Brunton, E. K., Silveira, C., Rosenberg, J., Schiefer, M. A., Riddell, J. and Nazarpour,
K. (2019) Temporal modulation of the response of sensory fibers to paired-pulse
stimulation. IEEE Transactions on Neural Systems and Rehabilitation Engineering,
27(9), pp. 1676-1683.
There may be differences between this version and the published version. You are
advised to consult the publisher’s version if you wish to cite from it.
http://eprints.gla.ac.uk/203625/
Deposited on: 18 November 2019
Enlighten – Research publications by members of the University of Glasgow
http://eprints.gla.ac.uk
SUBMITTED TO IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING 1
Temporal Modulation of the Response of Sensory
Fibers to Paired-Pulse Stimulation
Emma K. Brunton, Member, IEEE, Carolina Silveira, Student Member, IEEE, Joshua Rosenberg, Matthew A.
Schiefer, Member, IEEE, John Riddell, and Kianoush Nazarpour, Senior Member, IEEE
Abstract—Multi-channel nerve cuff electrode arrays can pro-
vide sensory feedback to prosthesis users. To develop efficacious
stimulation protocols an understanding of the impact that spatio-
temporal patterned stimulation can have on the response of the
sensory fibers is crucial. We used experimental and modelling
methods to investigate the response of nerve fibers to paired-
pulse stimulation. Nerve cuff electrode arrays were implanted
for stimulation of the sciatic nerves of rats and the sensory
compound action potentials were recorded from the L
4
dorsal
root. A model of the nerve cuff electrode array and sciatic
nerve was also developed. The experimental and modelling results
were compared. Experiments showed that it took 8 ms for
the sensory fibers to completely recover from a conditioning
stimulus, regardless of the relative position of the electrodes used
for stimulation. The results demonstrate that the electrodes on
the cuff cannot be considered independent. Additionally, at the
stimulus level used here, there is a large overlap in the fibers
that were activated by the different electrodes. If a stimulus
paradigm considered the electrodes as independent, stimuli from
the different electrodes would need to be interleaved, and the
intervals between the stimuli should be greater than 8 ms.
I. INTRODUCTION
A sense of touch is vital when it comes to interacting and
experiencing the world around us [1], [2]. In the case of
limb difference (loss or absence of limb), mechatronic hands
have advanced over the past decades, however, the addition
of sensory perception is still in its infancy [2]. Providing
prosthetic hand users with sensory perception has been shown
not only to greatly improve control of the hand, but also
promote a sense of embodiment and reduce phantom limb pain
[1], [3]. Substituting sensation with external devices has been
shown to help in laboratory settings, however, none of these
devices have been widely adopted [2]. Electrical stimulation
of the nerves in the residual limb has the potential to provide
sensory information from a prosthetic hand [4]–[10].
A number of devices that interface directly with the pe-
ripheral nerves have been developed to provide electrode
This work is supported by the UK Engineering and Physical Sciences
Research Council (EPSRC) research grants EP/M025977/1, EP/N023080/1
and EP/R004242/1.
Corresponding authors are E. Brunton and K. Nazarpour. Emails:
{emma.brunton,kianoush.nazarpour}@newcastle.ac.uk
E. Brunton and C. Silveira are with the School of Engineering, Newcastle
University, Newcastle-upon-Tyne, UK.
J. Rosenberg is with the Department of Biomedical Engineering, Case
Western Reserve University, Cleveland, Ohio, USA.
M. Schiefer is with the Malcom Randall VA Medical Center, Gainesville,
Florida, USA.
J. Riddell is with the Institute of Neuroscience and Psychology, University
of Glasgow, UK.
K. Nazarpour is with the School of Engineering and the Institute of
Neuroscience, Newcastle University, UK.
stimulation [3], [11]. These neural interfaces include intrafas-
cicular electrodes that penetrate the nerves (TIMEs [6], [12],
[13], LIFEs [14], [15]) and cuff electrode arrays that wrap
around the nerve without penetrating it (Spiral cuffs [1], [16],
FINEs [17], [18]). Generally, nerve cuff electrode arrays do
not stimulate as selectively as intrafasciular electrode arrays.
However, they have been shown to provide a stable interface
with the nerve, and to evoke realistic sensory sensations [1].
