|Home | About | Journals | Submit | Contact Us | Français|
Rodents begin to use bilaterally coordinated, rhythmic sweeping of their vibrissae (“whisking”) for environmental exploration around two weeks after birth. Whether and how vibrissal control circuitry changes after birth is unknown, and relevant premotor circuitry remains poorly characterized. Using a modified rabies virus transsynaptic tracing strategy, we labeled neurons synapsing directly onto vibrissa facial motor neurons (vFMNs). Sources of potential excitatory, inhibitory, and modulatory vFMN premotor neurons, and differences between the premotor circuitry for vFMNs innervating intrinsic versus extrinsic vibrissal muscles, were systematically characterized. The emergence of whisking is accompanied by the addition of “new” sets of bilateral excitatory inputs to vFMNs from neurons in the lateral paragigantocellularis (LPGi). Furthermore, descending axons from the motor cortex directly innervate LPGi premotor neurons. Thus, neural modules well suited to facilitate the bilateral coordination and cortical control of whisking are added to premotor circuitry in parallel with the emergence of this exploratory behavior.
Human infants acquire and refine controlled movements gradually, perhaps most conspicuously as they gain the ability to walk (Dominici et al., 2011; Gerber et al., 2010). An emerging idea is that these behavioral changes reflect the addition of new inputs from premotor control modules to lower level motor control circuits, including motor neurons (Dominici et al., 2011). However, the identity of these added inputs are unknown and it is currently impractical to map them with high resolution in human subjects.
Rodents also acquire and refine movements during postnatal development and they are amenable to high-resolution neural circuit tracing methods. One well-known example of a developmentally emergent behavior is exploratory whisking: the rhythmic sweeping movements of vibrissae (or whiskers) that rodents used to detect the texture, shape, and location of objects in their environment (Diamond et al., 2008; Kleinfeld et al., 2006). Whisking is usually bilaterally coordinated (Mitchinson et al., 2007; Sellien et al., 2005). Notably, although neonatal rodents can unilaterally retract their vibrissae as early as P4, they only begin to exhibit bilateral whisking between P11 – P14 (Grant et al., 2012; Landers and Philip Zeigler, 2006; Welker, 1964; and our own observations in mouse), raising the possibility that this behavioral progression depends on changes to the central circuits that control vibrissal movements.
The final neural control of whisking is mediated by motor neurons in the lateral facial nucleus (FN), and these vibrissal facial motor neurons (vFMNs) form synapses with the intrinsic muscles surrounding the vibrissae or with the extrinsic muscles that extend throughout the mystacial pad (Ashwell and Watson, 1983; Dorfl, 1982; Hill et al., 2008; Wineski, 1985). Because vFMNs form neuromuscular junctions at late embryonic stages and unilateral vibrissal movements are evident by P4 (Ashwell and Watson, 1983; Grant et al., 2012), the emergence of bilaterally coordinated whisking presumably involves the addition of new control inputs to vFMNs as the animals mature. Furthermore, because whisking patterns and dynamics are regulated by motor cortex (Berg et al., 2005; Berg and Kleinfeld, 2003; Brecht et al., 2004; Carvell et al., 1996; Donoghue and Parham, 1983; Hill et al., 2011; Li and Waters, 1991; Matyas et al., 2010), and both task-dependent whisking and vibrissae-dependent behaviors require the motor cortex (Huber et al., 2012), these added inputs must somehow link cortical signals to the vFMNs. Prior studies have mapped neurons that project axons to or through the lateral FN in adult rodents using conventional retrograde tracing techniques (Hattox et al., 2002; Isokawa-Akesson and Komisaruk, 1987), but these techniques could not selectively target neurons that synapse on vFMNs, nor could they reliably reveal how premotor inputs change during the critical early postnatal window when whisking first emerges.
To begin to address whether and how vibrissal control circuitry changes over early postnatal life in the mouse, we employed a monosynaptic rabies virus based technique (Arenkiel and Ehlers, 2009; Callaway, 2008; Wickersham et al., 2007b) adapted to selectively trace neurons that directly synapse on the vFMNs (hereafter referred to collectively as the vFMN premotor circuitry). We mapped and compared the vFMN premotor circuitry prior to and immediately after the onset of whisking, and discovered that new modules are added to this circuitry in parallel with the emergence of the whisking behavior.
We sought to exploit monosynaptic viral tracing methods to selectively label those neurons that make synapses onto vFMNs. Previous efforts led to the development of a glycoprotein-G deleted mutant rabies virus (ΔG-RV) that cannot be transported into presynaptic neurons unless the infected source cell expresses rabies glycoprotein-G (rabies-G) to complement the mutant virus (Wickersham et al., 2007b). Once inside presynaptic neurons, which do not express the rabies-G, the deficient virus ceases to spread further, thereby only tracing neurons that are monosynaptically connected to the source cell (Wickersham et al., 2007b). To take advantage of this method for our study, we first generated a knock-in mouse line at the ROSA26 locus in which rabies-G is conditionally expressed in a Cre-dependent manner. Briefly, a ubiquitous CAG promoter followed by a “loxp-neo-STOP-loxp” cassette and a cDNA encoding “rabies-G-IRES-TVA” (i.e CAG-loxp-STOP-loxp-rabies-G-IRES-TVA) was inserted into the intron of the ROSA26 locus (Figure 1A; for the purpose of this study, TVA expression is irrelevant and thus will not be further mentioned). We refer to this new line as RΦGT mouse, where Φ stands for the loxp-neo-STOP-loxp cassette.
