Postural corrections of the upper limb are required in tasks ranging from handling an umbrella in the changing wind to securing a wriggling baby. One complication in this process is the mechanical interaction between the different segments of the arm where torque applied at one joint induces motion at multiple joints. Previous studies have shown the long-latency reflexes of shoulder muscles (50–100 ms after a limb perturbation) account for these mechanical interactions by integrating information about motion of both the shoulder and elbow. It is less clear whether long-latency reflexes of elbow muscles exhibit a similar capability and what is the relation between the responses of shoulder and elbow muscles. The present study utilized joint-based loads tailored to the subjects' arm dynamics to induce well-controlled displacements of their shoulder and elbow. Our results demonstrate that the long-latency reflexes of shoulder and elbow muscles integrate motion from both joints: the shoulder and elbow flexors respond to extension at both joints, whereas the shoulder and elbow extensors respond to flexion at both joints. This general pattern accounts for the inherent flexion-extension coupling of the two joints arising from the arm's intersegmental dynamics and is consistent with spindle-based reciprocal excitation of shoulder and elbow flexors, reciprocal excitation of shoulder and elbow extensors, and across-joint inhibition between the flexors and extensors.
- heteronymous reflex
the mechanical interactions between the different segments of a moving limb (Craig 2005) present an intrinsic complexity for motor behavior and, consequently, an important topic for motor control theories. Research over the past 30 years has established that healthy individuals perform self-initiated arm movements by generating motor commands that anticipate the mechanical interactions between the different joints of the upper limb (de Rugy et al. 2006; Galloway and Koshland 2002; Hollerbach and Flash 1982; Lackner and Dizio 1994; Naito et al. 2014; Sainburg et al. 1999; Shadmehr and Mussa-Ivaldi 1994). A clear example of this predictive capability occurs when healthy subjects perform fast flexion movements of just their elbow while their elbow and shoulder joints are free to move (Gottlieb et al. 1996; Gribble and Ostry 1999). The agonist burst by their elbow flexor is accompanied by a burst of activity from their shoulder flexor to counteract the elbow's interaction torque onto the shoulder joint. Absent the co-occurring burst of shoulder flexor activity, the shoulder joint would be passively displaced into extension, since torque applied at one joint creates an interaction torque of the opposite sign at the adjacent joint. A particular mix of flexion torque at both the shoulder and elbow is required to balance out these interactions and selectively flex the elbow. Likewise, flexion torque at both joints (albeit a different pattern) is required to selectively flex the shoulder (Gottlieb et al. 1996; Gribble and Ostry 1999).
A growing body of work demonstrates that corrective actions to external loads, like when handling an umbrella in the changing wind or securing a wriggling baby, also account for the arm's intersegmental dynamics (for review see Kurtzer 2015; Pruszynski and Scott 2012). In an example of corrective action that mirrors self-initiated focal elbow movement, shoulder extensor activity is evoked by an external perturbation that suddenly displaces the elbow into flexion (Kurtzer et al. 2008). The increased activity by the shoulder extensor muscle is appropriate to counter the underlying torque applied to the shoulder and elbow joint that caused motion at just the elbow; remember, motion at just one joint occurs with a unique pattern of torque applied to both joints. Because the shoulder muscle is neither stretched nor shortened by elbow motion, its evoked response must reflect the integration of sensory information from a nonlocal source such as a muscle crossing the elbow joint. The ability of somatosensory-based corrections to integrate information from multiple joints is first expressed ∼50–100 ms after arm displacement. This epoch of time is termed the long-latency reflex (LLR) and is the fastest feedback response that accounts for the arm's intersegmental dynamics. The nervous system can generate an even faster burst of muscle activity called the short-latency reflex (SLR; ∼20–50 ms postperturbation), but this response is exclusively linked to local changes in joint motion; i.e., sudden elbow displacement does not evoke a SLR in the shoulder muscle (Kurtzer et al. 2008). Distinct processing abilities of the SLR and LLR reflect their different neural substrates within the spinal cord and primary motor cortex, respectively (for further treatment see discussion, Neural substrates for the reflex circuits of the shoulder and elbow).
As described above, the impact of elbow motion on shoulder muscle LLRs has been revealed by selectively displacing the elbow only (Cluff and Scott 2013; Crevecoeur et al. 2012; Kurtzer et al. 2008, 2009, 2013; Nashed et al. 2015). Several of these studies (Kurtzer et al. 2008, 2009, 2013; Nashed et al. 2015) also displaced the shoulder by a given amount with two different amounts of elbow motion. Taken together, the shoulder extensor LLR responded to flexion of the shoulder and elbow, whereas the shoulder flexor LLR responded to extension of the shoulder and elbow.
It remains unclear whether LLRs of elbow monoarticulars and elbow-shoulder biarticulars integrate multijoint motion appropriate for the arm's dynamics or only respond to local processing of joint motion. Obtaining that information is critical to understanding the neural organization of reflexes for shoulder-elbow coordination because each muscle group is an essential and distinct contributor to controlling the arm. Moreover, determining the neural organization of reflexes for shoulder-elbow coordination is important given our great skill in these actions, their central role in daily activities, and the extensive research effort in uncovering their organization and adaptation during self-initiated action (for review see Shadmehr and Wise 2005).
Multijoint LLRs appropriate for the arm's intersegmental dynamics would involve the elbow flexor responding to elbow and shoulder extension as well as the elbow extensor responding to elbow and shoulder flexion. These responses would parallel the multijoint LLRs of the shoulder muscles, where the shoulder flexor responds to extension at both joints and the shoulder extensor responds to flexion at both joints, to account for the fact that torque applied at one joint creates an interaction torque of the opposite sign at the adjacent joint (for further explanation, see results, Experiment 1: Pattern of limb motion). This straightforward prediction is consistent with the repeated demonstration of multijoint LLRs for different arm muscles, such as finger responses to displacements of other fingers (Cole et al. 1984), wrist responses to elbow displacements (Koshland et al. 1991; Latash 2000), and elbow responses to wrist displacements (Gielen et al. 1988), but has not been adequately tested since previous studies were compromised by poorly controlled patterns of joint displacement (Lacquaniti and Soechting 1984, 1986a, 1986b; Soechting and Lacquaniti 1988). The researchers opted to induce a wide range of joint motions rather than inducing specific patterns of joint motion. Accordingly, the different directions of induced shoulder-elbow motion should be evenly spaced (rather than uneven sampling/clustering of particular movement directions) and have a uniform magnitude of net displacement (rather than large motions for some directions and small motions for other directions). Without these safeguards, the sampling of joint motion is intrinsically biased and interferes with the interpretation of how information is combined across different joints.
