In this work, we investigated a novel approach for a human-machine interface where the activity of descending tracts in the spinal cord is proposed as a means of generating motor control signals. The main focus of this study was to show: 1) the feasibility of long-term recording from the spinal cord descending tracts, 2) that these signals can be recorded with similar fidelity without deterioration of the signal-to-noise ratio after the first few weeks of acute tissue response, and 3) show that the information encoded in these signals are correlated to the motor function. Rousche and Normann [57
] have shown reliable recordings from the cat sensory cortex for several months using arrays. Recent studies reported that stable signals could be recorded over long term using these electrodes in non-human primates [58
] and humans [9
]. However, all of these investigations involved electrodes implanted in the cerebral cortex. This study proposes an alternative method for human-machine interfacing by shifting the paradigm from the brain to the spinal cord.
] and Sacrey [61
] have shown similarities in forelimb movements and velocity profiles between humans and rats, suggesting that the rat is a suitable model to study the neural control of forelimbs. The RST at the cervical level contains information required for controlling forelimb movements in rats and other species [35
]. The same implant site in the dorsolateral funiculus of the human spinal cord would primarily record the neural activity of the lateral corticospinal tract (rather than the RST), which is the primary descending pathway in humans. Segments C5-C6 were chosen as the implant site for recording RST signals in rats since it has been reported that RST axons and its collaterals present in C5-T1 spinal cord segments terminate onto the motor neuron pools at these levels that are involved in the control of the forelimb movements [62
]. Note that electrode implants at a higher level may have compromised diaphragm function leading to respiratory complications.
The dorsolateral funiculus occupies a large portion of the cervical spinal cord across species. In the rat, the RST tract is found in the dorsolateral funiculus just underneath a thin layer of sensory tract, i.e.
dorsal spinocerebellar tract and thus accessible with electrodes of 1
mm length. A surface electrode would not be suitable as it would be recording mostly sensory information. Therefore, commercially available MEAs with penetrating shanks were used as the electrode of choice. The rat spinal cord is too small for implantation of a 100-channel standard Blackrock array. A custom-made 15-channel array with 1
mm electrode shanks was selected instead. Increasing the number of channels would have enabled us to sample a larger set of signals, but it would also increase the size of the MEA hence the trauma to the spinal cord.
Microelectrodes implanted into the white matter of the spinal cord are surrounded by a dense population of axons. Multi-unit activities recorded by these micro-electrodes may thus be insensitive to micro motions. This may be the most important factor for the stability of the recordings. The outcome measures to evaluate long-term stability in the recordings were the variability of the SNR and the mean amplitude of the signal during behavior. The mean signal amplitudes and SNRs were found to vary during the implant duration (Figure ). However, the signals do not deteriorate substantially or disappear after the acute tissue response, or change in amplitude during the recording time, some of which are common problems with single neuron recordings from the cortex. Some signal amplitude change can also be attributed to the variations in the way the rats perform the behavior. The SNR values in general were low in our case, which may be due to the background neural activity that was considered as the baseline noise in our analysis. The signal mean for the duration of the entire behavior is a more robust parameter to quantify the activity rather than the signal peak, which fluctuates substantially from trial to trial. Using the standard deviations of signal and noise also produced modest yet robust values for SNR.
It is still debated whether the cortical area M1 encodes forelimb dynamics [63
] or kinematics such as position, velocity, and acceleration [70
]. Vargas-Irwin [74
] showed that M1 encodes sufficient information to reconstruct some of the kinematic features such as joint angles of arm, wrist, and hand during a reach-and-grasp task. Miller and Sinkjaer [75
] showed that the magnocellular portion of the red nucleus (RNm), where the RST originates [76
] encodes the dynamics of limb muscle activity. The spinal cord circuitry is responsible for integrating these signals to activate various groups of muscles. Therefore, it is plausible that the signals in these spinal cord motor tracts encode both dynamic and kinematic parameters of a voluntary movement.
In this study, we concentrated only on forelimb kinematics. The forelimb motor function was hypothesized to have a stronger correlation to the spinal cord motor signals since these signals are closer to the target muscles in the signal pathway than the higher centers in the brain. The degree of success in reconstructing limb kinematics, determined by the correlation coefficient, was the outcome measure. High degrees of correlation were obtained for all the reconstructed kinematic functions in all the animals suggesting that some aspects of the forelimb kinematics are encoded by the spinal cord signals.
The performance of any neural prostheses would depend on the reproducibility of the control signals. The size of the implanted electrode, insertion technique, and the glial response [77
] are some of the major factors that determine the stability of the interface over the long term. Recent reports by Bamford [86
] showed that spinal cord tissue tolerated microwire implants up to one month post-implant. However, Utah arrays used here were significantly larger and rigid in structure compared to the single microwires. Despite the evoked immune response and the tissue encapsulation observed around the electrode shanks, animals used in the study were not at any time during the recording period paralyzed due to the implant. Paralysis was observed in some cases where the spinal cord was damaged during the surgery. Those animals were euthanized and no data were included in the analysis from them. The immune response also did not significantly degrade the neural signals during our recording period as shown by moderate changes in the mean signal amplitude between the first and the last recording sessions (Figure ).
Limitations of the study
One of the major limitations of the study so far has been the inability to record for more than three months. This is mainly due to the lack of a suitable electrode for the rat spinal cord. The use of surface electrodes is attractive because of reduced invasiveness compared to the penetrating electrodes but the anatomical locations of the spinal cord motor tracts does not allow us to use these electrodes. The signals recorded from a surface electrode would contain sensory contamination from the spinocerebellar tracts residing close to the spinal cord surface on the dorsolateral side. Penetrating Utah MEAs were used for study since they were commercially available and the electrode shanks were arranged in well-defined positions. There was also not a large variation between electrode sizes and impedances due the fabrication process. These MEAs were able to record stable signals in our study up to 3
months. However, histological studies showed that they caused significant damage around the implant site. These MEAs also had a solid and thick base which contorted the shape of the cord as seen in histological pictures.
We tested custom-made 25
μm diameter Pt-Ir microwire electrodes implanted bilaterally in a separate study [56
]. Using these electrodes, we were able to show that the contralateral cord did not contain observable activity during a task (Figure ). These microwires were individually inserted into the cord and the surgical procedure was significantly more tedious and longer than MEA implants. However, the histological analysis in these animals showed that the glial scarring and the tissue response were significantly lesser compared to the MEA implants. Only mild astrocytosis and gliosis were observed in the microwire implants. We have not been able to record for more than four weeks in these microwire implanted animals because of wire breakage.
Another major limitation of this study was that the reaching kinematics was reconstructed within individual trials and not generalizable across a set of test trials using the beta coefficients from a training set. The variation in beta coefficients might be because of the slight differences in how the animal reaches for the food pellet between trials. Coefficient variations may have also been due to the use of a single camera to capture the forelimb kinematics in one plane only. Our laboratory has started to use a three camera system with faster frame grabbing rate to improve the behavioral imaging.