We then confirmed that the RΦGT mouse does not facilitate “leaky” spreading of ΔG-RV particles in the absence of Cre transgene (Figures S2A–S2D). Specifically, injecting ΔG-RV-GFP (abbreviated as ΔG-GFP) into the mystacial pad in neonatal RΦGT mice resulted in infection and labeling of vFMNs, as well as some trigeminal sensory neurons which innvervate vibrissa follicles and project axons into brainstem, but did not label other cells in the CNS (Figures S2A–S2D). These experiments confirmed that, in the absence of Cre, this approach only labels neurons that have axon terminals in the mystacial pad. We then crossed RΦGT mice with a transgenic mouse line that expresses Cre under the control of choline-acetyl-transferase gene (Chat::Cre), allowing rabies-G to be expressed in all motor neurons, which are cholinergic and consequently can complement the deficient virus (Figure 1A). This double transgenic mouse allows for the selective labeling of vFMNs and their presynaptic partners following an injection of ΔG-RV into the mystacial pad (Figure 1A).
Using this approach, we first examined the vFMN premotor circuitry in neonatal Chat::Cre; RΦGT mice prior to the onset of whisking (<P11). Injection of ΔG-GFP into regions adjacent to the left B2 and C2 vibrissae in P1 pups (Figure 1B) led to the infection of the intrinsic muscles that surround the B1–2 and C1–3 vibrissae as well as small numbers of extrinsic muscle fibers between row B and row C (Figure 1B and data not shown). One week later (i.e., at P8), we sectioned and imaged each brain. As expected, labeled vFMNs were located in the lateral part of the FN (Figure 1D). These injections also resulted in a wide distribution of labeled cells spanning from the caudal medulla to the rostral brainstem (Movie S1 scans through the entire set of serial coronal sections from a representative brain after P1→P8 transsynaptic tracing, giving an overview of the wide distributions of neurons presynaptic to vFMNs). A more detailed description follows below and in Table S1.
We consistently observed labeled premotor neurons either within or adjacent to the three ipsilateral brainstem trigeminal nuclei that process sensory-related information from the vibrissae. In the most caudal part of the medulla, ΔG-GFP tracing labeled a group of neurons located in the dorsal medullary reticular formation (MdD) adjacent to the spinal caudalis (SpC) that extended radially oriented dendrites into the SpC (Figure 1E and Figure S1B). Within the rostral part of spinal interpolaris (SpIr), a group of neurons with large cell bodies and circumferentially oriented dendrites were labeled (Figures 1I–1J and Figure S1A). Furthermore, a cluster of smaller neurons adjacent to the spinal oralis (SpO) was traced by ΔG-GFP (Figure 1L and Figures S1A–S1B). Finally, GFP-labeled neurons were also found in the ipsilateral spinal vestibular nucleus (Ve, Figure 1I), which presumably relay information about head orientation onto vFMNs.
GFP-labeled neurons were found in the intermediate reticular nuclei (IRt) and the gigantocellular reticular nuclei (Gi) both ipsilaterally and contralaterally, and in the midline raphe nuclei (Figures 1F–1H). In the IRt, numerous GFP-labeled neurons were located immediately adjacent to the nucleus ambiguus (NA) (arrow in Figure 1F, arrowheads in Figure 1G and Figures S21I–S2L). Notably, extensive labeling was found in the pre-Bötzinger (preBötC) and Bötzinger complex (BötC) (Figures 1I, 1K and Figures S1C–S1D), two regions that contain respiratory central pattern generator (CPG) neurons (Ezure, 1990; Smith et al., 2009; Smith et al., 1991). Although some respiratory CPG neurons are known to express somatostatin (Sst) and NK1R (Gray et al., 2001; Tan et al., 2008), most of the GFP-labeled premotor neurons in the preBötC area did not express Sst or NK1R (5% Sst+, 8% NK1R+, Figures 2K, 2L and Table1).
Scattered GFP-labeled cells were detected in the ventral pontine reticular formation (PnV) (Movie S1). Dorsal to the principal trigeminal nucleus (PrV), we observed numerous GFP-labeled neurons in the Kölliker-Fuse (KF, Figure 1M), a region that is also known to regulate respiratory rhythms (Dutschmann and Herbert, 2006; Dutschmann et al., 2004). Finally, in the midbrain, GFP-labeled neurons were found in deeper layers of the superior colliculus (SC) and in mesencephalic reticular nucleus (MRN) (Figure 1N).