Testing of multi-channel cuffs in humans has mainly fo-
cused on the ability to modulate the perceived sensation by
tuning the frequency, amplitude, or pulse width delivered by
a single electrode [1], [2], [4], [6]. Spatio-temporal patterns
of electrical stimulation delivered from multiple electrodes
has the potential to provide patients with different sensations
experienced concurrently. In cochlear [19], [20] and retinal
implants [21] interactions between electrodes can greatly alter
the resultant percept. For example, stimulating the nerve with
two electrodes simultaneously results in significant interac-
tions between adjacent electrodes due to the vector summation
of their electric fields [20], [22], [23]. As a result, cochlear
implants employ strategies that interleave stimuli to avoid
electrode interactions [19], [20]. Additionally, even after the
electric field applied by an electrode has been removed, the
nerve still needs to recover [24]. This can result in changes
to the response of nerve fibers to the same stimulus even if
stimuli are applied asynchronously. Therefore, it cannot be
assumed that asynchronous stimuli will produce independent
percepts and stimulus paradigms that consider electrode chan-
nel independence will need to carefully consider the timings
between sequential stimuli. In stimulus paradigms that move
beyond electrode channel independence, interactions between
electrodes could be taken advantage of [20]. This is the case
in current steering, also known as field shaping, where the
electric fields generated by two or more electrodes stimulated
simultaneously are combined to target a specific population
of fibers within the nerve [16], [25]–[27]. These studies
show that knowledge of the spatio-temporal interactions of an
electrode array is essential for developing effective patterns of
stimulation in a sensory prosthesis.
Spatio-temporal interaction studies to date in peripheral
nerves, e.g. the sciatic nerve, have been limited to their effects
on motor fibers, as the response of these fibers can be inferred
from twitch force [28]–[30] and ankle torque [17], [31] mea-
surements. These studies have shown that both intrafascicular
[28], [29], [32] and nerve cuff electrode arrays [16], [17],
[33] can be used to selectively stimulate motor fibers from
different branches of the sciatic nerve. In addition, they have
Page 1 of 11
http://mc.manuscriptcentral.com/tnsre-embs
Transactions on Neural Systems and Rehabilitation Engineering
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
SUBMITTED TO IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING 2
shown that through interleaving multi-site stimulation, fatigue-
resistant and ripple-free motor responses can be generated
[30], [33]. Due to the difficulty of isolating sensory fibers,
little work has been done to determine if sensory fibers will
behave in the same way. However, we expect that sensory
fibers will behave in a similar fashion to motor fibers, albeit
with a lower threshold to generate an action potential [1].
Modelling of the electric fields generated in combination
with simulations of axon populations are invaluable in the
study of spatio-temporal interactions. This is because exami-
nation of a large range of parameters would not be feasible
to test in clinical or preclinical studies [34]. Models allow for
the effects of the location of the electrode contact in relation
to fascicles or nodes of Ranvier to be investigated [31], [35].
This would be near impossible to test in-vivo. Models can also
provide greater insight into what state both the fast and slow
acting voltage-gated sodium channels are in [36].
We examine how spatio-temporally patterned stimulation of
the sciatic nerve affects the sensory responses on the L
4
dorsal
root. We compare the results from laboratory experiments and
computer modelling. We characterize the effects of varying
both the delay between sequential stimuli, and the spatial
location of the electrodes on the compound action potentials
(CAPs) generated at L
4
dorsal root. We address two questions:
(1) whether or not stimulation from different electrodes on the
multi-channel cuff could be considered independent; and (2) if
the electrodes can not be considered independent, what inter-
stimulus interval is required so that these interactions do not
have an effect on the response of the sensory fibers.
II. METHODS
We describe the experimental and simulation studies used
to investigate the spatio-temporal interactions of stimuli that
were delivered with a multi-channel cuff electrode array.
A. Animal Preparation
All procedures were performed under appropriate licences
issued by the UK Home Office under the Animals (Scientific
Procedures, Act, 1986) and were approved by the Animal
Welfare and Ethical Review Board of Newcastle University.
Four Sprague Dawley rats were used in this study weighing
from 400 to 475 grams. Anaesthesia was induced in a box with
3% isoflurane in Oxygen. After anaesthesia was induced, the
animal was moved onto a surgical table where anaesthesia
was maintained through a mask. To help maintain anaesthetic
depth, a subcutaneous injection of meloxicam was given at
a dose of 1 mg/kg. Anaesthetic depth was assessed through
monitoring of the animal’s heart and breathing rates and
its responses to noxious toe pinches. Anaesthetic level was
adjusted as needed throughout the procedure. Fluids were
delivered through a tail vein cannula at 0.2 ml/hour (20 ml
0.9% NaCl and 5% glucose, with 0.05 ml KCl).
