of the medial motor column (lateral MMC), also termed the hypaxial motor column (HMC) (Bonanomi and Pfaff, 2010; Dasen et al., 2008). In contrast to the limb-innervating LMC, which is subdivided into lateral LMC and medial LMC, the thoracic HMC is not known to be subdivided. AER transplantation or FGF bead implantation in the thoracic flank induces supernumerary limbs (Abud et al., 1996; Crossley et al., 1996) innervated by motor axons, which form nerves and innervate limb tissue. These nerves have an overall topology that is intermediate in comparison with that of nerves at hindlimb and forelimb levels (Lance-Jones and Landmesser, 1981; Turney et al., 2003). It has not previously been determined whether thoracic supernumerary limbs have proper DV compartments and whether thoracic motor axons actively choose between dorsal and ventral limb.

We hypothesized that wild type supernumerary thoracic limbs would have normal DV polarity while ll supernumerary limbs would have a dorsal transformation, and thus that motor projection patterns in control and ll thoracic supernumerary limbs could address whether changing the DV character of the interlimb mesenchyme influences projection patterns and choice decisions. Thoracic supernumerary limbs were induced by inserting FGF beads in the flank ectoderm. The resulting outgrowths were smaller in ll than in wild type embryos (Supplementary Fig. 4). Nonetheless we were able to examine the extent of the dorsal transformation at the time of DV motor axon choice. In controls, half of the limb is dorsal (Plzf+ Lmx1+ EphA4+ ephrin-A5–) and half ventral (Plzf+ Lmx1– EphA4– ephrin-A5+) (Figs. 5A–C). In ll, the majority of the limb is dorsal (Figs. 5A–B) with a ventral proximal remnant (Fig. 5A–C). By contrast in ll, the dorsal area encompasses most of the supernumerary limb (79% ± 4%; Fig. 5E, Table 4, P = 0.0009 wild type vs. ll). Thus, as in forelimbs, the dorsal limb transformation in ll thoracic supernumerary limbs is incomplete. Moreover, supernumerary ll limbs protrude at an angle that is 27° more acute than in controls (ll 74° ± 10°, wild type 101° ± 6°, P=0.05, Fig. 5F), similar to the altered body angles in ll forelimbs.

Figure 5. FGF-induced thoracic ll supernumerary limbs have an extensive but incomplete dorsal transformation and abnormal innervation.

A) FGF-induced thoracic ll limbs have altered DV polarity. In wild type supernumerary HH24–25 limbs (i-iv), Plzf (i) is expressed in all limb territory and aligns proximally with Lmx1 (ii). In supernumerary ll HH24–25 limbs (v-viii), extended dorsal (Plzf+ Lmx1+) and reduced ventral (Plzf+ Lmx1–) domains are found. Incoming axons (iii, vii) are visualized by detecting neurofilaments (NF).

(B) In supernumerary wild type limbs (i-iii), EphA4 (i) is a proximal dorsal limb marker and its proximal border coincides with that of Lmx1+ (ii). In FGF-induced ll limbs (iv-vi) the proximal border of EphA4 (iv) coincides with that of Lmx1+ (v) only dorsally; the proximal ventral tissue does not have dorsal character.

(C) In supernumerary wild type limbs (i-iii), ephrin-A5 (i) marks the ventral limb and does not overlap with Lmx1 (ii). In most supernumerary ll limbs (iv-vi), most limb territory is dorsal (Plzf+ Lmx1+ ephrin-A5–) but a small ventral proximal (Plzf+ Lmx1– ephrin-A5+) territory persists.

(D) At HH25, NF+ axons branch and invade FGF-induced limbs in both wild type (ii) and ll mutant (iv-v) while contralateral operated sides have no limbs and no nerve branching (i, iii). In wild type FGF-induced limbs, the dorsal nerve branch invades the dorsal Lmx1+ territory and the ventral branch invades the Lmx1– territory (ii). In ll FGF-induced limbs (iv-v), most limb territory is Lmx1+, the dorsal nerve branch invades Lmx1+ territory while the ventral branch (arrow) straddles the border between Lmx1+ and Lmx1– territory.

