a
GCL
b
IPL
INL
OPL
ONL
OS
RPE
Supplementary Figure 1: Histological comparison of healthy and degenerate rat retina.
(a) Histological cross-section of a healthy rat retina. Photoreceptor outer segments (OS) are in
contact with the retinal pigment epithelium (RPE). Photoreceptor somas located in the outer
nuclear layer (ONL) transmit neural signals to cells in the inner nuclear layer (INL) via synapses located in the outer plexiform layer (OPL). From there, signals are relayed to the ganglion
cell layer (GCL) via synapses in the inner plexiform layer (IPL). (b) Histological cross-section
of a P140 RCS rat retina. The OS, ONL and OPL have degenerated and the INL is now in
contact with the RPE. The INL, IPL and GCL are preserved to a large extent. Scale bar: 50
µm.
Nature Medicine: doi:10.1038/nm.3851
a
d
implant outline
400 µV
100
Time (ms)
e
800
400
200
0
f
−200
−1200
c
−800
−400
0
PC1 (a.u.)
0.4
0.4
OFF cells
0.3
0.2
0.1
0
0
100
200
300
400
Time after stimulation pulse (ms)
500
100
200
300
400
Time after stimulation pulse (ms)
500
0.3
0.2
0.1
0
0
0.9
0.8
#spikes/reversal/5 ms bin
PC2 (a.u.)
600
#spikes/trial/5ms bin
50
#spikes/trial/5ms bin
0
b
RGC vRF
ON cells
200 µV
0.7
0.6
0.5
axonal tract
0.4
0.3
soma
dendritic arbor
0.2
0.1
0.0
0
100
200
300
400
500
600
Time (ms)
700
800
900
Supplementary Figure 2: Data processing. (a) We estimated electrical stimulation artifacts
by averaging over many trials (black dashed trace), aligning and pointwise subtracting from
the raw recordings (orange). Artifact removal is often imperfect over the duration of the light
pulse (shaded) therefore we blanked data in software, resulting in the purple trace. Action
potentials are subsequently detected by thresholding prior to (b) undergoing principal component analysis (PCA) and expectation-maximization clustering. (c) Putative neural activity is
correlated with the recorded electrical activity over the implant, thereby creating electrophysiological images (EIs) of the ganglion cells. Triangulation of this voltage signature localizes
RGCs over the recording array. We discarded from the analysis cells exhibiting abnormal EIs,
indicative of improper spike sorting (for example, RGCs with multiple axons). (d) Finally, we
correlated the firing patterns with the stimulus. White noise analysis shows receptive field
mosaics underneath the implant for both ON and OFF cells, an indicator that the retina was in
good physiological condition. (e) For eRF mapping, we recorded responses from each RGC
(yellow dot) to each pixel stimulation. (f) For alternating grating measurements, spiking
patterns are correlated with the grating contrast reversal. The peristimulus time histogram
shown here exhibits response to both phases of the grating, characteristic of non-linear
summation in subunits of the RGC receptive field.
Nature Medicine: doi:10.1038/nm.3851
1000
a
t = 10 ms
15 ms
20 ms
25 ms
30 ms
35 ms
45 ms
55 ms
65 ms
b
c
Supplementary Figure 3: Time course of the eRFs. (a) The most commonly observed type
of eRF had a localized component with latency 20 – 45 ms. (b) Some eRFs had a localized
component with a shorter (< 15 ms) latency. (c) For eRFs with center and diffuse components, the central localized response had a short latency (< 15 ms), and the diffuse latency
was 20-45 ms. In all figures, the red dot indicates the estimated position of the RGC soma on
the array (see Supplementary Fig 2c). Scale bar: 500 µm.
Nature Medicine: doi:10.1038/nm.3851
a
Fraction of cells (a.u.)
3.0
2.0
1.0
0.0
b
0
50
100
150
200
50
100
150
200
Fraction of cells (a.u.)
3.0
2.0
1.0
0.0
0
Grating stripe width at threshold (µm)
Supplementary Figure 4: Distribution of the responses to alternating gratings projected
with visible light. Histograms and kernel density estimates of the stimulation thresholds for
alternating gratings in two different preparations. (a) The peak in the threshold distribution
occurs at 28 µm (n = 163 RGCs). (b) The peak in the threshold distribution occurs at 48 µm
(n = 115 RGCs).
Nature Medicine: doi:10.1038/nm.3851
a
b
30 µV
Corneal potential
100 ms
c
eVEP
7.5 µV
60%
contrast
50 ms
40%
contrast
200 µm
gratings
I < 30%
150 µm
gratings
frame 1
frame 2
signal
Supplementary Figure 5: Responses to grating alternations are not due to a luminance
misbalance. (a) Illustration of the two phases of the large gratings overlapping with the
implant. The total stimulation current was assessed via a corneal electrode, thereby comparing the stimulus misbalance (ΔI) between the two phases (b) In all conditions, even with the
large gratings, this misbalance was below 30%. To verify whether this change in the total
current could elicit responses to the alternating gratings, we recorded the VEP responses to
full-field stimulation with contrast modulation (c) Animals responded to 60% contrast, but did
not respond to 40% contrast. Therefore, strong responses to the gratings with 200 µm and
narrower stripes could not be due to the luminance misbalance, but rather originate from the
spatial modulation of the stimulus.
Nature Medicine: doi:10.1038/nm.3851
a
2.5 µV
implant
stimulus
b
150 µm
gratings
10 µV
Photovoltaic
Visible
Visible
NIR
50 ms
WT rat
50 ms
RCS rat
Supplementary Figure 6: Control experiments. (a) Lack of VEP responses to NIR stimulation
outside of the implant (red) and a strong response to the visible light in the same location in a
WT animal (blue). (b) Lack of VEP response to alternating grating with 150 µm/stripe projected with visible light onto the implant in RCS rat (blue) and a robust response to patterns
projected with NIR illumination (red).
Nature Medicine: doi:10.1038/nm.3851
Normalized VEP amplitude
1
0.8
Polynomial
Gaussian
Gamma
0.6
0.4
0
50 64 µm
100
Grating size (µm)
150
200
Supplementary Figure 7: Visual acuity estimation. We obtained the acuity value by calculating the intersection of a fit of the data points and the noise level. For prosthetic visual
acuity, we used 75, 100, 150 and 200 µm data points. We defined the noise level as the
amplitude of the signal in response to alternations of the grating with 25μm/stripe, which are
not resolved. A second order polynomial fitting yielded the most conservative estimate of the
acuity among the different functions tested (sigmoidal, Gaussian and polynomial, see Methods). We calculated the uncertainty of the measurement from the fitting parameter covariance and the uncertainty in the noise level (see Methods).
Nature Medicine: doi:10.1038/nm.3851
Type
Receptive fields
In vitro gratings
In vitro frequency
In-vivo full field
In-vivo acuity
WT
n = 4, 92 cells
n = 2, 278 cells
n = 2, 178 cells
n=9
n=7
RCS
n = 3, 48 cells
n = 4, 109 cells
n = 2, 45 cells
n=3
n=7
Table 1: Data summary. Number of cells and animals contributing to each experiment.
Nature Medicine: doi:10.1038/nm.3851