The numbers of GFP-labeled neurons in every structure containing vibrissa premotor neurons throughout the brainstem from 4 different animals are listed in Table S1. To measure the correlations of the transsynaptic tracing results among different animals, serial sections collected from each experimental brain were aligned, and each labeled neuron’s spatial coordinates (see Experimental Procedures) were extracted to generate a 3D reconstructed model of the transsynaptically traced circuitry for each animal (Figure 1O and Movie S2). Pair-wise cross-correlation comparison (see Experimental Procedures) showed that the spatial distributions of the labeled premotor neurons were highly similar among the different animals (pair-wise correlation coefficiency r = 0.75~0.88, Figure 1P).
The consistent anatomical locations of GFP-labeled vFMN premotor neurons across different animals suggested to us that rabies mediated monosynaptic circuit tracing is reproducible and therefore potentially suitable to discover whether connectivity patterns change over development. We conducted two additional experiments to better confirm this impression.
Because the Chat::Cre driver results in expression of rabies-G also in all neurons that transiently or persistently express Chat (i.e., all central cholinergic neurons), it is important to determine that vFMNs were the only ΔG-GFP labeled neurons derived from a Chat::Cre lineage in our experiments. We therefore performed ΔG-GFP transsynaptic tracing in triple transgenic “Chat::Cre; RΦGT; Rosa Φ tomato” mice. In these mice, Rosa Φ tomato (Arenkiel et al., 2011) serves as red-fluorescence reporter for all cells that transiently or persistently express Chat::Cre activity. Importantly, no “premotor” neurons were double-labeled with GFP and tomato, even though subsets of GFP-labeled neurons were located in close proximity to tomato-positive cells in many regions (Figures S2I–S2P). These results indicate that none of the premotor neurons we observed following ΔG-GFP injections in the mystacial pad were derived from a cholinergic lineage, and confirmed that no secondary spreading into “pre-premotor” neurons occurred under our experimental conditions.
Because the deficient rabies virus is an excellent retrograde tracer that can infect wild-type neurons from their axons but cannot subsequently “jump” synapses without complementation (Etessami et al., 2000; Wickersham et al., 2007a), we also used ΔG-GFP as a “conventional” retrograde tracer to label inputs to the FN in wild-type neonatal mice. Injections of ΔG-GFP were made directly into the left lateral FN in wild-type pups at P6 (n = 4) and the distributions of labeled cells were analyzed at P9. In principle, such direct injections should label not only those neurons that synapse with vFMNs but also those with axons that extend through the injection site. Indeed, results from the direct retrograde tracing encompassed all locations uncovered by the transsynaptic tracing (Figure S3), and also resulted in GFP-labeling at additional locations.
These additional loci included the parvocellular reticular nucleus (PCRt), solitary nucleus (Sol), peri-aqueductal gray (PAG), and regions adjacent to the trigeminal motor nucleus (Figures S3B, S3D and S3F). There were also significantly more labeling throughout the IRt and Gi, and midline raphe nucleus (Figures S3B–S3D). The numbers of all labeled neurons in all locations from 4 independently injected animals were counted and listed in Table S1.
To identify the potential functional nature of inputs onto vFMNs from different sources, we used in situ hybridization to examine the expression of molecular markers for glutamatergic (vGlut2), GABAergic (Gad1 and Gad2), glycinergic (Glyt2) or serotonergic (Tph2) transmission in GFP-labeled premotor neurons. Each of the 4 probes was hybridized to the entire serial sections of 3 different ΔG-GFP traced brains (n = 3 animals per probe, P1→P8 transsyanptic tracing from the mystacial pad). Figures 2A–2J show the representative images of one selected in situ probe together with GFP expression in each of the nuclei traced by ΔG-GFP (Results of the other 3 probes for each region are shown in Figures S4A–S4O). Detailed quantitative results of the average percentage of marker-expressing cells among GFP-labeled neurons for each premotor structure are summarized in Table 1. Briefly, among the three pools of sensory-related premotor neurons, those located in the SpO and SpIr are mostly glutamatergic (vGlut2+), whereas those located in the MdD near SpC are primarily glycinergic (GlyT2+), many of which also likely co-express GABAergic markers (Gad1/2+). Premotor neurons in the various reticular nuclei (IRt, Gi, preBötC and BötC) and in Ve are mixtures of vGlut2+, GlyT2+ and GAD1/2+ cells. Additionally, most (>70%) GFP-labeled neurons in midline raphe nuclei were labeled with markers for GABAergic transmission (Gad1/2+), whereas the remainder (28%) expressed serotonergic (Tph2+) markers. Finally, GFP-labeled cells rostral to the FN from pons and midbrain, including those located in KF, SC and MRN, primarily expressed glutamatergic markers (vGlut2+) (Table 1).
