Modularity of Endpoint Force Patterns Evoked Using Intraspinal Microstimulation in Treadmill Trained and/or Neurotrophin-Treated Chronic Spinal Cats

Abstract

Chronic spinal cats with neurotrophin-secreting fibroblasts (NTF) transplants recover locomotor function. To ascertain possible mechanisms, intraspinal microstimulation was used to examine the lumbar spinal cord motor output of four groups of chronic spinal cats: untrained cats with unmodified-fibroblasts graft (Op-control) or NTF graft and locomotor-trained cats with unmodified-fibroblasts graft (Trained) or NTF graft (Combination). Forces generated via intraspinal microstimulation at different hindlimb positions were recorded and interpolated, generating representations of force patterns at the paw. Electromyographs (EMGs) of hindlimb muscles, medial gastrocnemius, tibialis anterior, vastus lateralis, and biceps femoris posterior, were also collected to examine relationships between activated muscles and force pattern types. The same four force pattern types obtained in spinal-intact cats were found in chronic spinal cats. Proportions of force patterns in spinal cats differed significantly from those in intact cats, but no significant differences in proportions were observed among individual spinal groups (Op-control, NTF, Trained, and Combination). However, the proportions of force patterns differed significantly between trained (Trained and Combination) and untrained groups (Op-control and NTF). Thus the frequency of expression of some response types was modified by injury and to a lesser extent by training. Force pattern laminar distribution differed in spinal cats compared with intact, with more responses obtained dorsally (0–1,000 μm) and fewer ventrally (3,200–5,200 μm). EMG analysis demonstrated that muscle activity highly predicted some force pattern types and was independent of hindlimb position. We conclude that spinal motor output modularity is preserved after injury.

INTRODUCTION

We have previously reported that stimulation of the spinal cord, specifically the gray matter, produces a small number of stereotypic responses, measured as isometric force patterns at the paw, in both anesthetized and decerebrated spinal-intact cats (Lemay and Grill 2004). Our results confirmed earlier results in the spinalized frog (Giszter et al. 1993) and rat (Tresch and Bizzi 1999) that point to a modular organization of the spinal motor output (Bizzi et al. 2002). Further experiments in the frog suggest that these responses are used during natural behaviors to produce unperturbed motions and that reactions to perturbations are summations of the individual movement primitives (Kargo and Giszter 2000).

Activation of this interneuronal circuitry has been proposed by various groups as a way of restoring locomotor movements (Mushahwar et al. 2002; Tai et al. 2003). Mushahwar and colleagues have demonstrated that locomotor like movements can be generated with as few as two to three intraspinal electrodes in both spinal-intact (Mushahwar and Horch 2000) and chronic spinal cats (Mushahwar et al. 2002). Similar results were obtained by Tai et al. in anesthetized spinal-intact cats (Tai et al. 2003). Both groups have stressed that a limited number of movement types are obtained with intraspinal microstimulation and that the sites producing these responses are topographically organized in the cord: flexion movements are generated with stimulation in the dorsal portion of the cord, whereas extension responses are more prevalent with stimulation into more ventral sites. Recently, Barthélemy et al. (2006) have reported a similar anatomical stratification in flexor/extensor responses obtained with intraspinal microstimulation in subacute spinal cats and chronic locomotor-trained cats.

To determine the feasibility of using intraspinal microstimulation to elicit motor output following chronic spinal cord injury, it is necessary to understand the types of changes in motor responses that occur in the spinal cord below the level of injury. It is also necessary to understand how the responses to stimulation may be affected by other intervention strategies such as treadmill locomotor training (Barbeau and Rossignol 1987; Lovely et al. 1986), pharmacological intervention (Barbeau et al. 1993; Barbeau and Rossignol 1991; Rossignol et al. 2001), and cellular transplants that promote axonal growth and neuroprotection (Howland et al. 1995; Liu et al. 1999b) that are likely to be used in combination with electrical stimulation to maximize locomotor recovery. Under the motor primitive hypothesis, plasticity following injury would not be expected to affect the types of motor responses but could be expected to produce changes in the relative proportions of the responses by affecting the neural circuitry, e.g., the balance of the excitatory/inhibitory supraspinal input to the “hard-wired” primitives (Giszter et al. 1993). Some of the effects of spinalization on motor output due to stimulation have been examined by a number of groups. The motor responses to stimulation of lumbar cord seem relatively unchanged after chronic spinalization (Barthélemy et al. 2006; Saigal et al. 2004) with the topographical localization of the responses being modestly affected by chronic injury (Saigal et al. 2004), more significantly with chronic injury and locomotor training, and even more significantly by administration of clonidine, an α2-noradrenergic agonist that promotes locomotion in spinal cats (Barthélemy et al. 2006).

However, the effects of transplant or of chronic spinalization by itself on the motor output elicited via intraspinal microstimulation have not been investigated in the cat. Here we determined the effects of locomotor training and/or neurotrophin secreting cellular transplants on the distribution, types, and frequency of force patterns elicited via intraspinal microstimulation in the lumbar spinal cord of chronically spinalized cats. Because neurotrophins have a profound effect on the locomotor recovery observed in these animals (Boyce et al. 2007) and provide the untrained animals with locomotor ability similar to that achieved with training (Belanger et al. 1996; Lovely et al. 1986), we investigated whether this recovery correlates with changes in the spinal output to microstimulation of the lumbar gray. Some of these results have been reported previously in abstract form (Boyce and Lemay 2005; Boyce et al. 2004; Lemay and Boyce 2003).

METHODS

Surgical preparation

Sixteen female domestic short-haired cats were used in this study. All animal care and procedures were according to National Institutes of Health guidelines and were reviewed and approved by the Institutional Animal Care and Use Committee of Drexel University. The procedures for treadmill training and cell transplants are described in detail elsewhere (Boyce et al. 2007). Briefly, after acclimating the cats to a motorized treadmill and acquiring sufficient pretransection locomotion data, the cats received a complete spinal transection at the T₁₁–T₁₂ level. All cats received a transplant of unmodified or neurotrophin secreting fibroblasts into the cavity site at the time of transection. To prevent rejection of the graft, all cats were immunosuppressed for the duration of the study. On the day of surgery, the fibroblasts were harvested using aseptic techniques (Liu et al. 1997, 1998), suspended in Vitrogen (Cohesion Technologies, Encinitas CA), a collagen matrix adjusted to a pH of 7.2–7.4, and injected into the cavity site. The dura was then closed with 6.0 Prolene, muscles closed in layers, and skin closed using 5.0 Prolene.

