Lateral motor column axons execute a ternary trajectory choice between limb and body tissues

Background Neuronal topographic map formation requires appropriate selection of axonal trajectories at intermediate choice points prior to target innervation. Axons of neurons in the spinal cord lateral motor column (LMC), as defined by a transcription factor code, are thought to innervate limb target tissues exclusively. Axons of the medial and lateral LMC divisions appear to execute a binary decision at the base of the limb as they choose between ventral and dorsal limb trajectories. The cellular logic that guides motor axon trajectory choices into non-limb tissues such as the ventral flank remains unclear. Results We determined the spinal cord motor column origin of motor nerves that innervate ventral flank tissues at hindlimb level. We found unexpectedly that a subset of medial LMC axons innervates ventral non-limb mesenchyme at hindlimb level, rather than entering ventral limb mesenchyme. We also found that in a conditional BmprIa mutant where all ventral hindlimb mesenchyme is converted to a dorsal identity, all medial LMC axons are redirected into the ventral flank, while lateral LMC axons innervate the bidorsal limb. Conclusion We have found that medial LMC neurons innervate both ventral flank and limb targets. While normally only a subset of medial LMC axons innervate the flank, all are capable of doing so. Furthermore, LMC axons execute a ternary, rather than binary, choice at the base of the limb between ventral flank, ventral limb and dorsal limb trajectories. When making this choice, medial and lateral LMC axons exhibit different and asymmetric relative preferences for these three trajectories. These data redefine the LMC as a motor column that innervates both limb and body tissues.


Background
The precision of neural circuits requires stereotypic patterns of neuronal connectivity, which are often organized as topographic maps. Studies of how the point-to-point connections between spinal cord motor neurons and their targets are generated have revealed much about how such maps are constructed [1][2][3]. One principle to emerge from these studies is that functionally related neurons extend axons along shared trajectories that are precisely and accurately subdivided at intermediate choice points [4,5]. Deciphering the logic underlying these trajectory choices is thus critical to understanding how these connectivity patterns are established.
The overall map between motor neurons and their peripheral targets is well described [3,[6][7][8]. The motor neurons are organized mediolaterally into medial and lateral motor columns (MMC and LMC) and their divisions. Medial MMC neurons innervate dorsal axial muscles. At thoracic levels, lateral MMC neurons innervate ventral body wall muscles, and autonomic motor neurons of the intermediolateral column (IML, column of Terni in birds) innervate sympathetic postganglionic neurons. At limb levels, lateral LMC neurons innervate dorsal limb muscles while medial LMC neurons innervate ventral limb muscles. The motor columns are further subdivided into pools, and each motor pool innervates an individual muscle [3]. These descriptions are generally accurate, although possibly incomplete, as the identity of motor neurons that innervate many muscles, particularly outside of the limb, has not been examined in detail.
Current models suggest that shared intrinsic properties of neurons in a columnar division guide common trajectory decisions required of all axons projecting from that division [3]. With the exception of some cervical motor populations, all motor axons exit the ventral root of the spinal cord along a ventrolateral trajectory. Divisional populations deviate from this shared path at progressively more distal choice points. At limb axial levels the final divisional decision is made in the motor plexus at the base of the limb, where LMC axons assume trajectories into either dorsal or ventral limb mesenchyme.
The discovery that combinations of LIM homeodomain transcription factors identify motor neurons within columnar divisions led to the idea that the columnar identity of a motor neuron as defined by this transcription factor code is predictive of its axonal trajectory [3,9]. Consistent with this, neurons whose cell bodies are misplaced within morphological columns relative to their axonal trajectories nonetheless express columnar transcription factors appropriate to their trajectories [9]. Gain and loss of function genetic experiments demonstrate that these transcription factors impose specific pathfinding behaviors on motor axons [10,11], and do so largely by regulating expression of axon guidance receptors [12,13].
Experiments designed to test how trajectories into limb mesenchyme are chosen by axons of LMC neurons have invariably found both medial and lateral LMC axons entering the limb. This is the case for limbs that are surgically truncated, rotated or duplicated along the dorsalventral axis [4,[14][15][16][17]; for mutants in which limb mesenchymal dorsoventral identity or guidance cue expression is altered [11,[18][19][20]; for mutants in which LMC axonal guidance receptor expression is altered [12,[19][20][21][22]; and for embryos in which spontaneous neural activity is inhibited [23]. These results, in combination with descriptions of the neuromuscular topographic map [6][7][8], have led to the idea that at the base of the limb, LMC axons have only two options: they make a binary choice between trajectories into either dorsal or ventral limb mesenchyme [3,[24][25][26].
The contributions of both intrinsic properties of LMC motor neurons and extrinsic guidance cues to this trajectory choice have been further defined. Surgical experiments demonstrate that the choice is active, and is regulated locally by signals generated at the choice point [4,[14][15][16]. Molecular genetic experiments have identified primarily [12,18,19,27,28], but not exclusively [20], repulsive ligand:receptor combinations that influence this choice. The emerging model is that the LIM homeodomain code imparts selective insensitivity to peripheral guidance cues until axons arrive at the base of the limb [3,13]. Differential sensitivity to guidance cues reflecting limb dorsoventral pattern then guides medial and lateral LMC axons as they choose between dorsal and ventral limb trajectories [11,12,[18][19][20]28].
While appealing, it is not clear if this model can account for the selection of trajectories by all motor axons that reach the base of the limb. Several nerves (iliohypogastric, ilioinguinal and genitofemoral) that project from the most rostral lumbar segments (L1-L2) of the spinal cord innervate lateral abdominal body wall (external oblique, internal oblique and transverse) and genital (cremaster) muscles [29,30]. The motor axons that ultimately form these nerves extend through the plexus into ventral flank mesenchyme. These trajectories are not typical of medial MMC, IML, medial LMC or lateral LMC axons, which are all present at this axial level.
We investigated the columnar identity of the ventral flank nerves and found that they originate within the medial LMC. This result indicates there is a third trajectory, in addition to the limb trajectories, available to LMC axons, one that binary choice models do not accommodate. To investigate how peripheral cues influence LMC axon trajectory decisions, we removed the ventral limb mesenchyme using a conditional bone morphogenetic protein receptor Ia genetic model (BmprIa flox/-) [31], in which all hindlimb mesenchyme has a dorsal identity. When medial LMC axons are confronted with this bidorsal limb mesenchyme, they do not enter the limb, and are redirected into the ventral flank. In contrast, lateral LMC axons enter the bidorsal hindlimb and populate both dorsal and ventral nerve branches. Our results show that motor axons have a ternary, not binary, choice of trajectories at the base of the hindlimb, and reveal that medial and lateral LMC axons have different and asymmetric preferences for trajectories that enter ventral flank, ventral limb or dorsal limb mesenchyme. Our findings redefine LMC columnar identity as predictive of trajectories that include both body and limb tissues.