The short-comings described above are an expected result of applying force pulses of equal magnitude and direction in extrinsic space given the nonuniform inertia of the arm. In contrast, in this study we conducted two experiments using joint-based loads to induce well-controlled patterns of joint motion.1 The first experiment applied loads tailored to each subject's arm (paradigm extended from Experiment 3, Kurtzer et al. 2008) to induce eight shoulder-elbow displacements nearly equal in magnitude and even spacing of direction. Thereby, we examined the relative sensitivity of shoulder, elbow, and biarticular LLRs to shoulder and elbow displacement in the same experiment. A second experiment imposed flexion and extension displacements of the subject's shoulder while the elbow position was physically clamped by a brace (paradigm analogous to the control experiment of Crevecoeur et al. 2012). Thereby, we examined how elbow muscle LLRs were impacted by displacement of only the shoulder joint.
Our results reconfirm that the LLRs of shoulder muscles integrate motion from both joints (the flexor responds to extension at both joints and the extensor responds to flexion at both joints) and that the relative sensitivity to shoulder-elbow displacement was appropriate to counter the underlying shoulder torque. Furthermore, our results clearly demonstrate that LLRs of elbow muscles integrate motion from both joints (the flexor responds to extension at both joints and the extensor responds to flexion at both joints) and that the relative sensitivity to shoulder-elbow displacement was appropriate to counter the underlying elbow torque. LLRs for the biarticulars also responded to multijoint motion with sensitivity situated between the shoulder and elbow muscles as appropriate for their anatomical action. Hence, multijoint responses are a general feature of long-latency reflexes by muscles controlling the shoulder and elbow. This pattern accounts for the inherent flexion-extension coupling of the shoulder and elbow arising from the arm's intersegmental dynamics such that the muscular responses counter the underlying torque perturbations. We propose a general model of sensorimotor connectivity to account for the patterns of corrective muscle activity and discuss the likely neural substrates.
MATERIALS AND METHODS
A total of 15 healthy young subjects (9 male, 6 female) participated in the experiments following informed consent to procedures approved by the ethics committee at New York Institute of Technology College of Osteopathic Medicine. The average age of the subjects was 26 yr. Experiments 1 and 2 utilized 8 and 7 individuals, respectively. Each experiment lasted 90–120 min, and subjects were paid for their time.
Subjects interacted with a robotic exoskeleton (KINARM; BKIN Technologies, Kingston, ON, Canada). The device was adjusted to the size of their right arm, supported its weight, enabled flexion/extension movements of the shoulder and elbow in the horizontal plane, and could selectively apply torques to each joint. Shoulder angle is measured relative to the frontal plane and elbow angle is measured between the forearm and upper arm; 0° is full extension. Visual targets and a hand-aligned cursor were presented in the horizontal plane via a virtual reality system while a metal partition obscured direct vision of the subject's arm. During experiment 1, the shoulder and elbow were free to move as described. During experiment 2, the elbow joint was locked in a fixed position so that only shoulder motion was possible. The locking mechanism was a metal plate bolted between the linkage of parallel bars on the robotic exoskeleton that allows elbow motion (Scott 1999).
We recorded surface electromyography (EMG) of arm muscles following procedures described previously (Pruszynski et al. 2008). In experiment 1, we recorded activity from six major muscles controlling the shoulder and/or elbow: a shoulder flexor (pectoralis major), shoulder extensor (posterior deltoid), elbow flexor (brachioradialis), elbow extensor (triceps lateral), shoulder-elbow flexor (biceps brachii), and shoulder-elbow extensor (triceps longus). In experiment 2, we recorded activity from an elbow flexor (brachioradialis) and elbow extensor (triceps lateral). The skin surface overlying each muscle of interest was lightly abraded with alcohol, bipolar Ag-AgCl electrodes (no. FT007; MVAP Medical Supplies, Newbury Park, CA) were affixed over the muscle belly, and a ground electrode was placed on the subject's acromion. EMG signals were routed and amplified using a Bortec AMT-8 (Bortec Biomedical, Calgary, AB, Canada).
Each experiment had two phases. The probe phase at the beginning of the session provided an individualized estimate of the arm's inertia and familiarized the subject with the overall task. The subsequent test phase displaced the arm in particular patterns of motion based on the inertial estimate. During the probe phase, the applied torque pulse had a magnitude of 2 Nm. In experiment 1, this perturbation was directed to eight different combinations of shoulder and elbow torque having equal separation in joint-torque space (see Fig. 1A). Thereby, the subject's arm underwent eight different combinations of shoulder-elbow motion. The induced motion pattern from these perturbation was highly nonuniform (in spacing and magnitude) due to the anisotropy of the arm. Perturbations were repeated five times each for a total of 40 trials in the probe phase. During experiment 2, the probe torques were only directed to shoulder flexion or extension since the elbow was locked in place. Perturbations were repeated 8 times for a total of 16 trials in the probe phase. Few trials were required in the probe phase since the induced motion was highly similar across trials.
Mechanical interactions across the different joints of the arm create a complex relation between the applied torques and resulting motion. The complete equations of motion for the arm and robot have been previously provided (Gritsenko et al. 2011; Scott 1999). An implicit and more intuitive description is achieved by examining the covariation of kinetics and kinematics (Gottlieb et al. 1997; Graham et al. 2003). The relation between the change in torque and change in motion can be approximated with a linear matrix on a brief timescale since the velocity-dependent and position-dependent nonlinearity of limb dynamics are relatively small before neural feedback makes an impact. Accordingly, we used a covariance matrix (MATLAB release 2010a, lscov function, a least-squared algorithm) to relate the two-dimensional (2-D) applied torque and 2-D joint displacement 50 ms from the perturbation onset (Kurtzer et al. 2008):
By inverting the subject-specific matrix, one can obtain the necessary torque for inducing a particular pattern of joint motion.