We further explored whether we could use this modified rabies transsynaptic tracing method to identify differences in the premotor circuitry associated with vibrissa retraction versus protraction in neonatal mice. Those vFMNs that innervate the intrinsic muscles (vFMNs-in) are responsible for vibrissa protraction and project to the mystacial pad mostly through the buccal branch of the facial nerve, whereas those vFMNs that innervate the caudal extrinsic muscle nasolabialis (vFMNs-ex) are partly responsible for vibrissa retraction and course through the zygomatic facial nerve (Dorfl, 1985; Hill et al., 2008; Klein and Rhoades, 1985) (Figure 3A). Therefore, cutting the zygomatic nerve prior to ΔG-GFP injection into mystacial pad should enable transsynaptic tracing primarily from vFMNs-in (Figure 3A, left). In contrast, injecting ΔG-GFP into the caudal nasolabialis muscle should selectively label subsets of vFMNs-ex and their associated inputs in Chat::Cre; RΦGT pups (P1→P8 tracing)(Figure 3A, right).
Examining the FN confirmed the relatively selective targeting of different vFMN subpopulations with such manipulations. GFP-labeled neurons following viral injections into the caudal nasolabialis muscle were found in the dorsolateral part of the FN (Figure 3C), whereas GFP-labeled neurons following injection near vibrissae after zygmoatic nerve transections were located in the lateral and ventrolateral part of the FN (Figure 3B). These localizations are in close agreement with previous studies showing the relative positions of the vFMNs-ex and vFMNs-in (Klein and Rhoades, 1985; Semba and Egger, 1986). The number of labeled vFMNs-ex (average 85 per animal) was about half that of labeled vFMNs-in (average 172 per animal). A comparison of the viral tracing results for vFMNs-in (n = 4) and vFMNs-ex (n = 4) revealed that the overall spatial distributions of premotor neurons for both groups of vFMNs were very similar (Figure 3).
Notably, three differences in the pattern of premotor pools to the two vFMN-subpopulations were observed. First, significantly fewer neurons were labeled in the IRt when traced from vFMNs-ex (arrows in Figures 3F and 3G, when the number of labeled cells was divided by number of infected vFMNs, the results were 0.65 ± 0.08 labeled IRt cells per vFMN-in, versus 0.34 ± 0.11 per vFMN-ex, p < 0.03). Second, markedly more neurons were labeled in the SpIr when traced from vFMNs-ex (Figures 3H–3K, 0.14 ± 0.08 SpIr neurons per vFMN-in, versus 1.79 ± 0.21 SpIr neurons per vFMN-ex, p < 0.01). Third, we did not observe any GFP-labeled neurons in the Ve when tracing from intrinsic muscles, while they were consistently labeled from extrinsic muscles (arrow, Figures 3H and 3I). This latter finding suggests that vestibular derived sensory information exclusively feeds onto vFMNs-ex that control vibrissa retraction. This is consistent with the observation that horizontal rotations of the head are accompanied by ipsilateral vibrissa retraction in the direction of head turning, and the finding that rodents are capable of such unilateral vibrissal movements as early as P4 (Grant et al., 2012; Towal and Hartmann, 2006).
To map the vFMN premotor circuitry near the time in development when whisking is first observed (P11 – P14) (Grant et al., 2012; Landers and Philip Zeigler, 2006; Welker, 1964; and our observations in mouse), we injected ΔG-GFP at P8 in Chat::Cre; RΦGT mice and analyzed the resulting distribution of GFP-labeled cells at P15 (Figure 4A). Qualitatively, the infection efficiency of vFMNs from mystacial pad injection at P8 was lower than that seen with similar injections made at P1, and most of the infected vFMNs from such mystacial pad injection were vFMNs innervating intrinsic muscles (i.e., vFMNs-in), as judged by their positions in the ventrolateral FN (Figure 4F). Further, the vFMNs were highly susceptible to the virus-induced pathology at this stage, as evident by the presumed clearance of dead vFMNs by infiltrating glial cells at 7 days post infection (Figure 4F, stars). Nonetheless, the spatial distributions of labeled premotor neurons were still highly correlated among different animals at this stage (correlation coeffeciency r = 0.71~0.85, Figures 4H–4I). Movie S3 shows an entire series of sections through the brainstem from such P8→P15 tracing, Movie S4 is the 3D model showing distribution of pre-motor neurons, and Table S2 listed the numbers of labeled neurons in different brainstem structures from 4 different animals.
Given the reduced efficacy of GFP-labeling following ΔG-GFP injections made at P8, we focused our comparison with P1→P8 results on the “locations” of GFP-labeled premotor neurons rather than on their absolute numbers. Most P8→P15 labeled premotor neurons were found in the same nuclei as those observed before the onset of whisking (Compare Movies S1, S2 with Movie S3, S4). For example, at both ages we found GFP-labeled neurons in the MdD (Movie S3), IRt and Gi (Figures S5A-S5B), preBötC, BötC and KF (Figures 4B–4D), SpO (Figure S5C), ventral pons (Figure S5D), SC and MRN (Movies S3 and S4). Furthermore, following virus injection selectively into the extrinsic nasolabialis muscles, we also found labeled cells in the SpIr and Ve, as observed following injections made earlier in development (Figures S5E–S5F).