Medial LMC axons project to the ventral flank mesenchyme
To determine the columnar identity of the motor neurons whose axons project into the ventral flank at hindlimb axial levels, we retrogradely labeled these neurons by horseradish peroxidase (HRP) or lysinated tetramethylrhodamine dextran (RDA) injection [24] into ventral flank mesenchyme of E13.5 mouse embryos (Figure 1). HRP+ and RDA+ cell bodies are restricted to rostral lumbar regions, consistent with the described trajectories of the relevant nerves. Retrogradely labeled neurons express predominantly the medial LMC marker Isl1, but not the medial MMC marker Lim3 or the lateral LMC marker Lim1 (Figure 1c,d) [9]. They also express FoxP1, which marks LMC but not MMC motor neurons (Figure 1e; Jeremy Dasen and Thomas Jessell, Columbia University, personal communication) [32]. The ventral flank nerves are thus composed of motor axons that originate from medial LMC neurons.
To determine whether these neurons exist within a single motor pool, we asked whether the retrogradely labeled neurons express additional transcription factor markers that label subsets of medial LMC neurons [33,34]. We could readily detect labeled Nkx6.1+ and labeled Nkx6.1neurons, and could occasionally detect labeled Er81+ neurons (Additional file 1). Together these data indicate that neurons that project along ventral flank trajectories likely derive from multiple motor pools, which is consistent with the observation that the nerves innervate multiple target muscles.
A subset of medial LMC neurons, as defined by their LIM homeodomain transcription factor code, thus unexpectedly project not to the limb, but to the ventral flank, after leaving the anterior plexus. This raises the possibility that this non-limb trajectory might be available to other medial LMC or perhaps all LMC axons. If so, then additional medial LMC neurons might also project to the ventral flank mesenchyme in appropriate mutant contexts.

