The results presented here show that, six weeks after complete transection of the spinal cord at low thoracic segments, there was an attenuated expression of markers of presynaptic terminals in the ventral quadrants of L3/L4 spinal segments, possibly indicating impoverished innervation of motoneuron pools. Surprisingly, expression of synaptophysin around large neurons of lamina IX and synaptic zinc labeling along fibers of the ventral funiculi, clearly weaker in the spinal than in intact animals, was accompanied by an increased BDNF level in perikarya of large neurons of lamina IX, as well as by a higher number of BDNF IR processes and fibers of altered morphology.
Five-week treadmill locomotor training improved the motor abilities of the spinal rats, confirming beneficial effects of adequate proprio- and exteroceptive, rhythmical stimulation of the hindlimbs and pressure stimulation of the tail . It was accompanied by changes in molecular correlates of compensatory plasticity examined in the neuronal network located caudally to the transection. Namely, an increased number of synaptophysin-labeled nerve terminals upon ventral horn neurons was concomitant with altered levels of zinc-ergic terminals, which tended to increase in the ventral funiculi.
Interestingly, training caused selective enhancement of BDNF IR in perikarya in two out of five subpopulations of cells located in the motor nuclei, whereas expression of BDNF in processes and fibers in the ventral horn tended to be normalized by the training.
The effect of training of spinal animals on the distribution of markers of presynaptic terminals
A clear decrease of synaptophysin expression around the large neurons in the motor nuclei of spinal animals and its up-regulation in spinal-trained rats may be interpreted as an enrichment of synapses on the motoneuronal perikarya owing to exercise. Our data do not provide direct evidence on the type of synapses which were enriched. Serotonergic and noradrenergic terminals of the descending fibers retract permanently following complete transection, and this effect excludes them from the pool of boutons under consideration. However, the location and morphology of synaptophysin-positive terminals surrounding perikarya of the large neurons of lamina IX resembled large cholinergic C-boutons [30, 31]. A dramatic, sustained decrease of vesicular acetylcholine transporter (VAChT) in the terminals contacting sacrocaudal motoneurons following spinal cord transection at S2 segment, reported by Kitzman , indicates that cholinergic projection is vulnerable to the damage and perhaps might be restored by means of locomotor exercise. Indeed, our recent, preliminary data showed that locomotor training of the spinal rats caused an increase of the number of VAChT IR boutons in the triceps surae motoneuron pools in the lumbar segments . These cholinergic terminals may originate from a restricted group of partition interneurons located in the medial part of lamina VII . These interneurons terminate on perikarya and on proximal dendrites of motoneurons and, as documented recently, they potently regulate the excitability of motoneurons during locomotion . Namely, motoneurons are more likely to discharge if excitatory postsynaptic potential (EPSP) amplitude increases and afterhyperpolarization (AHP) decreases; if both occur, stepping of spinal animals was shown to be facilitated . Thus, a decrease of cholinergic input, together with a serotoninergic one, resulting in increased AHP, may contribute to the failure in stepping observed in our experiments.
Other types of terminals contacting motoneuron soma and their proximal dendrites can also be influenced by spinal cord transection and/or by locomotor training. In particular, morphological and biochemical correlates of inhibitory neurotransmission are affected. Both treatments were reported to influence inhibitory GABA and/or glycinergic synapses . Vesicular GABA transporter (VGAT)-expressing terminals, contacting soma and the proximal dendrites of sacro-caudal motoneurons were reported not to increase in number but to increase in size during 12 post-transection weeks . Moreover, a beneficial effect of training on the stepping ability in spinal cats was attributed to attenuation of GABA and glycine synthesis, upregulated after spinal cord transection in the segments caudal to the lesion . Also, in intact rats, four weeks of locomotor training caused a decrease of GABA and glycine levels in the tissue of the L3-L5 segments .
The number of glutamatergic terminals expressing vesicular glutamate transporter (VGLUT-2) and found in apposition to the soma and proximal dendrites of motoneurons, markedly increased beginning from the 2nd post-injury week . As a result, the ratio between vesicular transporters of glutamate and GABA (VGLUT-2/VGAT) on the soma of sacrocaudal motoneurons changed dynamically in time after spinal cord transection, finally showing a tendency to spontaneous normalization 12 weeks after injury .
