of 9
Supplemental Document
Optical phased array neural probes for
beam-steering in brain tissue: supplement
W
ESLEY
D. S
ACHER
,
1,2,3,
,†
F
U
-D
ER
C
HEN
,
2,3,†
H
OMEIRA
M
ORADI
-C
HAMEH
,
4
X
INYU
L
IU
,
1
I
LAN
F
ELTS
A
LMOG
,
2
T
HOMAS
L
ORDELLO
,
2
M
ICHAEL
C
HANG
,
4
A
ZADEH
N
ADERIAN
,
4
T
REVOR
M.
F
OWLER
,
1
E
RAN
S
EGEV
,
1
T
IANYUAN
X
UE
,
2
S
ARA
M
AHALLATI
,
4
T
AUFIK
A. V
ALIANTE
,
4,5,6
L
AURENT
C. M
OREAUX
,
1
J
OYCE
K. S.
P
OON
,
2,3
AND
M
ICHAEL
L. R
OUKES
1
1
Division of Physics, Mathematics, and Astronomy, California Institute of Technology, Pasadena,
California 91125, USA
2
Department of Electrical and Computer Engineering, University of Toronto, 10 King’s College Rd.,
Toronto, Ontario M5S 3G4, Canada
3
Max Planck Institute of Microstructure Physics, Weinberg 2, 06120, Halle, Germany
4
Krembil Research Institute, Division of Clinical and Computational Neuroscience, University Health
Network, Toronto, Ontario, Canada
5
Division of Neurosurgery, Department of Surgery, Toronto Western Hospital, University of Toronto,
Toronto, Ontario, Canada
6
Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, Ontario, Canada
Corresponding author:
wesley.sacher@mpi-halle.mpg.de
These authors contributed equally to this work.
This supplement published with Optica Publishing Group on 18 February 2022 by The Authors
under the terms of the Creative Commons Attribution 4.0 License in the format provided by the
authors and unedited. Further distribution of this work must maintain attribution to the author(s)
and the published article’s title, journal citation, and DOI.
Supplement DOI: https://doi.org/10.6084/m9.figshare.17185418
Parent Article DOI: https://doi.org/10.1364/OL.441609
Optical phased array neural probes
for beam-steering in brain tissue:
supplemental document
1. OPTICAL PHASED ARRAY BEAM PROFILE SIMULATIONS
We applied the beam propagation method (BPM) to simulate the optical scattering in tissue [
1
].
The model simulates the forward propagation of a beam in tissue with an iterative two-step
process: 1) a two-dimensional phase mask is applied to the propagating beam, the phase mask
has a small phase variance about the mean phase, which is equal to 0, and 2) beam propagation
between the phase masks is assumed to only diffract, and it is modeled with the angular spectrum
method [2]. The scalar electric field,
E
, is given by
E
(
k
x
,
k
y
,
z
+
d
) =
E
(
k
x
,
k
y
,
z
)
e
in
k
2
k
2
x
+
k
2
y
d
,
(S1)
where
d
is the propagation distance,
n
is the average refractive index of the medium, and
k
is
the wavenumber (
k
x
and
k
y
are
x
and
y
components of the corresponding wavevector). The
amplitude of any wave with
k
2
x
+
k
2
y
>
k
is set to 0 as it is evanescent. The model is applicable
to brain tissue since the beam is mostly forward scattered and its simulation results have been
validated with the analytical solutions from the radiative transfer equation [1].
The patterns of the phase masks determine the scattering properties of the medium. We
followed the design strategy in [
1
] to create the phase mask patterns. Each phase mask introduced
a spatially varying random phase to the fields with the statistical properties governed by
σ
p
and
σ
x
.
σ
p
adjusted the variance of the phase introduced by each phase mask, which affected the
scattering coefficient. Each phase mask was smoothened by a Gaussian filter with variance
σ
x
to control the phase correlation between pixels, which affected the anisotropy value. We also
multiplied the beam profile with the attenuation factor after each propagation step. For the
scattering simulation, we set the scattering coefficient (
μ
s
) to 200
cm
1
, the anisotropy of the
tissue (
g
) to 0.83, and the attenuation coefficient (
μ
a
) to 0.62 cm
1
[3].
The field 3
μ
m
above the optical phased array (OPA) was calculated and used as the launch
field for the above propagation simulations. The SiO
2
cladding thickness above the OPA was 1
μ
m. To calculate the launch field, the emission field from a single grating emitter array element of
the OPA was first obtained (3
μ
m
above the grating) using finite difference time domain (FDTD)
simulations, with the simulations performed for transverse magnetic (TM) polarized light in the
waveguides. Then an array of single grating emitter field profiles was generated via laterally
shifting the position of each subsequent emitter by the array pitch and applying phase shifts to
each field profile corresponding to the wavelength and delay line parameters of the array element.