BmprIa flox/mutant embryos have completely bidorsal hindlimbs
We reasoned that a mutant lacking ventral limb mesenchyme, the predominant medial LMC axonal target, is one scenario in which additional ventral flank projections might be observed, as medial LMC axons would, of necessity, innervate an alternative mesenchymal territory. Brn4cre Tg/-, BmprIa flox/-(BmprIa flox/-) mice carry a conditional null allele of the BmprIa gene that is inactivated to homozygosity by Cre recombinase expressed under the control of a transgenic Brn4 promoter [31]. In these mice the ventral hindlimb mesenchyme is transformed to an apparently dorsal identity, while forelimb dorsoventral polarity is normal [31,35,36]. While BmprIa flox/mutant hindlimbs are small, they nonetheless include proximal and intermediate limb tissue [31]. This mutant is thus a promising candidate in which to look for additional LMC axons innervating the ventral flank.
The dorsoventral projection choice is controlled by local signals [15,16,37]. Thus, we first confirmed that at the time of the choice in the BmprIa flox/mutant no residual ventral limb mesenchyme is present near the choice point, as previous studies did not establish this [31,36]. We defined a set of molecular markers for evaluating the dorsoventral identity of the proximal limb mesenchyme. Our criteria were: (1) that each marker is expressed either throughout the dorsoventral extent of the limb or is restricted to dorsal or ventral limb; and (2) that the markers share a proximal boundary adjacent to the plexus as axons begin entering the limb mesenchyme (E10.5-11.0 in hindlimb, E10.0-10.5 in forelimb) [11,16]. The markers in combination (the 'dorsoventral code') should define the limb mesenchyme as dorsal or ventral with respect to the projection choice and account for all topography along the limb dorsoventral axis.
We used these markers to examine the dorsoventral character of BmprIa flox/mutant limb mesenchyme. All of the Plzf+ hindlimb mesenchyme is Lmx1b+ (Figure 2b) and is also Σephrin-A- (Figure 2d). EphA4 is present throughout the dorsoventral extent of the hindlimb mesenchyme and, notably, its proximal ventral expression boundary coincides with that of Lmx1b (Figure 2c). Taken together, these results provide evidence that all of the mutant hindlimb mesenchyme is dorsal (Plzf+ Lmx1b+ Σephrin-A-) Medial LMC axons innervate the ventral flank mesenchyme and that at the base of the limb it is proximally dorsal (EphA4+). By contrast, BmprIa flox/mutant forelimb mesenchyme has normal dorsoventral polarity, consistent with previous reports (Additional file 2) [31]. Thus, relative to the dorsoventral projection choice, the dorsal transformation of ventral mesenchyme in BmprIa flox/hindlimbs is complete.

A dorsoventral axon branch point forms in BmprIa flox/mutant hindlimbs
We next examined the early axonal trajectories into the bidorsal BmprIa flox/mutant limbs in mice also carrying the Lim1 tlz allele, which directs tau-lacZ reporter expression only to axons of lateral LMC neurons [11]. At rostral axial levels of normal hindlimbs, neurofilament (NF) immunostaining, which labels all axons, marks a LacZ+ dorsal limb branch invading Lmx1b+ mesenchyme, a LacZ-ventral limb branch entering Lmx1b-mesenchyme and a LacZ-ventral flank branch (Figure 3a,b). Along the complete rostrocaudal extent of the BmprIa flox/hindlimbs, two LacZ+ branches invade Lmx1b+ limb mesenchyme, and a LacZ-branch projects ventrally, into the Lmx1b-flank (Figure 3c,d). Quantification of LacZ immunoreactivity in each branch supports these observations (Figure 3e). In BmprIa flox/forelimbs, there is no difference in the nerve trajectories compared to normal forelimbs (Additional file 3). Together, these data demonstrate that three branches form with normal timing in the bidorsal limbs, but that lateral LMC axons populate both limb branches equally, and are absent from the ventral flank branch. These data also provide evidence that lateral LMC axons perceive both halves of the BmprIa flox/mutant hindlimb as equivalently dorsal.