Experimental sequence and motor task.
In experiment 1, we aimed to displace the arm by a fixed amount (1° in 50 ms) in eight equally separated directions of joint space (Fig. 1B). The applied torque was highly nonuniform in order to create uniform motion of an arm having inertial anisotropy. In experiment 2, we aimed to displace the shoulder by a fixed amount (1° in 50 ms).
The arm was perturbed near the middle of its workspace, typically a shoulder angle of 60° and elbow angle of 75°. A torque pulse was applied to their arm after the hand-aligned cursor (0.5-cm radius) remained in the center of a visual target (3-cm radius) for a random time interval (probe phase = 1–1.5 s; testing phase = 1–3.5 s). Torque pulses had a 75- to 80-ms duration with a 10-ms rise and fall time. In experiment 1, background activity by the shoulder and elbow muscles was created by subjects countering a background torque applied to the shoulder and elbow: a constant flexor torque at both joints, or a constant extensor torque at both joints. The torque magnitude could range between 1 and 2 Nm depending on the subject's level of comfort. Thirty trials were collected for each condition, leading to a total of 480 trials: 2 types of background muscle activity × 8 perturbation directions × 30 repeats. In experiment 2, background elbow flexor or elbow extensor activity was generated by having the subject isometrically exert elbow flexion or elbow extension at 10% of maximal voluntary effort. Thirty trials were collected for each condition, leading to a total of 120 trials: 2 types of background muscle activity × 2 perturbation directions × 30 repeats. Subjects were encouraged to ignore the transient perturbation so that we could assess the automatic aspect of the evoked muscle activity (Calancie and Bawa 1985; Crago et al. 1976; Hammond 1956; Lee and Tatton 1982; Lewis et al. 2006; Pruszynski et al. 2008).
Angular positions of the shoulder and elbow were low-pass filtered (25-Hz, 2-pass, 6th-order Butterworth). The pattern of joint displacements (at 50 ms postperturbation) was assessed by the consistency of their magnitude and spacing across conditions. Displacement magnitude (D) is the vector length of the shoulder-elbow displacement.
Uniformity of displacement magnitude is the standard deviation of displacement magnitude across the eight conditions (n) normalized by their average magnitude (D̄). For perfect uniformity (a circular distribution), this measure would be 0, whereas increasing nonuniformity would yield increasing values (a more elliptical distribution).
We also examined the directional spacing of the joint displacements. The angular direction of each joint displacement is given by the arc tangent of the induced shoulder and elbow motion. Angular separation (A) is the change in movement direction between adjacent conditions.
Uniformity of the angular separation between all eight conditions (n) was quantified by their standard deviation, where Ā is the average angular separation.
Perfect uniformity of angular separation between different joint displacements would yield 0, whereas increasing nonuniformity would yield increasing values.
Surface electrical activity of the muscles was amplified (gain = 1–5 K), digitally sampled at 1,000 Hz, bandpass filtered (25–250 Hz), and rectified. The EMG signals were then normalized by the muscle's mean activity during the hold period with an activating torque. Evoked activity was relative to the steady-state activity 50–100 ms before the perturbation. The evoked activity of interest spans a window of 50–100 ms postperturbation. It is alternately called the long-latency reflex (LLR), long-latency response, M2/3, and R2/3 (Crago et al. 1976; Lee and Tatton. 1982; Kurtzer et al., 2008; Pruszynski et al. 2008). In this work we retain the long-latency reflex (LLR) designation. In addition to the LLR we also examined an earlier epoch termed short-latency reflex (SLR) or R1 (20–45 ms postperturbation). These epochs are consistent with previous studies, including our own (Calancie and Bawa 1985; Crago et al. 1976; Jaeger et al. 1982; Lee and Tatton 1982; Lewis et al. 2006; Kurtzer et al. 2008; Pruszynski et al. 2008).
Planar regressions characterized how the evoked muscle activity was sensitive to shoulder and elbow motion.
The orientation of the best-fitting plane indicates the relative sensitivity of the response to shoulder and elbow motion. We term this relative sensitivity the “preferred motion direction” (PMD).
Note that shoulder and elbow motion form the x- and y-axes, flexion motion is positive, and extension motion is negative. Accordingly, a PMD of 0° would indicate muscle responses solely linked to shoulder flexion, whereas a PMD of 180° would indicate muscle responses solely linked to shoulder extension. PMDs of 90° and 270° would indicate muscle responses solely linked to elbow flexion and elbow extension, respectively. We tested whether the SLR and LLR of the four monoarticular muscles were aligned to one of these cardinal axes, deemed a “local PMD.” Alignment would occur if a muscle spanning just one joint generated a response based on sensory information only from its own stretch. Alternatively, a bias away from the cardinal axis would occur if the muscle response was based on its own stretch and stretch of muscles crossing an adjacent joint.
We also contrasted the observed PMDs with the directional preference associated with the underlying torque, i.e., the ideal sensitivity to shoulder and elbow motion that would account for the arm's intersegmental dynamics. For each muscle we simply assumed that the ideal response was proportional to the applied torque it opposed (e.g., shoulder flexor activity solely linked to applied shoulder extensor torque). This torque component from the eight torque combinations was regressed against the associated joint motion to yield the “ideal PMD” for a muscle to counter the underlying torque. The ideal PMDs are described more fully in results (Experiment 1: Patterns of limb motion) and summarized in Table 1.
A final measure was the magnitude of response in the SLR and LLR epochs. We examined the steepness of the plane fit, which was the norm of the shoulder and elbow scaling factors.
We examined differences in the pattern of PMDs using circular statistics (Baschelet 1981). A Rayleigh's test determined if PMDs from different subjects had a unimodal distribution. If the 95% confidence interval of the measured PMDs did not overlap with the local PMD or ideal PMD, then the bias was considered statistically significant.
Signed-rank tests contrasted the uniformity of joint motion in experiment 1. Paired t-tests contrasted the responses magnitude in experiment 1 and the evoked responses in experiment 2. P < 0.05 was considered statistically significant.