Notably, P8→P15 transynaptic tracing with unilateral injections of ΔG-GFP in the mystacial pad labeled a “new” group of neurons in the rostral part of the lateral paragigantocellularis (LPGi) both ipsilateral and contralateral to the injection site (arrows in Figure 4F, high mag shown in 4G, also see Movie S3). This group of neurons was never labeled in P1→P8 transsynaptic tracing (Figure 4E), despite the fact that P1-injected viruses infected significantly more vFMNs and premotor neurons than did P8-injected viruses. Moreover, injecting ΔG-RV into the extrinsic nabialis muscle in P8 mice after transecting the buccal nerve to prevent infection of vFMNs-in also labeled neurons in LPGi, although these labeled cells localized to a slightly more caudal position (Figures 4J–4L). Thus both vFMNs-in and vFMNs-ex receive “new” sets of synaptic inputs from LPGi neurons around the time when the whisking behavior first emerges.
To further confirm the P8→P15 transsynaptic tracing results, we performed direct injection of ΔG-GFP virus into the lateral FN at P12 and analyzed the brains at P15. Results from the direct retrograde tracing, although highly variable, included all anatomical loci identified by transsynaptic tracing (Figure S6 and Table S2). In addition, direct injection of virus into FN also labeled neurons in the PCRt, Sol, PAG, and in regions adjacent to the trigeminal motor nucleus (Figures S6B–S6D, S6F and Table S2); all of these additional loci were similarly labeled in the P6→P9 direct FN infection experiments (Figure S3 and Table S1).
What might be the synaptic function of these “newly-added” LPGi premotor neurons on vFMNs? Using two-color in situ-immuno co-localization methods, we found that GFP-labeled LPGi neurons express vGlut2 (Figure 5A), but not Gad1 (a GABAergic marker) (Figure 5B), Tph2 (a marker for serotonergic neurons) (Figure 5C), or TH (tyrosine hydroxylase, Figure 5D), and therefore likely form glutamatergic synapses onto vFMNs. Because unilateral virus injections labeled neurons in both left and the right LPGi (Figure 4F, 4J), we wondered whether individual LPGi neurons project bilaterally to vFMNs. Following injections of ΔG-GFP and ΔG-mCherry into the left and right mystacial pads, respectively, we found that many LPGi neurons were double-labeled, indicating that individual LPGi neurons innervate vFMNs bilaterally (Figure 5E). Importantly, no other double-labeled cells were found in our dual-color tracing experiments, strongly suggesting that LPGi neurons are likely to play an especially important role in coordinating vFMN activity bilaterally.
We used optogenetic-assisted electrophysiological recording methods (Petreanu et al., 2007; Petreanu et al., 2009) in brainstem slices to further characterize the functional nature of the synapses that LPGi axons make on vFMNs. To this end, we injected deficient rabies expressing channelrhodopsin and mCherry (ΔG-ChR2-mcherry) (Osakada et al., 2011) into the left lateral FN to retrogradely infect LPGi premotor neurons on both sides of the brainstem (Figure 5F–5G). Because many LPGi neurons project bilaterally, ΔG-ChR2 labeled LPGi neurons also send axons to the FN contralateral to the injection site. Thus, focal illumination of the contralateral FN with blue light should activate ChR2 on the axon terminals of LPGi neurons and trigger neurotransmitter release (Figure 5G). Using this strategy, we performed whole-cell voltage-clamp (VH = −75mV) recordings from randomly selected “putative” vFMNs (i.e., neurons located in the lateral-most part of the FN) contralateral to the injection site (Figure 5G). Brief (10ms) illumination in the presence of TTX and 4-AP evoked large, fast inward synaptic currents in half (5/10) of the recorded vFMNs (Figures 5H–5I). Since ΔG-ChR2 likely only infected subsets of LPGi neurons (that innervate subsets of vFMNs), we did not expect all randomly recorded vFMNs to respond to photo-stimulation. These findings indicate that LPGi neurons provide robust excitatory synaptic inputs onto vFMNs, which along with their bilateral projection patterns could enable them to drive bilaterally synchronized whisking.
