Journal of Neurophysiology

Bilateral oscillatory hip movements induce windup of multijoint lower extremity spastic reflexes in chronic spinal cord injury

Tanya Onushko, Allison Hyngstrom, Brian D. Schmit


After spinal cord injury (SCI), alterations in intrinsic motoneuron properties have been shown to be partly responsible for spastic reflex behaviors in human SCI. In particular, a dysregulation of voltage-dependent depolarizing persistent inward currents (PICs) may permit sustained muscle contraction after the removal of a brief excitatory stimulus. Windup, in which the motor response increases with repeated activation, is an indicator of PICs. Although windup of homonymous stretch reflexes has been shown, multijoint muscle activity is often observed following imposed limb movements and may exhibit a similar windup phenomenon. The purpose of this study was to identify and quantify windup of multijoint reflex responses to repeated imposed hip oscillations. Ten chronic SCI subjects participated in this study. A custom-built servomotor apparatus was used to oscillate the legs about the hip joint bilaterally and unilaterally from 10° of extension to 40° flexion for 10 consecutive cycles. Surface electromyograms (EMGs) and joint torques were recorded from both legs. Consistent with a windup response, hip and knee flexion/extension and ankle plantarflexion torque and EMG responses varied according to movement cycle number. The temporal patterns of windup depended on the muscle groups that were activated, which may suggest a difference in the response of neurons in different spinal pathways. Furthermore, because windup was seen in muscles that were not being stretched, these results imply that changes in interneuronal properties are also likely to be associated with windup of spastic reflexes in human SCI.

  • spasticity

involuntary muscle spasms are a prominent component of the spastic syndrome that develops in chronic human spinal cord injury (SCI). Individuals with SCI will typically present with uncontrolled hyperexcitable reflexes, such as flexor and extensor spasms, in which brief stimuli can evoke forceful, prolonged responses throughout muscles in the lower extremities lasting from a few seconds to tens of seconds (Benz et al. 2005). Particular to extensor spasms, hip proprioceptive stimuli elicit long-lasting, coordinated responses of hip flexion, knee extension, and ankle plantarflexion (Schmit and Benz 2002) and typically occur when patients shift from a seated to supine position (Macht and Kuhn 1948) or by extension of the leg (Schmit and Benz 2002). Although not recognized as part of the extensor spasm, flexing the hip will also cause a coordinated response, generally composed of hip extension, knee flexion, and ankle extension with responses equally as long as those seen in extensor spasms (Schmit and Benz 2002). Extensor spasms, which are most common in SCI patients, are functionally relevant, since they not only interfere with patient mobility but also cause pain and discomfort (Little et al. 1989). Accordingly, it is important to understand the underlying mechanisms associated with these prolonged responses to prescribe appropriate treatment options to improve function.

The neuropathophysiology behind involuntary muscle spasms in human SCI still remains unclear, but one hypothesis is that changes in the intrinsic electrical properties of spinal circuitry contribute to the spastic syndrome (Bennett et al. 2004; Nielsen et al. 2007). In an intact nervous system, descending monoaminergic input from the brain stem to the spinal cord modulates spinal neuronal excitability through the facilitation of persistent inward currents (PICs) found in both motoneurons (MNs) (Hounsgaard et al. 1988; Lee and Heckman 1999) and motor-related interneurons (INs) (Chen et al. 2001; Theiss et al. 2007). In mammals, it is believed that PICs are primarily generated by dendritic voltage-dependent Na+ and Ca2+ channels that, when activated by a brief excitatory stimulus, can amplify synaptic currents (Hultborn et al. 2003; Lee and Heckman 2000) and produce plateau potentials resulting in sustained firing even after the stimulus is removed (Bennett et al. 1998a; Crone et al. 1988; Schwindt and Crill 1980). PICs are associated with a phenomenon called “windup” in which repeated excitatory inputs can facilitate MN activity without increases in synaptic input (Bennett et al. 1998a; Russo and Hounsgaard 1994; Svirskis and Hounsgaard 1997).

