of 16
Patterned photostimulation via
visible-wavelength photonic probes
for deep brain optogenetics
Eran Segev
Jacob Reimer
Laurent C. Moreaux
Trevor M. Fowler
Derrick Chi
Wesley D. Sacher
Maisie Lo
Karl Deisseroth
Andreas S. Tolias
Andrei Faraon
Michael L. Roukes
Eran Segev, Jacob Reimer, Laurent C. Moreaux, Trevor M. Fowler, Derrick Chi, Wesley D. Sacher,
Maisie Lo, Karl Deisseroth, Andreas S. Tolias, Andrei Faraon, Michael L. Roukes,
Patterned
photostimulation via visible-wavelength photonic probes for deep brain optogenetics,
Neurophoton.
4
(1), 011002 (2016), doi: 10.1117/1.NPh.4.1.011002.
Downloaded From: http://neurophotonics.spiedigitallibrary.org/ on 01/03/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
Patterned photostimulation via visible-wavelength
photonic probes for deep brain optogenetics
Eran Segev,
a,b
Jacob Reimer,
c
Laurent C. Moreaux,
b
Trevor M. Fowler,
a,b
Derrick Chi,
a,b
Wesley D. Sacher,
a,b
Maisie Lo,
d
Karl Deisseroth,
d,e
Andreas S. Tolias,
c
Andrei Faraon,
a,f
and Michael L. Roukes
a,b,
*
a
Kavli Nanoscience Institute, California Institute of Technology, Pasadena, California 91125, United States
b
California Institute of Technology, Departments of Physics, Applied Physics, and Bioengineering, 1200 East California Boulevard, MC149-33,
Pasadena, California 91125, United States
c
Baylor College of Medicine, Department of Neuroscience, One Baylor Plaza, Suite S553, Houston, Texas 77030, United States
d
Stanford University, Department of Bioengineering, Stanford, West 250, Clark Center, 318 Campus Drive West, California 94305, United States
e
Stanford University, Howard Hughes Medical Institute, Department of Psychiatry and Behavioral Sciences, West 083, Clark Center, 318 Campus
Drive West, Stanford, California 94305, United States
f
California Institute of Technology, Departments of Applied Physics and Medical Engineering, 1200 East California Boulevard, MC107-81,
Pasadena, California 91125, United States
Abstract.
Optogenetic methods developed over the past decade enable unprecedented optical activation and
silencing of specific neuronal cell types. However, light scattering in neural tissue precludes illuminating areas
deep within the brain via free-space optics; this has impeded employing optogenetics universally. Here, we report
an approach surmounting this significant limitation. We realize implantable, ultranarrow, silicon-based photonic
probes enabling the delivery of complex illumination patterns deep within brain tissue. Our approach combines
methods from integrated nanophotonics and microelectromechanical systems, to yield photonic probes that are
robust, scalable, and readily producible
en masse
. Their minute cross sections minimize tissue displacement upon
probe implantation. We functionally validate one probe design
in vivo
with mice expressing channelrhodopsin-2.
Highly local optogenetic neural activation is demonstrated by recording the induced response
both by
extracellular electrical recordings in the hippocampus and by two-photon functional imaging in the cortex of
mice coexpressing GCaMP6.
©
The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution
or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.
[DOI:
10.1117/1.NPh.4.1.011002
]
Keywords: optogenetics; photonic probes; visible photonics.
Paper 16055SSR received Aug. 2, 2016; accepted for publication Oct. 24, 2016; published online Dec. 6, 2016.
1 Introduction
An overarching technological goal in the field of optogenetics is
the development of new methods for stimulating neural circuits
with very high spatiotemporal precision. Ongoing efforts seek
to address large functional ensembles of neurons, i.e.,
brain
circuits,
through realization of tools providing fine enough
resolution to interrogate and control each constituent neuron indi-
vidually and independently. Significant advances in the develop-
ment of excitatory and inhibitory opsins have been made over the
past decade that now permit direct optical control of cellular
processes.
1
To realize the full potential of these technologies,
complementary methods for delivering light with cellular preci-
sion
in vivo
are now essential.
2
,
3
Existing, state-of-the-art ap-
proaches involve the use of spatially patterned light, projected
via free-space optics, to stimulate small and transparent organ-
isms
4
6
or excite neurons within superficial layers of the
cortex.
7
,
8
However, light scattering and absorption in neural tis-
sue, characterized by the optical attenuation length, cause ballistic
light penetration to be extremely short.
9
This makes it impossible
to employ free-space optical methods to probe brain regions
deeper than about
2mm
. This statement holds true even if
we take into account methods for two-photon and three-photon
excitation, and recent efforts made to develop opsins that operate
in the red or near infrared. With these limitations in mind, we
advance here an alternative approach, involving implantable pho-
tonic devices, as the most promising paradigm for delivering and
projecting high-resolution patterned light at
arbitrary depths
and with minimal perturbation in the brain.