(E) Analysis of supernumerary limb areas in wild type (white) and FGF-induced ll embryos (gray). The percentage of (Plzf+ Lmx1+) dorsal limb tissue is significantly larger in ll (wild type 48 ± 2%, n=4 embryos; mutant 79 ± 4%, n=5 embryos, P=0.0009) but there still is a (Plzf+ Lmx1–) ventral proximal limb tissue remnant.

(F) The angle between the supernumerary limb and body, defined as in Figure 1C, is smaller in ll limbs (gray) than in wild type (white) (wild type 101 ± 6°, n=4; mutant 74 ± 10°, n=5, P=0.05).

(G) A summary of the expression data. Plzf is a general limb marker. Half of the wild type limb and most of the ll limb are dorsal (Plzf+ Lmx1+). EphA4 and ephrinA5 are dorsal and ventral limb markers.

Dorsal is to the top, medial to the left. The SC is dorsal and medial (D).

Table 4.

Dorsal and ventral areas in FGF-induced supernumerary ll limbs.

	% Dorsal area	% Ventral area
WT	48 ± 2	52 ± 2
ll	79 ± 4	21 ± 4

Measurements were taken as indicated in the legend of Table 2. n=4 FGF-induced limbs in both wild type and ll embryos.

We also examined nerve patterns in relation to dorsal and ventral limb territories as defined by limb markers (Fig. 5A–D). In wild type the supernumerary limb nerve splits into dorsal and ventral branches at the base of the limb (Lance-Jones and Landmesser, 1981; Turney et al., 2003), the bifurcation point coincides with proximal Lmx1 expression (Fig. 5D), and dorsal and ventral branches invade Lmx1+ and Lmx1– territory respectively. In ll a spinal nerve reaches the limb and a few axons penetrate the limb (Fig. 5D). While the limb nerve also splits into dorsal and ventral branches, the ventral branch traces the border between Lmx1+ and Lmx1- territory (Fig. 5D ii-iii, v-vi, arrow), similar to the pattern at limb axial levels. In contrast, no branching occurs on the unoperated side in either wild type or ll (Fig. 5D i, iv).

Thus the dorsal transformation in ll supernumerary limbs is extensive yet incomplete and perturbs the orientation of the limb relative to the body axis (Fig. 5G). Surprisingly, ll axons form a limb nerve that bifurcates into dorsal and ventral branches upon encountering supernumerary limb tissue. However the fidelity of this DV axon choice cannot be assessed with molecular markers in supernumerary limbs since the HMC has no known subdivisions. Furthermore, the ventral branch tracks the border between dorsalized limb and the ventral limb remnant, indicating that some lateral MMC axons are apparently sensitive to DV targeting cues derived from the dorsal limb and suggesting that the lateral MMC might be subdivided. Thus, as at forelimb level, the geometry of limb nerves is abnormal and the effect on the nerve spatial pattern is even more notable considering that the limb-body angle is smaller in ll than in wild type.

Spinal cord neuronal populations are largely normal in ll

ll mutants have abnormal limbs but a morphologically normal spinal cord (Grieshammer et al., 1996; Lanser and Fallon, 1984). Whereas no defects were observed in the generation of motor neurons (Lanser et al., 1986; Lanser and Fallon, 1984), the allocation of motor neurons to columnar divisions and the generation and positioning of non-motor populations were not examined. Since we did observe significant changes in the geometry of ll motor nerves, we investigated the SC neural architecture in ll in order address whether ll motor axon projection decisions might be influenced by changes in the motor neurons, rather than peripheral tissue. We asked whether the ll spinal cord is properly patterned and whether all classes of spinal cord neurons are properly specified.