Medial LMC axons are redirected from ventral limb to ventral flank
To define directly the origin of the ventral flank motor branch in the BmprIa flox/mutant, we retrogradely labeled neurons projecting into the ventral flank. In both normal and BmprIa flox/mutant embryos, almost all HRP+ or RDA+ neurons express medial LMC markers (Figure 4ad), while the remainder expresses lateral LMC or medial MMC markers. A similar 95% of labeled neurons also express the additional LMC marker FoxP1 (Figure 4d,e). This result is consistent with the Lim1 tlz labeling experiments and provides evidence that in BmprIa flox/mutant embryos only medial LMC neurons innervate the ventral flank.
To address whether these medial LMC axons innervating the mutant flank include axons that normally innervate the limb, we further analyzed the distribution and marker expression of the ventral flank-labeled neurons. While retrogradely labeled neurons in control embryos are restricted to more rostral lumbar regions and span about 25% of the rostrocaudal length of the LMC, in BmprIa flox/mutant embryos they are distributed more broadly along approximately 50% of the length of the lumbar LMC (Figure 5). Scip, a POU domain transcription factor [43,44], marks caudal lumbar medial LMC neurons that in control embryos are frequently colabeled with HRP following ventral limb injections ( Figure 5) but infrequently (about 6% of labeled cells) following ventral flank injections ( Figure 5). However, about 24% of labeled motor neurons are also Scip+ following ventral flank injections in BmprIa flox/mutants ( Figure 5). The Scip+ population is equivalent in rostrocaudal extent and location in control and mutant embryos, as is the extent of the LMC itself, indicating that the differences are not due to an expansion of the Scip+ population, or gross changes in LMC organization ( Figure 5). Scip thus marks caudal lumbar medial LMC neurons that normally project to ventral limb mesenchyme but whose axons enter the ventral flank in BmprIa flox/mutant embryos. These results provide direct evidence that medial LMC axons are redirected from the ventral limb to the ventral flank when confronted with only dorsal limb mesenchyme.
While at hindlimb levels some medial LMC axons are rerouted to the ventral flank in BmprIa flox/mutants, other medial LMC axons might assume different trajectories, or perhaps stall at the base of the limb. NF immunostaining indicates that axons do not assume random trajectories in the peripheral non-limb mesenchyme. If axons stall, growth cones should accumulate in the hindlimb plexus. To address this possibility, we examined vesicular acetylcholine transporter (VAChT) expression [45], which preferentially labels distal regions of axons, shortly after the dorsoventral projection choice. We did not detect significant differences in VAChT staining between forelimbs or hindlimbs of normal and BmprIa flox/embryos (Additional file 4). These results provide evidence that axons extend through the hindlimb motor plexus of BmprIa flox/mutants.
To visualize the trajectories of medial LMC axons we anterogradely labeled medial LMC neurons by injecting HRP in the ventral medial spinal cord of E12.5 embryos. We only analyzed trajectories if an embryo had HRP+ medial LMC and not lateral LMC neurons (5 normal and 5 mutant embryos of 67 total embryos injected; Figure  6b). In normal embryos HRP labels primarily axons entering ventral limb and some entering ventral flank ( Figure  6c BmprIa flox/mutant embryos, and do not perceive ventral flank and dorsal limb mesenchyme as equally permissive.

Extra flank nerves are present in embryos with bidorsal limbs
To compare directly the trajectories of the lumbar motor axons in normal and BmprIa flox/mutant embryos, we crossed an Hb9-GFP transgenic allele [46] into these mice. The Hb9 promoter drives green fluorescent protein (GFP) expression in motor neurons, but not in sensory neurons, allowing direct visualization of the motor axons. We scored the distribution of GFP+ axons projecting along ventral flank, ventral limb and dorsal limb trajectories in sections of E11.5-E11.75 embryos (Figure 7e). Axons entering either limb branch spanned similar extents of the embryonic anteroposterior axis in control and mutant embryos, consistent with the observation that the anteroposterior extent of the mutant limbs is not smaller at the time of axon arrival (data not shown). In contrast, axons assuming flank trajectories spanned twice as much territory in mutant embryos compared to controls. Perhaps not surprisingly, the anteroposterior extent of both limb and flank-projecting axons we detected by direct visualization was greater than that detected by retrograde labeling ( Figure 5). This likely reflects the difficulty of quantitative retrograde labeling, and the conservative nature of the retrograde tracer injections necessary not to contaminate inappropriate nerve branches.
We also examined the trajectories of GFP+ axons that project along the lower abdominal wall in freshly dissected, intact E13.5 embryos (Figure 7). In control embryos (N = 8/8 embryos, 16/16 limbs) three nerves were detected projecting abdominally from the rostral lumbar spinal cord when viewed from a peritoneal aspect (Figure 7a). The two most rostral of these, the iliohypogastric and ilioinguinal nerves, project laterally from L1 and L2, and follow trajectories that parallel those of the more rostral subcostal and intercostal nerves. The third, the genitofemoral nerve, projects laterally upon leaving the rostral plexus in caudal L3 and curves anteriorly towards the ilioinguinal nerve. The ilioinguinal and genitofemoral trajectories together describe a roughly halfmoon pattern with the ilioinguinal transcribing the meridian and the genitofemoral the perimeter. The genitofemoral leaves the plexus coincident with the femoral and obturator nerves that extend distally into the limb. More caudally no lumbar nerves follow flank trajectories after leaving the sciatic plexus, although many nerves extend into the limb (Figure 7b).
In BmprIa flox/mutant embryos (N = 3 embryos, 6 limbs) there are differences from controls both in the distribution and number of flank projections. The genitofemoral nerve projects more directly laterally, which results in more even spacing among the three major flank nerves (Figure 7c). In addition, one to three excess nerves project from rostral lumbar segments along abdominal trajectories, although with variable projection patterns that differ even within a single embryo. These fibers are located between the ilioinguinal and genitofemoral nerves or caudal to the genitofemoral nerve. A substantial population of novel GFP+ axons also extends abdominally from the sciatic plexus (Figure 7d). Nerves projecting along the femoral and obturator trajectories are visible, although reduced in size, and sciatic projections into the limb are also present. These data provide clear evidence that the ventral flank axons normally derive from rostral lumbar regions, and that when confronted with only dorsal limb mesenchyme, motor axons originating along the length of the lumbar spinal cord aberrantly choose ventral flank trajectories.

