We used an intact in vitro spinal cord preparation to demonstrate the axon diversity of LCNs in lamina I. We provide morphological evidence for their possible involvement in segmental interlaminar and propriospinal sensory information processing by showing that LCNs frequently possess a number of short and putative long propriospinal branches. To the best of our knowledge, for the first time, we provided extensive 3-D reconstructions of several lamina I neurons that allow novel topological analyses. Besides the complexity of LCN axons, these analyses also revealed heterogeneity in the overlap between major axon branches, in the distribution of axon varicosities, and in the estimated AP propagation times for different points of the axon.
Our sample in this study is biased in some respects. First, we used young animals, because the viability of the in vitro intact spinal cord preparation in our experiments proved to be best at P14–P24. For this reason, one cannot exclude that axons of more mature LCNs may be different. This, however, seems unlikely, in that lamina I neurons with very similar local axon branching pattern have been described for adult rats (Li et al., 2000). Second, because of the decreased visibility through regions rich in myelinated fibers, the medial part of the dorsal horn surface (dorsal root entry zone) was excluded from the search for neurons. Thus, conclusions of this study are based on neurons from the lateral two-thirds of lamina I. Third, when we visualized the surface of the dorsal horn in our intact spinal cord preparation, detection of the laminar border between lamina I and substantia gelatinosa relied solely on the appearance of the uniform, densely packed layer of small lamina II neurons. For this reason, to ensure that recorded neurons were in lamina I, we selected larger neurons in the most superficial (first to appear) cell layer of the dorsal horn. In this way we unavoidably excluded small lamina I neurons. The description of axons of small lamina I and lamina II neurons will be the subject of another study. Finally, full analysis of all investigated parameters was possible only in 3-D reconstructed neurons. However, the number of such reconstructions is relatively small compared with the total of the LCN sample, resulting in an additional source of potential variance.
Somatodendritic architecture of LCNs
Based on their somatodendritic architecture, lamina I neurons in the rat have been classified into four categories: fusiform, pyramidal, flattened, and multipolar (Lima and Coimbra, 1986). Lamina I LCNs in this study, with few exceptions, fell into the flattened and multipolar somatodendritic categories. This is in good agreement with our earlier study on local collaterals of lamina I ALT-PNs (Szucs et al., 2010) in which none of the filled large multipolar neurons had an ascending axon in the contralateral anterolateral white matter.
Although lamina I neurons have the bulk of their dendritic arbors confined within this lamina (Gobel, 1978), we also showed in this study that multipolar neurons issue prominent ventrally protruding dendrites (Lima and Coimbra, 1986) that not only enter lamina II, as reported for human lamina I neurons (Schoenen, 1982), but also reach lamina III. Therefore, it is reasonable to assume that, in addition to the common sources of input to lamina I neurons, multipolar cells integrate information from deeper laminae (e.g., from local axons and primary afferents terminating in laminae II and III) and should be treated as a separate group. Along this line, one cannot exclude the possibility that multipolar lamina I LCNs may even be activated monosynaptically by primary afferents that convey information to deeper laminae (e.g., A-beta afferents or A-delta hair follicle afferents; Todd, 2010). A detailed description of primary afferent input to lamina I LCNs that may answer this question is the subject of an ongoing study in our laboratory.
Similar to ALT-PNs (Szucs et al., 2010), a large percentage of LCN axons originated from one of the primary dendrites, although the mean distance from the soma in case of LCNs was slightly less. This, however, may be simply related to the fact that dendritic spread of LCNs was generally smaller than that of ALT-PNs reported from our previous study (Szucs et al., 2010). The high proportion of axons with dendritic origin among LCNs suggests that this axon initiation type is a common feature of large lamina I neurons and is in agreement with earlier reports (Cheunsuang and Morris, 2000; Hylden et al., 1986). Dendritic origin of the main axon has also been observed in cat preganglionic sympathetic neurons (Morgan, 2001) and in motoneurons (Duflocq et al., 2011). This anatomical variation seems to be frequent in spinal cord neurons, and determining its functional importance will require further in-depth knowledge of the ultrastructure and ion-channel composition of the axon initial segment and the dendrites giving rise to them (Duflocq et al., 2011).
Overlaps between LCN axons and ipsilateral collaterals of lamina I ALT-PNs
LCNs reported in this study all had collaterals located in the superficial dorsal laminae (I–II, occasionally entering III and IV), but only about one-third of ALT-PNs (32.5%; lateral and mixed collateral types) have ipsilateral collaterals in the same region (Szucs et al., 2010). This suggests that the influence of LCNs in laminae I–II is greater than that of the ALT-PNs. In the DLF, the contribution to rostrocaudally oriented varicose axon collaterals from the two groups is more balanced: 39% of LCNs and 40% of ALT-PNs (lateral and mixed collateral types) have collaterals in this region. Thus, both types of lamina I neurons may relay local segmental information to neighboring segments and to neurons of the lateral spinal nucleus. Finally, information from lamina I to ventral laminae (V–VIII) is mostly relayed by collaterals of ventral- and mixed collateral-type ALT-PNs (30%), insofar as only a few LCNs (n = 2; 2.4%) had axons reaching lamina VII.
