Real-Time Prosthetic Digit Actuation by Optical Read-out of Activity-Dependent Calcium Signals in an Ex Vivo Peripheral Nerve

Abstract

Improved neural interfacing strategies are needed for the full articulation of advanced prostheses. To address limitations of existing control interface designs, the work of our laboratory has presented an optical approach to reading activity from individual nerve fibers using activity-dependent calcium transients. Here, we demonstrate the feasibility of such signals to control prosthesis actuation by using the axonal fluorescence signal in an ex vivo mouse nerve to drive a prosthetic digit in real-time. Additionally, signals of varying action potential frequency are streamed post hoc to the prosthesis, showing graded motor output and the potential for proportional neural control. This proof-of-concept work is a novel demonstration of the functional use of activity-dependent optical read-out in the nerve.

I. Introduction

Upper-limb prosthetic devices have developed significantly in recent years [1,2]; yet full and intuitive control of multiple degrees-of-motion remains a challenge due to inadequate control interfaces. Perhaps the largest limitation of current neural interface strategies is the lack of sufficient control sites with which to decipher and control the numerous degrees of freedom needed for full arm/hand replacements. Myoelectric control techniques using surface electromyographic (EMG) signals are limited in the number of independent control signals that can be acquired and are therefore not well-suited for the control of complex hand articulations. While recent advances in implantable myoelectric sensors (IMES) may enable independent signal extraction from specific muscles [3], a peripheral nerve interface (PNI) or brain-machine interface (BMI) have the potential to overcome the shortcomings of EMG by providing both control and sensory feedback directly from/to the nervous system. Due to the small axon-level signals from electrode-based PNI designs, these interfaces have had difficulty achieving the specificity needed for control signal read-out. Cuff electrode strategies like the flat nerve interface electrode (FINE) have demonstrated selective induction of sensory percepts, but deciphering minute signals for highly specific axonal read-out with electrode-arrays has proved elusive [4–7].

Due to the inherent difficulty of electrodes to decipher highly specific (single axon level) activity in a long-term interface, we have investigated an optical approach to neural interrogation at the peripheral nerve level. An optical system can be used to read changes in activity-associated light output from single neurons or neuronal processes. Recent groups have demonstrated optical read-out of activity with calcium reporters in the in vivo mammalian cortex for BMI applications [8–10]. However, surgical intervention in the brain may not be practical for certain limb rehabilitation cases. Our group has focused on the peripheral nerve interface [11], where neural processes to the limb are available in a more spatially concentrated arrangement relative to cortex. We have characterized activity-dependent fluorescent signals in peripheral nerve axons ex vivo using a calcium-sensitive fluorophore [12] as well as with a virally transduced GCaMP sensor [13]. In this work, the gradation in signal amplitude as a function of action potential frequency was shown.

In the present study, we demonstrate the concept of using optical read-out of activity dependent calcium signals as a mode of control for externally-powered prosthesis actuation. Given that fluorescent calcium transients can provide information on the presence and intensity of neural activity, monitoring these signals in motor axons may allow a neuroprosthetic system to monitor the motor command intended for an absent biological actuator and drive the prosthetic device in its place. In this proof-of-concept work, the finger of an electrically powered prosthetic hand is commanded in real-time by axonal activity via an optical calcium-associated signal. In addition to this real-time actuation, graded frequency-modulated signals are streamed post hoc to the prosthetic hand to demonstrate varying degrees of finger flexion in response to graded neural command intensity.

II. Methods

A. Nerve Preparation

The sciatic nerve and its tibial nerve branch are excised from adult wild type mice and loaded from the tibial end with a synthetic calcium indicator (2 mM Calcium Green-1 Dextran, ex/em = 506/531 nm) dissolved in a buffer containing 130 mM KCl and 30 mM MOPS, pH 7.2 (see Supplementary Figure 1, Fontaine et al, 2017 [12]). The tibial end is freshly cut in a zero-calcium buffer (Mouse Saline with Ca²⁺ replaced with Mg²⁺, and 1mM EGTA added) to ensure open axon cylinders before being suctioned into a tight-fit electrode with the dye buffer to facilitate longitudinal axonal dye-loading via diffusion and/or axoplasmic transport. The suction electrode on the tibial nerve also serves to record electrical activity within the nerve. The sciatic end of the nerve is drawn into a suction electrode for electrical stimulation of compound action potentials (CAPs). The nerve sample is stored initially in a mouse saline solution (in mM: 126 NaCl, 5 KCl, 1.8 CaCl₂, 1 MgCl₂, 10 MOPS Buffer pH 7.2, 30 glucose) during experimental preparation and is perfused with a modified Tyrode’s solution (in mM: 126 NaCl, 3 KCl, 2 MgSO₄, 20 NaHCO₃, 1.2 NaH₂PO₄, 2 CaCl₂, 30 glucose) bubbled with 95%0₂/5%CO₂, for the duration of the two-hour dye loading period and subsequent imaging/electrophysiology.