Only lateral LMC axons enter the bidorsal BmprIa flox/hindlimb
The anterograde and Lim1 tlz labeling results together imply that lateral LMC neurons are the only motor neurons that innervate the bidorsal limb mesenchyme. To address this directly, we retrogradely labeled neurons innervating the BmprIa flox/mutant hindlimbs. HRP injections into dorsal hindlimb of either normal or BmprIa flox/mutant embryos at E13.5 label 95% to 97% lateral LMC neurons (Figure 8a,e). HRP injection into the ventral hindlimb of normal embryos labels 95% medial LMC neurons, as expected (Figure 8c,f). In contrast, ventral hindlimb injection of BmprIa flox/mutants labels lateral (94%), and not medial (6%), LMC neurons (Figure 8c,f). These data provide additional evidence that the motor components of both nerve branches that enter BmprIa flox/mutant hindlimb mesenchyme are composed solely of lateral LMC axons. Our anterograde labeling results (Figure 6) further support the idea that the failure to detect medial LMC axons by retrograde labeling from BmprIa flox/ mutant hindlimbs is not due to a selective inability of the medial LMC axons to take up or transport the label, but rather is due to their absence from the limb mesenchyme.

Altered LMC trajectories are due to changes in the BmprIa flox/hindlimb
Because the Brn4-cre transgene is expressed throughout the spinal cord, in addition to the limb ectoderm [31,35], we asked whether removing BmprIa from only the spinal cord might alter the LMC axonal trajectories. We first used molecular markers to label ventral progenitor or postmitotic motor neuron and interneuron populations at both brachial and lumbar levels of the spinal cord in BmprIa flox/ mutants. We observed no significant difference between BmprIa flox/mutant and normal sibling progenitor or motor neuron populations at either axial level (Additional file 5). We next examined LMC motor projection   in (a,c). (e,f) Quantification of (e) dorsal and (f) ventral retrograde labeling data. Lim1+ Isl1-HRP+ lateral LMC or Lim1-Isl1+ HRP+ medial LMC cells were counted in triply immunostained sections after retrograde labeling of control and BmprIa flox/hindlimbs, and are shown as percentages of total HRP+ cells. Dorsal limb HRP injection does not label significantly different populations in normal and mutant (normal: lateral LMC 97% ± 3%, n = 3 embryos, N > 150 HRP+ neurons; mutant: lateral LMC 95% ± 2%, n = 8 embryos, N > 350 HRP+ neurons; P = 0.6). Ventral limb HRP injection does label significantly different cell populations (normal: medial LMC 95% ± 1%, n = 6 embryos, N > 300 HRP+ neurons; mutant: medial LMC 6% ± 2%, n = 9 embryos, N > 400 HRP+ neurons; P < 0.00001). *P < 0.00001. patterns to BmprIa flox/forelimbs, which have normal dorsoventral patterning. Lim1 tlz lateral LMC marking and retrograde labeling analyses from dorsal and ventral forelimb mesenchyme all show that the columnar origins of these axons are normal (Additional files 3 and 6). Thus removing BmprIa from the spinal cord does not alter the LMC dorsoventral trajectory choice at forelimb levels.
To address directly whether removing BmprIa from lumbar LMC neurons alters their trajectories, we examined hindlimb LMC axonal projection patterns in Hb9 cre/+ , BmprIa flox/embryos. In these animals Cre is expressed in motor neurons and some ventral interneurons as they are becoming postmitotic, but is not expressed in the limbs (Additional file 7) [47]. It thus should remove BmprIa from most, if not all, LMC neurons prior to arrival of the axons at the base of the hindlimb. We found by retrograde labeling from dorsal or ventral hindlimb that there was no significant difference in the dorsoventral trajectory choice between Hb9 cre/+ , BmprIa flox/mutant and control embryos ( Figure 9). Moreover, in Hb9 cre/+ , BmprIa flox/embryos the organization of the ventral spinal cord and dorsoventral pattern of the limb mesenchyme are normal (Additional file 7 and data not shown).
Multiple lines of evidence therefore indicate that inactivating BmprIa selectively in motor neurons does not perturb the limb motor axon projection patterns. These data include the normal projection patterns to Brn4-cre Tg/-, BmprIa flox/mutant forelimbs and to Hb9 cre/+ , BmprIa flox/mutant hindlimbs, as well as the normal generation of LMC neurons on both genetic backgrounds. These results provide strong support for the idea that the hindlimb projection defects observed in the Brn4-cre Tg/-, BmprIa flox/mutant are caused by altered guidance cues in the limb mesenchyme.