Experiment 1: Patterns of limb motion.
Representative data from the probe phase of experiment 1 are presented in Fig. 1A. The eight applied torque pulses have a circular distribution since each has the same magnitude as its neighbor and the same spacing with its neighbor. In contrast, the induced displacements of the subject's shoulder and elbow (50 ms postperturbation) have an elliptical distribution with larger changes in magnitude for some directions than others. Peak elbow motion is greater than peak shoulder motion, 232% (SD 15%) on average, reflecting the inertial anisotropy of the arm. Moreover, the displacement ellipse is rotated into the top left and bottom right quadrants so that most displacements involve flexion at one joint and extension at the other joint. Torque applied to one joint creates motion of the same sign at that joint and motion of the opposite sign at the opposing joint. This is exemplified by shoulder flexion torque displacing the shoulder into flexion and the elbow into extension (Fig. 1A, white circles) while elbow extension torque displaces the elbow into extension and the shoulder into flexion (gray circles).
The mechanical interactions that relate a circular distribution of applied torques and elliptical distribution of arm displacements also require an elliptical distribution of applied torques to induce a circular distribution of arm displacements. Larger shoulder than elbow torque was needed to overcome the larger inertia of the upper segment relative to the forearm, and large combined flexion torque and combined extension torque were needed to overcome the intrinsic flexion-extension coupling between the two joints. Accordingly, the proscribed torque from the inverted linear matrix (see methods, Applied torques) involved a peak shoulder torque 231% (SD 17%) larger than the peak elbow torque and a distribution of shoulder-elbow torques rotated into the top right and bottom left quadrants of joint-torque space (Fig. 1B), i.e., opposite the intrinsic coupling of joint motion (Fig. 1A).
The resulting joint motion was nearly uniform in magnitude, and spacing indicating that the linear matrix provided a good approximation of two-joint dynamics on a brief timescale. This approach to a uniform distribution of displacement magnitudes and angular separations was quantitatively confirmed by contrasting the standard deviation of normalized magnitude (signed rank, P < 0.001) and the standard deviation of angular separation (signed rank, P < 0.01) between the probe and test phases (Fig. 1C). The same analysis was conducted on the motion data provided in Soechting and Lacquaniti (1988) (gray lines in Fig. 1C); their Figs. 7 and 8 depict the group-average displacements of the elbow and shoulder to upward and downward force perturbations and was their most complete experiment on this topic. The current methodology using joint torques tailored to the subject's arm dynamics was a clear improvement from the previous method of using the same force pulses for each subject.
The experiment tested whether monoarticular shoulder and elbow muscles respond to perturbations based on local joint motion. Local joint responses would involve a shoulder extensor muscle only increasing its activity with increased shoulder flexion and an elbow extensor muscle only increasing its activity with increased elbow flexion. These patterns are presented in Fig. 2, left. Increases and decreases in muscle activity are shown for each joint displacement along with its preferred motion direction. Shoulder extensor and elbow extensor responses based only on local joint motion would exhibit PMDs oriented to 0° and 90°, respectively.
Alternatively, monoarticular muscles could respond to a combination of shoulder and elbow motion appropriate for countering the underlying torque. Accordingly, a shoulder extensor muscle would express activity when the shoulder is displaced into flexion (Fig. 1B, right, gray circles) and when the elbow is displaced into flexion (Fig. 1B, right, white circles), because both require compensation of the underlying shoulder flexion torque (Fig. 1B, left). Figure 2A, right, shows increases and decreases in activity based on shoulder torque for all the joint displacements. A planar regression of shoulder flexor torque against shoulder and elbow displacement indicates that the shoulder extensor's ideal PMD is 18°; the weighting of shoulder and elbow motion appropriate to counter the underlying torque.
An elbow extensor responding to shoulder-elbow motion appropriate to counter the underlying torque would express activity when the elbow is displaced into flexion and when the shoulder is displaced into flexion (Fig. 1B, right, gray and white circles). Both displacements require compensation of an underlying elbow flexion torque (Fig. 1B, left). Figure 2B, right, shows increases and decreases in elbow extensor activity based on elbow torque for all the joint displacements. A planar regression of elbow flexor torque against shoulder and elbow displacement indicates that the elbow extensor's ideal PMD is 45°.
The local and ideal PMDs for all four monoarticulars are summarized in Table 1. Note that these PMDs are in the flexion-flexion and extension-extension quadrants opposite the intrinsic flexion-extension coupling of the shoulder and elbow (Fig. 1A).
Experiment 1: Patterns of evoked muscle activity.
The overall pattern of arm muscle activity (Figs. 3–5) was similar to previous observations. Limb displacements evoked arm muscle activity as early as 20 ms after the perturbation, although bursts in the SLR epoch were typically small compared with subsequent bursts in the LLR epoch. Evoked activity gradually changed with the different displacement directions rather than uniformly responding to all perturbations or selectively responding to a single perturbation. Evoked activity included increases and decreases with greater increases than decreases due to a relatively low background activity creating a floor effect. Last, the evoked activity peaked within the LLR epoch consistent with the instruction to ignore the transient perturbation.
The response pattern by the six arm muscles is separately described below; summary results from these muscles are presented in Table 1. The LLR of the shoulder extensor muscle (Fig. 3A, outer panels) showed increased activity with perturbations that flexed the shoulder and decreased activity with perturbations that extended the shoulder. Increases and decreases in activity also followed elbow flexion and extension, respectively. Plane fits of the change in LLR compared with the initial shoulder and elbow displacement indicate that the PMDs had a unimodal distribution (P < 0.001, z = 6.9) oriented toward shoulder flexion-elbow flexion (Fig. 3A, center panel). The shoulder extensor's LLR had a mean PMD of 22° (95% CI = 1°), which significantly differed from the muscle's local PMD of 0° (i.e., maximally excited by shoulder flexion). The mean PMD was strongly biased toward the ideal PMD of 18° (i.e., multijoint response appropriate to counter the applied shoulder flexor torque) but was not in full alignment (P < 0.05).