It is known that exploratory whisking movements can be elicited and/or are modulated by cortical activity (Berg et al., 2005; Berg and Kleinfeld, 2003; Brecht et al., 2004; Carvell et al., 1996; Donoghue and Parham, 1983; Hill et al., 2011; Li and Waters, 1991; Matyas et al., 2010). Recent study in mouse further showed that task-related whisking behavior depends on motor cortex (Huber et al., 2012). However, vFMNs themselves receive only sparse synaptic inputs from the motor cortical axons (Grinevich et al., 2005). Therefore, we wondered whether transsynaptically labeled LPGi neurons are direct targets of cortical neurons, an arrangement that could enable descending signals from the motor cortex to affect vFMN activity and thus initiate and modulate the whisking behavior. To test this idea, we injected AAV-expressing GFP (AAV-GFP) into the right vibrissa motor cortex (M1), and then performed transsynaptic tracing by injecting ΔG-mCherry into left B2/C2 vibrissae at P8 (Figure 6A). We found GFP-labeled collaterals of motor cortical axons traversed through regions where mCherry-labeled LPGi premotor neurons were located (Figures 6B–6D). We used anti-vGlut1 to stain presynaptic terminals and found that motor cortical axons appeared to make numerous synapses onto LPGi premotor neurons (Figures 6D–6G). Similarly, AAV-GFP viruses injected into the primary somatosensory cortex (S1) (Figure 6A) also labeled axon collaterals coursing through the LPGi region (Figure 6H). However, S1 axons appeared to form sparser vGlut1-positive terminals on LPGi neurons (Figures 6I–6J). We counted the numbers of putative motor or somatosensory cortical synapses (i.e. vGlut1-positive boutons) onto every transsynaptically labeled LPGi neurons (N = 3 different animals for each case). On average, M1 axons formed twice as many vGlut1+ boutons onto LPGi neurons than did S1 axons (Figure 6K).
To ascertain the functions of these putative cortical synapses on LPGi neurons, we obtained whole-cell voltage-clamp recordings from ΔG-mCherry transsynaptically labeled LPGi neurons while activating motor cortical inputs using optogenetic methods (Petreanu et al., 2007; Petreanu et al., 2009). First, AAV-Channelrhodopsin (AAV-ChR2) was injected into M1 at P2 (Figure 6L), ΔG-mCherry was then injected into the vibrissae at P8, and coronal brainstem slices were prepared at P14 (Figures 6L–6M). Focal illumination of ChR2-expressing cortical axons in LPGi in the presence of TTX and 4-AP evoked fast inward synaptic currents in 7 out of 25 transsynaptically labeled LPGi neurons (VH = −75mV; Figures 6N–6O), consistent with the notion that motor cortical axons make excitatory synapses on LPGi neurons. Taken together, these results indicate that LPGi neurons are in a key position to relay motor cortical signals bilaterally to vFMNs.
Here we used monosynaptic rabies virus based tracing, in situ molecular analysis of virallylabeled neurons, and optogenetic-assisted electrophysiological recordings to map and functionally characterize the premotor circuitry for vibrissal control in the neonatal mouse before and after a critical developmental landmark, namely the emergence of whisking behavior. We further assigned putative functions to vibrissal premotor inputs by using in situ hybridization to assess the neurotransmitter phenotypes of virally-labeled premotor neurons. The premotor wiring diagram with transmitter signs is summarized in Figure 7. In parallel with the developmental onset of whisking, we detected the addition of a new premotor module that is well suited to bilaterally coordinate and potentially synchronize vibrissal movements, and facilitate cortical modulation of whisking, which are two characterictic features of this exploratory behavior.
This is the first study to selectively trace inputs to vFMNs during the first two weeks of postnatal life, when exploratory whisking first emerges. Although prior studies in adult rats have used conventional retrograde tracers (i.e., HRP or CTB) to trace inputs to the lateral FN (Hattox et al., 2002; Isokawa-Akesson and Komisaruk, 1987), the injection of these tracers directly into the FN is likely to have labeled not only inputs to vFMNs but also other neurons that either project axons through the injection site or that make synapses with facial motor neurons other than vFMNs. In contrast, the transsynaptic tracing method used here made it possible to infect specific groups of vFMNs with “monosynaptic” rabies viruses and subsequently label their inputs more precisely.
The different specificities of these approaches likely account for why conventional methods label a wider range of loci – including the PCRt, Sol, PAG, and peri-trigeminal motor nucleus – than observed with the monosynaptic rabies method. In fact, these additional loci were also labeled when we injected ΔG-RV directly into the lateral FN, an approach that is likely to label other inputs to the FN besides those that terminate on vFMNs. Additionally, previous conventional tracing of inputs to the FN also labeled numerous neurons in the midbrain red nucleus (Hattox et al., 2002; Isokawa-Akesson and Komisaruk, 1987), which were not labeled in our transsynaptic tracing. A possible explanation for this discrepancy is that diffusion of tracer from the lateral FN may have spread into the juxtaposed rubrospinal tract, resulting in false-positive labeling of neurons in the red nucleus.