Discussion
We investigated the relationship between the columnar identity of motor neurons, as defined by a transcription factor code, and the projection decisions made by their axons as they reach the base of the limb. We found that the ventral flank mesenchyme is innervated by neurons of the medial division of the lateral motor column, which were previously thought to project only into the limb mesenchyme. We also found that while axons of lateral LMC neurons enter the BmprIa flox/bidorsal limb mesenchyme and assume both dorsal and ventral trajectories, those of medial LMC neurons do not enter the limb mesenchyme, and are directed into the ventral flank mesenchyme (Figure 10a). These results support the idea that at hindlimb levels LMC axons project to both flank and limb tissues, and while doing so they normally choose between three, not two, mesenchymal trajectories: into ventral flank, into ventral limb, or into dorsal limb. Our data also provide evidence that the relative preference for each trajectory differs significantly between medial LMC axons and lateral LMC axons. We discuss these ideas in the context of previous studies investigating the logic that underlies these projection decisions.

Ventral flank mesenchyme is innervated by medial LMC neurons
We used multiple approaches to identify the motor neurons that extend axons into the ventral flank mesenchyme at hindlimb levels. All of them reveal that medial LMC neurons normally innervate both ventral limb and ventral flank mesenchyme. We therefore propose a change in the predictive value of the LIM homeodomain code regarding LMC motor axons [3,9]: Lim1-Isl1+ Lim3-medial LMC neurons should project to both ventral flank and ventral limb.
The axons that enter the hindlimb ventral flank are likely ultimately to form the motor components of the iliohypogastric, ilioinguinal and genitofemoral nerves [29,30]. In mammals these nerves arise from segments T12-L2 and they are the three most rostral nerves that project ventrally from the rostral hindlimb motor plexus. The motor components of the iliohypogastric and ilioinguinal nerves innervate lower abdominal muscles, while the sexually dimorphic genitofemoral nerve innervates the male cremaster muscle, which controls the position of the testes within the scrotum [48][49][50]. Previous experiments have not directly addressed the columnar origin of these nerves. However, retrograde labeling of the genitofemoral nerve from the cremaster muscle in rats marked a longitudinal distribution of somas in rostral lumbar regions [49], while ventral abdominal muscle labeling experiments consistently mark rostral lumbar motor pools [51][52][53]. Electrophysiological studies in humans demonstrate upper motor control of the genitofemoral nerve, and some degree of voluntary control [54]. These results are consistent with our identification of the ventral flank-projecting axons as somatic motor axons projecting from the rostral portion of the medial LMC.

Relationship of medial LMC and lateral MMC neurons
Intriguingly, our observations support the idea that the trajectory preferences of lumbar medial LMC axons and thoracic lateral MMC axons are similarly regulated [9]. These neurons have the same LIM homeodomain transcription factor code and similar mediolateral somatic settling positions, and the ventral flank trajectory we documented for some medial LMC axons parallels that of lateral MMC axons. These axons make a series of similar projection decisions, ignoring both the dorsal trajectory of the medial MMC and the ventromedial trajectory of the autonomic motor neurons, before assuming a distal trajectory into ventral lateral plate mesodermal derivatives The motor innervation of Hb9 cre/+ , BmprIa flox/hindlimbs is normal Figure 9 The motor innervation of Hb9 cre/+ , BmprIa flox/hindlimbs is normal. Retrograde labeling was used to determine the columnar origin of axons innervating Hb9 cre/+ , BmprIa flox/hindlimbs.   9,13]. Our results also suggest that lateral MMC axons might perceive ventral flank and limb mesenchyme as similarly permissive, which might explain why they readily innervate ectopic thoracic limbs [55] or normal limbs when the axons are caudally displaced [2,4]. However, these populations are not a simple continuum, as LMC, but not lateral MMC, neurons express FoxP1 and Raldh2, for example [32,56]. It will be interesting to learn more about what these populations have in common, and how they diverged.

