Supplementary Information for:
Neural coding during active somatosensation revealed using illusory touch Daniel H. O’Connor1, *,†, S. Andrew Hires1, *, Zengcai V. Guo1, Nuo Li, Jianing Yu, Qian-Quan
Sun1, #, Daniel Huber1, %, Karel Svoboda1
1
Janelia Farm Research Campus, HHMI, Ashburn VA 20147
†
Current address: The Solomon H. Snyder Department of Neuroscience & Brain Science
Institute, The Johns Hopkins University School of Medicine, Baltimore, MD 21205
#
Current address: Department of Zoology and Physiology, University of Wyoming, Laramie,
Wyoming 82071
%
Current address: Department of Basic Neurosciences, University of Geneva, CH-1211 Geneva,
Switzerland.
*
These authors contributed equally to this work.
Nature Neuroscience: doi:10.1038/nn.3419
YES
Lick
(‘yes’)
b
θ at touch onset
c
Non-whisking
60
40
20
0
−20
2
0
60
0
Counts / deg Occupancy (s)
Number of sessions
10
0
30
20
−1
NO
YES
NO
YES
0
Exploration bias
NO
No lick
(‘no’)
Whisker θ (deg)
a
NO
YES
θROI
1
sFigure 1 | Whisking strategy underlying the object location discrimination task.
sFigure 1
a. Schematic of the basic object location discrimination task. On NO trials the object was
further away from the resting position of the whisker and required withholding a lick
response (i.e., a ‘no’ response). On YES trials the object was closer and required a lick
response (a ‘yes’ response).
b. Whisking during object location discrimination. Left, whisker θ at touch onset (295
touches; YES trials, blue; NO trials, red; one example session, different from Fig. 1c).
Right, distribution of whisker positions during non-whisking (black; whisking amplitude,
θamp < 2.5 degrees; Methods) and whisking periods (θamp > 2.5 degrees). Occupancy (in
seconds) is the time spent at a particular θ (bin size, one degree).
c. Exploration bias for 148 behavioral sessions. This measures the preference of the θROI for
one pole position compared to the other. The two pole positions were defined as the
distribution of θ at touch onset for all exploration period touches, normalized to 1. These
pole position distributions were multiplied by the whisker occupancy during active
whisking. These two products (p1, p2) were combined (p1-p2)/(p1+p2) to generate the
exploration bias for the behavioral session. Mice preferentially explored one pole location
(exploration bias ≠ 0). Exploration bias less than 0 indicates a tendency to explore the NO
location. Exploration bias greater than 0 indicates a tendency to explore the YES
location.
Nature Neuroscience: doi:10.1038/nn.3419
a
d
% DR/R0
0.1
IS image
AAV-FLEX-ChR2
Layer 4 Cre mouse
Vascular structure
0
-0.1
250 μm
Scnn1a / Six3
hSyn1
b
Cre
250μm
ChR2:tdTomato
e
Layer 4, C2-column ChR2 expression
250 μm
250 μm
100 µm
f
c
‘C2’
‘E3’
E3
Cell Attached
Si Probe
C2
C2
200 µm
200 µm
sFigure
2
sFigure 2 | Targeting layer 4 of the C2 barrel column for ChR2 expression and
recording.
a. Viral gene delivery targeted by intrinsic signal imaging. Intrinsic signal image (left) taken
through the intact skull and used to localize the C2 (or E3) barrel column next to an
image of the vascular landmarks (middle). Subsequent to intrinsic signal imaging-based
mapping, Cre-dependent virus is injected into the relevant barrel in L4 Cre mice (right).
b. Analysis of ChR2 expression. Left, flattened and cytochrome oxidase-stained section
through barrel cortex, showing ChR2-tdTomato fluorescence in the C2 barrel column.
Right, a coronal section showing ChR2-tdTomato fluorescence (magenta) and nuclei
(DAPI, in blue).
Nature Neuroscience: doi:10.1038/nn.3419
c. Labeling accuracy for AAV injections targeted to C2 (red) or E3 (yellow). The contour
lines encompass the labeled somata. Dotted lines show the best estimate for the two cases
in which there was some ambiguity about barrel identity due to imperfect histology. The
barrel map schematic is based on data from 55.
d. Reconstructing recording locations. Flattened and cytochrome oxidase-stained section
through barrel cortex, overlaid with DiI fluorescence introduced by a cell-attached
recording electrode. Estimated L4 cell locations (2) marked with red arrows.
e. Composite of two flattened and cytochrome oxidase-stained sections through barrel
cortex with DiI overlay from silicon probe recording. Estimated neuronal locations (3)
are marked with red arrows.
f. Location of recordings during object location discrimination behavior (Figs. 1, 2).