The monosynaptic rabies tracing method employed here is not without limitations. Although prior studies have established that rabies virions produced from an infected neuron spread predominantly to its presynaptic partners (Kelly and Strick, 2000; Ugolini, 1995), at longer time points post infection, virus particles that are shed by dying neurons may infect nearby axon terminals and glial cells (e.g. Figures S2G–S2H show labeled GFAP-positive glial cells in FN). Although we cannot entirely rule out the possibility of false-positive labeled neurons, we used relatively short survival times, and our transsynaptic tracing method labeled fewer loci than observed in other studies using conventional tracers, as described above. Furthermore, the LPGi premotor neurons are situated immediately outside the FN, and in fact direct injection of ΔG-RV into lateral FN in P6→P9 retrograde tracing experiments labeled these neurons (Figures S3 and Tables S1) likely due to their proximity to FN. Had virus leaked from dying neurons in our trans-sysnaptic tracing experiment and infected nearby neurons, we would have also labeled LPGi neurons in our P1→P8 trans-synaptic tracing, yet this did not happen. Instead, LPGi neurons were only labeled in P8→P15 trans-synaptic tracing, arguing against non-synaptic spreading. A second potential limitation is that rabies virus jumps synapses in a stochastic fashion, and thus may fail to label neurons that make only very sparse inputs to the infected cell. This feature of the rabies tracing method could explain why we recovered only small numbers of labeled neurons in the cortex, and only when we injected larger volumes of the virus into multiple vibrissae (Figures S1E–S1H). Notwithstanding these limitations, the monosynaptic rabies virus used here presents a useful strategy for selectively tracing inputs to motor neurons through intramuscular injections (Stepien et al., 2010), an approach that greatly simplifies mapping of premotor circuitry at early developmental stages when precise targeting of injections to brainstem motor nuclei can be challenging. Here, this approach enabled us to detect developmental changes in the vFMN premotor circuitry.
The in situ-GFP co-localization analysis performed here represents an initial step in characterizing the functional properties of the various synaptic inputs onto vFMNs (Table 1 and Figure 7). For example, we found that GFP-labeled LPGi neurons express glutamatergic markers, and whole cell recordings from vFMNs combined with optogenetic stimulation of LPGi axons show that LPGi axon terminals can evoke robust fast inward synaptic currents in vFMNs, consistent with fast excitatory synaptic transmission involving ionotropic glutamate receptors. In contrast, a previous study showed that direct electrical stimulation in the LPGi region elicited vibrissal movements, and ascribed these behavioral effects to the actions of serotonin (5-HT) because of the presence of 5-HT fibers in LPGi and the observation that 5-HT application can elicit periodic firing in vFMNs (Hattox et al., 2003). The present findings suggest that the glutamatergic synapses that LPGi premotor neurons form on vFMNs also contribute significantly to vibrissal movements evoked by electrical stimulation of the LPGi. Although a full characterization of the functional nature of various components of the vFMN premotor circuitry is beyond the scope of the current study, the transmitter phenotype analysis we performed here highlights the remarkable diversity of the synaptic inputs onto vFMNs.
The developmental addition of bilateral-projecting neurons in the LPGi to the vFMN premotor circuitry provides a plausible anatomical substrate for the emergence of whisking, which is characterized by cortically modulated, bilaterally coordinated movements of the vibrissae. When rodents whisk freely in air, vibrissae on both sides of the head move synchronously at similar amplitudes and frequencies (Gao et al., 2001; Sellien et al., 2005). Even when asymmetric bilateral vibrissal movements occur as a result of contact with objects on one side of the head or head turning, both sides usually continue to whisk at the same frequency and/or in phase, and hence are still bilaterally coordinated (except in object-localization tasks in head-fixed mice) (Mitchinson et al., 2007; O'Connor et al., 2010; Sachdev et al., 2003; Towal and Hartmann, 2006). Electrophysiological recordings show that LPGi axons provide potent excitatory input bilaterally to vFMNs, suggesting that LPGi premotor neurons are well suited to synchronize and coordinate bilateral vFMN firing. Furthermore, cortical axons form numerous synapses on LPGi neurons, suggesting that LPGi premotor neurons are especially well positioned to relay cortical command to vFMNs to facilitate cortical modulation of whisking patterns, kinematics, and behaviors (Berg et al., 2005; Berg and Kleinfeld, 2003; Brecht et al., 2004; Carvell et al., 1996; Donoghue and Parham, 1983; Hill et al., 2011; Huber et al., 2012; Li and Waters, 1991; Matyas et al., 2010).
The developmental addition of new premotor modules, as seen here for murine vibrissal circuitry, may be a general theme underlying the postnatal acquisition and refinement of behaviors in vertebrates. For example, juvenile zebra finches initially produce babbling-like subsongs, but abruptly begin to produce “plastic” songs with identifiable syllables and phrases near the end of the seventh week after hatching (Aronov et al., 2008; Immelman, 1969). This behavioral transition follows the formation of synaptic connections between the song premotor nucleus HVC and the song motor nucleus RA, suggesting that the emergence of plastic songs depends on the developmental addition of the HVC module to the premotor pathway (Aronov et al., 2008; Mooney, 1992; Mooney and Rao, 1994). Human infants also transition from infant stepping to toddler walking, from babbling to talking, and from grasping-like hand movements to fine finger movements (Gerber et al., 2010; Kuhl and Meltzoff, 1996; Wallace and Whishaw, 2003), and these behavioral changes likely require the addition of new control modules to the relevant motor pathways (Dominici et al., 2011). Therefore, vertebrate nervous system development may first involve the formation of neural circuitry for primitive movements, followed later by the addition of premotor modules that enable the generation of behaviors necessary for environmental and social exploration.