Immediately following spinal injury in animals, MN excitability is greatly decreased with concomitant loss of PICs (Bennett et al. 1999; Hounsgaard et al. 1988; Hyngstrom et al. 2008b), and this has been associated with an acute interruption of monoaminergic inputs to the cord. Despite the loss of descending drive, a reemergence of PICs occurs weeks to months following injury, along with a concurrent increase in MN excitability and emergence of spasms (Bennett et al. 2001; Li and Bennett 2003). Constitutive activity in serotonergic (Murray et al. 2010) and adrenergic receptors (Rank et al. 2011) are in part responsible for PIC recovery, and the abnormal regulation of recovered PICs is likely to be involved in the long-lasting muscle spasms observed in chronic spinalized animals (Bennett et al. 2001, 2004; Li et al. 2004; Murray et al. 2010). These changes in MN properties in spinal cord-injured animals are consistent with clinical descriptions of involuntary spasms in human SCI (Collins et al. 2001; Gorassini et al. 2004; Hornby et al. 2003, 2006; Nickolls et al. 2004). Alterations in intrinsic properties of INs, which have not been investigated as extensively as with MNs, also demonstrate electrophysiological characteristics similar to PICs in MNs (Russo and Hounsgaard 1996; Smith and Perrier 2006; Theiss et al. 2007; Ziskind-Conhaim et al. 2008), including windup (Russo and Hounsgaard 1994). Since MNs receive much of their synaptic input from INs, and IN axons show the ability to regenerate post-SCI (Fenrich and Rose 2009), it is plausible that changes in IN PICs would have downstream effects on motor behaviors as well, especially since the spastic syndrome is in large part composed of long-lasting, multijoint flexor and extensor spasms. Thus, after spinal injury, changes in the excitability of interneuronal networks may affect reflex muscle activity and the associated emergence of long-lasting motor output (Jankowska and Hammer 2002).

In human SCI, windup has been examined with repeated stretch of the ankle (Hornby et al. 2006) and in flexor reflex responses to repeated, brief stimulation of the skin of the foot (Hornby et al. 2003), and results have contributed to the understanding of clinical behaviors of single-joint clonus and flexor spasms. However, windup behavior has not been explored in extensor spasms, and it is plausible that underlying plateau potentials may play a large role in the prolonged, nonlinear responses associated with this spastic multijoint reflex. In this report, we use the term “windup,” similar to the phrase “warm-up” (Bennett et al. 1998a; Mendell 1966), to describe the behavior of increasing motor output (in this study quantified by torque and EMG) in response to repeated excitatory synaptic inputs. In the current study we investigated whether the multijoint reflexes associated with imposed hip movement in human SCI (i.e., extensor spasms; Schmit and Benz 2002) would exhibit windup. We hypothesized that windup responses in joint torque and muscle activity would occur throughout the lower extremities. Specifically, peak torque and muscle activity acquired through surface electromyograms (EMGs) would be greater with subsequent hip movements in a nonlinear manner, implicating PICs in extensor spasms.

Results from this study have been published in abstract form (Onushko et al. 2008).


Participants included 10 individuals with chronic (>6 mo), clinically complete [American Spinal Injury Association (ASIA) classification A], and incomplete SCI (ASIA B, C, or D). During the time of this study, three subjects were prescribed antispastic medication (e.g., baclofen, a GABA agonist) to control the frequency and intensity of their spasms. Subject information is summarized in Table 1. Exclusion criteria for this study included lower extremity nerve injury or injury below spinal cord segments innervating the hip region, significant medical complications due to skin breakdown, urinary tract infection, heterotopic calcification, significant osteoporosis, other concurrent illnesses limiting the capacity to conform to study requirements, or the inability to give informed consent. In addition, five subjects with no reported neurological injury volunteered for this study. Informed consent was obtained before study participation, and all procedures were conducted in accordance with the Helsinki Declaration of 1975 and were approved by the Institutional Review Board of Marquette University.

View this table:
Table 1.

Subject clinical characteristics

The details of the experimental setup have been previously described in detail (Onushko and Schmit 2007). Briefly, study participants lay supine with both legs secured within custom-designed, adjustable leg braces. Since multijoint extensor reflexes are triggered by imposed hip movements (Schmit and Benz 2002), the knee and ankle joints were held isometrically within the leg braces in slightly flexed positions (∼22° knee flexion and 13° ankle plantarflexion) to isolate the reflex responses to hip joint rotations. The leg braces were attached to a novel robotic apparatus that used servomotor drive systems (Kollmorgen, Northampton, MA) to generate oscillatory movements of the legs about the hip joints. Torque transducers (S. Himmelstein and Company, Hoffman Estates, IL) measured sagittal plane hip, knee, and ankle torque from both legs while hip position was monitored using optical encoders (US Digital, Vancouver, WA). Custom-written LabVIEW software (National Instruments, Austin, TX) was used to control hip trajectory and acquire torque and position signals. All signals were low-pass filtered (500 Hz) before data acquisition and sampled at 1,000 Hz using a data acquisition card (National Instruments) and a personal computer.

Disposable, pregelled Ag-AgCl surface electrodes (Vermed Medical, Bellows Falls, VT) were used to measure EMG of the participant's legs. Electrodes were placed in a bipolar arrangement over the cleaned muscle belly of the following muscles of both legs: vastus medialis (VM) and lateralis (VL), rectus femoris (RF), medial hamstrings (MH), medial gastrocnemius (MG), soleus (Sol), and tibialis anterior (TA). EMG signals were amplified (×1,000), band-pass filtered (10–1,000 Hz; Bortec Medical AMT-16, Calgary, Alberta, Canada), and sampled (1,000 Hz) using the same data acquisition card (National Instruments) and personal computer used to acquire the torque/position signals.