Finally, the OPA field was the sum of the arrayed field profiles. This process was repeated for
each simulated wavelength. The nominal OPA dimensions reported in the manuscript were used
for the simulations. In addition, we corrected for the amplitude distribution across the output
ports of the OPA star coupler by multiplying the emitter field profiles by an envelope function.
We set the simulation domain volume to be
500
μ
m
×
500
μ
m
×
250
μ
m
to avoid any edge effects
while containing the entire simulation in the computer memory.
The same simulation model was used to study the OPA beam profiles in a non-scattering
medium (water) with all phase masks set to a uniform value of 0. For verification, we compared
cross-sections of beam profiles obtained from the BPM simulations and the conventional diffrac-
tion integral method. No significant differences were observed between the two methods at
propagation distances of up to 200
μ
m.
After the BPM simulations, we performed a series of steps to determine the 2D beam profile
that would be expected from a top-down microscope as in the experimental apparatus in Fig. 1(e).
First, we stacked the beam profile cross-sections to form a 3D model of the beam in tissue. We
then rotated the model so that the beam propagation axis was parallel with the horizontal plane.
Bicubic spline interpolation was used to interpolate the pixel values on the new grid of the rotated
coordinate system. Lastly, a 2D beam profile was obtained by filtering and adding together the
transverse planes of the 3D beam profile model, i.e., all planes parallel to the microscope focal
plane. To approximately emulate the microscope resolution limit, we applied a Gaussian filter to
each transverse plane with filter size equal to the Gaussian propagation beam waist,
w
(
z
) =
w
0
1
+
(
λ
z
π
w
2
0
)
2
,
(S2)
where
w
0
is the point spread function of the objective used in the experimental apparatus [Fig.
1(e)], and
z
is the distance of the transverse plane from the focal plane. For each simulation, the
transverse plane with the highest intensity at the input to the scattering medium was selected as
the focal plane.
The simulated and measured top-down beam profile FWHM values and peak-to-background
ratios in Figs. 2, 3, and S2 were calculated along concentric arcs centered on the OPA emitting
region; the radius of each arc was equal to the propagation distance. For the FWHM beam width
calculations, the maximum was simply the maximum intensity along the arc, while for peak-to-
background ratio calculations, to reduce the impact of noise, the peak intensity was defined as
the average of the top 1% of intensity data points along the arc. The background intensity for
the peak-to-background ratio calculations was defined as the intensity in the troughs between
adjacent lobes. As the OPA beam profiles typically had 3 lobes (Figs. 2 - 3), the background
intensity was calculated considering only the higher of the 2 troughs between the 3 lobes. The
trough intensity was calculated as the average of the lowest 1% of intensity data points along the
arc in the trough. The arcs spanned the central lobe and the two adjacent troughs.
2. EXPERIMENTAL APPARATUS
The scanning system in Fig. 1(a) is detailed in [
4
], with the exception that a wavelength-tunable
laser was used here rather than a fixed wavelength laser. Briefly, light from the wavelength-
tunable laser (TOPTICA Photonics Inc., DLC DL pro tunable laser system with integrated optical
isolator, fiber coupler, motorized wavelength tuning, 484.3 - 491 nm wavelength tuning range)
was coupled into a single-mode fiber (460-HP, Nufern Inc.), which was connected to an inline fiber
polarization controller. The laser light was launched into free space using a fiber collimator, and
this free-space laser beam was gated by a mechanical shutter, directed through a variable neutral
density filter (for control of optical power), and input into the scanning system, as shown in Fig.
1(a). The scanning system included a 2-axis micro-electro-mechanical system (MEMS) mirror
(A7B2.1-3600AL, Mirrorcle Technologies Inc.), two biconvex lenses with 35- and 150-mm focal
lengths, and a 20
×
objective lens (Plan Apochromat, 20-mm working distance, 0.42 numerical
aperture, Mitutoyo Corporation). Actuation of the MEMS mirror enabled addressing of individual
cores of the image fiber bundle (Fujikura FIGH-06-300S). The fiber bundle was optically coupled
to and packaged together with the OPA neural probe using the method described in [
4
]. The
packaged probe was attached to a 4-axis micro-manipulator (QUAD, Sutter Instrument Company)
for immersing the probe in the fluorescein solution (10
μ
M concentration, pH
>
9) and inserting
the probe into the brain slices. Since the OPA beam profiles were polarization-dependent, the
fiber bundle was fixed in position during the experiments (to avoid polarization fluctuations due
to movement of the fiber bundle). The input light to the neural probe chip was TM-polarized.