After transection, the animals were returned to their cages. After a 4- to 5-day period of recovery, the animals were allowed to freely move around the room during the day. They were returned to their cages at night. Cats were either trained to locomote daily on a treadmill (15 min/day, 5 day/wk) or not trained. Four experimental groups of spinal cats were thus generated: cats receiving no intervention (Op-control, n = 3), cats receiving control fibroblasts but daily treadmill training (Trained, n = 5), cats receiving neurotrophic-factor-secreting cells but no training (NTF, n = 5), and finally cats that received daily treadmill training and NTF-secreting graft (Combination, n = 3). All cats were evaluated on the treadmill once every 2 wk by filming and digitizing the kinematics of locomotion induced with perineal stimulation (see Boyce et al. 2007 for details).

Op-control animals were unable to generate plantar steps. Cats receiving daily treadmill training (Trained) regained the ability to step, i.e., perform ≥10 consecutive plantar weight-bearing steps at speeds ≤0.8 m/s. The group of cats that was not trained but received transplants of BDNF/NT-3-secreting fibroblasts (NTF) recovered stepping in as little as 2 wk. Finally in cats that received a combination of BDNF/NT-3 transplant and daily training (Combination), stepping was enhanced (longer steps) compared with either treatment alone. Histological examination verified the absence of descending fiber growth through the lesion, suggesting that the locomotor recovery observed was independent of axonal growth through the lesion.

Spinal mapping experiments were performed for cats in all four groups at 12–14 wk posttransection. On the day of spinal mapping, animals were anesthetized with ketamine HCl (Ketaset, 25 mg/kg im) given in combination with atropine (0.05 mg/kg im). Hair on the lumbosacral region, medial aspect of the left hindlimb, ventral part of the neck and the top of the head was removed using clippers. Animals were then intubated and anesthesia maintained using isoflurane (1.5–3% in oxygen). The cephalic vein was catheterized to administer fluid/drugs during the procedure. Animals were ventilated to maintain expired CO₂ at 3–4%, body temperature was maintained between 37 and 39°C using thermal pads, warm 0.9% saline with 8.4 mg/ml sodium bicarbonate and 5% dextrose added was administered IV (10–15 mg·kg⁻¹·h⁻¹), and carotid blood pressure was monitored throughout the experiment. The L₃ to S₁ vertebrae were cleared, and the L₄ to L₇ spinal column dorsal processes removed by laminectomy. Prior to the laminectomy and every 6 h after the initial dose, Dexamethazone (2 mg/kg iv) was administered to minimize spinal swelling.

EMG electrodes were inserted into representative knee flexor (biceps femoris posterior), ankle flexor (tibialis anterior), knee extensor (vastus lateralis), and ankle extensor (medial gastrocnemius) muscles of the hindlimb ipsilateral to the stimulation site. Proper localization of the electrodes into the muscles of interest was verified via direct electrical stimulation of the muscle and postmortem dissection. To eliminate possible mechanical coupling of the contralateral hindlimb responses onto the force sensor, the contralateral hindlimb was denervated by severing the femoral, obturator, and sciatic nerves. The cat was then placed in a stereotaxic frame and the right hind paw attached with tie-wraps and a shell to a small bar mounted on a six-axis force transducer (nano17, ATI Industrial Automation) by a rotational joint. The force transducer was mounted at the endpoint of a 2 DOF robot that allowed positioning of the paw through a 20 × 20-cm workspace. Warm mineral oil was applied to the cord to prevent drying, the dura was opened and the lumbar segments identified by examining the vertebral exit levels of the dorsal roots. A mid-collicular decerebration followed, the brain rostral to the transection was removed (including cortex and thalamus) and the skull packed with Surgicel, Avitene, and agar to control bleeding. Dextran was administered if needed to maintain blood pressure, and anesthesia was discontinued once the decerebration was completed. Spinal mapping started ≥1 h postdecerebration.

Spinal cord motor output mapping

The L₅–L₇ spinal cord was mapped from the dorsal root entry zone to the midline mediolaterally and from the cord surface to 5,200 μm deep along dorsoventral penetrations (in 200 μm increments). Using platinum iridium microelectrodes (No. UEPSEJLEBN1M, FHC, Bowdoinham, ME), biphasic current pulses (0.5 s train, 10 to 100 μA amplitude, 100 μs pulse duration, and 40 Hz frequency) were delivered to the spinal gray matter to elicit motor responses from the limb. When a robust (force magnitude >0.3–0.4 N at the paw) repeatable response was obtained at any depth along a track, forces acting at the hind paw from stimulation at that spinal site were recorded at multiple paw positions with the stimulation parameters kept constant. In general, forces >0.3–0.4 N were elicited with a stimulus amplitude of 10 μA. In instances where forces were small, typically in the ventral portion of the cord, the stimulation amplitude was increased ≤100 μA. Stimulation amplitude was kept low to minimize the volume of neural tissue activated, but we did not explore using <10 μA. Recording the forces acting at the hind paw entailed moving the paw to nine positions in a 6 × 6 cm grid (or 12 positions on a 9 × 6 cm grid) centered at a mid-stance position in the sagittal plane. In one cat, we limited the number of hindlimb positions to three to sample a larger number of stimulation sites. The isometric forces produced at the hind paw were sampled at 2,500 Hz, along with the amplified and filtered (10–1,000 Hz) EMG signals. The mid-stance position force measurement was repeated after forces at the other eight grid positions were collected to ensure that forces were stable over time.

Responses were stable during the collection of force patterns and the force vectors at mid-stance before and after the measurement of forces at other positions differed by 1.6 ± 28.5° in direction and 0.07 ± 0.43 N in magnitude (mean ± SD of the initial minus final forces, n = 141). The average difference for the forces was not significantly different from zero, with the confidence interval ranging from 0.0 to 0.14 N. V-test (Batschelet 1981) of the differences in direction rejected the null hypothesis of randomness in the direction (P < 0.0001), indicating a significant grouping around zero for the differences in direction. We also verified that the small limb movements occurring at the onset of stimulation due to the kinematic redundancy of the three links system (thigh, shank, foot) did not affect the endpoint's force orientation by comparing the force vector direction at the very onset of stimulation (from 50 to 100 ms poststimulation onset, i.e., prior to movement onset) to the direction at a later point during the stimulus period (from 200 to 300 ms poststimulation onset, i.e., following movement). Differences in orientation were not significant (average difference was 1.6 ± 28.4°, P > 0.05 using paired t-test, SD reported).