The use of animals was approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Colorado Health Sciences Center, with accreditation by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC). All experiments were performed in accordance with IACUC regulations and approved protocol.

B. Electrophysiology

CAPs are generated and recorded throughout the experiment (MultiClamp 700B Amplifier, Axon Instruments, PCLAMP10 Software) using 50 μs square pulses to confirm and monitor nerve viability. The stimulation voltage threshold for maximum CAP amplitude is determined, and a modestly supra-threshold voltage is subsequently used to elicit calcium signals. CAP amplitudes were monitored throughout the duration of the incubation period, to confirm stable nerve health.

C. Electrode and Chamber Fabrication

Electrodes were custom fabricated; capillary glass was heat extruded and the annulus custom-sized to the nerve diameter with a diamond scribe and fire polishing. The electrical leads were made from Teflon-coated silver wire, the ends of which were stripped and ‘chlorided’ using bleach. The chamber was designed in SolidWorks software and produced using a 3D printer.

D. Optical Imaging/Recording

Dye labeled axons were imaged in a region of nerve near the tibial recording electrode. The nerve was gently pressed to the optical glass of the chamber with low-tension silk strings attached to a small weight for imaging on an inverted microscope. Placement of the small ‘harp-like’ device did not affect the CAP. Fluorescence imaging was performed on a spinning disk confocal microscope (Intelligent Imaging Innovations, Marianas). A 515nm laser line was used to excite the Calcium Green-1. Pixels were generally binned (2x2) to improve the frame read-out time for fast imaging. To record calcium transients, time-lapse images were acquired at 12-20Hz, during which the nerve was stimulated by an electrical stimulator triggered via TTL pulses from the microscope. Fluorescence was imaged onto an EMCCD camera (Photometrics Evolve) through a 525/50nm emission filter. Images were collected with a 63X, 1.4NA oil-immersion objective lens. Photobleaching of the signal was kept minimal by the reduction of laser power and exposure, and any mild decay due to photobleaching was not removed.

E. Control Interface and Method

A standard laptop computer served as the interface between the optical recording and the prosthetic hand. Time-lapse images from the microscope were acquired using SlideBook 6.0 software (Intelligent Imaging Innovations). SlideBook sent the raw image data to a custom Matlab program (Mathworks, MA) which controlled the motors of the prosthetic hand. A setup function within the Matlab script established the serial communication between the computer and the prosthetic hand. A second function received the time-lapse captures from SlideBook and translated the image data into an optical signal by averaging nodal ROI pixel intensities at each frame. The optical signal was used to control the velocity of the prosthetic finger motion. The amplitude of the signal above or below the predefined optical signal threshold determined the speed and direction of the finger motion. Velocity gains were adjusted to achieve a full range of motion.

F. Prosthetic Hand Design

The electronics in the original Bebionic v2 hand (RSL Steeper, UK) were replaced with a custom motor controller system (Sigenics Inc., Chicago, IL) and included a central Arduino controller board and six satellite boards referred to as ‘penny boards’ (as they were the size of a penny). Each penny board was connected by a four-wire I2C bus with each board associated with an individual actuator and finger. Motor commands indicating the speed and direction of motion for the driven finger were sent from the Matlab script to the Arduino (SparkFun Electronics, Boulder, CO) which converted the serial commands into I2C communication. Position encoder values from the prosthetic finger motor were recorded simultaneously and converted to joint angle measurements post hoc. The Bebionic v2 hand is a commercially available five degree-of-actuation (DoA), multi-functional prosthesis with a manually positioned thumb abduction joint (Figure 1a). The prosthetic hand was modified by installing the penny boards in order to allow control of individual digits (Figure 1b). The speed of the digits remained the same as the original commercial device (1 second to open/close 90 degrees, or 15 rpm). All other features of the prosthesis remained intact.

Figure 1.

(a) Commercially available Bebionic v2 hand (b) Modified Bebionic hand used for finger actuation experiments. Custom electronics were installed in order to control individual motors within the prosthesis.