Experimental protocol.

Multijoint reflexes were elicited through controlled sinusoidal hip oscillations imposed by the robotic apparatus while subjects were instructed to remain relaxed. Hip oscillations were repeated for 10 consecutive cycles to elicit windup of reflex responses. The range of hip motion was ∼50° for all tests (40° of hip flexion to 0–10° of hip extension; hip extension was dependent on the subject's range of motion). Furthermore, we investigated whether the facilitation of multijoint spastic reflexes would be altered by varying stretch-related synaptic input by oscillating the legs about the hip joint bilaterally (bilateral 180° out of phase, bilateral in phase) and unilaterally [unilateral leg movements with the contralateral leg (i.e., left leg) held fixed either at end-range hip flexion (flexed) or at end-range hip extension (extended)]. Hip oscillations were performed at two movement frequencies (0.50 and 0.75 Hz) to test for velocity dependence. Tests were presented in random order and were repeated three times for each condition (total of 24 tests), with 2–5 min allowed between tests to minimize adaptation of the reflexes. At the completion of the experiment, two additional hip movements were performed to estimate the torque due to gravity, passive joint resistance, and inertial properties of the participant's legs. The torque contribution from gravity and passive joint resistance was approximated by slowly moving each leg through the entire range of motion at 2°/s in 5° increments, pausing for 5 s between each increment. The inertial properties of the leg and leg brace were estimated by rapidly oscillating (1.5 Hz) each leg from 25° to 10° of hip flexion for 10 cycles. Oscillating the hip within midrange minimized the potential for eliciting reflexes. For all tests, EMG and torque data were acquired for the duration of the movements.

Data analysis.

Active muscle torques for the hip, knee, and ankle joints were corrected for the biomechanical properties of the leg and leg brace using a method previously described (Onushko and Schmit 2007). All torque data were low-passed filtered (5 Hz) using a second-order Butterworth filter (filtfilt function in MATLAB; The Mathworks, Natick, MA). The torques due to gravity and passive resistance was estimated by fitting a third-order polynomial curve to the torque measurements taken during the slow, incremental hip movements, and then those polynomial coefficients were used to calculate the gravitational and passive resistance torques from the measured trial data (i.e., in-phase, out-of-phase, flexed, and extended hip oscillations). The inertial properties of the leg and leg brace were calculated by subtracting the gravitational/passive resistance torque and then estimating the inertial constant of the leg and leg brace using a linear regression analysis. In addition, a mechanical artifact was present within the system that was not associated with biomechanical properties of the legs. The artifact was estimated using an ensemble average of torque measurements recorded from neurologically healthy subjects who completed the study in its entirety (using the said method to subtract the inertial and gravity/passive torque components before ensemble averaging). Since the healthy control subjects do not experience responses to the imposed hip oscillations (refer to Fig. 1), the ensemble torque provides a good estimate of the artifact (on average <5 Nm). Our findings of no response in neurologically intact subjects are consistent with previous studies (Onushko and Schmit 2007). The active muscle torque (i.e., torque produced by active muscle contraction) was calculated by subtracting the gravitational/passive torque, inertial torque, and artifact from the measured trial torque data and was used for all subsequent analyses.

Fig. 1.

Data from a single neurologically healthy subject during passive hip movements. Passive hip, knee, and ankle torque recordings from neurologically healthy individuals were used for estimating the artifact from the hip movements.

To identify windup of the multijoint reflex responses to the movements, peak flexion and extension torques from the hip, knee, and ankle joints were found for the first seven consecutive hip oscillations. Torque data were separated into flexion and extension components before the peak torque per cycle was identified. The peak torque data per cycle were normalized to the mean of the seven cycles for each subject and then averaged across all subjects. The data were normalized to account for differences in the amplitude of the spastic reflex activity among the subjects. Differences in the peak torques from repeated hip oscillations were statistically compared using a multifactor ANOVA [main factors: stretch number (cycles 1–7), frequency (0.50 and 0.75 Hz), and movement type (in phase, out of phase, flexed, and extended); random factor: subject] and Bonferroni correction to determine differences among successive stretches (α = 0.05).

Windup was identified within the EMG data using a similar technique. EMG data were first notch filtered (59–61 Hz) to remove line noise and then band-pass filtered (10–300 Hz) using fourth-order Butterworth filters (MATLAB). For analysis, the root mean square (RMS) of the filtered EMG data was calculated using a 100-ms sliding window. The peak EMG response was found per cycle for the first seven successive hip oscillations during the time when the muscle was active for the movement cycle. EMG data for a single trial of any particular muscle were excluded if the peak EMG did not exhibit suprathreshold activity for at least two consecutive movement cycles (threshold = mean of baseline + 3 SD). A multifactor ANOVA was used to statistically compare the effects of movement type, frequency, stretch number, and subject on windup of spastic reflexes. Bonferroni correction was used for post hoc comparisons between individual stretches. The level of significance was set at α = 0.05 for all tests.