Possible roles of the local axon of LCNs
The extensive branching of LCN axons suggest that, besides the anatomical divergence of primary afferent fibers, LCNs may provide further divergence of processed primary afferent information after integrating it with other sources of input. Furthermore, LCN axons occupy most dorsal laminae (I–II and occasionally III–IV), and this feature, theoretically, allows relaying of C-fiber information to deeper laminae. Indeed, Braz and Basbaum (2009) reported neurons in deep laminae (III–V) receiving polysynaptic input from unmyelinated primary afferents. Lamina I LCNs may be direct sources of such polysynaptic input or provide it through contacting lamina II islet and stalked neurons that were shown to project to deeper laminae (Eckert et al., 2003). At the same time, multipolar LCNs may be monosynaptically activated by primary afferents that terminate in deeper laminae. Consequently, neurons in laminae III–V may integrate direct primary afferent information with indirect processed form of the same information through lamina I LCNs.
Because of their extensive local axon, relatively few LCNs could influence the superficial dorsal horn along several segments, in a sustained tonic manner based on their firing pattern and the frequent occurrence of rhythmic intrinsic firing. The large number of varicosities and the highly branched, extensive axon of LCNs also imply that these neurons may be involved in volume transmission. In the case of inhibitory LCNs, rhythmic intrinsic activity may result in continuous GABA release, leading to metabotropic GABAB receptor activation in neurons of several neighboring segments.
Although the majority of the tested LCN axons proved to be VGAT positive, for several reasons, it is likely that not all LCNs recorded in this study are inhibitory. About 75% of neurons in lamina I of the rat spinal cord are neither GABA nor glycine immunoreactive, and these cells (many of which are LCNs) are thought to be glutamatergic (Polgar et al., 2003). Soma areas of inhibitory LCNs fell in the lower half of the total LCN soma area range, so LCNs with larger somata may be the excitatory ones. In addition, in a previous study, we recorded synaptically connected lamina I neuron pairs, in which the presynaptic neurons were always located in lamina I, and those connections were, without exception, excitatory (Luz et al., 2010). Furthermore, about one-third of the LCNs in this study expressed functional NK1 receptors (data not shown here), which were shown to be expressed in high percentage on lamina I projection neurons (Spike et al., 2003; Todd et al., 2000) and in excitatory interneurons (Littlewood et al., 1995), although the existence of NK1 receptor positive/GABAergic neurons has also been proposed for the spinal (Ferrini at el., 2007) and the medullary (Wang et al., 2000) dorsal horn of rats.
We have demonstrated that, although the general appearance of the complex local axon is similar in all LCNs, axon varicosity distribution and branching pattern show certain heterogeneity. This may be related to the segmental position or to the location of a particular LCN in the somatotopic map. The relatively low number of 3-D reconstructed axons that could be used for such topological analyses, however, did not allow us to establish such correlations.
Functional consequences of the organization of LCN axons: branching, varicosity distribution, and spike propagation
The complex axons of LCNs with large numbers of varicosities and a tonic firing pattern, together with the observation that some of these neurons are rhythmically active (data not shown here), suggest that LCNs could provide tonic regulation and could distribute integrated information in a pattern governed by inherent spatial and temporal properties of the axon. For a complex axon with thousands of varicosities, forming networks in a highly somatotopic spinal dorsal horn, temporal dispersion of synaptic output is of crucial functional importance. The propagation time of an AP, initiated in the axon initial segment, is determined mostly by path length, axon diameter, myelination, and branching (Debanne et al., 2011). Path distance to a particular point of the axon is strongly dependent on branching. The alternating branching of primary collaterals from the main axon into the rostral and caudal directions, frequently observed in this study, seems to be an efficient solution for maximally filling the target space (laminae I–III) and, at the same time, equalizing path distance for the rostral and caudal portions of the axon tree. This setup, however, in a system with strong spatial boundaries, such as the spinal dorsal horn, results in an overlap between major branches. Target regions supplied by multiple overlapping branches may be activated repeatedly, so it will be important to understand conduction properties of LCN axons to evaluate their role in the spinal network.
A recent article comparing 3-D reconstructed basket and spiny cell axons used uniform conduction velocities to estimate, among other things, temporal dispersion in the axonal trees (Budd et al., 2010). The authors pointed out that this approach might underestimate temporal economy, because, for example, myelinated primary axon collaterals could reduce latency to child branches with the same amount of axon for wiring. Myelinated long-range branches of GABAergic interneurons, running in the stratum moleculare between innervation regions, have been shown recently in the hippocampus (Jinno et al., 2007). This study also demonstrated the presence of myelin around thicker initial parts of the axon of LCNs, and, indeed, when we used a diameter threshold to distinguish myelinated parts of LCN axons, temporal dispersion of AP propagation in the tree was reduced. This effect seemed to be more prominent in cells with remote varicosity fields connected by thicker, probably myelinated, axon pieces.