III. Results

A. Real-Time Control

A single axon was selected for the real-time control, which yielded a visible fluorescence response to the emulated motor command. The calcium response, consistent with many observed nodal responses, originated at the node of Ranvier center and propagated bi-directionally into the internodal region of the axon. The nodal region, which was taken as the motor control signal, showed an approximately 12% change in fluorescence intensity. This signal amplitude was comparable to those achieved in prior work at the same action potential frequency (100 Hz) [12]. Under the velocity control paradigm employed here, the digit drove flexion for the duration of the supra-threshold optical signal, and extended back to its resting state as the signal diminished below threshold (Figure 2).

Figure 2.

Real-time prosthetic digit actuation by action potential evoked calcium fluorescence signal in a peripheral nerve axon. (a) Confocal images of a CalciumGreen-1-Dextran-loaded axon node of Ranvier used to control finger actuation, shown before, during and after the activity-induced fluorescent signal (scale bar 10μm). (b) Quantitative trace of the calcium-fluorescence signal in response to the 1s, 100Hz train of action potentials (black bar). (c) Prosthetic hand’s middle finger flexes and extends under control of the optical signal from panel b. Virtual red dot denotes the tip of the middle finger driven in the experiment. (d) Corresponding finger joint angle illustrates digit flexion occurring during supra-threshold optical control signal.

B. Proportional Control Demonstration

Previously recorded signals collected over a range of action potential frequencies were used post-hoc to drive prosthetic finger actuation. As characterized in earlier work [12] average fluorescence amplitudes of sustained stimulus are linearly modulated by the frequency of action potentials. Such graded signals can thus encode intensity of the motor command. The prosthetic finger actuated to varying degrees of flexion, depending on its action potential frequency-modulated control signal ranging from 25-125Hz (Figure 3).

Figure 3.

Motor flexion of the prosthetic digit is graded by the action potential frequency of the optical calcium signal. (a) Graded calcium-fluorescence transients in an axon node of Ranvier in response to a range of action potential frequencies. (b) Resulting finger joint angles of the prosthetic finger as driven with control signals from panel a.

Previous work [14,15] suggests that the range of action potential frequencies used to drive the prosthesis in this study is physiologically relevant, as motor units are capable of firing at or higher than the upper-bound frequency in these experiments. Thus, the real-time control signal which derived from a 1 second, 100 Hz action potential burst would likely correspond to a relatively short, high-force motor command.

IV. Discussion

In the present work we have demonstrated prosthesis actuation under real-time control of neural activity in a peripheral nerve axon. The neural activity is transduced to an optically emitting signal via a fluorescent calcium sensor, which was loaded into the axonal cylinder via diffusion and or axoplasmic transport. By interfacing microscope software with custom Matlab script, fluorescence intensity data could be streamed into a real-time signal of neural activity and command prosthetic finger motor control. The configuration was used to drive the prosthetic finger using a velocity control paradigm to achieve basic flexion and extension of the digit. Furthermore, via pre-recorded signals, the prosthesis was driven to varying flexion joint angles by neural signals of differing action potential frequency, illustrating the potential for neural intensity modulated motor output.

Prior studies have optically stimulated peripheral nerve axons for functional modulation of motor units [16–18] and pain pathways [19] using the light-activated ChannelRhodopsin2 (ChR2); however, there is an absence of literature describing the use of optically obtained signals of peripheral axon activity. While the present experiments demonstrate this using an ex vivo model, in vivo translation is feasible using the genetically encoded calcium indicator GCaMP, which has been used extensively to image neuronal activity in in vivo animal models [8–10, 20–24].

The experiments presented here provide a relatively rudimentary proof-of-concept for a control mechanism which may be used for future multiple degree-of-freedom control. The future development of an operational PNI using this technique will require the continued improvement and integration of several technologies. Fiber-coupled microscope systems using optical fiber bundles are in development [25, 26], which may enable the appropriate light deliverance to and from in vivo neural targets. Sufficient scanning speed within the neural tissue may be achieved with electro-wetting lens technology [27] for rapid depth dimension scanning. Piezoelectric actuators, microscanners or other devices, situated at the extracorporeal end of the fiber would provide lateral scanning. Lasers for multi-photon interrogation must be improved to achieve maximal tissue penetration, while ultimately becoming more compact and portable. With adequate scanning speed and range within the nerve, a set of nodes would be serially scanned to monitor their activity. Detectable nodes within the scan range would first be tested to identify their physiological role by running through a protocol of visualized/imagined movements with the subject. Nodal signals would be pre-calibrated to map signal amplitude to appropriate prosthetic motor output, by deriving the frequency-amplitude slope. Ideally, at least one nodal response could be identified and pre-calibrated for each digit. By virtue of the analog nature of a calcium response to trains of action potentials, pertinent nodal activities could be scanned in a serial fashion with reasonable sampling rate per node, given sufficient scan speed.