Evoked activity of the shoulder flexor muscle (Fig. 3B, outer panels) mirrored its antagonist. Displacing the shoulder into extension increased its LLR. Displacing the shoulder into flexion decreased its LLR. Moreover, displacement of the elbow into extension and flexion led to increased and decreased LLRs. The resulting PMDs of the shoulder flexor muscle were unimodally distributed (P < 0.001, z = 6.9) and oriented toward shoulder extension-elbow extension (Fig. 2B, center panel). The mean PMD of 206° (95% CI = 6°) significantly differed from the muscle's local PMD of 180° (i.e., maximally excited by shoulder extension). The bias was toward the ideal PMD of 198°, although it was not in full alignment (P < 0.05).
The elbow extensor muscle (Fig. 4A, outer panels) displayed an increased LLR following elbow flexion and a decreased LLR following elbow extension. Shoulder displacement also modified its LLR with increases and decreases during shoulder flexion and extension, respectively. PMDs of the elbow extensor's LLR had a unimodal distribution (P < 0.001, z = 6.9) and orientation toward shoulder flexion-elbow flexion (Fig. 4A, center panel). The mean PMD of 36° (95% CI = 6°) significantly differed from the muscle's local PMD of 90° (i.e., maximally excited by elbow flexion). The PMD approached the ideal PMD of 45° but was not in full alignment (P < 0.05).
The elbow flexor muscle expressed LLR activity that mirrored its antagonist (Fig. 4B, outer panels). LLR increases and decreases followed elbow extension and elbow flexion, respectively. Increased LLR also followed shoulder extension; decreased LLR was quite small with shoulder flexion. The elbow flexor's LLR expressed PMDs with a unimodal distribution (P < 0.001, z = 6.8) and orientation toward shoulder extension-elbow extension (Fig. 4B, center panel). The average PMD of 242° (95% CI = 9°) significantly differed from the muscle's local PMD of 270° (i.e., maximally excited by elbow extension). This approached the ideal PMD of 225° but was not fully aligned (P < 0.05).
Biarticular muscles expressed LLRs that depended on motion of both the shoulder and elbow. The biarticular extensor muscle expressed the greatest LLR to flexion at both joints (Fig. 5A, outer panels); its PMDs were unimodally distributed (P < 0.001, z = 7.0) and directed to 31° (95% CI = 4°; Fig. 5A, center panel). Conversely, the biarticular flexor muscle expressed the greatest LLR to extension at both joints (Fig. 5B, outer panels); its PMDs were unimodally distributed (P < 0.005, z = 4.6) and directed to 226° (95% CI = 12°; Fig. 5B, center panel). The mean PMD of the biarticular extensor muscle lies between the mean PMDs of the shoulder extensor and elbow extensor muscle, and the mean PMD of the biarticular flexor muscle lies between shoulder flexor muscle and elbow flexor muscle. There is no firm characterization of the biarticular's moment arms at both the shoulder and elbow, so we cannot make a clear prediction about their local or ideal PMDs.
We also examined arm muscle activity in the SLR epoch. Evoked activity in this earlier epoch was notably different than during the LLR epoch (see Table 1). The plane fits were significantly flatter (t-test, P < 0.05, t > 3.1). The average response magnitude expressed during the SLR epoch was 27% the size expressed in the LLR epoch. Moreover, the shoulder extensor, shoulder flexor, and elbow flexor had mean PMDs of −6° (95% CI = 12°), 168° (95% CI = 15°), and 273° (95% CI = 11°) during the SLR epoch. These did not differ from their muscles' local PMDs of 0°, 180°, and 270°. The elbow extensor had a mean PMD of 47° (95% CI = 32°), which did differ from its local PMD of 90° but not its ideal PMD of 45°.
Experiment 2: Patterns of evoked muscle activity.
During experiment 2, the elbow joint was physically braced such that there was effectively no motion at the elbow (Fig. 6A). Still the elbow muscles expressed reciprocal responses to different direction of imposed shoulder motion (Fig. 6, B and C). Shoulder flexion evoked increased activity in the elbow extensor muscle and decreased activity in the elbow flexor muscle. Shoulder extension evoked decreased activity in the elbow extensor muscle and increased activity in the elbow flexor muscle. The change in activity between conditions was significant in LLR epoch (t-test, P < 0.001, t > 5.2) but not in the SLR epoch (t-test, P > 0.7, t < 0.4).
The present study examined the evoked activity of muscles controlling the shoulder and elbow to shoulder-elbow motion. The first experiment displaced the subject's arm in eight directions of shoulder-elbow motion by nearly equal magnitudes and spacing. The second experiment displaced the subject's shoulder into flexion and extension while the elbow position was clamped by a brace. Evoked activity by the shoulder monoarticulars was consistent with our earlier findings (Crevecoeur et al. 2012; Kurtzer et al. 2008, 2009, 2013; Nashed et al. 2015). The LLR of the shoulder flexor responded to extension of both joints, and the LLR of the shoulder extensor responded to flexion at both joints. In contrast, the SLRs of shoulder muscles only responded to displacement of the shoulder joint. We found a similar pattern for the elbow muscles. The LLR of the elbow flexor responded to extension of both joints, the LLR of the elbow extensor responded to flexion at both joints, and the SLR of both elbow muscles only responded to elbow displacement. Importantly, these patterns are appropriate for the inherent flexion-extension coupling of the shoulder and elbow arising from the arm's intersegmental dynamics. A shoulder extensor response countering a shoulder flexor torque would be present when the shoulder or elbow is flexed (Fig. 3A), since both situations result from perturbations that include a shoulder extensor torque (Figs. 1B and 2A). An elbow extensor response countering an elbow flexor torque would also be present in these situations (Fig. 4A), since the perturbations include an elbow extensor torque (Figs. 1B and 2B). Accordingly, the LLRs of the shoulder and elbow monoarticular muscles were maximally activated by combinations of shoulder-elbow motion linked to the underlying shoulder and elbow torque, respectively, and this directional tuning was biased into the quadrant flexion-flexion and extension-extension motion. That is, the ideal preferred motion directions of the monoarticulars were opposite the intrinsic flexion-extension coupling of the shoulder and elbow. The upper limb's biarticulars also responded to motion of both joints during the SLR and LLR as expected for muscles that spanned both joints.