An incicision in the skin was made over the L
2
to L
6
vertebrae (Fig. 1a). Muscle tissue was thoroughly cleared from
around the L
6
spinous process for placement of the ground
electrode. The L
6
spinous process was left in place and a
tungsten wire was wrapped around it to act as a ground
electrode for recordings. The wire was then secured with
dental acrylic. To expose the L
4
dorsal root a restricted lateral-
medial laminectomy was performed. The opening was then
covered in saline and gauze to keep the tissue wet while the
rest of the surgery was carried out.
A concentric nerve cuff electrode (Microprobes for Life-
science, USA) was implanted on the proximal side of the
sciatic nerve following procedures described previously [37],
[38]. Briefly, an incision was made in the skin approximately
0.5 cm caudal and parallel to the right femur. The two planes of
the biceps femoris muscle were dissected to expose the sciatic
nerve (Fig. 1a). The nerve was freed from the surrounding
tissue in preparation for implantation of the cuff electrode
array. Two tungsten wire hooks were placed in the tibialis
anterior (TA) muscle to monitor electromygraphy (EMG).
The cuff electrode arrays had an inner diameter of 1 mm
with sixteen channels arranged in four rings of four contacts
(Fig. 1b). Each ring was separated by 0.75 mm. Each contact
was made from 100 µm platinum wire and had a surface area
of approximately 0.0629 mm
2
. All other electrodes were made
in-house from tungsten insulated wire of 125 µm diameter
(Advent Research Materials, UK).
After the cuff electrode array was secured with Kwik-Cast
(World Precision Instruments, USA), the muscles and skin
were closed above the nerve cuff with tissue glue and the
gauze and saline were removed from the opening above the
spinal cord. The dura was cut to expose all the spinal roots.
The L
4
dorsal root was identified after locating the L
4
dorsal
root ganglion. The L
4
dorsal root was then separated from the
others, lifted and placed across tungsten wire hook electrodes
using a glass hook. The tungsten hooks were separated by
approximately 1 mm, and connected to form a bipolar pair
with an electrode located 2 mm away (Fig. 1c). The root was
placed over three hooks. Only one bipolar channel recorded,
if it was not long enough to be placed over the four hooks
without stretching. Otherwise, the root was placed over four
hooks and two bipolar channels were recorded. The opening
was then filled with paraffin oil to insulate the recording
electrodes from the surrounding tissue.
B. Neural Recording
CAPs were recorded using bipolar hook electrodes placed
on the L
4
dorsal root. While the rat sciatic nerve contains
sensory fibers that project to the dorsal root ganglia from L
3
to L
6
, we chose to record from L
4
due to space restrictions,
and that 98-99% of sciative nerve neurons project to either L
4
or L
5
[39]. However, this does means that we may not have
been capturing the complete effects of the stimulation. The
electroneurographic (ENG) signals were bandpassed filtered
between 10 and 5000 Hz, and amplified using a differen-
tial amplifier (A-M systems
TM
, USA). The output from the
amplifier was connected to an analogue input of a Cerebus
Neural Signal Processor (Blackrock Microsystems, USA) and
sampled at a rate of 30 kHz.
C. Neural Stimulation
Stimuli were delivered to the sciatic nerve through each
electrode on the 16-channel cuff electrode array using a Ceres-
Page 2 of 11
http://mc.manuscriptcentral.com/tnsre-embs
Transactions on Neural Systems and Rehabilitation Engineering
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
SUBMITTED TO IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING 3
14
10
2
6
Ø=1mm
L=4.25 mm
Th=0.2mm
0.75mm
Ø=0.1mm
0.2-0.5mm
+
-
L
4
Dorsal root
-
+
Tungsten
Hook
1 mm
Connector
To Amplier
0 -10
500
Conditioning
(electrode 1)
Test
(electrode 2)
Normalization
(electrode 2)
B
Sciatic
Nerve
1
3
7
5
9
11
13
15
4
8
12
16
e
n
e
n+1
e
n+2
e
n+4
C
E
D
...