Nature Neuroscience: doi:10.1038/nn.3419
a
b
c
50 mW
50 mW
20
trials
5 mV
33 mW
33 mW
L2/3
10 ms
L4
L5
20 mW
VPM
20 mW
10 ms
0.5 50 mW
0
1
0.5 33 mW
0
1
0.5 20 mW
0
-10
0
10
20
Time from pulse onset (ms)
f
6
L5
5
L2/3
4
3
2
L4
100 66 40
20
Relative power (% max)
sFigure 3 | Calibration of ChR2-based photostimulation.
Evoked spikes / pulse
Evoked spikes / pulse
1
Latency from light onset (ms)
e
d
0.8
0.6
0.4
L2/3
L4
0.2
0
L5
100 66 40 20
Relative power (% max)
sFigure 3
a. Schematic of the recording configuration. Loose cell-attached recordings were made
from neurons in all cortical layers (n=85 neurons total; N=10 mice). L6 neurons did not
produce evoked spikes in response to photostimulation (cf. Fig. 4c) and they are not
shown. Responses to photostimulation of L4 neurons were measured.
b. Example action potential responses from a single L4 neuron to short pulses of blue light
at different power levels (50, 33, and 20 mW).
c. Spike raster showing trials from the neuron in b.
d. Peristimulus time histogram for the neuron shown in b.
e. Latency of evoked spikes as function of power, across cortical layers. Plot symbols show
the mean across neurons. Because of the trade-off between maximum intensity and pulse
duration, to pool neurons we occasionally report “Relative power” as a percentage of the
maximum power (see Methods).
f. Number of evoked spikes as function of power, across cortical layers. Plot symbols show
the mean across neurons. Error bars show bootstrap SEM across neurons. L2/3 activity
Nature Neuroscience: doi:10.1038/nn.3419
appears less sparse than expected based on previous in vivo imaging and
electrophysiology experiments 12. It is possible that differences in the recruitment of
feedforward inhibition between photostimulation and natural touch may underlie the
differential recruitment of L2/3 neurons. For example, cross-whisker inhibition is
expected in multi-whisker behavior 12 and may promote sparseness in L2/3 responses. It
is also possible that the somewhat more synchronous excitation of L4 by
photostimulation vs. natural touch (Fig. 4) may drive more efficient excitation of L2/3
neurons.
Nature Neuroscience: doi:10.1038/nn.3419
sFigure 4 | Reliability of L4 ChR2-positive neuron responses during naturalistic
photostimulation.
a. Extracellular voltage traces from an L4 neuron via loose-seal cell attached recording
during naturalistic photostimulation. Photostimulation trains were chosen from a set of
whisker crossings obtained from the object location discrimination task (in which mice
determined the pulse train pattern by their whisking; Methods).
b. Evoked spike latency as a function of interstimulus interval for individual cells (grey) and
the population mean (black, n=6 cells).
c. The probability of photostimulation evoking a spike as a function of interstimulus
interval. Color conventions as in panel b.
Nature Neuroscience: doi:10.1038/nn.3419
d. The probability of photostimulation evoking a spike as a function of the sequential
photostimulus number within a train (first, second, etc). Color conventions as in panel b. Nature Neuroscience: doi:10.1038/nn.3419
NO
YES
No lick
(‘no’)
Lick
(‘yes’)
b
θ at first touch Non-whisking
60
YES
40
20
NO
0
Whisker θ (deg)
a
-20
# of sessions
6 4 2 0
YES
NO
6 4 2 0
Occupancy (s)
d
C2, ChR2+
Fooling Index
c
−1
0
1
Exploration bias
Anterior
θROI
300 200 100 0
Contacts/deg
0.3
0.2
0.1
0
−0.1
sFigure 5 | Illusory touch occurs robustly when the YES location is anterior and the NO
location is posterior.
sFigure 5
a. Schematic of the experiment and the four trial types. As in Fig. 5, except that the YES
and NO locations were switched with respect to the experiments of Fig. 5.
Correspondingly, the θROI was anterior, as mice chose to focus whisking on the rewarded
location, even though it was further from the resting position of the whisker.
b. Quantification of whisker occupancy and contact probability, as in Fig. 1, sFig. 1b. .
Left, whisker θ at touch onset (1,490 touches). Black line shows whisker occupancy
when the mouse is not whisking (i.e. the ‘resting position’ of the whisker). Occupancy (in
seconds) is the time spent at a particular θ (bin size, one degree).
c. Quantification of whisking exploration bias, as in sFig. 1c.
d. Photostimulation increases the fraction of yes responses in NO trials. Error bars, SEM.