LMC axons make a three-way projection decision at the base of the limb
Our data reveal that a trajectory into ventral flank mesenchyme is an option for LMC axons, which raises the question of why its significance was not previously appreciated. One likely reason is that relatively few motor axons normally enter the ventral flank mesenchyme, and thus they have not been studied as extensively as the axons that innervate the limb. Furthermore, previous experiments investigating initial limb innervation decisions always found that LMC axons entered the limb mesenchyme [2,11,12,15,16,[19][20][21][22][23]. In addition, when motor innervation patterns were examined in apparently bidorsal chick hindlimbs generated surgically, they were found to be normal [17], in contrast to the situation in the bidorsal BmprIa flox/mutant hindlimbs. This might reflect a bona fide difference in the behavior of chick and mouse motor axons, although more likely small amounts of ventral limb mesenchyme remained near the dorsoventral choice point following the surgical manipulations, leading to a normal projection choice [37]. Together, these data led to the idea that at the base of the limb LMC axons make a binary choice between either dorsal or ventral limb trajectories [24,25]. Nonetheless, our data reveal that the motor axons execute a ternary, not a binary, choice of trajectories at hindlimb axial levels.
Whether lateral LMC axons enter ventral flank mesenchyme in existing mutants is not clear. We did not find lateral LMC axons making this choice when the hindlimbs are bidorsal. Similarly, no lateral LMC axons were described entering the ventral flank in Lmx1b -/mice, which have biventral hindlimbs [11], or in EphA4 -/embryos, in which apparently all LMC axons are directed to ventral limb mesenchyme [21]. However, lateral LMC axons assuming this trajectory might have gone unnoticed, as the ventral flank nerve branch was not directly examined in the Lmx1b -/and EphA4 -/mutants. One interesting question is why LacZ+ lateral LMC axons are not substantially redirected into the ventral flank mesenchyme of Lim1 tlz/+ , Lmx1b -/embryos, as they are repulsed by ephrin-A ligands that are uniformly high across the hindlimb of this mutant [11,12]. Perhaps attractive cues in limb mesenchyme such as glial-derived neurotrophic factor (GDNF) [20] can overcome these repellant signals, or an additional repulsive cue in the ventral flank makes this territory even less permissive than ventral limb mesenchyme to lateral LMC axons. Regardless, it appears that axon entry into limb mesenchyme is not a default trajectory for LMC axons, but rather involves an active choice.
The comparative behavior of medial and lateral LMC axons is informative about the regulatory logic guiding their trajectory choices. One might anticipate that peripheral guidance cues influence medial and lateral LMC axons in a similar but opposite fashion. If this were so, since lateral LMC axons populate dorsal and ventral nerve branches in equal proportion in Lmx1b -/mutants with biventral hindlimbs [11], then medial LMC axons should project equally to dorsal and ventral limb in BmprIa flox/mutants with bidorsal hindlimbs. However, medial LMC axons instead project to the ventral flank, indicating that medial and lateral LMC axonal trajectories are guided asymmetrically.
We can order the relative preference of the three trajectories from the perspective of axons in either LMC division (Figure 10b). From the perspective of medial LMC axons, the normal preference for most is to innervate ventral limb mesenchyme, with the remainder innervating the ventral flank. In the absence of ventral limb mesenchyme, ventral flank mesenchyme is strongly preferred over dorsal limb. From the perspective of lateral LMC axons, the normal preference is to innervate dorsal limb mesenchyme, and in its absence, to innervate ventral limb mesenchyme. The least favored trajectory appears to be into the ventral flank. Thus, the relative trajectory preferences for LMC axons are neither identical nor mirror images, and appear for medial LMC as: ventral limb > ventral flank > dorsal limb, and for lateral LMC as: dorsal limb > ventral limb > ventral flank.
The trajectory assumed by LMC axons might follow from one ternary choice between three options, or from sequential binary decisions. All three hindlimb LMC trajectories diverge at roughly the same proximodistal position within the hindlimb plexus ( Figure 2). This contrasts with the trajectory assumed by some medial LMC axons that extend from the brachial plexus into forelimb level ventral flank mesenchyme, and that ultimately innervate the latissimus dorsi and cutaneus maximus muscles [57,58]. The forelimb ventral flank branch clearly diverges from the ventral limb branch after the axons of both branches have entered the ventral limb mesenchyme (for example, Additional file 3). This morphology implies that the decision to assume the forelimb flank trajectory is taken after the decision to enter the ventral limb mesenchyme. In contrast, the morphology of the hindlimb plexus favors a single ternary choice.