RΦGT mice: A “CAG-loxp-neo-loxp-rabies-G-IRES-TVA” cassette was inserted into the rosa26 locus to generate the RΦGT mice. Chat::Cre mouse line was purchased from the Jackson Laboratories (stock#006410). All Chat::Cre; RΦGT mice used in this study carry one Chat::Cre allele and one RΦGT allele (i.e. heterozygous for both alleles).
Chat::Cre; RΦGT mice were anesthetized by hypothermia (P1) or with ketamine/xylazine (50 mg/kg and 5 mg/kg, i.p.) (P8) and were injected with ΔG-mCherry or ΔG-GFP into the mystacial pad at P1 or P8 and brains were collected at 7 days post infection. To selectively infect motor neurons innervating intrinsic muscles (vFMN-in), the zygomatic branch was first transected at the location anterior to the eye, followed by injection of ΔG-GFP virus into the mystacial pad. To selectively infect motor neurons innervating extrinsic muscles (vFMN-ex), ΔG-GFP was injected into the caudal end of the nasolabialis muscle just under the dermis. For direct retrograde tracing with ΔG-RV, ΔG-GFP was stereotaxically injected into left side of the lateral FN of wild-type mice (C57/BL6) at P6 or P12 and brains were collected 3 days post infection. For AAV2-GFP (University of Pennsylvania Vector Core) and ΔG-mCherry double tracing experiments, AAV2-GFP was stereotaxically injected into right side of the motor or somatosensory cortex at P7 and ΔG-mCherry was injected into left mystacial pad as described above at P8.
The x, y, z coordinates of labeled neurons from all serial sections were obtained using the IMARIS Spots function, and 3D models were reconstructed using IMARIS software. To compare the spatial distribution of labeled neurons across individuals, the space covered by the 3D data points was evenly divided into identical-sized cubes. Dividing the brain into a range of cubes from 10x10x10 (1000 cubes) to 50x50x50 (125,000 cubes) yields essentially identical cross-correlation coefficiencies between animal pairs. Specifically, the number of labeled neurons inside each cube was counted, and similarity between two mice was measured by the Pearson correlation coefficient. Standardization, correlation calculation and heatmap plotting were performed using the statistical software R (http://www.r-project.org).
were performed according to standard procedures.
To examine motor cortex-LPGi connection, Chat::Cre; RΦGT mice were injected with AAV2-ChR2-YFP (University of Pennsylvania Vector Core) and ΔG-mCherry into right side of the motor cortex at P2 and left side of the mystacial pad at P8, respectively. Brainstem slices were prepared at P14. To examine LPGi-vFMNs connection, ΔG-ChR2-mCherry (Osakada et al., 2011) was stereotaxically injected into the left side of the lateral FN of wild-type (C57BL/6) mice at P12. Recordings were performed from the contralateral (right side) uninjected FN.
Coronal brainstem slices were incubated for 14 minutes in a 34°C bath of carbogenated modified ACSF containing (in mM): 92 N-methyl-D-glucamine, 2.5 KCl, 1.2 NaH2PO4, 30 NaHCO3, 20 HEPES, 25 dextrose, 5 sodium ascorbate, 2 thiourea, 3 sodium pyruvate, 10 MgSO4, 0.5 CaCl2. All recordings were made in whole cell voltage-clamp configuration using a Multiclamp 700B amplifier, the output of which was digitized at 10kHz (Digidata 1440A), and all recordings were made in room temperature ACSF containing 1 µM TTX (Sigma) + 100 µM 4-AP (Tocris). ChR2-expressing axon terminals were stimulated with brief pulses (10 ms, though 2 ms pulses were sufficient) of 473 nm laser light (Shanghai Laser and Optics BL473T3-150) presented directly over the recording location with a 200 µm jacketed fiber optic (Thor Labs). (See Supplementary materials for more details).
We thank Ed Callaway, Yingchun Ni and Zhigang He for providing various ΔG rabies viruses. We thank Bao-Xia Han for general technical support, and the members of the Wang lab and Mooney lab for discussions and suggestions. We thank Martin Deschenes, Jeff Moore and David Kleinfeld for sharing results prior to publication. We thank Drs Stephen Lisberger, Rebecca Yang, Martin Deschenes, Jeff Moore, and David Kleinfeld for critically reading the manuscript. This work is supported by NIH grants DA028302 and DE019440 to F.W., NS079929 to R.M, and NS078294 to B.R.A., who is also supported by the McNair Foundation. J.T is in part supported by Duke Institute of Brain Sciences.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.