Nature Neuroscience: doi:10.1038/nn.3419
b
NO
YES
Left lick
(‘no’)
Right lick
(‘yes’)
θ at first touch
40
Non-whisking
NO
20
0
YES
4
0
80
40
0 8
Counts / deg Occupancy (s)
c
NO
# of sessions
6 4 2 0
−1
0
YES
1
Exploration bias
θROI
-20
Whisker θ (deg)
a
NO
YES
NO
YES
θROI
sFigure 6
sFigure 6 | Whisking strategy in the symmetric response task is the same as in the go/no-go
task.
a. Symmetric response task; both object locations were indicated by licking at one of two
lickports (lick left / lick right).
b. Whisking strategy in the symmetric response task. Same units as Fig. 1c.
c. Exploration bias for 18 symmetric response sessions (see Methods). Mice preferentially
explored one pole location (exploration bias ≠ 0).
Nature Neuroscience: doi:10.1038/nn.3419
a
b
Pure synthetic touch
Fraction yes response
Trial count (X 1000)
1
8
6
4
2
0
Stim: Touch: -
+
-
+
0.8
0.6
0.4
0.2
0
Stim:
Touch:
+
+
- +
- -
NO trials
NO trials
Fraction yes responses
1
0.8
0.6
0.4
0.2
0
0
Trials w/o photostim
2
4
6
Contact number
YES trials
1
0.8
0.6
0.4
0.2
0
Trials w/o touch
0
5
10
15
Photostimulus number
0.2
0.1
0
0.5
1
Fraction NO trials with touch
Hit rate minus false alarm rate
f
0.3
r = −0.63
p = 0.013
Fooling Index
- +
++
d
Fraction yes responses
c
e
- +
+ +
C2, ChR2-
1
0.5
0
−0.5
1
10
20
30
Number of 1 ms pulses at 40 Hz
sFigure 7
Nature Neuroscience: doi:10.1038/nn.3419
40
sFigure 7 | Interactions between illusory and real touch.
a. We examined interactions between activity evoked by photostimulation and touch,
capitalizing on trial-to-trial variability in behavior and the large numbers of trials in our
experiments (mean, 3007 trials over 9 sessions per mouse). We sorted behavioral trials
into four groups based on the presence of touch contact (yellow explosion) and/or
photostimulation (blue bolt). The histogram shows the number of trials of each type, for
NO trials.
b. Fraction of yes responses depending on touch and photostimulation. Each line
corresponds to one mouse. Black lines, lick / no lick task; green lines, symmetric
response task. Real touch on NO trials increased ‘yes’ responses (p<0.001, paired onetailed t-test). In trials without contact, photostimulation also drove ‘yes’ responses
(p<0.001), to levels comparable to NO trials with contact (p=0.099). However, in NO
trials with contact, photostimulation had less additional effect (p=0.002, paired one-tailed
t-test on difference in photostimulation effect). These data indicate that photostimulation
and touch occlude each other, and thus provide evidence that real touch and illusory
touch are perceptually similar.
c. Rapid saturation of perception-driven choice with the number of stimuli (contacts or
photostimuli) might underlie the occlusion shown in panel b. On trials without
photostimulation, the fraction of ‘yes’ responses on NO trials showed a large increase
with the first contact, but only small additional increases thereafter. Fraction of yes
responses therefore saturates rapidly with the number of touches. Error bars, STD.
d. On trials without touch, photostimulation-driven ‘yes’ responses showed equally rapid
saturation (compared with touch-driven ‘yes’ responses shown in panel c). Error bars,
STD.
e. Across-animal analysis of the Fooling Index vs. the fraction of NO trials with touch
during the exploration window. Animals that touched more often were less fooled, due to
occlusion of fooling by touch.
f. Detection of light pulse trains over the C2 column is difficult or impossible for a sham
infected (ChR2-) mouse. Light detection (75 mW) psychometric-style curve for a highly
trained ChR2-negative mouse. Thin lines show individual sessions, thick lines show the
mean. Mouse was a Tg(Etv1-cre)GM225Gsat mouse with a ‘sham’ AAV injection; virus
was injected in the C2 column but the mouse was confirmed ChR2-negative, for
unknown reasons, by histology.
Nature Neuroscience: doi:10.1038/nn.3419
a
θ at 1st stim
- 5 ms
2
40
θamp
Occupancy
Trials
60
1
20
0
*
d
‘Delayed’ light
-20 0 20 40
Whisker θ (deg)
10 deg
0
θ
b
200 ms
‘Shuffled’ light
stim
10°
0.5 s
θ
θamp = 18 deg
25 deg
12 deg
10 deg
stim
< θamp > = 18 + 25 + 12 deg
3
Trials
Crossings, trial N-5
Light, trial N
e
0.1
0
N
N−5
Light source trial
Trial count
Fooling Index
50
0.3
0.2
f
40
30
20
10
0
0
5
10
stim
< θamp >
15
20
Answer window
1
{
c
Cum. frac. of light pulses
θ
20 ms
0.8
0.6
0.4
0.2
Not whisking
Whisking
Virtual pole
0
−1
0
1
2
Time from start of pole movement (s)
sFigure 8 | Illusory touch does not require precisely timed or sequenced spikes, but only
occurs only during whisking
sFigure
8 6) with Δt ≥ 20 ms (- 5 ms,
a. Distribution of θ at first stimulus for delayed light
trials (Fig.