All these observations indicate that synapses located on the perikarya and on proximal dendrites of motoneurons, which decreased in number in spinal rats and increased owing to the training, may represent limited phenotypes and may change the pattern of innervation of motoneurons.
The synaptic changes observed in our experiment may stem not only from increasing the number of synapses around motoneurons, but also from enlargement of synaptic territories owing to the training. Enlargement of presynaptic territory of C-terminals has been observed in pathological conditions in transgenic G93A SOD1 mice, during disease progression . In our study we were not able to distinguish between the two mechanisms, as immunofluorescent deposits in spinalized trained animals formed a uniform layer on large areas of the motoneuron membrane. Nevertheless, both mechanisms can lead to changes in synaptic connectivity either through formation of new synapses, or by strengthening the existing ones.
Widespread distribution of zinc-containing terminals in the spinal cord observed in the intact rats in this study fully confirmed the pattern described previously in rodents with the use of the same method for synaptic zinc visualization [40–42]. It has been suggested that a majority of zinc-containing neurons located in the ventral horn belong to a propriospinal system projecting segmentally or intersegmentally in the ventral gray matter . Here, we reported that after complete transection of the spinal cord, the density of synaptic zinc staining decreased in the ventral funiculi in the segments caudal to the injury site. This observation may partly reflect the changes in both ascending axons and descending proprioceptive fibers originating caudal to spinal transection, which were shown to contribute to the response of motoneurons to stimulation of the ventrolateral funiculi after spinalization . A lack of detectable changes in the dense and convoluted system of zinc-ergic innervation of the ventral horn in the lumbar segments might result from a relatively small, diffused fraction of degenerating fibers bearing zinc-ergic terminals in an extremely rich population of zinc-ergic endings. This observation also indicates that the majority of terminals contacting motoneurons that disappear in the synaptophysin staining are not zinc-ergic ones or that the retracting fibers are replaced by other zinc-ergic terminals. Interestingly, locomotor training of spinal rats produced an increase in density of zinc-ergic terminals in the ventral funiculi, leading to its normalization. This observation strengthens the possibility of reorganization of the neuronal network after post-injury locomotor training that includes axonal sprouting of the ventral propriospinal system and modification of the dendritic tree.
What type of transmission is involved in zinc-ergic network reorganization? In the spinal cord, a majority of zinc-ergic terminals were shown to be GABAergic , although synaptic zinc is also present in a subset of glycinergic terminals, as well as in glutamatergic boutons [43, 44]. Notably, synaptic zinc released during neurotransmission has direct and indirect actions: it may diminish excitatory neurotransmission, as an inhibitor of NMDA receptors, or act bidirectionally on inhibitory neurotransmission by modulating GABA and Gly receptors, as well as other receptors . A tendency to overall increased inhibitory neurotransmission following spinal cord injury, observed by Tillakaratne and co-authors [32, 38] speaks in favor of preferential degeneration of excitatory zinc-ergic nerve terminals. If so, the sprouting of zinc-ergic axons after locomotor training should involve terminals co-releasing glutamate to re-establish the balance between the excitatory and inhibitory inputs.
Formation of inhibitory, preferentially GABAergic synapses [46–49], as well as of glutamatergic synapses , was repeatedly reported to be promoted by BDNF, shaping synaptic plasticity. Assuming that a vast majority of synaptic changes in the isolated spinal segments involves such innervation, localization and levels of BDNF immunoreactivity were analyzed to evaluate a relation of BDNF responses to detected synaptic changes.