Although in our simulation of AP propagation in the complex axon of LCNs we considered the presence of myelin, this is still a simplified approach. Unfortunately, experimental data on other important factors (e.g., voltage-gated ion channel types and densities), with the exception of some parts of substantia gelatinosa neuron axons (Safronov, 1999; Safronov et al., 1997), are not available for different parts of the axons of spinal neurons. The diameter threshold for myelinated axons (0.35 μm) was chosen from measurements in a single neuron, but it is in agreement with the diameter reported for unmyelinated cortical axons (0.08–0.4 μm; Berbel and Innocenti, 1988; Westrum and Blackstad, 1962). It should also be noted that our simulated AP propagation times in LCN axons are probably overestimations, because the constants used for calculating nonmyelinated segment conduction velocities are based on measurements at room temperature. In a living organism, the AP propagation times must, therefore, be shorter. Our goal, however, was to prove that the adequately positioned myelinated regions of the axon significantly alter the AP propagation times in remote parts of the axon tree.
APs traveling through several sudden diameter irregularities such as axon varicosities have been suggested to suffer additional delay (Manor et al., 1991), and passing through several branch points may even result in a failure of propagation (Debanne et al., 1997). In the case of LCNs with thousands of varicosities, this could mean a significant further increase in propagation times and may even cut off some parts of the axon tree from signal invasion temporarily.
Subthreshold signal propagation within LCN axons could also lead to robust functional differences in different parts of the tree. Recent direct recordings from axonal structures in the hippocampus and neocortex suggested that subthreshold graded signals propagate down the axon over distances of up to 1 mm. At certain synapses, these analog axonal signals were shown to modulate AP-dependent transmitter release (Alle and Geiger, 2008). The same mechanism in LCN axons could modulate synaptic output at varicosities in the vicinity of the cell body as opposed to remote varicosities. For example, the SP-induced depolarization observed in almost half of the LCNs (data not shown here) could also propagate down the LCN axon and may boost transmitter release. Given the complexity of LCN axons and their accessibility resulting from their superficial location, they could be ideal targets for studies on axon conduction by using novel imaging techniques, such as high-speed fluorescent sodium imaging with high temporal and spatial resolution (Fleidervish et al., 2010; Foust et al., 2010), in vitro or in vivo.
Long propriospinal connections of lamina I LCNs
Although LCNs clearly do not project through the lateral funiculus on the contralateral side, for several reasons it seems likely that they may also function as short and long propriospinal neurons. As we report here, a large percentage of LCNs (86% of flattened and 69% of multipolar neurons) had long, solitary axon branches, often with a myelinated appearance, in the DF, DLF, and Lissauer's tract. These branches had no preferential direction and often ran for two or three segments in the rostral or caudal direction before fading below visibility and were similar to those described from earlier anatomical studies (Cajal, 1909; Lenhossek, 1895; Szentagothai, 1964).
The presence of such collaterals is also in line with electrophysiological observations. About one-third of nociceptor-driven lamina I neurons could be antidromically activated by stimulation of Lissauer's or deeper tracts from up to three segments rostral to their origin (Cervero et al., 1979). The conduction velocity of the axons of these neurons suggested that they are small and myelinated (Cervero et al., 1979). In addition, unmyelinated propriospinal axons were also observed in the DLF at the sacral level (Chung et al., 1988). These latter fibers, as we showed earlier (Szucs et al., 2010) and also in this study (see blue neuron, cell ID: L292_E4, in Fig. 3), could also include lateral collaterals of ALT-PNs. Given all this, it is reasonable to say that a substantial proportion of axons in the DLF is indeed propriospinal fibers and forms massive intersegmental connections. These connections are formed not only by axons of lamina II–III neurons, resulting in a closed system (Szentagothai, 1964), but also by collaterals of lamina I neurons, some of which serve as major output elements of the dorsal horn circuitry.
Lenhossek (1895), Cajal (1909), and Szentagothai (1964) described lamina II–III neurons with several segment-spanning, rostrocaudally oriented axons in Lissauer's tract, and intersegmental integration was also proposed recently for dorsal horn cholinergic neurons, with similar rostrocaudal axonal organization (Mesnage et al., 2011). However, some LCNs in our study possess morphological features that would also allow them to establish not only short but also long propriospinal connections. Solitary branches in the DF, DLF, and Lissauer's tract never crossed the midline and ran rostrally until they faded or reached the end of the spinal cord block, providing an anatomical substrate for such long propriospinal projections. Indeed, previous reports suggested that about one-fourth of lamina I neurons have long, ascending propriospinal projections that extend from the lumbar to at least midthoracic spinal levels but do not reach the brain (Bice and Beal, b) Thus, the long, solitary branches observed in some of our LCNs may be long propriospinal branches, indicating that at least some LCNs may have influence on distal spinal cord regions. To test this hypothesis further, careful morphometric analysis of LCN axons will be needed combined with retrograde labeling from distal spinal cord segments. Eventually, understanding of local connections and outputs of lamina I neurons will allow designing better strategies for intercepting pain signals at the spinal cord level.