In general, a nonlinear increase was observed in the torque and EMG recordings, followed by either saturation or depression in the response. Figure 2 illustrates an example of multijoint windup from a single subject. Hip flexor and knee extensor torques increased from the first movement of the leg into hip extension, and the amplitude of the response to each subsequent stretch exceeded the previous response through the third stretch, which then modulated to the same or lower level for the remainder of the hip oscillations (Fig. 2A). Muscle activity from the RF and VM followed a similar trend to the torque data, increasing in amplitude during the initial two stretches. Although RF and VM activity slightly decreased following the fifth stretch, EMG activity modulated at a heightened level compared with the first stretch for the duration of the movements (Fig. 2A). Hip extensor and knee flexor torques successively increased in amplitude for the first two stretches of the leg into hip flexion and remained elevated for the duration of the trial (Fig. 2B). Activity at the ankle joint increased with repeated hip perturbations as well (Fig. 2C), although reflex responses in the ankle were more variable than responses observed at the other joints. MG and Sol activity generally remained above baseline, increasing in activity by the second rotation of the leg into hip extension, but little modulation was observed during the hip oscillations. In addition, TA EMG increased above baseline level in response to successive stretches of the hip, but it was modulated by the hip movements, with peak TA EMG coinciding with hip extension. Knee (VM) and ankle (Sol) muscles (i.e., muscles not crossing the hip joint) exhibited a delay in activation in response to stretches of the hip in which reflex activity was not elicited until after the second or third hip oscillation in this subject (refer to Fig. 2, A and C).

Fig. 2.

Example of multijoint windup in a single subject. Multijoint reflex responses (right leg only) from a spinal cord-injured (SCI) study participant (subject E) during the out-of-phase hip oscillations at 0.75 Hz for the first 7 cycles are shown. A: windup of hip flexors/knee extensor torque (half-wave rectified) and root mean square (RMS) electromyogram (EMG) of rectus femoris (RF), vastus medialis (VM), vastus lateralis (VL). B: hip extensor/knee flexor torque (half-wave rectified) and RMS EMG of medial hamstring (MH) during reflex responses. C: reflex activity of ankle dorsiflexor/plantarflexor torque and EMG of medial gastrocnemius (MG), soleus (Sol), and tibialis anterior (TA). Increasing positive torque values represent plantarflexion, and decreasing positive torque values represent dorsiflexion.

The effects at the hip joint from repeatedly moving the hips are illustrated in Fig. 3 for all movement conditions across all subjects. There were significant differences among the peak hip torque data with subsequent hip extensor stretches (hip flexion torque: ANOVA, P < 0.001) and hip flexor stretches (hip extension torque: ANOVA, P < 0.001). In addition, hip extension torque typically remained elevated after the third or fourth stretch, whereas hip flexion torque typically decreased after the fourth stretch. Although more variability was observed within the EMG data, peak RF and MH activity also showed significant differences with subsequent stretches (RF: ANOVA, P < 0.01; MH: ANOVA, P < 0.001). Except for the out-of-phase condition, peak RF EMG increased through the first two stretches, and peak MH EMG increased until the second stretch but generally remained elevated through subsequent stretches (refer to Fig. 3).

Fig. 3.

Windup of hip torque and EMG. Average peak hip torque (top) and peak RF and MH RMS EMG (bottom) are shown for each of the movement types for the 0.75-Hz movement frequency (right leg data only). Peak data were normalized to the mean of each subject's data. Significant differences between individual responses to stretches are indicated (*P < 0.05; **P < 0.01; ***P < 0.001). There was no statistically significant difference among the movement types (P > 0.05).

Reflex responses were also observed at the knee joint in response to repeated hip movements. Normalized peak knee torque responses showed significant differences with subsequent hip movements (knee extension and flexion torques: ANOVA, P < 0.001; Fig. 4). Knee extension torque increased through the first three consecutive joint rotations during the out-of-phase, in-phase, and flexed hip movements, and in general, the reflex amplitude was slightly elevated during the subsequent stretches. Knee extension torque during the extended movement condition elicited a slightly different pattern, where the activity rapidly increased during the second hip perturbation and then gradually declined throughout the remaining hip movements. Because RF is a biarticular muscle that acts at the hip and knee joint, it is possible that knee extensor torques are a result of RF stretch-related activity; however, since VM also demonstrated significant increased activity through the repeated hip oscillations (Bonferroni, 1st response vs. 4th response, P < 0.001), RF may have only partially contributed to knee extension torques in some cases. Significant increases in the group mean peak VL EMG activity were not observed (ANOVA, P > 0.05). A continuing increase in knee flexion torque (refer to Fig. 4, top right) occurred in response to repeated hip extensor stretches and followed a similar pattern to the responses observed in hip extension torques (refer to Fig. 3, top right), which might be accounted for by MH activity.