Neural progenitors are organized along the DV axis of the ventricular SC in domains that give rise to distinct populations of postmitotic neurons. We first used molecular markers (Briscoe et al., 2000; Jessell, 2000) to characterize the generation and distribution of three progenitor populations in ll. Motor neuron progenitors (pMN) are marked by the expression of Olig2 (Novitch et al., 2001) and are situated dorsally to Nkx2.2+ V3 ventral interneuron progenitors (pV3) and ventrally to Nkx6.1+ ventral interneuron progenitors V2 (pV2) (Fig. 6A). In ll embryos, pMN, pV3 and pV2 are present and correctly positioned (Fig. 6A). While the ll pV3 population is increased by 39% relative to controls (Table 5, Fig. 6B–C), the increase is not statistically significant (P=0.17). Motor neurons, marked by pre- and early post-mitotic marker MNR2 expression, are generated in normal numbers and correctly positioned in the ll ventral SC (Figs. 6A, C). The positions and numbers of progenitors of other populations of dorsal and ventral interneurons were similar in wild type and ll embryos (Fig. 6A–C). These data show that DV patterning of the ventricular SC is essentially normal in ll.

Figure 6. Spinal cord progenitor populations are largely normal in ll.

(A) All five ventral spinal cord neuron progenitor populations (Briscoe et al., 2000) are present at HH22 and situated in their normal relative positions in ll. Dorsal neuron progenitors are also present in both wild type and ll.

(B) Diagram of the SC neuron progenitor populations. A transcription factor code identifies progenitor populations of V3 interneurons (pV3), motor neurons (pMN) and V2 interneurons (pV2).

(C) pV3, pMN and pV2 were counted using the code in (B). pMN and pV2 are similar in wild type and ll but pV3 is increased by 39% in ll (Table 5).

Table 5.

Spinal cord progenitor populations in ll.

	Number of cells /section
	pV3	pMN	pV2
WT	49 ± 6	45 ± 6	29 ± 3
ll	68 ± 11	42 ± 6	30 ± 4

The figures represent mean ± SEM cells counted per section. n=3 wild type and 3 ll embryos; at least four sections per embryo were counted.

pV3, pMN, pV2 are progenitors of V3 interneurons, motor neurons and V2 interneurons. pV3 are identified as Nkx2.2+ Nkx6.1+ Olig2–; pMN are Nkx2.2– Nkx6.1+ Olig2+; pV2 are Nkx2.2– Nkx6.1+ Olig2–.

We next compared the distribution and numbers of postmitotic SC neurons in wild type and ll embryos (Fig. 7). We found no difference between ll and wild type embryos (Fig. 7A) with regard to the relative positions of D1-D6 dorsal interneurons (Muller et al., 2002) (Fig. 7B). We also found no difference (Fig. 7C) with regard to the presence and relative positions of ventral interneurons (V0-V3) and motor neurons (MN) (Briscoe et al., 2000; Jessell, 2000) (Fig. 7D). Since the V3 progenitor population is expanded in ll (Figs. 6A, 6D), we evaluated the V3 postmitotic neurons more closely (Fig. 7C) by measuring the area of the ventral SC occupied by cells expressing the transcription factor Sim1, which marks V3 interneurons (Briscoe et al., 1999). Indeed, the V3 population occupied an area that is significantly increased by 23% in ll compared to control embryos (Fig. 7E, P=0.03).

Figure 7. Spinal cord postmitotic neuron populations are largely normal in ll.

(A) ll spinal cords generate all dorsal D1-D6 interneuron populations, identified by the combinatorial expression of homeodomain transcription factors (Muller et al., 2002).

(B) Diagram of SC dorsal postmitotic neuron populations.

(C) ll mutant spinal cords generate all ventral interneuron populations (Briscoe et al., 2000).

(D) Diagram of the SC dorsal postmitotic neuron populations.

(E) Areas expressing Sim1 mRNA were measured. The Sim1+ postmitotic V3 population in ll is increased by 23% (ll 1477 ± 75 μm², n=7 embryos; wild type 1203 ± 89 μm², n=7 embryos; P=0.03).

(F-H) ll spinal cords have all three (medial LMC, lateral LMC, medial MMC) brachial motor columns and each have a proportion similar to that in wild type spinal cords at stages HH 25 and HH 27 (Table 7).

(I) At HH 25, the total number of brachial LMC motor neurons does not differ between wild-type and ll. By HH27, ll spinal cords have a deficit of total motor neurons (P=0.004, Table 6), consistent with a lack of target-derived trophic factors.

Limbs in mutant embryos of Figs. 6 and 7 were not FGF-rescued.

* Denotes P<0.001 (two-tailed t-test).

We also assessed the number of motor neurons in ll as compared to wild type embryos. While the total number of motor neurons in ll embryos was previously determined (Lanser et al., 1986), the columnar specification (medial LMC, lateral LMC, medial MMC) of these neurons was not analyzed. Consistent with previous observations (Lanser et al., 1986), we found that the total number of motor neurons is slightly decreased in ll at HH 25 and is significantly decreased by 22% at HH 27 (Fig. 7I, Table 6, P=0.004). Since motor neuron survival depends on target tissue (Hamburger, 1975), the progressive decrease of motor neuron numbers in ll embryos could be accounted for by the virtual absence of limb target tissue in the embryos. We also found that in ll all three columnar classes of motor neurons (Tsuchida et al., 1994) are present in the appropriate relative proportions (Fig. 7F–H, Table 7). Additionally, we found no significant differences between wild type and ll thoracic spinal cord progenitor and postmitotic populations (data not shown). Therefore, in ll both the generation of MN classes and the DV patterning of the postmitotic SC are normal, with the exception of the ventral-most V3 interneuron population. Thus, the ll spinal cord is essentially normal, indicating that the abnormal ll motor innervation of the limb is caused by changes in target tissues, not in motor neurons.

Table 6.

Total number of MN in ll.

	HH25 Neurons	HH27 Neurons
WT	18,949 ± 1972	19,229 ± 699
ll	16,679 ± 1701	15,093 ± 405
P	0.43	0.004

n = 3 wild type and 3 ll embryos (HH25) and 4 wild type and 4 ll embryos (HH27). In each embryo, motor neurons were counted in every fifth section throughout the brachial spinal cord. The total number of brachial motor neurons is reported as mean ± SEM. P values calculated by 2-tailed t-test.

Table 7.

MN columnar division identities in ll.

	% LMCl		% LMCm		% MMCm
	HH25	HH27	HH25	HH27	HH25	HH27
WT	32 ±1.6	30 ± 1.3	41 ± 1.9	46 ± 1.0	26 ± 1.5	24 ± 0.6
ll	33 ±1.7	32 ± 1.2	42 ± 1.6	45 ± 1.7	25 ± 2.1	24 ± 1.4

Percentages refer to the number of motor neurons within a division relative to total number of motor neurons (Table 6) and are reported as mean ± SEM. Medial MMC neurons are Lim1– Isl1+ Lim3+, lateral LMC neurons are Lim1+ Isl1– Lim3–, and medial LMC neurons are Lim1– Isl1+ Lim3–.

DISCUSSION

We found the dorsal transformation in ll FGF-induced limbs, though extensive, is incomplete and perturbs the orientation of the limb relative to the body axis. ll axons form a limb nerve that bifurcates into dorsal and ventral branches upon encountering limb tissue, resulting in abnormal nerve patterns. The ventral branch does not innervate the dorsally transformed ventral tissue but tracks the border between dorsal limb and the ventral limb remnant. While the medial and lateral LMC axons segregate correctly, the geometry of limb nerves is abnormal. Thus, ll stands in marked contrast with other mutants with perturbed limb DV polarity, such as bidorsal BmprIa^flox/- and biventral Lmx1b^−/− (Kania et al., 2000; Luria and Laufer, 2007), in which the geometry of limb nerves is always normal even though the medial and lateral LMC axons segregate incorrectly. The strikingly abnormal geometry of ll nerves may be explained by the displacement of the border between dorsal and ventral compartments, indicating that the geometry of limb nerves is controlled by the position of the DV compartment boundary.

Molecular code analysis shows ll limbs have an extensive dorsal transformation

The DV limb code (Luria and Laufer, 2007) is valid in chick as well as mouse, and can aid in analyses of mutants with perturbed limb DV polarity (Bell et al., 1998; Loomis et al., 1996; Mic et al., 2002; Parr and McMahon, 1995; Wang et al., 2004). We used this code to analyze ll FGF-induced forelimbs and thoracic supernumerary limbs, and found the dorsalization of ll limbs is extensive but incomplete. This observation contrasts with previous analyses, which found ll FGF-induced limbs to be bidorsal and did not describe ventral limb remnants (Grieshammer et al., 1996; Noramly et al., 1996; Ros et al., 1996). Interestingly, since more ventral tissue is present in rescued (20%) than in non-rescued ll limbs (11%), FGF may promote growth of ventral tissue more effectively than that of dorsal tissue. Thus ll lateral plate mesoderm is dorsally transformed at both limb and interlimb anterior-posterior axial levels.

The fidelity of ll motor projections is preserved and reflects an active DV choice

The ll motor projection phenotype illuminates how medial LMC axons are matched to ventral limb. First, medial LMC axons do not invade the dorsal territory and project towards the ventral remnant possibly because the ventral limb remnant suffices to present medial LMC and lateral LMC axons with an almost normal choice between ventral and dorsal limb. Motor axons choose correctly even if most limb territory is ablated, if the remaining target tissue contains both ventral and dorsal tissue (Tosney and Landmesser, 1984, 1985a, b). While in limb partial ablation experiments, the limb remnant dorsal:ventral ratio was approximately 1:1, in ll this ratio is 4:1. Nevertheless, the DV segregation of medial and lateral LMC axons occurs correctly in ll, indicating the fidelity of DV motor axon targeting depends on the presence of both target tissues, not on their relative amounts. This rules out passive deployment of medial LMC axons to the ventral limb (Lance-Jones, 1986; Summerbell and Stirling, 1981) and indicates a strong preference of medial LMC axons for ventral limb (Ferguson, 1983; Whitelaw and Hollyday, 1983). Second, medial LMC axons may avoid dorsal tissue, since the ventral nerve branch tracks the border between Lmx1+ and Lmx1- tissue. This is consistent with the larger angle between the ventral and dorsal nerve branches in ll, and with the rerouting of medial LMC axons outside the limb in mouse conditional BmprIa^flox/- mutants with completely bidorsal limbs (Luria and Laufer, 2007). If so, the amount of ventral tissue left in ll is irrelevant, as medial LMC axons refrain from projecting into dorsalized ventral limb because they avoid dorsal limb.

Eph:ephrin signaling might mediate the shift in the branch projection position in ll mutants. When faced with the option of entering dorsal territory that expresses ephrin-B2 or ventral territory that does not, EphB1+ medial LMC axons choose the less repellant limb compartment (Luria et al., 2008). Attractive cues encoded by reverse ephrin-Eph signaling pairs of molecules also function in LMC(l) and LMC(m) axons (Dudanova and Klein, 2011; Kao and Kania, 2011) and might contribute to determining projection position. These cues are encoded by ephrins expressed in the axons, such as ephrin-B2 in LMC(m) neurons (Luria et al., 2008) and cognate Eph receptors preferentially expressed in ventral limb (EphB2 and EphB3) (Kania and Jessell, 2003). However in Lmx1b^−/− or conditional BmprIa^−/− mutants, all limb territory either lacks or expresses ephrin-B2 (Luria et al., 2008; Luria and Laufer, 2007) or EphA proteins (Kania and Jessell, 2003). In these cases there is no more or less repulsive or attractive option, and LMC axons appear to use a different system for determining their entry trajectories: the permissive axonal highways described by (Tosney and Landmesser, 1985a). In this view, there is a hierarchy of route guiding options. First, when present, differentially repulsive or attractive and instructive Eph:ephrin signaling determines the branch paths, with repulsive cues acting more strongly than attractive ones. Attractive cues may be encoded by reverse ephrin-Eph signaling pairs of molecules functioning in LMC(l) and LMC(m) axons (Dudanova and Klein, 2011; Kao and Kania, 2011), such as ephrins expressed in the axons (Luria et al., 2008) and cognate Eph receptors expressed in dorsal limb (EphA4 and ΣEphAs (Kania and Jessell, 2003)) or ventral limb (EphB2, EphB3 (Luria et al., 2008)). Second, permissive routes, which are normally dominated by the Eph signaling systems, come into play in their absence. Alternatively, while both LMC axon segregation and pathway positioning are under the control of the limb DV patterning system, only the segregation choice might involve instructive interactions between motor axons and peripheral guidance cues. Thus, the location of permissive pathways might be the only player in determining the route of the axons at the base of the limb. Interestingly, axon projection into the dorsal limb follows a trajectory centered in the dorsal limb mesenchyme, rather than following a compartment border. Whether this reflects a difference in the hierarchical relationships of these systems or distinct mechanisms of establishing these pathways is an interesting question.

Motor projections to ll limbs have an abnormal geometry controlled by the DV boundary

A disorganized plexus, with fan-like axonal structures, was previously described in ll FGF-induced limbs and interpreted as evidence that axons invade such limbs in a disorderly manner (Munoz-Sanjuan et al., 2000). However, few developmental stages and few embryos with very incomplete rescues were examined, while retrograde nerve labeling was not attempted. Moreover, axonal patterns were examined only with whole-mount antibody stains, a method of lower resolution than antibody stains on serially sectioned tissue. In contrast, we observed that axons invade the limbs in a stereotyped manner: the limb nerve splits into a dorsal and a ventral branch, the dorsal branch enters the dorsalized limbs while the ventral branch straddles the border between the dorsal limb and the ventral limb remnant (Fig. 8). Moreover, the ll ventral branch is situated at a wider angle relative the dorsal branch. Additionally, the ll dorsal nerve branch does not have a normal trajectory and is situated at a smaller angle relative to the limb nerve compared to wild type, possibly reflecting the perturbed position of the ll limb relative to the body axis. Thus, the ll motor innervation pattern is a combination of accurate segregation, independent of the dorsal:ventral ratio, and abnormal geometry, which is dependent on the dorsal:ventral ratio, and both are a consequence of the shift of the DV border.

Figure 8. Summary of wild type and ll motor axon projection patterns.

(A) In ll forelimbs, limb nerves branch abnormally while motor axon targeting is normal. The mutant ventral branch tracks the border between Lmx1+ and Lmx1– territory.

(B) In both wild type and ll thoracic supernumerary limbs, limb nerves bifurcate. The mutant ventral branch tracks the border between Lmx1+ and Lmx1– territory, although both branches originate from the thoracic medial MMC.

Molecular differences exist between medial LMC axons that innervate the ventral flank or the ventral limb

PSA-NCAM is present on LMC axons that innervate the dorsal and the ventral limb but not the ventral flank. Medial LMC axons innervate both limb and flank targets at both forelimb (Haase et al., 2002) and hindlimb level (Luria and Laufer, 2007). Molecular differences between limb-innervating and flank-innervating axons, such as the differential expression of transcription factors Scip and Pea3, exist within the medial LMC (Haase et al., 2002; Luria and Laufer, 2007). Here we show that medial LMC axons innervating the ventral limb are PSA-NCAM+ while medial LMC axons innervating the flank are PSA-NCAM–. Thus PSA-NCAM may play a role in the differential targeting of medial LMC subgroups. This is consistent with the fact that although PSA-NCAM has been reported to be present only in the dorsal limb nerve derived from lateral LMC, blocking PSA-NCAM function resulted in projection errors of both lateral LMC and medial LMC axons (Tang et al., 1992; Tang et al., 1994).