The effect of spinal cord transection on BDNF distribution and level
The transection itself caused an overall increase of BDNF IR in neuronal perikarya and in processes and fibers of the ventral horn in the L3/L4 segments. This observation is consistent with that of Zvarova and co-authors , who detected, using the technique of ELISA, a higher level of BDNF in the whole tissue homogenate of selected thoraco-lumbo-sacral segments, one and six weeks after complete spinal cord transection at low thoracic level. In that experiment, one week after injury, BDNF level in the L4-5 segments was higher by about 40% than in intact animals and that increase attenuated to 17% by six weeks after transection. Li and co-authors also reported recently that the number of BDNF IR neurons in the ventral horn was increased by over 60% by the end of the first week after complete spinal cord transection at low thoracic segments and that two weeks later it returned to control level . Thus, surprisingly, in spinal rats, which do not demonstrate spontaneous locomotor recovery, it is not an overall BDNF level deficit which seems to be a limiting factor in functional improvement. We assume that it is a limited BDNF availability in the synaptic cleft, which results from disturbances in BDNF synaptic release, and/or altered expression of TrkB receptors, particularly TrkB truncated forms, shown to limit BDNF signaling in vivo . Support for this hypothesis stems from the studies that showed an increase of expression of truncated TrkB, detected four weeks after spinal hemisection  and seven weeks after contusion of the spinal cord . On the other hand, local synaptic accumulation of BDNF released from overloaded terminals might desensitize TrkB full length (FL) receptors, downregulating neurotrophin signaling, as shown by us in in vitro model . In addition, a deficit of zinc ions, which can decrease transactivation of the synaptic TrkB by a neuronal activity-regulated mechanism [56, 57], may attenuate TrkB signaling. These disturbances may affect the strengthening of synaptic connections owing to desynchronized firing of the presynaptic and postsynaptic neurons , discussed by Petruska et al. .
It is worth consideration that the effect of other types of spinal cord injury on BDNF mRNA and protein levels in the lumbar spinal cord of the rat was different from that after complete transection [14, 59, 13, 60, 61]. Thus, the extent of spinal cord injury substantially influences the expression of BDNF mRNA and protein in the region caudal to the injury site, suggesting the role of descending pathways in this regulation. However, Garraway and Mendell  attributed these physiological differences to cellular changes characteristic of these two types of injury rather than to an interruption of descending inputs, as they have been observed both caudally and rostrally to the lesion site.
An hypothesis of increased excitability of the central pattern generator (CPG) in chronic spinal animals might be useful to explain up-regulation of BDNF caudally to the lesion site . An increase of the BDNF level in the ventral horn neurons, which was sustained several weeks after transection, may be indicative of the compensatory response of the regions deprived of the descending innervation but still receiving peripheral inputs. First, since a majority of ventral horn neurons can synthesize BDNF [8, 64], its higher level reflects the recovery potential of these neurons. Second, as BDNF expression is activity-dependent, it may also be indicative of an increased drive to ventral horn neurons.
A cutaneous input, indirectly activating motoneurons via interneurons, could be a good candidate responsible for an increased drive to spinal neurons [65–67]. The motoneurons innervating hindlimb muscles were frequently driven by cutaneous input from the dorsal aspect of the paraplegic hindlimbs, which could keep BDNF up-regulated. If BDNF release processes were undisturbed or enhanced, we could expect that BDNF would elevate presynaptic transmitter release, as reviewed by Poo . Our observation that, in the spinal animals, the enhanced levels of BDNF appeared in shorter and thinner fibers and processes than in the intact and spinal trained ones, may result from their shrinkage or generation of sprouts after spinalization and be indicative of altered presynaptic mechanisms. Also, it may explain an impoverishment in neural networks, as indicated by a reduced synaptophysin expression around the large neurons in the motor nuclei, together with a decreased number of zinc grains along the processes and fibers in the ventral funiculi. Postsynaptic responses to altered levels of neurotransmitters may be further modified by BDNF. BDNF may differentially affect an effectiveness of neurotransmission, depending on the type of synapse and postsynaptic cell, as shown for GABAergic synapses on GABAergic neurons, where BDNF decreased the efficacy of inhibitory transmission .
The effect of training in spinal animals on BDNF distribution and level
Surprisingly, training did not influence the overall levels of BDNF immunoreactivity either in perikarya or in processes and fibers of the ventral quadrant, compared with those of the spinal non-trained animals. However, two subsets of the ventral horn cells tended to respond with higher BDNF levels to the locomotor training. One of them, with soma size ranging between 100 and 400 μm2, represented a mixed population, which could consist of, for example, γ-motoneurons , interneurons, or even glial cells. The other cells, with the perikarya exceeding 1000 μm2 presumably corresponded to α-motoneurons . Our previous studies showed that locomotor training caused up-regulation of BDNF mRNA in practically all types of cells, differentiated by size, in the intact spinal cord . A question arises about the physiological meaning of an increased, selective expression of BDNF in some cells of the ventral horn in trained spinal rats.