Fig. 4.

Windup of knee torque and EMG. Average peak knee torque (top) and peak VM and VL RMS EMG (bottom) are shown for each of the movement types for the 0.75-Hz movement frequency. Peak data were normalized to the mean of each subject's data set. Significant differences between individual responses to stretches are indicated (*P < 0.05; **P < 0.01; ***P < 0.001). There was no statistically significant difference among the movement types (P > 0.05).

Increased activity at the ankle was also noted during repeated stretches at the hip. Figure 5 illustrates normalized group mean ankle torque and EMG responses. Significant increases in plantarflexion torque (ANOVA, P < 0.001) were observed during the repeated stretches. Because of higher variability, the increased dorsiflexion torque during the initial two stretches was not significant (Bonferroni, P > 0.05). More notably, though, responses in dorsiflexion torque began to decrease following the second hip oscillation (Bonferroni, 1st response vs. 5th through 7th responses, P < 0.05), whereas plantarflexor torque continually increased (Bonferroni, 1st response vs. 2nd through 7th responses, P < 0.001). MG and Sol EMG activity did not fully resemble the pattern of plantarflexion torque, likely as a result of greater variability within the EMG data (Fig. 5A). TA activity followed a similar pattern to the dorsiflexion torque, with an increase in activity over the course of the first few consecutive hip oscillations followed by a depression in reflex activity (Fig. 5B).

Fig. 5.

Windup of ankle torque and EMG. Average peak torque and RMS EMG from the ankle plantarflexor (A) and ankle dorsiflexor (B) are shown for each of the movement types for the 0.75-Hz movement frequency. Peak data were normalized to the mean of each subject's data set. Significant differences between individual responses to stretches are indicated (*P < 0.05; **P < 0.01; ***P < 0.001). There was no statistically significant difference among the movement types (P > 0.05).

The type of hip movement did not significantly affect windup amplitude during repeated hip oscillations (torque: P > 0.05; EMG: P > 0.05). However, out-of-phase hip movements tended to elicit greater responses (for VM, VL, RF, MG, and TA EMG) during the first stretch compared with the other movement types, and the peak EMG responses were smaller over the first few stretches during the in-phase hip movements compared with the other movement conditions, which was more noticeable in hip flexor/knee extensor responses than in hip extensor/knee flexor responses. Furthermore, compared with the bilateral hip movements, responses during unilateral hip perturbations tended to increase more rapidly during the initial hip oscillations. Multijoint reflex windup was not sensitive to the movement frequency, with no significant effects on the amplitude of the joint torques (ANOVA, P > 0.05) or EMG activity (ANOVA, P > 0.05).

Increased muscle activity was also observed in the nonmoving leg of three SCI subjects (E, F, and J) during the unilateral tests. Figure 6 shows sample RMS EMG and torque data for the nonmoving leg from subject J during the extended (A) and flexed (B) movement conditions. The muscle activity pattern in the nonmoving leg resembled the windup seen in the oscillating leg and was also observed in the recorded ankle muscles. In addition, the hip, knee, and ankle torque also exhibited an increase in amplitude with subsequent stretches of the contralateral (i.e., oscillating) leg (Fig. 6). Subjects E and F showed similar windup responses in the nonmoving leg (data are not shown).


In the present study, repeated stretches of hip musculature in subjects with chronic SCI induced motor responses consistent with windup not only in the primary muscles stretched but also in muscles throughout the lower extremities. Subsequent stretching produced a windup response that was manifested as whole limb spasms of increasing EMG magnitude and joint torque. Although the spastic syndrome post-SCI is likely multifactorial, involving changes in both cellular intrinsic electrical properties and reflex pathway excitability (e.g., changes in reflex recruitment or alterations in presynaptic inhibition; Nielsen et al. 2007), the temporal characteristics of multijoint reflex activation seen in this study were consistent with previous examples of stretch reflex windup (Hornby et al. 2006) and windup of flexor reflexes (Hornby et al. 2003). In addition to being the first study to quantify windup in muscles not being stretched (VM, VL, MG, Sol, and TA), this study is the first to demonstrate in humans differences in temporal patterns of windup between functionally different muscle groups, suggesting that modulation in the excitability of this reflex pathway may be related to motor function. The findings from this study further support a strong role of changes in the excitability of motor-related circuitry in spastic reflex activity such as extensor spasms. Possible mechanisms of changes in network excitability contributing to the windup behavior are discussed below.

Possible cellular mechanisms of SCI windup response.