Model of reflex connectivity of the elbow and shoulder muscles.
We propose a general model of reflex connectivity based on muscle stretch signals to account for the general pattern of evoked activity in the shoulder and elbow monoarticulars. Muscle spindles are embedded throughout the body's skeletomotor muscles and transduce the static and changing stretch of their host muscle (for review see Loeb 1984). Our general model of monoarticular reflexes diagrams the net impact of spindle afferents onto motor neurons as an excitatory or inhibitory linkage. Note that it is not intended to be a precise mathematical representation of reflex processing (Dufresne et al. 1979; Mugge et al. 2010; Stein et al. 1995) or to include all somatosensory afferents that may contribute (Matthews 1988; Proske and Gandevia 2012). We address several important details, such as afferent type(s), central circuit(s), and the role of biarticulars and limitations, in subsequent sections.
The simplest reflex circuit involves processing from and to the same muscle (Fig. 7A). Stretching the muscle stimulates its muscle's spindles, which then excite the same muscle's motor neurons. This organization yields four separate circuits for the four monoarticular muscles of the arm and accounts for the fact that stretching each muscle evokes excitatory activity in the same muscle during the SLR and LLR epochs. For example, shoulder flexion stretches the shoulder extensor muscle and leads to an increase in its activity (Fig. 3A). Likewise, elbow flexion stretches the elbow extensor muscle and leads to an increase in its activity (Fig. 4A). A similar pattern applies to the other two monoarticulars; shoulder extension and elbow extension increase the activity of the shoulder flexor (Fig. 3B) and elbow flexor (Fig. 4B), respectively.
The next reflex circuit involves processing between a pair of muscles having opposite actions at the same joint (Fig. 7B). Stretch of one muscle inhibits the motor neurons of the opposing/shortened muscle, i.e., reciprocal inhibition. This organization yields two circuits for the two pairs of monoarticular muscles and accounts for the fact that stretching a muscle can evoke inhibitory activity in the opposing/shortened muscle during the SLR and LLR epochs. Accordingly, displacement of the shoulder joint into extension stretches the shoulder flexor muscle, which decreases the activity of the opposing/shortened shoulder extensor muscle (Fig. 3A). Displacement of the elbow joint into extension stretches the elbow flexor muscle, which decreases the activity of the elbow extensor muscle (Fig. 4A). The other two monoarticulars possess a similar pattern; shoulder flexion results in decreased shoulder flexor activity (Fig. 3B), and elbow flexion results in decreased elbow flexor activity (Fig. 4B).
Two types of reflex circuits involve processing between shoulder and elbow muscles. Reciprocal excitation links the shoulder and elbow flexors and also links the shoulder and elbow extensors (Fig. 7C, left). Reciprocal inhibition links the flexors and extensors of the two joints (Fig. 7C, right). Importantly, this pattern of reflex connectivity mirrors the arm's intersegmental dynamics where flexion at one joint induces extension at the other and extension at one joint induces flexion at the other.
Reflexes between shoulder and elbow muscles yields a total of four circuits for the four muscle pairs and accounts for their excitatory and inhibitory affects during the LLR epoch. Reciprocal excitation between the extensors is evident when a displacement flexes either joint. Elbow flexion stretches the elbow extensor muscle which activates its spindle afferents and thereby evokes increased LLR activity in the shoulder extensor muscle (Fig. 3A). In a complementary manner, shoulder flexion stretches the shoulder extensor, which activates its spindle afferent and thereby evokes increased LLR activity in the elbow extensor muscle (Fig. 4A). Reciprocal excitation between the flexors follows the same pattern: elbow extension stretches the elbow flexor, which leads to an excitatory LLR by the shoulder flexor (Fig. 3B); shoulder extension stretches the shoulder flexor, which leads to an excitatory LLR by the elbow flexor (Fig. 4B).
Reciprocal inhibition between flexors and extensors of the opposing joints is evident when elbow extension (stretching the elbow flexor) leads to an inhibitory LLR by the shoulder extensor (Fig. 3A) and elbow flexion (stretching the elbow extensor) leads to an inhibitory LLR by the shoulder flexor (Fig. 3B). In the other direction, shoulder flexion (stretching the shoulder extensor) leads to an inhibitory LLR by the elbow flexor (Fig. 4B) and shoulder extension (stretching the shoulder flexor) leads to an inhibitory LLR by the elbow extensor (Fig. 4A).
Taken together, a general model of reflex connectivity driven by muscle stretch signals is consistent with the general pattern of evoked activity by the shoulder and elbow monoarticulars.
Neural substrates for the reflex circuits of the shoulder and elbow.
Two of the reflex circuits, simple stretch reflex and reciprocal inhibition of agonist-antagonist pairs, are long established and commonly described in textbooks on motor neurophysiology (examples include Bear et al. 2007; Kandel et al. 2013; Purves et al. 2008). The other two reflex circuits, reciprocal excitation and inhibition between shoulder and elbow muscles, are less established but consistent with the evidence. We found that that LLRs of shoulder and elbow monoarticulars responded to both shoulder and elbow motion and that their SLRs only responded to motion of a single joint. Previous studies that also utilized physical displacements of the upper limb reported multijoint integration by the LLR and that the SLR (when present) only depended on local muscle stretch (Cole et al. 1984; Gielen et al. 1988; Koshland et al. 1991; Kurtzer et al. 2008; Latash 2000; Soechting and Lacquaniti 1988). Hence, the arm's LLR to limb displacement appears to involve all the circuits described above, whereas the arm's SLR to limb displacement appears to only involve the circuits concerned with local joint information.2
The different processing capability of the arm's SLR and LLR reflects their different neural substrates. The rapidity of the SLR (20–45 ms) unambiguously indicates the impact of group I/fast-conducting afferents on spinal cord networks. Its muscle afferents include primary muscle spindles relaying information about the rate of muscle stretch and static muscle stretch. Activity expressed during the LLR (50–100 ms) is more ambiguous, because multiple spinal and supraspinal substrates have the opportunity to contribute. In other words, the more delayed event could reflect the impact of information conducted slowly over short distances or conducted quickly over long distances.