The workspace was divided into triangles and forces in the sagittal plane produced at each of the vertices were interpolated to produce a vector plot of the forces at the paw throughout the space covered by the measurements. The endpoint forces were represented as two-dimensional vectors in the sagittal plane with the magnitude of the vector representing the active force magnitude (force due to muscle activation) and its direction representing the direction of the force. One force vector (F_x, F_y), as well as one position (x, y) coordinate was obtained from the forces at each corner of each triangle. By combining the three corners of a triangle, a system of six unknowns and six equations was produced, i.e

(1)

By using the a_i,j parameters associated with a triangle, the forces within that triangle were estimated. The triangulation was constructed to minimize the distance between the points interpolated and the vertices of each triangle.

Hindlimb muscle force patterns were also collected by directly stimulating each of the implanted muscles through the EMG electrodes after we collected spinal responses. Muscles were activated using trains of biphasic pulses (0.5 s trains, 1 to 4 mA amplitude, 100 μs pulse duration, and 40 Hz frequency) and forces at the paw were recorded for the 9–12 hindlimb positions. These measurements were obtained in two animals for eight of the muscles implanted: two tibialis anterior, two biceps femoris, two vasti, and two medial gastrocnemius. These forces were interpolated across the workspace as previously described.

Data were collected only while the preparation was stable. When the cat exhibited prolonged spontaneous muscle activity, or if blood pressure fell <60 mmHg, the experiment was terminated. The animal was euthanized via an intravenous administration of sodium pentobarbital and then intracardially perfused with 0.9% saline followed by 4% buffered paraformaldehyde. The spinal cord was then harvested for anatomical analysis.

Data analysis

FORCE PATTERN CLASSIFICATION.

Classification of force pattern types was done via cluster analysis. First, the active force vectors produced at a given stimulation site (1 per hindlimb position) were normalized such that the largest force vector measured within each force pattern had a magnitude of 1. The squared Euclidean distance between cases was used as the partitioning criterion. A case was defined as the set of forces at each of the nine positions for one pattern, i.e

(2)

with n representing the number of spinal sites at which forces were recorded. The squared Euclidean distance between cases was defined as , i.e., the squared sum of the differences in the x and y forces measured at each position between case i and case j.

To determine how many different types of force patterns were produced, force responses were partitioned using Ward's linkage hierarchical cluster analysis (Gordon 1999), and cases were grouped based on the results of the agglomeration schedule. Once the number of types of force patterns was determined, the force patterns obtained with a reduced number of limb configurations were classified by visual inspection into one of the types found with cluster analysis.

FORCE PATTERN PROPORTION AND LAMINAR DISTRIBUTION.

Once the number of force pattern types was determined and each pattern classified, the proportion and laminar distribution of force pattern types were compared between the four experimental groups, and to previously published data for the spinal-intact cat (Lemay and Grill 2004).

We first verified that similar portions of the cord were sampled in our four experimental groups. The segmental (L₅–L₇) and mediolateral distributions of the penetrations made were compared between the different groups using χ² analysis for the segmental distribution and ANOVA for the mediolateral distribution. The distribution of the dorsoventral depths at which patterns were measured was also compared between groups using ANOVA.

The frequency of each force pattern type was calculated as a percentage of the total number of patterns obtained in each group. χ² analysis was used to determine significant differences in the proportion of force pattern types between the five groups.

To determine a possible topographical organization in force pattern types, their laminar, rostrocaudal, and mediolateral distributions in the lumbar cord were examined. Responses in the spinal cord were divided into three regions: upper laminae, defined as surface (0) to 1,000 μm, intermediate (1,200–3,000 μm), and lower laminae (3,200–5,200 μm). The percentage of responses obtained at each of these three regions was compared between the groups via χ² analysis. The rostrocaudal segmental organization was also examined by comparing the distribution of force pattern types with respect to segmental level for each group using χ² analysis. The organization of the force pattern types with respect to mediolateral position was examined using regression analyses.

Finally, ANOVA was used to compare the stimulus amplitude and response magnitude as a function of penetration depth for the different groups. Stimulus threshold was compared between the spinal-intact (Lemay and Grill 2004) and chronic spinal cats using one sample t-test.

EMG ANALYSIS.

To process raw muscle activity (EMG) for analysis, the individual muscles' EMGs at each position were first rectified then divided into 10 ms bins and averaged over each bin (also see Lemay and Grill 2004 for further details). The time periods with no EMG activity were then extracted (i.e., before stimulus onset and once signal amplitudes were back to noise level). At each hindlimb endpoint, the binned EMG for each muscle was divided by the sum of the largest binned EMG from all muscles at that endpoint position to give its normalized relative amplitude. These normalized bins ranged from 0 to 1 and described the relative amplitude of each muscle's EMG with respect to the total EMG signal over time. Muscles with a normalized EMG <0.1 were considered inactive and not significant contributors to the force pattern. From these data, both the muscles that were active and inactive during stimulus delivery were tabulated. Multinomial logistic regression analysis was then used to determine relationships between the active muscles and force pattern type obtained with intraspinal microstimulation. Full factorial multinomial logistic regression model of the relationship between the force pattern type (independent variable) and muscle activity (dependent variable) was performed to evaluate the main effect of the active muscles and their interactions on the force pattern type.

The effect of hindlimb position on muscle activation during stimulus delivery was also examined. To achieve this, the normalized binned EMG for each active muscle at each endpoint position was concatenated to give the EMG vector at that position. The differences between the EMG vectors at the eight positions surrounding the mid-stance position and the vector at the mid-stance position were then obtained. The averages of these vectors were then compared with the deviations between EMG vectors at the mid-stance position obtained at the beginning and end of a force pattern measurement. Position was considered to significantly affect EMG levels if the average of the difference between the EMG vector at that position and the mid-stance position was larger than the value delineating the interval containing 95% of the differences in EMG vectors measured at the mid-stance position at the beginning and end of each force pattern measurement.