to correct for the difference in the delay between photostimulus and spike versus touch
and spike; cf. Fig. 4). Red line, whisker occupancy.
b. We next determined if the precise sequence of inter-photostimulation intervals mattered
for illusory touch. We interleaved the standard experiment (Δt = 5 ms) with trials in
which a photostimulus pattern corresponded to a whisking pattern measured in a previous
trial (N-5,where N is the current trial; ‘shuffled’ light). Under these conditions the
statistical features of the photostimulus pattern are matched across the different types of
stimulated trials, but the patterns of whisker crossings does not predict the pattern of
photostimuli within a trial.
c. On the shuffled photostimulation trials mice were still fooled, although the effect was
smaller than for the standard experiment (Fooling Index 0.12 vs. 0.20; 1 of 3 mice p <
Nature Neuroscience: doi:10.1038/nn.3419
0.05, two-tailed permutation test). These experiments show that illusory touch does not
require L4 activity to match the precise pattern of virtual pole crossings.
d. Schematic showing the measurement of mean θamp at the time of stimulation. Blue circles
indicate light pulse times. Asterisk indicates response lick time.
e. Mean θamp at the time of stimulation. Red line shows the criterion (θamp; 2.5 deg) used to
separate trials into ‘whisking’ and ‘not whisking’.
f. Distribution of light pulses across times within the trial, for trials with different
categorizations. ‘Virtual pole’ trials refer to the standard Δt = 5 ms trials. ‘Answer
window’ is the time in which licks were scored as responses.
Nature Neuroscience: doi:10.1038/nn.3419
a
Ignored
Activity
Tactile
exploration
θROI
Spike count in L4
Spike count, L4
d
c
Probability
b
Time
Object location
sFigure 9 | Schematic of how L4 spike count codes for object location.
sFigure 9
a. Mice report perception of active touch with an object when neural activity occurs during
epochs of tactile exploration, and within somatotopic locations matching the moving
body part (the C2 whisker). Thus, perceptual reports of touch are gated by sensory
expectation.
b. Objects at different locations (grey scaled circles) with respect to the whisking region of
interest (θROI) will produce different numbers of whisker-object contacts and a different
pattern of forces/moments, with more numerous contacts and higher average and peak
forces/moments for closer (lighter grey) objects.
c. Hypothetical distributions of spike count in L4 for trials where the object is in different
locations (grey scale, matching panel ‘b’). Because of more numerous contacts as well as
higher forces/moments, the spike count in L4 will be higher for objects closer to the
center of the θROI (lighter grey).
d. The distributions in panel ‘c’ yield a monotonic spike count code for object location.
Nature Neuroscience: doi:10.1038/nn.3419
Nature Neuroscience: doi:10.1038/nn.3419
sFigure 10 | Comparison of Scnn1a-Tg3-Cre and Six3Cre mouse lines.
a. Fooling Index for the two mouse lines used in this study. We detected no difference
between the two lines in illusory touch behavior. Therefore, for all analyses we pooled
mice from the two lines. Similarity of behavior is likely due to powerful recruitment of
inhibition in both lines, despite expression of ChR2 in GABAergic cells in the Six3Cre
but not Scnn1a-Tg3-Cre lines. Error bars, SEM.
b. Brain slice recordings from barrel cortex L4 neurons of Six3Cre and Scnn1a-Tg3-Cre
mice expressing AAV2/5-hSyn1-FLEX-hChR2-tdTomato. Recorded neurons were
ChR2-negative but within the ChR2-expressing barrel. EPSCs (grey) and IPSCs (black)
were recorded in response to wide-field blue light illumination using whole-cell voltage
clamp at -70 mV and 0 mW, respectively, using a cesium-based internal solution.
Recordings were conducted at room temperature.
c. The excitation/inhibition ratio, measured by integrated current (charge transfer), was
larger in Scnn1a-Tg3-Cre mice, in which Cre expression does not occur in GABAergic
neurons, compared with Six3Cre mice, in which GABAergic neurons do express Cre
(one-tailed permutation test). Error bars, SEM.
d. The inhibition onset latency is longer in Scnn1a-Tg3-Cre mice (one-tailed permutation
test). Error bars, SEM.
Nature Neuroscience: doi:10.1038/nn.3419