The maximum optical power available at the input to the scanning system was about 2 mW,
and the loss of the scanning system (measured from the free-space input of the scanning system
to the distal facet of the fiber bundle) was typically 40 - 60
%
(with the neutral density filter
set to its minimum loss). The transmission of the OPA neural probe chip (from the facet of an
optimally aligned single-mode fiber to the free-space OPA output) was typically about -20 dB.
As described in [
4
], the edge couplers accounted for roughly 10 dB of this loss, and improved
edge coupler designs may greatly improve the optical transmission of OPA neural probes. In
addition, deviations of the fiber bundle core positions from a regular pitch and fiber misalignment
during the optical packaging procedure resulted in significant variations in the coupling efficiency
between the fiber bundle cores and the edge couplers of the neural probe. As a result, the total
transmission of the OPAs (from the input to the scanning system to the free-space OPA output)
varied from about -23 dB to -40 dB, with the majority of the OPAs having transmissions between
about -23 to -33 dB.
The experimental apparatus used for OPA beam characterization and
in vitro
testing of the neu-
ral probe [Figs. 1(e) and 5(a)] included a Nikon Eclipse FN1 upright epifluorescence microscope
2
with an sCMOS camera (Zyla 4.2 PLUS, Andor Technology Ltd.) and a 10
×
objective lens (Plan
Apochromat, 34-mm working distance, 0.28 numerical aperture, Mitutoyo Corporation). The
images captured by the microscope were inverted (i.e., vertically mirrored). No image processing
to correct the image inversion was applied, and the beam profile images [Figs. 2(a), 3(b), and 4(a)]
and epifluorescence brain slice images [Figs. 5(b), 5(d), and S6(a)] are inverted. For the fluorescein
and Thy1-GCaMP6s mouse brain slice imaging, an EGFP filter cube (49002, Chroma Technology
Corporation) was used in the epifluorescence microscope, and for the VGAT-ChR2-EYFP mouse
brain slice imaging, an EYFP filter cube (49003, Chroma Technology Corporation) was used. The
camera exposure time was: 50 ms for the fluorescein beam profile images in Fig. 2(a), 500 ms for
the
in vitro
brain slice beam profile images in Fig. 3(b), and 25 ms for the
in vitro
brain slice calcium
imaging in Fig. 4. The microelectrode array (MEA) was a perforated design with 60 titanium ni-
tride electrodes, 30
μ
m
electrode diameter, and 100
μ
m
electrode pitch (60pMEA 100/30iR-Ti-pr-6
mm high plastic ring, Multi-Channel Systems). MEA electrical activity recordings were performed
using a MEA-1060-Up-BC amplifier and the MC Rack software (Multi-Channel Systems). The
sampling rate for the MEA recordings was 25 kHz.
3. BRAIN SLICE PREPARATION
All experimental procedures described here were reviewed and approved by the animal care
committees of the University Health Network in accordance with the guidelines of the Canadian
Council on Animal Care. Brain slices were prepared from 40 - 80 days old VGAT-ChR2-EYFP
(The Jackson Laboratory, stock number 014548) and Thy1-GCaMP6s (The Jackson Laboratory,
stock number 025776) mice for the
in vitro
beam profile/optogenetic stimulation and the
in vitro
calcium imaging experiments, respectively; the brain slice preparation is detailed in [
4
]. 350
μ
m
thick sagittal slices from the cerebellum were used for the
in vitro
beam profile and optogenetic
stimulation experiments (Figs. 3 and 5, Visualization 2). A
300
450
μ
m
thick horizontal slice
from hippocampus was also tested during the beam profile experiments (Visualization 3), but the
labeling was significantly more non-uniform compared to the slices from the cerebellum. The
calcium imaging experiment used a
350
450
μ
m
thick horizontal slice from the hippocampus. For
the optogenetic stimulation and imaging experiments, brain slices were transferred to the MEA
chamber, Figs. 1(e) and 5(a), and perfused with a constant flow of rodent artificial cerebrospinal
fluid (ACSF) [
5
], which was continuously aerated with carbogen. During imaging of the Thy1-
GCaMP6s mouse brain slice, KCl was added to the ACSF to increase the excitability of the neurons
and the amount of spontaneous neuronal activity; the KCl concentration in the ACSF was 30 mM.
4. OPTOGENETIC STIMULATION PROTOCOL
For the single-OPA optogenetic stimulation experiment shown in Fig. 5(b,c), 10 optical pulses
with a pulse width of 30 ms and a period of 200 ms were applied to the brain slice. A recovery
period of 10 s was used after each stimulation pulse train. We repeated the pulse train 10 times.