An extensive research effort, using single-unit recordings (Cheney and Fetz 1984; Herter et al. 2009; Pruszynski et al. 2014), electroencephalography (EEG) recordings (MacKinnon et al. 2000), single-pulse transcranial magnetic stimulation (TMS; Day et al. 1991), repetitive TMS (Tsuji and Rothwell 2002), and cortical stroke victims (Marsden et al. 1977; Trumbower et al. 2013), indicates that primary motor cortex contributes to the LLR (for review see Pruszynski and Scott 2012; Shemmell et al. 2010). This is particularly important because primary motor cortex is a key node for funneling diverse sources of information from premotor and parietal cortex, basal ganglia, and the cerebellum (Porter and Lemon 1993). Accordingly, primary motor cortex likely underlies the sophisticated capabilities exhibited by the LLR (exemplified by multijoint integration) that are relevant to optimal feedback control (Pruszynski and Scott 2012; Scott 2004).
Recent evidence demonstrates that primary motor cortex supports the LLR's ability for multijoint integration (Pruszynski et al. 2011a). One experiment applied TMS over the scalp of human subjects in addition to arm perturbations. TMS was delivered to the primary motor cortex in a suitable location to activate cortical circuits controlling the shoulder muscle. When TMS was timed to evoke a muscle response coincident with the LLR evoked by elbow displacements, the observed muscle activity was greater than the sum of the two stimuli separately, which implies their neural interaction. That is, both elbow displacement and TMS engaged a cortical linkage between elbow muscle afferents and motor outflow to shoulder muscles. The augmentation was not observed if TMS was timed to evoke a muscle response coincident with the SLR. A complementary experiment examined the responses of single neurons within the primary motor cortex of behaving monkeys to perturbations of their arm. The authors examined a subset of cortical neurons that expressed a steady-state/postural response similar to shoulder muscles (“shoulder muscle-like” neurons). When the torque perturbation selectively displaced the elbow, these neurons quickly expressed a response appropriate to the underlying torque. Moreover, the integration of elbow information by cortical neurons preceded the shoulder's LLR by a delay consistent with a causal role of primary motor cortex. Interestingly, cerebellar damage leads to small LLRs with normal directional tuning, suggesting that its prominent impact on primary motor cortex is in scaling the magnitude of LLRs rather than patterning the multijoint integration (Kurtzer et al. 2013).
Given the length of the peripheral–central loop, muscle stretch information for a cortically mediated LLR must be transmitted by fast-conducting afferents. Similarly, fast-conducting afferents would have to underlie any brain stem and midbrain contributions to the LLR. The current evidence for reticular formation's role is indirect via an association with the startle reflex when subject generate a vigorous response to the perturbation (Ravichandran et al. 2013). No studies to date have perturbed the arm and directly examined perturbation-related activity in the reticular formation. One study has reported robust perturbation responses in the red nucleus of awake monkeys (Herter et al. 2015), but the rubrospinal tract is relatively small in humans (Nathan and Smith 1982), which suggests that it plays a minor role. Notably, individuals suffering cortical stoke exhibit abnormal coupling of arm muscle activity during voluntary movement (Dewald and Beer 2001) and in response to passive displacement (Sangani et al. 2008; Trumbower et al. 2010). These “muscle synergies” may reflect the abnormal activation of multijoint circuits for the LLR within the brain stem.
A final possible contributor to the arm's LLR is spinal processing of group II/slow-conducting signals from secondary muscle afferents. There is evidence that a group II spinal network is engaged with single-joint perturbations (Corna et al. 1995; Grey et al. 2001; Hendrie and Lee 1978; Meskers et al. 2010), but no study has examined their involvement with multijoint perturbations. For a more detailed treatment on the likely neural substrates by LLR's multijoint response, see Kurtzer (2015).
Limits of the proposed model.
The proposed model exclusively focuses on muscle spindles to relay information about limb position and motion. This neglects force feedback information from Golgi tendon organs (for review see Jami 1992) that project to spinal and supraspinal structures, including somatosensory cortex, and can help create well-formed behavior according to simulation studies (Buhrmann and Di Paolo 2014; Raphael et al. 2010). Relatively few studies have empirically tested the involvement of Golgi tendon organs in upper limb reflexes (Lewis et al. 2010; Mugge et al. 2010), but their evidence is positive, and a more complete model should include their contribution. Cutaneous sensors provide another mechanism for registering limb position and motion, since stretch patterns of the skin covary with joint movement, joint motion via skin stretch is encoded by slowly adapting II receptors/Ruffini corpuscles (Edin 2004), and skin stretch influences proprioceptive judgments of limb position (Collins et al. 2005; Edin and Johansson 1995). Evidence that skin sensors contribute to the LLR is patchy. Some studies report a blunting of the LLR with skin anesthesia (Loo and McCloskey 1985), whereas others have failed to find an impact (Bawa and McKenzie 1981). Given this ambiguity and that muscle stretch alone is sufficient to evoke SLRs and LLRs, either by mechanically displacing the tendon (Cody and Plant 1989) or puling on the muscle via magnetic force on an inserted pellet (Wolpaw and Colburn 1978), our current focus on muscle spindles seems reasonable.
Another important limitation of the model is its neglect of the shoulder-elbow biarticulars of the arm. The biarticular flexor and extensor exhibited preferred motion directions between the shoulder and elbow muscles and certainly contributed to the global response of the arm. Primary motor cortex may even have augmented influence over biarticulars (Gritsenko et al. 2011) since their flexion (or extension) moment at the shoulder and elbow mirrors the arm's flexion-extension interaction torques. However, we cannot determine a specific linkage with the arm's monoarticulars because no study has examined the biarticular's moment arm at both the shoulder and the elbow. The lack of information is surprising since there are studies on their moment arm at the elbow (Murray et al. 1995) and shoulder (Wood et al. 1989) separately along with modeling studies that include shoulder-elbow biarticulars (Franklin et al. 2007; Hu et al. 2012). Without precise anatomical information we cannot assess how afferent information from the biarticular sensors is routed to monoarticulars or how monoarticulars route sensory information to the biarticulars.