Much of the current evidence regarding alterations in intrinsic electrical properties and subsequent windup behavior following injury to the nervous system has focused on MNs (human: Gorassini et al. 2004; McPherson et al. 2008; Mottram et al. 2009; animal: Bennett et al. 1998a, 1998b, 2001; Lee and Heckman 2001; Li et al. 2004). Indirect measurements from this study are consistent with a plateau-driven windup response for several reasons. First, the time course of windup responses measured during repeated hip oscillations in the current study was similar to that of windup observed in MNs exhibiting plateau behavior (Bennett et al. 1998a, 1998b; Svirskis and Hounsgaard 1997). The stimulus interval is critical in generating windup responses to repeated stretches of the hindlimb, as shown in the decerebrate cat MN, generating a progressively larger amplitude response within 3- to 6-s intervals (Bennett et al. 1998a). Similar interstimulus intervals were used in the current study to elicit windup (0.66 and 1 s). Moreover, during repeated stretches, a nonlinear increase in torque and EMG amplitude was observed, which favors the idea of postsynaptic alterations in MNs, whereas a linear increase would likely suggest passive transmission of synaptic inputs. Second, the speed and magnitude of the stretch were consistent across all repetitions. It is possible to have increases in torque due to additional motor unit recruitment with each stretch, especially in the absence of synaptic inhibition from the stretch of the antagonist musculature. However, torque and EMG values typically returned to baseline between stretches (refer to Fig, 5), which would indicate a relative silence in the corresponding motoneuron pools. The analog to this in the Bennett et al. (1998a) study in animals is the electrical stimulation of the common peroneal nerve (mediating reciprocal inhibition on the triceps surae motoneuron pool).

A notable difference between the results of Bennett et al. (1998a) and results from the present study is that we found that in some cases (Fig. 5) the peak torque measurements returned to baseline levels over a time course of seconds. This is in contrast to the findings of Bennett et al. (1998a), who showed the membrane potential “plateauing” (see Bennett et al. 1998a, Fig. 4B, single-cell example) and no return to baseline membrane potential. There are several plausible explanations. First, in the Bennett et al. (1998a) study, the mean time course of the plateaus initiated from the windup is not reported, although it is indicated that several seconds needed to pass before the cell could “reset” and repeated measurements could be made. Second, they reported that several cells did not exhibit tonic activation of the plateaus. The authors noted that those cells with a lack of a tonic response (i.e., not lasting more than a few seconds) tended to have a higher threshold for recruitment. Lee and Heckman (1998) also demonstrated the differences in the time course of the plateau based on motor unit type. In their study, “partially bistable” motoneurons maintained the plateau on the order of 1–2 s and corresponded to cells with higher rheobases. Therefore, a return to baseline demonstrated by Lee and Heckman (within a reasonable time frame as the animal data) does not necessarily negate the role of PICs or plateaus in windup.

MNs, the “final common pathway,” make an attractive candidate for accounting for the windup response, but windup may also depend on excitatory interneuronal inputs to the MN. Changes in the regulation of the intrinsic electrical properties of locomotor or stretch-related IN populations due to SCI could result in IN participation in the windup response. For example, locomotor-related INs have firing patterns that rely heavily on the regenerative intrinsic electrical properties of the cell (Kettunen et al. 2005; Kiehn 2006; Kyriakatos and El Manira 2007). Moreover, they receive and integrate stretch-related input from hip afferents (Edgley and Jankowska 1987; Harrison and Jankowska 1985; Kriellaars et al. 1994). In this study, stretching of the hip musculature by repeated hip movements, similar to a locomotor pattern (i.e., out-of-phase movements), could have provided an excitatory drive to locomotor-related INs, leading to a windup response or entrainment of central pattern generators (CPGs) in the human spinal cord. Windup of INs is supported by the data from three subjects in which windup was produced in the stationary leg (refer to Fig. 6). In addition, the delay in the muscle activity of the ankle could also suggest that subthreshold windup occurs in INs, since it has been shown that MNs need activation of PICs to fire repetitively and exhibit windup (Bennett et al. 1998b; Lee and Heckman 2001). Repeated stretching of hip musculature may have induced subthreshold depolarizations that windup in the INs first and then proceed to windup MNs of ankle muscles. In addition, of the subjects who exhibited windup in muscles of the ankle, four subjects showed no EMG response during the first stretch of the hip (for the out-of-phase hip movement). Although this result still suggests multijoint windup is occurring through windup of MNs and/or INs, a sample population of four subjects is not strong enough to rule out the possibility that other sensory pathways (e.g., cutaneous afferents) may be eliciting these responses as well in the other subjects. However, a recent study done in the spinalized cat revealed a broadening of the movement-related receptive fields of the MNs, in which stretch input from hip muscles generated large currents in knee and ankle extensor MN pools (Hyngstrom et al. 2008a), which complements the findings from the current study.

Fig. 6.