For the multi-OPA optogenetic stimulation experiments shown in Figs. 5(e) and S6, multiple
“stimulation trials” were performed, and for each trial, illumination was applied to the brain slice
from one of the 4 OPAs in Figs. 5(d) and S6. Each trial corresponds to a point in Figs. 5(e) or S6,
and the optical power was varied between trials. Each stimulation trial consisted of 3 sets of 10
optical pulses; the optical pulse width was 50 ms, the period was 200 ms, and the recovery period
between each set of 10 optical pulses was 10 s. 7-8 trials with different optical power settings were
performed for each OPA by using the variable neutral density filter in the experimental apparatus
to reduce the input optical power to the scanning system in steps of approximately 10-20
%
of the
maximum power. The trials were performed in the following order: 1) the optical power was
set to the maximum value, 2) a trial was performed for each OPA sequentially (from OPA 1 to
4), 3) the optical power was reduced by a 10-20
%
increment, 4) the next set of OPA trials was
performed. This process was repeated until 7-8 trials were performed for each OPA; at the lowest
power setting,
<
1 spike per optical pulse (on average) was observed. This procedure ensured
that the stimulated electrical response caused by different OPAs could not have been simply due
to variations in optical power between the OPAs. The recovery period between trials of the same
optical input power but different OPAs was typically 20-60 s. The recovery period between trials
where the input power was changed was 2 minutes and limited by the time required to adjust the
variable neutral density filter.
3
5. SPIKE SORTING OF MICROELECTRODE ARRAY RECORDINGS
Spike sorting of the microelectrode array recordings was performed to analyze the data from the
optogenetic stimulation (Figs. 5 and S6). The full electrical traces from each experiment, spanning
all optical pulse trains and the recovery periods between them, were analyzed. We performed
spike sorting with the Spyking Circus package [
6
]. We first selected the electrode channels where
neuronal activity was detected in response to the optical stimulation. A bandpass filter with a
passband from 300 to 3000 Hz was applied to the signal from each electrode channel. Then, any
negative spike with amplitude larger than
6
×
the mean absolute deviation was selected as a valid
spike. Each spike waveform was extracted from a 3 ms time window centered at the spike peak.
Then, to extract spike templates, a subset of 10000 spikes was selected; for data sets with
<
10000
spikes, all spikes were selected. The selected spikes were projected to a lower dimension using
Principal Component Analysis. We selected the first 5 prominent components as the basis. Then
a density-based spike clustering algorithm was applied to cluster spikes with similar waveforms
[
6
]. The median of the spike waveform in the same cluster was defined as the spike template of
the cluster. Template matching was performed to decompose all spikes as a linear combination of
the templates, addressing the problem of overlapping spikes [
6
]. Lastly, we performed a manual
inspection of the sorted spikes in the phy GUI interface [
7
]. We selected the clusters that meet the
following four criteria: 1) isolation distance > 10, 2) likelihood ratio < 0.3, 3) signal to noise ratio
(SNR) > 2 [
8
10
], and 4) the percentage of spikes with interspike intervals < 2 ms was less than 2%
when only considering the spikes in the optical stimulation windows. Only sorted spikes within
an optical stimulation pulse window were accounted for in the spike rate calculations.
6. ADDITIONAL NOTES ON MICROELECTRODE ARRAY RECORDINGS
Figure S5 shows our analysis of the spike gain and spike latency for the optogenetic stimulation
experiment in Fig. 5(d,e). We observed a long latency time between the start of each optical
stimulation pulse and the first evoked spikes (>15 ms) and that spikes were evoked up to 300 -
400
μ
m away from the OPAs. These features suggest that most of the spikes we recorded were
likely excited indirectly through synaptic connections to directly stimulated neurons [
9
] and that
we were stimulating a network of neurons.
As our primary goal was to observe the change in spike rate with illumination patterns from the
neural probe, expression of ChR2 in excitatory neurons (rather than inhibitory neurons as in the
VGAT-ChR2-EYFP mice in this work) may have worked to our advantage. The baseline activity
of the brain slices was low, and in our typical experiments, only
<
25%
of the channels recorded
spikes. During optogenetic stimulation, we were likely stimulating a variety of interconnected
inhibitory neurons in the cerebellar slices (including Purkinje cells, Golgi cells, Basket cells and
Stellate cells). Stimulating inhibitory neurons can further suppress activity in connected neurons,
which could have otherwise generated spikes in response to the optical stimulation. In addition,
Purkinje cells and their synapses can span over 100
μ
m [
11
]. The dimensions of these cells is
another possible reason why the experiment in Fig. 5(e) required switching between OPAs 250
μ
m apart to observe repeatable changes in the activity patterns on the MEA.
4