...
e
n+3
0.2 0.1 0.2
measurements in ms
A
Stimulating Site
D
Recording Site
Stimulation
Sites
Recording Site
x
Fig. 1. Experiment Setup: (a) Animal preparation. A multi-channel cuff electrode array was implanted on the sciatic nerve to deliver stimulations, recordings
were made from the L4 dorsal root, x marks the approximate location of the current return before the skin was closed; (b) Illustration of the nerve cuff
electrode array arrangement indicating an example of the electrode pair combinations used as described in Table 1; (c) Recordings of compound action
potentials were made using hook electrodes placed on the L
4
dorsal root; (d) The paired-pulse paradigm used for stimulation; (e) A finite element model of
the nerve cuff electrode array and rat sciatic nerve.
tim R96 (Blackrock Microsystems, USA). The experiment was
conducted in two parts. First, the threshold current required to
elicit a CAP that could be identified on a single-trial basis
on an oscilloscope was found. The stimuli used to determine
threshold and all subsequent stimulations were monopolar,
biphasic, cathodic first, current pulses with a pulse width of
200 µs and an inter-pulse-interval of 100 µs. The current return
path was a tungsten wire placed in the skin, above the sciatic
nerve. All parameters were kept constant except for the current
amplitude. The current amplitude was initially set to 40 µA
and stepped up or down at intervals of 5 µA. When close
to the threshold current, the step size was reduced to 1 µA.
After finding threshold a current that generated the maximum
CAP was found by finding a current that when the amplitude
was increased further, no detectable increase in CAP could
be seen. The current was then increased beyond this level to
ensure that the maximum CAP was recorded. Each electrode
was stimulated 10 times at threshold, 120% of threshold and
at a current amplitude that generated the maximum response.
The recordings were averaged across the trials. In Animal 2,
electrode 14 was broken. This electrode was removed from all
animals so that equal comparisons could be made.
For part 2, the nerve was stimulated with a pair of electrodes
using current amplitudes of 120% of threshold. A paired-pulse
stimulation paradigm was used where a first “conditioning”
pulse was sent from one electrode, e
n
, (n = 1, 2, ..., 12). A
second “test” pulse was then sent from a second electrode that
could be in one of five possible locations relative to the first
electrode (e
n
, e
n+1
, e
n+2
, e
n+3
or e
n+4
) as illustrated in Fig.
1b. These five possible locations are described in more detail
in Table 1 and were labelled as: “origin”, 90 degrees, 180
degrees, 270 degrees and 0 degrees. In Animal 1, the origin
location was not tested.
The time period between the conditioning and test pulses
was varied from 0 to 10 ms in 1 ms steps. Preceding the
conditioning pulse by 0.5 seconds, a single “normalization”
pulse, identical to the test pulse was delivered, as illustrated
in Fig. 1d. The normalization pulse was used to normalize the
CAP to account for any changes in the nerve’s responsiveness
over time. Each stimulus combination was repeated 10 times.
D. Analysis of the ENG recordings
The ENG recordings were analysed offline in MATLAB
TM
.
In the interest of consistency between all animals, only one
bipolar channel was used for data analysis. A synchronisation
signal from the Cerestim was used to segment the data. The
recordings from the 10 repeats for each stimulus combination
was then averaged over a period of 0 to 20 ms, where 0
ms corresponded to the detection of the rising edge of the
synchronisation signal. The peak-to-peak response of the CAP
was calculated by first finding the minimum and maximum
potentials recorded over the time period of 1.5 to 3 ms after
the initiation of the test pulse. The minimum potential was
subtracted from the maximum. The time period was chosen to
contain the entire CAP and exclude the stimulus artefact.
E. Simulations
A semi-infinite nerve was modelled with a length of 60 mm
(in the z-direction) and simulated using 64 FEMS developed
by SimNeurex LLC (Gainesville, FL). The nerve contained
Page 3 of 11
http://mc.manuscriptcentral.com/tnsre-embs
Transactions on Neural Systems and Rehabilitation Engineering
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
SUBMITTED TO IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING 4
TABLE I
DESCRIPTION OF THE FIVE POSITIONS THAT THE ELECTRODES THAT DELIVERED THE CONDITIONING AND TEST PULSES COULD BE LOCATED IN.
Label Description Conditioning Test
origin The conditioning and test pulses are delivered by the same electrode e
n
e
n
90 degrees The conditioning and test pulses are delivered by electrodes that are separated by 90 degrees. They could
be located on the same ring or on adjacent rings.
e
n
e
n+1
180 degrees The conditioning and test pulses are delivered by electrodes that are separated by 180 degrees. They
could be located on the same ring or on adjacent rings.
e
n
e
n+2
270 degrees The conditioning and test pulses are delivered by electrodes that are separated by 270 degrees. They
could be located on the same ring or on adjacent rings.
e
n
e
n+3
0 degrees The conditioning and test pulses are delivered by electrodes that are separated by 0 degrees and are
located on adjacent rings.
e
n
e
n+4
two fascicles based on histology obtained from the proximal
end of the rat sciatic nerve [40]. The larger fascicle was 0.61
mm in diameter while the smaller fascicle was 0.35 mm. Both
fascicles were modelled as an endoneurium contained within a
perineurial sheath that was equal to 3% of the fascicle diameter
[41]. A nerve cuff electrode array was centered on the nerve.
The cuff electrode array was modelled as a silicone sleeve
with inner diameter of 1 mm and 4.25 mm in length, with a 0.2
mm thick wall, approximating the array used in experiments.
A total of 16 platinum electrodes were included, simuting the
used cuff electrode array. Adjacent rows of electrodes were
0.75 mm apart. The diameter of each electrode was 100 µm.
The arc-length of each electrode was varied from 0.2 to 0.5
mm in 0.1 mm steps (Figure 1E). A 1 mA cathodic current was
applied to each electrode independently. The fields generated
by multiple electrodes were summed. The nerve-cuff complex
was centred in a saline volume measuring 100 × 100 × 200
mm
3
. The outer borders of the saline were set as sinks.
Electrical conductivities of all materials can be found in [18].
Using the DC Conduction solver with a stopping threshold
of 0.5% error and an adaptive mesher, each model required
approximately 5-10 minutes and 100,000-200,000 tetrahedra
to converge to a solution. The potential (voltage) field within
each fascicle was exported to MATLAB. The exported fields
were used to linearly interpolate the extracellular potential
along axons. Specifically, 1000 axons were randomly posi-
tioned within each fascicle. Each axon contained 41 nodes of
Ranvier. The diameter of the axons ranged from 4 to 15 µm
with a bimodal distribution with peaks at 4 and 9 µm [42].
We simulated the axons once the extracellular potential was
interpolated along the randomly positioned and sized axons.
The double cable axon model was used [43]. This model
is based on a mammalian motor axon, rather than sensory
fibers. For the experiments we used a pulse amplitude that
was 120% of the threshold. For the simulations we assumed
that at threshold 10% of the axons fired an action potential.
Thus, we first determined the stimulation threshold for every
axon. Stimulation thresholds were then sorted in ascending
order. The threshold was determined as the pulse amplitude
required to generate an action potential in 10% of the axons.
We then simulated stimulus pulses at 120% of the threshold.
The time delays between each stimulus pulse was stepped from
0 to 10 ms in steps of 1 ms. This sweep was repeated for
every combination of the 16 electrodes at the five possible
angles, within the 4 families of electrode lengths, and for
every axon, totalling 1024 possible combinations for each
axon. Additionally, axons were simulated with only one active
electrode to isolate the timing characteristics of an action
potential produced by that electrode during post-processing.
All simulations were run at the Ohio Supercomputer Center
[44]. Each combination of parameters was run in parallel on
a 28-core machine and required three hours of wall-time per
electrode combination for a total of 768 hours of computation
time. Symmetry of the electrodes within the cuff array allowed
us to eliminate half of the simulations, considering only the
combinations of electrodes in the first and second row with
any of the other electrodes. To reduce the amount of storage
from the simulation results, the voltage at each time point was
only stored for every fifth node of Ranvier.
F. Analysis of Simulated Data
The output from the simulations comprised the voltage
against time data for every axon and electrode parameter
combination. The voltage values represented the extracellular
voltage at every 5th node of Ranvier in 5 µs time steps for
the duration of the stimulus. In the case where the pulse
amplitude was below threshold, a zero was stored to indicate
that no action potentials would have occurred. In the case
where an action potential was generated in an axon (from here
on called an active axon), the peak voltage was stored. This
analysis was repeated for every axon for each of the electrode
combinations. To investigate the effect of a conditioning pulse
on the response of the fibers, we counted the number of active
axons in response to the second (test) pulse and compared
this to the number of active axons when a single pulse was
delivered on the same electrode, i.e. the one that delivered the
test pulse, without a condition pulse. The ratio between active
axons with and without a conditioning pulse was calculated
for each time delay.
III. RESULTS
The effect of the spatio-temporal interactions between elec-
trodes on the recovery of the response of sensory fibers was
investigated both experimentally and with modelling. In all
four animals the CAPs were recorded at the L
4
dorsal root
in response to stimulation of the sciatic nerve. A paired-pulse
paradigm was used to investigate the recovery of the response
of sensory fibers, where 11 different temporal spacings and 5
different spatial positions were examined in all animals except
Animal 1, where the origin spatial condition was not tested.
Page 4 of 11
http://mc.manuscriptcentral.com/tnsre-embs
Transactions on Neural Systems and Rehabilitation Engineering
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60