Example of windup in the nonmoving leg of a single subject. RMS EMG and torque data from the stationary leg (i.e., leg not being oscillated) from subject J during the unilateral tests for the 0.75-Hz movement frequency are shown. A: extended test. The left leg was held at 10° hip extension (dashed line at bottom) while the right leg oscillated from 40° hip flexion to 10° hip extension. B: flexed test. The left leg was held at 40° hip flexion (dashed line at bottom) during contralateral leg oscillations. For all torque traces, flexion torques are decreasing values and extension torques are increasing values.

The loss of group Ia IN activity could also indirectly contribute to windup of MNs (Nielsen et al. 2007). In the animal model, Ia inhibition is important for modulating PIC amplitude, and conversely, the loss of Ia IN activity has been shown to elicit windup in MNs in a decerebrate preparation (Hyngstrom et al. 2007). In SCI and other neurological disorders, it has been recognized that Ia inhibitory pathways can be disrupted and or even lost (Nielsen et al. 2007). Loss of reciprocal inhibition in neurological conditions such as SCI would prevent stretch-related inhibition of PIC activity in MNs, thereby facilitating more PIC-related behaviors such as windup. However, the distribution of the monosynaptic Ia afferent and corresponding INs onto MNs is relatively focused (Eccles et al. 1957; Nichols 1999) and would not likely have resulted in the multisegmental responses seen in the present study. For example, Ia input would not be carried from stretch of the hip flexors onto muscles of the ankle, which SCI subjects demonstrated in this study (Fig. 5).

Alternatively, recent evidence has shown that the hyperactivity observed in chronic spinal injury may be due to a lowered expression of the potassium chloride cotransporter (KCC2) in MNs and spinal networks (Boulenguez et al. 2010). In the intact cord, KCC2 keeps the intracellular chloride ion concentration low; however, Boulenguez et al. (2010) have shown that KCC2 is decreased in chronic spinal cord-injured rats, limiting chloride transport out of the cell, which, compared with the intact spinal cord, results in a depolarizing shift in the equilibrium potential of chloride below the level of the lesion. The authors suggested that the shift to a more positive membrane potential could enable sensory inputs to produce long-lasting depolarizations, which, in turn, activates PICs. Furthermore, recent evidence from a study by Murray et al. (2010) has demonstrated that 5-HT2C receptors are constitutively open in chronic SCI, which then leads to increased intracellular Ca2+ concentration and recovery of large Ca2+ PICs. If similar mechanisms occur in human SCI, it is possible that while repeatedly moving the hips, stretch-related input may have been sufficient to initiate PICs in MNs or INs, and over the course of several hip movements, weaker sensory synaptic information could have initiated PICs in muscles not being stretched. However, the amount of non-stretch-related input was not measured, and it is unknown what contribution it had to the observed responses.

Finally, but not necessarily exclusively, group II interneuronal pathways carry stretch-related information from the whole limb (Jankowska 1992, 2001) and have been implicated previously in the widening of a given MN's “movement receptive field” in acute SCI (Hyngstrom et al. 2008a). Group II INs receive convergent information from other afferent pathways (Jankowska and Hammer 2002) that could thus also contribute to windup. The multisegmental stretch-related distribution of the windup response in the current study, the delay in the windup response in muscles not being stretched, and the lack of velocity sensitivity in the magnitude of the response support the role for the involvement of group II IN windup in human spastic reflexes. PIC behavior is necessarily regulated in the neurologically intact state, and thus it is likely that SCI disrupts control of PICs in both MNs and INs. Because of the indirect nature of the measurements of windup in this human study, we are limited in determining the relative contribution of each cell population (i.e., interneuron vs. motoneuron) to windup.

Although PICs make a strong candidate for the windup observed during imposed hip movements in the current study, other postsynaptic mechanisms may also play a role in maintained muscle activity that was observed in SCI subjects. For example, after SCI there is increased concentration of extracellular glutamate that results in the upregulation of glutamate receptor expression (Liu et al. 1991), which could increase neuronal excitability through synaptic inputs. In addition, NMDA receptors have been shown to induce windup in dorsal horn neurons in response to nociceptive inputs (Mendell 1966; for review see Daw et al. 1993), and windup can be reduced by blocking NMDA receptors through the application of specific NMDA antagonists (Daw et al. 1993). Although it is possible that in the current study NMDA receptors could have contributed to the windup phenomena, NMDA currents have been reported to only last for up to 500 ms in spinal neurons (Dale and Grillner 1986), and previous studies in human SCI have shown that prolonged spastic reflex responses last several seconds after the end of the movement (Hornby et al. 2006), suggesting that NMDA receptors play only a minor role in neuronal hyperexcitability associated with spastic reflex activity. Furthermore, exogenous application of NMDA after spinal cord transection in rats decreases rather than increases spinal reflex sensitivity (Krenz and Weaver 1998), suggesting that other mechanisms are likely to contribute to exaggerated reflex behaviors post-SCI. An alternative explanation for the windup response is an increase in spinal excitability through increased descending input from the brain stem via monoaminergic neurons projecting to the spinal cord. However, this seems unlikely, since passive movement of a limb does not increase tonic activation of brain stem pathways (Jacobs et al. 2002). In addition, windup responses were observed in the ankle in subjects with complete lesions (e.g., subjects A and B), further suggesting no supraspinal influences on spinal excitability. Although other cellular mechanisms could contribute to the underlying the windup response in spastic multijoint reflexes seen in the current study, we are limited in our interpretation of the results on account of using indirect measurements of motor output.