A model for the three-way choice
The existence of both attractive and repulsive cues that guide lateral LMC axons provides a framework for thinking about how medial LMC axons are directed to ventral limb, and how a subset of them is guided to the ventral flank. In one model (Figure 10c), attractive guidance cues are present throughout the limb mesenchyme, but not in the ventral flank, and a step gradient of a repulsive cue is also present. The dorsal limb mesenchyme expresses the repellant at the highest level, the ventral limb mesenchyme at an intermediate level and the ventral flank mesenchyme at the lowest level. All medial LMC neurons express receptors for these cues, with a subset expressing the repellant receptor at higher levels. Upon arrival at the base of the limb, the growth cones integrate the attractive and repulsive cues and assume the most permissive trajectory. So long as the repellant activity in the dorsal limb mesenchyme dominates, all medial LMC axons are directed ventrally. In contrast, the moderate repellant level in ventral limb mesenchyme is permissive to most medial LMC axons, although the subset of axons with the highest level of the repellant receptor is directed more ventrally, into the flank mesenchyme. This model predicts that the limb dorsoventral patterning system controls repulsive cue levels, and thus in a completely dorsalized limb, such as in the BmprIa mutant, the entire limb mesenchyme expresses high levels of repulsive cue, and directs all medial LMC axons to ventral flank tissue.
If a subset of medial LMC neurons expresses more repulsive receptor, this leads to the question of how this subset is specified. The neurons that normally contribute to the ventral flank nerves are located in rostral segments of the lumbar spinal cord. Higher receptor expression levels therefore might be established by the system that patterns the anterior-posterior axis of the spinal cord [59], perhaps influenced by differential expression of pool-specific Hox genes [44] that act prior to overt segregation of the cell bodies into motor pools. Interestingly, the distribution of peripheral guidance cues might be established via another axial patterning system, since the dorsal limb, ventral limb and ventral flank mesenchyme are mediolaterallyarrayed derivatives of the lateral plate mesoderm [60]. While this model can account for the observed behavior of the medial LMC axons, others are equally plausible. Whatever the mechanism, it must make all medial LMC axons competent to assume ventral flank trajectories, while compelling only a few to do so.

Conclusion
We found that neurons of the LMC, long thought to innervate only limb tissues, actually innervate tissues beyond the limb, such as the ventral flank. To select appropriate trajectories, LMC axons execute a three-way choice between ventral flank, ventral limb and dorsal limb tar-gets. These findings provide new insights into the cellular logic used in guiding trajectory decisions, uncover a novel function for the LMC and illuminate the relationship between LMC axons and those of other motor columns.
Noon of the day a mating plug was observed was designated embryonic day 0.5 (E0.5). Stage matching was based on limb bud developmental stage [63], since limb stage and nerve invasion are correlated [4,64].
For anterograde labeling, explanted E12.5 embryos were dorsally laminectomized, and a pulled glass needle was introduced from a dorsomedial aspect towards the LMC. Approximately 10-40 nl (empirically determined to label specifically the most medial LMC neurons) of HRP or RDA was injected into every second lumbar spinal cord segment. Hb9-GFP Tg transgene [46] GFP fluorescence at E12.5 is weaker in medial than in lateral LMC neurons, and some injections were guided under UV illumination to this GFP low region. Labeling accuracy was determined in cryosections co-immunostained for HRP or RDA and NF or LMC division markers. Of 67 injected embryos, 10 (5 normal, 5 mutant) were further analyzed, because the vast majority of HRP+ cells were medial LMC neurons. In some embryos a few labeled axons appeared to be from sensory or medial MMC neurons.

Quantification
Neurons retrogradely labeled by HRP or RDA and colabeled with LMC markers were counted [33]. Four or more sections from three or more embryos encompassing at least 150, and typically more than 250, retrogradely labeled cells were counted for each condition and marker combination. Statistical comparisons were performed using a two-tailed heteroscedastic Student's t-test, with a significance limit of P < 0.05. Values are reported as mean ± standard error of the mean (SEM).

Imaging and morphometric analysis
Expression patterns were compared using multicolor immunofluorescence imaging or overlaying of RNA in situ hybridization or EphA4-AP staining patterns from adjacent cryosections [12]. For Lim1 tlz nerves, immunofluorescence intensity of NF+ and LacZ+ nerve branches were compared using the ImageJ program as described [11,23]. The rostrocaudal extent of motor columns, pools and nerve branches was determined by scoring alternate transverse sections for marker presence. The length of the region was then calculated by multiplying the number of sections between the first and last positive section by the section thickness. Typically >80% of internal sections were positive for a given nerve branch.

Additional file 2
Additional file 3 Additional file 4