A final issue concerns the flexibility of the proposed model. Reciprocal excitation of shoulder and elbow flexors, reciprocal excitation of shoulder and elbow extensors, and reciprocal inhibition between flexors and extensors mirrors the inherent flexion-extension coupling of the shoulder and elbow. Accordingly, the general model is broadly appropriate for countering an imposed torque on an unencumbered arm and consistent with the shoulder's multijoint responses we have previously demonstrated during posture maintenance (Crevecoeur et al. 2012, 2013; Kurtzer at al. 2008; Nashed et al. 2015) and reaching to relatively simple targets (i.e., circles and ellipses) (Kurtzer et al. 2009; Nashed et al. 2012). Extensive research has shown that LLR can be modified to the spatial-temporal demands of the task (Calancie and Bawa 1985; Crago et al. 1976; Crevecoeur et al. 2013; Hammond 1956; Kurtzer et al. 2014; Lee and Tatton 1982; Lewis et al. 2006; Nashed et al. 2012; Pruszynski et al. 2008; Weiler et al. 2015), mechanical stability of the environment (Akazawa et al. 1983; Doemges and Rack 1992; Perreault et al. 2008; Krutky et al. 2010), and alterations in limb dynamics (Ahmadi-Pajouh et al. 2012; Cluff and Scott 2013). These changes reflect altered weighting of sensory afferents into motor commands and could extend beyond the general model we propose (e.g., reciprocal inhibition of shoulder and elbow extensors). Perhaps this general model would be useful as a guide to test the limits of LLR modifiability. Note that accumulating evidence indicates that LLR reflects the summation of two independent components: an automatic component (scaled by the background activity of the muscle) and a voluntary-related component (scaled by the voluntary goal) (Kurtzer et al. 2014; Lewis et al. 2005; Pruszynski et al. 2011b; Rothwell et al. 1980). The current experiment examined the automatic component, since subjects were instructed to ignore the transient perturbation.
Broader context and conclusion.
Previous studies have demonstrated multijoint integration by the shoulder's long-latency reflexes appropriate for the arm's mechanical properties (Cluff and Scott 2013; Crevecoeur et al. 2012; Kurtzer et al. 2008, 2009, 2013; Nashed et al. 2015). Given this extensive background, we predicted that elbow muscles should respond to shoulder displacement appropriate for the arm's intersegmental dynamics: elbow flexors (extensors) would express excitatory responses to shoulder extension (flexion). The present study unambiguously demonstrated this to be case and helps solidify that multijoint integration is a core feature of the arm's long-latency reflex. Providing this information in a paradigm also appropriate to assay multijoint responses by shoulder and biarticular muscles is critical to understanding coordination of the shoulder and elbow, one of the most extensively studied human behaviors. In this respect, previous studies on the shoulder provided less than half the story. The present work helps us to fit the pieces together and introduce a general model of reflex connectivity between shoulder and elbow monoarticulars that broadly accounts for the observed patterns. To our knowledge, this is the first such proposal for shoulder-elbow long-latency reflexes but remains incomplete.
Other capabilities exhibited by the long-latency reflex including modifiability with task goals (Hammond 1956; Pruszynski et al. 2008; Rothwell et al. 1980), sensory predictions (Crevecoeur and Scott 2013), and adaptation to novel dynamics (Ahmadi-Pajouh et al. 2012; Cluff and Scott 2013). The important role played by the long-latency reflex in sophisticated and flexible feedback control can be fruitfully understood as the output of an optimal feedback controller. These controllers possess an internal model of the body/environment, a state estimator that accounts for internal noise and prior history, and feedback gains sculpted to a global criteria of performance/cost function. Optimal feedback controllers have been proposed for behaviors ranging from whole body postural control to eye movements (Loeb et al. 1990; Munuera et al. 2009; Scott 2004, 2015; Shadmehr and Krakauer 2008; Todorov 2002). These controllers contrast with other models that utilize heuristic solutions that are efficient and effective but far less than optimal (de Rugy et al. 2012; Zhang et al. 2013). Future work is needed to determine what is a better characterization of the arm's long-latency reflexes.
This work was supported by National Institute of Child Health and Human Development Grant 5R24HD050821-09.
No conflicts of interest, financial or otherwise, are declared by the authors.
I.L.K. conception and design of research; I.L.K., J.M., N.P., and K.S. performed experiments; I.L.K. analyzed data; I.L.K. interpreted results of experiments; I.L.K. prepared figures; I.L.K. drafted manuscript; I.L.K. edited and revised manuscript; I.L.K. approved final version of manuscript.
We thank Dr. Tyler Cluff for a careful reading of the manuscript.
↵1 The earlier studies by Kurtzer and colleagues (e.g., Crevecoeur et al. 2012; Kurtzer et al. 2008, 2009; Nashed et al. 2105) applied the same pattern of torque for all subjects to study the shoulder muscle responses. The design took advantage of a notable property of two-joint dynamics. When the elbow is positioned at 90°, then 1) torques having the same sign and magnitude at the shoulder and elbow will cause initial acceleration of only the elbow, and 2) an applied shoulder torque will induce the same initial shoulder acceleration as an applied elbow torque if the two torques are opposite in sign and equal in magnitude. These patterns are obtained regardless of the length and mass of the individual's forearm and upper arm because the complex nonlinear dynamic equations of the shoulder and elbow (from Soechting and Lacquaniti 1988) simplify to θ̈e = Te/If + (Te − Ts)/Ia and θ̈s = (Ts − Te)/Ia, where θ̈e and θ̈s are elbow and shoulder acceleration, respectively; Te and Ts are elbow and shoulder torque, respectively; and If and Ia are forearm and arm inertia, respectively. The equations also reveal that a design of fixed torques is not suited to study the responses of elbow muscles, since the comparable conditions must be tailored to the inertia of the subject's arm.
↵2 Studies utilizing tendon taps (Lewis and McNair 2010) and electrical nerve stimulation of the nerve (Cavallari and Katz 1989; Cavallari et al. 1992) have reported across-muscle/heteronymous responses at short latency. The discrepancy between different findings using different techniques should be rectified by future studies.
- Copyright © 2016 the American Physiological Society