Differential electrical states related to patterns of muscle function.

An unexpected result from this study was that the temporal pattern of the windup response differed based on hip phasing between muscle groups. Although the windup between strict “flexor” and “extensor” groups was not literally identified in our study, there were two distinct groupings, based on coincident activity and temporal changes in the response, consisting of 1) hip flexors, ankle dorsiflexors, and knee extensors and 2) hip extensors, knee flexors, and ankle plantarflexors. Recent work by Endo and Kiehn (2008) has demonstrated that an asymmetry in conductances exists between extensor and flexor MN pools in that extensor MN pools are dominated more by inhibition than are flexor MN pools in the neonatal mouse spinal cord. Although we found an asymmetric response in different muscle groups within our study, the asymmetry in Endo and Kiehn's work is different, with the windup responses seen during the hip movements. During repeated stretches of the hip flexors, hip flexion, knee extension (particularly in the extended condition), and ankle dorsiflexion torque typically decreased following the third hip perturbation (see Figs. 35). In contrast, hip extension, knee flexion, and ankle plantarflexion torques generally remained elevated throughout the course of the hip oscillations when hip extensor muscles were stretched (see Figs. 35). This difference in flexor and extensor muscle groups seen in the current study could be due to the fact that tests were done in chronic SCI subjects, since the loss of supraspinal inhibition may result in more disinhibition over the extensor MN pool than the flexor MN pools. However, the asymmetry in the windup responses was consistent with previous studies investigating windup in human SCI (Hornby et al. 2003, 2006). Depression in active muscle torque responses during repeated stretches of hip flexor muscles is consistent with the time course of windup of the flexion reflex in human SCI, where the responses decrease following the second or third stimulus (Hornby et al. 2003). Stretch of the hip extensor muscles resulted in hip extension, knee flexion, and ankle plantarflexion torques that generally remained elevated throughout the course of the hip oscillations (see Figs. 35), reflecting the time course of stretch reflex windup (Hornby et al. 2006). EMG activity from RF, MH, and VM followed a similar trend to the torque data; however, muscle activity at the ankle was more variable, and the pattern of facilitation of the ankle EMG recordings was not as consistent, as the torque data. Sol and MG EMG activity rapidly increased through the initial three stretches but then decreased to levels slightly above the initial response (Fig. 5A), similar to TA activity (Fig. 5B). Since the ankle was held at a constant joint angle, variance among subjects could be due to excitatory polysynaptic pathways and lack of Ia IN inhibition.

Functional consequences of uncontrolled PIC behavior.

The role of PICs in neurosystem behavior is still unclear, but evidence suggests that PICs are important for increasing the gain of the MN pool through amplification of synaptic input and bistable firing behavior of certain cell types. The modulation of PICs is multifactorial, involving supraspinal centers and local inhibitory circuitry (Heckman et al. 2008). Loss of regulation of PICs, as may happen post-SCI, could interfere with volitional control of movement by causing uncontrollable spasms or by causing large errors during repetitive movements that induce windup (e.g., during walking). From a rehabilitation standpoint, this would suggest that it might be beneficial to restore the regulation of PICs. However, if stretch of the hip is entraining CPG pathways in human SCI, the additional power due to windup of these pathways may be useful for providing them with limb propulsion. Future studies could investigate ways to facilitate inhibitory pathways through either electrical stimulation or pharmacology, with the caveat that PICs provide sufficient power needed to complete motor tasks such as transferring or walking (Dietz 2008). Importantly, methods that more clearly delineate excitability of INs from MNs are needed to determine their relative contributions to the windup response. This information could more precisely direct treatment strategies.

Intersubject variability.

Although the small number of subjects participating in the study might limit the interpretation, we believe that the consistency of the response across the heterogeneous clinical presentation actually increases the generalizability of the results. In particular, it is interesting that, regardless of completeness of injury (i.e., complete or incomplete) and level of injury, the windup response occurred throughout the lower extremities in response to stretch of a single joint. In addition, in all of the subjects, we showed a delay in activation (ankle muscles) that could suggest a subthreshold windup or involvement of windup in interneurons. However, no conclusive evidence can be shown regarding level of injury, completeness of injury, or antispasticity medication due to the limited sample size.


No conflicts of interest, financial or otherwise, are declared by the author(s).


View Abstract