We identify five critical requirements for realizing widely
useful, implantable photonic devices, which we term
visible-
wavelength photonic probes
: (i) the probes should provide a
multiplicity of microscale illumination sources (hereafter
emit-
ter pixels,
or
E-pixels
), each individually controllable and
capable of delivering fine illumination, with cellular-scale
cross-section dimensions. Ideally, emission from these micro-
scale E-pixels should have minimal spatial overlap, while col-
lectively covering the entire brain volume of interest. (ii) This
patterned illumination must be delivered with sufficient inten-
sity to activate optogenetic effectors (actuators/silencers) within
the interrogated region. (iii) Associated thermal perturbations of
neural tissue at, or adjacent to, the implanted devices must
minimally affect neural circuits. Recent studies show that tem-
perature elevation of as small as 1°C can change the neural firing
rate and behavior of mice.
10
(iv) The cross-sectional dimensions
of the probes must be made as small as possible
to reduce dis-
placement of brain tissue upon implantation, to minimize tissue
damage, and to suppress potential immunological response.
11
(v) Finally, photonic nanoprobe fabrication should be compat-
ible with, and ultimately transferrable to, foundry (factory)-
based methods for mass production. This will permit wide
*Address all correspondence to: Michael L. Roukes, E-mail:
roukes@caltech
.edu
Neurophotonics
011002-1
Jan
Mar 2017
Vol. 4(1)
Neurophotonics 4(1), 011002 (Jan
Mar 2017)
Downloaded From: http://neurophotonics.spiedigitallibrary.org/ on 01/03/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
deployment of this new technology in the near-term to the
neuroscience community. Here, we present a class of photonic
probes satisfying these requirements; they are based on inte-
grated, silicon-based nanophotonic components adapted to
operate at visible wavelengths and embedded onto implantable
silicon probes patterned by microelectromechanical systems
(MEMS) processes.
2 Implantable Photonic Neural Probes
Various architectures for implantable optical probes have
recently been proposed.
12
For example, one approach relies
upon multiple optical fibers to excite individually addressed illu-
mination points, each driven by a dedicated laser source.
13
,
14
Given the complexity of coupling many fibers to a probe,
this approach is capable of providing only a few illumination
points. To surmount this issue, coupling a fiber-bundle to on-
chip photonic waveguides has been proposed,
15
but neither
in vitro
nor
in vivo
validation of this particular approach has
been reported. Another approach implements modal multiplex-
ing to address several illumination points along an implantable
multimode optical fiber.
16
However, this approach necessitates a
rather large distance between illumination points (
>
200
μ
m
)to
avoid overlap between adjacent illumination beams. Neither
approach is readily upscalable to many emission points, nor
easily produced
en masse
.
An alternative approach involves the integration of micro-
scale light emitting diodes (
μ
LEDs
) directly onto the probe
shanks.
17
19
A variation on this theme integrates laser diodes
upon the probe head,
20
22
with their light output routed by
on-chip integrated photonic waveguides to emission points
located along the shanks. In both cases, however, the power dis-
sipated by these active
μ
LED
devices must be strictly limited,
given that neuronal activity thresholds are highly sensitive to
minute temperature variations.
10
,
23
Minimizing the total heat
delivered to brain tissue by the probe, which is dominated by
the heat generated by the
μ
LEDs
in these architectures, can sig-
nificantly restrict the number of active illumination sources that
can be integrated. Unless the efficiency of
μ
LED
or laser diode
sources is dramatically increased, it will not be feasible to
include more than a limited number of active light emitters
on implantable photonic probes.
Here, we present a new paradigm for photonic probes that
employs wavelength division multiplexing (WDM).
24
It pro-
vides the potential for massive upscaling of the number of
E-pixels that can be incorporated and individually addressed
within implantable, ultra-compact neural probes. The technique
of WDM employs a multiplicity of independent data streams,
each imprinted on individual carrier wavelengths (spectral chan-
nels), that are combined (i.e., multiplexed) and transmitted via
a single optical fiber. At the receiving end, these multispectral
signals are subsequently demultiplexed and delivered to their
intended destinations. In our application of WDM, each tempo-
rally modulated carrier wavelength is delivered to an indepen-
dent E-pixel at a specific, spatial location located along an
implantable photonic probe shank. Spectral separation is
achieved by photonic circuitry for WDM integrated within
the probe head. Our technique is exceptionally well suited
for optogenetic effectors, because currently employed opsins
respond to a relatively broad spectrum of light, typically span-
ning
50 nm
.
25
,
26
This permits accommodating many spectral
channels within the opsin absorption band. Additionally, this
unique assignment of different wavelengths to specifically
located E-pixels can be accomplished solely using
passive
components, which neither requires power to operate, nor gen-
erate additional heat.