RESULTS

Force pattern classification

In 16 chronic spinal cats, 168 endpoint force patterns were obtained via intraspinal microstimulation (96 penetrations) within the L₅–L₇ spinal cord. Of these, 147 were obtained using nine hindlimb positions, 11 using a more elaborate grid (12 hindlimb positions), and 10 using a reduced grid (3–7 positions, 2 cats). Cluster analysis of the 147 complete (9 positions) active endpoint force responses yielded the same four force pattern types (see Fig. 1) obtained in the spinal-intact cat: caudal flexion (CF), caudal extension (CE), rostral flexion (RF), and rostral extension (RE) (Lemay and Grill 2004). Forces in the caudal flexion pattern were oriented up and toward the back of the animal, in a rostral flexion pattern they were oriented up and toward the front, in rostral extension they were oriented forward and down, and in caudal extension they were oriented backward and down. These four types accounted for 84% of the variance in the force response data. Only one convergent force pattern, a caudal extensor, was generated, and no diverging force pattern was observed.

FIG. 1.

Summary of the types of endpoint force patterns obtained via intraspinal microstimulation of the gray matter within the L₅–L₇ cord of spinal cats. Four types of force pattern types were obtained, rostral flexion (RF), caudal flexion (CF), rostral extension (RE), and caudal extension (CE). A drawing of the cat hindlimb (not to scale) is shown in A.

Extent of spinal cord area mapped

The L₅–L₇ spinal cord was sampled from the dorsal root entry zone to the midline and from the surface to 5,200 μm depth in 200 μm increments. The sites sampled are presented in Fig. 2 along with the maximal stimulus amplitude used at each site and the force magnitude obtained at that stimulation level. To confirm the uniformity of sampling throughout the L₅–L₇ segments between the spinal groups, χ² analyses and ANOVAs were performed. The analyses performed examined the relationships between segmental distribution of the penetrations and groups (χ²), between laterality of the penetrations and groups (ANOVA) and between depths of sites where force patterns were found and groups (ANOVA). These analyses confirmed that there was no significant difference in sampling topography between groups (P > 0.05 for all the tests mentioned). While the L₆ and L₇ segments were more extensively mapped than the L₅ segment (Fig. 2), one-way ANOVA found no significant difference in distribution of penetrations by segments between groups (P > 0.05). As stated in the preceding text, similar results were obtained for the laterality of the penetrations sampled and the depths of the force patterns. Thus similar portions of the lumbar spinal cord were sampled in each of the four groups. This similarity in the portion of the cord investigated in each group made it reasonable to compare the proportions of force pattern types obtained in each group.

FIG. 2.

Stimulation intensity (gray level) and force response magnitude (circle size) at each stimulation site in the L₅–L₇ cord for untrained cats with unmodified-fibroblasts graft (Op-control) or neurotrophin-secreting fibroblasts (NTF) graft and locomotor-trained cats with unmodified-fibroblasts graft (Trained) or NTF graft (Combination). The cord was extensively sampled dorsoventral- and mediolaterally in each of the 4 experimental groups (colored triangles). While the L₅ segment was sampled less frequently than the L₆ and L₇ segments, regression analyses confirmed that sampling within each segment was uniform and did not differ significantly between groups (P > 0.05). In addition, the stimulus level required to generate measurable endpoint force changed with penetration depth but was not influenced by group.

Proportion of each force pattern type

The proportions of each of the active force pattern types obtained with ipsilateral intraspinal microstimulation were compared among the four experimental groups. The proportions of each force pattern type for the groups (Op-control, Trained, NTF, and Combination) were represented in pie chart form (Fig. 3). In the Op-control group, 34 patterns were elicited from 21 penetrations from three cats. Sixty-eight patterns were obtained from 31 penetrations in five trained cats while 43 patterns were produced from 30 penetrations in five cats from the NTF group. In three cats with combination treatment, spinal mapping yielded 23 force patterns from 14 penetrations. The proportions of CE, CF, RE, and RF force patterns obtained from cats in each of the four experimental groups were compared with the proportion of these force pattern types obtained in the spinal-intact cat (Fig. 3A).

FIG. 3.

A comparison of the proportions of each of the endpoint force pattern types obtained from spinal-intact cats (A) and cats from the 4 experimental groups (B–E). The proportions of force pattern types in Op-control (B), Trained (C), NTF (D), and Combination (E), were significantly different from those in the spinal-intact cat (A, χ², P < 0.05). Force pattern type proportions were not significantly different among the 4 spinal groups. Force pattern type proportions in cats receiving daily training were significantly different from those in cats that received no training (χ², P < 0.05). There was no effect of the neurotrophic factor secreting transplant on force pattern type proportions (χ², P > 0.05). Data for spinal-intact cats from Lemay and Grill (2004).

The proportions of each of the force pattern types obtained from each of the spinal cat groups (Op-control, Trained, NTF, and Combination groups) were significantly different from that of spinal-intact cats (χ², P < 0.05), but no significant differences were observed between spinal cat groups (χ², P > 0.05). There was a striking increase in the proportion of RF responses in the chronic spinal cat (χ², P < 0.01, all spinal cats). Although RF responses were rare (1 in 67 or 1.4%) in spinal-intact cats (Lemay and Grill 2004), they occurred frequently (94 in 168 or 55.9%) in spinalized cats (Table 1). In addition, CF responses were significantly reduced in chronic spinal cats compared with spinal-intact cats (χ², P < 0.01, all spinal cats). The number of RE and CE responses was relatively unchanged in the spinal cats compared with the spinal-intact cats (χ², P > 0.05) except that no RE type responses were obtained from cats in the combination group. Thus chronic spinalization altered the proportion of endpoint force patterns elicited via intraspinal microstimulation irrespective of treatment. As noted in the preceding text, no difference in the proportion of force pattern type emerged between the four experimental groups (χ², P = 0.054).

TABLE 1.

Summary of endpoint force pattern type

Group	Force Pattern Type
	RF	RE	CF	CE
Op-control	26	1	1	6
Trained	30	5	21	12
NTF	28	2	8	5
Combination	10	0	8	5
Total	94	8	38	28

Summary of the number of each endpoint force pattern types obtained via intraspinal microstimulation within the L₅–L₇ cord of spinal cats for the four experimental groups. RF and RE, rostral flexion and extension, respectively; CF and CE, caudal flexion and extension, respectively; Op-control, cats receiving no intervention; Trained, cats receiving control fibroblasts and daily treadmill training; NTF, cats receiving neurotrophic-factor-secreting cells and no training; Combination, cats receiving NTF-secreting graft and treadmill training.

We also combined experimental groups based on training or transplant paradigm to determine any separate effect of training or neurotrophin administration on the proportion of force pattern type expressed with stimulation. There was a significant difference in the proportion of force pattern types obtained in the subset of cats receiving daily treadmill training compared with those that received no training (χ², P < 0.05). However, no significant difference in the proportion of force pattern types was found between cats with or without the neurotrophin secreting graft.

Thus neither of the interventions used, training or neurotrophins, was sufficient to restore the proportion of endpoint force pattern types to preinjury levels. Training, however, influenced the proportion of the types of endpoint force patterns while neurotrophic factor administration had no effect on the proportion of the types of force patterns elicited.

The variability in the number of each of the force pattern types obtained from each cat was also compared (Table 2). The number of RF, RE, and CE force patterns did not differ significantly between animals (χ², P > 0.05). However the number of CF force patterns obtained from the fifth trained cat (Train-5) was significantly greater than that obtained from the other cats. Overall, similar numbers of each of the force pattern types was obtained in the 16 animals, justifying our grouping of the animals.

TABLE 2.

Force pattern types obtained via intraspinal microstimulation of the L₅–L₇ cord

Cat ID	Force Pattern Type
	RF	RE	CF	CE
OC-1	7	1	0	3
OC-2	10	0	1	3
OC-3	9	0	0	0
Train-1	5	0	2	3
Train-2	4	1	0	2
Train-3	3	0	6	5
Train-4	13	2	0	0
Train-5	5	2	13	2
NTF-1	11	0	0	1
NTF-2	5	0	0	0
NTF-3	2	0	0	0
NTF-4	9	0	0	0
NTF-5	1	2	8	4
Combo-1	0	0	5	1
Combo-2	6	0	1	4
Combo-3	4	0	2	0

The number of force pattern types obtained via intraspinal microstimulation of the L₅–L₇ cord for each spinal cat studied. There were no significant differences in the number of RF, RE, and CE force patterns between the individual cats (χ², P > 0.05). However, the number of CF force patterns obtained from the fifth trained cat (Train-5) was significantly greater than from the other animals (χ², P ≤ 0.05).

Segmental response topography

Table 3 illustrates the number of penetration sites and the number of force patterns obtained at each lumbar level for each group of spinal cats. The majority of force patterns were collected from penetrations made in the L₆ and L₇ segments. χ² analysis of force pattern type distribution for spinal cats, with respect to rostrocaudal segmental level, showed no effects of segmental level. χ² analyses for each of the individual spinal groups showed small significant effects of spinal levels on proportions of force pattern types for the NTF (P = 0.046) and Combination (P = 0.043) groups but not for the other groups. A logistic loglinear model (Christensen 1997) with pattern type as the dependent variable and spinal group and segmental levels as factors, performed using SPSS 11 (SPSS, Chicago, IL) for Macintosh (Apple Computing, Cupertino, CA), showed no significant correlation between segmental level and pattern type. Overall the distribution of force pattern types did not seem to vary much with spinal level.

TABLE 3.

Summary of the microstimulation data, per lumbar region, for each of the four experimental groups

Group	n	Lumbar Region	Penetrations	No. of Force Patterns	Force Patterns, %
Op-control	3	L5	1	2	5.9
		L6	10	18	52.9
		L7	10	14	41.2
Trained	5	L5	2	5	7.4
		L6	19	40	58.8
		L7	10	23	33.8
NTF	5	L5	7	9	20.9
		L6	14	19	44.2
		L7	9	15	34.9
Combination	3	L5	4	6	26.1
		L6	4	8	34.8
		L7	6	9	39.1

Further, there were few noticeable differences in segmental response localization in cats from the experimental groups, compared with spinal-intact cats. The largest difference observed was for RF responses (•), which were rare and restricted to the dorsal laminae in spinal-intact cats (Fig. 4A), but occurred frequently in chronically injured cats (Fig. 4, B–E). These responses were widely distributed in the chronic spinal cat. The incidence of CF responses (▴) was reduced in spinal cats, but they were still primarily located in dorsal laminae. The number of CE responses (▵) was unchanged, and these responses were predominantly present within intermediate and ventral laminae of the L₅–L₇ spinal cord, similar to the spinal-intact cat. RE responses (○) were found at dorsal and ventral sites as was the case in the spinal-intact cat (Fig. 4).

FIG. 4.

Location of spinal sites that generated endpoint force patterns on intraspinal microstimulation of the L₅–L₇ cord. Data from the spinal-intact cat (A) represent ipsilateral microstimulation sites and were adapted from Lemay and Grill (2004). Compared with the spinal-intact group (A), there was a noticeable increase in rostral flexion responses (•) and decrease in caudal flexion responses (▴) in Op-control (B), trained (C), NTF (D), and combination (E) groups at all lumbar segments.

Therefore chronic spinal cord injury did not grossly alter the topography of endpoint force patterns elicited via intraspinal microstimulation in the L₅, L₆, or L₇ segments compared with the spinal-intact spinal cord but did affect the proportion at which each of the force pattern types was expressed.

Laminar distribution of force pattern types

To further examine the effect of chronic spinalization on the location of responses within the L₅–L₇ spinal cord, the percentages of endpoint force responses obtained at dorsal (0–1,000 μm), intermediate (1,200–3,000 μm), and ventral (3,200–5,200 μm) laminar regions for the pooled L₅–L₇ cord were compared (Fig. 5). χ² analysis demonstrated that the percentages of force patterns at dorsal laminae were significantly greater than at ventral laminae in chronic spinal cats (all spinal cats grouped; χ², P < 0.05) compared with the percentages obtained in spinal-intact cats at these regions. There was no difference in the percentage of response at intermediate laminae between the spinal-intact and chronic spinal groups of cats (Fig. 5). Thus chronic spinal cord injury shifted the distribution of microstimulation sites in the L₅–L₇ spinal cord that produced endpoint force patterns to more dorsal laminae.

FIG. 5.

Distribution of endpoint force patterns within dorsal (0–1,000 μm), intermediate (1,200–3,000 μm), and ventral (3,200–5,200 μm) laminar regions of the lumbar spinal cord of spinal and spinal-intact cats derived from pooling the patterns obtained via microstimulation of the L₅–L₇ cord. The percentage of force patterns elicited in spinal-intact cats at dorsal and ventral laminae was less than that obtained from the spinal cats (χ², P < 0.05). No significant difference in force pattern distribution was obtained among Op-control, Trained, NTF, and Combination groups of cats (χ², P > 0.05). Data for spinal-intact cats (intact) modified from Lemay and Grill (2004).

The percentage of force patterns elicited at dorsal, intermediate and ventral depths was also compared between the four individual groups of spinal cats. For the NTF group of cats, there was a trend toward an increase in the percentage of responses obtained at dorsal laminae (0–1,000 μm) compared with Op-control, Trained, and Combination groups (χ², P = 0.054). In NTF cats, there was also a trend toward a reduction in the number of responses elicited at the deepest laminae (χ², P = 0.053). No differences were found within intermediate laminae. Therefore there were no significant differences in the laminar distribution of force patterns obtained from intraspinal microstimulation of the L₅–L₇ spinal cord in spinal cats.

Mediolateral distribution of force pattern types

The effect of the mediolateral position of the stimulation site on the force pattern type produced was investigated using regression analyses. Linear regression using mediolateral force pattern position as independent and force pattern type as the dependent showed no significant effect (P > 0.05) of the stimulation site's lateral position on the force pattern type obtained.

Muscle combinations and force pattern type

Multinomial logistic regression analysis was used to determine possible correlation between the limited sample of active muscles and the endpoint force pattern types produced by intraspinal microstimulation of the L₅–L₇ spinal cord. Because there was no effect of experimental groups on the muscle combinations used in the generation of the endpoint force patterns collected (multinomial logistic regression between groups and muscle combinations), the data from Op-control, Trained, NTF, and Combination groups were pooled to increase the power of the regression model.

A full factorial regression model was used to predict the endpoint force pattern type (dependant variable) based on the active muscles (i.e., normalized binned EMG >0.1, see methods; Table 4). The model predicted RE and RF force patterns within 71.4 and 97.5% accuracy, respectively, but was not a good predictor of CE and CF patterns, indicating that our small sample of muscles could be used to predict the RE and RF force pattern types from the active EMG but was too limited to fully predict force pattern outcome.

TABLE 4.

Classification: pooled data from all four spinal groups

Observed	Predicted
	CE	CF	RE	RF	Correct, %
CE	1	0	1	18	5.0
CF	0	0	1	30	0.0
RE	0	0	5	2	71.4
RF	1	0	1	79	97.5
Overall Percentage	1.4	0	5.8	92.8	61.2

Classification of endpoint force pattern types, CE, CF, RE, and RF based on hindlimb muscles active during intraspinal microstimulation of the L₅–L₇ spinal cord. The regression analysis was performed on pooled data from all spinal cats. Vastus lateralis, biceps femoris posterior, medial gastrocnemius, and tibialis anterior muscle activity were used as predictors of force pattern types.

Endpoint position effects on EMGs

The cat's hindlimb was placed at 9 to 12 positions in the sagittal plane during the mapping of a force pattern. To examine the effect of hindlimb (endpoint) position on hindlimb muscle activity, the differences between the EMG vectors obtained at the endpoint positions surrounding the mid-stance position and the ones at the mid-stance position were examined. EMG vectors for 139 spinal stimulation sites were analyzed. One muscle was active in 20 of these vectors, two muscles were active in 42 vectors, three muscles were active in 49 vectors, and four muscles were active in 28 vectors.

Average differences between the EMG vectors at the surrounding positions and the one at the mid-stance position were considered significantly different from zero if they were outside the interval containing 95% of the differences between EMG vectors at the mid-stance positions taken at the beginning and end of each force pattern measurements. We had repeated EMGs measured at the mid-stance position for 98 spinal sites and the distribution had over 16,000 points. The average difference was 0.00 ± 0.09, but the distribution was highly peaked so that 95% of the values were within 1 SD as opposed to 2 SD in the normal distribution.

We found that of 139 spinal sites, only 15 showed some modulation of the EMG activity with position, and even in those cases, only one to two positions showed a significant difference, i.e., >0.09. Therefore changes in limb configuration seem to have little influence on the relative activation of muscles with respect to one another.

Stimulus amplitude at a given force output

The threshold for activation of the lumbar spinal cord was low at superficial penetrations (0–1,000 μm). At this depth, the stimulus amplitude required to activate the spinal cord, and elicit an endpoint force was <50 μA (Fig. 6A). With deeper penetrations, it was increasingly difficult to activate the spinal cord. To compensate for this, the stimulus amplitude used was increased to generate a response at the endpoint. Therefore at intermediate (1,200–3,000 μm) and ventral (3,200–5,200 μm) penetrations, the amplitude required was significantly greater (ANOVA, P < 0.05). This trend was observed for all four groups of spinal cats. The endpoint force produced with increasing stimulation amplitude was not altered at any of the depths tested, however (Fig. 6B). This suggests that the threshold required for activation of the spinal cord varied with depth in the chronic spinal cat and was lower than in spinal-intact cats for dorsal and intermediate penetrations (t-test, P < 0.05).

FIG. 6.

Average stimulus amplitude (A) utilized to elicit similar endpoint force (B) during intraspinal microstimulation of the lumbar spinal cord. The average stimulus amplitude required to activate the spinal cord at dorsal (0 –1,000 μm) and intermediate (1,200–3,000 μm) penetrations was low (<50 μA; inset A, average for all spinal cats). However, at ventral (3,200–5,200 μm) depths, the threshold to generate endpoint force with intraspinal microstimulation increased, requiring increased stimulation amplitude (ANOVA, P < 0.05). There were no group differences at each depth (A). In addition the average force elicited with intraspinal microstimulation at each depth was similar for all groups (B).

DISCUSSION

The goal of this study was to characterize the changes in endpoint forces elicited by intraspinal microstimulation of the cat lumbar spinal cord following spinal transection and treatments that restored treadmill locomotion. As previously observed in other spinalized species (Giszter et al. 1993; Tresch and Bizzi 1999) and in spinal-intact cats (Lemay and Grill 2004), we found that the hindlimb endpoint forces grouped into a small number of directionally arranged patterns, suggesting a modular organization to the spinal cord motor output. Spinalization had no effect on the types of force patterns expressed by microstimulation, but stimulation threshold was lowered and the proportion at which each force pattern type was expressed changed. Neurotrophins released by transplants of genetically modified fibroblasts (Boyce et al. 2007; Himes et al. 2001; Liu et al. 1999a) had no effect on these changes, but locomotor training had a small effect on the frequency of expression of the different types.

Modularity in force patterns evoked by intraspinal microstimulation in spinal cats and its relationship to natural behavior responses

The same four force pattern types were obtained in spinal-intact cats and chronic spinal cats using intraspinal microstimulation of the lumbar spinal cord (Fig. 1, A–D). This result suggests that as proposed by the “primitive” hypothesis, the endpoint force patterns are “hard-wired” into the spinal cord and may be used to accelerate the learning of walking (Giszter et al. 2000) and simplify the supraspinal movement control by acting as building blocks of natural movement (Kargo and Giszter 2000).

As in the frog (Giszter et al. 1993), the force responses obtained with intraspinal microstimulation match well with the limb endpoint forces in natural movements. Ground reaction forces observed in response to postural disturbances in the cat (Ting and Macpherson 2005; Torres-Oviedo et al. 2006) show modularity in both the force vector orientation and the muscle synergies used to counteract the disturbances. The ground reaction force vectors in the sagittal plane show preferred direction in rostral flexion and caudal extension (Torres-Oviedo et al. 2006), similar to two of our responses (RF and CE). However, no caudal flexion or rostral extension was observed for any of the postural disturbances (translations and rotations) used in those studies (Ting and Macpherson 2005). These types of responses may not be necessary with the types of disturbances used by Ting and Macpherson (2005), but caudal flexion is highly reminiscent of the flexion withdrawal reflex (Sherrington 1910) and rostral extension would provide useful ground reaction forces for braking.

Torres-Oviedo et al. (2006) also examined the muscle synergies activated during the postural corrective response to linear and translational perturbations in adult cats. Only four muscle synergies accounted for the corrective response. There was also a significant correlation between muscle synergy activation in a particular direction and the force vectors produced for the same perturbation, suggesting that the muscle synergies coded for limb endpoint force. Thus muscle synergies further simplify supraspinal control of limb movements during natural behaviors such as postural control (Ting and Macpherson 2005) and hindlimb movements (Hart and Giszter 2004). In our study, experimental groups showed a significant correlation between muscle combinations and endpoint force patterns for some of the force pattern types but not all, which may be due to our limited sampling of hindlimb muscles.

In these experiments, the relative activation of each muscle within a synergy did not seem to be affected by changes in limb configuration. Possible variations in EMG electrode pick-up with changes in muscle length prevent us from making any conclusion about possible variations in activation levels due to changes in limb configuration and limit our conclusions to the balance of activity level within the active muscles. This contradicts our previous observation in spinal-intact cats (Lemay and Grill 2004) but confirms the observation that deafferentation had no effect on muscle activation in the frog (Loeb et al. 1993).

Our conclusions in the spinal-intact cats were based on a cluster analysis of the EMG vectors and may have been biased by the analysis method. The use of similar clustering techniques on the data in spinal cats also divides the data into two or more clusters for most spinal sites, suggesting a modulation of the relative activation of the muscles involved with position. The current analysis is more statistically stringent and does not show significant modulation of the relative activation levels of the muscles within a synergy with changes in position. These results and the ones in the spinal frog would suggest that the balance in the activation of muscles within a synergy is independent of limb configuration/muscle length changes.

Modularity in force patterns evoked by intraspinal microstimulation in other intraspinal microstimulation studies

Other reports with intraspinal microstimulation of the spinal cord have concentrated on describing their responses in terms of the movements evoked rather than on forces at the endpoint as reported here. Barthélemy et al. (2006) show clear rostral flexion or caudal extension movements in spinal cats. Because some movements produced contact with the treadmill mounted under the paw, it is difficult to evaluate if some of the extensor movements might have been rostral extensor. Tai et al. (2003) also report modularity in the movements generated via intraspinal microstimulation of the L₅–S₁ spinal cord in spinal-intact anesthetized cats. In the sagittal plane, the authors report flexion and extension movements (in addition to abduction and adduction in the sagittal plane). The extension movement presented is again a caudal extensor. One flexor movement appears to be a rostral flexor as the toe moves forward and up (Fig. 3 of Tai et al. 2003) and the other a caudal flexor as the toe moves slightly backward and up (Fig. 8 of Tai et al. 2003). The authors do not make a distinction between the two types. In chronic spinal cats, Saigal et al. (2004) showed caudal and rostral flexor movements in addition to weight-bearing extension. From their previous work on anesthetized spinal-intact animals, the extension movements appear to be directed backwards (Mushahwar et al. 2002), although movements in all directions could be obtained in spinal-intact anesthetized animals through stimulation of the intermediate zone of the lumbar gray (Aoyagi et al. 2004a). Thus the modularity in force pattern types seen in our results is similar to the modularity in movement direction seen with intraspinal microstimulation in other laboratories (Mushahwar et al. 2002; Saigal et al. 2004; Tai et al. 2003) and the reaction forces to postural disturbances in natural behavior (Ting and Macpherson 2005; Torres-Oviedo et al. 2006).

The modularity may arise from the properties of the neuromuscular system as suggested by some authors (Aoyagi et al. 2004b) or from co-activation of muscles based on the topography of their motoneuron pools (Mushahwar et al. 2004). The hypothesis that force patterns result from co-activation of muscles whose motoneurons are located in proximity to one another in the cord and activated simultaneously has been argued against using a topological model of the motoneuron pool organization in the cat (Lemay et al. 2007) and also in the frog (Giszter et al. 1993). The force pattern types produced using this strategy did not resemble the ones obtained experimentally with stimulation of the lumbar intermediate gray in either species. As shown by Aoyagi et al. (2004a), stimulation of major muscle groups, nerves, or roots all produced movements in a limited number of preferred directions. It is thus likely that the neuromuscular system plays a significant role in determining the force pattern orientation. It is also worth noting that some directions that can be obtained with pure muscle or nerve activation, such as horizontal movements, were never observed (rostral force pattern) or rarely (caudal force pattern) observed with intraspinal microstimulation in our experiments. It thus seems that the spinal cord is organized to autonomously use only a subset of the force/movement orientations that can be produced by the muscular system. This is in agreement with the limited behaviors that can be obtained in spinal animals versus the repertoire of movements available in spinal-intact animals.