A possible explanation may be found in the experiments showing that the amplitude of monosynaptic Hoffmann reflex in the soleus muscle increases by a factor of two when preceded by tactile stimulation of the tail in the awake, intact rat . It is thus feasible that, in spinal animals, treadmill walking, which produces adequate activation of proprioceptive input from the plantar aspect of the feet when supplemented with the pressure of the tail, may favor activation of extensor motoneurons [65, 72]. This effect would explain the selective up-regulation of BDNF in the neurons of motor nuclei in spinal trained animals. Indeed, our recent experiment showing an increase in the number of cholinergic boutons apposing the extensor but not the flexor motoneurons after training, speaks in favor of this possibility .
Training caused changes in the distribution of BDNF in the processes and fibers, leading to its normalization. More numerous, but shorter and thinner BDNF IR processes and fibers, were detected in the spinal animals, versus longer and thicker ones in the spinal trained rats. This observation, when associated with selective up-regulation of BDNF in some cells owing to the training, might point to their dendritic origin. Indeed, as identified with double immunolabeling, many of the BDNF IR profiles were dendrites. They can represent a dendritic tree of spinal motoneurons, enriched owing to locomotor training. Such enrichment was described by Gazula and co-workers after five days of physical training in spinal animals . In contrast to the observations by Ying and co-workers  showing that BDNF protein levels were increased in proportion to the amount of voluntary exercise performed after hemisection of the spinal cord, we did not find such a relationship.
Taken together, our data indicate that locomotor training caused redistribution of BDNF to selected groups of spinal cells rather than influenced the general level of BDNF caudally to the site of transection. This observation may indicate that selective up-regulation of BDNF promotes rewiring of the spinal neuronal network activated by kinesthetic stimuli associated with locomotion and by pressure stimulation of the tail. Moreover, it indicates that, in functional reorganization, not only the availability of a releasable pool of BDNF, but also its proper patterning, is crucial.
Co-localization of BDNF with pre- and postsynaptic markers
The number of co-localized profiles of synaptophysin and BDNF in the ventral horn was small. This result was surprising in view of the recent data pointing to a role of BDNF in the regulation of the expression of presynaptic proteins involved in synaptic vesicle fusion and in synapse formation . Three days of intensive, voluntary locomotor exercise were reported to lead to an increase of synapsin I and synaptophysin proteins level in the hippocampus, which was abolished when BDNF signaling via TrkB receptor was blocked [50, 74]. It is worth stressing that Pozzo-Miller and colleagues observed a reduction of synaptophysin in hippocampal synaptosomes in BDNF knockout mice .
An unexpectedly small number of co-localized profiles of synaptophysin and BDNF in the ventral horn may derive from the fact that both proteins, present in synapses, occupy different subcellular compartments. Synaptophysin is a component of all synaptic vesicles, whereas BDNF is present only in a subpopulation of endosomes. As a result, relatively small deposits of BDNF IF (see Figure 12) may be overshadowed by intensive synaptophysin IF profiles of numerous synaptic vesicles, which build large spots of signal in terminals and their clusters.
We also questioned the identity of the BDNF IR bouton-like accumulations around large neurons of lamina IX. Rare co-localization of synaptophysin and BDNF IR in these structures suggested that the latter do not represent BDNF-driven active terminals. Indeed, some of these accumulations corresponded to dendritic structures, as revealed by strong MAP-2 staining with double-labeled (BDNF/MAP-2) profiles localized in the proximity of large neurons. A deficiency in synaptophysin IR around large neurons of lamina IX after spinalization and its high increase in terminals owing to training were surprising in view of similarly strong expression of BDNF in this region in both groups. This discrepancy may partly result from the fact that, as shown by us, some of the reappearing terminals in trained rats do not carry BDNF, albeit it does not exclude their dependence on BDNF.