The photonic neural probes described herein provide a first
proof-of-concept of our paradigm. The prototype devices we
report here comprise implantable shanks, initially with nine
E-pixels, which are spectrally addressed through
one
sin-
gle-mode optical fiber. We implement the E-pixels themselves
using large diffractive grating couplers that produce beams with
low divergence angles, as small as 1.7 deg. This low-light diver-
gence offers beam cross-section dimensions that are comparable
to the size of neural cell bodies
even after traversing several
hundreds of micrometers. Other recent implementations of
implantable probes based on photonic technology
27
do not pro-
vide a route toward the goal of generating complex illumination
patterns with narrow illumination beams at arbitrary locations
within the brain.
3 Probe Architecture and Fabrication
The overall structure of our prototype photonic probes is pat-
terned using standard nanophotonic and MEMS fabrication
processes (Appendix
A
, Fig.
6
). The implantable, needle-like
probe shanks have widths of
90
μ
m
near the probe head
decreasing to only
20
μ
m
near the tip, with a uniform thick-
ness of
18
μ
m
throughout. The shank tips are wedge-shaped
[Fig.
1(b)
], with tips having a
1
μ
m
radius of curvature;
this ensures smooth penetration of brain tissue with minimal
dimpling.
28
Our approach yields implantable probes with overall
cross sections representing the state-of-the-art for optical probes.
They are far smaller than the optical fibers or endoscopes
currently implanted for optogenetic experiments (Appendix
A
,
Fig.
9
). The shanks of the prototype probes reported here have
a pitch of
200
μ
m
and lengths of either 3 or 5 mm, yet they
remain straight after fabrication through our careful engineering
of the ubiquitous internal stresses present within thin-film
multilayers (Appendix
A
).
The probe head [Fig.
1(a)
] contains integrated nanophotonic
circuitry required to couple multispectral light delivered from
a single external optical fiber onto the probe chip and, sub-
sequently, to route the individual spectral components (chan-
nels) to specific emitters on the shank(s). E-pixels arrays
[Fig.
1(b)
] can be placed at any location along the implantable
shanks; in the first prototypes reported here, we include nine
E-pixels, spaced on a
200-
μ
m
pitch. It is straightforward to
achieve
50
μ
m
spacing between adjacent E-pixels without
changing our fabrication protocols
29
(Appendix
A
).
4 Nanophotonic Circuitry
The visible-wavelength photonic circuitry on the probe is
fabricated from an optical multilayer comprising a 200-nm
thick silicon nitride (
Si
3
N
4
) layer encapsulated between two
layers of silicon dioxide (
SiO
2
), to yield a total thickness of
2.8
μ
m
. This multilayer is grown upon a commercially available
silicon-on-insulator (SOI) substrate, itself comprising a
15-
μ
m
-thick Si (structural) layer atop, a
2-
μ
m
buried oxide
(BOX) layer, above a
300-
μ
m
-thick Si wafer. The photonic
circuit is comprised of grating couplers, photonic waveguides,
and arrayed waveguide gratings (AWGs). In Appendix
A
,we
describe a microprism coupling method (hereafter,
μ
-prism)
bridging the external input fiber
s terminus to the on-chip gra-
ting coupler. This efficiently couples the light to the photonic
waveguides on-chip. Once on-chip, this multispectral light is
Neurophotonics
011002-2
Jan
Mar 2017
Vol. 4(1)
Segev et al.: Patterned photostimulation via visible-wavelength photonic probes...
Downloaded From: http://neurophotonics.spiedigitallibrary.org/ on 01/03/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
routed by a single waveguide to an AWG located on the probe
head [Fig.
1(a)
]. The AWGs function as passive optical demul-
tiplexers that spectrally separate the incoming wavelength-mul-
tiplexed signal to the nine output waveguides. These separated
post-AWG signals are subsequently routed by separate photonic
waveguides onto the shank, and then to their termini at individ-
ual E-pixels. These E-pixels (described in Appendix
B
) com-
prise small-footprint diffractive grating couplers patterned on
the surface of the shanks [Fig.
2(a)
inset]; light routed to each
is emitted off-shank, almost perpendicularly, into adjacent neu-
ral tissue. Our prototype devices require a single optical fiber per
AWG or shank. Future designs will incorporate a hierarchical
on-probe photonic circuit, in which a master AWG drives sub-
sequent AWGs, to reduce the total number of required optical
fibers to one.
The critical integrated photonic elements on the probes
the
AWGs and the grating couplers
require spatiotemporally
coherent light for their operation. We drive them with multispec-
tral light generated and modulated off-probe, and then delivered
to the probe head by a single external optical fiber. The ratio of
the incident power delivered by the fiber, to the total power
emitted by the E-pixels, defines the probe insertion loss (IL).