The topographical localization of responses reported by the various groups using intraspinal microstimulation is also remarkably similar to our results. Sites producing flexion are located dorsally, either in the dorsal columns or dorsal gray (Barthélemy et al. 2006; Mushahwar et al. 2002; Tai et al. 2003), whereas sites producing extension are located more ventrally (ventral gray or white matter) (Mushahwar et al. 2002). We saw no clear distribution in the rostrocaudal localization of responses, as other groups have (Barthélemy et al. 2006; Tai et al. 2003). However, the dorsoventral distribution of responses was segregated. Although we still observed a number of extensor responses in the dorsal or intermediate gray, chronic spinalization clearly segregated the localization of the flexor and extensor responses to the dorsal and ventral cord, respectively. This result was in agreement with results of Mushahwar et al. (2004), who showed silencing of some spinal sites producing extension when the decerebrated animal was acutely spinalized. Predominantly flexor responses were also obtained from microstimulation experiments performed in the chronic spinal rat, although the flexion responses were strictly caudally oriented (Tresch and Bizzi 1999). Removal of the supraspinal influence thus seems to bias the spinal motor output toward flexion.

The proportion of each type of endpoint force pattern was similar in Op-control, Trained, NTF, and Combination cats but significantly different from that of spinal-intact cats. Therefore chronic injury altered the proportions of the type of force pattern elicited with intraspinal microstimulation of the lumbar cord, and treatments that improved locomotion in spinal cats had limited effects on the proportions of force pattern types. This was surprising given the profound effects on locomotor recovery obtained with neurotrophins alone or with exercise (Boyce et al. 2007). Barthélemy et al. (2006) show greater changes in the number of dorsal column sites producing locomotor responses than in the number of gray matter sites producing nonlocomotor responses after chronic spinalization and treadmill training. Because dorsal column stimulation likely activates a larger volume of neural tissue than focal gray matter stimulation, both theirs and our results would suggest that the changes responsible for the locomotor improvements in animals receiving training or neurotrophic factors are distributed throughout the lumbar area.

Neural elements activated and role of biomechanics

The volume of neural tissue activated via intraspinal microstimulation is relatively large [50 μm radius sphere at 10 μA; 250 μm radius sphere at 50 μA (Gustafsson and Jankowska 1976), and subject to interpretation (see Lemay and Grill 2004) for further discussion]. Flexion responses mostly occurred in response to stimulation in laminae II–IV, known regions of termination of cutaneous afferents (Levinsson et al. 2002). Stimulation in this area may activate the incoming afferents or first-order inter/projection-neuron populations, as the thresholds for pre- and postsynaptic elements are similar (Baldissera et al. 1972; Gustafsson and Jankowska 1976; Jankowska et al. 1975; McIntyre and Grill 2002). Interestingly, flexion responses can still be evoked in deafferented animals when stimulating in the dorsal regions of the cord (Tresch and Bizzi 1999), suggesting that it is the postsynaptic elements that are responsible for the observed responses. A mix of extension and flexion responses were evoked by stimulation in lamina VII, which contains a number of first-order inhibitory interneurons that mediate reciprocal inhibition of motoneurons and are activated by the Ia and Ib afferents. This intermediate region appeared quieter in spinal animals, although we found no significant differences in the proportion of responses found at those depths. Responses at the more ventral depths were mostly extensor responses as reported by the Mushahwar group (Mushahwar et al. 2002).

While the femur was fixed in our previous force pattern measurements (Lemay and Grill 2004), these force patterns measurements were taken with the femur free, which could be suggested as an explanation for the proportion of rostral flexor patterns observed. Additional experiments with seven decerebrated spinal-intact animals performed with a similar robotic apparatus and the femur free showed an increase in caudal flexor response over caudal extension responses but no change in the proportion of rostral flexor responses (Lemay and Grill, unpublished data). Of 20 force patterns evoked by ipsilateral stimulation, we obtained 16 caudal flexor (80%), 3 caudal extensor (15%), 1 rostral extensor (5%), and no rostral flexor (0%). With the femur fixed, we obtained (for ipsilateral stimulation) 61% caudal flexor (31/51), 31% caudal extensor (16/51), 6% rostral extensor (3/51), and 2% rostral flexor (1/51) (Lemay and Grill 2004). Thus changes in the mechanics of the leg do not appear to be the main reason for the change in proportion of rostral flexor pattern obtained. The spinal circuitry in the L₅–L₇ cord is affected by the spinalization in a way that promotes rostral flexion responses. The changes may be due to a decrease in activation threshold in the dorsal laminae. By using less current in spinalized animals, we are activating a smaller volume and may have then uncovered a predominance of rostral flexor responses encoding that was masked by the higher activation levels needed in spinal-intact animals. Caudal flexors are more a protection response (they are similar to a flexion withdrawal response) and may be encoded at a gross level in the cord so that stimulation of a large volume of neural tissue may predominantly activate this protection mechanism; while rostral flexors, which can more readily be used in a locomotor pattern, by driving the limb forward, may be encoded in a finer topographical organization.

Conclusions

The endpoint force patterns elicited via intraspinal microstimulation are hypothesized to be movement “primitives” (Giszter et al. 1993) that can be used as building blocks of movements. We demonstrated that although their proportions are significantly altered with spinal cord injury and training, the basic patterns are maintained following injury. Untrained animals that received neurotrophins recovered locomotor capacity, although we found that neurotrophins had no significant effect on the types or proportions of force pattern type evoked with stimulation. In addition, Op-control cats were incapable of locomotor function yet they exhibited force pattern proportions that were not statistically different from those groups of spinal cats capable of locomotion. It appears that the manifestation of these primitives in the lumbar area does not necessarily correlate with locomotor function. It may be that modules of locomotion are preserved postspinalization but that the circuitry responsible for activating these modules is located rostrally (Langlet et al. 2005). Our results would indicate that the building blocks of movements are maintained regardless of loss of supraspinal input and that their relative proportions may be influenced by some of the treatments.

These findings give greater insight into the types of motor changes that occur in the spinal cord with chronic injury and following the rehabilitative interventions used, treadmill locomotor training, and neurotrophin administration. The experimental paradigm used, however, limits any possible conclusions to the L₅–L₇ cord because this is the region that was mapped and so may not be generalized to the global changes occurring in the spinal cord caudal to a chronic injury.

GRANTS

This research was supported by National Institute of Neurological Disorders and Stroke Grants NS-41975 and NS-24707, Pennsylvania Department of Health, and The Christopher and Dana Reeve Foundation.