The thoracic lateral MMC may be subdivided

The ll thoracic spinal cord appears normal while the supernumerary limb exhibits a dorsal transformation. Thoracic motor axons project to both dorsal territory and ventral limb remnant. A subdivision of the spinal cord, the lateral division of the medial motor column (MMC(l)), which at thoracic levels is called the hypaxial motor column (HMC) (Agalliu et al., 2009; Dasen et al., 2008; Stifani, 2014), projects axons through the ventral root and innervates the thoracic body wall muscles (Bonanomi and Pfaff, 2010; Hollyday et al., 1977) as well as wild type thoracic supernumerary limbs (Hollyday et al., 1977). Nevertheless, there is no apparent topographic difference in the cellular origins of the axons from the dorsal or ventral nerve branches in thoracic supernumerary limbs (Jennifer Brown and Thomas Jessell, personal communication). While the identity of thoracic HMC neurons is transcriptionally specified by Isl1/2 and Hb9 (Dasen et al., 2008; Rousso et al., 2008) accompanied by downregulating Lhx3 (Bonanomi and Pfaff, 2010), the guidance of HMC axons employs the TAG1 surface molecule to control the fasciculation and initial guidance decisions (Suter et al., 2020), and Nrp1 and Nrp2 as receptors for peripheral Sema3A guidance molecule (Huber et al., 2005; Moret et al., 2007; Nedelec et al., 2012). In ll, the limb nerve that innervates the supernumerary limb bifurcates into a dorsal and a ventral branch. Similar to forelimb levels, the ventral nerve branch does not invade the dorsally transformed supernumerary limb and the geometry of nerve branches in supernumerary limbs is thus also determined by the border between Lmx1+ and Lmx1- territory. This phenotype raises the possibility that the geometry of the two nerve branches in supernumerary limbs reflects a hitherto unidentified distinction between HMC neurons, despite the absence of overt differences in the HMC columnar organization. Alternatively, the altered branch geometry might reflect purely changes in local limb mesenchyme interactions that lead to the establishment of altered permissive pathways. In this view subdivision of the HMC would be unnecessary and branch choice randomized due to the lack of axonal input to this decision.

Limb and spinal cord abnormalities indicate signaling pathways that may be affected by the ll mutation

ll is a single recessive mutation (Fallon et al., 1983; Prahlad et al., 1979) and has eluded efforts aimed at identifying its molecular nature (Dealy et al., 1997; Ferrari et al., 1999; Lizarraga et al., 1999). In ll, SC neuronal populations are normally generated and maintained postmitotically, with the exception of the V3 interneuron postmitotic population. V3 interneurons are the neuronal population situated closest to the floor plate, a source of SC neuronal DV patterning (Briscoe and Ericson, 2001; Ericson et al., 1996; Jessell, 2000; Yamada et al., 1993). Since the floor plate patterns ventral SC populations and the ventral-most V3 population is expanded in ll, the ll mutation may enhance the function of the floor plate, which secretes the morphogen Shh and can be functionally replaced by it (Ericson et al., 1996). The V3 increase could be explained by an increase in the concentration of Shh protein secreted by the floor plate. Alternatively, V3 precursors may transduce the Shh signal more effectively. Thus the ll mutation might enhance the SC patterning function of the floor plate, possibly by enhancing Shh signaling in a manner similar to Tectonic (Reiter and Skarnes, 2006) but opposed to Rab23 (Eggenschwiler et al., 2001).

The ll mutation affects limb DV polarity by increasing the extent of dorsal territory throughout the anterior-posterior body axis. BMP signaling controls the formation of ventral body and ventral limbs: in some BMP pathway mutants, the ventral body wall is absent or reduced (Zakin et al., 2005). Moreover, the conditional mouse mutant BmprIa^flox/- is not only bidorsal (Luria and Laufer, 2007) but also has a defective or absent AER in affected hindlimbs (Ahn et al., 2001), reminiscent of the absence of AER formation in ll limbs. Since ventral limb territory is reduced in ll, it is possible that the ll mutation partially compromises the BMP signaling pathway.

In summary, we have shed light on the behavior of motor axons in limbless. We found that the limbless mutation results in an extensive but incomplete dorsal limb transformation but does not affect motor neurons. In ll limb-projecting motor axons segregate correctly but form nerves with an abnormal geometry. These results show that the DV motor axon choice depends solely on the absolute presence of dorsal and ventral tissue, not on their relative ratio and that the geometry of motor axon projections to the limb is dependent on, and controlled by, the position of limb compartment boundaries.

Supplementary Material

MMC1

Supplementary Figure 1. FGF-soaked beads rescue ll forelimb outgrowth.

(A) At HH27 (E5.5), ll embryos have a forelimb outgrowth on the operated right side; the unoperated side has no limb (ii). The ll limb outgrowth comprises at least the proximal limb. A wild type sibling of the same age has well-developed forelimbs (i).

(B) At HH35 (E8.5), a cartilage stain (wild type and ll n=4 embryos) in a wild type embryo (i) reveals the forelimb bones: the humerus in the proximal limb, radius and ulna in the middle segment, digits in the distal segment. (ii) A sibling ll embryo has no limb bones in the unoperated side but (iii) and has a forelimb outgrowth containing a humerus in the operated side. The FGFsoaked bead and the attaching wire are embedded in the posterior limb tissue (iii).

(A, B) All views are dorsal. Anterior (head) is to the top, posterior (tail) to the bottom. Thick arrows (Aii, Biii) indicate FGF-soaked beads. Thin arrows (Bi, Biii) indicate humeral cartilages.

Supplementary Figure 2. Nerve branching patterns in rescued ll limbs.

(A) In ll, most rescued limb territory is dorsal and marked by Lmx1+ (pink). Axons are NF+. (i-iii) In most ll embryos, axons segregate into a dorsal branch and a ventral branch (iii, arrows). The dorsal nerve branch invades the Lmx1+ territory while the ventral branch tracks the border between Lmx1+ and Lmx1– territory. (iv-vi) In two embryos, a second phenotype appears: axons are not segregated into branches but form a fan-like structure (vi, arrow).

(B) At HH27, PSA-NCAM is present on axons that invade the forelimb, but not the ventral flank. (i-vi) In intact wild type embryos (i-iii) and wild type embryos with the AER replaced by an FGF-soaked bead (AER– FGF+; iv-vi), three nerve branches derived from the LMC exist: a first branch invading the Lmx1+ dorsal limb, a second branch invading the Lmx1–ventral limb, and a third branch invading the flank (arrows in iii, vi). In both wild type (i-iii) and AER– FGF+ (iv-vi) embryos, most axons in the dorsal branch and some axons in the ventral branch that innervates the limb are PSA-NCAM+, while the axons of the ventral branch invading the flank are PSA-NCAM–(arrows in iii, vi).

(vii-ix) In ll embryos, only two nerve branches are present: a first branch invading the Lmx1+ dorsal limb and a second branch invading the small Lmx1– ventral limb remnant; there is no third branch invading the flank outside the limb. In ll embryos (vii-ix) most axons of dorsal branch PSA-NCAM+. In the only ventral branch that is present and invades the small ventral limb remnant, some axons are PSA-NCAM+ while others are PSA-NCAM– (arrow in ix). This is consistent with the possibility that the third PSA-NCAM– branch, that would otherwise innervate the flank, is fused to the second branch that innervates the ventral limb remnant.

Dorsal is to the top, proximal limb is to the left, distal to the right.

Supplementary Figure 3. FGF bead implantation does not affect limb motor innervation.

(A, B) Dorsal nerves originate from lateral LMC in both wild type (n=5 embryos) and AER– FGF+ (n=4 embryos). (A) Diagram of the experiment in (B). (B) In both wild type (i) and AER– FGF+ (ii) SC, mostly Isl1– lateral LMC cells are HRP+.

(C, D) Ventral nerves originate from medial LMC in both wild type (n=4 embryos) and AER–FGF+ (n=4 embryos). (C) Diagram of the experiment in (D). (D) In both wild type (i) and AER–FGF+ (ii) SC, mostly Lim1– medial LMC cells are HRP+.

The outlines of lateral and medial LMC motor subcolumns are marked with a dotted white line.

Supplementary Figure 4. FGF-soaked beads induce thoracic ll supernumerary limb outgrowths.

(A) At HH27 (E5.5), a wild type embryo has well-developed forelimbs and an FGF-induced supernumerary limb (arrow).

(B) A sibling ll embryo of the same age has an FGF-induced supernumerary thoracic limb outgrowth on the operated right side (arrow) while the unoperated side has no limb.

Highlights.

The geometry of limb motor innervation is controlled by the dorsal-ventral compartment boundary

In chick limbless FGF-rescued limbs, the angle between dorsal and ventral nerves is large, changing motor nerve geometry

The dorsal-ventral motor axon targeting is not affected in rescued limbless limbs

The double-dorsal transformation of rescued limbless limbs is extensive but incomplete, and varies with time

Ventral V3 spinal interneurons are expanded in limbless, hinting at the molecular nature of the unidentified limbless gene