The total IL of these first unoptimized prototypes is about
20 dB
. Roughly,
16 dB
of this arises from coupling loss
into and out of the probe, dominated by nonideal coupling
between the fiber and the on-chip photonic circuitry. The various
losses present are fully delineated in Appendix
B
, Fig.
11
.In
future device generations, these ILs can be reduced significantly
through advanced engineering design and, especially, by use of
the highly optimized fabrication processes available at commer-
cial photonic foundries.
30
We emphasize that the majority of
these losses, 18 dB in our current prototypes occurs within the
probe base, rather than at the point of emission, as is the case
for
μ
LEDs
.
Validation of the capability of our photonic probes to stimu-
late neural activity, as described in the next section, has been
achieved with E-pixel emission power that ranged between
5to
10
μ
W
. With IL of 20 dB, incident laser power of 1 mW
per E-pixel is required. Such power is readily available over
the relevant wavelength range with supercontinuum lasers.
5 Characterization of Single E-Pixel
Illumination
Our measurement and simulations results demonstrate that
E-pixels emit beams with a propagation direction angle of
2 to 30 deg from the normal to the probe surface [Fig.
2(a)
].
The exact angle of each individual E-pixel can be engineered
during the probe design phase by setting the period of the gra-
ting couplers. Once probes are fabricated, this angle is fixed.
The low divergence of the beams minimizes overlap between
adjacent beams, while preserving light intensity over significant
propagation distances from the E-pixel. We have capitalized on
the highly collimated photonic probe beamshape to enable local
optogenetic activation of neurons.
The beam profile at the surface of the E-pixel is <
6
μ
m
(FWHM) along both transverse axes of the beam [Fig.
2(a)
inset]. We characterize the beam profile versus distance from
probe shank by: (i) imaging in a fluorescein solution [Figs.
2(b)
and
13
], (ii) imaging in,
300-
μ
m
thick, adult mouse brain
slices soaked overnight in a fluorescein solution [Fig.
2(c)
],
and (iii) comparison with numerical simulations [Figs.
2(d)
,
14
, and
15
]. We find Fresnel diffraction determines the beam
intensity profile up to a distance of
70
μ
m
from the probe;
beyond that, it is characterized by far-field Fraunhofer diffrac-
tion. (Appendix
B
; Fig.
15
). The minimal beam width, observed
at the transition between the Fresnel and far-field regions at a
distance of
90
μ
m
[Fig.
2(d)
], is
10
μ
m
in the fluorescein
solution,
17
μ
m
at a distance of
70
μ
m
in the brain slice,
and <
5
μ
m
at distances smaller than
100
μ
m
in our simulations.
Light scattering results in a slightly larger beam divergence in
tissue [Fig.
2(c)
] than in the fluorescein solution [Fig.
2(b)
].
However, all beam widths measured up to a distance of
200
μ
m
are less than, or of order, the size of an individual neuronal cell
body. This property of E-pixel illumination permits reducing
the E-pixel pitch to
50
μ
m
, while still maintaining negligible
overlap between adjacent beams.
6 Multibeam Illumination
Our use of WDM makes it possible to independently address on-
shank E-pixels by separate temporally modulated multispectral
components of the light delivered to the probe. Figure
3(a)
shows an illustration of our WDM approach. Coherent light
from a broadband (multispectral) source is split into
N
discrete
Fig. 1
Prototype photonic probe architecture. (a) Optical micrograph
showing photonic probes operating at visible wavelengths. This
specific design contains three 3-mm-long,
18
-
μ
m-thick shanks that
taper in width from
90
μ
m, at the head, down to
20
μ
m, near the tip.
The photonic elements comprise three AWG demultiplexers; one
per shank. Each is driven by a single-input waveguide (left) and
subsequently drives a multiplicity of output waveguides (right) that
traverse the shanks, carrying light to their ultimate destinations on
the shank tips. At their termini, the photonic waveguides drive grating
couplers (termed E-pixels), which couple light off-shank into brain tis-
sue. All the on-chip photonic elements are patterned from a 200-nm
thick silicon nitride layer, which is deposited on top of the oxidized
silicon structural layer used to form the probe body. (b) Photograph
showing the top view of two
10
μ
m
×
10
μ
m grating couplers that con-
stitute the E-pixels near the tip of a shank. The tapered waveguides
transform the small optical cross section of the submicron waveguides
to the larger neuronal-scale spot size delivered by the E-pixels.
(c) Side view of a photonic probe. Although the shank thickness is
18
μ
m, the thicker
320
μ
m probe head (on left) facilitates handling
and mounting of the device.
Neurophotonics
011002-3
Jan
Mar 2017
Vol. 4(1)
Segev et al.: Patterned photostimulation via visible-wavelength photonic probes...
Downloaded From: http://neurophotonics.spiedigitallibrary.org/ on 01/03/2017 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx